Women in Commutative Algebra

Association for Women in Mathematics Series
Claudia Miller
Janet Striuli
Emily E. Witt Editors
Women in
Commutative
Algebra
Proceedings of the 2019 WICA Workshop
Volume 29
Series Editor
Kristin Lauter
Facebook
Seattle, WA, USA
Focusing on the groundbreaking work of women in mathematics past, present, and
future, Springer’s Association for Women in Mathematics Series presents the latest
research and proceedings of conferences worldwide organized by the Association
for Women in Mathematics (AWM). All works are peer-reviewed to meet the highest
standards of scientific literature, while presenting topics at the cutting edge of pure
and applied mathematics, as well as in the areas of mathematical education and
history. Since its inception in 1971, The Association for Women in Mathematics
has been a non-profit organization designed to help encourage women and girls
to study and pursue active careers in mathematics and the mathematical sciences
and to promote equal opportunity and equal treatment of women and girls in
the mathematical sciences. Currently, the organization represents more than 3000
members and 200 institutions constituting a broad spectrum of the mathematical
community in the United States and around the world.
Titles from this series are indexed by Scopus.
More information about this series at https://link.springer.com/bookseries/13764

Claudia Miller • Janet Striuli • Emily E. Witt
Editors
Women in Commutative
Algebra
Proceedings of the 2019 WICA Workshop
Editors
Claudia Miller Janet Striuli
Department of Mathematics Department of Mathematics
Syracuse University Fairfield University
Syracuse, NY, USA Fairfield, CT, USA
Emily E. Witt
Department of Mathematics
University of Kansas
Lawrence, KS, USA
ISSN 2364-5733 ISSN 2364-5741 (electronic)

ISBN 978-3-030-91985-6 ISBN 978-3-030-91986-3 (eBook)
https://doi.org/10.1007/978-3-030-91986-3
Mathematics Subject Classification: 13-XX, 14-XX, 18-XX
© The Author(s) and the Association for Women in Mathematics 2021

This work is subject to copyright. All rights are solely and exclusively licensed by the Publisher, whether
the whole or part of the material is concerned, specifically the rights of translation, reprinting, reuse
of illustrations, recitation, broadcasting, reproduction on microfilms or in any other physical way, and
transmission or information storage and retrieval, electronic adaptation, computer software, or by similar
or dissimilar methodology now known or hereafter developed.
The use of general descriptive names, registered names, trademarks, service marks, etc. in this publication
does not imply, even in the absence of a specific statement, that such names are exempt from the relevant
protective laws and regulations and therefore free for general use.
The publisher, the authors and the editors are safe to assume that the advice and information in this book
are believed to be true and accurate at the date of publication. Neither the publisher nor the authors or
the editors give a warranty, expressed or implied, with respect to the material contained herein or for any
errors or omissions that may have been made. The publisher remains neutral with regard to jurisdictional
claims in published maps and institutional affiliations.
This Springer imprint is published by the registered company Springer Nature Switzerland AG
The registered company address is: Gewerbestrasse 11, 6330 Cham, Switzerland
Preface
This volume is the result of research activities that took place during the first
workshop for Women in Commutative Algebra at Banff International Research
Station for Mathematical Innovation and Discovery. The workshop brought together
researchers from a diverse list of institutions and fostered research collaborations
among women at different career stages and from different research backgrounds.
This volume has several purposes. First and foremost, it is a celebration of the
high-level research activities that took place at the workshop. Further it wants to
testify to the successful model behind the high research achievements, a model
put in place by several fields in mathematics and, with Emily E. Witt, brought
to commutative algebra by the hard work of Karen Smith, Sandra Spiroff, Irena
Swanson, to whom we are extremely grateful. We have intended this volume
to support and expand the goal of the workshop in re-enforcing the network of
collaborations among women in commutative algebra by displaying in one place
those connections, and bringing in new ones. The volume indeed contains articles
or research advances that were made by the research groups, as well as some
contributions related to the area of commutative algebra and, survey articles.
Commutative algebra is the study of the properties of rings that historically rose
in algebraic and arithmetic geometry. With the development of several techniques
and a rich theory, commutative algebra has become a thriving research area that
feeds to and from several fields of mathematics such has topology and combina-
torics, beyond the classical algebraic and arithmetic geometry. It would not be fair
not to mention that significant advances have been made recently in the field with
the the breakthrough of new techniques in positive and mixed characteristic methods
and homological algebra, and the consequent solution of long-standing conjectures.
The volume reflects the ripple effect of such breakthroughs that have inspired
a great deal of activities in commutative algebra. Our volume delivers readings
that span from case studies to survey articles and cover a wide range of topics
in commutative algebra. The study of characteristic p rings is present in this
volume through the classification of Frobenius forms in certain dimension, the
Frobenius singularities of certain varieties, and through a comprehensive survey
of the Hilbert–Kunz function; further, the reader can find results of a homological
v
vi Preface
flavor in the articles that deliver resolutions of powers of the homogeneous maximal
ideal of graded Koszul algebras, the construction of the truncated free resolution
for the residue field, or the notion of Tor-independent modules; finally, the volume
contains several articles that expand on the connection between homological and
combinatorial invariants.
Syracuse, NY, USA Claudia Miller

Fairfield, CT, USA Janet Striuli
Lawrence, KS, USA Emily E. Witt
Acknowledgments
We would like to thank Emily E. Witt’s co-organizers, Karen Smith, Sandra Spiroff,
and Irena Swanson, of the Women in Commutative Algebra workshop at the Banff
International Research Station for gathering such a great group of 3 researchers
vii
Contents
On Gerko’s Strongly Tor-independent Modules . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1

Hannah Altmann and Keri Sather-Wagstaff
Properties of the Toric Rings of a Chordal Bipartite Family of Graphs . . . 11
Laura Ballard
An Illustrated View of Differential Operators of a Reduced
Quotient of an Affine Semigroup Ring. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 49
Christine Berkesch, C-Y. Jean Chan, Patricia Klein, Laura Felicia
Matusevich, Janet Page, and Janet Vassilev
A Hypergraph Characterization of Nearly Complete Intersections . . . . . . . . 95
Chiara Bondi, Courtney R. Gibbons, Yuye Ke, Spencer Martin, Shrunal
Pothagoni, and Ada Stelzer
The Shape of Hilbert–Kunz Functions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 111
C-Y. Jean Chan
Standard Monomial Theory and Toric Degenerations of
Richardson Varieties in Flag Varieties . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 165
Narasimha Chary Bonala, Oliver Clarke, and Fatemeh Mohammadi
Simplicial Resolutions for the Second Power of Square-Free
Monomial Ideals. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 193
Susan M. Cooper, Sabine El Khoury, Sara Faridi, Sarah Mayes-Tang,
Susan Morey, Liana M. Şega, and Sandra Spiroff
Cohen–Macaulay Fiber Cones and Defining Ideal of Rees
Algebras of Modules . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 207
Alessandra Costantini
Principal Matrices of Numerical Semigroups . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 233
Papri Dey and Hema Srinivasan
ix
x Contents
A Survey on the Koszul Homology Algebra . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 257

Rachel N. Diethorn
Canonical Resolutions over Koszul Algebras . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 281
Eleonore Faber, Martina Juhnke-Kubitzke, Haydee Lindo, Claudia Miller,
Rebecca R. G., and Alexandra Seceleanu
Well Ordered Covers, Simplicial Bouquets, and Subadditivity of
Betti Numbers of Square-Free Monomial Ideals . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 303
Sara Faridi and Mayada Shahada
A Survey on the Eisenbud-Green-Harris Conjecture . . . . . . . . . . . . . . . . . . . . . . . . 327
Sema Güntürkün
The Variety Defined by the Matrix of Diagonals is F -Pure . . . . . . . . . . . . . . . . . 343
Zhibek Kadyrsizova
Classification of Frobenius Forms in Five Variables . . . . . . . . . . . . . . . . . . . . . . . . . 353
Zhibek Kadyrsizova, Janet Page, Jyoti Singh, Karen E. Smith,
Adela Vraciu, and Emily E. Witt
Projective Dimension of Hypergraphs . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 369
Kuei-Nuan Lin and Sonja Mapes
A Truncated Minimal Free Resolution of the Residue Field . . . . . . . . . . . . . . . . 399
Van C. Nguyen and Oana Veliche
On Gerko’s Strongly Tor-independent
Modules
Hannah Altmann and Keri Sather-Wagstaff
Keywords Differential graded algebras · Semidualizing modules · Syzygies ·

Tor-independent modules
1 Introduction
We are interested in how existence of certain sequences of modules over a local

ring (R, mR ) imposes restrictions on R. Specifically, we investigate what Gerko [6]
calls strongly Tor-independent R-modules: A sequence N1 , . . . , Nn of R-modules
is strongly Tor-independent provided TorR 1 (Nj1 ⊗R · · · ⊗R Njt , Njt+1 ) = 0
for all distinct j1 , . . . , jt+1 . Gerko is led to this notion in his study of Foxby’s
semidualizing modules [5] and Christensen’s semidualizing complexes [3]. In
particular, Gerko [6, Theorem 4.5] proves that if R is artinian and possesses a
sequence of strongly Tor-independent modules of length n, then mnR = 0. (Note that
Gerko’s result only assumes the modules are finitely generated and strongly Tor-
independent, not necessarily semidualizing.) This generalizes readily from artinian
rings to Cohen–Macaulay rings; see Proposition 5.1 below.
Our goal in this paper is to prove the following non-Cohen–Macaulay comple-
ment to Gerko’s result.
Theorem 1.1 Assume (R, mR ) is a local ring. If N1 , . . . , Nn are non-free, strongly
Tor-independent R-modules, then n ecodepth(R).
H. Altmann ()
College of Arts and Sciences, Dakota State University, Madison, SD, USA
e-mail: hannah.altmann@dsu.edu
K. Sather-Wagstaff
School of Mathematical and Statistical Sciences, Clemson University, Clemson, SC, USA
e-mail: ssather@clemson.edu
© The Author(s), under exclusive license to Springer Nature Switzerland AG 2021 1

C. Miller et al. (eds.), Women in Commutative Algebra, Association for Women in
Mathematics Series 29, https://doi.org/10.1007/978-3-030-91986-3_1
2 H. Altmann and K. Sather-Wagstaff
Here ecodepth(R) = β0R (mR ) − depth(R) is the embedding codepth of R, where

β0R (mR ) is the minimal number of generators of mR . Note that our result does not
recover Gerko’s, but compliments it. Our proof is the subject of Sect. 5 below.
Part of the proof of our result is modeled on Gerko’s proof with one crucial
difference: where Gerko works over an artinian ring, we work over a finite dimen-
sional DG algebra. See Sects. 2 and 3 for background material and foundational
results, including our DG version of Gerko’s notion of strong Tor-independence.
Theorem 4.7 is our main result in the DG context, which is the culmination of
Sect. 4. Our proof relies on a DG syzygy construction of Avramov et al. [2].
2 DG Homological Algebra
Let R be a nonzero commutative Noetherian ring with identity. We work with

R-complexes indexed homologically, so for us an R-complex X has differential
∂iX : Xi → Xi−1 . The supremum and infimum of X are respectively
sup(X) = sup{i ∈ Z | Xi = 0} inf(X) = inf{i ∈ Z | Xi = 0}.
The amplitude of X is amp(X) = sup(X) − inf(X). Frequently we consider these

invariants applied to the total homology H (X), e.g., as sup(H (X)).
As we noted in the introduction, the proof of Theorem 1.1 uses DG techniques
which we summarize next. See, e.g., [1, 4] for more details.
A differential graded (DG) R-algebra is an R-complex A equipped with an R-
linear chain map A ⊗R A → A denoted a ⊗ a → aa that is unital, associative, and
graded commutative. We simply write DG algebra when R = Z. The chain map
condition here implies that this multiplication is also distributive and satisfies the
Leibniz Rule: ∂(aa ) = ∂(a)a + (−1)|a| a∂(a ) where |a| is the homological degree
of a. We say that A is positively graded provided Ai = 0 for all i < 0. For example,
the trivial R-complex R is a positively graded DG R-algebra, so too is every Koszul
complex over R, using the wedge product. The underlying algebra associated to A
is the R-algebra A = i∈Z Ai .
If R is local, then a positively graded DG R-algebra A is local provided H0 (A)
is Noetherian, each H0 (A)-module Hi (A) is finitely generated for all i 0, and the
ring H0 (A) is a local R-algebra.
Let A be a DG R-algebra. A DG A-module is an R-complex X equipped with
an R-linear chain map A ⊗R X → X denoted a ⊗ x → ax that is unital and
associative. For instance DG R-modules are precisely R-complexes. We say that X
is homologically bounded if amp(H (X)) < ∞, and we say that X is homologically
finite if H (X) is finitely generated over H0 (A). We write n X for the nth shift of X
n
obtained by ( n X)i = Xi−n and ∂i X = (−1)n ∂i−n X . Quasiisomorphisms between
R-complexes, i.e., chain maps that induce isomorphisms on the level of homology,
are identified with the symbol .
Let A be positively graded and let X be a DG A-module such that inf(X) >
−∞. We say that X is semifree if the underlying A -module X is free. In this case
On Gerko’s Strongly Tor-independent Modules 3
a semibasis for X is a set of homogeneous elements of X that is a basis for X

over A . A semifree resolution of a DG A-module Y with inf(H (Y )) > −∞ is

a quasiisomorphism F − → Y such that F is semifree. The derived tensor product

of DG A-modules Y and Z is Y ⊗L A Z F ⊗A Z where F − → Y is a semifree

resolution of Y . We say that Y is perfect if it has a semifree resolution F −
→ Y such
that F has a finite semibasis.
Let A be a local DG R-algebra, and let Y be a homogically finite DG A-
module. By [2, Proposition B.7] Y has a minimal semifree resolution, i.e., a semifree

resolution F −→ Y such that the semibasis for F is finite in each homological degree
and ∂ F (F ) ⊆ mA F .
3 Perfect DG Modules and Tensor Products
Throughout this section, let A be a positively graded commutative homologically

bounded DG algebra, say amp(H (A)) = s, and assume that A 0.
Most of this section focuses on four foundational results on perfect DG modules.
Lemma 3.1 Let L be a non-zero semifree DG A-module with a semibasis B
concentrated in a single degree n. Then L ∼
= n A(B) . In particular, inf(H (L)) = n
and sup(H (L)) = s + n and amp(H (L)) = s.
Proof It suffices to prove that L ∼ = n A(B) . Apply an appropriate shift to assume
without loss of generality that n = 0.
The semifree/semibasis
assumptions tell us that every element x ∈ L has
the form finite
e∈B ae e; the linear independence of the semibasis tells us that this
representation is essentially unique. Since A is positively graded, we have L−1 = 0,
so ∂ L (e) = 0 for all e ∈ B. Hence, the Leibniz rule for L implies that
finite

finite
finite
finite
∂ L
ae e = ∂ A (ae )e + (−1)|ae | ae ∂ L (e) = ∂ A (ae )e.
e∈B i i i
From this, it follows that the map A(B) → L given by the identity on B is an
isomorphism.

Proposition 3.2 Let L be a non-zero semifree DG A-module with a semibasis B
concentrated in degrees n, n + 1, . . . , n + m where n, m ∈ Z and m 0. Then
inf(H (L)) n and sup(H (L)) s + n + m, so amp(H (L)) s + m.
Proof It suffices to show that inf(H (L)) n and sup(H (L)) s + n + m. We
induct on m. The base case m = 0 follows from Lemma 3.1.
For the induction step, assume that m 1 and that the result holds for semifree
DG A-modules with semibasis concentrated in degrees n, n + 1, . . . , n + m − 1. Set
B = {e ∈ B | |e| < n + m}
and let L denote the semifree submodule of L spanned over A by B . (See the
first paragraph of the proof of [2, Proposition 4.2] for further details.) Note that L
has semibasis B concentrated in degrees n, n + 1, . . . , n + m − 1. In particular,
our induction assumption applies to L to give inf(H (L )) n and sup(H (L ))
s + n + m − 1.
If L = L , then we are done by our induction assumption. So assume that L = L .
Then the quotient L/L is semifree and non-zero with semibasis concentrated
in degree n + m. So, Lemma 3.1 implies that inf(H (L/L )) = n + m and
sup(H (L/L )) = s + n + m. Now, consider the short exact sequence
0 → L → L → L/L → 0. (1)
The desired conclusions for L follow from the associated long exact sequence in
homology.

Now, we use the preceding two results to analyze derived tensor products.
Lemma 3.3 Let L be a non-zero semifree DG A-module with a semibasis B
concentrated in a single degree, say n, and let Y be a homologically bounded DG
AY Y
A-module. Then L⊗L n (B) . In particular, inf(H (L⊗L Y )) = inf(H (Y ))+n
A
and sup(H (L ⊗A Y )) = sup(H (Y )) + n and amp(H (L ⊗L
L
A Y )) = amp(H (Y )).
Proof Immediate from Lemma 3.1.

Proposition 3.4 Let L be a non-zero semifree DG A-module with a semibasis B
concentrated in degrees n, n+1,. . . n+m where n, m ∈ Z and m 0, and let Y be a
homologically bounded DG A-module. Then inf(H (L ⊗L A Y )) inf(H (Y )) + n and
sup(H (L ⊗LA Y )) sup(H (Y )) + n + m, so amp(H (L ⊗ L Y )) amp(H (Y )) + m.
A
Proof As in the proof of Proposition 3.2, we induct on m. The base case m = 0
follows from Lemma 3.3.
For the induction step, assume m 1 and the result holds for semifree DG
A-modules with semibasis concentrated in degrees n, n + 1, . . . , n + m − 1 and
Y ∈ Db (A). We work with the notation from the proof of Proposition 3.2, and we
assume that L = L . The exact sequence (1) of semi-free DG modules gives rise to
the following distinguished triangle in D(A).
L ⊗ L
A Y → L ⊗A Y → (L/L ) ⊗A Y →
L L
Another long exact sequence argument gives the desired conclusion.

We close this section with our DG version of strongly Tor-independent modules.
Definition 3.5 The DG A-modules K1 , . . . , Kn are said to be strongly Tor-
independent if for any subset I ⊂ {1, . . . , n} we have amp(H (⊗L
i∈I Ki )) s.
Remark 3.6 It is worth noting that the definition of K1 , . . . , Kn being strongly Tor-
independent includes amp(H (Ki ))) s for all i = 1, . . . , n. Also, if K1 , . . . , Kn
are strongly Tor-independent, then so is any reordering by the commutativity of

tensor products.
4 Syzygies and Strongly Tor-independent DG Modules
Throughout this section, let (A, mA ) be a local homologically bounded DG algebra,

say amp(H (A)) = s, and assume that A 0 and mA = A+ . It follows that A0 is a
field.
The purpose of this section is to provide a DG version of part of a result of
Gerko [6, Theorem 4.5]. Key to this is the following slight modification of the
syzygy construction of Avramov et al. mentioned in the introduction.
Construction 4.1 Let K be a homologically finite DG A-module. Let F K be a
minimal semifree resolution of K, and let E be a semibasis for F . Let F (p) be the
semifree DG A-submodule of F spanned by Ep := ∪mp Em .
Set t = sup(H (K)), and consider the soft truncation K = τr (F ) for a
fixed integer r t. Note that the natural morphism F → K is a surjective

quasiisomorphism of DG A-modules, so we have K F K. Next, set L = F (r) ,
which is semifree with a finite semibasis Er . Furthermore, the composition π of
the natural morphisms L = F (r) → F → K is surjective because the morphism

F → K is surjective, the morphism L → F is surjective in degrees r, and
Ki = 0 for all i > r. Set Syzr (K) = ker(π ) ⊆ L and let α : Syzr (K) → L be the
inclusion map.
Proposition 4.2 Let K be a homologically finite DG A-module. With the notation
of Construction 4.1, there is a short exact sequence of morphisms of DG A-modules
α
0 → Syzr (K) −
π
→L−
→K→0 (2)
K and Im(α) ⊆ A+ L.
such that L is semifree with a finite semibasis and where K
Proof Argue as in the proof of [2, Proposition 4.2].

Our proof of Theorem 1.1 hinges on the behavior for syzygies documented in the
following four results.
Lemma 4.3 Let K be a homologically finite DG A-module with amp(H (K))
s and K = Syzr (K) where r sup(H (K)). Then sup(H (K )) s + r and
inf(H (K )) r. Therefore, amp(H (K )) s.
Proof Use the notation from Construction 4.1. Then sup(H (L)) s + r by
= sup(H (K)) r r +s.
Proposition 3.2. Also, by definition we have sup(H (K))
The long exact sequence in homology coming from (2) implies sup(H (K )) s +r.
Also, inf(H (K )) inf(K ) r because πi is an isomorphism for all i < r
by Construction 4.1. So, amp(H (K )) = sup(H (K ))−inf(H (K )) s +r −r = s.

Proposition 4.4 Let K be a homologically finite DG A-module and set K =

Syzr (K) where r sup(H (K)). Let Y be a homologically bounded DG A-module
and assume that K, Y are strongly Tor-independent. Then sup(H (K ⊗L A Y ))
sup(H (Y ))+r and inf(H (K ⊗L A Y )) inf(H (Y ))+r. So, amp(H (K ⊗L Y )) s;
A
in particular, K , Y are strongly Tor-independent.

Proof Let G − and L be as in Construc-
→ Y be a semifree resolution of Y . Let K
tion 4.1. As K, Y are strongly Tor-independent we have sup(H (K ⊗A G)) s.
Also, Proposition 3.4 implies sup(H (L ⊗A G)) sup(H (Y )) + r. To conclude
the proof, consider the short exact sequence
⊗A G → 0
0 → K ⊗A G → L ⊗A G → K (3)
and argue as in the proof of Lemma 4.3.

Proposition 4.5 Let K1 , K2 , . . . , Kn be strongly Tor-independent, homologically
finite DG A-modules for n ∈ Z+ and Ki = Syzri (Ki ) where ri sup(H (Ki )). Then
K1 , . . . , Km
,K
m+1 , . . . , Kn are strongly Tor-independent for all m = 1, . . . , n.
Proof Induct on m using Proposition 4.4.

Proposition 4.6 Let K1 , K2 , . . . , Kj be strongly Tor-independent DG A-modules,
and set Ki = Syzri (Ki ) where ri sup(H (Ki )) for i = 1, 2, . . . , j . If mnA = 0,
n−j
then mH (A) H (⊗L
i=1,...,j Ki ) = 0.
Proof Shift Ki if necessary to assume without loss of generality that inf(H (Ki )) =

→ Ki be semifree resolutions, and
0 for i = 1, . . . , j . For i = 1, . . . , j let Gi −
consider the following diagram with notation as in Construction 4.1.
(4)
Notice, Im(αi ) ⊆ Ki ⊆ mA Li for i = 1, 2, . . . , j .

Set G = ⊗i=1,...,j −1 Gi and consider the following commutative diagram
where θ = ⊗i=1,...,j −1 αi and β is induced by θ ⊗ αj .

Claim: H (β) is 1-1. Notice that Hi (G ⊗A Gj ) = 0 for all i < r1 + . . . + rj , so it

suffices to show that Hi (β) is 1-1 for all i r1 + . . . + rj . To this end it suffices to
show Hi (G ⊗ αj ) and Hi (θ ⊗ Lj ) are 1-1 for all i r1 + . . . + rj . First we show
this for Hi (G ⊗ αj ). Consider the short exact sequence
G⊗αj
j → 0.
0 → G ⊗A Gj −−−→ G ⊗A Lj → G ⊗A K (5)
Proposition 4.4 implies
j )) r1 + . . . + rj −1 + sup(H (K
sup(H (G ⊗A K j )) r1 + . . . + rj .
Thus, the long exact sequence in homology associated to (5) implies Hi (G ⊗A αj )

is 1-1 for all i r1 + . . . + rj as desired.
Next, we show Hi (θ ⊗ Lj ) is 1-1 for i r1 + . . . + rj . Consider the exact
sequence
θ⊗L
j
0 → G ⊗A Lj −−−→ (⊗i=1,...,j −1 Li ) ⊗A Lj → (⊗i=1,...,j −1 K i ) ⊗A Lj → 0.
(6)
The first inequality in the next display follows from Proposition 3.4
i ) ⊗A Lj )) sup(H (⊗i=1,...,j −1 K
sup(H ((⊗i=1,...,j −1 K i )) + rj
r1 + . . . + rj −1 + rj .
Thus, the long exact sequence in homology associated to (6) implies Hi (θ ⊗ Lj ) is

1-1 for all i r1 + . . . + rj . This establishes the claim.
n−j
To complete the proof it remains to show mH (A) H ((⊗L L
i=1,...,j −1 Ki )⊗A Kj ) = 0.
L
Since H (β) is 1-1, we have H ((⊗L i=1,...,j −1 Ki ) ⊗A Kj ) isomorphic to a submodule
j n−j
of H (mA ((⊗i=1,...,j −1 Li ) ⊗A Lj )). So it suffices to show that mH (A) annihilates
j
H (mA ((⊗i=1,...,j −1 Li ) ⊗A Lj )); this annihilation holds because mnA = 0.

Here is the aforementioned version of part of [6, Theorem 4.5].
Theorem 4.7 Let K1 , . . . , Kn be strongly Tor-independent non-perfect DG A-mo-
dules. Then mnA = 0, therefore, n s.
Proof Suppose mnA = 0. Proposition 4.6 implies that 0 = m0H (A) H (⊗L
i=1,...,n Ki ) =

H (⊗i=1,...,n Ki ). Since each Ki has a minimal resolution for i = 1, . . . , n, we must
L
have H (Kl ) = 0 for some l. Hence, Kl has a semifree basis concentrated in a

finite number of degrees. This contradicts our assumption that Ki is not perfect for
i = 1, . . . , n. Therefore, mnA = 0.
Now we show n s. Soft truncate A to get A A such that sup(A ) = s. Thus,
s+1
mA = 0. The sequence of n strongly Tor-independent non-perfect DG A-modules
gives rise to a sequence of n strongly Tor-independent non-perfect DG A -modules.

Since mnA = 0 and ms+1
A = 0, we have n s.

5 Proof of Theorem 1.1
Induct on depth(R).
Base Case: depth(R) = 0. Let K denote the Koszul complex over R on a minimal
generating sequence for mR . The condition depth(R) = 0 implies
amp(H (K)) = ecodepth(R) = amp(K). (7)
Claim: The sequence K ⊗LR N1 , . . . , K ⊗R Nn is a strongly Tor-independent

L
sequence of DG K-modules.
To establish the claim we compute derived tensor
products where both L are indexed by i ∈ I :
L L
K (K ⊗L
R Ni ) K ⊗R (
L
R Ni ).
From this we get the first equality in the next display.

L L
amp(H ( K (K ⊗L
R Ni ))) = amp(H (K ⊗R (
L
R Ni )))

= amp(H (K ⊗R ( i∈I Ni )))

amp(K ⊗R ( i∈I Ni ))
= amp(K)
= amp(H (K))
The second equality comes from the strong Tor-independence of the original
sequence. The inequality and the third equality are routine, and the final equality
is by (7). This establishes the claim.
A construction of Avramov provides a local homologically bounded DG algebra
(A, mA ) such that A K 0 and mA = A+ ; see [7, 8]. The strongly Tor-
independent sequence K ⊗L R N1 , . . . , K ⊗R Nn over K gives rise to a strongly
L
Tor-independent sequence M1 , . . . , Mn over A. Now, Theorem 4.7 and (7) imply

n amp(H (A)) = amp(H (K)) = ecodepth(R). This concludes the proof of the
Base Case.
Inductive Step: Assume depth(R) > 0 and the result holds for local rings S with
depth(S) = depth(R)−1. For i = 1, . . . , n let Ni be the first syzygy of Ni . Since the
sequence N1 , . . . , Nn is strongly Tor-independent, so is the sequence N1 , . . . , Nn .

Moreover, strong Tor-independence implies that i∈I Ni is a submodule of a free
R-module, for each subset i ∈ {1, . . . , n}.
Use prime avoidance to find an R-regular element x ∈ mR −m2R . Set R = R/xR.

that depth(R) = depth(R) − 1 and ecodepth(R) = ecodepth(R). The fact that
Note
each i∈I Ni is a submodule of a free R-module implies that x is also i∈I Ni -
regular. It is straightforward to show that the sequence R ⊗L
R N1 , . . . , R ⊗R Nn is
L
strongly Tor-independent over R. By our induction hypothesis we have
n ecodepth(R) = ecodepth(R)
as desired.

We conclude with the generalization of Gerko’s result [6, Theorem 4.5] from
artinian rings to Cohen–Macaulay rings mentioned in the introduction. In prepara-
tion, recall that the Loewy length of a finite length R-module M is
R (M) = min{i 0 | miR M = 0}.
The generalized Loewy length of R is then
R (R) = min{
R (R/x) | x is a system of parameters of R}.
Notice that when R is artinian, i.e., when R has finite length as an R-module, the
generalized Loewy length of R equals the Loewy length of R, so the symbol
R (R)
is unambiguous.
Proposition 5.1 Assume that R is Cohen–Macaulay and that K1 , . . . , Kn are non-
free, finitely generated, strongly Tor-independent R-modules. Then n
R (R).
Proof We induct on d = dim(R). In the base case d = 0, the ring R is artinian,
so Gerko’s result [6, Theorem 4.5] says that mnR = 0. By the definition of Loewy
length, this is exactly the desired conclusion.
For the induction step, assume that d 1, and that our result holds for Cohen–
Macaulay local rings of dimension d − 1. Let x = x1 , . . . , xd be a system of
parameters of R such that
R (R) =
R/x (R/x). Since R is Cohen–Macaulay,

this is a maximal R-regular sequence. Furthermore, the definition of generalized
Loewy length implies that
R/x1 (R/x1 )
R/x (R/x) =
R (R).
Replace the modules Ki with their first syzygies if necessary to assume without
loss of generality that x1 is Ki -regular for i = 1, . . . , n. From this, it is straightfor-
ward to use the assumptions on the Ki to conclude that K1 /x1 K1 , . . . , Kn /x1 Kn are
non-free, finitely generated, strongly Tor-independent R/x1 -modules. Thus, our
induction hypothesis implies that n
R/x1 (R/x1 )
R (R), as desired.
Acknowledgement We are grateful to the anonymous referee for their thoughtful comments.
References
1. L. L. Avramov, H.-B. Foxby, and S. Halperin, Differential graded homological algebra, in

preparation.
2. L. L. Avramov, S. B. Iyengar, S. Nasseh, and S. Sather-Wagstaff, Homology over trivial
extensions of commutative DG algebras, Comm. Algebra 47 (2019), no. 6, 2341–2356, see also
arxiv:1508.00748. MR 3957101
3. L. W. Christensen, Semi-dualizing complexes and their Auslander categories, Trans. Amer.
Math. Soc. 353 (2001), no. 5, 1839–1883. MR 2002a:13017
4. Y. Félix, S. Halperin, and J.-C. Thomas, Rational homotopy theory, Graduate Texts in
Mathematics, vol. 205, Springer-Verlag, New York, 2001. MR 1802847
5. H.-B. Foxby, Gorenstein modules and related modules, Math. Scand. 31 (1972), 267–284
(1973). MR 48 #6094
6. A. A. Gerko, On the structure of the set of semidualizing complexes, Illinois J. Math. 48 (2004),
no. 3, 965–976. MR 2114263
7. A. R. Kustin, Classification of the Tor-algebras of codimension four almost complete intersec-
tions, Trans. Amer. Math. Soc. 339 (1993), no. 1, 61–85. MR 1132435
8. S. Nasseh and S. K. Sather-Wagstaff, Applications of differential graded algebra techniques in
commutative algebra, preprint (2020), arXiv:2011.02065.
Properties of the Toric Rings of a
Chordal Bipartite Family of Graphs
Laura Ballard
1 Introduction
In recent decades, there has been a growing interest in the investigation of algebraic
invariants associated to combinatorial structures. Toric ideals of graphs (and the
associated edge rings), a special case of the classical notion of a toric ideal,
have been studied by various authors with regard to invariants such as depth,
dimension, projective dimension, regularity, graded Betti numbers, Hilbert series,
and multiplicity, usually for particular families of graphs (see for example [2, 3, 5, 7–
10, 12, 14, 16–19, 22, 23, 26]). We note in Remarks 2.6 and 2.14 that the family we
consider does not overlap at all or for large n with those considered in [5, 8, 9], and
[23]; it is more obviously distinct from other families that have been studied. We
think it fitting to mention that the recent book by Herzog et al. [15] also investigates
toric ideals of graphs as well as binomial ideals coming from other combinatorial
structures.
In this work, we consider a family of graphs with iterated subfamilies and develop
algebraic properties of the toric rings associated to the family which depend only on
the number of vertices (equivalently, the number of edges) in the associated graphs.
In the development of this project, we were particularly inspired by the work of
Jennifer Biermann, Augustine O’Keefe, and Adam Van Tuyl in [3], where they
establish a lower bound for the regularity of the toric ideal of any finite simple graph
and an upper bound for the regularity of the toric ideal of a chordal bipartite graph.
Our goal is to construct as “simple” a family of graphs as possible that still yields
The author was partially supported by the National Science Foundation (DMS-1003384).
L. Ballard ()
Mathematics Department, Syracuse University, Syracuse, NY, USA
e-mail: lballard@syr.edu

12 L. Ballard
interesting toric ideals. It is our hope that our process and results will lead to further
generalizations of properties of toric ideals for other (perhaps broader) families of
graphs, or for graphs containing or arising from such graphs.
Herein, we introduce the infinite family F of chordal bipartite graphs Gtn , where
n determines the number of edges and vertices and t determines the structure of the
graph, and establish some algebraic properties of the toric rings R(n, t) associated to
the graphs Gtn . The use of bipartite graphs makes each R(n, t) normal and Cohen–
Macaulay by [25] and [15]; we use the latter in Sect. 3. Our main results prove to be
independent of t and depend only on n.
In Sect. 2, we construct the family F of graphs Gtn from a family of ladder-like
structures Ltn so that the toric ideals of the Gtn are generalized determinantal ideals
of the Ltn . The ladder-like structures associated to a subfamily F1 ⊂ F, introduced
in Example 2.4, are in fact two-sided ladders (for large n), so that the family of rings
R(n, t) is a generalization of the family of ladder determinantal rings coming from
F1 . While the rings arising from F1 come from a distributive lattice and have easily
derived properties (see for example [15]), we show that the rings associated to F do
not naturally arise from any lattice in general, and merit closer study.
In Sect. 3, we establish some algebraic properties of the R(n, t), particularly
Krull dimension, projective dimension, multiplicity, and regularity. To do so, we
prove that the determinantal generators of the defining ideal IGtn are a Gröbner basis
(it follows immediately from [15] that R(n, t) is Koszul) and work with the initial
ideal in> IGtn . We also develop a system of parameters Xn that allows us to work
with Artinian reductions in part of our treatment, and their Hilbert series.
Our first result gives an alternate proof for the Krull dimension of the toric
ring R(n, t) = S(n)/IGtn , already known due to a result of Villarreal for bipartite
graphs [27, Prop 3.2]. Here, the ring S(n) = k[x0 , x2 , x3 , . . . , x2n+3 , x2n+4 ] is the
polynomial ring over the edges of Gtn and IGtn is the toric ideal of Gtn .
Theorem 1.1 (Theorem 3.4) The dimension of R(n, t) is
dim R(n, t) = n + 3.
As a corollary, since R(n, t) comes from a bipartite graph and is hence Cohen–
Macaulay (Corollary 2.16), we obtain the projective dimension of R(n, t).
Corollary 1.2 (Corollary 3.5) The projective dimension of R(n, t) over S(n) is
pdS(n) R(n, t) = n + 1.
We then develop a linear system of parameters for R(n, t), using differences of
elements on antidiagonals of the ladder-like structure Ltn .
Proposition 1.3 (Proposition 3.10) Let R(n, t) = S(n)/IGtn . Then the image of
Xn = x0 , x2 − x3 , x4 − x5 , . . . , x2n − x2n+1 , x2n+2 − x2n+3 , x2n+4

Properties of the Toric Rings of a Chordal Bipartite Family of Graphs 13
in R(n, t) is a system of parameters for R(n, t).

Since R(n, t) is Cohen–Macaulay, the linear system of parameters above is a regular
sequence (Corollary 3.12).
With the aim of obtaining the multiplicity and regularity of R(n, t), we form an
t). We
Artinian quotient of R(n, t) by the regular sequence above and call it R(n,
t) does not denote the completion, and explain the choice of notation
note that R(n,
in Definition 3.7.
t) established in Lemma 3.13, we
Using a convenient vector space basis for R(n,
t).
show the coefficients of the Hilbert series for R(n,
Theorem 1.4 (Theorem 3.16) If R(n, t) = t)
S(n)/IGtn and R(n, ∼
=
R(n, t)/(Xn ), we have
⎧
⎪
⎪ 1 i=0
⎪
⎪
⎪
⎨1 i
t))i =
dimk (R(n, (n + j − 2(i − 1)) 1 i n/2 + 1
⎪
⎪ i!
⎪
⎪ j =1
⎪
⎩0 i > n/2 + 1.
As a corollary, we obtain the regularity of R(n, t), which is equal to the top nonzero
t).
degree of R(n,
Corollary 1.5 (Corollary 3.18) For Gtn ∈ F,
reg R(n, t) = n/2 + 1.
We include an alternate graph-theoretic proof of the result above at the end of this
work. Beginning with an upper bound from [3] (or equivalently for our purposes,
one from [14]) and then identifying the initial ideal in> IGtn with the edge ideal of a
graph, we use results from [4] (allowing us to use in> IGtn instead of IGtn ) and then
[13] for a lower bound which agrees with our upper bound.
From a recursion established in Lemma 3.15, we go on to prove a Fibonacci
relationship between the lengths of the Artinian rings R(n,t) in Proposition 3.19,
and obtain the multiplicity of R(n, t) as a corollary. In the following, we drop t for
convenience.
Corollary 1.6 (Corollary 3.21) For n 2, there is an equality of multiplicities
e(R(n)) = e(R(n − 1)) + e(R(n − 2)).
In particular,
√ √ n+3
(1 + 5)n+3 − (1 − 5)
e(R(n)) = F (n + 3) = √ .
2n+3 5
14 L. Ballard
For more background, detail, and motivation, we refer the reader to [1], but note
that different notation and indexing conventions have been employed in this work.
Throughout, k is a field.
2 The Family of Toric Rings
In the following, we define a family of toric rings R(n, t) coming from an iterative
chordal bipartite family of graphs, F. We show that although one subfamily of these
rings comes from join-meet ideals of a (distributive) lattice and has some easily
derived algebraic invariants, this is not true in general. The reader may find the
definition of the toric ideal of a graph in Sect. 2.2, when it becomes relevant to the
discussion. We recall for the reader that a chordal bipartite graph is a bipartite graph
in which every cycle of length greater than or equal to six has a chord.
2.1 The Family F of Graphs
Below, we define the family F of chordal bipartite graphs iteratively from a family
of ladder-like structures Ltn . We note that the quantities involved in the following
definition follow patterns as follows:
n n/2 + 2 n/2 + 2
0 2 2
1 2 3
2 3 3
3 3 4
.. .. ..
. . .
Definition 2.1 For each n 0 and each t ∈ Fn+1 2 , we construct a ladder-like

structure Ltn with (n/2 + 2) rows and (n/2 + 2) columns and with nonzero
entries in the set {x0 , x2 , x3 , . . . , x2n+4 }. To do so, we use the notation
t ∈ Fn2 for
the first n entries of t, that is, all except the last entry. The construction is as follows,
where throughout, indices of entries in Ltn are strictly increasing from left to right in
each row and from top to bottom in each column. We note that Ltn does not depend
on t for n < 2, but does for n 2.
• For n = 0, the ladder-like structure L00 = L10 is
x0 x2
x3 x4
• For n = 1, to create Lt1 (regardless of what t is in F22 ), we add another column

with the entries x5 and x6 to the right of Lt0 to obtain
x0 x2 x5
x3 x4 x6
• For 2 n ≡ 0 mod 2 (≡ 1 mod 2), to create Ltn , we add another row (column)
with the entries x2n+3 , x2n+4 below (to the right of) Ltn−1 in the following way:
◦ The entry x2n+4 is in the ultimate row and column, row n/2 + 2 and column
n/2 + 2.
◦ The entry x2n+3 is in the new row (column) in a position directly below (to
the right of) another nonzero entry in Ltn .
· If the last entry of t is 0, x2n+3 is directly beneath (to the right of) the first
nonzero entry in the previous row (column).
· If the last entry of t is 1, x2n+3 is directly beneath (to the right of) the second
nonzero entry in the previous row (column).
In this way, the entries in t determine the choice at each stage for the placement of
x2n+3 .
Remark 2.2 We note a few things about this construction for n ≡ 0 mod 2 (≡ 1
mod 2), which may be examined in the examples below:
• We note that x2n+4 is directly beneath (to the right of) x2n+2 .
• We note that the only entries in row n/2 + 1 (column n/2 + 1) of Ltn−1 are
x2n−1 , x2n , and x2n+2 , so that the choices listed for placement of x2n+3 are the
only cases. In particular, tn+1 = 0 if and only if x2n+3 is directly beneath (to the
right of) x2n−1 , and tn+1 = 1 if and only if x2n+3 is directly beneath (to the right
of) x2n .
• Finally, we note that the only entries in column n/2 + 2 (row n/2 + 2) of Ltn
are x2n+1 , x2n+2 , and x2n+4 , and that the only entries in row n/2 + 2 (column
n/2 + 2) of Ltn are x2n+3 and x2n+4 .
Example 2.3 We have
x0 x2 x5 x0 x2 x5
(1,1,1) (0,0,0)
L2 = x3 x4 x6 L2 = x3 x4 x6
x7 x8 x7 x8
In either of the cases above, we could go on to construct Lt3 and Lt4 in the
following way: For n = 3, place x10 to the right of x8 and place x9 to the right
of either x5 or x6 , depending whether the last entry of
t is 0 or 1, respectively. Then
for n = 4, place x12 below x10 and place x11 below either x7 or x8 , depending
whether the last entry of t is 0 or 1, respectively.
16 L. Ballard
(1,1,...,1)
Example 2.4 In fact, when the entries of t are all ones, we see that Ln has
a ladder shape (is a two-sided ladder for n 3), shown below in the case when
2 n ≡ 0 mod 2:
x0 x2 x5
x3 x4 x6 x9
x7 x8 x10 x13
x11 x12 x14 x17
..
x15 x16 x18 .
..
x19 x20 . x2n+1
.. ..
. . x2n+2
x2n+3 x2n+4 .
We denote the subfamily of graphs coming from t = (1, 1, . . . , 1) by F1 ⊂ F.

When the entries of t are all zeros, Ln(0,0,...,0) has the following structure, shown
below in the case when 2 n ≡ 0 mod 2:
x0 x2 x5 x9 x13 x17 x21 · · · x2n+1

x3 x4 x6
x7 x8 x10
x11 x12 x14
x15 x16 x18
x19 x20 x22
.
x23 x24 . .
.. ..
. . x2n+2
x2n+3 x2n+4 .
(1,0,1,0,1,1,0,0,1,1,1,0,0,0,1,0,0)
For a more varied example, we have L16 below:
x0 x2 x5 x9
x3 x4 x6
x7 x8 x10 x13 x17
x11 x12 x14
x15 x16 x18 x21 x25 x29 x33
x19 x20 x22
x23 x24 x26
x27 x28 x30
x31 x32 x34
x35 x36 .
Definition 2.5 If we associate a vertex to each row and each column and an edge to
each nonzero entry of Ltn , we have a finite simple connected bipartite graph Gtn . The
set Vr of vertices corresponding to rows and the set Vc of vertices corresponding to
columns form a bipartition of the vertices of Gtn . We say a graph G is in F if G = Gtn
for some n 0 and some t ∈ Fn+1 2 .
Remark 2.6 We note that by construction Gtn has no vertices of degree one, since
each row and each column of Ltn has more than one nonzero entry. This ensures that
for large n our family is distinct from that studied in [5], since a Ferrers graph with
bipartitation V1 and V2 with no vertices of degree one must have at least two vertices
in V1 of degree |V2 | and at least two vertices in V2 of degree |V1 |, impossible for our
graphs when n 3, as the reader may verify. We also use the fact that Gtn has no
vertices of degree one for an alternate proof of the regularity of R(n, t) at the end of
this work.
(1,1,...,1)
Example 2.7 When n = 5, G5 ∈ F1 is
13 8 1 7 10
5
0 2
5 3 1 2 4
3 4
9
6
14 11 2 12
We develop properties of the Ltn which allow us to show in Sect. 2.2 that certain
minors of the Ltn are generators for the toric rings of the corresponding graphs Gtn .
Definition 2.8 For this work, a distinguished minor of Ltn is a 2-minor involving
only (nonzero) entries of the ladder-like structure Ltn , coming from a 2 × 2 subarray
of Ltn .
Proposition 2.9 For each i 1 and each f ∈ Fi+1
2 , the entry x2i+3 and the entry
f
x2i+4 each appear in exactly two distinguished minors of Li . For i ≡ 0 mod 2
(≡ 1 mod 2), these minors are of the form
18 L. Ballard
s2i := x2i+1 x2i+3 − xj2i x2i+4
coming from the subarray

xj2i x2i+1 xj2i x2i+3
x2i+3 x2i+4 x2i+1 x2i+4
for some j2i ∈ {0, 2, 3, . . . , 2i − 2} and
s2i+1 := x2i+2 x2i+3 − xj2i+1 x2i+4
coming from the subarray

xj2i+1 x2i+2 xj2i+1 x2i+3
x2i+3 x2i+4 x2i+2 x2i+4
for some j2i+1 ∈ {2i − 1, 2i}, and the only distinguished minor of Ltn with indices
all less than 5 is s1 := x2 x3 − x0 x4 .
Proof The last statement is clear by Definition 2.1; we prove the remaining
statements by induction on i. For i = 1, we have the distinguished minors s2 =
x3 x5 − x0 x6 and s3 = x4 x5 − x2 x6 coming from the subarrays

x0 x5
x3 x6
and

x2 x5
x4 x6
where j2 = 0 ∈ {0} and j3 = 2 ∈ {1, 2}, so we have our base case. Now suppose
the statement is true for i with 1 i < n, and let n ≡ 0 mod 2 (≡ 1 mod 2) and
t ∈ Fn+1
2 .
Case 1: If tn+1 = 0, then by Remark 2.2, x2n+3 is in the same column (row)
as x2n−1 . By induction, we have the distinguished minor s2n−2 = x2n−1 x2n+1 −
xj2n−2 x2n+2 coming from the subarray

xj2n−2 x2n+1 xj2n−2 x2n−1
.
x2n−1 x2n+2 x2n+1 x2n+2
Then in fact we have a subarray of the form

⎡ ⎤
xj2n−2 x2n+1
⎣x2n−1 x2n+2 ⎦ xj2n−2 x2n−1 x2n+3
,
x2n+1 x2n+2 x2n+4
x2n+3 x2n+4
so that we have the distinguished minors
s2n = x2n+1 x2n+3 − xj2n−2 x2n+4

s2n+1 = x2n+2 x2n+3 − x2n−1 x2n+4
with
j2n = j2n−2 ∈ {0, 2, 3, . . . , 2n − 4} ⊂ {0, 2, 3, . . . , 2n − 2}
by induction and with
j2n+1 = 2n − 1 ∈ {2n − 1, 2n}.
Since the only entries in row n/2 + 2 (column n/2 + 2) of Ltn are x2n+3 and
x2n+4 and since the only entries in column n/2 + 2 (row n/2 + 2) of Ltn are
x2n+1 , x2n+2 , and x2n+4 by Remark 2.2, these are the only distinguished minors of
Ltn containing either x2n+3 or x2n+4 , as desired.
Case 2 for tn+1 = 1 is analogous and yields
j2n = j2n−1 ∈ {2n − 3, 2n − 2} ⊂ {0, 2, 3, . . . , 2n − 2}
and
j2n+1 = 2n ∈ {2n − 1, 2n}.

Definition 2.10 Define the integers j2i , j2i+1 for j2 , . . . , j2n+1 as in the statement
of Proposition 2.9. We note in the remark below some properties of the jk .
Remark 2.11 From the proof of Proposition 2.9, we note that j2 = 0, j3 = 2, and
that for i 2, we have the following:
ti+1 = 0 ⇐⇒ j2i = j2i−2 ⇐⇒ j2i+1 = 2i − 1

ti+1 = 1 ⇐⇒ j2i = j2i−1 ⇐⇒ j2i+1 = 2i.
For the sake of later proofs, we extend the notion of jk naturally to s1 = x2 x3 −x0 x4
and say that j1 = 0, and note the following properties of the jk for 1 k 2n+1:
• We have j2i ∈ {j2i−2 , j2i−1 } and j2i 2i − 2. Indeed, for i = 1, j2 = j1 = 0,
and for i 2, this is clear from the statement above.
• We have j2i+1 ∈ {2i − 1, 2i}. Indeed, for i = 0, j1 = 0 ∈ {−1, 0}, for i = 1,
j3 = 2 ∈ {1, 2}, and for i 2, this follows from the statement above.
• The j2i form a non-decreasing sequence. Indeed, for i 2, either j2i = j2i−2 or
j2i = j2i−1 2i − 3 > 2i − 4 j2i−2 .
20 L. Ballard
Remark 2.12 We also note from the proof above that the following is a subarray of
Ltn for all i ≡ 0 mod 2 (≡ 1 mod 2) such that 1 i n, which we use in the
proof of the proposition below:
⎡ ⎤
xj2i x2i+1
⎣xj ⎦ xj2i xj2i+1 x2i+3
2i+1 x2i+2
x2i+1 x2i+2 x2i+4
x2i+3 x2i+4
Proposition 2.13 For n 0, each graph Gtn ∈ F is chordal bipartite with vertex
bipartition Vr ∪ Vc of cardinalities
n
|Vr | = +2
2
n
|Vc | = + 2.
2
Proof We already know by Definition 2.5 that every graph Gtn is bipartite for
n 0, with the bipartition above coming from the rows and columns of Ltn .
The cardinalities of the vertex sets follow from Remark 2.2. We prove the chordal
f
bipartite property by induction on n. It is clear for i = 0 and i = 1 that Gi is
i+1
chordal bipartite for f ∈ F2 , since these graphs have fewer than six vertices. Now
f
suppose Gi is chordal bipartite for i with 1 i < n ≡ 0 mod 2 (≡ 1 mod 2),
and consider Gtn for t ∈ Fn+1
2 . We know that the following array (or its transpose)
is a subarray of Ltn by Remark 2.12, and we include for reference the corresponding
subgraph of Gtn with vertices labeled by row and column.
⎡ ⎤
xj2n x2n+1
⎣xj ⎦
2n+1 x2n+2
x2n+3 x2n+4
2 2 +1 2 +3
1 2 3
2 +1 2 +2 2 +4
2
We know the only difference between Gtn and Gtn−1 is one vertex r3 correspond-
ing to row n/2 + 2 (column n/2 + 2) and two edges x2n+3 = {r3 , c1 } and
x2n+4 = {r3 , c2 }, where c1 corresponds to the column (row) containing x2n+3 and
c2 corresponds to column n/2 + 2 (row n/2 + 2). By Remark 2.2, deg r3 = 2,
since the only entries in row n/2 + 2 (column n/2 + 2) are x2n+3 and x2n+4 .
Then any even cycle containing r3 must also contain x2n+3 and x2n+4 . Similarly, by
the same remark, the only other edges with endpoint c2 are x2n+1 and x2n+2 , the
entries added to make Ltn−1 , so we know that any even cycle containing x2n+4 and
x2n+3 must contain either x2n+1 or x2n+2 . We see that any even cycle containing r3
and x2n+1 is either a 4-cycle or has xj2n as a chord, and any even cycle containing
r3 and x2n+2 is either a 4-cycle or has xj2n+1 as a chord. Thus every graph Gtn is
chordal bipartite for n 0, with the bipartition above.

Remark 2.14 We note that the previous proposition ensures that our graphs are
distinct from those studied in [8, 9], and [23], which are not chordal bipartite except
for the first family in [8], in which every four-cycle shares exactly one edge with
every other four-cycle (also distinct from our family except for the trivial case with
only one four-cycle, corresponding to Gt0 ).
2.2 Toric Rings for F
In this section, we develop the toric ring R(n, t) for each of the chordal bipartite
graphs Gtn in the family F. We first show that the toric ideal IGtn of the graph Gtn
is the same as the ideal I (n, t) generated by the distinguished minors of Ltn . We
then demonstrate that for some n and t, these ideals do not arise from the join-meet
ideals of lattices in a natural way, so that results in lattice theory do not apply to the
general family F in an obvious way.
We first define the toric ideal of a graph. For any graph G with vertex set V and
edge set E, there is a natural map π : k[E] → k[V ] taking an edge to the product
of its endpoints. The polynomial subring in k[V ] generated by the images of the
edges under the map π is denoted k[G] and is called the edge ring of G. The kernel
of π is denoted IG and is called the toric ideal of G; the ring k[G] is isomorphic
to k[E]/IG . In this work, we consider the toric ring k[E]/IG and the toric ideal IG
for our particular graphs. It is known in general that IG is generated by binomial
expressions coming from even closed walks in G [27, Prop 3.1] and that the toric
ideal of a chordal bipartite graph is generated by quadratic binomials coming from
the 4-cycles of G (see [20, Th 1.2]).
Let S(n) = k[x0 , x2 , x3 , . . . , x2n+4 ] be the polynomial ring in the edges of Gtn .
The edge ring for Gtn ∈ F is denoted by k[Gtn ] and is isomorphic to the toric ring
S(n)
R(n, t) := ,
IGtn
22 L. Ballard
where IGtn is the toric ideal of Gtn [15, 5.3]. For the general construction of a toric
ideal, we refer the reader to [24, Ch 4]. Our goal is to show that the toric ideal IGtn
of Gtn is equal to
I (n, t) = ({distinguished minors of Ltn }).
Proposition 2.15 Let S(n) = k[x0 , x2 , x3 , . . . , x2n+4 ]. For Gtn ∈ F, we have
S(n)
R(n, t) = ,
I (n, t)
where
I (n, t) = ({distinguished minors of Ltn }).
Proof To prove this, we need only show that I (n, t) is the toric ideal IGtn of
the graph Gtn . By Definition 2.5, it is clear that the distinguished minors of Ltn
are in IGtn , corresponding to the 4-cycles of Gtn . Since G is chordal bipartite by
Proposition 2.13, these are the only generators of IGtn [15, Cor 5.15].

Corollary 2.16 The rings R(n, t) are normal Cohen–Macaulay rings.
Proof By Definition 2.5 and Proposition 2.15, the ring R(n, t) is the toric ring of a
finite simple connected bipartite graph, and hence by Corollary 5.26 in [15], R(n, t)
is Cohen–Macaulay for each n and t. The fact that each R(n, t) is normal follows
from [25, Th 5.9, 7.1].

Because we know the distinguished minors of Ltn , we are now able to character-
ize the generators for the toric ideal R(n, t) of Gtn .
Remark 2.17 By Proposition 2.9, the generators s1 , . . . , s2n+1 for IGtn may be
summarized as follows. For integers i such that 1 i n, set
s1 = x2 x3 − xj1 x4
s2i = x2i+1 x2i+3 − xj2i x2i+4
s2i+1 = x2i+2 x2i+3 − xj2i+1 x2i+4 ,
where the nonnegative integers jk are as in Remark 2.11, that is, j1 = j2 = 0,

j3 = 2, and for i 2, we have
ti+1 = 0 ⇐⇒ j2i = j2i−2 ⇐⇒ j2i+1 = 2i − 1

ti+1 = 1 ⇐⇒ j2i = j2i−1 ⇐⇒ j2i+1 = 2i.
We note that the number of generators depends on n and that the jk depend on t, but
we may ignore dependence on t when working with general jk . We sometimes call
s1 , . . . , s2n+1 the standard generators of IGtn , and show in Sect. 2.3 that for certain
n and t, they are not equal to the usual generators for the join-meet ideal of any
lattice D.
Example 2.18 We consider the toric ideal of a graph in F1 . For n = 5 and t =
(1, 1, . . . , 1), by Remark 2.11 we have j1 = j2 = 0, j3 = 2, j2i = j2i−1 and
j2i+1 = 2i for i 2, so that
k[x0 , x2 , . . . , x14 ]
R(5, (1, 1, . . . , 1)) = ,
IG(1,1,...,1)
5
(1,1,...,1)
where IG(1,1,...,1) is generated by the distinguished minors of L5 :
5
s1 = x2 x3 − x0 x4 s2 = x3 x5 − x0 x6
s3 = x4 x5 − x2 x6 s4 = x5 x7 − x2 x8
s5 = x6 x7 − x4 x8 s6 = x7 x9 − x4 x10
s7 = x8 x9 − x6 x10 s8 = x9 x11 − x6 x12
s9 = x10 x11 − x8 x12 s10 = x11 x13 − x8 x14
s11 = x12 x13 − x10 x14 .
2.3 Distinction from Join-Meet Ideals of Lattices
We saw in Example 2.4 and Proposition 2.15 that if Gtn ∈ F1 ⊂ F, then IGtn is a
ladder determinantal ideal for n 2. It is known that a ladder determinantal ideal is
equal to the join-meet ideal of a (distributive) lattice (indeed, with a natural partial
ordering which decreases along rows and columns of Ltn we obtain such a lattice).
Some algebraic information such as regularity and projective dimension may be
easily derived for some join-meet ideals of distributive lattices (see, for example,
Chapter 6 of [15]). We spend some time in this section establishing that not all
rings R(n, t) ∈ F arise from a lattice in a natural way (see Remark 2.20), and so
there does not seem to be any obvious way to obtain our results in Sect. 3 from
the literature on join-meet ideals of distributive lattices or on ladder determinantal
ideals. The results in Sect. 3 may be viewed as an extension of what may already be
derived for the family F1 from the existing literature.
The following five lemmas serve to provide machinery to show that there is at
least one ring in the family F, namely R(5, (1, 1, 1, 1, 1, 0)), whose toric ideal does
not come from a lattice on the set {x0 , x2 , . . . , x14 } in any obvious way. That is, we
show that the standard generators of IGt , the sk from Remark 2.17, are not equal to
5
the standard generators (see Definition 2.19) for any lattice D on {x0 , x2 , . . . , x14 }.
24 L. Ballard
Before we begin, we introduce some definitions and notation that we use

extensively throughout.
Definition 2.19 The join-meet ideal of a lattice is defined from the join (least upper
bound) x ∨ y and meet (greatest lower bound) x ∧ y of each pair of incomparable
elements x, y ∈ L. In this work, a standard generator of the join-meet ideal of a
lattice D is a nonzero element of one of the following four forms:
xa xb − (xa ∨ xb )(xa ∧ xb ) = xa xb − (xa ∧ xb )(xa ∨ xb )

(xa ∨ xb )(xa ∧ xb ) − xa xb = (xa ∧ xb )(xa ∨ xb ) − xa xb
for xa , xb ∈ L. We sometimes refer to such an element as a standard generator of

D (the join-meet ideal is defined by analogous generators in the literature, though
sometimes a, b ∈ L instead of xa and xb ). We note that for a standard generator,
the pair {xa , xb } is an incomparable pair, and the pair {(xa ∨ xb ), (xa ∧ xb )} is a
comparable pair.
Though we are in a commutative ring, we provide all possible orderings for
factors within the terms of a standard generator to emphasize that either factor of
the monomial
(xa ∨ xb )(xa ∧ xb ) = (xa ∧ xb )(xa ∨ xb )
may be the join or the meet of xa and xb .

Remark 2.20 We give an explanation for why it makes sense to focus only on the
standard generators of a join-meet ideal. We recall that the standard generators sk
for IGtn from Remark 2.17 come from distinct 2 × 2 arrays within the ladder-like
structure Ltn and recognize that either monomial of sk determines its 2 × 2 array.
Then an element of the form ab − cd in IGt with a, b, c, d ∈ {x0 , x2 , x3 , . . . , x14 }
5
must be equal to ±si for some i, since a nontrivial sum of sk with coefficients in
{−1, 1} either has more than two terms or is equal to si for some i, and other
coefficients would be extraneous. Then any generating set for IGt where each
5
element has the form ab − cd in IGt with a, b, c, d ∈ {x0 , x2 , x3 , . . . , x14 } must
5
consist of all the sk (up to sign). We conclude that it is natural to check whether the
sk are standard generators of a lattice D, instead of non-standard generators.
Definition 2.21 Given a standard generator s = uz − wv of a lattice D, where
u, v, w, z ∈ D, let Fs ∈ F2 be defined as follows:
• If Fs = 0, the elements in the first (positive) monomial of s are not comparable
in D (so the elements in the second (negative) monomial of s are comparable in
D).
• If Fs = 1, the elements in the negative monomial of s are not comparable in D
(so the elements in the positive monomial of s are comparable in D).
For a given list s1 , s2 , . . . , sm of standard generators of a lattice D, we use
F = (F1 , . . . , Fm ) ∈ Fm
2,
where Fj = Fsj , to encode the comparability of the variables in these generators.

We note that exactly one of Fj = 0 or Fj = 1 happens for each j ; we are merely
encoding which monomial in sj corresponds to xa xb , and which to (xa ∨ xb )(xa ∧
xb ) = (xa ∧ xb )(xa ∨ xb ).
Notation 2.22 We use the notation u > {w, v} if u > w and u > v in a lattice D,
and {w, v} > z if w > z and v > z in D.
In the first lemma, we begin by showing what restrictions we must have on
a lattice whose join-meet ideal contains the 2-minors of the following array as
standard generators:
ab e
cdf
Lemma 2.23 Suppose
s1 = bc − ad
s2 = ce − af
s3 = de − bf
are standard generators of a lattice D. Let F ∈ F32 be defined for these three
elements as in Definition 2.21. Then up to relabeling of variables,
F ∈ {{0, 0, 0}, {0, 0, 1}, {0, 1, 1}}.
Proof We first note that some of the cases we consider are equivalent. If we relabel
variables according to the permutation (ac)(bd)(ef ), we see that
F = {i, j, k} ≡ {1 − i, 1 − j, 1 − k}.
This limits the cases we need to consider to
F ∈ {{0, 0, 0}, {0, 0, 1}, {0, 1, 0}, {0, 1, 1}}.
That is, we only need to show that the case F = {0, 1, 0} is impossible.
Let F1 = 0. Then without loss of generality, up to reversing the order in the lattice
(which does not affect the join-meet ideal), we have a > {b, c} > d. If F2 = 1, we
have e > {a, f } > c, so e > {b, f } > d and hence F3 = 1. We conclude that the
case F = {0, 1, 0} is impossible.

26 L. Ballard
In the second lemma, we show what restrictions we must have on a lattice whose
join-meet ideal contains the 2-minors of the ladder
ab e
cdf
g h
as standard generators, and which meets certain comparability conditions.

Lemma 2.24 Suppose
s1 = bc − ad
s2 = ce − af
s3 = de − bf
s4 = eg − bh
s5 = f g − dh
are standard generators of a lattice D, and that {a, g},{a, h},{c, g}, and {c, h} are
comparable pairs in D. Let F ∈ F52 be defined for these five elements as in 2.21.
Then up to relabeling of variables, F = {0, 0, 0, 0, 0}.
Proof We first note that with natural relabeling, both {s1 , s2 , s3 } and {s3 , s4 , s5 }
satisfy the hypotheses of Lemma 2.23, so if we let F be defined as in Definition 2.21,
this limits the cases we need to consider to 5-tuples whose first three elements and
whose last three elements satisfy the conclusion of Lemma 2.23. We note that some
of the cases we consider are equivalent, allowing us to reduce to eighteen cases. If
we relabel variables according to the permutation (ac)(bd)(ef ), we see that F =
{i, j, k, l, m} ≡ {1−i, 1−j, 1−k, m, l}, and if we relabel the variables according to
the permutation (be)(df )(gh), we have F = {i, j, k, l, m} ≡ {j, i, 1 − k, 1 − l, 1 −
m}. The permutation (ah)(cg)(bf ) yields F = {i, j, k, l, m} ≡ {m, l, k, j, i}. Then
we have eighteen cases in four equivalence classes, where the only equivalence class
that satisfies the conclusion of the lemma is emboldened.
{0, 0, 0, 0, 0} ≡ {1, 1, 1, 0, 0} ≡ {1, 1, 0, 1, 1} ≡ {0, 0, 1, 1, 1}
{0, 0, 0, 0, 1} ≡ {1, 1, 1, 1, 0} ≡ {1, 1, 0, 0, 1} ≡ {0, 0, 1, 1, 0} ≡ {0, 1, 1, 0, 0}

≡ {1, 0, 0, 0, 0}
≡ {0, 1, 1, 1, 1}
≡ {1, 0, 0, 1, 1}
{0, 0, 0, 1, 1} ≡ {1, 1, 1, 1, 1} ≡ {1, 1, 0, 0, 0} ≡ {0, 0, 1, 0, 0}

{0, 1, 1, 1, 0} ≡ {1, 0, 0, 0, 1}
Case {0, 0, 0, 0, 1}: Since F1 = 0, without loss of generality (reversing the order on
the entire lattice if needed) we have a > {b, c} > d. Then F2 = F3 = F4 = 0 and
F5 = 1, with the ordering chosen, yield
a > {c, e} > f

b > {d, e} > f
b > {e, g} > h
g > {d, h} > f.
If c > g, then a > {b, c} > g > d, but then bc − ad is not a standard generator
of D, and this is a contradiction. If c < g, then c < g < b so that both {b, c} and
{a, d} from s1 are comparable pairs, but this is a contradiction. We conclude that the
case {0, 0, 0, 0, 1} is impossible.
In the case F = {0, 0, 0, 1, 1}, the comparability of {c, g} forces either the
comparability of {b, c} (a contradiction), or d < {b, c} < a < g, up to reversing
the order in the lattice, since s1 is a standard generator. Since {a, h} is a comparable
pair, it immediately follows that either {b, h} is comparable (a contradiction) or
e < {b, h} < a < g, which is a contradiction since s4 is a standard generator,
as the reader may verify. Because of the comparability of the pair {c, g}, the case
F = {0, 1, 1, 1, 0} forces comparability of {f, g} or {b, c} and hence yields a
contradiction.
These cases are compatible with the given relabelings and thus conclude our
proof.

In the third lemma, we show what restrictions we must have on a lattice whose
ab e
cdf i
g hj
as standard generators and which meets certain comparability conditions.
Lemma 2.25 Suppose
s1 = bc − ad
s2 = ce − af
s3 = de − bf
s4 = eg − bh
s5 = f g − dh
s6 = gi − dj
28 L. Ballard
s7 = hi − fj
are standard generators of a lattice D, and that {a, g}, {a, h}, {c, g}, {c, h}, {b, i},
{b, j }, {e, i}, and {e, j } are comparable pairs in D. Let F ∈ F72 be defined for
these seven elements as in Definition 2.21. Then up to relabeling of variables, F =
{0, 0, 0, 0, 0, 0, 0}.
Proof We first note that with natural relabeling, the subsets {s1 , s2 , s3 , s4 , s5 }
and {s3 , s4 , s5 , s6 , s7 } satisfy the hypotheses of Lemma 2.24, so if we let F be
defined as in Definition 2.21, this limits the cases we need to consider to 7-tuples
whose first five elements and whose last five elements satisfy the conclusion of
Lemma 2.24. The only possible cases are {0, 0, 0, 0, 0, 0, 0} and {0, 0, 1, 1, 1, 0, 0}.
If we relabel variables according to the permutation (be)(df )(gh), we see that
F = {i, j, k, l, m, n, o} ≡ {j, i, 1 − k, 1 − l, 1 − m, o, n}, so that these two cases
are equivalent. Then up to relabeling of variables, F = {0, 0, 0, 0, 0, 0, 0}.

In the fourth lemma, we show what restrictions we must have on a lattice whose
ab e
cdf i
g hj
k l
as standard generators, and which meets certain comparability conditions.

Lemma 2.26 Suppose
s1 = bc − ad
s2 = ce − af
s3 = de − bf
s4 = eg − bh
s5 = f g − dh
s6 = gi − dj
s7 = hi − fj
s8 = ik − f l
s9 = j k − hl
are standard generators of a lattice D, and that {a, g}, {a, h}, {c, g}, {c, h},
{b, i}, {b, j }, {e, i}, {e, j }, {d, k}, {d, l}, {g, k}, and {g, l} are comparable pairs
in D. Let F ∈ F92 be defined for these nine elements as in Definition 2.21. Then
F = {0, 0, 0, 0, 0, 0, 0, 0, 0}.
Proof We first note that with natural relabeling, both {s1 , s2 , s3 , s4 , s5 , s6 , s7 } and
{s3 , s4 , s5 , s6 , s7 , s8 , s9 } satisfy the hypotheses of Lemma 2.25, so if we let F be
defined as in Definition 2.21, this limits the cases we need to consider to 9-tuples
whose first seven entries and whose last seven entries satisfy the conclusion of
Lemma 2.25. We see by Lemma 2.25 that F = {0, 0, 0, 0, 0, 0, 0, 0, 0}.

We now have the machinery necessary to show that for t = (1, 1, 1, 1, 1, 0), IGt
5
does not come from a lattice. In our proof, we use the previous four lemmas and the
(1,1,1,1,1,0)
fact that IGt is generated by the distinguished minors of L5 :
5
x0 x2 x5
x3 x4 x6 x9 x13
x7 x8 x10
x11 x12 x14
Proposition 2.27 Let n = 5 and t = (1, 1, 1, 1, 1, 0). Then the set of standard
generators for IGt is not equal to the complete set of standard generators (up to
5
sign) of any (classical) lattice.
Proof By Remark 2.17 and choice of n = 5 and t = (1, 1, 1, 1, 1, 0), the generators
of IGt are
5
s1 = x2 x3 − x0 x4 s2 = x3 x5 − x0 x6
s3 = x4 x5 − x2 x6 s4 = x5 x7 − x2 x8
s5 = x6 x7 − x4 x8 s6 = x7 x9 − x4 x10
s7 = x8 x9 − x6 x10 s8 = x9 x11 − x6 x12
s9 = x10 x11 − x8 x12 s10 = x11 x13 − x6 x14
s11 = x12 x13 − x9 x14
Suppose a lattice D exists whose complete set of standard generators (up to

sign) equals {s1 , . . . , s11 }. We note that if the monomial xi xj does not appear in
any of the sk , then {xi , xj } is a comparable pair, since otherwise ±(xi xj − (xi ∨
xj )(xi ∧ xj )) would be in the set of standard generators of D. Thus the pairs
{x0 , x7 }, {x0 , x8 }, {x3 , x7 }, {x3 , x8 }, {x2 , x9 }, {x2 , x10 }, {x5 , x9 }, {x5 , x10 }, {x4 , x11 },
{x4 , x12 }, {x7 , x11 }, {x7 , x12 }, {x10 , x13 }, and {x10 , x14 } are comparable pairs in D.
Let F ∈ F11 2 be defined as in Definition 2.21. Then with natural relabeling of the
first nine relations, this lattice satisfies the hypotheses of Lemma 2.26, so the only
cases we need to consider are F = {0, 0, 0, 0, 0, 0, 0, 0, 0, a, b}.
Since F1 = 0, without loss of generality, we have x0 > {x2 , x3 } > x4 . Then with
the ordering chosen, the fact that F3 = F5 = F7 = F9 = 0 yields
x2 > {x4 , x5 } > x6

30 L. Ballard
x4 > {x6 , x7 } > x8

x6 > {x8 , x9 } > x10
x8 > {x10 , x11 } > x12 .
The reader may verify that b = 0 and b = 1 both yield contradictions based
on inspecting the standard generator s11 in light of the comparability of the pairs
{x10 , x13 } and {x10 , x14 }, respectively, using the same technique employed in Case
{0, 0, 0, 0, 1} of the proof of Lemma 2.24.
We conclude that there is no lattice whose complete set of standard generators
(up to sign) equals the set of standard generators of IG(1,1,1,1,1,0) .

5
3 Properties of the Family of Toric Rings
In Sect. 2, we defined a family of toric rings, the rings R(n, t) coming from the
family F, and we demonstrated some context for these rings in the area of graph
theory. Now we investigate some of the algebraic properties of each R(n, t). We
develop proofs to establish dimension, regularity, and multiplicity.
3.1 Dimension and System of Parameters
We use the degree reverse lexicographic monomial order with x0 > x2 > x3 > · · ·
throughout this section, and denote it by >. We show that the standard generators sk
given in Remark 2.17 are a Gröbner basis for IGtn with respect to >.
Lemma 3.1 If s1 , . . . , s2n+1 are as in Remark 2.17, then B = {s1 , . . . , s2n+1 } is a
Gröbner basis for IGtn with respect to >.
Proof This is a straightforward computation using Buchberger’s Criterion and
properties of the jk from Remark 2.11. By Remark 2.17, for 1 i n the ideal
IGtn is generated by
s1 = x2 x3 − xj1 x4
s2i = x2i+1 x2i+3 − xj2i x2i+4
s2i+1 = x2i+2 x2i+3 − xj2i+1 x2i+4 ,
where j1 = j2 = 0, j3 = 2, and for i 2, we have
ti+1 = 0 ⇐⇒ j2i = j2i−2 ⇐⇒ j2i+1 = 2i − 1

ti+1 = 1 ⇐⇒ j2i = j2i−1 ⇐⇒ j2i+1 = 2i.
If we adopt S-polynomial notation Si,j for the S-polynomial of si and sj , then

the cases to consider are
S2i−1,2i and S2i,2i+1 for 1 i n
S2i,2i+2 for 1 i n − 1.
To give a flavor of the computation involved, we show the case S2i−1,2i for 1
i n, and leave the remaining cases to the reader. We show in each subcase that
S2i−1,2i is equal to a sum of basis elements with coefficients in S(n), so that the
reduced form of S2i−1,2i is zero in each subcase. We have
S2i−1,2i = x2i+3 (x2i x2i+1 − xj2i−1 x2i+2 ) − x2i (x2i+1 x2i+3 − xj2i x2i+4 )
= −xj2i−1 x2i+2 x2i+3 + xj2i x2i x2i+4
Case 1: If i 2 and ti+1 = 0, then j2i = j2i−2 and j2i+1 = 2i − 1, so we have
S2i−1,2i = −xj2i−1 x2i+2 x2i+3 + xj2i−2 x2i x2i+4
Case 1.1: If in addition i 3 and ti = 0, then j2i−2 = j2i−4 and j2i−1 =

2i − 3, so we have
S2i−1,2i = −x2i−3 x2i+2 x2i+3 + xj2i−4 x2i x2i+4

= −x2i−3 (x2i+2 x2i+3 − xj2i+1 x2i+4 ) − x2i+4 (x2i−3 x2i−1 − xj2i−4 x2i )
= −x2i−3 s2i+1 − x2i+4 s2i−4 .
Case 1.2: If in addition i = 2 or i 3 and ti = 1, then j2i−2 = j2i−3 and

j2i−1 = 2i − 2, so we have
S2i−1,2i = −x2i−2 x2i+2 x2i+3 + xj2i−3 x2i x2i+4

= −x2i−2 (x2i+2 x2i+3 − xj2i+1 x2i+4 ) − x2i+4 (x2i−2 x2i−1 − xj2i−3 x2i )
= −x2i−2 s2i+1 − x2i+4 s2i−3 .
Case 2: If i = 1 or if i 2 and ti+1 = 1, then j2i = j2i−1 and j2i+1 = 2i, so we

have
S2i−1,2i = −xj2i x2i+2 x2i+3 + xj2i x2i x2i+4

= −xj2i s2i+1 .
This concludes the case S2i−1,2i for 1 i n − 1. The remaining cases are similar.

32 L. Ballard
Corollary 3.2 The ring R(n, t) is Koszul for all n and all t.
Proof Since IGtn has a quadratic Gröbner basis, the ring R(n, t) is Koszul for all n
and all t due to [15, Th 2.28].

Corollary 3.3 The initial ideal for IGtn with respect to the degree reverse lexico-
graphic monomial order > is
in> IGtn = (x2 x3 , {x2i+1 x2i+3 , x2i+2 x2i+3 | 1 i n}).
We note that in> IGtn does not depend on t, which will be useful for the following
sections.
Since Gtn is bipartite and has n + 4 vertices, the Krull dimension of R(n, t) is
already known to be n + 3 [27, Prop 3.2]. We provide an alternate proof of the Krull
dimension using the initial ideal in> IGtn from Corollary 3.3 and direct computation.
As a corollary, we obtain the projective dimension of R(n, t). We note that the
Krull dimension, like the initial ideal, does not depend on t. We refer the reader to
Remark 2.17 for a reminder of how to think of the toric ring
S(n) k[x0 , x2 , x3 , . . . , x2n+4 ]

R(n, t) = =
IGtn IGtn
in the context of this work.

Theorem 3.4 The Krull dimension of R(n, t) is
dim R(n, t) = n + 3.
Proof Let > be the degree reverse lexicographic monomial order with
x0 > x2 > x3 > · · · > x2n+4 .
By Corollary 3.3, the initial ideal of IGtn with respect to > is
in> IGtn = (x2 x3 , {x2i+1 x2i+3 , x2i+2 x2i+3 | 1 i n}).
Since S(n)/(in> IGtn ) and R(n, t) = S(n)/IGtn are known to have the same Krull
dimension (see for example [6, Props 9.3.4 and 9.3.12]), it suffices to prove that
dim S(n)/(in> IGtn ) = n + 3.
To see that the dimension is at least n + 3, we construct a chain of prime ideals

in S(n) containing in> IGtn . Since every monomial generator of in> IGtn contains a
variable of odd index, we begin with Pn = ({xk | k odd, 2 < k < 2n + 4}), a prime
ideal containing in> IGtn . Then we have the chain of prime ideals Pn Pn + (x0 )
Pn + (x0 , x2 ) Pn + (x0 , x2 , x4 ) · · · Pn + ({x2i | 0 i n + 2}), so that
dim S(n)/(in> IGtn ) n + 3.
To see that the dimension is at most n + 3, we find a sequence of n + 3 elements

in S(n)/(in> IGtn ) such that the quotient by the ideal they generate has dimension
zero. Let
in S(n), and take the quotient of S(n)/(in> IGtn ) by the image of Xn to obtain the
following. In the last step, we rewrite the quotient of S(n) and (in> IGtn ) + (Xn )
by (Xn ) by setting x0 and x2n+4 equal to zero and replacing x2i with x2i+1 for
1 i n + 1:
S(n)/(in> IGtn ) S(n)

∼
=
((in> IGtn ) + (Xn ))/(in> IGtn ) ((in> IGtn ) + (Xn ))
∼ S(n)/(Xn )
=
((in> IGtn ) + (Xn ))/(Xn )
∼ k[x3 , x5 , . . . , x2n+1 , x2n+3 ]

= .
(x3 , {x2i+1 x2i+3 , x2i+3
2 2 |1i n})
Since

(x32 , {x2i+1 x2i+3 , x2i+3
2 | 1 i n}) = (x3 , x5 , . . . , x2n+3 ),
the above ring has dimension zero. Thus,
dim S(n)/(in> IGtn ) n + 3.
We conclude that dim R(n, t) = dim S(n)/(in> IGtn ) = n + 3.

Corollary 3.5 The projective dimension of R(n, t) over S(n) is
pdS(n) R(n, t) = n + 1.
Proof We know the Krull dimension of the polynomial ring S(n) is 2n + 4. The
result follows from the fact that R(n, t) is Cohen–Macaulay (Corollary 2.16) and
from the graded version of the Auslander-Buchsbaum formula.

Remark 3.6 The proof of the previous theorem shows that the image of
in S(n)/(in> IGtn ) is a system of parameters for S(n)/(in> IGtn ). We prove in the

next theorem that the image of Xn in R(n, t) (which we call Xn ) is also a system of
34 L. Ballard
parameters for R(n, t). Before doing so, we introduce some notation and a definition
which allows us to better grapple with the quotient ring R(n, t)/(Xn ).
Definition 3.7 Consider the isomorphism
R(n, t) S(n)/(IGtn ) S(n)/(Xn )

= ∼
=
(Xn ) ((IGtn ) + (Xn ))/(I Gtn ) ((I ) + (Xn ))/(Xn )
Gtn
We view taking the quotient by Xn as setting x0 and x2n+4 equal to zero and
replacing x2i with x2i+1 for 1 i n + 1 to obtain
:= k[x3 , x5 , . . . , x2n+1 , x2n+3 ] ∼

S(n) = S(n)/(Xn ).
By the same process (detailed below), we obtain the ideal I ∼

Gtn = (IGtn +(Xn ))/(Xn ).
We further define the quotient
t) := S(n)/
R(n, I ∼
Gtn = R(n, t)/(Xn ).
We find this notation natural since it is often used for the removal of variables, and
the quotient by Xn may be viewed as identifying and removing variables. Since this
work has no completions in it, there should be no conflict of notation.
Definition 3.8 To define IGtn in particular, we recall the standard generators of
IGtn and introduce further notation to describe the generators of I ∼
Gtn = (IGtn +
(Xn ))/(Xn ). By Remark 2.17, the standard generators of IGtn are
s1 = x2 x3 − xj1 x4
s2i = x2i+1 x2i+3 − xj2i x2i+4
s2i+1 = x2i+2 x2i+3 − xj2i+1 x2i+4 ,
for 1 i n, where the nonnegative integers jk are as in Remark 2.11.

Letι be the largest index such that j2ι = 0. By Remark 2.11, we see that the j2i
are defined recursively and form a non-decreasing sequence. Then
j1 = j2 = j4 = j6 = · · · = j2ι = 0,
and since we view taking the quotient by Xn as setting x0 and x2n+4 equal to zero
and replacing x2i with x2i+1 for 1 i n + 1, we define I
Gtn by replacing xjk with
xJk (defined below) for 1 k < 2n to obtain
s1 = x32 − xJ1 x5

s
2i = x2i+1 x2i+3 − xJ2i x2i+5
s
2i+1 = x2i+3 − xJ2i+1 x2i+5
2
s
2n = x2n+1 x2n+3
s
2n+1 = x2n+3
2
for 1 i < n, where

⎧
⎪
⎪ if k is even and k 2
ι, or if k = 1
⎨0
xJk = xjk +1 if 2
ι < k < 2n and jk is even
⎪
⎪
⎩x if jk is odd
jk
We note that Jk k for each 1 k < 2n, since jk k − 1 by Remark 2.11.

By properties of the original jk from Remark 2.11, we know that xJ1 = xJ2 = 0,
J3 = 3, and for 2 i < n,
ti+1 = 0 ⇐⇒ xJ2i = xJ2i−2 ⇐⇒ J2i+1 = 2i − 1

ti+1 = 1 ⇐⇒ xJ2i = xJ2i−1 ⇐⇒ J2i+1 = 2i + 1.
(0,0,0)
Example 3.9 We construct the ring R(2, (0, 0, 0)). For the graph G2 ∈ F, we
have the toric ring
k[x0 , x2 , x3 , . . . , x8 ]
R(2, (0, 0, 0)) =
(x2 x3 − x0 x4 , x3 x5 − x0 x6 , x4 x5 − x2 x6 , x5 x7 − x0 x8 , x6 x7 − x3 x8 )
coming from the ladder-like structure
x0 x2 x5
(0,0,0)
L2 = x3 x4 x6
x7 x8
from Example 2.3. We know
X2 = x0 , x2 − x3 , x4 − x5 , . . . , x4 − x5 , x6 − x7 , x8 ,
so that R(2, (0, 0, 0))/(X2 ) is isomorphic to
k[x0 , x2 , x3 , x4 , x5 , x6 , x7 , x8 ]
(x2 x3 − x0 x4 , x3 x5 − x0 x6 , x4 x5 − x2 x6 , x5 x7 − x0 x8 , x6 x7 − x3 x8 , x0 , x2 − x3 , . . ., x8 )
∼ k[x3 , x5 , x7 ]
= = R(2,
(0, 0, 0)).
(x3 , x3 x5 , x52 − x3 x7 , x5 x7 , x72 )
2
Now we show that Xn is also a system of parameters for R(n, t), and not just for
the quotient by the initial ideal.
36 L. Ballard
Proposition 3.10 Let R(n, t) = S(n)/IGtn and let
so that the image of Xn in S(n)/(in> IGtn ) is the system of parameters from

Remark 3.6. Then the image of Xn in R(n, t) is a system of parameters for R(n, t).
Proof Let Xn be defined as above. Then by Theorem 3.4 and Definition 3.7 we
t) = 0. We have for n = 0 that
need only show that dim R(n,
t) = k[x3 ] ,
R(0,
(x32 )
for n = 1
t) = k[x3 , x5 ]
R(1, ,
(x32 , x3 x5 , x52 )
and for n > 1

t) = S(n) = k[x3 , x5 , . . . , x2n+1 , x2n+3 ] ,
R(n,
I
Gtn s1 , s
({ 2i , s
2i+1 | 1 i n})
where
s1 = x32
s
2i = x2i+1 x2i+3 − xJ2i x2i+5
s
2i+1 = x2i+3 − xJ2i+1 x2i+5
2
s
2n = x2n+1 x2n+3
s
2n+1 = x2n+3
2
for 1 i < n from Definition 3.8. We know dim R(n, I

t) = dim S(n)/ Gtn =

dim S(n)/ I
Gt . We claim that
n

I
Gtn = (x3 , x5 , . . . , x2n+1 , x2n+3 ) .
This is clear for n ∈ {0, 1}. For n > 1, we prove this by induction. Since s1 = x32

and s
2n+1 = x 2 are in
IG t , we have x , x
3 2n+3 ∈ I
Gt . Since
2n+3 n n

s3 = x52 − x3 x7 ∈ I
Gtn ⊆ I
Gtn

and x3 ∈ I t
Gn , we get x5
2 ∈ It
Gn , so that x5 ∈ I
Gtn . Now suppose x2i−1 , x2i+1 ∈

I
Gtn for 2 i < n. We have

s
2i+1 = x2i+3 − xJ2i+1 x2i+5 ∈ I
2
Gtn ⊆ I
Gtn .

But xJ2i+1 ∈ {x2i−1 , x2i+1 } by Definition 3.8 and {x2i−1 , x2i+1 } ⊆ I Gtn by

induction, so that x 2 ∈ I
2i+3 n

Gt , and hence x2i+3 ∈ IGt . We conclude that
n

(x3 , x5 , . . . , x2n+1 , x2n+3 ) ⊆ I
Gtn ⊆ (x3 , x5 , . . . , x2n+1 , x2n+3 ) ,

so we have equality. Since S(n)/ I ∼ ∼
Gtn = k has dimension zero, so does R(n, t) =
R(n, t)/(Xn ). Thus, the image of Xn is a system of parameters for R(n, t).

Remark 3.11 We note that as a consequence of the proof of the preceding theorem,
t) is Artinian, which will be relevant in Sect. 3.2.
the ring R(n,
Corollary 3.12 The image of
in R(n, t) = S(n)/IGtn is a regular sequence for R(n, t).

Proof We know by Proposition 3.10 that the image of Xn in R(n, t) is a linear
system of parameters. Since the rings R(n, t) are Cohen–Macaulay (Corollary 2.16),
we are done.

3.2 Length, Multiplicity, and Regularity
In this section, we determine the multiplicity and Castelnuovo-Mumford regularity

of the toric rings R(n, t) coming from the associated graphs Gtn ∈ F by computing
the length of the Artinian rings
t) ∼
R(n, = R(n, t)/(Xn )
from Definition 3.7. We know by Corollary 3.12 that Xn is a linear regular sequence
for R(n, t), which allows us to compute the multiplicity of the rings R(n, t). As
a corollary of Theorem 3.16, which establishes the Hilbert function for R(n, t),
we obtain the multiplicity and regularity of R(n, t). We also develop an alternate
38 L. Ballard
graph-theoretic proof for the regularity of R(n, t), which is included at the end of
this section.
We begin with a lemma establishing a vector space basis for R(n,t), which we
use extensively for our results.
Lemma 3.13 The image of all squarefree monomials with only odd indices whose
indices are at least four apart, together with the image of 1k , forms a vector space
t) over k.
basis for R(n,
t) and then find the initial ideal
Proof We recall for the reader the definition of R(n,

of IGn and use Macaulay’s Basis Theorem to show that the desired representatives
t
t) as a vector space over k.

form a basis for R(n,
From Definition 3.7, we have
R(n, t)/(Xn ) ∼ t) = S(n)/

= R(n, I Gtn ,
where
= k[x3 , x5 , . . . , x2n+1 , x2n+3 ]

S(n)
and where I Gtn has generators s

1 , · · · , s
2n+1 as given in Definition 3.8.
We first show that the image of the monomials with the desired property is a basis
by the initial ideal in> I
in the quotient of S(n) Gtn . By Macaulay’s Basis Theorem,
which is Theorem 1.5.7 in [21], the image of these monomials in R(n, t) is also a
basis.
To find the initial ideal of IGtn , we establish that the given generators s k are a

Gröbner basis for IGtn with respect to the degree reverse lexicographic order >.
This is a relatively straightforward computation using Buchberger’s Criterion, with
separate cases for when xJk = 0.
If we adopt S-polynomial notation Si,j for the S-polynomial of si and sj , then
the cases to consider are
S1,2 , S2,3 , S2,4 , S2n−2,2n , S2n−1,2n , S2n,2n+1
S2i−1,2i for 1 < i < n
S2i,2i+1 for 1 < i < n
S2i,2i+2 for 1 < i < n − 1.
To give a flavor of the computation involved, we show the case S2i,2i+2 for 1 < i <
n − 1, and leave the remaining cases to the reader. We show in each subcase that
so
S2i,2i+2 is equal to zero or to a sum of basis elements with coefficients in S(n),
that the reduced form of S2i,2i+2 is zero in each subcase. We have
S2i,2i+2 = x2i+5 (x2i+1 x2i+3 − xJ2i x2i+5 ) − x2i+1 (x2i+3 x2i+5 − xJ2i+2 x2i+7 )
= −xJ2i x2i+5
2
+ x2i+1 xJ2i+2 x2i+7
Case 1: If ti+2 = 0, then xJ2i+2 = xJ2i and J2i+3 = 2i + 1, so we have
S2i,2i+2 = −xJ2i x2i+5

2
+ xJ2i x2i+1 x2i+7
= −xJ2i s
2i+3
We note that if xJ2i = xJ2i+2 = 0, then S2i,2i+2 = 0.
Case 2: If ti+2 = 1, then xJ2i+2 = xJ2i+1 and J2i+3 = 2i + 3, so we have
S2i,2i+2 = −xJ2i x2i+5

2
+ x2i+1 xJ2i+1 x2i+7 .
Case 2.1: If in addition ti+1 = 0, then xJ2i = xJ2i−2 and J2i+1 = 2i − 1, so

we have
S2i,2i+2 = −xJ2i−2 x2i+5

2
+ x2i+1 x2i−1 x2i+7
= −xJ2i−2 (x2i+5
2
− xJ2i+3 x2i+7 ) + x2i+7 (x2i−1 x2i+1 − xJ2i−2 x2i+3 )
= −xJ2i−2 s
2i+3 + x2i+7 s
2i−2 .
We note that if xJ2i = xJ2i−2 = 0, then S2i,2i+2 = x2i+7 s

2i−2 .
Case 2.2: If in addition ti+1 = 1, then xJ2i = xJ2i−1 and J2i+1 = 2i + 1, so

we have
S2i,2i+2 = −xJ2i−1 x2i+5

2
+ x2i+1
2
x2i+7
= −xJ2i−1 (x2i+5
2
− xJ2i+3 x2i+7 ) + x2i+7 (x2i+1
2
− xJ2i−1 x2i+3 )
= −xJ2i−1 s
2i+3 + x2i+7 s
2i−1 .
This concludes the case S2i,2i+2 for 1 < i < n − 1. The remaining cases are similar.
Then the given generators sk are a Gröbner Basis for I
Gtn , so that the initial ideal
is
in> (I
Gtn ) = (x3 , {x2i+1 x2i+3 , x2i+3 | 1 i n})
2 2
= k[x3 , x5 , . . . , x2n+1 , x2n+3 ]. Since in> (I

in the ring S(n) Gtn ) consists precisely
and all degree two products of variables whose
of all squares of variables in S(n)
indices differ by exactly two, it follows that the image of the squarefree monomials
whose indices are at least four apart, together with the image of 1k , forms a basis for
40 L. Ballard

S(n)
. By Macaulay’s t) =
Basis Theorem, the image of these monomials in R(n,
in> I
Gtn

S(n)
It
is also a basis.

Gn
We use the lemma above to establish facts about the vector space dimensions of
t), which are applied further below to establish length and
degree i pieces of R(n,
multiplicity.
Notation 3.14 Throughout this section, we use dn,i := dimk (R(n, t))i for the

vector space dimension of the degree i piece of R(n, t), that is, for the ith coefficient
t). By Lemma 3.13, these are independent of t.
in the Hilbert series of R(n,
We establish a recursive relationship between these dimensions by introducing a
short exact sequence of vector spaces.
Lemma 3.15 For n 2 and i 1, the vector space dimension dn,i =
t))i satisfies the recursive relationship
dimk (R(n,
dn,i = dn−1,i + dn−2,i−1 .
Proof We use the vector space basis defined in Lemma 3.13. We note that the basis
elements described are actually monomial representatives (which do not depend on
t) of equivalence classes (which do depend on t), but we suppress this and speak as
if they are monomials, not depending on t. We then take the liberty of suppressing t
in what follows, for convenience. We recall for the reader that
= k[x3 , x5 , . . . , x2n−3 , x2n−1 , x2n+1 , x2n+3 ]

S(n)

S(n − 1) = k[x3 , x5 , . . . , x2n−3 , x2n−1 , x2n+1 ]

S(n − 2) = k[x3 , x5 , . . . , x2n−3 , x2n−1 ]

Let x2n+3 : (R(n i be multiplication by x2n+3 , and let
− 2))i−1 → (R(n))
x i → (R(n
2n+3 : (R(n)) − 1))i be defined for a basis element b by

b if x2n+3 b
x
2n+3 (b) =
0 if x2n+3 | b.
We note that these vector space maps are well-defined, since 1k or a squarefree
monomial with odd indices at least four apart has an output of 0, 1, or a monomial
with the same properties. The following sequence of vector spaces is exact
so that
dn,i = dn−1,i + dn−2,i−1 .

Applying Lemma 3.15 and induction, we achieve the following closed formula
t).
for the coefficients of the Hilbert series of R(n,
t) ∼
Theorem 3.16 If R(n, t) = S(n)/IGtn and R(n, = R(n, t)/(Xn ), we have
⎧
⎪
⎪ 1 i=0
⎪
⎪
⎪
⎨1 i
t))i =
dimk (R(n, (n + j − 2(i − 1)) 1 i n/2 + 1
⎪
⎪ i!
⎪
⎪ j =1
⎪
⎩0 i > n/2 + 1.
Proof We establish the base cases i, n ∈ {0, 1} and proceed by induction, using
Notation 3.14. It is clear that dn,0 = 1, generated by 1k . By Lemma 3.13 and by
t) is a graded quotient, every nonzero element of positive degree
the fact that R(n,
i can be represented uniquely as a sum of degree i squarefree monomials with odd
t))1 is generated by the
indices whose indices are at least four apart. Then (R(n,
images of all the odd variables
x3 , x2(1)+3 , . . . , x2n+3
in S(n), so that
1
1
dn,1 = n + 1 = (n + j − 2(1 − 1))
1!
j =1
matches the given formula.

Now we establish the base cases n = 0 and n = 1. We recognize that the first
monomial of degree two with odd indices at least four apart is x3 x7 , which does not
exist until n = 2, so we have
⎧
⎪
⎪1 i = 0
⎨
d0,i = 1 i = 1
⎪
⎪
⎩0 i > 1
and
⎧
⎪
⎨1 i = 0
⎪
d1,i = 2 i=1
⎪
⎪
⎩0 i > 1
42 L. Ballard
which match the given formula.

This gives us the following table of base cases for dn,i , which match the given
formula:
n\i 0 1 2 3 4 5 6 7 ···
0 1 1 0 0 0 0 0 0 ···
1 1 2 0 0 0 0 0 0 ···
2 1 3
3 1 4
4 1 5
.. .. ..
. . .
We recall by Lemma 3.15 that we have the recursive relationship
dn,i = dn−1,i + dn−2,i−1
for n 2 and i 1. We proceed by induction. Suppose that N 2 and 2 I <

N/2 + 1 such that the dimension formula holds for all i when n < N. We recognize
that we have 1 I (N − 1)/2 + 1 and 1 I − 1 (N − 2)/2 + 1, so that by
our recursion and by induction, we have
dN,I = dN −1,I + dN −2,I −1

−1
1
I I
1
= (N − 1 + j − 2(I − 1)) + (N − 2 + j − 2(I − 2))
I! (I − 1)!
j =1 j =1
I −1
I −1
1 1
= (N − 2(I − 1)) (N + j − 2(I − 1)) + (I ) (N + j − 2(I − 1))
I! I!
j =1 j =1
1
I
= (N + j − 2(I − 1)),
I!
j =1
as desired.
For the special case where I = N/2 + 1, there is only one possible monic
monomial of degree N/2 + 1 with odd indices at least four apart,

N/2+1
x4j −1 ,
j =1
so that dN,N/2+1 = 1. This matches the formula, since

1
(N/2+1)
1
(N/2+1)
(N + j − 2(N/2)) = j = 1.
(N/2 + 1)! (N/2 + 1)!
j =1 j =1
The remaining case is I > N/2 + 1. In this case, we have I > (N − 1)/2 + 1 and
I −1 > (N −2)/2+1, so that dN,I = 0 by our recursive formula and induction.
Remark 3.17 We note from the proof of the theorem above a few facts for future
t)) = 1 + 1 = 2 and
(R(1,
reference. By our base cases, we have
(R(0, t)) =
1 + 2 = 3. Taking the Fibonacci sequence F (n) with F (0) = 0 and F (1) = 1, we
have F (2) = 1, F (3) = 2, and F (4) = 3, so that
t)) = F (3)

(R(0,
t)) = F (4).

(R(1,
These facts become useful in Proposition 3.19. We also note that dn,n/2+1 = 1 when
n is even, which we use in the following corollary.
It follows quickly from Theorem 3.16 that the regularity of R(n, t) is n/2 + 1.
For an alternate proof of the regularity of R(n, t) which uses different machinery
and more graph-theoretic properties, see the end of this section.
Corollary 3.18 For Gtn ∈ F,
reg R(n, t) = n/2 + 1.
t) is Artinian by Remark 3.11, it is clear that reg R(n,

Proof Since R(n, t) is the top
t). By Theorem 3.16 and Remark 3.17, we only need show
nonzero degree of R(n,
that dn,n/2+1 = 0 when n is odd. We have
1
(n+1)/2
dn,n/2+1 = dn, n+1 = (j + 1) = 0.
2 ((n + 1)/2)!
j =1
t) is an Artinian quotient of R(n, t) by a linear regular sequence, we

Since R(n,
conclude that
t) = n/2 + 1.
reg R(n, t) = reg R(n,

t),
In the following, we first compute the lengths of the dimension zero rings R(n,
and then show a closed form for the multiplicity of our original rings R(n, t) by
t) and applying
using a Fibonacci relationship between the lengths of the rings R(n,
Binet’s formula for F (n), the nth number in the Fibonacci sequence:
44 L. Ballard
√ n √
(1 + 5) − (1 − 5)n
F (n) = √ .
2n 5
In the theorem and corollaries which follow, we suppress t for convenience, since
the statements are independent of t.
satisfy the recursive formula
Proposition 3.19 The lengths of the rings R(n)
=
(R(n

(R(n))
− 1)) +
(R(n − 2))
for n 2. Consequently, if F (n) is the Fibonacci sequence, with F (0) = 0 and

F (1) = 1, then
√ √
= F (n + 3) = (1 + 5)n+3 − (1 − 5)n+3

(R(n)) √ .
2n+3 5
Proof Again, we use Notation 3.14. By the recursive relationship from

Lemma 3.15, since dn,0 = 1 in general, and since dn,i = 0 in general for i > n/2+1
by Theorem 3.16, we have for n 2 that
n/2+1
n/2+1
!
=

(R(n)) dn,i = 1 + dn−1,i + dn−2,i−1
i=0 i=1
n/2+1
(n−2)/2+1

= dn−1,i + dn−2,i
i=0 i=0

=
(R(n
− 1)) +
(R(n − 2)).
Now we show the second statement. For our base cases, we see from Remark 3.17
= F (3) = F (0 + 3) and that
(R(1))
that
(R(0)) = F (4) = F (1 + 3).

Now suppose that
(R(n − 1)) = F (n + 2) and
(R(n − 2)) = F (n + 1).
Then we have
=
(R(n

(R(n))
− 1)) +
(R(n − 2)) = F (n + 3) ,
follows directly from Binet’s formula for

as desired. The closed form for
(R(n))
the Fibonacci sequence.

Corollary 3.20 For n 2, there is an equality of multiplicities
= e(R(n
e(R(n))
− 1)) + e(R(n − 2)).
and hence the

Proof We have established the length of the Artinian rings R(n),

multiplicity e(R(n)).

Corollary 3.21 For n 2, there is an equality of multiplicities
e(R(n)) = e(R(n − 1)) + e(R(n − 2)).
In particular,
√ √ n+3
(1 + 5)n+3 − (1 − 5)
e(R(n)) = F (n + 3) = √ .
2n+3 5
= R(n)/(Xn ), which by
Proof To obtain the multiplicity of R(n), we look at R(n)
Remark 3.11 and Corollary 3.12 is the Artinian quotient of R(n) by a linear regular
sequence. By a standard result, we may calculate length along the obvious short
exact sequences coming from multiplication by elements of our regular sequence to
obtain the equality
HilbR(n) (t)(1 − t)d = Hilb

R(n)
(t),
where d is the Krull dimension of R(n). Defining multiplicity as in and preceding

[24, Thm 16.7], it follows immediately that
"
e(R(n)) = HilbR(n) (t)(1 − t)d "t=1 = Hilb
R(n)

(1) =
(R(n)).
We are done by Proposition 3.19.

We reintroduce t and spend the remainder of this section providing an alternate
graph-theoretic proof for the regularity of R(n, t).
Alternate Proof of Corollary 3.18 We show that reg R(n, t) = n/2 + 1 by
proving that reg IGtn = n/2 + 2. We first show that
reg IGtn n/2 + 2.
We recall by Proposition 2.13 that the graph Gtn is chordal bipartite with vertex
bipartition V1 ∪ V2 of cardinalities
n
|V1 | = +2
2
n
|V2 | = + 2,
2
and that Gtn does not have any vertices of degree one by Remark 2.6. Then by
Theorem 4.9 of [3], we have
# n n $ n
reg IGtn min + 2, +2 = + 2.
2 2 2
46 L. Ballard
% &
We note that we may equivalently prove reg R(n, t) n2 + 1 by choosing the
%n&
2 + 2 edges whose indices are equivalent to zero modulo 4, one from each row
of Ltn , to obtain an edge matching (different from an induced matching, below) and
then applying [14, Th 1].
We now show that reg IGtn n/2 + 2. Since IGtn is homogeneous and in> IGtn
consists of squarefree monomials by Corollary 3.3, we have by Corollary 2.7 of [4]
that reg in> IGtn = reg IGtn , so it suffices to prove that reg in> IGtn n/2 + 2. The
ideal in> IGtn can be viewed as the edge ideal of a simple graph, a “comb” with n + 1
tines, with consecutive odd variables corresponding to vertices along the spine, as
pictured below:
2 4 6 8 2 2 2
3 5 7 9 ... 2 1 2 3
We know from Theorem 6.5 of [13] that the regularity of an edge ideal is
bounded below by the number of edges in any induced matching plus one, so we
choose n/2 + 1 edges (tines) corresponding to certain odd variables that create
an induced matching. By beginning with the x3 -tine and choosing every other tine
corresponding to the variables
x3 , x3+4(1) , . . . , x3+4(n/2) ,
we obtain n/2 + 1 edges that are an induced matching, so we have
reg in> IGtn n/2 + 2,
as desired.
We conclude that reg IGtn = n/2 + 2, and hence that reg R(n, t) = n/2 + 1.

Acknowledgments Macaulay2 [11] was used for computation and hypothesis formation. We
would like to thank Syracuse University for its support and hospitality and Claudia Miller for
her valuable input on the original project in [1] and this condensed version. We would also like to
thank the referees for noticing errors in the statement and proof of Theorem 3.16 as well as in the
proof of Lemma 3.1, and for highlighting with clarity the connection between Theorem 3.4 and
work done by Rafael Villarreal. We acknowledge the partial support of an NSF grant.
References
1. L. Ballard. Properties of the Toric Rings of a Chordal Bipartite Family of Graphs. PhD thesis,
Syracuse University, 2020.
2. S. K. Beyarslan, H. T. Há, and A. O’Keefe. Algebraic properties of toric rings of graphs.
Communications in Algebra, 47(1):1–16, 2019.
3. J. Biermann, A. O’Keefe, and A. Van Tuyl. Bounds on the regularity of toric ideals of graphs.
Advances in Applied Mathematics, 85:84–102, 2017.
4. A. Conca and M. Varbaro. Square-free Gröbner degenerations. Inventiones Mathematicae,
221(3):713–730, 2020.
5. A. Corso and U. Nagel. Monomial and toric ideals associated to Ferrers graphs. Transactions
of the American Mathematical Society, 361(3):1371–1395, 2009.
6. D. Cox, J. Little, and D. O’Shea. Ideals, varieties, and algorithms. Undergraduate Texts
in Mathematics. Springer, New York, third edition, 2007. An introduction to computational
algebraic geometry and commutative algebra.
7. A. D’Alí. Toric ideals associated with gap-free graphs. Journal of Pure and Applied Algebra,
219(9):3862–3872, 2015.
8. G. Favacchio, G. Keiper, and A. Van Tuyl. Regularity and h-polynomials of toric ideals of
graphs. Proceedings of the American Mathematical Society, 148:4665–4677, 2020.
9. F. Galetto, J. Hofscheier, G. Keiper, C. Kohne, M. E. Uribe Paczka, and A. Van Tuyl. Betti
numbers of toric ideals of graphs: a case study. Journal of Algebra and its Applications,
18(12):1950226, 2019.
10. I. Gitler and C. E. Valencia. Multiplicities of edge subrings. Discrete Mathematics,
302(1):107–123, 2005.
11. D. R. Grayson and M. E. Stillman. Macaulay2, a software system for research in algebraic
geometry. Available at http://www.math.uiuc.edu/Macaulay2/.
12. Z. Greif and J. McCullough. Green-Lazarsfeld condition for toric edge ideals of bipartite
graphs. Journal of Algebra, 562:1–27, 2020.
13. H. T. Hà and A. Van Tuyl. Monomial ideals, edge ideals of hypergraphs, and their graded Betti
numbers. Journal of Algebraic Combinatorics, 27(2):215–245, 2008.
14. J. Herzog and T. Hibi. The regularity of edge rings and matching numbers. Mathematics,
8(1):39, 2020.
15. J. Herzog, T. Hibi, and H. Ohsugi. Binomial ideals, volume 279 of Graduate Texts in
Mathematics. Springer, Cham, 2018.
16. T. Hibi, A. Higashitani, K. Kimura, and A. O’Keefe. Depth of edge rings arising from finite
graphs. Proceedings of the American Mathematical Society, 139(11):3807–3813, 2011.
17. T. Hibi and L. Katthän. Edge rings satisfying Serre’s condition (R1 ). Proceedings of the
American Mathematical Society, 142(7):2537–2541, 2014.
18. T. Hibi, K. Matsuda, and H. Ohsugi. Strongly Koszul edge rings. Acta Mathematica
Vietnamica, 41(1):69–76, 2016.
19. T. Hibi, K. Matsuda, and A. Tsuchiya. Edge rings with 3-linear resolutions. Proceedings of
the American Mathematical Society, 147(8):3225–3232, 2019.
20. T. Hibi and H. Ohsugi. Toric ideals generated by quadratic binomials. Journal of Algebra,
218(2):509–527, 1999.
21. M. Kreuzer and L. Robbiano. Computational Commutative Algebra 1. Springer, Berlin,
Heidelberg, 2000.
22. K. Mori, H. Ohsugi, and A. Tsuchiya. Edge rings with q-linear resolutions. Preprint,
arxiv.org/abs/2010.02854.
23. R. Nandi and R. Nanduri. On Betti numbers of toric algebras of certain bipartite graphs.
Journal of Algebra and Its Applications, 18(12):1950231, 2019.
24. I. Peeva. Graded Syzygies. Springer London, London, 2011.
25. A. Simis, W. V. Vasconcelos, and R. H. Villarreal. On the ideal theory of graphs. Journal of
Algebra, 167(2):389–416, 1994.
26. C. Tatakis and A. Thoma. On the universal Gröbner bases of toric ideals of graphs. Journal of
Combinatorial Theory, Series A, 118(5):1540–1548, 2011.
27. R. H. Villarreal. Rees algebras of edge ideals. Communications in Algebra, 23(9):3513–3524,
1995.
An Illustrated View of Differential
Operators of a Reduced Quotient of an
Affine Semigroup Ring
Christine Berkesch, C-Y. Jean Chan, Patricia Klein,

Laura Felicia Matusevich, Janet Page, and Janet Vassilev
Keywords Differential operators · Quotient · Reduced · Semigroup ring ·

Toric · Toric variety
1 Introduction
In this paper, we provide illustrative examples and visualizations of some differ-

ential operators on the quotient of an affine semigroup ring by a radical monomial
ideal. These examples motivate our work in [1]. For a finitely-generated algebra R
over a field, let D(R) denote the ring of differential operators of R and use ∗ to
denote an action of a differential operator. A foundational result in this area relates
CB was partially supported by NSF DMS 2001101.

LFM was partially supported by the Simons Foundation.
C. Berkesch () · P. Klein

University of Minnesota, Minneapolis, MN, USA
e-mail: cberkesc@umn.edu; klein847@umn.edu
C-Y. J. Chan
Department of Mathematics, Central Michigan University, Mt. Pleasant, MI, USA
e-mail: chan1cj@cmich.edu
L. F. Matusevich
Texas A & M University, College Station, TX, USA
e-mail: laura@math.tamu.edu
J. Page
University of Michigan, Ann Arbor, MI, USA
e-mail: jrpage@umich.edu
J. Vassilev
University of New Mexico, Albuquerque, NM, USA
e-mail: jvassil@math.unm.edu

50 C. Berkesch et al.
the ring of differential operators of an arbitrary quotient of a polynomial ring by an

ideal J to differential operators on that polynomial ring and the idealizer of J .
Proposition 1.1 ([8, Propostion 1.6]) Let S be the coordinate ring of a nonsingu-
lar affine variety over an algebraically closed field of characteristic 0, and let J be
an S-ideal. Then there is an isomorphism

S ∼ I(J )
D = , (1.1)
J J D(S)
where I(J ) := {δ ∈ D(R) | δ ∗ J ⊂ J }, which is called the idealizer of J .

The phrasing of [8, Proposition 1.6] given above differs from the original
by making use of the equivalence of the conditions, with notation as above, of
δJ D(S) ⊂ J D(S) and δ∗J ⊂ J [9, Lemma 2.3.1]. Traves [11] uses Proposition 1.1
to give concrete descriptions of rings of differential operators of Stanley–Reisner
rings, and Saito and Traves [9] use the same to compute rings of differential
operators of affine semigroup rings. For a non-regular affine semigroup ring RA
over an algebraically closed field of characteristic 0, Proposition 1.1 fails, even for
a radical monomial ideal J in RA . However, the differential operators on RA that
induce maps on RA /J are precisely those in I(J ). Further, there is an embedding of
rings

I(J ) RA
→ D ,
D(RA , J ) J
so the description of D(RA ) by Saito and Traves can still be used to compute this
subring of D(RA /J ).
One primary goal of this article is to visually illustrate the computation of
I(J )/D(RA , J ) when RA is an affine semigroup ring and J is a radical monomial
ideal in RA ; in [1], we provide an explicit formula for this computation. In particular,
the pictures we provide explain how to compute differential operators of the form
D(I, J ) := {δ ∈ D(R) | δ ∗ I ⊆ J },
where I and J are subsets of the ring R, and ∗ denotes an action by a differential
operator. These sets were originally instrumental in both the work of Smith and
Stafford [8] and of Musson [5, Section 1], and they are essential building blocks of
our computations, as well.
Our second major goal in this paper is to compare J D(RA ) and D(RA , J ) for
an affine semigroup ring RA and radical monomial ideal J of RA . Towards this
end, the first set of examples we consider consists of quotients of the coordinate
rings of rational normal curves. These quotients are all isomorphic to C[x, y]/xy,
which is handled in [11]. From the standpoint of comparing J D(RA ) and D(RA , J ),
what we will see is that J D(RA ) = D(RA , J ) for the rational normal curves of
degrees 1 and 2 but fail to coincide in all degrees larger than 2. For rational normal
An Illustrated View of Differential Operators 51
curves of degree at most 2, J is principally generated and so is J D(RA ). In this

case, it is straightforward to see (by definition) that D(RA , J ) is also principal and
is isomorphic to J D(RA ). However, for degree n > 2, J has n − 1 generators.
We will see that this number of generators greatly impacts I(J )/J D(RA ) but not
I(J )/D(RA , J ).
In [1], we consider cases where
J D(RA ) = D(RA , J ), (1.2)
for the particular ideal J generated by all monomials corresponding to the interior of
the semigroup, i.e., J = ωRA . When RA is Gorenstein and normal, (1.2) holds since
ωRA is principal. We show that the converse is also true; that is, (1.2) is equivalent
to RA being a Gorenstein ring.
To provide intuition beyond the two-dimensional case, we also offer a
three-dimensional normal affine semigroup ring RA for which we compute
I(J )/D(RA , J ) for two choices of J . Then, we return to the two-dimensional
setting to consider a scored but not normal example, as well as a non-scored
example, computing differential operators for quotients of both rings.
Outline Section 2 fixes notation to be used throughout the article and describes
the main result of [9]. Sections 3, 4, and 5 describe I(J )/D(RA , J ) for rational
normal cones of degrees 2, 3, and higher, respectively. Section 6 considers a three-
dimensional normal affine semigroup ring modulo two different choices of radical
monomial ideal J , and Sect. 7 computes I(J )/D(RA , J ) for two different non-
normal two-dimensional affine semigroup rings, where J = ωRA is the radical
monomial ideal corresponding to the interior of the semigroup NA.
2 Background and Notation
In this section, we fix notation and conventions to be assumed throughout the article.
Although the results we discuss in this paper hold over any algebraically closed field
of characteristic 0, we will use in this illustrated view the field of complex numbers
for the sake of simplicity. Having fixed notation, we will then state and discuss [9,
Theorem 3.2.2].
Definition 2.1 Let A be a k ×
matrix with entries in Z. Let NA denote the
semigroup in Zk that is generated by the columns of A. The affine semigroup ring
determined by A is
'
RA = C[NA] = C · ta ,
a∈NA
where ta = t1a1 t2a2 · · · tkak for a = (a1 , a2 , . . . , ak ) ∈ NA. Throughout this article,
we assume that the group generated by the columns of A is the full ambient lattice,
so ZA = Zk , and also that the real positive cone over A, R0 A, is pointed, meaning
that it is strongly convex.
Definition 2.2 A semigroup NA is normal if NA = R0 A ∩ ZA. A semigroup NA
is scored if the difference (R0 A∩ZA)\NA consists of a finite union of hyperplane
sections of R0 A ∩ ZA that are all parallel to facets of the cone R0 A. An affine
semigroup ring C[NA] is said to be normal, or scored, if NA is normal, or scored,
respectively. Note that normal semigroups are scored.
When we write a facet σ of R0 A (or A or NA), we will always mean the integer
points in the corresponding facet of R0 A. When A is normal, this is the same as
the semigroup generated by the columns of A that lie in the corresponding facet of
R0 A.
Throughout, we use − to indicate ideals in the commutative rings RA or the
polynomial ring C[θ ] = C[θ1 , . . . , θk ]. It will be clear from the context if the ideals
are in RA or in C[θ ].
Notice that the Zk -graded prime ideals in RA are in one-to-one correspondence
with the faces of A (or R0 A), as a face τ of A corresponds to the multigraded
prime RA -ideal
( " )
Pτ := td " d ∈ NA \ τ .
In this paper, we compute rings of the form I(J )/D(RA , J ), where J is a radical
monomial ideal in RA ; as such, J is always as intersection of primes of the form Pτ ,
for various faces τ of A.
We are mainly following the description of Saito and Traves [9], although
Musson and Van den Bergh describe the ring of differential operators of a toric ring
C[NA] first in [4, 6] and [7] when viewed as a subring of the ring of differential oper-
ators of the Laurent polynomials, i.e., D(C[Zk ]) = C{t1±1 , . . . , tk±1 , ∂1 , . . . , ∂k }/ ∼,
where ∂i denotes the differential operator ∂t∂ i and ∼ denotes the usual relations on
the free associative algebra C{t1±1 , . . . , tk±1 , ∂1 , . . . , ∂k } that describe the behavior
of differential operators. To explain the differential operators of C[NA] as presented
by Saito and Traves, note first that D(C[Zk ]) carries a Zk -grading, where ei denotes
the i-th column of the identity matrix I and deg(ti ) = ei = − deg(∂i ). Note also
that if ai is a column of A, then deg(tai ) = ai . Set θj = tj ∂j for 1 j k, and set
(d) := {a ∈ NA | a + d ∈
/ NA} = NA \ (−d + NA).
We note also that for any tm ∈ RA and f (θ ) ∈ C[θ ], f (θ ) ∗ tm = f (m)tm . The

idealizer of (d) is defined to be
I((d)) := f (θ ) ∈ C[θ ] = C[θ1 , θ2 . . . , θk ] | f (a) = 0 for all a ∈ (d),

which is viewed as an ideal in the ring C[θ ], where θi = ti ∂i ∈ D(C[Zd ]) is of

degree 0. We will soon see that I((d)) consists of f (θ ) such that td f (θ ) ∈ D(RA ).
To compute I((d)) for a normal semigroup ring, consider a facet σ of A,
recalling that by this we mean all lattice points on the corresponding facet of the
cone R0 A. The primitive integral support function hσ is a uniquely determined
linear form on Rk such that:
(1) hσ (R0 A) 0,
(2) hσ (Rσ ) = 0, and
(3) hσ (Zk ) = Z.
We write σ1 , σ2 , . . . , σm for the facets of A, so that for the remainder of the paper,
we set
hj := hj (θ ) = hσj (θ ).
For a non-negative integer n, we will use the following notation to denote this
descending factorial:

n
(α, n)! := (α − i) = α(α − 1) · · · (α − n),
i=0
where α is a function, which could be constant or already evaluated. For example,
hj (−d)−1
!
hj , hj (−d) − 1) ! = (hj − i)
i=0
will be a common expression throughout this article, as it is a factor in the generator

of the idealizer I((d)). To streamline our presentation, we make the convention
that (α, n)! = 1 for all n < 0.
Theorem 2.3 ([9, Theorem 3.2.2]) If RA is a pointed, normal affine semigroup
ring with ZA = Zk , then
'
D(RA ) = td · I((d)),
d∈Z
d
* +
' !
= t ·
d
hj , hj (−d) − 1 ! , (2.1)
d∈Z
d hj (d)<0
where the outer product runs over those j = 1, 2, . . . , m with hj (d) < 0.
We now turn to an example that illustrates Theorem 2.3, as well as our notation.
d
-d
d
-d
Fig. 1 Half-lines of vanishing for various d; here, d = (−1, 2) and (−2, −1)
Example 2.4 Let R = C[s, st, st 2 , st 3 ], where deg(s) = 1 = deg(t). Let ∂1 and ∂2
denote the partial derivatives with respect to s and t, and let θ1 = s∂1 and θ2 = t∂2 .
By Theorem 2.3, D(R)d , for some d = (d1 , d2 ) ∈ Z2 , is generated by an element
in the form of s d1 t d2 · f (θ1 , θ2 ), where f (θ1 , θ2 ) is a polynomial in θ1 and θ2 .
To understand the philosophy from [9], fix d = (d1 , d2 ) ∈ Z2 , and consider a
differential operator on R = k[s, st, st 2 , st 3 ] of the form δ = s d1 t d2 f (θ1 , θ2 ). By
definition, δ ∗ (s m1 t m2 ) ∈ R for any monomial s m1 t m2 ∈ R. We can check that
δ ∗ s m1 t m2 = f (m)s d1 +m1 t d2 +m2 .
In particular, if s d1 +m1 t d2 +m2 ∈/ R then we must have that δ ∗ s m1 t m2 = 0 so that

f (m) = 0. In summary, δ ∗ (s 1 t m2 ) may be nonzero if and only if s d1 +m1 t d2 +m2 ∈
m
R.
These multidegrees m = (m1 , m2 ) where δ ∗ s m1 t m2 vanishes are exactly the
points in (d). We illustrate examples of (d) for specific d in Fig. 1 by the integral
points that are on the dotted lines in the first quadrant, including those on the positive
horizontal axis.
To explicitly compute the polynomial f = f (θ1 , θ2 ) for a fixed d ∈ Z2 , we
will use the primitive integral support functions for the two facets of A, which are
h1 = θ2 and h2 = 3θ1 − θ2 . Both of these must divide f , along with all of the linear
forms representing the dotted lines in Fig. 1. Specifically, I((−1, 2)) is generated
by
f (θ1 , θ2 ) = h2 (h2 − 1)(h2 − 2)(h2 − 3)(h2 − 4) = (h2 , 4)!,
and I((−2, −1)) is generated by
f (θ1 , θ2 ) = h1 h2 (h2 − 1)(h2 − 2)(h2 − 3)(h2 − 4) = (h1 , 0)!(h2 , 4)!.

For d = (−1, 2), we have h2 (−d) = 5 while h1 (−d) = −2 < 0, so there are no
linear forms involving h1 in this generator. Similarly for d = (−2, −1), h1 (−d) =
1, h2 (−d) = 5. In particular, (h1 , h1 (−d) − 1)! = (h1 , 0)! = h1 and (h2 , h2 (−d) −
1)! = (h2 , 4)! = (h2 − 4)(h2 − 3)(h2 − 2)(h2 − 1)h2 .
As shown in Example 2.4, I((d)) is an ideal in C[θ ], and any polynomial
f (θ ) ∈ C[θ ] has multidegree 0; for any monomial tm ∈ RA , f (θ ) ∗ tm = f (m)tm ,
which belongs to RA . So to determine if the d-th graded piece of D(RA ) applied
to a monomial lands in RA or an RA -ideal J , it is enough to test the membership
of any monomial td+m and then adjust the θ -portion of the differential operator
appropriately. Membership failure for RA only occurs if m lies in NA but outside
of the cone −d + R0 A, whereas for J = RA , failure is more likely to occur.
Since the monomials whose exponents lie on a face of −d + R0 A will lie on the
corresponding face τ of A, if an associated prime of J is a prime ideal associated to
a face that contains τ , then membership failure will also occur for m on this face.
In either case, there are only finitely many potential linear forms to be determined,
and they are of the form hσ (θ ) − hσ (−d) where hσ is the primary support function
of a facet σ .
In [2, 11, 12], Eriksson, Traves, and Tripp separately computed the ring of
differential operators of a Stanley–Reisner ring over an arbitrary field, i.e., the
quotient of any polynomial ring over a field by a squarefree monomial ideal.
We include here the ring of differential operators of an ordinary double point
R = C[x, y]/xy using the viewpoint presented by the above authors, as it exhibits
behavior akin to our computations in this article. In [5], Musson also considered the
differential operators on an ordinary double point.
Example 2.5 The ring of differential operators on C[x, y] is the Weyl algebra
W = C{x, y, ∂x , ∂y }, which is the free associative algebra generated by x, y, ∂x , ∂y ,
subject to the relations:
{xy − yx, ∂x ∂y − ∂y ∂x , ∂x y − y∂x , ∂y x − x∂y , ∂x x − x∂x + 1, ∂y y − y∂y + 1}.
The Weyl algebra W is a graded ring with deg(x) = deg(y) = 1 = − deg(∂x ) =

− deg(∂y ).
For the ordinary double point ring R = C[x, y]/xy, Traves shows in [11,
Theorem 3.5] that the idealizer of xy in W is also graded and generated by
j j
{1, x m , y n , x m y n , x m ∂xi , y n ∂y , x m y n ∂xi ∂y | i m > 0, j n > 0}.
Notice in particular that x m ∂xm and y n ∂yn both have degree 0. Setting θx = x∂x and
θy = y∂y , from the Weyl algebra relations, it follows that

i−1 −1
j
!
∂xi = x −i · (θx −
) = x −i (θx , i − 1)! and ∂y = y −j ·
j
(θy −
) = θy , j − 1 !.

=0
=0
j ! j
Hence x m ∂xi = x m−i · (θx , i − 1)! and y n ∂y = y n−j · θy , j − 1 !, and x m y n ∂xi ∂y
has multidegree (m − i, n − j ) in W . In fact,
' !
W = x m y n · (θx , m − 1)! θy , n − 1 !,
(m,n)∈Z
2
which is a presentation of the Weyl algebra using the Saito–Traves approach from
Theorem 2.3. From this viewpoint, Traves showed that
⎧
⎪
⎪x y · C[θx , θy ]
m n if m, n 0,
⎪
⎪
⎨x m y n · (θ , −m)! if m < 0, n 0,
x
I(xy)(m,n) = !
⎪
⎪ x y · θy , −n !
m n if n < 0, m 0,
⎪
⎪ !
⎩ m n
x y · (θx , −m)! θy , −n ! if m, n < 0.
Further, xyW can be expressed as a multigraded W -ideal that is contained in

I(xy) as follows:
⎧
⎪
⎪x y · C[θx , θy ]
m n
⎪ if m, n > 0,
⎪
⎨x m y n · (θ , −m)!
x if m 0, n > 0,
xyW(m,n) = !
⎪x m y n · θy , −n !
⎪ if n 0, m > 0,
⎪
⎪ !
⎩ m n
x y · (θx , −m)! θy , −n ! if m, n 0.
Now, applying Proposition 1.1 yields a computation for the ring of differential
operators for the ordinary double point R = C[x, y]/xy:
⎧
⎪
⎪ 0 if mn = 0,
⎪
⎪
⎪
⎪ C[θx , θy ]
⎪
⎪ if m = n = 0,
⎪
⎨ θx θy
C[x, y]
D = (θx , −m)! (2.2)
xy ⎪
⎪ xm · if m = 0, n = 0,
(m,n) ⎪
⎪ (θx , −m)!θ! y
⎪
⎪
⎪
⎪y n · θ y , −n !
⎪
⎩ ! if n = 0, m = 0.
θy , −m !θx
We next fix some notation to be used in the remainder of this paper. Whenever
the dimension of the semigroup ring under consideration is k = 2, instead of using
t1 , t2 as our variables, we will use s, t. Further, when considering subsets of R2 and
R3 that contain the lattice points that describe a set of monomials in our semigroup,
such as lines or planes, we will describe them with the variables x, y, and z, for
example, the line y = 2x − 1 in R2 or the plane x − z = 2 in R3 .
Consider the matrix

1 1 1 ··· 1
An = .
0 1 2 ··· n
We call RAn = C[NAn ] = C[s, st, st 2 , . . . st n ] the ring of the rational normal curve
of degree n, since it is the coordinate ring of the affine cone of the projective rational
normal curve. This ring will be the subject of the next three sections. The two facets
of An are

1
σ1 = N = {(x, y) ∈ N2 | x 0, y = 0} and
0

1
σ2 = N = {(x, y) ∈ N2 | x 0, y = nx}.
n
The prime ideal associated to σ1 is P1 = st, st 2 , . . . , st n , and the prime ideal

associated to σ2 is P2 = s, st, . . . , st n−1 . We will consider the radical ideal

st, st 2 , . . . , st n−1 if n > 1,
J = P1 ∩ P2 =
s 2 t if n = 1.
∼ C[x, y]/xy for all n > 0. At the end of Sect. 5, we revisit

Observe that RAn /J =
Example 2.5 and present a C-algebra isomorphism between D(C[x, y]/xy) and
D(RAn /J ) and compare it to our calculations for I(J )/D(RAn , J ).
3 Differential Operators on the Rational Normal Curve of

Degree 2
In this section, we compute the idealizer, I(I ), along with the subset of differential
operators D(RA2 , I ) for the ideals I = P1 = st, st 2 and I = J = st over
the ring of the rational normal curve of degree 2, RA2 = C[s, st, st 2 ]. To aid our
computations we include illustrations of the lattice representing the multidegrees in
the plane broken down into four chambers where the operators will be determined
by similar expressions. The facets of A2 are
σ1 = {(x, y) ∈ N2 | x 0, y = 0} and σ2 = {(x, y) ∈ N2 | x, y 0, y = 2x},
which have primitive integral support functions
h1 = θ2 and h2 = 2θ1 − θ2 .
Figure 2 illustrates the integer lattice, divided into four chambers that are colored as
follows:
Fig. 2 Chambers of D(RA2 )
C1 : The red multidegrees correspond to monomials in J , the gray multidegrees

correspond to monomials in P1 \ J and the blue multidegrees correspond to
monomials in RA2 \ P1 ,
C2 : The yellow multidegrees are the d with h1 (d) 0 and h2 (d) < 0,
C3 : The violet multidegrees are the d with both h1 (d) < 0 and h2 (d) < 0, and
C4 : The green multidegrees are the d with h1 (d) < 0 and h2 (d) 0.
Still following the convention (h, n)! = 1 if n < 0, by Theorem 2.3, the graded
pieces of D(RA2 ) are
D(RA2 )d = s d1 t d2 · (h1 , h1 (−d) − 1)! (h2 , h2 (−d) − 1)! .
Broken down by chambers, this amounts to:

⎧
⎪
⎪s d1 t d2 · C[θ ] if d ∈ C1,
⎪
⎪
⎨s d1 t d2 · (h , −2d + d − 1)! if d ∈ C2,
2 1 2
D(RA2 )d =
⎪
⎪s t · (h1 , −d2 − 1)! (h2 , −2d1 + d2 − 1)!
d 1 d 2 if d ∈ C3,
⎪
⎪
⎩ d1 d2
s t · (h1 , −d2 − 1)! if d ∈ C4.
Example 3.1 We first compute the graded pieces of the sets of differential operators
I(P1 ) and D(RA2 , P1 ). Later, since C[x] ∼
= RA2 /P1 , we will exhibit a C-algebra
isomorphism between D(C[x]) and D(RA2 /P1 ).
To begin, recall that
I(P1 ) = {δ ∈ D(RA2 ) | δ ∗ P1 ⊆ P1 }
and
D(RA2 , P1 ) = {δ ∈ D(RA2 ) | δ ∗ R ⊆ P1 }.
Now if d ∈ C1 and s m1 t m2 ∈ P1 or if d ∈ C1 \ σ1 and s m1 t m2 ∈ RA2 , then for any

g(θ ) ∈ C[θ ],
(a) (b)
Fig. 3 Lines parallel to the facets. (a) Lines parallel to σ2 . (b) Lines parallel to σ1
s d1 t d2 · g(θ ) ∗ s m1 t m2 = g(m)s d1 +m1 t d2 +m2 ∈ P1 , so
I(P1 )d = D(RA2 )d for all d ∈ C1 and D(RA2 , P1 )d = D(RA2 )d for all d ∈ C1\σ1 .
With the aid of Fig. 3, we will explain how to determine I(P1 ) and D(RA2 , P1 )
in the other chambers. The red lattice points in Fig. 3, indicate the monomials in P1 .
First, note that if d ∈ C2 and s m1 t m2 ∈ P1 or d ∈ C2 \ (−σ1 ) and
s t m2 ∈ RA2 , and m lies on one of the lines y = 2x − r shown in Fig. 3a, then
m 1
s d1 t d2 (h2 , −2d1 + d2 − 1)!, the generator of D(RA2 )d , applied to such a monomial

will either be a constant times a monomial represented by a red lattice point on one
of the lines y = 2x − j for 0 j r or, when (h2 (m), −2d1 + d2 − 1)! = 0, it
will be 0. Hence,
I(P1 )d = D(RA2 )d for d ∈ C2 and D(R, P1 )d = D(RA2 )d for d ∈ C2 \ (−σ1 ).
Now if d ∈ C4 and s m1 t m2 ∈ P1 or d ∈ C4 ∪ σ1 and s m1 t m2 ∈ RA2 and m lies

on one of the lines y = r shown in Fig. 3b, then s d1 t d2 (h1 , −d2 − 1)!, the generator
of D(RA2 )d , applied to such a monomial will either be a constant times a monomial
represented by a red lattice point on one of the lines y = j for 0 j r or, when
(h1 (m), −2d1 + d2 − 1)! = 0, it will be 0. However, the monomials represented by
the lattice points on y = 0 do not lie in P1 . The monomials represented by the red
lattice points on the line y = −d2 are precisely the monomials whose image is a
term represented by a lattice point on σ1 . Hence, we need to further right-multiply
any operator in D(RA2 )d by θ2 + d2 , so that
I(P1 )d = s d1 t d2 · (h1 , −d2 )! for d ∈ C4 and

D(RA2 , P1 ) = s d1 t d2 · (h1 , −d2 )! for d ∈ C4 ∪ σ1 .
Using similar reasoning, we can easily see that
I(P1 )d = s d1 t d2 · (h1 , −d2 )! (h2 , −2d1 + d2 − 1)! for d ∈ C3 and

D(RA2 , P1 )d = s d1 t d2 · (h1 , −d2 )! (h2 , −2d1 + d2 − 1)! for d ∈ C3 ∪ (−σ1 ).
Putting these all together, the graded pieces of I(P1 ) and D(RA2 , P1 ) are as
follows:
⎧
⎪
⎪ s d1 t d2 · C[θ ] if d ∈ C1 = NA2 ,
⎪
⎪
⎨s d1 t d2 · (h , −2d + d − 1)! if d ∈ C2,
2 1 2
I(P1 )d =
⎪
⎪ s t · (h1 , −d2 )! (h2 , −2d1 + d2 − 1)!
d 1 d 2 if d ∈ C3,
⎪
⎪
⎩ d1 d2
s t · (h1 , −d2 )! if d ∈ C4,
⎧
⎪
⎪ s d1 t d2 · C[θ ] if s d1 t d2 ∈ P1 ,
⎪
⎪
⎨s d1 t d2 · (h2 , −2d1 + d2 − 1)! if d ∈ C2 \ (−σ1 ),
D(RA2 , P1 )d =
⎪
⎪ s d1 t d2 · (h1 , −d2 )! (h2 , −2d1 + d2 − 1)! if d ∈ C3 ∪ (−σ1 ),
⎪
⎪
⎩ d1 d2
s t · (h1 , −d2 )! if d ∈ C4 ∪ σ1 .
Now taking the quotient, we obtain:

⎧
⎪
⎪ 0 if d ∈
/ Zσ1 ,
⎪
⎪
I(P1 ) ⎨ d1 C[θ ]
= s · h1 if d ∈ σ1 ,
D(RA2 , P1 ) d ⎪
⎪
⎪
⎪ (h2 , −2d1 − 1)!
⎩s d1 · if d ∈ (−σ1 \ 0).
h1 (h2 , −2d1 − 1)!
Viewing D(C[x]) as a Z-graded algebra over C[θx ], we note that

∞
∞

D(C[x]) = ∂xd · C[θx ] ⊕ x d C[θx ] = x d (θx , −d − 1)! · C[θx ].
d=1 d=0 d∈Z
The map
(h2 , −2d − 1)!
φ: x d · (θx , −d − 1)! · C[θx ] → sd · ,
h1 (h2 , −2d − 1)!
d∈Z d∈Z
which is defined on the generators by
φ(x d (θx , −d − 1)!) = s d (h2 , −2d − 1)! + h1 (h2 , −2d − 1)!,
although a C-vector space isomorphism, does not produce a ring isomorphism. The
graded pieces in negative degree are generated by polynomials in θ whose degrees
are twice as large as large as the degrees in θx given in the Weyl algebra. In fact,
⎧
⎪
⎪ if d ∈
/ Zσ1 ,
⎪0
⎪
⎪
⎪ C[θ ]
⎪ s d1 · , -
⎪ if d ∈ σ1 ,
⎪
⎪ h1
⎪
⎨
D(RA2 /P1 ) = 2
, -
⎪
⎪ h2
⎪
⎪ , −d 1 − 1 !
⎪
⎪ 2
⎪
⎪ s ·,
d 1 - if d ∈ (−σ1 \ 0).
⎪
⎪ h1 h2
⎪
⎩ , −d1 − 1 !
2 2
Therefore, we can produce an isomorphism of graded rings ψ : D(C[x]) →

D(RA2 /P1 ) defined for any m ∈ N by
, -
−m h2 h 1 h2
ψ(x ) = s ,
m m
and ψ(∂xm ) =s · ,m − 1 ! + ,m − 1 ! .
2 2 2
Example 3.2 We will now compute the graded pieces of the sets of differential
operators I(J ) and D(RA2 , J ), as well as the graded pieces of J D(RA2 ). Recall
that
I(J ) = {δ ∈ D(RA2 ) | δ ∗ J ⊆ J } and D(RA2 , J ) = {δ ∈ D(RA2 ) | δ ∗ R ⊆ J }.
We will soon give a general formula for the graded pieces of I(J ) and D(RA2 , J );
however, for illustrative purposes, first consider the graded piece of D(RA2 ) at
(−1, 0): s −1 · (h2 , 1)!. Applying s −1 · (h2 , 1)! to a monomial whose exponent
lies in the two parallel half-lines σ2 and y = 2x − 1 in NA will yield 0, which
certainly lives inside J . However, when we let s −1 · (h2 , 1)! act on a monomial
whose exponent is a member of the half-lines y = 2x − 2 or y = 0 lying inside
NA2 , we obtain an integer multiple of a monomial whose exponent lies in one of the
facets of A2 , σ2 or σ1 respectively, and these are not in J . The remaining monomials
in J yield another element of J when they are acted upon by any operator in
s −1 · (h2 , 1)! .
In Fig. 4, the two light blue lines indicate the two half-lines representing
the multidegrees of monomials in RA2 that, after application of an element in
D(RA2 )(−1,0) , fails to yield an element in J . To correct for this lack of membership
in J for the monomials on y = 2x − 2, every element of D(RA2 )(−1,0) should be
multiplied by (h2 − 2); applying s −1 · (h2 , 2)! to these monomials yields 0. The
application of s −1 · (h2 , 2)! to the remaining monomials in J will output a term
inside J . Thus,
I(J )(−1,0) = s −1 · (h2 , 2)!.

Fig. 4 Vanishing for d = (−1, 0)
Similarly, for every operator δ ∈ D(RA2 )(−1,0) , δ(h2 − 2)h1 ∗ s m1 t m2 = 0 for every
m on y = 2x − 2 or in σ1 . Also, δ(h2 − 2)h1 ∗ s m1 t m2 ∈ J for all other m in NA, so
D(RA2 , J )(−1,0) = s −1 · h1 (h2 , 2)!.
In fact, using a similar argument applied to any d ∈ C2 in the case of I(J )d and
d ∈ C2 \ (−σ1 ) for D(RA2 , J )d , applying an operator in D(RA2 )d to a monomial
with exponent on the half-lines y = 2x−j for 0 j < h2 (−d) will give 0; whereas,
these operators applied to a monomial with multidegrees on y = 2x + h2 (d) yields
a constant multiple of a monomial with exponent in σ2 . Hence, right-multiplying
s d1 t d2 · (h2 , h2 (−d) − 1)! by h2 + h2 (d) produces an operator that, when applied to
monomials with multidegrees on the lines y = 2x + h2 (d), becomes 0, and
I(J )d = s d1 t d2 · (h2 , −2d1 + d2 )! for all d ∈ C2, and

D(RA2 , J )d = s d1 t d2 · (h2 , −2d1 + d2 )! for all d ∈ C2 \ (−σ1 ).
We will discuss the multidegrees d ∈ C2 ∩ (−σ1 ) momentarily, when we turn to

C3.
Determining the graded piece of multidegree d for both I(J ) in C4 and
D(RA2 , J ) in C4 \ (−σ2 ) is quite similar to the arguments we used to determine
I(J )d for d in C2 and D(RA2 , J )d for d in C2 \ (σ1 ), respectively. We will
briefly describe I(J )(−1,−2) and D(RA2 , J )(−1,−2) with the aid of Fig. 5 and then
immediately describe the general case. Recall that D(RA2 )(−1,−2) = s −1 t −2 ·
(h1 , 1)!. Similar to the argument for degree d = (−1, 0) above, the monomials
corresponding to the multidegrees which lie on the two light blue lines (y = 2 and
σ2 ) in Fig. 5 are the only exponents of monomials that fail to land inside J after the
application of s −1 t −2 · (h1 , 1)!.
Fig. 5 Vanishing for d = (−1, −2)
Fig. 6 Vanishing for d = (−1, −1)
To correct this deficiency, right-multiply by (h1 − 2) for I(J )(−1,−2) and (h1 −
2)h2 for D(RA2 , J )(−1,−2) . Notice that applying the operator s −1 t −2 · (h1 , 2)! to
a monomial corresponding to d ∈ NA2 along the half-lines y = 2 or the operator
s −1 t −2 ·(h1 , 2)!h2 to a monomial corresponding to d ∈ NA2 along y = 2 or y = 2x
now yields 0, and no problems are created for the remaining monomials in J or RA2 ,
respectively. Thus,
I(J )(−1,−2) = s −1 t −2 · (h1 , 2)! and D(RA2 , J )(−1,−2) = s −1 t −2 · (h1 , 2)!h2 .
In fact,
I(J )d = s d1 t d2 · (h1 , −d2 )! for all d ∈ C4, and

D(RA2 , J )d = s d1 t d2 · (h1 , −d2 )! for all d ∈ C4 \ (−σ2 ).
We will return to D(RA2 , J )d for d ∈ C4 ∩ (−σ2 ) when we turn to C3.

Determining I(J )d for d ∈ C3 or D(RA2 , J )d for d ∈ C3 ∪ (−σ1 ) ∪ (−σ2 ) again
is akin to the arguments we gave above for D(RA2 , J )(−1,0) and D(RA2 , J )(−1,−2) .
With the aid of Fig. 6, we will briefly describe
I(J )(−1,−1) and D(RA2 , J )(−1,−1) .
This explanation can easily be extended from d = (−1, −1) to all multidegrees
d in C3 (or in C3 ∪ (−σ1 ) ∪ (−σ2 ) in the case of D(RA2 , J )). If the operator
s −1 t −1 · h1 h2 , which is the generator for D(RA2 )(−1,−1) , is applied to any of the
monomials corresponding to multidegree d ∈ NA2 along the two light blue half-
lines in Fig. 6 (the portion of y = 1 or y = 2x − 1 in C1), we obtain an integer
multiple of a monomial with exponent in the facets σ1 or σ2 , respectively, which is
not in J , and no problems are created for the remaining monomials in J (or RA2 for
D(RA2 , J )).
Hence, right-multiplying by (h1 − 1)(h2 − 1) yields a new operator s −1 t −1 ·
(h1 , 1)! (h2 , 1)! that will send to 0 all monomials with multidegrees d ∈ NA2 along
the half-lines y = 2x − 1 and y = 1. No problems are created for the remaining
monomials in RA2 and we obtain
I(J )(−1,−1) = D(RA2 , J )(−1,−1) = s −1 t −1 · (h1 , 1)! (h2 , 1)!.
In fact,
I(J )d = D(RA2 , J )d = s d1 t d2 · (h1 , −d2 ))! (h2 , −2d1 + d2 )! for all d ∈ C3,
and
D(RA2 , J )d = s d1 t d2 · (h1 , −d2 ))! (h2 , −2d1 + d2 )! for all d ∈ (−σ1 ) ∪ (−σ2 ).
Hence, the graded pieces of I(J ) and D(RA2 , J ) are as follows:

⎧
⎪
⎪ s d1 t d2 · C[θ ] if d ∈ C1 = NA2 ,
⎪
⎪
⎨s d1 t d2 · (h2 , −2d1 + d2 )! if d ∈ C2,
I(J )d =
⎪s d 1 t d 2
⎪ · (h1 , −d2 )! (h2 , −2d1 + d2 )! if d ∈ C3,
⎪
⎪
⎩ d1 d2
s t · (h1 , −d2 )! if d ∈ C4,
⎧
⎪
⎪s d1 t d2 · C[θ ] if s d1 t d2 ∈ J,
⎪
⎪
⎪
⎪
⎨s t · (h2 , −2d1 + d2 )!
⎪ if d ∈ (C2 \ (−σ1 )) ∪ (σ2 \ {0}),
d1 d2
D(RA2 , J )d = s d1 t d2 ·
⎪
⎪
⎪
⎪ (h1 , −d2 )! (h2 , −2d1 + d2 )! if d ∈ C3 ∪ (−σ1 ) ∪ (−σ2 ),
⎪
⎪
⎪
⎩s d1 t d2 · (h , −d )!
1 2 if d ∈ (C4 \ (−σ2 )) ∪ (σ1 \ {0}).
Now taking the quotient, we obtain:

⎧
⎪
⎪ 0 if d ∈
/ (Zσ1 ∪ Zσ2 ),
⎪
⎪
⎪
⎪ C[θ ]
⎪
⎪ s d1 t d2 · if d = 0,
⎪
⎪ h1 h2
⎪
⎪ C[θ ]
⎪
⎪
⎪
⎨s t
d1 d2 · if d ∈ σ1 \ {0},
I(J ) h1
= C[θ ]
D(RA2 , J ) ⎪
⎪ s d1 t d2 · if d ∈ σ2 \ {0},
d ⎪
⎪ h2
⎪
⎪
⎪
⎪ (h2 , −2d1 )!
⎪
⎪ s d1 t d2 · if d ∈ (−σ1 ),
⎪
⎪ h1 (h2 , −2d1 )!
⎪
⎪ (h1 , −d2 )!
⎪
⎩s d 1 t d 2 · if d ∈ (−σ2 ).
h2 (h1 , −d2 )!
As both J and D(RA2 ) are graded, we can similarly determine J D(RA2 ). Our
goal for the remainder of the section is to compute the graded pieces of J D(RA2 ),
in order to observe that, in this case, D(RA2 , J ) = J D(RA2 ).
To begin, note that for all s d1 t d2 ∈ Z2 ,
D(RA2 )(d1 −1,d2 −1) = s d1 −1 t d2 −1 · (h1 , −d2 )! (h2 , −2d1 + d2 )!.
For s m1 t m2 ∈ J the graded piece at the multidegree (d1 − m1 , d2 − m2 ) will be
s d1 −m1 t d2 −m2 · (h1 , −(d2 − m2 ) − 1)! (h2 , −2(d1 − m1 ) + d2 − m2 − 1)!.
Note that since m1 and m2 are both positive,
s m1 t m2 s d1 −m1 t d2 −m2 = s d1 t d2 ,
and
(h1 , −(d2 − c2 ) − 1)! (h2 , −2(d1 − c1 ) + d2 − c2 − 1)!
is contained in (h1 , −d2 )! (h2 , −2d1 + d2 )!. Thus, to determine (J D(RA2 ))d for
any d ∈ Z2 , it is enough to (left) multiply D(RA2 )(d1 −1,d2 −1) by st. Hence, from
our previous computation, it follows that
⎧
⎪
⎪s d1 t d2 · C[θ ] if s d1 t d2 ∈ J,
⎪
⎪
⎪
⎪
⎪ d1 d2
⎨s t · (h2 , −2d1 + d2 )! if d ∈ (C2 \ (−σ1 )) ∪ (σ1 \ {0}),
(J D(RA2 ))d = s d1 t d2 · (h1 , −d2 )!·
⎪
⎪
⎪
⎪ (h2 , −2d1 + d2 )! if d ∈ C3 ∪ (−σ1 ) ∪ (−σ2 ),
⎪
⎪
⎪
⎩s d1 t d2 · (h1 , −d2 )! if d ∈ (C4 \ (−σ2 )) ∪ (σ1 \ {0}).
Combining all cases together, it follows that for an arbitrary d ∈ Z2 , there is an

equality
(J D(RA2 ))d = s d1 t d2 · (h1 , h1 (−d))! (h2 , h2 (−d))! = D(RA2 , J )d .
4 Differential Operators on the Rational Normal Curve of

Degree 3
In this section, we determine some subsets of the ring of differential operators for
the ring of the rational normal curve of degree 3 determined by its interior ideal
J . We contrast these computations with the degree 2 case that was determined in
Sect. 3. Although the description follows the same reasoning as the degree 2 setting,
we ultimately get quite different behavior for the operators that end up in J D(RA3 ).
The ring of the rational normal curve of degree 3 is RA3 = C[s, st, st 2 , st 3 ], and
we will compute I(J )/D(RA3 , J ), where J = st, st 2 is a radical ideal. The facets
of A3 are
σ1 = {(x, y) ∈ N2 | x 0, y = 0} and σ2 = {(x, y) ∈ N2 | x, y 0, y = 3x},
h1 = θ2 and h2 = 3θ1 − θ2 .
Figure 7 illustrates the integer lattice, divided into four chambers that are colored as
follows:
C1 : The red multidegrees correspond to monomials in J , and the blue multidegrees
correspond to monomials in RA2 \ J ,
C2 : The yellow multidegrees are the d with h1 (d) 0 and h2 (d) < 0,
C3 : The violet multidegrees are the d with both h1 (d) < 0 and h2 (d) < 0, and
C4 : The green multidegrees are the d with h1 (d) < 0 and h2 (d) 0.
Still following the convention (h, n)! = 1 if n < 0, by Theorem 2.3, the graded
pieces of D(RA3 ) are
D(RA3 )d = s d1 t d2 · (h1 , h1 (−d) − 1)! (h2 , h2 (−d) − 1)! .
Broken down by chambers, this amounts to:

⎧
⎪
⎪s d1 t d2 · C[θ ] if d ∈ C1,
⎪
⎪
⎨s d1 t d2 · (h , −3d + d − 1)! if d ∈ C2,
2 1 2
D(RA3 )d =
⎪
⎪s t · (h1 , −d2 − 1)! (h2 , −3d1 + d2 − 1)!
d 1 d 2 if d ∈ C3,
⎪
⎪
⎩ d1 d2
s t · (h1 , −d2 − 1)! if d ∈ C4.
Determining the graded pieces of I(J ) and D(RA3 , J ) in D(RA3 ) is very

similar to our computations in Sect. 3. Here we include some visualizations for
d ∈ {(−1, 0), (−1, −3), (−1, −1), (−1, −2)} in Fig. 8 to aid our in our description
of how to obtain I(J )d and D(RA3 , J )d . Then, we will list the expressions for I(J )d
and D(RA3 , J )d by chamber, as we did in Sect. 3.
For each d in the illustrations in Fig. 8, when we apply elements of D(RA3 )d to
any of the monomials represented by lattice points along the light blue lines, we
obtain an integer multiple of a monomial whose exponent lies on a facet. As in
Sect. 3, these lines determine the linear multiples hi − hi (d) we must append to
I((d)) to obtain either I(J )d or D(RA3 , J ). Note that the light blue lines hi (x) −
hi (d) = 0 that have contain a red dot determine the hi − hi (d) we right-multiply by
to obtain I(J )d and that the light blue lines hi (x) − hi (d) = 0 that have non-empty
intersection with the entire cone determine the hi − hi (d) we right-multiply by to
obtain D(RA3 , J )d . Below we summarize the graded pieces of I(J ) and D(RA3 , J ),
as we did in Sect. 3:
Fig. 7 Chambers of D(RA3 )

Fig. 8 Vanishing at various d. (a) d = (−1, 0). (b) d = (−1, −3). (c) d = (−1, −1). (d) d =
(−1, −2)
⎧
⎪
⎪ s d1 t d2 · C[θ ] if d ∈ NA3 ,
⎪
⎪
⎨s d1 t d2 · (h1 , −d2 )! if d ∈ C4,
I(J )d =
⎪s d1 t d2
⎪ · (h2 , −3d1 + d2 )! if d ∈ C2,
⎪
⎪
⎩ d1 d2
s t · (h1 , −d2 )! (h2 , −3d1 + d2 )! if d ∈ C3,
⎧
⎪
⎪ s d1 t d2 · C[θ ] if s d1 t d2 ∈ J,
⎪
⎪
⎪
⎪
⎨s t · (h1 , −d2 )!
⎪ if d ∈ (C4 \ (−σ2 )) ∪ (σ1 \ {0}),
d1 d2
D(RA3 , J )d = s d1 t d2· (h2 , −3d1 + d2 )! if d ∈ (C2 \ −σ1 ) ∪ (σ2 \ {0}),

⎪
⎪
⎪
⎪ s t · (h1 , −d2 )!·
d d
⎪
⎪
1 2
⎪
⎩ (h2 , −3d1 + d2 )! if d ∈ C3 ∪ (−σ1 ) ∪ (−σ2 ).
(4.1)
Taking the quotient, we obtain:

⎧
⎪
⎪ 0 if d ∈
/ (Zσ1 ∪ Zσ2 ),
⎪
⎪
⎪
⎪ C[θ ]
⎪
⎪ s d1 t d2 · if d = 0,
⎪
⎪ h1 h2
⎪
⎪ C[θ ]
⎪
⎪
⎪
⎨s t
d1 d2 · if d ∈ (σ1 \ {0}),
I(J ) h1
= C[θ ]
D(RA3 , J ) ⎪
⎪ s d1 t d2 · if d ∈ (σ2 \ {0}),
d ⎪
⎪ h2
⎪
⎪
⎪
⎪ (h2 , −3d1 )!
⎪
⎪ s d1 t d2 · if d ∈ (−σ1 \ {0}),
⎪
⎪ h1 (h2 , −3d1 )!
⎪
⎪ (h1 , −d2 )!
⎪
⎩s d1 t d2 · if d ∈ (−σ2 \ {0}).
h2 (h1 , −d2 )!
In the remainder of the section, we compute the graded pieces of J D(RA3 ) to

show that they are not equal to the graded pieces of D(RA3 , J ). This means that
Proposition 1.1 does not hold for this ring. To begin the computation, if s m1 t m2 ∈ J ,
then
D(RA3 )(d1 −m1 ,d2 −m2 )

=s d1 −m1 t d2 −m2 · (h1 , −(d2 −m2 ) − 1)! (h2 , −3(d1 −m1 )+d2 − m2 − 1)!.
Further, since m1 , m2 1, s m1 t m2 s d1 −m1 t d2 −m2 = s d1 t d2 , and there is a

containment
(h1 , −(d2 − m2 ) − 1)! (h2 , −3(d1 − m1 ) + d2 − m2 − 1)!

⊆ (h1 , −d2 )! (h2 , −3d1 + d2 + 1)!, (h1 , −d2 + 1)! (h2 , −3d1 + d2 )!.
Thus, to determine the graded piece of (J D(RA3 ))d for any d ∈ Z2 , it is enough to
consider D(RA3 ) in multidegrees (d1 − 1, d2 − 1) and (d1 − 1, d2 − 2), where
D(RA3 )(d1 −1,d2 −1) = s d1 −1 t d2 −1 · (h1 , −d2 )! (h2 , −3d1 + d2 + 1)! and
D(RA3 )(d1 −1,d2 −2) = s d1 −1 t d2 −2 · (h1 , −d2 + 1)! (h2 , −3d1 + d2 )!.
We will now break down this computation by chambers from Fig. 7, leaving off
some half-lines along the way and addressing them as special cases later on. For C1,
Fig. 9a helps to visualize that if s d1 t d2 ∈ J , then at least one of the two multidegrees
(d1 − 1, d2 − 1) or (d1 − 1, d2 − 2) lives in NA. Since D(RA3 )m = s m1 t m2 · C[θ ]
for all m ∈ NA, (J D(RA3 ))d = s d1 t d2 · C[θ ] for s d1 t d2 ∈ J . We will consider the
blue multidegrees in C1 with their neighbors in C2 and C4.
Now consider the graded pieces of (J D(RA3 ))d for d in σ2 or C2, excluding the
two half-lines in that chamber given by −σ1 and y = 1. Since we excluded the
Fig. 9 Visualizing (J D(RA3 ))d . (a) (J D(RA3 ))d for td ∈ J . (b) (J D(RA3 ))d for d ∈ C2. (c)
(J D(RA3 ))d for d ∈ C4. (d) (J D(RA3 ))d for d ∈ C3. (e) (J D(RA3 ))d for d on exceptional lines
in C2 and C4
x-axis and y = 1, both (d1 − 1, d2 − 1) and (d1 − 1, d2 − 2) lie in C2 for all d under
consideration, see Fig. 9b. Since
D(RA3 )(d1 −1,d2 −1) = s d1 −1 t d2 −1 · (h2 , −3d1 + d2 + 1)! and

D(RA3 )(d1 −1,d2 −2) = s d1 −1 t d2 −2 · (h2 , −3d1 + d2 )!,
and there is a containment (h2 , −3d1 + d2 + 1)! ⊆ (h2 , −3d1 + d2 )!, for
such d,
(J D(RA3 ))d = s d1 t d2 · (h2 , −3d1 + d2 )! .
Now consider (J D(RA3 ))d for d in σ1 or in C4, excluding those multidegrees

that lie on −σ2 or the half-line y = 3x − 1. Since we excluded the multidegrees on
−σ2 and y = 3x − 1, both (d1 − 1, d2 − 1) and (d1 − 1, d2 − 2) also lie in C4, as
seen in Fig. 9c. Since
D(RA3 )(d1 −1,d2 −1) = s d1 −1 t d2 −1 · (h1 , −d2 )! and

D(RA3 )(d1 −1,d2 −2) = s d1 −1 t d2 −2 · (h1 , −d2 + 1)!,
and there is a containment (h1 , −d2 + 1)! ⊆ (h1 , −d2 )!, it follows that for
such d,
(J D(RA3 ))d = s d1 t d2 · (h1 , −d2 )!.
For (J D(RA3 ))d for d in C3, −σ1 , or −σ2 , see Fig. 9d,
D(RA3 )(d1 −1,d2 −1) = s d1 −1 t d2 −1 · (h1 , −d2 )! (h2 , −3d1 + d2 + 1)!,

D(RA3 )(d1 −1,d2 −2) = s d1 −1 t d2 −2 · (h1 , −d2 + 1)! (h2 , −3d1 + d2 )!.
Since the ideals
(h1 , −d2 )! (h2 , −3d1 + d2 + 1)! and (h1 , −d2 + 1)! (h2 , −3d1 + d2 )!
are incomparable, it follows that for such d,
(J D(RA3 ))d
=s d1 t d2 · (h1 , −d2 )! (h2 , −3d1 +d2 +1)!, (h1 , −d2 +1)! (h2 , −3d1 + d2 )!.
The multidegrees d that we have not yet considered for (J D(RA3 ))d are those in
C2 along y = 1 and in C4 along y = 3x − 1. For such d, one of (d1 − 1, d2 − 1) or
(d1 − 1, d2 − 2) belongs to C3, see Fig. 9e. First, for d in C2 along the line y = 1,
(d1 − 1, −1) ∈ C3, and
D(RA3 )(d1 −1,−1) = s d1 −1 t −1 · h1 (h2 , −3d1 + 1)!.

Combining the fact that D(RA3 )(d1 −1,0) = s d1 −1 · (h2 , −3d1 + 2)!, we compute
that
(J D(RA3 ))d = s d1 t · (h2 , −3d1 + 2)!, h1 (h2 , −3d1 + 1)!.
Second, for d ∈ C4 along the line y = 3x − 1,
D(RA3 )(d1 −1,3d1 −2) = s d1 −1 t 3d1 −2 · (h1 , −3d1 + 1)!h2 .
Connecting with the fact that
D(RA3 )(d1 −1,3d1 −3) = s d1 −1 t 3d1 −3 · (h1 , −3d1 + 2)!,
we determine that
(J D(RA3 ))d = s d1 t 3d1 −1 · (h1 , −3d1 + 1)!h2 , (h1 , −3d1 + 2)!.
Having now computed J D(RA3 ) in all multidegrees, we have that

⎧
⎪
⎪ s d1 t d2 · C[θ ] if s d1 t d2 ∈ J,
⎪
⎪
⎪
⎪
⎨s t · (h2 , −3d1 + d2 )!
⎪ if d ∈ R2,
d1 d2
(J D(RA3 ))d = s d1 t d2· (h1 , −d2 )! (h2 , −3d1 + d2 + 1)!,

⎪
⎪
⎪
⎪ (h1 , −d2 + 1)! (h2 , −3d1 + d2 )! if d ∈ R3,
⎪
⎪
⎪
⎩s d1 t d2 · (h , −d )!
1 2 if d ∈ R4,
where
R2 = (C2 \ (σ1 ∪ {y = 1})) ∪ (σ2 \ {0}),

R3 = C3 ∪ (−σ1 ∪ −σ2 ) ∪ (NA3 ), and
R4 = (C4 \ (σ2 ∪ {y = 3x − 1})) ∪ (σ1 \ {0}).
Now compare this with D(RA3 , J ) from (4.1); it becomes clear that (J D(RA3 ))d =
D(RA3 , J )d whenever d belongs to any of the following: −σ1 , −σ2 , C3, C2 along
the half-line {y = 1}, or C4 along the half-line {y = 3x − 1}. In particular,
I(J )/D(RA3 , J ) = I(J )/J D(RA3 ), in contrast to Proposition 1.1.
5 Differential Operators on a Rational Normal Curve
The techniques used in Sects. 3 and 4 can also! be applied to compute I(J )d ,
D(RAn , J )d , (J D(RAn ))d and I(J )/D(RAn , J ) d , where the radical ideal J =
st, st 2 , . . . , st n−1 is again the intersection of the primes defined by the facets of
An . We denote the facets of An by
σ1 = {(x, y) ∈ N2 | x 0, y = 0} and σ2 = {(x, y) ∈ N2 | x, y 0, y = nx},
h1 = θ2 and h2 = nθ1 − θ2 .
Again, we divide Z2 into chambers, analogous to those used in Sects. 3 and 4, so

C1 = {d ∈ Z2 | d ∈ NA},
C2 = {d ∈ Z2 | h1 (d) 0, h2 (d) < 0},
C3 = {d ∈ Z2 | h1 (d) < 0, h2 (d) < 0}, and
C4 = {d ∈ Z2 | h1 (d) < 0, h2 (d) 0}.
These computations yield the following formulas:
⎧
⎪
⎪s d1 t d2 · C[θ ] if d ∈ NAn ,
⎪
⎪
⎨s d1 t d2 · (h1 , −d2 )! if d ∈ C4,
I(J )d =
⎪
⎪s d1 t d2 · (h2 , −nd1 + d2 )! if d ∈ C2,
⎪
⎪
⎩ d1 d2
s t · (h1 , −d2 )! (h2 , −nd1 + d2 )! if d ∈ C3,
⎧
⎪
⎪ s d1 t d2 · C[θ ] if s d1 t d2 ∈ J,
⎪
⎪
⎨s d1 t d2 · (h1 , −d2 )! if d ∈ C4 ,
D(RAn , J )d =
⎪
⎪ s d1 t d2 · (h2 , −nd1 + d2 )! if d ∈ C2 ,
⎪
⎪
⎩ d1 d2
s t · (h1 , −d2 )! (h2 , −nd1 + d2 )! if d ∈ C3 ,
where
C2 = (C2 \ (−σ1 )) ∪ (σ2 \ {0}),

C3 = C3 ∪ (−σ1 ∪ −σ2 ),
C4 = (C4 \ (−σ2 )) ∪ (σ1 \ {0});
⎧
⎪
⎪ s d1 t d2 · C[θ ] if s d1 t d2 ∈ J,
⎪
⎪
⎪
⎪
⎨s t · (h1 , −d2 )!
⎪ d1 d2 if d ∈ C4 ,
(J D(RAn ))d = s d1 t d2 · (h2 , −nx + y)! if d ∈ C2 ,
⎪
⎪
⎪
⎪ s d1 t d2 · (h1 , −d2 + j )! ·
⎪
⎪
⎪
⎩ (h , −nd + d + n − 2 − j )!n−2
2 1 2 j =0 if d in C3 ,
where
⎛ ⎞
0
n−2
C2 = C2 \ ⎝ {y = j }⎠ ,
j =0
⎡ ⎛ ⎞⎤ ⎡ ⎛ ⎞⎤
0
n−2 0
n−2
C3 = C3 ∪ ⎣C2 ∩ ⎝ {y = j }⎠⎦ ∪ ⎣C4 ∩ ⎝ {y = nx − j }⎠⎦ ,
j =0 j =0
⎛ ⎞
0
n−2
C4 = C4 \ ⎝

{y = nx − j }⎠ .
j =0
Recall from Sect. 4 that J D(RA3 ) = D(RA3 , J ); we see that this is true for all
rings of rational normal curves RAn = C[s, st, . . . , st n ] with n 3. Specifically,
comparing D(RAn , J )d and (J D(RAn ))d for various d ∈ Z2 , the graded pieces
differ whenever d belongs to σ1 , σ2 , C3, the half-lines in C2 inside
{(x, y) ∈ R2 | x < 0, y = i, 0 i n − 2},
or the half-lines in C4 inside
{(x, y) ∈ R2 | x < 0, y = nx − i, 0 i n − 2}.
Figure 10 has these multidegrees d in blue for the case n = 7.

As with the rational normal cones in degrees 2 and 3 from Sects. 3 and 4, for
d ∈ Z2 ,
Fig. 10 (J D(RA7 ))d and D(RA7 , J )d differ at blue d

⎧
⎪
⎪ 0 if d ∈ Z2 \ (Zσ1 ∪ Zσ2 ),
⎪
⎪
⎪
⎪ C[θ ]
⎪
⎪ if d = 0,
I(J ) ⎨ h1 h2
= (h2 , −nd1 )! (5.1)
D(RAn , J ) d ⎪⎪ s d1 · if d ∈ Zσ1 \ {0},
⎪
⎪ h1 (h2 , −nd1 )!
⎪
⎪
⎪
⎪ (h1 , −d2 )!
⎩s t ·
d 1 d 2 if d ∈ Zσ2 \ {0}.
h2 (h1 , −d2 )!
Example 2.5 considered the ordinary double point C[x, y]/xy, which is
isomorphic to RA!n /J for all n 1. Comparing D(C[x, y]/xy))d from (2.2) and
I(J )/D(RAn , J ) d from (5.1), we see that there is an isomorphism between the
graded components,
! viewed as C-vector spaces, given by ϕ : D (C[x, y]/xy) →
D RAn /J with, for m ∈ Z,

C[θx , θy ] C[θ ]
ϕ = ,
θx θy h1 h2

(θx , −m)! (h2 , −nm)!
ϕ x m
= sm · , and
(θx , −m)!θy (h2 , −nm)!h1
!
θy , −m ! (h1 , −nm)!
ϕ y ·
m ! = s m t nm · .
θy , −m !θx (h1 , −nm)!h2
However, this isomorphism is not an isomorphism of rings. Again as in Exam-

ple 3.1, the degree of the polynomial generator in θ1 and θ2 in multidegree d is n
times the degree of the polynomial generator of in θx and θy . In fact, noting that
along Zσ2 , d2 = nd1 , we have
⎧
⎪
⎪ 0 if d ∈ Z2 \ (Zσ1 ∪ Zσ2 ),
⎪
⎪
⎪
⎪ C[θ ]
⎪
⎪ , - if d = 0,
⎪
⎪
⎪
⎪
h 1 h2
⎪
⎪
⎪
⎪ n n , -
⎪
⎪
⎪
⎪
h2
, −d1 !
! ⎨
n
D RAn /J d = s d1 · , - if d ∈ Zσ1 \ {0}, (5.2)
⎪
⎪ h1 h2
⎪
⎪ , −d1 !
⎪
⎪ n ,n
⎪
⎪ -
⎪
⎪ h1
⎪
⎪ , −d1 !
⎪
⎪ n
⎪
⎪ s d1 t d2 · , - if d ∈ Zσ2 \ {0}.
⎪
⎪
⎪
⎪ h2 h1
⎩ , −d 1 !
n n
Now comparing (2.2) and (5.2), we see there is an isomorphism of the ring of
differential operators between the two rings given by, for each m ∈ Z,
, -
−m h2 h2
ψ((θx , −m)! + (θx , −m)!θy ) = s · , −m ! + , −m !h1 ,
n n
, -
! ! −m −nm h1 h1
ψ( θy , −m ! + θy , −m !θx ) = s t · , −m ! + , −m !h2 ,
n n
ψ(x m ) = s m , and ψ(y m ) = s m t nm .
6 Higher Dimensional Examples
Thus far, we have considered differential operators determined by radical ideals in

two-dimensional semigroup rings. In this section, we turn to looking at some of the
differential operators in the three-dimensional semigroup ring given by
⎡ ⎤
1111
A = ⎣0 1 0 1 ⎦
0011
and describe some subsets of differential operators of RA given by two different

radical ideals J . For both choices of J , we will compute the graded pieces I(J )d ,
D(RA , J )d and (I(J )/D(RA , J ))d . For the given matrix A,
RA = C[NA] = C[t1 , t1 t2 , t1 t3 , t1 t2 t3 ].
The facets of the cone R0 A are
σ1 = N{e1 , e1 + e2 }, σ3 = N{e1 + e3 , e1 + e2 + e3 },
σ2 = N{e1 , e1 + e3 }, σ4 = N{e1 + e2 , e1 + e2 + e3 },
and the corresponding primitive integral support functions are
h1 = h1 (θ ) = θ3 , h3 = h3 (θ ) = θ1 −θ3 , h2 = h2 (θ ) = θ2 , h4 = h4 (θ ) = θ1 −θ2 .
The prime ideals associated to the facets are
Pσ1 = t1 t3 , t1 t2 t3 , Pσ2 = t1 t2 , t1 t2 t3 , Pσ3 = t1 , t1 t2 , Pσ4 = t1 , t1 t3 ,
and the prime ideals associated to the rays (or 1-dimensional faces) of the cone are
Pσ1 ∩σ2 = t1 t2 , t1 t3 , t1 t2 t3 ,

Pσ2 ∩σ3 = t1 , t1 t2 , t1 t2 t3 ,
Pσ3 ∩σ4 = t1 , t1 t2 , t1 t3 ,

Pσ1 ∩σ4 = t1 , t1 t3 , t1 t2 t3 .
In contrast with the two dimensional cases considered in Sects. 3, 4, and 5, the
increased dimension of the semigroup NA allows for many more choices for radical
monomial ideal J in RA . We will compute I(J )/D(RA , J ) for two such choices,
J = Pσ1 ∩ Pσ2 ∩ Pσ3 ∩ Pσ4 = t12 t2 t3 and

J = Pσ1 ∩ Pσ2 ∩ Pσ3 ∩σ4 = t12 t2 t3 , t12 t22 t3 , t12 t2 t32 .
For both choices of J , to compute the graded pieces of I(J ), we will divide up Z3
into 14 chambers, depending on various combinations of signs of hi (d). Table 1
describes the 14 chambers in a list, while Fig. 11 illustrates them, with chamber C1
(given by NA) in the top right with some of the lattice points of NA shown. Note
the x-axis is the vertical axis in this picture.
We must modify the chambers slightly to compute the graded pieces of
D(RA , J ), as we now consider which differential operators, when applied to an
element in RA , yield an element in J . Although all of the operators in the graded
pieces of I(J ) lying on the facets send J into J , it is not necessarily the case that
these operators applied to an element of RA will output an element in J . So we will
switch the lattice points on the facets that correspond to monomials that do not lie
in J to lie in the adjacent chambers by interchanging with >. We will describe
these new regions within the examples.
Example 6.1 First consider the ideal
J = Pσ1 ∩ Pσ2 ∩ Pσ3 ∩ Pσ4 = t12 t2 t3
in RA = C[t1 , t1 t2 , t1 t3 , t1 t2 t3 ].
In Fig. 12, the multidegrees in the gray cone whose vertex lies at (2, 1, 1)
correspond to the monomials that lie in J , and the monomials in the outer cone
lie in RA \ J . Note also that the view of the cone that we see in Fig. 12 is from the
side of the xz-plane.
Since none of the points on the facets of the cone have monomials that lie in J ,
the 14 regions that we consider in determining D(RA , J ) are listed in Table 2.
If d is in C1 (which are the lattice points corresponding to points in the
semigroup) when considering the idealizer or R1 (which are the lattice points
corresponding to the monomials in J ) when considering D(RA , J ), then D(RA )d is
generated by multiplication by td . Multiplying any element of J by td remains in J
making the graded pieces of I(J )d = td · C[θ ] and D(RA , J )d = td · C[θ ].
Since the multidegrees in the chambers and regions corresponding to C2−C5 and
R2 − R5 will all need operators adjusted only for monomials in J whose exponents
are parallel to a single facet in A, we will only describe the process to determine the
78
Table 1 The chambers used to compute I(J ) in Examples 6.1 and 6.2
Chamber Halfspace inequalities Lattice point inequalities
C1 {d ∈ Z3 | h1 (d), h2 (d), h3 (d), h4 (d) 0} NA
C2 {d ∈ Z3 | h1 (d), h3 (d) 0, h2 (d) > 0, h4 (d) < 0} {d ∈ Z3 | d2 > d1 d3 0}
C3 {d ∈ Z3 | h2 (d), h4 (d) 0, h1 (d) > 0, h3 (d) < 0} {d ∈ Z3 | d3 > d1 d2 0}
C4 {d ∈ Z3 | h1 (d), h3 (d) 0, h4 (d) > 0, h2 (d) < 0} {d ∈ Z3 | d1 d3 0 > d2 }
C5 {d ∈ Z3 | h2 (d), h4 (d) 0, h3 (d) > 0, h1 (d) < 0} {d ∈ Z3 | d1 d2 0 > d3 }
C6 {d ∈ Z3 | h1 (d), h2 (d) 0, h3 (d), h4 (d) < 0} {d ∈ Z3 | d3 0, d2 0, d2 > d1 , d3 > d1 }
C7 {d ∈ Z3 | h2 (d), h3 (d) 0, h1 (d), h4 (d) < 0} {d ∈ Z3 | d2 0 > d3 , d2 > d1 d3 }
C8 {d ∈ Z3 | h1 (d), h4 (d) 0, h2 (d), h3 (d) < 0} {d ∈ Z3 | d3 0 > d2 , d3 > d1 d2 }
C9 {d ∈ Z3 | h3 (d), h4 (d) 0, h1 (d), h2 (d) < 0} {d ∈ Z3 | 0 > d2 , 0 > d3 , d1 d2 , d1 d3 }
C10 {d ∈ Z3 | h1 (d), h3 (d), h4 (d) < 0, h2 (d) 0} {d ∈ Z3 | d2 0 > d3 > d1 }
C11 {d ∈ Z3 | h2 (d), h3 (d), h4 (d) < 0, h1 (d) 0} {d ∈ Z3 | d3 0 > d2 > d1 }
C12 {d ∈ Z3 | h1 (d), h2 (d), h3 (d) < 0, h4 (d) 0} {d ∈ Z3 | 0 > d3 > d1 d2 }
C13 {d ∈ Z3 | h1 (d), h2 (d), h4 (d) < 0, h3 (d) 0} {d ∈ Z3 | 0 > d2 > d1 d3 }
C14 {d ∈ Z3 | h1 (d), h2 (d), h3 (d), h4 (d) < 0} −Int(NA)
C. Berkesch et al.
Fig. 11 Chambers for C[t1 , t1 t2 , t1 t3 , t1 t2 t3 ]
Fig. 12 Cone of RA with ideal J = t12 t2 t3
graded pieces of I(J ) and D(RA , J ) for chamber C2 and region R2, respectively,
as precisely the same type of argument holds for the other chambers and regions. If
d is in C2 in the case of the idealizer or in R2 in the case of D(RA , J ), consider the
lattice points corresponding to the monomials in J on the plane x − y + h4 (d) = 0.
When an element in I((d)) is applied to such a monomial in J , the result is a
constant times a monomial with exponent on the facet σ4 , which is not in J . Hence
we need to multiply I((d)) by h4 + h4 (d), so for d in C2,
I(J )d = td · (h4 , h4 (−d))!
and for d in R2,
D(RA , J )d = td · (h4 , h4 (−d))!.
Figure 13a illustrates the plane in C2 that determines the linear form that we
multiply by I((d)) to obtain I(J )d .
Since the multidegrees in the chambers and regions corresponding to C6−C9 and
R6 − R6 will all need operators adjusted only for monomials in J whose exponents
80
Table 2 Regions to compute D(RA , J ) in Example 6.1

Region Halfspace inequalities Lattice point inequalities
R1 {d ∈ Z3 | h1 (d), h2 (d), h3 (d), h4 (d) > 0} Int(NA)
R2 {d ∈ Z3 | h1 (d), h2 (d), h3 (d) > 0, h4 (d) 0} {d ∈ Z3 | d2 d1 > d3 > 0}
R3 {d ∈ Z3 | h1 (d), h2 (d), h4 (d) > 0, h3 (d) 0} {d ∈ Z3 | d3 d1 > d2 > 0}
R4 {d ∈ Z3 | h1 (d), h3 (d), h4 (d) > 0, h2 (d) 0} {d ∈ Z3 | d1 > d3 > 0 d2 }
R5 {d ∈ Z3 | h2 (d), h3 (d), h4 (d) > 0, h1 (d) 0} {d ∈ Z3 | d1 > d2 > 0 d3 }
R6 {d ∈ Z3 | h1 (d), h2 (d) > 0, h3 (d), h4 (d) 0} {d ∈ Z3 | d2 > 0, d3 > 0, d2 d1 , d3 d1 }
R7 {d ∈ Z3 | h2 (d), h3 (d) > 0, h1 (d), h4 (d) 0} {d ∈ Z3 | d2 > 0 d3 , d2 d1 > d3 }
R8 {d ∈ Z3 | h1 (d), h4 (d) > 0, h2 (d), h3 (d) 0} {d ∈ Z3 | d3 > 0 d2 , d3 d1 > d2 }
R9 {d ∈ Z3 | h3 (d), h4 (d) > 0, h1 (d), h2 (d) 0} {d ∈ Z3 | d1 > d2 , d1 > d3 , 0 d2 , 0 d3 }
R10 {d ∈ Z3 | h1 (d), h3 (d) 0, h4 (d) < 0, h2 (d) > 0} {d ∈ Z3 | d2 > 0 d3 d1 }
R11 {d ∈ Z3 | h2 (d), h4 (d) 0, h3 (d) < 0, h1 (d) > 0} {d ∈ Z3 | d3 > 0 d2 d1 }
R12 {d ∈ Z3 | h1 (d), h3 (d) 0, h2 (d) < 0, h4 (d) > 0} {d ∈ Z3 | 0 d3 d1 > d2 }
R13 {d ∈ Z3 | h2 (d), h4 (d) 0, h1 (d) < 0, h3 (d) > 0} {d ∈ Z3 | 0 d2 d1 > d3 }
R14 {d ∈ Z3 | h1 (d), h2 (d), h3 (d), h4 (d) 0} −NA
C. Berkesch et al.
Fig. 13 Vanishing planes. (a) d = (1, 2, 1) ∈ C2 and {x − y + h4 (d) = 0}. (b) d = (−1, 0, 0) ∈
C6, {x − y + h4 (d) = 0}, and {x − z + h4 (d) = 0}. (c) d = (−2, 0, −1) ∈ C10, {z + h1 (d) = 0},
{x − z + h3 (d) = 0}, and {x − y + h4 (d) = 0}
are parallel to two facets of A, we will only describe the process to determine the
graded pieces of I(J ) and D(RA , J ) for chamber C6 and region R6, respectively, as
precisely the same type of argument holds for the other chambers and regions. If d is
in C6 in the case of the idealizer or R6 in the case of D(RA , J ), consider the lattice
points corresponding to the monomials in J on the planes x − y + h4 (d) = 0 and
x − z + h3 (d) = 0. When an element in I((d)) is applied to such monomial in J ,
the result is a constant times a monomial with exponent on x − y = 0 or x − z = 0,
which is not in J . Hence we need to multiply I((d)) by (h3 + h3 (d))(h4 + h4 (d)),
so for d in C6,
I(J )d = td · (h3 , h3 (−d))! (h4 , h4 (−d))!,
and for d in R6,
D(RA , J )d = td · (h3 , h3 (−d))! (h4 , h4 (−d))!.
Figure 13b illustrates the two planes in C6 that determine the linear forms that we
multiply I((d)) by to obtain I(J )d .
Since the multidegrees in the chambers and regions corresponding to C10 − C13
and R10 − R13 will all need operators adjusted only for monomials in J whose
exponents are parallel to three facets of A, we will only describe the process to
determine the graded piece of I(J ) and D(RA , J ) for chamber C10 and region R10,
respectively, as precisely the same type of argument holds for the other chambers
and regions. If d is in C10 in the case of the idealizer or R10 in the case of D(RA , J ),
consider the lattice points corresponding to the monomials in J on the planes z +
h1 (d) = 0, x − z + h3 (d) = 0 and x − y + h4 (d) = 0. When an element of I((d))
is applied to such a monomial in J , the result is a constant times a monomial whose
exponent is in σ1 , σ3 , or σ4 , which is not in J . Hence we need to multiply I((d))
by (h1 + h1 (d))(h3 + h3 (d))(h4 + h4 (d)), so for d in C10,
I(J )d = td · (h1 , h1 (−d))! (h3 , h3 (−d))! (h4 , h4 (−d))!,
and for d in R10,
D(RA , J )d = td · (h1 , h1 (−d))! (h3 , h3 (−d))! (h4 , h4 (−d))!.
Figure 13c illustrates the three planes in C10 that determine the linear forms that we
multiply by I((d)) to obtain I(J )d .
If d is in C14 in the case of the idealizer or R14 in the case of D(RA , J ), consider
the lattice points corresponding to the monomials in J on the planes z + h1 (d) = 0,
y+h2 (d) = 0, x−z+h3 (d) = 0 and x−y+h4 (d) = 0. When an element of I((d))
is applied to such a monomial in J , the result is a constant times a monomial with
exponent on one of the facets σi of A, which is not in J . Hence we need to multiply
I((d)) by (h1 + h1 (d))(h2 + h2 (d))(h3 + h3 (d))(h4 + h4 (d)), so for d in C14,
I(J )d = td · (h1 , h1 (−d))! (h2 , h2 (−d))! (h3 , h3 (−d))! (h4 , h4 (−d))!,
and for d in R14,
D(RA , J )d = td · (h1 , h1 (−d))! (h2 , h2 (−d))! (h3 , h3 (−d))! (h4 , h4 (−d))!.
Combining the information from all 14 chambers C1-C14 and regions R1-R14,
the general formula for the graded piece of the idealizer of J at d ∈ Z3 is
* +

I(J )d = t · d
(hi , hi (−d))! ,
hi (d)<0
and the general formula for the graded piece of the D(RA , J ) at d ∈ Z3 is
* +

D(RA , J )d = t · d
(hi , hi (−d))! .
hi (d)0
For any d ∈ Z3 ,
* +
3
(hi , hi (−d))!
I(J ) hi (d)<0
=t ·*d +.
D(RA , J ) d 3
(hi , hi (−d))!
hi (d)0
Example 6.2 Now consider the ideal
J = Pσ1 ∩ Pσ2 ∩ Pσ3 ∩σ4 = t12 t2 t3 , t12 t22 t3 , t12 t2 t32

Fig. 14 Cone of RA with J = Pσ1 ∩ Pσ2 ∩ Pσ3 ∩σ4
in RA = C[t1 , t1 t2 , t1 t3 , t1 t2 t3 ]. In Fig. 14, we give two views of the cone with the
lattice points in J shaded in grey. Note that multidegrees of the monomials in J lie
in the interior of the cone and on the interiors of the two faces σ3 and σ4 , so this is
the grey portion of the figure.
Since the only points on the facets of the cone that have monomials that lie in
J lie in the interiors of σ3 and σ4 , the 14 regions that we consider in determining
D(RA , J ) appear in Table 3.
If d is in C1 = NA, when considering the idealizer or R1, which are the
exponents of the monomials in J , when considering D(RA , J ), then D(RA )d is
generated by td . The product of any element of J and td is also an element of J ,
which means that the graded pieces are I(J )d = td ·C[θ ] and D(RA , J )d = td ·C[θ ].
Since the multidegrees in C2−C5 and R2−R5 potentially affect monomials in J
whose exponents are parallel to a single facet of A, we will only describe the process
to determine the graded pieces of I(J ) and D(RA , J ) for chamber C2 and region R2,
and regions. The reader may refer to Fig. 13a to help visualize the argument below.
exponents of monomials in J on the plane x − y + h4 (d) = 0. When an element of
I((d)) is applied to such a monomial in J , the result is a constant times a monomial
on the interior of the facet σ4 , which is in J . Hence, for d in C2,
I(J )d = td · (h4 , h4 (−d) − 1)!,
and for d in R2,
D(RA , J )d = td · (h4 , h4 (−d) − 1)!.

and R6 − R9 can will all need operators adjusted only for monomials in J whose
exponents are parallel to two facets of A, we will only describe the process to
and regions. The reader may refer to Fig. 13b to help visualize the argument below.
Table 3 Regions for D(RA , J ) in Example 6.2
84
Region Halfspace inequalities Lattice point inequalities
R1 {d ∈ Z3 | h13(d), h2 (d), h3 (d) > 0, h4 (d) 0} ∪

{d ∈ Z | h1 (d), h2 (d), h4 (d) > 0, h3 (d) 0} {d ∈ Z3 | xd ∈ J }
R2 {d ∈ Z3 | h1 (d), h2 (d) > 0, h3 (d) 0, h4 (d) < 0} {d ∈ Z3 | d2 > d1 d3 > 0}

R3 {d ∈ Z3 | h1 (d), h2 (d) > 0, h4 (d) 0, h3 (d) < 0} {d ∈ Z3 | d3 > d1 d2 > 0}
R4 {d ∈ Z3 | h13(d), h3 (d), h4 (d) > 0, h2 (d) 0} ∪

{d ∈ Z | h3 (d) = 0, h1 (d), h4 (d) > 0, h2 (d) < 0}
{d ∈ Z3 | d13 > d3 > 0 d2 } ∪
{d ∈ Z | d1 = d3 > 0 > d2 }
R5 {d ∈ Z3 | h23(d), h3 (d), h4 (d) > 0, h1 (d) 0} ∪

{d ∈ Z | h4 (d) = 0, h2 (d), h3 (d) > 0, h1 (d) < 0}
{d ∈ Z3 | d13 d2 > 0 d3 } ∪
{d ∈ C | d1 = d2 > 0 > d3 }
R6 {d ∈ Z3 | h13(d), h2 (d) > 0 and h3 (d), h4 (d) < 0} ∪

{d ∈ Z | h1 (d) > 0, h2 (d) > 0, h3 (d) = h4 (d) = 0}
{d ∈ Z3 | d3 > 0, d2 > 0, d32 > d1 , d3 > d1 } ∪
{d ∈ Z | d1 = d2 = d3 > 0}
R7 {d ∈ Z3 | h2 (d), h33(d) > 0, h1 (d) 0, h4 (d) < 0} ∪

{d ∈ Z | h2 (d) > 0, h3 (d) = 0, h1 (d) < 0}
{d ∈ Z3 | d2 > 0 d33, d2 > d1 > d3 } ∪
{d ∈ Z | d2 > 0 > d1 = d3 }
R8 {d ∈ Z3 | h1 (d), h43(d) > 0, h2 (d) 0, h3 (d) < 0} ∪

{d ∈ Z | h1 (d) > 0, h4 (d) = 0, h2 (d) < 0}
{d ∈ Z3 | d3 > 0 d23, d3 > d1 > d2 } ∪
{d ∈ Z | d3 > 0 > d1 = d2 }
{d ∈ Z3 |3h3 (d), h4 (d) > 0, h1 (d), h2 (d) 0} ∪ {d ∈ Z3 | d13 > d2 , d3 , 0 d2 , d3 } ∪

R9 {d ∈ Z | h13(d), h2 (d) < 0, h4 (d) > 0, h3 (d) = 0} ∪ {d ∈ Z | 0 3> d1 = d2 , 0 > d1 > d3 } ∪
{d ∈ Z | h1 (d), h2 (d < 0, h3 (d) > 0, h4 (d) = 0} {d ∈ Z | 0 > d1 = d3 , 0 > d1 > d2 }
R10 {d ∈ Z3 | h1 (d) 0, h3 (d), h4 (d) < 0, h2 (d) > 0} {d ∈ Z3 | d2 > 0 d3 > d1 }
R11 {d ∈ Z3 | h2 (d) 0, h4 (d), h3 (d) < 0, h1 (d) > 0} {d ∈ Z3 | d3 > 0 d2 > d1 }
R12 {d ∈ Z3 | h1 (d), h3 (d) < 0, h2 (d) 0, h4 (d) 0} {d ∈ Z3 | 0 d2 > d1 d3 }
R13 {d ∈ Z3 | h2 (d), h4 (d) < 0, h1 (d) 0, h3 (d) 0} {d ∈ Z3 | 0 d3 > d1 d2 }
R13 {d ∈ Z3 | h1 (d), h2 (d), h4 (d) 0, h3 (d) > 0} {d ∈ Z3 | 0 d2 d1 > d3 }
R14 {d ∈ Z3 | h13(d), h2 (d)h3 (d) 0, h4 (d) < 0} ∪

{d ∈ Z | h1 (d), h2 (d)h4 (d) 0, h3 (d) < 0}
{d ∈ Z3 | d13 < d2 0, d1 d3 0} ∪
{d ∈ Z | d1 d2 0, d1 < d3 0}
C. Berkesch et al.
If d is in C6 in the case of the idealizer or R6 in the case of D(RA , J ), consider the

multidegrees of monomials in J on the planes x−z+h3 (d) = 0 or x−y+h4 (d) = 0.
When an element of I((d)) is applied to such a monomial in J , the result is a
constant times a monomial on σ3 or σ4 . This is not in J only if this monomial’s
exponent is in the intersection of the two planes. Hence, we need to multiply I((d))
by either (h3 + h3 (d)) or (h4 + h4 (d)), so that for d in C6,
I(J )d = td · (h4 , h4 (−d) − 1)! (h3 , h3 (−d) − 1)! · (h3 + h3 (d)), (h4 + h4 (d)),
and for d in R6,
D(RA , J )d = td ·
(h4 , h4 (−d) − 1)! (h3 , h3 (−d) − 1)! · (h3 + h3 (d)), (h4 + h4 (d)).
and R10 − R13 can will all need operators adjusted only for monomials in J whose
exponents are parallel to a three facets of A, we will only describe the process to
and regions. The reader may refer to Fig. 13c to help visualize the argument below. If
d is in C10 in the case of the idealizer or R10 in the case of D(RA , J ), consider the
multidegrees of monomials in J on the planes z+h1 (d) = 0, x −z+h3 (d) = 0, and
x − y + h4 (d) = 0. When an element in I((d)) is applied to such a monomial in J ,
the result is a constant times a monomial on σ1 , which is not in J , or a constant times
a monomial in σ3 or σ4 , which is not in J only if the monomial is in the intersection
of σ3 and σ4 . Hence, we need to multiply I((d)) by either (h1 +h1 (d))(h3 +h3 (d))
or (h1 + h1 (d))(h4 + h4 (d)), so that for d in C10,
I(J )d = td ·
(h1 , h1 (−d))! (h3 , h3 (−d) − 1)! (h4 , h4 (−d) − 1)! · (h3 + h3 (d)), (h4 + h4 (d)),
and for d in R10,
D(RA , J )d = td ·
(h1 , h1 (−d))! (h3 , h3 (−d) − 1)! (h4 , h4 (−d) − 1)! · (h3 + h3 (d)), (h4 + h4 (d)).
the multidegrees of the monomials in J on the planes z + h1 (d) = 0, y + h2 (d) = 0,
x − z + h3 (d) = 0, and x − y + h4 (d) = 0. When an element of I((d)) is
applied to such a monomial in J , the result is a constant times a monomial in σ1
or σ2 , which is not in J , or a constant times a monomial in σ3 or σ4 , which is not
in J if this monomial lies in both planes parallel to these faces. Hence we need to
multiply I((d)) by (h1 + h1 (d))(h2 + h2 (d))(h3 + h3 (d))(h4 + h4 (d) − 1) or
(h1 + h1 (d))(h2 + h2 (d))(h3 + h3 (d) − 1)(h4 + h4 (d)), so that for d in C14,
I(J )d = td ·
* 3 +

4
2
(hi , hi (−d) − 1)! · (hi + hi (d)) , (h4 + h4 (d)) (hi + hi (d))
i=1 i=1 i=1
and for d in R14,
D(RA , J )d = td ·
* 3 2 +

4
(hi , hi (−d) − 1)! · (hi + hi (d)) , (h4 + h4 (d)) (hi + hi (d)) .
i=1 i=1 i=1
Putting all the information together from all 14 chambers C1–C14 and regions
R1–R14, the general formula for the graded piece of the idealizer of J at d ∈ Z3 is
I(J )d = td ·
⎛ ⎞ ⎛ ⎞
* +

4
⎜ ⎟ ⎜ ⎟
⎜
(hi , hi (−d) − 1)! · ⎝ ⎟ ⎜
(hi + hi (d))⎠ , ⎝ ⎟
(hi + hi (d))⎠ ,
i=1 hi (d)<0 hi (d)<0
for i=4 for i=3
and the general formula for the graded piece of D(RA , J ) at d ∈ Z3 is
D(RA , J )d = td ·
⎛ ⎞ ⎛ ⎞
* +

4
⎜ ⎟ ⎜ ⎟
(hi , hi (−d) − 1)! · ⎜
⎝ (hi + hi (d))⎟ ⎜
⎠,⎝ (hi + hi (d))⎟
⎠ .
i=1 hi (d)0 hi (d)0
for i=4 for i=3
We also have for any d ∈ Z3 ,

I(J )
= td ·
D(RA , J ) d
⎛ ⎞ ⎛ ⎞
* +
3
4
⎜ 3 ⎟ ⎜ 3 ⎟
(hi , hi (−d) − 1)! · ⎝ (hi + hi (d))⎠ , ⎝ (hi + hi (d))⎠
i=1 hi (d)<0 hi (d)<0
for i=4 for i=3
⎛ ⎞ ⎛ ⎞ .
* +
3
4
⎜ 3 ⎟ ⎜ 3 ⎟
(hi , hi (−d) − 1)! · ⎝ (hi + hi (d))⎠ , ⎝ (hi + hi (d))⎠
i=1 hi (d)0 hi (d)0
for i=4 for i=3
7 Non-normal Examples
Thus far, we have restricted our attention to normal semigroup rings. However,
Saito and Traves determined the ring of differential operators for all saturated affine
semigroup rings in [9, Theorem 3.3.1] (see also [10, Theorem 2.1]), and we can use
this work to compute I(J ), D(RA , J ) and I(J )/D(RA , J ) for non-normal RA and
a graded radical RA -ideal J . We broaden slightly from normal to scored semigroup
rings, see Definition 2.2. We include two examples; one is scored and the other is
not. However, the two quotient rings modulo the interior ideals are isomorphic and
we will see that the expressions for I(J )/D(RA , J ) in both rings are isomorphic.

230
Example 7.1 Consider the matrix A = , which is associated to the
001
semigroup ring RA = C[NA] = C[s 2 , s 3 , t]. Since RA is not normal, note that
integer points in the faces of A and NA differ from those in R0 A. We let
σ1 = N{e2 } and σ2 = N{e1
be the integer points in the facets of the cone R0 A. The primitive integral support
functions of NA
h1 = h1 (θ ) = θ1 and h2 = h2 (θ ) = θ2 .
The prime ideals associated to the facets of A are
P1 = s 2 , s 3 and P2 = t.
Set J = P1 ∩ P2 = s 2 t, s 3 t. Note that RA is a Gorenstein ring that is not normal,

and, in this case, the ideal J , which is the interior, is not the canonical module of
RA .
We will again use chambers of Z2 to aid in our computations of the graded pieces
of I(J )/D(RA , J ); in particular, we show that I(J )/D(RA , J ) is only nonzero in
multidegree d if d ∈ σ1 ∪ σ2 , as occurred in the normal cases computed thus far.
In Fig. 15, the red multidegrees correspond to the monomials in J , and the blue
multidegrees are exponents for monomials in the semigroup ring RA that do not lie
in J . The monomials at the orange multidegrees behave somewhat more like the
red and blue, in that, for all such d, I(J )d = D(RA )d . Thus, together the red, blue,
and orange multidegrees form chamber C1. The yellow multidegrees constitute C2,
the violet points form C3, and both the green and dark green make C4. The vertical
lines {x = −1} and {x = 1} contain multidegrees d for which extra care is needed
to determine I(J )d and D(RA , J )d .
Eriksen showed in [3, Proposition 2.1] that the ring of differential operators of a
numerical semigroup ring R = C[t ] is
Fig. 15 Chambers for C[s 2 , s 3 , t]
* +
'
D(R) = d1 d2
s t · (θ − γ ) ,
d∈Z γ ∈(d)
where (d) = {γ ∈ | γ + d ∈
/ }. We encode these polynomial generators using
Gd (θ ), where
⎧
⎪
⎪1 if d1 ∈ N2, 3,
⎪
⎪
⎪
⎨h1 if d1 = 1,
Gd (θ ) :=
⎪
⎪h1 − 1 if d1 = −1,
⎪
⎪
⎩ (h1 , 1 − d1 )!
⎪
if d1 −2.
(h1 − 1)(h1 + d1 )
By Theorem 2.3, for any d ∈ Z2 ,
D(RA )d = s d1 t d2 · Gd (θ ) (h2 , −d2 − 1)! .
To determine I(J ) or D(RA , J ), we must determine the lines of multidegrees that

contain operators that, when applied to monomials in J or R, produce elements in
R \ J . Note that given m ∈ NA,
s d1 t d2 Gd (θ ) (h2 , −d2 − 1)! ∗ s m1 t m2 = Gd (m)(h2 (m), −d2 − 1)! · s d1 +m1 t d2 +m2 .

(7.1)
If d ∈ C1 and s m1 t m2 ∈ J , then the result of (7.1) will remain in J . Thus, for all
d ∈ C1, D(RA , J )d = s d1 t d2 · Gd (θ ).
Next, for d ∈ C2, when m ∈ NA ∩ {x = −d1 }, then the result of (7.1) is a
constant times a monomial in σ1 , which does not belong to J . Hence, for all d ∈ C2,
I(J )d = s d1 t d2 · Gd (θ )(h1 + d1 ).

Further, if d ∈ C2 ∪ (σ1 \ {0}), then
D(RA , J )d = s d1 t d2 · Gd (θ )(h1 + d1 ).
For d ∈ C4, when m ∈ NA ∩ {y = −d2 }, then the result of (7.1) is a constant

times a monomial in σ2 , which does not belong to J . Hence, for all d ∈ C4,
I(J )d = s d1 t d2 · Gd (θ ) (h2 , −d2 )!.
Further, if d ∈ C4 ∪ (σ2 \ {0}), then
D(RA , J )d = s d1 t d2 · (h2 , −d2 )!.
Finally, for d ∈ C3, if m ∈ {x = −d1 } ∩ {c | s c1 t c1 ∈ J } or if m ∈ {y =

−d2 }∩NA, respectively, then the result of (7.1) is either a constant times a monomial
in σ1 or σ2 , respectively. Hence,
I(J )d = s d1 t d2 · Gd (θ )(h1 + d1 ) (h2 , −d2 )!.
Finally, if d ∈ C3 ∪ (−σ1 ) ∪ (−σ2 ), then
D(RA , J )d = s d1 t d2 · Gd (θ )(h1 + d1 ) (h2 , −d2 )!.
Gathering these computations,

⎧
⎪
⎪
d1 d2 · Gd (θ ) if d ∈ C1,
⎪s t
⎪
⎨s d 1 t d 2 · Gd (θ )(h1 + d1 ) if d ∈ C2,
I(J )d =
⎪s d 1 t d 2
⎪ · Gd (θ )(h1 + d1 ) (h2 , −d2 )! if d ∈ C3,
⎪
⎪
⎩ d1 d2
s t · Gd (θ ) (h2 , −d2 )! if d ∈ C4,
and
⎧
⎪
⎪ s d1 t d2 · Gd (θ ) if d ∈ C1 \ (σ1 ∪ σ2 ),
⎪
⎪
⎪
⎪
⎨s t · Gd (θ )(h1 + d1 )
⎪ if d ∈ (C2 \ σ2 ) ∪ (σ1 ∩ {d2 > 0}),
d1 d2
D(RA , J )d = s d1 t d2 · Gd (θ )·
⎪
⎪
⎪
⎪ (h1 + d1 ) (h2 , −d2 )! if d ∈ C3 ∪ {0} ∪ (−σ1 ) ∪ (−σ2 ),
⎪
⎪
⎪
⎩s d1 t d2 · G (θ ) (h , −d )!
d 2 2 if d ∈ (C4 \ σ1 ) ∪ (σ2 ∩ {d1 > 0}).
Taking the quotient of the above multigraded modules, we then obtain a description
of the graded pieces of I(J )/D(RA , J ) as follows:
⎧
⎪
⎪ 0 if d ∈
/ Zσ1 ∪ Zσ2 ,
⎪
⎪
⎪
⎪ C[θ ]
⎪
⎪ if d = 0,
⎪
⎪ h h
⎪
⎪ 1 2
I(J ) ⎨ Gd (θ )
= s t · if d1 = 0, d2 = 0,
d 1 d2 ,
D(RA , J ) d ⎪ Gd (θ )h2
⎪
⎪ Gd (θ )
⎪
⎪ s d1 t d2 · if d1 = 0, d2 > 0,
⎪
⎪ Gd (θ )h1
⎪
⎪
⎪
⎪ Gd (θ ) (h2 , −d2 )!
⎪
⎩s d1 t d2 · if d1 = 0, d2 < 0.
h1 Gd (θ ) (h2 , −d2 )!
In this case, J D(RA ) = D(RA , J ), so (1.2) happens to hold in this example. To

see this, compute (J D(RA ))d by looking at I((d − (2, 1)) and I((d − (3, 1)) for
any d ∈ Z3 . For an example of how showing this equality works, when (d1 − 2, d2 −
1) and (d1 − 3, d2 − 1) are both in C2 \ σ2 , then for d1 < 0,
(h1 , 3 − d1 )!
G(d1 −2,d2 −1) (θ ) = and
(h1 − 1)(h1 + d1 − 2)
(h1 , 4 − d1 )!
G(d1 −3,d2 −1) (θ ) = .
(h1 − 1)(h1 + d1 − 3)
Both of these polynomials are divisible by Gd (θ )(h1 + d1 ):
(h1 , 3 − d1 )!
G(d1 −2,d2 −1) (θ ) =
(h1 − 1)(h1 + d1 − 2)
= Gd (θ )(h1 + d1 )(h1 + d1 − 3)
and
(h1 , 4 − d1 )!
G(d1 −2,d2 −1) (θ ) =
(h1 − 1)(h1 + d1 − 2)
= Gd (θ )(h1 + d1 )(h1 + d1 − 2)(h1 + d1 − 4).
Further,
1 = (h1 + d1 − 3)2 − (h1 + d1 − 2)(h1 + d1 − 4),
which implies that

6 7
Gd (θ )(h1 + d1 ) = G(d1 −2,d2 −1) (θ ), G(d1 −3,d2 −1) (θ ) .
Hence for d ∈ C2, there is an equality D(RA , J )d = s d1 t d2 · Gd (θ )(h1 + d1 ) =

(J D(RA ))d .

= 1230
Example 7.2 Consider the matrix A , which is associated to the
1001
semigroup ring RA = C[NA] = C[st, s 2 , s 3 , t], which is not normal, scored, or
Gorenstein. As in Example 7.1, we denote the integer points in the facets of the
by
cone R0 A
σ1 = N{e2 } and σ2 = N{e1 }.
are
The primitive integral support functions of NA
h1 = h1 (θ ) = θ1 and h2 = h2 (θ ) = θ2 .
are
The prime ideals associated to the facets of A
P1 = st, s 2 , s 3 and P2 = st, t.
Finally, set J = P1 ∩ P2 = st, s 2 t.

Note that RA /J from Example 7.1 is isomorphic to RA/J. Hence, we would
also expect that I(J )/D(RA ,!J ) ∼
= I(J )/D(RA, J). We will determine the graded

pieces of I(J ) and D RA, J , as well as I(J)/D(RA, J); then, we will exhibit an

isomorphism between the graded pieces of I(J)/D(RA, J) and I(J )/D(RA , J ).
We will again use chambers of Z2 to aid in our computations of the graded pieces
of I(J )/D(RA , J ). In Fig. 16, the red multidegrees correspond to monomials in
J, while the blue ones (both light and dark) are the remaining d ∈ Z2 for which
D(RA)d = I(J)d ; together, these red and blue multidegrees form chamber C1. The
remaining multidegrees are split into chambers as in the previous examples: yellow
points form C2, violet points make C3, and green points give C4.
In [9, Examples 3.2.7 and 3.3.4], Saito and Traves found that for any d ∈ Z2 ,
D(RA)d = s d1 t d2 · I((d)), where
Fig. 16 Chambers for C[st, s 2 , s 3 , t]

⎧* +
⎪
⎪ 32
⎪
⎪ (hi , hi (−d) − 1)! if d ∈
/ e1 − NA,
⎪
⎪
⎨ i=1
I((d)) = 32 (7.2)
⎪
⎪ (hi , hi (−d) − 1)!·
⎪
⎪i=1
⎪
⎪
⎩ h1 + d1 − 1, h2 + d2
if d ∈ e1 − NA.
To compute I(J), we first determine which elements of I((d)) are already in I(J).
To do this, we first break up (7.2) into cases by chamber:
⎧
⎪
⎪ s d1 t d2 · I((d)) if d ∈ C1,
⎪
⎪
⎪
⎪
⎨s t · (h1 , h1 (−d))!
⎪ if d ∈ C2 \ (−σ2 ),
d1 d2
I(J)d = s d1 t d2 · (h1 , h1 (−d))! · h1 + d1 − 1, h2 if d ∈ (−σ2 ),

⎪
⎪
⎪
⎪ s t · (h1 , h1 (−d))! (h2 , h2 (−d))!
d d if d ∈ C3,
⎪
⎪
1 2
⎪
⎩s d1 t d2 · (h , h (−d))!
2 2 if d ∈ C4.
Similarly,
⎧
⎪
⎪ s d1 t d2 · C[θ ] if s d1 t d2 ∈ J,
⎪
⎪
⎪
⎪
⎨s t · (h1 , h1 (−d))!
⎪ if d ∈ (C2 \ (−σ2 )) ∪ (σ1 \ {0}),
d1 d2
D(RA, J)d = · (h2 , h2 (−d))!

s d1 t d2 if d ∈ C4 \ (−σ1 ),
⎪
⎪
⎪
⎪ s t · (h1 , h1 (−d))!·
d d
⎪
⎪
1 2
⎪
⎩ (h , h (−d))!
2 2 if d ∈ C3 ∪ (−σ1 ) ∪ (−σ2 ).
Putting these together, the graded piece of I(J)/D(RA, J) at d ∈ Z2 is

⎧
⎪
⎪ 0 if d ∈
/ Zσ1 ∪ Zσ2 ,
⎪
⎪
⎪
⎪ C[θ ]
⎪
⎪ s d1 t d2 · if d ∈ σ1 \ {0},
⎪
⎪ h1
⎪
⎪ C[θ ]
⎪
⎪
⎪
⎪ s d1 t d2 · if d ∈ σ2 \ {0, (1, 0)},
⎪
⎪ h2
⎪
⎨ d d C[θ ]
I(J) · if d = 0,
= s t
1 2

D(RA, J ) d ⎪ θ1 θ2
⎪
⎪ h1 , h2
⎪
⎪ s d1 t d2 · if d = (1, 0),
⎪
⎪ h2
⎪
⎪
⎪
⎪ (h1 , h1 (−d))!h1 + d1 − 1, h2
⎪
⎪ s d1 t d2 · if d ∈ (−σ2 ) \ {0},
⎪
⎪ (h1 , h1 (−d))!h2
⎪
⎪
⎪
⎪ (h2 , h2 (−d))!
⎩s d1 t d2 · if d ∈ (−σ1 ) \ {0}.
h1 (h2 , h2 (−d))!
Looking at the graded pieces of I(J )/D(RA , J ) from Example 7.1 and noting
that Gd (θ ) = C[θ ] for d ∈ (σ1 ) ∪ (σ2 \ {0, (1, 0)}, it follows that (I(J )/D(RA , J ))d
!
is identical to I(J)/D(RA, J) d for all d except d = (1, 0) and d ∈ (−σ2 ) \ {0}.
So we only need to exhibit that the graded pieces are isomorphic for d ∈ {(1, 0)} ∪
(−σ2 ) \ {0}. Note that for d1 < 0 and d1 = −1,
, -
(h1 , 1 − d1 )!
(h − 1)(h1 + d1 ) ∼ (h1 , 1 − d1 )!
, 1 -=
(h1 , 1 − d1 )! h2 (h1 , 1 − d1 )! h2
(h1 − 1)(h1 + d1 )
by the Third Isomorphism Theorem, and
(h1 , h1 (−d))!h1 + d1 − 1, h2 ∼ (h1 , 1 − d1 )!

=
(h1 , h1 (−d))!h2 (h1 , 1 − d1 )! h2
by the Second Isomorphism Theorem. A similar argument gives an isomorphism

in the remaining multidegrees. Combining the information on all graded pieces, we
have explicitly shown that
I(J) I(J )
!∼
= .
D RA, J D (RA , J )
Acknowledgments We are grateful to the organizers, Karen Smith, Sandra Spiroff, Irena Swan-
son, and Emily E. Witt, of Workshop: Women in Commutative Algebra (WICA), at which this
work began. Additionally, we are thankful to BIRS for their hospitality in hosting this meeting.
We also owe thanks to Jack Jeffries, who pointed us to an example that improved our work. We
truly appreciate the comments of the referee which added extra insight into our work and greatly
improved the paper.
References
1. Christine Berkesch, C-Y. Jean Chan, Patricia Klein, Laura Matusevich, Janet Page, and Janet
Vassilev, Differential operators on quotients of normal affine semigroup rings, In progress,
2021.
2. Anders Eriksson, The ring of differential operators of a Stanley-Reisner ring, Comm. Algebra
26 (1998), no. 12, 4007–4013.
3. Eivind Eriksen, Differential operators on monomial curves, J. Algebra 264 (2003), 186–198.
4. Ian M. Musson, Rings of differential operators on invariant rings of tori, Trans. Amer. Math.
Soc. 303 (1987), no. 2, 805–827.
5. , Rings of differential operators and zero divisors, J. Algebra 125 (1989), no. 2, 489–
501. MR 1018959
6. , Differential operators on toric varieties, J. Pure Appl. Algebra 95 (1994), no. 3, 303–
315.
7. Ian M. Musson and Michel Van den Bergh, Invariants under tori of rings of differential
operators and related topics, Mem. Amer. Math. Soc. 136 (1998), no. 650, viii+85.
8. S.P. Smith and J.T. Stafford, Differential operators on an affine curve, Proc. London Math. Soc.
56 (1988), no. 3, 229–259.
9. Mutsumi Saito and William Traves, Differential algebras on semigroup algebras, Contemp.
Math. 286 (2001), 207–226.
10. , Finite generation of rings of differential operators of semigroup algebras, J. Algebra
278 (2004), 76–103.
11. William Traves, Differential operators on monomial rings, J. Pure and Appl. Algebra 136
(1999), 183–197.
12. J.R. Tripp, Differential operators on Stanley-Reisner rings, Trans. of the A.M.S 349 (1997),
no. 6, 2507–2523.
A Hypergraph Characterization of
Nearly Complete Intersections
Chiara Bondi, Courtney R. Gibbons, Yuye Ke, Spencer Martin,

Shrunal Pothagoni, and Ada Stelzer
Keywords Betti diagrams · Nearly complete intersections · Monomial ideals ·

Hypergraph ideals
1 Introduction
A standard tool for studying a homogeneous ideal I over a polynomial ring R is the
minimal free resolution of the module R/I . The rank of the i-th module in the free
resolution is called the i-th Betti number of R/I and denoted βi (R/I ), and these
invariants, along with their graded counterparts (denoted βij (R/I ), the ij -th Betti
number of R/I is the number of degree j minimal generators of the i-th syzygy of
R/I ), have been the focus of much study. When I is a complete intersection ideal
with g generators, its free resolution is the Koszul
! complex on its generators, and its
i-th Betti number is given by βi (R/I ) = gi .
This work was initiated as part of the virtual COURAGE REU in Summer 2020, supported by
Clemson University’s School of Mathematical and Statistical Sciences. All of the authors thank
the referee for their helpful and kind comments and uncountably many women in commutative
algebra for inspiration, advice, and friendship.
C. Bondi · C. R. Gibbons ()

Hamilton College, Clinton, NY, USA
e-mail: crgibbon@hamilton.edu
Y. Ke
The Ohio State University, Columbus, OH, USA
S. Martin
University of Virginia, Charlottesville, VA, USA
S. Pothagoni
George Mason University, Fairfax, VA, USA
A. Stelzer
Lawrence University, Appleton, WI, USA

96 C. Bondi et al.
The Buchsbaum-Eisenbud and Horrocks rank conjecture ! (see [2, 1.4]) states that
an ideal I of height c will always satisfy βi (R/I ) ci . In the case of a complete
intersection, it happens that g = c and there is an equality. When g > c and I is not a
complete intersection, much less is known. In [8], Walker proved that the sum of the
Betti numbers of an arbitrary ideal I is at least 2c if the characteristic of the base field
is not 2. In the case of monomial ideals, [1], Boocher and Seiner showed that if I is
not a complete intersection ideal, then the sum of its Betti numbers βi (R/I ) is at
least 2c + 2c−1 . Boocher and Seiner’s proof centers on reducing arbitrary monomial
ideals to a new class they called nearly complete intersections. They define a nearly
complete intersection to be a square-free monomial ideal I (not itself a complete
intersection) such that for any variable x in the support of I , the ideal I (x = 1)
(formed by substituting in 1 for x in the generators of I ) is a complete intersection
ideal. Geometrically, one can view a nearly complete intersection as a monomial
ideal over a homogeneous ring whose standard affine patches are given by complete
intersection ideals.
In [6], Miller and Stone succeeded in classifying nearly complete intersections
generated in degree two as the edge ideals of a certain set of simple graphs.
Theorem 1.1 (Miller-Stone [6]) Let G be a connected graph with at least 3
vertices. The edge ideal I of G is not a nearly complete intersection (generated
in degree 2) if and only if there exist vertices v1 , v2 , v3 , v4 , v5 in G such that v1 is a
leaf in their induced subgraph H and H has a spanning tree isomorphic to one of
the following:
1 or 1
Our main result, Theorem 3.1, generalizes Miller and Stone’s work by classifying
all nearly complete intersections as the edge ideals of particularly nice hypergraphs,
eliminating the degree restriction. In Sect. 2, we extend Miller and Stone’s notion of
vertex inversion to hypergraphs in order to describe nearly complete intersections
(hyper)graphically. We prove that this operation commutes with taking induced
sub-hypergraphs, laying the foundation for a complete characterization of NCI-
hypergraphs in Sect. 3. Section 4 contains a decomposition of the graded Betti
numbers of an arbitrary nearly complete intersection I , using an iterated mapping
cone construction to reduce the general case to degree 2; we also comment on the
effect higher degree generators have on homological invariants of these ideals.
Finally, we include an appendix to discuss an alternative representation of nearly
complete intersections as vertex-weighted graphs and share a webtool developed to
make the construction and manipulation of these graphs easier.
A Hypergraph Characterization of Nearly Complete Intersections 97
2 Hypergraphs and Squarefree Monomial Ideals
A simple hypergraph G is a pair (V , E) where V is a set and E ⊆ 2V is a subset of

the power set of V . We call an element v ∈ V a vertex of G and an element e ∈ E
of cardinality k a k-edge of G.
When two elements e, f ∈ E have the property that e ⊆ f , we say that e is a
subedge of f . If f has no subedges except for itself in G, then we call f minimal.
If every k-edge of G is minimal, then we call G a minimal hypergraph. Given an
arbitrary hypergraph G, we can always remove nonminimal edges to get a unique
minimal hypergraph, which we denote Min(G). In this paper, all hypergraphs are
assumed to be minimal unless otherwise indicated. We will often refer to 2-edges as
“edges” and k-edges for k 3 as “hyperedges”. When 1-edges occur, we identify
them as such.
Given a hypergraph G = (V , E), its skeleton is the simple graph S = (V , ES ),
where ES is the set of all 2-edges in E. We define the 2-neighbor set N(v) for a
vertex v in V to be its neighbors in the skeleton of G, i.e. N(v) = {w|{v, w} ∈
ES }. The next definitions provide two distinct yet closely related ways to take sub-
hypergraphs of G, both of which will be important in this paper.
Definition 2.1 An induced sub-hypergraph of G = (V , E) is a minimal hyper-

graph G = (V , E ) where V ⊆ V and E = E ∩ 2V , the set of all e ∈ E
containing only vertices in V . We denote this sub-hypergraph by G = IndV (G)
Definition 2.2 A weak induced sub-hypergraph of G = (V , E) is a (not necessar-
ily minimal) hypergraph G = (V , E ) where V ⊆ V and E = {e ∩ V |e ∈ E}.
Instead of removing k-edges that aren’t strictly contained in V , we restrict each one
down to the vertices in V . We denote this sub-hypergraph by G = WeakIndV (G).
Next, we describe how to represent squarefree monomial ideals as minimal
hypergraphs. Let R = k[x1 , . . . , xn ] be a standard graded k-algebra, where k is
a field. Unless stated otherwise, all ideals are assumed to be squarefree monomial
ideals over R. The minimal generating set of such an ideal naturally leads to the
following definition:
Definition 2.3 Let I be a squarefree monomial ideal with minimal monomial
generating set {m1 , . . . , m
} over R. The hypergraph of I (denoted G(I )) is the
minimal hypergraph (V , E), where V is the support of I (i.e. all variables of R
appearing in some m
) and E = {m1 , . . . , m
}.
Conversely, given a minimal hypergraph G = (V , E) where V = {v1 , . . . , vn }
and E = {m1 , . . . , m
}, the (hyper)edge ideal of G is the squarefree monomial
ideal I (G) = (m1 , . . . , m
) over the ring R = k[v1 , . . . , vn ]. Since the minimal
monomial generators of a squarefree monomial ideal are unique, this yields a one-
to-one correspondence between minimal hypergraphs and squarefree monomial
ideals.
98 C. Bondi et al.
Example 2.4 Let I = (x1 , x2 x3 , x3 x4 x5 ). Then the hypergraph G(I ) consists of a

1-edge around vertex x1 , an edge connecting vertices x2 and x3 , and a hyperedge
containing vertices x3 , x4 , and x5 .
2 5
1 3
Since nearly complete intersection ideals are squarefree monomial ideals by

definition, they can be represented by hypergraphs, and much of this paper is
devoted to establishing the structure of these representations. We begin with a
trivial definition before translating the algebraic definition of a nearly complete
intersection into hypergraph theory.
Definition 2.5 A hypergraph G is an NCI-hypergraph if its (hyper)edge ideal I (G)
is a nearly complete intersection. Similarly, G is a CI-hypergraph if I (G) is a
complete intersection ideal.
Lemma 2.6 If G = (V , E) is a CI-hypergraph, then each vertex v is contained in
at most one k-edge e in E.
Proof If some v is contained in both e1 and e2 , then I (G) must have two minimal
generators both divisible by v, and therefore I (G) cannot be a complete intersection
ideal.

As ideals, nearly complete intersections are defined using a localization operation
that amounts to substituting x = 1 for a variable x in the support of I . This operation
can be written in the language of hypergraphs we have developed so far.
Definition 2.7 Let G = (V , E) be a hypergraph and let v ∈ V . The vertex
inversion of G at v is given by Iv (G) = Min(WeakIndV \{v} (G))
Remark 2.8 Let I be a square-free monomial ideal. An inversion at any vertex xi in
the hypergraph of I yields the hypergraph of I (xi = 1). From this, it immediately
follows that a hypergraph G is an NCI-hypergraph iff Iv (G) is a CI-hypergraph for
every vertex v in G.
In later sections we will frequently use the key fact that the operations of vertex
inversion and taking induced subhypergraphs commute. Our rigorous proof of this
fact is followed by Fig. 1 which demonstrates the basic idea.
Lemma 2.9 Let G be a hypergraph, let V be a vertex subset of G, and let v ∈ V .
Then Iv (IndV (G)) = IndV (Iv (G)).
2 3 2 3
5
Invert at 5
1 4 1 4
Take induced sub-

Take induced sub-
hypergraph on
hypergraph on 1 2
1 2 5
2 2
5
Invert at 5
1 1
Fig. 1 Illustration of Lemma 2.9
Proof By the definition of vertex inversion and the observation that induced and
weak induced subgraphs commute, we have
Iv (IndV (G)) = Min(WeakIndV \{v} (IndV (G)))

= Min(IndV (WeakIndV \{v} (G))).
It remains to show that this expression is equal to the induced sub-hypergraph

IndV (Min(WeakIndV \{v} (G))). To do so, we show that Min and IndV commute
in general. Consider the poset P of (hyper)edges in a hypergraph G ordered by
inclusion. Then the poset P of (hyper)edges in IndV (G) is simply some subset of
P, so any minimal elements of P contained in P remain minimal in P . Similarly,
if e ∈ P is not minimal in P, then there is e ⊆ e in P. But by the definition of
IndV (G), we have e ∈ P , so e is not minimal in P . Thus e ∈ P is minimal if and
only if it is minimal in P, so IndV and Min do indeed commute.
Therefore,
Iv (IndV (G)) = Min(WeakIndV \{v} (IndV (G)))

= Min(IndV (WeakIndV \{v} (G)))
= IndV (Min(WeakIndV \{v} (G)))
= IndV (Iv (G)).

100 C. Bondi et al.
Before proving results on the hypergraphs of nearly complete intersections, we

introduce a special class of hypergraphs that have an almost trivial hyperedge
structure.
Definition 2.10 A hypergraph G is joinable if it satisfies the following properties:
(J1) No two hyperedges of G intersect.
(J2) If e = x1 . . . xn is a hyperedge of G, then N(x1 ) = . . . = N(xn ).
In a joinable hypergraph, hyperedges are only permitted to group disjoint sets of
vertices with identical 2-neighbor sets together, creating “fat vertices” that enable
us to view them as simple graphs.
Definition 2.11 Given a joinable hypergraph G, the join of G at a hyperedge h is
the hypergraph G = Joinh (G) defined by replacing all vertices in h with a single
vertex v. (Recall that a 2-edge is not a hyperedge; see Fig. 4.) If all elements of h
are connected to a vertex w in G, then v is connected to w in G .
If no hyperedge is specified, Join(G) refers to the simple graph obtained by
iteratively taking the join of G at each of its hyperedges. Note that the order in
which the hyperedges are joined into vertices does not matter, so this operation is
well-defined. A slightly modified version of the join operation defines a bijection
between joinable hypergraphs and vertex-weighted graphs, which makes it possible
to describe most of our results in the language of weighted graphs. For more details
on this process, see the appendix.
3 NCI-hypergraphs
The goal of this section is to prove the following theorem, which is a complete
characterization of nearly complete intersections in terms of their corresponding
hypergraphs. Recall that a CI-hypergraph is a collection of disjoint (hyper)edges,
and that an NCI-hypergraph is a hypergraph G such that Iv (G) is a CI-hypergraph
for every vertex v of G.
Theorem 3.1 An ideal I is a nearly complete intersection if and only if the
corresponding hypergraph G = G(I ) satisfies the following properties:
• G is joinable;
• The skeleton of G is an NCI-graph.
We prove that NCI-hypergraphs are joinable directly, verifying conditions J1
and J2 separately. The fact that J1 holds for these hypergraphs is essentially a
restatement of [1, Lemma 4.1 and Lemma 4.3] using our framework. We present
our own proof to illustrate how the ideas from that paper can be restated graphically.
The fundamental fact needed is Lemma 2.9, which allows us to easily reduce these
statements to the local level in proofs by contraposition.
Lemma 3.2 Let G be an NCI-hypergraph. Then no two hyperedges of G intersect.
Invert at
(a)
1
Invert at Invert at
1
(b)
Invert at
(c)
Fig. 2 Proof illustrations. (a) Illustration of first paragraph of the proof of Lemma 3.2. (b)
Illustration of second paragraph of the proof of Lemma 3.2. (c) Illustration of the proof of
Lemma 3.3
Proof Let G be a simple hypergraph in which two hyperedges e1 and e2 intersect,

and let H = Inde1 ∪e2 (G) be the induced subhypergraph on all elements of e1 and
e2 . The proof breaks into three cases. First, suppose that the intersection e1 ∩ e2
contains at least two vertices v and w. If we invert H at v, the resulting hypergraph
must contain both e1 \ v and e2 \ v, which intersect at w (see Fig. 2a). This implies
that Iv (H ) is not a CI-hypergraph, and by Lemma 2.9 this graph is identical to
Inde1 ∪e2 (Iv (G)), which is a subgraph of Iv (G). It follows that G cannot be an
NCI-hypergraph in this case (see Fig. 2a).
Now, suppose instead that e1 and e2 intersect in a single vertex v—let e1 =
(a1 , . . . , an , v) and e2 = (b1 , . . . , bm , v). By the previous case we know that e1
and e2 must be the only hyperedges in H , and also that v cannot be an element of
any edge in H . In the case where there are edges in H , we can relabel vertices so
that we have the edge (a1 , b1 ) and then invert H at v. We are left with a hypergraph
containing the hyperedge e1 \ v and edge (a1 , b1 ) (among other things), which is not
the hypergraph of a complete intersection. If there are no edges in H , then inverting
H at a1 yields a graph with hyperedges e1 \ a1 and e2 , which does not represent a
complete intersection either. Applying Lemma 2.9 again proves that G cannot be an
NCI-hypergraph (see Fig. 2b).

102 C. Bondi et al.
Condition J2 (elements of hyperedges have the same 2-neighbors) is shown to be

necessary in identical fashion:
Lemma 3.3 If H is an NCI-hypergraph and e = (x1 , . . . , xn ) is a hyperedge of H ,
then N (x1 ) = . . . = N(xn ).
Proof Let G be a simple hypergraph, e = (x1 , . . . , xn ) a hyperedge of G, and v a
vertex of G connected to x1 by an edge but not to xn . Let H = Inde∪{v} (G) be the
subhypergraph induced by {x1 , . . . , xn , v} and consider its inversion Ixn (H ) at xn
(see Fig. 2c).
The resulting hypergraph must contain both (x1 , v) and (x1 , . . . , xn−1 ) and
thus does not represent a complete intersection. Lemma 2.9 then implies that the
inversion of G at xn does not represent a complete intersection either, so G cannot
be an NCI-hypergraph.

To prove Theorem 3.1, it now suffices to show that if G = (V , E) is a joinable
NCI-hypergraph, then adding or removing a hyperedge from G gives another NCI-
hypergraph. The proof of each case here is brief.
Proof of Theorem 3.1 First, let e be a hyperedge of G. Since G is joinable, only
2-edges intersect with e. The cardinality of e is still at least 2 in the inversion Iv (G)
for any v ∈ V , so e does not delete any elements of E when Min is applied. It
follows that Iv (G − {e}) is a subgraph of Iv (G), which implies that Iv (G − {e})
represents a complete intersection for all v ∈ V . Thus G−{e} is an NCI-hypergraph
as desired.
In the other direction, suppose that W = (v1 , . . . , vn ) ⊂ V is an independent
set of vertices (n 3) such that G = (V , E ∪ {W }) is still a joinable hypergraph.
Since N (v1 ) = . . . = N(vn ) and G is an NCI-hypergraph by assumption, all of the
vi must become isolated after inversion at any v ∈ V . This implies that in Iv (G )
the vi are either isolated as in G or grouped into the single hyperedge W . In either
case every vertex is contained in at most one k-edge in this inversion, so G is still
an NCI-hypergraph.
The proof of Theorem 3.1 follows quickly now. If G is an NCI-hypergraph,
then it must be joinable by Lemmas 3.2 and 3.3, and the above discussion implies
that its skeleton must also be an NCI-hypergraph, so G satisfies the hypotheses
of Theorem 3.1. Conversely, if G satisfies the hypotheses of the theorem then its
skeleton is an NCI-hypergraph by assumption, and we can add on its hyperedges
one by one without changing this fact.

Given an arbitrary ideal I , it is fairly easy to check whether or not G(I ) is a
joinable hypergraph, so the utility of Theorem 3.1 hinges entirely on our ability to
identify nearly complete intersections generated in degree 2. In [6], Stone and Miller
provide a characterization of the graphs of these ideals using two small forbidden
subgraphs, completing our description of NCI-hypergraphs. We close this section
with a lemma describing the effect of the join operation on NCI-hypergraphs, which
will be useful later.
Lemma 3.4 Let G = (V , E) be an NCI-hypergraph and let h ∈ E be a hyperedge.

Then Joinh (G) is an NCI-hypergraph or a CI-hypergraph.
Proof Let v ∈ h, and let V = (V \h) ∪ v. Then Joinh (G) = ∼ IndV (G). We wish
to show that Iv (IndV (G)) is a CI-hypergraph. By the commutativity of vertex
inversion with taking induced sub-hypergraphs, we have that Iw (IndV (G)) =
IndV (Iw (G)) for any w ∈ V . Since G is an NCI-hypergraph, Iw (G) is always
a CI-hypergraph, meaning IndV (Iw (G)) is also a CI-hypergraph. Hence, inverting
at any vertex in Joinh (G) yields a CI-hypergraph. Thus, Joinh (G) must be an NCI-
hypergraph or a CI-hypergraph.

4 Resolutions of NCIs
Using the established structure of NCI-hypergraphs, we can now extract information

about their Betti numbers and minimal free resolutions. We explicitly describe the
effect of the generators of degree 3 and higher on the Betti table of a nearly complete
intersection. In this section, we gently abuse terminology by referring to the skeleton
of a nearly complete intersection ideal I ; by this we mean the degree 2 squarefree
monomial ideal corresponding to the skeleton of the induced hypergraph G(I ).
Given the minimal free resolution of the skeleton of a nearly complete inter-
section, the minimal free resolution of the entire nearly complete intersection is
obtained through the use of an iterated mapping cone construction. To prove the
minimality of this constructed resolution, it suffices to show that each step is a Betti
splitting of our ideal as defined in [7]. We take this opportunity to remind the reader
that the Betti numbers of I are slightly offset from the Betti numbers of R/I , with
βij (I ) = βi+1 j (R/I ).
Definition 4.1 (from [7]) I = J + K is a Betti splitting if for all i, j 0 we have
βij (I ) = βij (J ) + βij (K) + βi−1,j (J ∩ K)
Note that if I is generated by {m1 , . . . , mn }, any bipartition of the mi provides a

candidate decomposition J + K. From the definition, it is not immediately obvious
how to check if a given partition yields a Betti splitting without explicitly computing
the Betti tables. Fortunately, Eliahou and Kervaire provided a more applicable
condition for proving a partition is a Betti splitting in the ungraded case [3]. This
result was extended by Fatabbi to show that such Betti splittings respect graded Betti
numbers [5]. Let G(I ) represent the generators of I .
Definition 4.2 (from [3]) We say that J + K = I is a splitting of I if G(J )
and G(K) form a partition of G(I ) and there exists a splitting function G(J ∩
K) → G(J ) × G(K) sending w to (φ(w), ψ(w)) and satisfying the following two
properties:
(S1) For all w ∈ G(J ∩ K), w = lcm(φ(w), ψ(w)).
104 C. Bondi et al.
(S2) For every subset G ⊂ G(J ∩ K), both lcm(φ(G )) and lcm(ψ(G )) strictly
divide lcm(G ).
Defining an appropriate splitting function is more feasible than attempting to
explicitly compute free resolutions of arbitrary nearly complete intersections. We
prove that any nearly complete intersection I with at least one generator h of degree
at least 3 has a Betti splitting I = (h) + K. Applying this decomposition iteratively
allows us to break the Betti table of I down into a sum of three sets of tables: the
table of the skeleton of I , trivial tables coming from each principal ideal (h), and
tables of intersection ideals (h) ∩ K. We also prove that (h) ∩ K is a multiple of a
complete intersection.
Theorem 4.3 Let I = (h) + K be a nearly complete intersection where h is a
monomial of degree at least 3. Then (h) + K is a Betti splitting of I .
Proof If K is generated by (m1 , . . . , mn ), then a (not necessarily minimal) set
of generators of (h) ∩ K is given by {lcm(hmi )}ni=1 . From our knowledge of
the hypergraph structure of G(I ), we know that each mi has one of two forms:
either it can be written as xy, where x lies in h and y is connected to every
vertex of h, or mi = e represents a (hyper)edge of G(I ) that does not intersect
h. In the first case, lcm(h, mi ) = lcm(h, xy) = hy, while in the second case
lcm(h, mi ) = lcm(h, e) = he. Some of these generators may generate each other,
in which case we remove those terms to get a minimal generating set where all
generators are of one of these forms.
We can now explicitly construct a splitting function G((h) ∩ K) → G((h)) ×
G(K). Define the function w → (φ(w), ψ(w)) as follows: φ(w) = h, ψ(he) = e,
and ψ(hy) = xy, where x is some fixed vertex chosen in h. We now verify the two
properties of this splitting function in turn.
(S1) We need to show that for all w generating the intersection, w =
lcm(φ(w), ψ(w)). If w = he, then φ(w) = h and ψ(w) = e, so their
least common multiple is indeed he = w. If w = hy, then φ(w) = h and
ψ(w) = xy, and since x ∈ h and y ∈ / h by definition their least common
multiple is hy = w.
(S2) We need to show that for every subset G of generators in the intersection,
both lcm(φ(G )) and lcm(ψ(G )) strictly divide lcm(G ). We know
that lcm(φ(G )) is always just h, which certainly divides lcm(G ).
The set ψ(G ) consists of m distinct edges e1 , . . . , em and n distinct
terms of the form xy1 , . . . , xyn . The lcm of all these terms must be
lcm(x, e1 , . . . , em , y1 , . . . , yn ). This strictly divides lcm(G ), which is the
same lcm with x replaced by all of h.

Take an arbitrary nearly complete intersection I , call its skeleton S, and denote
the hyperedges of its hypergraph by h1 , h2 , . . . , hn . For 1 k n, let Kj denote
the ideal generated by all generators of I except for h1 , . . . , hj . Applying the above
theorem to I repeatedly yields the following:

n
n
βij (I ) = βij (S) + βij ((ha )) + βi−1 j ((hb ) ∩ Kb ). (1)
a=1 b=1
The minimal free resolution of R/(ha ) is
0 ← R ← R(− deg(ha )) ← 0 (2)
so its Betti table is trivial:

1 (i, j ) = (0, 0) or (1, deg(ha )),
βij ((ha )) =
0 otherwise.
The following theorem can be used to compute the Betti numbers of (hi ) ∩ Ki .
Theorem 4.4 Let I be a nearly complete intersection and h be a generator of
degree at least 3, so I = (h) + K is a Betti splitting of I . Then (h) ∩ K is of
the form hJ , where J is a complete intersection.
Proof If (m1 , . . . , mk ) is the minimal monomial generating set for K, then
(lcm(m1 , h), . . . , lcm(mk , h)) is a monomial generating set for (h) ∩ K. Moreover,
lcm(mi , h) = hmi for some square-free monomial mi coprime to h. Let J =
(m1 , . . . , mk ). We wish to show that J is a complete intersection by reinterpreting
its construction as operations on a hypergraph.
Let H = (V , E) be a hypergraph and define the hypergraph operation Lv (H )
to be Lv (H ) = (V , E ) where E = {e ∪ {v}|e ∈ E}. Then we may translate the
construction of J into hypergraph operations as follows:
Considering the monomial h as a formal variable corresponds to taking the join
of the hypergraph of K at h. Computing the ideal generated by lcm(mi , h) then
corresponds to computing Lh (H ) on the resulting hypergraph. Finally, factoring
out the h term is equivalent to inverting at h and then multiplying the resulting edge
ideal by h.
Hence, if H is the hypergraph of K, we get that the hypergraph of J is given
by (Ih ◦ Lh ◦ Joinh )(H ). But Ih = Min ◦ WeakIndV \{v} and WeakIndV \{h} ◦
Lh = WeakIndV \{h} . Hence, the hypergraph of J is given by (Ih ◦ Joinh )(H ). By
Lemma 3.4, the join of an NCI-hypergraph at a hyperedge yields an NCI-hypergraph
or a CI-hypergraph, and in either case inverting at any vertex in the support of
the resulting hypergraph yields a CI-hypergraph. Thus (Ih ◦ Joinh )(H ) is a CI-
hypergraph, so J is a complete intersection.

Since (h) ∩ K = hJ for some complete intersection J with monomial generators
f1 , f2 , . . . , fd , we can explicitly calculate the free resolution of R/ hJ . The first
syzygy is generated by hf1 , hf2 , . . . , hfd since the support of h and the support
of J are disjoint. The second syzygy is straightforward to calculate explicitly, and
it is exactly the second syzygy R/J . It follows that the remaining syzygies are the
106 C. Bondi et al.
corresponding Koszul relations of R/J , and thus the projective dimension of (h)∩K
is the number of minimal generators of hJ .
The skeleton S is much more complicated, and we note only that our decom-
position technique cannot be used to break it down further, as the principal ideal
generated by a 2-edge of I does not work with the splitting function defined in
Theorem 4.3. In this paper we will not prove results about the structure of βij (S).
Example 4.5 Consider the nearly complete intersection I given by the hypergraph
below (over the ring R = k[x1 , . . . , x8 ]). The hyperedges correspond to the
monomials H = {x1 x2 x3 , x4 x5 x6 }, the skeleton corresponds to the monomials in
S = {xi x7 , xi x8 , x7 x8 | 1 i 6}, and the ideal I is generated by the union S ∪ H .
1 4
2 5
3 6
With this setup, h1 = x1 x2 x3 , h2 = x4 x5 x6 , K1 = (S) + (h2 ), and K2 = (S). For

the first step of the decomposition in Equation 1, we write βij (I ) = βij (K1 ) +
βij (h1 ) + βi−1 j (K1 ∩ (h1 )) and compute the intersection ideal K1 ∩ (h1 ) =
h1 (x7 , x8 , h2 ). In the second step, we decompose βij (K1 ) = βij (S) + βij (h2 ) +
βi−1 j (K2 ∩ (h2 )) and find that K2 ∩ (h2 ) = K2 ∩ (h2 ) = h2 (x7 , x8 ). Both
intersection ideals are monomial multiples of complete intersections, as guaranteed
by Theorem 4.4, so four of the five Betti numbers in this decomposition are easy
to compute. Assuming we have computed β(K1 ) as in step 2, the component Betti
tables from step 1 sum to the Betti table for I as follows:
⎛ ⎞
13 42 70 70 42 14 2
⎜ 1 2 1 − − − −⎟
β(I ) = ⎜
⎝ − − − − − − −⎠
⎟ from β(K1 )
− − − − − − −
⎛ ⎞
−−−−−−−
⎜ 1 − − − − − −⎟
+⎜⎝− − − − − − −⎠
⎟ from β(h1 )
−−−−−−−
⎛ ⎞
−−−−−−−
⎜− 2 1 − − − −⎟
+⎜⎝− − − − − − −⎠
⎟ from the shift of β(K1 ∩ (h1 ))
− 1 2 1 −−−
⎛ ⎞
13 42 70 70 42 14 2
⎜2 4 2 − − − −⎟
=⎜
⎝−
⎟.
− − − − − −⎠
− 1 2 1 − − −
Note in particular that the projective dimension of every component other than
S in Example 4.5 is less than 4, while the dimension of the skeleton is 7. This
suggests that the projective dimension of a nearly complete intersection is primarily
determined by the skeleton of its corresponding hypergraph. The following theorem
captures some of this intuition:
Theorem 4.6 Let I = K + (h) be a nearly complete intersection with deg(h) 3,
and let pd denote the projective dimension of an ideal. Then pd(I ) pd(K) + 1.
Proof Suppose I = K + (h) is as above. Then K + (h) is a Betti splitting of I ,
which implies that
pd(I ) = max(pd(K), pd((h)), pd(K ∩ (h)) + 1)
We know that pd((h)) = 1, so it suffices to show that pd(K ∩ (h)) pd(K).

We accomplish this by demonstrating that pd(K ∩ (h)) = pd(K(x = 1)) for any
variable x dividing h and then noting that pd(K(x = 1)) cannot exceed pd(K).
By Theorem 4.4, K ∩ (h) = hJ where J is a complete intersection. Similarly,
K(x = 1) is a complete intersection. Since both ideals are complete intersections,
their minimal free resolutions are given by the Koszul complexes on their minimal
monomial generators. This implies in particular that pd(J ) is equal to the number
of (hyper)edges in G(J ) and pd(K(x = 1)) is equal to the number of (hyper)edges
in G(K(x = 1)).
By the proof of Theorem 4.4, we have that G(J ) = Ih (Joinh (G(K))), and
the proof of Lemma 3.4 implies further that G(J ) ∼ = Ix (IndV (G(K))) where
V = (V \h) ∪ {x} and V is the collection of vertices in G(K). Since IndV and
Ix commute by Lemma 2.9, we can write G(J ) ∼ = IndV (Ix (G(K))). Inverting
at x in G(K) kills all edges connected to every vertex of the hyperedge h because
N (x) = N (y) for every vertex y in h, so IndV (Ix (G(K))) has the same number of
(hyper)edges as Ix (G(K)). Thus G(J ) and G(K(x = 1)) have the same number
of (hyper)edges, implying that their projective dimensions are equal.
Since considering the ideal K(x = 1) is analogous to localizing K with respect
to the powers of x, it follows that pd(K(x = 1)) pd(K) and the proof is complete.

Although we only have that pd(K(x = 1)) pd(K) for any variable x belonging
to h, we conjecture that the inequality is always strict, which would imply that
pd(K + (h)) = pd(K). An important consequence of this conjecture is that
the projective dimension of a nearly complete intersection should be completely
determined by the skeleton of its hypergraph.
108 C. Bondi et al.
Not all homological properties of nearly complete intersections appear to be

as well-behaved as projective dimension, however. The regularity of these ideals,
for example, may depend on either the skeleton or the higher-degree terms. The
involvement of higher-degree terms is not surprising; the regularity of any monomial
ideal is bounded below by the maximal degree of its generators, and hence any
nearly complete intersection containing a k-edge must have regularity at least k.
More surprising perhaps is the fact that the skeleton of a nearly complete intersection
may have arbitrarily large regularity, as demonstrated by the following class of
examples generated in degree 2.
Theorem 4.7 Let Gn = (Vn , En ) where Vn = {c, a1 , b1 , . . . , an , bn } and E =
{(c, ai ), (c, bi ), (ai , bi )|1 i n}. Then reg(R/I (Gn )) = n.
For this proof, we use the notions of ind-match and min-match from [4].
Let G = (V , E) be a graph. We say that a matching of G is a set of edges
M ⊆ E so that no two edges in M share a vertex. The matching number of G,
denoted match(G), is the largest cardinality of a matching of G.
Now let H be a collection of graphs. An H-subgraph of G is a subgraph of G
whose connected components belong to H. An induced H-subgraph of G is an
H-subgraph which is an induced subgraph of G. An H-subgraph H = (V , E )
of G is called maximal if the subgraph induced on V \V contains no nonempty
H-subgraphs.
The authors of [4] define ind-match and min-match as follows:
ind-matchH (G) := max {match(H ) | H is an induced H-subgraph of G};

min-matchH (G) := min {match(H ) | H is a maximal H-subgraph of G}.
Proof Use the inequality
ind-match{K2 ,C5 } (G) reg(R/I (G)) min-match(G){K2 ,C5 }
given in [4, Corollary 3.9]. Let H = {K2 , C5 }. Since Gn contains no subgraphs

isomorphic to C5 , it follows that the only H-subgraphs of Gn are subgraphs
consisting of disjoint copies of K2 . The only maximal H-subgraphs of Gn are
H = (V , E) where V and E take one of the following 3 forms:
(1) V0 = {ai , bi | 1 i n} and E0 = {{ai , bi } | 1 i n}.
(2) Vbi = V0 \ {bi } ∪ {c} and Ebi = E0 \{{ai , bi }} ∪ {{ai , c}}.
(3) Vai = V0 \ {ai } ∪ {c} and Eai = E0 \{{ai , bi }} ∪ {{bi , c}}.
In any of these cases, the matching number of the maximal H-subgraph is n, so
min-match{K2 ,C5 } (Gn ) = n.
Similarly, the subgraph induced on {a1 , b1 , . . . , an , bn } has maximal matching
number among all induced H-subgraphs. Thus, reg(R/I (Gn )) = n. See Fig. 3 for a
graphical representation of G4 and its corresponding matching set.

Fig. 3 G4 with the

1 1
corresponding matching set
in red
4 2
4 2
3 3
5 Appendix: Vertex-Weighted Graphs
Theorem 3.1 characterizes NCI-hypergraphs as having an extremely simple hyper-

edge structure, a visual fact quantified by our discussion of the Betti tables of NCIs
in the body of the paper. In particular, NCI-hypergraphs must be joinable, with
hyperedges that resemble “fat vertices.” This notion naturally leads us to connect
joinable hypergraphs with vertex-weighted graphs.
Definition 5.1 Given a joinable hypergraph G, the join of G at a hyperedge h =
(v1 , . . . , vn ) is the hypergraph G = Joinh (G) defined by replacing all vertices in h
with a single vertex v of weight n. If all elements of h are connected to a vertex w
in G, then v is connected to w in G .
This is identical to the definition of join in the main paper, except we also label
the new vertex with a weight using the cardinality of the k-edge it came from. If we
take the join of G at all of its hyperedges, we are left with a unique vertex-weighted
graph. In fact, join defines a bijection between joinable hypergraphs and vertex-
weighted graphs, as can be seen by defining its inverse operation. Given a vertex-
weighted graph G, the splay of G is the hypergraph G obtained by replacing each
vertex v ∈ G of weight n > 1 with the n vertices v1 , . . . , vn . Each vi is connected
to every element of N(v), and the hyperedge e = (v1 , . . . , vn ) is included in G
as well. The definition of vertex inversion on a vertex-weighted graph is induced
directly from the definition of vertex inversion on the associated joinable hypergraph
via the join and splay operations (Fig. 4).
Consistent with [6], we define a CI-weighted graph to consist of isolated edges
between vertices of weight 1 and isolated vertices of weight greater than 1. It follows
immediately that the splay of a weighted graph G is a CI-hypergraph if and only if
G is a CI-weighted graph. This connection is enough for us to generalize the Miller-
Stone characterization [6] of NCI-graphs in degree 2 to weighted graphs.
Lemma 5.2 A weighted graph G on at least 3 vertices is an NCI-weighted graph if
and only if it satisfies the following properties:
110 C. Bondi et al.
Fig. 4 A joinable hypergraph with its corresponding vertex-weighted graph representation
• G does not contain the path on 4 vertices as a subgraph, where one end has
weight greater than 1;
• G satisfies the Miller-Stone condition when viewed as an unweighted graph.
NCI-weighted graphs and NCI-hypergraphs exist in close correspondence: the
splay of an NCI-weighted graph is an NCI-hypergraph, and the join of an NCI-
hypergraph is an NCI-weighted graph. Representing nearly complete intersections
as weighted graphs drastically reduces the number of edges involved, making it
easier to see underlying structures and properties.
The reader who likes to tinker with these ideas may enjoy our graph builder tool:
https://crgibbons.github.io/files/WeightedGraphBuilder.
References
1. Adam Boocher and James Seiner. Lower bounds for Betti numbers of monomial ideals. J.
Algebra, 508:445–460, 2018.
2. David A. Buchsbaum and David Eisenbud. Algebra structures for finite free resolutions, and
some structure theorems for ideals of codimension 3. Amer. J. Math., 99(3):447–485, 1977.
3. Shalom Eliahou and Michel Kervaire. Minimal resolutions of some monomial ideals. J. Algebra,
129(1):1–25, 1990.
4. S. A. Seyed Fakhari and S. Yassemi. Improved bounds for the regularity of edge ideals of graphs.
Collectanea Mathematica, 69(2):249–262, September 2017.
5. G. Fatabbi. On the resolution of ideals of fat points. J. Algebra, 242(1):92–108, 2001.
6. Branden Stone and Charles Miller. Classifying nearly complete intersection ideals generated in
degree two. https://arxiv.org/abs/2101.07901, 2020, Preprint.
7. Adam Van Tuyl. A beginner’s guide to edge and cover ideals. In Monomial ideals, computations
and applications, volume 2083 of Lecture Notes in Math., pages 63–94. Springer, Heidelberg,
2013.
8. Mark E. Walker. Total Betti numbers of modules of finite projective dimension. Ann. of Math.
(2), 186(2):641–646, 2017.
The Shape of Hilbert–Kunz Functions
C-Y. Jean Chan
Keywords Hilbert–Kunz function · Hilbert–Kunz multiplicity · Characteristic

p · Frobenius homomorphism · Representation ring · Divisor class group ·
Harder-Narasimhan filtration · Local Riemann–Roch formula · Cohen–Macaulay
cones · Affine semigroup ring · Conic divisor · Ehrhart’s theorem ·
Quasipolynomial
1 Motivation and Outline
In the 1960s, Kunz [62] introduced a function in order to study the regularity of
integral domains. Monsky named this function after Hilbert and Kunz in [72].
Despite its close resemblance with the usual Hilbert-Samuel functions, Hilbert–
Kunz functions, in many ways, behave very differently and are highly unpredictable.
The main aim of this article is to provide an overview of the development of
Hilbert–Kunz functions in the recent decades, and to link Hilbert–Kunz theory to
Ehrhart theory that may potentially provide accessible tools to investigate Hilbert–
Kunz functions for certain families of interesting rings and varieties.
Han and Monsky [48], Pardue [82]
In the 1990s, a series of works were done for algebraic curves by Han,
Monsky [48], Pardue [82], Buchweitz, Chen [19], and Monsky [73–75]. They
considered projective plane curves whose homogeneous coordinate rings are of
the form of k[x, y, z]/(g). Hypersurfaces in higher dimensions were considered
by Han and Monsky [47, 48], Chang [27], Chiang and Hung [29]. In addition,
Seibert [94] worked on the Cohen–Macaulay rings of finite representation type.
These studies revealed some unpredictable behaviors and mysterious nature of
Hilbert–Kunz functions. Starting from the early 1990s, they have been investigated
in waves of studies from different points of view with a variety of machinery.
C-Y. J. Chan ()

Department of Mathematics, Central Michigan University, Mt. Pleasant, MI, USA
e-mail: chan1cj@cmich.edu
112 C-Y. J. Chan
For instance, Han and Monsky developed the theory of representation rings that
provided a means to compute these functions for diagonal hypersurfaces. Brenner,
Frakhruddin and Trivedi studied the subject using sheaf theory [10, 11, 41] and
gave a systematic treatment for the case of smooth algebraic curves. Kurano
linked algebraic intersection theory with Hilbert–Kunz function and produced an
expression of the function in terms of local Chern characters [65, 66]. Bruns and
Watanabe offered a strong insight into Hilbert–Kunz function and multiplicity and
proved that the computations in the setting of normal affine semigroup rings can be
understood via Ehrhart theory [16, 112]. Nevertheless, it is in general difficult to
compute Hilbert–Kunz function and its associated multiplicity. However, in light of
Ehrhart theory and effective computer algebra systems developed in recent years, it
is likely that we will be able to formulate accessible questions regarding Hilbert–
Kunz functions.
In comparison, many more studies have been done for Hilbert–Kunz multiplicity
than its functional counterpart. The literature concerning Hilbert–Kunz multiplicity
alone is a rich entity, and it is beyond the scope of this manuscript to give a
comprehensive account. The discussions in this paper are mainly focused on the
functions. While making no attempt to be complete, we mention some results on the
multiplicity that are relevant to Hilbert–Kunz functions and link as many relevant
scholarly works as possible.
Valuable overviews of Hilbert–Kunz theory can be found in [13] by Brenner that
offers a very rich source for the subject, and in [57] by Huneke which provides
alternative approaches to many results different from the original proofs.
The outline of the paper is as follows. Sect. 2 introduces the definitions and a
brief history of the theory. While presenting interesting questions arising from the
literature, we also address how Hilbert–Kunz theory is related to other important
notions and studies in commutative algebra. Sect. 3 reviews some of the techniques
applied to the studies of Hilbert–Kunz functions. These include representation rings
and p-fractals [47, 48, 80, 81, 102], divisor class groups [24, 59], the cohomology
of vector bundles [10], intersection theory [65, 66], and cellular decompositions on
the fundamental domain [16]. We pay close attention to the key steps and extract
the crucial facts that contribute to establishing the targeted results. Readers may
skip the subsection regarding each technique and return to it as needed. We are
hopeful that our sketches may be helpful to those who wish to access the ideas in
these proofs. Sect. 4 is dedicated to normal affine semigroup rings. In this setting,
Hilbert–Kunz function is closely related to the lattice point enumerator as shown
first in Watanabe [112]. Bruns [16] builds a pathway that bridges the Hilbert–
Kunz and Ehrhart theories in a rigorous manner. This idea will be elaborated in
this section. Sect. 5 includes examples done by counting lattice points or estimating
with Macaulay2. Most examples presented in Sect. 5 and comments about the
techniques made throughout Sects. 3, 4, 5 are due to the author’s own studies and
observations. They can serve as starting points for further rigorous investigations.
The Shape of Hilbert–Kunz Functions 113
2 History in Brief
Throughout this paper, p denotes a prime number and e denotes a nonnegative

integer which is often the argument of functions under consideration. Let (R, m)
be a d-dimensional local ring of positive characteristic p. For any ideal I , let I [p ]
e
e
denote the ideal generated by elements of the form of a for all a ∈ I . Since R has
p
characteristic p, I [p ] can in fact be generated by the pe -th power of the elements in

e
any given generating set of I . The ideal I [p ] is called the pe -th Frobenius power of
e
I and is an m-primary ideal if I is m-primary. All modules in this paper are finitely
generated.
The Frobenius map f : R → R takes r ∈ R to f (r) = r p and is a ring
homomorphism. Naturally one may consider R as a module, denoted by 1R, over
itself via restriction of scalars along f . When R is reduced, f is injective so R
is isomorphic to its image R p . In that case, one can equivalently consider R as
a module over the subring R p which is how Kunz viewed it. Thus as an abelian
group, 1R is equal to R but its module structure is given by r · a = r p a for all r ∈ R
and a ∈ 1R. We say R is F -finite if the Frobenius map f is a finite morphism. (See
Miller [68] for a general background regarding the Frobenius endomorphism.)
In 1969, Kunz proved that the flatness of R as an R p -module characterizes its
regularity. Precisely, Kunz [62] proved that R is regular if and only if it is a reduced
flat module over R p . In fact his proof can be extended to show that the following
are all equivalent without the reduced condition: (1) R is regular; (2) the composed
homomorphism f e is flat for all positive integers e; (3) f e is flat for some positive
integer e (c.f. [61, Theorem 21.2]). Kunz achieved this by studying the numbers
(R/m[p ] ) as e increases and by applying Cohen’s structure theorem for complete

e
local rings. In particular, we have

Theorem 2.1 (Kunz [62]) Let R be a Noetherian local ring of dimension d. Then
R is regular if and only if
(R/m[p ] ) = (pe )d .
e
Let M be a finitely generated R-module and I an m-primary ideal. In 1983,

Monsky [72] called the function ϕM,I (e) : Z0 → Z
ϕM,I (e) : e −→
A (M/I [p ] M)
e
the Hilbert–Kunz function of M with respect to I . Although it depends on both M

and I , when there is no ambiguity, we will simply write it as ϕM (e). In particular,
by the Hilbert–Kunz function of the ring R, we mean M = R and I = m, the
maximal ideal. For a finitely generated R-module M of dimension d, Monsky also
considered the limit
ϕM,I (e)
eH K (M, I ) := lim (2.1)
e→∞ (p e )d
and proved the following results :

114 C-Y. J. Chan
Theorem 2.2 (Monsky [72]) Let (R, m) be a Noetherian local ring of positive
characteristic p and dim R = d. Let I be an m-primary ideal and M a finitely
generated R-module of dimension d. Then
(a) The limit in (2.1) exists and is always a positive real number.
(b) The Hilbert–Kunz function is of the form of
ϕM,I (e) = eH K (M, I ) (pe )d + O((pe )d−1 ).
(c) If R is a one-dimensional complete local domain and dim M = 1, then

ϕM,I (e) = eH K (M, I ) · pe + δe , where eH K (M, I ) is an integer and δe is
an eventually periodic function.
In this paper, we write g(x) = O(f (x)) if there is a constant C independent of
x such that |g(x)| C|f (x)| for x large enough (or x " 1).
Monsky named the limit in (2.1) the Hilbert–Kunz multiplicity of M with respect
to I . The limit is exactly the coefficient of the dominating term (pe )d in ϕM,I (e).
If I = m, we drop I from the notation, and simply write eH K (M) and ϕM (e). By
replacing R by R/annih(M), modulo the annihilator of M, we may always assume
that dim M = dim R and thus Theorem 2.2 still holds.
Next we recall the familiar Hilbert-Samuel functions which concerns the length
of M modulo the usual powers of an m-primary ideal I :
hM,I (n) :=
(M/I n M).
The Hilbert-Samuel multiplicity is defined as
hM,I (n)
i(M, I ) := d! lim .
n→∞ nd
There exists a polynomial pM,I (n) with rational coefficients so that hM,I (n) =
pM,I (n) for all large enough n. More precisely,
1
pM,I (n) = i(M, I ) nd + (lower degree terms in n).
d!
Also commonly known and studied are the Hilbert functions for graded rings or
modules, and Buchsbaum-Rim function which is the module version of Hilbert-
Samuel function via Rees rings (c.f. [18, 26, 87]). Similar to Hilbert-Samuel
function, Hilbert function and Buchsbaum-Rim function are polynomials of n for
large n, and their leading coefficients are always rational numbers.
However, unlike the typical Hilbert-type functions just mentioned, the behavior
of Hilbert–Kunz functions is rather unpredictable as seen in the next two exam-
ples. In Sect. 3, upon the consideration of Theorem 3.2, we will provide some
observations that set apart Hilbert–Kunz functions from Hilbert and Hilbert-Samuel
functions. These observations also show that it is often necessary to consider the
higher (co)homology modules when it comes to Hilbert–Kunz functions.
Example 2.3 (Kunz [62, Example 4.6(b)]) Set R = κ[[x, y]]/(y 4 − x 3 y) (cor-
responding to four different lines) where κ is an algebraically closed field. Then
ϕR (e) = 4pe − 3, if p ≡ 1 (mod 3); and ϕR (e) = 4pe + δe , if p ≡ −1 (mod 3)
where δe = −3 if e is even and −4 if e is odd.
Example 2.4 (Monsky [72, p. 46]) Set R = Z/p[[x, y]]/(x 5 − y 5 ) and p ≡ ±2
(mod 5). Then ϕR (e) = 5pe + δe where δe = −4 if e is even and −6 if e is odd.
There is no effective algorithm for computing Hilbert–Kunz multiplicity.
Hilbert–Kunz multiplicity was long thought to be rational, but this was proved
not to be the case (see [13]). In general, the question regarding the rationality of
Hilbert–Kunz multiplicity has attracted serious research efforts since the notion
was introduced in [72] and the debate was once a popular pastime for decades.
The question regarding how to effectively compute Hilbert–Kunz function remains
widely open.
Hilbert–Kunz function was initially defined for local rings. One may apply the
same definition for graded rings over a field κ of positive characteristic with respect
to their graded maximal ideals, or affine semigroup rings over such κ with respect
to the maximal ideals generated by elements in the semigroup. Which case we are
dealing with will be clear from the context in each section or example.
2.1 Multiplicity
Although this article focuses on Hilbert–Kunz functions, it would be incomplete

without touching upon some developments on the multiplicity.
By Theorem 2.1, being regular for a local ring R is equivalent to having a nice
Hilbert–Kunz function, which in turn implies eH K (R) = 1. Then Kunz asked if
the property eH K (R, m) = 1 is sufficient for R being regular. This was proved
by Watanabe and Yoshida [113] for unmixed local rings, and also by Huneke and
Yao [60] without using the tight closure techniques as done in [113].
A series of studies regarding various levels of singularities and the estimation of
lower bounds for Hilbert–Kunz multiplicities can be found in the works of Watan-
abe, Yoshida, Blickle, Enescu, Shimomoto, Aberbach, Celikbas, Dao, Huneke, and
Zhang [2, 3, 9, 22, 37, 114–116].
Another algebraic notion related to singularities is tight closure, introduced by
Hochster and Huneke [52, 53, 56]. The subjects of tight closure and Hilbert–Kunz
multiplicities share a close relationship. In fact, under some mild conditions, the
tight closure of an m-primary ideal I is the largest ideal containing I that has
the same Hilbert–Kunz multiplicity as I [53]. The introduction of [40] by Eto
and Yoshida offers a nice comparison between Hilbert–Kunz and Hilbert-Samuel
multiplicities. It describes a parallel relationship of integral closure to Hilbert-
Samuel multiplicity versus tight closure to Hilbert–Kunz multiplicity. The literature
116 C-Y. J. Chan
on the relationship between these two notions is extremely rich. Before leaving
this topic, we point out yet another connection between the two theories. By
understanding the semistability of vector bundles on projective curves, Brenner
brought out the geometric interpretation of tight closures and successfully solved
some open problems in the tight closure theory (c.f. [12]). This technique has
been proved to be effective also in Hilbert–Kunz theory which will be reviewed
in Sect. 3.3.
As we will see in the next section, it is sometimes natural to consider the Euler
characteristic after applying the Frobenius functor, and the Hilbert–Kunz function
can be expressed in terms of the Euler characteristic. By the singular Riemann–
Roch theorem, Kurano made a connection between the Hilbert–Kunz function with
local Chern characters and proved that the Hilbert–Kunz multiplicity characterizes
rings that are numerically equivalent to Roberts rings. (See Sect. 3.4 for numerical
equivalence.) In the 1960s, Serre defined the intersection multiplicity in terms of the
Tor-functors [95]. Several conjectures followed, some of which remain unsolved.
Roberts proved the vanishing conjecture for complete intersections and isolated
singularities by using intersection theory [85]. Independently, this was also proved
by Gillet and Soulé using K-theory [45, 46]. Kurano called a ring R a Roberts ring
if the only nontrivial component in the Todd class of R belongs to the subgroup
of codimension zero in the Chow group [64]. This condition is satisfied by rings
for which the vanishing theorem holds in Roberts’ result. If R is further assumed
to be a homomorphic image of a regular local ring and also Cohen–Macaulay,
then it is numerically equivalent to a Roberts ring if and only if the Hilbert–Kunz
multiplicity satisfies the condition eH K (I ) =
(R/I ) for any m-primary ideal I of
finite projective dimension [65].
Going back to the mystery of the multiplicity, Monsky initially suspected
that eH K (R) should always be rational [72]. Even though all the examples with
explicitly known Hilbert–Kunz multiplicities do take on rational values, the question
itself remained stubbornly unresolved. Later, Monsky [78] conjectured otherwise.
Eventually Brenner [13] constructed a module that has an irrational multiplicity
leading to the existence of three dimensional rings with irrational Hilbert–Kunz
multiplicity.
Still it is natural to ask under what conditions or for what family of rings the
Hilbert–Kunz multiplicities are rational. Some of the known cases include
(1) Regular local rings (eH K (R) = 1, Kunz [62]).
(2) Complete local domains of dimension one (eH K (R) ∈ Z, Monsky [72]).
(3) Algebraic curves—two-dimensional standard graded algebra over an alge-
braically closed field:
• R is normal (Brenner [11] and Trivedi [103] independently).
• R = k[x, y, z]/(f ) for plane cubics (deg f = 3) (Monsky [73, 76, 77, 79],
Buchweitz and Chen [19], Pardue [82]).
(4) Hypersurfaces
• Diagonal hypersurfaces (Han and Monsky [47, 48])
• Some special families (Kunz [63]; Monsky and Teixeira [81, 102])
(5) F -finite Cohen–Macaulay rings of finite Cohen–Macaulay type (Seibert [94]).

(6) Full flag varieties and elliptic curves (Fakhruddin and Trivedi [41]).
(7) Stanley-Reisner rings and binomial hypersurfaces (Conca [30]).
(8) Segre product of polynomial rings (Eto and Yoshida [40]).
(9) Segre product of any finite number of projective curves (Trivedi [107]).
(10) Normal affine semigroup rings (Watanabe [112]).
In [112], the value
(R/I [p ] ) is obtained by counting the lattice points within a
e
certain relevant region defined by a normal affine semigroup. This approach inspires
our discussions in Sects. 4 and 5 for the Hilbert–Kunz functions.
It would be interesting to know what other families, especially in higher dimen-
sions, produce rational eH K (R). More generally, what are the hidden conditions or
properties common to these examples?
2.2 Functions
From now on, we will use q to denote pe unless we want to emphasize its explicit
dependence on e. If ϕ(e) is expressed as a function in pe , we write (q) to highlight
this feature, namely, ϕ(e) = (q). We will also use α in place of eH K . Hilbert–
Kunz functions have the following functional form for q " 1:
By Theorem 2.1 (Kunz [62]), if R is a regular local ring of dimension d, then
ϕR,m (e) = ϕR (e) = R (q) = q d . (2.2)
By Theorem 2.2(b) (Monsky [72]), for an arbitrary Noetherian local ring R and
finitely generated R-module M with respect to ideal I ,
ϕM,I (e) = M,I (q) = α q d + O(q d−1 ). (2.3)
Huneke, McDermott and Monsky [59] refines (2.3) for an excellent normal
domain R with perfect residue field to the following form which identifies the next
order term
ϕM,I (e) = M,I (q) = α q d + β q d−1 + O(q d−2 ), (2.4)
where β is called the second coefficient. It is written as βI (M) if we wish to

emphasize that β is a function of M with respect to I or β(M) if I is the maximal
ideal. The normality condition on the ring was further relaxed by Kurano and the
author [24], and independently Hochster and Yao [55].
As noted in [59], the third coefficient in (2.4) in general does not exist.
For example, as computed by Han and Monsky [48], the domain R =
Z/5Z[x1 , x2 , x3 , x4 ]/(x14 + x24 + x34 + x44 ) has ϕR (e) = 168
61 (5 ) − 61 (3 ). The end
n 3 107 n
61 (3 ) is O(5 ) as expected in (2.4) but is not in the form of c(5 ) + O(1)

term − 107 n n n
118 C-Y. J. Chan
with some constant c. In Sect. 3.2, we will point out that the normality condition
(R1)+(S2) can be loosened to be (R1 ), a condition similar to (R1). However, as
seen in Examples 2.3 and 2.4, this new condition cannot be further relaxed.
In the earlier studies of Hilbert–Kunz theory, the entire function was computed
in many cases. These include certain projective plane curves and hypersurfaces in
the work of Kunz, Han, Monsky, Pardue, Buchweitz, and Chen [19, 30, 48, 63, 73–
75, 82], Stanley-Reisner rings by Conca [30], and Cohen–Macaulay rings of finite
representation types by Seibert [94]. The outcome of these investigations shows that
Hilbert–Kunz functions are periodic in some cases but not always.
Despite this evidence, it is still interesting to ask how close the Hilbert–Kunz
functions are to the form of a polynomial in q, or the next best possibility, a
quasipolynomial in q. By a quasipolynomial, we mean a function in the form of
a polynomial whose coefficients are periodic functions. The functions obtained in
Examples 2.3 and 2.4 are quasipolynomials. Some cases have been studied and com-
puted, for instance Conca [30], Seibert [93], Miller, Robinson and Swanson [70, 89],
but a systematic treatment is still lacking. This will be discussed in Sects. 4 and 5
for normal affine semigroup rings.
Before moving on to the next section for some established techniques applied
in the study of Hilbert–Kunz functions, we briefly mention some more recent
developments related to Hilbert–Kunz theory.
In recent years, Hilbert–Kunz multiplicity has been generalized to be taken with
respect to ideals that are not primary to the maximal ideal. This is done by using
the length of the local cohomology at the maximal ideal. This notion, known as the
generalized Hilbert–Kunz multiplicity, was proved to exist and further developed
by Epstein and Yao [38], Hernández and Jeffries [50], and Dao and Smirnov [34]. It
should be noted that this is different from another generalized version for monomial
ideals, primary to a maximal ideal, considered by Conca, Miller, Robinson and
Swanson [30, 70, 89] (see Sect. 3.6).
A potential extension of Hilbert–Kunz multiplicity to the characteristic zero set-
ting has been proposed. In this regard, limit Hilbert–Kunz multiplicity is considered.
When the limit exists, its uniformity is also of interest; see Application 3.5 for more
details and relevant references.
It is also known, from the work of Huneke and Leuschke [58], Singh [98],
Yao [118], and Aberbach [1], that the theory of F -signature is closely related to that
of Hilbert–Kunz multiplicity. In particular, it is proved by Tucker, Polstra, Caminata
and De Stefani [21, 83, 110] that F -signature functions have the same approximate
functional forms as Hilbert–Kunz functions. In the case of affine semigroup rings,
the work of Watanabe [112], Bruns [16], Singh [98], and Von Korff [111] show that
the same effective methods can be used to compute these two sets of functions and
multiplicities. We will present the techniques of [16] in Sect. 3.5.
3 Techniques in Hilbert–Kunz Functions
In this section, we review some of the techniques that have been applied repeatedly
in investigating Hilbert–Kunz functions. In doing so, we wish to address the depth
of Hilbert–Kunz theory, and to make the technical proofs more accessible. The
review here should be taken as complementary reading to the original proofs. Our
summaries are not meant to replace them.
To keep the paper self-contained, we will define the terms commonly used in
this paper to avoid confusion and ambiguity. However, for the definitions of those
notions that require separate background, we will refer to appropriate references so
that the discussions remain focused on the key ideas. In these cases, we provide a
means of processing them to help the readers move forward.
Unexplained notation and terminology can be looked up in the following
references: Cutkosky [33], Hartshorne [49] and Shafarevich [96, 97] for algebraic
geometry; Fulton [43] and Roberts [87] for intersection theory. For an algebraic
approach to locally free sheaves as modules and Chern characters, we recommend
Part II, especially Chap. 9, of [87]. An affine semigroup ring can sometimes be
viewed as a subalgebra generated by finitely many monomials of a polynomial ring
over a field, but we highly recommend some basic knowledge in their connection
to toric varieties as presented in Cox et al. [32]. Stanley [101] and Miller and
Sturmfels [69] are excellent sources for combinatorics theory.
Whenever possible, we adopt the notation from the original papers from which
we are presenting the techniques. There may be some overlapping notation. But we
believe the confusion is minimal as they appear in different contexts.
We begin by presenting two functors arising from the Frobenius map.
Let R be a Noetherian local ring of positive characteristic p with perfect residue
field. Recall that the Frobenius homomorphism is the map f : R → R, f (r) =
r p . The Frobenius homomorphism induces two functors. One is the functor by
restriction of scalars:
M → 1M
where 1M is the module over R via f . Precisely, for any m ∈ M, if (m)1 denotes
the element m considered as an element in 1M, then r · (m)1 = (r p m)1 for any
r ∈ R. The functor by the restriction of scalars M → 1M is an exact functor (c.f.
[87, Sect. 7.3]). In what follows, we assume that R has a perfect residue field and
that 1R is a finitely generated R-module, that is, R is F -finite. Note that the ring R
is F -finite if R is a complete local ring with perfect residue field or a localization
of a ring of finite type over a perfect field. If R is F -finite, the rank of 1R is pd if
rank is defined (e.g. R is a domain). If R is regular, then by Kunz’s theorem, 1R is
flat so it is also free. For any positive integer e, the module eM is obtained by the
self-composition of the functor; or equivalently, eM is the module over R via the
composite function f e = f ◦ · · · ◦ f .
120 C-Y. J. Chan
The other functor is the Frobenius functor F (also known as the Peskine-Szpiro
functor) from the category of left R-modules to itself via tensor product over f ,
F(M) := 1R ⊗R M.
This is considered as an extension of scalars along f . The resulting R-module

structure on F(M) is via the left factor. More precisely, for any a, r ∈ R and m ∈ M,
a(r ⊗m) = ar ⊗m and 1⊗rm = r p ⊗m. As left R-modules, F(R) ∼ = R. Therefore,
F(R/I ) = 1 R ⊗R R/I = R/f (I ) = R/I [p]
with its ordinary left scalar multiplication. One can iterate this process and see that
Fe (R) = R and Fe (R/I ) = R/I [p ] .
e
Remark 3.1 Due to commutativity, 1R ⊗R M = ∼ M ⊗R 1R, but the module structure

of M ⊗R R is via the right factor. That is, for any a, r ∈ R and m ∈ M, a(m ⊗ r) =
1
m ⊗ ra and mr ⊗ 1 = m ⊗ r p .
The Peskine-Szipiro functor is covariant, additive, and right exact from the
category of left R-modules to itself. When F is applied to a complex of free modules
of finite ranks, all free modules remain the same up to isomorphism since tensor
product commutes with direct sum, but the entries of the matrices of the boundary
maps are raised to the p-th power. A notable theorem by Peskine and Szpiro states
that if G• is a finite complex of finitely generated free R-modules, then G• is acyclic
if and only if F(G• ) is acyclic (c.f. [18, Theorem 8.7]).
Next we present an analogous Hilbert function for Frobenius powers. Its form is
proved by induction on dimension similarly to the case of usual Hilbert functions,
but the induction steps are more complicated (c.f. Seibert [93] and Roberts [87]).
Let C be a category of finitely generated R-modules of dimension no bigger than
a nonnegative integer n. We say that C satisfies the property (†) if whenever
M , M, M in C form a short exact sequence 0 → M → M → M → 0, then M
is in C if and only if M and M are in C. The property (†) implies that M ∈ C if
and only if R/p ∈ C for all prime ideals in the support of M. Since M and 1 M have
the same support, it follows that if M ∈ C, then 1M ∈ C.
Theorem 3.2 ([93] [87, Theorems 7.3.2 & 7.3.3]) Let C be a category of finitely
generated R-modules of dimension no bigger than a nonnegative integer n. Assume
that C satisfies the property (†). Let g be a function from C to Z.
(a) If g is additive on short exact sequences, then for any M ∈ C, there exists a
polynomial in pe with rational coefficients a0 , . . . , an such that
g(e M) = an (pe )n + an−1 (pe )n−1 + · · · + a0
for all integer e 0.

(b) If g(M) g(M ) + g(M ) whenever 0 → M → M → M → 0 is exact in
C, then there exists a real number c(M) such that
g(e M) = c(M)(pe )n + O((pe )n−1 )
for all integers e 0.

In Seibert [93] and Roberts [87, Sect. 7.3], one can find interesting applications
of Theorem 3.2 to the existence of Hilbert–Kunz multiplicity originally proved by
Monsky [72] and the asymptotic Euler characteristic χ∞ defined by Dutta [35]. The
latter is now known as Dutta multiplicity.
Now we provide some elementary observations based on Theorem 3.2(a). Let
I be an m-primary ideal with a finite free resolution. Then R/I has a finite free
resolution G• , i.e., G• → R/I → 0 is exact. Now let C be the category of finitely
generated R-modules. The Euler characteristic χG• of a module M ∈ C is defined
to be the following alternating sum of lengths of the homology modules

χG• (M) := (−1)i
(Hi (G• ⊗R M).
i
The Euler characteristic is additive on short exact sequences. Since restriction of

scalars along f is exact, therefore, by Theorem 3.2(a), χG• (e M)) is a polynomial
in pe with rational coefficients.
We observe that
G• ⊗R eM ∼
= G• ⊗R e(R ⊗R M) ∼
= (G• ⊗R eR) ⊗R M ∼
= Fe (G• ) ⊗R M.
The last equality holds for G• ⊗R eR ∼

= eR ⊗R G• = Fe (G• ) by Remark 3.1. Upon
taking M = R, we have that

χG• (e R) = i (−1)i
(Hi (Fe (G• )))
=
(H0 (Fe (G• ))) (3.1)
=
(R/I [p ] ) = ϕR,I (e).
n
In the display above, the second equality holds since F preserves acyclicity of finite
free complexes due to the result of Peskine and Szpiro mentioned above. Hence the
higher homology modules all vanish. The third equality is due to the fact that F is
right exact so H0 (Fe (G• ) = Fe (R/I ). Applying Theorem 3.2(a) to χG• in (3.1),
we obtain the following corollary for the Hilbert–Kunz function.
Corollary 3.3 Let (R, m) be a local ring of characteristic p with perfect residue
field and I an m-primary ideal. Assume that I has finite projective dimension. Then
ϕR,I (e) is a polynomial in pe with rational coefficients.
For ϕM,I (e), if M is a finitely generated flat R-module, then by a parallel
argument as in (3.1), we have χG• (e M) = ϕM,I (e). But a finitely generated flat
module over a local ring is free. So ϕM,I (e) is just a multiple of ϕR,I (e).
Corollary 3.3 imposes a strong condition: I has finite projective dimension.
Commonly interesting ideals in any given ring R do not necessarily have finite
122 C-Y. J. Chan
projective dimension unless R is regular. If R is regular, then by Corollary 3.3,

we have ϕR,I (e) = χG• (Fe (R)) = ad (pe )d + ad−1 (pe )d−1 + · · · + a0 for any m-
primary ideal I . As with classical Hilbert polynomials, an immediate challenge is
how one can identify the coefficients. Suppose we can use anything at our disposal,
then we can jump directly to Theorem 3.10 which shows that each coefficient ai is
determined by a certain class in the Chow group Ai (R). But the Chow group of a
regular ring has only the top piece, namely Ai (R) = 0 for i = d − 1, . . . , 0. Hence
ϕR,I (e) = ad (pe )d and indeed ad =
(R/I ) by [65, Theorem 6.4], recovering
Kunz’s original theorem when I = m. This is a long detour back to where we
started. We note that even if Hilbert–Kunz function is eventually a polynomial in
pe , determining the values of the coefficients is a very difficult task.
In general for an arbitrary finitely generated module M, the higher homology
modules in

χG• (e M) = (−1)i
(Hi (Fe (G• )) ⊗R M) (3.2)
i
do not always vanish. In such a case, one has to study and control the higher
homology modules in order to properly describe the desired length of the zeroth
homology that equals ϕM,I (e). Sections 3.2 and 3.3 will demonstrate how this can
be done in their respective cases.
For an estimate of the length of higher homology modules in a latter discussion,
we quote a theorem from [59], which can be traced back to [54, 86, 93] and later
strengthened by [28]. Let d = dim M. It states that if G• : 0 → Gn → · · · →
G0 → 0 is a finite complex of free modules such that each homology module
Hi (G• ) has finite length, then for e 0,

(Hn−t (Fe (G• ) ⊗R M)) = O((pe )min(d,t) ). (3.3)
In what follows, we will explore how each path leads us to approximate or

compute Hilbert–Kunz functions.
3.1 Via Representation Rings and p-Fractals
In this subsection, we describe representation rings and their connections to Hilbert–

Kunz theory. Then we briefly describe the next set of activity that it led to, namely
the highly technical theory of p-fractals. At the end we mention a conjecture about
the behavior of Hilbert–Kunz multiplicity when the characteristic p increases.
Pioneered by Example 4.3 in Kunz [63] and Monsky’s initial paper [72], a
series of studies on the Hilbert–Kunz function of hypersurfaces followed in the
1990s. These include works by Buchweitz, Chang, Chen, Han, Monsky and Par-
due [19, 48, 73–75, 82] and for Cohen–Macaulay rings of finite representation type
by Seibert [94]. Especially remarkable is the paper “Some surprising Hilbert–Kunz
functions” by Han and Monsky [48]. It introduced the idea of using representation
rings in the setting of the affine coordinate rings of the diagonal hypersurfaces,

κ[x1 , . . . , xs ]/( i xidi ), which became the first systematic method for computing
Hilbert–Kunz functions. It also identified these rings as the first known family whose
Hilbert–Kunz functions are eventually periodic. For the detailed structure of this
special representation ring, we refer to [47, 48, 102] but we describe the basics here.
Out of this grew the method of p-fractals of Monsky and Teixeira for handling other
hypersurfaces, which we mention subsequently; for further details see [80, 102].
Representation Rings Here we describe the relevant representation rings and how
they apply to computing Hilbert–Kunz functions. Let κ[T ] be a polynomial ring in
one variable and C the category of finitely generated modules over κ[T ] annihilated
by a power of T . Then C satisfies the property (†) defined previously, namely, for
any short exact sequence
0 → M → M → M → 0,
M is in C if and only if M and M are in C . We further define the Grothendieck

group K0 (C ) of C as the free abelian group generated by the isomorphism classes
[M] of objects M in C modulo the subgroup generated by [M] − [M ] − [M ] for
any M, M , M satisfying the above short exact sequence.
Let [M] and [N] represent two classes in K0 (C ). Consider the tensor product of
M and N over κ. We define an action of T on M ⊗κ N by T (m ⊗ n) = T (m) ⊗ n +
m ⊗ T (n) for any m ∈ M and n ∈ N. Thus M ⊗κ N is in C . It can be verified that
K0 (C ) is a commutative ring with the binary operations + and · induced by ⊕ and ⊗
respectively, and that the classes of the zero module and κ[T ]/(T ) are the respective
additive and multiplicative identities ([48, Theorem 1.5]). The Grothendieck group
K0 (C ) with this commutative ring structure is called the representation ring in [48],
where it is denoted by .
Since every module M in C is finitely generated and annihilated by a power of T ,
by the structure theorem for modules over a P.I.D., M can be decomposed uniquely
up to isomorphism into a finite direct sum as M ∼ = ⊕j κ[T ]/(T nj ). Let δj denote
the class represented by κ[T ]/(T ) in . Then {δj }∞
j
j =1 forms a basis for as an
abelian group. Moreover, one can define a map α : → Z that takes [M] to the
number of indecomposable summands in the unique decomposition just mentioned.
The definition of α here is equivalent to that in [48] where for technical reasons α is
defined via a different basis λj = (−1)j (δj +1 − δj ).
We observe that if V = κ[T ]/(T j ), then dimκ (V /T V ) = 1. Since M ∼ =
⊕j κ[T ]/(T nj ) is a unique decomposition, and α([M]) equals the minimal number
of generators of M as a κ[T ]-module, we have α([M]) = dimκ (M/T M).
Now we may present the connection between the representation ring and the
Hilbert–Kunz function of a diagonal hypersurface, i.e., a ring of the form
R = κ[x1 , . . . , xs ]/(x1d1 + · · · + xsds ).

124 C-Y. J. Chan
Without loss of generality, we assume di > 0. First we take q as an integer larger

q
than all the di ’s and view Mi = k[xi ]/(xi ) as a κ[T ]-module with T acting as
di
multiplication by xi . As a κ-vector space,
Mi = κ · 1 + κ · xi + κ · xi2 + · · · + κxidi −1 + κ · xidi + κ · xidi +1 + · · · + κ · xi

q−1
.
Then, there are invariant κ[T ]-submodules generated by 1, xi , xi2 , . . . , xidi −1 respec-
tively. Write q = ki di + ai with 0 ai < di . The invariant submodule generated
by xic , with c = 0, . . . , ai − 1 are annihilated by T ki +1 but no smaller power. This
is true because T ki · xic = xiki di · xic = 0 in Mi if c < ai , but T ki +1 · xic =
k d +a +(d −a )+c
= xi · xid−ai +c = 0 in Mi and 0 < di − ai di − ai + c
q
xi i i i i i
di − 1 if 0 c ai − 1. Similarly, the invariant submodules generated by xic ,
c = ai , . . . , di − 1, are annihilated by T ki but no smaller power. Hence as a κ[T ]-
module, Mi can be decomposed into
8 9ai 8 9di −ai
Mi ∼
= κ[T ]/(T ki +1 ) ⊕ κ[T ]/(T ki ) .
Hence we can conclude that [Mi ] = ai δki +1 + (di − ai )δki in .

Next we assume that κ is a field of positive characteristic p and recall that q =
pe . Observe that
M1 ⊗ κ · · · ⊗ κ Ms ∼
= k[x1 , · · · , xs ]/(x1 , . . . , xs ) = k[x1 , · · · , xs ]/m[q] ,
q q
where m is the ideal (x1 , . . . , xs ). By the definition of the action T on each Mi and
the multiplication structure in , M1 ⊗κ · · ·⊗κ Ms is a κ[T ]-module where T acts by
multiplication by x1d1 +· · ·+xsds . On the other hand, if we let M denote M1 ⊗κ · · ·⊗κ
Ms , then M/T M is obviously a module over R = k[x1 , · · · , xs ]/(x1d1 + · · · + xsds ).
Moreover, as R-modules,
M/T M ∼
= k[x1 , · · · , xs ]/(x1 , . . . , xs , x1d + · · · + xsds ) ∼
= R/m[q] ,
q q
and
R (R/m[q] ) = dimκ (M/T M). As noted earlier, α([M]) = dimκ (M/T M).
Hence the Hilbert–Kunz function can be obtained via α as

R (R/m[q] ) = dimκ (M/T M) = α([M]) = α([M1 ⊗κ · · · ⊗κ Ms ])
831 ] · · · [Ms ])
= α([M 9 (3.4)
d
=α i=1 (ai δki +1 + (di − ai )δki ) .
Han and Monsky developed an intricate combinatorial method of computing

the indecomposable summands in the decomposition of tensor products of basis
elements (i.e., products of δj ’s) in to calculate the value of α in the last equality in
(3.4). This yields a method for computing Hilbert–Kunz functions. Using this, they
also proved that the Hilbert–Kunz function of the rings in this family is eventually
periodic and that the multiplicity is a rational number [48, Theorem 5.7].
We note that the eventually periodic functions of the family in this subsection are
not in the form of quasipolynomials as will be discussed in Sect. 4. This can be seen
from the following examples in [48].
Example 3.4 (Han-Monsky [48, p. 135])

(1) R = κ[x1 , x2 , x3 , x4 ]/( i xi4 ) where κ = Z/5Z, then ϕR = 168
61 · 5 − 61 · 3 .
3e 107 e
2
(2) R = κ[x1 , . . . , x5 ]/( i xi ) where κ = Z/3Z, then ϕR = 19 · 3 − 19 · 5 .
23 4e 4 e
Using the technique developed by Han and Monsky [48], Chiang and
Hung [29] extended the result from diagonal hypersurfaces to rings of the form
m
κ[x1 , . . . , xs ]/(g) where g = di
i=1 (Xi ) and Xi is a product of elements of a
subset of {x1 , . . . , xs } such that at least one Xi is a single variable.
p-Fractals Building on the theory of representation rings, Monsky and Teix-
eira [80, 81, 102] develop the theory of p-fractals which takes them beyond diagonal
hypersurfaces. The theory of p-fractals provides a means to understand and allows
the computations of Hilbert–Kunz series in may situations. Here we give a brief
overview of the results, but do not delve into this method due to its technical
complexity.
First, recall that Hilbert–Kunz series is the generating function of Hilbert–Kunz
function

HKSR (t) = ϕR (e) · t e .
e
Seibert [94] first proves that if R is Cohen–Macaulay of finite Cohen–Macaulay

type, and R is F -finite (a finitely generated module via the Frobenius morphism),
then HKSR (t) is rational, i.e., a quotient of two polynomials, and that eH K (R) is a
rational number. Hilbert–Kunz series has also been studied in [76, 81, 102].
The combined techniques of representation rings and p-fractals enable Monsky
and Teixeira to prove the rationality of Hilbert–Kunz series for a large family
of rings of the form of a quotient of a power series ring by a principal ideal:
κ[[x1 , . . . , xs ]]/(g) for a finite field |κ| and certain power series g [81, Theorem 4.4].
Using this, in some examples, the Hilbert–Kunz multiplicities can be calculated
and appear to be rational numbers [81, Section 5]. Furthermore, the Hilbert–Kunz
functions are eventually periodic for the case s = 3, g = x3D − h(x1 , x2 ) and h a
nonzero element in the maximal ideal of κ[[x1 , x2 ]] [76, Theorem 6.11]. (See also
the ending comment in [81, p. 255] regarding the finiteness condition on the field κ
and the vanishing of the (p e )-term of the Hilbert–Kunz function. We will address
these interesting points in Sect. 3.3 after the proof of Theorem 3.8 and in Sect. 3.4,
respectively.) In [76, Theorem, p. 351], Monsky gives a concrete description of the
Hilbert–Kunz function for the case g = x3D − x1 x2 (x1 + x2 )(x1 + λx2 ), for λ = 0
in κ, and p ≡ ±1 (mod D).
126 C-Y. J. Chan
Next we present an attempt to define Hilbert–Kunz multiplicity in the character-

istic zero case and report some concrete progress based on Han-Monsky’s technique
of representation rings.
Application 3.5 The notion of limit Hilbert–Kunz multiplicity arises by reducing a
ring of characteristic zero modulo primes p. Precisely, if R is a Z-algebra essentially
of finite type over Z and I is an ideal, let Rp be a reduction of R mod p and Ip the
extended ideal. If
(Rp /Ip ) is finite and nonzero for almost all p, then one can
define
∞
eH K (I, R) := lim eH K (Ip , Rp )
p→∞
if this limit exists. The above is called the limit Hilbert–Kunz multiplicity of I .
The existence of the limit is not known except in a few cases. These include when
eH K (Ip , Rp ) is constant for almost all p as in [10, 19, 30, 39, 73, 82, 112, 114], and
when non-constant, for homogeneous coordinate rings of smooth projective curves
by Trivedi [104] and for diagonal hypersurfaces by Gessel and Monsky in [44]. The
example below illustrates the last case.
Based on Han and Monsky’s scheme, Chang [27] computes precisely the
Hilbert–Kunz function of the ring R = κ[x1 , x2 , x3 , x4 ]/(x14 + x24 + x34 + x44 )
with respect to the maximal ideal (x1 , . . . , x4 ) for κ = Z/pZ. The Hilbert–Kunz
multiplicity in this case is also calculated in Gessel and Monsky [44] to be

8 2p 2 + 2p + 3 8 2p 2 − 2p + 3
eH K (R) = , p ≡ 1 (mod 4); and , p ≡ 3 (mod 4)
3 2p 2 + 2p + 1 3 2p 2 − 2p + 1
In this example, eH K (R) obviously depends on the characteristic p and yet one has

∞ 8 2p2 ± 2p + 3 8
eH K (m, R) = lim =
p→∞ 3 2p2 ± 2p + 1 3
which is independent of p.
Assuming that eH ∞ (I, R) exists, one can also ask for any fixed e 1 if the
K
following equality holds,
[pe ]
∞
(Rp /Ip )
eH K (I, R) = lim . (3.5)
p→∞ (pe )d
Brenner et al. [15] provide an affirmative answer for the case of homogeneous
coordinate rings of smooth projective curves. Their proof is based on the formula
of the Hilbert–Kunz multiplicity provided by the Harder-Narasimhan filtrations as
described in Sect. 3.3 and the approach in [104] which handles the case of larger
values of e. Then building on the proof of [44], it is proved in [15] that the above
question has affirmative answer for coordinate rings of diagonal hypersurfaces.
More recent developments can be found in [71, 84, 99, 106, 108, 109].
3.2 Via the Divisor Class Group
Assume that (R, m) is an excellent normal local domain of dim R = d with a

perfect residue field. Here we sketch the main ideas of Huneke, McDermott and
Monsky [59] in proving that the Hilbert–Kunz function of M with respect to an
m-primary ideal I has the following form as in (2.4)
ϕM,I (e) = α (pe )d + β (pe )d−1 + O((pe )d−2 ). (3.6)
We remark that in general the d in (3.6) can be replaced by dim M as noted in

the statement after Theorem 2.2. For the purpose of discussions in this subsection,
we fix d to be dim R. Thus the α in (3.6) is zero for those M with dim M < d. And
similarly, β = 0 if dim M < d − 1.
Below we state two key lemmas. Lemma 3.6 is from [59] and is a consequence
of (3.3) that estimates the length of higher homology modules mentioned after
Corollary 3.3. Lemma 3.7 is implicit in [59] and is a crucial fact regarding the
convergence rate of a sequence.
Lemma 3.6 ([59, Lemma 1.1]) Let R be a local ring of positive characteristic p
and I and m-primary ideal of R. If T is a finitely generated torsion R-module with
dim T = u, then
(TorR [pe ] , T )) = O((p e )u ).
1 (R/I
Lemma 3.7 Let s > t be fixed positive integers. Let {ηe }∞ e=1 be a sequence of real
numbers that satisfies a recurrent relation ηe+1 = ps ηe + O((pe )t ) for e " 1. Then
there exists a real number a such that ηe = a · (pe )s + O((pe )t ) for e " 1.
ηe
Proof Set ρe = (pe )s . Then by straightforward computation, one sees ρe+1 = ρe +
O(( p1e )s−t ) which means |ρe+1 − ρe | C · ( p1e )s−t for some constant C and e " 0.
Let m > n be integers. Then
|ρm − ρn | |ρm − ρm−1 | + · · · + |ρn+2 − ρn+1 | + |ρn+1 − ρn |.

C[( pm−1
1
)s−t + · · · + ( pn+1 ) + ( p1n )s−t ]
1 s−t
= C · ( ps−t
1 n
) [1 + ( ps−t
1
) + · · · + ( ps−t
1 m−n−1
) ]
s−t
= C pps−t −1 [( ps−t
1 n
) − ( ps−t
1 m
) ]
which can be made as small as possible for large enough m, n. Hence {ρe } is a
Cauchy sequence and thus it converges to some real number a. Furthermore, if we
fix n = e " 1, then
ps−t 1 1 ps−t 1 e
lim |ρm − ρe | lim C [( s−t )e − ( s−t )m ] = C s−t ( ) .
m→∞ m→∞ ps−t −1 p p p − 1 ps−t
Note that lim |ρm − ρe | = |a − ρe |. Hence |a − ρe | O(( ps−t

1 e
) ) which implies
m→∞
ηe = a(pe )s + O((pe )t ) as desired.

128 C-Y. J. Chan
The work in [59] shows that the value of the second coefficient β is intimately
related to the divisor class of the module, denoted by c(·). The proof of (3.6) goes
through the following three main steps. Step 1 deals with torsion free modules and
compares them with free modules of the same rank. Step 2 applies the outcome of
Step 1 to 1R to obtain ϕR,I (e). Step 3 reduces arbitrary modules to torsion free ones
and then applies Steps 1, 2 and Lemma 3.6 to obtain ϕM,I (e) for arbitrary M.
Step 1 We focus on torsion free modules and prove that if M has rank r, then
there exists a real constant τ (M) such that ϕM,I (e) = r ϕR,I (e) + τ (M)(pe )d−1 +
O((pe )d−2 ). That is, up to O((pe )d−2 ), the Hilbert–Kunz function of M differs
from that of a free module of the same rank by a constant multiple of (pe )d−1
for e " 1. We give a brief outline of the proof for this statement. For a
nonzero ideal J (equivalently torsion free module of rank one), if its divisor class
c(J ) = 0, then R/J is torsion and dim(R/J ) d − 2. The short exact sequence
0 → J → R → R/J → 0 implies that ϕJ,I (e) = ϕR,I (e) − ϕR/J,I (e) +
(TorR [pe ] , R/J )). So using [72, Lemma 1.2] and Lemma 3.6 above, one can
1 (R/I
prove that ϕJ,I (e) = ϕR,I (e) + O((pe )d−2 ) ([59, Lemma 1.2]). A similar result
for a torsion free module M of rank r with c(M) = 0 can be achieved, namely
ϕM,I (e) = rϕM,I (e) + O((pe )d−2 ) ([59, Theorem 1.4]). Next, if M and N are
torsion free and c(M) = c(N), then their Hilbert–Kunz functions are equal up to
O((pe )d−2 ) and also
(Tor1 (R/I [p ] , M) = O((pe )d−2 ) [59, Lemma 1.5]. (This
e
latter statement will be needed in Step 3.) We note that since two modules are
often fit into the same short exact sequence in order to compare their Hilbert–
Kunz functions (as just done for J and R in the above), bounding the length
of the Tor-functor within a desired range becomes crucial for the success of the
argument. Finally for an arbitrary torsion free module of rank r, one considers
δe = ϕM,I (e) − rϕR,I (e) and proves that this difference satisfies a recurrence
relation: δe+1 = (pe )d−1 δe + O((pe )d−2 ) [59, Theorem 1.8]. Thus by Lemma 3.7,
there exists a real number, denoted by τ (M), such that
δe = τ (M)(pe )d−1 + O((pe )d−2 ).
By the definition of δe , we have
ϕM,I (e) = r ϕR,I (e) + τ (M)(pe )d−1 + O((pe )d−2 )
where the number τ (M) depends only on the class of M and is additive. In fact
M → τ (M) gives a well-defined map on the divisor class group of R to R. In
particular, τ (M) = 0 if c(M) = 0 (c.f. [59, Theorem 1.9] [24, Theorem 4.1]).
Step 2 In this step, we prove ϕR,I (e) has the desired form (3.6) by taking M = 1R
and repeating a similar approximation as in Step 1. In fact, 1R is a finitely generated
R-module by the hypothesis of R. Notice that
ϕ1R,I (e) =
R (1R/I [p ]1R) =
R (R/I [p
e e+1 ]
) = ϕR,I (e + 1).
On the other hand, since R is a domain and R is F -finite, 1R is a torsion free R-

module of rank pd . Thus by Step 1, we have
ϕR,I (e + 1) = ϕ1R,I (e) = pd ϕR,I (e) + τ (1 R)(pe )d−1 + O((pe )d−2 ). (3.7)
Let τ = τ (1R). We set
ue = ϕR,I (e) − β(pe )d−1
for some β whose value will be clear in the following. Using (3.7), we calculate
: ; : ;
ue+1 − pd ue = ϕR,I (e + 1) − β pd−1 (pe )d−1 − pd ϕR,I (e) − β (pe )d−1
: d ;
= p ϕR,I (e) + τ (pe )d−1 + O((pe )d−2 )
: ;
− p ϕR,I (e) + (pd − pd−1 )β (pe )d−1 + O((pe )d−2
= [τ + (pd − pd−1 )β] (pe )d−1 + O((pe )d−2 ).
τ
Now by setting β = , the first term following the last equality in the
pd−1 − pd
display above vanishes and hence ue+1 − pd ue = O((pe )d−2 ). Thus applying
Lemma 3.7, there exists a real number α such that
ue = α (pe )d + O((pe )d−2 ).
Recovering ϕR,I (e) from ue , we then have the desired form
ϕR,I (e) = α (pe )d + β (pe )d−1 + O((pe )d−2 .
Comparing to Monsky’s original result for ϕR,I (e), the leading coefficient α must
be the Hilbert–Kunz multiplicity eH K (R, I ) as expected.
Step 3 For an arbitrary module M, let T be the submodule of torsion elements in
M. We have a short exact sequence 0 −→ T −→ M −→ M/T −→ 0 where M/T
is a torsion free module, denoted M . Tensoring the sequence by R/m[p ] , we obtain
e
a long exact sequence

e e e e
· · · −→ Tor1 (M , R/m[p ] ) −→ T /m[p ] T −→ M/m[p ] M −→ M /m[p ] M −→ 0.
By [59, Lemma 1.5], we know

(Tor1 (M , R/m[p ] )) = O((pe )d−2 ). Thus
e
ϕM,I (e) = ϕM ,I (e) + ϕT ,I (e) + O((pe )d−2 )

= r ϕR,I (e) + τ (M ) (pe )d−1 + ϕT ,I (e) + O((pe )d−2 )
where r = rank M . The proof is completed by noticing that ϕR,I (e) has the desired
form from Step 2 and since dim T d − 1, we have ϕT ,I (e) = β(T ) (pe )d−1 +
O((pe )d−2 ) for some β(T ) ∈ R by [72, Lemma 1.2 and Theorem 1.8].
130 C-Y. J. Chan
Instead of the normal (R1) + (S2) condition, Kurano and the author consider a
weaker condition in [24]. The ring R satisfies (R1 ) if the localization of R is a field
at any prime ideal of Krull dimension d, and is a DVR at any prime ideal of Krull
dimension d −1. The (R1 ) condition is similar to, but not the same as the usual (R1).
It can be shown that (2.4) holds for excellent local rings that satisfy (R1 ) but are not
necessarily integral domains. The proof is done by reducing to the assumption that
R is a normal domain ([24, Theorem 3.2]). (See also [55] for a different approach.)
On the other hand, we observe that each step of the proof in [59] just outlined is
interesting in its own right. These steps work more generally than just in a normal
setting. If R is not normal, the divisor class group is no longer well-defined. The
immediate challenge is the description of τ that leads to the second coefficient β.
The Chow group is a natural substitute for the divisor class group in the non-
normal case. We now describe how to replace divisor classes of modules by cycle
classes. For a finitely generated module M, there always exists a finite filtration
of submodules, called a prime filtration, such that the quotient of two consecutive
submodules is isomorphic to a quotient of R by a prime ideal. Let p1 , · · · , ps be the
prime ideals of codimension 0 or 1 that occur in such a prime filtration. Then these
prime ideals define a cycle class [M] = [A/p1 ] + · · · + [A/ps ] in the Chow group
A∗ (R) (see Sect. 3.4 for the definition). Since A∗ (R) = ⊕di=0 Ai (R) according to its
prime ideal generators, we have [M] ∈ Ad (R)⊕Ad−1 (R). Even though a collection
of such prime ideals pi ’s may not be unique for prime filtrations are not unique,
the class [M] in the Chow group is independent of the choice of filtration and is
additive as proved in [23, Theorem 1 and Corollary 1]. This implies that each finitely
generated module M has a unique cycle class defined in the Chow group obtained by
a filtration. Hence the definition of the map τ can be extended to a homomorphism
from Ad (R) ⊕ Ad−1 (R) to R. The proof in [59] is extended step by step to the case
where R is an integral domain satisfying (R1 ) in [24, Sect. 5]. The cycle classes
of the modules affect τ and β in the same way as in the normal situation. This
extension also inspires the consideration of an additive error of the Hilbert–Kunz
function which is not additive on short exact sequences. With the τ map mentioned
above, one sees that the additive error always arises from torsion submodules and is
determined by their classes in the Chow group [24, Sect. 4].
The vanishing of the second coefficient β has been of interest since its discovery.
We will return to this topic in Sect. 3.4. See also Theorem 3.9 for an example where
β does not vanish.
The divisor class group will appear again in Sect. 3.5 when we review the BG
decomposition of normal affine semigroups [16]. The modules M and M/T define
the same divisor class (or cycle class) in the divisor class group (resp. Chow
group). From the above sketch, we notice that the leading coefficient of ϕM,I (e)
is determined by the rank of M but the second coefficient β (or the additive
error) depends on the class of M (resp. classes of torsion submodules). So in
order to understand the second coefficient, or the remaining terms of the Hilbert–
Kunz function, the divisor class group (or Chow group) cannot be overlooked (see
Remark 3.15).
3.3 Via Sheaf Theory
In this subsection, R is a standard graded κ-algebra. Sheaf theoretic approaches

were first considered independently by Fakhruddin and Trivedi [41, 103], and by
Brenner [10, 11]. The general idea is that ϕR,I (e) is identified as the alternating sum
of the lengths of sheaf cohomology modules. Then they carefully study the sheaves
occurring in the sequences arising from the resolution of R/I to describe ϕR,I (e).
In this subsection, we present one of these approaches, following the argument
in [11, Sect. 6] (equivalent to that by Trivedi in [103]). In that work, Brenner applies
the theory built for locally free sheaves on smooth projective curves to obtain
the Hilbert–Kunz functions of two-dimensional normal domains that are standard
graded κ-algebras with an algebraically closed field κ of positive characteristic p.
In this case, the Hilbert–Kunz function has the following form
ϕR (e) = R (q) = α q 2 + γ (q) (3.8)
where α is a rational number [10, 103], and γ (q) is a bounded function taking on
rational values and is eventually periodic when κ is the algebraic closure of a finite
field [11]. For detailed background in this subsection, one can also consult Brenner’s
lecture [12].
Let R+ denote the graded maximal ideal of R and I be a graded ideal primary
to R+ generated by f1 , . . . , fs of degree d1 , . . . , ds respectively. We consider
Y = Proj(R) which is assumed to be a smooth projective curve. Equivalently, R is
normal and dim R = 2. With these, one obtains the following short exact sequence
of coherent sheaves on Y
f1 ,...,fn
0 −→ Syz(f1 , . . . , fn )(m) −→ ⊕ni=1 OY (m − di ) −→ OY (m) −→ 0 (3.9)
where m ∈ Z indicates the twist of the structure sheaf OY and Syz(f1 , . . . , fn ) is

known as the syzygy sheaf (or syzygy bundle if locally free).
We explain below why Syz(f1 , . . . , fn ) is a locally free sheaf. Let K be an R-
f1 ,...,fn
module such that the sequence 0 → K → R n −→ R → R/I → 0 is exact, or
equivalently, R/I is the zero-th cohomology of the following complex
f1 ,...,fn
0 −→ K −→ ⊕ni=1 R(−di ) −→ R −→ 0. (3.10)
Since I is R+ -primary, R/I is only supported at R+ . Therefore for any prime ideals
p not equal to R+ , the localization (R/I )p = 0. This shows that the above complex
(3.10) is exact locally at every prime ideal p ∈ Proj(R). That is, the following
sequence is exact and it obviously splits since Rp is free
0 −→ Kp −→ ⊕ni=1 Rp (−di ) −→ Rp −→ 0. (3.11)

132 C-Y. J. Chan
Hence, as a direct summand of a finitely generated free module over a local ring,
Kp is a free module. Taking the sheafification of (3.10), we have the following exact
sequence of locally free sheaves on Proj(R)
−→ ⊕n OY (−di ) −→ OY −→ 0
0 −→ K (3.12)
i=1
and K is precisely the syzygy bundle Syz(f1 , . . . , fn ) under consideration. The

complex (3.9) is obtained by twisting (3.12) by an integer m; equivalently by
tensoring with O(m). Since (3.11) is split exact on stalks, so is (3.9). Thus the
complex remains exact when applying any additive functor, including the tensor
product.
Next we consider the absolute Frobenius morphism F : Y → Y which is the
identity on the points of Y and furthermore, on every local ring of sections, it is the
Frobenius homomorphism f . The Frobenius pull-back of a sheaf of modules on Y
is obtained by base change along the Frobenius homomorphism, i.e.,
= 1R
F∗ (M)
⊗R M = F(M).
As (3.9) is split exact on stalks, the Frobenius pull-back of this sequence remains
exact. The resulting locally free sheaves are of the same rank twisted by the
appropriate degree, but the multiplication maps are raised to the p-th power. One
may iterate this e times and the sequence remains exact:
pe pe
pe pe f1 ,...,fn
0 −→ (Syz(f1 , . . . , fn ))(m) −→ ⊕ni=1 OY (m − pe di ) −→
OY (m) −→ 0.
(3.13)
Taking global sections is a left exact functor. Since R is normal, the global sections
of twists of the structure sheaf can be realized as the graded pieces of degree m:
8 9 pe
f1 ,...,fn
pe
pe
0 −→ Y, (Syz(f1 , . . . , fne ))(m) −→ ⊕ni=1 Rm−pe di −→ Rm −→ · · · .
The cokernel at Rm is the graded piece of degree m in R/I [p ] . By definition, the

e
Hilbert–Kunz function is
8 9 8 9
e e pe pe
ϕR,I (e) =
R/I [p ] =
(R/I [p ] )m = dimκ ((Y, OY (m))/(f1 , . . . , fn )).
m0 m0
(3.14)
Thus for each m 1, we have
8 9 n
pe pe
(R/I [p ] )m =h0 (OY (m))−

e
h0 (OY (m−pe di ))+h0 (Syz(f1 , . . . , fn )(m)),
i=1
(3.15)
where h0 (·) denotes the κ-dimension of the 0-th cohomology module which is the
module of global sections of the sheaf in the argument.
We remark that (3.14) is a finite sum since the sum is only defined for m 0. In
addition, the alternating sum in (3.15) is zero for m " 0 due to Serre’s vanishing
theorem. In fact, by [12, Lemma 9.4], it suffices to consider m within the certain
range as described in the summation presented in (3.18) in order to compute the
right hand side of (3.14).
To analyze the terms in (3.15), one reduces to the situation of considering
semistable locally free sheaves. This is done by carefully applying the strong
Harder-Narashimhan filtration on locally free sheaves. To further explain the
concepts, we first recall some definitions and their properties from [10, 11] (see also
[12, Chap. 5 and 9]). In the remainder of this subsection, S denotes a locally < free
sheaf on Y of rank r. The degree and slope of S are defined by deg S := deg r (S)
and μ(S) := deg S/r. The slope has the property that μ(S1 ⊗S2 ) = μ(S1 )+μ(S2 ).
A locally free sheaf S is semistable if μ(T) μ(S) for every locally free
subsheaf T ⊆ S. For every locally free sheaf on Y , there exists a unique Harder-
Narasimhan filtration. This is a finite filtration of locally free subsheaves S1 ⊂
S2 · · · ⊂ St = S such that the quotients Sk /Sk−1 are semistable and of decreasing
slope: μ(Sk /Sk−1 ) > μ(Sk+1 /Sk ). Naturally the largest and smallest numbers in
the sequence are called the maximal and minimal slopes of S. They are denoted
by μmax (S) and μmin (S) respectively. The sheaf S is semistable if and only if
μ(S) = μmin (S) = μmax (S).
The Frobenius pull-back of a semistable locally free sheaf S is not necessarily
semistable. Let Sq denote the q-th Frobenius pull-back of S. A locally free sheaf S
is said to be strongly semistable if the pull-back of Sq is semistable for any e 1.
The existence of a filtration with such nice factors is due to a theorem of
Langer [67] (see also [11]): there exists a Frobenius power q such that the quotients
in the Harder-Narasimhan filtration of the pull back Sq are strongly semistable. This
is called the strong Harder-Narasimhan filtration (of Sq ), denoted by
0 ⊂ (Sq )1 ⊂ · · · ⊂ (Sq )t = Sq . (3.16)
We use (Sq )k to indicate the members in the filtration of Sq and to distinguish

it from the pull-back (Sk )q of Sk whose quotient may not be semistable. For q

q " 1, we have the Harder-Narasimhan filtration of Sq
q /q q /q
0 ⊂ (Sq )1 ⊂ · · · ⊂ (Sq )t = (Sq )q /q = Sq . (3.17)
q /q q /q !q /q
We observe that the quotient in (3.17) (Sq )k /(Sq )k−1 = (Sq )k /(Sq )k−1
is the pull-back of (Sq )k /(Sq )k−1 . Thus these quotients are semistable since
(Sq )k /(Sq )k−1 is strongly semistable which follows from Langer’s Theorem.
The slopes of these quotients are decreasing since μ((Sq )k /(Sq )k−1 ) >

μ((Sq )k+1 /(Sq )k ). Therefore (3.17) is a Harder-Narasimhan filtration of Sq ,
as a (q /q)-th Frobenius pull-back of Sq . Using (3.16), for q " 1, we consider
the normalized slope of the quotients in the Harder-Narasimhan filtration of Sq and
define
134 C-Y. J. Chan
μ((Sq )k /(Sq )k−1 )

μ̄k = μ̄k (S) = .
q
Similarly to the above argument, for any q q " 0, if we take the Harder-

Narasimhan filtration of Sq from (3.17), then
q /q q /q
μ((Sq )k /(Sq )k−1 ) (q /q)μ((Sq )k /(Sq )k−1 ) μ((Sq )k /(Sq )k−1 )
μ̄k (Sq ) = = = .
q q q

This shows that μ̄k (S q ) is independent of q " 1.
With the above, we define the Hilbert–Kunz slope

t
μH K (S) = rk μ̄2k ,
k=1
where rk = rk((Sq )k )/(Sq )k−1 ). Note that Hilbert–Kunz slope is a positive rational
number. In [10], the Hilbert–Kunz multiplicity of R with respect to a homogeneous
R+ -primary ideal I = (f1 , . . . , fn ) with deg fi = di is expressed in terms of μH K
as

1
n
eH K (R, I ) = μH K (Syz(f1 , · · · , fn )) − (deg Y )2 di2 ,
2 deg Y
i=1
where deg Y = deg(OY (1)) is the degree of the curve.

We fix some notation before stating the main results regarding the global sections
of locally free sheaves in [11]. Similarly to h0 (·), h1 (·) denotes the κ-dimension of
the first homology of a sheaf. We define vk = −μ̄k / deg Y and write qvk =
qvk + πk with the eventually periodic function πk = πk (q). For simplicity, we
sometimes drop the argument q. Let σ v1 and ρ " vt be rational numbers. We
set qρ = qρ + π .
For q = pe " 1, using Serre duality, the κ-dimension of the global sections of
the twisted sheaf Sq (m) can be expressed in the following form
qρ−1

h0 (Sq (m))
m=qσ
q2
= 2 deg8Y (μH K (S) + 2ρ9deg ⇐S) deg Y + ρ 2 rk(S) deg(Y )2 )
8 9
+q ρ rk(S) + deg S 1 − g − deg Y + qπ(deg S + ρ rk(S) deg Y )
deg Y 2
8 9 t
deg Y
+ rk(S)π (π − 1) deg2 Y + 1 − g − rk πk (πk − 1) +1−g
2
⎛ deg ω
k=1 ⎞
qvk + deg Y
⎜ t ⎟
+ ⎝ h1 (((Sq )k /(Sq )k−1 )(m))⎠ ,
k=1 m=qvk
(3.18)
where g is the genus and ω is the canonical sheaf of the curve Y respectively.
The proof of the above expression in [11, Theorem 3.2] utilizes the fact that the
rank and degree are additive on short exact sequences and that the Hilbert–Kunz
slope is additive on the quotients in the strong Harder-Harasimhan filtration.
A main theorem in [11] is stated as follows.
Theorem 3.8 (Brenner [11, Theorem 4.2]) Assume that κ is an algebraically
closed field of positive characteristic p. Let Y be a smooth projective curve over
κ. Let S denote a locally free sheaf on Y and Sq is the q-th Frobenius pull-back of
S. Let σ v1 and ρ " vt denote rational numbers. Then we have
qρ−1

h0 (Sq (m)) = αq 2 + β(q)q + γ (q),
m=qσ
where α is a rational number and β(q) is an eventually periodic function and γ (q)
is a bounded function (both with rational values). Moreover, if κ is the algebraic
closure of a finite field, then γ (q) is also an eventually periodic function.
We outline the proof of Theorem 3.8 paying special attention to why γ (q) is
eventually periodic when the ground field κ is the algebraic closure of a finite field.
Beginning of the Sketch The first statement on the summation of the global sections
is a consequence of (3.18) once the last summand is understood. One observes
that all the values in (3.18) including the Hilbert–Kunz slope μH K (S) are rational
numbers. Thus the leading coefficient α of q 2 is rational and β(q) and γ (q) are
both rational valued. The function β(q) depends on π which is a periodic function
of q " 1. Hence it is eventually periodic. The boundedness of γ (q) is a result of
[11, Lemma 4.1] which proves the existence of an upper bound of the sum of h1
terms (3.18).
Below, we point out two facts in the proof of [11, Theorem 4.2] that lead to
the periodicity of γ (q) under the assumption that κ is an algebraic closure of a
finite field. For a given rational number v, write m(q) = qv = qv + π(q) with
0 π(q) < 1. Notice that π(q) is an eventually periodic function. Let q̃ denote its
period.
Fact 1 deg(Sq (m(q)) is eventually periodic with period q̃.
We consider a subset M of N of the type M = {q0 q̃
|
∈ N} where q0 = pe0 for
some e0 and satisfies 1 q0 < q̃.
Fact 2 Sq q̃ (m(q)) = Sq (m(q))q̃ ⊗ O((−q̃ + 1)π(q)) for all q ∈ M.
From Fact 1, all Sq (m(q)) with q ∈ M have the same degree. Furthermore,
due to the fact that there are only finitely many semistable locally free sheaves with
the same fixed degree on Y defined over the finite field of which κ is an algebraic
closure, the following set
S := {Sq (m(q))|q ∈ M}
136 C-Y. J. Chan
is a finite set. Then, Fact 2 gives a recursive relation as q varies in M. Since S is a

finite set, this relation leads to an eventually periodic pattern. Other sheaves, namely
Sq (m(q)) where q ∈ / M, can be identified as Sq (m(q) + s) = Sq (m(q)) ⊗ O(s)
for some q ∈ M and s taking values in a fixed bounded interval. Thus they will
occur eventually periodically as some element from S tensored by OY (s) with s
in the same fixed bounded interval. The h1 terms in γ (q) are determined by the
sheaves Sq (m(q)) just described. Hence they inherit the periodic behavior. End of
the sketch.
Now we return to the function ϕR,I (e). By (3.12), (3.14) and (3.15), it can be
expressed in terms of the Frobenius pull-back
⎛ ⎞

n
ϕR,I (e) = R,I (q) = h0 (Oq (m)) − ⎝ h0 ((O(−di ))q (m))⎠
m0 i=1 m0
!
+ h0 Sq (m) ,
m0
(3.19)
where S = Syz(f1 , . . . , fn ). The last summand in (3.19) can be written using
(3.18). The range of the sum is given in [12, Lemma 9.4]. Hence by Theorem 3.8,
the Hilbert–Kunz function has the form
R,I (q) = αq d + βq d−1 + γ (q).
Furthermore, we observe that the coefficients of q in (3.18) are linear combinations

of ranks and degrees which are additive on short exact sequences. Thus they will
cancel each other in the sum (3.19). Hence the linear coefficient β always vanishes
in the case of smooth projective curves. This gives the Hilbert–Kunz function the
desired form R,I (q) = αq 2 + γ (q) in (3.8).
By Theorem 3.8, we know that γ (q) is bounded. In addition, if κ is the
algebraic closure of a finite field, γ (q) is an eventually periodic function (see [11,
Theorem 6.1]). We note that the proof of (3.8) and the periodicity of γ (q) has been
known prior to [11] for other cases such as in [14, 19, 31, 41, 73, 76, 80, 81, 102]
where the finiteness condition on the field is not needed. On the other hand, if κ is
not the algebraic closure of a finite field, i.e., if κ is transcendental over any finite
subfield, then whether or not γ (q) is eventually periodic is an open question in
general.
A similar approach for smooth algebraic curves can be found in Trivedi [103].
The above sheaf theoretic approach depends heavily on the smoothness condition on
the curve Y . Without smoothness, the Frobenius functor is not an exact functor, and
torsion free sheaves are not necessarily locally free. Monsky [77] extends Brenner
and Trivedi’s method to irreducible projective plane curves without smoothness,
and then a linear term appears in R,I (q). The following Theorem 3.9, quoted
from [12, Theorem 9.10], summarizes the very interesting results of [77] which
show that appearance of β reflects the existence of singularities. In this theorem,

the Hilbert–Kunz functions R,I (q) are taken with respect to I = (x, y, z) and
arbitrary f . Then in [79], Monsky focused on nodal curves recovering a result
of Pardue [82] when I = (x, y, z), and also extended the result to cases of
arbitrary I primary to (x, y, z). More specifically, Monsky [79] applied Brenner
and Trivedi’s method, together with the theory for indecomposable vector bundles
developed by Burban [20], and obtained a sharp result. Precisely, one has R,I (q) =
αq 2 + βq + γ (q) for any ideal I primary to (x, y, z). The leading and second
coefficients α and β (nonzero) are constant with explicit formulas given in the paper.
Moreover, when p = 3, γ (q) is a periodic function depending on q modulo 3;
otherwise γ (q) is a constant. Thus R,I (q) is a polynomial function when p = 3.
Theorem 3.9 (Monsky [77, Theorems I & II]) Let C = Proj(R) be an irreducible
projective plane curve where R = κ[x, y, z]/(f ) with a homogeneous f of
degree d. Then
3d a2 a
)(pe )2 + be∗ · pe + γ (pe ),
e e e
ϕR (e) =
(R/(x p , y p , zp , f ) = ( +
4 4d d
where a ∈ Z[ p1 ], 0 < a < d, and be∗ is a periodic integer-valued function. The

∗
function be∗ can be written as be∗ = Q βQ (e) where the sum runs over the singular
points Q of C and βQ ∗ (e) is an integer-valued periodic function for each Q. The
function γ (pe ) is bounded and is eventually period if κ is an algebraic closure of a

finite field.
We remark that the coefficient of pe in the above theorem does not neces-
sarily vanish and in fact it is described by the singular points on C. Monsky’s
papers [77, 79] provide some examples and are interesting resources for those who
are interested in the technique presented in this subsection.
Finally we would like to point out that this sheaf theoretic approach has also been
applied to study the uniformity of the limit Hilbert–Kunz function as described in
Application 3.5.
3.4 Via Local Chern Characters
In this subsection, we present an application of algebraic intersection theory to

the study of Hilbert–Kunz functions. We start with a brief introduction to some
fundamental notions of this theory in the local ring setting. Then we present
Kurano’s theorem that expresses R,I (q) in terms of local Chern characters,
followed by discussions on the vanishing property of the second coefficient, and the
existence of Hilbert–Kunz functions in a polynomial form with desired coefficients.
Let (R, m) be a Noetherian local domain of positive characteristic p and dim R =
d. We assume also that R is a homomorphic image of a regular local ring (in order to
138 C-Y. J. Chan
define its Todd class). Let G• be a bounded complex of free modules of finite ranks
such that all the homology modules have finite length. We call such G• a perfect
complex supported at m, although we will abbreviate this to perfect complex here.
The local Chern character ch(G• ) is a fundamental piece of machinery in
intersection theory and is often defined in the setting of projective schemes. Here
we describe ch(G• ) and other relevant terms following Roberts [87] where the
definitions over local rings (in the setting of affine schemes) are provided in addition
to their projective version. (See also Fulton [43] for the general theory.)
We begin by defining the Chow group of R which is decomposed into a direct
sum of subgroups: A∗ (R) = ⊕di=0 Ai (R) where each Ai (R) is a quotient of a
free group Zi (R) modulo rational equivalence and Zi (R) is generated by prime
ideals p of dim R/p = i. We use [R/p] to denote an element in Zi (R) and,
by abuse of notation, its equivalence class in Ai (R). For any prime ideal q of
R/q = i + 1 and x not in q, we define div(q, x) in Zi (R) as div(q, x) =
dim
p
(R/(q + (x))p )[R/p] where the summation is taken over all prime ideals
p of dimension i containing q and x. Notice that this is a finite sum since all
the p in the sum are minimal prime ideals of R/(q + (x)) and there are only
finitely many of them. Rational equivalence is the equivalence relation on Zi (R)
induced by setting all div(q, x) = 0. Applying this definition to R/m in place of
R gives the Chow group of R/m as a free group of rank one. Indeed A∗ (R/m) =
A0 (R/m) = Z · [R/m]. Its tensor product with the rational number field is denoted
A∗ (R/m)Q := A∗ (R/m) ⊗Z Q. (c.f. [87, Sect. 1.1]).
Let G• be a perfect complex as defined above. The local Chern character ch(G• )
is a map from A∗ (R) to A∗ (R/m)Q . For any class α in A∗ (R), the notation
ch(G• )(α) means applying sufficiently many hyperplane sections arising from the
complex G• to the class α. Many details are involved to assure that ch(G• ) is a
well-defined map. For this we refer the readers to Roberts [87, Sect. 11.5] which
will further lead to appropriate references for the precise definitions in each step
along the way. Here we mention some properties about the Chern characters.
First, the local Chern character is additive. Second, it decomposes as ch(G• ) =
chd (G• )+chd−1 (G• )+· · ·+ch0 (G• ). Each chi (G• ) results in i hyperplane sections
so the dimension will drop exactly by i. Hence for any α ∈ A∗ (R), if we consider
its decomposition in A∗ (R) as α = αd + αd−1 + · · · + α0 , then chi (G• )(αj ) = 0 for

i = j , and thus ch(G• )(α) = di=0 chi (G• )(αi ). We note also that the local Chern
character can be defined for a bounded complex supported at a larger spectrum;
that is, for those bounded complexes whose homology modules need not have finite
length. In this case, the target Chow group will also be larger. This fact will be used
in the definition of the Todd class below.
Now we describe the Todd class. Following Roberts [87, Sect. 12.4], it is defined
for bounded complexes of finitely generated modules. Let M• be such a complex
supported at an ideal a. By assumption, R is a homomorphic image of a regular
local ring, say S. Viewing M• as a complex over S, we take a finite free resolution
H• of M• over S. Consider the class [S] defined by the zero ideal in the Chow group
A∗ (S). The Todd class of M• is given by td(M• ) = ch(H• )([S]) where ch(H• ) is
defined in a more general sense than previously described and as a result, td(M• )
is a class in A∗ (R/a)Q . It is important to note that the definition of the Todd class
is independent of the choice of the regular local ring S. The Todd class can also be
defined for an R-module M by considering it as a complex concentrated in degree
0. In the special case of M• being a perfect complex, [87, Theorem 12.4.2] relates
the Todd class to the Euler characteristic
td(M• ) = χ (M• )[R/m], (3.20)
which is part of the local Riemann–Roch formula to be described next.

Local Chern characters do not have well-behaved functorial properties, namely,
they do not commute with push-forwards or pull-backs. Todd classes are introduced
to fill in this gap via Riemann–Roch Theorem (c.f. Serre [95, Introduction]). The
Riemann–Roch Theorem can be stated in many different forms depending on the
context. The local Riemann–Roch Formula ([87, Sect. 12.6]) relates local Chern
characters to the Euler characteristic. It states that
td(G• ⊗R M) = ch(G• )(td(M)). (3.21)
By assumption, G• is perfect which implies that G• ⊗R M is also perfect. Thus

(3.20) and (3.21) together give
χ (G• ⊗R M)[R/m] = ch(G• )(td(M)). (3.22)
Notice that the local Chern character ch(G• ) maps into the rational Chow group
of R/m, or precisely Q · [R/m]. Thus its image may be identified with a rational
number.
Now replacing M by e R yields
χG• (e R) = ch(G• )(td(e R)). (3.23)
(See also Fulton [43, Example 18.3.12]).

Theorem 3.10 (Kurano [65, 66]) Let R be a local ring that satisfies the following
conditions:
(i) R is the homomorphic image of a regular local ring whose residue field is
perfect;
(ii) R is F -finite;
(iii) R is a Cohen–Macaulay ring.
Let I be an m-primary ideal of finite projective dimension and let G• be a finite free
resolution of R/I . Then
ϕR,I (e) = (ch(G• )(cd ))(pe )d + (ch(G• )(cd−1 ))(pe )d−1 + · · · + (ch(G• )(c0 )),
140 C-Y. J. Chan
where ci is the i-th Todd class of R in the i-th component of the Chow group, namely
td(R) = cd + · · · + c0 ∈ A∗ (R).
With the prior preparation, we now bring up two main facts that lead to the results
above: (1) one has tdi (eR) = pie ci ∈ Ai (R) ([66, Lemma 2.2(iii)]); and (2) for any
i, the i-th Chern character maps ci to a rational number but all other cj to zero. In
other words, ch(G• )(ci ) = chi (G• )(ci ) is a rational number in A∗ (R/m)Q ∼ = Q.
Therefore by (3.1) and (3.23)
ϕR,I (e) = χG• (eR)

= ch(G• )(td(eR))
!
= di=0 chi (G• ) (pe )d cd + (pe )d−1 cd−1 + · · · + c0

= di=0 (chi (G• )(ci ))(pe )i .
Or equivalently with q = pe ,
R,I (q) = (ch(G• )(cd )q d + (ch(G• )(cd−1 ))q d−1 + · · · + (ch(G• )(c0 )). (3.24)
This is a polynomial in q of degree d and all the coefficients are rational numbers.
The leading coefficient, which never vanishes, is the Hilbert–Kunz multiplicity.
However, the second coefficient βI (R) = chd−1 (G• )(cd−1 ) can sometimes vanish,
for instance, in the case of a two-dimensional normal domain as observed in [59]
and [11]. More generally, from Kurano [66, Corollary 1.4], we learn that if R
satisfies the conditions (i) and (ii) above, and R is normal and Q-Gorenstein (i.e., the
canonical module defines a torsion element in the divisor class group), then βI (R)
vanishes. This is also true for the second coefficient in the function M,I (q) of a
module M.
In [24, Theorem 3.5], the vanishing property of β is characterized by the classes
in the Chow group. For example, by the technique in [88] the vanishing of βI (R)
for every I is equivalent to the fact that the second top Todd class τd−1 = tdd−1 (R)
is numerically equivalent to zero, i.e., ch(G• )(τd−1 ) = 0 for any perfect complex
G• . Furthermore, if the localization Rp is Gorenstein for all minimal prime ideals
p, then βI (R) = 0 if and only if τd−1 and its canonical module ωR are numerically
equivalent, namely, ch(G• )(τd−1 ) = ch(G• )(ωR ) for any G• that is again perfect.
Theorem 3.10 also makes it possible to prove the existence of Cohen–Macaulay
local rings such that the Hilbert–Kunz functions have polynomial expressions as in
(3.24) and their rational coefficients have the desired positive, negative or vanishing
properties ([25, Theorem 1.1]). The theorem states that if 0 , 1 , . . . , d are integers
such that i = 0 for i d/2, i = −1, 0, or, 1 for d/2 < i < d and d = 1,
then there exists a d-dimensional Cohen–Macaulay local ring R of characteristic
p, an m-primary ideal I of R of finite projective dimension, and positive rational
numbers β0 , β1 , . . . , βd such that R,I (q) = di=0 i βi q i for all i > 0. In [25],
a convex cone in a finite dimensional vector space, named Cohen–Macaulay cone,
spanned by maximal Cohen–Macaulay modules is introduced to carry out the proof.
This theorem proves the existence but does not offer a constructive method to build
a ring with the desired Hilbert–Kunz function.
In general, it is difficult to construct rings that have specific forms of Hilbert–
Kunz functions. This is true even in the setting of affine semigroup rings. (See also
Remark 5.5.) Investigations in this direction can provide not only desired Hilbert–
Kunz functions but also, as shown in Theorem 3.10, a possible approach to access
local Chern characters whose values are equally, if not more, challenging to obtain.
3.5 Via Bruns-Gubeladze (BG) Decomposition
This subsection describes a cellular decomposition on Rd and its fundamental

domain when R is a normal affine semigroup ring. There are one-to-one correspon-
dences between the set of full dimensional cells in this decomposition, the set of
conic divisor classes, and the set of rank one modules as direct summands of an
extension ring of R. Then we present Bruns’s ideas of using these correspondences
to calculate Hilbert–Kunz functions.
Let κ be an algebraically closed field of positive characteristic p and R be a d-
dimensional normal κ-subalgebra of the polynomial ring κ[t1 , · · · , td ] generated by
finitely many monomials. Then the exponents of monomials in R form a finitely
generated monoid M in Zd . Such an R is called an affine semigroup ring, denoted
by R = κ[M]. We use Q0 M (and R0 M) to denote the extended rational cone
spanned by M in Qd (resp. Rd ), and ZM to denote the free abelian group generated
by M. Then ZM has rank d since dim R = d. We assume ZM equals the ambient
group Zd . The assumption that R is a normal domain is equivalent to a condition on
the semigroup: ZM ∩ Q0 M = M (c.f. [51]). Let n1 M = { xn : x ∈ M} and consider
1
R n = κ[ n1 M] as an extension ring of R. Sect. 4 is devoted to the affine semigroup
rings where relevant definitions and terms will be stated in more detail.
The idea of a cellular decomposition of Rd /Zd as a quotient group grew out
of Bruns and Gubeladze’s work [17] that investigates the minimal number of
generators and depth of the divisorial ideal classes of a normal affine semigroup ring
in general. Later it was constructed precisely with the lattice structure in Bruns [16]
which we now refer to as BG decomposition. Some readers might find it helpful to
read Sect. 4 before proceeding to the current subsection. Here as part of the section
for techniques, we present the main theorems on the BG decomposition in [16] that
leads to the computation of the Hilbert–Kunz function of a normal affine semigroup
ring R with respect to the maximal ideal generated by all monomials other than 1.
Let σ1 , . . . , σ
be the linear functionals that define the support hyperplanes for M
so that σi (M) 0 for all i = 1, . . . ,
. Since rank ZM = dim κ[M] = d, we have
d and the set of support hyperplanes is uniquely determined if it is irredundant.

The map σ = (σ1 , . . . , σ
) : M → Z
is called the standard embedding (c.f. [17])
which has also been used in Hochster [51] and Stanley [100]. Such a σ transforms
a normal semigroup to an isomorphic subsemigroup in Z
that is said to be full in
142 C-Y. J. Chan
[51] or pure in [17]. While normality is independent of the embeddings, “fullness”

(resp. “pureness”) is a notion relative to the embedding. But we will not get into the
details on this issue here.
Obviously σ can be extended to define a linear homomorphism on Rd and
σ (Rd ) ∼
= Rd ⊆ R
. We obtain a cell decomposition of Rd by the hyperplanes
Hi,z = {x ∈ Rd |σi (x) = z}, i = 1, . . . ,

, and z ∈ Z,
which then induces a cell decomposition ¯ on Rd /Zd . For concreteness, both and
¯ will both be called the BG decomposition.
A conic divisorial ideal in [16, 17] is defined to be
C(y) = κ · (Zd ∩ (y + R+ M)), y ∈ Rd ,
and it should be considered as the ideal of R generated by the monomials

corresponding to the lattice points on the right hand side. Conic divisorial ideals
are among the divisorial ideals that generate the divisor class group Cl(R). The set
of conic divisor classes in Cl(R) contains all torsion elements but it can be larger if
Cl(R) is nontorsion ([17, p. 141], [16, Corollary1.3]). The following bijections are
the key to the discussions in this subsection.
Theorem 3.11 (Bruns [16, Corollary 1.2]) The following sets are in a bijective
correspondence:
(a) the set of conic divisor classes ;
(b) the fibers of the map Rd /Zd → R
/σ (Zd ) induced by x → σ (x);
¯
(c) the full dimensional cells of .
Let γ̄ be a full-dimensional cell in ¯ represented by some full-dimensional γ in
. Fix an element y in γ , the upper closure of γ is defined to be
γ = {x ∈ Rd |σ (x) = σ (y)}.
This closure, however, is independent of y. Theorem 3.11 shows that all the conic
ideals C(x) with x ∈ γ coincide in Cl(R). Thus we may use Cγ to denote such a
1
conic ideal class. We recall R n = k[ n1 M] for any positive integer n. For each residue
1
class c ∈ ( n1 ZM)/ZM, let Ic be the ideal generated by R n ∩k ·c. Since M is normal,
Ic is equivalently generated by monomials corresponding to R0 M ∩ (c + Z0 M),
and so it defines the same conic divisor class as C(−c) which must be isomorphic to
Cγ for some full dimensional cell γ by Theorem 3.11. This shows that the set of Ic
corresponds to a subset of the conic divisor classes. Then by [17, Proposition 3.6]
we know that the set of Ic and the set of the conic divisor classes are in a one-to-one
correspondence.
1
By [17, Theorem 3.2(a)], R n can be decomposed as an R-module into the
direct sum of Ic over all c ∈ ( n ZM)/ZM with rankR Ic = 1. This implies that
1
1
the rank of R n is the number of the residue classes in ( n1 ZM)/ZM. Furthermore,
each Ic corresponds to a unique conic divisor class Cγ , but there may be multiple
Ic ’s that correspond to the same Cγ . Let νγ (n) be the multiplicity with which the
1
isomorphism class of Cγ occurs in the decomposition of R n as an R-module. From
the previous discussion, it is not difficult to see that νγ (n) = #{Ic |c ∈ γ }.
Theorem 3.12 ([16, Theorem 3.1]) Let γ be a full-dimensional cell in . Then
1 d
νγ (n) = #(γ ∩ Z ).
n
Furthermore, there exists a quasi-polynomial qγ (n) : Z → Z with rational

coefficients such that qγ (n) = νγ (n) for all n 1. In particular,
qγ (n) = vol(γ ) nd + bγ nd−1 + q̃γ (n), n ∈ Z,
where vol(γ ) is the volume of γ , bγ is a constant and q̃γ (n) is a quasi-polynomial

of degree d − 2.
The first statement in Theorem 3.12 holds due to the fact that Ic ∼ = C(−c) if and
only if (−c + Zd ) ∩ γ = ∅. Given this, the relation of νγ (n) to a quasipolynomial
qγ (n) is not a surprise. In fact, since all the vertices of γ have rational coordinates,
we observe that
1 d
#(γ ∩ Z ) = #((n · γ ) ∩ Zd ),
n
where (n · γ ) is the region in Rd whose vertices are n times those of γ .

Therefore, νγ (n) counts the number of lattice points in the dilated γ . This is
exactly the generalized Ehrhart function which will be described in Sect. 4. And
that νγ (n) is a quasipolynomial is a direct consequence of the a polycell version of
Theorem 4.2 which will be explained also in the next section. It has been known
that the leading coefficient of the generalized Ehrhart polynomial can be described
by the volume and hence it is a constant. The remaining coefficients are periodic
functions in n. However, in the setting of the previous Theorem 3.12, Bruns further
refines the results from Theorem 4.2 and identifies that the second coefficient is
also a constant. This refinement leads to the existence of the second coefficient of
the Hilbert–Kunz function in the following Theorem 3.13 that matches the more
general case of normal domains proved in [59] as discussed in Sect. 3.2.
The BG decomposition takes place for any positive integer n. However, if the
coefficient field κ has positive characteristic p, we may restrict n to be powers of
p to obtain Hilbert–Kunz functions. In what follows, we use μR (·) to denote the
minimal number of generators of an R-module.
Theorem 3.13 ([16, Corollary 3.2(a)]) Let κ be an algebraically closed field of
characteristic p > 0 and M a normal finitely generated monoid of rank d. Set
144 C-Y. J. Chan
R = κ[M] and let m be the maximal ideal of R generated by the monomials whose
exponents are in M\{0}. Let γ be a full-dimensional cell as described above and
Cγ be the conic ideal corresponding to γ . Then the Hilbert–Kunz function of R with
respect to m is a quasipolynomial with constant leading and second coefficients.
Precisely

ϕR (e) = μR (Cγ ) νγ (pe ) (3.25)
γ
for all positive integers e.

1
Proof Recall q = pe and R q = κ[ q1 M]. Since κ is perfect, we have the following
κ-algebra isomorphism induced by the Frobenius map,
1 1
R/m[q] ∼
= R q /mR q .
1 1 1
Obviously dimκ R/m[q] = dimκ R q /mR q μR (R q ). By Nakayama’s Lemma,
1 1
the reverse of the last inequality also holds. Hence we have dimκ R q /mR q =
1
μR (R q ).
We recall also that Ic ’s are the rank one direct summands in the decomposition
1
of R q as R-modules. Therefore, we have
1
dimκ R/m[q] = μR (R q ) = μR (⊕c Ic )

= μ (C ), (since Ic ∼ = Cγ for some γ )
c R γ
= μR (C γ ) · #(Ic isomorphic to Cγ )
γ̄
= γ̄ μR (Cγ ) νγ (q).
Hence

R (q) =
R (R/m[q] ) = dimκ R/m[q] = μR (Cγ ) νγ (q).
γ̄
Moreover since μR (Cγ ) is a constant independent from q, it is clear that R (q)

has all the properties enjoyed by νγ (q) in Theorem 3.12, namely, R (q) is a
quasipolynomial and that its leading and the second coefficients are both constants.

We will return to (3.25) after the discussion of Ehrhart Theory in Sect. 4 (see
Remark 4.4).
The next example describes the BG decomposition of the semigroup generating
the rational normal cone of degree 2. This is a special case of Example 5.1 with
g = 2.
Fig. 1 and the equivalence classes in . (a) and . (b) Open cell γ (enlarged). (c) Closure γ
(enlarged)
Fig. 2 P by
Example 3.14 Let R = k[s, st, st 2 ] where M is the semigroup in Z2 generated by

(1, 0), (1, 1), (1, 2). Then σ is the linear map σ = (d2 , 2d1 − d2 ) : M → Z2 . Note
that in this case, M generates a simplicial cone in R2 . The slanted grids in Fig. 1
below show the decomposition of R2 . Full dimensional cells that are equivalent
in are labeled by % and & respectively just to show a few examples. Note that
there are two distinct classes. We also show the graph of a single open cell γ and its
closure γ .
Let P = {x ∈ R0 M|x ∈ / d + R0 M for any nonzero d ∈ M}. Then P can
be tessellated by finitely many full-dimensional cells. Each non-equivalent cell may
occur in a different number of copies. (See Fig. 2.) For some γ , P may contain its
entire closure but not always. Such decomposition is unique in the sense that each
full dimensional cell in ¯ is identified with a unique conic ideal class. The closure
of these semi-open rational polycells γ are convex.
We write γ1 for a cell labeled by % and γ2 for one labeled by &. By
Theorem 3.13,
ϕR (e) = μR (Cγ1 ) νγ1 (pe ) + μR (Cγ2 ) νγ2 (pe ).

146 C-Y. J. Chan
Next we determine the values for μR (Cγ1 ) and μR (Cγ2 ) which are independent from
e. In fact, since Cγ is a class defined by Ic for some c ∈ γ , to obtain the value
for μR (Cγ ), it suffices to consider the minimal number of generators of a divisorial
ideal Ic . Let c = (− 23 , − 23 ) ∈ γ1 , then Ic ∼
= C(−c) ∼
= R · st and μR (Cγ1 ) = 1.
∼ ∼
And if c = (0, − 3 ) ∈ γ2 , then Ic = C(−c) = R · st + R · st 2 and μR (Cγ1 ) = 2.
1
In Sect. 4, we will see that the Hilbert–Kunz function of an affine semigroup ring
R is equivalent to an Ehrhart function which counts the number of lattice points
in a certain dilated rational polycell P arising from R which is nonconvex and not
necessarily closed. We will discuss ϕR (e) from this aspect there. On the other hand,
when it comes to counting lattice points in a polycell (or a polytyope), there is no
previously known canonical method about dissecting the polycell (resp. polytope)
before proceeding to count. Via conic divisor classes, Bruns [16] decomposes P into
finitely many cells and each appears finitely many times. The BG decomposition is
unique for any given affine semigroup. The periodic behavior of the Hilbert–Kunz
function is mainly due to the fact that the vertices of full-dimensional open cells
γ have rational coordinates. Even though the polycell P is not convex, the full-
dimensional cells in the BG decomposition are always convex. These facts can be
very useful.
In order to fully express ϕR (e) using BG decompositions, it is equally important
to know μ(Cγ ) for each γ in ¯ (see Remark 4.4 and Question 5.9). The significance
of μ(Cγ ) can also be seen in [17] in which Bruns and Gubeladze use it to measure
the number of Cohen–Macaulay divisor classes and prove that there are only finitely
many such divisor classes.
Finally we make a remark on the divisor class group.
Remark 3.15 We have seen that the divisor class group appears often in these
techniques. First it is used in proving the existence of β in Sect. 3.2. Then we see that
the representation of the divisor classes play the most crucial role in the discussion
of the periodic behaviors in the current subsection. The divisor classes also occur
implicitly in Sect. 3.3, because the divisor class group is generated by the twists
of the structure sheaves which are the main object dealt with in Theorem 3.8. So
perhaps some level of finiteness condition on the divisor class group is responsible
for the (periodic) behavior of the function. This remains to be investigated.
3.6 Via Combinatorics
A lot of attention has been given to quotients of polynomial rings by monomial

or binomial ideals. In this subsection, we give a very brief account in this
direction without developing their details. These can be traced back to Conca [30]
who considered generalized Hilbert–Kunz functions and utilized Gröbner basis to
calculate the multiplicity of binomial hypersurfaces. Watanabe [112], followed by
Eto [39], approximated the multiplicity by the volume of the relevant polytope. (Our
discussions on affine semigroup ring settings were initially inspired by [112].) Eto
and Yoshida [40] expressed the multiplicity in terms of the Stirling numbers.
Combining the techniques of Segre products as done in [40] and Gröbner bases,
Miller and Swanson [70] compute the Hilbert–Kunz function of the rings of the
form of R = κ[X]/I2 (X) where X is an m × n matrix of indeterminates, κ[X] is the
polynomial ring joining all indeterminates in X, and I2 (X) is the ideal generated by
all the 2 × 2 minors of X. In [70], it is proved that the Hilbert–Kunz functions, for
arbitrary m and n, are true polynomials and are expressed in a recursive manner, but
a closed form is provided only for the special case m = 2. This study is extended
by Robinson and Swanson [89] who give explicit closed polynomial forms for all
positive integers m and n.
Before ending this subsection, we mention an interesting article by Batsukh and
Brenner [5] in which the notion of binoid is introduced as a commutative monoid
with an absorbing element ∞. For rings of combinatorial natures such as Stanley-
Reisner and toric rings and those just described above, several questions arise. These
include the rationality of multiplicity, the interpretation of the notions in the case of
characteristic 0, and the dependence of the results on the characteristic. Batsuhk
and Brenner propose a unifying method evolved around binoids to answer all these
questions simultaneously. It will be interesting to further investigate if this new
approach may be applied to obtain or approximate Hilbert–Kunz functions.
4 Normal Affine Semigroup Rings and Ehrhart Theory
In this section, we describe the generalized Ehrhart function and relate it to

the Hilbert–Kunz function of an affine semigroup ring. Below, we give a precise
definition of affine semigroup rings focusing on identifying monoid elements with
monomials of a Laurent polynomial ring. In this way, affine semigroup rings under
consideration in Sect. 3.5 are naturally subrings of polynomial rings. Furthermore,
it can be understood as the coordinate ring of an affine toric variety.
A monoid in Zd is a semigroup that contains an identity element. Let M denote
a monoid that can be generated by finitely many elements in Zd . We call such M an
affine semigroup. An affine semigroup M is positive if M ∩ −M = 0; namely, the
only invertible element in M is the identity element. Let Q0 M denote the rational
cone generated by M. Saying an affine semigroup is positive is equivalent to saying
that its rational cone is strongly convex or pointed. For the discussions within this
paper, an affine semigroup always stands for a monoid that is finitely generated
and positive. Let ZM denote the subgroup of Zd generated by M. Obviously M ⊆
ZM ∩ Q0 M. An affine semigroup M is normal if equality holds, that is, for any
h ∈ ZM, we have h ∈ M whenever nh ∈ M for some positive integer n. Pictorially,
M is normal means that there is no “hole” when comparing M against ZM ∩Q0 M.
We will assume that M is normal in this section although some of the results might
be true in more general settings.
148 C-Y. J. Chan
Let κ be an algebraically closed field of any characteristic. We write κ[M] for

the affine semigroup ring generated by M over κ. By identifying (i1 , . . . , id ) ∈
Zd with t1i1 · · · tdid , we regard κ[M] as a κ-subalgebra in the Laurent polynomial
ring S = κ[t1 , . . . , td , t1−1 , . . . , td−1 ]. For the purpose of our discussion, it is more
convenient to work with affine semigroup rings under such identification. However,
for notational simplicity, we still write κ[M] in place of the formal κ[tM ]. It is
known that κ[M] is a normal ring if and only if the semigroup M is normal [51].
Without loss of generality, we also assume that ZM has rank d (otherwise, replace d
by the rank of ZM), and that ZM equals the ambient group Zd . Thus dim κ[M] = d.
Let R = κ[M] be an affine semigroup ring as described above. Then R can
be viewed as a k-algebra generated by finitely many monomials in a Laurent
polynomial ring S. Clearly R is an integral domain. Sometimes R is also called a
toric ring, since it is the coordinate ring of an affine toric variety that is constructed
from lattice points in M (c.f. [32, 42, 69]). For instance, a convex polyhedral cone
σ ⊂ (Rd )∨ defines an affine toric variety X = Spec R where R = κ[σ ∨ ∩ Zd ]. In
this convention, σ ∨ = R0 M is an extended cone of the rational cone Q0 M. It can
be proved that R is isomorphic to the quotient of a polynomial modulo a prime ideal
generated by binomials. Conversely a prime ideal generated by binomials always
defines a toric ring and thus a semigroup ring (c.f. [32, Chap. 1]).
Affine semigroup rings have all the above classical interpretations making them
an interesting family of rings to study for any topic in both algebra and geometry.
Next, after a brief introduction of Ehrhart theory, we will see how the polytopes and
their enclosed lattice points interact with ideals of finite colength.
An Ehrhart function concerns the number of lattice points contained in a dilated
polytope. A polytope, denoted by , is the convex hull of finitely many points in
RN . Precisely, for a given finite set of points {v0 , . . . , vg } ⊂ RN , the convex hull
of v0 , . . . , vg is given by = {λ0 v0 + · · · + λg vr |0 λi 1, λ0 + · · · λg =
1} ⊂ RN . The dimension of is equal to dimR Span{v1 − v0 , . . . , vg − v0 }. A
d-polytope refers to a d-dimensional polytope. And n denotes the convex hull of
{nv0 , . . . , nvg }; that is also a dilation of . The lattice-point enumerator (c.f. [6,
p.27]) for the n-th dilation of is defined to be
8 9
E (n) = # n ∩ ZN .
We say that is an integral polytope if all its vertices have integer coordinates
and it is a rational polytope if its vertices have rational coordinates. The following
theorem can be found in [6, Theorems 3.8 and 5.6].
Theorem 4.1 (Ehrhart [36]) Let be an integral d-polytope in ZN and n be a
positive integer.
(a) E (n) is a polynomial function in n of degree d.
(b) If we write E (n) = ad nd + ad−1 nd−1 + · · · + a0 , then
ad = the volume of the polytope ,

1
ad−1 = · E(∂(n·)∩ZN ) (n),
2
a0 = the Euler characteristic of ,
where ∂ indicates the boundary of the polytope.

Whether or not other coefficients carry any significant information is unknown.
Ehrhart’s Theorem can be generalized to rational polytopes. A function f (n) on
Z or Z0 is called a quasipolynomial of period r̃ if it can be written in polynomial
form in n whose coefficients are periodic functions in n with r̃ as the least common
multiple of their periodic lengths. Equivalently, there exist r̃ polynomials f1 , . . . , fr̃
such that f (n) = fi (n) if n ≡ i (mod r̃).
Theorem 4.2 (Generalized Ehrhart’s Theorem [36]) Assume that is a convex
rational d-polytope. Then E (n) is a quasipolynomial in n of degree d. Moreover,
the period divides the least common multiple of the denominators of the coordinates
of the vertices of .
Naturally, E (n) is called the Ehrhart polynomial or quasipolynomial according
to whether is an integral or rational polytope.
The proof of Theorem 4.2 follows very similarly to that of Theorem 4.1, but
as pointed out in Beck and Robin [6], “ the arithmetic structures of Ehrhart
quasipolynomials is much more subtle and less well known than that of Ehrhart
polynomials”. A detailed proof following the instructions of [6, Section 3.7] is
documented in Barco [4]. (See Sam [91] for an alternative approach.)
Note that for any polytope , all the faces in its boundary ∂ must be convex.
Let ◦ = − ∂ be the interior of . By the inclusion and exclusion principle
and applying Theorem 4.1 repeatedly, one can prove that Ehrhart’s Theorem holds
for ◦ with E◦ (n) defined analogously, and also for non-convex polytopes and
their interiors as well. Similarly, Theorem 4.2 which concerns rational polytopes
holds for the boundary ∂, interior ◦ and their non-covex counterparts (c.f. [101,
Section 4.6.2]).
For the following discussion, we define a polycell to be a union of finitely many
polytopes, but not necessarily containing all the faces on the boundary of the union.
We also require that if any two polytopes intersect, they do so only at their faces.
Thus, a polycell is not necessarily convex; neither is it closed nor open. We use the
term a semi-open polycell to emphasize that it contains only part of its boundary. The
Ehrhart Theorem and its generalized version both hold for the semi-open polycells
by the inclusion and exclusion principle.
Next we relate the counting of lattice points to ϕR (e). We begin by doing so
to Hilbert-Samuel functions. We use as a shorthand for the cone R0 M in Rd .
Let a be an ideal generated by monomials in R = κ[M] such that R/a has finite
length. The Hilbert-Samuel function h(n) =
(R/an ) can be obtained by counting=
the lattice points in the complement of the union of the shifted cones Ln = (d +
), where the union runs over the d’s corresponding to the generators td of an .
150 C-Y. J. Chan
Since
(R/an ) is finite, \Ln is a bounded semi-open region. Hence there are only
finitely many lattice points in \Ln . Furthermore, as the number of generators of
an increases as the power n increases, so does the number
= of vertices of Ln . Now
let L be the convex hull containing all the n1 Ln := ( n1 d + ) for n 1. One
sees that h(n) is equal to the number of points in ( \ n1 Ln ) ∩ n1 Zd and a lattice
cube in n1 Zd has volume exactly n1d . Thus multiplying n1d to h(n) gives the Riemann
sum that approximates the volume of \L so that when n → ∞, the quantity h(n)nd
converges to the volume of \L. Hence the Hilbert-Samuel multiplicity i(R, a) of
R with respect to a can be computed using the following volume formula (c.f. [87,
Exercise 2.8])
h(n)
i(R, a) = d! lim = d! Vol( \L). (4.1)
n→∞ nd
Now we assume that char κ = p > 0 and consider the Frobenius powers of
the maximal ideal m consisting of all monomials in R other than 1. (Similar con-
siderations apply to arbitrary ideals of finite colength.) Suppose m is= generated by
td1 , . . . , tdh . Then m[p ] is generated by (td1 )p , . . . , (tdh )p . Let L = hj=1 (dj + )
e e e
be the union of h polyhedral cones. Then

8 > 9

(R/m[p ] ) = dimκ R/m[p ] = # pe · ( \L)
e e
Zd .
This gives the Hilbert–Kunz function a combinatorial interpretation. More precisely,

if we let P = \L. Then P is a semi-open polycell that does not contain its
“upper right” boundary faces. In particular, P is not necessarily (in fact, almost
never) convex and pe · P ∩ Zd consists of all the monomials in R not in m[p ] .
e
The monomials that generate R/m[p ] as a κ vector space are exactly those whose
e
exponents are contained in pe · P, the polycell P dilated by pe . Hence the values of

the Hilbert–Kunz function at e is
ϕR (e) = dimk R/m[p ] = #(pe · P ∩ Zd ).

e
(4.2)
To illustrate the ideas above, we use R = k[s, st, st 2 ] from Example 3.14. Then
ϕR (e) =
(R/m[p ] ) can be obtained by counting the number of lattice points
e
belonging to the pe -dilated semi-open polycell as shown in Fig. 3 below.

A complete computation of the Hilbert–Kunz function of this example will be
presented in Example 5.1.
In general, it is not difficult to see that
8 9
1
# p ·P∩Z
e d
=# P∩ e ·Z .
d
(4.3)
p
Hence similarly to the Hilbert-Samuel multiplicity, one can argue that the limit of
ϕR (e) divided by (pe )d tends to the volume of P. This is a geometric illustration of
Fig. 3 p e · P ∩ Z2 (The labels will be needed in Example 5.1). (a) p = 3, e = 1. (b) p = 3, e = 2
how Watanabe [112] proved the rationality of Hilbert–Kunz multiplicity for affine
semigroup rings, although his proof does not involve Ehrhart theory.
It is important to note that the coordinates of the vertices of P are not
necessarily all integers. Therefore as e varies, we are examining about the values of
generalized Ehrhart functions which are eventually periodic with predictable period
by Theorem 4.2. (See Remark 5.3(a) for comments on periods.)
Next we reflect on the discussion above on Ehrhart theory, Hilbert-Samuel
and Hilbert–Kunz functions. The value h(n) of the Hilbert-Samuel
= function is the
number of lattice points in \Ln . Recall that Ln = (d + ) where the union
runs over the monomial generators td of an . The vertices of \Ln are integral but
its shape changes as n increases due to the increasing number of generators. This
prevents us from taking the advantage of the Ehrhart’s Theorem. On the other hand,
when working with the Frobenius powers of m, the shape of pe · P remains rigid
and is always similar to P. Each value of the Hilbert–Kunz function ϕR (e) is thus
exactly the value provided by the Ehrhart function EP (pe ) for the pe -th dilation
of P. This allows us to apply the generalized Ehrhart’s Theorem 4.2 for the rational
polycell to its full potential. Hence
ϕR (e) = EP (pe ), or equivalently, R (q) = EP (q).
We then conclude that R (q) is a quasipolynomial in q and its value is given by the
lattice points in a dilated polycell.
We now return to the BG decomposition. We see from (4.3) that to compute
EP (pe ), one may restrict the BG decomposition to P and decompose P into
a union of convex polycells. In this way, Bruns refined the generalized Ehrhart
theorem in the case of affine semigroup rings and proved that the second coefficient
is also a constant.
152 C-Y. J. Chan
Fig. 4 P by with labeled

cells
In summary, the discussions above lead to the following identity for e " 1

μR (γ )νCγ (q) = EP (q) (4.4)
γ̄
where the sum runs over the full-dimensional cells in . ¯ More precisely, (4.4)
gives us two equivalent expressions for R (q). As a corollary to Theorem 3.13,
Theorem 4.3 below refines Theorem 4.2 by identifying that the second coefficient is
always a constant for the case of normal affine semigroup rings.
Theorem 4.3 (Bruns [16]) The Hilbert–Kunz function of R with respect to m is a
quasi-polynomial. Precisely it has the form
R (q) = Vol(P)q d + βq d−1 + (a quasipolynomial in lower degrees),
where β is a constant, and Vol(P) denotes the volume of the polycell P determined
by the shape of M in Zd .
Roughly speaking, by Theorem 3.13, the leading coefficient is eH K =

μ
Cγ R (γ )·Vol(C γ ) = Vol(P), and the second coefficient β = γ̄ ∈¯ μR (γ )·bγ
is a constant as expected where bγ is the second coefficient in νγ (n).
We now elaborate on the content of (4.4). First recall the ring R = κ[s, st, st 2 ]
from Example 3.14 and the BG decomposition of the polycell P with labeled cells
in Fig. 4.
Thus (4.4) shows that
EP (pe ) = ϕR (e) = μR (Cγ1 ) νγ1 (pe ) + μR (Cγ1 ) νγ1 (pe ) = νγ1 (pe ) + 2 νγ1 (pe )
where γ1 is equivalent to the cell labeled by % that occurs once and γ2 by & that
occurs twice. Moreover, each νγi (pe ) = Eγi (pe ) is an Ehrhart quasipolynomial
of γi .
The following remarks point out some subtleties in (4.4).
Remark 4.4
(a) Despite the fact that νγ (q) = Eγ (q), a general one-to-one correspondence
between the set of points counted by the left hand side in (4.4) and the
actual lattice points in P, counted by EP (q), is not immediately obvious for

arbitrary P. One can carefully match them in the above Example 3.14. Those
in the interiors (of γi ) are trivial, but the correspondence at the boundaries is
much more subtle, especially in higher dimensions. This is the place where
the BG decomposition lends us power, by relating the lattice points to those
enclosed by the closure of the full-dimensional cells without doing the actual
correspondence.
(b) Since P is tessellated by the full-dimensional equivalence classes γ̄ , (4.4)
suggests that the minimal generators μR (Cγ ) can be obtained by the number
of times that γ̄ appears in P. This is clear in Example 3.14 (see Fig. 4), but full
generality would probably require some careful consideration.
We end the section by summarizing several families of rings whose Hilbert–Kunz
function enjoys the following functional form with constants α and β,
R (q) = αq d + βq d−1 + γ (q).
(1) R is an excellent normal local domain or an excellent local ring satisfying (R1 ).
In this case, γ (q) = O(q d−2 ). (See Sect. 3.2.)
(2) R is a two-dimensional normal domain that is a standard graded algebra over
an algebraically closed field κ. Indeed, β = 0 in this case. Moveover, γ (q) is
a bounded function in general, and is eventually periodic if κ is the algebraic
closure of a finite field. (See Sect. 3.3.)
(3) R is a normal affine semigroup ring. Then, γ (q) is eventually a quasipolyno-
mial. (See Theorem 4.3.)
(4) R = κ[X]/I2 (X) where I2 (X) is the ideal generated by all 2 × 2 minors of the
m × n matrix X of indeterminates. In this case, R (q) is a polynomial. (See
Sect. 3.6.)
5 Examples in Normal Affine Semigroup Rings
In this section, we present several examples where the Hilbert–Kunz functions

are obtained as generalized Erhart quasipolynomials by counting the lattice points
inside the semi-open rational polycell P as described in Sect. 4. First we count the
points for the family of rational normal cones of arbitrary degree in the most intuitive
way. Then we use another 2-dimensional example to illustrate a Macaulay2
simulation. We also present some results and discussions on the homogeneous
coordinate ring of Hirzebruch surfaces and its relative variations. The calculations
presented here are not performed with BG decompositions. However, we provide
some comments for those who are interested in pursuing this direction. Throughout
the section, we assume e 1.
154 C-Y. J. Chan
Example 5.1 Consider C g = Spec Rg ⊂ Cg+1 , the rational normal cone of degree
g in characteristic p: Rg = k[s, st, · · · , st g ] or k[x0 , · · · , xg ]/J where J denotes

x0 x1 · · · xg−1
the ideal generated by the maximal minors of the matrix . The
x1 x2 · · · xg
graph in the previous Fig. 3 illustrates C 2 . The Hilbert–Kunz function of Rg with
respect to the maximal graded ideal is
g+1 1
ϕRg (e) = ( )(pe )2 + (−ve2 + ve g − g + 1)
2 2
where ve ∈ {0, . . . , g − 1} is the congruence class of pe − 1 modulo g.

We present a computation by an elementary approach. Observe Fig. 3b for e = 2.
The number of the lattice points in p2 · P ∩ Z2 can be obtained by counting those
in &OT V without the right vertical boundary and the smaller regions &T U W ,
&U V Q without the upper and right boundaries. For arbitrary g, there will be exactly
g identical smaller triangular shapes. The count in &OT V , denoted Ae , is given by
Ae = 1 + (1 + g) + (1 + 2g) + · · · + (1 + (pe − 1)g)

= pe + g [1 + 2 + · · · + (pe − 1)]
= g2 (pe )2 + (1 − g2 )(pe ).
The number of the lattice points contained in the identical smaller triangles &T U W
and &U V Q, denoted Be , is
Be = (pe − 1) + (pe − 1 − g) + (pe − 1 − 2g) + · · · + (pe − 1 − ug),
where u is a non-negative integer such that pe − 1 = ug + ve with ve as described

above. After replacing u by (pe − 1 − ve )/g, Be is simplified to be
Be = (u + 1)(pe − 1) − g · (1+u)u 2
= 2g
1
(pe )2 + ( 12 − g1 )(pe ) − 2g
1
(ve2 − ve g + g − 1).
Thus the Hilbert–Kunz function of Rg can be expressed as
ϕR (e) = Ae + g · Be
= g2 (p e )2 + (1 − g2 )(p e ) + g[ 2g
1
(p e )2 + ( 12 − g1 )(p e ) − 1 2
2g (ve − ve g + g − 1)]
1+g
= e 2
2 (p ) + 0 · (p e ) − 1 2
2 (ve − ve g + g − 1)
with ve ≡ pe − 1 (mod g) as described.

It is a good exercise to carry out the computation for ϕRg (e) by the BG
decomposition. There will be g non-isomorphic conic divisors and each corresponds
to a 2-dimensional semiopen polycell γ . In fact, all of them are parallelograms
of the same shape and size, and for each i = 1, . . . , g, there exists exactly
one cell that appears i times in P. However, since the polytopes are rational,
polycells presenting non-isomorphic conic divisors do not necessarily have the same
Ehrhart quasipolynomials. This adds complications but interesting twists to this
exercise which leads to a better understanding of BG decompositions and Ehrhart
quasipolynomials.
Example 5.2 R = κ[s 2 t, st, st 2 ] with char κ = p. In this example, we describe an
interpolation process using the values of ϕR (e) provided by Macaulay2. In this
example for p = 2 and e = 1, we use the following code:
p=2
k=ZZ/p
T=k[s,t]
S=k[x,y,z]
f=map(T, R, {s^2*t, s*t, s*t^2})
J=ker f
R=S/J
e=1
q=pê
Ie=ideal(x^q, y^q, z^q)
degree(R/Ie)
toRR(200, oo/4ê)
The second to last command produces the values of ϕR (e) while the last
command toRR(200, oo/4ê) divides the value by (2e )2 and converts it to
the precision of 200 decimals. By Theorem 4.3, we know that ϕR (e) must be in the
form of
ϕR (e) = a (2e )2 + b (2e ) + ce
with rational constants a and b, and eventually periodic function ce . Thus the
sequence of outputs obtained by iterating the codes above with increasing e must
converge to the desired leading coefficient a. Since a is a rational number, the
sequence will eventually stabilize with apparent repeating decimals and one can
guess an expression for a as a quotient of two integers.
Once the precise value a is available, we can calculate b by iterating similar
Macaulay2 codes with the last two input commands changed to
degree(R/Ie) - a * q^2
toRR(200, oo/2ê)
We do so until the sequence of outputs stabilizes with apparent repeating decimals.
Then do the simple computation to obtain b in the form of a rational number. For
the final term, do the Macaulay2 iterations with the last two input commands
changed to
degree(R/Ie) - a * q^2 - b * q
toRR(200, oo)
156 C-Y. J. Chan
In this step, the data may form multiple groups and each group will converge to a
rational number. In higher dimensional cases, we expect such a property will hold
for the coefficients corresponding to terms of degree equal to dim R − 2. But for the
remaining terms, the trend of the data might not be so clear.
We perform this interpolation for small primes p = 2, 3, 5, 7, 11. The numerical
outputs suggest the following results
5 2 2 5 2
R (q) = q − , if p = 3; and R (q) = q , if p = 3. (5.1)
3 3 3
By doing the lattice point count as done in Example 5.1, one can show that the
Hilbert–Kunz functions of R for arbitrary primes are indeed given by (5.1) as
suggested by Macaulay2. The BG decomposition can be performed in a similar
way as commented in Example 5.1. We leave these exercises to the interested
readers.
We make two remarks regarding the above two examples.
Remark 5.3
(a) In Example 5.1, Rg (q) (i.e., ϕRg (e)) is a polynomial in q (resp. pe ) if and only
if the class of pe − 1 modulo g is independent of e. For instance, g|p − 1 or
g = p. In Example 5.2, R (q) is a polynomial for any given prime p even
though the vertices of the polycell P defined in Sect. 4 are obviously rational,
but not integral in this example.
By the generalized Ehrhart Theorem 4.2, the period of an Ehrhart quasipoly-
nomial EP (n) divides the least common multiple of the denominators of the
vertices. Since Hilbert–Kunz function is specialized only at the powers of a
prime p, the period may be smaller or even 1, but does not necessarily divide
the least common multiple. These are clearly displayed in the previous two
examples and the next one. For instance, in Example 5.1, taking p = 3 and
g = 5, then the least common multiple of the vertices is 5 but the period of the
Hilbert–Kunz function is 4.
(b) In both Examples 5.1 and 5.2, the second coefficient β vanishes. This is not a
coincidence because a simplicial cone defines a quotient singularity which can
be realized as a polynomial invariant of a finite group. Therefore, its divisor
group is a torsion group and hence β = 0 (c.f. [8, 59],[42, 2.2],[66]).
Next, suggested by U. Walther, we consider the family of Hirzebruch surfaces
Xa . Let Pa be the integral polytope with vertices: (0, 0), (1, 0), (0, 1), (1, a + 1)
in R2 . A Hirzebruch surface is a toric variety constructed from the normal fan
a associated to Pa . One can also view Xa as the Zariski closure of the set
{(1, s, t, st, . . . , st a+1 )|(s, t) ∈ (C∗ )2 } in Pa+2 . Notice that the powers of the
monomials in s, t are exactly the lattice points in Pa (c.f. [32, Example 2.3.16]).
The ring R = κ[M] in Example 5.4 is the coordinate ring of the affine cone of Xa
with M = {(1, v)|v ∈ Pa ∩ Z2 }.
Example 5.4 Set κ = Z/p. Let Sa be the affine semigroup ring arising from Xa
Sa = κ[u, us, ut, ust, ust 2 , . . . , ust a+1 ].
We compute the Hilbert–Kunz function of Sa with respect to the maximal ideal

generated by all monomials other than 1. By performing the same calculations using
Macaulay2 as done in Example 5.2, the simulations suggest
(1) a = 1,
if p = 2, S1 (q) = 74 q 3 − 18 q 2 − 14 q;
if p 3, S1 (q) = 74 q 3 − 18 q 2 − 14 q − 38 .
(2) a = 2,
if p=3, S2 (q) = 20 3

9 q − 13 q 2 ;
?
− 49 , q ≡ 2 (mod 3)
if p = 3, S2 (q) = 20 3
9 q − 13 q 2 + .
− 89 , q ≡ 1 (mod 3)
More cases are documented in Schalk [92]. The knowledge about the possible values
of the period from Theorem 4.2 is very helpful for the interpolation process of the
quasipolynomials. The vertices of the polycell within which we count the lattice
points must be the intersection of those parallel to any three support hyperplanes
of the affine cone of the semigroup generating R. By straightforward calculation,
we know that the least multiple of the denominator of these vertices is a + 1. Thus
the period of Sa (q) is bounded above by a + 1, although not necessarily dividing
a + 1, as indicated in Remark 5.3(a).
The Hilbert–Kunz function of Hirzebruch surfaces has also been considered via
the pair (X, L) by Trivedi [105] using sheaf theoretic method where X is regarded
as a nonsingular ruled space over P1 and L is an ample line bundle on X. A closed
form of the Hilbert–Kunz function is given in terms of formulas in the data provided
by the ample line bundle L.
Saikali [90] considered a variation of Sa where the semigroup ring is a k-algebra
generated by the monomials contained in the cone over the polytope Pb,h with the
vertices (0, 0, 1), (b, 0, 1), (0, h, 1) and (b + h, h, 1). Setting b = h = 1 gives the
special case in Example 5.4 with a = 1 up to an isomorphism, i.e., (Pa , 1) = Pb,h
when a = b = h = 1. For Sa in Example 5.4, all the nonzero lattice points are on the
boundary of (1, Pa ) for any a while Pb,h has lattice points in the interior if h > 1. So
the shape of the corresponding polycell P is a lot more complicated in [90] in which
the lattice points are counted by viewing P as the union of a large convex polytope
and multiple prickly smaller shapes similar to the way Example 5.1 is considered.
A closed form of the Hilbert–Kunz function is obtained by analyzing complicated
counting functions with the assistance of the 3D graphing software GEOGEBRA.
158 C-Y. J. Chan
The BG decompositions for Sa in Example 5.4 and for the variations in [90]
are much more subtle since these are nonsimplicial three-dimensional polycells and
they have not been explored in great detail.
We observe the following general facts. Let R be a normal afffine semigroup ring
and R+ be the maximal ideal consisting of all the monomials other than 1.
Remark 5.5 The Hilbert–Kunz multiplicity of an affine semigroup ring, as it can be
expressed by the volume of P, is independent of the characteristic p. Obviously the
Hilbert–Kunz function is not. Not only does the shape of ϕR (e) depend on p, but,
even in polynomial form, the function varies as p does.
For a given Sa , if q is constant modulo a + 1, then R (q) is in a polynomial
form. Otherwise, it is likely to be a quasipolynomial. This indicates that the
construction of a normal semigroup ring with a desired form of Hilbert–Kunz
function is, in theory, plausible. In practice though, in spite of Theorem 4.2 that
identifies the possible period of the Ehrhart quasipolynomial from the coordinates
of the vertices, the exact period is not immediately obvious (see references following
Question 5.8).
Furthermore, since the Ehrhart quasipolynomial of a polytope with integral
vertices is a polynomial, it is possible to construct affine semigroup rings whose
Hilbert–Kunz functions are true polynomials. In fact, this happens if all the cells
of have integral vertices. Bruns pointed out in [16, p. 71] that this is possible
if the affine semigroup M has a unimodular configuration. As we can see from
the examples above, the integral condition on the full dimensional cells are not
necessary for Hilbert–Kunz functions to be of true polynomial forms. While the
configuration of semigroups are independent of the characteristic, the functional
form of Hilbert–Kunz functions in general does depend on the characteristic.
Remark 5.6 The open cells in of the BG decomposition determine the ultimate
shape of ϕR (e). We say that a function F (e) is a polynomial up to a periodic function
if it is in the form of
F (e) = ad (pe )d + · · · + a1 (pe ) + δe
where ad , . . . , a1 are constants and δe is a periodic function. All the functions

presented in this section are in such a form. Next we call ϕR (e + 1) − ϕR (e)
the difference function of ϕR (e). By a straightforward computation, we can show
that if the difference function is a polynomial up to a periodic function, then so is
ϕR (e). This is supported by Saikali’s analysis in [90]. It is also related to the shape
of the full dimensional cells in the BG decomposition. Not all full-dimensional
cells have its Ehrhart quasipolynomial as a polynomial up to a periodic function.
It is reasonable to expect that the Hilbert–Kunz functions can be characterized
by the BG decomposition. In another words, we expect that the functional form
of Hilbert–Kunz function can be determined by the combinatorial structure of the
affine semigroup ring.
By studying the BG decompositions, one might be able to understand better the

periodic behavior of ϕR (e) as pointed out in Remarks 5.5 and 5.6. In addition, we
may challenge ourselves to find a geometric interpretation of the coefficients of
the lower degree terms, and search for useful ways to determine the period of the
quasipolynomial.
We end this paper with the following questions. An answer to Question 5.9 will
provide an effective algorithm of accessing the Hilbert–Kunz functions of normal
affine semigroup rings.
Question 5.7 As in Theorem 4.1 for the leading and the second coefficients, can
one give a meaningful geometric interpretation to the coefficient of each degree in
the polynomial form of ϕR (e)?
Question 5.8 How can the period of ϕR (e) be effectively and efficiently estimated?
An Ehrhart version of Question 5.8 is posted in Beck and Robins [6, 3.39]; see
also Woods [117], and Beck, Sam and Woods [7].
Let R = k[M] be a normal affine semigroup ring in Sect. 3.5. Note that the BG
decomposition is determined by the semigroup structure of M.
Question 5.9 Can an effective algorithm be developed for the function νγ (n)
and the minimum generators μR (Cγ ) for any full dimensional class γ in ?
Furthermore, does the semigroup structure of M characterize the BG decomposition
and vice versa?
Counting lattice points inside a polytope, though a simple and rather primitive
task, is not as easy as it appears. The desire of understanding Hilbert–Kunz function
now leads us to ask for what class of polytope can E (n) be explicitly calculated.
Starting from an affine semigroup and using the BG decomposition, it seems that
one may branch out into Hilbert–Kunz or Ehrhart theory. The results obtained from
either direction may very likely enhance each other.
Acknowledgments My sincere gratitude goes to Kazuhiko Kurano for showing me the connection
between Hilbert–Kunz functions and Ehrhart Theory. To Joseph Gubeladze for pointing to the
beautiful source [6]. To Uli Walther for discussions on the Macaulay 2 codes. To Holger Brenner
and Claudia Miller for insightful conversations offering support and many crucial references.
Special thanks to I-Chiau Huang and Academia Sinica of Taiwan who have sponsored multiple
visits throughout the course while the main ideas in this article were formed. I am grateful for the
anonymous referees for providing detailed comments and corrections that greatly improved the
manuscript.
References
1. I. ABERBACH, The existence of the F -signature for rings with large Q-Gorenstein locus, J.
Algebra 319 (2008) 2994–3005.
2. I. ABERBACH AND F. ENESCU, Lower bounds for Hilbert–Kunz multiplicities in local rings
of fixed dimension, Special volume in honor of Melvin Hochster. Michigan Math. J. 57 (2008)
1–16.
160 C-Y. J. Chan
3. I. ABERBACH AND F. ENESCU, New estimates of Hilbert–Kunz multiplicities for local rings
of fixed dimension, Nagoya Math. J. 212 (2013) 59–85.
4. M. BARCO, Dilation and Lattice Point Count, Central Michigan University Masters Plan B
Paper (Spring 2012) 25 pp.
5. B. BATSUKH AND H. BRENNER, Hilbert–Kunz Multiplicity of Binoids, arXiv:1710.05761
(2017).
6. M. BECK AND S. ROBINS, Computing the Continuous Discretely, Springer (2007).
7. M. BECK, S. SAM AND K. WOOD, Maximal periods of (Ehrhart) quasi-polynomials, J.
Combin. Theory Ser. A 115 (2008) 517–525.
8. D. BENSON, Polynomial Invariants of Finite Groups, Cambridge University Press, Cambridge
(1993).
9. M. BLICKLE AND F. ENESCU, On rings with small Hilbert–Kunz multiplicity, Proc. Amer.
Math. Soc. 132 (2004) 2505–2509.
10. H. BRENNER, The rationality of the Hilbert–Kunz multiplicity in graded dimension two, Math.
Ann. 334 (2006) 91–110.
11. H. BRENNER, The Hilbert–Kunz function in graded dimension two, Comm. Algebra 35 (2007)
3199–3213.
12. H. BRENNER, Tight closure and vector bundles, Three Lecturers on Commutative Algebra,
Univ. Lecture Ser. 42 Amer. Math. Soc., Providence, RI (2008).
13. H. BRENNER, Irrational Hilbert–Kunz multiplicity, arXiv:1305.5873vl [math.AG] (2013).
14. H. BRENNER AND G. HEIN, Restriction of the cotangent bundle to elliptic curves and
Hilbert–Kunz functions, Manuscripta Math. 119 (2006) 17–36.
15. H. BRENNER, J. LI, AND C. MILLER, A direct limit for limit Hilbert–Kunz multiplicity for
smooth projective curves, J. Algebra 372 (2012) 488–504.
16. W. BRUNS, Conic divisor classes over a normal monoid algebra, Commutative algebra and
algebraic geometry, Contemp. Math. 390 (2005) 63–71.
17. W. BRUNS AND J. GUBELADZE, Divisorial Linear Algebra of Normal Semigroup Rings
Algebr. Represent. Theory 6 (2003) 139–168.
18. W. BRUNS AND J. HERZOG, Cohen–Macaulay Rings, Cambridge Studies in Advanced
Mathematics 39, Cambridge University Press (1993).
19. R.-O. BUCHWEITZ AND Q. CHEN, Hilbert–Kunz functions of cubic curves and surfaces, J.
Algebra 197 (1997) 246–267.
20. I. BURBAN, Frobenius morphism and vector bundles on cycles of projective lines, Comm.
Algebra 40 (2012) 2983–2988.
21. A. CAMINATA AND A. DE STEFANI, F -signature function of quotient singularities, J.
Algebra 523 (2019) 311–341.
22. O. CELIKBAS, H. DAO, C. HUNEKE, AND Y. ZHANG, Bounds on the Hilbert–Kunz
multiplicity, Nagoya Math. J. 205 (2012) 149–165.
23. C-Y. J. CHAN, Filtrations of Modules, the Chow Group and the Grothendieck Group, J.
Algebra 219 (1999) 330–344.
24. C-Y. J. CHAN AND K. KURANO, Hilbert–Kunz functions over rings regular in codimension
one, Comm. Algebra 44 (2016) 141–163.
25. C-Y. J. CHAN AND K. KURANO, Cohen–Macaulay cones spanned by maximal Cohen–
Macaulay modules, Trans. Amer. Math. Soc. 368 (2016) 939–964.
26. C-Y. J. CHAN, J.-C. LIU AND B. ULRICH, The Buchsbaum-Rim multiplicity and the Hilbert-
Samuel multiplicity, J. Algebra 319 (2008), 4413–4425.
27. SH.-T. CHANG, The asymptotic behavior of Hilbert–Kunz functions and their generalizations,
Thesis, University of Michigan, 1993.
28. SH.-T. CHANG, Hilbert–Kunz functions and Frobenius functors, Trans. Amer. Math. Soc. 349
(1997) 640–659.
29. L. CHIANG AND Y. HUNG, On Hilbert–Kunz Functions of Some Hypersurfaces, J. Alge-
bra 199 (1998) 499–527.
30. A. CONCA, Hilbert–Kunz function of monomial ideals and binomial hypersurfaces,
Manuscripta math. 90 (1996) 287–300.
31. M. CONTESSA, On the Hilbert–Kunz function and Koszul homology, J. Algebra 175 (1995)
757–766.
32. D. COX, J. LITTLE AND H. SCHENCK, Toric Varieties, Graduate Studies in Mathematics 124,
AMS, Providence, 2011.
33. S. D. Cutkosky, Introduction to Algebraic Geoemtry, Graduate Studies in Mathematics 188,
34. L. DAO AND I. SMIRNOV, On the generalized Hilbert–Kunz function and multiplicity, Israel
J. Math. 237 (2020) 155–184.
35. S. DUTTA, Frobenius and multiplicities, J. Algebra 85 (1983) no. 2 424–448.
36. E. EHRHART, Sur les polyèdres rationnels homothétiques à n dimensions, C. R. Acad. Sci.
Paris 254 (1962) 616–618.
37. F. ENESCU AND K. SHIMOMOTO, On the upper semi-continuity of the Hilbert–Kunz
multiplicity, J. Algebra 285 (2005) 222–237.
38. N. EPSTEIN AND Y. YAO, Some extensions of Hilbert–Kunz multiplicity, Collect. Math. 68
(2017) 69–85.
39. K. ETO, Multiplicity and Hilbert–Kunz Multiplicity of Monoid Rings, Tokyo J. Math. 25
(2002) 241–245.
40. K. ETO AND K.-I. YOSHIDA, Notes on Hilbert–Kunz multiplicity of Rees algebras, Comm.
Algebra 31 (2003) 5943–5976.
41. N. FAKHRUDDIN AND V. TRIVEDI Hilbert–Kunz functions and multiplicities for full flag
varieties and elliptic curves, J. Pure App. Algebra 181 (2003) 23–52.
42. W. FULTON, Introduction to Toric Varieties, Princeton Univ. Press, Princeton, NJ (1993)
43. W. FULTON, Intersection Theory, second edition, Springer, Berlin (1998).
di
44. I. GESSEL AND P. MONSKY, The limit as p → ∞ of the Hilbert–Kunz multiplicity of xi ,
arXiv:1007.2004 (2010).
45. H. GILLET AND C. SOULÉ, K-théorie et nullité des multiplicités d’intersection, C. R. Acad.
Sci. Paris Sér. I Math. 300 (1985), 71–74.
46. H. GILLET AND C. SOULÉ, Intersection theory using Adams operations, Invent. Math. 90
(1987) 243–277.
47. C. HAN, The Hilbert–Kunz function of a diagonal hypersurface, Thesis, Brandeis University,
1992.
48. C. HAN AND P. MONSKY, Some surprising Hilbert–Kunz functions, Math. Z. 214 (1993)
119–135.
49. R. HARTSHORNE, Algebraic Geometry, Graduate Texts in Mathematics 52 Springer-Verlag,
New York-Heidelberg (1977).
50. D. HERNÁNDEZ AND J. JEFFRIES, Local Okounkov bodies and limits in prime characteristic,
Math. Ann. 372 (2018) 139–178.
51. M. HOCHSTER, Rings of invariants of toric, Cohen–Macaulay rings generated by monomials,
and polytopes, Ann. Math. 96 (1972) 318–337.
52. M. HOCHSTER, Tight closure theory and characteristic p methods. With an appendix to
Graham J. Leuschke. In Trends in Commutative Algebra 51 Math. Sci. Res. Inst. Pub., pages
181–210, Cambridge (2004).
53. M. HOCHSTER AND C. HUNEKE, Tight closure, invariant theory, and the Briançcon-Skoda
theorem. J. Amer. Math. Soc. (1990) 31–116.
54. M. HOCHSTER AND C. HUNEKE, Phantom homology, Mem. Amer. Math. Soc. 103 (1993).
55. M. HOCHSTER AND Y. YAO, Second coefficients of Hilbert–Kunz functions for domains,
preliminary preprint: http://www.math.lsa.umich.edu/~hochster/hk.pdf.
56. C. HUNEKE, Tight closure, parameter, ideals, and geometry. In Six Lectures on Commutative
Algebra, pages 187–329, Birkhäuser (1998).
57. C. HUNEKE, Hilbert–Kunz multiplicity and the F-signature, Commutative Algebra, 485–525,
Springer, New York (2013).
58. C. HUNEKE, G. LEUSCHKE, Two theorems about maximal Cohen–Macaulay modules, Math.
Ann. 324 (2002) 391–404.
162 C-Y. J. Chan
59. C. HUNEKE, M. MCDERMOTT AND P. MONSKY, Hilbert–Kunz functions for normal rings,
Math. Res. Letters 11 (2004) 539–546.
60. C. HUNEKE AND Y. YAO, Unmixed local rings with minimal Hilbert–Kunz multiplicity are
regular, Proc. Amer. Math. Soc. 130 (2002) 661–665.
61. S. B. IYENGAR, G. J. LEUSCHKE, A. LEYKIN, C. MILLER, E. MILLER, A. SINGH, U.
WALTHER, Twenty-Four Hours of Local Cohomology, Graduate Studies in Mathematics 87,
62. E. KUNZ, Characterizations of regular local rings for characteristic p, Amer. J. Math. 91
(1969) 772–784.
63. E. KUNZ, On Noetherian rings of characteristic p, Amer. J. Math. 98 (1976) 999–1013.
64. K. KURANO, On Roberts rings, J. Math. Soc. Japan 53 (2001) 333–355.
65. K. KURANO, Numerical equivalence defined on Chow groups of Noetherian local rings,
Invent. Math. 157 (2004) 575–619.
66. K. KURANO, The singular Riemann–Roch theorem and Hilbert–Kunz functions, J. Algebra
304 (2006) 487–499.
67. A. LANGER, Semistable sheaves in positive characteristic, Ann. Math. 159 (2004) 251–276.
68. C. MILLER, The Frobenius endomorphism and homological dimensions, Contemporary
Mathematics 331 (2003) 207–234.
69. E. MILLER AND B. STURMFELS, Combinatorial Commutative Algebra, Graduate Texts in
Math. 227, Springer, New York, 2005.
70. L. MILLER AND I. SWANSON Hilbert–Kunz functions of 2 × 2 determinantal Rings, Illinois
J. Math. 57 (2013) 251–277.
71. M. MONDAL AND V. TRIVEDI, Hilbert–Kunz density function and asymptotic Hilbert–Kunz
multiplicity for projective varieties, J. Algebra 520 (2019) 479–516.
72. P. MONSKY, The Hilbert–Kunz function, Math. Ann. 263 (1983) 43–49.
73. P. MONSKY, The Hilbert–Kunz function of a characteristic 2 cubic, J. Algebra 197 (1997)
268–277.
74. P. MONSKY, Hilbert–Kunz functions in a family: point-S4 quartics, J. Algebra 208 (1998)
343–358.
75. P. MONSKY, Hilbert–Kunz functions in a family: line-S4 quartics, J. Algebra 208 (1998) 359–
371.
76. P. MONSKY, On the Hilbert–Kunz function of zD − p4 (x, y), J. Algebra 291 (2005) 350–372.
77. P. MONSKY, Hilbert–Kunz functions for irreducible plane curves, J. Algebra 316 (2007) 326–
345.
78. P. MONSKY, Rationality of Hilbert–Kunz multiplicities; a likely counterexample, Michigan
Math. J. 57 (2008) 605–613.
79. P. MONSKY, Hilbert–Kunz theory for nodal cubics, via sheaves, J. Algebra 346 (2011) 180–
188.
80. P. MONSKY AND P. TEIXEIRA, p-fractals and power series. I. Some 2 variables results, J.
Algebra 280 (2004) 505–536.
81. P. MONSKY AND P. TEIXEIRA, p-fractals and power series. II. Some applications to Hilbert–
Kunz theory, J. Algebra 304 (2006) 237–255.
82. K. PARDUE, Nonstandard Borel-Fixed Ideals, Doctoral Thesis, Brandeis University (1994).
83. T. POLSTRA AND K. TUCKER F -signature and Hilbert–Kunz multiplicity: a combined
approach and comparison, Algebra Number Theory 12 (2018) 61–97.
84. F. PÉREZ, K. TUCKER, Y. YAO, Uniformity in reduction to characteristic p, in preparation.
85. P. ROBERTS, The vanishing of intersection multiplicities and perfect complexes, Bull. Amer.
Math. Soc. 13 (1985) 127–130.
86. P. ROBERTS, Le théorème d’intersection, C. R. Acad. Sci. Paris Sér. I Math. 304 (1987)
177–180.
87. P. ROBERTS, Multiplicities and Chern Classes in Local Algebra, Cambridge Tracts in
Mathematics 133, Cambridge University Press (1998).
88. P. ROBERTS AND V. SRINIVAS, Modules of finite length and finite projective dimension,
Invent. Math. 151 (2003) 1–27.
89. M. ROBINSON AND I. SWANSON, Explicit Hilbert–Kunz function of 2 × 2 determinantal

rings, Pacific J. Math. 275 (2015) 433–442.
90. N. SAIKALI, Ehrhart Theory on Normal Affine Semigroup Rings and Its Application to
Hilbert-Kuns Functions, Doctoral Thesis –Central Michigan University. 2018. 63 pp.
91. S. SAM, A bijective proof for a theorem of Ehrhart, Amer. Math. Monthly 116 (2009) 409–
426.
92. A. SCHALK, Hilbert–Kunz Functions for Certain Families of the Quotients of Polynomial
Rings, Central Michigan University Masters Plan B Paper (Spring 2013) 30 pp.
93. G. SEIBERT, Complexes with homology of finite length and Frobenius functors, J. Algebra
125 (1989) 278—287.
94. G. SEIBERT, The Hilbert–Kunz function of rings of finite Cohen–Macaulay type, Arch. Math.
69 (1997) 286–296.
95. J.-P. SERRE, Algèbre Locale—Multiplicités, Lecture Notes in Mathematics 11 Springer-
Verlag, New York, Berlin, Heidelberg (1961).
96. I. SHAFAREVICH, Basic Algebraic Geometry. 1. Varieties in Projective space, third ed.,
Springer, Heidelberg (2013).
97. I. SHAFAREVICH, Basic Algebraic Geometry. 2. Schemes and complex manifolds, third ed.,
Springer, Heidelberg (2013).
98. A. SINGH, The F -signature of an affine semigroup ring, J. Pure Appl. Algebra 196 (2005)
313–321.
99. I. SMIRNOV, On semicontinuity of multiplicities in families, Doc. Math. 25 (2020) 381–400.
100. R.P. STANLEY, Hilbert functions of graded algebras, Adv. in Math. 28 (1978) 57–83.
101. R.P. STANLEY, Enumerative Combinatorics, Vol. I, 2nd Ed., Cambridge Uni. Press (2012).
102. P. TEIXEIRA, p-Fractals and Hilbert–Kunz Series, Thesis, Brandeis University, 2002.
103. V. TRIVEDI, Semistability and Hilbert–Kunz multiplicity for curves, J. Algebra 284 (2005)
627–644.
104. V. TRIVEDI, Hilbert–Kunz multiplicity and reduction mod p, Nagoya Math. J. 185 (2007)
123–141.
105. V. TRIVEDI, Hilbert–Kunz functions of a Hirzebruch surface, J. Algebra 457 (2016) 405–430.
106. V. TRIVEDI, Asymptotic Hilbert–Kunz multiplicity, J. Algebra 492 (2017) 498–523.
107. V. TRIVEDI, Hilbert–Kunz density function and Hilbert–Kunz multiplicity, Trans. Amer.
Math. Soc. 370 (2018) 8403–8428.
108. V. TRIVEDI, Toward Hilbert–Kunz density functions in characteristic 0, Nagoya Math. J. 235
(2019) 3331–3338.
109. V. TRIVEDI AND K.-I. WATANABE, Hilbert–Kunz density functions and F -thresholds, J.
Algebra 567 (2021) 533–563.
110. K. TUCKER, F -signature exists, Invent. Math. 190 (2012) 488–504.
111. M. VON KORFF, F -signature of affine toric varieties, arXiv:1110.0552 (2011).
112. K.-I. WATANABE, Hilbert–Kunz multiplicity of toric rings, Proc. of The Inst. of Natural
Sciences 35 (2000) 173–177.
113. K.-I. WATANABE AND K.-I. YOSHIDA, Hilbert–Kunz multiplicity and an inequality between
multiplicity and colength, J. Algebra 230 (2000) 295–317.
114. K.-I. WATANABE AND K.-I. YOSHIDA, Hilbert–Kunz multiplicity of two dimensional local
rings, Nagoya Math. J. 162 (2001) 87–110.
115. K.-I. WATANABE AND K.-I. YOSHIDA, Minimal relative Hilbert–Kunz multiplicity, Illinois J.
Math. 428 (2004) 273–294.
116. K.-I. WATANABE AND K.-I. YOSHIDA, Hilbert–Kunz multiplicity of three-dimensional local
rings, Nagoya Math. J. 177 (2005) 47–75.
117. K. WOODS, Computing the period of an Ehrhart quasipolynomial, Electron. J. Combin. 12
Research Paper 34 (2005) 12 pp.
118. Y. YAO, Observations on the F -signature of local rings of characteristic p, J. Algebra 299
(2006) 198–218.
Standard Monomial Theory and Toric
Degenerations of Richardson Varieties
in Flag Varieties
Narasimha Chary Bonala, Oliver Clarke, and Fatemeh Mohammadi
Keywords Gröbner basis · Flag varieties · Schubert varieties · Richardson

varieties · Semi-standard Young tableaux · Toric degenerations
1 Introduction
The geometry of the flag variety heavily depends on the study of its Schubert
varieties. For example, they provide an excellent way of understanding the mul-
tiplicative structure of the cohomology ring of the flag variety. In this context, it
is essential to understand how Schubert varieties intersect in a general position. A
Richardson variety in the flag variety is the intersection of a Schubert variety and
an opposite Schubert variety. In [9] and [19], the fundamental properties of these
varieties are studied, including their irreducibility. Many geometric properties of
the flag variety and its subvarieties can be understood through standard monomial
theory (SMT). For example, vanishing results of cohomology groups, normality
and other singularities of Schubert varieties, see e.g. [20, Chapter 3]. Also, the
relationship between K-theory of the flag variety and SMT is established in [17].
Let K[PJ ] be the polynomial ring on the Plücker variables PJ for non-empty
subsets J of {1, . . . , n} and let In ⊂ K[PJ ] be the Plücker ideal of the flag variety
Fln . The homogeneous coordinate ring of Fln is given by K[PJ ]/In . We say that
N. Chary Bonala
Ruhr-Universität Bochum, Fakultät für Mathematik, Bochum, Germany
e-mail: narasimha.bonala@rub.de
O. Clarke
University of Bristol, School of Mathematics, Bristol, UK
e-mail: oliver.clarke@bristol.ac.uk
F. Mohammadi ()
Department of Mathematics: Algebra and Geometry, Ghent University, Ghent, Belgium
F. Mohammadi ()
Department of Mathematics and Statistics, The Arctic University of Norway, Tromsø, Norway
e-mail: fatemeh.mohammadi@ugent.be
166 N. Chary Bonala et al.
a monomial P = PJ1 . . . PJd is standard for Fln if J1 · · · Jd . A standard

monomial basis of K[PJ ]/In is a subset of standard monomials that forms a basis
for K[PJ ]/In as a vector space. Hodge in [13] provided a combinatorial rule to
choose such basis for Grassmannians in terms of semi-standard Young tableaux.
He also proved that such a basis is compatible with Schubert varieties. More
precisely, the basis elements that remain non-zero after restriction form a basis for
the quotient ring associated to the Schubert varieties. Hodge’s work is generalised to
flag varieties by Lakshmibai, Musili and Seshadri, see e.g. [20] for a more detailed
exposition.
In this paper, we investigate when the standard monomials directly restrict to
a monomial basis for the Richardson variety, thus providing a particularly simple
rule for determining standard monomial bases for particular Richardson varieties.
A standard monomial P = PJ1 . . . PJd restricts to a non-zero function on the
Richardson variety Xw v if and only if v J w for all i. We say that such a
i
monomial restricts to the Richardson variety Xw v and we often write these restricted
monomials as semi-standard Young tableaux, see Sect. 2.6. For Richardson varieties
in the Grassmannians, the restricted standard monomials always form a monomial
basis. However, this is not true for an arbitrary Richardson variety in the flag variety,
see Example 3.3. And so the conventional method for determining a monomial
basis for the coordinate ring of a Richardson variety inside the flag variety is
fairly complicated, see Sect. 2.7. Other combinatorial methods for calculating these
monomials have been explored using so-called key tableaux [21]. We restrict our
attention to the family of Richardson varieties Xw v , where (v, w) ∈ T from (3.1).
n
Our main result is the following:
Theorem 1.1 (Theorem 3.12) The restriction of standard monomials for the flag
v , for all
variety forms a standard monomial basis for the Richardson variety Xw
(v, w) ∈ Tn from (3.1).
We also observe a surprising relation between the pairs in Tn and toric
degenerations of Richardson varieties. A toric degeneration of a variety X is a flat
family f : X → A1 , where the special fiber (over zero) is a toric variety and all
other fibers are isomorphic to X. In particular, some of the algebraic invariants of
X, such as the Hilbert polynomial, will be the same for all the fibers. Hence, we
can do the computations on the toric fiber. The study of toric degenerations of flag
varieties was started in [10] by Gonciulea and Lakshmibai using standard monomial
theory. In [15], Kogan and Miller obtained toric degenerations of flag varieties using
geometric methods. Moreover, in [2] Caldero constructed such degenerations using
tools from representation theory. In [14], Kim studied the Gröbner degenerations
of Richardson varieties inside the flag variety, where the special fiber is the toric
variety of the Gelfand-Tsetlin polytope; this is a generalisation of the results of
[15]. We notice that the corresponding ideals of many such degenerations contain
monomials and so their corresponding varieties are not toric. Hence, we aim to
characterise such toric ideals. In particular, we explicitly describe degenerations of
Richardson varieties inside flag varieties, and provide a complete characterisation
for permutations leading to monomial-free ideals.
Standard Monomial Theory and Toric Degenerations of Richardson Varieties in. . . 167
Theorem 1.2 (Theorem 5.2) Every pair of permutations (v, w) in Tn from (3.1)
v in the flag variety.
gives rise to a toric degeneration of the Richardson variety Xw
The toric degenerations that we construct are Gröbner degenerations of the
ideal of the Richardson variety with respect to a weight vector which we define
in Sects. 2.4 and 2.5. Our goal is to study the initial ideal in(I (Xw v )) for each
v
Richardson variety Xw with respect to this weight vector. To do this, we consider
the result for the flag variety which is proved in [8, 18]. More precisely, the
corresponding initial ideal is toric (binomial and prime), see Theorem 2.3. We write
the initial of the Plücker ideal as in(In ) = ker(φn ), which is equal to the kernel of
a monomial map φn , see (2.4). For each Richardson variety Xw v we construct the
restriction of in(In ) to the variables {PJ : J ∈ Tw }, which are the variables which
v
do not vanish on the Richardson variety. We can readily obtain the generators of
the restriction in(In )|Twv using Lemma 2.5, in particular this ideal is generated by
degree two monomials and binomials. Our degeneration is known as the diagonal
Gröbner degeneration. In Remark 5.3, we consider the degenerations in [14] which
correspond to the antidiagonal Gröbner degeneration. We show that our methods
can be used to generalise these results and find further toric degenerations.
We have summarised our approach to understanding the relationships between
the aforementioned ideals, namely in(In )|Twv , in(I (Xw v )) and ker(φ |v ) in the
n w
following diagram.
(1.1)
Our method is to consider all possible pairs of permutations (v, w) and determine
whether the equality labelled by ‘?’ holds. We give the following complete
classification:
Theorem 1.3 (Theorem 4.1) The ideal in(In )|Twv is monomial-free if and only if
(v, w) ∈ Tn . In particular, if (v, w) ∈ Tn then in(In )|Twv is a toric ideal and
coincides with the kernel of φn |vw .
To prove this theorem, we exploit the explicit description of a generating set for
in(In )|Twv in Sect. 2.5 and the inductive structural results for elements of Tn in
Sect. 3.1. In order to determine whether we obtain a toric degeneration of the
Richardson variety, we check whether the left hand square in (1.1) commutes,
i.e. we check whether in(I (Xw v )) = in(I )| v . In Theorem 5.2 we show that the
n Tw
square does indeed commute whenever the standard monomials for the flag variety
restrict to a monomial basis for the coordinate ring for the Richardson variety,
i.e. whenever (v, w) ∈ Tn and our main theorem holds. In these cases we obtain
toric degenerations of Richardson varieties.
1.1 Outline of the Paper
In Sect. 2, we fix our notation throughout the paper and we recall the notion
of Richardson varieties and Gröbner degenerations. In Sect. 3, we study standard
monomial bases of Richardson varieties. We first construct the set Tn and introduce
the notion of a block structure for its elements. This is our main tool to prove
Theorem 3.12. In Sect. 4, we classify monomial-free ideals of from in(In )|Twv , see
Theorem 4.1. Section 5 contains the proofs of our results on toric degenerations
of Richardson varieties, in particular Theorem 5.2. In Remark 5.3 we perform
calculations for all Richardson varieties in Fl4 . We show how our methods can be
used to study initial ideals with respect to different weight vectors. In particular, we
outline how our method can refine the results from [14].
2 Preliminaries
Throughout we fix an algebraically closed field K and write [n] for the set
{1, . . . , n}. We denote the symmetric group on n symbols by Sn and for any
w ∈ Sn we write w = (w1 , . . . , wn ), where wi = w(i) for each i ∈ [n]. We
fix w0 := (n, n − 1, . . . , 2, 1) for the longest product of adjacent transpositions in
Sn . The permutations of Sn act naturally on the left of subsets of [n]. So, for each
I = {i1 , . . . , ik } ⊂ [n], we have w0 I = {n+1−i1 , . . . , n+1−ik } which is obtained
by applying the permutation w0 element-wise to I . We use for the natural partial
order on the subsets of [n] given by
{i1 < · · · < is } {j1 < · · · < jt } if s t and i1 j1 , . . . , it jt .
We recall the Bruhat order on Sn , which is given by
(v1 , . . . , vn ) (w1 , . . . , wn ) if {v1 , . . . , vk } {w1 , . . . , wk } for all k ∈ [n].
It is also convenient for us to define a comparison operator between subsets and

permutations:
{i1 , . . . , ik } (w1 , . . . , wn ) if {i1 , . . . , ik } {w1 , . . . , wk },
(v1 , . . . , vn ) {i1 , . . . , ik } if {v1 , . . . , vk } {i1 , . . . , ik }.
Remark 2.1 The comparison between subsets and permutations can be phrased
purely in terms of the Bruhat order as follows. For each subset I = {i1 < · · · <
is } ⊆ [n], let {j1 < · · · < jn−s } = [n]\I denote its complement. Then for any pair
of permutations v, w ∈ Sn we have:
I w ⇐⇒ (i1 , i2 . . . , is , j1 , . . . , jn−s ) w and

v I ⇐⇒ v (is , is−1 . . . , i1 , jn−s , . . . , j1 ).
2.1 Flag Varieties
A full flag is a sequence of vector subspaces of Kn :
{0} = V0 ⊂ V1 ⊂ · · · ⊂ Vn−1 ⊂ Vn = Kn
where dimK (Vi ) = i. The set of all full flags is called the flag variety and denoted
by Fln , which is naturally embedded in a product of Grassmannians. Here, we
consider the structure of algebraic variety on Fln induced from the product of
Grassmannians. We view the full flag variety Fln as a homogeneous space for the
group SL(n, K) of complex n × n matrices with determinant one. Precisely, there
is a natural transitive action of SL(n, K) on the flag variety Fln which identifies the
variety Fln with the set of left cosets SL(n, K)/B, where B is the stabiliser of the
standard flag 0 ⊂ e1 ⊂ · · · ⊂ e1 , . . . , en = Kn . Here, note that B is the subgroup
of SL(n, K) consisting of upper triangular matrices. Given a permutation w ∈ Sn ,
we denote by σw the n × n permutation matrix with 1’s in positions (w(i), i) for all
i. By the Bruhat decomposition, we can write the aforementioned set of cosets as
@
Fln = SL(n, K)/B = Bσw B/B.
w∈Sn
The spaces Bσw B/B are all affine and are called Bruhat cells. Similarly, for the
subgroup of lower triangular matrices B − , the homogeneous space Fln can be
decomposed as
@
Fln = B − σv B/B.
v∈Sn
2.2 Richardson Varieties
Let v, w ∈ Sn . We define the Richardson variety Xw v associated to v, w as the
intersection of Schubert variety Xw and opposite Schubert variety Xv inside the

flag variety Fln . More precisely, the Schubert and opposite Schubert varieties are
defined as the Zariski closure of the corresponding cells in the aforementioned
decomposition, namely:
Xw = Bσw B/B ⊆ Fln and Xv = B − σv B/B ⊆ Fln .
Note that Xwv is nonempty if and only if v w with respect to the Bruhat order, see
Sect. 2.6. Moreover, the dimension of Xw v is given by dim(X v ) = N(w) − N(v),

w
where N (w) is the inversion number of w, i.e. the total number of pairs (i, j ) ∈
[n] × [n] such that i < j and w(i) > w(j ). which we denote by N(w).
We also note that the opposite Schubert variety Xv can be observed as a translate
w0 Xw0 v of the Schubert variety Xw0 v since B − = σw0 Bσw0 . Moreover, Xwid = X
w
and Xw0 = X .
v v
2.3 Ideals of Flag Varieties and Richardson Varieties
Every point in the flag variety Fln can be represented by an (n − 1) × n matrix

X = (xi,j ) of full rank. Let K[xi,j ] be the polynomial ring on the variables xi,j . The
ideal of the flag variety Fln , denoted by In , is the kernel of the polynomial map
ϕn : K[PJ : ∅ = J [n]] → K[xi,j ] (2.1)
sending each variable PJ to the determinant of the submatrix of X with row indices
1, . . . , |J | and column indices in J . We call the variables PJ of the ring Plücker
variables and their images ϕn (PJ ) Plücker forms. We also call In the Plücker ideal
of the flag variety Fln .
Given v w in Sn , we define the collection of subsets Twv = {J ⊂ [n] : v
J w} and its complement Swv = {J ⊆ [n]}\Twv . The comparison of subsets and
elements of Sn , along with the Bruhat order on Sn is given in Sect. 2.6. Then the
associated ideal of the Richardson variety Xw v is
v
I (Xw )=(In +PJ : J ∈ Swv )∩K[PJ : J ∈ Twv ]=(I (Xw )+I (Xv ))∩K[PJ : J ∈ Twv ].
(2.2)
We now give an example of the subsets Swv and Twv (see [16, §3.4] for more
details).
Example 2.2 Let n = 4. Consider the permutations v = (2314) and w = (4231).
The subsets of [n] in Twv of size one are given by those entries that lie between
v1 = 2 and w1 = 4, which are 2, 3 and 4. The subsets of size two are those that
lie between {v1 , v2 } = 23 and {w1 , w2 } = 24 which are 23 and 24. The subsets of
size three are those which lie between {v1 , v2 , v3 } = 123 and {w1 , w2 , w3 } = 234
which are all possible three-subsets. So we have:
(2314) (2314)
T(4231) = {2, 3, 4, 23, 24, 123, 124, 134, 234} and S(4231) = {1, 12, 13, 14, 34}.
2.4 Gröbner Degenerations of In
We first fix our notation throughout this section. We fix the (n − 1) × n matrix M
with entries:
Mi,j = (i − 1)(n − j + 1) (2.3)

For instance, when n = 5 we have the following matrix 4 × 5 matrix:

⎡ ⎤
0 0 00 0
⎢5 4 32 1⎥
M=⎢
⎣10
⎥
8 64 2⎦
15 12 96 3
Let X = (xi,j ) be an (n−1)×n matrix of indeterminates. For each k-subset J of [n],

the initial term of the Plücker form ϕn (PJ ) ∈ K[xij ] denoted by inM (PJ ) is the sum
of all terms in ϕn (PJ ) of the lowest weight, where the weight of a monomial m is
the sum of entries in M corresponding to the variables in m. By [3, Proposition 2.7],
the initial term inM (PJ ) is the leading diagonal term of the minor ϕn (PJ ) for each
subset J ⊆ [n]. Explicitly, if J = {j1 < · · · < jk } then we have inM (PJ ) =
x1,j1 x2,j2 . . . xk,jk . The weight of each variable PJ is defined as the weight of each
term of inM (PJ ) with respect to M, and it is called the weight induced by M. We
write wM for the weight vector induced by M on the Plücker variables.
Throughout this note, we will write in(In ) for the initial ideal of In with respect
to wM . In the following theorem, we summarise some of the important properties of
in(In ) from [8]. See also Theorem 14.16 in [18] in which in(In ) is realised as a Hibi
ideal [12] associated to the poset whose underlying set consists of Plücker variables.
Theorem 2.3 (Theorem 3.3 and Corollary 4.13 in [8]) The ideal in(In ) is gener-
ated by quadratic binomials. Moreover, it is toric and it is equal to the kernel of the
monomial map:
φn : K[PJ : ∅ = J [n]] → K[xij ] with PJ → inM (PJ ). (2.4)
v)
2.5 Gröbner Degenerations of I (Xw
For the Richardson variety Xw v we project the weight vector w

M induced by the
matrix M in (2.3) to the coordinates corresponding to the variables in the polynomial
ring K[PJ : J ∈ Twv ] and study its corresponding initial ideal inw (I (Xw v )) and its
relation to the kernel of the monomial following map obtained by restricting the
map φn from (2.4) to the polynomial ring K[PJ : J ∈ Twv ] as follows:
φn |vw : K[PJ : J ∈ Twv ] → K[xij ] with PJ → inM (PJ ). (2.5)
To simplify our notation we will omit the weight vector wM from the initial ideals
and write in(I (Xw v )). We also introduce the following notation to simplify the
description of our ideals.

Notation 2.4 Let G ⊂ K[PJ : ∅ = J [n]] a collection of polynomials and T
be a collection of subsets of [n]. We identify T with the characteristic vector of T c ,

i.e. TJ = 1 if J ∈ T otherwise TJ = 0. For each g ∈ G we write g = α cα Pα and
define

ĝ = cα Pα and G|T = {ĝ : g ∈ G} ⊆ K[PJ : J ∈ T ].
T ·α=0
We call G|T the restriction of the ideal G to T . It is useful to think of G|T
as the set obtained from G by setting the variables {PJ : J ∈ T } to zero. We
say that the variable PJ vanishes in the ideal G|T if J ∈ T . Similarly, we say
that a polynomial g ∈ K[PJ : J ∈ T ] vanishes in the restricted ideal G|T if
g ∈ PJ : J ∈ T . The ideal G|T can be computed in Macaulay2 [11] as
an elimination ideal using the following command
eliminate(G + PJ : J ∈ T , {PJ : J ∈ T }).
Lemma 2.5 With the notation above we have:

(i) G|T = G ∪ {PJ : J ∈ T } ∩ K[PJ : J ∈ T ].
(ii) Let v, w ∈ Sn . Then the ideal in(In )|Twv is generated by quadratic binomials.
Proof Part (i) is [7, Lemma 6.3]. To prove (ii) we first note that by Theorem 2.3,
there exists a set of quadratic binomials G generating the ideal in(In ). Hence, the
assertion follows immediately from (i). In particular, we have that
in(In )|Twv = G|Twv = G ∪ {PJ : J ∈ Swv } ∩ K[PJ : J ∈ Twv ] (2.6)
which completes the proof of (ii).

2.6 Permutations, Tableaux and Their Combinatorial

Properties
In this section we introduce and prove some basic facts about semi-standard Young
tableaux and their defining chains. We use these to study standard monomial bases
for Richardson varieties in Sect. 3. We begin by recalling, from the beginning of
Sect. 2, that denotes the Bruhat order on Sn and a comparison operator between
subsets of [n] and permutations.
A semi-standard Young tableau T is a sequence of subsets I1 , . . . , Id of [n] such
that I1 · · · Id . Each subset Ij is called a column of T and we will write
this T = [I1 . . . Id ]. For each pair of permutations v, w, we define SSY Td (v, w)
to be the collection of all semi-standard Young tableau T = [I1 . . . Id ] such that
v Ik w for all k ∈ [d]
Example 2.6 It is often convenient to draw T in a diagram, for example if T =
[I1 I2 I3 ] = [125, 246, 35] then the corresponding diagram has columns I1 , I2 , I3
and is drawn:
123
T = 245
56
Note that such diagrams are defined by: columns with weakly decreasing length,
weakly increasing entries in each row and strictly increasing entries in each column.
Definition 2.7 Let T be a semi-standard Young tableau with columns I1 , . . . , Id .
Let u = (u1 , . . . , ud ) be a sequence of permutations and write uk =
(uk,1 , . . . , uk,n ) ∈ Sn for each k ∈ [d]. We say that u is a defining chain for T
if the permutations are monotonically increasing u1 u2 · · · ud with respect
to the Bruhat order and for each k ∈ [d] we have Ik = {uk,1 , . . . , uk,|Ik | }.
There is a natural partial order on the set of defining chains for a given semi-
standard Young tableau T . Let π = (π 1 , . . . , π t ) and σ = (σ 1 , . . . , σ t ) be defining
chains for T . We say π σ if π k σ k for all k ∈ [t]. It turns out that there exists
a unique minimum w− (T ) = (w1− , . . . , wd− ) and a unique maximum w+ (T ) =
(w1+ , . . . , wd+ ) defining chains for T . When the tableau is not clear from the context,
we write wi+ (T ) for wi+ and wi− (T ) for wi− .
The following notation is particularly useful for describing the permutations w1+
and w2− .
Notation 2.8 Let P = (P1 , P2 , . . . , Pk ) be a k-partition of [n] where Pi are non-
empty and disjoint subsets of [n]. Write Pi = {pi,1 < pi,2 < · · · < pi,|Pi | } for each
i, and define the permutations:
↑ ↑ ↑
(P1 , P2 , . . . , Pk )=(p1,1 , p1,2 , . . . , p1,|P1 | , p2,1 , . . . , p2,|P2 | , p3,1 , . . . , pk,|Pk | ),
↓ ↓ ↓
(P1 , P2 , . . . , Pk )=(p1,|P1 | , p1,|P1 |−1 , . . . , p1,1 , p2,|P2 | , . . . , p2,1 , p3,|P3 | , . . . , pk,1 ).
Note that the set Pk is determined uniquely by P1 , . . . , Pk−1 . So we write

↑ ↑ ↓ ↓
(P1 , . . . , Pk−1 , ↑) and (P1 , . . . , Pk−1 , ↓) for the above permutations respectively.
If any of the parts Pi = {pi,1 } are singleton sets then we omit the arrow on that part
from the notation.
We proceed by proving some basic properties of these permutations from
partitions and their relationship to minimum and maximum defining chains.
Proposition 2.9 Suppose P1 , P2 , P3 is a 3-partition of [n]. Let v, w be permuta-
tions such that P1 w and v P1 ∪ P2 .
↑ ↑ ↑
• If (P1 , P2 , P3 ) w then P1 ∪ P2 w.
↓ ↓ ↓
• If v (P1 , P2 , P3 ) then v P1 .
Proof For the permutation w, the proof follows from the fact that if P1 w and
↑ ↑ ↑
P1 ∪ P2 w then (P1 , P2 , P3 ) w. For the permutation v, the proof follows
↓ ↓ ↓
from the fact that if v P1 and v P1 ∪ P2 then v (P1 , P2 , P3 ).

Proposition 2.10 Let T = [I J ] be a semi-standard Young tableau with two
↑ ↓ ↓
columns I and J . We have that w2− = (J ↑ , I− , ↑) and w1+ = (I \I+ , I+ , ↓) for
some subsets I+ , I− ⊆ I .
Proof For any permutation w = (w1 , . . . , wn ) ∈ Sn and k n, we write w([k]) =
{w1 , . . . , wk }. We define s = |J | and t = |I | for the size of the columns of T .
Note that w1− = (I ↑ , ↑) is the smallest permutation such that w1− ([t]) = I , i.e. for
all permutations v with v([t]) = I we have w1− v. Similarly w2+ = (J ↓ , ↓) is
the greatest permutation such that w2+ ([s]) = J . It follows from the definition that
w2− is the smallest permutation such that w1− w2− and w2− ([s]) = J . It easily
follows that w2− has the desired form. Similarly, by definition, w1+ is the greatest
permutation such that w1+ w2+ and w1+ ([t]) = I . And so w1+ has the desired
form.

2.7 Standard Monomials
The Plücker algebra of the flag variety is given by K[PJ ]J ⊂[n] /In , where In is the
Plücker ideal. A monomial P = PJ1 . . . PJd is called standard for Fln if J1 · · ·
Jd . A standard monomial basis of a Richardson variety is a collection of standard
monomials which form a basis for the corresponding Plücker algebra. We refer to
[20, §2.2] for more details. To simplify our notation, we identify the semi-standard
Young tableau T = [J1 , . . . , Jd ] with the monomial PJ1 . . . PJd ∈ K[PJ ], following
the notation of Sect. 2.6. The standard monomials for Richardson varieties can be
determined by minimum and maximum defining chains, as follows (see e.g. [17,
Theorem 34]).
Theorem 2.11 Let v w be permutations. The collection of semi-standard Young
− +
| w and v w1 forms a monomial basis for Xw , where
tableaux T such that w|T v
− − + +
w− (T ) = (w1 , . . . , w|T | ) and w+ (T ) = (w1 , . . . , w|T | ) are the unique minimum
and maximum defining chains for T , respectively.
3 Standard Monomials
The description of the standard monomials for Richardson varieties in Theorem 2.11
can be combinatorially difficult to determine. The goal of this section is to prove
Theorem 3.12 which is our main result and gives a very simple description of the
standard monomials for the Richardson varieties Xw v with (v, w) ∈ T . We note
n
that, in contrast to the Grassmannian, the monomials associated to the tableaux
SSY Td (v, w) may not constitute a monomial basis for the Richardson variety Xw v
inside the flag variety, see Example 3.3. However, if (v, w) ∈ Tn then Theorem 3.12
shows that the semi-standard Young tableaux SSY Td (v, w) do in fact form a
standard monomial basis for the Richardson variety Xw v.
We will now introduce the set of pairs of permutations Tn inductively. In order to

define this set we require the following. For any permutation w = (w1 , . . . , wn ) ∈
Sn with wt = n for some t ∈ [n], we define the induced permutation w =
(w1 , . . . , wt−1 , wt+1 , . . . , wn ) ∈ Sn−1 .
Definition 3.1 (Compatible Pairs) Let v, w ∈ Sn with n = vt = wt and n − 1 =
vs = ws . We say that (v, w) is a compatible pair if either (i) t = t or (ii) t < t and
in this case the following conditions hold:
s t, t s, n = wt > wt +1 > · · · > wt , and n = vt > vt−1 > · · · > vt .
We define the set of pairs of permutations Tn ⊆ Sn × Sn inductively by
T1 = {(id, id)}, Tn+1 = {(v, w) ∈ Sn+1 × Sn+1 : (v, w) ∈ Tn and (v, w) compatible}.
(3.1)
!
Example 3.2 Consider (1, 3, 2), (2, 3, 1) in T3 . We find all (v, w) in T4 with v =
(1, 3, 2) and w = (2, 3, 1). Firstly, if we have 4 = vt = wt for some t then we have
the following pairs:
! !
(1, 3, 2, 4), (2, 3, 1, 4) , (1, 3, 4, 2), (2, 3, 4, 1) ,
! !
(1, 4, 3, 2), (2, 4, 3, 1) , (4, 1, 3, 2), (4, 2, 3, 1) .
Secondly, for a compatible pair (v, w) such that vt = 4, wt = 4, assume that t = t .

Since v w we have that t > t . So we get the pair: ((1, 3, 4, 2), (2, 4, 3, 1)).
Example 3.3 Consider Fl3 and w = (3, 1, 2) ∈ S3 . Note that P23 P1 = P13 P2 −
P12 P3 . Consider the tableaux below:
13 12
T1 = and T2 = .
2 3
So we have P23 P1 = T2 − T1 . We have that P23 vanishes on Xw , since {2, 3}

{3, 1} and by the defining ideal of the Schubert variety Xw . And so T1 and T2 are
equal in the coordinate ring of the Schubert variety Xw , in particular T1 and T2 are
linearly dependent. Therefore, the set SSY T1 (id, w) is not a monomial basis for the
Schubert variety Xw = Xw id .
3.1 Block Structure on Compatible Permutations
To prove Theorem 3.12, we need to introduce a block structure on the pairs (v, w) ∈
Tn .
Definition 3.4 Let (v, w) ∈ Tn and vd = we = n for some e, d ∈ [n]. A block
of (v, w) is a pair of consecutive subsets of v and w on the same indices which
j
we write as (v, w)i = ({vi , vi+1 , . . . , vj }, {wi , wi+1 , . . . , wj }) for some i j
and satisfies one of the following criteria. Note that the persistence and expansion
criteria are defined inductively on n.
j
• (Creation) If i = j = e = d then (v, w)i is a block.
j
• (Persistence) Assume that n ∈ / {vi , . . . , vj } and n ∈
/ {wi , . . . , wj }. If (v, w)i is a
j
block for (v, w) then (v, w)i is a block for (v, w).
• (Expansion) Assume that n ∈ {vi , . . . , vj } and n ∈ {wi , . . . , wj }. In addition,
assume that i < d and e < j . If
j −1
(v, w)i = ({vi , . . . , vd−1 , vd+1 . . . , vj }, {wi , . . . , we−1 , we+1 , . . . , wj })
j
is a block for (v, w) then (v, w)i is a block for (v, w).
j
The size of a block (v, w)i is equal to j − i + 1.
Example 3.5 Here we give two examples illustrating properties of blocks.
• Let v = (3, 5, 6, 4, 1, 2) and w = (4, 6, 5, 3, 2, 1) then there are three distinct
blocks
(v, w)65 = ({1, 2}, {2, 1}), (v, w)41 = ({3, 5, 6, 4}, {4, 6, 5, 3}),
(v, w)32 = ({5, 6}, {6, 5}).
We see that blocks are either disjoint: such as (v, w)65 and (v, w)41 , or subsets of
one another: such as (v, w)41 and (v, w)32 .
• For v = (1, 2, 4, 5, 3) and w = (2, 4, 5, 3, 1) the only block is (v, w)51 . Note
that, 1 < 2 < 4 < 5 is an increasing sequence in v and 5 > 3 > 1 is a decreasing
sequence in w.
j
Definition 3.6 We say two distinct blocks (v, w)i and (v, w)
k are crossing if i <
k < j <
or k < i <
< j . Otherwise, they are called non-crossing and must be
either disjoint: i.e. j < k or
< i, or contained in one another: i.e. i < k <
< j
or k < i < j <
.
Proposition 3.7 Let (v, w) ∈ Tn , then the following hold:
j
(a) For any block (v, w)i we have {vi , . . . , vj } = {wi , . . . , wj } and vi = wj =
min{vi , . . . , vj }.
(b) Any pair of distinct blocks are non-crossing.

Proof For (a) we proceed by double induction: first on n and then on the size of the
block which is j − i + 1. If the block has size 1, i.e. i = j , then the result holds
j
trivially. So let us assume that j − i 1. If (v, w)i is a block for (v, w) then the
j
results follows by induction on n. Otherwise if (v, w)i is not a block of (v, w) then
j
(v, w)i must occur by the expansion criterion and so by induction the result holds.
Part (b) follows easily by induction n, noting that if two distinct blocks are
crossing then each has size at least two and must arise, either by the persistence
or expansion criteria, from a pair of crossing blocks of (v, w).

Definition 3.8 Let (v, w) ∈ Tn . By Proposition 3.7(b) we have that the blocks
containing vd = we = n are totally ordered by inclusion. The smallest such block
is called the maximum block of (v, w).
j
Proposition 3.9 Let (v, w) ∈ Tn and (v, w)i be the maximum block of (v, w).
Write vd = we = n for some d, e ∈ [n]. Then we have that vi < vi+1 < · · · < vd
and we > we+1 > · · · > wj .
Proof Throughout the proof we write we = vd = n − 1 for some e , d ∈ [n]. We
proceed by double induction, first by induction on n and then by induction on the
size of the maximum block. For any n, if the maximum block has size one then the
result trivially holds. For the induction step, assume the maximum block has size at
least two. It immediately follows that e < d, so by compatibility of v and w, we
have: ve < · · · < vd , we > · · · > wd , d e and e d.
Let us write (v, w)k
−1 for the maximum block of (v, w) for some k,
∈ [n],
which is the smallest block containing n − 1. Since d e and e d, by the
j
expansion criterion, we have that (v, w)
k is a block containing n. Since (v, w)i is
j
the maximum block of (v, w) we have that (v, w)
k is contained in (v, w)i , i.e. k i
and j
. Since n is contained in the maximum block of (v, w), we must have that
j −1
the maximum block is obtained from the block (v, w)i in (v, w) by the expansion
j −1
criterion. Therefore, (v, w)i is a block contained in the maximum block of (v, w).
However, it can be shown directly from the definition that the maximum block does
j −1
not properly contain any other blocks. Therefore, (v, w)i = (v, w)k
−1 is the
maximum block and so by induction we have vi < · · · < vd and we > · · · > wj .
By compatibility we have ve < · · · < vd and we > · · · > wd . Since d e and
e d, it follows that vi < · · · < vd and we > · · · > wj .

j
Proposition 3.10 Let (v, w) ∈ Tn and (v, w)i be the maximum block of (v, w).
Then we have {vi , vi+1 , . . . , vj } = {wi , wi+1 , . . . , wj } = {vi , vi+1 , . . . , n} =
{wi , wj +1 , . . . , n}.
Proof By Proposition 3.7(a), it suffices to show that {vi , vi+1 , . . . , vj } =
{vi , vi+1 , . . . , n}. We proceed by induction on the size of the maximum block
j
j − 1 + 1. If the maximum block has size 1, then (v, w)i = ({n}{n}) and the result
holds trivially.
Assume that the maximum block has size j − i + 1 ≥ 2 and write vd = we = n

and we = vd = n − 1 for some d, d , e, e ∈ [n]. Since (v, w) ∈ Tn we have that
v ≤ w and so e ≤ d. If e = d, then (v, w)dd is a block that is properly contained in
j
the maximum block (v, w)i , a contradiction. So we must have e < d. In particular,
j −1
e < j and i < d. By the expansion criterion, we have that (v, w)i is a block for
(v, w).
j −1
We proceed by showing that (v, w)i is the maximum block for (v, w). To
do this, we first show that n − 1 ∈ {vi . . . , vj }. Assume by contradiction that
n − 1 ∈ {vi , . . . , vj } = {wi . . . , wj }. Since (v, w) are compatible, we have
that d ≥ e and e ≤ e. Therefore d ≥ j + 1 and e ≤ i − 1. By the
j −1
persistence criterion, we have that (v, w)i is a block for (v, w). We write
j −1
(v, w)i = ({ṽi , . . . , ṽj −1 }, {w̃i , . . . , w̃j −1 }). Since i < d ≤ j < d , we
have that {ṽi , . . . , ṽj −1 } = {ṽi , . . . , ṽj } \ {n}. Since e < i ≤ e < j , we have
that {w̃i , . . . , w̃j −1 } = {w̃i+1 , . . . , w̃j +1 } \ {n}. By Proposition 3.7(a) applied
j
to (v, w)i we have that {vi , . . . , vj } = {wi , . . . , wj }. Since (v, w) ∈ Tn , we
j −1
have that (v, w) ∈ Tn−2 . So, by Proposition 3.7(a) applied to (v, w)i , we
have that {ṽi , . . . , ṽj −1 } = {w̃i , . . . , w̃j −1 }. However, by the above we have
wi ∈ {ṽi , . . . , ṽj −1 } but wi ∈ {w̃i , . . . , w̃j −1 }, a contradiction.
j −1
Next, we show that (v, w)i does not properly contain another block. Assume
j −1
by contradiction that (v, w)i is a block containing n−1 and is properly contained
j −1
in (v, w)i . By compatibility, we have that d e and ve < ve+1 < · · · < vd = n.
It follows that d d − 1, and so j d. Similarly, by compatibility, we have
e d and n = we > we−1 > · · · > wd . If follows that e e + 1, and so i e.
j
By the expansion criterion, we have that (v, w)i is a block for (v, w) that is strictly
contained in the maximum block, a contradiction.
j −1
So we have shown that (v, w)i is the maximum block of (v, w). The result
follows immediately by induction.

3.2 Proof of Main Result
In this section we prove Theorem 3.12. To do this we require the following

construction. Let (v, w) ∈ Tn be a pair of permutations and write we = vd = n for
some integers e and d. Assume that e < d. Define the permutations
w = (w1 , . . . , we−1 , we+1 , we , we+2 , . . . , wn ) and

v = (v1 , . . . , vd−2 , vd , vd−1 , vd+1 , . . . , vn ).
Lemma 3.11 For a given pair (v, w) ∈ Tn , we have (v , w), (v, w ) ∈ Tn .

Proof We note that v = v and w = w so it suffices to check that (v , w) and

(v, w ) are compatible pairs. Define e , d ∈ [n] such that we = vd = n − 1. Since
(v, w) are compatible, we have d d − 1 and e e + 1. If e = d − 1 then we
have that the position of n in each of the pairs (v, w ) and (v , w) is the same and so
each is a compatible pair. If e < d − 1 then we have that e + 1 d − 1 d and
so (v, w ) is a compatible pair, and similarly we have d − 1 e + 1 e and so
(v , w) is a compatible pair.

This construction is useful for our inductive argument in the proof of Theo-
rem 3.12. In particular, we will induct on the dimension of the Richardson variety
Xwv which can be read combinatorially from the inversion numbers N(v) and N(w)
of the permutations v and w respectively. Note that N(v ) = N(v) + 1 and

N (w ) = N (w) − 1. Therefore, dim(Xw v ) = dim(X v ) = dim(X v ) − 1.
w w
Theorem 3.12 Let d 1 be a natural number and (v, w) ∈ Tn be a pair of

v in degree d is equal
permutations. Then the number of standard monomials for Xw
to |SSY Td (v, w)|.
Proof Let (v, w) ∈ Tn be a pair of permutations. We note that for d = 1 the result
holds immediately. Since the ideal of the Richardson variety Xw v is generated by
homogeneous quadrics, it suffices to show that any semi-standard Young tableau

T with two columns I, J such that v I, J w is standard for the Richardson
variety Xwv . We do this by showing that v w + and w − w.
1 2
We note that if s = t then the defining permutations have a particularly simple
description, i.e. w1+ = (I ↓ , ↓) v and w2− = (J ↑ , ↑) w and the result
immediately follows. For the remaining cases, the proof proceeds by induction on
n. If n = 1 then the result is trivial. For each n > 1 we proceed by induction on the
dimension of the Richardson variety Xw v . If the dimension is zero then we have that
v = w and the result is trivial. Let us assume that v < w. We write
I = {i1 < · · · < it }, J = {j1 < · · · < js }, w = (w1 , . . . , wn ), v = (v1 , . . . , vn ).
Let e, d ∈ [n] be integers such that we = vd = n.

We proceed by taking cases on e and d.
Case 1 Assume that both e and d lie in one of the sets: {1, . . . , s}, {s + 1, . . . , t} or
{t + 1, . . . , n}. We define the tableau T with columns I , J as follows.
• If e, d ∈ {1, . . . , s} then define I = I \n and J = J \n.
• If e, d ∈ {s + 1, . . . , t} then define I = I \n and J = J .
• If e, d ∈ {t + 1, . . . , n} then define I = I and J = J .
By construction we have v I , J w. Since (v, w) ∈ Tn , by induction on n we
have T is standard for Xw . We write w − = (w − − + +
v
1 , w 2 ) and w + = (w 1 , w 2 ) for the

minimum and maximum defining sequences for T in Sn−1 . Since T is standard for
Xw we have w− +
v
2 w and v w 1 .
↑ ↓
By Proposition 2.10 we have that w − ↑ + ↓
2 = (J , I− , ↑) and w 1 = ((I \I+ ) , I+ , ↓)

in Sn−1 for some subsets I+ , I− ⊆ I . It follows by the same proposition that:
↑
• If e, d ∈ {1, . . . , s} then w2− = (J ∪ {n}↑ , I− , ↑) and w1+ = ((I \I+ ) ∪
↓
{n}↓ , I+ , ↓),
• If e, d ∈ {s + 1, . . . , t} then w2− = (J ↑ , I− ∪ {n}↑ , ↑) and w1+ = ((I \I+ )↓ , I+ ∪
{n}↓ , ↓),
↑ ↓
• If e, d ∈ {t + 1, . . . , n} then w2− = (J ↑ , I− , ↑) and w1+ = ((I \I+ )↓ , I+ , ↓).
Since w− + − +
2 w and v w 1 , it follows that w2 w and v w1 .
For the remaining cases note that we have e < d. So we recall the permutations
w = (w1 , . . . , we−1 , we+1 , we , we+2 , . . . , wn ) and

v = (v1 , . . . , vd−2 , vd , vd−1 , vd+1 , . . . , vn ).
Note that by Lemma 3.11 we have that (v , w), (v, w ) ∈ Tn .

Case 2 Assume s e t and t + 1 d. If d > t + 1 then it follows that
v I, J w and so by induction we have w2− w and v v w1+ . Therefore,
we may assume that d = t + 1. Similarly, if e < t then it follows that v I, J w
and so by induction we have w2− w w and v w1+ . So we may assume that
e = t.
Claim Either v I or I w .
To prove the claim, we proceed by taking cases on I , either n ∈ I or n ∈
/ I.
Case i Assume n ∈ I . Since wt = n, it follows that I w and so we will show
that v I . Let (v, w)i be the maximum block. Since e = t and d = t + 1 we have
j
that i t and j t +1. By Proposition 3.9 we have that vi < · · · < vt < vt+1 = n.
By Proposition 3.10, the elements appearing in the maximum block are precisely
{vi , vi+1 , . . . , vj } = {vi , vi + 1, . . . , n}. Therefore, for all k < i we have vk < vi
and so vt = max{v1 , . . . , vt }. Since n ∈ I , it follows easily that v I .
Case ii Assume n ∈ / I . Since vt+1 = n, it follows that v I and so we will

show that I w . Similarly to the above case, we consider the maximum block
j
(v, w)i . By Proposition 3.9 we have that n = wt > · · · > vj . By Proposition 3.10,
the elements appearing in the maximum block are precisely {wi , wi+1 , . . . , wj } =
{wj , wj + 1, . . . , n}. Therefore, for all k > j we have that wk < wj and so wt+1 =
max{wt+1 , . . . , wn }. Since n ∈ / I , it follows easily that I w . And so we proved
the claim.
If v I then we have v I, J w and so by induction we have T is standard
for Xwv . Hence, v v w + and w − w. Therefore, T is standard for X v . On
1 2 w
the other hand, if I w then we have v I, J w and so by induction we have
T is standard for Xw v . Hence, v w + and w − w w. Therefore, T is standard
1 2
v
for Xw .
Case 3 Assume e s and s + 1 d t. This case is identical to Case 2 by

considering the subset J instead of I .
Case 4 Assume e s and t + 1 d. We can assume as, similarly to Case 2,
that d = t + 1 and e = s. Note that s < t and so v v w w. If either
v I or J w then we can deduce the result by induction, so we will assume
that both conditions do not hold. Equivalently, we will assume that n ∈ J and
j
n∈ / I . Let (v, w)i be the maximum block of (v, w). By Proposition 3.9 we have
vi < · · · < vd and we > · · · > wj and by Proposition 3.7(a) for all k < i and k > j
we have vk < vi and wk < wj . Let us consider the size of the maximum block.
Since n ∈ {w1 , . . . , ws } we have that i s. Since n ∈ {vt+1 , . . . , vn } we have that
j t + 1. In particular, the indices of the maximum block span {s, s + 1, . . . , t + 1}.
We now show that v w1+ and w2− w. Recall by Proposition 2.10 that w1+ =
(I \I˜↓ , I˜↓ , ↓) and w2− = (J ↑ , I ↑ , ↑) for some subsets I˜ and I of I . Since vs <
· · · < vt are the largest elements of {v1 , . . . , vt } and v I , it follows that v w1+ .
Similarly, since ws > · · · > wt+1 are the largest elements in {w1 , . . . , wt } and
I, J w, it follows that w2− w.

Remark 3.13 Schubert and opposite Schubert varieties are special examples of
Richardson varieties. For these cases Theorem 3.12 has a particularly simple
combinatorial description. A Schubert variety is a Richardson varieties Xw v such
that v = id. It is easy to show that (id, w) ∈ Tn if and only if w is a 312-avoiding

permutation. On the other hand, opposite Schubert varieties are Richardson varieties
Xwv such that w = w = (n, n − 1, . . . , 1). In this case (v, w ) ∈ T if and only if
0 0 n
v is 213-avoiding.
4 Monomial-Free Ideals
We recall the definition of the ideal in(In )|Twv from Sects. 2.4 and 2.5 and the
collection of pairs of permutations Tn ⊆ Sn × Sn from (3.1). Our main result in
this section is the following which gives a complete characterisation for monomial-
free ideals of form in(In )|Twv .
Theorem 4.1 The ideal in(In )|Twv is monomial-free if and only if (v, w) ∈ Tn .
Proof The proof follows directly from Lemmas 4.3, 4.4 and 4.5. In particular,
Lemmas 4.3 and 4.4 show that if in(In )|Twv is monomial-free then in(In−1 )|Twv is
monomial-free and (v, w) is a compatible pair. And so, by induction on n, we have
that if in(In )|Twv is monomial-free then (v, w) ∈ Tn . On the other hand, Lemma 4.5
shows that if (v, w) ∈ Tn then in(In )|Twv is monomial-free.

Example 4.2 Let v = (1, 3, 2) and w = (3, 1, 2), which are non-compatible
permutations. Note that the ideal in(I2 )|Twv = 0, in particular it is monomial-free.
We also have in(I3 ) = P13 P2 − P23 P1 , Since Twv = {1, 2, 3, 13}, it follows that
the ideal in(I3 )|Twv = P13 P2 contains a monomial. This monomial arises from the
generator of in(I3 ), where P23 P1 vanishes in in(I3 )|Twv .
We now proceed to prove the lemmas used in the proof of Theorem 4.1. We will
first show, in Lemma 4.3, that
in(In+1 )|Twv monomial-free -⇒ in(In )|Twv monomial-free.
However, the converse does not hold, see Example 4.2. We will show that
compatibility is an essential ingredient in showing that in(In+1 )|Twv is monomial-
free. In Lemma 4.5, we will show that the converse holds with the added assumption
of compatibility
in(In )|Twv monomial-free and (v, w) compatible -⇒ in(In+1 )|Twv monomial-free.
Lemma 4.3 Let v, w ∈ Sn+1 with v w. If in(In+1 )|Twv is monomial-free then

in(In )|Twv is monomial-free.
Proof Suppose that in(In )|Twv contains a monomial PI PJ which arises from the
binomial PI PJ − PI PJ in in(In ). Now we construct a monomial in in(In+1 )|Twv .
Assume that |I | = |I | |J | = |J |. Let 1 t t n + 1 with vt = wt = n + 1.
We take cases on t, t , |I | and |J |.
Case 1 Assume that t > |I |. Then PI PJ − PI PJ is a binomial in in(In+1 ). It is
clear that PI PJ does not vanish in in(In+1 )|Twv and PI PJ vanishes in in(In+1 )|Twv .
Hence, PI PJ is a monomial in in(In+1 )|Twv .
Case 2 Assume |J | < t |I | < t. Note that v I, J, I ∪ {n + 1} and I, J, I ∪
{n + 1} w. So, PI , PJ and PI ∪{n+1} do not vanish in Fn+1 |vw . Since PI PJ
v
vanishes in Fn |w , we may proceed by taking cases on which of the following hold:
v I , v J , w I or w J . For each case we define sets I˜, I˜ , J˜, J˜ such
that: PI˜ PJ˜ − PI˜ PJ˜ is a binomial in inWD (Fn+1 ), PI˜ PJ˜ does not vanish, and PI˜ PJ˜
vanishes in Fn+1 |vw . It follows that Fn+1 |vw contains the monomial PI˜ PJ˜ .
Case 2.1 If either v I , v J or w J , then define I˜ = I, I˜ = I, J˜ = J and
J˜ = J . So, PI˜ PJ˜ does not vanish in Fn+1 |vw . However, either v I˜ , v J˜ or
w J˜ holds, respectively. And so PI˜ PJ˜ vanishes in Fn+1 |vw .
Case 2.2 If w I , then define I˜ = I ∪ {n + 1}, I˜ = I ∪ {n + 1}, J˜ = J and
J˜ = J . So, PI˜ PJ˜ does not vanish in Fn+1 |vw . However, we have that w I˜ .
Hence, PI˜ vanishes in Fn+1 |vw .
Case 3 Assume that |J | < t t |I |. Define I˜ = I ∪ {n + 1}, J˜ = J, I˜ =
I ∪ {n + 1}, and J˜ = J . Observe that PI˜ PJ˜ − PI˜ PJ˜ is a binomial in inWD (Fn+1 ).
Since v I˜, J˜ w, we have that PI˜ PJ˜ does not vanish in Fn+1 |vw . Since PI PJ
vanishes in Fn |w , we have that at least one of v I , v J , w I or w J
v
holds. It follows that at least one of v I , v J , w I or w J holds

respectively. And so PI˜ PJ˜ vanishes in Fn+1 |vw . Therefore PI˜ PJ˜ is a monomial in
Fn+1 |vw .
Case 4 Assume t ≤ |J | |I | < t. Note that v I, J, I ∪ {n + 1}, J ∪ {n + 1} and
I, J, I ∪ {n + 1}, J ∪ {n + 1} w. So, PI , PJ , PI ∪{n+1} and PJ ∪{n+1} do not vanish
v
in Fn+1 |vw . Since PI PJ vanishes in Fn |w , we may proceed similarly to Case 2 by
taking cases on which of the following hold: v I , v J , w I or w J .
Case 4.1 If v I or v J , then define I˜ = I, J˜ = J, I˜ = I , and J˜ = J .
It follows that v I˜ or v J˜ holds, respectively. Hence, PI˜ PJ˜ is a monomial in
Fn+1 |vw .
Case 4.2 If w I or w J , then define I˜ = I ∪ {n + 1}, J˜ = J ∪ {n + 1}, I˜ =
I ∪ {n + 1}, and J˜ = J ∪ {n + 1}. It follows that w I˜ or w J˜ holds,
respectively. Hence, PI˜ PJ˜ is a monomial in Fn+1 |vw .
Case 5 Assume that t ≤ |J | < t |I |. Note that v I ∪ {n + 1}, J, J ∪ {n + 1}
and I ∪ {n + 1}, J, J ∪ {n + 1} w. So, PI ∪{n+1} , PJ and PJ ∪{n+1} do not vanish
v
in Fn+1 |vw . Since PI PJ vanishes in Fn |w , we may proceed similarly to Case 2 by
taking cases on which of the following hold: v I , v J , w I or w J .
Case 5.1 If either v I , v J or w I , then define I˜ = I ∪ {n + 1}, J˜ =
J, I˜ = I ∪ {n + 1}, and J˜ = J . It follows that either v I˜ , v J˜ or w I˜
holds, respectively. Hence, PI˜ PJ˜ is a monomial in Fn+1 |vw .
Case 5.2 If w J , then define I˜ = I ∪ {n + 1}, J˜ = J ∪ {n + 1}, I˜ = I ∪ {n + 1},
and J˜ = J ∪{n+1}. It follows that w J˜ , hence, P ˜ P ˜ is a monomial in Fn+1 |vw .
I J
Case 6 Assume that t |J |. Define
I˜ = I ∪ {n + 1}, J˜ = J ∪ {n + 1}, I˜ = I ∪ {n + 1} and J˜ = J ∪ {n + 1}.
Then PI˜ PJ˜ − PI˜ PJ˜ is a binomial in inWD (Fn+1 ). Since v ≤ I, J ≤ w, then
v ≤ I˜, J˜ ≤ w so PI˜ PJ˜ does not vanish in Fn+1 |vw . Since PI PJ vanishes in Fn |w ,
v

we have that at least one of v I , v J , w I or w J holds. It follows that
at least one of v I , v J , w I or w J holds respectively. And so PI˜ PJ˜
vanishes in Fn+1 |vw . Therefore PI˜ PJ˜ is a monomial in Fn+1 |vw .

For the following lemma, recall that for any pair of permutations (v, w) ∈
Sn+1 × Sn+1 , we denote vt = wt = n + 1 and vs = ws = n. In addition we
write (w1 , . . . , wk )↑ for the ordered list whose elements are {w1 , . . . , wk } taken in
increasing order.
Lemma 4.4 Let v, w ∈ Sn+1 with v w. If in(In+1 )|Twv is monomial-free then
(v, w) is a compatible pair.
Proof Note that if t = t then (v, w) is compatible. Since v w, we may assume

that t < t. We have that in(In+1 )|Twv is monomial-free, so by Lemma 4.3, we have
that in(In )|Twv is monomial-free. We prove that if (v, w) ∈ Sn+1 × Sn+1 is not
compatible and v < w then in(In+1 )|Twv contains a monomial. Let v
= n−1 = w
.
Case 1 Assume that t < s . By compatibility we have t t < s s. Suppose
that s = s and
> s. Let
I = {w1 , . . . , ws } = {i1 < i2 < · · · < is −2 < n < n + 1},

J = {i1 < · · · < is −3 < n − 1}.
Note that v w and s = s <

therefore is −2 < n − 1. And so we have
v I, J w. Let
I = {i1 < · · · < is −3 < n − 1 < n < n + 1}, J = {i1 < · · · < is −2 }.
By construction it is clear that PI PJ −PI PJ is a binomial in in(In+1 ). Since n−1 ∈ /

{w1 , . . . , ws }, it follows that I w. And so PI PJ is a monomial in in(In+1 )|Twv .
So we may now assume that either
s or s < s. However, if s < s then, by
induction, we have
s. So we assume
s.
Case 1.1 Let
> s . Since t t < s and t = t, we have t < s − 1. Take

(w1 , . . . , wt −1 , w
)↑ if t > 1,
I = (w1 , . . . , ws )↑, J =
w
if t = 1,

(w1 , . . . , wt −1 , w
, wt , wt +2 , . . . , ws )↑ if t > 1,
I = and
(w
, wt , wt +2 , . . . , ws )↑ if t = 1

(w1 , . . . , wt −1 , wt +1 )↑ if t > 1,
J =
wt +1 if t = 1.
It is clear from the construction that PI PJ − PI PJ is a binomial in in(In+1 ) and

v I, J w. So PI PJ does not vanish in in(In+1 )|Twv . Since I w, then PI
vanishes in in(In+1 )|Twv . Therefore, PI PJ is a monomial in in(In+1 )|Twv .
Case 1.2 Let
< s . Let k = max{
, t} and vr = max{vi : 1 i k, i = t}.
Now define
I = (v1 , . . . , vk )↑, J = (v1 , . . . , vr−1 , vr+1 , . . . , vk−1 , n − 1)↑,
I = (v1 , . . . , vr−1 , vr+1 , . . . , vk , n − 1, )↑ and J = (v1 , . . . , vk−1 )↑

Consider the tableaux for PI PJ and PI PJ . Note that all rows are the same except
for the (k − 1)th row, and in this row we interchange n − 1 and vr . Since k < s s,
n−1 ∈ / I and PI PJ − PI PJ is a binomial in in(In+1 ), it follows that v I, J w
and so PI PJ does not vanish in in(In+1 )|Twv . Since k < s and n − 1 ∈ I , we see
that I w. So PI vanishes in in(In+1 )|Twv .
Case 2 Assume that t > s. Then we have t t > s s . Let vr = max{vi : 1
i t , i = s}. We define

(v1 , . . . , vs−1 , n + 1, vs+1 , . . . , vt )↑ if s > 1,
I= J = (v1 , . . . , vt −1 )↑,
(n, vs+1 , . . . , vt )↑ if s = 1,
I = (v1 , . . . , vr−1 , vr+1 , . . . , vt , n + 1)↑ and

(v1 , . . . , vs−1 , vs+1 , . . . , vt −1 , vr )↑ if s > 1,
J =
(vs+1 , . . . , vt −1 , vr )↑ if s = 1.
/ J and
By construction, PI PJ − PI PJ is a binomial in in(In+1 ). Since n − 1, n ∈

t > s, we have v J and PI PJ is a monomial in in(In+1 )|Twv .
Case 3 Assume that there exists t k < t such that vk > vk+1 . We take

(v1 , . . . , vk−1 , vk+1 , n)↑ if k > 1,
I= J = (v1 , . . . , vk )↑,
(vk+1 , n)↑ if k = 1,

(v1 , . . . , vk−1 , vk , n)↑ if k > 1,
I = and J = (v1 , . . . , vk−1 , vk+1 )↑ .
(vk , n)↑ if k = 1,
Then PI PJ − PI PJ is a binomial in in(In+1 ). Since vk+1 > vk , we have v

I, J and v J . Now we show that I w. Since v w and vk+1 > vk , we have
(v1 , v2 , . . . , vk−1 , vk+1 )↑ (w1 , . . . , wt −1 , wt +1 , . . . , wk , wk+1 )↑ .
Then (v1 , v2 , . . . , vk−1 , vk+1 , n) ↑ (w1 , . . . , wk , wk+1 ) ↑. Then we see that

PI PJ is non-zero and PJ vanishes in in(In )|Twv . Hence, PI PJ is a monomial in
in(In+1 )|Twv .
Case 4 Assume that there exists t < k t with wk < wk+1 . In this case, we
choose:

(w1 , . . . , wt −1 , wt +1 , . . . , wk−1 , wk+1 )↑ if t > 1,
I = (w1 , . . . , wk )↑, J =
(w2 , . . . , wk−1 , wk+1 )↑ if t = 1,
I = (w1 , . . . , wk−1 , wk+1 )↑ and

(w1 ,. . . , wt −1 , wt +1 , . . . , wk−1 , wk )↑ if t > 1,
J =
(w2 , . . . , wk−1 , wk ) if t = 1.
It is easy to see that PI PJ − PI PJ is a binomial in in(In+1 ). Since

wk < wk+1 , we have v I w and J w. Now we show that
v J . Since v w and wk+1 > wk , we have (v1 , v2 , . . . , vk−1 ) ↑
(w1 , . . . , wt −1 , wt +1 , . . . , wk−1 , wk+1 ) ↑. Note that PI PJ is non-zero and PI
vanishes in in(In+1 )|Twv . Then PI PJ is a monomial in in(In+1 )|Twv .
Hence, we conclude that if (v, w) is not compatible, then (v, w) ∈ Tn+1 , as
desired.

Lemma 4.5 Let v, w ∈ Sn+1 . If in(In )|Twv is monomial-free and (v, w) is a
compatible pair then in(In+1 )|Twv is monomial-free.
Proof We proceed by double induction, first on n and secondly on the dimension of
the Richardson variety. By Lemma 4.4, we may assume that for all v, w ∈ Sk where
k n we have in(Ik )|Twv is monomial-free if and only if (v, w) is a compatible pair
and in(Ik−1 )|Twv is monomial-free. Fix v, w ∈ Sn+1 and let PI PJ − PI PJ be a
binomial in the ideal in(In+1 ) and assume that PI PJ does not vanish in in(In+1 )|Twv .
Without loss of generality we assume that |I | = |I | |J | = |J |. We will show
that PI PJ does not vanish in in(In+1 )|Twv by taking cases on t, t .
Case 1 Assume that t, t both lie in one of the following sets: {1, . . . , |J |}, {|J | +
1, . . . , |I |} or {|I | + 1, . . . , n + 1}. For each of these cases we deduce immediately
which sets I, I , J, J contain n + 1. By removing n + 1 from these sets we obtain
I , I , J , J . We have that PI PJ − PI PJ is a binomial in in(In ). Since PI PJ does
not vanish in in(In+1 )|Twv , by construction we have that PI PJ does not vanish in
in(In )|Twv . Since (v, w) ∈ Tn we have that PI PJ does not vanish in in(In )|Twv . By
construction it follows that PI PJ does not vanish in in(In+1 )|Twv .
Case 2 Assume that t ∈ {|J | + 1, . . . , |I |} and t ∈ {|I | + 1, . . . , n + 1}.
Case 2.1 Assume that n + 1 ∈ I . By Case i. of the claim on page 180, we have that
v = (v1 , . . . , vt−2 , vt , vt−1 , vt+1 , . . . , vn+1 ) I . Since |J | < |I | we have v J .
By Lemma 3.11 we have that (v , w) ∈ Tn+1 and so by induction on the dimension
PI PJ does not vanish in in(In+1 )|T v . Therefore, v < v I , J w and so
w
PI PJ does not vanish in in(In+1 )|Twv .
Case 2.2 Assume that n + 1 ∈ / I . By Case ii. of the claim on page 180, we have that
I w = (w1 , . . . , wt −1 , wt +1 , wt , vt+1 , . . . , wn+1 ). Since |J | < |I | we have
J w . By Lemma 3.11 we have that (v, w ) ∈ Tn+1 and so by induction on the
dimension PI PJ does not vanish in in(In+1 )|T v . Therefore, v I , J w < w
w
and so PI PJ does not vanish in in(In+1 )|Twv .
Case 3 Assume that t ∈ {1, . . . , |J |} and t ∈ {|J | + 1, . . . , |I |}. This case is

identical to Case 2, where we can use a similar argument to show that either v J
if n + 1 ∈ J or J w if n + 1 ∈
/ J.
Case 4 Assume that t ∈ {1, . . . , |J |} and t ∈ {|I | + 1, . . . , n + 1}. If t < |J | or
t > |I | + 1 then we can use the construction of v , w as above and conclude the
result by induction on the dimension of Xw v . So we may assume that t = |J | and
t = |I | + 1. In this case we write I = {i1 < · · · < i|I | } and J = {j1 < · · · < j|J | }.
j
Let (v, w)i be the maximum block of (v, w). Then we have by Proposition 3.9
that vi < · · · < vt = n + 1 and by Proposition 3.7 we have vk < vi for all k < i.
Since t = |J | it follows that i |J |. And so ordering the first t elements of v we
get
{v1 , . . . , vt } = {
v1 < · · · <
vt } = {
v1 < · · · <
vi−1 < vi < · · · < vt }.
And so for all k we have vk ik and vk jk , i.e. the ordering of the first |I |
elements of v coincides with ordering of the first |J | elements of v. Since the
tableaux representing PI PJ and PI PJ are row-wise equal, it follows that v I
and v J .
Next we show that I w and J w. By Proposition 3.9 we have wt > · · · >
wj and by Proposition 3.7 we have that wk < wj for all k > j .
Claim For any K ⊆ [n + 1]: K w if and only if K c := [n + 1]\K ww0 =
(wn+1 , wn , . . . , w1 ).
To prove the claim, it is an easy observation that K w if and only if
(K ↑ , (K c )↑ ) w. Then (K ↑ , (K c )↑ ) w if and only if (K ↑ , (K c )↑ )w0 ww0 .
Explicitly we have (K ↑ , (K c )↑ )w0 = ((K c )↓ , K ↓ ). And so (K ↑ , (K c )↑ )w0
ww0 if and only if K c ww0 . This completes the proof of the claim.
Since the tableaux representing PI PJ and PI PJ are row-wise equal, it follows
that the tableaux representing PI c PJ c and PI c PJ c are also row-wise equal. By the
claim above we have that ww0 I c and ww0 J c . And so we have I w and

J w.
5 Toric Degenerations
Recall the ideals in(In )|Twv , in(I (Xwv )) and ker (φ |v ) from Sects. 2.4 and 2.5. In
n w
this section, we will study the relationships between these ideals with the goal of
understanding the initial ideal in(I (Xw v )). When this ideal is toric, we obtain a
toric degeneration of the Richardson variety Xw v inside the flag variety. We recall,
by Theorem 2.3, that the initial ideal in(In ) is quadratically generated and is the
kernel of the monomial map φn in (2.4). We will see that if the ideal in(In )|Twv is
monomial-free then in(I (Xw v )) is quadratically generated. Furthermore, we prove
that if in(In )|Twv is monomial-free then in(In )|Twv = in(I (Xw v )) and in(I )| v is a
n Tw
toric ideal. We prove this by showing that if in(In )|Twv is monomial-free then it is
equal to the ideal ker(φn |vw ).
Lemma 5.1 We have the following:
(i) The ideal in(In )|Twv is monomial-free if and only if it coincide with ker (φn |vw ).
(ii) in(In )|Twv ⊆ in(I (Xwv )).
Proof By Theorem 2.3, there exists a set G of quadratic binomials which generate
the initial ideal in(In ) and by Lemma 2.5(ii), the ideal in(In )|Twv is generated by
G|Twv . See (2.6).
(i) First note that φn |vw is a monomial map, hence its kernel does not contain
any monomials. So, if the ideal in(In )|Twv contains a monomial then it is not
equal to ker (φn |vw ). Now assume that the ideal in(In )|Twv does not contain any
monomials, therefore the set G|Twv does not contain any monomials. Since all
binomials m1 − m2 ∈ G|Twv lie in in(In ) and contain only the non-vanishing
Plücker variables PJ for J ∈ Twv , therefore m1 − m2 ∈ ker (φn |vw ). And so we
have in(In )|Twv ⊆ ker (φn |vw ). Thus the proof of (i) follows.
(ii) Since in(In )|Twv = G|Twv , we take ĝ ∈ G|Twv . So we have that g ∈ G and there
exists f ∈ I (Xw v ) such that in(f ) = g. The terms of ĝ are precisely the non-
vanishing terms of the initial terms of f . Therefore, ĝ = in(fˆ) ∈ in(I (Xw v )).
This completes the proof of lemma.

Theorem 5.2 If the ideal in(In )|Twv is monomial-free, then in(I (Xw v )) is a toric
ideal and it provides a toric degeneration of the Richardson variety Xwv.
Proof Let us consider M ⊆ R := K[PJ : J ∈ Twv ] be a collection of monomials

which are linearly independent in R/ in(I (Xw v )). If the image of M in R/ in(I )| v
n Tw
is a linearly dependent subset, thenwe have m∈M cm m ∈ in(In )|Twv for some
cm ∈ K. So by Lemma 5.1 we have m∈M cm m ∈ in(I (Xw v )), and so the image of
M in R/ in(I (Xw )) is linearly dependent, a contradiction. Moreover, since in(In )|Twv

v
is monomial-free we have that in(In )|Twv = ker (φn |vw ). Hence, for all d 1, any
standard monomial basis for R/ in(I (Xw v )) of degree d is linearly independent in
R/ in(In )|Tw = R/ ker (φn |w ). Note that ker(φn |vw ) is generated by binomials which
v v
correspond to pairs of row-wise equal tableaux, whose columns I satisfy v I

w. It is easy to see that every tableau is row-wise equal to a unique semi-standard
Young tableau. Therefore, the semi-standard Young tableaux SSY Td (v, w) form a
monomial basis for the degree d part of R/ ker(φn |vw ).
Note that any Gröbner degeneration gives rise to a flat family, so the Hilbert
polynomials of all fibers are identical. By Theorem 3.12, the semi-standard Young
tableaux SSY Td (v, w) form a standard monomial basis for R/I (Xw v ). Thus, the
dimension of the degree d part of R/ in(I (Xw ) is equal to |SSY Td (v, w)|. And so
v
in(In )|Twv = ker(φn |vw ) = in(I (Xw

v )).

Remark 5.3 Our methods can be used to produce other toric degenerations of
Richardson varieties with respect to different weight vectors. For example, in [14],
Kim considers a weight on the polynomial ring K[xi,j ] such that the leading term of
any minor ϕn (PJ ), see (2.1), is the antidiagonal term. Explicitly, if M is the weight
on K[xi,j ] then
inM (ϕn (PJ )) = x1,jt x2,jt−1 . . . xt,j1 for every J = {j1 < · · · < jt }.
We write wM for the weight on the ring K[PJ : J ⊆ [n]] induced by M . In

[14], the pairs of permutations (v, w) are labelled by collections of so-called pipe
dreams. Each pair of reduced pipe dreams associated to (v, w) gives rise to a face of
the Gelfand-Tsetlin polytope. If (v, w) is labelled by a unique pair of reduced pipe
dreams, then the initial ideal inwM (I (Xw v )) is toric. However, if there are multiple
pairs of pipe dreams associated to (v, w), then this method cannot differentiate
between toric and non-toric initial ideals. For each (v, w) ∈ S4 × S4 we have
calculated the ideals inwM (I4 )|Twv . We have confirmed that Theorem 5.2 holds in all
v ))
these cases, i.e. if inwM (I4 )|Twv is monomial-free then the initial ideal inwM (I (Xw
is toric. Our calculations are displayed in Table 1. The symbol ∗ appears in the
table beside pairs of permutations for which the description by pipe dreams does
not determine whether the corresponding ideal is toric or non-toric.
In many cases, it is possible to give an explicit description of the polytopes
associated to toric degenerations. Any toric variety whose ideal is of the form
v )), for some v and w, is a toric subvariety of the toric variety asso-
inwM (I (Xw
ciated to the Gelfand-Tsetlin polytope. Therefore, the toric polytope associated to
v )) is a face of the Gelfand-Tsetlin polytope. Suppose that in
inwM (I (Xw v
wM (I (Xw ))
is toric. On the one hand, if there is a unique pair of reduced pipe dreams associated
to (v, w), then this face of the Gelfand-Tsetlin polytope is determined uniquely and
is described in terms of Gelfand-Tsetlin patterns in [14]. On the other hand, if there
does not exist a unique pair of pipe dreams associated to (v, w), then determining
the face of the Gelfand-Tsetlin polytope is a difficult computational task.
Computing polytopes of toric degenerations of the flag variety Fln for large n
is already very difficult. For example, all toric degenerations via Gröbner degener-
ations have been calculated up to Fl5 , see [1]. One approach to understand these
toric polytopes, and by extension their faces, is to first consider the Grassmannian
and its toric degenerations studied in [6]. The vertices of the corresponding toric
polytopes can be read directly from the monomial map, analogous to (2.4). In [5],
the authors study these polytopes using combinatorial mutations which preserve
many important properties of the polytope, such as its Ehrhart function. However,
in the forthcoming work [4], the authors note that the monomial map (2.4) does
not immediately give rise to the toric polytope. Instead, they give a combinatorial
analogue to: embedding products of projective varieties into higher dimensional
projective spaces, for polytopes. We give an example of this procedure below.
Example 5.4 Let v = (2, 3, 4, 1) and w = (4, 2, 3, 1). Let us consider the
antidiagonal term order from Remark 5.3. We have that ((2, 3, 4, 1), (4, 2, 3, 1)) is
v degenerates to a toric variety,
contained in Table 1, and so the Richardson variety Xw
which we will call T . However, the entry has an asterisk beside it in Table 1, so it is
v with respect
Table 1 The list of pairs of permutations (v, w) leading to toric degenerations of Xw
to the antidiagonal term order in Fl4
((1, 2, 3, 4), (1, 4, 2, 3)) ∗ ((2, 3, 1, 4), (4, 3, 1, 2)) ∗ ((1, 2, 3, 4), (1, 2, 4, 3)) ∗
((1, 2, 3, 4), (1, 4, 3, 2)) ∗ ((2, 3, 1, 4), (4, 3, 2, 1)) ∗ ((1, 2, 3, 4), (1, 3, 2, 4)) ∗
((1, 2, 3, 4), (3, 1, 2, 4)) ∗ ((2, 3, 4, 1), (4, 2, 3, 1)) ∗ ((1, 2, 3, 4), (2, 1, 3, 4)) ∗
((1, 2, 3, 4), (3, 2, 1, 4)) ∗ ((2, 3, 4, 1), (4, 3, 2, 1)) ∗ ((1, 2, 3, 4), (2, 1, 4, 3)) ∗
((1, 2, 3, 4), (4, 1, 2, 3)) ∗ ((3, 1, 2, 4), (4, 1, 2, 3)) ∗ ((1, 2, 4, 3), (2, 1, 4, 3)) ∗
((1, 2, 3, 4), (4, 1, 3, 2)) ∗ ((3, 1, 2, 4), (4, 1, 3, 2)) ∗ ((1, 3, 4, 2), (1, 4, 3, 2)) ∗
((1, 2, 3, 4), (4, 2, 1, 3)) ∗ ((3, 1, 2, 4), (4, 2, 1, 3)) ∗ ((1, 4, 2, 3), (1, 4, 3, 2)) ∗
((1, 2, 3, 4), (4, 3, 1, 2)) ∗ ((3, 1, 2, 4), (4, 3, 1, 2)) ∗ ((2, 1, 3, 4), (2, 1, 4, 3)) ∗
((1, 2, 3, 4), (4, 3, 2, 1)) ∗ ((3, 1, 2, 4), (4, 3, 2, 1)) ∗ ((2, 3, 1, 4), (3, 2, 1, 4)) ∗
((1, 3, 2, 4), (1, 4, 2, 3)) ∗ ((3, 1, 4, 2), (4, 1, 3, 2)) ∗ ((2, 3, 4, 1), (2, 4, 3, 1)) ∗
((1, 3, 2, 4), (1, 4, 3, 2)) ∗ ((3, 2, 1, 4), (4, 2, 1, 3)) ∗ ((2, 3, 4, 1), (3, 2, 4, 1)) ∗
((2, 1, 3, 4), (3, 1, 2, 4)) ∗ ((3, 2, 1, 4), (4, 3, 1, 2)) ∗ ((3, 1, 2, 4), (3, 2, 1, 4)) ∗
((2, 1, 3, 4), (3, 2, 1, 4)) ∗ ((3, 2, 1, 4), (4, 3, 2, 1)) ∗ ((3, 4, 1, 2), (3, 4, 2, 1)) ∗
((2, 1, 3, 4), (4, 1, 2, 3)) ∗ ((3, 2, 4, 1), (4, 2, 3, 1)) ∗ ((3, 4, 1, 2), (4, 3, 1, 2)) ∗
((2, 1, 3, 4), (4, 1, 3, 2)) ∗ ((3, 2, 4, 1), (4, 3, 2, 1)) ∗ ((3, 4, 2, 1), (4, 3, 2, 1)) ∗
((2, 1, 3, 4), (4, 2, 1, 3)) ∗ ((4, 1, 2, 3), (4, 3, 1, 2)) ∗ ((4, 1, 2, 3), (4, 1, 3, 2)) ∗
((2, 1, 3, 4), (4, 3, 1, 2)) ∗ ((4, 1, 2, 3), (4, 3, 2, 1)) ∗ ((4, 1, 2, 3), (4, 2, 1, 3)) ∗
((2, 1, 3, 4), (4, 3, 2, 1)) ∗ ((4, 2, 1, 3), (4, 3, 1, 2)) ∗ ((4, 2, 3, 1), (4, 3, 2, 1)) ∗
((2, 3, 1, 4), (2, 4, 1, 3)) ∗ ((4, 2, 1, 3), (4, 3, 2, 1)) ∗ ((4, 3, 1, 2), (4, 3, 2, 1)) ∗
((2, 3, 1, 4), (4, 2, 1, 3)) ∗
not possible to immediately determine the polytope associated to this degeneration

of Xw v using pipe dreams from [14]. Let us calculate the polytope associated to this
degeneration following [4].

Recall that, via the Plücker embedding, the Flag variety Fl4 is a subvariety of
the product of projective spaces P3 × P5 × P3 . The coordinates for each projective
space are PI where I ⊆ [n] is a fixed size. For instance [P1 , P2 , P3 , P4 ] are
the homogeneous coordinates for the first copy of P3 . The Richardson variety
Xw v and its toric degeneration T naturally live in the product of projective spaces
P × P1 × P0 , with coordinates given by the non-vanishing variables which are

2
[P2 , P3 , P4 ], [P23 , P24 ] and [P234 ], respectively. The ideal of T ⊆ P2 × P1 × P0 is

the kernel of the monomial map ϕ4 , see Remark 5.3, restricted to the non-vanishing
variables. We can write ϕ4 as the following integer matrix, note that only the non-
zero rows are included:
P2 P3 P4 P23 P24 P234

⎛ ⎞
x2 1
x3 ⎜ 1 1 ⎟
⎜ ⎟
x ⎜ 1 1 1 ⎟
A= 4 ⎜ ⎟.
y2 ⎜ 1 1 ⎟
⎜ ⎟
y3 ⎝ 1 ⎠
z2 1
Fig. 1 Projection of the

polytope P of the toric variety
in Example 5.4. The
illustration includes all lattice
points of the polytope, which
are labelled by the
corresponding coordinates
of P5
The image of P2 × P1 × P0 under the Segre embedding is a toric subvariety Y ⊆

P5 . We label the coordinates for P5 by the corresponding products of variables:
P2 P23 P234 , P2 P24 P234 and so on. The monomial map, with matrix S, associated
to Y sends each coordinate of P5 to the product of Plücker variables which it is
indexed by. Explicitly, the columns of S are the exponent vectors of the products of
variables:
P2 P23 P234 P2 P24 P234 P3 P23 P234 P3 P24 P234 P4 P23 P234 P4 P24 P234
⎛ ⎞
P2 1 1
⎜ ⎟
P3 ⎜ 1 1 ⎟
⎜ ⎟
⎜ ⎟
P4 ⎜ 1 1 ⎟
S= ⎜ ⎟.
P23 ⎜ 1 1 1 ⎟
⎜ ⎟
⎜ ⎟
P24 ⎝ 1 1 1 ⎠
P234 1 1 1 1 1 1
Consider the image of T under the Segre embedding. The ideal of T embedded in
P5 is the kernel of the following monomial map, whose matrix is the product of A
and S:
P2 P23 P234 P2 P24 P234 P3 P23 P234 P3 P24 P234 P4 P23 P234 P4 P24 P234
⎛ ⎞
x2 1 1
⎜ ⎟
x3 ⎜ 1 2 1 1 ⎟
⎜ ⎟
⎜ ⎟
x4 ⎜ 1 2 1 2 2 3 ⎟
AS = ⎜ ⎟.
y2 ⎜ 1 1 1 1 1 1 ⎟
⎜ ⎟
⎜ ⎟
y3 ⎝ 1 1 1 1 1 1 ⎠
z2 1 1 1 1 1 1
The polytope P corresponding to this toric variety is the convex hull of the columns
of AS. It is easy to see that P is 2-dimensional by projecting it to the coordinates
indexed by x2 , x3 and x4 . Moreover, P lives in the 2-dimensional affine subspace
{(i, j, k) : i + j + k = 3}. See Fig. 1.
Acknowledgments NC was supported by the SFB/TRR 191 “Symplectic structures in Geometry,

Algebra and Dynamics”. He gratefully acknowledges support from the Max Planck Institute for
Mathematics in Bonn, and the EPSRC Fellowship EP/R023379/1 who supported his multiple visits
to Bristol. OC was supported by EPSRC Doctoral Training Partnership award EP/N509619/1. FM
was supported by EPSRC Fellowship EP/R023379/1, the BOF grant BOF/STA/201909/038, and
the FWO (project no. G023721N and G0F5921N).
References
1. L. Bossinger, S. Lamboglia, K. Mincheva, and F. Mohammadi. Computing toric degenerations

of flag varieties. In Combinatorial algebraic geometry, pages 247–281. Springer, 2017.
2. P. Caldero. Toric degenerations of Schubert varieties. Transformation Groups, 7(1):51–60,
2002.
3. N. Chary Bonala, O. Clarke, and F. Mohammadi. Standard monomial theory and toric
degenerations of Richardson varieties in the Grassmannian. To appear in Journal of Algebraic
Combinatorics, arXiv preprint arXiv:2103.16208, 2021.
4. O. Clarke, A. Higashitani, and F. Mohammadi. Block diagonal polytopes for flag varieties and
their combinatorial mutations. In preparation, 2021.
5. O. Clarke, A. Higashitani, and F. Mohammadi. Combinatorial mutations and block diagonal
polytopes. Collectanea Mathematica, pages 1–31, 2021.
6. O. Clarke and F. Mohammadi. Toric degenerations of Grassmannians and Schubert varieties
from matching field tableaux. Journal of Algebra, 559:646–678, 2020.
7. O. Clarke and F. Mohammadi. Standard monomial theory and toric degenerations of Schubert
varieties from matching field tableaux. Journal of Symbolic Computation, 104:683–723, 2021.
8. O. Clarke and F. Mohammadi. Toric degenerations of flag varieties from matching field
tableaux. Journal of Pure and Applied Algebra, 225(8):106624, 2021.
9. V. Deodhar. On some geometric aspects of Bruhat orderings. I. A finer decomposition of
Bruhat cells. Inventiones mathematicae, 79(3):499–511, 1985.
10. N. Gonciulea and V. Lakshmibai. Degenerations of flag and Schubert varieties to toric varieties.
Transformation Groups, 1(3):215–248, 1996.
11. D. R. Grayson and M. E. Stillman. Macaulay2, a software system for research in algebraic
12. T. Hibi. Distributive lattices, affine semigroup rings and algebras with straightening laws. In
Commutative Algebra and Combinatorics, pages 93–109, 1987.
13. W. V. D. Hodge. Some enumerative results in the theory of forms. In Mathematical
Proceedings of the Cambridge Philosophical Society, volume 39, pages 22–30, 1943.
14. G. Kim. Richardson varieties in a toric degeneration of the flag variety. Thesis (Ph.D.)—
University of Michigan. 82 pp. ISBN: 978-1339-03949-7, 2015.
15. M. Kogan and E. Miller. Toric degeneration of Schubert varieties and Gelfand-Tsetlin
polytopes. Advances in Mathematics, 193(1):1–17, 2005.
16. V. Kreiman and V. Lakshmibai. Richardson varieties in the Grassmannian. arXiv preprint
math/0203278, 2002.
17. V. Lakshmibai and P. Littelmann. Richardson varieties and equivariant K-theory. Journal of
Algebra, 260(1):230–260, 2003.
18. E. Miller and B. Sturmfels. Combinatorial commutative algebra, volume 227. Springer
Science & Business Media, 2004.
19. R. Richardson. Intersections of double cosets in algebraic groups. Indagationes Mathematicae,
3(1):69–77, 1992.
20. C. S. Seshadri. Introduction to the theory of standard monomials, volume 46. Springer.
21. M. Willis. A direct way to find the right key of a semistandard young tableau. Annals of
Combinatorics, 17, 10 2011.
Simplicial Resolutions for the Second
Power of Square-Free Monomial Ideals
Susan M. Cooper, Sabine El Khoury, Sara Faridi, Sarah Mayes-Tang,

Susan Morey, Liana M. Şega, and Sandra Spiroff
1 Introduction
The question of finding, or even effectively bounding, the Betti numbers of an ideal
in a commutative ring is a difficult one. Even more complicated is using the structure
of an ideal I to find information about the Betti numbers of its powers I r : predicting
something as basic as the minimal number of generators of I r is a difficult problem.
S. M. Cooper
Department of Mathematics, University of Manitoba, Winnipeg, MB, Canada
e-mail: susan.cooper@umanitoba.ca
S. El Khoury
Department of Mathematics, American University of Beirut, Beirut, Lebanon
e-mail: se24@aub.edu.lb
S. Faridi ()
Department of Mathematics & Statistics, Dalhousie University, Halifax, NS, Canada
e-mail: faridi@dal.ca
S. Mayes-Tang
Department of Mathematics, University of Toronto, Toronto, ON, Canada
e-mail: smt@math.toronto.edu
S. Morey
Department of Mathematics, Texas State University, San Marcos, TX, USA
e-mail: morey@txstate.edu
L. M. Şega
Department of Mathematics and Statistics, University of Missouri - Kansas City, Kansas City,
MO, USA
e-mail: segal@umkc.edu
S. Spiroff
Department of Mathematics, University of Mississippi, University, MS, USA
e-mail: spiroff@olemiss.edu
194 S. M. Cooper et al.
Taylor’s thesis [10] described a free resolution of any ideal minimally generated
by q monomials using the simplicial chain complex of a simplex with q vertices.
Taylor’s construction, though often far from minimal, produces a resolution of every
q !
monomial ideal I . It gives upper bounds i+1 βi (I ) for the Betti numbers
q !
of I where i+1 is the number of i-faces of a q-simplex. If I is generated by q
monomials and r is a positive integer, then the number of generators of I r generally
grows exponentially and as a result, so do the bounds on the Betti numbers of I r
given by Taylor’s resolution.
In this paper, we focus on the case where r = 2 and I is a square-free monomial
ideal!with q generators. In this case, we know that I 2 can be generated by at most
q+1
monomials, and hence has a Taylor resolution supported on a simplex with
2 !
at most q+1 2 vertices. The question that we address in this paper is: can we find a
subcomplex of this simplex whose simplicial chain complex yields a free resolution
of I 2 ? Such a resolution would be closer to minimal than the Taylor resolution.
!
We answer this question by constructing a simplicial complex on q+1 2 vertices
which we call L2q in honor of the Lyubeznik resolution [9] which was our inspiration.
!
While L2q has the same number of vertices as the q+1 2 -simplex, it is significantly
smaller because it has far fewer faces. For a given square-free monomial ideal I , we
can use further deletions of L2q specific to the generators of I to show that L2q has an
induced subcomplex L2 (I ) which supports a free resolution of I 2 . As a result, we
find (sharp) upper bounds on the Betti numbers of the second power of any square-
free monomial ideal. These bounds are often significantly smaller than the bounds
provided by the Taylor resolution (see Sect. 4).
Section 2 lays out the notation and terminology used in the paper including the
construction of simplicial resolutions. In Sect. 3, we describe the complexes L2q
(Definition 3.1) and L2 (I ) (Definition 3.4) and prove that L2 (I ) supports a free
resolution of I 2 when I is a square-free monomial ideal (Theorem 3.9). Section 4
provides results on the bounds on the Betti numbers that follow from the main
results.
This paper is part of a larger project [3] to study resolutions of powers of
monomial ideals, which the authors started during the 2019 Banff workshop
“Women in Commutative Algebra”.
2 Background
Throughout this paper we let S = k[x1 , . . . , xn ] be a polynomial ring over a field

k. In this section we briefly recall some necessary background about simplicial
complexes.
A simplicial complex over a vertex set V is a set of subsets of V such that
if F ∈ and G ⊆ F then G ∈ . An element σ of is called a face and
the maximal faces under inclusion are called facets. A simplicial complex can be
uniquely determined by its facets, and we use the notation
Simplicial Resolutions for the Second Power of Square-Free Monomial Ideals 195
= F0 , . . . , Fq
to describe a simplicial complex whose facets are F0 , . . . , Fq .

The dimension of a face F in is dim(F ) = |F | − 1, and the dimension of
is the maximum of the dimensions of its faces.
A simplicial complex with one facet is called a simplex.
If W ⊆ V , the subcomplex
W = {σ ∈ | σ ⊆ W }
is called the induced subcomplex of on W .

If is a simplicial complex with vertex v, then when we delete v from we
obtain the simplicial complex
\ {v} = {σ ∈ | v ∈
/ σ }.
A facet F of is said to be a leaf if it is the only facet of , or there is a different

facet G of , called a joint, such that
F ∩H ⊆G
for all facets H = F . The joint G in this definition is not unique ([4]). A simplicial
complex is a quasi-forest if the facets of can be ordered as F0 , . . . , Fq such
that for i = 0, . . . , q, the facet Fi is a leaf of the simplicial complex F0 , . . . , Fi .
A connected quasi-forest is called a quasi-tree ([11]).
Example 2.1 The simplicial complex below is a quasi-tree, with leaf order:
F0 , F1 , F2 , F3 , meaning that each Fi is a leaf of F0 , . . . , Fi . Note that in this
case, the joint of Fi is F0 for every i 1.
This complex is in fact L23 as we will see later in Example 3.2.

If I is minimally generated by monomials m1 , . . . , mq in S, a minimal free
resolution of I is a (unique up to isomorphism) exact sequence of free S-modules
0 → S βp → S βp−1 → · · · → S β1 → S β0 → I → 0
where p ∈ N, β0 = q and for each i ∈ {1, . . . , p}, βi is the smallest possible rank
of a free module in the i-th spot of any free resolution of I . The βi , called the Betti
numbers of I , are invariants of the ideal I .
Finding ways to describe a free resolution of a given ideal is an open and active
area of research. For monomial ideals, combinatorics plays a big role. In her thesis
in the 1960’s, Diana Taylor introduced a method of labeling the faces of a simplex
with monomials, and then used this labeling to turn the simplicial chain complex
of into a free resolution of a monomial ideal. This technique has been generalized
to other simplicial complexes by Bayer and Sturmfels [1], among others.
More precisely, if I is minimally generated by monomials m1 , . . . , mq and is
a simplicial complex on q vertices v1 , . . . , vq , we label each vertex vi with the
monomial mi , and we label each face of with the least common multiple of
the labels of its vertices. Then, if the labeling of satisfies certain properties, the
simplicial chain complex of can be “homogenized” using the monomial labels on
the faces to give a free resolution of I . In this case, we say that supports a free
resolution of I and the resulting free resolution is called a simplicial resolution of
I . Peeva’s book [8] details this method for simplicial as well as other topological
resolutions.
Example 2.2 Let I = (x 2 , y 2 , z2 , xy, xz, yz). In the picture below, we label the
simplicial complex in Example 2.1 using the generators of I . To make the picture
less busy, we have included the labels of the vertices and the facets only.
Our main result Theorem 3.9 will prove that, indeed, does support a free
resolution of I . This in particular implies that βi (I ) is bounded above by the number
of i-faces of , which is the rank of the i-th chain group of . That is,
β0 (I ) 6, β1 (I ) 9, β2 (I ) 4, βi (I ) = 0 if i > 2.
We calculate, using Macaulay2 [7], that the actual Betti numbers of I are:
β0 (I ) = 6, β1 (I ) = 8, β2 (I ) = 3, βi (I ) = 0 if i > 2.
A major question in the theory of combinatorial resolutions is to determine

whether a given simplicial complex supports a free resolution of a given monomial
ideal. Taylor proved that a simplex with q vertices always supports a free resolution
of an ideal with q generators, or in other words, every monomial ideal has a Taylor
q !
resolution. As a result i+1 (the number of i-faces of a simplex with q vertices) is
an upper bound for βi (I ) if I is any monomial ideal with q generators. We denote
the q-simplex labeled with the q generators of I by Taylor(I ).
Taylor’s resolution is usually far from minimal. However, if I is a monomial ideal
with a free resolution supported on a (labeled) simplicial complex , then has to
be a subcomplex of Taylor(I ). As a result, the question of finding smaller simplicial
resolutions of I turns into a question of finding smaller subcomplexes of Taylor(I )
which support a resolution of I .
One of the best known tools to identify such subcomplexes of the Taylor complex
is due to Bayer et al. [2], and reduces the problem to checking acyclicity of induced
subcomplexes. This criterion was adapted in [5] to the class of simplicial trees, and
then in [3] to quasi-trees. Theorem 2.3 is this latter adaptation, and will be used in
the rest of the paper.
If is a subcomplex of Taylor(I ) and m is a monomial in S, let m be the
subcomplex of induced on the vertices of whose labels divide m, and let
LCM(I ) denote the set of monomials that are least common multiples of arbitrary
subsets of the minimal monomial generating set of I .
Theorem 2.3 ([3] Criterion for Quasi-Trees Supporting Resolutions) Let be
a quasi-tree whose vertices are labeled with the monomial generating set of a
monomial ideal I in the polynomial ring S over a field k. Then supports a
resolution of I if and only if for every monomial m in LCM(I ), m is empty or
connected.
If I is minimally generated by q monomials, then I 2 is minimally generated by
q+1!
at most 2 monomials. Our goal in this paper is to find a (smaller) subcomplex
!
of the q+1 2
2 -simplex which produces a free resolution of I , and only depends on
q. The quasi-tree Lq , introduced in the next section, is such a candidate: it has
2
!
exactly q+12 vertices, and for any given ideal I with q generators, it has an induced
subcomplex L2 (I ) contained in Taylor(I 2 ) which supports a free resolution of I 2 .
3 The Quasi-Trees L2q and L2 (I )
For an integer q 1 we now give a description of a simplicial complex L2q ,

!
a subcomplex of the q+1 2 -simplex. We will show that if I is a monomial ideal
generated by q square-free monomials, an induced subcomplex of L2q , which we
denote by L2 (I ), always supports a free resolution of I 2 . The complex L2q is a far
!
smaller subcomplex of the q+1 2 -simplex, and its construction is motivated by the
monomial orderings used to build the Lyubeznik complex [9].
Definition 3.1 For an integer q 3, the simplicial complex L2q over the vertex set
{
i,j : 1 i j q} is defined by its facets as:
L2q = {
i,j : 1 j q}1iq , {
i,j : 1 i < j q},
where we define
j,i for j > i by the equality
j,i =
i,j . For q = 1 and q = 2 we
use the same construction but note that {
i,j : 1 i < j q} is empty for q = 1
and is a face but not a facet for q = 2.
When q = 1, the ideals I and I r for all r 2 are principal and L2q is a point.
When q = 2, the complex L22 has only 2 facets, see Example 3.2. Note that L2q
! !
has q+1 vertices, which is the number of vertices of the q+1
2 -simplex, and, when
2 !
q > 2, it has q + 1 facets, where one facet has dimension q2 − 1 and the remaining
q facets have dimension q − 1.
Example 3.2 The complexes L23 and L22 are shown on the left and right, respectively.
Proposition 3.3 For q 1, L2q is a quasi-tree.
Proof If q = 1, then L2q is a simplex of dimension 0, and so is a quasi-tree. If q = 2,

there are only two facets, namely F1 and F2 (as depicted above), and F2 is a leaf of
F1 , F2 with joint F1 , so L22 is a quasi-tree. For q 3, order the facets of L2q by
F0 = {
i,j : 1 i < j q}, and Fi = {
i,j : 1 j q} for 1 i q. By
definition, if i = k are nonzero, then Fi ∩ Fk = {
i,k } ⊆ F0 . Thus each Fi is a leaf
of F0 , . . . , Fi with joint F0 , and we are done.

Given a square-free monomial ideal I , we now define a labeled induced
subcomplex of L2q , denoted L2 (I ), which is obtained by deleting vertices from L2q .
Definition 3.4 (L2 (I )) For an ideal I minimally generated by the square-free

monomials m1 , . . . , mq , we define L2 (I ) to be a labeled induced subcomplex of
L2q formed by the following rules:
(1) Label each vertex of

i,j of L2q with the monomial mi mj .
(2) If for any indices i, j, u, v ∈ [q] where [q] = {1, . . . , q} with {i, j } = {u, v}
we have mi mj | mu mv , then
• If mi mj = mu mv and i = min{i, j, u, v}, then delete the vertex
i,j .
• If mi mj = mu mv , then delete the vertex
u,v .
(3) Label each of the remaining faces with the least common multiple of the labels
of its vertices.
The remaining labeled subcomplex of L2q is called L2 (I ), and is a subcomplex of
Taylor(I 2 ).
Remark 3.5 It follows from Proposition 3.7 below that if m2i divides mu mv then
u = v = i, hence the vertices
i,i are not deleted in the construction of L2 (I ).
In Step 2 above, when there is equality, the choice was made to eliminate the vertex
i,j with minimum index i so that one has a well-defined definition for L2 (I ); in
fact, one could show that a different choice of elimination would also serve our
purposes.
Example 3.6 Let I = (abe, bc, cdf, ad). Setting m1 = abe, m2 = bc and m3 =
cdf , m4 = ad, we first label all vertices of L24 with the products mi mj , but then
note that m2 m4 | m1 m3 .
So the (labeled) facets of L2 (I ) are the following five:
Facet Dimension
{m21 , m1 m2 , m1 m4 } 2
{m22 , m1 m2 , m2 m3 , m2 m4 } 3
{m23 , m2 m3 , m3 m4 } 2
{m4 , m1 m4 , m2 m4 , m3 m4 }
2 3
{m1 m2 , m1 m4 , m2 m3 , m2 m4 , m3 m4 } 4
In particular, L2 (I ) is a four-dimensional complex labeled with the generators

of I 2 .
We now present two preliminary results needed for the proof that when the ideal
I is square-free, L2 (I ) supports a free resolution of I 2 .
Proposition 3.7 Let m1 , . . . , mq be a minimal square-free monomial generating

set for an ideal I , let r be a positive integer, and suppose that for some i ∈ [q] and
1 u1 · · · ur q,
mri | mu1 · · · mur or mu1 · · · mur | mri .
Then u1 = · · · = ur = i.
Proof If for all, or some, of j ∈ [r] we have uj = i, then those copies of mi can be
deleted from each side of the division, so one can assume, without loss of generality
that i = 1 < u1 · · · ur q. Suppose that
j j
b
m1 = x1a1 · · · xnan and muj = x1 1 · · · xnbn ,
j
where av , bv ∈ {0, 1} for j ∈ [r] and v ∈ [n]. It follows that:
• if mr1 | mu1 · · · mur , then for every index v ∈ [n] where av = 0, we have rav = r
and so bv1 = · · · = bvr = 1. Therefore, we have m1 | muj for j ∈ [r]. This is a
contradiction since these monomials are minimal generators of I .
• if mu1 · · · mur | mr1 , then for each nonzero exponent bv1 of mu1 we must have
av = 0, and so mu1 | m1 , again a contradiction.

Proposition 3.8 Let I be an ideal minimally generated by square-free monomials
m1 , . . . , mq with q 2. Then for every i ∈ [q] there is a j ∈ [q] \ {i} such that
mu mv mi mj for any choice of u, v ∈ [q] \ {i, j }.
In particular, mi mj is a minimal generator of I 2 .

Proof Suppose, by way of contradiction, that there exists i ∈ [q] such that for every
j ∈ [q] \ {i} there exist u, v ∈ [q] \ {i, j } such that mu mv | mi mj .
With i as above, there exist functions ϕ, ψ : [q] \ {i} → [q] \ {i} such that
mϕ(j ) mψ(j ) | mi mj for all j ∈ [q] \ {i}. (1)
For each k 0, let ϕ k denote the composition ϕ ◦ ϕ ◦ · · · ◦ ϕ (k times). (When k = 0,

ϕ 0 is the identity function.) Let a ∈ [q]\{i}. For each w 1, set bw = ψ(ϕ w−1 (a)).
Apply (1) with j = ϕ k−1 (a) to get:
mϕ k (a) mbk | mi mϕ k−1 (a) for all k 1 .
From this, it is easy to see that

"

k "
k−1
"
mϕ k (a) · mbw " mi mϕ k−1 (a) · m bw for all k 2 .
"
w=1 w=1
Inductively, we thus obtain

"

k "
k−s
"
mϕ k (a) · mbw " mi mϕ k−s (a) ·
s
m bw for all k 2 and k > s 1.
"
w=1 w=1
(2)
Assume ϕ k (a) = ϕ k−s (a) for some k 2 and some s with k > s 1. After
simplifying in (2) we obtain
"

k "
"
mbw " msi .
"
w=k−s+1
For s = 1, this implies mbk | mi , but since bk = i, this contradicts the minimality
of the generating set. If s > 1 this is a contradiction according to Proposition 3.7.
Therefore, we have shown that the integers ϕ(a), ϕ 2 (a), . . . are distinct. This is a
contradiction, since ϕ k (a) ∈ [q]
{i} for all k, and [q]
{i} is a finite set.

We are now ready to prove the main result of the paper.
Theorem 3.9 (Main Result) Let I be a square-free monomial ideal. Then L2 (I )
supports a free resolution of I 2 .
Proof Suppose I is minimally generated by the square-free monomials
m1 , . . . , mq .
The simplicial complex L2 (I ) is an induced subcomplex of the quasi-tree L2q
(Proposition 3.3), and is therefore a quasi-forest itself (see [3, 6]). Let V denote
the set of vertices of L2 (I ). In view of Theorem 2.3, to show that L2 (I ) supports
a resolution of I 2 , we need to show that, for every m ∈ LCM(I 2 ), L2 (I )m is
connected, where L2 (I )m is the induced subcomplex of the complex L2 (I ) on the
set Vm = {
i,j ∈ V : mi mj | m}.
Suppose m ∈ LCM(I 2 ). If q = 1, then L2 (I )m is either empty or a point. If
q = 2, then I 2 = (m21 , m1 m2 , m22 ) and L2 (I ), as pictured in Example 3.2, has two
facets connected by the vertex
1,2 . If m ∈ {m21 , m22 }, then L2 (I )m is a point, and
hence connected. Otherwise, m1 m2 | m, so the vertext
1,2 will be in L2 (I )m . If
either
1,1 or
2,2 are in L2 (I )m , they will be connected to
1,2 . Therefore L2 (I )m
is connected.
Now assuming q 3, we use the notation introduced in the proof of
Proposition 3.3 for the facets of L2q , namely F0 , . . . , Fq . The facets of L2 (I )m are
the maximal sets among the sets F0 ∩ Vm , . . . , Fq ∩ Vm .
If m = m2i for some i ∈ [q], then Proposition 3.7 shows that L2 (I )m is one
point, and hence is connected. Assume now that m = m2i for all i ∈ [q], and hence
F0 ∩ Vm = ∅. To show that L2 (I )m is connected, it suffices to show that, for each
i ∈ [q] such that Fi ∩ Vm = ∅, the intersection between Fi ∩ Vm and F0 ∩ Vm is
nonempty. Note that any vertex in Fi ∩ Vm other than
i,i is also in F0 ∩ Vm . We
thus need to show that if

i,i ∈ Vm for some i ∈ [q], then there exists b ∈ [q] with
b = i such that
i,b ∈ Vm .
Assume
i,i ∈ Vm , hence m2i | m. Set
A = {j ∈ [q] : mj | m} .
Note that i ∈ A. Since m = m2i , we see that |A| 2. By Proposition 3.8 applied to
the ideal generated by the monomials mj with j ∈ A, there exists b ∈ A
{i} such
that
mu mv does not divide mi mb for all u, v ∈ A \ {i, b} . (3)
Since b ∈ A, we have mb | m. We claim that mi mb | m as well. Indeed, since

mb is a square-free monomial, setting m = m2i n, one has
mb | m ⇒ mb | m2i n ⇒ mb | mi n ⇒ mi mb | m2i n ⇒ mi mb | m . (4)
In order to conclude
i,b ∈ Vm , we need to show that
i,b ∈ V , that is,
i,b is
a vertex of L2 (I ). If
i,b ∈
/ V , then we must have mu mv | mi mb for some u, v ∈
[q]
{i, b}. Since mi mb | m, we further have mu | m and mv | m, hence u, v ∈ A.
This contradicts (3) above.

Remark 3.10 Given any q 2, there are square free monomial ideals I with q
generators such that L2 (I ) = L2q and the resolution supported on L2 (I ) is minimal.
The ideal I = (xabc, yade, zbdf, wcef ) is such an example when q = 4, (see [3]).
4 A Bound on the Betti Numbers of I 2
We now consider bounds on the Betti numbers of the second power of a square-free
monomial ideal I , as provided by the simplicial complex L2 (I ). Since I 2 has a free
resolution supported on L2 (I ), βd (I 2 ) is bounded above by the number of d-faces
of L2 (I ), which itself is bounded above by the number of d-faces of L2q .
It can be seen from the proof of Corollary 4.1 below that the right-hand term
of the inequality (a) below is precisely the number of d-faces of L2q . Note that the
bound in (a) depends only on the number of generators q, and not on I itself. The
right-hand term of the inequality (b) below is equal to the number of d-dimensional
faces of L2 (I ), which provides a more precise bound that is dependent on the ideal I .
Theorem 4.1 Let I be a square-free monomial ideal minimally generated by q
monomials, where q ≥ 2. Then for each d 0 the d th Betti number βd (I 2 ) satisfies
1
2 (q − q)
2 q −1
(a) βd (I )
2
+q .
d +1 d
Furthermore, setting s to be the minimal number of generators of I 2 and ti to be the

number of vertices of the form
i,j that were deleted from L2q when forming L2 (I ),
then
q
s−q q − 1 − ti
(b) βd (I )
2
+ .
d +1 d
i=1
By Remark 3.10, the bound in (a) is sharp.

Proof We begin by proving inequality (b). Theorem 3.9 gives, for each d ≥ 0,
that βd (I 2 ) is bounded above by the number of d-dimensional faces of L2 (I ). We
compute this number next.
The faces of L2 (I ) are of two types:
(1) Faces that do not contain any vertex of the form
i,i for i ∈ [q].
(2) Faces that contain a vertex
i,i for some i ∈ [q], and, as a consequence, all the
other vertices have the form
i,j with j ∈ [q]
{i}.
!
Let s denote the minimal number of generators of I 2 and set t = q+1 − s.
q+1!
2
Since 2 is the number of vertices of Lq , the integer t is precisely the number of
2
vertices that are deleted in the construction of L2 (I ), as described in Definition 3.4.

As noted in Remark 3.5, all the deleted vertices
i,j must satisfy ! i = j , hence the
number of vertices
i,j of L2 (I ) with i, j ∈ [q] and i = j is q2 − t, which is equal
to s − q.
To construct a d-dimensional face of type (1), we need to choose d + 1 vertices
among the vertices
i,j of L2 (I ) with i, j ∈ [q] and i = j . As noted above, there are
s−q !
s − q such vertices. Thus, the number of d-dimensional faces of type (1) is d+1 .
Fix i ∈ [q]. To construct a d-dimensional face of type (2) that contains
i,i , we
need to choose d vertices among the vertices
i,j of L2 (I ) that satisfy j = i. There
are q − 1 − ti such vertices, where ti denotes the number of vertices
i,j of L2q
! in L (I ). Thus the number of d-dimensional faces of type (2) is
that 2
q areq−1−t
deleted
i
i=1 d .
Putting the two computations above together, wehave that the ! number of
s−q ! q
d-dimensional faces of L2 (I ) is equal to d+1 + i=1 q−1−t d
i
, yielding the
inequality (b). !
Note that inequality (a) follows from (b) by setting ti = 0 for all i and s = q+12 .
In view of our computation above, the right-hand side of inequality (a) is precisely
the number of d-dimensional faces of L2q .

For comparison, the fact that Taylor(I 2 ) supports a free resolution of I 2 gives an
inequality
1
2 (q + q)
2
βd (I 2 ) ,
d +1
where the binomial on the right side denotes the number of d-faces of a 12 (q 2 + q)-
simplex, which is the largest possible size for Taylor(I 2 ).
To get an idea how much Corollary 4.1 improves on this bound, we present the
following table, for q = 4:
d 0 1 2 3 4 5 6
d-faces of largest possible Taylor(I 2 )
10 45 120 210 252 210 120
10 !
d+1
d-faces of L2q
10 27 32 19 6 1 0
6 ! 3!
d+1 + 4 d
To put this in context, we examine two specific ideals with 4 generators, and use
Macaulay2 to find the Betti numbers of these ideals.
Example 4.2 For the ideal J = (x, y, z, w) Macaulay2 gives the following Betti
table for J 2 :
d 0 1 2 3
βd (J 2 ) 10 20 15 4
These Betti numbers should be compared with the bounds in the table above.
Now let I = (abe, bc, cdf, ad) be the ideal Example 3.6. The Betti numbers of
I 2 as calculated by Macaulay2 are the following.
d 0 1 23
βd (I 2 ) 9 14 6 0
In this case we should compare these Betti numbers with the bounds given by the
Taylor complex with 9 vertices and the bounds given by the Corollary 4.1(b). For
the given ideal, we saw that L2 (I ) has 9 vertices, and m1 m3 is an eliminated vertex,
hence s = 9, t2 = t4 = 0 and t1 = t3 = 1 in Corollary 4.1(b). We have:
d 0 1 2 3 4 5 6
d-faces of Taylor(I 2 )
9 36 84 126 126 84 36
9 !
d+1
d-faces of L2 (I )
9 20 18 7 1 0 0
5 ! 3! 2!
d+1 + 2 d + 2 d
Acknowledgments The bulk of this work was done during the 2019 Banff workshop “Women
in Commutative Algebra”. We are grateful to the organizers, the funding agencies (NSF DMS-
1934391), and to the Banff International Research Station for their hospitality.
Author Şega and Spiroff were partially supported by grants from the Simons Foundation
(#354594, #584932, respectively), and authors Cooper and Faridi were supported by the Natural
Sciences and Engineering Research Council of Canada (NSERC).
References
1. D. Bayer, B. Sturmfels: Cellular resolutions of monomial modules. J. Reine Angew. Math. 503
(1998) 123–140.
2. D. Bayer, I. Peeva, B. Sturmfels: Monomial resolutions. Math. Res. Lett. 5, no. 1–2 (1998)
31–46.
3. S. M. Cooper, S. El Khoury, S. Faridi, S. Mayes-Tang, S. Morey, L. M. Şega and S. Spiroff:
Powers of graphs & applications to resolutions of powers of monomial ideals. Preprint (2022).
arXiv:210.07703.
4. S. Faridi: The facet ideal of a simplicial complex. Manuscripta Math. 109, no. 2, 159–174
(2014).
5. S. Faridi: Monomial resolutions supported by simplicial trees. J. Commut. Algebra 6, no. 3
(2014).
6. S. Faridi, B. Hersey: Resolutions of monomial ideals of projective dimension 1. Comm.
Algebra 45, no. 12, 5453–5464 (2017).
7. D.R. Grayson, M.E. Stillman: Macaulay2, a software system for research in algebraic
8. I. Peeva: Graded syzygies. Algebra and Applications 14. Springer-Verlag London, Ltd.,
London, (2011).
9. G. Lyubeznik: A new explicit finite free resolution of ideals generate by monomials in an R-
sequence, J. Pure Appl. Algebra 51, 193–195 (1998).
10. D. Taylor: Ideals generated by monomials in an R-sequence. Thesis, University of Chicago
(1966).
11. X. Zheng: Resolutions of facet ideals. Comm. Algebra 32 no. 6 (2004) 2301–2324.
Cohen–Macaulay Fiber Cones and
Defining Ideal of Rees Algebras of
Modules
Alessandra Costantini
Keywords Fiber cones of modules · Generic Bourbaki ideals · Deformation ·

Iterated Jacobian duals
1 Introduction
In this paper we study the Rees algebra R(E) and the fiber cone F(E) of a finitely
generated R-module E, where R is either a Noetherian local ring or a standard
graded ring. The Rees algebra R(E) is defined as the symmetric algebra S(E)
modulo its R-torsion submodule. The fiber cone of F(E) is then obtained by
tensoring the Rees algebra with the residue field k of R.
Much of the motivation to study Rees algebras and fiber cones comes from
algebraic geometry. Indeed, Rees algebras arise for instance as homogeneous
coordinate rings of blow-ups of schemes along one or more subschemes, or as
bihomogeneous coordinate rings of graphs of rational maps between varieties in
projective spaces. Correspondingly, the fiber cone is the homogeneous coordinate
ring of the special fiber of the blow-up at the unique closed point, or of the image
of the given rational map. In many situations, these are Rees algebras and fiber
cones of modules which are not ideals, for instance when considering a sequence
of successive blow-ups of a scheme along two or more disjoint subschemes,
or the Gauss map on an algebraic variety. In addition, Rees algebras and fiber
cones of modules arise when studying how algebraic varieties and morphisms
change as they vary in families, in connection with the theory of multiplicity and
Whitney equisingularity [39, 40]. Algebraically, this relates to the study of integral
dependence of ideals and modules [12, 20, 21, 29, 31, 32, 37, 41, 42].
In this paper we have two main goals. The first is to understand the Cohen–
Macaulay property of the fiber cone F(E) of a module E. In [38], Simis, Ulrich
A. Costantini ()
University of California Riverside, Riverside, CA, USA
e-mail: alessanc@ucr.edu
208 A. Costantini
and Vasconcelos introduced the notion of generic Bourbaki ideals to reduce the
study of the Cohen–Macaulay property of Rees algebras of modules to the case
of ideals, exploiting the remarkable fact that torsion-free module of rank one are
isomorphic to ideals of positive grade (see Sect. 2 for details). However, to the
best of our knowledge there is no known general technique to study the Cohen–
Macaulayness of fiber cones of modules that are not ideals, and this property is
understood only for a few classes of modules (see for instance [8, 23, 24]).
We address this issue using generic Bourbaki ideals, and show that they allow
to reduce the study of the Cohen–Macaulay property of fiber cones of modules to
the case of ideals, at least in the case when enough information on the Rees algebra
R(E) of E is available. More precisely, our main result, Theorem 4.8, is a refined
version of the following theorem.
Theorem 1.1 Let R be a Noetherian local ring, E a finite R-module with a rank,
and let I be a generic Bourbaki ideal of E.
(a) If F(E) is Cohen–Macaulay, then F(I ) is Cohen–Macaulay.
(b) Assume that after a generic extension the Rees algebra R(E) is a deformation
of R(I ). If F(I ) is Cohen–Macaulay, then F(E) is Cohen–Macaulay.
The deformation condition in Theorem 1.1(b) is satisfied in particular whenever
the Rees algebra of E is Cohen–Macaulay. In general, the Cohen–Macaulayness of
F(E) and of R(E) are not related to each other. Indeed, suppose that R is Cohen–
Macaulay and that E isomorphic to an R-ideal I of positive grade. Then the Rees
algebra R(E) is isomorphic to the subalgebra
R(I ) = R[I t] = ⊕j 0 I j t j
of the polynomial ring R[t], and is known to be Cohen–Macaulay whenever the

associated graded ring G(I ) = ⊕j 0 I j /I j +1 is Cohen–Macaulay and some
additional numerical conditions are satisfied [14, 15, 19, 36]. However, one can
construct perfect ideals I of height two over a power series ring over a field so that
F(I ) is Cohen–Macaulay while R(I ) is not (see the introduction of [4]). Moreover,
for some ideals I defining monomial space curves one has that R(I ) is Cohen–
Macaulay while F(I ) is not [9].
Nevertheless, in some circumstances the Cohen–Macaulay property of the Rees
algebra R(I ) of an ideal I implies that of the fiber cone F(I ) (see for instance
[4, 25]), so it makes sense to investigate similar connections in the case of Rees
algebras and fiber cones of modules as well. Our Theorems 4.9 and 4.11 provide
classes of modules whose Rees algebra and fiber cone are both Cohen–Macaulay,
generalizing previous work of Corso, Ghezzi, Polini and Ulrich on the fiber cone of
an ideal (see [4, 3.1 and 3.4]).
A class of modules with Cohen–Macaulay fiber cone but non-Cohen–Macaulay
Rees algebra is instead given in Theorem 5.6. In this case, the deformation condition
in Theorem 1.1(b) is guaranteed by Theorem 3.1, which is a crucial technical result
Cohen–Macaulay Fiber Cones and Defining Ideal of Rees Algebras of Modules 209
in this work, as it in fact extends the applicability of generic Bourbaki ideals to the
study of Rees algebras which are not necessarily Cohen–Macaulay, nor even S2 .
The second main goal of this work is to study the defining ideal of Rees algebras
of modules. Recall that for an R-module E = Ra1 + . . . + Ran the defining ideal
of R(E) is the kernel of the natural homogeneous epimorphism
φ : R[T1 , . . . , Tn ] −→ R(E)
Ti → ai ∈ [R(E)]1
Determining the defining ideal is usually a difficult task, but it becomes

treatable for Rees algebras of ideals or modules whose free resolutions have a
rich structure. Using generic Bourbaki ideals, in [38, 4.11] Simis, Ulrich and
Vasconcelos determined the defining ideal of R(E) in the case when E is a module
of projective dimension one with linear presentation matrix over a polynomial ring
k[X1 , . . . , Xd ], where k is a field. Their proof ultimately relies on the fact that a
generic Bourbaki ideal I of E has Cohen–Macaulay Rees algebra R(I ), which
allows to deduce the shape of the defining ideal of R(E) from that of R(I ). In
fact, their proof only requires that after a generic extension R(E) is a deformation
of R(I ).
With a similar approach, the deformation condition of Theorem 3.1 allows us
to describe the defining ideal of the Rees algebra of an almost linearly presented
module E of projective dimension one over k[X1 , . . . , Xd ] (see Theorem 5.6). This
condition means that all entries in a presentation matrix of E are linear, except
possibly those in one column, which are assumed to be homogeneous of degree
m 1.
Our result generalizes work of Boswell and Mukundan [2, 5.3] on the Rees
algebra of almost linearly presented perfect ideals of height two. While this
manuscript was being written, in his Ph.D. thesis [46] Matthew Weaver extended
Boswell and Mukundan’s techniques to linearly presented perfect ideals of height
two over a hypersurface ring R = k[X1 , . . . , Xd ]/(f ) , and used our methods to
determine the defining ideal of the Rees algebra of linearly presented modules of
projective dimension one. His work suggests potential applications to the case of
Rees algebras and fiber cones of modules of projective dimension one over complete
intersection rings, which include the module of Kähler differentials of such a ring R.
This is particularly interesting from a geometrical perspective, since its fiber cone
is the homogeneous coordinate ring of the tangential variety to the algebraic variety
defined by the ring R.
We now briefly describe how this paper is structured.
In Sect. 2 we give the necessary background on Rees algebras and fiber cones of
modules and set up the notation that will be used throughout the paper. In particular,
we briefly review the construction and main properties of generic Bourbaki ideals
from [38], as well as Boswell and Mukundan’s construction of iterated Jacobian
duals [2], which we will need later in Sect. 5.
210 A. Costantini
Section 3 contains our main technical result, namely the deformation condition
of Theorem 3.1, which is going to be crucial throughout the paper and in particular
in the proofs of Theorems 4.8, 5.3 and 5.6.
In Sect. 4 we study the Cohen–Macaulay property of fiber cones of modules
via generic Bourbaki ideal. Our main results are Theorem 4.8, which reduces
the problem to the case of fiber cones of ideals, as well as Theorem 4.9 and
Theorem 4.11, which produce modules with Cohen–Macaulay fiber cones.
Section 5 is dedicated to the study of the defining ideal of Rees algebras of mod-
ules. Besides the aforementioned Theorem 5.6 on almost linearly presented modules
of projective dimension one, another key result in this section is Theorem 5.3, which
characterizes the fiber type property of a module over a standard graded k-algebra,
where k is a field.
2 Preliminaries
In this section we recall the definitions and main properties of Rees algebras and
fiber cones of modules, and review the construction of generic Bourbaki ideals.
2.1 Rees Algebras and Fiber Cones of Modules
Unless otherwise specified, throughout this work, R will be a Noetherian local ring
and all modules will be assumed to have a rank. Recall that a finite R-module E has
a rank, rank E = e, if E ⊗R Quot(R) ∼ = (Quot(R))e , or, equivalently, if Ep ∼
= Rpe
for all p ∈ Ass(R).
This is not a restrictive assumption. In fact, if R is a domain all finite R-
modules have a rank. Moreover, every module with a finite free resolution over
any Noetherian ring has a rank. Most importantly, torsion-free modules of rank one
are isomorphic to ideals of positive grade, which is crucial for our purposes.
ϕ
In this setting, let R s −→ R n E be any presentation of E = Ra1 +. . .+Ran .
Then, the natural homogeneous epimorphism
φ : R[T1 , . . . , Tn ] −→ S(E)
Ti → ai ∈ E = [S(E)]1
onto the symmetric algebra of E induces an isomorphism
S(E) ∼
= R[T1 , . . . , Tn ]/L,
where the ideal L is generated by linear forms

1 , . . . ,
s in R[T1 , . . . , Tn ] so that
[T1 , . . . , Tn ] · ϕ = [
1 , . . . ,
s ].
This definition is independent of the choice of the presentation matrix ϕ (see for
instance [3, Section 1.6]).
The Rees algebra R(E) of E is the quotient of S(E) modulo its R-torsion
submodule. In particular,
R(E) ∼
= R[T1 , . . . , Tn ]/J
for some ideal J, called the defining ideal of R(E). Notice that by construction
J ⊇ L and the module E is said to be of linear type if equality holds, since in this
case J is generated by linear equations.
Let k be the residue field of R. The fiber cone (or special fiber ring) of E is
defined as
F(E) := R(E) ⊗R k
(see [11, 2.3]). It can be described as
F(E) ∼
= k[T1 , . . . , Tn ]/I
for some ideal I in k[T1 , . . . , Tn ]. The Krull dimension

(E) := dim F(E) is
called the analytic spread of E (see [11, 2.3]) and satisfies the inequality
e
(E) dim R + e − 1
whenever dim R > 0 and rank E = e (see [38, 2.3]).

Similarly as for powers of an ideal I , one defines the power E j of a module
E as the j -th graded component of the Rees algebra R(E). A reduction of E is a
submodule U ⊆ E so that E r+1 = U E r for some integer r 0. The least such
r is denoted by rU (E). A reduction U of E is a minimal reduction if it is minimal
with respect to inclusion and the reduction number of E is
r(E) := min {rU (E) | U is a minimal reduction of E}
(see [11, 2.3]). Moreover, if k is infinite then any minimal reduction of E is

generated by
(E) elements, and any general
(E) elements in E generate a minimal
reduction U of E with rU (E) = r(E).
2.2 Generic Bourbaki Ideals
Generic Bourbaki ideals were introduced by Simis, Ulrich and Vasconcelos in [38]
as a tool to study the Cohen–Macaulay property of Rees algebras of modules. Most
212 A. Costantini
of our technical work in this paper will consist in providing modifications of the
known theory of generic Bourbaki ideals in order to extend their applicability to
new situations. For this reason, we recall their construction and main properties
below, referring the reader to [38] for the proofs.
Notation 2.1 ([38, 3.3]) Let (R, m) be a Noetherian local ring, E a finite R-module
with rank E = e > 0. Let U = Ra1 + · · · + Ran be a submodule of E for some
ai ∈ E, and consider a set of indeterminates
Z = {Zij | 1 i n, 1 j e − 1}.

Denote R := R[Z] and E := E⊗R R . For 1 j e−1, let xj = ni=1 Zij ai ∈

E and F = e−1
j =1 R xj . Also, denote R = R(Z) = R[Z]m R[Z] , E = E ⊗R

R and F = F ⊗R R .
In the setting of Notation 2.1, the existence of generic Bourbaki ideals is
guaranteed by the following result, and exploits the fundamental fact that torsion-
free modules of rank one are isomorphic to ideals of positive grade.
Theorem and Definition 2.2 ([38, 3.2 and 3.3]) Let R be a Noetherian local ring,
and E a finite R-module with rank E = e > 0, U ⊆ E a submodule. Also, assume
that:
(i) E is torsion-free.
(ii) Ep is free for all p ∈ Spec(R) with depth Rp 1.
(iii) grade(E/U ) 2.
Then, for R , E and F as in Notation 2.1, F is a free R -module of rank e − 1 and
E /F is isomorphic to an R -ideal J with grade J > 0. Also, E /F is isomorphic
to an R -ideal I , called a generic Bourbaki ideal of E with respect to U . If U = E,
I is simply called a generic Bourbaki ideal of E.
Generic Bourbaki ideals of E with respect to a submodule U are essentially
unique. Indeed, if K is another ideal constructed as in Definition 2.2 using variables
Y , then the ideals generated by I and K in T = R(Z, Y ) coincide up to
multiplication by a unit in Quot(T ), and are equal whenever I and K have grade
at least 2 (see [38, 3.4]).
Notice that assumption (iii) in Theorem 2.2 is automatically satisfied if U is a
minimal reduction of E. Moreover, if in this case I ∼ = E /F is a generic Bourbaki
∼
ideal with respect to U , then the ideal K = U /F is a minimal reduction of

I . Sometimes it is possible to relate the reduction number of E and the reduction

number of I , as described in part (d) of the following theorem, which summarizes
the main properties of generic Bourbaki ideals.
Theorem 2.3 ([38, 3.5]) In the setting of Notation 2.1, let U be a reduction of E.
Let I be a generic Bourbaki ideal of E with respect to U , and let K ∼ = U /F .
Then the following statements hold.
(a) R(E) is Cohen–Macaulay if and only if R(I ) is Cohen–Macaulay.

(b) E is of linear type and grade R(E)+ e if and only if I is of linear type, if
and only if J is of linear type.
(c) If any of condition (a) or (b) hold, then R(E )/(F ) ∼
= R(I ) and x1 , . . . , xe−1
of F form a regular sequence on R(E ).
(d) If R(E )/(F ) ∼
= R(I ), then K is a reduction of I with rK (I ) = rU (E). In this
case, if in addition the residue field of R is infinite and U = E, then r(E) =
r(I ).
Condition (c) above says that R(E ) is a deformation of R(I ). This is in fact the
key property that allows to transfer properties from R(E) to R(I ) and backwards.
The following result characterizes the deformation property along a Bourbaki exact
sequence.
Theorem 2.4 ([38, 3.11]) Let R be a Noetherian ring, E a finite R-module with
rank E = e > 0. Let 0 → F → E → I → 0 be an exact sequence where F is a
free R-module with free basis x1 , . . . , xe−1 and I is an R-ideal. The following are
equivalent.
(a) R(E)/(F ) is R-torsion free.
(b) R(E)/(F ) ∼
= R(I ).
(c) R(E)/(F ) ∼
= R(I ) and x1 , . . . , xe−1 of F form a regular sequence on R(E).
Moreover, if I is of linear type, then so is E and the equivalent conditions above
hold.
For our purposes, it will often be convenient to think of the rings R and R as
the result of an iterative process, where at each step only n variables are adjoined.
This is formalized in the following notation.
Notation 2.5 Let R be a Noetherian ring, E a finite R-module with positive
rank, U = Ra1 + · · · + Ran a submodule of E for some ai ∈ E. Let
Z1 , . . . Zn be indeterminates, R := R[Z1 , . . . , Zn ], E
:= E ⊗R R,
U :=
n

U ⊗R R, and x :=
i=1 Zi ai ∈ U . If R is local with maximal ideal m, let
mR.
S := R(Z1 , . . . , Zn ) = R
In fact, the rings R and R as in Notation 2.1 are respectively obtained from R
by iterating the construction of the rings R and S as in 2.5 e − 1 times. Moreover,
in Notation 2.3 the Cohen–Macaulay property is transferred from R(E) to R(I ) and
backwards using the following two results iteratively.
Theorem 2.6 ([38, 3.6 and 3.8]) In the setting of Notation 2.5, assume that
rank E = e 2 and that E/U is a torsion R-module. Let E := E/ Rx
and

R := R(E )/(x). Then,
(a) x is regular on R(E ).
(b) The kernel of the natural epimorphism π : R R(E) is K = H 0 (R) and
UR

coincides with the R-torsion submodule of R.
(c) If U is a reduction of E and grade R(E)+ 2, then π is an isomorphism.
214 A. Costantini
Theorem 2.7 ([38, 3.7]) In the setting of Notation 2.5, assume that R is local, that
rank E = e 2 and that U is a reduction of E. Let E denote (E ⊗R S)/xS
and R := R(E ⊗R S)/(x). If R(E) satisfies S2 , then the natural epimorphism
π : R R(E) is an isomorphism, and x is regular on R(E ⊗R S). In particular,
R(E) satisfies S2 .
Notice that formation of Rees algebras of finite modules commutes with flat
∼
extensions (see [11, 1.3]). Hence, one has that R(E) as well as
= R(E) ⊗R R,
R(E ⊗R S) ∼ = R(E) ⊗R S. Therefore, tensoring with the residue field k yields
isomorphisms R(E) ⊗R k ∼ and F(E ⊗R S) ∼
= F(E) ⊗R R, = F(E) ⊗R S.
2.3 Iterated Jacobian Duals
When studying the defining ideal of Rees algebras, the most challenging aspect
usually consists in identifying its non-linear part. In many cases of interest [2, 18,
22, 26, 27, 30, 35, 43, 44], this can be done by examining some auxiliary matrices
associated with ϕ, namely the Jacobian dual or the iterated Jacobian duals of ϕ,
introduced by Vasconcelos [44] and by Boswell and Mukundan [2] respectively. We
briefly recall these notions here, as we will use them intensively in Sect. 5.
Although both definitions make sense over any Noetherian ring, for our purposes
we assume that R = k[Y1 , . . . , Yd ] is a standard graded polynomial ring over a field
k. Let S = R[T1 , . . . , Tn ] be bigraded, and set Y = Y1 , . . . , Yd , T = T1 , . . . , Tn .
Let
ϕ
R s −→ R n
be an n×s matrix whose entries are homogeneous of constant Y -degrees δ1 , . . . , δs

along each column and assume that I1 (ϕ) ⊆ (Y ).
Theorem and Definition 2.8 ([44]) With R, S and ϕ as above, let M = coker(ϕ)
and let
1 , . . . ,
s be linear forms in the Ti variables, generating the defining ideal
of the symmetric algebra S(M). Then,
(a) There exists a d × s matrix B(ϕ) whose entries are linear in the Ti variables and
homogeneous of constant Y -degrees δ1 − 1, . . . , δs − 1 along each column,
satisfying
[
1 , . . . ,
s ] = [T ] · ϕ = [Y ] · B(ϕ).
B(ϕ) is called a Jacobian dual of ϕ.

(b) B(ϕ) is not necessarily unique, but it is if the entries of ϕ are all linear.
Moreover, by Cramer’s rule it follows that L + Id (B(ϕ)) ⊆ J.
For a matrix A, let (Y · A) denote the ideal generated by the entries of the row
vector [Y ] · A.
Theorem and Definition 2.9 ([2, 4.1, 4.2 and 4.5]) With R, S and ϕ as above, let
B1 (ϕ) = B(ϕ) for some Jacobian dual B(ϕ) of ϕ. Assume that matrices Bj (ϕ) with
d rows have been inductively constructed for 1 j i, such that each Bj (ϕ)
has homogeneous entries of constant Y -degrees and T -degrees along each column.
There exists a matrix Ci whose entries in S are homogeneous of constant Y -degrees
and T -degrees in each column, such that Bi+1 (ϕ) := [Bi (ϕ) | Ci ] satisfies
(Y · Bi (ϕ)) + (Id (Bi (ϕ)) ∩ (Y )) = (Y · Bi (ϕ)) + (Y · Ci ).
A matrix Bi (ϕ) as above is called an i-th iterated Jacobian dual of ϕ. Moreover,

for all i 1:
(a) The ideal (Y · B(ϕ)) + Id (Bi (ϕ)) only depends on ϕ.
(b) (Y · Bi (ϕ)) + Id (Bi (ϕ)) = (Y · B(ϕ)) + Id (Bi (ϕ)) ⊆ (Y · B(ϕ)) + Id (Bi+1 (ϕ)).
In particular, there exists an N > 0 so that (Y · B(ϕ)) + Id (Bi (ϕ)) = (Y ·
B(ϕ)) + Id (Bi+1 (ϕ)) for all i N .
(c) (Y · B(ϕ)) + Id (Bi (ϕ)) ⊆ ((Y · B(ϕ)) : (Y )i ).
3 A Deformation Condition for the Rees Algebra of a Module
As described in Sect. 2, transferring properties from a module E to a generic

Bourbaki ideal I of E and backwards depends on whether the Rees algebra R(E )
is a deformation of R(I ). Theorems 2.3 and 2.4 show that this always occurs when
I is of linear type, or if R(E) or R(I ) are known to be Cohen–Macaulay. On the
other hand, it is interesting to find alternative conditions on E or I that guarantee
this deformation property.
Inspired by [38, 3.7] (which was stated here as Theorem 2.7), the following
result provides a new deformation condition, which applies to ideals and modules
not necessarily of linear type and whose Rees algebras are not necessarily Cohen–
Macaulay (see for instance Theorems 4.8 and 5.6).
Theorem 3.1 Let R be a Noetherian local ring, E a finite R-module with rank E =
e 2 and let U = Ra1 + · · · + Ran be a reduction of E. Let S and x be as in
Notation 2.5 and denote E := (E ⊗R S)/Sx.
Assume that depth R(E q ) 2 for all q ∈ Spec(S) such that E q is not of linear
type. Then, the natural epimorphism
π : R(E ⊗R S)/(x) R(E)
is an isomorphism, and x is regular on R(E ⊗R S).

Proof We modify the proof of [38, 3.7]. Since x is regular on R(E ⊗R S) by
Theorem 2.6(a), we only need to show that K = ker(π ) is zero. In fact, we only need
to prove this locally at primes q ∈ Spec(S) such that E q is not of linear type. Indeed,
216 A. Costantini
∼ S(E q ) is isomorphic to S((E ⊗R S)q )/(x)

if E q is of linear type, then R(E q ) =
by construction, whence Kq = 0.
Notice that, after localizing S at a minimal prime p of S/(0 : K) if needed,
we may assume that K vanishes locally on the punctured spectrum of S. Indeed,
replacing S with Sp changes the shape of S, but does not affect the assumptions
nor the rest of the proof. Also, for a non-maximal prime q ∈ Spec(Sp ) we have
(0 : Kq ) = Sq , so Kq = 0 as claimed.
Now, let R denote R(E ⊗R S)/(x) and let M = (m, R(E ⊗R S)+ ) be the
unique homogeneous maximal ideal of R(E ⊗R S). Since K vanishes locally on
the punctured spectrum, K is annihilated by a power of m. Also, by Theorem 2.6
it follows that K is annihilated by a power of U R, and hence by a power of
ER = (R)+ , since E is integral over U . Hence, K ⊆ HM 0 (R).
Thus, for all q ∈ Spec(S) Kq ⊆ HMq (Rq ) and it suffices to show that
0
HM0 (R ) = 0 whenever E is not of linear type. Consider the long exact sequence
q q q
of local cohomology induced by the exact sequence
0 → Kq → Rq → R(E q ) → 0 .
Since by assumption depth R(E)q ) 2, then HM i (R ) ∼ H i (K ) for

q q = Mq q
1 (K ) ∼
i = 0, 1. In particular, since Kq ⊆ HMq (Rq ), it follows that 0 = HM
0
q q =
1 (R ). Therefore, the exact sequence
HMq q
x
0 → R((E ⊗R S)q )(−1) −→ R((E ⊗R S)q ) −→ Rq → 0
induces the exact sequence

x
0 → HM
0
q
(Rq ) −→ HM
1
q
(R((E ⊗R S)q ))(−1) −→ HM
1
q
(R((E ⊗R S)q )) → 0 .
1 (R((E ⊗ S) )) is finitely
Now, similarly as in [38, 3.7], one can show that HMq R q
generated, as a consequence of the graded version of the Local Duality Theo-
rem. Therefore, by the graded version of Nakayama’s Lemma, it follows that
HM1 (R((E ⊗ S) )) = 0, whence also H 0 (R ) = 0.

q R q Mq q
4 Cohen–Macaulay Property of Fiber Cones of Modules
In this section we examine the Cohen–Macaulay property of fiber cones of modules.

We first show that the construction of generic Bourbaki ideals allows to reduce
the problem to the case of ideals, as long as the passage to a generic Bourbaki
ideal induces a deformation between the Rees algebras (see Theorem 4.8). We then
provide sufficient conditions for the fiber cone of a module to be Cohen–Macaulay,
generalizing known results of Corso, Ghezzi, Polini and Ulrich for the fiber cone of
ideals [4, 3.1 and 3.4].
Let I be a generic Bourbaki ideal of E. The proof of [38, 3.5] (which was
stated here as Theorem 2.3) suggests that, in order to transfer the Cohen–Macaulay
property from F(E) to F(I ) and backwards, the natural map
π : F(E ) ∼
= R(E ) ⊗R k R(I ) ⊗R k ∼
= F(I )
needs to be an isomorphism. Hence, it suffices to provide conditions on the module

E or on the ideal I so that this isomorphism is guaranteed. Our first goal in this
direction is to prove that an analogous statement as that of Theorem 2.6 holds for
fiber cones. This is done through the next two propositions.
Proposition 4.1 Let (R, m, k) be a Noetherian local ring, E a finite R-module with
rank E = e 2, and let U be a submodule of E such that E/U is torsion. In the
setting of Notation 2.5, let L be the kernel of the natural epimorphism

π : (F(E) ⊗R R)/(x) Rx)
R(E/ ⊗R k.
Then,

(a) L ⊆ HU0 ((F(E) ⊗R R)/(x)).
(b) If in addition U is a reduction of E and depth F(E) > 0, then x is regular on
F(E) ⊗R R.

Proof Let R denote R(E)/(x). By Theorem 2.6(b), there is an exact sequence
ι π
Rx)
0 → K −→ R −→ R(E/ →0
where K = H 0 (R). Tensoring with the residue field k, it then follows that
UR

L = (ι ⊗ k)(H 0 (R) ⊗R k) ⊆ HU0 ((F(E) ⊗R R)/(x)).
UR
n proves (a). For part (b), notice that if U = Ra1 + · · · + Ran , then x =
This
i=1 Zi ai is a generic linear combination of the generators of the ideal U F(E)
of F(E). Since grade U F(E) = depth F(E) > 0, then x is regular on F(E) ⊗R R
(see, for instance, [13, p.1] or [28, (6.13), p.17]).

Proposition 4.2 Let (R, m, k) be a Noetherian local ring, E a finite R-module with
rank E = e 2, and let U be a reduction of E. With the notation of Theorem 4.1,
assume that depth F(E) 2. Then,

π : (F(E) ⊗R R)/(x) Rx)
R(E/ ⊗R k.

is an isomorphism, and x is regular on F(E) ⊗R R.
218 A. Costantini
Proof Let R denote R(E ⊗R R)/(x). By Proposition 4.1 it follows that L =

ker(π ) = (ι ⊗ k)(H 0 (R) ⊗R k) ⊆ HU0 ((F(E) ⊗R R)/(x)) and that x is regular
UR
on F(E) ⊗R R. In particular, there is an exact sequence

0 → (F(E) ⊗R R)(−1)
x
−→ (F(E) ⊗R R)/(x)
−→ F(E) ⊗R R → 0.
= 0 since
Now, notice that HU1 (F(E) ⊗R R)
= grade EF(E) ⊗R R
grade U F(E) ⊗R R depth F(E) 2.
Hence, the long exact sequence of local cohomology implies that

HU0 ((F(E) ⊗R R)/(x)) = 0.
Thus, L = 0 and π is an isomorphism.

By applying Proposition 4.2 repeatedly, we obtain the following useful corollary.
Corollary 4.3 Let R be a Noetherian local ring, and let E be a finite R-module
with rank E = e. In the setting of Notation 2.1, let I be a generic Bourbaki ideal of
E with respect to a reduction U of E. Assume that depth F(E) e, then the natural
epimorphism
π : F(E )/(F ) F(I )
is an isomorphism and F F(E ) is generated by a regular sequence of linear

forms.
We now proceed to set up the technical framework in order for the Cohen–
Macaulay property to be transferred from F(I ) back to F(E). The key result is
Theorem 4.6, whose proof relies on the next two lemmas. Lemma 4.4 states that x
with respect to the ideal E(F(E) ⊗R R)
is a filter-regular element on F(E) ⊗R R
(see for instance [33, p. 13]).
Lemma 4.4 Let R be a Noetherian local ring, E a finite R-module with rank E =
e 2, and let U be a reduction of E. Then, in the setting of Notation 2.5,

SuppF(E)⊗R R (0 : F(E)⊗R R x) ⊆ V (E(F(E) ⊗R R)).
Proof Let q ∈ Spec(F(E) ⊗R R) \ V (E(F(E) ⊗R R))

and let p = q ∩ F(E). Since
U = R a1 +. . .+R an is a reduction of E, it follows that q U (F(E)⊗R R). Hence,
pU F(E), which means that at least one of the ai is a unit in F(E)p . Therefore,
x = ni=1 Zi ai is a nonzerodivisor in F(E)p [Z1 , . . . , Zn ], hence also in its further
q . Thus, (0 :
localization (F(E)[Z1 , . . . , Zn ])q = (F(E) ⊗R R) F(E)⊗R R x)q = 0.

Lemma 4.5 Let R be a positively graded Noetherian ring with R0 local, and let x
be a homogeneous non-unit element of R. Let M be a finite graded R-module, and
assume that dim (0 : M x) < depth (M/xM). Then, x is a nonzerodivisor on M.
0 (M) = 0, which would follow from Nakayama’s
Proof It suffices to show that H(x)
Lemma once we prove that H(x) 0 (M)/xH 0 (M) = 0. For this, it suffices to show
(x)
0 (M)/xH 0 (M)) = ∅.
that Ass(H(x) (x)
Consider the short exact sequences
0 → 0 : M x → M → M/(0 : M x) → 0 (4.1)
and
x
0 → M/(0 : M x) −→ M → M/xM → 0. (4.2)
Since (0 : M x) = H(x)
0 (0 : x), it follows that H 1 (0 : x) = 0. Hence, the long
M (x) M
0 (M) surjects
exact sequence of local cohomology induced by (4.1) implies that H(x)
0 (M/(0 : x)). Therefore, the long exact sequence of local cohomology
onto H(x) M
induced by (4.2)
x
0 → H(x)
0
(M/(0 : M x)) −→ H(x)
0
(M) → H(x)
0
(M/xM)
in turn induces an exact sequence

x
0
H(x) (M) −→ H(x)
0
(M) → H(x)
0
(M/xM) ⊆ M/xM.
0 (M)/xH 0 (M) embeds into M/xM.

In particular, H(x) (x)
Also, notice that (0 : x) = 0 if and only if H(x) 0 (M) = 0, hence if
and only if H(x) 0 (M)/xH 0 (M) = 0 by Nakayama’s Lemma. Therefore,

(x)
Supp(0 : M x) = Supp(H(x) 0 (M)/xH 0 (M)). Hence, if there exists some p ∈
(x)
Ass(H(x)0 (M)/xH 0 (M)), then dim(R/p) dim(0 : x). On the other hand,
(x) M
since Ass(H(x)0 (M)/xH 0 (M)) ⊆ Ass(M/xM), we also have that dim(R/p)
(x)
depth(M/xM). But then
dim(0 : M x) = dim(H(x)
0 0
(M)/xH(x) (M)) depth(M/xM),
0 (M)/xH 0 (M)) =
which contradicts the assumption. So, it must be that Ass(H(x) (x)
∅, as we wanted to prove.

Theorem 4.6 In the setting of Notation 2.5, assume that R is local and that
rank E = e 2. Let U be a reduction of E, and denote E := (E ⊗R S)/Sx.
Assume that one of the two following conditions hold:
220 A. Costantini
(i) R(E) satisfies S2 , or

(ii) depth R(E q ) 2 for all q ∈ Spec(S) such that E q is not of linear type.
Then, the natural epimorphism
π : F(E ⊗R S)/(x) F(E)
is an isomorphism. Moreover, x is regular on F(E ⊗R S) if depth F(E) > 0.

Proof Assumption (i) and Theorem 2.7 together imply that the natural epimorphism
π : R(E ⊗R S)/(x) R(E)
is an isomorphism. The same conclusion holds if assumption (ii) is satisfied, thanks

to Theorem 3.1. Hence, π : F(E ⊗R S)/(x) F(E) is an isomorphism as well.
In particular, if in addition depth F(E) > 0 then depth (F(E ⊗R S)/(x)) > 0.
Moreover, by Lemma 4.4 we know that (0 : F(E⊗R S) x) is an Artinian F(E ⊗R S)-
module. Hence, x is regular thanks to Lemma 4.5.

Corollary 4.7 Let R be a Noetherian local ring, and let E be a finite R-module
with rank E = e. In the setting of Notation 2.1 I be a generic Bourbaki ideal of E
with respect to a reduction U of E.
(a) Assume that either R(I ) is S2 , or depth R(Iq ) 2 for all q ∈ Spec(R ) so
that Iq is not of linear type. Then, the natural epimorphism
π : F(E )/(F ) F(I )
is an isomorphism.
(b) If in addition F(I ) is Cohen–Macaulay, then F F(E ) is generated by a
regular sequence of linear forms.
Proof Let T = R(Z1,1 , . . . , Zn,1 , . . . , Z1,e−2 , . . . , Zn,e−2 ). Then, R can be
rewritten as R = T (Z1,e−1 , . . . , Zn,e−1 ). If the assumptions in (a) are satisfied,
then from Theorem 2.7 or Theorem 3.1 respectively it follows that
88 E ⊗ T 9 9
R(I ) ∼ ⊗T R /(R xe−1 )
R
=R e−2
j =1 T xj
88 E ⊗ T 9 9
⊗T R . In particular,
R
and that xe−1 is a regular element on R e−2
j =1 T xj
E ⊗ T
replacing R with T , the T -module e−2
R
satisfies the same assumption in (a)
j =1 T xj
as I . Hence, iterating the argument, the deformation property for the Rees algebras
considered at each step implies that the assumptions (i) or (ii) in Theorem 4.6 are
satisfied at each iteration.
Thus, R(E )/(F ) ∼= R(I ), and F R(E ) is generated by a regular sequence

on R(E ). Hence, by iteration of Theorem 4.6, it follows that F(E )/(F ) ∼

= F(I ).
Now, if furthermore F(I ) is Cohen–Macaulay, then the proof of Theorem 4.6
implies that also F F(E ) is generated by a regular sequence on F(E ). That the
generators of F F(E ) are linear forms in F(E ) is clear by construction.

We are now ready to state and prove our main result.
Theorem 4.8 Let R be a Noetherian local ring, E a finite R-module with rank E =
e, U a reduction of E. Let I be a generic Bourbaki ideal of E with respect to U .
(a) If F(E) is Cohen–Macaulay, then F(I ) is Cohen–Macaulay.
(b) Assume that either R(I ) is S2 , or depth R(Iq ) 2 for all q ∈ Spec(R ) so
that Iq is not of linear type. If F(I ) is Cohen–Macaulay, then F(E) is Cohen–
Macaulay.
Proof We may assume that e 2. Suppose that F(E) is Cohen–Macaulay, then
depth F(E) =
(E) e. Hence, by Corollary 4.3 it follows that depth F(I ) =
(E) − e + 1. The latter equals

(I ) by [38, 3.10], hence F(I ) is Cohen–Macaulay.
Conversely, if the assumptions in (b) hold, by Corollary 4.3 it follows that
depth F(E) =
(I ) + e − 1 =
(E). Therefore, F(E) is Cohen–Macaulay.

From the proofs of Corollary 4.3 and Theorem 4.8 it follows that F(E) is
Cohen–Macaulay whenever F(I ) is Cohen–Macaulay and R(E ) is a deformation
of R(I ). In particular, finding conditions other than those in Theorem 2.7 or
Theorem 3.1 to guarantee this deformation property for the Rees algebras would
provide alternative versions of Theorem 4.8. In fact, in order to transfer the Cohen–
Macaulay property from F(E) to F(I ) and backwards one would only need
F(E ) to be a deformation of F(I ). Finding conditions for this to occur without
any prior knowledge of the Rees algebra would potentially allow to use generic
Bourbaki ideals also in the case when the Cohen–Macaulayness of F(I ) is possibly
unrelated to that of R(I ) (see for instance [4, 5, 10, 16, 17, 25, 34, 45]).
4.1 Modules with Cohen–Macaulay Fiber Cone
Theorem 4.8 above implies that the fiber cone F(E) of E is Cohen–Macaulay
whenever both the Rees algebras R(I ) and the fiber cone F(I ) of a generic
Bourbaki ideal I of E are Cohen–Macaulay. The goal of this section is to provide
specific classes of modules with this property. A class of modules with Cohen–
Macaulay fiber cone and non-Cohen–Macaulay Rees algebra will be provided later
in Theorem 5.6.
Our first result regards modules of projective dimension one, and extends a
known result proved by Corso, Ghezzi, Polini and Ulrich for perfect ideals of
height two (see [4, 3.4]). Recall that a module E of rank e satisfies condition Gs
if μ(Ep ) dimRp − e + 1 for every p ∈ Spec(R) with 1 dim Rp s − 1.
222 A. Costantini
Moreover, by [38, 3.2] if E satisfies Gs then so does a generic Bourbaki ideal I

of E.
Theorem 4.9 Let R be a local Cohen–Macaulay ring, and let E a finite, torsion-
free R-module with projdim(E) = 1, with
(E) =
. Assume that E satisfies
G
−e+1 . If R(E) is Cohen–Macaulay, then F(E) is Cohen–Macaulay.
Proof Since E is a torsion-free module of projective dimension one which satisfies
G
−e+1 , then E admits a generic Bourbaki ideal I , which is perfect of height 2
(see for instance the proof of [38, 4.7]). If e = 1, then the conclusion follows
from [4, 3.4]. Otherwise, notice that R(I ) is Cohen–Macaulay by Theorem 2.3,
whence F(I ) is Cohen–Macaulay by [4, 3.4]. Hence, F(E) is Cohen–Macaulay by
Theorem 4.8.

Corollary 4.10 Let R be a local Cohen–Macaulay ring, and let E a finite, torsion-
free R-module with projdim E = 1, with
(E) =
. Let n = μ(E) and let
ϕ
0 → R n−e −→ R n → E → 0
be a minimal free resolution of E. Assume that E satisfies G

−e+1 and that one of
the following equivalent conditions hold.
(i) r(E)
− e.
(ii) r(Ep )
− e for every prime p with dim Rp =
(Ep ) − e + 1 =
− e + 1.
(iii) After elementary row operations, In−
(ϕ) is generated by the maximal minors
of the last n −
rows of ϕ.
Then, F(E) is Cohen–Macaulay.
Proof By [38, 4.7], each of the conditions (i)-(iii) is equivalent to R(E) being
Cohen–Macaulay. Hence, the conclusion follows from Theorem 4.9.

The following result was proved for fiber cones of ideals in [4, 3.1].
Theorem 4.11 Let R be a Cohen–Macaulay local ring with infinite residue field.
Let E be a finite, torsion-free R-module with rank E = e > 0,
(E) =
and
r(E) = r. Assume that E satisfies G
−e+1 and
− e + 1 2. Let I be a generic
Bourbaki ideal of E, and g = ht(I ). Suppose that one of the following conditions
holds.
(i) If μ(E)
+ 2, then F(E) has at most two homogeneous generating relations
in degrees max{ r,
− e − g + 1 }.
(ii) If μ(E) =
in degrees
− e − g + 1.
If R(E) is Cohen–Macaulay, then F(E) is Cohen–Macaulay.
Proof By Theorem 2.2, E admits a generic Bourbaki ideal I with ht I = g 1,
μ(I ) = μ(E) − e + 1 and r(I ) r, which satisfies G
−e+1 , i.e. G
(I ) (see
[38, 3.10]). If e = 1 the conclusion follows from [4, 3.1], since G(I ) is Cohen–
Macaulay whenever R(I ) is by [14, Proposition 1.1]. So we may assume that e 2.

We show that both R(I ) and F(I ) are Cohen–Macaulay, whence F(E) is Cohen–
Macaulay by Theorem 4.8.
Since by assumption R(E) is Cohen–Macaulay, then R(I ) is Cohen–Macaulay
by Theorem 2.3. Hence also the associated graded ring G(I ) is Cohen–Macaulay by
[14, Proposition 1.1]. Also, by Corollary 4.7 there is a homogeneous isomorphism
F(E )/(F ) ∼
= F(I ). Therefore, if condition (i) holds, then whenever μ(I )
+ 2 − e + 1 =
(I ) + 2 it follows that F(I ) has at most two homogeneous
generating relations in degrees at most max{ r,
− e − g + 1 }, hence in degrees at
most max{ r(I ),
(I ) − g }. Similarly, in the situation of assumption (ii), whenever
μ(I ) =
(I ) + 1 it follows that F(I ) has at most two homogeneous generating
relations in degrees at most
− e − g + 1 =
(I ) − g. Hence, F(I ) is Cohen–
Macaulay by [4, 3.1].

In particular, with some additional assumptions on the module E, we can give
more explicit sufficient conditions for F(E) to be Cohen–Macaulay. The next two
corollaries exploit results on the Cohen–Macaulay property of Rees rings from [7]
and recover [4, 2.9] in the case when E is an ideal of grade at least two.
<Recall that a module E∗ is called orientable if E has a rank e > 0 and
( e E)∗∗ ∼ = R, where (−) denotes the functor HomR (−, R)).
Corollary 4.12 Let R be a local Gorenstein ring of dimension d with infinite
residue field. Let E be a finite, torsion-free, orientable R-module, with rankE =
e > 0 and
(E) =
. Let g be the height of a generic Bourbaki ideal of E, and
assume that the following conditions hold.
(a) E satisfies G
−e+1 .
(b) r(E) k for some integer 1 k
− e.
# d − g − j + 2 for 1 j
− e − k − g + 1
(c) depth E j
d −
+ e + k − j for
− e − k − g + 2 j k
j +1 j
(1) If g = 2, ExtRp (Ep , Rp ) = 0 for
− e − k j
− e − 3 and for all
p ∈ Spec(R) with dimRp =
− e such that Ep is not free.
Assume furthermore that one of the following two conditions holds.
(i) If μ(E)
in degrees max{ r,
− e − g + 1 }.
(ii) If μ(E) =
in degrees
− e − g + 1 .
Proof Assumptions (a)-(d) together imply that R(E) is Cohen–Macaulay, thanks
to [7, 4.3]. Hence, if either condition (i) or (ii) hold it follows that F(E) is Cohen–
Macaulay by Theorem 4.11.

Recall that an R-module E is called an ideal module if E = 0 is finitely
generated, torsion-free and so that E ∗∗ is free, where −∗ denotes the functor
224 A. Costantini
HomR (−, R). Equivalently, E is an ideal module if and only if E embeds into a
finite free module G with grade(G/E) 2 (see [38, 5.1]).
Corollary 4.13 Let R be a local Cohen–Macaulay ring, and let E be an ideal
module with rank E = e and
(E) =
. Assume that the following conditions
hold.
(a) r(E) k, where k is an integer such that 1 k
− e.
(b) E is free locally in codimension
− e − min{2, k}, and satisfies G
−e+1 .
(c) depth(E j ) d −
+ e + k − j for 1 j k.
Assume furthermore that one of the following two conditions holds.
(i) If μ(E)
in degrees max{ r,
− e − g + 1 }.
(ii) If μ(E) =
in degrees
− e − g + 1 .
Proof From assumptions (a)–(c) it follows that R(E) is Cohen–Macaulay, thanks to
[7, 4.10]. Hence, if either condition (i) or (ii) hold, then F(E) is Cohen–Macaulay
by Theorem 4.11.

5 Defining ideal of Rees Algebras
In this section we use generic Bourbaki ideals to understand the defining ideal of
Rees algebras of modules. The idea is not completely novel, as it appears in the
proof of [38, 4.11], where the authors determine the defining ideal for the Rees
algebra of a module E of projective dimension one having a linear presentation. In
their case, the Rees algebra of a generic Bourbaki ideal I of E is Cohen–Macaulay,
however the latter condition will not be guaranteed nor required in the situations
considered in this section.
The first key observation is that it is always possible to relate a presentation
matrix of E to a presentation matrix of a generic Bourbaki ideal I of E.
Remark 5.1 (See also [38, p. 617]) Let (R, m, k) be a Noetherian local ring and let
ϕ
E be a finite R-module with a minimal presentation R s −→ R n → E → 0 .
(a) With Z and xj as in Notation 2.1, by possibly multiplying ϕ from the left by an
invertible matrix with coefficients in k(Z), we may assume that ϕ presents E
with respect to
a minimal generating set of the form x1 , . . . , xe−1 , ae , . . . , an .
A
Then, ϕ = , where A and ψ are submatrices of size (e − 1) × s and
ψ
(n − e + 1) × s respectively. By construction, ψ is a presentation of I , and is
minimal since μ(I ) = μ(E) − e + 1 = n − e + 1.
(b) Assume that R = Sm , where S is a standard graded algebra over a field and m
is its unique homogeneous maximal ideal. If the entries of ϕ are homogeneous
polynomials of constant degrees δ1 , . . . , δs along each column, then the entries
of ψ are homogeneous polynomials of constant degrees δ1 , . . . , δs along each
column.
Let R be a standard graded algebra over a field k and let E be a finite R-module.
Then, the fiber cone F(E) of E has a particularly useful description as a subring of
a polynomial ring over k, which is summarized in the following remark.
Remark 5.2 Let R be a standard graded algebra over a field k and homogeneous
maximal ideal m. Let E = Ra1 +. . .+Ran be finite R-module minimally generated
by elements of the same degree. On the polynomial ring S = R[T1 , . . . Tn ] define a
bigrading by setting deg Ri = (i, 0) and deg Ti = (0, 1) . Then, the Rees algebra
R(E) ∼= S/J has a natural bigraded structure induced by the bigrading on S and
is generated in degrees (0, 1). Moreover, m R(E) ∼
= [R(E)](>0,−) . Hence, the fiber
cone F(E) satisfies
F(E) ∼
= R(E)/mR(E) ∼
= [R(E)](0,−) ⊆ k[T1 , . . . , Tn ].
As a consequence, the homogeneous epimorphism
k[T1 , . . . , Tn ] = S ⊗R k R(E) ⊗R k = F(E)
has kernel I ∼
= J(0,−) .
Notice also that for an R-module E as in Remark 5.2 the defining ideal L of
the symmetric algebra S(E) satisfies L = J(−,1) . In particular, if IR[T1 , . . . , Tn ]
denotes the extension of I to the ring R[T1 , . . . , Tn ], then
J ⊇ L + IR[T1 , . . . , Tn ].
E is said to be of fiber type if the latter inclusion is an equality, or equivalently, if

J is generated in bidegrees (−, 1) and (0, −). The following theorem characterizes
the fiber type property of modules.
Theorem 5.3 Let R = Sm where S is a standard graded algebra over a field
with unique homogeneous maximal ideal m. Let E be a finite R-module with
rank E = e 2, minimally generated by elements a1 , . . . , an that are images
in R of homogeneous elements of the same degree in S. Assume that E is torsion-
free and that Ep is free for all p ∈ Spec R with depth R 1, and let I be a generic
Bourbaki ideal of E constructed with respect to the generators a1 , . . . , an .
Assume that one of the following conditions hold.
(i) R(I ) satisfies S2 ; or
(ii) depth R(Iq ) 2 for all q ∈ Spec(R ) so that Iq is not of linear type.
Then, E is of fiber type if and only if I is of fiber type.
226 A. Costantini
Proof Let R[T1 , . . . , Tn ] R(E) be the natural epimorphism mapping Ti to ai

n
for all i. As in Notation 2.1, for 1 j e − 1 let xj = Zij ai . For every j
i=1

n
let Xj = Zij Ti ⊆ R [T1 , . . . , Tn ] and notice that Xj is mapped to xj via the
i=1
natural epimorphism R [T1 , . . . , Tn ] R(E ). Let ϕ be a minimal presentation of
E with respect to the generators a1 , . . . , an , let ψ be a minimal presentation of I
constructed as in Remark 5.1, and use these presentations to construct the symmetric
algebras S(E) and S(I ) respectively. Let LE , JE and IE denote the defining ideals
of S(E), R(E) and F(E) respectively. Similarly, let LI , JI and II denote the
defining ideals of S(I ), R(I ) and F(I ) respectively.
By construction it then follows that LI = LE R + (X1 , . . . , Xe−1 ), as well as
JI = JE R + (X1 , . . . , Xe−1 ). Since S is standard graded, this implies that
II = [JI ](0,−) = [JE ](0,−) R + (X1 , . . . , Xe−1 ) = IE R + (X1 , . . . , Xe−1 ).
Hence, I is of fiber type if and only if
JI = LI + [JI ](0,−) R = LE + [JE ](0,−) R + (X1 , . . . , Xe−1 ).
Now, if either assumption (i) or (ii) are satisfied, by Theorem 2.7 or Theorem 3.1
it follows that X1 , . . . , Xe−1 form a regular sequence modulo JE R . Therefore, it
follows that
8 9
JE R = LE R + [JE ](0,−) R + (X1 , . . . , Xe−1 ) ∩ JE R
= LE R + [JE ](0,−) R + (X1 , . . . , Xe−1 ) JE R
By the graded version of Nakayama’s Lemma, this means that
JE R = LE R + [JE ](0,−) R ,
which can occur if and only if in R[T1 , . . . , Tn ] one has JE = LE + [JE ](0,−) ,
i.e. if and only if E is of fiber type.

5.1 Almost Linearly Presented Modules of Projective

Dimension One
In this section we describe the Rees algebra and fiber cone of almost linearly
presented modules of projective dimension one. Throughout we will consider the
situation of Setting 5.4 below.
Setting 5.4 Let R = k[Y1 , . . . , Yd ] be a polynomial ring over a field k, where

d 2. Let E be a finite R-module, minimally generated by homogeneous elements
of the same degree. Assume also that E has projective dimension one and satisfies
Gd . Then, has positive rank e and admits a minimal free resolution of the form
ϕ
0 → R n−e −→ R n → E → 0, where n = μ(E). Assume that ϕ is almost linear,
i.e. has linear entries, except possibly for those in the last column, which are
homogeneous of degree m 1.
In the situation of Setting 5.4, after localizing at the unique homogeneous
maximal ideal, by Theorem 2.2 E admits a generic Bourbaki ideal I , which is
perfect of grade 2. Let ψ be a minimal presentation of I obtained from ϕ as in
Remark 5.1. By construction, ψ is also almost linear. In particular, the defining
ideal of R(I ) is described by the following theorem of Boswell and Mukundan [2,
5.3 and 5.6].
Theorem 5.5 Let R = k[Y1 , . . . , Yd ] be a standard graded polynomial ring over a
field k. Let I be a perfect ideal of height 2 admitting an almost linear presentation
ψ. Assume that I satisfies Gd and that μ(I ) = d + 1. Then, the defining ideal of the
Rees algebra R(I ) is
J = (Y · B(ψ)) + Id (Bm (ψ)) = (Y · B(ψ)) : (Y1 , . . . , Yd )m ,
where m is the degree of the non-linear column of ψ and Bm (ψ) is the mth-iterated
Jacobian dual of ψ as in Definition 2.9. Moreover:
(i) R(I ) is almost Cohen–Macaulay, i.e. depth R(I ) d − 1, and it is Cohen–
Macaulay if and only if m = 1.
(ii) F(I ) is Cohen–Macaulay.
We now generalize Theorem 5.5 to almost linearly presented modules of
projective dimension one.
Theorem 5.6 Under the assumptions of Setting 5.4, set Y = [Y1 , . . . , Yd ] and
assume that n = d + e. Then, the defining ideal of R(E) is
J = ((Y · B(ϕ)) : (Y )m ) = (Y · B(ϕ)) + Id (Bm (ϕ)),
where m is the degree of the non-linear column of ϕ and Bm (ϕ) denotes an m-th
iterated Jacobian dual as in Definition 2.9. Moreover:
(i) R(E) is almost Cohen–Macaulay, and it is Cohen–Macaulay if and only if
m = 1.
(ii) F(E) is Cohen–Macaulay.
Proof We modify the proof of [38, 4.11]. Let a1 , . . . , an be a minimal generating
set for E corresponding to the presentation ϕ, and let R[T1 , . . . , Tn ] R(E)
be the natural epimorphism, mapping Ti to ai for all i. Localizing at the unique
homogeneous maximal ideal, we may assume that R is local and that E admits a
228 A. Costantini
generic Bourbaki ideal I , which is perfect of grade 2 and such that μ(I ) = n − e +
1 = d + 1. If e = 1, then E ∼ = I and the statement follows from Theorem 5.5.
So, assume
n that e 2. With xj as in Notation 2.1, for 1 j e − 1 set
Xj = i=1 Z ij T i , and note that Xj is mapped to xj under the epimorphism
R [T1 , . . . , Tn ] R(E ). Set T = [T1 , . . . , Tn ]. As in Remark 5.1, we can
construct a minimal almost linear presentation ψ of I , such that

0
[Y ] · B(ϕ) ≡ [T ] · modulo (X1 , . . . , Xe−1 ).
ψ

0
Let B(ψ) be a Jacobian dual of ψ defined by [T ] · = [Y ] · B(ψ). Then, by
ψ
Theorem 5.5, the defining ideal of R(I ) is
JI = (Y · B(ψ)) + Id (Bm (ψ)) = (Y · B(ψ)) : (Y )m , (5.1)
where m is the degree of the non-linear column of ϕ. Moreover, R(I ) is almost

Cohen–Macaulay, and Cohen–Macaulay if and only if the entries of ψ are all linear,
while F(I ) is Cohen–Macaulay. In particular, by Theorem 2.3 it follows that R(E)
is Cohen–Macaulay if and only if ϕ is linear.
To prove the remaining statements, notice that Eq is of linear type for all
primes q in the punctured spectrum of R (this is because E has projective
dimension one and satisfies Gd , by [1, Propositions 3 and 4]). Hence, also Iq
is of linear type for the same primes q. Moreover, the discussion above shows
that depth R(I ) dim R(I ) − 1 = d 2. Hence, inducting on e and using The-
orem 3.1 in the case when e = 2, we obtain that R(I ) ∼ = R(E )/(F ) and

x1 , . . . , xe−1 form a regular sequence on R(E ). Thus, X1 , . . . , Xe−1 form a
regular sequence modulo JR . This shows that R(E ) is almost Cohen–Macaulay,
whence R(E) is almost Cohen–Macaulay. It also implies that
JI = JR + (X1 , . . . , Xe−1 )
Hence, from (5.1) and Lemma 5.7 below it follows that
JR + (X1 , . . . , Xe−1 ) = (Y · B(ϕ)) + Id (Bm (ϕ)) + (X1 , . . . , Xe−1 ).
On the other hand, since E is of linear type locally on the punctured spectrum of
R, it follows that
J ⊇ (Y · B(ϕ)) : (Y )m ⊇ (Y · B(ϕ)) + Id (Bm (ϕ)),
where the last inclusion follows from Theorem 2.9(c). Therefore, since
X1 , . . . , Xe−1 form a regular sequence modulo JR , in R [T1 , . . . , Tn ] we have:
J= ((Y · B(ϕ)) + Id (Bm (ϕ)) + (X1 , . . . , Xe−1 )) ∩ J

= ((Y · B(ϕ)) + Id (Bm (ϕ))) + (X1 , . . . , Xe−1 ) J.
By the graded version of Nakayama’s Lemma, this means that
J = (Y · B(ϕ)) + Id (Bm (ϕ)) = (Y · B(ϕ)) : (Y )m ,
as claimed. Finally, since F(I ) is Cohen–Macaulay and depth R(I ) 2, from

Theorem 4.8(b) it follows that F(E) is Cohen–Macaulay.

Lemma 5.7 Let R = k[Y1 , . . . , Yd ](Y1 ,...,Yd ) , and denote Y = [Y1 , . . . , Yd ]. Let ϕ,
ψ, B(ψ), and X1 , . . . , Xe−1 be as in the proof of Theorem 5.6. Then, for all i and
for any Jacobian dual B(ϕ) of ϕ, in R [T1 , . . . , Tn ] we have
(Y · B(ϕ)) + Id (Bi (ϕ)) + (X1 , . . . , Xe−1 ) = (Y · B(ψ)) + Id (Bi (ψ)).

0
Proof Choose B(ψ) such that [Y ] · B(ψ) = [T ] · , as in the proof of
ψ
Theorem 5.6. Then, in R [T1 , . . . , Tn ] we have
[Y ] · B(ϕ) ≡ [Y ] · B(ψ) modulo (X1 , . . . , Xe−1 ).
So, the statement is proved for i = 1. Now, let i + 1 2 and assume that the
statement holds for Bi (ϕ). Let Ci be a matrix as in Definition 2.9. Since
(Y · Bi (ϕ)) + (Id (Bi (ϕ)) ∩ (Y )) = (Y · Bi (ϕ)) + (Y · Ci )
and the Bi (ϕ) are bigraded, going modulo (X1 , . . . , Xe−1 ), in R [T1 , . . . , Tn ] we
have
(Y · Bi (ψ)) + (Id (Bi (ψ)) ∩ (Y )) = (Y · Bi (ψ)) + (Y · Ci ),
where Ci denotes the image of Ci modulo (X1 , . . . , Xe−1 ). Now, let Bi+1 (ψ) =
[Bi (ψ)) | Ci ]. Then, in R [T1 , . . . , Tn ] we have that
(Y · B(ϕ)) + Id (Bi+1 (ϕ)) + (X1 , . . . , Xe−1 ) = (Y · B(ψ)) + Id (Bi+1 (ψ)),
as we aimed to show.

We remark that the equality J = (Y · B(ϕ)) : (Y1 , . . . , Yd )m for the defining
ideal of the Rees algebra of a module E as in Theorem 5.6 could be obtained without
using generic Bourbaki ideals, by modifying the proof of [22, 6.1(a)]. In fact, even if
[22, 6.1(a)] is stated for perfect ideals of height two, its proof only uses the structure
of the presentation matrix, and one would only need to adjust the ranks to prove
the statement for modules of projective dimension one. More generally, up to this
230 A. Costantini
minor adjustment in the proof, [22, 6.1(a)] shows that if E is a finite module over
k[Y1 , . . . , Yd ] minimally generated by homogeneous elements of the same degree,
d the defining ideal of R(E) is J = (Y · B(ϕ)) : (Y ) , where N = 1 +
then N
i=1 (i − 1) and the i are the degrees of the columns of ϕ.

Similarly, a good portion of the proof of Theorem 5.5 could be adjusted to
the case of modules of projective dimension one by modifying the ranks of the
presentation. However, we would not be able to generalize the whole statement of
Theorem 5.5 using this method. Indeed, the proof of the equality J = (Y · B(ϕ)) +
Id (Bi (ϕ)) in the case of perfect ideals of height two crucially makes use of the ideal
structure of the cokernel of the presentation matrix (see the proof of [2, 5.3]).
Acknowledgments Most of the work presented in this manuscript was part of the author’s Ph.D.
thesis [6]. The author wishes to thank her advisor Bernd Ulrich for his insightful comments on
some of the results here presented and for assigning Lemma 4.5 as a homework problem in one of
his courses. Also, part of the content of Sect. 4 was motivated by a question of Jonathan Montaño,
whom we thank very much for fruitful conversations on the topic. Finally, the author is grateful
to the anonymous referee for their useful comments, which improved the exposition of parts of
Sect. 4.
References
1. Avramov, L.: Complete intersections and symmetric algebras. J. Algebra 73, 248–263 (1981)
2. Boswell, J.A., Mukundan, V.: Rees algebras of almost linearly presented ideals. J. Algebra 460,
102–127 (2016)
3. Bruns, W., Herzog, J.: Cohen–Macaulay rings. Cambridge studies in advanced mathematics
39, Cambridge University Press (1993)
4. Corso, A., Ghezzi, L., Polini, C., Ulrich, B.: Cohen–Macaulayness of special fiber rings.
Comm. Algebra 31, 3713–3734 (2001)
5. Corso, A., Polini, C., Vasconcelos, W.: Multiplicity of the special fiber of blowups. Math. Proc.
Camb. Phil. Soc. 140, 207–219 (2006)
6. Costantini, A.: Rees algebras and fiber cones of modules. Purdue University Graduate School.
Thesis. (2019) https://doi.org/10.25394/PGS.9107936.v1
7. Costantini, A.: Residual intersections and modules with Cohen–Macaulay Rees algebras
(2020) https://arxiv.org/abs/1811.08402
8. Cox, D., Lin, K.-N., Sosa, G.: Multi-Rees algebras and toric dynamical systems. Proc. Amer.
Math. Soc. 147, 4605–4616 (2019)
9. D’Anna, M., Guerrieri, A., Heinzer, W.: Ideals having one-dimensional fiber cone. In D.
Anderson, I. Papick, Ideal Theoretic Methods in Commutative Algebra, Lecture Notes in Pure
and Appl. Math. 220,155–170, New York: Dekker (2001)
10. D’Cruz, C., Raghavan, K.N., Verma, J.K.: Cohen–Macaulay fiber cones. Commutative algebra,
algebraic geometry and computational methods (Hanoi, 1996), 233–246, Springer, Singapore
(1999)
11. Eisenbud, D., Huneke, C., Ulrich, B.: What is the Rees algebra of a module?, Proc. Amer.
Math. Soc. 131, 701–708 (2003)
12. Flenner, H., Manaresi, M.: A numerical characterization of reduction ideals. Math. Z. 238,
205–214 (2001)
13. Hochster, M.: Properties of Noetherian rings stable under general grade reduction. Arch. Math.
24, 393–396 (1973)
14. Huneke, C.: On the associated graded ring of an ideal. Illinois J. Math. 26, 121–137 (1982)
15. Ikeda, S., Trung, N.V.: When is the Rees algebra Cohen–Macaulay? Comm. Algebra 17, 2893–
2922 (1989)
16. Jayanthan, A.V., Verma, J.K.: Hilbert coefficients and depth of fiber cones. J. Pure Appl.
Algebra 201, 97–115 (2005)
17. Jayanthan, A.V., Verma, J.K.: Fiber cones of ideals with almost minimal multiplicity. Nagoya
Math. J. 177, 155–179 (2005)
18. Johnson, M.: Second analytic deviation one ideals and their Rees algebras. J. Pure Appl.
Algebra 119, 171–183 (1997)
19. Johnston, B., Katz, D.: Castelnuovo regularity and graded rings associated to an ideal. Proc.
Amer. Math. Soc. 123, 727–734 (1995)
20. Kirby, D., Rees, D.: Multiplicities in graded rings I: The general theory. Contemp. Math. 159,
209–267 (1994)
21. Kleiman, S., Thorup, A.: A geometric theory of the Buchsbaum-Rim multiplicity. J. Algebra
167, 168–231 (1994)
22. Kustin, A., Polini, C., Ulrich, B.: The equations defining blowup algebras of height three
Gorenstein ideals. Algebra Number Theory 11, 1489–1525 (2017)
23. Lin, K.-N., Polini, C.: Rees algebras of truncations of complete intersections, J. Algebra 410,
36–52 (2014)
24. Miranda-Neto, C.B.: On special fiber rings of modules. Canad. J. Math. 72, 225–242 (2020)
25. Montaño, J.: Artin-Nagata properties, minimal multiplicities, and depth of fiber cones. J.
Algebra 425, 423–449 (2015)
26. Morey, S.: Equations of blowups of ideals of codimension two and three. J. Pure Appl. Algebra
109, 197–211 (1996)
27. Morey, S., Ulrich, B.: Rees algebras of ideals of low codimension, Proc. Amer. Math. Soc. 124,
3653–3661 (1996)
28. Nagata, M.: Local rings. Interscience Tracts in Pure and Applied Mathematics, No. 13,
Interscience Publishers a division of John Wiley & Sons, New York-London (1962)
29. Northcott, D.G., Rees, D.: Reductions of ideals in local rings. Math. Proc. Camb. Phil. Soc. 50,
421–444 (1954)
30. Polini, C., Ulrich, B.: Necessary and sufficient conditions for the Cohen–Macaulayness of
blowup algebras. Compositio Math. 119, 185–207 (1999)
31. Rees, D.: Amao’s theorem and reduction criteria. J. London Math. Soc. 32, 404–410 (1985)
32. Rees, D.: Reduction of modules. Math. Proc. Cambridge Phil. Soc. 101, 431–449 (1987)
33. Rossi, M., Valla, G.: Hilbert functions of filtered modules. Lect. Notes Unione Mat. Ital. 9,
Springer-Verlag, Berlin; UMI, Bologna (2010)
34. Shah, K.: On the Cohen–Macaulayness of the fiber cone of an ideal. J. Algebra 143, 156–172
(1991)
35. Simis, A., Ulrich, B., Vasconcelos, W.: Jacobian dual fibrations. Amer. J. Math. 115, 47–75
(1993)
36. Simis, A., Ulrich, B., Vasconcelos, W.: Cohen–Macaulay Rees algebras and degrees of
polynomial relations, Math. Ann. 301, 421–444 (1995)
37. Simis, A., Ulrich, B., Vasconcelos, W.: Codimension, multiplicity and integral extensions.
Math. Proc. Camb. Phil. Soc. 130, 237–257 (2001)
38. Simis, A., Ulrich, B., Vasconcelos, W.: Rees algebras of modules, Proc. London Math. Soc.
87, 610–646 (2003)
39. Teissier, B.: Cycles évanescents, sections planes et conditions de Whitney. Astérisque 7-8,
285–362 (1973)
40. Teissier, B.: Résolution simultanée, I et II, Séminaire sur les singularités des surfaces 1976–77,
Lecture Notes in Mathematics 777, 71–146, Springer-Verlag, Berlin, (1980)
41. Ulrich, B., Validashti, J.: A criterion for integral dependence of modules. Math. Res. Lett. 14,
1041–1054 (2007)
42. Ulrich, B., Validashti, J.: Numerical criteria for integral dependence. Math. Proc. Camb. Phil.
Soc. 151, 95–102 (2011)
232 A. Costantini
43. Ulrich, B., Vasconcelos, W.: The equations of Rees Algebras of ideals with linear presentation.
Math. Z. 214, 79–92 (1993)
44. Vasconcelos, W.: On the equations of Rees algebras. J. Reine Angew. Math. 418, 189–218
(1991)
45. Viet, D. Q.: On the Cohen–Macaulayness of fiber cones. Proc. Amer. Math. Soc. 136, 4185–
4195 (2008)
46. Weaver, M.: Rees algebras of ideals and modules over hypersurface and complete intersection
rings. Purdue University Graduate School. Thesis, in preparation.
Principal Matrices of Numerical
Semigroups
Papri Dey and Hema Srinivasan
Keywords Semigroup Rings · Gluing · Structure
1 Introduction
Consider a sequence a = {a1 , . . . , an } of positive integers and the monoid a

generated by a. When the gcd(a1 , . . . , an ) = 1, a is called a numerical semigroup
which contains zero and all but finitely many positive integers. Let k be an arbitrary
field. Consider the k- algebra homomorphism φa : k[x1 , . . . , xn ] → k[t] given
by φa (xi ) = t ai . The image of this map φa is the semigroup ring k[a] which is
isomorphic to the coordinate ring k[a] = k[x1 , . . . , xn ]/Ia where ker φa = Ia .
The semigroup ring k[a] and the ideal Ia become homogeneous with the weighting
deg(xi ) = ai , 1 i n. The ideal Ia is also the toric ideal in k[x1 , . . . , xn ]
and is generated by a finite set of binomials in xi . Since the semigroup ring is an
integral domain of dimension one, the ideal Ia is a prime ideal of height n − 1.
The embedding dimension of the semigroup ring is n if and only if Ia ⊂ m2 ,
m = (x1 , . . . , xn ). A semigroup is called complete intersection or Gorenstein if the
corresponding semigroup ring is complete intersection or Gorenstein respectively.
Given a = {a1 , . . . , an } ⊂ N, foreach i, 1 i n, there exists a smallest
positive integer ri such that ri ai = j =i rij aj . The n × n matrix D(a) := (rij )
where rii := −ri is called a principal matrix associated to a.
Although the diagonal entries −ri are uniquely determined, rij are not always
unique. In that sense, there can be more than one principal matrix D(a) for a given
a. Since D(a)a T = 0, rank of D(a) n − 1. The sequence of positive integers
a can be recovered from a given principal matrix A of rank n − 1 by factoring out
the g.c.d of the entries of any nonzero column of the adjoint of A and changing
the signs to be positive. Each of the n rows of the principal matrix, corresponds to a
P. Dey · H. Srinivasan ()

Department of Mathematics, University of Missouri, Columbia, MO, USA
e-mail: pdbdn@missouri.edu; SrinivasanH@missouri.edu
234 P. Dey and H. Srinivasan
binomial αi , 1 i n, known as critical binomials [9] as they are part of a minimal

generating set of Ia . Indeed, the principal matrix will have maximal rank n − 1, if
the (αi , 1 i n, xj ) is m-primary. As such, the principal matrix of a numerical
semigroup contains some of the essence of the information about the semigroup.
This article grew out of our attempts to uncover information about the semigroup or
its structure from the principal matrix.
This leads to the notion of a pseudo principal matrix, which looks like a principal
matrix of a, except that the diagonal entries −ri are not the smallest possible. We
prove in Corollary 2.12 that an n × n pseudo principal matrix of size n must have
rank at least n2 . Similar concepts have been studied in [4] and a related notion of
generalized critical binomial matrix in [7].
We determine the structure of a pseudo principal matrix depending on its rank
and use it to characterize the numerical semigroups a from their principal matrices
of any size in Theorems 2.13 and 2.15. In particular, numerical semigroups of
embedding dimension 4 and 5 have been discussed in terms of their principal
matrices in detail in Sect. 3. For the embedding dimension n = 4 we prove
that the principal matrix D(a) has rank 2 if and only if a = {a1 , a2 , a3 , a4 } =
d{c1 , c2 } e{c3 , c4 } and D(a) is the concatenation of two 2 × 2 matrices D(c1 , c2 )
and D(c3 , c4 ). For the embedding dimension n = 5 we prove that the principal
matrix D(a) has rank 3 if and only if either a = {a1 , a2 , a3 , a4 , a5 } = d{c1 , c2 , c3 }
e{c4 , c5 } and D(a) has 3 + 2 block structure or a = {a1 , a2 , a3 , a4 , a5 } = a1
d{c2 , c3 } e{c4 , c5 } and D(a) has 1 + 2 + 2 block structure. On the other hand,
we show that when n = 4, if the rank of D(a) is 3, then there are exactly three
possibilities for A = D(a):(i) AT X = 0 has a solution in positive integers or (ii)
a = d{c1 , c2 } {b1 , b2 } with bi ∈ c1 , c2 or (iii). a = d{b1 , b2 , b3 } b4 is a simple-
split gluing and is either a complete intersection or an almost complete intersection.
Furthermore, we also give a sufficient condition for a pseudo principal matrix of
rank n − 1 to be a principal matrix, in Proposition 4.4. This is a direct generalization
to n of the 4 × 4 case in [1] and [4].
As a result of the general result in embedding dimension 4 one finds that there is
a finite number ν(a) of a set of positive integers minimally generating a semigroup
with a given principal matrix up to gluing. In the last Sect. 5, we list some examples
to illustrate many of the results.
2 Principal Matrix and Look Alikes
For I ⊂ N, |I | denotes the number of elements in I . We write [r] for the

set {1, 2, . . . , r} of first r positive integers and |x| for the absolute value of x.
Determinant of a matrix T is denoted by det T and the adjoint or adjugate of T
as Adj T . For any set I we write I − {a} is the subset of all elements in I except
a. When the context is clear we write a for the set {a1 , . . . , an } as well as for the
vector a ∈ Nn .
Principal Matrices of Numerical Semigroups 235
Let A be a matrix. Given I, J ⊂ N, with |I | = |J |, AIJ is the determinant

of the submatrix of A consisting of the rows in I and the columns in J . If I ⊂
{1, 2, · · · , n}, then I denotes the determinant of the submatrix of A consisting of
the rows and columns in I . Thus, I = AII . We denote the gcd of two integers a
and b by (a, b).
Definition 2.1 Let a = {a1 , . . .⎛, an } be a set of⎞positive integers minimally
−c1 a12 . . . a1n
⎜ .. .. .. ⎟
generating a semigroup a. A = ⎜ ⎝ . . . ⎟ is called a principal matrix
⎠
..
a a n1 . −c
n2 n
of a if Aa = 0 and ci is the smallest positive integer such that ci ai ∈ a − {ai }, 1
i n}. We denote this by A = D(a).

The relations ci ai = aij aj are also called critical relations and the
ci 3 j =i aij
associated binomials xi − j =i xj are called the critical binomials of a. Clearly
ci are uniquely determined by a and a minimally generates the semigroup a if and
only if ci 2 for all i.
The rows of a principal matrix are relatively prime. If n > 2, then the principal
matrix has rank at least 2. For, if the matrix has rank one, then all the rows are
multiples of the first row. But then, all the entries on the diagonal are negative and
the entries off the diagonal are non negative. This is impossible if n > 2.
Although the diagonal of a principal matrix is unique, the rest of the entries can
be different. In fact, it is possible for a to have two principal matrices of different
ranks.
⎛ ⎞ ⎛ ⎞
−3 2 0 0 0 −3 2 0 0 0
⎜ 3 −2 0 0 0 ⎟ ⎜ 3 −2 0 0 0 ⎟
⎜ ⎟ ⎜ ⎟
⎜ ⎟ ⎜ ⎟
Example 2.2 ⎜ 0 0 −3 2 0 ⎟ and ⎜ 2 0 −3 0 1 ⎟
⎜ ⎟ ⎜ ⎟
⎝ 0 0 3 −2 0 ⎠ ⎝ 0 0 3 −2 0 ⎠
5 0 1 0 −4 5 0 1 0 −4
are both principal matrices for {22, 33, 26, 39, 34} where one of them has rank 3
and another has rank 4.
In fact, any {a{2, 3} b{2, 3} {2c}} where 3b ∈ a, c will exhibit this phenomenon.
Nevertheless, the principal matrix does say something about the nature of the set a.
It’s not necessary that any matrix of the above form (which has negative diagonal
entries and positive off-diagonal entries) is a principal matrix associated with a
semigroup.
⎡ ⎤
−4 0 11
⎢1 −5 40⎥
Example 2.3 A = ⎢
⎣0
⎥ looks like a Principal matrix. Taking the gcd
4 −51⎦
3 1 −2
0
⎡ ⎤
7
⎢11⎥
of the first column of its adjoint, we see that A ⎢ ⎥
⎣12⎦ = 0. But then, A is not
16
principal because the second row should be [3, −3, 1, 0] and −3 > −5.
We will call such matrices pseudo principal.
Definition 2.4 An n×n integral matrix A = (aij ) is called pseudo principal matrix,
if aij 0, i = j, aii < 0, 1 i, j n and there exist positive integers a1 , . . . , an
⎡ ⎤
a1
⎢ ⎥
such that A ⎣ ... ⎦ = 0. We say A = P (a), to denote that A is a pseudo principal
an
matrix with Aa = 0 for a = {a1 , . . . an } ⊂ N.
For a pseudo
principal matrix to be principal, aii needs to be minimal with respect
to aii ai = j =i aij aj and a minimally generate the semigroup a. In an actual
principal matrix aii < −1, for all i. That is the reason we require aii < −1 in a
pseudo principal matrix. However, many of our results here will go through even
if aii = −1. It is true that if A is a pseudo principal matrix none of whose entries
are zeros and whose product with 1T is zero, then AT is a pseudo principal matrix
(see [7] for details). We will prove in Sect. 4, Theorem 4.4 a condition for a pseudo
principal matrix to be principal which allows some zero entries. It is not a serious
restriction that the matrix is integral, for if it is over rational numbers, we can clear
the denominator and then the matrix will satisfy the same conditions required to be
a pseudo principal matrix. Although we begin this study looking for the structures
of principal matrices, many of the results seem to be good for the general pseudo
principal matrices.
⎤ a set {a1 , . . . , an } of relatively prime positive integers, we denote by a =
⎡ Given
a1
⎢ .. ⎥
⎣ . ⎦. We denote the Principal matrix of a by D(a).
an
Principal matrices of numerical semigroups of embedding dimension 3 are
completely characterized by the results of Herzog [6]. Principal matrices of aT =
⎡ ⎤
−c1 c2 0
[a1 , a2 , a3 ] are either D(a) = ⎣ c1 −c2 0 ⎦ and a is a complete intersection
a31 a32 −c3
or it is an almost complete intersection such that the principal matrix with all the
2 × 2 minors are not zero. In the former case, {a1 , a2 } a3 = d{c2 , c1 } a3 and
a3 = a31 c1 + a32 c2 .
We will generalize this in theorems 2.13, 2.15 and 4.4.

As another generalization of this result, we have the notion of gluing.
Definition 2.5 Let b = {b1 , . . . , bs }, c = {c1 , . . . , cn−s } minimally generate b
and c respectively. Then we say a is a gluing of b and c if there are relatively
prime positive integers d and e such that a = d{b} e{c}, d ∈ c, d ∈ / c, e ∈
b, e ∈
/ b.
When both b and c have more than one element, the concatenation of the principal
matrices D(b) and D(c) may not always be principal matrix of D(a). It is indeed
a pseudo principal matrix of a. When one of them, say c has exactly one element,
it is called a simple split and the principal matrix can reflect this, as can be seen
from Theorem 2.15. The concept of gluing for numerical semigroups has been
introduced in [8] and studied for instance in [5]. A numerical semigroup is a
complete intersection if and only if it is the gluing of two complete intersection
numerical semigroups by the result of Delorme [2]. An n × n principal matrix of a
glued semigroup can have maximal rank n − 1, see 5 of Example 5.1.
Lemma 2.6 Let A = (aij ) be any m × n. Let ψ(t) = ti=1 [i] . Suppose [i] = 0,
for all i < r.
⎡ Then there exists a series of row operations ⎤ on A to arrive at
a11 a12 ... a1q ...
⎢ 0 A12 . . .⎥
⎢ 12 ... 1q ⎥
⎢ . . .. ⎥
⎢ . . .. .. ⎥
⎢ . . . . . ⎥
⎢ ⎥
⎢ 0 0
⎢ . . . [r] . . . A[r][r−1],q . . .⎥
⎥
Â(r) = ⎢ . .
.. ⎥
. . . . If A is a matrix of integers, then
⎢ .. .. .. .. ⎥
⎢ ⎥
⎢ 0 0 . . . A[r−1],i . . . A[r−1],i . . .⎥
⎢ [r] [r−1],q ⎥
⎢ . . .. ⎥
⎢ . . .. .. ⎥
⎣ . . . . . ⎦
[r−1],n [r−1],n
0 0 . . . A[r] . . . A[r−1],q . . .
so is Â(r).
In particular, if A is a pseudo principal matrix, with [i] = 0, for
all i < r, then there is an invertible
⎡ ⎤ matrix M with MA = Â(r) =
−c1 a12 ... a1q ...
⎢ 0 A12 . . .⎥
⎢ 12 ... 1q ⎥
⎢ . .. ⎥
⎢ . .. .. .. ⎥
⎢ . . . . . ⎥
⎢ [r] ⎥
⎢ 0 0 . . . [r] . . . A[r−1],q . . .⎥
⎢ ⎥
⎢ .. .. .. .. .. ⎥.
⎢ . . . . . ⎥
⎢ ⎥
⎢ 0 0 . . . A[r−1],i . . . A[r−1],i . . .⎥
⎢ [r] [r−1],q ⎥
⎢ . .. ⎥
⎢ . .. .. .. ⎥
⎣ . . . . . ⎦
[r−1],n [r−1],n
0 0 . . . A[r] . . . A[r−1],q ...
Further, if I = {1, . . . , s} = [s], then
I Â(r) = [s] Â(r) = si=1 [i] (A) for s r and [s] Â(r) =
α(s)[s] , α(s) = 0,
Proof We proceed with row operations as in Gaussian elimination, except we will
not do any row interchanges. Indeed, we will not need to do so, thanks to the
hypothesis [i] = 0, i < r. For i 2, by adding −ai1 /a11 times the first row
to the ith⎡ row and then multiplying ⎤ the ith row by a11 , we arrive at the matrix
a11 a12 . . . a1q . . .
⎢ 0 12 . . . A12 . . .⎥
⎢ 1q ⎥
⎢ . . . .. .. ⎥
⎢
Â (1) = ⎢ . . .
. . . ⎥
. . ⎥. Suppose we have obtained by induction,
⎢ ⎥
⎣ 0 A1i 1i
12 . . . A1q . . . ⎦
0 A1n
12 . . . A1n . . .
1q
⎡ ⎤
a11 a12 ... a1q ...
⎢ 0 12 ... A12 . . .⎥
⎢ 1q ⎥
⎢ [p] ⎥
⎢ 0 0 [p] . . . A[p−1],q . . .⎥
⎢ ⎥
⎢ .. .. .. .. .. ⎥
⎢ . . . . . ⎥
⎢ ⎥
⎢ 0 0 . . . [r−1] . . . A[r−1] . . .⎥
Â (r − 1) = ⎢ [r−2],q ⎥
⎢ . . .. .. .. ⎥
⎢ .. .. . ⎥
⎢ . . ⎥
⎢ [r−2],i [r−2],i ⎥
⎢ 0 0 . . . A[r−1] . . . A[r−2],q . . .⎥
⎢ ⎥
⎢ .. .. .. .. .. ⎥
⎣ . . . . . ⎦
[r−2],n [r−2],n
0 0 . . . A[r−1] . . . A[r−2],q ...
Now, since [r−1] =
0, we can continue the row operations to make the entries
in the r − 1st column below.
We use the following relations, which can be verified by direct computation of
determinants and say, Plucker relations,
[r−1] [r−2],k [r−2],k [r−2],r−1 [r−2] [r−2],r,k

A[r−1] A[r−2],q − A[r−1] A[r−2],q = A[r−2] A[r−2],r,q , k, q r − 1
We add a proof in the appendix.

Then we get the matrix,
⎡ ⎤
a11 a12 ... a1q ...
⎢ 0 12 ... A12 . . .⎥
⎢ 1q ⎥
⎢ . . .. .. .. ⎥
⎢ .. .. . ⎥
⎢ . . ⎥
⎢ [r−1] ⎥
⎢ 0 0 . . . [r−1] . . . A[r−2],q . . .⎥
⎢ ⎥
⎢
Â(r) = ⎢ 0 0 ...0 [r] . . . A[r]
[r−1],q . . .⎥⎥
⎢ . . . .. .. ⎥
⎢ .. .. .. . ⎥
⎢ . ⎥
⎢ [r−1],i [r−1],i ⎥
⎢ 0 0 . . . 0 A[r] . . . A[r−1],q . . .⎥
⎢ ⎥
⎢ .. .. .. .. .. ⎥
⎣ . . . . . ⎦
[r−1],n [r−1],n
0 0 . . . 0 A[r] . . . A[r−1],q ...
Finally, in the row operations, we did not permute the rows at all and every
time we multiplied the sth row onwards by [s−1] and divided by [s−2] . Hence
comparing the principal minors, we get
[s] Â(r) = si=1 [i] , s r
and
[s] Â(r) = α(s)[s] , α(s) = 0

Theorem 2.7 Let A = P (a) be a pseudo principal matrix. Then,

I,p
1. AI,q 0, p = q if and only if |I | is even.
2. If I = 0, then I,j = 0 for all j .
3. I 0 if and only if |I | is even.
Proof The proof is by induction on |I |. In what follows, we will take the number 0
to be both positive and negative.
Without loss of generality, we can assume I = {1, . . . , t} for some t = |I |.
I,j −cI aI k
Suppose |I | = 1. Then I = −cI < 0 and AI,k = det 0.
aj I aj k
Let |I | = 2 so that I = {1, 2}. Since 1 = 0, so after a row operation, of adding
c1−1 times the first row to the kth row, and then multiply the rows by −c1 , we get to
⎡ ⎤
−c1 a12 a13 . . .
⎢ 12 ⎥
Â(2) = ⎣ 0 12 A13 . . .⎦
.. .. .. ..
. . . .
Since Â(2)a = 0 with a ∈ Z+ and A1,2
1,j 0, j 3, we conclude 12 0. So,
I 0, |I | = 2.
By induction, we assume
1. Sign AI,k = (−1)|I | , j = k ∈
I,j
/ I , |I | r − 2
2. If I = 0, then I,j = 0 for all j for |I | r − 1
3. Sign I = (−1)|I | , |I | r − 1
We will prove 1 for r − 1 and 2 and 3 for r.
Let |I | = r − 1. Again, we may assume I = {1, 2, . . . , r − 1}.
Consider AI,kI,j . Expanding along the rth column which is {ak1 , . . . , akr−1 , akj },
we get,

r−1
([r−1]−i),i
AI,k
I,j = akj I + aki (−1)r−i A([r−1]−i),j (−1)r−1−i
i=1

r−1
([r−1]−i),i
= akj I − aki A([r−1]−i),j
i=1
(I −i),i
Now, sign of akj I = (−1)r−1 and sign of −A(I −i),j = (−1)r−1 , j = i
Hence sign AI,k |I |
I,j = (−1) , j = k, as desired.
Next suppose [t] = 0 for some t r. Let s be the smallest number such that
[s] = 0. By lemma 2.6, Â(s)a = 0, for a ∈ (Z+ )n . We have already shown that
[s−1]j
for all j = k s, the sign of A[s−1]k = (−1)|I | . Since [s] = 0, we must have
[s−1]j
A[s−1]k = 0 and hence the entire sth row of Â(s) is zero. Since the principal minors
of A and Â(s) are non zero multiples of each other, we must have [t] = 0, t s.
Thus, we have shown that 2 holds for |I | = r.
It remains to prove 3, for |I | = r. Let I = {1, 2, . . . r}. If I = 0, we are done.
If it is not zero, then 1,2,...,r−1 is not zero. By lemma 2.6, A is row equivalent to
Â(r) whose r-th row is [0 . . . 0, I . . . A1,...r−1,r 1,...r−1,r
1,...r−1,j . . .]. Since the sign of A1,...r−1,j =
(−1)r−1 , as we proved above, we get the sign of I = (−1)r and the induction is
complete.

Remark 2.8 As a corollary to the proof of the above theorem, we see that
[r] = 0 -⇒ [t] = 0, t r
Lemma 2.9 Let A = P (a) be a pseudo principal matrix. Suppose 1,...r−1 = 0.

Then there exists a sequence of row operations on A to arrive at Â(r) whose (p, q)th
entry is
[p]
|A[p−1],q |, p r, q > p
−|[p] |, p = q r
0, q r − 1, q < p
[r−1],p
|A[r−1],q |, r p, q, p = q
−|[r − 1], p|, r p = q
In other words,
⎡ ⎤
−c1 a12 ... a1q ...
⎢ 0 −| | |A12 . . .⎥
⎢ 12 ... 1q | ⎥
⎢ . .. ⎥
⎢ . .. .. .. ⎥
⎢ . . . . . ⎥
⎢ ⎥
⎢ 0
⎢ 0 . . . − |[r] | . . . |A[r]
[r−1],q | . . .⎥
⎥
Â(r) = ⎢ .
⎢ ..
.
.. .
.. .
.. .. ⎥
.
⎥
⎢ ⎥
⎢ 0 . . . − |[r−1],i | . . . |A[r−1],q | . . .⎥
[r−1],i
⎢ 0 ⎥
⎢ . .. ⎥
⎢ . .. .. .. ⎥
⎣ . . . . . ⎦
[r−1],n
0 0 . . . − |[r−1],n | . . . |A[r−1],q | ...
Proof Since 1,...,r−1 = 0, we get from theorem 2.7 1,..t = 0, t r − 1.

1,2,...r−1,p
Hence by Lemma 2.6, we can get to an Â(r). Finally, by theorem 2.7, A1,2...r−1,q
1,2,...r−1,p [p] 1,2,...r−1,p
and A1,2...r−1,p have opposite signs and A1,2...r−1,q and A1,2...r−1,p have opposite
signs. So we arrive at Â(r) as desired.

Lemma 2.10 Let A = P (a) be a pseudo principal matrix. Suppose [r−1] = 0
and [r] = 0. Then arq = 0, q r + 1 and ari aiq = 0, q r + 1, for all i < r.
Proof We have Â(r) as in the lemma 2.9. Since Â(r)a = 0 and a has positive
entries, considering the rth row of Â(r), we get

n
−|[r] |ar + |A[r]
[r−1],q |aq = 0
q=r+1
Since [r] = 0,

n
|A[r]
[r−1],q |aq = 0
q=r+1
and hence
A[r]
[r−1],q = 0, q r + 1
But

r−1
A[r]
[r−1],q = arq [r−1] −
[r−1]−i,i
ari A[r−1]−i,q = 0.
i=1
Since all the terms in this sum have the same sign (−1)r−1 , each of the term is
zero. But [r−1] = 0. So, arq = 0, q r + 1.
[r−1]−i,i
Further, if ari = 0 for some i < r, then A[r−1]−i,q = 0.
Hence applying the above argument, with a reordering as, 1, 2, i − 1, i + 1, ..r −
1, i, we get aiq = 0, q r + 1.

Corollary 2.11 Let A be an n × n pseudo principal matrix of rank r. Then A has a
principal minor of size r that is not zero. If rank A n − 2, then there is a principal
minor of size r that is zero.
Proof Suppose t − 1 is the maximal size of a nonzero principal minor of A. After
reordering a, we can apply lemma 2.9 to get Â(t). In Â(t), all principal minors of
size t must vanish. Hence all the diagonal entries of Â(t), in rows t are zero. But
by theorem 2.7, this implies the entire rows after t − 1 are zeros and the rank of A
= rank Â(t) is t − 1. Thus, there is always a nonzero maximal minor of size rank A.
Now suppose rank A = r and all principal minors of size r are not zero. Then
we get Â(r) which has nonzero diagonal entries. But

T1 T2
Â(r) =
0n−r×r−1 T4
where T1 is an r − 1 × r − 1 upper triangular matrix with nonzero diagonal entries.

Hence we must have rank T4 = 1 and hence every row after t r is a multiple of
the r-th row. However, the entries on the diagonal I are of opposite signs to the rest
of the entries in the row and there is at least one nonzero entry above the diagonal
in each row. So, there can only be two nonzero rows in T4 . Hence r − 1 + 2 = n or
r = n − 1. Hence if rank A n − 2, there must be a principal minor of size r that
is zero.

Corollary 2.12 If A = P (a) or D(a), then rank A n2 .
Proof If rank A = n − 1, it is n2 . Let r = rank A n − 2. Then by corollary 2.11,
there is a nonzero principal minor of size r. Reorder a to make this the top left
minor. So, we have the matrix Â(r + 1). Since all the r + 1 minors are zero, we
1,...r,p
get A1,..r,q = 0. Let r + 1 p < q. Then this determinant is the sum of terms of
the same sign. Hence they must all be zero. So, apq 1,...r = 0. But 1,...r = 0. So,
apq = 0, r + 1 p < q. Hence

T1 T 2
A= ,
T3 T4
where T1 is r ×r and T4 is n−r ×n−r lower triangular matrix with negative entries
on the diagonal and hence is of rank n − r. So, rank A n − r. So, r n/2.

In the following theorem, we give the structure of a principal matrix of rank less
than n − 1.
Theorem 2.13 Let A = P (a) be an n × n pseudo principal matrix. Suppose the
rank A = n − p n − 2. Then there exists integers r1 , r2 , . . . rp 2 such that

ri m, and after a possible reordering a = r1 r2 . . . rp rp+1 , where
|ri | = ri , 1 i p and

D 0
A=
A1 A2
where D is the concatenation of P (ri ), for all 1 i p. We say such a principal

matrix D(a) and the semigroup < a > is of type r1 + r2 + . . . + rp + 1 + 1 + . . .,
where r1 + . . . + rp + 1 + 1 + . . . = n. If A = D(a) is principal, then D is the
concatenation of D(ri ), for all 1 i p.
Proof Let rank A = r. Since r n − 2, there is a principal minor of size r that is
zero. Suppose t is the smallest size of a principal minor that is zero. By reordering
ai if necessary, we may take [t] = 0 and [t−1] = 0. By the lemma 2.10, we get
atj = 0, j t + 1. Now, if atk = 0,for some k < t, then we get akj = 0, j t + 1.
By reordering if necessary, we can take atj = 0, j s − 1, atj = 0, j s. Now,
if aij = 0, for some i, s i t − 1, j s − 1, then we may relabel a to make
the ith row the tth row, by the minimality of t and get aj k = 0, k t + 1. So, there
exists an s, such that aij = 0, s i t, j s − 1, aij = 0, s i t, j t + 1.
By looking at the rows s, . . . , t and Aa = 0, we find that the t − s + 1 × t − s + 1
principal minor {s,...,t} = 0. Since t is minimal, s = 1. Let t = r1 . By reversing
the order, we conclude that the bottom right r1 × r1 submatrix of A is a pseudo
principal matrix of {an−r1 +1 , . . . , an } and a zero matrix of size r1 × n − r1 as the
bottom left. Then the first n − r1 + 1 rows of A form a matrix B which again B with
Ba = 0. We can repeat this with B. Thus, a = (a1 , . . . an−r1 ) (an−r1 +1 , . . . , an ).
Let r1 = (an−r1 +1 , . . . , an ).

B11 B12
A=
0r1 ×(n−r1 ) P (r1 )
Rank of P (r1 ) = r1 − 1. So, the n − r1 × n − r1 matrix B11 has rank n − r1 − 2.

Let t1 be the smallest size of a principal minor in B11 which is zero. Repeating the
argument above, and relabeling a, we arrive at
⎡ ⎤
B21 B22 B23
A = ⎣ 0 P (r2 ) 0 ⎦
0 0 P (r1 )
Continuing this, we get the result. In each iteration, we get an ri and with the rank
ri − 1 and we stop when the matrix Bi1 is invertible. Thus, the rank of A = n − p.
If A = D(a) is principal, then so are P (ri ) = D(ri ).

Remark 2.14 In fact, we proved the following: Let A = (aij ) be an r × n matrix
with aii < 0, aij 0, i = j with AaT = 0, a ∈ (Z+ )n . Suppose [r] = 0 and
I = 0, for any |I | < r. Then A = [Tr×r 0r×n−r ]
When a pseudo principal matrix has rank n − 1 and there is a vanishing principal
minor of size n − 1, we have a similar structure.
Theorem 2.15 Suppose A = P (a) is an n × n pseudo principal matrix of rank
n − 1 and that there is a principal minor of size n − 1 that vanishes. Then after a
possible reordering of ai ’s, there is an s 1 such that the matrix

D 0n−s×s
A=
Cs×n−s Bs×s
where D is of rank n − s − 1 and is a pseudo principal matrix. If A is a principal

matrix, then D = D(a1 . . . , an−s+1 )
Proof We may take [n−1] = 0, [n−2] = 0. So, as in the first part of the proof
of theorem 2.13, we get that there is an s such that aij = 0, s i n − 1, j
s −1, aij = 0, s i n−1, j = n. After a reordering, A has the desired form.

Remark 2.16 In the section on examples, we give examples to show that it is
possible to have types 2 + 3 and 2 + 2 + 1 for 5 × 5 principal matrices with rank 3
and 2 + 4, 2 + 3 + 1, 3 + 3, 2 + 2 + 1 + 1 for 6 × 6 principal matrix of rank 4.
In the Sect. 4 we consider the case of a pseudo principal matrix of rank n − 1 none
of whose principal minors of size n − 1 vanish and give a sufficient condition when
it is principal.
3 Embedding Dimension 4 and 5
In this section, we consider the lower embedding dimensions. In the embedding

dimension 3, we have a complete classification of the numerical semigroups as
complete intersection or the ideal of 2 × 2 minors of a 2 × 3 matrix by [6]. So,
we know the principal matrix of (a1 , a2 , a3 ) in a suitable order is either A =
⎡ ⎤ ⎡ ⎤
−c1 c2 0 −c1 a12 a13
⎣ c1 −c2 0 ⎦ or A = ⎣ a21 −c2 a23 ⎦.
a31 a32 −c3 a31 a32 −c3
When the embedding dimension n = 4, we either have the rank as 2 or 3 and if
the embedding dimension is n = 5, the rank is ether 4 or 3. In either case if the rank
is not n − 1, it is n − 2.
Numerical semigroups of embedding dimension 4 have been analyzed in terms
of their principal matrices. Villarael [9] has characterized the almost complete inter-
section numerical semigroups of embedding dimension 4 by their principal matrices
and Bresinsky [1] gave the picture of the principal matrix for the Gorenstein non
complete intersections numerical semigroup in embedding dimension 4. In this
section, for the embedding dimension n = 4 we prove that the principal matrix
D(a) has rank n − 2 = 2 if and only if a = (a1 , a2 , a3 , a4 ) = d(c1 , c2 ) e(c3 , c4 )
and D(a) is the concatenation of two 2 × 2 matrices D(c1 , c2 ) and D(c3 , c4 ). In
particular, except for a finite number of cases, it must be a gluing of two complete
intersections and is therefore a complete intersection. Further, if the rank is 3, then
one of these possibilities must be true for A = D(a)
1. AT X = 0 has a solution in positive integers.
2. a = d{c1 , c2 } {b1 , b2 } with bi ∈< c1 , c2 >
3. a = d{b1 , b2 , b3 } b4 is a gluing of two numerical semigroups and a is either a
complete intersection or an almost complete intersection.
For the embedding dimension n = 5 we prove that the principal matrix D(a) has
rank n−2 = 3 if and only if either a = (a1 , a2 , a3 , a4 , a5 ) = d(c1 , c2 , c3 ) e(c4 , c5 )
and D(a) has 3 + 2 block structure or a = (a1 , a2 , a3 , a4 , a5 ) = a1 d(c2 , c3 )
e(c4 , c5 ) and D(a) has 1 + 2 + 2 block structure. Using theorem 2.15, one can also
analyze the case of maximal rank 4 as we did in embedding dimension 5.
Theorem 3.1 Suppose a = {a1 , a2 , a3 , a4 } are relatively prime positive integers
minimally generating a semigroup < a >. Suppose that the principal matrix A of
a has rank 2. Then after a permutation of the entries if necessary, a = d{c1 , c2 }
e{c3 , c4 } with (d, e) = (c1 , c2 ) = (c3 , c4 ) = 1 and the principal matrix has the
following block structure
⎡ ⎤
−c1 c2 0 0
⎢ c1 −c2 0 0 ⎥
A=⎢
⎣ 0
⎥
0 −c3 c4 ⎦
0 0 c3 −c4
⎡ ⎤
−c1 a12 a13 a14
⎢ a21 −c2 a23 a24 ⎥
Proof Let A = ⎢ ⎥
⎣ a31 a32 −c3 a34 ⎦. Since the rank of A is 2, by corollary 2.11,
a41 a42 a43 −c4
one of the principal minors of size 2 must vanish. By reordering the ai , we may
choose 12 = 0. By lemma 2.10, and the fact that it is a principal matrix, we get
{a1 , a2 } = d{c1 , c2 } where (c1 , c2 ) = 1.
⎡ ⎤
−c1 c2 0 0
⎢ c1 −c2 0 0 ⎥
A=⎢
⎣
⎥
∗ −c3 ∗ ⎦
∗ ∗ −c4
Now, since the rank of A is 2, we get that the last two rows must be multiples of
each other. Since −c3 < 0, a34 > 0, −c4 < 0, the fourth row must be a negative
multiple of the third row. But then a4i = −xa3i , i = 1, 2. So, a31 = a32 = a41 =
a42 = 0. That implies, c3 a3 = a34 a4 . So, a3 = c4 e, a4 = c3 e, where e = (a3 , a4 ).
Thus, {a1 , a2 , a3 , a4 } = d{c2 , , c1 } e{c4 , c3 }

Corollary 3.2 If A is a 4 × 4 pseudo principal matrix of rank 2, then after some
rearrangement of rows and columns, A looks like
⎡ ⎤
−c1 c2 0 0
⎢ ac1 −ac2 0 0 ⎥
A=⎢
⎣ 0
⎥
0 −c3 c4 ⎦
0 0 bc3 −bc4
Proof This is precisely what we proved in the above theorem. For we did not use
the fact A is a principal matrix for the concatenation argument. ( We only uses it to
get the decomposition of a as a disjoint union. )

Definition 3.3 We say a = {a1 , a2 . . . , an } is ordinarily decomposable if a = db

ec where (d, e) = 1 and (b) = (c) = 1 and the principal matrix D(a) is the
concatenation of the principal matrices D(b) and D(c).
An ordinary decomposition is gluing if d ∈< c > −c and e ∈< b > −b
What we just proved was if the principal matrix of a = (a1 , a2 , a3 , a4 ) has rank
2 then it is decomposable as 2 and 2. As a corollary to theorem 2.13, we have
Corollary 3.4 If a has no ordinary decomposition, then A has rank n − 1.
Proof It follows from the Theorem 2.13 as p 2.

The converse of Corollary 3.4 is not true. For an example consider the sequence
a = {6, 8, 10, 13}. It can be thought of as gluing of 2{3, 4, 5} 13{1}. But the rank
⎡ ⎤
−3 1 1 0
⎢ 1 −2 1 0 ⎥
of D(a) = ⎢ ⎥
⎣ 2 1 −2 0 ⎦ is 3. However, when n = 4, rank of D(a) < 3 implies
3 1 0 −2
it is a decomposition.
Theorem 3.5 Let C = d{a1 , a2 } e{b1 , b2 } be a decomposition with (d, e) = 1
with the principal matrix of C a concatenation of D(a1 , a2 ) and D(b1 , b2 ). Then
d > bk , and e > ak for k = 1, 2

Proof Consider da1 , eb1 , eb2 . In what follows, we may assume (e, a1 ) = 1 for
otherwise, we can simply divide by the common factor for this triad. If da1 is not
in the semigroup generated by b1 , b2 , then this triad is not a complete intersection.
This means, the b2 (eb1 ) − b1 (eb2 ) is not a principal relation. A contradiction to
our assumption on the principal matrix of C. So, da1 ∈< b1 , b2 >. Thus, da1 =
p q
pb1 + qb2 so that x1e − x3 x4 is a relation. But then, e a2 . If e = a2 , then since
x3 b2 − x4 b1 is a minimal relation, we must have d ∈ b1 , b2 . But then C is not
minimal. So, e > a2 . Similarly, e > a1 and d > bj for all j .

Theorem 3.6 Suppose a = {a1 , a2 , a3 , a4 , a5 } are relatively prime positive inte-
gers minimally generating a semigroup < a >. Suppose that the principal matrix
A of a has rank 3. Then after a permutation of the entries if necessary, a =
d{c1 , c2 , c3 } e{c4 , c5 } with (d, e) = (c1 , c2 , c3 ) = (c4 , c5 ) = 1 and the principal
matrix has the following block structures
⎡ ⎤
−c1 a12 a13 0 0
⎢a −c2 0 ⎥
⎢ 21 a23 0 ⎥
⎢ ⎥
A = ⎢ a31 a32 −c3 0 0 ⎥
⎢ ⎥
⎣ 0 0 0 c4 a45 ⎦
0 0 0 a54 −c5
or a = a1 d{c2 , c3 } e{c4 , c5 } with (c2 , c3 ) = (c4 , c5 ) = (a1 , d, e) = 1 and

⎡ ⎤
−c1 a12 a13 a14 a15
⎢ 0 −c2 0 ⎥
⎢ c3 0 ⎥
⎢ ⎥
A=⎢ 0 c2 −c3 0 0 ⎥
⎢ ⎥
⎣ 0 0 0 −c4 a45 ⎦
0 0 0 a54 −c5
⎡ ⎤
−c1 a12 a13 a14 a15
⎢ a −c a a a ⎥
⎢ 21 2 23 24 25 ⎥
⎢ ⎥
Proof Let principal matrix A = ⎢ a31 a32 −c3 a34 a35 ⎥. As the rank of A = 3,
⎢ ⎥
⎣ a41 a42 a43 −c4 a45 ⎦
a51 a52 a53 a54 −c5
so there is a principal minor of size 3 that is zero by Corollary 2.11. If 12 = 0,
then by Lemma 2.10, we get a1i = a2i = 0, i = 3, 4, 5. Since the rank of A is 3, we
must have 3,4,5 = 0 and 34 = 0. In either case, we get a ij k = 0, ij = 0.
Now, without loss of generality, let us reorder and assume that 12 = 0 and
123 = 0.
So, by lemma 2.10, we get a3j = 0, j = 4, 5. Further if a31 a32 = 0, then we get
⎡ ⎤
−c1 a12 a13 0 0
⎢a −c2 0 ⎥
⎢ 21 a23 0 ⎥
⎢ ⎥
A = ⎢ a31 a32 −c3 0 0 ⎥
⎢ ⎥
⎣ ∗ ∗ ∗ −c4 a45 ⎦
∗ ∗ ∗ a54 −c5
So, the fifth row is a linear combination of the first four rows. This means a54 =
−xc4 , −c5 = xa45 . So, 45 = 0. reordering and applying lemma 2.10, we get
a4i = a5i = 0, i = 1, 2, 3. Thus, if a31 a32 = 0,
⎡ ⎤
−c1 a12 a13 0 0
⎢a −c2 0 ⎥
⎢ 21 a23 0 ⎥
⎢ ⎥
A = ⎢ a31 a32 −c3 0 0 ⎥
⎢ ⎥
⎣ 0 0 0 −c4 a45 ⎦
0 0 0 a54 −c5
Now, if one of a31 or a32 = 0, say a31 = 0, then we get 23 = 0 and hence
we come to the case by reordering, where 145 = 0 and 14 = 0. If a53 a52 = 0,
we get back to the first case and have the block form above but as concatenation
of D(a1 , a4 , a5 ) and D(a3 , a2 ). If one of a52 or a53 is zero, then after reordering if
necessary, we arrive at
⎡ ⎤
−c1 a12 a13 a14 a15
⎢ 0 −c2 0 ⎥
⎢ c3 0 ⎥
⎢ ⎥
A=⎢ 0 c2 −c3 0 0 ⎥
⎢ ⎥
⎣ 0 0 0 −c4 a45 ⎦
0 0 0 a54 −c5
and (a1 , a2 , a3 , a4 , a5 ) = a1 d{c2 , c3 } e{c4 , c5 } where (c2 , c3 ) = (c4 , c5 ) =

(d, e, a1 ) = 1

4 Sufficient Condition for Principal Matrix
Let P be a pseudo principal matrix with P a = 0. If P has at least one vanishing

principal minor of size n − 1, then we saw structures for a and P in Theorems 2.13
and 2.15. In this section we consider the case when none of these principal minors
other than P vanish.
Lemma 4.1 Let A be a pseudo principal matrix of rank n − 1 such that Aa = 0.
None of the principal minors of size n − 1 are zero if and only if AT X = 0 has a
solution in positive integers.

Proof Suppose none of the principal minors of size n−1 are zero. Then the diagonal
entries of the Adj A are not zero. Since the columns of the adjoint are multiples of
a, we have Adj A = (ai dj )n×n . By Theorem 2.7, the sign of the diagonal entries
are all (−1)n−1 . So, we have di are all positive or all negative. In either
case, from
Adj AA = 0I , we get nj=1 ai dj aj k = 0 for all i and k. So, we get nj=1 |dj |aj k =
0 for all k or [|d1 |, . . . |dn |]A = 0, where |di | > 0. Conversely, since rank of AT
is n − 1, if AT X = 0 has a solution in positive integers, then that solution must be
a multiple of di ’s. So, di = 0, 1 i n which in turn implies none of the n − 1
minors vanish.

Using Lemma 4.1 and Theorem 6.3(c) in [7] we conclude that
Corollary 4.2 If A is a pseudo principal matrix of rank n − 1 such that AT X = 0
has solutions in positive integers, then AT is a pseudo principal matrix.

For a Pseudo principal matrix A of rank n − 1, there is a b ∈ Nn such that
Ab = 0. So there is a unique a ∈ N with the entries in a are relatively prime such
that Aa = 0. We denote this by D −1 A = a.
⎛ ⎞
−4 1 1 1
⎜ 2 −4 1 1 ⎟
Example 4.3 Matrix A = ⎜ ⎟
⎝ 1 1 −3 1 ⎠ is a pseudo principal matrix.
1 2 1 −3
D −1 A = {20, 24, 25, 31}. And A is indeed the principal matrix of {20, 24, 25, 31}.
Thus, in this case, D(D −1 (A)) = A. AT is also pseudo principal but D −1 AT =

{1, 1, 1, 1} and clearly D(D −1 (AT )) = AT .

⎡ ⎤
−c1 a12 . . . a1n
⎢ a21 −c2 . . . a2n ⎥
⎢ ⎥
Theorem 4.4 Let n 3 and A = ⎢ . .. .. ⎥ be a pseudo princial matrix
⎣ .. . ... . ⎦
an1 an2 . . . −cn
of rank n − 1 with ci 2. Suppose, all the columns add upto zero and the first
column of the adjoint is relatively prime and none of the columns in A have more
than one zero. Then A is the principal matrix of D −1 A, i.e., D(D −1 A) = A and the
associated numerical semigroup has embedding dimension n.

⎡ ⎤
a1
⎢ .. ⎥
Proof By hypothesis there exist a = ⎣ . ⎦ ∈ Nn such that Aa = 0. As the rank
an
of A is n − 1, we can take a = D −1 A and hence (a1 , . . . , an ) = 1. So, there exist
integers λ1 , . . . , λn such that λ1 a1 + · · · + λn an = 1. Further, by assumption, the
column sums of A are zero and hence Adj A = (−1)n−1 [a, . . . , a].
In order to show that the matrix A is the principal matrix D(a) it is sufficient to
show that the relation in each row of A is a principal relation. We show this for the
first row and the same proof works for the rest of the rows.
n
Suppose that b11 a1 = i=2 b1j aj is a relation with ⎡ b11 2 and b1j
⎤
−b11 b12 . . . b1n
⎢ a21 −c2 . . . a2n ⎥
⎢ ⎥
0, j = 2, . . . , n. We will show that b11 c1 . Let B = ⎢ . .. .. ⎥.
⎣ .. . ... . ⎦
an1 an2 . . . −cn
So, Ba = 0 and hence det(B) = 0. Therefore, there ⎡ exist x i , i = 1, .⎤. . , n such that
−b11 b12 . . . b1n
⎢ .. .. . ⎥
: ; ⎢ . . . . . .. ⎥
x1 = 1 x2 . . . xn B = 0. For, j 2, let Bj = ⎢ ⎥
⎢ λ1 λ2 . . . λn ⎥ by replacing
⎣ ⎦
.. .. ..
. . ... .
the j -th row of B by λi ’s.
⎛ ⎞
0
⎜ .. ⎟
⎜.⎟
⎜ ⎟ : ;T
Since Bj a = ⎜ ⎟
⎜1⎟, the j th column of the adjoint of Bj is det Bj a1 a2 . . . an .
⎜.⎟
⎝ .. ⎠
0
: ;
Using x1 x2 . . . xn B = 0 one gets
" " " "

"−c . . . . . . a2n "" "−c . . . . . . a2n ""
" 2 " 2
" . " " . . ""
" . . "
. " " . .
" . ... ... . " " . . . . . . . . ""
" "
" " " "
det Bj a1 = (−1)j +1+j −2 det " b12. . . . . . b1n " = xj det " aj 2 . . . . . . aj n " = −(−1)
n−1 x a
j 1
" " " "
" . .. "" " . .. ""
" . " .
" . ... ... . " " . ... ... . "
" " " "
"a " "a . . . . . . −c "
n2 . . . . . . −cn n2 n
So, xj = (−1)n−1 det Bj is an integer.

{1,...,j −1,j +1,...,n}
By Cramer’s rule, a1 det Bj = (−1)j +1 B{2,...,n} =(−1)j +1 (−1)
n−2+n−j B {2...n}\j,1 .
{2...n}\j,j
{2...n}\j,1
By theorem 2.7, sign of B{2...n}\j,j is n−2. We get, the sign of det Bj = (−1)n−1 .
Hence, xj 0.
Since [x1 , . . . xn ]B = 0, n 3, at least three of the xj ’s are non zero. We
know, x1 = 1. Say, wlog, x2 , x3 = 0. If xj = 0 for some j 4, then 0 =
b1j + i=j,i2 aij xi . But then a2j = a3j = 0, which is impossible. So, xj = 0 for
any j and are allpositive integers.

Now, b11 = i2 ai1 xi i2 ai1 = c1 .
Thus, c1 is the smallest possible entry in the diagonal. Similarly, all of the
diagonal entries are the smallest possible and hence A is a principal matrix.

Example 4.5 Both the conditions on the columns of A in the proposition, 4.4
⎛ ⎞
−5 1 1 4
⎜ 4 −3 0 0 ⎟
are necessary. For instance, ⎜ ⎟
⎝ 1 1 −2 0 ⎠ is a pseudo principal matrix, of
0 1 1 −4
(24, 32, 28, 15). It does have all the columns summing up to zero, but has two zeros
in the last column. It is not principal. For, the first entry −5 can be −4. Indeed,
⎡ ⎤
−4 0 1 1
⎢ 1 −5 4 0 ⎥
4(24) = 3(32). In the Example 2.3, A = ⎢ ⎥
⎣ 0 4 −5 1 ⎦, we have all the
3 1 0 −2
columns sum up to zero, no column with two zeros but then the n − 1 = 3 minors
of any of the three rows are not relatively prime.

5 Examples
We would like to acknowledge that we have used the MonomialAlgebras package

[3] to compute the binomial ideal Ia associated with a. We have written codes in
Macaulay2 to compute a principal matrix D(a) to come up with various examples
in this section.
Example 5.1 Embedding Dimension n = 4:

(1) Consider the semigroup a = 17{4, 7} 13{3, 8}. A principal matrix is D(a) =
⎡ ⎤
−5 1 3 1
⎢ 2 −3 3 1 ⎥
⎢ ⎥
⎣ 0 0 −8 3 ⎦ with rank 3. This has the minimal number of generators 4,
0 0 8 −3
thus it is an almost complete intersection, not a gluing of the two semigroups
< 4, 7 > and < 3, 8 > and hence is not symmetric.
(2) Consider the semigroup a = 10{4, 7} 13{3, 8}. A principal matrix is
⎡ ⎤
−7 4 0 0
⎢ 7 −4 0 0 ⎥
D(a) = ⎢ ⎥
⎣ 0 0 −8 3 ⎦ with rank 2. Recall by Theorem 3.5, we must
0 0 8 −3
have d 9, e 8. The list of all possible values of d, e such that
the corresponding semigroup of the form d{4, 7} e{3, 8} is not a gluing
is d = {10, 13} and e = {9, 10, 13, 17}. The values of d, e such that
the corresponding semigroup d{4, 7} e{3, 8} (not guling) has the same
principal matrix of rank 2 like D(a) mentioned above are as follows. (d, e) =
{(10, 9), (10, 13), (10, 17), (13, 9), (13, 17)}.
(3) Consider the semigroups of the form d{5, 8} e{3, 7}. If d = 11, e = 17,
then the semigroup a = 11{5, 8} 17{3, 7} has a principal matrix D(a) =
⎡ ⎤
−8 5 0 0
⎢ 8 −5 0 0 ⎥
⎢ ⎥
⎣ 0 0 −7 3 ⎦ with rank 2. It is not a gluing, the minimal number of
0 0 7 −3
generators is 8. Recall by Theorem 3.5, we must have d 8, e 9. The list of
all possible values of d, e such that the corresponding semigroup is not a gluing
is d = {8, 11} and e = {9, 11, 12, 14, 17, 19, 22, 27}. The values of d, e such
that the corresponding semigroup d{5, 8} e{3, 7} (not guling) has the same
principal matrix of rank 2 like D(a) mentioned above are as follows. (d, e) =
{(8, 17), (8, 19), (8, 27), (11, 14), (11, 17), (11, 19), (11, 27)}. An example of
a gluing semigroup with same principal matrix is 13{5, 8} 18{3, 7}. The
⎡ ⎤
−2 0 2 1
⎢ 0 −5 1 7 ⎥
semigroup 13{5, 8} 10{3, 7} has a principal matrix D(a) = ⎢ ⎣ 0 0 −7 3 ⎦
⎥
0 0 7 −3
with rank 3. It is a gluing and hence a complete intersection.
(4) Consider the semigroup a = 11{3, 4} 13{3, 4}. A principal matrix is D(a) =
⎡ ⎤
−4 3 0 0
⎢ 4 −3 0 0 ⎥
⎢ ⎥
⎣ 0 0 −4 3 ⎦ with rank 2, symmetric, 3 generators, thus a complete
0 0 4 −3
intersection.
(5) Consider the semigroup a = 7{2, 3} 6{3, 4}. This is a gluing of two
complete intersections and is therefore a complete intersection and the principal
⎡ ⎤
−3 2 0 0
⎢ 3 −2 0 0 ⎥
matrix has two possibilities. Either D(a) = ⎢ ⎥
⎣ 0 0 −4 3 ⎦ with rank 2 or
0 0 4 −3
⎡ ⎤
−3 0 1 1
⎢ 0 −2 1 1 ⎥
D(a) = ⎢ ⎥
⎣ 0 0 −4 3 ⎦ with rank 3. Since it is symmetric and a complete
0 0 4 −3
intersection, the principal matrix does not have the standard form for the
symmetric non complete intersection semigroups.

Example 5.2 Embedding Dimension n = 5
(1) Consider
⎡ ⎤ a = 11{4, 7} 13{5, 7, 9}. A principal matrix is D(a) =
the semigroup
−7 4 0 0 0
⎢ 7 −4 0 0 0 ⎥
⎢ ⎥
⎢ ⎥
⎢ 0 0 −5 1 2 ⎥ with rank 3, and the minimal number of generators is 12.
⎢ ⎥
⎣ 0 0 1 −2 1 ⎦
0 0 4 1 −3
(2) Consider the⎡semigroup a = 15{2, ⎤ 3} 14{2, 5} {43}. A principal matrix
−3 2 0 0 0
⎢ 3 −2 0 0 0 ⎥
⎢ ⎥
⎢ ⎥
is D(a) = ⎢ 0 0 −5 2 0 ⎥ with rank 3, and the minimal number of
⎢ ⎥
⎣ 0 0 5 −2 0 ⎦
1 0 2 0 −2
generators is 7.
(3) Consider the semigroup
⎡ a = 14{2, ⎤ 3} 12{3, 4} {43}. A principal matrix
−3 2 0 0 0
⎢ 3 −2 0 0 0 ⎥
⎢ ⎥
⎢ ⎥
is D(a) = ⎢ 0 0 −4 3 0 ⎥ with rank 3 and the principal matrix
⎢ ⎥
⎣ 0 0 4 −3 0 ⎦
1 0 0 3 −4
has
⎡ the expected ⎤ for rank 3. Another principal matrix is D(a) =
form
−3 0 1 1 0
⎢ 0 −2 1 1 0 ⎥
⎢ ⎥
⎢ ⎥
⎢ 0 0 −4 3 0 ⎥ with rank 4.
⎢ ⎥
⎣ 0 0 4 −3 0 ⎦
1 0 0 3 −4
This is an example where principal matrix is not only not unique but the rank
also varies.

Example 5.3 Embedding Dimension n = 6:

(1) Case 2 + 2 + 2: Consider the ⎡ semigroup a = 11{2, 3}⎤ 13{2, 3} 17{2, 3}. A
−3 2 0 0 0 0
⎢ 3 −2 0 0 0 0 ⎥
⎢ ⎥
⎢ ⎥
⎢ 0 0 −3 2 0 0 ⎥
principal matrix is D(a) = ⎢ ⎥ with rank 3. It also has
⎢ 0 0 3 −2 0 0 ⎥
⎢ ⎥
⎣ 0 0 0 0 −3 2 ⎦
0 0 0 0 3 −2
⎡ ⎤
−3 2 0 0 0 0
⎢ 3 −2 0 0 0 0 ⎥
⎢ ⎥
⎢ ⎥
⎢ 2 0 −3 0 1 0 ⎥
another principal matrix D(a) = ⎢ ⎥ with rank 4. The
⎢ 2 0 0 −2 1 0 ⎥
⎢ ⎥
⎣ 0 0 0 0 −3 2 ⎦
0 0 0 0 3 −2
minimal number of generators 10.
(2) Case 2 + 1 + 1 + 2: Consider⎡ the semigroup a = {10, ⎤ 15, 14, 21, 18, 27}.
−3 2 0 0 0 0
⎢ 3 −2 0 0 0 0 ⎥
⎢ ⎥
⎢ ⎥
⎢ 1 0 −2 0 1 0 ⎥
A principal matrix is D(a) = ⎢ ⎥ with rank 4 and the
⎢ 0 1 0 −2 0 1 ⎥
⎢ ⎥
⎣ 0 0 0 0 −3 2 ⎦
0 0 0 0 3 −2
minimal number of generators 12. It is easy to generate such examples. If we
simply find three suitable larger) numbers a, b, c, with 3a ∈< b, c >, then
a = b{2, 3} a{2, 3} c{2, 3} should have two such principal matrices.
(3) Case 4 + 2: Consider the semigroup a = 7{5, 6, 7, 9} 9{4, 6}. A principal
matrix is D(a) =
⎡ ⎤
−3 1 0 1 0 0
⎢1 −2 1 0 0 0⎥
⎢ ⎥
⎢1 0 −2 1 0 0⎥
⎢ ⎥ with rank 5. Another principal matrix is D(a) =
⎢0 0 0 −2 2 1⎥
⎢ ⎥
⎣0 0 0 0 −3 2⎦
0 0 0 0 3 −2
⎡ ⎤
−3 1 0 1 0 0
⎢1 −2 1 0 0 0⎥
⎢ ⎥
⎢1 0 −2 1 0 0⎥
⎢ ⎥ with rank 4. Note that this decomposition does not
⎢1 1 1 −2 0 0⎥
⎢ ⎥
⎣0 0 0 0 −3 2⎦
0 0 0 0 3 −2
guarantee for the existence of a principal matrix of rank 4. For example,
consider the semigroup a = 7{5, 6, 7, 9} 9{5, 6}. A principal matrix
⎡ ⎤
−3 1 0 1 0 0
⎢ 1 −2 1 0 0 0 ⎥
⎢ ⎥
⎢ ⎥
⎢ 1 0 −2 1 0 0 ⎥
is D(a) = ⎢ ⎥. This semigroup does not have any
⎢ 1 1 1 −2 0 0 ⎥
⎢ ⎥
⎣ 0 0 0 2 −4 1 ⎦
0 0 0 1 1 −2
principal matrix with rank 4.
(4) Case 2 + 3 + 1: Consider the ⎡ semigroup a = 13{2,
⎤ 3} 7{5, 6, 7} 43. A
−3 1 0 1 0 0
⎢ 1 −2 1 0 0 0⎥
⎢ ⎥
⎢ 1 0 −2 1 0 0⎥
principal matrix is D(a) = ⎢ ⎥ with rank 5 and another
⎢ 1 1 1 −2 0 0⎥
⎣ 0 0 0 0 −3 2 ⎦
0 0 0 0 3 −2
⎡ ⎤
−3 2 0 0 0 0
⎢ 3 −2 0 0 0 0⎥
⎢ ⎥
⎢ 0 0 −4 1 2 0⎥
principal matrix is D(a) = ⎢
⎢
⎥ with rank 4.
⎢ 0 0 1 −2 1 0⎥ ⎥
⎣ 1 0 1 0 −3 2⎦
2 0 1 1 0 −3
(5) Case 3 + 3: Consider the semigroup a = 10{5, 6, 7} 11{5, 6, 7}. A principal
matrix is D(a) =
⎡ ⎤
−4 1 2 0 0 0
⎢ 1 −2 1 0 0 0⎥
⎢ ⎥
⎢ 2 0 −3 2 0 0⎥
⎢ ⎥ with rank 5 and another principal matrix is D(a) =
⎢ 0 0 0 −4 1 2⎥
⎢ ⎥
⎣ 0 0 0 1 −2 1⎦
0 0 0 3 1 −3
⎡ ⎤
−4 1 2 0 0 0
⎢1 −2 1 0 0 0⎥
⎢ ⎥
⎢3 1 −3 0 0 0⎥
⎢ ⎥ with rank 4.

⎢0 0 0 −4 1 2⎥
⎢ ⎥
⎣0 0 0 1 −2 1⎦
0 0 0 3 1 −3
6 Appendix
Let A = (aij ) be an m × n matrix. Then for all k, q
[r−1] [r−2],k [r−2],k [r−2],r−1 [r−2] [r−1],k

A[r−1] A[r−2],q − A[r−1] A[r−2],q = A[r−2] A[r−1],q
Proof Both sides are equal to zero if k < r−1 or q < r−1. So, we take k, q r−1.
Let r − 1 = j .
j −1

[j ] [j −1],k [j −1] [j −1]−i,j
A[j ] A[j −1],q = (A[j −1] ajj + (−1)i+j aij A[j −1] )
i=1
j −1

[j −1] [j −1]−i,k
(A[j −1] akq + (−1)i+j aiq A[j −1] )
i=1
j −1

[j −1],k [j −1],j [j −1] [j −1]−i,k
A[j ] A[j −1],q = (A[j −1] akj + (−1)i+j aij A[j −1] )
i=1
j −1

[j −1]
(A[j −1] aj q + (−1)i+j aiq A[j −1]−i,j
i=1
Thus,
[j ] [j −1],k [j −1],k [j −1],j
A[j ] A[j −1],q − A[j ] A[j −1],q =
[j −1] [j −1]
A[j −1] [(ajj akq − akj aj q )A[j −1] +
j −1 [j −1]−i,k
+ i=1 (−1)
i+j (a a
jj iq − aij aj q )A[j −1] +
j −1 [j −1]−i,j
i=1 (−1)
i+j (a a
kq ij − akj aiq )A[j −1] +
j −1 s+t a a (A[j −1]−s,j A[j −1]−t,k [j −1]−t,j [j −1]−s,k
s,t=1 (−1) sj tq [j −1] [j −1] − A[j −1] A[j −1] )
But,
[j −1]−s,j [j −1]−t,k [j −1]−t,j [j −1]−s,k [j −1] [j −1]−{s,t},j,k
A[j −1] A[j −1] − A[j −1] A[j −1] = A[j −1] A[j −1]
by Plucker relations.
And,
j −1
[j −1]−s,j [j −1]−t,k [j −1]−t,j [j −1]−s,k
(−1)s+t asj atq (A[j −1] A[j −1] − A[j −1] A[j −1] )
s,t=1
j −1
[j −1] [j −1]−{s,t},j,k
= (−1)s+t (asj atq − atj asq )A[j −1] A[j −1]
s>t=1
So,
[j ] [j −1],k [j −1],k [j −1],j
A[j ] A[j −1],q − A[j ] A[j −1],q =
[j −1] [j −1]
= A[j −1] [((ajj akq − akj aj q )A[j −1]
j −1 [j −1]−i,k j −1
+ i=1 (−1)
i+j (a a
jj iq − aij aj q )A[j −1] + i=1 (−1)
i+j (a a
kq ij − akj aiq )
[j −1]−i,j j −1 [j −1] [j −1]−{s,t},j,k

A[j −1] + s>t=1 (−1)
s+t (a a
sj tq − atj asq )A[j −1] A[j −1]
[j −1] [j ],k

= A[j −1] A[j ]q
References
1. H. Bresinsky, Symmetric semigroups of integers generated by 4 elements, Manuscr. Math. 17

(1975), 205–219.
2. C. Delorme, Sous-monoïdes d’intersection complète de N , Ann. Sci. École Norm. Sup. (4) 9
(1976), 145–154.
3. David Eisenbud and Janko Boehm and Max Nitsche, MonomialAlgebras: A Macaulay2
package. Version 2.3, A Macaulay2 package available at https://github.com/Macaulay2/M2/
tree/master/M2/Macaulay2/packages.
4. P. Gimenez and H. Srinivasan, A note on Gorenstein monomial curves, Bull. Braz. Math. Soc.
New Series 45 (2014), 671–678.
5. P. Gimenez and H. Srinivasan, The structure of the minimal free resolution of semigroup rings
obtained by gluing, J. Pure Appl. Alg. 223 (2019), 1411–1426.
6. J. Herzog, Generators and relations of abelian semigroups and semigroup rings, Manuscr.
Math. 3 (1970), 175–193.
7. O’Carroll, Liam and Planas-Vilanova, Francesc and Villarreal, Rafael H Degree and algebraic
properties of lattice and matrix ideals, SIAM Journal on Discrete Mathematics, 28,(2014),394–
427.
8. J. C. Rosales and P. A. García-Sánchez, Numerical Semigroups, Developments in Mathematics
20, Springer, 2009.
9. R. H. Villarreal, Monomial Algebras, Monographs and Textbooks in Pure and Applied
Mathematics 238, Marcel Dekker, 2001.
A Survey on the Koszul Homology
Algebra
Rachel N. Diethorn
Keywords Koszul homology · Complete intersections · Gorenstein rings ·

Golod rings · Koszul algebras · Betti numbers
1 Introduction
The properties of a local (or graded) ring R are often encoded in the Koszul
homology algebra H (R); and in certain cases, the key to understanding the ring
is to understand the algebra structure on its Koszul homology. Roughly speaking,
the simpler the algebra structure on H (R), the nicer the ring R. For example, a local
ring R with residue field k is regular if and only if H (R) = k; this follows from
the characterization of regular local rings given by Eilenberg and by Auslander and
Buchsbaum (see for example [30, Theorem 1.4.13] or [56, Lemma 5]). Complete
intersection rings, Gorenstein rings, and Golod rings are also characterized by their
Koszul homology algebras; we discuss these classical characterizations in Sect. 4.
Such characterizations have been quite useful in a wide range of applications in
commutative algebra, particularly so in the study of Golod rings, and very recently
in calculating Poincaré series over certain classes of rings; see for example [54] and
[41]. We highlight a few such applications at the end of Sect. 5.
The classical characterizations discussed in this survey have also played impor-
tant roles in the development of complete classifications of the Koszul homology
algebras of quotients of regular local rings by certain ideals. Such classifications
were given by Avramov, Kustin, and Miller in [10] and Weyman in [57] for ideals
of grade 3, by Kustin and Miller in [42] for Gorenstein ideals of grade 4, and by
Kustin in [40] for almost complete intersections of grade 4. We discuss these results
and a few of their applications in Sect. 5.
R. N. Diethorn ()
Yale University, Department of Mathematics, New Haven, CT, USA
e-mail: rachel.diethorn@yale.edu
258 R. N. Diethorn
While there are complete classifications of the algebra structures on Koszul

homologies in certain cases, for a general ring, the structure can be quite compli-
cated. This is evidenced by the paper [53], in which Roos determines completely
the Poincaré series (and the graded Betti numbers) of 104 rings with embedding
dimension 4 whose defining ideals are generated by quadratic forms, and examines
the different homological properties of their Koszul homology algebras. Even for
Koszul1 algebras, whose Koszul homology algebras are known to be restricted
in various ways, some of which we describe in Sect. 6, Roos shows that the
structure of H (R) can be surprisingly complicated. We highlight one such result
in Theorem 6.23 and one remarkable example in Example 6.11, and leave the rest
of [53] for the reader to explore.
Although Koszul algebras are not known to be characterized completely by the
algebra structures on their Koszul homologies, the strong connections between the
Koszul property of R and the Koszul homology algebra of R has provided an active
and vibrant area of research in recent years. The results of and questions posed by
Avramov, Conca, and Iyengar in their papers [7] and [8] were catalysts for much
of this progress, and thus are the focus of Sect. 6 of this survey. In particular we
discuss three questions about the Koszul homology algebras of Koszul algebras;
the first is a question of Avramov about the generation of H (R) as an algebra; the
second and third are questions of Avramov, Conca, and Iyengar about bounds on
the Betti numbers of R, and about subadditivity of syzygies of R, respectively. We
observe that the latter two questions have equivalent formulations in terms of Koszul
homology, and we outline the progress made towards answering these questions and
what remains unanswered.
A recurring theme throughout this survey is the use of the Koszul homology
algebra H (R) as a tool for understanding the ring R; of course this requires having
effective methods for studying Koszul homology. One such approach, which has
been used in the proofs of several important results that we state in this paper, is that
of minimal models, which is developed by Avramov in [5, Section 7.2]. Another
approach, which was first developed by Herzog in [32], is to find, explicitly, the
generators of Koszul homology modules. One of the key ingredients of Herzog’s
approach is an isomorphism that relates Koszul homologies Hi (R) to the minimal
resolution of R. We discuss this isomorphism and state Herzog’s result in Sect. 3,
and we refer back to this machinery often throughout the paper.
2 Preliminaries
In this section we recall the definition of the Koszul homology

algebra of a graded
ring (the definition in the local case is similar). Let R = i0 Ri be a standard
1 We use the word “Koszul” in two different ways throughout this survey: Koszul homology
algebras H (R), which we define in Sect. 2, and Koszul algebras R which we define in Sect. 6.
A Survey on the Koszul Homology Algebra 259
graded k-algebra with k a field, and fix a minimal set of generators x = x1 , . . . , xn

of R1 . In this survey we focus our attention on the Koszul complex K(x), which is
the complex with modules
Ci
Ki = (R n )
for 0 i n and Ki = 0 for all i > n, where we fix a basis dx1 , . . . , dxn of the
free module R n , and differentials

i
∂iK (dxj1 · · · dxji ) =
(−1)
+1 xj
dxj1 · · · dxj
· · · dxji ,

=1
where here we shorten K(x) and K(x)i to K and Ki , respectively. We ! note that
K0 = R, K1 = R n , and for 2 i n, Ki is a free R-module of rank ni with basis
{dxj1 · · · dxji | 1 j1 < · · · < ji n}.
Given our fixed minimal set of generators x of R1 , we define K(R) and Hi (R) to
be the Koszul complex K(x) and its i th homology Hi (K(x)), respectively; although,
it is worth noting that for any other minimal set of generators y of R1 , the resulting
Koszul homology modules Hi (K(y)) are isomorphic to Hi (K(x)) (see for example
[36, Remark 1.4]). Recall that Koszul homology satisfies depth sensitivity; that is,
the last nonvanishing Koszul homology module is Hn−g (R) where g = depth R.
For more background on the Koszul complex, see for example [47].
We denote by H (R), the sum of the Koszul homology modules
'
H (R) = Hi (R).
i0
The Koszul complex is a differential graded algebra, or a DG algebra; that is, it is

an algebra whose differential satisfies the Leibniz rule:
∂ K (ab) = ∂ K (a)b + (−1)|a| a∂ K (b),
where |a| denotes the homological degree of a. One can easily verify this fact
using the definition given above. The DG algebra structure on K(R) induces a k-
algebra structure on H (R). Indeed, given two cycles a and b in K(R), the equality
∂ K (ab) = 0 follows directly from the Leibniz rule; thus, ab is a cycle, and one can
define an algebra structure on H (R) by
[a] · [b] := [ab], (2.1)

260 R. N. Diethorn
where [ · ] denotes the homology class. From (2.1) it is easy to see that H (R) inherits
graded commutativity; that is, y · z = (−1)|y||z| z · y for all y and z in H (R), from
K(R). We call H (R) the Koszul homology algebra of R (even though it is defined
only up to non-canonical isomorphism of k-algebras).
3 Canonical Bases
Let k be a field and R a standard graded k-algebra. Then R ∼ = Q/I where Q =

k[x1 , . . . , xn ] is a standard graded polynomial ring with I a homogeneous ideal. The
Koszul homology algebra H (R) is closely connected to the minimal free resolution
F of R over Q via the isomorphsim
Hi (R) = Hi (R ⊗ K(Q)) = TorQ ∼

i (R, k) = Hi (F ⊗ k) = Fi ⊗ k. (3.1)
Q Q Q
As such, Koszul homology has been used classically as a tool for studying the
minimal free resolution of R and vice versa. We refer to this isomorphism often
throughout this survey.
In [32], Herzog uses the isomorphism above to explicitly describe canonical k-
bases of the Koszul homology modules, in the case that k is a field of characteristic
zero. The basis elements are described in terms of Jacobians that arise from the
differentials in the minimal graded free resolution F of R, as we see in the following
result.
To set notation for the theorem, for each i we let bi be the rank of Fi , and fix
bi−1 i i−1
a homogeneous basis e1i , . . . , ebi i of Fi such that ∂ F (eji ) =
=1 f
,j e
. For
z ∈ K(Q), we denote by z the image of z in R ⊗ K(Q). In the theorem, cj1 ,...,ji is
the inverse of a constant that tracks the degrees of the corresponding basis elements
of F .
Theorem 3.1 ([32, Corollary 2]) For each i = 1, . . . , n, a k-basis of Hi (R) is
given by [zj ] for j = 1, . . . , bi , where
8 9

b1
bi−1 ∂ fjii−1 ,j , fji−1
i−2 ,ji−1
, . . . , f 2 ,f1
j1 ,j2 1,j1
zj = ··· cj1 ,...,ji ! dx
1 . . . dx
i
L j1 =1 ji−1 =1
∂ x
1 , x
2 , . . . , x
i
and L = {1
1 < · · · <
i n}.

This result of Herzog was generalized in [20] and [24]; however, the description
of the basis elements given in both generalizations require that k is a field of
characteristic zero. In [34], Herzog and Maleki give a different description of a basis
for Hi (R) that does not depend on the characteristic of the field. This description
is given in terms of certain operators on Q, which are defined as follows. For
f ∈ (x1 , . . . , xn ) and for r = 1, . . . , n, let
f (0, . . . , 0, xr , . . . , xn ) − f (0, . . . , 0, xr+1 , . . . , xn )

d r (f ) = .
xr
It is clear that these operators are k-linear maps and that they depend on the order
of the variables.
Theorem 3.2 ([34, Theorem 1.3]) For each i = 1, . . . , n, a k-basis of Hi (R) is
given by [z̄j ] for j = 1, . . . , bi , where

b1
bi−1
zj = ··· d
i (fjii−1 ,j )d
i−1 (fji−1
i−2 ,ji−1
) . . . d
2 (fj21 ,j2 )d
1
L j1 =1 ji−1 =1
1
(f1,j 1
)dx
1 . . . dx
i
and L = {1
1 < · · · <
i n}

It is important to note that both theorems above give generators for the modules
Hi (R), but do not directly provide information about the algebra structure on H (R).
However, such bases have been used as tools to understand the Koszul homology
algebra; see for example Theorem 6.7, where Theorem 3.2 is used to study the
Koszul homology algebras of certain Koszul algebras. The theorems above have
also been particularly useful in obtaining criteria for Golodness, which we discuss
in Sect. 4.2 of this survey.
We finish this section by observing that just as the isomorphism (3.1) can be
used to describe Koszul homologies Hi (R) in terms of the data in the minimal
free resolution of R, as was done in Theorems 3.1 and 3.2, it can also be used to
describe the minimal resolution of R in terms of its Koszul homologies. The latter
was the focus of [1] and [2], in which Aramova and Herzog describe explicitly how
to construct the minimal free resolution of R, once bases of Koszul homologies are
known. These constructions provide our first concrete example of how the Koszul
homologies of R can be used as a tool for learning about the ring R, and we will see
several more examples throughout this survey.
4 Classical Characterizations
In this section we discuss classical characterizations of complete intersection

rings, Gorenstein rings, and Golod rings by their Koszul homology algebras. Such
characterizations have played a tremendous role in the effort to understand these
classes of rings. We highlight a few direct applications of these characterizations
throughout the section.
262 R. N. Diethorn
4.1 Complete Intersection and Gorenstein Rings
In this section we state the classical results that characterize complete intersection
and Gorenstein rings in terms of their Koszul homology algebras. We begin with
the characterization of complete intersection rings due to Assmus and Tate; the
equivalence of the first three conditions is due to Assmus, and the equivalence of
the first and last is due to Tate.
Theorem 4.1 ([3, Theorem 2.7], [56, Theorem 6]) The following conditions are
equivalent:
(1) R is a local complete intersection;
(2) H (R) is isomorphic to the exterior algebra on H1 (R);
(3) H2 (R) = (H1 (R))2 ;
(4) H (R) is generated by H1 (R) as a k-algebra.

Soto generalizes Theorem 4.1 slightly in [55, Proposition 3]; the result asserts
that R is a complete intersection if and only if H (R) is a free graded k-algebra.
Another characterization of complete intersection rings by their Koszul homology
algebras is given by Bruns. In order to state this characterization we first recall that
there is a natural map
C
λ: H1 (R) −→ H (R)
which extends the identity map on H1 (R); this fact follows from the universal
property of the exterior algebra on H1 (R), or from the graded commutativity of
H (R). Thus it follows from Theorem 4.1 that R is a complete intersection ring if
and only if the map λ is surjective. The result of Bruns below describes complete
intersections by the injectivity of this map.
Theorem 4.2 ([15, Theorem 2]) If R is a local ring containing a field, then the
following statements are true:
(1) H1 (R)i = 0 for i > edim R − dim R;
(2) R is a complete intersection if and only if λ is injective.

This result is applied in the proof of Theorem 6.16, which provides a condition
under which the defining ideal of a Koszul algebra is a complete intersection.
An analagous characterization of Gorenstein rings in terms of their Koszul
homology algebras involves the condition known as Ponicaré duality, which we
define below.
n
Definition 4.3 A graded algebra H = i=0 Hi satisfies Poincaré duality if for
each i = 0, . . . , n, the H0 -homomorphism
Hi −→ HomH0 (Hn−i , Hn )
h −→ φh
with φh (x) = hx, is an isomorphism.

Now we state the result of Avramov and Golod which asserts that Gorenstein
rings are characterized by their Koszul homology algebras.
Theorem 4.4 ([9]) Let R be a local ring, and let n = edim R − depth R. The
following conditions are equivalent:
1. R is a Gorenstein ring;
2. H (R) satisfies Poincaré duality;
3. the k-linear map Hn−1 (R) −→ Homk (H1 (R), Hn (R)) induced by the multipli-
cation on H (R) is injective.

We highlight applications of Theorem 4.4 at the end of Sect. 4.2 and in Sect. 5.
4.2 Golod Rings
In this section we discuss a classical result which characterizes Golod rings by their
Koszul homology algebras, and a more recent result which stems from this classical
theorem. At the end of the section, we highlight an application of Theorem 4.4
which provides a connection between the Golod and Gorenstein properties. We
begin with a definition of a Golod ring.
Definition 4.5 A local ring R is a Golod ring if its Poincaré series satisfies the
equality
(1 + t)edim R
PkR (t) = n , (4.1)
1 − j =1 rankk Hj (R)t j +1
where n = edim R − depth R.

A classical result of Serre asserts that, for any local ring, the formula in (4.5)
provides a coefficient-wise upper bound for its Poincaré series; see for example [28],
[43, Theorem 1.3], [5, Proposition 3.3.2], or [30, Corollary 4.2.4]. Thus, roughly
speaking, the resolution of the residue field k of a Golod local ring has the fastest
growth possible.
In [28], Golod introduced some higher order operations, called Massey oper-
ations, on Koszul homology, and used them to characterize rings satisfying the
equality (4.5), which we now call Golod rings. For a more detailed treatment of
Massey operations, see for example [30, Section 4.2].
Definition 4.6 We say that K(R) admits a trivial Massey operation if for some
k-basis B = {hλ }λ∈ of H1 (R), there is a function
264 R. N. Diethorn
∞
D
μ: Bn → K(R)
n=1
such that μ(hλ ) = zλ is a cycle with [zλ ] = hλ and

p−1
∂ K μ(hλ1 , . . . , hλp ) = μ(hλ1 , . . . , hλj )μ(hλj +1 , . . . , hλp ) (4.2)
j =1
where a = (−1)|a|+1 a.

We state Golod’s classical result using the modern terminology of Golod rings,
as follows.
Theorem 4.7 ([28]) A local ring R is Golod if and only if K(R) admits a trivial
Massey operation.

Thus Golod rings are characterized by having highly trivial Koszul homology
algebras. Indeed, if K(R) admits a trivial Massey operation, then (4.2) implies that
the product on H (R) is trivial; that is, H1 (R) · H1 (R) = 0. Note, however, that a
trivial product on H (R) is not enough to imply that R is Golod; the higher Massey
products must also be trivial. In [38], Katthän produces an example which illustrates
this fact.
Theorem 4.8 ([38, Theorem 3.1]) Let Q = k[x1 , x2 , y1 , y2 , z] be a polynomial
ring with k a field, let I be the ideal
I = (x1 x22 , y1 y22 , z3 , x1 x2 y1 y2 , y22 z2 , x22 z2 , x1 y1 z, x22 y22 z),
and let R = Q/I . Then the product on H (R) is trivial, but R is not Golod.

For the remainder of this section, we recognize a few direct applications of
the characterizations above. First, we look at an application of Theorem 4.7 and
Theorem 3.1 to Golod rings. In [33] Herzog and Huneke exploit the canonical bases
of Koszul homologies given in Theorem 3.1 and the characterization of Golod rings
in Theorem 4.7 to provide a differential condition for Golodness, and in doing so,
they are able to produce large classes of Golod rings, including quotients by powers
and symbolic powers of ideals. The differential condition is given in the following
result; see also [34, Sections 2 & 3], [31, Theorem 3.5], and [24, Proposition 4.4]
for similar applications.
Theorem 4.9 Let Q be a standard graded polynomial ring over a field of charac-
teristic zero, let I ⊆ Q be a homogeneous ideal, and let R = Q/I . Let ∂(I ) denote
the ideal generated by the partial derivatives of the elements of I . If (∂(I ))2 ⊆ I ,
then R is a Golod ring.

Now we look at an application of Theorem 4.4 by Avramov and Levin in [44]
which establishes a connection between Golod homorphisms and Gorenstein rings.
The following theorem asserts that factoring a Gorenstein ring by its socle produces
a Golod homomorphism; that is, a relative version of the Golod property. We discuss
an application of Theorem 4.10 to calculating Poincaré series in the following
section.
Theorem 4.10 ([44, Theorem 2]) Let R be a local Gorenstein ring with edim R >
1 and dim R = 0. Then
R −→ R/(0 : m)
is a Golod homomorphism.

Throughout this section, we have seen several classes of local rings which
are completely characterized by the algebra structures on their Koszul homology
algebras, and we conclude by emphasizing that, even for an arbitrary local ring,
the Koszul homology algebra is such a sensitive invariant that it encodes all of the
numerical information about the free resolution of its residue field, as we see in the
following result of Avramov.
Theorem 4.11 ([4, Corollary 5.10]) The Poincaré series PkR (t) of a local ring R
with residue field k depends only on the algebra structure on H (R) and its higher
order Massey operations.

5 Classifications of Koszul Homology Algebras
For a general ring, the algebra structure on its Koszul homology can be quite
complicated (see for example [53]); however, there are certain classes of rings whose
Koszul homology algebras are completely classified. We outline those cases in this
section.
We begin with the case of R = Q/I , where Q is a local ring with residue
field k, whose defining ideal I is perfect of grade 3; that is, the length of the
longest Q-regular sequence contained in I and the projective dimension of R
over Q are both equal to 3. In this case, Weyman [57] and Avramov, Kustin, and
Miller [10] (see also [6]) provide complete classifications of the algebra TorQ
• (R, k).
Restricting to the case where Q is regular, this gives the following classification of
the possible multiplicative structures on the Koszul homology algebra H (R) via the
isomorphism (3.1). Notice that in this case the last nonvanishing Koszul homology
module is H3 (R) by the Auslander-Buchsbaum formula and depth sensitivity of the
Koszul complex.
Theorem 5.1 ([10, Theorem 2.12]) Let R = Q/I with Q a regular local ring. If
the projective dimension of R over Q is 3, then there are nonnegative integers p, q,
and r and bases {ei }, {fi }, and {gi } of H1 (R), H2 (R), and H3 (R), respectively, such
that the multiplication on H (R) is given by one of the following:
266 R. N. Diethorn
CI: f1 = e2 e3 , f2 = e3 e1 , f3 = e1 e2 , ei fj = δij g1 for 1 i, j 3

TE: f1 = e2 e3 , f2 = e3 e1 , f3 = e1 e2
B: e1 e2 = f3 , e1 f1 = g1 , e2 f2 = g1
G(r): ei fi = g1 for 1 i r and r 2
H(p, q): ep+1 ei = fi for 1 i p and ep+1 fp+i = gi for 1 i q
with ej ei = −ei ej , ei2 = 0, and ei fj = fj ei for all i and j , and with the products
of basis elements that are not listed above being zero.

The class listed as CI is precisely the class where I is a complete intersection;
the class listed as G(r) is the class where I is Gorenstein, but not a complete
intersection, and r indicates the minimal number of generators of I ; the class listed
as TE consists of rings that are neither Golod, nor complete intersections.
The classification in Theorem 5.1 builds on the classification for perfect grade
3 almost complete intersection ideals given by Buchsbaum and Eisenbud in [16,
Theorem 5.3]; see also [14] for another special case of this classification.
Such classifications for the Tor algebra are also known for grade 4 Gorenstein
ideals and grade 4 almost complete intersection ideals; the former is due to Kustin
and Miller [42, Theorem 2.2], and the latter is due to Kustin [40, Theorem
1.5]. Again, restricting to the case where the ambient ring is regular, we present
the former case here, and we note that the proof applies the characterization of
Gorenstein rings in Theorem 4.4.
Theorem 5.2 ([42, Theorem 2.2]) Let R = Q/I with Q a regular local ring, and
assume that every element in k has a square root in k. If I is a grade 4 Gorenstein
ideal that is not a complete intersection, then there are bases e1 , . . . , en for H1 (R);
f1 , . . . , fn−1 , f1 , . . . , fn−1
for H2 (R); g1 , . . . , gn for H3 (R); and h for H4 (R)
such that the multiplication Hi (R) · H4−i (R) is given by
ei gj = δij h, fi fj = δij h, fi fj = fi fj = 0,
with all other products given by one of the following cases:

(1) H1 (R) · H1 (R) = 0 and H1 (R) · H2 (R) = 0
(2) All products in H1 (R) · H1 (R) and H1 (R) · H2 (R) are zero except:
e1 e2 = f3 , e1 e3 = −f2 , e2 e3 = f1 ,
e1 f2 = −e2 f1 = g3 , −e1 f3 =e3 f1 = g2 , e2 f3 = −e3 f2 = g1
(3) There is an integer p such that ep+1 ei = fi , ei fi gp+1 , and ep+1 fi = −gi for
1 i p, and all other products in H1 (R) · H1 (R) and H1 (R) · H2 (R) are
zero.

Not surprisingly, we see that the algebra structure in the grade 4 case is more
complicated than that in the grade 3 case.
The classifications of Koszul homology algebras in Theorems 5.1 and 5.2 have
been instrumental in the study of rationality of Poincaré series, which has been a
topic of intense interest over the years and remains a central problem in homological
commutative ring theory today; see for example [49, Section 6] or [5, Section 4.3]
for a complete treatment of this problem. For example, Theorem 5.1 is an important
ingredient in the proof of [10, Theorem 6.4], which determines several classes of
rings over which the Poincaré series of a finitely generated module is rational;
analagously, [37, Theorem A] applies the classification in Theorem 5.2. In fact,
much of the recent progress on rationality of Poincaré series relies on the classical
characterizations of Koszul homology discussed in Sect. 4. As a direct application
of Theorem 4.10, Rossi and Şega calculate the Poincaré series of a class of Artinian
Gorenstein rings in [54, Proposition 6.2] and show that they are indeed rational.
For similar applications of the Koszul homology algebra structure in calculating
Poincaré series, see [54, Theorem 5.1] and [41, Theorem 7.1].
We finish this section by recognizing one more recent application of the
classifications above. In [17], Christensen and Veliche determine minimal cases for
which powers of the maximal ideal in a local ring are not Golod. We state their
result in embedding dimension 3, which uses the classification in Theorem 5.1; the
embedding dimension 4 case [17, Proposition 5.2] uses Theorem 5.2.
Theorem 5.3 ([17, Theorem 4.2]) Let R be an Artinian Gorenstein local ring of
embedding dimension 3 and socle degree 3 with maximal ideal m. The following
conditions are equivalent:
(1) R is a complete intersection;
(2) R is compressed and Koszul;
(3) R/m3 belongs to the class TE;
(4) R/m3 is not Golod.

Such applications demonstrate the usefulness of the Koszul homology algebra
H (R) as a tool for learning about R, and discovering its properties.
6 Recent Progess: Koszul Algebras
In this section, we focus on recent progress on the Koszul homology algebras

of Koszul algebras; that is, k-algebras over which k has a linear resolution. One
can check that the defining ideals of Koszul algebras are generated by quadratics;
however, not all algebras defined by quadratics are Koszul (see for example [18,
Remark 1.10]). Classical examples of Koszul algebras include quotients of poly-
nomial rings by quadratic monomial ideals (e.g. edge ideals), Veronese algebras,
and Segre product algebras. Unlike the classes of rings discussed in Sect. 4, Koszul
algebras are not known to be characterized solely by the algebra structures on their
Koszul homologies; nonetheless, the Koszul property of R is closely connected to
the algebra structure on its Koszul homology H (R).
To present these connections, we assume throughout this section that R = Q/I
is a standard graded k-algebra with Q = k[x1 , . . . , xn ] and k a field (although some
268 R. N. Diethorn
results can be stated more generally), and we view the Koszul homology algebra
H (R) of a Koszul algebra R as a bigraded algebra
'
H (R) = Hi (R)j ,
i,j
where i is the homological degree, and j is the internal degree given by the grading
on R. Given the isomorphism (3.1), the rank of Hi (R)j is given by the graded
Betti number βij of R. As such, we can view these bigraded pieces of the Koszul
homology algebra in the following table, which comes from the Betti table of R:
PP
PP i
j − i PPP
0 1 2 3 ...
P
0 H0,0 H1,1 H2,2 H3,3 ...
1 H0,1 H1,2 H2,3 H3,4 ...
2 H0,2 H1,3 H2,4 H3,5 ...
3 H0,3 H1,4 H2,5 H3,6 ...
.. .. .. .. ..
. . . . .
where Hi,j = Hi (R)j .

It is customary to call row one of the Betti table the linear strand and the other
rows nonlinear strands. More precisely, we have the following definition.
Definition 6.1 We call the subspace
'
Hi (R)i+1
i
of H (R) the linear strand of H (R).

Now we recall some useful terminology for describing the shapes of Betti tables,
which we will use throughout this section. In [7], Avramov, Conca, and Iyengar
introduce the sequence
ti (R) = sup{j ∈ Z | Hi (R)j = 0},
which encodes important information about R. For example, the integer ti (R) − i is
the height of the ith column of the Betti table. Notice that the regularity of R
reg(R) = sup{ti (R) − i}

i0
measures the height of the Betti table above. The slope of R,

? E
ti (R) − t0 (R)
slope(R) = sup ,
i1 i
introduced and studied in [7], measures the slope of the Betti table. The last number
we will use in this section to describe the shapes of Betti tables is
m(R) = min{i ∈ Z | ti (R) ti+1 (R)},
which is introduced in [8].

The first result we state is a special case of a more general result of Avramov,
Conca and Iyengar in [7, Main Theorem]; the inequality in (2) follows from results
of Backelin in [11] and of Kempf in [39, Lemma 4].
Theorem 6.2 If R is Koszul, then R satisfies the following conditions:
(1) slope(R) = 2;
(2) ti (R) 2i for all i 0;
(3) reg(R) pdQ R.

The general result from which Theorem 6.2 follows is one about the slope and
regularity of R where Q is Koszul, with no assumptions on R, and the proof is
based on the theory of minimal models; see for example [5, Section 7.2]. We will
see several other applications of minimal models throughout this section.
It follows from Theorem 6.2 that if R is Koszul, then Hi (R)j = 0 for all j > 2i,
which we record in Theorem 6.3, and one can easily check that Hi (R)i = 0 for all
i 1; in other words, the Betti table above has the form:
PP
PP i
j − i PPP
0 1 2 3 ...
P
0 H0,0 0 0 0 ...
1 0 H1,2 H2,3 H3,4 ...
2 0 0 H2,4 H3,5 ...
3 0 0 0 H3,6 ...
.. .. .. .. ..
. . . . .
Theorem 6.2, along with the other results of [7] and [8], sparked an intensive
investigation of the Koszul homology algebra of a Koszul algebra, which we outline
throughout the rest of this section.
6.1 Generation by the Linear Strand
In this section we discuss a question of Avramov about the generation of the Koszul
homology algebra of a Koszul algebra. We begin with the following result of
270 R. N. Diethorn
Avramov, Conca, and Iyengar which records a consequence of Theorem 6.2 and
describes the algebra structure on the main diagonal of the Betti table. This result
prompted Question 6.4, which we discuss throughout this section.
Theorem 6.3 ([8, Theorem 5.1]) If R is Koszul, then
(1) Hi (R)j = 0 for all j > 2i, and
(2) Hi (R)2i = (H1 (R)2 )i for all i 0.

In fact, Avramov, Conca, and Iyengar prove a more general version of Theo-
rem 6.3 where the Koszul hypothesis is relaxed to only require linearity in the
resolution of k up to degree n. Under this relaxed hypothesis, the equalities in the
theorem hold for all i n − 1. Again, the proof is based on the theory of minimal
models.
By Theorem 6.3, one can see that the elements of the Koszul homology algebra
which lie on the main diagonal are contained in the subalgebra generated by the
linear strand. This observation led Avramov to ask the following question.
Question 6.4 If R is Koszul, is the Koszul homology algebra of R generated as a
k-algebra by the linear strand?

In [12, Theorem 3.1], the authors extend Theorem 6.3 to the next diagonal of the
Betti table; their result is stated as follows.
Theorem 6.5 ([12, Theorem 3.1]) If R is Koszul, then
Hi (R)2i−1 = (H1 (R)2 )i−2 H2 (R)3
for all i 2.

However, the answer to Question 6.4 is negative in general. The first counterex-
ample was found computationally by Eisenbud and Caviglia on Macaulay2 [29],
and this example led Conca and Iyengar to consider quotients by edge ideals of n-
cycles. From this, Boocher, D’Alí, Grifo, Montaño, and Sammartano were able to
produce the family of counterexamples in (3) of the following theorem. The proofs
of both Theorems 6.5 and 6.6 continue the trend of using the minimal model of R
over Q to study H (R).
Theorem 6.6 ([12, Theorem 3.15]) Let Q = k[x1 , . . . , xn ], let I be the edge ideal
associated to a graph G on the vertices x1 , . . . , xn , and let R = Q/I .
(1) If G is an n-path, then H (R) is generated by H1 (R)2 and H2 (R)3 .
(2) If G is an n-cycle with n ≡ 1(mod 3), then H (R) is generated by H1 (R)2 and
H2 (R)3 .
(3) If G is an n-cycle with n ≡ 1(mod 3), then for any 0 = z ∈ H 2n (R)n , H (R)
3
is generated by H1 (R)2 , H2 (R)3 , and z.

In [23, Theorem 4.2] the author applies Theorem 3.2, and extends Theo-
rem 6.6(1) to give a broader class of edge ideals for which Question 6.4 has a
positive answer.
Theorem 6.7 ([23, Theorem 4.2]) Let Q = k[x1 , . . . , xn ] with k a field, let I be
the edge ideal associated to a forest on the vertices x1 , . . . , xn , and let R = Q/I .
Then H (R) is generated by the linear strand.

The following result of Conca, Katthän, and Reiner gives another class of Koszul
algebras whose Koszul homology algebras are generated by the linear strand.

Theorem 6.8 ([19, Corollary 2.4]) Let Q = k[x1 , . . . , xn ] = i Qi with k a field
of characteristic zero, and let R = i Q2i be the second Veronese subalgebra.
Then H (R) is generated by the linear strand.

As the authors of [19] point out, Question 6.4 remains open for other classes of
Koszul algebras, such as higher Veronese subalgebras and Segre product algebras.
Furthermore, the authors of [12] ask the following weaker version of Question 6.4.
Question 6.9 ([12, Question 3.16]) Is there a Koszul algebra R which is a domain
and whose Koszul homology H (R) is not generated as a k-algebra by the linear
strand?

The answer to this question is affirmative; the example below was communicated
to the authors of [12] by McCullough.
Example 6.10 Let k be a field and let
R = k[ae, af, ag, ah, bg, bh, ce, cg, ch, de, df, dg].
Notice that R is the toric edge ring of a bipartite graph, hence a domain, and its
defining ideal is generated by quadratic binomials; indeed,
R∼
= k[X1 , X2 , X3 , X4 , X5 , X6 , X7 , X8 , X9 , X10 , X11 , X12 ]/I,
where
I = X3 X11 −X2 X12 , X8 X10 −X7 X12 , X3 X10 −X1 X12 , X2 X10 − X1 X11 ,
!
X6 X8 −X5 X9 , X4 X8 −X3 X9 , X4 X7 − X1 X9 , X3 X7 − X1 X8 , X4 X5 − X3 X6 ,
thus R is Koszul by [52]. Using Macaulay2 [29], we see that the Betti table of R is
01 2 3 4 5
0 1- - - - -
1 -9 11 - - -
2 - - 10 26 15 1
3 - - - - - 1
272 R. N. Diethorn
and it is easy to see that the Koszul homology algebra must have a generator in
bidegree (5, 7) for degree reasons. Thus H (R) is not generated by the linear strand.

So far we have seen that Koszulness of R does not imply generation by the linear
strand in H (R). It is also the case that generation by the linear strand in H (R) does
not imply Koszulness of R, as we see in the example below.
Example 6.11 ([21, 7.4]) Let k be a field, and let
R = k[X, Y, Z, U ]/(X2 + XY, XZ + Y U, XU, Y 2 , Z 2 , ZU + U 2 )
(see [53, Case 55]). H (R) is generated by the linear strand; indeed, it has only 6
generators in bidegree (1, 2) and 4 generators in bidegree (2, 3). However, R is not
Koszul; the resolution of k is not linear.

However, the authors of [21] provide stronger conditions on the algebra structure
of H (R) that are enough to imply Koszulness of R. They prove that if H (R)
is generated by either a single element of bidegree (1, 2), or by a special set of
elements in the linear strand, then R is Koszul. More precisely, their result is as
follows.
Theorem 6.12 ([21, Theorem 6.1]) Assume one of the following conditions
holds:
(1) There exists an element [z] of bidegree (1, 2) such that every element in the
nonlinear strands of H (R) is a multiple of [z].
(2) R3 = 0 and there is a set of cycles Z representing elements in the linear
strand with the property that zz = 0 for all z, z ∈ Z whose homology classes
generate the nonlinear strands of H (R).
Then R is Koszul.

In fact, they prove that the condition in (1) implies that R is absolutely Koszul, a
condition which implies Koszulness.
6.2 Upper Bounds on Betti Numbers
In this section we discuss the following question of Avramov, Conca, and Iyengar
about upper bounds of the Betti numbers of R, and equivalently, by (3.1), the ranks
of the Koszul homology modules of Koszul algebras.
Question 6.13 ([7, Question 6.5]) If R = Q/I is Koszul, and I is minimally
generated by g quadrics, does the inequality

Q g
βi (R) ,
i
!
equivalently, dim k Hi (R) gi , hold for all i? In particular, is the projective
dimension of R over Q at most g?

Standard arguments show that the answer to this question is positive for Koszul
algebras whose defining ideals are generated by monomials or have Gröbner bases
of quadrics. Furthermore, the answer to this question is positive for Koszul algebras
whose defining ideals are minimally generated by g 4 quadrics and for Koszul
almost complete intersection algebras; that is, algebras with ht I = g − 1, as we
outline below. Otherwise, Question 6.13 remains open.
The case where g 3 is addressed by Boocher, Hassanzadeh, and Iyengar in the
following result.
Theorem 6.14 ([13, Theorem 4.5]) If I is generated by 3 quadrics, then the
following conditions are equivalent.
(1) R is Koszul;
(2) H (R) is generated by the linear strand;
(3) The Betti table of R over Q is one of the following:
0123
0123
0 1- - - 012 0123
01- - -
1 -3- - 01- - 01- - -
1 -31-
2 - -3- 1 -32 1 -331
2 - -21
3 - - -1

In [13] the authors note that the first Betti table listed in Theorem 6.14
corresponds to a complete intersection of three quadrics; the second is the Betti
table of the ring k[x, y, x]/(x 2 , y 2 , xz); the third corresponds to the ideal of minors
of a 3 × 2 matrix of linear forms (e.g. k[x, y]/(x, y)2 ); the last is the Betti table of
a ring with a linear resolution (e.g. k[x, y, z]/(x 2 , xy, xz)).
The implication (1) -⇒ (3) in Theorem 6.14 gives an affirmative answer to
Question 6.13 in the case where g 3, and by the isomorphism (3.1), it determines
the ranks of the bigraded pieces of H (R) for such algebras; the implication (1) -⇒
(2) provides another class of Koszul algebras for which Question 6.4 has a positive
answer.
The proofs of the implications (2) -⇒ (1) and (3) -⇒ (1) use D’Alí’s
classification of Koszul algebras defined by 3 quadrics in [22], which we state below.
Theorem 6.15 ([22, Theorem 3.1]) Let k be an algebraically closed field of
characteristic different from 2, let Q be a polynomial ring over k, and let R = Q/I
with I a quadratic ideal of Q. If dimk R2 = 3, then R is Koszul if and only if it is not
isomorphic as a graded k-algebra (up to trivial fiber extension) to any of these:
(1) k[x, y, z]/(y 2 + xy, xy + z2 , xz)
(2) k[x, y, z]/(y 2 , xy + z2 , xz)
(3) k[x, y, z]/(y 2 , xy + yz + z2 , xz).
274 R. N. Diethorn
In particular, if R is a non-Koszul algebra with edim R = 3 defined by 3

quadrics, then its Betti table is
0123
01- - -
1 -3- -
2 - -42

The proof of Theorem 6.14 is also based on some more general results on the
algebra structure on H(R). The authors of [13, Theorem 3.3] prove that the diagonal
subalgebra (R) = i Hi (R)2i of H (R) is a quotient of the exterior algebra on
H1 (R)2 by quadratic relations that depend only on the first syzygies of I . As a
consequence of this, they find that the linear strand and the main diagonal of the
Betti table for a Koszul algebra satisfy the inequality in Question 6.13, as we see in
the following result. We note that the proof of (2) also uses the characterization of
complete intersections given by Bruns in Theorem 4.2.
Theorem 6.16 ([13, Corollary 3.4, Proposition 4.2]) If R is a Koszul algebra
whose defining ideal is minimally generated by g quadrics, then
!
(1) dim k Hi (R)i+1 gi for 2 i g, and if equality holds for i = 2, then I has
height 1 and a linear
! resolution of length g.
(2) dim k Hi (R)2i gi for 2 i g, and if equality holds for some i, then I is a
complete intersection.

We note that the inequality in Theorem 6.16 (2) clearly holds for quadratic
complete intersections, and the fact that it holds for Koszul algebras which are
not complete intersections follows from the contrapositive of the statement in [13,
Corollary 3.4]. In fact, this inequality also follows from the proofs of [7, Theorem
3.1] and [7, Corollary 3.2].
Building on the work of Boocher, Hassanzadeh, and Iyengar on the g 3 case,
Mantero and Mastroeni describe the Betti tables of Koszul algebras whose defining
ideals are generated by 4 quadrics in [45]; their result is as follows.
Theorem 6.17 ([45, Main Theorem]) Let R = Q/I be a Koszul algebra with I
minimally generated by 4 quadrics and ht I = 2. Then the Betti table of R over Q
is one of the following:
0123 01234 0123 4

0123
01- - - 01- - - - 01- - - -
01- - -
1 -43- 1 -431- 1 -42- -
1 -441

2 - -11 2 - -331 2 - -44 1
Theorem 6.17 gives an affirmative answer to 6.13 in the case that g = 4 and
ht I = 2; the case where ht I = 4 (i.e. I is a complete intersection) is clear, the case
where ht I = 1 follows from Theorem 6.2, and the case where ht I = 3 is settled by
Mastroeni in [46, Main Theorem]. Together these results give an affirmative answer
to Question 6.13 when g = 4. However, Question 6.13 remains open for g > 4.
6.3 Subadditivity of Syzygies
The focus of this section is the following conjecture of Avramov, Conca, and Iyengar
from [8] concerning the subadditivity of syzygies, and equivalently the subadditivity
of Koszul homologies.
Conjecture 6.18 ([8, Conjecture 6.4]) If R is a Koszul algebra, then the following
inequality holds
ta+b (R) ta (R) + tb (R)
whenever a + b pdQ R.

This conjecture is based on the following result, which asserts that a weaker
inequality holds.
Theorem 6.19 ([8, Theorem 6.2]) Let a and b be non-negative integers such that
a+b!
a is invertible in k and max{a, b} m(R). If R is Koszul, then the following
inequalities hold
ta+1 (R) ta (R) + 2

ta+b (R) ta (R) + tb (R) + 1
for b 2.

In fact, both Conjecture 6.18 and Theorem 6.19 are stated more generally in [8];
rather than requiring that R is Koszul, it is only required that the minimal resolution
of k is linear up to a certain degree.
To our knowledge, Conjecture 6.18 remains open; however, the following results
of Eisenbud, Huneke, and Ulrich and of Herzog and Srinivasan provide supporting
evidence for its validity.
Theorem 6.20 Conjecture 6.18 holds in the following cases:
(1) ([25, Corollary 4.1]) dimR 1, depthR = 0, and a + b = rankk R1 ;
(2) ([27, Corollary 2.1], [35, Corollary 4]) R has monomial relations and b = 1;

We note that the quadratic monomial case in Theorem 6.20 (2) is treated in [27]
and the general monomial case is treated in [35].
Furthermore, in [35] Herzog and Srinivasan prove the following case of Conjec-
ture 6.18 (without the Koszul assumption), which improves a result of McCullough
in [48].
276 R. N. Diethorn
Theorem 6.21 ([35, Corollary 3], [48, Theorem 4.4]) Let pdQ R = p. Then the
following inequality holds
tp (R) tp−1 (R) + t1 (R).
We finish this section by noting that the question of when the subadditivity
property holds is an interesting one, even for non-Koszul algebras. The subadditivity
property holds for complete intersection ideals, but not for Gorenstein ideals; these
are recent results of McCullough and Seceleanu in [50]. Although the subadditivity
property holds for certain classes of monomial ideals (see for example [26]), the
question remains open for general monomial ideals (as well as for Koszul algebras,
as discussed above).
6.4 Towards a Characterization
Although the Koszul property of R has not been shown to be characterized com-
pletely by the algebra structure on H (R), recent work of Myers in [51] introduces
a property of H (R) that is equivalent to the Koszul property of R under some extra
conditions. Myers defines H (R) to be strand-Koszul if H (R) = j −i=n Hi (R)j
is Koszul in the classical sense; that is, if k has a linear resolution over H (R). With
this terminology, he proves the following result.
Theorem 6.22 ([51, Theorem C]) Let R be a standard graded k-algebra with k a
field. If H (R) is strand-Koszul, then R is Koszul. Furthermore, the converse holds
if R satisfies one of the following conditions:
(1) R has embedding dimension 3;
(2) R is Golod;
(3) The defining ideal of R is minimally generated by 3 elements;
(4) R is a quadratic complete intersection;
(5) The defining ideal of R is the edge ideal of a path on 3 vertices.

Despite the connections we have seen between the Koszul property and the alge-
bra structure on Koszul homology throughout this section and the progress towards
a characterization highlighted in the theorem above, results of Roos in [53] present
potential challenges, or perhaps even obstructions, to finding a characterization of
Koszul algebras by the algebra structures on their Koszul homologies alone, as we
see in the following theorem.
Theorem 6.23 ([53, Section 3]) The Koszul homology algebras of the Koszul
algebras
R71 = k[x, y, z, u]/(x 2 , y 2 , z2 , u2 , xy, zu, yz − xu)

and
R71v16 = k[x, y, z, u]/(xz + u2 , xy, xu, x 2 , y 2 + z2 , zu, yz),
are isomorphic as bigraded k-vector spaces; however, they are not isomorphic as
k-algebras.

Roos uses the DGAlgebras package written by Moore to describe and compare
the Koszul homology algebra structures of the Koszul algebras A = R71 and B =
R71v16 defined above. Roos shows that the Poincaré series PkA (t) and PkB (t), and the
Betti tables of A and B over Q = k[x, y, z, u] are both the same, hence H (A) and
H (B) are isomorphic as bigraded k-vector spaces; however, the algebra structures
on H (A) and H (B) are different. In fact, Roos concludes that the Eilenberg-Moore
spectral sequence degenerates over H (A) but not over H (B); thus, by [51, Theorem
B], H (A) is strand-Koszul and H (B) is not.
We end by noting that the algebra B = R71v16 provides an example of a Koszul
algebra whose Koszul homology algebra is not strand-Koszul; thus, the converse of
Theorem 6.22 does not hold in general.
Acknowledgments The author thanks the anonymous referee for many helpful and insightful
comments that greatly improved this survey.
References
1. A. Aramova and J. Herzog, Free resolutions and Koszul homology, J. Pure Appl. Algebra, 105
(1995), pp. 1–16.
2. , Koszul cycles and Eliahou-Kervaire type resolutions, J. Algebra, 181 (1996), pp. 347–
370.
3. E. F. Assmus, Jr., On the homology of local rings, Illinois J. Math., 3 (1959), pp. 187–199.
4. L. L. Avramov, The Hopf algebra of a local ring, Izv. Akad. Nauk SSSR Ser. Mat., 38 (1974),
pp. 253–277.
5. L. L. Avramov, Infinite free resolutions, in Six lectures on commutative algebra, Mod.
Birkhäuser Class., Birkhäuser Verlag, Basel, 2010, pp. 1–118.
6. , A cohomological study of local rings of embedding codepth 3, J. Pure Appl. Algebra,
216 (2012), pp. 2489–2506.
7. L. L. Avramov, A. Conca, and S. B. Iyengar, Free resolutions over commutative Koszul
algebras, Math. Res. Lett., 17 (2010), pp. 197–210.
8. , Subadditivity of syzygies of Koszul algebras, Math. Ann., 361 (2015), pp. 511–534.
9. L. L. Avramov and E. S. Golod, On the homology algebra of the Koszul complex of a local
Gorenstein ring, Math. Notes 9, (1971), pp. 30–32.
10. L. L. Avramov, A. R. Kustin, and M. Miller, Poincaré series of modules over local rings of
small embedding codepth or small linking number, J. Algebra, 118 (1988), pp. 162–204.
11. J. Backelin, Relations between rates of growth of homologies, Research Reports in Mathemat-
ics 25, Department of Mathematics, Stockholm University, (1988).
12. A. Boocher, A. D’Alì, E. Grifo, J. Montaño, and A. Sammartano, Edge ideals and DG algebra
resolutions, Matematiche (Catania), 70 (2015), pp. 215–238.
278 R. N. Diethorn
13. A. Boocher, S. H. Hassanzadeh, and S. B. Iyengar, Koszul algebras defined by three relations,
in Homological and computational methods in commutative algebra, vol. 20 of Springer
INdAM Ser., Springer, Cham, 2017, pp. 53–68.
14. A. E. Brown, A structure theorem for a class of grade three perfect ideals, J. Algebra, 105
(1987), pp. 308–327.
15. W. Bruns, On the Koszul algebra of a local ring, Illinois J. Math., 37 (1993), pp. 278–283.
16. D. A. Buchsbaum and D. Eisenbud, Algebra structures for finite free resolutions, and some
structure theorems for ideals of codimension 3, Amer. J. Math., 99 (1977), pp. 447–485.
17. L. W. Christensen and O. Veliche, The Golod property of powers of the maximal ideal of a
local ring, Arch. Math. (Basel), 110 (2018), pp. 549–562.
18. A. Conca, Koszul algebras and their syzygies, arXiv e-prints, (2013), p. arXiv:1310.2496.
19. A. Conca, L. Katthän, and V. Reiner, The Koszul homology algebra of the second Veronese is
generated by the lowest strand, J. Algebra, 571 (2021), pp. 179–189.
20. A. Corso, S. Goto, C. Huneke, C. Polini, and B. Ulrich, Iterated socles and integral dependence
in regular rings, Trans. Amer. Math. Soc., 370 (2018), pp. 53–72.
21. A. Croll, R. Dellaca, A. Gupta, J. Hoffmeier, V. Mukundan, L. M. Şega, G. Sosa, P. Thompson,
and D. Rangel Tracy, Detecting Koszulness and related homological properties from the
algebra structure of Koszul homology, Nagoya Math. J., 238 (2020), pp. 47–85.
22. A. D’Alì, The Koszul property for spaces of quadrics of codimension three, J. Algebra, 490
(2017), pp. 256–282.
23. R. N. Diethorn, Koszul homology of quotients by edge ideals, arXiv e-prints, (2019),
p. arXiv:1908.10848.
24. R. N. Diethorn, Generators of Koszul homology with coefficients in a J-closed module, Journal
of Pure and Applied Algebra, 224 (2020), p. 106387.
25. D. Eisenbud, C. Huneke, and B. Ulrich, The regularity of Tor and graded Betti numbers, Amer.
J. Math., 128 (2006), pp. 573–605.
26. S. Faridi, Lattice complements and the subadditivity of syzygies of simplicial forests, J.
Commut. Algebra, 11 (2019), pp. 535–546.
27. O. Fernández-Ramos and P. Gimenez, Regularity 3 in edge ideals associated to bipartite
graphs, J. Algebraic Combin., 39 (2014), pp. 919–937.
28. E. S. Golod, Homologies of some local rings, Dokl. Akad. Nauk SSSR, 144 (1962), pp. 479–
482.
29. D. R. Grayson and M. E. Stillman, Macaulay2, a software system for research in algebraic
30. T. H. Gulliksen and G. Levin, Homology of local rings, Queen’s Paper in Pure and Applied
Mathematics, No. 20, Queen’s University, Kingston, Ont., 1969.
31. A. Gupta, Ascent and descent of the Golod property along algebra retracts, J. Algebra, 480
(2017), pp. 124–143.
32. J. Herzog, Canonical Koszul cycles, in International Seminar on Algebra and its Applications
(Spanish) (México City, 1991), vol. 6 of Aportaciones Mat. Notas Investigación, Soc. Mat.
Mexicana, México, 1992, pp. 33–41.
33. J. Herzog and C. Huneke, Ordinary and symbolic powers are Golod, Adv. Math., 246 (2013),
pp. 89–99.
34. J. Herzog and R. A. Maleki, Koszul cycles and Golod rings, Manuscripta Mathematica, 157
(2018), pp. 483–495.
35. J. Herzog and H. Srinivasan, On the subadditivity problem for maximal shifts in free
resolutions, in Commutative algebra and noncommutative algebraic geometry. Vol. II, vol. 68
of Math. Sci. Res. Inst. Publ., Cambridge Univ. Press, New York, 2015, pp. 245–249.
36. C. Huneke, The Koszul homology of an ideal, Adv. in Math., 56 (1985), pp. 295–318.
37. C. Jacobsson, A. R. Kustin, and M. Miller, The Poincaré series of a codimension four
Gorenstein ring is rational, J. Pure Appl. Algebra, 38 (1985), pp. 255–275.
38. L. Katthän, A non-Golod ring with a trivial product on its Koszul homology, J. Algebra, 479
(2017), pp. 244–262.
39. G. R. Kempf, Some wonderful rings in algebraic geometry, J. Algebra, 134 (1990), pp. 222–
224.
40. A. R. Kustin, Classification of the Tor-algebras of codimension four almost complete intersec-
tions, Trans. Amer. Math. Soc., 339 (1993), pp. 61–85.
41. A. R. Kustin, L. M. Şega, and A. Vraciu, Poincaré series of compressed local Artinian rings
with odd top socle degree, J. Algebra, 505 (2018), pp. 383–419.
42. A. R. Kustin and M. Miller, Classification of the Tor-algebras of codimension four Gorenstein
local rings, Math. Z., 190 (1985), pp. 341–355.
43. G. Levin, Lectures on Golod Homomorphisms, Research Reports of the University of
Stockholm (1976), no. 15.
44. G. L. Levin and L. L. Avramov, Factoring out the socle of a Gorenstein ring, J. Algebra, 55
(1978), pp. 74–83.
45. P. Mantero and M. Mastroeni, Betti numbers of Koszul algebras defined by four quadrics, J.
Pure Appl. Algebra, 225 (2021), pp. 106504, 16.
46. M. Mastroeni, Koszul almost complete intersections, J. Algebra, 501 (2018), pp. 285–302.
47. H. Matsumura, Commutative ring theory, vol. 8 of Cambridge Studies in Advanced Mathe-
matics, Cambridge University Press, Cambridge, 1986. Translated from the Japanese by M.
Reid.
48. J. McCullough, A polynomial bound on the regularity of an ideal in terms of half of the syzygies,
Math. Res. Lett., 19 (2012), pp. 555–565.
49. J. McCullough and I. Peeva, Infinite graded free resolutions, Commutative algebra and
noncommutative algebraic geometry. Vol. I, Math. Sci. Res. Inst. Publ., vol. 67, Cambridge
Univ. Press, New York, 2015, pp. 215–257.
50. J. McCullough and A. Seceleanu, Quadratic Gorenstein algebras with many surprising
properties, Arch. Math. (Basel), 115 (2020), pp. 509–521.
51. J. Myers, Linear resolutions over Koszul complexes and Koszul homology algebras, J. Algebra,
572 (2021), pp. 163–194.
52. H. Ohsugi and T. Hibi, Koszul bipartite graphs, Adv. in Appl. Math., 22 (1999), pp. 25–28.
53. J.-E. Roos, Homological properties of the homology algebra of the Koszul complex of a local
ring: examples and questions, J. Algebra, 465 (2016), pp. 399–436.
54. M. E. Rossi and L. M. Şega, Poincaré series of modules over compressed Gorenstein local
rings, Adv. Math., 259 (2014), pp. 421–447.
55. J. J. M. Soto, Quasi-isomorphisms of Koszul complexes, Proc. Amer. Math. Soc., 128 (2000),
pp. 713–715.
56. J. Tate, Homology of Noetherian rings and local rings, Illinois J. Math., 1 (1957), pp. 14–27.
57. J. Weyman, On the structure of free resolutions of length 3, J. Algebra, 126 (1989), pp. 1–33.
Canonical Resolutions over Koszul
Algebras
Eleonore Faber, Martina Juhnke-Kubitzke, Haydee Lindo, Claudia Miller,

Rebecca R. G., and Alexandra Seceleanu
Keywords Koszul algebra · Minimal free resolution · Betti numbers
1 Introduction
Koszul algebras show up naturally and abundantly in algebra and topology. They
were first introduced by Priddy in 1970 as algebras for which the bar resolution,
which is normally far from minimal, admits a reduction to a comparatively small
subcomplex; see [23]. Priddy’s work explained contemporaneous ideas on restricted
Lie algebras in the work of May and, separately, that of Bousfield, Curtis, Kan,
Quillen, Rector, and Schlesinger; see [2, 17]. Priddy was an algebraic topologist,
but Koszul algebras have since been linked to several fundamental concepts across
mathematics where they appear naturally and are studied extensively in fields
E. Faber
School of Mathematics, University of Leeds, Leeds, UK
e-mail: e.m.faber@leeds.ac.uk
M. Juhnke-Kubitzke
Department of Mathematics, University of Osnabrück, Osnabrück, Germany
e-mail: juhnke-kubitzke@uni-osnabrueck.de
H. Lindo
Department of Mathematics, Harvey Mudd College, Claremont, CA, USA
e-mail: hlindo@hmc.edu
C. Miller
Department of Mathematics, Syracuse University, Syracuse, NY, USA
e-mail: clamille@syr.edu
Rebecca R. G.
Department of Mathematical Sciences, George Mason University, Fairfax, VA, USA
e-mail: rrebhuhn@gmu.edu
A. Seceleanu ()
Department of Mathematics, University of Nebraska–Lincoln, Lincoln, NE, USA
e-mail: aseceleanu@unl.edu
282 E. Faber et al.
as diverse as topology [14], representation theory [6], commutative algebra [12],

algebraic geometry [4], noncommutative geometry [16], and number theory [21].
For a general overview, see the monograph by Polishchuk and Positselski [22].
An interesting feature of Koszul algebras is that they appear in pairs: every
Koszul algebra A has a dual algebra A! which is also a Koszul algebra (see
Sect. 2). The prototypical example of such a Koszul pair is a polynomial algebra
S over a field, together with the corresponding exterior algebra . The associated
theory of Koszul duality is a generalization of the duality underlying the Bernstein–
Gelfand–Gelfand correspondence [4] describing coherent sheaves on projective
space in terms of modules over the exterior algebra. This exemplifies the philosophy
that facts relating the symmetric and exterior algebras often have Koszul duality
counterparts.
In this paper we extend Priddy’s methods of constructing free resolutions
over standard graded Koszul algebras and generalize Buchsbaum and Eisenbud’s
resolutions in [3] to resolve powers of the homogeneous maximal ideal over Koszul
algebras. Resolutions over Koszul algebras have previously appeared in works of
Green and Martínez-Villa [15, Theorem 5.6], Martínez-Villa and Zacharia [20,
Proposition 3.2] and in other sources referenced below. Our approach has the
advantage of producing explicit minimal resolutions.
In particular, in [23] Priddy exploits a natural differential on A ⊗k A! to give an
explicit construction for the linear minimal graded free resolution of the residue
field of a graded Koszul algebra; see Definition 2.5. In this paper, we extend
this construction to a family of acyclic complexes that yields highly structured
resolutions of the powers of the homogeneous maximal ideal over standard graded
Koszul algebras; see Definition 4.1. Since these complexes are typically not
minimal, we also seek to determine their minimal counterparts. To achieve this, we
take inspiration from results that describe structured resolutions over a polynomial
ring S constructed starting from the Koszul complex (exterior algebra) and its
generalizations; see [3]. We provide analogs of these results using any pair of Koszul
dual algebras, A and A! , instead of S and .
Our main result generalizes the canonical resolutions for the powers of the homo-
geneous maximal ideal constructed over a polynomial ring S by Buchsbaum and
Eisenbud in [3] to obtain minimal free resolutions for powers of the homogeneous
maximal ideal of a graded Koszul algebra. In contrast to the situation over S, these
are in general infinite resolutions. This allows us to obtain an explicit formula for
by βi,j (ma ) =
the graded Betti numbers defined a dimk Tori (ma , k)j and the graded
Poincaré series Pma (y, z) = i,j 0 βi,j (m )y zi . The following is a combination
A j
of Theorem 5.3 and Corollary 5.5.

Theorem If A is a graded Koszul algebra with homogeneous maximal ideal m, the
complexes

∂n ∂n−1 ∂1 εa
LA
a : · · · → Ln,a −
A
→ LA
n−1,a −−→ . . . −
→ LA
0,a −
→ ma → 0,
Canonical Resolutions over Koszul Algebras 283
defined in Eq. (5.3) with the augmentation map εa defined in Eq. (5.4) are minimal
free resolutions of the powers ma with a 1.
The nonzero graded Betti numbers of the powers of m are given by

a
A
βn,n+a (ma ) = (−1)i+1 dimk ((A! )∗n+i ) dimk (Aa−i ).
i=1
and the graded Poincaré series is

−a
a (y, z) = −(−z)
A
Pm HA! ∗ (yz)HA/ma (−yz).
In particular, the minimal graded resolution of ma is a-linear.

The Betti numbers presented in the theorem are recovered in a more restricted
setting in the recent paper [25] investigating resolutions of monomial ideals over
strongly Koszul algebras via different techniques.
The paper is structured as follows. In Sect. 2, we provide background on Koszul
algebras and the Priddy complex. In Sect. 3, we explain how to obtain a free solution
of a module M from a resolution of the ring over its enveloping algebra. In Sect. 4,
we rewrite this resolution as the totalization of a double complex, in the case that
the module is a power of the homogeneous maximal ideal and the ring is a Koszul
algebra. In Sect. 5, we give the minimal resolution and Betti numbers for the powers
of the homogeneous maximal ideal over a Koszul algebra. In Sect. 6 we apply our
construction to several specific Koszul algebras A to obtain explicit formulas for the
Betti numbers of ma .
2 Koszul Algebras and the Priddy Resolution
Throughout k is a field and A is a graded k-algebra having finite-dimensional graded

components with Ai = 0 for i < 0 and A0 = k. We further assume that A is
standard graded, that is, A is generated by A1 as an algebra over A0 = k.
Definition 2.1 ([23, Chapter 2]) We say that A is Koszul if k = A/A>0 admits a
linear graded free resolution over A, i.e., a graded free resolution P• in which Pi is
generated in degree i.

Classes of graded Koszul algebras arise from: quadratic complete intersections
[24], quotients of a polynomial ring by quadratic monomial ideals [12], quotients
of a polynomial ring by homogeneous ideals which have a quadratic Gröbner basis,
and from Koszul filtrations [9]; see, for example, the survey paper by Conca [7].
We now focus on quadratic algebras in order to define Koszul duality.
Definition 2.2 ([22, Chapter 1, Section 2]) Let A be a standard graded k-algebra.
We say that A is quadratic if A = T (V )/Q, where V is a k-vector space, T (V ) is
the tensor algebra of V , and Q is a quadratic ideal of T (V ).
284 E. Faber et al.
If A is a quadratic algebra, its quadratic dual algebra is defined by
T (V ∗ )
A! =
Q⊥
where V ∗ = Homk (V , k) and Q⊥ is the quadratic ideal generated by the orthogonal

complement to Q2 in T (V ∗ )2 = V ∗ ⊗k V ∗ with respect to the natural pairing
between V ⊗ V and V ∗ ⊗ V ∗ given by
v1 ⊗ v2 , v1∗ ⊗ v2∗ = v1 , v1∗ v2 , v2∗ .
Choosing dual bases x1 , . . . , xd and x1∗ , . . . , xd∗ for V and V ∗ respectively yields
that T (V ) = kx1 , . . . , xd and T (V ∗ ) = kx1∗ , . . . , xd∗ are polynomial rings in
noncommuting variables of degrees |xi | = 1 and |xi∗ | = −1. This allows one to
compute Q⊥ given a quadratic ideal Q ⊆ T (V ) using linear algebra, as described
for example in [18, Section 8].

Graded Koszul algebras are quadratic (see, for example, [22, Chapter 2, Defini-
tion 1]) and the duality of quadratic algebras restricts well to the class of Koszul
algebras since A and A! are Koszul simultaneously [22, Chapter 2, Corollary 3.2 ].
Moreover, (A! )! = A.
Example 2.3 The main example of Koszul dual algebras is given by the symmetric
algebra on a vector space V
kx1 , . . . , xd
S = k[x1 , . . . , xd ] =
(xi xj − xj xi , 1 i < j d)
and the exterior algebra on V ∗
kx1∗ , . . . , xd∗
S! = = .
((xi∗ )2 , xi∗ xj∗ + xj∗ xi∗ , 1 i j d)
Example 2.4 For the following commutative Koszul algebra
k[x, y, z] kx, y, z
A= = 2 ,
2 2
(x , xy, y ) (x , xy, y , xz − zx, xy − yx, yz − zy)
2
the dual algebra is given by
kx ∗ , y ∗ , z∗
A! = .
((z∗ )2 , x ∗ z∗ + z∗ x ∗ , y ∗ z∗ + z∗ y ∗ )
This pair of algebras are further discussed in Example 6.1.

Definition 2.5 The Priddy complex [23] of a quadratic algebra A is the complex
P•A whose i-th term is given by
∗
PiA = A ⊗k (A! )i ,

and the differential is defined by right multiplication by the trace element di=0 xi ⊗
xi∗ , where multiplication by xi∗ ∈ A! on (A! )∗ is defined as the dual of multiplication
by xi∗ on A! .

2.6 The importance of the Priddy complex lies in the fact that P•A is acyclic if and
only if A is Koszul; see [22, Chapter 2, Corollary 3.2]. Moreover, when A is Koszul
the Priddy complex, also called the generalized Koszul resolution, is a minimal free
resolution of the residue field k of A. This will be the base case in the proof that our
construction in Sect. 4 is a resolution.

2.7 Duality of Koszul algebras extends to an equivalence of derived categories that
goes back to [5] and was developed further in [6]. Let T = A⊗k A! which is an A-A! -
bimodule. For complexes N• of A! -modules and M• of A-modules define functors
L(N• ) = T ⊗A! N• ∼ = A ⊗k N• and R(M• ) = HomA (T , M• ) ∼ = Homk (A! , M• ) ∼=
! ∗
(A ) ⊗k M• . It is shown in [6, Theorem 2.12.1] that these functors induce an
equivalence of categories
L : D ↑ (A! ) D ↓ (A) : R
where D ↑ (A! ) stands for the derived category of complexes N• of graded A! -

modules with Ni,j = 0 for i " 0 or i + j 0 0 and D ↓ (A) is the derived category
of complexes M• of graded A-modules with Mi,j = 0 for i 0 0 or i + j " 0.
3 Resolutions via the Enveloping Algebra
Let A be a (not necessarily commutative) k-algebra where k is a field. In this section,

we review how one obtains a free resolution of any A-module M from a resolution
of A over its enveloping algebra. In general, one obtains a resolution that is far from
minimal. We remedy this in Sects. 4 and 5 over Koszul algebras A for the modules
ma (and hence A/ma ).
3.1 Given a k-algebra A, its enveloping algebra is given by Ae = A ⊗k Aop . A left
Ae -module structure is equivalent to an A-A-bimodule structure via
(a ⊗ b) · m = a · m · b (3.1)
We consider A as an Ae -module via the action in (3.1), where a, b, m ∈ A and

a · m · b represents internal multiplication in A.
286 E. Faber et al.
Consider a graded free resolution1 of A over Ae and note that any free left Ae -
module F can be rewritten as
F = Ae ⊗k V = A ⊗k Aop ⊗k V ∼
= A ⊗k V ⊗k A
for some vector space V , where the rightmost expression is thought of as an A-A-
bimodule via the outside two factors. Thus the resolution will be of the form
ε
· · · → A ⊗k V2 ⊗k A → A ⊗k V1 ⊗k A → A ⊗k A −
→ A → 0, (3.2)
where the augmentation ε from A ⊗k A to A is given by multiplication across the

tensor.
We observe that A ⊗k k ⊗k A ∼= A ⊗k A. Thus setting V0 = k, we may write the
resolution as a quasi-isomorphism of Ae -modules

A ⊗k V• ⊗k A −
→A
Next we show how to construct an A-free resolution for arbitrary A-modules M

using (3.2). This is well known; we include it because the construction is the basis
of our next step in Sect. 4. In the case of Koszul algebras, it can also be seen using
Koszul duality, and, more generally, it follows when the resolution of A comes from
an acyclic twisting cochain; see the remarks following the proof.

Proposition 3.2 If M is a graded A-module and A ⊗k V• ⊗k A − → A is a graded

Ae -free resolution of A, then the induced map A ⊗k V• ⊗k M − → M is a graded
A-free resolution of M where the A-module structure on the latter tensor product is
via the first factor.

Proof First note that both A ⊗k V• ⊗k A and A are A-A-bimodules in the obvious
ways. Furthermore, the complex A ⊗k V• ⊗k A (considered as an A-module via its
righthand factor) and the trivial complex A both consist of free A-modules (although
the latter is not a free Ae -module). Therefore the quasi-isomorphism A ⊗k V• ⊗k

A −→ A is actually a homotopy equivalence of A-modules. Hence it remains a
quasi-isomorphism after tensoring over A on the right with arbitrary A-modules. To
see this, note that the augmented complex of free A-modules (3.2) is contractible,
that is, homotopy equivalent to 0 (equivalently, it is split exact over A). But this
complex is the mapping cone of the chain map A ⊗k V• ⊗k A → A.
Therefore, upon tensoring (3.2) on the right over A with a left A-module M, one
obtains a quasi-isomorphism of left A-modules

(A ⊗k V• ⊗k A) ⊗A M −
→ A ⊗A M
1 One can always take the bar resolution of A over its enveloping algebra, but that is usually far
from minimal.

A ⊗k V• ⊗k M −
→M
As the original resolution of A was a map of A-A-bimodules, this one is still a map
of A-modules (via the lefthand factor of the tensor product). Viewing V• ⊗k M as a
(rather large) k-vector space, one sees that the complex on the left consists of free
A-modules, giving a free A-resolution of M.
Remark 3.3 Under the assumption that A is a Koszul algebra, we include here an
alternate proof of Proposition 3.2 using Koszul duality. Using the functors from the
equivalence described in 2.7, for a graded A-module M, one gets that L(R(M))
M. On the other hand, one computes that R(M) is the complex
0 → (A! )∗ ⊗k M0 → (A! )∗ ⊗k M1 → · · · → (A! )∗ ⊗k Mi → · · ·
Furthermore, it is clear from the definition that L((A! )∗ ) is simply the Priddy
complex P•A . Applying this to the complex above termwise and totalizing gives
that L(R(M)) equals the totalization of
0 → P•A ⊗k M0 → P•A ⊗k M1 → · · · → P•A ⊗k Mi → · · ·
which is exactly the complex described in Proposition 3.2. In the case that M =
A/ma , one gets the double complex XA
a described in Corollary 4.5.

Remark 3.4 More generally, we now briefly describe this from the perspective of
acyclic twisting cochains. Although these go far back, for recent quite general
versions of the duality they afford, modeled on that of Dwyer, Greenlees, and
Iyengar in [10] and generalizing Koszul duality, and for descriptions of how it
specializes to the situation of Koszul algebras, as well as the terms used below,
see Avramov’s paper [1], especially Theorem 4.7.
Let A be an augmented dg (differential graded) algebra. When there is an
augmented dg coalgebra C with a map τ : C → A of degree −1 that is a twisting
cochain, that is, a Maurer-Cartan equation
∂A τ + τ ∂C + μ(τ ⊗ τ ) = 0,
holds, where : C → C ⊗ C is the diagonal map and μ is the multiplication map,

then one can form tensor products whose differential is “twisted” by τ yielding a
natural map A ⊗ τ Cτ ⊗ A → A. If this is a quasi-isomorphism, then τ is called
acyclic, in which case the induced map A ⊗ τ Cτ ⊗ M → M is a quasi-isomorphism
for all dg A-modules M and a duality generalizing Koszul duality holds. An example
is given by the bar construction C = BA with the canonical map τ : BA → A, but
in the case of Koszul algebras one can get by with a much smaller complex using
Priddy’s construction.

Remark 3.5 Suppose A is local (or standard graded) k-algebra with (homogeneous)
maximal ideal m. The resolutions obtained in Proposition 3.2 are in general not
288 E. Faber et al.
minimal (respectively, minimal graded) resolutions even when one starts with a
minimal (respectively, minimal graded) resolution of A over Ae . For example, for
the explicit resolution XA
a given in Corollary 4.5, although ∂ is minimal, ∂ is
clearly not.

4 The Case of Koszul Algebras
In this section, under the further assumption that A is a Koszul k-algebra, we write
the A-free resolution of A/ma obtained in Remark 3.3 as the totalization of a certain
double complex.
First we recall the minimal graded resolution of A over Ae following the
presentation in [26, Section 3]. It is a symmetrization of the resolution of k over
A found by Priddy in [23], which is presented in Definition 2.5.
Definition 4.1 Let A be a Koszul k-algebra with dual Koszul algebra A! . Let
(A! )∗ = Homk (A! , k); thus (A! )∗ is an A! -module where the action of A! on (A! )∗
is the dual of the action of A! on itself. Define free A-modules
Fn = A ⊗k (A! )∗n ⊗k A
and differentials
∂n = (∂ )n + (−1)n (∂ )n
where

d
∂ = right multiplication by xi ⊗ xi∗ ⊗ 1
i=0
(which will form our vertical maps) and

d
∂ = left multiplication by 1 ⊗ xi∗ ⊗ xi
i=0
(which will form our horizontal maps).2 Then the complex
∂n ε
FA : · · · → Fn −
→ Fn−1 → · · · F0 −
→A→0 (4.1)
2 There is a misprint in [26] with regards to this map.

Fig. 1 The minimal resolution of a Koszul algebra A over Ae
augmented by the multiplication map ε from F0 = A ⊗k k ⊗k A = ∼ A ⊗k A to A is

the minimal graded free resolution of A over Ae [26, Proposition 3.1]. The complex
is FA is the totalization of the double complex with differentials ∂ and ∂ depicted
in Fig. 1. In particular, Fn is the sum of the modules on the n-th antidiagonal of this
double complex.

Remark 4.2 Here is the explicit connection with Priddy’s resolution: tensoring (4.1)
on the right over A with k gives Priddy’s minimal resolution of k as a left A-module
in Definition 2.5, also called the generalized Koszul resolution. Tensoring (4.1) on
the left gives the minimal resolution of k as a right A-module.

4.3 Considering the graded strands of (4.1), one can write this complex as a
totalization of an anticommutative double complex, which we also call FA , of free
A-modules given by the free A-modules
Fij = A ⊗k (A! )∗i ⊗k Aj (4.2)
where we are using the first tensor factor as “coefficients” and the maps ∂ and ±∂
of Definition 4.1 become the vertical and horizontal maps, respectively in Fig. 1.
Note that i is the homological degree in the complex FA .

4.4 One can interpret the complex FA in the language of the functors introduced
in 2.7 as FA = L(R(A)), where A is viewed as a complex concentrated in
homological degree 0. The discussion in 4.1 shows there is a quasi-isomorphism
L(R(A)) A. This was previously shown in [6, Thm 2.12.1] and is a particular
case of Remark 3.3.

290 E. Faber et al.
Next we apply the discussion from Sect. 3 to the A-module A/ma and arrange its
resolution into a double complex similarly to the one shown in Fig. 1 above.
Corollary 4.5 Totalization of the truncation of the double complex (4.2) obtained
by removing the columns with index j a 1 gives a graded A-free resolution
! ∗
a = A ⊗k (A ) ⊗k Aa−1 −
XA → A/ma . (4.3)
Proof Applying Proposition 3.2 by tensoring the resolution of A over Ae on the

right over A with A/ma gives a graded A-free resolution
! ∗
XA
a = A ⊗k (A ) ⊗k A/m −
a
→ A/ma
By means of the k-vector space identification
A/ma = Aa−1
the resolution becomes

! ∗
a = A ⊗k (A ) ⊗k Aa−1 −
XA → A/ma
Viewing graded strands, one can write this as a totalization of an anticommutative

double complex of free A-modules given by the terms Fij = A ⊗k (A! )∗i ⊗k Aj with
i 0, 0 j a − 1 of (4.2) with the differentials inherited from those described
in Definition 4.1.

We display the diagram for the double complex that yields the A-free graded
resolution XA a
a of A/m in Corollary 4.5 in Fig. 2. In the language of 2.7 this
resolution can be described as XA
a = L(R(A/m )).
a
Remark 4.6 The resolution XA a is minimal for a = 1 in which case Xa recovers the
A
Priddy complex without its first term. However for a 2 this resolution is typically
non minimal as the rows are split acyclic; see 5.1. The goal of Sect. 5 is to produce
a minimal free resolution for A/ma using XA a.

5 Minimal Resolutions for Powers of the Maximal Ideal
We introduce complexes LA a inspired by work of Buchsbaum and Eisenbud [3].

These will turn out to be the minimal resolutions for the powers of the homogeneous
maximal ideal of a graded Koszul algebra.
Fig. 2 The A-free graded resolution XA

a of A/m
a
5.1 We define free A-modules analogous to the Schur modules used by Buchsbaum
and Eisenbud in their resolutions (the case where A is a polynomial ring) in [3]. First
note that the rows of the double complex (4.2) except the bottom one are exact; in
fact, they can be viewed as the result of applying the exact base change A ⊗k −
to the strands of the dual Priddy complex, all of which are exact except the one
whose homology is k (in the case where A is a polynomial ring, it is applied to the
strands of the tautological Koszul complex; see [19, Section 1.4]). These complexes
are contractible, as they consist of free A-modules, and so all kernels, images, and
cokernels of the differentials are free as well.
Define for a > 0 the following free A-modules

!∗ (−1)n+1 ∂ !∗
LA
n,a = im A ⊗k A n+1 ⊗k Aa−1 −−−−−−→ A ⊗k A n ⊗k Aa (5.1)

∗ (−1)n ∂ ∗
= ker A ⊗k A! n ⊗k Aa −−−−→ A ⊗k A! n−1 ⊗k Aa+1 . (5.2)
The vertical differentials ∂ in Fig. 2 induce maps on these modules, which we

again denote by ∂ , to yield a complex

∂n ∂n−1 ∂1
a : · · · → Ln,a −
LA A
→ LA
n−1,a −−→ . . . −
→ LA
0,a . (5.3)
This complex is minimal in the sense that ∂ (LA n,a ) ⊆ mLn−1,a for all n since the
A
same property holds for the columns of the complex in Fig. 1 viewed as complexes
with differential ∂ . The construction of the complex LAa is depicted in Fig. 3.
Lemma 5.2 The complex LA

a can be augmented by the evaluation map
292 E. Faber et al.
Fig. 3 The construction of the complex LA

a
! ∗
εa : LA
0,a = A ⊗k A 0 ⊗k Aa → m
a
(5.4)
which is the restriction of the multiplication map

∗
ε : A ⊗k A! 0 ⊗k A → A sending r ⊗ v ⊗ s → rvs.
Proof As stated in Definition 4.1, ε is an augmentation map FA • → A. We verify

explicitly that ε satisfies the required property ε ◦ (∂ − ∂ ) = 0 below:
ε ◦ (∂ − ∂ )(r ⊗ v ⊗ s) = ε(rv ⊗ 1 ⊗ s − r ⊗ 1 ⊗ vs) = rvs − rvs = 0.
Since ∂ |LA = 0 it follows from the computation above that ε◦∂ (LA
1,a ) = 0, hence
1,a
the complex LA
a can be augmented to

∂n ∂n−1 ∂1 εa
· · · → LA
n,a −
→ LA
n−1,a −−→ . . . −
→ LA
0,a −
→ ma → 0.

The following is the main result of our paper. The case when A is an exterior
algebra has appeared previously in [11, Corollary 5.3]. The proof therein uses the
self-injectivity of the exterior algebra in a crucial manner, and therefore does not
seem to extend to all Koszul algebras.
Theorem 5.3 If A is a Koszul algebra, the complexes LA a defined in Eq. (5.3) with
the augmentation map εa defined in (5.4) are minimal free resolutions for the powers
ma of the maximal ideal with a 1.

Proof The complexes LA a are minimal by the discussion in 5.1. The proof of the
remaining claims is by induction on a 1.
The definition of Ln,1 in (5.1) shows that there are isomorphisms
∼ !∗ ∼ !∗
n,1 = A ⊗k A n+1 ⊗k k = A ⊗k A n+1 ,
LA
since the map ∂ is injective on FA

n+1,0 for n 0, the leftmost column of the double
complex in Fig. 1 (this column considered by itself is in fact XA 1 ). Therefore there
is an isomorphism of complexes LA ∼
= A ) [−1], where (XA )
1 (X 1 1 1 1 denotes the
truncation of the complex X1 by removing the homological degree 0 component.
A
Since XA
1 = P• is just the Priddy complex (upon noting that ∂ : (X1 )1 → (X1 )0
A A A
∗
agrees with ε under the identification (X1 )0 = A ⊗k A 0 ⊗k A0 ∼
A !
= A), we see
ε ε
that (X1 )1 −
A → m is a resolution of m by 2.6 and the base case that LA 1 −
→ m is a
minimal resolution of m follows.
For arbitrary a 2, (5.1) gives a short exact sequence of complexes
a−1 → P• ⊗k Aa−1 → La [−1] → 0,

0 → LA A A
∗
where PnA ⊗k Aa−1 = A ⊗k A! n ⊗k Aa−1 is the (a − 1)-st column of the double
complex in Fig. 2. The notation signifies that this column can be viewed as the
Priddy complex P•A tensored with Aa−1 . From the long exact sequence in homology
induced by the short exact sequence of complexes displayed above we deduce

0 i1
a)=
Hi (LA !
ker H0 (LA
a−1 ) → H0 (P•A ⊗k Aa−1 ) i = 0.
a ) = m . Indeed, the induced map in homology

It remains to show that H0 (LA a
∼
a−1 ) → H0 (P• ⊗k Aa−1 ) = H0 (P• ) ⊗k Aa−1
H0 (LA A A
can be recovered as the bottom map in the following commutative diagram
LA
0,a−1 A ⊗k (A! )∗0 ⊗k Aa−1
εa−1 ε0 ⊗idAa−1

ma−1 / k ⊗k Aa−1 ∼
= ma−1 /ma .
294 E. Faber et al.
Commutativity of the diagram yields that for r ⊗ s ⊗ v ∈ LA

0,a the induced map in
homology is given by
εa (r ⊗ s ⊗ v) = rsv → rsv,
where r is the coset of r in k = A/m. Thus we obtain the desired identification

8 9
ker H0 (LA
a−1 ) → H0 (PA
• ⊗ k Aa−1 )
= εa (Span{r ⊗ s ⊗ v | r ∈ m, s ∈ k, v ∈ Aa−1 })
∼
= ma .

Remark 5.4 Recall that rows of the double complex (4.2) except the bottom one
are exact; in fact, they can be viewed as the result of applying a base change to
the strands of the dual Priddy complex; see 5.1. These complexes are contractible,
as they consist of free R-modules. Hence for n 1 the (n + a)-th row of Fig. 2,
counting from the bottom (as the 0th row), is quasi-isomorphic to LA n,a and the
lower rows (numbered 1 through a) are split exact. The acyclic assembly lemma

[27, Lemma 2.7.3] yields quasi-isomorphisms (XA a )1 [−1] −
→ LA a for a 1. As

shown in Corollary 4.5 there are quasi-isomorphisms XA a − → A/ma , hence also

(XAa )1 −→ ma . Transitivity yields a quasi-isomorphism LAa −→ ma .
This approach gives an alternate proof for our main result, but it only determines
the augmentation map up to an isomorphism on its target. We prefer the more
explicit approach of Theorem 5.3, which specifies the augmentation map εa .

The following corollary of Theorem 5.3 gives an explicit formula for the Betti
numbers of powers of the maximal ideal of a Koszul algebra. That ma has an
a-linear minimal free resolution also follows from [8, Theorem 3.2]. Once the
linearity of this resolution has been established, [22, Chapter 2, Corollary 3.2
(iiiM)] gives an alternate interpretation for the Betti numbers of ma in terms of
the graded components of a quadratic dual module for the A-module ma . However
this description seems less amenable to explicit computations than our methods.
The next result utilizes the Hilbert series

a−1
H(A! )∗ (t) = dimk (A! )∗
· t
and HA/ma (t) = dimk Aj · t j .

0 j =0
Corollary 5.5 If (A, m) is a Koszul algebra, the nonzero graded Betti numbers of
the powers of m are given by

a
A
βn,n+a (ma ) = (−1)i+1 dimk ((A! )∗n+i ) dimk (Aa−i ).
i=1
In particular, the minimal graded resolution of ma is a-linear and its graded

Poincaré series is
−a
a (y, z) = −(−z)
A
Pm H(A! )∗ (yz)HA/ma (−yz).
Consequently, the nonzero Betti numbers of A/ma are given by

⎧
⎨a (−1)i+1 dim ((A! )∗ n > 0, j = n + a − 1
i=1 k n+i−1 ) dimk (Aa−i )
A
βn,j (A/ma ) =
⎩1 n = j = 0.
Proof The fact that minimal free resolution of ma is a-linear follows from the
Theorem 5.3 and the description of the differential ∂ of the complex (5.3) in view
of the fact that there is a splitting of the map ∂ to each Ln,a identifying a basis of
it with part of a basis of the last column of XA a . Consider the rows of the truncated
complex XA a when augmented to the relevant L n,a as follows.
0 −→ A ⊗k (A! )∗n+a ⊗k A0 −→ · · · −→ A ⊗k (A! )∗n+1 ⊗k Aa−1 −→ LA

n,a −→ 0
The exactness of this complex, as explained in Remark 5.4, yields the identities

a 8 9
A
βn,n+a (ma ) = rankA (LA
n,a ) = (−1)i+1 rankA A ⊗k (A! )∗n+i ⊗k Aa−i
i=1

a
= (−1)i+1 dimk (A! )∗n+i dimk Aa−i
i=1
and the vanishing of the remaining Betti numbers is due to the fact that the minimal
resolution in Theorem 5.3 is a-linear.
Lastly, the coefficients of the series
⎛ ⎞
−H(A! )∗ (yz)HA/ma (−yz) ⎜ ⎟ n n+a
= ⎜ (−1)j −a dimk (A! )∗
dimk Aj ⎟
(−z)a ⎝ ⎠z y ,
n0 0j a−1
j +
=n+a
A
agree with the preceding expression for βn,n+a (ma ) by setting j = a − i,
= n + i.

In contrast to Theorem 5.3, for non Koszul algebras the A-free resolution of
A/ma afforded by Corollary 4.5 cannot be minimized by the procedure presented in
this section. We illustrate the obstructions by means of the following example.
Example 5.6 Let A = k[x]/(x 3 ), which is a non Koszul (also non quadratic)
algebra. The enveloping algebra is
296 E. Faber et al.
Ae = A ⊗k A = k[x]/(x 3 ) ⊗k k[y]/(y 3 ) ∼
= k[x, y]/(x 3 , y 3 )
and the Ae -module structure induced on A by the (surjective) multiplication map

→ A yields the isomorphism A ∼
ε
Ae = A ⊗k A − = Ae /(x − y). Therefore A has the
following two-periodic resolution over the complete intersection Ae
x−y x 2 +xy+y 2 x−y ε

· · · → Ae −−→ Ae −−−−−−→ Ae −−→ Ae −
→ A → 0.
Rewriting this complex in the form of Sect. 3 gives
∂ ∂ ε
· · · → A ⊗k V2 ⊗k A −
→ A ⊗k V1 ⊗k A −
→ A ⊗k V0 ⊗k A −
→ A → 0,
where each Vi is a one dimensional vector space with basis {ei } and for i > 0

x ⊗ ei−1 ⊗ 1 − 1 ⊗ ei−1 ⊗ y for i odd
∂(1 ⊗ ei ⊗ 1) =
x2 ⊗ ei−1 ⊗ 1 + x ⊗ ei−1 ⊗ y + 1 ⊗ ei−1 ⊗ y2 for i even.
The conclusion of Corollary 4.5 still holds and indicates that the truncated
complexes XA a
a are (non minimal) free resolutions for A/m . But by contrast to the
Koszul case, we see that arranging by grading as in (4.2) yields a diagram that is
not a bicomplex and whose rows are no longer exact (or even complexes!), and so
in the truncated complex (4.3) the rows are no longer acyclic. Correspondingly, the
modules Ln,a one could define are no longer free. Thus there is no clear way to
minimize the complex XA a in a similar manner to the technique used in this section,
except for the case a = 1 where XA a is already minimal.

6 Examples
In this section we provide examples which illustrate our constructions for certain
Koszul algebras. For simplicity, all our examples are commutative algebras defined
by quadratic monomial ideals, but of course there are plenty of noncommutative
examples as well. This class is known to yield Koszul algebras by [12].
Example 6.1 Consider the following pair of dual Koszul algebras from Example 2.4
k[x, y, z] kx ∗ , y ∗ , z∗
A= and A! = .
(x 2 , xy, y 2 ) ((z∗ )2 , x ∗ z∗ + z∗ x ∗ , y ∗ z∗ + z∗ y ∗ )
The graded pieces (A! )−n are spanned by the words of length n on the alphabet
{x ∗ , y ∗ , z∗ } where the first letter is x ∗ , y ∗ or z∗ and the other n − 1 are x ∗ or y ∗ ,
whence dimk (A! )−n = 3 · 2n−1 for n 1. For n 1, An is spanned by monomials
of the form (z∗ )n , x ∗ (z∗ )n−1 , and y ∗ (z∗ )n−1 so that dimk A−n = 3. Thus in this
case both A and A! are infinite dimensional k-algebras.
The Priddy complex P•A (2.5) consists of terms of the form
P0 = A ⊗k k
n−1
Pn = A ⊗k k 3·2 for n 1
and the resolution of A over Ae viewed as a double complex (4.2) has terms
i−1
3·2i−1 A3·2 ⊗A A i 1, j = 0
Fi,j = A ⊗k k ⊗k Aj = i−1
A3·2 ⊗A A3 i 1, j 1.
Corollary 5.5 reveals that the Betti numbers of ma are independent of a. Indeed for
a 1 and n 0 we have

a−1
βn,n+a (ma ) = (−1)i+1 · 9 · 2n+i−1 + (−1)a+1 · 3 · 2n+a−1
i=1
1 − (−2)a−1
= 9 · 2n · + (−1)a−1 · 3 · 2a+n−1
3
= 3 · 2n .
Example 6.2 Consider the following commutative Koszul algebra
k[x, y, z] kx, y, z
A= = .
(xy, xz) (xy, xz, xz − zx, xy − yx, yz − zy)
The Koszul dual algebra is given by
kx ∗ , y ∗ , z∗
A! =
((x ∗ )2 , (y ∗ )2 , (z∗ )2 , y ∗ z∗ + z∗ y ∗ )
and its Hilbert function satisfies the Fibonacci recurrence
dim A!−n−2 = dim A!−n + dim A!−n−1 .
Indeed, setting u(n) to be the number of monomials in A! of degree −n ending in x

and v(n) to be the number of monomials in A! of degree −n not ending in x ∗ , yields
u(n) = v(n − 1) and v(n) = 2u(n − 1) + u(n − 2). The second expression follows
because the number of monomials ending in y ∗ or z∗ where the previous letter is x ∗
is 2u(n − 1) and the number of monomials ending in y ∗ z∗ (or equivalently, z∗ y ∗ )
where the previous letter is x is u(n − 2). Thus this leads to
298 E. Faber et al.
dim A!−n−2 = u(n + 2) + v(n + 2) = v(n + 1) + 2u(n + 1) + u(n)

= v(n + 1) + u(n + 1) + v(n) + u(n) = dim A!−n−1 + dim A!−n .
This shows that the Betti numbers of m are the Fibonnacci numbers starting with
β0A (m) = 3 and β1A (m) = 5.
The identity above in turn implies that the terms of the double complex as well
as the free modules in the resolution of ma satisfy similar recurrences
rank Fn+2,a = rank Fn+1,a + rank Fn,a , rank Ln+2,a = rank Ln+1,a + rank Ln,a .
We conclude that the Fibonacci recurrence holds for Betti numbers

A
βn+2,n+2+a (ma ) = βn+1,n+1+a
A
(ma ) + βn,n+a
A
(ma ) for a 1, n 0
subject, if a 2, to the initial conditions β0A (ma ) = a + 4 and β1A (ma ) = 2a + 4.

Solving the above recurrence yields closed formulas for the Betti numbers of ma
with a 2 as follows
√ n √ n
a + 4 3a + 4 1+ 5 a + 4 3a + 4 1− 5
A
βn,n+a (ma ) = + √ + − √ .
2 2 5 2 2 2 5 2
We now give an infinite resolution counterpart to a family of square-free

monomial ideals that have appeared as ideals of the polynomial ring in work of
Galetto [13].
Example 6.3 (See also [25, Example 3.18]) Consider the dual pair of Koszul
algebras
k[x1 , . . . , xd ] kx1∗ , . . . , xd∗

A= , and A! = ,
(x12 , . . . , xd2 ) (xi∗ xj∗ + xj∗ xi∗ , 1 i < j d)
d! i+d−1!
where dimk (Aj ) = j and dimk (A!−i ) = d−1 . Thus the terms in the double
complex (4.2) are
i+d−1 d
Fi,j = A ⊗k k ( d−1 ) ⊗k k (j ) .
Notice that for a d the ideal ma of A can be described as the ideal generated by
all square-free monomials of degree a in A, while for a > d we have ma = 0. We
compute the Betti numbers of this family of ideals using Corollary 5.5 as follows

a
n+i+d −1 d
βn,n+a (ma ) = (−1)i+1 (6.1)
d −1 a−i
i=1

n+i+d −1
Note that (−1)i+1 is equal to (−1)n−1 times the coefficient of t n+i
d −1
1 d
in the Taylor expansion of the rational function (1+t)d around 0. Similarly,
a−i
is the coefficient of t a−i in the binomial expansion of (1 + t)d . Since (1+t)
1
d · (1 +
t)d = 1, for n + a > 0, the coefficient of t n+a in their product is 0, i.e.

a
n+i+d −1 d
(−1) n+i
= 0.
d −1 a−i
i=−n
However, when i < a − d, the second binomial coefficient is 0, so this can be

restated as

a
n+i+d −1 d
(−1) n+i
= 0.
d −1 a−i
i=a−d
Combined with (6.1), the identity above leads to the more compact formula
! d !
0 i n+i+d−1
i=a−d (−1) 1ad
βn,n+a (m ) =
a d−1 a−i
0 a d + 1.
This is consistent with ma = 0 for a > d and can be easier to evaluate than (6.1)
for some values of a. For example, setting a = d yields

n+d −1
βn,n+d (md ) = .
d −1
Acknowledgments Our work started at the 2019 workshop “Women in Commutative Algebra”
hosted by Banff International Research Station. We thank the organizers of this workshop for
bringing our team together. We acknowledge the excellent working conditions provided by BIRS
and the support of the National Science Foundation for travel through grant DMS-1934391. We
thank the Association for Women in Mathematics for funding from grant NSF-HRD 1500481.
In addition, we have the following individual acknowledgements for support: Faber was
supported by the European Union’s Horizon 2020 research and innovation programme under the
Marie Skłodowska-Curie grant agreement No 789580. Miller was partially supported by the NSF
DMS-1003384. R.G.’s travel was partially supported by an AMS-Simons Travel Grant. Seceleanu
was partially supported by NSF DMS-1601024.
We thank Liana Şega for helpful comments and for bringing [8] to our attention and Ben Briggs
for answering a question and pointing us to [26].
Lastly, we thank the referee for a thorough reading and especially for guiding us to other points
of view, cf. Remarks 3.3 and 3.4, as well as for pointing out the papers [15] and [20].
300 E. Faber et al.
References
1. Luchezar L. Avramov, (Contravariant) Koszul duality for DG algebras, Algebras, quivers and
representations, Abel Symp., vol. 8, Springer, Heidelberg, 2013, pp. 13–58.
2. A. K. Bousfield, E. B. Curtis, D. M. Kan, D. G. Quillen, D. L. Rector, and J. W. Schlesinger,
The mod − p lower central series and the Adams spectral sequence, Topology 5 (1966), 331–
342.
3. David A. Buchsbaum and David Eisenbud, Generic free resolutions and a family of generically
perfect ideals, Advances in Math. 18 (1975), no. 3, 245–301.
4. I. N. Bernšteı̆n, I. M. Gel’fand, and S. I. Gel’fand, Algebraic vector bundles on Pn and problems
of linear algebra, Funktsional. Anal. i Prilozhen. 12 (1978), no. 3, 66–67.
5. A. A. Beı̆linson, V. A. Ginsburg, and V. V. Schechtman, Koszul duality, J. Geom. Phys. 5
(1988), no. 3, 317–350.
6. Alexander Beilinson, Victor Ginzburg, and Wolfgang Soergel, Koszul duality patterns in
representation theory, J. Amer. Math. Soc. 9 (1996), no. 2, 473–527.
7. Aldo Conca, Koszul algebras and their syzygies, Combinatorial algebraic geometry, Lecture
Notes in Math., vol. 2108, Springer, Cham, 2014, pp. 1–31.
8. Liana M. Şega, Homological properties of powers of the maximal ideal of a local ring, J.
Algebra 241 (2001), no. 2, 827–858.
9. Aldo Conca, Ngô Viêt Trung, and Giuseppe Valla, Koszul property for points in projective
spaces, Math. Scand. 89 (2001), no. 2, 201–216.
10. W. G. Dwyer, J. P. C. Greenlees, and S. Iyengar, Duality in algebra and topology, Adv. Math.
200 (2006), no. 2, 357–402.
11. David Eisenbud, Gunnar Fløystad, and Frank-Olaf Schreyer, Sheaf cohomology and free
resolutions over exterior algebras, Trans. Amer. Math. Soc. 355 (2003), no. 11, 4397–4426.
12. R. Fröberg, Koszul algebras, Advances in commutative ring theory (Fez, 1997), Lecture Notes
in Pure and Appl. Math., vol. 205, Dekker, New York, 1999, pp. 337–350.
13. Federico Galetto, On the ideal generated by all squarefree monomials of a given degree, J.
Commut. Algebra 12 (2020), no. 2, 199–215.
14. Mark Goresky, Robert Kottwitz, and Robert MacPherson, Equivariant cohomology, Koszul
duality, and the localization theorem, Invent. Math. 131 (1998), no. 1, 25–83.
15. Edward L. Green and Roberto Martínez Villa, Koszul and Yoneda algebras, Representation
theory of algebras (Cocoyoc, 1994), CMS Conf. Proc., vol. 18, Amer. Math. Soc., Providence,
RI, 1996, pp. 247–297.
16. Yu. I. Manin, Quantum groups and noncommutative geometry, Université de Montréal, Centre
de Recherches Mathématiques, Montreal, QC, 1988.
17. J. P. May, The cohomology of restricted Lie algebras and of Hopf algebras, J. Algebra 3 (1966),
123–146.
18. Jason McCullough and Irena Peeva, Infinite graded free resolutions, Commutative algebra and
noncommutative algebraic geometry. Vol. I, Math. Sci. Res. Inst. Publ., vol. 67, Cambridge
Univ. Press, New York, 2015, pp. 215–257.
19. Claudia Miller and Hamidreza Rahmati, Free resolutions of Artinian compressed algebras, J.
Algebra 497 (2018), 270–301.
20. Roberto Martínez-Villa and Dan Zacharia, Approximations with modules having linear resolu-
tions, J. Algebra 266 (2003), no. 2, 671–697.
21. Leonid Positselski, Galois cohomology of a number field is Koszul, J. Number Theory 145
(2014), 126–152.
22. Alexander Polishchuk and Leonid Positselski, Quadratic algebras, University Lecture Series,
vol. 37, American Mathematical Society, Providence, RI, 2005.
23. Stewart B. Priddy, Koszul resolutions, Trans. Amer. Math. Soc. 152 (1970), 39–60.
24. John Tate, Homology of Noetherian rings and local rings, Illinois J. Math. 1 (1957), 14–27.
25. K. VandeBogert, Iterated mapping cones for strongly Koszul algebras,
arXiv:2104.00037.
26. Michel Van den Bergh, Noncommutative homology of some three-dimensional quantum spaces,
Proceedings of Conference on Algebraic Geometry and Ring Theory in honor of Michael Artin,
Part III (Antwerp, 1992), vol. 8, 1994, pp. 213–230.
27. Charles A. Weibel, An introduction to homological algebra, Cambridge Studies in Advanced
Mathematics, vol. 38, Cambridge University Press, Cambridge, 1994.
Well Ordered Covers, Simplicial
Bouquets, and Subadditivity of Betti
Numbers of Square-Free Monomial
Ideals
Sara Faridi and Mayada Shahada
1 Introduction
This paper grew out of investigations into the subadditivity property of syzygies of
monomial ideals. For a homogeneous ideal I of a polynomial ring S, suppose the
maximum degree j such that βi,j (S/I ) = 0 is denoted by ti . The subadditivity
property is said to hold if we have ta+b ta + tb for all positive values of a and b.
While the subadditivity property or related inequalities are known to hold in
many special cases—certain cases for ideals of codimension 1 ([5, Corol-
lary 4.1]); some Koszul rings [1]; when I monomial ideal and a = 1 [15]; certain
homological degrees for Gorenstein algebras [6]; when a = 1, 2, 3 and I monomial
ideal generated in degree 2 [2, 11]; facet ideals of simplicial forests [9]—the
problem is open for monomial ideals and is known to fail (see Caviglia’s example
in [5, Example 4.5]) for general homogeneous ideals.
In the case of monomial ideals, Betti numbers can be calculated as ranks of
homology modules of topological objects. In particular, the order complex of the
lcm lattice of I (the poset of least common multiples of the minimal monomial
generating set of I ordered by division) can be used for this purpose. A nonvanishing
Betti number, in this context, corresponds by a result of Baclawski [3] to a
“complemented” lcm lattice (see Sect. 4).
S. Faridi ()
Department of Mathematics & Statistics, Dalhousie University, Halifax, NS, Canada
e-mail: faridi@dal.ca
M. Shahada
Department of Mathematics, College of Science, University of Bahrain, Sakheer, Kingdom of
Bahrain
e-mail: mshahada@uob.edu.bh
304 S. Faridi and M. Shahada
This approach was initiated by the first author in [9]: the existence of com-
plements which have nonvanishing Betti numbers in the “right” homological
degrees implies the subadditivity property (Question 4.2). This approach was further
pursued by both authors in [10], where Question 4.2 was translated into the context
of homological cycles in simplicial complexes breaking into smaller ones.
This paper takes yet a different angle. The idea of complements really comes
down to the following: take a monomial m in the lcm lattice of I . Among all the
complements of m in the lcm lattice, can you pick one, say m , which behaves
desirably? If m is a square-free monomial, this question is even simpler: m and m
correspond to subsets A and A of the variables {x1 , . . . , xn } such that A ∪ A =
{x1 , . . . , xn } and the product of the variables in A ∩ A is not in I . Can we consider
the subideals induced on A and A and extract properties from/for them?
The main object of study using this approach will be “well ordered covers” of
ideals (Definition 3.2). The existence of a well ordered cover of size i is known,
via the Lyubeznik resolution, to guarantee a nonvanishing i th Betti number [8]. In
this paper we investigate when complements in the lcm lattice produce subideals
with well ordered covers of sizes a and b with a + b = i, in order to result in the
subadditivity property.
Moreover, we introduce strongly disjoint sets of simplicial bouquets, which we
show are always well ordered covers, and we demonstrate how they can be broken
up to prove subadditivity in certain homological degrees. An advantage of simplicial
bouquets is that they are rather easy to spot in simplicial complexes, as they do not
rely on an ordering. Rather, one or more orderings are inherent in the definition (see
Theorem 5.4).
In Sect. 2 we set up the background on simplicial resolutions. Section 3 intro-
duces the reader to well ordered covers of monomials. Section 4 describes the
subadditivity property and contains one of the main results of the paper (Theo-
rem 4.7) which considers well ordered covers under complementation. In Sect. 5
we introduce simplicial bouquets and show that certain types of simplicial bouquets
are well ordered facet covers. We then apply the results of Sect. 4 to simplicial
complexes that contain strongly disjoint sets of bouquets, and show that the
subadditivity property holds in degrees that come from the sizes of the bouquets
(Theorem 5.7). Section 6 offers ways to optimize the order of monomials in a well
ordered cover to get the best possible subadditivity results.
With some numerical manipulation, the results of this paper can be adapted to
non-square-free monomial ideals via polarization [12], a method that transforms a
monomial ideal into a square-free one which retains many of the algebraic properties
of the original ideal, including the minimal free resolution.
Well Ordered Covers, Simplicial Bouquets, and Subadditivity of Betti Numbers. . . 305
2 Background
Throughout let S = k[x1 , . . . , xn ] be a polynomial ring over a field k.
2.1 Simplicial Complexes and Facet Ideals
A simplicial complex on a finite vertex set V () is a set of subsets of V ()

such that {v} ∈ for every v ∈ V () and if F ∈ , then for every G ⊆ F , we
have G ∈ . The elements of are called faces; the maximal faces with respect
to inclusion are called facets, and a simplicial complex contained in is called a
subcomplex of . The set of all facets in defines and is denoted by Facets().
If Facets() = {F1 , . . . , Fq }, then we write = F1 , . . . , Fq .
A subcollection of is a subcomplex of whose facets are also facets of . If
A ⊆ V (), then the induced subcollection [A] is the simplicial complex defined
as
[A] = F ∈ Facets() : F ⊆ A .
We say a facet F contains a free vertex v if F is the only facet of containing v.

Given a simplicial complex on vertices {x1 , . . . , xn }, we can define the facet
ideal of as
!
F() = xi : F is a facet of
xi ∈F
which is an ideal of S = k[x1 , . . . , xn ]. Conversely, given a square-free monomial

ideal I ⊂ S, the facet complex of I is the simplicial complex
6 7
F(I ) = F : xi is a generator of I .
xi ∈F
Example 2.1 For I = (xy, yz, zu), the simplicial complex F(I ) is below.
2.2 Simplicial Resolutions
Any monomial ideal I of S = k[x1 , . . . , xn ] admits a minimal graded free

resolution, which is an exact sequence of free S-modules
0 → ⊕j ∈N S(−j )βp,j → ⊕j ∈N S(−j )βp−1,j → · · · → ⊕j ∈N S(−j )β1,j → S.
For each i and j , the rank βi,j (S/I ) of the free S-modules appearing above are
called the graded Betti numbers of the S-module S/I , and the total Betti number
in homological degree i is

βi (S/I ) = βi,j (S/I ).
j
If I is generated by monomials, the graded Betti numbers can be further refined

into sums of multigraded Betti numbers. For a monomial m in S, the multigraded
Betti number of S/I is of the form βi,m (S/I ) and we have

βi,j (S/I ) = βi,m (S/I ) (2.1)
where the sum is taken over all monomials m of degree j that are least common
multiples of subsets of the minimal monomial generating set of I .
The multigraded Betti numbers βi,m (S/I ) are related to the combinatorics of the
ideal I . Given a monomial ideal I minimally generated by m1 , . . . , mq , one can
consider a simplicial complex on q vertices {v1 , . . . , vq }, where each vertex vi is
labeled with the monomial generator mi , and each face τ of is labeled with the
monomial
!
lcm(τ ) = lcm mi : vi ∈ τ .
We say that supports a free resolution of I if the simplicial chain complex

of can be “homogenized”, using the monomial labels of the faces, to produce
a free resolution of I . For details of homogenization of a chain complex see [20,
Section 55]. The resulting free resolution is called a simplicial resolution.
Taylor’s resolution ([21], see also [20, Construction 26.5]) is an example of a
simplicial resolution where the underlying simplicial complex is a full simplex T(I )
over the vertex set labeled with the monomial generators {m1 , . . . , mq } of I , called
the Taylor complex of I . It is known that all simplicial complexes supporting a
free resolution of I are subcomplexes of the Taylor complex, in other words: all
simplicial resolutions are contained in the Taylor resolution ([19]).
If is a simplicial complex supporting a free resolution of I , and m is a
monomial in S, the simplicial subcomplex <m is defined as
<m = {τ ∈ : lcm(τ ) strictly divides m}.

Example 2.2 Consider I = (xy, yz, zu) in k[x, y, z, u]. The Taylor complex T(I )
and a subcomplex T(I )<xyzu are shown in the following figures.
If supports a free resolution of a monomial ideal I , then for a fixed integer i,

the Betti number βi (S/I ) is bounded above by the number of (i − 1)-faces of .
Therefore, the more we “shrink” the supporting complex , the better we bound the
Betti numbers.
In particular, as stated by Bayer and Sturmfels [4], the multigraded Betti numbers
of I can be determined by the dimensions of reduced homologies of subcomplexes
of . The statement below is from Peeva’s textbook [20].
Theorem 2.3 ([20], Theorem 57.6) Let I be a proper monomial ideal of S which
is minimally generated by the monomials m1 , . . . , mq , and suppose that I has a
free resolution supported on a simplicial complex . For i > 0 and a monomial m
of positive degree, the multigraded Betti numbers of I are given by
?
dimk H̃i−2 (<m , k) if m | lcm(m1 , . . . , mq )
βi,m (S/I ) =
0 otherwise.
Example 2.4 For I = (xy, yz, zu) in Example 2.2, T(I )<xyzu is acyclic and hence
βi,xyzu (S/I ) = 0 for all i.
For a monomial ideal I of S which is minimally generated by the of monomials
set G = {m1 , . . . , mq }, the lcm lattice of I , denoted by LCM(I ), is the bounded
lattice whose elements are the least common of subsets of G ordered by divisibility.
The top element of LCM(I ) is 1̂ = lcm(m1 , . . . , mq ) and the bottom element is
0̂ = 1 regarded as the lcm of the empty set. The least common multiple of elements
in LCM(I ) is their join.
Example 2.5 The following is LCM(I ) for I = (xy, yz, zu) from Example 2.2.
3 (Well Ordered) Covers
Let be a simplicial complex. A set D ⊆ Facets() is called a facet cover of if

every vertex v of belongs to some facet F in D. A facet cover is called minimal
if no proper subset of it is a facet cover of . For instance, the simplicial complex
in Example 2.1 has the set {{x, y}, {z, u}} as a minimal facet cover.
We can translate a minimal facet cover of a simplicial complex to its facet ideal. If
I = (m1 , . . . , mq ) is a square-free monomial ideal and = F(I ) = F1 , . . . , Fq
so that each mi is the product of the vertices in Fi , then we say mi1 , . . . , mit is a
(minimal) cover of I to imply that Fi1 , . . . , Fit is a (minimal) facet cover of .
For the sake of simplicity, assume that every variable xi appears in at least one
of the generators of I . If M = {mi1 , . . . , mit } is a minimal cover of I , then one can
see that for every j ∈ {1, . . . , t}
ij , . . . , mit ) = lcm(mi1 , . . . , mit ) = x1 . . . xn .

lcm(mi1 , . . . , m
This implies that if T(I ) is the Taylor complex of I , then T(I )<x1 ...xn contains
the boundary of a (t − 1)-simplex, which by Theorem 2.3 means that we could
potentially have βt,x1 ...xn (S/I ) = 0.
In other words, the existence of a minimal cover M of length t indicates that
we might have a “top degree” Betti number βt,n . By ordering the generators of I
we can make M into a “well ordered” cover, which, using a Lyubeznik resolution
(a simplicial resolution that is based on ordering the generators of an ideal [18]),
guarantees the nonvanishing of βt,n (Theorem 3.4).
Definition 3.1 ([8], Definition 3.1) Let be a simplicial complex. A sequence of
F1 , . . . , Fs of facets of is called a well ordered facet cover if it is minimal facet
cover of , and for every facet F ∈ / {F1 , . . . , Fs } of there exists j s − 1 such
that
Fj ⊆ F ∪ Fj +1 ∪ · · · ∪ Fs .
The definition below is an equivalent version of Definition 3.1 stated for

monomial ideals.
Definition 3.2 (Well Ordered Cover) A sequence m1 , . . . , ms of generators of a
square-free monomial ideal I is called a well ordered cover of I if {m1 , . . . , ms }
is a minimal cover of I and for every generator m ∈
/ {m1 , . . . , ms } of I there exists
j s − 1 such that
mj | lcm(m , mj +1 , . . . , ms ).
Example 3.3 (1) For I = (abz, bcz, xyz, axz), abz, bcz, xyz is a well ordered
cover since abz divides lcm(axz, bcz, xyz).
(2) For I = (xy, yz, zu), {xy, zu} is a minimal cover that cannot be made into a
well ordered cover of I since xy lcm(yz, zu) and zu lcm(yz, xy).
Notice that I has no well ordered cover since {xy, zu} is the only possible
minimal cover and it is not well ordered.
A class of examples of well ordered facet covers is (simplicial) bouquets, which
will be discussed in Sect. 5.
As the following theorem shows, well ordered covers are facets in a Lyubeznik
complex, and hence ensure nonvanishing multigraded Betti numbers.
Theorem 3.4 ([8]) Let M : m1 , . . . , ms be a well ordered cover of a square-free
monomial ideal I . Then there is a total order < on the set of generators of I such
that M is a facet of the Lyubeznik simplicial complex and hence βs,m (S/I ) = 0
where m = lcm(m1 , . . . , ms ).
The converse of the above theorem holds in some cases, for example when I is
the facet ideal of a simplicial forest , every nonzero Betti number of I corresponds
to a well ordered facet cover of an induced subcollection of ([8]).
4 The Subadditivity Property
In this section, we will explore how we can use well ordered covers to consider
the subadditivity property for the maximal degrees of syzygies of square-free
monomial ideals. Let
ta = max{j : βa,j (S/I ) = 0}.
We say that I satisfies the subadditivity property if for all a, b > 0 with a + b is at
most the projective dimension of S/I ,
ta+b ta + tb .
In what follows, we will be moving back and forth between a square-free

monomial ideal I and its facet complex = F(I ). For a monomial m ∈ LCM(I )
where m = xi1 · · · xir , by I[m] we mean the facet ideal of the induced subcollection
[m] = [{xi1 ,...,xir }] or in other words
I[m] = F([m] ) = F([{xi1 ,...,xir }] ).
If we set and to be the Taylor complexes of I and I[m] , respectively, then

Theorem 2.3 indicates that
βi,m (S/I ) = βi,m (S/I[m] ) (4.1)
. For an integer i, (2.1) and (4.1) show that

as <m = <m
βi (S/I ) = 0 ⇐⇒ βi,m (S/I ) = 0 for some m ∈ LCM(I )

⇐⇒ βi (S/I[m] ) = 0 for some m ∈ LCM(I ).
In the rightmost statement in the previous line, the monomial m is at the “top” of the
lcm lattice of I[m] . This useful observation tells us that questions about multidegree
Betti numbers can be reduced to questions about “top degree” Betti numbers. In the
same vein, the subadditivity question can always be rephrased as a “top degree” one.
We state this version for square-free monomials below.
Question 4.1 (Top Degree Subadditivity) Suppose I = (m1 , . . . , mq ) is a
square-free monomial ideal in a polynomial ring S, and
lcm(m1 , . . . , mq ) = xi1 · · · xir .
Suppose βi,r (S/I ) = 0, and a, b > 0 are such that i = a + b. Then can we show
that ta + tb r = ta+b ?
In fact, the ambient ring does not really matter, so for the sake of simplicity we
can assume in Question 4.1 that r = n and xi1 · · · xir = x1 · · · xn . If I is a square-
free monomial ideal in S = k[x1 , . . . , xn ], then since there is only one monomial of
degree n in LCM(I ), from (2.1) we have
βi,n (S/I ) = βi,x1 ···xn (S/I ).
So from now on we will start from this setting.

For a square-free monomial ideal I in the variables x1 , . . . , xn , two monomials
m, m ∈ LCM(I ) are called lattice complements if lcm(m, m ) = x1 · · · xn and
gcd(m, m ) ∈/ I . As a potential way to examine the subadditivity property, the first
author raised the following question in [9] (see also [10]):
Question 4.2 (Betti Numbers of Lattice Complements [9], Question 1.1) If
an ideal I is generated by square-free monomials in variables x1 , . . . , xn , and
βi,n (S/I ) = 0, a, b > 0 and i = a + b, are there complements m and m in
LCM(I ) with βa,m (S/I ) = 0 and βb,m (S/I ) = 0?
The existence of lattice complements will establish the subadditivity property of
I simply since
ta + tb deg(m) + deg(m ) n = ta+b .
A well ordered cover seems to be a promising place to look for lattice comple-
ments as the following example shows.
Example 4.3 Let = F(I ) = xy, yz, xz, za, ab, bc.
According to Macaulay2 [13], S/I has the following Betti table:
01 2 34
total : 16 10 7 2
0: 1 . . . .
1: . 6 6 1 .
2: . . 4 62
Here β4,xyzabc (S/I ) = 0 and hence t4 = 6. It is easy to check that the set
M = {ab, xy, bc, xz} (highlighted above) is a well ordered cover. We consider the
following two cases:
(1) a = 1 and b = 3. Then, using the above Betti table t1 = 2 and t3 = 5. Here we
have t4 < t1 + t3 = 7. Define the monomials m, m ∈ LCM(I ) as lcm of some
subsets of M as follows:
m = ab = lcm(ab) and m = bcxyz = lcm(xy, bc, xz).
Then,
β1,ab (S/I ) = dimk H̃−1 ({φ}, k) = 0
and
β3,xyzbc (S/I ) = dimk H̃1 (xy, yz, xz, bc, k) = 0.
Note that lcm(m, m ) = xyzabc and gcd(m, m ) = b ∈

/ I . As a result, m
and m are lattice complements and
t1 + t3 = deg(m) + deg(m ) > 6 = t4 .
(2) a = b = 2. From the Betti table we have t2 = 4 and t4 < 2t2 = 8. As in the
above case, we take
m = abxy = lcm(ab, xy) and m = bcxz = lcm(bc, xz).

Here, we also have lcm(m, m ) = xyzabc and gcd(m, m ) = bx ∈

/ I . So m
and m are lattice complements, and we have
β2,xyab (S/I ) = dimk H̃0 (xy, ab, k) = 0
and
β2,xzbc (S/I ) = dimk H̃0 (xz, bc, k) = 0.
As a result,
2t2 = deg(m) + deg(m ) > 6 = t4 .
Proposition 4.4 Let M : m1 , . . . , ms be a well ordered cover of a square-free

monomial ideal I . Define for each 1 a s − 1 the monomials
m = lcm(m1 , . . . , ma ) and m = lcm(ma+1 , . . . , ms ).
Then m and m are lattice complements in LCM(I ).

Proof Suppose I = (m1 , . . . , ms , n1 , . . . , nk ) and
lcm(m1 , . . . , ms ) = lcm(m, m ) = x1 · · · xn .
Suppose that m and m are not lattice complements, so gcd(m, m ) ∈ I . In

particular, there is a generator q of I such that
q | lcm(m1 , . . . , ma ) and q | lcm(ma+1 , . . . , ms ).
Suppose that q = mi for some i = 1, . . . , s. If i a, then as mi divides

lcm(ma+1 , . . . , ms ) we must have
i , . . . , ma , ma+1 , . . . , ms ) = lcm(m1 , . . . , ms ),
lcm(m1 , . . . , m
which contradicts M being a minimal cover of I .

The case i a + 1 also leads to the same contradiction. Suppose that q = ni
for some i ∈ {1, . . . , k}. Since M is a well ordered cover of , then there exists
s − 1 such that
m
| lcm(ni , m
+1 , . . . , ms .)
If
a, then
+ 1 a + 1 and as ni | lcm(ma+1 , . . . , ms ) we have
m
| lcm(m
+1 , . . . , ma+1 , ma+2 , . . . , ms ).
Therefore,

, . . . , ms ) = lcm(m1 , . . . , ms ).
lcm(m1 , . . . , m
This also contradicts the minimality of the cover M. If

a + 1, then since
ni | lcm(m1 , . . . , ma ),
m
| lcm(ni , m
+1 , . . . , ms )
| lcm(m1 , . . . , ma , m
+1 , . . . , ms ).
Similarly,

, m
+1 , . . . , ms ) = lcm(m1 , . . . , ms ),
lcm(m1 , . . . , ma , . . . , m
a contradiction. Thus, gcd(m, m ) ∈

/ I and hence m and m are lattice complements
in I , as required.

In fact more is true: the second half of the well ordered cover in Proposition 4.4
is itself a well ordered cover.
Proposition 4.5 Let M : m1 , . . . , ms be a well ordered cover of a square-free
monomial ideal I . Define for each 1 a s − 1 the monomials
Then
(1) {m1 , . . . , ma } is a minimal cover of I[m] .
(2) ma+1 , . . . , ms is a well ordered cover of I[m ] .
(3) βs−a,m (S/I ) = 0.
Proof Clearly, {m1 , . . . , ma } and {ma+1 , . . . , ms } are covers of I[m] and I[m ]
respectively. Suppose that {m1 , . . . , ma } is not minimal, i.e. there exists a proper
subset
{mi1 , . . . , mik } ⊂ {m1 , . . . , ma },
that is a cover of I[m] . In particular, there exists h ∈ {1, . . . , a} \ {i1 , . . . , ik } such

that
mh | lcm(mi1 , · · · , mik ).
Then
h , · · · , ms ) = lcm(m1 , · · · , ms ).
lcm(m1 , · · · , m
This contradicts the minimality of the cover M. Thus, {m1 , . . . , ma } is a minimal

cover of I[m] . Using a similar argument, {ma+1 , . . . , ms } is also a minimal cover of
I[m ] .
In order to show that ma+1 , . . . , ms is a well ordered cover of I[m ] , if n is a
minimal generator of I[m ] such that n ∈/ {ma+1 , . . . , ms }, we need to find
in the
set {a + 1, . . . , s − 1} such that
m
| lcm(n, m
+1 , · · · , ms ).
By Proposition 4.4, m and m are lattice complements. As a result, n cannot be

a minimal generator of I[m] ; in particular, n ∈ / {m1 , . . . , ma }. Since M is a well
ordered cover and n ∈
/ {m1 , . . . , ms }, then there exists 1
s − 1 such that
m
| lcm(n, m
+1 , · · · , ms ).
Suppose that
a, so
+ 1 a + 1. Since n is a minimal generator of I[m ] ,
n | lcm(ma+1 , · · · , ms ). Therefore,
m
| lcm(n, m
+1 , · · · , ma+1 , · · · , ms )
| lcm(m
+1 , · · · , ms ).
Thus,

, · · · , ms ) = lcm(m1 , · · · , ms ),
lcm(m1 , · · · , m
contradicting the minimality of the cover M. Hence a + 1

s − 1 which
makes ma+1 , . . . , ms a well ordered cover of I[m ] . Using Theorem 3.4,
βs−a,m (S/I[m ] ) = 0, and hence βs−a,m (S/I ) = 0.

Example 4.6 If = F(I ) = xy, yz, xz, za, ab, bc, then ab, xy, bc, xz is a well
ordered cover.
Then, by Proposition 4.5 we have:
β1,xz (S/I ) = 0, β2,bcxz (S/I ) = 0, β3,bcxyz (S/I ) = 0, β4,abcxyz (S/I ) = 0.

Under certain extra conditions, the first part of the well ordered cover is also
well ordered, which gives us the subadditivity property in those cases. Note that
the subadditivity property in the case a = 1 below is also covered by a theorem of
Herzog and Srinivasan [15].
Theorem 4.7 (Well Ordered Covers and Subadditivity) Let M : m1 , . . . , ms
be a well ordered cover of a square-free monomial ideal I . Define for each a in
{1, . . . , s − 1} the monomials
Suppose either of the following conditions holds:

(1) I[m] = (m1 , . . . , ma ) (e.g. when a = 1); or
(2) gcd(m, m ) = 1.
Then m and m are lattice complements in I = F() such that βa,m (S/I ) = 0. In
particular
ts ta + ts−a .
Proof By Proposition 4.5 we already know that {m1 , . . . , ma } is a minimal cover

for I[m] . We claim that if either of the two conditions above hold, then it is a well
ordered cover. If I[m] = (m1 , . . . , ma ), then this is trivial, since there is no generator
other than m1 , . . . , ma . Suppose gcd(m, m ) = 1, and let n be in the minimal
monomial generating set of I[m] and n ∈ / {m1 , . . . , ma }. Then, n is a minimal
generator of I as well. Moreover m1 , . . . , ms is a well ordered cover of I , and
so for some j s − 1 we have mj | lcm(n, mj +1 , . . . , ms ). On the other hand,
since n | m and gcd(m, m ) = 1, we must have j a − 1 and
mj | lcm(n, mj +1 , . . . , ma )
which implies that m1 , . . . , ma is a well ordered cover of I[m] . Therefore, by

Theorem 3.4, βa,m (S/I[m] ) = 0, and hence βa,m (S/I ) = 0.
From Proposition 4.5 it follows that ma+1 , . . . , ms is a well ordered cover for
I[m ] and βs−a,m (S/I[m ] ) = 0, hence βs−a,m (S/I ) = 0.
Now, since m and m are complements (Proposition 4.4), we have that
ts ta + ts−a ,
which ends our argument.

5 Simplicial Bouquets
As proved in [8, Proposition 4.3], an example of a well ordered facet cover for
a simple graph is a strongly disjoint set of bouquets (see [22, Definition 1.7] and
also [17, Definitions 2.1 and 2.3]). Bouquets are in general much easier to identify
in graphs than well ordered edge covers, as one does not need to worry about
the order. In this section we define a simplicial counterpart for a strongly disjoint
set of bouquets of graphs, and show that they form well ordered facet covers and
hence guarantee nonvanishing Betti numbers. Similar to graph bouquets, simplicial
bouquets are easy to spot in a picture, and strongly disjoint sets of simplicial
bouquets often come with more than one guaranteed order. We then apply the results
of the previous section to examine the subadditivity property in the presence of such
simplicial bouquets. It must be noted that our definition of a simplicial bouquet is
very close to hypergraph bouquets developed in [16, Definition 3.1], and almost the
same as the hypergraph bouquets in [7, Definition 2.1].
Definition 5.1 (Simplicial Bouquet) Let be a simplicial complex. A simplicial
bouquet is a subcollection B = G1 , . . . , Gt of such that each facet Gi has at
least one free vertex in B, and
>
t
Gi = ∅.
i=1
The nonempty intersection of the facets of B is called the root of B and denoted by
Root(B).
A simplicial bouquet in a graph coincides with the usual definition of bouquets
in graphs (see [22, Definition 1.7]).
The distance between two distinct facets F, F of , denoted by dist (F, F ),
is the minimum length α of sequences of facets of
F = F0 , F 1 , . . . , F α = F where Fi−1 ∩ Fi = ∅,
or ∞ if there is no such sequence. We say that F and F are 3-disjoint in if

dist (F, F ) 3. A subset E ⊂ Facets() is said to be pairwise 3-disjoint if
every pair of distinct facets F, F ∈ E are 3-disjoint in (see [14, Definitions 2.2
and 6.3]).
Definition 5.2 ((Strongly Disjoint) Set of Bouquets) Let be a simplicial com-
plex. For a set B = {B1 , B2 , . . . , Bd } of simplicial bouquets of , define
Facets(B) = Facets(B1 ) ∪ · · · ∪ Facets(Bd ) and V (B) = V (B1 ) ∪ · · · ∪ V (Bd ).
Then B is called strongly disjoint in if:

(1) V (Bi ) ∩ V (Bj ) = ∅ for all i = j , and
(2) we can choose a facet Gi from each Bi ∈ B so that the set {G1 , . . . , Gd } is
pairwise 3-disjoint in .
We say that contains a strongly disjoint set of bouquets if there exists a
strongly disjoint set of bouquets B = {B1 , . . . , Bq } of such that:
(1) V () = V (B), and
(2) if F ∈ Facets() \ Facets(B) and F ∩ G = ∅ for some G ∈ Facets(Bi ) and
i ∈ {1, . . . , d}, then (G \ Root(Bi )) ⊆ F .

Example 5.3 Let I = (abc, bcd, cdf, def, eg, fg, gh, hi, gi, f i, gx, gy). It is
easy to check that = F(I ) contains a strongly disjoint set of bouquets {B1 , B2 }
where the bouquets
B1 = bcd, abc and B2 = gy, gx, ge, gf, gh, gi
are highlighted below, and the set of facets {abc, gx} is pairwise 3-disjoint.
The following statement is a generalization of [8, Proposition 4.3].

Theorem 5.4 (Strongly Disjoint Set of Bouquets and Well Ordered Facet
Covers) Let be a simplicial complex which contains a strongly disjoint set of
bouquets B = {B1 , . . . , Bd }. For each q ∈ {1, . . . , d} suppose
q q q
Facets(Bq ) = {G1 , G2 , . . . , Gbq , Gq }
where {G1 , . . . , Gd } is pairwise 3-disjoint in . Then for any permutation
k1 , k2 , . . . , k d
of the integers 1, . . . , d, the sequence of facets
Gk11 , . . . , Gkb1k , Gk1 , Gk12 , . . . , Gkb2k , Gk2 , . . . , Gk1d , . . . , Gkbdk , Gkd

F GH 1 I F GH 2 I F GH d I
Facets(Bk1 ) Facets(Bk2 ) Facets(Bkd )
form a well ordered facet cover of .

Proof Observe that we can identify each monomial generator x1 . . . xa of F()

with the facet {x1 , . . . , xa } of . The condition that V () = V (B) and that every
facet of B has a free vertex guarantees that Facets(B) is a minimal facet cover of
. Suppose F ∈ Facets() \ Facets(B). Since the set {Gk1 , . . . , Gkd } is pairwise
3-disjoint in , F can intersect at most one of {Gk1 , . . . , Gkd }, and so at least one
=
vertex of F does not belong to di=1 Gki . Therefore for some q ∈ {1, . . . , d} and
k
j ∈ {1, . . . , bkq } we have Gjq ∩ F = ∅ and so
k
F ⊇ Gjq \ Root(Bkq ).
Hence
k k
Gjq = (Gjq \ Root(Bkq )) ∪ Root(Bkq )
⊆ F ∪ Gkq
k k k
⊆ F ∪ Gjq+1 ∪ · · · ∪ Gbqk ∪ Gkq ∪ G1q+1 ∪ · · · ∪ Gkd
q
which implies that the sequence above is a well ordered facet cover of .

Equivalently, strongly disjoint simplicial bouquets produce well ordered covers
for facet ideals.
Example 5.5 For the ideal in Example 5.3, Theorem 5.4 states that both
bcd, abc, gy, ge, gf, gh, gi, gx and gy, ge, gf, gh, gi, gx, bcd, abc
are well ordered covers. Note that there are many other options for well ordered
covers, obtained by reordering all facets (except for the last one) of each bouquet,
or picking another set of pairwise 3-disjoint facets from the set of bouquets.
The following is a generalization of [17, Theorem 3.1]. From here onwards we
assume S is a polynomial ring over a field generated by variables which are vertices
of the simplicial complex .
Corollary 5.6 (Betti Numbers from Simplicial Bouquets) Let be a simplicial
complex, W ⊆ V () and suppose the induced subcollection [W ] contains strongly
disjoint set of bouquets B with | Facets(B)| = i and |W | = |V (B)| = j . Then
βi,j (S/F()) = 0.
Proof As in Sect. 4, βi,j (S/F()) = 0 if and only if βi,j (S/F([W ] )) = 0 for
some W ⊆ V () with |W | = j . The statement is now a direct consequence of
Theorems 3.4 and 5.4.

Theorem 5.7 (Subadditivity from Simplicial Bouquets) Let be a simplicial
complex and I = F(). Suppose contains a strongly disjoint set of bouquets B.
Let B = B B be a partition of B into disjoint subsets B and B , and let
b = | Facets(B )|, b = | Facets(B )|, and b = b + b = | Facets(B)|.
Then the monomials

m= x and m = x

x∈V (B ) x∈V (B )
are lattice complements in LCM(I ) and
βb ,m (S/I ) = 0 and βb ,m (S/I ) = 0
In particular,
tb tb + tb .
Proof As any bouquet of B is either in B or in B , and no two distinct bouquets

share any vertices, gcd(m, m ) = 1. The claim now follows directly from
Theorem 4.7 and Theorem 5.4.

Example 5.8 Let I = (ax, ay, bz, bv, bw, cu, cg, yz, az) and G = F(I ). It is
easy to check that B = {ax, ay, bz, bv, bw, cu, cg} (highlighted below) is
a strongly disjoint set of bouquets in G where {ax, bv, cu} ⊂ Facets(B) is a set of
pairwise 3-disjoint in G.
According to Corollary 5.6 (or ([17, Theorem 3.1] since this is the case of
bouquets in graphs) β7,10 (S/I ) = 0, so that t7 = 10. Using the following order
of B
ay, ax , bz, bw, bv, cg, cu

FGHI F GH I F GH I
Facets(ax,ay) Facets(bz,bv,bw) Facets(cu,cg)
and setting m = axy = lcm(ay, ax) and m = bzvwcug = lcm(bz, bw, bv, cg, cu),
by Theorem 5.7 we have β2,m (S/I ) = 0 and β5,m (S/I ) = 0. On the other hand,
using the following reordering of B
ay, ax , cg, cu, bz, bw, bv

FGHI FGHI F GH I
Facets(ax,ay) Facets(cu,cg) Facets(bz,bv,bw)
and setting
m = acxyug = lcm(ay, ax, cg, cu) and m = bzvw = lcm(bz, bw, bv),
by Theorem 5.7 we also get β4,m (S/I ) = 0 and β3,m (S/I ) = 0. As a result,
t7 < t2 + t5 = 12 and t7 < t3 + t4 = 13
which can be confirmed by using the Betti table for S/I .
0 1 2 3 4 5 67
total : 1 9 28 44 40 22 71
0:1 . . . . . . .
1: . 9 10 3 . . . .
2: . . 18 33 20 4 . .
3: . . . 8 20 18 71
Example 5.9 For I and B in Example 5.3, since | Facets(B1 )| = 2 and

| Facets(B2 )| = 6, and the two monomials abcd and efghixy are complements
in LCM(I ), it follows that β6,7 (S/I ) = 0 and β2,4 (S/I ) = 0 so that t8 t2 + t6 .
6 Reordering Well Ordered Covers
As we saw in the Sect. 4, the tail end of every well ordered cover is itself a well
ordered cover, and therefore guarantees a nonvanishing Betti number in a certain
homological degree. For example, let
m1 , . . . , ms (6.1)
be a well ordered cover of I . As a result of Proposition 4.5 we have that
βi,m (S/I ) = 0 for 1 i s and m = lcm(ms−i+1 , . . . , ms ). (6.2)
On the other hand, we also know by Proposition 4.4 that for a fixed i, setting
m = lcm(m1 , . . . , ms−i ) and m is as in (6.2), then m and m are complements. So,
if we additionally have
βs−i,m (S/I ) = 0 (6.3)

then (6.2) and (6.3) together would imply that
ts ti + ts−i . (6.4)
In this section we consider extra conditions under which (6.3) would hold. One
particular situation is when the monomials m1 , . . . , ms−i in the well ordered cover
in (6.1) could be shifted to the end of the ordering, so that
ms−i+1 , . . . , ms , m1 , . . . , ms−i
is also a well ordered cover of I . In this case, by Proposition 4.5 we would have
(6.3) automatically, and therefore more cases of the subadditivity inequality (6.4)
could easily follow.
Example 6.1 Consider the ideal I and = F(I ) as in Example 5.3 and consider
the well ordered cover
m1 m2 m3 m4 m5 m6 m7 m8
M: gy , gx , ge , , , , ,
gf bcd gh gi abc
Name n1 = cdf , n2 = def , n3 = f i and n4 = hi and for each k = 1, 2, 3, 4,

set
αk = max{j : mj | lcm(nk , mj +1 , · · · , m8 )}.
Then α1 = α2 = 5, α3 = 4 and α4 = 6. Let

= min(α1 , α2 , α3 , α4 ) = 4. Using
,
we can modify the order in M as follows:
m4 m5 m6 m7 m8 m1 m2 m3
M : , , , , , gy , gx , ge
gf bcd gh gi abc
It is easy to see that M is also a well ordered cover of I .

The example above is a special case of a simple observation, stated in Proposi-
tion 6.2, which is often powerful enough to prove the subadditivity inequality (6.4)
for all i.
Proposition 6.2 (Reordering Well Ordered Covers) Let
I = (m1 , . . . , ms , n1 , . . . , nk )
be a square-free monomial ideal which has m1 , . . . , ms as a well ordered cover.

Define for each 1 i k,
J K
αi = max j : mj | lcm(ni , mj +1 , · · · , ms ) . (6.5)
If
= min(α1 , . . . , αk ) > 1, then
(1) for every i ∈ {2, . . . ,

} the sequence
mi , mi+1 , . . . , ms , m1 , . . . , mi−1
is a well ordered cover of I ;

(2) ts ts−i + ti for 1 i
− 1.
Proof Statement 1 follows directly from the definition of well ordered covers. For
the second statement, by part 1, for 1 i <
, we have
mi+1 , . . . , ms , m1 , . . . , mi
is a well ordered cover for I . By (6.5)
mi+1 , . . . , ms
is a well ordered cover of I[m] where m = lcm(mi+1 , . . . , ms ). By Proposition 4.5
m1 , . . . , mi
is a well ordered cover of I[m ] where
m = lcm(m1 , . . . , mi ).
By Proposition 4.4, m and m are complements in LCM(I ), and by Theorem 3.4
βs−i,m = 0 and βi,m = 0,
which together imply that ts ts−i + ti .

Example 6.3 The ideal I from Example 5.3 has the following well ordered cover
M : gy , gx , ge , gf , bcd , gh , gi , abc.
In particular β8,11 (S/I ) = 0 and t8 = 11. The reordering in Proposition 6.2 of M

yields the following well ordered cover
M : gf , bcd , gh , gi , abc , gy , gx , ge,
where
= 4. By part 2 of Proposition 6.2 and using the new well ordered cover M ,
we have:
(1) t8 t7 +t1 . Here we take m = abcdfghixy = lcm(gf, bcd, gh, gi, abc, gy, gx)
and m = ge.
(2) t8 t6 + t2 . Here we take m = abcdfghiy = lcm(gf, bcd, gh, gi, abc, gy)
and m = gex = lcm(gx, ge). Note that this inequality was also done in
Example 5.9 using simplicial bouquets.
(3) t8 t5 + t3 . Here we take m = abcdfghi = lcm(gf, bcd, gh, gi, abc) and
m = gexy = lcm(gy, gx, ge).
The only remaining case is a = b = 4. If we take
m = bcdfghi = lcm(gf, bcd, gh, gi) and m = abcgexy = lcm(abc, gy, gx, ge),
then {abc, gy, gx, ge} is a well ordered cover of I[m ] by Proposition 4.5, and
it is easy to check that {gf, bcd, gh, gi} is a well ordered cover of I[m] . Hence
β4,m (S/I ) = 0 and β4,m (S/I ) = 0 by Theorem 3.4. Also as m and m are lattice
complements (by Proposition 4.4), we get t4 + t4 deg(m) + deg(m ) > 11 = t8 .
Acknowledgments The authors are grateful to the referees for their comments, which greatly
improved this paper. Sara Faridi’s research is supported by an NSERC Discovery Grant.
References
1. Avramov, L., Conca, A., Iyengar, S.: Subadditivity of syzygies of Koszul algebras. Math. Ann.,
361, no.1–2, 511–534 (2015)
2. Abedelfatah, A., Nevo, E.: On vanishing patterns in j -strands of edge ideals. J. Algebraic
Combin. 46, no. 2, 287–295 (2017)
3. Baclawski, K.: Galois connections and the Leray spectral sequence. Advances in Math. 25, no.
3, 191–215 (1977)
4. Bayer, D., Sturmfels, B.: Monomial resolutions. Math. Res. Lett. 5, no. 1–2, 31–46 (1998)
5. Eisenbud, D., Huneke, C., Ulrich, B.: The regularity of Tor and graded Betti numbers. Amer.
J. Math. 128, no. 3, 573–605 (2006)
6. El Khoury, S., Srinivasan, H.: A note on the subadditivity of Syzygies. Journal of Algebra and
its Applications, vol.16, no.9, 1750177 (2017)
7. Erey, N.: Bouquets, Vertex Covers and Edge Ideals. Journal of Algebra and its Applications,
vol.16, no.5, 1750084 (2017)
8. Erey, N., Faridi, S.: Betti numbers of monomial ideals via facet covers. J. Pure Appl. Algebra
220, no. 5, 1990–2000 (2016)
9. Faridi, S.: Lattice complements and the subadditivity of syzygies of simplicial forests. Journal
of Commutative Algebra, Volume 11, Number 4, 535–546 (2019)
10. Faridi, S., Shahada, M.: Breaking up Simplicial Homology and Subadditivity of Syzygies. J.
Algebraic Combin., (2021). https://doi.org/10.1007/s10801-021-01073-3
11. Fernandez-Ramos, O., Gimenez, P.: Regularity 3 in edge ideals associated to bipartite graphs.
J. Algebraic Combin., 39 (2014)
12. Fröberg, R.: On Stanley-Reisner rings. Topics in algebra, Banach Center Publications, 26 Part
2, 57–70 (1990)
13. Grayson, D., Stillman, M. : Macaulay2, a software system for research in algebraic geometry.
available at http://www.math.uiuc.edu/Macaulay2/.
14. Ha, H. T., Van Tuyl, A.: Monomial ideals, edge ideals of hypergraphs, and their graded Betti
numbers. J. Algebraic Combin., 27, 215–245 (2008)
15. Herzog, J., Srinivasan, H.: On the subadditivity problem for maximal shifts in free resolutions.
Commutative Algebra and Noncommutative Algebraic Geometry, II MSRI Publications
Volume 68, (2015)

16. Khosh-Ahang, F., Moradi, S.: Codismantlability and projective dimension of the Stanley-
Reisner ring of special hypergraphs. Proc. Indian Acad. Sci. Math. Sci. 128, no. 1, Paper No.
7, 10 pp (2018)
17. Kimura, K.: Non-Vanishingness of Betti Numbers of Edge Ideals. Harmony of Grobner Bases
and the Modern Industrial Society, World Scientific, Hackensack 153–168 (2012)
18. Lyubeznik, G.: A new explicit finite free resolution of ideals generate by monomials in an
R-sequence. J. Pure Appl. Algebra 51, 193–195 (1998)
19. Mermin, J.: Three simplicial Resolutions. Progress in commutative algebra 1, 127–141, de
Gruyter, Berlin (2012)
20. Peeva, I.: Graded syzygies. Algebra and Applications, 14, Springer-Verlag London, Ltd.,
London (2011)
21. Taylor, D.: Ideals generated by monomials in an R-sequence, Thesis, University of Chicago
(1966)
22. Zheng, X.: Resolutions of facet ideals. Comm. Algebra, 32, 2301–2324 (2004)
A Survey on the Eisenbud-Green-Harris
Conjecture
Sema Güntürkün
Keywords Hilbert functions · Lex-plus-powers ideals · Lexicographic ideals ·

Regular sequences
1 Introduction
Let I be a homogeneous ideal given in a standard graded polynomial ring R in n

variables over a field K, and let Id denote the degree d graded component of I.
Assuming the K-dimension of Id is known, it sounds a quite simple question to
ask what one can say about the dimension of the graded component of I in degree
d + 1, and yet it attracts a lot of attention in commutative algebra and algebraic
geometry. An answer to this question was given by Macaulay’s breakthrough work
[34] by providing a numerical bound for the growth of Hilbert function depending
on the value at the preceding degree. He showed that Hilbert functions of special
monomial ideals, called lexicographic ideals, describe all possible Hilbert functions
of homogeneous ideals in R. Macaulay’s result led to other classical results on
Hilbert functions such as Gotzmann’s Persistence Theorem and Green’s Hyperplane
Restriction Theorem (see [4] for nice treatments of all these theorems).
Generalizations of Macaulay’s result on the extremal behavior of lexicographic
ideals for Hilbert functions allows to relate a homogeneous ideal containing the
powers of variables with a monomial ideal containing the same powers of variables
(see [13, 30, 31]).
In their Higher Castelnuovo Theory paper, Eisenbud, Green and Harris con-
jectured a further generalization of Macaulay’s theorem for homogeneous ideals
containing a regular sequence in certain degrees (see Conjecture 3.3). Eisenbud-
Green-Harris (EGH) conjecture, motivated by Cayley-Bacharach theorems, sug-
gests a refinement of Macaulay’s bound on the growth of the Hilbert function
S. Güntürkün ()
Amherst College, Department of Mathematics and Statistics, Amherst, MA, USA
e-mail: sgunturkun@amherst.edu
328 S. Güntürkün
by involving the information of degrees of the regular sequence contained in the

ideal minimally. Although there are notable works done on Eisenbud-Green-Harris
(EGH) conjecture, it is still widely open after more than 25 years.
A survey on lex-plus-powers ideals by Francisco and Richert [19] provides a very
good source to understand these special monomial ideals thoroughly, and it also
discusses the EGH conjecture and its equivalent variations in details. Since there
have been significant progress on the EGH conjecture since [19], the main intent of
our survey paper is to contribute the literature by providing the current state of the
EGH conjecture and to assemble the results that have been obtained so far.
As plan of this paper, in Sect. 2 we state some preliminaries and review
Macaulay’s results on Hilbert function. Section 3 lays out the Eisenbud-Green-
Harris (EGH) conjecture and its variations. In Sect. 4, we present the results obtained
on the EGH conjecture by grouping them in terms of their approaches. Finally, in
Sect. 5 we point out the open cases of the EGH conjecture, and we recall a closely
related conjecture known as the Lex-Plus-Powers conjecture. We conclude the final
section with some applications of EGH.
2 Preliminaries and Macaulay’s Theorem on Hilbert

Functions
We let R be the polynomial ring K[x1 , . . . , xn ] over a field K with standard grading
R = ⊕i0 Ri where Ri is the i-th graded component.
We fix the lexicographic order as x1 >lex x2 >lex . . . >lex xn . Then we define the
monomial order between two monomials of the same degree as x1a1 x2a2 · · · xnan >lex
x1b1 x2b2 · · · xnbn if ai > bi where i is the smallest index such that ai = bi . For the
sake of simplicity, we use > for >lex .
Definition 2.1 Let I be a monomial ideal in R minimally generated by monomials
m1 , . . . , mk . We call I a lexicographic ideal or simply a lex ideal if it satisfies the
following property: for any monomial m in R with deg m = deg mi and m > mi
for some i = 1, . . . , k, then m ∈ I as well.

We next define another special type of monomial ideal in our context.
Definition 2.2 For given 2 a1 a2 . . . an , we call a monomial ideal L a
lex-plus-powers ideal associated with degree (a1 , . . . , an ) if it can be written as
L = (x1a1 , . . . , xnan ) + J
where J is a lex ideal in R.

For any homogeneous ideal I we define the Hilbert function of R/I as
HR/I (t) = dimK (R/I )t = dimK Rt − dimK It

A Survey on the EGH Conjecture 329
n−1+t !
where Rt is the degree t graded component of R with dimK Rt = t , and It is
the degree t graded component of the ideal I .
Example 2.3 In K[x, y, z], consider the monomial ideal I = (x 3 , xy, y 4 , yz, z2 ).
Then simple computations give us the graded components of R/I ; (R/I )1 = K −
Span{x, y, z}, (R/I )2 = K − Span{x 2 , xz, y 2 }, (R/I )3 = K − Span{x 2 z, y 3 } and
dimK (R/I )i = 0 for i 4. Therefore, one expresses the Hilbert function of R/I
as (h0 , h1 , h2 , h3 ) = (1, 3, 3, 2) where hi = HR/I (i) and hi = 0 for i 4.

As we see in the above example, it is possible to compute the Hilbert function
of monomial ideals even by hand, however for arbitrary homogeneous ideals it
becomes challenging to calculate without using a software such as Macaulay 2 (see
[33] package LexIdeals, command hilbertFunct).
We call a sequence f1 , . . . , fr of forms in R = K[x1 , . . . , xn ] a regular
sequence of length r if, for each i = 1, . . . , r, fi is a non-zero-divisor on the ring
R/(f1 , . . . , fi−1 ). If r = n, then the regular sequence has full length, in this case, it
is referred as a maximal regular sequence.
Remark 2.4 Let c be a homogeneous ideal in R generated by a regular sequence
f1 , . . . , fn with deg fi = ai for i = 1, . . . , n. We call c a complete intersection
ideal of type (a1 , . . . , an ) and the ring R/c is called a complete intersection ring.
Then the Hilbert function of R/c is HR/c (i) = HR/(x a1 ,...,x an ) (i) for all i 0.
1 n

n
Furthermore, HR/(x a1 ,...,x an ) (i) = HR/(x a1 ,...,x an ) (s − i) where s = (ai − 1).
1 n 1 n
i=1

The following proposition provides a very useful relation between the Hilbert
function of an ideal I and the Hilbert function of another ideal generated by a regular
sequence contained in I under the liaison (see [15, Theorem 3]).
Proposition 2.5 Let I be a homogeneous ideal and c ⊆ I a complete intersection

n
ideal, and s = (ai − 1). Then, for all j 0,
i=1
HR/I (j ) = HR/c (j ) − HR/(c:I ) (s − j )
For a given two positive integers d and i, the i-th Macaulay representation of d
(also known as the i-th binomial expansion of d), denoted d (a) , is given by

mi mi−1 m2 m1
d= + + ... + +
i i−1 2 1
where mi > mi−1 > . . . > m1 0 are uniquely determined and are called the i-th
Macaulay coefficients of d. In this case, we let

i mi + 1 mi−1 + 1 m2 + 1 m1 + 1
d := + + ... + + .
i+1 i 3 2
330 S. Güntürkün
! ! ! !
To give a simple example, let d = 21, i = 4. Then 21(4) = 64 + 43 + 22 + 11 ,
! ! ! !
therefore 214 = 75 + 54 + 33 + 22 = 28.
We next state Macaulay’s well-known theorem on Hilbert functions.
Theorem 2.6 ([4, 34]) Let I be a homogeneous ideal in the polynomial ring R.
(a) There is a lex ideal in R with the same Hilbert function, and this lex ideal is
uniquely determined.
(b) [Macaulay’s bound] If HR/I (i) = d then
HR/I (i + 1) d i .
!
Example 2.7 Notice that HR (i) = n+i−1 = d and so the i-th Macaulay
! i !
representation of d is d (i) = n−1+i
i . Computing HR (i + 1) = n+i
i+1 shows that R
attains exactly Macaulay’s bound d i .

The following important theorem shows when a homogeneous ideal carries a
similar behavior of attaining Macaulay’s bound as in Example 2.7.
Theorem 2.8 (Gotzmann’s Persistence Theorem [4, 24]) Let I be a homoge-
neous ideal in R generated by forms of degree d. If the Hilbert function of
R/I achieves Macaulay’s bound in the next degree d + 1, that is HR/I (d + 1) =
HR/I (d)d , then
HR/I (j ) = HR/I (j − 1)d for all j d.
Another classical result on the growth of Hilbert functions worth to mention is

given by Green. This result was also used to give an elegant proof of Macaulay’s
theorem.
Theorem 2.9 (Green’s Hyperplane Restriction Theorem [4, 22]) Let I be a
homogeneous ideal in R, and let t 1 be a given degree. Then

mt − 1 mt−1 − 1 m2 − 1 m1 − 1
HR/I +(
) (t) + + ... + +
t t −1 2 1
where

mt mt−1 m2 m1
HR/I (t)(t) = + + ... + +
t t −1 2 1
is the t-th Macaulay representation of HR/I (t).

3 The EGH Conjecture
A generalization of Macaulay’s results was considered by studying homogeneous

ideals in S = R/(x1a1 , . . . , xnan ) instead of homogeneous ideals in the polynomial
ring R = K[x1 , . . . , xn ]. The existence of the lex ideal in S with the same Hilbert
function was shown by Kruskal [31] and Katona [30] when a1 = a2 = . . . =
an = 2 and more generally when 2 a1 . . . an was done by Clements and
Lindström [13] and also by Greene-Kleitman [23]. These results were obtained in
set theoretical and combinatorial settings.
A question can be raised for a similar behavior for the homogeneous ideals in
R/(f1 , . . . , fn ) where f1 , . . . , fn is a regular sequence in R. In [16], Eisenbud,
Green and Harris initially stated the following conjecture for the case of regular
sequence of quadrics.
Conjecture 3.1 Given homogeneous ideal I in R containing a full length regular
sequence of quadratic forms. Let HR/I (i) = h and the i-th Macaulay representation
of h be

mi mi−1 m2 m1
h(i) = + + ... + +
i i−1 2 1
where mi > mi−1 > . . . m1 0.

Then

mi mi−1 m1
HR/I (i + 1) + + ... + .
i+1 i 2
The new bound proposed by Conjecture 3.1 ! is m

finer
! than Macaulay’s bound. We
m !
can see this using the binomial identity m+1
k = k + k−1 ,

i mi + 1 mi−1 + 1 m1 + 1
h = + + ... +
i+1 i 2

mi mi mi−1 mi−1 m1 m1
= + + + + ... + +
i+1 i i i−1 2 1

mi mi−1 m1
= h(i) + + + ... +
i+1 i 2

mi mi−1 m1
> + + ... + .
i+1 i 2
Example 3.1 Suppose I ⊆ K[x1 , . . . , x7 ] is a homogeneous ideal containing a

regular sequence of quadratic forms f1 , . . . , f7 and HR/I (2) = 17. The possible
growth for the Hilbert function in degree 3 by Macaulay’s bound is HR/I (3) 38,
but Conjecture 3.1 claims that HR/I (3) 21.

332 S. Güntürkün
If the homogeneous ideal I is generated by generic quadrics, it is already known

that Conjecture 3.1 is true by Herzog and Popescu [27] when the characteristic is
zero. When K has arbitrary characteristic, this was shown by Gasharov [20].
The main motivation behind Conjecture 3.1 about homogeneous ideals contain-
ing quadratic regular sequence was another conjecture, known as the Generalized
Cayley-Bacharach conjecture, stated in [16] in more geometric perspective.
Conjecture 3.2 (Generalized Cayley-Bacharach Conjecture for Quadrics) Let be
a complete intersection of quadrics in Pn . Any hypersurface X ⊂ Pn of degree d
containing a subscheme ⊂ of degree strictly greater than 2n − 2n−d must
contain .

Conjecture 3.1 implies the Generalized Cayley-Bacharach conjecture for quadrics.
In the same article, Eisenbud, Green and Harris dropped the quadratic condition
on the regular sequence, and further conjectured the same statement for homoge-
neous ideals containing regular sequences with any degrees 2 a1 a2 . . . an .
Conjecture 3.3 (Eisenbud-Green-Harris (EGH) Conjecture, [16]) Let I be a
homogeneous ideal in R = K[x1 , . . . , xn ] containing a regular sequence f1 , . . . , fn
with degrees a1 , . . . , an such that 2 a1 . . . an . Then there is a lex-plus-
powers ideal L = (x1a1 , . . . , xnan ) + J with a lex ideal J in R such that
HR/I (i) = HR/L (i) for all i 0.
From now on, we will refer to this conjecture as the EGH conjecture. We will
also use EGH(a1 ,...,an ),n to emphasize the degrees of the regular sequence and also
that it is a full length-n regular sequence in R = K[x1 , . . . , xn ].
Notice that in Remark 2.4, we observed a very trivial version of this statement
when I = (f1 , . . . , fn ) and L = (x1a1 , . . . , xnan ).
Another statement of the Generalized Cayley-Bacharach conjecture that does
not require quadrics was given in [17, Conjecture CB12]. In 2013, Geramita and
Kreuzer reformulated this version of Generalized Cayley-Bacharach conjecture for
arbitrary degrees by dividing it into intervals. They also strengthened the Conjecture
CB12 in [21, Conjecture 3.5]. In P3 , they provided a proof for it. In Pn , they
confirmed [21, Conjecture 3.5] for some intervals. The EGH conjecture which is
the concern of this paper is stronger than [17, Conjecture CB12] as well.
One of the variations of the EGH conjecture in the literature is when one allows
to have a regular sequence that is not of full length.
Conjecture 3.4 (EGHn,(a1 ,...,ar ),r ) Let I be a homogeneous ideal in R =
K[x1 , . . . , xn ] containing a regular sequence of length r < n with degrees
a1 , . . . , ar such that 2 a1 . . . ar . Then there is a lex-plus-powers ideal
L = (x1a1 , . . . , xrar ) + J in R with the same Hilbert function as I .

Remark 3.2 The equivalence between Conjectures 3.3 and 3.4 was discussed by
Caviglia and Maclagan [9]. For 2 a1 . . . ar fixed, they showed that if
EGH(a1 ,...,an ),n holds for all 2 a1 . . . an where ai = ai for i = 1, . . . , r
then EGHn,(a1 ,...,ar ),r holds (see [9, Propositions 9]). They also showed that if
EGH(a1 ,...,ar ),r holds then EGHn,(a1 ,...,ar ),r holds for all n r (see [9, Propositions
10]).

4 Results on the EGH Conjecture
Richert [37] proved that the EGH conjecture is true for R = K[x, y]. Thus, for any
homogeneous ideal I in two variables containing a regular sequence f1 , f2 with
degrees 2 a1 a2 , we have a lex plus powers ideal L = (x a1 , y a2 ) + J such that
dimK Ii = dimK Li for all i 0.
For K[x1 , . . . , xn ] with n > 2, we put together the known results on EGH
depending on the approaches were used.
4.1 EGH Depending on the Degrees (a1 , . . . , an )
Let n 2. For a fixed degree d 1, when the Hilbert function of R/I at degree d is
known, the EGH conjecture proposes a maximal growth for degree d +1. One of the
adopted approaches in the literature focuses on the growth at certain degree. Hence,
the following definition states a partial version of EGH conjecture for consecutive
degrees.
Definition 4.1 (EGH(a1 ,...,an ),n (d)) For any homogeneous ideal I in R containing
a regular sequence f1 , . . . , fn of degrees 2 a1 . . . an respectively, if there
exists a lex-plus-powers ideal L associated with degrees a1 , . . . , an such that
HR/I (d) = HR/L (d) and HR/I (d + 1) = HR/L (d + 1),
then we say that EGH(a1 ,...,an ),n (d) holds.

The following proposition is given by Francisco [18] for the almost complete
intersection ideals.
Proposition 4.2 Let I = (f1 , . . . , fn , g) be a homogeneous ideal where f1 , . . . , fn
is a regular sequence with degrees a1 , . . . , an and deg g = d a1 . Then
EGH(a1 ,...,an ),n (d) is true for I .

In [6], for an almost complete intersection I = (f1 , f2 , f3 , g) ⊂ K[x1 , . . . , xn ]
where f1 , f2 , f3 is a regular sequence of length three with deg fi = ai , i = 1, 2, 3
and deg g = d a1 +a2 +a3 −3, Caviglia-De Stefani showed HR/I (i) HR/L (i)
for all i 0 where L = (x1a1 , x2a2 , x3a3 , m) with m is the largest monomial of degree
334 S. Güntürkün
d with respect to lexicographic order that is not in (x1a1 , x2a2 , x3a3 ). Their work on
such almost complete intersections also recovered the result of [21] for P3 .
To show that EGH(d,...,d),n (d) holds it suffices to show that the statement in
Definition 4.1 holds for the homogeneous ideals I = (f1 , . . . , fn , g1 , . . . , gm ) ⊆
K[x1 , . . . , xn ] generated by degree d forms f1 , . . . , fn , g1 , . . . , gm where
f1 , . . . , fn form a regular sequence (see [25, Lemma 2.6].) In other words, it is
enough to show the statement for the ideals where not only the regular sequence
f1 , . . . , fn in the generators have degree d, but also rest of the generators g1 , . . . , gm
have degree d too. Thus, focusing on the case d = 2 we have the following remark.
Remark 4.3 In order to show that EGH(2,...,2),n (2) is true, it suffices to study the
ideals generated by only quadrics containing a maximal regular sequence.

In [9], Caviglia-Maclagan provided the following lemma about this weaker
version of the EGH. Due to its importance as a tool for studying the EGH conjecture,
we would like to present its proof given in [9, Lemma 12].

n
Lemma 4.4 Given 2 a1 . . . an , set s = (ai − 1). Let d 1. Then
i=1
EGH(a1 ,...,an ),n (d) holds if and only if EGH(a1 ,...,an ),n (s − d − 1) holds.
Furthermore, if EGH(a1 ,...,an ),n (d) holds for all 0 d s−1 2 then EGH(a1 ,...,an ),n
holds.

Proof Suppose that EGH(a1 ,...,an ),n (d) holds. Then given any homogeneous ideal
containing a regular sequence with degrees a1 , . . . , an , there is lex-plus-powers
ideal associated with degrees a1 , . . . , an such that Hilbert functions of both ideals
agree at degrees d and d +1. Let I be a homogeneous ideal in R containing a regular
sequence f1 , . . . , fn with deg fi = ai , for i = 1, . . . , n. Then by Proposition 2.5
we get
HR/I (j ) = HR/(f1 ,...,fn ) (j ) − HR/((f1 ,...,fn ):I ) (s − j ).
Since the colon ideal ((f1 , . . . , fn ) : I ) contains the regular sequence f1 , . . . , fn ,

then by assumption there is a lex-plus-powers ideal L = (x1a1 , . . . , xnan ) + J such
that
HR/((f1 ,...,fn ):I ) (d) = HR/L (d) and HR/((f1 ,...,fn ):I ) (d + 1) = HR/L (d + 1).
(1)
On the other hand, we also have
HR/L (j ) = HR/(x a1 ,...,x an ) (j ) − HR/((x a1 ,...,x an ):L ) (s − j ). (2)

1 n 1 n
Thus, for j = s − d − 1 and j = s − d,
HR/I (j ) = HR/(f1 ,...,fn ) (j ) − HR/((f1 ,...,fn ):I ) (s − j )

= HR/(f1 ,...,fn ) (j ) − HR/(x a1 ,...,x an ) (s − j ) + HR/((x a1 ,...,x an ):L ) (j )

1 n 1 n
= HR/((x a1 ,...,x an ):L ) (j ).

1 n
The second equality follows from (1) and (2). Then the last equality is by
Remark 2.4 and Proposition 2.5.
Finally, since the ideal ((x1a1 , . . . , xnan ) : L ) contains the regular sequence
a1
x1 , . . . , xnan , by Clement-Lindström result mentioned previously, there exists
a lex-plus-powers ideal L associated with degrees a1 , . . . , an such that
HR/((x a1 ,...,x an ):L ) (i) = HR/L (i) for all i 0.
1 n
Hence,
HR/I (s − d − 1) = HR/L (s − d − 1) and HR/I (s − d) = HR/L (s − d).

Remark 4.5 We know HR/I (0) = 1. If HR/I (1) = n, then any lex-plus-powers
ideal L = (x1a1 , . . . , xnan ) + J where the lex ideal J does not contain any linear form
has HR/L (0) = 1 and HR/L (1) = n. If HR/I (1) = r < n, that is I has n − r
linear generators, then it is enough to pick the lex ideal J containing x1 , . . . , xn−r ,
then HR/L (1) = r as well. Therefore, we see that EGH(a1 ,a2 ,...,an ),n (0) is always
true.

−1
j
Theorem 4.6 ([7, 9]) For 2 a1 . . . an such that aj (ai − 1), the
i=1
EGH(a1 ,...,an ),n is true.

The strict inequality was shown by Caviglia-Maclagan in [9]. Their proof used
an inductive argument on n, and followed from Lemma 4.4 and Remark 3.2. Very
recently, Caviglia-De Stefani [7] extended this degree growth condition by including
the equality. Their work actually provided a stronger case. They showed that if a
homogeneous ideal I ⊆ R contains a regular sequence f1 , . . . , fn−1 with degrees
2 a1 . . . an−1 satisfies the EGHn,(a1 ,...,an−1 ),n−1 , then I + (fn ) satisfies the
EGH conjecture for any fn where f1 , . . . , fn form a regular sequence and deg fn

n−1
(ai − 1) (see [7, Theorem 3.6]).
i=1
The result of Caviglia-Maclagan with the recent improvement by Caviglia-De
Stefani provides an affirmative answer for the EGH conjecture for a large case in
terms of the degrees a1 , . . . , an . One of the interesting case that is not covered by
this result is when a1 = a2 = . . . = an , more specifically as in Conjecture 3.1 when
n 4. We will focus on the quadratic case ai = 2 for all i = 1, . . . , n separately
(see Sect. 4.4).
In [14], Cooper approached the EGH conjecture for n = 3 in a geometric setting
by investigating the Hilbert functions of the subsets of complete intersections in P2
and P3 . She showed that the EGH(a1 ,a2 ,a3 ),3 is true for the degrees (2, a, a) for a 2
and (3, a, a) for a 3.
336 S. Güntürkün
Another result for n = 3 for the Gorenstein ideals containing a regular sequence
with degrees 2 a1 a2 a3 was proven by Chong [12].
4.2 EGH via Liaison
Chong’s work covers EGH beyond Gorenstein ideals in K[x, y, z]. It uses the
linkage theory and studies a special subclass of licci ideals. First, we would like
to review some definitions and concepts related to linkage theory for ideals in
R = K[x1 , . . . , xn ] to present Chong’s result.
Let I, I ⊆ R be homogeneous ideals of height r n. If there exists a regular
sequence g1 , . . . , gr such that the complete intersection c = (g1 , . . . , gr ) ⊆ I ∩ I ,
and I = c : I and I = c : I , then we say that I and I are linked (algebraically)
c
via c. We express this linkage as I ∼ I . If I minimally contains a regular sequence
f1 , . . . , fr with degrees 2 a1 . . . ar and if the link c is a complete
intersection of type (a1 , . . . , ar ) then we say c is a minimal link.
c c c
Suppose that there is a finite sequence of links I ∼1 I1 ∼2 · · · ∼t It where It is
a complete intersection, we say that I is in the linkage class (a.k.a. liaison class) of
the complete intersection It . An ideal in the linkage class of a complete intersection
is called licci. We next state the work of Chong on this.
Theorem 4.7 ([12]) Let 2 a1 . . . an , and I ⊆ R be a homogeneous ideal
c c
containing a regular sequence of degrees a1 , . . . , an . If I is licci such that I ∼1 I1 ∼2
c
· · · ∼t It where each link ci has type āi for i = 1, . . . , n with ā1 ā2 . . . ān ,
and c1 is a minimal link, that is ā1 = (a1 , . . . , an ), then there is a lex-plus-powers
ideal associated with degrees a1 , . . . , an with the same Hilbert function as I .

In the same paper [12], Chong also proved that EGHn,(a1 ,...,ar ),r holds for the licci
ideals where the types of the links satisfy the ascending condition and the first link
is minimal. His result on Gorenstein ideals when n = 3 we mentioned in previous
subsection is a consequence of this result.
4.3 EGH via the Structure of the Regular Sequence
Let I ⊆ R = K[x1 , . . . , xn ] be a homogeneous ideal containing a regular sequence

f1 , . . . , fn with degree 2 a1 . . . an , respectively. By Clements-Lindström’s
result, we already know that EGH(a1 ,...,an ),n is true when fi = xiai for all i =
1, . . . , n.
Mermin and Murai [35] proved a special case of EGHn,(a1 ,...,ar ),r , r < n,
when char K = 0. For the homogeneous ideals containing a regular sequence
f1 , . . . , fr formed by monomials with degrees 2 a1 . . . ar , they showed
the existence of lex-plus-powers ideal associated with (a1 , . . . , ar ) with the same
Hilbert function.
Another notable work regarding the structure of the regular sequence contained
in the ideal is done by Abedelfatah in [1]. He showed that if a homogeneous ideal
I containing a regular sequence f1 , . . . , fn such that deg fi = ai , i = 1, . . . , n and
each fi splits into linear factors completely, then I has the same Hilbert function of
a lex-plus-powers ideal (x1a1 , . . . , xnan ) + J in R.
Shortly after, Abedelfatah extended this result in [2].
Theorem 4.8 Let a be an ideal generated by the product of linear forms and con-
tain a regular sequence f1 , . . . , fr with degrees a1 , . . . , ar . Let I be a homogeneous
ideal in R such that (f1 . . . , fr ) ⊂ a ⊂ I then the Hilbert function of I is the same
as the lex-plus-powers ideal containing powers x1a1 , . . . , xrar .

The previous result in [1] is simply the case when a = (f1 , . . . , fr ). Let I ⊂ R be
a height r monomial ideal containing a regular sequence of degrees a1 . . . ar ,
then this theorem of Abedelfatah confirms that I has the same Hilbert functions
as a lex-plus-powers containing x1a1 , . . . , xrar . This also improves another related
result given by Caviglia-Constantinescu-Varbaro in [5] for height r monomial ideals
generated by quadrics.
4.4 When a1 = . . . = an = 2
Finally we focus on the case when the regular sequence is formed by quadrics
as originally conjectured by the Eisendbud-Green-Harris as in Conjecture 3.1. For
simplicity, we refer it as EGH(2,2,...,2),n = EGH2̄,n .
We have already mentioned the cases when n = 2 by Richert [37] as his result
shows EGH for any degree when n = 2. Moreover, in K[x, y, z], we have seen
that the EGH conjecture for the degrees (2, 2, 2) was covered by Cooper [14] and
Caviglia-De Stefani [7] separately, and quadratic monomial ideal case by [5].
In terms of the weaker version of the EGH conjecture for consecutive degrees
given in Definition 4.1, using Proposition 4.2 given by Francisco, EGH2̄,n (2) is true
for almost complete intersections I = (f1 , . . . , fn , g) where deg fi = 2 for all
i = 1, . . . , n and deg g = 2. More precisely, HR/I (3) HR/L (3) where L is the
lex-plus-powers ideal containing the squares of the variables.
An analogous result on EGH2̄,n (2) for the ideals generated by a quadratic regular
sequence plus two more generators is given in the following theorem.
Theorem 4.9 ([25]) For n 5, EGH2̄,n (2) holds for homogeneous ideals mini-
mally generated by a regular sequence of quadrics and two more generators whose
degrees are at least 2.

Thanks to [25, Lemma 2.6], which is mentioned in Remark 4.3, to prove Theo-
rem 4.9 it was enough to show the statement for an ideal I generated by n + 2
quadrics containing a maximal regular sequence. More precisely, it sufficed to show
338 S. Güntürkün
the Hilbert function of the lex-plus-powers ideal (x12 , . . . , xn2 , x1 x2 , x1 x3 ) in degree

3 is greater than or equal to HR/I (3). This was shown by analyzing the linear
relations among the generators of the ideal I .
For a homogeneous ideal I = (f1 , . . . , fn , g1 , . . . , gm ) containing quadratic
regular sequence f1 , . . . , fn , it is easy to see that if each gi has degree > 2 then
they don’t contribute the dimension in degree 2 and dimK I2 = n. Therefore, any
lex plus power ideal L = (x12 , . . . , xn2 ) + J where J is generated by monomials of
degree > 2 gives dimK L2 = n as well.
We finish this section by presenting the known cases of the original EGH
conjecture for n 4.
Theorem 4.10 EGH2̄,n is true when
(a) n = 4 by Chen [11],
(b) n = 5 by the author and Hochster [25].

n
Proof Notice that when a1 = . . . = an = 2 we get s = (ai − 1) = n. Then by
i=1
Lemma 4.4, we get EGH2̄,n (d) holds if and only if EGH2̄,n (n − 1 − d) holds.
Using this symmetry, when n = 4, by Remark 4.5 we trivially have EGH2̄,4 (0),
therefore we have EGH2̄,4 (3). It is enough to show EGH2̄,4 (1) and therefore we also
get EGH2̄,4 (2).
Similarly, when n = 5, we know EGH2̄,5 (0) holds, so does EGH2̄,5 (4). Then
we need to show only EGH2̄,5 (1) and EGH2̄,5 (2) because EGH2̄,5 (1) implies
EGH2̄,5 (3).
By [11, Proposition 2.1], we know that EGH2̄,n (1) holds for any n 2. Thus
this completes the proof of (a).
On the other hand, the proof of (b) is done as well since EGH2̄,5 (2) is true as a
result of Theorem 4.9 and Remark 4.3.

5 Open Cases of EGH and More Connections
Although there has been a significant progress on the EGH conjecture, it is fair to
say that the conjecture is still broadly open. In this section, we discuss the open
cases, and also state another well known conjecture related to the EGH conjecture.
Besides Richert’s result on EGH when n = 2 in [37], we still do not know if the
EGH conjecture is true when n 3 without assuming any conditions on the degrees
or on the homogeneous ideal.
Question 5.1 Is EGH(a1 ,a2 ,a3 ),3 true for any given degrees 2 a1 a2 a3 ?

We have seen that the works by Cooper, Caviglia-Maclagan and Caviglia-De
Stefani cover many cases of (a1 , a2 , a3 ) already. On the other hand, Chong’s and
Abedelfatah’s results require certain conditions on the homogeneous ideals. As a
result, we can conclude that it is not known if EGH(a1 ,a2 ,a3 ),3 is true for the ideals in
the following scenario: Let I be a homogeneous ideal containing a regular sequence

f1 , f2 , f3 such that
• the degrees deg fi = ai , i = 1, 2, 3 satisfy 4 a1 a2 a3 and a2 a3 <
a1 + a2 − 2,
• I is not a Gorenstein ideal, and
• f1 , f2 , f3 cannot be split into linear factors.
For example, EGH(4,4,5),3 and EGH(4,5,5),3 are two open cases with small
degrees.
Remark 5.2 If we focus on the original EGH conjecture, that is, when a1 = . . . =
an = 2, EGH2̄,n is still open when n 6.

For a given homogeneous ideal I in R = K[x1 , . . . , xn ], the graded Betti number
of I is βi,j (R/I ) = dimK (TorR i (R/I, K))j . Another well-known conjecture
motivated by the EGH conjecture is given by Evans and Charalambous if these
graded Betti numbers are also concerned (see lex-plus-powers ideals survey [19]).
This conjecture can be also considered analogous to Bigatti-Hulett-Pardue Theorem
[3, 28, 36] which is a generalization of the Macaulay’s theorem for the graded Betti
numbers, more precisely, it shows the extremal behavior of lex ideals for the graded
Betti numbers among the homogeneous ideals with the same Hilbert function.
Conjecture 5.1 (Lex-Plus-Powers (LPP) Conjecture) Let I be a homogeneous ideal
containing a regular sequence of degrees 2 a1 . . . an in R = K[x1 , . . . , xn ].
Suppose that there exists a lex-plus-powers ideal L with the same Hilbert function
as I . Then the graded Betti numbers of R/I cannot exceed those of R/L. That is,
βi,j (R/I ) βi,j (R/L) for all i and j.
Just like the EGH conjecture, the LPP conjecture remains widely open. Nev-
ertheless, there have been remarkable results obtained. In [37], Richert showed
the equivalence of the EGH conjecture and the LPP conjecture when n = 2, 3.
Therefore, the LPP conjectures holds when n = 2.
Similar to Conjecture 3.4 where ideal contains a regular sequence of length n,
one can restate the LPP conjecture by allowing non-maximal regular sequences
f1 , . . . , fr with r < n. Caviglia and Kummini [8] showed that this case can be
also reduced to Artinian case like the EGH conjecture. In [35], the LPP conjecture
is shown to be true when the homogeneous ideal containing monomial regular
sequence. Thus, when the regular sequence has full length then the LPP conjecture
is true for the homogeneous ideals containing x1a1 , . . . , xnan .
In [10], when characteristic of K is 0, it was shown that the LPP conjecture holds
for the homogeneous ideals containing a regular sequence with degrees 2 a1

i−1
. . . an if ai (aj − 1) + 1 for i 3.
j =1
340 S. Güntürkün
When i = 1, the numbers β1,j (R/I ) tells us how many generators the
homogeneous ideal I has in degree j . For a given homogeneous I containing a
regular sequence with degrees 0 a1 . . . an , the EGH conjecture claims the
existence of the lex-plus-powers ideal L associated with degrees 0 a1 . . . an
such that dimK Ij = dimK Lj for all j 0. Then it is well-known that this implies
β1,j (R/L) β1,j (R/I ). Thus, the EGH conjecture covers this particular case of
the LPP conjecture when i = 1.
Remark 5.3 Let I = (f1 , . . . , fn , g1 , g2 ) ⊆ K[x1 , . . . , xn ] be a homogeneous
quadratic ideal where f1 , . . . , fn is a regular sequence. Then, by Theorem 4.9,
we see that EGH2̄,n (2) holds for such quadratic ideals. Therefore, we get
dimK I3 n2 + 2n − 5 = dimK L3 , where L is the lex-plus-powers ideal
(x12 , . . . , xn2 , x1 x2 , x1 x3 ). This shows us that the number of the independent linear
relations among the generators f1 , . . . , fn , g1 , g2 is always at most 5. In other
words, we obtain β2,3 (R/I ) 5 = β2,3 (R/L) as well.

Richert and Sabourin, in [38], showed that the following conjecture, a special
case of the LPP conjecture when i = n, is equivalent the EGH conjecture.
Conjecture 5.2 Let I be a homogeneous ideal in R containing a regular sequence
of degrees 2 a1 . . . an and let L be a lex-plus-powers ideal associated with
degrees ai such that HR/I (i) = HR/L (i) for all i 0. Then the dimension of the
socle of I is at most the dimension of socle of L in every degree. In other words,
βn,j (R/L) βn,j (R/I ) for all j 0.

The LPP conjecture seems much harder than the EGH conjecture due to its strong
claim on every graded Betti numbers, yet focusing on certain Betti numbers as its
special cases opens up many new directions to work on.
Finally, there are some interesting applications of the EGH conjecture worth
to mention. In [26], Harima-Wachi-Watanabe show that, assuming the EGH con-
jecture is true, every graded complete intersection has the Sperner property,
which simply says for a graded complete intersection A = R/(f1 , . . . , fn ),
max{μ(I ) | I is an ideal in A} = max{dimK Ai | i = 0, 1, 2, . . .}, where μ(I ) is
the number of minimal generators of I . It is also known that the Sperner property is
related to the so-called Weak Lefschetz property.
Due to geometric background of the EGH conjecture as a result of its connection
to Cayley-Bacharach theory, EGH has applications in more geometric settings as
well. For instance, a recent work by Jorgenson [29] points out that an affirmative
answer for EGH has an implication on the sequence of secant indices of Veronese
varieties of Pn (see Question 3.2 in [29].)
Another application of the EGH conjecture coincides with a very famous
problem on decomposing real polynomials in n variables as a sum of squares of
real polynomials. The cone of real polynomials that can be decomposed as a sum
of squares of real polynomials is simply referred as SOS cone. In a recent work
by Laplagne and Valdettaro[32], they show that, when EGH holds, for a strictly
positive polynomial on the boundary of the SOS cone, they provide bounds for the
maximum number of polynomials that can appear in a SOS decomposition and the
maximum rank of the matrices in the Gram spectrahedron.
Acknowledgments The author thanks Mel Hochster for introducing and proposing to work on
the EGH conjecture during her postdoctoral research. She is deeply grateful for all of their
conversations. The author thanks the referee for their valuable feedback and comments. She also
thanks Martin Kreuzer for pointing out their work.
References
1. Abedelfatah, A.: On the Eisenbud-Green-Harris conjecture. Proc. Amer. Math. Soc. 143, no.
1, 105–115 (2015)
2. Abedelfatah, A.:Hilbert functions of monomial ideals containing a regular sequence. Israel J.
Math. 214, no. 2, 857–865 (2016)
3. Bigatti, A.: Upper bounds for the betti numbers of a given Hilbert function. Comm. Algebra
21, no. 7, 2317–2334 (1993)
4. Bruns, W., Herzog, J.: Cohen-Macaulay Rings. Cambridge Studies in Advanced Mathematics.
39, Cambridge University Press, Cambridge (1993)
5. Caviglia, G., Constantinescu, A., Varbaro, M.: On a conjecture by Kalai. Israel J. Math. 204,
no. 1, 469–475 (2014)
6. Caviglia, G., De Stefani, A.: A Cayley-Bacharach theorem for points in Pn . Bull. Lond. Math.
Soc., 53, no. 4, 1185–1195 (2021)
7. Caviglia, G. De Stefani, A.: The Eisenbud-Green-Harris conjecture for fast-growing degree
sequences. Preprint. arXiv:2007.15467v2 (2020)
8. Caviglia, G., Kummini, M.: Poset embeddings of Hilbert functions and Betti numbers. J.
Algebra 410, 244–257 (2014)
9. Caviglia, G., Maclagan, D.: Some cases of the Eisenbud-Green-Harris conjecture. Math. Res.
Lett. 15, no. 3, 427–433 (2008)
10. Caviglia, G., Sammartano, A.: On the lex-plus-powers conjecture. Adv. Math. 340, 284–299
(2018)
11. Chen, R.-X.: Some special cases of the Eisenbud-Green-Harris conjecture. Illinois J. Math. 56,
no. 3, 661–675 (2012)
12. Chong, K.F.E.: An Application of liaison theory to the Eisenbud-Green-Harris conjecture. J.
Algebra 445, 221–231 (2016)
13. Clements, G., Lindström, B.: A generalization of a combinatorial theorem of Macaulay. J.
Combinatorial Theory 7, 230–238 (1969)
14. Cooper, S. M.: Subsets of complete intersections and the EGH conjecture. Progress in
Commutative Algebra 1, de Gruyter, Berlin 167–198 (2012)
15. Davis, E. D., Geramita A. V., Orecchia, F.: Gorenstein algebras and the Cayley– Bacharach
theorem. Proc. Amer. Math. Soc. 93, no. 4, 593–597 (1985)
16. Eisenbud, D., Green, M., Harris, J.: Higher Castelnuovo theory. Journées de Géométrie
Algébrique d’Orsay (Orsay, 1992), Astérisque 218, 187–202 (1993)
17. Eisenbud, D., Green, M., Harris, J.: Cayley-Bacharach theorems and conjectures. Bull. Amer.
Math. Soc. (N.S.) 33, no. 3, 295–324 (1996)
18. Francisco, C.: Almost complete intersections and the lex-plus-powers conjecture. J. Algebra
276, no. 2, 737–760 (2004)
19. Francisco, C. A., Richert, B. P.: Lex-plus-powers ideals. Syzygies and Hilbert functions, Lect.
Notes Pure Appl. Math. 254, 113–144 (2007)
20. Gasharov, V.: Hilbert functions and homogeneous generic forms II. Compositio Math. 116, no.
2, 167–172 (1999)
342 S. Güntürkün
21. Geramita, A., Kreuzer, M.: On the uniformity of zero-dimensional complete intersections. J.
Algebra, 391, 82–92 (2013)
22. Green, M.: Restrictions of linear series to hyperplanes, and some results of Macaulay and
Gotzmann. In Algebraic curves and projective geometry (Trento, 1988), Lecture Notes in Math.
1389, 76–86. Springer, Berlin (1989)
23. Greene, C., Kleitman, D.: Proof techniques in the theory of finite sets. Studies in combinatorics,
MAA Stud. Math., Math. Assoc. America 17, 22–79 (1978)
24. Gotzmann, G.: Eine Bedingung für die Flachheit und das Hilbertpolynom eines graduierten
Ringes. Math. Z. 158, 61–70 (1978)
25. Gunturkun, S., Hochster, M.: The Eisenbud-Green-Harris conjecture for defect two quadratic
ideals. Math. Res. Letters 27, no. 5, 1341–1365 (2020)
26. Harima, T., Wachi, A., Watanabe, J.: The EGH conjecture and the Sperner property of complete
intersections. Proc. Amer. Math. Soc. 145, no. 4, 1497–1503 (2017)
27. Herzog, J., Popescu, D.: Hilbert functions and generic forms. Compositio Math. 113, no. 1,
1–22 (1998)
28. Hulett, H.A.: Maximum betti numbers of homogeneous ideals with a given Hilbert function.
Comm. Algebra 21, no.7, 2335–2350 (1993)
29. Jorgenson, G.: Secant indices of projective varieties. Preprint arXiv:2003.08481 (2020)
30. Katona, G.: A theorem for finite sets. Theory of Graphs (P. Erdös and G. Katona, eds.),
Academic Press, New York 187–207 (1968)
31. Kruskal, J.: The number of simplices in a complex. Mathematical Optimization Techniques (R.
Bellman, ed.), University of California Press, Berkeley/Los Angeles 251–278 (1963)
32. Laplagne, S., Valdettaro, M.: Strictly positive polynomials in the boundary of the SOS cone.
Preprint arXiv:2012.05951 (2020)
33. Grayson, D. R., Stillman, M. E.: Macaulay2, a software system for research in algebraic
geometry. Available at http://www.math.uiuc.edu/Macaulay2/
34. Macaulay, F.: Some properties of enumeration in the theory of modular systems. Proc. London
Math. Soc. 26, 531–555 1927)
35. Mermin, J., Murai, S.: The lex-plus-powers conjecture holds for pure powers. Adv. Math. 226,
no. 4, 3511–3539 (2011)
36. Pardue, K.: Deformation classes of graded modules and maximal betti numbers. Illinois J.
Math. 40, no.4, 564–585 (1996)
37. Richert, B. P.: A study of the lex plus powers conjecture. J. Pure Appl. Algebra 186, no. 2,
169–183 (2004)
38. Richert, B. P., Sabourin, S.: The residuals of lex plus powers ideals and the Eisenbud-Green-
Harris conjecture. Illinois J. Math. 52, no. 4, 1355–1384 (2008)
The Variety Defined by the Matrix of
Diagonals is F -Pure
Zhibek Kadyrsizova
Keywords Frobenius · Singularities · F-purity · System of parameters
1 Introduction and Preliminaries
Let X = (xij )1i, j n be a square matrix of size n with indeterminate entries over
a field K and R = K[X] be the polynomial ring over K in {xij | 1 i, j n}.
Let D(X) be an n × n matrix whose j th column consists of the diagonal entries of
the matrix Xj −1 written from left to right with the convention that X0 is the identity
matrix of size n. Define P(X) = det(D(X)). In [8] and [9], H-W.Young studies the
varieties of nearly commuting matrices and derives their important properties such
as the decomposition into irreducible components through the use of the polynomial
P(X). We hope to understand better their Frobenius singularities and as it can be
seen in [7] this is also closely related to the singularities of the variety defined by
P(X). In this paper we prove that the latter is F -pure and find a homogeneous
system of parameters for it. Let us first define the necessary preliminaries. The
following notation is fixed for the rest of the paper. Let
⎡ ⎤
x11 x12 . . . x1n
X=⎣ ... ... ⎦
xn1 xn2 . . . xnn
and

X0 0
X̃ =
0 xnn
Z. Kadyrsizova ()
Department of Mathematics, Nazarbayev University, Nur-Sultan, Kazakhstan
e-mail: zhibek.kadyrsizova@nu.edu.kz
344 Z. Kadyrsizova
where
⎡ ⎤
x11 . . . x1,n−1
X0 = ⎣ ... ⎦.
xn−1,1 . . . xn−1,n−1
That is, X̃ is the matrix obtained from X by setting all the entries of its last column
and the last row to 0 except for the entry xnn . Let X0 be the matrix obtained from X
by deleting its last column and last row.
Lemma 1.1 ([9], Lemma 1.3) P(X) is an irreducible polynomial.

Lemma 1.2 ([9], Proof of Lemma 1.3) P(X̃) = P(X0 )cX0 (xnn ) where cX0 (t) is
the characteristic polynomial of X0 .

Remark 1.3 The above lemma is also true when we annihilate either only the last
column or only the last row of X with the exception of the entry xnn .

2 A System of Parameters
First, we find a homogeneous system of parameters on R/(P(X)). To do so we

prove several useful lemmas.
Lemma 2.1 Let A be a square matrix of size n with integer entries and with
det(A) = ±1. Then the following are equivalent
(a) P(A) = ±1.
(b) The diagonals of I, A, A2 , . . . , An−1 span Zn .
(c) The diagonals of the elements of Z[A, A−1 ] span Zn .
(d) There exist n integer powers of A with the property that their diagonals span
Zn .

Proof The equivalence of (a) and (b) is clear as the columns of the matrix D(A) are
the diagonals of I, A, A2 , . . . , An−1 .
By Cayley-Hamilton’s theorem we have that An ∈ n−1 i=0 ZA and
i hence I ∈
n −1
n−1
i=1 ZA . Therefore, since A is invertible, we also have that A ∈ i=0 ZAi as
i
−h −1
well as A for all integers h. Thus Z[A] = Z[A, A ]. Finally,
L M
det diag(I ) diag(A) . . . diag(An−1 ) = ±1 if and only if
diag(I ), diag(A), . . . , diag(An−1 ) span Zn if and only if
the diagonals of all the integer powers of A span Zn if and only if there exist n
integer powers of A with the property that their diagonals span Zn .

The Variety Defined by the Matrix of Diagonals is F -Pure 345
Lemma 2.2 Let

⎡ ⎤
1 1 ... 1 1
⎢1 1 ... 1 0⎥
A=⎢
⎣ ... ...
⎥.
⎦
1 0 ... 0 0
be a square matrix of size n 1 with the property that all the entries strictly below
the main anti-diagonal are 0 and the rest are equal to 1. Then P(A) is equal to
either 1 or -1.

Proof First observe that det(A) = ±1. Hence the matrix is invertible and it can be
shown that
⎡ ⎤
0 0 ... 0 1
⎢ 0 0 . . . 1 −1 ⎥
B = A−1 =⎢
⎣ ... ...
⎥.
⎦
1 −1 . . . 0 0
B has two non-zero anti-diagonals and the rest of the entries are equal to 0. To show
that P(A) = ±1 it is necessary and sufficient to show that there exist n powers of B
so that their diagonals span Zn , see Lemma 2.1. We claim that for this purpose it is
sufficient to take n odd powers of B.
Claim 2.3
⎡ ⎤
1 −1 0 ... 0 0
⎢ −1 −1 0 ⎥
⎢ 2 ... 0 ⎥
⎢ ⎥
B2 = ⎢ ... ... ⎥.
⎢ ⎥
⎣ 0 0 0 . . . 2 −1 ⎦
0 0 0 . . . −1 2
We show this by induction with the induction step equal to 2. Cases n = 2 and
n = 3 can be easily verified.
Write B as
⎡ ⎤
0 0...0 1
⎢0 −1 ⎥
⎢ ⎥
⎢ ⎥
⎢0 0 ⎥
⎢. .. ⎥
⎢. B0 ⎥
⎢. . ⎥
⎢ ⎥
⎣0 0 ⎦
1 −1 0 . . . 0 0
then
346 Z. Kadyrsizova
⎡ ⎤
1 −1 0 . . . . . . . . . 0 0 0
⎢ −1 0 ⎥
⎢
⎢ ⎡ ⎤ ⎥
⎥
⎢ 0 1 0 ... 0 0 ⎥
⎢ ⎥
⎢ ⎥
B =⎢
2
⎢ B02 + ⎣ 0 0 . . . 0 ⎦ ⎥.
⎥
⎢ . .. ⎥
⎢ . 0 0 ... 0 ⎥
⎢ . . ⎥
⎢ ⎥
⎣ 0 −1 ⎦
0 0 0.........0 − 1 2
Claim 2.4 For all 1 j n − 1, we have that

⎧
⎨ (−1)j +1 if k + l = n − j + 2,
2j −1
(B )kl = (−1)j if k + l = n + j + 1,
⎩
0 if k + l n − j + 1 or k + l n + j + 2.
We prove the claim by induction on j . When j = 1, it is true. Suppose that the

claim is true for all integers less than or equal to j . Then
B 2(j +1)−1 = B 2j +1 = B 2 B 2j −1 .
Therefore,

n
(B 2j +1 )kl = (B 2 B 2j −1 )kl = (B 2 )ks (B 2j −1 )sl =
s=1
For now assume that k > 1.
(B 2 )kk (B 2j −1 )kl + (B 2 )k,k+1 (B 2j −1 )k+1,l + (B 2 )k,k−1 (B 2j −1 )k−1,l .
Suppose first that k + l = n − j + 1. Then
(B 2j +1 )kl = (B 2 )kk · 0 + (−1)(−1)j +1 + (−1) · 0 = (−1)j +2 .
Next, suppose that k + l = n + j + 2
(B 2j +1 )kl = (B 2 )kk · 0 + (−1) · 0 + (−1)(−1)j = (−1)j +1 .
Finally, consider the case when k + l n − j or k + l n + j + 3. We have that
(B 2j +1 )kl = (B 2 )kk · 0 + (−1) · 0 + (−1) · 0 = 0.
Thus the formula for B 2j −1 is true.

Remark 2.5 The case k = 1 goes along the same lines as above with the exception
when we have that 1 + l = n + j + 2. Then l = n + j + 1 > n and this is not
possible.

Now we are ready to finish the proof of the lemma. Consider the matrix whose
columns are the diagonals of the odd powers of B written from left to right. Observe
that in each odd power the main diagonal meets only one of the sub-anti-diagonals
that we highlighted above. Therefore, we have that
L M
diag(B) diag(B 3 ) . . . diag(B 2j −1 ) . . . diag(B 2n−1 ) =
for matrices of odd sizes and for matrices of even sizes

⎡0 0 0 0 0 0 0 0 1⎤
∗ ∗⎥ ⎡ 0 −1 ⎤
⎢0 0 0 0 0 0 1 0 0 0 0 0 0
⎢0 0 ∗ ∗ ∗ ∗⎥ −1 ∗ ∗ ⎥
⎢ 0 0 0 ⎥ ⎢ 0 0 0 0 0
⎢ ⎥ ⎢ 0 ∗ ∗ ∗ ∗ ⎥
⎢ ... ... ⎥ ⎢ 0 0 0 ⎥
⎢0 0 ∗ ∗ ∗ ∗⎥ ⎢ ⎥
⎢ 0 0 1 ⎥ ⎢ ... ... ⎥
⎢0 0 1 ∗ ∗ ∗ ∗ ∗ ∗⎥ ⎢ 0 0 0 −1 ∗ ∗ ∗ ∗ ⎥
⎢ ⎥ ⎢ ⎥
⎢1 ∗ ∗ ∗ ... ∗ ∗ ∗ ∗⎥ ⎢ 0 −1 ∗ ∗ ∗ ∗ ∗ ∗ ⎥
⎢ ⎥, ⎢ ⎥.
⎢0 1 ∗ ∗ ∗ ∗ ∗ ∗ ∗⎥ ⎢ −1 ∗ ∗ ∗ ∗ ∗ ∗ ∗ ⎥
⎢ ⎥ ⎢ ⎥
⎢0 0 0 1 ∗ ∗ ∗ ∗ ∗⎥ ⎢ 0 0 −1 ∗ ∗ ∗ ∗ ∗ ⎥
⎢ ⎥ ⎢ ⎥
⎢ ... ... ⎥ ⎢ ... ... ⎥
⎢ ⎥ ⎢ ⎥
⎢0 0 0 0 0 ∗ ∗ ∗ ∗⎥ ⎢ 0 0 0 0 ∗ ∗ ∗ ∗ ⎥
⎣ ⎦ ⎣ ⎦
0 0 0 0 0 1 ∗ ∗ ∗ 0 0 0 0 −1 ∗ ∗ ∗
0 0 0 0 0 0 ∗ 1 ∗ 0 0 0 0 0 ∗ −1 ∗
Since every row of the matrices has a pivot position, we conclude that the
determinants of these matrices are not zero and thus the columns of each of them
span Zn . This finishes the proof of the lemma.

Theorem 2.6 Let X = (xij )1i, j n be a square matrix of size n with indeterminate
entries over a field K and R = K[X] be the polynomial ring over K in {xij | 1
i, j n}. Let
= {(i, j ) |1 i n, n + 1 − i j n}
and
= {(k, l) |1 k n − 1, 1 l n − k, (k, l) = (1, 1)} .
Then
J K
S = xij , x11 − xkl | (i, j ) ∈ , (k, l) ∈
is a homogeneous system of parameters and hence a regular sequence on R/(P(X)).

Proof We prove the lemma by induction on n.
Consider the first few small cases.
Let n = 2. In this case modulo (S) we have that
348 Z. Kadyrsizova

x 0
X = 11
0 0
and

1 x11
P(X) = det = −x11 .
1 0
Let n = 3. In this case

⎡ ⎤ ⎡ 2 2 ⎤
x11 x11 0 2x11 x11 0
X = ⎣ x11 0 0 ⎦ , X2 = ⎣ x11
2 x2 0 ⎦
11
0 0 0 0 0 0
and
⎡ 2 ⎤
1 x11 2x11
P(X) = det ⎣ 1 0 x11 2 ⎦ = x3 .
11
1 0 0
Let n = 4. In this case

⎡ ⎤ ⎡ 2 2 2 ⎤
x11 x11 x11 0 3x11 2x11 x11 0
⎢ x11 x11 0 0⎥⎥ ⎢ 2 2
2x11 2x11 2
x11 0⎥
X=⎢
⎣ x11 ⎦ , X2 = ⎢
⎣
⎥,
0 0 0 2 2
x11 x11 2
x11 0⎦
0 0 0 0 0 0 0 0
⎡ 3 5x 3 3x 3 ⎤ ⎡ 2 6x 3 ⎤
6x11 11 11 0 1 x11 3x11 11
⎢ 5x 3 4x 3 2x 3 0⎥ ⎢1 2 4x 3 ⎥
x11 2x11
⎢
X =⎣ 3 ⎥ , and P(X) = det ⎢ 11 ⎥
3 ⎦ = −x11 .
3 11 11 11 6
3 x3
3x11 2x11 11 0⎦ ⎣1 0 x112 x11
0 0 0 0 1 0 0 0
n(n−1)/2
Claim 2.7 P(X) = ±x11 in the quotient ring of R by the ideal generated by
{xij , x11 − xkl | (i, j ) ∈ , (k, l) ∈ }
Remark 2.8 In our proof of the claim we do not establish exactly the sign of the
image of P(X). We only show that it is either 1 or -1.

Consider the matrices X, X0 , X̃, and set the elements strictly below the main anti-
diagonal of X0 and xnn to 0, that is,
⎡ ⎤
x11 x12 . . . x1,n−2 x1,n−1 0
⎢ x 0⎥
⎢ 21 x22 . . . x2,n−2 0 ⎥
⎢ ⎥
X̃˜ = ⎢ ... ... 0 0⎥.
⎢ ⎥
⎣ xn−1,1 0 ... 0 0 0⎦
0 0 ... 0 0 0
Since P (X̃) = P (X0 )cX0 (xnn ) where cX0 is the characteristic polynomial of X0
(see Lemma 1.2), we have that P (X̃) ˜ = P (X̃ )(−1)(n−2)(n−1)/2 3n−1 x . Set
0 i=1 i,n−i
(n−1)(n−2)/2
the elements x11 − xkl to zero for all (k, l) ∈ . Then P (X̃0 ) = x11 P (Y0 ),
where
⎡ ⎤
1 1 ... 1 1
⎢1 1 ... 1 0⎥
Y0 = ⎢
⎣ ... ...
⎥.
⎦
1 0 ... 0 0
Hence
˜ = x (n−1)(n−2)/2 P (Y )(−1)(n−2)(n−1)/2 x = ±x n(n−1)/2 P (Y ).

n−1
P (X̃) 11 0 11 11 0
i=1
By Lemma 2.2 we have that P (Y0 ) = ±1, which finishes the proof of the claim.
Finally, we have that R/(P(X), S) ∼
n(n−1)/2
= K[x11 ]/(x11 ), which has Krull
dimension 0. Hence, S is indeed a system of parameters on R/(P(X)) and, since
the ring is a complete intersection, it is also a regular sequence.

3 The Variety Defined by P(X) Is F -Pure
Definition 3.1 Let S be a ring with positive prime characteristic p. Then S is called
F -pure if the Frobenius endomorphism F : S → S with F (s) = s p is pure, that is,
for every S-module M we have that F ⊗S idM : S ⊗ M → S ⊗ M is injective.
Next is Fedder’s criterion specialized for hypersurfaces and we use it to prove F -
purity of R/(P(X)).
Lemma 3.2 (Fedder’s Criterion, [2], Proposition 2.1) Let S be a polynomial ring
over a field K of positive prime characteristic p and let f ∈ S be a homogeneous
polynomial. Then S/(f ) is F -pure if the polynomial f p−1 has a non-zero monomial
term in which every indeterminate has degree at most p − 1.

We also need the fact that F -purity deforms for Gorenstein rings.
350 Z. Kadyrsizova
Lemma 3.3 ([2], Theorem 3.4(2)) Let S be a Gorenstein ring and let x ∈ S be a
non-zero divisor. Then if S/xS is F -pure, then so is S.

For more examples of F -pure rings and other related notions the reader may refer
to [1, 3, 4, 6]. Computer algebra system Macaulay2, [5], is a great tool in studying
rings and their properties.
Let S = {xij : 1 i n, n − i + 1 j n} ⊆ S. These are the
entries of the matrix X on and below the main anti-diagonal. It has been shown in
the previous section that S is part of a homogeneous system of parameters and a
regular sequence on R/(P(X)).
Theorem 3.4 Let K be a field of positive prime characteristic p. Then R/(P(X))
is F -pure for all n 1.

Proof We use the fact that F -purity deforms for Gorenstein rings, [2], and we have
that R/(P(X)) is a complete intersection and hence is Gorenstein. Therefore, it is
sufficient to show that we have an F -pure ring once we take the quotient of the ring
R/(P(X)) by the ideal generated by the regular sequence S .
Let us first take a look at what we have for few small values of n.
Case n = 1 is trivial. P(X) = 1 and R/(P(X))
= 0.
1 x11
Case n = 2: P(X) = det = x22 − x11 . Then R/(P(X)) ∼ =
1 x22
K[x11 , x12 , x21 ] is regular and hence F -pure.
Now we are ready to prove the general statement. We do it by induction on n.
First observe the following: if our statement is true for a fixed n, that is, if
R/(P(X), S ) is F -pure, then so is R/(P(X), S0 ), where
S0 = {xij : 2 i n, n − i + 2 j n} ⊆ S ,
that is, the entries strictly below the main anti-diagonal. Moreover, if a particular
monomial term of P(X) has a nonzero coefficient in R/(S ), then it is a monomial
term of P(X) with a nonzero coefficient in R/(S0 ). A partial converse is also true.
If P(X) in R/(S0 ) has a nonzero monomial term in the entries of X which are
strictly above the main anti-diagonal, then so does P(X) in R/(S ).
The first non-trivial case is n = 3. Let X̂ be the matrix X modulo the elements
of S0 , that is,
⎡ ⎤
x11 x12 x13
X̂ = ⎣ x21 x22 0 ⎦
x31 0 0
and
⎡ 2 +x x x x ⎤
1 x11 x11 12 21 13 31
P (X̂) = det ⎣ 1 x22 2 +x x
x22 12 21
⎦=
1 0 x13 x31
= −x11 x13 x31 + x11 x12 x21 − x12 x21 x22 + x11 x22
2
− x11
2
x22 .
Since P (X̂) has a monomial term x11 x12 x21 with a coefficient 1 modulo p, so
does P(X) in R/(S ).
Here is our induction hypothesis: for all k < n we have that P(X)p−1 in R/(S )
3 3k−i p−1
and in R/(S0 ) has a monomial term k−1 i=1 j =1 xij with coefficient ±1 modulo
p. In other words, this monomial term is the (p − 1)st power of the product of
the entries of X strictly above the main anti-diagonal. The basis of the induction is
verified above.
Now consider the matrices X0 , X̃ and X̃. ˜ As in the Theorem 1, since P (X̃) =
P (X0 )cX0 (xnn ) where cX0 is the characteristic polynomial of X0 (see Lemma 1.2)
˜ = P (X̃ )(−1)(n−2)(n−1)/2 3n−1 x
we have that P (X̃) 0 i,n−i . By induction hypothesis
3 3n−1−i i=1 p−1
P (X̃0 )p−1 has a monomial term n−2 i=1 j =1 xij with coefficient ±1.
3 3
˜ p−1 has a monomial term n−1 n−i x p−1 with coefficient ±1.
Hence P (X̃) i=1 j =1 ij
Therefore, by Fedder’s criterion we have that R/(P(X), S ) is F -pure and thus so
is R/(P(X)).

The next natural question that one can consider is whether the variety defined by
the polynomial P(X) is F -regular. It is certainly true when n = 1 and n = 2, but is
unknown for larger values of n.
Conjecture 3.5 Let K be a field of positive prime characteristic p. Then R/(P(X))
is F -regular for all n 1.

Acknowledgments The author is grateful to Mel Hochster for valuable discussions and comments
and thanks the referee for useful suggestions on improving the paper.
References
1. W. Bruns and J. Herzog, Cohen-Macaulay rings, second edition ed., Cambridge Univ. Press,
Cambridge, UK, 1998.
2. R. Fedder, F-purity and rational singularity, Trans. Amer. Math. Soc 278 (1983), no. 2, 461–
480.
3. , F-purity and rational singularity in graded complete intersection rings, Transactions
of the American Mathematical Society 301 (1987), no. 1, 47–62.
4. R. Fedder and K. Watanabe, A characterization of F-regularity in terms of F-purity, Com-
mutative Algebra, vol. 15, Math. Sci. Res. Inst., Berlin Heidelberg New York Springer, 1989,
pp. 227–245.
5. D. Grayson and M. Stillman, Macaulay 2: a computer algebra system for algebraic geometry
and commutative algebra, available at http://www.math.uiuc.edu/Macaulay2.
6. M. Hochster and J. L. Roberts, The purity of the Frobenius and local cohomology, Adv. in
Math. 21 (1976), 117–172.
7. Z. Kadyrsizova, Nearly commuting matrices, Journal of Algebra 497 (2018), 199–218.
8. Hsu-Wen Vincent Young, Components of algebraic sets of commuting and nearly commuting
matrices, Ph.D. thesis, University of Michigan, 2011.
9. , On matrix pairs with diagonal commutators, Journal of Algebra 570 (2021), 437 –
451.
Classification of Frobenius Forms in Five
Variables
Zhibek Kadyrsizova, Janet Page, Jyoti Singh, Karen E. Smith, Adela Vraciu,
and Emily E. Witt
Keywords Frobenius form · Quadratic form · Positive characteristic · Threefold
1 Introduction
Fix a field k of prime characteristic p. A Frobenius form over k is a homogeneous

polynomial in indeterminates x1 , . . . , xn of the form
pe pe pe
x1 L1 + x2 L2 + · · · + xn Ln (1)
where each Li is some linear form and e is a positive integer. Put differently, a
Frobenius form is a homogeneous polynomial of degree pe + 1 that is in the ideal
pe pe
generated by x1 , . . . , xn .
Frobenius forms can be compared to quadratic forms: if we allow e = 0 in the
expression (1) above, we get a quadratic form. Quadratic forms are well-studied in
the classical literature. For example, much is known about the geometry of quadric
Z. Kadyrsizova
Department of Mathematics, Nazarbayev University, Nur-Sultan, Kazakhstan
e-mail: zhibek.kadyrsizova@nu.edu.kz
J. Page · K. E. Smith
University of Michigan, Ann Arbor, MI, USA
e-mail: jrpage@umich.edu; kesmith@umich.edu
J. Singh
Visvesvaraya National Institute of Technology, Nagpur, India
e-mail: jyotijagrati@gmail.com
A. Vraciu
University of South Carolina, Columbia, SC, USA
e-mail: vraciu@math.sc.edu
E. E. Witt ()
Department of Mathematics, University of Kansas, Lawrence, KS, USA
e-mail: witt@ku.edu
354 Z. Kadyrsizova et al.
hypersurfaces, and, at least over a quadratically closed field, their classification up

to linear changes of coordinates is well-known. Both admit a convenient matrix
factorization, where the action of changing coordinates behaves similarly.
One special but very interesting case of Frobenius forms are those of degree
three (which are necessarily defined over a field of characteristic two). A smooth
cubic surface is always Frobenius split, it turns out, unless it is defined by a
Frobenius form; in particular, non-Frobenius split (smooth) cubic surfaces exist
only over fields of characteristic two [8, 5.5]. A detailed examination of non-
Frobenius split cubic surfaces of characteristic two, including the non-smooth ones,
was undertaken in [10]. In particular, the Frobenius forms of degree three in up
to four variables are classified there, up to projective equivalence. In this paper,
we extend that classification to arbitrary Frobenius forms in up to five variables.
Put differently, we classify the projective equivalence classes of three-dimensional
projective hypersurfaces defined by Frobenius forms—a class called extremal three-
folds in [11]. Section 4 describes this classification in detail, including a particularly
“sparse” equation representing each of the seven types of projective equivalence
classes of extremal three-folds.
Frobenius forms and the projective schemes they define have various “extreme”
properties, both algebraically and geometrically. For example, cubic surfaces
defined by Frobenius forms are characterized by the geometric property that
they “contain no triangles”—that is, any plane section consisting of three lines
must contain a point on all three [10, 5.1]. Analogous extremal configurations
of linear subvarieties occur more generally for extremal hypersurfaces of higher
degree and dimension; see [11, §8]. Algebraically, reduced Frobenius forms can
be characterized as those achieving the minimal possible F -pure threshold among
reduced forms of the same degree [11, 1.1].
2 Matrix Factorization of Frobenius Forms
Fix a field k of prime characteristic p. Let q denote an integral positive power of p.

The beauty of Frobenius forms is that, like quadratic forms, they admit a matrix
factorization: a Frobenius form h in n variables can be written as
⎡ ⎤
x1
L e e M ⎢ ⎥
pe ⎢x2 ⎥
h = x1p x2p · · · xn A ⎢ .. ⎥
⎣.⎦
xn
for some n × n matrix A with entries in k. Because e > 0, the matrix A representing
the Frobenius form h is unique.
When e = 0, we recover the case of quadratic forms. In the quadratic form case,
of course, A is not unique, but we can force uniqueness, for example, by insisting
that A be symmetric (when p = 2).
Classification of Frobenius Forms in Five Variables 355
There is an interesting story of Frobenius forms with “conjugate symmetric”

matrices, where conjugacy is defined using a Frobenius automorphism. By defi-
nition, a Hermitian Frobenius form over k is a Frobenius form whose matrix A
satisfies A1 = A[q] for some q = pe (in particular, such a matrix is defined
over Fq 2 ). This special class of Frobenius forms gives a characteristic p analog
of Hermitian forms over the complex numbers, and has been studied, for example,
in [1, 9, 15] and [12, §35]. Like quadratic forms over quadratically closed fields,
there is exactly one non-degenerate Hermitian Frobenius form, up to projective
equivalence, in each dimension [1, 4.1] over Fq 2 . As we will see, the classification
of more general Frobenius forms is more complicated.
To classify Frobenius forms up to projective equivalence, we need to understand
how linear changes of coordinates act on them.
Let g ∈ GLn (k) be any linear change of coordinates for the polynomial ring
k[x1 , . . . , xn ]. We represent the action of g on the variables as
⎡ ⎤ ⎡ ⎤
x1 x1
⎢x2 ⎥ ⎢x2 ⎥
⎢ ⎥ ⎢ ⎥
⎢ . ⎥ → g ⎢ . ⎥ ,
⎣ .. ⎦ ⎣ .. ⎦
xn xn
the usual matrix multiplication. In particular, when g acts1 on the Frobenius form h,
we have
⎛ ⎡ ⎤⎞ ⎡ ⎤
x1 x1
⎜L e M ⎢ x ⎥⎟ L e M ⎢ ⎥
⎜ e pe ⎢ 2 ⎥⎟ e pe [pe ] 1 ⎢x2 ⎥
g · h = g · ⎜ x1p x2p · · · xn A ⎢ . ⎥⎟ = x1p x2p · · · xn (g ) Ag ⎢ . ⎥ ,
⎝ ⎣ .. ⎦⎠ ⎣ .. ⎦
xn xn
where g [p ] denotes the matrix whose entries are the pe -th powers of the entries of g
e
and B 1 indicates the transpose of a matrix B. Thus, the action of g on the Frobenius
form h transforms the matrix A representing h into the matrix (g [p ] )1 Ag.
e
A Frobenius form is said to be nondegenerate if it cannot be written, after a

linear change of coordinates, in a smaller number of variables, and is otherwise
degenerate, similarly as for quadratic forms. We define its embedding dimension
to be the smallest number of variables needed to write f up to linear change of
coordinates.
The rank of a Frobenius form is defined as the rank of the representing matrix,
just as the rank of a quadratic form is the rank of the corresponding symmetric
matrix when p = 2. The rank of a Frobenius form is invariant under changes of
1 We write “g · h” to indicate the group action on polynomials, whereas adjacency, “gB,” will
indicate the usual matrix multiplication on an n × m matrix B.
coordinates, since the rank of a matrix is unchanged by multiplication on the left

and right by invertible matrices.
The rank of a Frobenius form is equal to the codimension of the singular locus
of the corresponding hypersurface [11, 5.3]. This is analogous to the corresponding
statement about quadratic forms of non-even characteristic, when we work with the
corresponding symmetric matrix.
There is precisely one full rank Frobenius form, up to change of coordinates,
over an algebraically closed field, in each fixed degree pe + 1 and embedding
dimension n:
Theorem 2.1 ([11, 6.1][2]) Every full rank Frobenius form over an algebraically
closed field k of characteristic p > 0 is represented, in suitable linear coordinates,
q+1 q+1
by the diagonal form x1 + · · · + xn , where q is an integral power of p.

Put differently, every smooth projective hypersurface defined by a Frobenius
q+1 q+1 q+1
form is projectively equivalent to one defined by x1 + x2 + · · · + xn . This
is analogous to the situation for quadratic forms over a quadratically closed field
of characteristic not two, though the proof is a bit more involved. In particular, we
do not have a complete understanding of the situation over non-closed fields, an
interesting open problem. We have no counterexample to the speculation that the
full rank Frobenius forms may be “diagonalizable” over a perfect field closed under
all degree pe extensions.
In the non-full rank case, there are finitely many projective equivalence classes of
Frobenius forms in each fixed degree and embedding dimension [11, 7.4]. Indeed,
the number of non-degenerate Frobenius forms of embedding dimension n (of fixed
degree) is bounded above by the n-th Fibonacci number. However, the paper [11]
stopped short of precisely classifying the Frobenius forms in each dimension, a task
essentially completed for Frobenius forms in four variables in [10]. The case of five
variables is treated here in Sect. 5. For the statement, see Sect. 4.
3 Quadratic Forms
To complete our story, we recall the classification of quadratic forms:

Proposition 3.1 Let f be a non-degenerate quadratic form in n variables over a
quadratically closed2 field. If the characteristic of k is two, then f is projectively
equivalent to either
1. x1 x2 + x3 x4 + · · · + xn−2 xn−1 + xn2 if n is odd, or
2. x1 x2 + x3 x4 + · · · + xn−3 xn−2 + xn−1 xn if n is even.
2 By quadratically closed, we mean that every degree two polynomial over k splits. In characteristic
two, this is a stronger assumption than requiring that the field contain the square root of every
element.
If the characteristic of k is not two, then f is projectively equivalent to
x12 + x22 + · · · + xn2 .
In particular, over a quadratically closed field, there is exactly one quadratic form
in each embedding dimension.

This classification is well known, but since we could not find a low-tech proof in
the modern literature for the case p = 2, we include one here. Alternate discussions
using more machinery can be found, for example, in [5, I.16], [7, 12.9] or [6, 7.32],
which also contain more refined classifications over non-closed fields. A lower-tech
proof can be found in the classic book of [4], but for the convenience of the reader
we include a direct and concise proof here.3
Proof The only quadratic form in one variable is x12 . Likewise, the two-variable
case is trivial: a degree two form in two variables must factor into two linear forms
over a quadratically closed field, so in suitable coordinates, the form is either x1 x2
or x12 (which is degenerate).

Case of Characteristic Not Two
It is straightforward to check (even without closure assumptions on k) that a suitable
choice of linear change of coordinates puts f in the form λ1 x12 + · · · + λn xn2 , where
the λi are nonzero (e.g., see [13]). So, if the ground field is quadratically closed, the
change of coordinates taking each xi → √1 xi normalizes the form to x12 +· · ·+xn2 .
λi
Case of Characteristic Two

Say that n 3. Since f is non-degenerate, it is not the square of a linear form. Thus
some square-free term, which we can assume to be x1 x2 , appears with nonzero
coefficient. Scaling, we may assume the coefficient of x1 x2 is 1.
Now write f in the form

n
n
L2 + x1 x2 + a1j x1 xj + a2j x2 xj + h1 (x3 , . . . , xn ) (2)
j =3 j =3
where L is a (possibly zero) linear form in x1 , x2 , and h1 is a quadratic form in

x3 , . . . , xn . Apply the linear change of coordinates sending x2 to x2 + nj=3 a1j xj ,
fixing the other variables. This transforms (2) into an expression which can be
written
3 While the stated classification of quadratic forms in characteristic two is well known, it appears
that an elementary proof is not. In his history of quadratic forms [14], Scharlau laments that
Dickson’s 1899 work [3], which is elementary but “rather involved,” is not better known, stating
“However, one must admit, that this paper—like most of Dickson’s work—is not very pleasant
to read. It is entirely algebraic.” We hope the reader will find our straightforward and entirely
algebraic proof more pleasant to read.

n

L2 + x1 x2 + a2j x2 xj + h2 (x3 , . . . , xn ), (3)
j =3
where again h2 is a quadratic form in x3 , . . . , xn .

nWith another change of coordinates, we may assume the summand
is nonzero, then we can assume a = 1
a
j =3 2j x2 xj is zero. Indeed, if some a2j 23
after renumbering and nscaling if necessary. Now apply the linear transformation
sending x3 to x3 + j =4 a2j x , fixing the other variables. This transforms the
j
expression (3) into
L2 + (x1 + x3 )x2 + h3 (x3 , . . . , xn ), (4)
where h3 is quadratic in x3 , . . . , xn . Next, the coordinate change taking x1 to x1 +x3 ,

fixing the other variables, transforms (4) into the form
L2 + x1 x2 + h4 (x3 , . . . , xn ), (5)
where h4 is quadratic.
Finally, by induction, we separately apply linear changes of coordinates to the
quadratic L2 + x1 x2 in {x1 , x2 } and the quadratic h4 in {x3 , . . . , xn } to put each into
the desired form. It is easy to see, then, that their sum has the desired form as well,
depending on the parity of n in the stated way. This completes the proof.

Remark 3.2 Our proof easily adapts to show the well-known basic fact that a
quadratic form over an arbitrary field of characteristic two is a sum of binary
quadratics in distinct variables (plus a quadratic in one variable if the embedding
dimension is odd). Alternatively, our proof adapts to prove Theorem 199 in [4] over
any perfect field of characteristic two.
Remark 3.3 Quadratic forms behave like Frobenius forms from the point of view
of achieving the minimal F -pure threshold. For a reduced form of degree d, it is
1
proved in [11, 1.1] that the F -pure threshold is at least d−1 , with equality if and
only if the form is a Frobenius or quadratic form. Another way in which Frobenius
forms and quadratic forms are similar is that the corresponding hypersurfaces both
contain many high-dimensional linear subvarieties; see [11, §8].
4 Frobenius Forms in Five Variables: Classification

Statements
Fix an algebraically closed field k of positive characteristic p. Let q be an integral

positive power of p.
Theorem 4.1 There are seven projective equivalence classes of Frobenius forms
of a fixed degree q + 1 and embedding dimension five. Specifically, these are
represented by the following forms:
q+1 q+1 q+1 q+1 q+1
1. x1 + x2 + x3 + x4 + x5 (rank 5)
q q q+1 q
2. x1 x5 + x2 x4 + x3 + x4 x2 (rank 4)
q q q+1 q
3. x1 x5 + x2 x4 + x3 + x4 x1 (rank 4)
q q q q
4. x1 x5 + x2 x4 + x3 x2 + x4 x1 (rank 4)
q q q q
5. x1 x5 + x2 x3 + x3 x2 + x4 x1 (rank 4)
q q q+1
6. x1 x5 + x2 x4 + x3 (rank 3)
q q q
7. x1 x5 + x2 x4 + x3 x2 (rank 3)
For completeness, we also describe degenerate Frobenius forms in five variables
in the following two theorems. The proofs are the same as in [10], although that
source considered only cubic Frobenius forms.
Theorem 4.2 There are five projective equivalence classes of Frobenius forms of a
fixed degree q +1 and embedding dimension four. Specifically, these are represented
by the following forms:
q+1 q+1 q+1 q+1
1. x1 + x2 + x3 + x4 (rank 4)
q q+1 q
2. x1 x4 + x2 + x3 x1 (rank 3)
q q q
3. x1 x4 + x2 x3 + x3 x1 (rank 3)
q q q
4. x1 x4 + x2 x3 + x3 x2 (rank 3)
q q
5. x1 x4 + x2 x3 (rank 2)
Theorem 4.3 There are three projective equivalence classes of nondegenerate
Frobenius forms in three variables in each fixed degree, represented by precisely
one of the following forms:
q+1 q+1 q+1
1. The diagonal form x1 + x2 + x3 (rank 3)
q q+1
2. The cuspidal form x1 x3 + x2 (rank 2)
q q
3. The reducible form x1 x3 + x2 x1 (rank 2)
Moreover, in two variables, each Frobenius form is projectively equivalent to
exactly one of the following:
q−1 q−1
1. The form x1 x2 (x1 +x2 ), defining q +1 distinct points 0, ∞ and the (q −1)-st
roots of unity in P1 .
q
2. The form x1 x2 , defining the union of a q-fold point and a reduced point.
q+1
3. The form x1 , defining a (q + 1)-fold point.
5 Classification of Frobenius Forms in Five Variables: Proofs
In this section, we prove Theorem 4.1.

For this, we recall the method in [11] for showing that there are only finitely many
Frobenius forms of any fixed degree and embedding dimension, up to projective
equivalence. A key point is that a Frobenius form is equivalent to one represented
by a “sparse matrix:”
Theorem 5.1 ([11]) 4 Fix any algebraically closed field of characteristic p > 0.
A Frobenius form of embedding dimension n and rank r can be represented by a
matrix A with the following properties:
1. All rows beyond the r-th are zero.
2. All columns beyond the n-th are zero.
3. There are exactly r nonzero entries (all of which are 1) occurring in positions
(1 j1 ), (2 j2 ), . . . , (r jr ), where j1 > j2 > · · · > jr .
In particular, we may assume that all columns of A are zero but for r of them,
which are the standard unit basis vectors er , . . . , e1 (in that order, and possibly
interspersed with zero columns). For example, any full rank Frobenius form is
projectively equivalent to a Frobenius form whose matrix is the anti-diagonal
matrix, that is, whose columns are en , . . . , e1 . A Frobenius form whose matrix
satisfies the three conditions of Theorem 5.1 will be called a sparse form.
Theorem 5.1 implies the following bounds:
Corollary 5.2
1. The rank of a Frobenius form of embedding dimension n is at least n2 .
2. The number of non-degenerate n × n matrices of rank r satisfying the three
r !
conditions in Theorem (5.1) is n−r .
3. The number of projective equivalence types of Frobenius! forms of rank r and
r
embedding dimension n (and fixed degree) is at most n−r .
Proof Choosing a sparse matrix to represent the Frobenius form, we can assume it
looks like
pe pe
x1 L1 + · · · + xr Lr ,
where L1 , . . . , Lr ∈ {x1 , . . . , xn } are variables that appear in reverse order, each

variable appearing at most once. All the variables xr+1 , . . . , xn must appear in the
list L1 , . . . , Lr (otherwise the embedding dimension is less than n). In particular,
n − r r. This proves (1).
4 An older version of [11] contained the statement of Theorem 5.1 explicitly; observe now that
it follows easily from Lemma 5.5 in the new version of [11]. The updated version of [11], in
fact, classifies Frobenius forms in all dimensions ([11, Thm. 7.1]), so [11] gives a new proof of
Theorem 4.1 as well.
For (2), we continue by observing that conditions (1) and (3) of sparseness
(Theorem 5.1) force
L1 = xn , L2 = xn−1 , . . . , Ln−r = xr+1 .
For the remaining linear forms {Ln−r+1 , . . . , Lr }, we can choose 2r − n variables

out of the remaining variables {x1 , . . . , xr }. There are

r r
=
2r − n n−r
such choices. So (3) follows as well.

It is easy to determine the embedding dimension of a sparse Frobenius form:
Lemma 5.3 A rank r Frobenius form of the type
q q q
x1 xj1 + x2 xj2 + · · · + xr xjr (6)
has embedding dimension equal to the number of distinct variables appearing in the
expression (6).

Proof Suppose that a Frobenius form f of the type (6) involves n variables and has
rank r. If its embedding dimension is not n, then f could be written as a polynomial
in n − 1 independent linear forms, y1 , . . . , yn−1 . Without loss of generality, we can
assume that the yi have the following very special property: there is an index j such
that every yi is a binomial linear form x
i + ai xj for some index
i = j and scalar
ai . To see this, write
⎡ ⎡ ⎤ ⎤
y1 x1
⎢ y2⎥ ⎢x2 ⎥
⎢ ⎥ ⎢ ⎥
⎢ ⎥=B⎢ . ⎥
..
⎣ ⎦
. ⎣ .. ⎦
yn−1 xn
where B is an (n − 1) × n matrix of full rank, and then left-multiply by the inverse

in GL(n − 1) of a full rank (n − 1) × (n − 1) submatrix of B. This replaces
{y1 , . . . , yn−1 } by a set of linear forms spanning the same space and with the desired
binomial form.
There are two cases to consider, depending on whether or not j ∈ {1, . . . , r}.
If j r, then without loss of generality j = 1, so that yi = ai x1 + xi+1 for each
i = 1, . . . , n − 1. Now if f is a Frobenius form in y1 , . . . , yn−1 , then
q q q q q q
x1 xj1 + x2 xj2 + · · · + xr xjr = y1 L1 + y2 L2 + · · · + yn−1 Ln−1 (7)
for some linear forms Li in the yi . Because the only terms on the right side of
q q q q
(7) that involve xr+1 , . . . , xn come from yr , . . . , yn−1 , respectively, it follows that
Lr = · · · = Ln−1 = 0. In this case,
q q q
f = y1 L1 + y2 L2 + · · · + yr−1 Lr−1 ,
so that f has rank less than r, contrary to our hypothesis.

If j > r, then without loss of generality j = n, so that yi = xi + ai xn for each
i = 1, . . . , n − 1. Now assuming
q q q q q q
x1 xj1 + x2 xj2 + · · · + xr xjr = y1 L1 + y2 L2 + · · · + yn−1 Ln−1 (8)
for some linear forms Li in the yj , again it follows that Lr+1 = · · · = Ln−1 = 0
q
by looking at the xi terms, with i > r. Moreover, since the only terms on the right
q q q q
side of (8) that involve x1 , . . . , xr come from y1 , . . . , yr , we see that Li = xji
for i r. But note that xn must appear among the variables xj1 , . . . , xjr , since the
original form f involves all n variables. So xn is one of the Li . This says that xn is
a linear combination of x1 + a1 xn , . . . , xn−1 + an−1 xn , a contradiction.
Combining the two cases, we conclude that f is not a form in y1 , . . . , yn−1 , and
so the embedding dimension of f is n.

Remark 5.4 The total number of projective equivalence classes of Frobenius forms
of embedding dimension n is bounded above by the n-th Fibonacci number [11,
7.4].5 This follows from Corollary 5.2 (3) simply by adding the bounds for each
relevant rank. This upper bound is sharp for n 4, but not in general. There
are distinct sparse matrices that define equivalent Frobenius forms starting in five
variables:

Example 5.4.1 Consider the Frobenius forms corresponding to the following matri-
ces:
⎡ ⎤ ⎡ ⎤
00001 00001
⎢0 0 0 1 0⎥ ⎢0 0 0 1 0⎥
⎢ ⎥ ⎢ ⎥
⎢ ⎥ ⎢ ⎥
⎢0 1 0 0 0⎥ and ⎢1 0 0 0 0⎥
⎢ ⎥ ⎢ ⎥
⎣0 0 0 0 0⎦ ⎣0 0 0 0 0⎦
00000 00000
q q q q q q
namely, x1 x5 + x2 x4 + x3 x2 and x1 x5 + x2 x4 + x3 x1 . These are rank three non-
degenerate Frobenius forms corresponding to distinct matrices satisfying the three
conditions of Theorem 5.1, but they are equivalent by the change of coordinates
which swaps x1 ↔ x2 and x4 ↔ x5 , but fixes x3 .

5 Inan updated version of [11], the authors have recently given a precise count of the number of
projective equivalence classes with a fixed embedding dimension.
The proof of Theorem 5.1 in [11] relied on noting that a sparse n × n matrix of
rank r will take one of two forms
⎡ ⎤ ⎡ ⎤
| 0 | | 0 |
⎣0 B e1 ⎦ or ⎣er B e1 ⎦
| 0 | | 0 |
where B is a sparse matrix of rank r − 1 in the first case, and a sparse matrix of rank
r − 2 in the second. In light of this, we make the following definition.
Definition 5.4.2 For n 3, we say a sparse n × n matrix of rank r < n has type a
if its first column is 0, and type b if its first column is er .
In particular, a Frobenius form of embedding dimension n and rank r is
equivalent, after a linear change of coordinates, to a polynomial of one of two forms:
(type a) h = x1 xn + h (x2 , . . . , xn−1 ) where h is a Frobenius form of rank r − 1
q
and embedding dimension n − 2. In this case, B is non-degenerate of rank r − 1.

(type b) h = x1 xn + xr x1 + h (x2 , . . . , xn−1 ) where h is a Frobenius form of
q q

rank r − 2. If h is non-degenerate in the n − 2 variables x2 , . . . xn−1 , then B is
non-degenerate of rank r − 2. Otherwise, h must be non-degenerate in the n − 3
variables x2 , . . . ,
xr , . . . , xn−1 , so that B is degenerate of rank r − 2.
The proof of Theorem 4.1 will now follow from the following two propositions.
Proposition 5.5.2 There are precisely n − 1 projective equivalence classes of
Frobenius forms of embedding dimension n and rank n − 1 (in fixed degree pe + 1).
The proof of Proposition 5.5.2 uses the following lemma:
Lemma 5.6 Let h1 and h2 be two Frobenius forms of the same degree, both of
embedding dimension n and rank r = n − 1, and both represented by matrices (say
A1 and A2 ) in sparse form. Then h1 and h2 are projectively equivalent if and only
if A1 and A2 are the same type, and their B matrices are projectively equivalent.
Proof Without loss of generality, by Theorem 5.1 we may assume
q q q q
h1 = x1 xn + x2 xj2 + x3 xj3 + · · · + xn−1 xjn−1 , and
q q q q
h2 = x1 xn + x2 xj2 + x3 xj3 + · · · + xn−1 xjn−1

where xji , xji ∈ {x1 , . . . , xn−1 } for each i. In particular, we note that the only term
q
containing xn in both h1 and h2 is x1 xn . Now suppose some change of coordinates
φ sends h1 to h2 . The singular locus of both h1 and h2 is defined by the vanishing of
x1 , x2 , . . . , xn−1 , so the ideal x1 , . . . , xn−1 is stable under φ. In particular, φ must
send xn to a linear form involving xn . Supposing that φ maps
x1 → λ11 x1 + · · · + λ1,n−1 xn−1 , and

xn → λn1 x1 + · · · + λnn xn with λnn = 0,
then φ sends h1 to a polynomial

q q q q
λ11 λnn x1 xn + · · · + λ1,n−1 λnn xn−1 xn + terms that do not involve xn .
In order for this to be equal to h2 , we must have λ1i = 0 for all i > 1. So φ sends
x1 → λ11 x1
where λ11 = 0. In particular, if h1 and h2 are equivalent, they must be equivalent

after modding out by x1 .
We can now see that if h1 and h2 are equivalent, then A1 and A2 have the same
type. Letting ∼ denote equivalence up to change of coordinates, if
⎡ ⎤ ⎡ ⎤
0 0 1 0 0 1
⎣0 B 0⎦ ∼ ⎣er−1 B 0⎦
0 0 0 0 0 0
then

B0 B 0
∼
0 0 0 0
which is the same as saying B ∼ B . By rank considerations this never happens

between type a and type b (see the discussion following Definition 5.4.2). Thus the
sparse matrices of projectively equivalent Frobenius forms whose rank is one less
than the embedding dimension must have the same type. Furthermore, the argument
above also shows that their “B” matrices are projectively equivalent.
Because the converse is obvious, the lemma is proved.

Proof (Proof of Proposition 5.5.2) Fix q. Let N(n, r) denote the number of
projective equivalence classes of Frobenius forms of degree q + 1, rank r, and
embedding dimension n. We want to show that N(n, n − 1) = n − 1. We will induce
on n.
One readily verifies that
N(1, 0) = 0, N(2, 1) = 1, and N(3, 2) = 2.
Since type a and type b matrices yield distinct classes for r = n − 1, the number of
classes N(n, n − 1) is equal to the number of classes of type a plus the number of
type b. The discussion of the types following Definition 5.4.2 informs us, therefore,
that for n 4
N (n, n − 1) = N(n − 2, n − 2) + N(n − 2, n − 3) + N(n − 3, n − 3).

Recalling that there is only one full rank form in each degree and dimension
(Theorem 2.1), it follows that
N (n, n − 1) = 1 + N(n − 2, n − 3) + 1 = N(n − 2, n − 3) + 2.
Finally, by induction on n,
N(n, n − 1) = (n − 3) + 2 = n − 1,
as desired.

In light of Proposition 5.5.2, we have now fully classified all Frobenius forms in
five variables of ranks four and five. By Corollary 5.2 (1), it only remains
! to analyze
the rank three case. Using Corollary 5.2 (2), we see that there are 32 = 3 sparse
forms of rank three in five variables, and we have seen in Example 5.4.1 that two
of them are projectively equivalent. To complete the classification, it remains to see
that the third sparse matrix produces a Frobenius form not equivalent to these. This
is accomplished by the following:
Proposition 5.7 The following rank three Frobenius forms in five variables are not
projectively equivalent:
q q q+1
f = x1 x5 + x2 x4 + x3
and
q q q
g = x1 x5 + x2 x4 + x3 x2 .
Proof To see this, note that g ∈ x1 , x2 . Thus, if f and g are projectively
equivalent, there must be some linear forms L1 and L2 such that f ∈ L1 , L2 .
Since any projective change of coordinates must respect the singular locus, any
form in x1 , x2 , x3 must be sent to another form in x1 , x2 , x3 , as the vanishing of
these coordinates defines the singular set of both hypersurfaces. This implies that
L1 , L2 , being the images of x1 and x2 under our linear change of coordinates, are
forms in x1 , x2 , x3 .( )
q+1
Now note that x3 , x4 , x5 = f, x4 , x5 ⊂ L1 , L2 , x4 , x5 . In particular,
√
x3 ∈ L1 , L2 , x4 , x5 = L1 , L2 , x4 , x5 , so that x3 ∈ L1 , L2 . Without loss
of generality, we may assume L1 , L2 = x3 , L2 , where L2 is a linear form in x1
and x2 , so that f ∈ x3 , L2 . Therefore,
q q
x1 x5 + x2 x4 ∈ x3 , L2 (9)
as well. Considering the image of the expression (9) under the natural quotient map
k[x1 , . . . , x5 ] → k[x1 , . . . , x5 ]/x3 , we see that
q q
x1 x5 + x2 x4 ∈ L2 (10)
q q
in a polynomial ring in four variables. But the polynomial x1 x5 +x2 x4 is irreducible
(for example, by Eisenstein’s criterion), so we arrive at a contradiction. This
contradiction ensures that f and g are not projectively equivalent.

This completes our classification of Frobenius forms in up to five variables.
Acknowledgments This paper grew out of discussions begun at the AWM-sponsored workshop
“Women in Commutative Algebra” at the Banff International Research Station in October 2019,
where additional funding was provided by NSF Conference Grant DMS #1934391 and AWM
ADVANCE Grant NSF-HRD #1500481. We are grateful to Eloísa Grifo and Jennifer Kenkel,
who also participated in those discussions. Zhibek Kadyrsizova was partially supported by
FDCRGP Grant 021220FD4151. Jyoti Singh was partially supported by SERB(DST) Grant
MTR/2021/000879. Karen E. Smith was partially supported by NSF Grant DMS #1801697 and
NSF Grant DMS #2101075. Emily E. Witt was partially supported by NSF CAREER Award DMS
#1945611.
References
1. R. C. Bose and I. M. Chakravarti, Hermitian varieties in a finite projective space PG(N, q 2 ),

Canad. J. Math. 18 (1966), 1161–1182.
2. A. Beauville, Sur les hypersurfaces dont les sections hyperplanes sont à module constant, The
Grothendieck Festschrift, Vol. I, Progr. Math., vol. 86, Birkhäuser Boston, Boston, MA, 1990,
With an appendix by David Eisenbud and Craig Huneke, pp. 121–133. MR 1086884
3. L.E. Dickson, Determination of the structure of all linear homogeneous groups in a Galois
field which are defined by a quadratic invariant, Amer. J. Math. 21 (1899), no. 3, 193–256.
MR 1505798
4. , Linear groups: With an exposition of the Galois field theory, B.G. Teubners Sammlung
von Lehrbüchern auf dem Gebiete der mathematischen Wissenschaften mit Einschluss ihrer
Anwendungen, B. G. Teubner, 1901.
5. J.A. Dieudonné, La géométrie des groupes classiques, Springer-Verlag, Berlin-New York,
1971, Troisième édition, Ergebnisse der Mathematik und ihrer Grenzgebiete, Band 5. MR
0310083
6. R. Elman, N. Karpenko, and A. Merkurjev, The algebraic and geometric theory of quadratic
forms, American Mathematical Society Colloquium Publications, vol. 56, American Mathe-
matical Society, Providence, RI, 2008. MR 2427530
7. L.C. Grove, Classical groups and geometric algebra, Graduate Studies in Mathematics, vol. 39,
American Mathematical Society, Providence, RI, 2002. MR 1859189
8. N. Hara, A characterization of rational singularities in terms of injectivity of Frobenius maps,
Amer. J. Math. 120 (1998), no. 5, 981–996. MR 1646049
9. M. Homma and S. J. Kim, The characterization of Hermitian surfaces by the number of points,
J. Geom. 107 (2016), no. 3, 509–521.
10. Z. Kadyrsizova, J. Kenkel, J. Page, J. Singh, K. E. Smith, A. Vraciu, and E. E. Witt, Cubic
surfaces of characteristic two, Trans. Amer. Math. Soc. 374 (2021), no. 9, 6251–6267.
11. Z. Kadyrsizova, J. Kenkel, J. Page, J. Singh, K. E. Smith, A. Vraciu, and E. E. Witt, Lower
bounds on the F-pure threshold and extremal singularities, Trans. Amer. Math. Soc. (2020).
12. J. Kollár, Szemerédi–Trotter-type theorems in dimension 3, Adv. Math. 271 (2015), 30–61.
13. T. Y. Lam, Introduction to quadratic forms over fields, Graduate Studies in Mathematics,
vol. 67, American Mathematical Society, Providence, RI, 2005. MR 2104929
14. W. Scharlau, On the history of the algebraic theory of quadratic forms, Quadratic forms and
their applications (Dublin, 1999), Contemp. Math., vol. 272, Amer. Math. Soc., Providence,
RI, 2000, pp. 229–259. MR 1803370
15. B. Segre, Forme e geometrie hermitiane, con particolare riguardo al caso finito, Ann. Mat.
Pura Appl. 70 (1965), no. 4, 1–201.
Projective Dimension of Hypergraphs
Kuei-Nuan Lin and Sonja Mapes
2020 Mathematics Subject Classification 13D02, 05E40
1 Introduction
Let R = K[x1 , . . . , xn ] be a polynomial ring over a field K. The minimal free

resolution of R/I for an ideal I ⊂ R is an exact sequence of the form
' '
0→ S(−j )βp,j (R/I ) → · · · → S(−j )β1,j (R/I ) → R → R/I → 0
j j
The exponents βi,j (R/I ) are invariants of R/I , called the Betti numbers of R/I .
In general, finding Betti numbers is still a wide open question. Two other invariants
that one can associate to a minimal resolution are the projective dimension of R/I ,
denoted pd(R/I ), which is defined as follows
pd(R/I ) = max{i | βi,j (R/I ) = 0},
and the (Castelnuovo-Mumford) regularity of R/I , denoted reg(R/I ), which is

defined as follows
reg (R/I ) = max{j − i | βi,j (R/I ) = 0}.
Those two invariants play important roles in algebraic geometry, commutative

algebra, and combinatorial algebra. In general, one finds the graded minimal free
K.-N. Lin ()

Department of Mathematics, The Penn State University, McKeesport, PA, USA
e-mail: kul20@psu.edu
S. Mapes
Department of Mathematics, University of Notre Dame, Notre Dame, IN, USA
e-mail: smapes1@nd.edu
370 K.-N. Lin and S. Mapes
resolution of an ideal to obtain those invariants, but the computation can be difficult
and computationally expensive.
Kimura, Terai and Yoshida define the dual hypergraph of a square-free monomial
ideal in order to compute its arithmetical rank [12](see Definition 2.1 for definition
of a hypergraph). Since then, there have been a couple of papers using this
combinatorial object to study various properties, for example, [8] and [16]. In
particular, Lin and Mantero use it to show that ideals with the same dual hypergraph
have the same Betti numbers and projective dimension [13] (Theorem 2.4 part 1),
which has found use in other papers, such as in [11].
The focus of this work is to use the properties of a hypergraph to compute the
projective dimension of the associated square-free monomial ideal without finding
the minimal free resolution of the ideal. This is a different focus than various
work by others which produce the full resolution, for example, the recent work
of Eagon, Miller, and Ordog [4]. More precisely, we find the projective dimension
of hypergraphs when their 1-skeleton is a string or cycle. This extends the work
of Lin and Mantero in [13]. A key development is the result of Lin and Mapes
in [15], which allows us to remove a large class of higher dimensional edges of a
hypergraph without impacting its projective dimension (Corollary 4.4 [15]). Given
this previous result we can restrict our study in this paper to specific cases which
are not previously dealt with. Our main results are summarized by the following
theorem.
Theorem 1.1 Let H be a hypergraph and F be an edge of H such that
dimF > 0.
1. Let HS be an open string of length μ. If H is the union of HS% and
& F such that
all vertices of F are on HS , then pd(H) = pd(HS ) = μ − μ3 or pd(H) =
pd(HS ) + 1 depending on the position of vertices of F .
2. Let HCμ be an open cycle of length μ. If H is the union of HCμ and F such that

all vertices of F are on HCμ , then pd(H) = pd(HCμ ) = μ − 1 − μ−2 3 .

In Sect. 2 of the paper we begin by giving the necessary background to

understand the results of this paper. We also establish a new technique for computing
the projective dimension using bounds on sub-ideals, which is inspired by methods
from [5], namely Betti splittings (Lemma 2.13). We then proceed with our results
concerning higher dimensional edges on strings and cycles in Sects. 3 and 4.
Through out this paper, ideals are monomial ideals in a polynomial ring R over
the field K.
2 Preliminaries
2.1 Hypergraph of a Square-Free Monomial Ideal
Kimura, Terai, and Yoshida associate a square-free monomial ideal with a hyper-
graph in [12], see Definition 2.1. Note that this construction is different from the
Projective Dimension of Hypergraphs 371
constructions associating ideals to hypergraphs coming from the study of edge

ideals. In particular relative to edge ideals, the hypergraph of Kimura, Terai, and
Yoshida might be more aptly named the “dual hypergraph”. The construction of
the dual hypergraph is first introduced by Berge in [1]. In the edge ideal case, one
associates a square-free monomial with a hypergraph by setting variables as vertices
and each monomial corresponds to an edge of the hypergraph (see for example
[7]). In the following definition, we actually associate variables with edges of the
hypergraph and vertices with the monomial generators of the ideal, and in practice
this is the dual hypergraph of the hypergraph in the edge ideal construction.
Definition 2.1 Let I be a square-free monomial ideal in a polynomial ring with n
variables with minimal monomial generating set {m1 , . . . , mμ }. Let V be the set
{1, . . . , μ}. We define H(I ) (or H when I is understood) to be the hypergraph
associated to I which is defined as {{j ∈ V : xi |mj } : i = 1, 2, . . . , n}. We call the
sets {j ∈ V : xi |mj } the edges of the hypergraph.

Note that if you start with a hypergraph you can create a monomial ideal by
assigning a variable to each edge, then each vertex (or element in V ) would be
assigned the monomial product of the variables corresponding the edges using
that vertex. The issue however that doing this will not always produce a minimal
generating set. To obtain a minimal generating set the hypergraph needs to be
separated, where H is separated if in addition for every 1 j1 < j2 μ, there
exist edges F1 and F2 in H so that j1 ∈ F1 ∩ (V − F2 ) and j2 ∈ F2 ∩ (V − F1 ). All
hypergraphs in this paper will be separated unless otherwise stated.
Example 2.2 Let
I =(m1 = abk, m2 = bcl, m3 = cdklm, m4 = dekn,

m5 = efgn, m6 = ghmn, m7 = hikl, m8 = ij k)
the Fig. 1 is the hypergraph associated to I via the Definition 2.1 where
H(I ) = {a = {1}, b = {1, 2}, c = {2, 3}, d = {3, 4}, e = {4, 5}, f = {5},
g = {5, 6}, h = {6, 7}, i = {7, 8}, j = {8}, k = {1, 3, 4, 7, 8}, l = {2, 3, 7},
m = {3, 6}, n = {4, 5, 6}}.

Some important terminology regarding these hypergraphs is the following. We
say a vertex i ∈ V of H is an open vertex if {i} is not in H, and otherwise i is
closed. In Fig. 1, we can see that the vertices labeled by a, f and j are all closed,
and the rest are open. Moreover, a hypergraph H with V = [μ] is a string if {i, i +1}
is in H for all i = 1, . . . , μ−1, and the only edges containing i are {i−1, i}, {i, i+1}
and possibly {i}. We say that a string is an open string if all vertices other than 1
and μ are open (Note that for H to be separated 1 and μ must be closed). Also, H
Fig. 1 Hypergraph of Example 2.2
is a μ-cycle if H = H̃ ∪ {μ, 1} where H̃ is a string. We say a cycle is an open cycle

if all the vertices are open. In Fig. 1 if we were to remove the edges corresponding
to the variables k, l, m, and n then we would be left with a string. If we further
removed the edge corresponding to the variable f , then it would be an open string.
Let Hi = {F ∈ H : |F | i + 1} denote the i-th dimensional subhypergraph of H
where |F | is the cardinality of the F . We call H1 the 1-skeleton of H.
Example 2.3 Some simple examples where V = {1, 2, 3} of open strings and open
cycles are as follows. Let H1 = {a = {1}, b = {1, 2}, c = {2, 3}, d = {3}}, then
H1 is an open string and the corresponding monomial ideal is I1 = (ab, bc, cd). If
we let H2 = {a = {1, 2}, b = {2, 3}, c = {1, 3}}, then this is an open cycle and the
corresponding monomial ideal is I2 = (ac, ab, bc).

Recently there have been a number of results determining both the projective
dimension and the regularity of certain square-free monomial ideals from the
associated hypergraph. As this paper focuses more on the projective dimension,
we include the statements of some of results that are useful for the rest of the paper
here (these appear separately in the literature but we list them all here as part of one
statement). We write pd(H) = pd(H(I )) = pd(R/I ) and reg(H) = reg(H(I )) =
reg(R/I ) where H(I ) is the hypergraph obtained from a square-free monomial ideal
I . We do this for the Betti numbers as well. Notice that for the content of this work,
we assume there is a unique variable associated to each face of H.
Theorem 2.4
1. [13, Proposition 2.2] If I1 and I2 are square-free monomial ideals associated to
the same separated hypergraph H, then the total Betti numbers of the two ideals
coincide.
2. [15, Corollary 4.4] Let F be an edge on the hypergraph H. If F is an union of
other edges of H, then pd(H) = pd(H\F ).
3. [14, Theorem 2.9 (c)]) If H ⊆ H are hypergraphs with μ(H ) = μ(H), then
pd(H ) pd(H) where μ(∗) denotes the number of vertices of ∗.
4. [9, Theorem 7.7.34, Corollary%7.7.35]
& An openNstring
O hypergraph with μ vertices
has projective dimension μ − μ3 , regularity μ3 , and βμ− μ ,μ (H) = 0 when
3
μ(H) = 0.
5. [10, Corollary 2.2] A hypergraph H with n1 + n2 vertices that is a disjoint union

of open string
% &hypergraphs,
% & H1 and H2 with n1 and n2 vertices has pd(H) =
n1 + n2 − n31 − n32 and β (H) = 0.
n1 n2
n1 +n2 − 3 − 3 ,n1 +n2
6. [9, Corollary 7.6.30] If H is an open cycle with μ vertices then pd(H) = μ −
1 − μ−2
3 .
7. [2, Corollary 20.19] If I = mI where m is a monomial of degree r then
pd(R/I ) = pd(R/I ), reg(R/I ) = r + reg(R/I ).
8. [2, Corollary 20.19] Let m be a monomial of degree r, then
reg(R/(I, m)) max{reg(R/I ), reg(R/(I : m) + r − 1}.
Remark 2.5 Note that Theorem 2.4 part 1 allows us to talk about the projective
dimension of a hypergraph rather than an ideal. We will use pd(H(I )) in the place
of pd(R/I ) throughout the paper. Moreover if an edge is an union of other edges in
a hypergraph, and we are considering the projective dimension of the hypergraph,
we can just ignore or remove the edge using Theorem 2.4 part 2. For example,
in Fig. 1, we can remove the edge k = {1, 3, 4, 7, 8} = {1} ∪ {3, 4} ∪ {7, 8} and
n = {4, 5, 6} = {4, 5} ∪ {5, 6}. If a hypergraph H = H(I ) is an union of two
disconnected hypergraphs G1 = H(I1 ) and G2 = H(I2 ), we have pd(H) =
pd(G1 )+pd(G2 ) and reg(H) = reg(G1 )+reg(G2 ) as one can construct the minimal
resolution of R/I using the tensor of minimal resolutions of R/I1 and R/I2 .

2.2 Colon Ideals: Key tool
One technique that is used in [13] and [14] which we will need here, is using the
short exact sequences obtained by looking at colon ideals. Specifically there are two
types of colon ideals that we are interested in, and we explain below what each
operation looks like on the associated hypergraphs.
Definition 2.6 Let H be a hypergraph, and I = I (H) be the standard square-free
monomial ideal associated to it in the polynomial ring R. Let G(I ) = {m1 , . . . , mμ }
be the minimal generating set of I . Let F be an edge in H and let xF ∈ R be the
variable associated to F . Also let v be a vertex in H and mv ∈ I be the monomial
generator associated to it.
• The hypergraph Hv : v = Qv is the hypergraph associated to the ideal Iv : mv
where Iv = (G(I )\mv ), and Hv = H(Iv ) is the hypergraph associated to the
ideal Iv .
• The hypergraph H : F , obtained by removing F in H, is the hypergraph
associated to the ideal I : xF .
• The hypergraph (H, xF ), obtained by adding a vertex corresponding to the
variable xF in H, is the hypergraph associated to the ideal (I, xF ).
Remark 2.7 Given a hypergraph H and a vertex v of H, we consider a short exact

sequence
0 → Hv : v = Qv → Hv → H → 0.
If pd(Hv ) > pd(Qv ), then with the projective dimension on the short exact sequence
that pd(Hv ) max{pd(Qv ), pd(H)} and
pd(H) max{pd(Hv ), pd(Qv ) + 1},
we can conclude that pd(H) = pd(Hv ). We will use this strategy in later sections.

The following result appearing in [14] will be very useful to us in this paper. We
put it here for the self-containment of this work and for the reader’s convenience.
We need some terminology first. The degree of a vertex is the number of faces
containing the vertex. Let v be a vertex in a hypergraph H, and let H1 , ..., Hr be
the connected components of Hv ; if one of them, say H1 , is a string hypergraph, we
call H1 a branch of H (from v).
Theorem 2.8 ([14]) Let H be a one-dimensional hypergraph, w a vertex with
degree at least 3 in H, and S be a branch departing from w with v1 , . . . , vn vertices.
Suppose vertices of S, v1 , . . . , vn−1 , are open and the end vertex vn , i.e., the leaf of
S, is the only closed vertex of S. Let E be the unique edge connecting w to v1 . Then
pd(H) = pd(H ), where H is the following hypergraph: (a) if n ≡ 1 mod 3, then
H = H : E; (b) if n ≡ 2 mod 3, then H = Hw .

Example 2.9 In Fig. 2, there are 4 hypergraphs, H, H1 = H : E, H2 = Hw2 ,
H3 = (H, xF ). S1 is a branch of length 1 departing from w1 and S2 is a branch of
length 2 departing from w2 . By Theorem 2.8, pd(H) = pd(H1 ) = pd(H2 ). Notice
that in H2 , the edge F becomes an one-dimensional edge.

2.3 Splittings: Key Tool
In [3] the notion of a splitting of a monomial ideal I was introduced.

Definition 2.10 ([3]) A monomial ideal I is splittable if I is the sum of two
nonzero monomial ideals J and K, i.e. I = J + K, such that
1. The generating set G(I ) of I , is the disjoint union of G(J ) and G(K).
2. There is a splitting function
G(J ∩ K) → G(J ) × G(K)

w→ (ψ(w), φ(w))
Fig. 2 Hypergraph operations
satisfying
(a) (S1) for all w ∈ G(J ∩ K), w = lcm(ψ(w), φ(w)).
(b) (S2) for every subset S ⊆ G(J ∩ K), both lcm(ψ(S)) and lcm(φ(S)) strictly
divide lcm(S).

If J and K satisfy the above properties they are called a splitting of I .

Now the key reason we are interested in splittings is the following result by both
Eliahou-Kervaire and separately Fatabbi.
Theorem 2.11 (Eliahou-Kervaire [3] Fatabbi [5]) Suppose I is a splittable
monomial ideal with splitting I = J + K. Then for all i, j 0
βi,j (I ) = βi,j (J ) + βi,j (K) + βi−1,j (J ∩ K).
It is important to note that not all monomial ideals admit splittings. What is
interesting is that there are sometimes monomial ideals that can be decomposed
into a sum of ideals J and K which satisfy the conclusions of the previous theorem.
This motivates the following definition by Francisco, Ha, and Van Tuyl in [6]
Definition 2.12 Let I, J and K be monomial ideals such that G(I ) is the disjoint
union of G(J ) and G(K). Then I = J + K is a Betti splitting if
βi,j (I ) = βi,j (J ) + βi,j (K) + βi−1,j (J ∩ K)
for all i ∈ N and all (multi)degrees j .

One complication however is that if one wants to use the existence of a Betti
splitting to prove something about a resolution, one must first know something
about the resolution in question. The key for us will be in dissecting the proof of
Fatabbi in order to prove that in some special cases, which may fail condition (S2)
in Definition 2.10, that a similar formula for (some) Betti numbers holds.
The following lemma is an adaptation of the proof of Fatabbi in a special case
where we do not have a splitting. In this case we can show that the necessary
conditions hold at the end of the resolution, so that we get a formula like that
of Theorem 2.11 for the last Betti numbers. In particular this allows us to prove
statements about projective dimension.
Lemma 2.13 Let I be a monomial ideal and I = J + K in the ring R =
K[x1 , . . . , xn ] over a field K. Suppose we have the following conditions on
projective dimension
1. pd(R/J ) < q
2. pd(R/K) = q,
and reg(R/K) < r. If βq,q+r (R/J ∩ K) = 0 and pd(R/J ∩ K) = q, then
pd(R/I ) = q + 1.

Proof We consider the short exact sequence
0 → J ∩ K → J ⊕ K → I → 0.
Let α(w) = (w, w) be the map from J ∩ K to J ⊕ K and π(u, v) = u − v be the

map from J ⊕ K to I . There is an induced homology sequence
· · · → TorR
q+1 (J ∩ K, K) → Torq+1 (J, K) ⊕ Torq+1 (K, K)
R R
q+1 (I, K) → Torq (J ∩ K, K)

→ TorR R
→ TorR
q (J, K) ⊕ Torq (K, K) → · · ·
R
We have pd(R/J ) < q, pd(R/K) = q, and pd(R/(J ∩ K)) = q, hence the short
exact sequence gives pd(R/I ) max{pd(R/(J ∩ K)) + 1, pd(R/J ⊕ K)} q + 1.
The homology sequence becomes
q+1 (I, K) → Torq (J ∩ K, K) → Torq (K, K) → · · ·

0 → 0 → TorR R R
Moreover, the q + r graded piece is the following:
q+1 (I, K)q+r → Torq (J ∩ K, K)q+r → Torq (K, K)q+r → · · ·

0 → TorR R R
Now using our assumption that reg(R/K) < r and pd(R/K) = q then we have
∼
TorRq (K, K)q+r = 0. This shows Torq+1 (I, K)q+r = Torq (J ∩ K, K)q+r = 0 by
R R
the fact that βq,q+r (R/J ∩ K) = 0 and hence pd(R/I ) = q + 1.

Remark 2.14 The above lemma can be translated in terms of associated hyper-
graphs. Let H = H(I ) be a hypergraph with underlying vertex set V , and let V1
and V2 be a partition of the vertices of H such that V1 ∪ V2 = V and V1 ∩ V2
is empty. Now define Ii to be the ideal generated by the generators of I indexed
by the elements in Vi for i = 1, 2. Let Gi = H(Ii ) for i = 1, 2 and H(I1 ∩ I2 )
be the hypergraphs corresponding the ideals Ii for i = 1, 2 and the ideal I1 ∩ I2 .
Suppose pd(G1 ) < q, pd(G2 ) = q, pd(H(I1 ∩ I2 )) = q, and reg(G2 ) < r and
reg(H(I1 ∩ I2 )) = r. Suppose βq,q+r (H(I1 ∩ I2 )) = 0, then pd(H) = q + 1.

3 Strings with Higher Dimensional Edges
In this section we are primarily interested in finding the projective dimension

of a square-free monomial ideal such that its hypergraph is a string with higher
dimensional edges attached to it. Our primary object is described below and we fix
the notation now for the easy reference later.
Notation 3.1 Let H be a hypergraph and F be an edge of H.
1. Let HSμ be a string hypergraph with vertex set V = {w1 , . . . , wμ }, and an
edge F with k > 1 vertices {v1 = wi1 , . . . , vk = wik } ⊆ {w1 , . . . , wμ } with
1 i1 < · · · < ik μ. When we say H is a string together with an edge
consisting of k vertices, we mean H = HSμ ∪ F .
2. Let n1 = i1 − 1, nj = ij − ij −1 − 1 for j = 2, . . . k, and nk+1 = μ − ik be the
number of vertices between vertices of F .
3. We write ni = 3li + ri where li are some non-negative integers, and 0 ri 2
for i = 1, . . . , k + 1.
Example 3.2 The hypergraph shown in Fig. 3 shows the string hypergraph with a
higher dimensional edge with the notation outlined above. In this case, μ = 11,
k = 4, n1 = 1, n2 = n3 = n4 = 2, and n5 = 0.


Because of Theorem 2.4 part 2, we will primarily focus on the higher dimensional
edges which are not unions of two or more edges. The vertex wi is assumed
to be open for all i unless otherwise stated. We find the k+1projective dimensions
by considering three different cases: k+1 r i < 2k, i=1 ri 2k + 1, and
k+1 i=1
i=1 ir = 2k in three propositions. We conclude this section with Theorem 3.14
which
k+1 covers all the previous results. Example 3.2 is an example of the cases when
i=1 ir < 2k.

First we will prove two propositions that deal with the case when k+1 i=1 ri < 2k,
k+1
or i=1 ri 2k + 1. In these two cases we will show that for a hypergraph H
satisfying the hypotheses of the propositions, that the projective dimension will be
the same as for HSμ . Before we can get to these propositions though we present
couple of technical computations first, and the computational processes will be used
in a similar manner in the later
proofs.
We often use the following
identities
without
justification A = 3 = 3A 3A+1
3 = 3A+2
3 and A − 1 = 3A−3
3 = 3A−1
3 =

3A−2
3 .
k+1
Lemma 3.3 Let ni = 3li +ri with 0 ri < 3 for i = 1, . . . , k+1. If i=1 ri < 2k,
then
P Q

k+1 n
k+1
k + k+1
i i=1 ni
(ni − )<k+ ni − .
3 3
i=1 i=1
Proof We simplify the left hand side of the inequality using ni = 3li + ri and
0 ri < 3.

k+1 n
k+1 R S
i 3li + ri
(ni − )= (ni − )
3 3
i=1 i=1

k+1
= (ni − li )
i=1

We simplify the right hand side of the inequality with the assumption of k+1 i=1 ri <
2k to obtain the conclusion as follows.
P Q P Q

k+1
k + k+1
k+1
k + k+1
i=1 ni i=1 (3li + ri )
k+ ni − =k+ ni −
3 3
i=1 i=1
P k+1 Q

k+1
k + 2k + i=1 (3li )
>k+ ni −
3
i=1

k+1
k+1
=k+ ni − k − li
i=1 i=1

k+1
= (ni − li )
i=1

Lemma 3.4 Let ni = 3li + ri with r1 = 2 and 0 ri < 3 for i = 1, . . . , k + 1. If
k+1
i=1 ri 2k + 1, then
R k+1 S
k+ i=1 ni % ni &
1. k + k+1 n
i=1 i − 3 = k+1
i=1 (ni − 3 )
R k+1 S
% ni &
2. n1 − n13−2 + k − 2 + k+1
k+ i=2 ni
n
i=2 i − 3 < k+1
i=1 (ni − 3 ).

Proof Notice that 2k + 2 k+1 i=1 ri 2k + 1 by the assumption 0
ri < 3.
Where the first inequality assumes that each ri = 2. This means that k+1
i=1 ri is
either 2k + 1 or 2k + 2, hence we must have
P Q P Q
k + k+1
i=1 ni k + k+1
i=1 (3li + ri )
=
3 3
P k+1 Q
k + k+1 (3l i ) + i=1 ri
= i=1
3

k+1
=k+ li
i=1
k+1
ni
=k+ .
3
i=1
The first equality follows immediately.

The inequality (2)requires the extra assumptionr1 = 2, since together with
the assumption that k+1 i=1 ri 2k + 1 it implies
k+1
ri 2k − 1. We write
k+1
i=2
n1 = 3l1 + 2 and ni = 3li + ri , and use the inequality i=2 ri 2k − 1 to obtain
the statement’s inequality.
R S P Q
n1 − 2
k+1
k + k+1
i=2 ni
n1 − +k−2+ ni −
3 3
i=2
R S P k+1 Q

k+1
k+ + ri )
3l1 i=2 (3li
= n1 − +k−2+ ni −
3 3
i=2
P k+1 Q

k+1
k + 2k − 1 + i=2 (3li )
n1 − l1 + k − 2 + ni −
3
i=2

k+1
k+1
=k−2+ ni − k + 1 − li
i=1 i=1

k+1 n
i
< (ni − ).
3
i=1

With these technical computations out of the way we are ready to state and prove
the first two propositions.

Proposition 3.5 We adapt Notation 3.1. If k+1 i=1 ri < 2k, then pd(H) = pd(HSμ ).

k+1 %μ& k+1
Proof Notice that μ = k + i=1 ni and pd(HSμ ) = μ − 3 = k + i=1 ni −
R k+1 S
k+ i=1 ni
3 . Now consider the short exact sequence
0 → (H : F ) → H → (H, xF ) → 0
where xF is the variable corresponding to F . We first observe that (H : F ) = HSμ

hence pd(HSμ ) pd(H) by Theorem 2.4 part 3. Let HVF = HSμ ∩ (V \VF ) be the
hypergraph obtained by removing all the vertices v1 , . . . , vk . Then HVF ∪ {xF } =
(H, xF ). Notice that HVF is union of k + 1-strings such that the i-th string has ni
% ni &
vertices so by Theorem 2.4 part 4 we get that pd(HVF ) = k+1 i=1 (ni − 3 ). Once
we show that pd(HVF ) < pd(HSμ ), then by the short exact sequence, we have
pd(HSμ ) pd(H) max{pd(HSμ ), pd(HVF ) + 1} = pd(HSμ ).
To show that pd(HVF ) < pd(HSμ ) it is sufficient to show that

P k+1 Q

k+1 n
k+1
k+ ni
i
(ni − )<k+ ni − i=1
3 3
i=1 i=1
which is shown in Lemma 3.3.

k+1
Proposition 3.6 We adapt Notation 3.1. If 2k + 1, then pd(H) =
i=1 ri
pd(HSμ ).

R k+1 S
k+ i=1 ni
Proof We first notice that pd(HSμ ) = k + k+1
i=1 ni − 3 = k+1
i=1 (ni −
% ni & k+1
3 ) by Theorem 2.4 part 4 and Lemma 3.4 (1). Since i=1 ri 2k + 1, ri < 3,
and k > 1, we have at most one ri is equal to 1. We may assume r1 = 2. Let

Hv1 = Hv be the hypergraph where we remove the vertex v1 = v from H and let
Hv : v1 = Qv be the hypergraph H(Iv : mv ) where Iv = I (Hv ) and mv is the
monomial corresponding to the vertex v. We have a short exact sequence
0 → Qv → Hv → H → 0.
We claim that in this case pd(HSμ ) = pd(Hv ) > pd(Qv ), then by Remark 2.7,
we conclude that pd(HSμ ) = pd(Hv ) = pd(H).
To see the proof of the claim, we use induction on k. When k = 2, Hv is a union
of two strings of length n2 + n3 + 1 and n1 . When n2 2 and n3 2, the string of
length n2 + n3 + 1 has two open strings with n2 − 1 and n3 − 1 open vertices. When
(n2 = 1 and r3 = 2), or (n3 = 1 and r2 = 2), the string of length n2 + n3 + 1 has
exactly 3 closed vertices at the ends of string and all other vertices are open. By the
work of [13], Theorem 2.4 part 4, we have either
n R S R S
1 n2 − 2 n3 − 2
pd(Hv ) = n1 − + n2 + n3 + 1 − 2 − − +1
3 3 3

3 n
i
= (ni − ) = pd(HSμ )
3
i=1
when n2 2, n3 2, r2 = 2 and r3 = 2, or
n R S R S
1 n2 − 2 n3 − 2
pd(Hv ) = n1 − + n2 + n3 + 1 − 2 − −
3 3 3

3 n
i
= (ni − ) = pd(HSμ )
3
i=1
when n2 2, n3 2, (r2 = 1 and r3 = 2) or (r2 = 2 and r3 = 1), or

n R S
1 n3 − 2
pd(Hv ) = n1 − + n2 + n3 + 1 − 1 −
3 3

3 n
i
= (ni − ) = pd(HSμ )
3
i=1
when n2 = 1 and r3 = 2, or
n R S
1 n2 − 2
pd(Hv ) = n1 − + n2 + n3 + 1 − 1 −
3 3

3 n
i
= (ni − ) = pd(HSμ )
3
i=1
when n3 = 1 and r2 = 2. On the other hand, Qv is a union of two isolated vertices

and two strings of length, n1 − 2, n2 − 2 + n3 + 1, hence we have
R S R S
n1 − 2 n2 + n3 − 1
pd(Qv ) = 2 + n1 − 2 − + n2 + n3 − 1 −
3 3

3 n
i
(ni − ) − 1 < pd(Hv ).
3
i=1
For the second inequality above, we use the fact that r2 + r3 3.

For the case when k > 2, we use the same exact sequence. Here the hypergraph

Hv is a union of a string of n1 vertices and a string of length μ = k − 1 + k+1 i=2 ni
k+1
with a k −2-dimensional edge of k −1 vertices such that i=2 ri % 2(k & −1)+1. By
the induction hypothesis and Theorem 2.4 part 4, pd(Hv ) = n1 − n31 +pd(HSμ ) =
k+1 % ni &
i=1 (ni − 3 ) = pd(HSμ ). On the other hand, Qv is a union of two isolated

closed vertices and two strings of length n1 − 2 and k − 3 + k+1 i=2 ni . Hence we
have
R S P Q
n1 − 2
k+1
k − 3 + k+1 i=2 ni
pd(Qv ) = 2 + n1 − 2 − +k−3+ ni −
3 3
i=2
R S P Q
n1 − 2
k+1
k + k+1
i=2 ni
= n1 − +k−2+ ni −
3 3
i=2

k+1 n
i
< (ni − ) = pd Hv
3
i=1
where the inequality is shown in Lemma 3.4 (2). We conclude pd(H) = pd(Hv ) =
pd(HSμ ).

Now we want to deal with the case when k+1 i=1 ri = 2k. In this case we get
two different outcomes, and it will be necessary to prove a number of lemmas that
will allow us to work with the special case. The primary strategy is using induction
on the number of vertices of the extra edge F , and the short exact sequences with
Theorem 2.8 to remove vertices or edges to reduce cases into smaller cases. Lemmas
are ordered in a way that later lemmas are built or proven with the previous lemmas’
conclusions.
First we will need to prove some results about when the spacing measured by the
ni is equivalent to 2 modulo 3. The following lemma deals with the case where one
of the ends of the string coincides with a vertex from F .
Lemma 3.7 We adapt Notation 3.1. Suppose the end vertices of the string are w1
and wμ where w1 is closed and wμ is open, and wμ = vk . If ni = 2 + 3li for
i = 1, . . . , k, then the projective dimension of H is
P k Q

k
k
k+ i=1 ni
2k + 2 li = k + ni −
3
i=1 i=1
and

k
reg(H) k + li .
i=1
Proof We use induction on k. When k = 1, we have wμ = v1 . In this case the

edge has only one vertex v1 which forces wμ = v1 to become a closed vertex. Also,
H becomes of length μ = n1 + 1,
a string so by Theorem 2.4 part 4, pd(H) =
n1 +1 n1 +1
n1 + 1 − 3 = 2 + 2l1 and reg(H) = 3 = 1 + l1 .
For the induction step, we consider the short exact sequence
0 → Hv1 : v1 = Qv1 → Hv1 → H → 0.

We will show pd(Hv1 ) = 2k + 2 ki=1 li > pd(Qv1 ), then by Remark 2.7, we

conclude pd(H) = pd(Hv1 ) = 2k + 2 ki=1 li .
Since k > 1, the vertex v1 corresponds to a monomial of degree 3. Notice that
Hv1 is the union of a string of length n1 and a hypergraph with exactly the same
structure of H (i.e. closed vertex on one end of string and an open vertex coinciding
with a vertex of the higher dimensional edge at the other end) such that it has an
edge with k − 1 vertices. By Theorem 2.4 part 4 and induction hypothesis, we have
n
k
k
1
pd(Hv1 ) = n1 − + 2(k − 1) + 2 li = 2k + 2 li
3
i=2 i=1
and
n
k
k
1
reg(Hv1 ) +k−1+ li = k + li .
3
i=2 i=1
Moreover, the hypergraph Qv1 is a union of three isolated vertices and two strings

of length n1 −2 and n2 −2+nk −1+k−2+ k−1 i=3 ni . Therefore by Theorem 2.4 part
k
4 with the fact that i=2 ri = 2(k − 1), we have
R S P Q
n1 − 2
k
k − 5 + ki=2 ni
pd(Qv ) = 3 + n1 − 2 − +k−5+ ni −
3 3
i=2
P k Q

k
k−5+ i=2 (3li + 2)
= 2l1 + k − 2 + (3li + 2) −
3
i=2
P k Q

k
k − 5 + 2(k − 1) + i=2 (3li )
= 2l1 + k − 2 + 2(k − 1) + (3li ) −
3
i=2

k
k
= 3k − 4 + 2 li − k + 3 = 2k + 2 li − 1
i=1 i=1
and
T U V W

k
n1 − 2 k − 5 + ki=2 ni
reg(Qv ) = + =k+ li − 2.
3 3
i=1
We have shown

k
pd(H) = pd(Hv1 ) = 2k + 2 li .
i=1
Using the short exact sequence and Theorem 2.4 part 8, we have and

k
k
k
reg(H) max{k + li , k + li − 2 + 3 − 1} = k + li .

i=1 i=1 i=1
In the next lemma we need to consider a specific case which is necessary for the
proof of Lemma 3.10. Specifically this will address Hv2 in Lemma 3.10 where H is
a string together with an edge consisting of k vertices. Those two special cases are
shown in Example 3.11 later.
Lemma 3.8 We adapt Notation 3.1. Assume w1 = v1 and wμ = vk , i.e. n1 =
nk+1 = 0, and both w1 = v1 and wμ = vk are open vertices. If ni = 2 + 3li for

i = 2, . . . , k. Then the projective dimension of Hv2 is 2(k − 1) + 2 ki=2 li and
k
reg(Hv2 ) k − 1 + i=2 li .
Proof We use induction on k. When k = 2, Hv2 is a string of length n2 + 1. By
n2 +1 3l2 +3
Theorem 2.4 part 4, pd(Hv2 ) = n2 + 1 − 3 = 3l2 + 3 − 3 = 2 + 2(l2 )

and reg(Hv2 ) = n23+1 = 1 + l2 .
For the induction step, we consider the short exact sequence
0 → Hv1 ,v2 : v1 → Hv1 ,v2 → Hv2 → 0,
where Hv1 ,v2 is the hypergraph obtained from H after removing vertices v1 and
v2 . The proof is almost identical to the Lemma 3.7 except that v1 corresponds to a
monomial of degree 2. Hv1 ,v2 is a union of a string of length n2 and a hypergraph
satisfying the assumptions of Lemma 3.7, and Hv1 ,v2 : v1 is a union of two isolated
and two strings of length n2 − 2 and nk − 1 when k = 3, and

vertices, k nk − 1 + k −
3 + k−1 ni when k > 3. We have reg(H v ,v : v1 ) = k − 1 + i=2 li − 1 and
i=3 1 2
pd(Hv1 ,v2 : v1 ) = 2(k−1)+2 ki=2 li −1. Hence we have pd(Hv2 ) = pd(Hv1 ,v2 ) =

2(k − 1) + 2 ki=2 li and reg(Hv2 ) max{k − 1 + ki=2 li , k − 1 + ki=2 li − 1 +
k
2 − 1} = k − 1 + i=2 li by Theorem 2.4 part 8.

Now we will use the splitting type result in Lemma 2.13 to finish our necessary
results for the hypergraphs which are a string together with an edge consisting of
k vertices, where the spacing between the vertices of the edge are equivalent to 2
modulo 3.
The next lemma deals with the intersection ideal in the special case where G1
will correspond to one vertex of a larger hypergraph H.
Lemma 3.9 We adapt Notation 3.1. Assume k > 2, w1 = v1 and wμ = vk , i.e.
n1 = nk+1 = 0, and both w1 = v1 and wμ = vk are open vertices. Furthermore,
ni = 2 + 3li for i = 2, . . . , k. Let G1 = {v2 } and G2 = Hv2 (which is just H
with v2 removed) and denote Ii = I (Gi ) as the ideals corresponding to Gi , then
I1 ∩ I2 = mv2 I where mv2 is the monomial corresponding to the vertex v2 and
H(I ) has 4 isolated vertices and 2 strings one of length:
• n2 − 3 when n2 > 2, or
• 0 if n2 = 2
and the other of length:

• k − 3 + nk − 1 + n3 − 2 + k−1
i=4 ni when k > 3, or
• n3 − 3 when k = 3 and when n3 > 2, or
• 0 when k = 3 and n3 = 2

Proof To see this consider the hypergraph H, for notational convenience, let us
denote the vertex neighboring v1 as wα , the vertices neighboring v2 as wβ1 and
wβ2 , and the vertex neighboring vk as wγ . Now removing v2 from H leaves us
with a hypergraph on the same vertex set excluding v2 and all vertices remain open
except wβ1 and wβ2 which become closed, together with a k − 1 edge F that has
{v1 , v3 , . . . , vk } as its vertex set (note this also describes the hypergraph Hv2 ).
Now we consider the intersection with the ideal generated by G1 . A first step
towards finding these new generators is to multiply each generator for G2 by the
monomial mv2 corresponding to v2 in the original H. The result on hypergraphs is
now we get a hypergraph, which we will denote as mv2 Hv2 , consisting of 2 strings
of only open vertices where one string is the part of H consisting of v1 to wβ1 and
the other is wβ2 to vk , and all the open vertices are in an edge corresponding to mv2 .
Note that the spacing measurements for mv2 Hv2 are the same as for Hv2 . The issue
here is that mv2 Hv2 is not separated (i.e. the generators are not minimal). Denote
the vertices neighboring each wβi as wβi . It is easy to see that removing vertices
wα , wβ1 , wβ2 , and wγ from mv2 Hv2 produces the desired separated hypergraph
corresponding to I1 ∩ I2 .
Notice that when n2 = 2 then wα = wβ1 and similarly if k = 3 and n3 = 2

then wβ2 = wγ . It is easy to see then that we get 4 isolated vertices corresponding
to the original v1 , wβ1 , wβ2 and vk . And the remaining chains of open strings reflect
removing v1 , wα , wβ1 and wβ1 from the chain of length n2 + 1 so the result is a
chain of length n2 − 3 when n2 > 2, or 0 when n2 = 2. Similarly, after removing
vk , wβ2 , wβ2 and wγ from a chain of length k − 2 + n3 + ki=4 ni results a chain of

length k − 3 + n3 − 2 + nk − 1 + k−1 i=4 ni when k > 3, or n3 − 3 when k = 3 and
n3 > 2, or 0 when k = 3 and n3 = 2.

Lemma 3.10 We adapt Notation 3.1. Assume w1 = v1 and wμ = vk , i.e. n1 =
nk+1 = 0, and both w1 = v1 and wμ = vk are open vertices. If ni = 2 + 3li for

i = 2, . . . , k. Then the projective dimension of H is 2(k − 1) + 2 ki=2 li + 1.

Proof If k = 2, then H is an open
cycleof length 2 + n2 , then by Theorem 2.4
part 6, pd(H) = 2 + n2 − 1 − 2+n32 −2 = 2 + 2l2 + 1. We now assume k >
2. By making vertices v1 , . . . , vk become closed, we obtain a hypergraph H and
pd(H) pd(H ) by Theorem 2.4 part 3. Let H be the hypergraph obtained from
H by removing the higher dimensional edge. Then by Theorem 2.4 part 2, we have
that pd(H ) = pd(H ) because all the vertices v1 , . . . , vk are closed in H . Note

that H is a string of length k + ki=2 ni with k − 1 open strings of n2 , . . . , nk open
vertices. By Theorem 3.4 in [13], the projective dimension is the sum of projective
dimension of each open string plus 1, hence we get

k k

k
ni
pd(H ) = ni − + 1 = 2(k − 1) + 2 li + 1.
3
i=2 i=2 i=2

Thus, we obtain pd(H) 2(k − 1) + 2 ki=2 li + 1.
Now we consider H = {v2 } ∪ Hv2 and we will show it satisfies the condition of
Remark 2.14 with V1 = {v2 } and V2 as the vertex set of Hv2 . Denote Ii as the ideal
generated by the generators of I (H) corresponding to Vi , and G1 = {v2 } = H(I1 )
and G2 = Hv2 = H(I2 ). First notice that pd(G1 ) = 1 and reg(G1 ) = 2, since the
degree of the generator corresponding to v2 is 3.Moreover G2 satisfies the condition
of Lemma 3.8, hence pd(G2 ) = 2(k − 1) + 2 ki=2 li = q and reg(G2 ) k − 1 +
k
i=2 li = r − 1. By Lemma 3.9, I1 ∩ I2 = mv2 I where H(I ) has 4 isolated
vertices and two strings of length n2 − 3 and k − 3 + nk − 1 + n3 − 2 + k−1 i=4 ni .
Then by ni = 2 + 3li for i = 2, . . . , k, Theorem 2.4 parts 4 and 7 we get
pd(H(I1 ∩ I2 ))
R S P Q
n2 − 3
k
k − 6 + ki=3 ni
= 4 + n2 − 3 − +k−6+ ni −
3 3
i=3
R S P Q
3l2 − 1
k
k − 6 + ki=3 (3li + 2)
= 3 + 3l2 − +k−6+ (3li + 2) −
3 3
i=3
P k Q

k
k − 6 + 2(k − 2) + i=3 (3li )
= 2l2 − 2 + k + 2(k − 2) + (3li ) −
3
i=3

k
k
= 2l2 − 2 + k + 2(k − 2) + (3li ) − k + 4 − (li )
i=3 i=3

k
= 2(k − 1) + 2 li = q
i=2
and
T U V W
n2 − 3 k − 6 + ki=3 ni
reg(H(I1 ∩ I2 )) = 3 + +
3 3
T U V W
3l2 − 1 k − 6 + 2(k − 2) + ki=3 (3li )
=3+ +
3 3

k
k
= 3 + l2 + k − 3 + li = k + li = r.
i=3 i=2
Moreover, by Theorem 2.4 parts 4 and 5, we have βq,q+r (H(I1 ∩I2 )) = 0. Hence

by Lemma 2.13, pd(H) = 2(k − 1) + 2 ki=2 li + 1. For the smaller cases, one can
check similarly.

The following example offers a view of the “splitting” in the proof of
Lemma 3.10.
Example 3.11 Let H be the whole hypergraph in the left side of Fig. 4. Let V1 be
the w4 = v2 (black part) of H and V2 = {w1 , w2 , w3 , w5 , w6 , w7 , w8 , w9 , w10 }
Fig. 4 Splitting hypergraph

(blue part) of H. Let G1 and G2 be the hypergraphs associated to the vertex sets
V1 and V2 . Then the edges {w3 , w4 }, {w4 , w5 }, and {w1 , w4 , w7 , w10 } (the purple
part) are the shared edges or edges of G1 and G2 . In the right of Fig. 4, we show the
hypergraphs for G1 and G2 separately.
Now with Lemma 3.10 we are ready to address the case when the sum of the ni
is 2k modulo 3. In this case Lemma 3.12 will be an instance of the special sub-case,
and Proposition 3.13 will give the general result.
Lemma 3.12 We adapt Notation 3.1. If r1 = 1 = rk+1 , and ri = 2 for all 1 < i <
k + 1, then pd(H) = pd(HSμ ) + 1.

R k+1 S
k+ i=1 ni
Proof First notice that pd(HSμ ) = k + k+1
i=1 ni − 3 and k+1i=1 ri = 2k.
We simplify pd(HSμ ) first.
P k+1 Q

k+1
k+ ni
pd(HSμ ) = k + ni − i=1
3
i=1
P k+1 Q

k+1
k + 2k + i=1 (3li )
= k + 2k + (3li ) −
3
i=1

k+1
= 2k + 2li .
i=1
Let xE be the variable corresponding to the edge Ex that connecting v1 = wi1

and the vertex of wi1 −1 and yE be the variable corresponding to the edge Ey that
connecting vk = wik and the vertex of wik +1 . We consider the short exact sequences:
0 → (H : Ex ) → H → (H, xE ) → 0,
and
0 → ((H : Ex ) : Ey ) → (H : Ex ) → ((H : Ex ), yE ) → 0.
Notice that ((H : Ex ) : Ey ) is a union of two isolated vertices, two strings

of length n1 − 2 and nk+1 − 2, and a hypergraph that satisfies assumptions of
Lemma 3.10. Now with assumptions of ri ’s, we have
S R R S
n1 − 2 nk+1 − 2
pd((H : Ex ) : Ey ) = 2 + n1 − 2 − + nk+1 − 2 −
3 3

k
+ 2(k − 1) + 2 li + 1
i=2

k
= 2 + 2l1 + 2lk+1 + 2(k − 1) + 2 li + 1
i=2

k+1
= 2k + 2li + 1 = pd(HSμ ) + 1.
i=1
union of an isolated vertex, a string of length n1 − 1, and a

Since (H, xE ) is a
string of length k−1+ k+1 i=2 ni with a k−2-dimensional edge such that n2 , . . . , nk+1

are the numbers of vertices between vertices of F and k+1 i=2 ri = 2k − 1. By
Proposition 3.6 and Theorem 2.4 part 4,
RS P Q
n1 − 1
k+1
k − 1 + k+1
i=2 ni
pd(H, xE ) = 1 + n1 − 1 − +k−1+ ni −
3 3
i=2
P k+1 Q

k+1
k − 1 + 2k − 1 + i=2 (3li )
= 2l1 + k + 2k − 1 + (3li ) −
3
i=2

k+1
< 2k + 2li + 1.
i=1
Moreover, ((H : Ex ), yE ) is a union of two isolated vertices, two strings of length

n1 − 2 and nk+1 − 1 and a hypergraph satisfies the assumptions of Lemma 3.7 with
a k − 2-dimensional edge. Then by Lemma 3.7 and Theorem 2.4 part 4,
S R R S
n1 − 2 nk+1 − 1
pd((H : Ex ), yE ) = 2 + n1 − 2 − + nk+1 − 1 −
3 3
P Q

k
k − 1 + ki=2 ni
+k−1+ ni −
3
i=2

k
= 1 + 2l1 + 2lk+1 + k + 2(k − 1) + (3li )
i=2
P k Q
k − 1 + 2k − 2 + i=2 (3li )
−
3

k+1
< 2k + 2li + 1 = pd((H : Ex ) : Ey ).
i=1
Since pd((H : Ex ) : Ey ) > pd((H : Ex ), yE ), we have
pd(H : Ex ) max{pd((H : Ex ) : Ey ), pd((H : Ex ), yE )} = pd((H : Ex ) : Ey )

and pd((H : Ex ) : Ey ) max{pd(H : Ex ), pd((H : Ex ), yE ) − 1}. This shows

pd(H : Ex ) = pd((H : Ex ) : Ey ). Similarly pd(H, xE ) < pd(H : Ex ) gives
pd(H) = pd(H : Ex ) = pd HSμ + 1.

k+1
We are now finally ready for the case, i=1 ri = 2k.

Proposition 3.13 We adapt Notation 3.1 and assume k+1 i=1 ri = 2k with r1 = 2. If
ri = 0 for all i, then pd(H) = (pd HSμ ) + 1, otherwise pd(H) = pd(HSμ ).
Proof We consider the short exact sequence
0 → Qv1 → Hv1 → H → 0.
We will show pd(Hv1 ) > pd(Qv1 ).

Moreover, we will show that when ri = 0 for all i, then pd(Hv1 ) = pd(HSμ ) + 1,
and otherwise pd(Hv1 ) = pd(HSμ ). These two claims with Remark 2.7 will prove
the proposition. We proceed by induction on k for both claims.
k+1
We observe that μ = k + k+1 i=1 ni and pd(HSμ ) = 2k + i=1 2li as in the proof
of Lemma 3.12.
When k = 2, observe that by Definition 2.6 and Discussion 2.8 in [14], Qv1 is a
union of two isolated vertices, and two strings of length n1 − 2 and n2 − 2 + n3 + 1.
Hence by Theorem 2.4 part 4, and r1 = 2 implies r2 + r3 = 2,
S
R R S
n1 − 2 n2 + n3 − 1
pd(Qv1 ) = 2 + n1 − 2 − + n2 + n3 − 1 −
3 3
= 2 + 2l1 + 2l2 + 2l3 + 1

4
<4+ 2li = pd(HSμ )
i=1
Notice Hv1 is a union of a string of length n1 and a string of length n2 + n3 + 1 with

two open strings having n2 − 1 and n3 − 1 open vertices. By Theorem 2.4 part 5 and
Theorem 3.4 in [13] with r2 + r3 = 2, we get
n R S R S
1 n2 − 1 − 1 n3 − 1 − 1
pd(Hv1 ) = n1 − + n2 + n3 + 1 − (2 + + )
3 3 3
= 2 + 2l1 + 2l2 + 2l3 + 3 = pd(HSμ ) + 1.
when ri = 0 for all i. When ri = 0 for some i, then

n R S R S
1 n2 − 1 − 1 n3 − 1 − 1
pd(Hv1 ) = n1 − + n2 + n3 + 1 − (2 + + )
3 3 3
= 2 + 2l1 + 2l2 + 2l3 + 2 = pd(HSμ ).
In both cases, we have pd(Hv1 ) > pd(Qv ) and this concludes the case when
k = 2.
Now suppose F is a k-dimensional edge with k + 1 vertices, v1 , . . . , vk+1 .
Suppose ri = 0 for all i then again by Definition 2.6 and discussion 2.8 in [14],
Qv1 is either
a union of two isolated vertices, a string of length n1 − 2 and a string of length
1.
k+2
i=2 ni + k − 2 when n2 2, or
2. Qv1 is a union of two isolated vertices, a string of length n1 − 2 and a string of

length k+2i=3 ni + k − 1 when n2 = 1.

For the later case, k+2
i=3 ri = 2k − 1, then by Theorem 2.4 part 4,
RS P Q
n1 − 2
k+2
k − 1 + k+2
i=3 ni
pd(Qv1 ) = 2 + n1 − 2 − +k−1+ ni −
3 3
i=3
P k+2 Q

k+2
k − 1 + 2k − 1 + 3li
= 2 + 2l1 + k − 1 + 2k − 1 + 3li − i=3
3
i=3

k+2
= 2l1 + 2k + (2li ) + 1
i=3

k+2
< 2k + 2 + (2li ) + 1.
i=1
For the first case, by Theorem 2.4 part 4, we have

SR P Q
n1 − 2
k+2
k − 2 + k+2 ni
pd(Qv1 ) = 2 + n1 − 2 − +k−2+ ni − i=2
3 3
i=2
P k+2 Q

k+2
k − 2 + 2k + 3li
= 2l1 + k + 2k + 3li − i=2
3
i=2

k+2
= 2k + 1 + 2li
i=1

k+2
< 2k + 2 + 2li + 1
i=1
with the fact r1 + . . . + rk+2 = 2k + 2 and r1 = 2.

Now in both cases consider that Hv1 is a union of a string of length n1 and a string
k+2
of length k+2i=2 ni + k with a k − 1-dimensional edge. Notice that i=2 ri = 2k
and ri = 0 for all i such that 2 i k + 2. By induction and Theorem 2.4 part 4,
we have
P Q
n
k+2
k + k+2
1 i=2 ni
pd(Hv1 ) = n1 − +k+ ni − +1
3 3
i=2

k+2
= 2k + 2 + 2li + 1 = pd(HSμ ) + 1
i=1
where we use the fact that r2 + . . . + rk+2 = 2k and r1 = 2. Therefore pd(Qv1 ) <
pd(Hv1 ), and pd(H) = pd(Hv1 ) = pd(HSμ ) + 1, satisfying the 2 claims.
Now suppose ri = 0 for some i > 1 then rj = 2 for all j = i. Notice again by
the discussions in [14] that Qv1 is either
a union of two isolated vertices, a string of length n1 − 2 and a string of length
1.
k+2
i=2 ni + k − 2 when n2 2, or
2. Qv1 is a union of two isolated vertices, a string of length n1 − 2 and a string of

length k+2i=3 ni + k − 2 when n2 = 0.

For the later case, we have k+2
i=3 ri = 2k. By Theorem 2.4 part 4,
R S P Q
n1 − 2
k+2
k − 2 + k+2
i=3 ni
pd(Qv1 ) = 2 + n1 − 2 − +k−2+ ni −
3 3
i=3
P k+2 Q

k+2
k − 2 + 2k + 3li
= 2l1 + k + 2k + 3li − i=3
3
i=3

k+2
= 2k + 1 + 2li
i=1

k+2
< 2k + 2 + 2li .
i=1
For the first case, by Theorem 2.4 part 4, we have

R S P Q
n1 − 2
k+2
k − 2 + k+2
i=2 ni
pd(Qv1 ) = 2 + n1 − 2 − +k−2+ ni −
3 3
i=2
P k+2 Q

k+2
k − 2 + 2k + 3li
= 2l1 + k + 2k + 3li − i=2
3
i=2

k+2
= 2k + 1 + 2li
i=1

k+2
< 2k + 2 + 2li .
i=1
with the fact that r2 + . . . + rk+2 = 2k and r1 = 2. Similar to the case where ri
is never 0, we get that Hv1 is a union of a string of length n1 and a string of length
k+2 k+2
i=2 ni + k with a k − 1-dimensional edge. Notice that i=2 ri = 2k and ri = 0
for some 2 i k + 2. So by induction and Theorem 2.4 part 4, we have
P k+2 Q
n
k+2
k+ ni
1
pd(Hv1 ) = n1 − +k+ ni − i=2
3 3
i=2

k+2
= 2k + 2 + 2li
i=1
= pd(HSμ )
where we use the fact that r2 + . . . + rk+2 = 2k and r1 = 2. Hence pd(Qv1 ) <
pd(Hv1 ), and pd(H) = pd(Hv1 ) = pd(HSμ ), thus finishing the proof.

Now tying together Proposition 3.5, Proposition 3.6, and Proposition 3.13 we can
prove the following result.
Theorem 3.14 Let HSμ be a string hypergraph such that it has μ vertices with all
vertices are open except the end vertices. We adapt Notation 3.1. If r1 +. . .+rk+1 =
2k and ri = 0 for all i, then pd(H) = pd(HSμ ) + 1 otherwise pd(H) = pd(HSμ ) =
% &
μ − μ3 .
Proof The final equality is coming from Theorem 2.4 part 4. For the cases when
r1 + . . . + rk+1 < 2k or r1 + . . . + rk+1 > 2k, we have pd(H) = pd(HSμ ) by
Proposition 3.5 and Proposition 3.6. We are left to consider the case, r1 + . . . +
rk+1 = 2k. If ri = 0 for some i, then rj = 2 for all j = i, hence Proposition 3.13
applies. If ri = 0 for all i, and r1 = rk+1 = 1, then Lemma 3.12 applies, otherwise,
we may assume r1 = 2 and Proposition 3.13 applies again.

Example 3.15 Three hypergraphs in Fig. 5 are strings attached with one extra blue
edges (checkerboard region) such that μ = 9, k = 3 and r1 + . . . + rk+1 = 2k = 6.
H1 and H2 satisfy ri = 0 for all i, but H3 has r4 = 0. Therefore pd(H1 ) =
pd(H2 ) = pd(HSμ ) + 1 = 9 − 3 + 1 = 7, and pd(H3 ) = pd(HSμ ) = 6 by
9
Theorem 3.14.
Fig. 5 Hypergraphs of Example 3.15
Remark 3.16 With Theorem 3.14, one can easily compute the projective dimension
of a string with extra edges. We first remove all the edges that are union of other
edges using Theorem 2.4 part 2. If we are left with one higher dimensional edge, we
can use Theorem 3.14 and Theorem 2.4 part 4 to compute the projective dimension.
Example 3.17 illustrates the process.

Example 3.17 Let H be the hypergraph shown in Fig. 6. We can remove all red
edges (dashed vertical line and the bricks regions) using Theorem 2.4 part 2 and
remove the blue edge (checkerboard regions)
using Theorem 3.14. The projective
dimension of the hypergraph is 11 − 11 3 = 8.

4 Cycles with Higher Dimensional Edges
Now we examine the case where we have added a higher dimensional edge to an
open cycle.
Notation 4.1 Let H be a hypergraph and F be an edge of H.
1. Let HCμ be an open cycle with vertex set V = {w1 , . . . , wμ }, and an edge F
with k > 1 vertices {v1 = wi1 , . . . , vk = wik } ⊆ {w1 , . . . , wμ } with 1 i1 <
· · · < ik μ. When we say H is an open cycle together with an edge consisting
of k vertices, we mean H = HCμ ∪ F .
2. Let nj = ij −ij −1 −1 for j = 1, . . . k be the number of vertices between vertices
of F .
3. We write ni = 3li + ri where li are some non-negative integers, and 0 ri 2

for i = 1, . . . , k + 1.
We start by showing that the induced hypergraph of HCμ on the complement
of VF has smaller projective dimension than that of HCμ when the sum of (the ni
modulo 3) is less than 2k − 1. This is necessary in the proof of Theorem 4.4.
Lemma 4.2 We adapt Notation 4.1. Let HVF = HCμ ∩ (V \VF ) be the hypergraph
obtained by removing all the vertices v1 , . . . , vk . If r1 + . . . + rk < 2k − 1, then
pd(HVF ) < pd HCμ .

k
Proof Notice that μ = k + i=1 ni and
R S
μ−2
pd HCμ = μ − 1 −
3
P Q

k
k − 2 + ki=1 ni
=k−1+ ni −
3
i=1
P k Q

k
k − 2 + 2k − 1 + i=1 3li
>k−1+ ni −
3
i=1

k
k
= ni − li
i=1 i=1
by Theorem 2.4 part

6 and the r1 + .
% assumption
& . . + rk < 2k − 1. On the other
hand, pd HVF = ki=1 (ni − n3i ) = ki=1 ni − ki=1 li because HVF is union of
k strings such that each string has ni vertices. Notice that this notation allows that
ni can be 0 for some i. Hence we have pd(HVF ) < pd HCμ .

Next, we show that when the sum of the ni modulo 3 is greater than 2k − 1 that
the projective dimension of H is the same as for the underlying cycle.

Lemma 4.3 We adapt Notation 4.1. If ki=1 ri 2k − 1, then pd(H) = pd(HCμ ).

k
Proof By Theorem 2.4 part 6 and the assumption 2k i=1 ri 2k − 1, we have
R S

k
k − 2 + ni=1 ni
pd(HCμ ) = k − 1 + ni −
3
i=1
P k k Q

k
k−2+ +
i=1 ri i=1 3li
=k−1+ ni −
3
i=1

k
k
= ni − li .
i=1 i=1

Since ki=1 ri 2k − 1, ri < 3, and k > 1, we have at most one ri such that ri = 1.
We may assume rk = 1 if there is one otherwise ri = 2 for all i. Let Hv1 = Hv be
the hypergraph removing the vertex v1 = v from H and let Hv : v1 = Qv . We have
a short exact sequence
0 → Qv → Hv → H → 0.
We will show that pd(Hv ) > pd(Qv ), then by Remark 2.7,

we conclude that pd(H) = pd(Hv ) = pd(HCμ ).
When k = 2, Hv is a string of length n1 + n2 + 1 with two open strings of n1 − 1
and n2 − 1 open vertices. Qv is the union of two closed vertices and an open string
of n1 − 2 + n2 − 2 + 1 vertices. By Theorem 2.4 part 4 and Theorem 3.4 in [13], we
have
R S R S
n1 − 2 n2 − 2
pd(Hv ) = n1 + n2 + 1 − (2 + + ) + 1 = 2l1 + 2l2 + 4,
3 3
and
R S
n1 + n2 − 3
pd(Qv ) = 2 + n1 + n2 − 3 − = 2l1 + 2l2 + 3
3
when r1 + r2 = 4, and
R S R S
n1 − 2 n2 − 2
pd(Hv ) = n1 + n2 + 1 − (2 + + ) + 1 = 2l1 + 2l2 + 3,
3 3
R S
n1 + n2 − 3
pd(Qv ) = 2 + n1 + n2 − 3 − = 2l1 + 2l2 + 2
3
when r1 + r2 = 3. R S
k k
k−1+ ki=1 ni
For k > 2, we show pd(Hv ) = k − 1 + i=1 ni − 3 = i=1 ni −
k k
i=1 li> pd(Qv ). Notice that Hv is a string of length μ − 1 = k − 1 + i=1 ni

attached with a k − 2-dimensional edge of k − 1 vertices. Moreover, ki=1 ri
2k − 1 = 2(k − 1) + 1. By Proposition 3.6,
P Q

k
k − 1 + ki=1 ni
k
k
pd(Hv ) = pd(HSμ−1 ) = k − 1 + ni − = ni − li .
3
i=1 i=1 i=1
On the other hand, Qv isthe union of two closed vertices and an open string of
length n1 − 2 + k − 3 + ki=2 ni hence by Theorem 2.4 part 4,
P Q

k
k − 5 + ki=1 ni
pd(Qv ) = 2 + k − 5 + ni −
3
i=1
P k Q

k
k − 5 + 2k − 1 + i=1 3li
k−3+ ni −
3
i=1

k
k
= ni − li − 1 < pd(Hv )
i=1 i=1
k
when i=1 ri 2k − 1.

Now we can show that the projective dimension is always preserved H and HCμ
differ by a single higher dimensional edge F .
Theorem 4.4 Let HCμ be the cycle hypergraph with μ open vertices. Let H =
HCμ ∪ F where F is a k − 1-dimensional edge with k vertices. Then pd(H) =

pd(HCμ ) = μ − 1 − μ−2
3 .

Proof The second equality is coming from Theorem 2.4 part 6. Let VF =
{v1 , . . . , vk }, which is a subset of the vertex set of HCμ , be the vertex set of F . Let

ni = 3li +ri be the spacing between the vi and defined as before. If ki=1 ri 2k−1
then
k the theorem holds by Lemma 4.3. So we need only consider the case when
i=1 ri < 2k − 1.
k
When i=1 ri < 2k − 1, we will use Lemma 4.2 and the short exact sequence
0 → (H : F ) = HCμ → H → (H, xF ) → 0
where xF is the variable corresponding to the edge F . The short exact sequence
gives pd(H) max{pd(HCμ ), pd(H, xF )}. Since (H, xF ) is the union of HVF and
an isolated vertex presenting the variable of xF , we have pd(H, xF ) = pd HVF +
1 pd HCμ by Lemma 4.2. By Theorem 2.4 part 3, we have to pd HCμ pd H.
Therefore we have
pd H max{pd(HCμ ), pd(H, xF )} = pd HCμ pd H.

Remark 4.5 As before, one can compute the projective dimension of an open cycle
with extra edges either using Theorem 2.4 part 2 or above theorem.

Remark 4.6 It is natural to conjecture that given an open cycle, no matter how
many higher dimensional edges are on the cycle the projective dimension of the
hypergraph is the same as the projective dimension of the open cycle. There are
more cases that one needs to consider for the proof of the conjecture. In particular,
in Theorem 3.14 we have a case for strings where the projective dimension does
not stay the same once a higher dimensional edge is removed. The proof of the case
when the projective dimension jumped up by one requires a lot of steps. In light of
this, new tools must be developed in order to prove the conjecture for cycles.

Acknowledgments The authors want to express heartfelt thanks to the reviewer for the sugges-
tions that greatly improved this paper.
References
1. Claude Berge, Hypergraphs. Combinatorics of finite sets. Translated from the French. North-
Holland Mathematical Library, 45. North-Holland Publishing Co., Amsterdam, (1989).
2. David Eisenbud, Commutative Algebra with a View Toward Algebraic Geometry, Graduate
Texts in Mathematics 150, Springer-Verlag New York (2004).
3. Shalom Eliahou and Michel Kervaire, Minimal resolutions of some monomial ideals. J. Algebra
129 (1990), 1–25.
4. John Eagon, Ezra Miller and Erika Ordog, Minimal resolutions of monomial ideals.
arXiv:1906.08837.
5. Giuliana Fatabbi, On the Resolution of Ideals of Fat Points. J. Algebra 242 (2001), 92–108.
6. Chris A. Francisco, Huy Tài Hà and Adam Van Tuyl, Splittings of monomial ideals. Proc.
Amer. Math. Soc. 137 (2009), 3271–3282.
7. Huy Tài Hà, Regularity of squarefree monomial ideals. Connections between algebra, combi-
natorics, and geometry, Springer Proc. Math. Stat., 76 Springer, New York (2014), 251–267.
8. Huy Tài Hà and Kuei-Nuan Lin, Normal 0-1 polytopes. SIAM J. Discrete Math. 29 (2015), no.
1, 210–223.
9. Sean Jacques, The Betti numbers of graph ideals. Ph.D. Thesis, The University of Sheffield
2004, arXiv.math.AC/0410107.
10. Sean Jacques and Mordechai Katzman, The Betti numbers of forests. arXiv.math.AC/0501226
11. Kyouko Kimura and Paolo Mantero, Arithmetical rank of strings and cycles. J. Commut.
Algebra 9 (2017), no. 1, 89–106.
12. Kyouko Kimura, Naoki Terai, and Ken-ichi Yoshida, Arithmetical rank of squarefree monomial
ideals of small arithmetic degree. J. Algebraic Combin. 29 (2009), no. 3, 389–404.
13. Kuei-Nuan Lin and Paolo Mantero, Projective dimension of string and cycle hypergraphs.
Comm. Algebra 44 (2016), no. 4, 1671–1694.
14. Kuei-Nuan Lin and Paolo Mantero, High projective dimension and 1-dimensional hypergraphs.
Int. J. Algebra Comput. 27, No. 6 (2017), 591–617.
15. Kuei-Nuan Lin and Sonja Mapes, Lattices and Hypergraphs associated to square-free mono-
mial ideals arXiv:1804.05919.
16. Kuei-Nuan Lin and Jason McCullough, Hypergraphs and regularity of squarefree monomial
ideals. Internat. J. Algebra Comput. 23 (2013), no. 7, 1573–1590.
A Truncated Minimal Free Resolution of
the Residue Field
Van C. Nguyen and Oana Veliche
Keywords Golod rings · Minimal free resolutions · Koszul complex · Tor

algebra · Massey products
1 Introduction
Throughout the paper, (R, m, k) is a local Noetherian ring with maximal ideal m
and residue field k, of embedding dimension n and codepth c = n − depth R. Let
K R be the Koszul complex of R on a minimal set of generators of m. Serre pointed
out that there is always a coefficient-wise inequality between the Poincaré series of
the R-module k and a rational series involving the ranks of the homologies of K R :
∞
(1 + t)n
k (t) :=
PR i (k, k)t
rankk TorR i c . (1)
1− R i+1
i=1 rankk Hi (K )t
i=0
In [8], Golod proved that a ring R attains this upper bound if and only if the
graded-commutative algebra H(K R ) has trivial multiplications and trivial Massey
operations; such a ring is now called a Golod ring. In the same paper, he also
constructed the minimal free resolution of the R-module k in terms of K R .
In [2, Corollary 5.10], Avramov proved that the Poincaré series PR k (t) is com-
pletely determined by the structure of the Koszul homology algebra A := H(K R )
as an algebra with Massey operations. One can now ask the following question:
V. C. Nguyen
Department of Mathematics, United States Naval Academy, Annapolis, MD, USA
e-mail: vnguyen@usna.edu
O. Veliche ()
Department of Mathematics, Northeastern University, Boston, MA, USA
e-mail: o.veliche@northeastern.edu
400 V. C. Nguyen and O. Veliche
Question 1.1 For any local ring, given the knowledge of its Koszul homology
algebra, with its products and Massey operations, how does one construct a minimal
free resolution of its residue field?
In this paper, we answer this question explicitly up to degree five. For any local
ring R, using the components of the Koszul complex K R as building blocks and
the graded-commutative structure of its homology A, we construct a minimal
free resolution of the R-module k, up to degree five, see Construction 3.1 and
Theorem 3.1:
∂5F ∂4F ∂3F ∂2F ∂1F

F : F5 −→ F4 −→ F3 −→ F2 −→ F1 −→ F0 → k → 0.
The higher degrees of the resolution can be extended similarly for some special
cases of R, but in general it requires an understanding of higher Massey products,
which remains elusive and is left for future projects. Indeed, Massey operations
can be represented by using the A∞ -algebra structure, which the Koszul homology
algebra A possesses, see for example, [12] by Lu, Palmieri, Wu, and Zhang.
Moreover, the A∞ -structure was also used by Burke in [6] to construct certain
projective resolutions. These connections suggest that perhaps the A∞ -algebra
structure of A may play a role in giving a complete answer to Question 1.1 from
this perspective.
To prepare for Construction 3.1, in Sect. 2 we analyze in detail the multiplicative
structure of the algebra A, up to degree four, in terms of bases of Ai and Ai · Aj ,
and necessary maps. Moreover, in degree four the Massey products appear for the
first time, as ternary Massey products A1 , A1 , A1 ; we give a description of the
elements of this set in Proposition 2.8.
In Sect. 4, we obtain several direct applications of Theorem 3.1. In Corollary 4.1,
we explicitly describe the Betti numbers βi := rankk TorR i (k, k), up to degree five,
in terms of the multiplicative invariants of the Koszul homology algebra A. Let
ai := rankk Ai and consider the difference of series
(1 + t)n
P(t) := − PR
k (t)
1 − ci=1 ai t i+1
that measures how far the ring R is from being Golod, that is, how far the Betti
numbers of R are from their maximum possible values. In Proposition 4.2, we
compute the first six coefficients of P(t) in terms of multiplicative invariants of A:
P0 = P1 = P2 = 0, P3 = q11 , P4 = (n + 1)q11 + q12 ,

n+1! !
P5 = 2 + 2a1 q11 + (n + 1)q12 + a − b,
where qij = rankk (Ai · Aj ), and a, b are described in Summary 2.5. For any local
k (t) =
ring R of embedding dimension n with rational Poincaré series of the form PR
A Truncated Minimal Free Resolution of the Residue Field 401
(1+t)n
d(t) , we express the denominator, up to degree five, as follows:
c !
d(t) = 1 − i=1 ai t
i+1 + q11 t 3 + (q11 + q12 )t 4 + (q12 − b + a)t 5 + f (t)t 6 ,
for some f (t) ∈ Z[t], see Proposition 4.5. We consider a few classes of rings with
such rational Poincaré series, see Corollary 4.10 and Examples 4.11 and 4.12, and
using the coefficients of the denominator we obtain algebraic invariants of their
Koszul homology algebra A. The Poincaré series PR k (t) can be described by using
the deviations εi ’s, see for example [4]. In Corollary 4.13, we give a description of
the first five deviations in terms of the algebraic invariants of A.
In Sect. 5, we illustrate Construction 3.1 through an example of a ring of codepth
4 examined in [2]. In that example, Avramov provided a nontrivial indecomposable
Massey product element in A1 , A1 , A1 that does not come from multiplications of
the homology. Using Proposition 2.8, we prove that, up to a scalar, this is the only
element with this property, modulo the products in homology.
2 Multiplicative Structure on the Homology of the Koszul

Algebra
Let (R, m, k) be a local ring of embedding dimension n and codepth c. The

differential graded algebra structure on the Koszul complex K = K R on a minimal
set of generators of the maximal ideal m induces a graded-commutative algebra
structure on the homology of K:
A := H(K) = A0 ⊕ A1 ⊕ A2 ⊕ · · · ⊕ Ac .
In this section we discuss this multiplicative structure up to degree four. We will use
this structure extensively in constructing a truncated minimal free resolution of the
residue field over R in the next section.
The following notation is used throughout the paper: the differential map on
K is denoted by ∂ K , a homogeneous element of degree i in the Koszul complex
K is denoted by π i , a representative in Ki of an element in the homology Ai is
denoted by pi , and elements of R are denoted by α’s, β’s, and γ ’s. The subscript of
a homogeneous element indicates its index in a tuple and the superscript indicates
its homological degree. Set
ai := rankk (Ai ) and qij := rankk (Ai · Aj ), for all 0 i, j c. (2.1)
It is clear that A0 = k. For each 1 i a1 we consider zi1 ∈ Ker ∂1K such that
{[zi1 ]}i=1,...,a1 is a basis of A1 . (2.2)

In particular, every element in Ker ∂1K can be written as a sum of elements in Im ∂2K
and a linear combination of {zi1 }i=1,...,a1 .
2.1 Products in Degree Two
Let A2 be a k-subspace of A2 such that
A2 = (A1 · A1 ) ⊕ A2 .
For each 1
a2 − q11 , we consider z
2 ∈ Ker ∂2K such that
{[z
2 ]}
=1,...,a2 −q11 is a basis of A2 . (2.3)
In particular, every element in Ker ∂2K can be written as a sum of elements in

Im ∂3K , a linear combination of {zi1 ∧ zj1 }i,j =1,...,a1 , and a linear combination of
{z
2 }
=1,...,a2 −q11 .
Consider the multiplication map:
φ1 : A1 ⊗ A1 → A2 , defined by φ1 ([x] ⊗ [y]) = [x] ∧ [y], (2.4)
for all [x], [y] ∈ A1 . As Im φ1 = A1 · A1 , we have rankk (Ker φ1 ) = a12 − q11 . For
all 1 i a1 and 1 s a12 − q11 , we choose psi
1 ∈ Ker ∂ K such that
1
#
a1 $
[zi1 ] ⊗ [1
psi ] is a basis of Ker φ1 . (2.5)
s=1,...,a12 −q11
i=1
a1
i=1 [zi ] ∧ [
For each s, we have 1 1]
psi = 0, so there exists
πs3 ∈ K3 such that

a1
πs3 ) =
∂3K ( zi1 ∧ p
si
1
. (2.6)
i=1
Lemma 2.1 Let p 1 ∈ Ker ∂1K be as in (2.5). The vectors

! si
{ [ psa1 ] } s=1,...,a 2 −q11 in Aa11 are linearly independent.
ps1 ], . . . , [
1 1
1
a12 −q11 !
Proof If αs is in R such that s=1 [1 ], . . . , [
ps1 1 ] ∧ [α ] = 0, then
psa 1 s
a12 −q11

[1
psi ] ∧ [αs ] = 0, for all 1 i a1 .
s=1
By tensoring with [zi1 ] and taking the sum over i we get

a12 −q11 a12 −q11 8 a1 9

a1
[zi1 ] ⊗ [1
psi ] ∧ [αs ] = [zi1 ] ⊗ [1
psi ] ∧ [αs ] = 0.
i=1 s=1 s=1 i=1
The desired conclusion now follows from (2.5).

Proposition 2.2 Let {[zi1 ]}i=1,...,a1 and {[z
2 ]}
=1,...,a2 −q11 be as in (2.2) and (2.3),
respectively. If βij and α
are in R such that

a1 −q11
a2
[zi1 ] ∧ [zj1 ] ∧ [βij ] + [z
2 ] ∧ [α
] = 0,
i,j =1
=1
then the following hold:

(a) [α
] = 0 for all 1
a2 − q11 ;
(b) There exist γs in R such that for all 1 i a1
a1 −q112

a1
[zj ] ∧ [βij ] =
1
[1
psi ] ∧ [γs ],
j =1 s=1
si
where p 1 is as in (2.5).
Proof
(a) The first sum in the hypothesized equality is in A1 · A1 . The elements
{[z
2 ]}
=1,...,a2 −q11 form a basis in A2 that completes a basis of A1 · A1 , thus
[α
] = 0 for all 1
a2 − q11 .
(b) By (a) it follows that

a1
a1 8
a1 9
0= [zi1 ] ∧ [zj1 ] ∧ [βij ] = [zi1 ] ∧ [zj1 ] ∧ [βij ] ,
i,j =1 i=1 j =1
1 1 8 9
which implies that ai=1 [zi ] ⊗ a1
[z
j =1 j
1 ] ∧ [β ] is in Ker φ . The desired
ij 1
conclusion now follows from (2.5).

2.2 Products in Degree Three
A3 = (A1 · A2 ) ⊕ A3 = (A1 · A1 · A1 + A1 · A2 ) ⊕ A3 .
For each 1 t a3 − q12 , we consider zt3 ∈ Ker ∂3K such that
{[zt3 ]}t=1,...,a3 −q12 is basis of A3 . (2.7)
In particular, every element in Ker ∂3K can be written as a sum of elements in

Im ∂4K , a linear combination of {zi1 ∧ zj1 ∧ zk1 }i,j, k=1,...,a1 , a linear combination of
{zi1 ∧ z
2 } i=1,...,a1 , and a linear combination of {zt3 }t=1,...,a3 −q12 .

=1,...,a2 −q11
Consider the map:
! !
ψ : A1 ⊗ A1 ⊗ A1 → (A1 · A1 ) ⊗ A1 ⊕ A1 ⊗ (A1 · A1 ) , (2.8)
defined by
8 9
ψ([x] ⊗ [y] ⊗ [z]) = [x] ∧ [y] ⊗ [z], [x] ⊗ [y] ∧ [z]
for all [x], [y], [z] ∈ A1 and set
b = rankk (Coker ψ). (2.9)
Remark that b = 0 if and only if ψ is surjective. If q11 = 0, then A1 · A1 = 0 and

hence b = 0.
Lemma 2.3 Let ai , qij be as in (2.1) and b be as in (2.9). The sequence
ψ ! ! μ
→ (A1 · A1 ) ⊗ A1 ⊕ A1 ⊗ A2 −
A1 ⊗ A1 ⊗ A1 − → A3 ,
where
8 9
ψ([x] ⊗ [y] ⊗ [z]) = [x] ∧ [y] ⊗ [z], [x] ⊗ [y] ∧ [z] , and
μ(([u] ⊗ [x], [y] ⊗ [v])) = [u] ∧ [x] − [y] ∧ [v],
for all [x], [y], [z] ∈ A1 , [u] ∈ (A1 ·A1 ) and [v] ∈ A2 , is a complex whose homology
has rank
a1 a2 − a1 q11 − q12 + b.
Proof We first show that μ ◦ ψ = 0.

88 99
μ(ψ([x] ⊗ [y] ⊗ [z])) = μ [x] ∧ [y] ⊗ [z], [x] ⊗ [y] ∧ [z]
= [x] ∧ [y] ∧ [z] − [x] ∧ [y] ∧ [z] = 0.

Hence, the sequence in the statement of the lemma is a complex. Moreover, we have
8 Ker μ 9
rankk = rankk (Ker μ) − rankk (Im ψ)
Im ψ
8 ! !9
= rankk (A1 · A1 ) ⊗ A1 ⊕ A1 ⊗ A2 − rankk (Im μ) − rankk (Im ψ)
8 ! !9
= a1 q11 + a1 a2 − q12 − rankk (A1 · A1 ) ⊗ A1 ⊕ A1 ⊗ (A1 · A1 ) + b
= a1 q11 + a1 a2 − q12 − 2a1 q11 + b

= a1 a2 − a1 q11 − q12 + b.
The homology of the complex in the statement has the desired rank.

To describe a basis of the homology of the complex in Lemma 2.3, it is enough
to observe that Im ψ is generated by elements of the form
([zi1 ] ∧ [zj1 ] ⊗ [zk1 ], [zi1 ] ⊗ [zj1 ] ∧ [zk1 ])

L M
for 1 i, j, k a1 and that in Ker μ/Im ψ we have ([zi1 ] ∧ [zj1 ] ⊗ [zk1 ], 0) =
L M
− (0, [zi1 ] ⊗ [zj1 ] ∧ [zk1 ]) , as [([zi1 ] ∧ [zj1 ] ⊗ [zk1 ], [zi1 ] ⊗ [zj1 ] ∧ [zk1 ])] = 0. We
ui
choose p 2 ∈ Ker ∂ K such that
2
#L 8
a1 9M$
0, [zi1 ] ⊗ [2
pui ] (2.10)
u=1,...,a1 a2 −a1 q11 −q12 +b,
i=1
is a basis of the homology

1 of1 the complex defined in Lemma 2.3. By definition of μ,
for each u we have ai=1 [zi ] ∧ [2 ] = 0 in A . Therefore, for each u there exists
pui 3

πu4 ∈ K4 such that

a1
πu4 ) =
∂4K ( zi1 ∧ p
ui
2
. (2.11)
i=1
ui
The following result gives a method for finding p 2 from (2.10).
Proposition 2.4 Consider the map φ2 : A1 ⊗ A2 → A3 , defined by
φ2 ([x] ⊗ [v]) = [x] ∧ [v],
for all [x] ∈ A1 , [v] ∈ A2 , and set

#
a1
" $
A= [zi1 ] ⊗ [1
psi ] ∧ [zj1 ] " 1 j a1 and 1 s a12 − q11 and
i=1
#
a1
" $
2 "
B= [zi1 ] ⊗ [
pui ] 1 u a1 a2 − a1 q11 − q12 + b ,
i=1
si
where p ui
1 and p 2 are as in (2.5) and (2.10) respectively. Then, the following hold:
(a) The set B is linearly independent and
Ker φ2 = (Spank A) ⊕ (Spank B).
(b) Spank A ⊆ A1 ⊗ (A1 · A1 ) and for b defined in (2.9),
b = a1 q11 − rankk (Spank A).
(c) If B is a k-subspace of Ker φ2 such that
Ker φ2 = (Spank A) ⊕ B,
then for every basis B of B the set [(0, B )] is a basis of the homology of the
complex in Lemma 2.3.
Proof
(a) The linear independence of B follows from (2.10). Next, we show that the
elements of A and B generate the kernel of φ2 . By definitions of p 1 and pui
2 , it
a1 si1
is clear that the listed elements are in the kernel of φ2 . Let i=1 [zi ] ⊗ [pi2 ] ∈
1 1
Ker φ2 , for some [pi2 ] ∈ A2 . Then [(0, ai=1 [zi ] ⊗ [pi2 ])] ∈ Ker μ, and thus by
definition (2.10) for all u, i, j there exist δu ∈ R and pij 1 ∈ Ker ∂ K such that
1
a1 a2 − a1 q11
8 a1 9 −q12 + b
8 a1 9
0, [zi1 ] ⊗ [pi2 ] = 0, [zi1 ] ⊗ [2
pui ] ∧ [δu ] (2.12)
i=1 u=1 i=1
a1 8
9
+ [zi1 ] ∧ [pij
1
] ⊗ [zj1 ], [zi1 ] ⊗ [pij
1
] ∧ [zj1 ] .
i,j =1
Comparing the first components of both sides of (2.12), for each j we have:

a1
0= [zi1 ] ∧ [pij
1
].
i=1
By (2.5), there exist εj s in R such that for all 1 i, j a1 we can write:

a12 −q11

[pij
1
]= [1
psi ] ∧ [εj s ].
s=1
Comparing the second components of both sides of (2.12), and using the above
1 ], for each i we have:
expression for [pij
a1 a2 − a1 q11
−q12 + b

a1
[pi2 ] = [2
pui ] ∧ [δu ] + [pij
1
] ∧ [zj1 ]
u=1 j =1
a1 a2 − a1 q11
−q12 + b a12 −q11 a1

= [2
pui ] ∧ [δu ] + [1
psi ] ∧ [zj1 ] ∧ [εj s ].
u=1 s=1 j =1
a1
Thus, i=1 [zi1 ] ⊗ [pi2 ] ∈ Ker φ2 is a linear combination of elements of A and
B. Next, we show that Spank A ∩ Spank B = {0}. By definition of p si1 we have
a 1 1
i=1 [zi ] ∧ [
psi ] = 0 for all 1 s a1 − q11 . By definition of ψ, for all s, j ,
1 2
we get
8
a1 9 8 a1 9
ψ [zi1 ] ⊗ [1
psi ] ⊗ [zj1 ] = 0, [zi1 ] ⊗ [1
psi ] ∧ [zj1 ] , (2.13)
i=1 i=1
hence (0, A) ⊆ Im ψ. On the other hand by (2.10), [(0, B)] is a basis of the
homology of the complex in Lemma 2.3. Therefore, Spank A and Spank B have
no nontrivial elements in common, and part (a) holds.
(b) The inclusion follows from the definition of A. Observe that rankk (Ker φ2 ) =
a1 a2 −q12 , and by the linear independence of B from part (a), rankk (Spank B) =
a1 a2 − q12 − (a1 q11 − b). Therefore, by part (a), rankk (Spank A) = a1 q11 − b.
(c) By parts (a) and (b), every basis B of B has a1 a2 − a1 q11 − q12 + b elements.
It is clear that (0, B ) ⊆ Ker μ, with μ as in Lemma 2.3 and that the set
[(0, B )] has at most a1 a2 − a1 q11 − q12 + b elements. It is enough to show
that it is a generating set for the homology of the complex in Lemma 2.3 to
conclude that [(0, B )] has exactly a1 a2 − a1 q11 − q12 + b elements, hence it
forms a basis for the homology. Since B ⊆ Ker φ2 , every element of B can be
written as a linear combination of elements of A and B . In particular, every
basis element in [(0, B)] can be written as a linear combination of elements
in [(0, A)] and [(0, B )]. However, each element in [(0, A)] is zero in the
homology, as remarked in (2.13), thus [(0, B )] is a generating set. Part (c) now
holds.

Remark 2.5 In practice, one can apply Proposition 2.4(b) to compute the value b,
ui
instead of using definition (2.9). Similarly, instead of using definition (2.10) for p 2 ,
si
by Proposition 2.4(c), one can find first p 1 as defined in (2.5), and then find a basis
of a space complementary to
X

a1
Spank A = Spank [zi1 ] ⊗ [1
psi ] ∧ [zj1 ]
i=1 1j a1 ,1sa12 −q11
in the kernel of the multiplication map φ2 , of the form

X

a1
[zi1 ] ⊗ [2
pui ] .
i=1 1ua1 a2 −a1 q11 −q12 +b
Proposition 2.6 Let {[zi1 ]}i=1,...,a1 , {[z

2 ]}
=1,...,a2 −q11 , and {[zt3 ]}t=1,...,a3 −q12 be as
in (2.2), (2.3), and (2.7), respectively. If γij k , βi
, and αt are in R such that

a1 −q11
a1 a2 −q12
a3
[zi1 ]∧[zj1 ]∧[zk1 ]∧[γij k ]+ [zi1 ]∧[z
2 ]∧[βi
]+ [zt3 ]∧[αt ] = 0,
i,j,k=1 i=1
=1 t=1
then the following hold:

(a) [αt ] = 0 for all 1 t a3 − q12 ;
(b) There exist δu and εj s in R such that

a1 −q11
a2
[zj1 ] ∧ [zk1 ] ∧ [γij k ] + [z
2 ] ∧ [βi
]
j,k=1
=1
a1 a2 − a1 q11
−q11
a1 a1
2
−q12 + b

= [1
psi ] ∧ [zj1 ] ∧ [εj s ] + [2
pui ] ∧ [δu ],
j =1 s=1 u=1
si
where p ui
1 and p 2 are as in (2.5) and (2.10) respectively.
Proof
(a) The first two sums in the hypothesized equality are in A1 · A2 . As
{[zt3 ]}t=1,...,a3 −q12 is a basis in A3 that completes a basis of A1 · A2 , this
implies [αt ] = 0 for all 1 t a3 − q12 .
(b) By (a) it follows that

a1 8
a1 −q11
a2 9
[zi ] ∧
1
[zj ] ∧ [zk ] ∧ [γij k ] +
1 1
[z
2 ] ∧ [βi
] = 0,
i=1 j,k=1
=1
1 1 a1 a2 −q11 2 !

hence ai=1 [zi ] ⊗ j,k=1 [zj ] ∧ [zk ] ∧ [γij k ] +
1 1

=1 [z
] ∧ [βi
] is in
Ker φ2 . The desired assertion now follows from Proposition 2.4(a).

2.3 Massey Products in Degree Four
Massey products occur in degrees four and higher. In degree four, one may obtain
only triple Massey products of elements of degree one. The Massey product of a
triplet [x], [y], [z] ∈ A1 satisfying
[x] ∧ [y] = 0 and [y] ∧ [z] = 0,
is a subset of A4 , defined as follows:
[x], [y], [z] = {[πxy

3
∧ z + x ∧ πyz
3
] | ∂3K (πxy
3
) = x ∧ y and ∂3K (πyz
3
) = y ∧ z},
3 , π3 ∈ K .
for some πxy yz 3
Remark 2.7 Let [x], [y], [z] ∈ A1 with [x] ∧ [y] = 0 and [y] ∧ [z] = 0. Choose
3 and π 3 in K such that ∂ K (π 3 ) = x ∧ y and ∂ K (π 3 ) = y ∧ z. Then, every
πxy yz 3 3 xy 3 yz
element of [x], [y], [z] is of the form
[(πxy
3
+ pxy
3
) ∧ z + x ∧ (πyz
3
+ pyz
3
)] = [πxy
3
∧ z + x ∧ πyz
3
]
+ [pxy
3
] ∧ [z] + [x] ∧ [pyz
3
],
3 and p 3 ∈ Ker ∂ K . Therefore,

for some pxy yz 3
[x], [y], [z] = [πxy

3
∧ z + x ∧ πyz
3
] + (A3 · [z]) + ([x] · A3 ).
The element πxy 3 ∧ z + x ∧ π 3 is called a representative of the triple Massey product

yz
[x], [y], [z]. Here, (A3 · [z] + [x] · A3 ) is called the indeterminacy of the Massey
operation, see e.g., May [14] and [2, Section 7].
Let A1 , A1 , A1 denote the set of elements in A4 which are Massey products of
triplets in A1 . The next result describes the elements of this set.
Proposition 2.8 Each element in A1 , A1 , A1 ⊆ A4 has a representative
a12 −q11 a12 −q11

πs3 ∧ ps1 such that [1
psi ∧ ps1 ] = 0, for all i = 1 . . . , a1 ,
s=1 s=1
where ps1 ∈ Ker ∂1K , and p

si
1 and
πs3 are defined in (2.5) and (2.6) respectively.
Proof Let [x], [y], [z] ∈ A1 such that [x] ∧ [y] = 0 and [y] ∧ [z] = 0. We write

a1
a1
[y] = [zi1 ] ∧ [αi ] for some αi ∈ R. Since [x] ∧ [y] = [zi1 ] ∧ [x] ∧ [−αi ] = 0,
i=1 i=1

a1
we have [zi1 ] ⊗ [x] ∧ [−αi ] ∈ Ker φ1 . Hence, by (2.5) there exist βs ∈ R such
i=1
a12 −q11
that [x] ∧ [−αi ] = s=1 [1 ] ∧ [β ] for all i, so
psi s
a12 −q11 8 a1 9

[x] ∧ [y] = [zi1 ∧ p
si
1
] ∧ [βs ]
s=1 i=1
a12 −q11 L
2 −q11
8 a1 9M

= [∂3K (
πs3 )] ∧ [βs ] = ∂3K
πs3 ∧ βs .
s=1 s=1
3 =
a12 −q11
Therefore, we may choose πxy s=1
πs3 ∧ βs . Similarly, using the equality
a12 −q11

3 =
[y] ∧ [z] = 0, we may choose πyz
πs3 ∧ γs , for some γs ∈ R, where
s=1
a12 −q11
[z]∧[αi ] = s=1 [1 ]∧[γ ] for
psi s all i. It follows that any element in A1 , A1 , A1
has a representative
a12 −q11

3
πxy ∧z+x ∧ πyz
3
=
πs3 ∧ (z ∧ βs − x ∧ γs ).
s=1
If we set ps1 = z ∧ βs − x ∧ γs , then for each 1 i a1 we have
a12 −q11 a12 −q11

[1
psi ∧ ps1 ] = [1
psi ] ∧ [z ∧ βs − x ∧ γs ]
s=1 s=1
−q11
8 a1
2
9 −q11
8 a1
2
9
= [1
psi ∧ βs ] ∧ [z] + [x] ∧ [1
psi ∧ γs ]
s=1 s=1
= −[x] ∧ [z] ∧ [αi ] + [x] ∧ [z] ∧ [αi ] = 0.
Hence the proposition holds.

2.4 Products in Degree Four
A4 = (A1 · A3 + A2 · A2 + Spank A1 , A1 , A1 ) ⊕ A4
and set
8 9
a = rankk A1 · A3 + A2 · A2 + Spank A1 , A1 , A1 . (2.14)
For some choice of zr4 ∈ Ker ∂4K , let
{[zr4 ]}r=1,...,a4 −a be a basis of A4 . (2.15)
In particular, every element in Ker ∂4K can be written as a sum of elements

in Im ∂5K , a linear combination of {zi1 ∧ pi3 }i=1,...,a1 , a linear combination of
{z
2 ∧ p
2 }
=1,...,a2 −q11 , a linear combination of {
πs3 ∧ ps1 }s=1,...,a 2 −q11 , and a linear
1
combination of {zr4 }r=1,...,a4 −a , where pi3 ∈ Ker ∂3K , p
2 ∈ Ker ∂2K and ps1 ∈ Ker ∂1K ,
such that ps1 is as in Proposition 2.8 and πs3 as in (2.6).
2.5 Summary
We summarize here all notations, introduced in this section, to be referred to

throughout the rest of the paper:
A = H(K R ) = A0 ⊕ A1 ⊕ A2 ⊕ · · ·
A2 = (A1 · A1 ) ⊕ A2
A3 = (A1 · A2 ) ⊕ A3
!
A4 = A1 · A3 + A2 · A2 + Spank (A1 , A1 , A1 ) ⊕ A4
ai = rankk Ai , for 1 i 4, (2.1)

qij = rankk Ai · Aj , for 1 i j 4, (2.1)
!
a = rankk A1 · A3 + A2 · A2 + Spank A1 , A1 , A1 , (2.14)
{[zi1 ]}i=1,...,a1 is a basis for A1 , (2.2)

{[z
2 ]}
=1,...,a2 −q11 is a basis for A2 , (2.3)
{[zt3 ]}t=1,...,a3 −q12 is a basis for A3 , (2.7)

{[zr4 ]}r=1,...,a4 −a is a basis for A4 , (2.15)
! !
ψ : A1 ⊗ A1 ⊗ A1 → (A1 · A1 ) ⊗ A1 ⊕ A1 ⊗ (A1 · A1 ) , (2.8)
8 9
ψ([x] ⊗ [y] ⊗ [z]) = [x] ∧ [y] ⊗ [z], [x] ⊗ [y] ∧ [z] for all [x], [y], [z] ∈ A1
b = rankk (Coker ψ), (2.9), see also Proposition 2.4(b)
φ1 : A1 ⊗ A1 → A2 , φ1 ([x] ⊗ [y]) = [x] ∧ [y], for all [x], [y] ∈ A1 , (2.4)

#
a1 $
[zi1 ] ⊗ [1
psi ] is a basis of Ker φ1 , (2.5)
s=1,...,a12 −q11
i=1

a1
πs3 ) =
∂3K ( zi1 ∧ p
si
1
, (2.6)
i=1
φ2 : A1 ⊗ A2 → A3 , φ2 ([x] ⊗ [y]) = [x] ∧ [y], for all [x] ∈ A1 , [y] ∈ A2 , Prop. 2.4
#
a1 $
[zi1 ] ⊗ [2
pui ] is a basis of B, where
1ua1 a2 −a1 q11 −q12 +b
i=1
#
a1 $
Ker φ2 = B ⊕ Spank [zi1 ] ⊗ [1
psi ] ∧ [zj1 ]
1j a1 ,1sa12 −q11
i=1

a1
πu4 ) =
∂4K ( zi1 ∧ p
ui
2
, (2.11).
i=1
si
Lemma 2.9 All the elements zi1 , z
2 , zt3 , zr4 , p 1 , ui
πs3 , p 2 , and
πu4 , as in Sum-
mary 2.5, are in mK.
Proof From the inclusions Ker ∂jK ⊆ mKj for all j 0, we obtain that
si
zi1 , z
2 , zt3 , zr4 , p ui
1 , and p 2 are in mK. The assertions for the preimages
πs3 and
πu4
K −1
follows from the inclusions (∂j ) (m Kj −1 ) ⊆ mKj for j = 3 and 4 respectively.
2
3 Truncated Minimal Free Resolution of the Residue Field
In this section, we construct the beginning of the minimal free resolution of the
residue field k over the local ring (R, m, k), by using the Koszul complex K of R
and the graded-commutative structure of the algebra A = H(K) described in Sect. 2.
3.1 Construction
We consider the following sequence of free R-modules:
∂5F ∂4F ∂3F ∂2F ∂1F

F : F5 −→ F4 −→ F3 −→ F2 −→ F1 −→ F0 ,
where
F0 :=K0
F1 :=K1
F2 :=K2 ⊕ K0a1
a −q11
F3 :=K3 ⊕ K1a1 ⊕ K0 2
a −q11 a −q12 a 2 −q11
F4 :=K4 ⊕ K2a1 ⊕ K1 2 ⊕ K0 3 ⊕ K0 1
a −q11 a −q12 a 2 −q11
F5 :=K5 ⊕ K3a1 ⊕ K2 2 ⊕ K1 3 ⊕ K0a4 −a ⊕ K1 1
a a2 −a1 q11 −q12 +b a a2 −a1 q11
⊕ K0 1 ⊕ K0 1 .
Using the elements described in Sect. 2, the differential maps are defined by:
∂1F : K1 → K0 , is given by ∂1F := ∂1K . (3.1)
!
∂2F : K2 ⊕ K0a1 → K1 , is given by ∂2F := ∂2K z1 ∧ , (3.2)
that is

π2
a1
∂2F := ∂2K (π 2 ) + zi1 ∧ αi .
(αi )i=1,...,a1 i=1

a −q ∂3K z1 ∧ −z2 ∧
∂3F : K3 ⊕K1a1 ⊕K0 2 11 → K2 ⊕K0a1 , is given by ∂3F := ,
0 (∂1K )a1 0
(3.3)
that is
⎛ ⎞ ⎛ −q11
⎞
π3
a1 a2
⎜ ⎟ ⎜∂3K (π 3 ) + zi1 ∧ πi1 − z

2 ∧ α
⎟
∂3F ⎝ (πi1 )i=1,...,a1 ⎠ := ⎜
⎝ i=1
=1
⎟.
⎠
(α
)
=1,...,a2 −q11 K 1
(∂ (π ))i=1,...,a 1 i 1
a −q11 a −q12 a 2 −q11 a −q11

∂4F : K4 ⊕K2a1 ⊕K1 2 ⊕K0 3 ⊕K0 1 → K3 ⊕K1a1 ⊕K0 2 , is given by
⎛ K 1 ⎞ (3.4)
∂4 z ∧ z2 ∧ z3 ∧ −
π 3∧
∂4F := ⎝ 0 (∂2K )a1 0 0 (p1 ∧)a1 ⎠ ,
0 K
0 (∂1 ) a 2 −q 11 0 0
that is
⎛ ⎞
π4
⎜ ⎟
⎜ (πi2 )i=1,...,a1 ⎟
⎜ ⎟
⎜ ⎟
∂4F ⎜(π
1 )
=1,...,a2 −q11 ⎟
⎜ ⎟
⎜ (αt )t=1,...,a −q ⎟
⎝ 3 12 ⎠
(βs )s=1,...,a 2 −q11

1
⎛ ⎞
−q11 −q12 a12 −q11
a1 a2 a3
⎜∂ K (π 4 ) + zi1 ∧ πi2 + z
2 ∧ π
1 + zt3 ∧ αt − πs3 ∧ βs ⎟

⎜ 4 ⎟
⎜ ⎟
⎜ i=1
=1 t=1 s=1 ⎟
⎜ ⎟
:= ⎜ a1 −q11
2
⎟.
⎜ ⎟
⎜ (∂2 (πi ) +
K 2
si ∧ βs )i=1,...,a1
p 1 ⎟
⎜ ⎟
⎝ s=1 ⎠
(∂1K (π
1 ))
=1,...,a2 −q11
a −q11 a −q12
X a −q11
X
K5 ⊕ K3a1 ⊕ K2 2 ⊕ K1 3 ⊕ K0a4 −a K4 ⊕ K2a1 ⊕ K1 2
∂5F : a 2 −q11 a a2 −a1 q11 −q12 +b a a2 −a1 q11 → a −q12 a 2 −q11
⊕K1 1 ⊕ K0 1 ⊕ K0 1 ⊕K0 3 ⊕ K0 1
(3.5)
is given by
⎛ ⎞
∂5K z1 ∧ −z2 ∧ z3 ∧ z4 ∧ −
π 3 ∧ − π 4∧ 0
⎜ 0 (∂3 )
K a 1 0 0 0 p ∧) (
(1 a 1 p ∧)a1 −(z2 ∧)a1 ⎟
2
⎜ (∂2K )a2 −q11
⎟
(z1 ∧)a2 −q11 ⎟ ,
∂5F := ⎜ 0 0 0 0 0 0
⎝0 0 0 (∂1K )a3 −q12 0 0 0 0 ⎠
a 2 −q
0 0 0 0 (∂1K ) 1 11 0 0 0
that is
⎛ ⎞
π5
⎜ ⎟
⎜ (πi3 )i=1,...,a1 ⎟
⎜ ⎟
⎜ (π 2) ⎟
⎜

=1,...,a2 −q11 ⎟
⎜ ⎟
⎜ (π 1 )
t t=1,...,a3 −q12 ⎟
∂5F ⎜
⎜
⎟ :=
⎟
⎜ (αr )r=1,...,a4 −a ⎟
⎜ ⎟
⎜ (π 1 ) ⎟
⎜ s s=1,...,a1 −q11
2 ⎟
⎜ ⎟
⎝(βu )u=1,...,a1 a2 −a1 q11 −q12 +b ⎠
(γ
i )
=1,...,a2 −q11 ; i=1,...,a1
⎛ a1 a2 − a1 q11 ⎞
a1 a2 −q11 a3 −q12 a4 −a a12 −q11 −q12 + b

⎜ K 5
∂5 (π ) + 1 3
zi ∧ π i − 2 2
z
∧ π
+ 3 1
zt ∧ π t + 4
zr ∧ αr − πs ∧ πs1 −
3 πu4 ∧ βu
⎟
⎜ ⎟
⎜ i=1
=1 t=1 r=1 s=1 u=1
⎟
⎜ ⎟
⎜ a12 −q11
a1 a2 − a1 q11
⎟
⎜ 8
−q12 + b

a2 −q11
9 ⎟
⎜ ∂3K (πi3 ) + 1 ∧ π 1 +
si
p 2 ∧β −
ui
p u z
2 ∧ γ
i ⎟
⎜ ⎟
s i=1,...,a1
⎜ s=1 u=1
=1
⎟.
⎜ a1 ⎟
⎜ 8 9
⎟
⎜ ∂2K (π
2 ) + zi1 ∧ γ
i

=1,...,a2 −q11 ⎟
⎜ i=1 ⎟
⎜ 8 9 ⎟
⎜ ∂1K (πt1 ) ⎟
⎜ t=1,...,a3 −q12 ⎟
⎝ ⎠
8 9
∂1K (πs1 )
s=1,...,a12 −q11
Theorem 3.1 Let (R, m, k) be a local ring. The sequence F constructed in 3.1 is a
truncated minimal free resolution of k over R, up to homological degree five.
Proof The minimality follows from Lemma 2.9. We show exactness at each degree
by using the Koszul relations in the complex K, and the basis elements and maps
defined in Sect. 2.
Exactness at Degree One: Im ∂2F ⊆ Ker ∂1F :

π2 a1 !
∂1F ◦ ∂2F = ∂1K ∂2K (π 2 ) + 1
i=1 zi ∧ αi = 0.
(αi )i=1,...,a1
Im ∂2F ⊇ Ker ∂1F : If π 1 ∈ Ker ∂1F , then there exist π 2 ∈ K2 and αi ∈ R such that

a1
π2
π =1
∂2K (π 2 ) + zi1 ∧ αi , therefore, π =1
∂2F .
i=1
(αi )i=1,...,a1
Exactness at Degree Two: Im ∂3F ⊆ Ker ∂2F :

⎛ ⎞
π3 a1 a2 −q11 2
⎜ ⎟ ∂3K (π 3 ) + i=1 zi ∧ πi −
1 1 z
∧ α
∂2F ◦ ∂3F ⎝ (πi1 )i=1,...,a1 ⎠ = ∂2

F
K 1

=1
(∂1 (πi ))i=1,...,a1
(α
)
=1,...,a2 −q11
8
a1 −q11
a2 9
a1
= ∂2K ∂3K (π 3 ) + zi1 ∧ πi1 − z
2 ∧ α
+ zi1 ∧ ∂1K (πi1 )
i=1
=1 i=1

a1
a1
=− zi1 ∧ ∂1K (πi1 ) + zi1 ∧ ∂1K (πi1 ) = 0.
i=1 i=1

π2 a 1
Im ∂3F ⊇ Ker ∂2F : If ∈ Ker ∂2F , then ∂2K (π 2 ) +
i=1 zi ∧ αi = 0
1
(αi )i=1,...,a1
1 1
in K1 . In particular, we have ai=1 [zi ] ∧ [αi ] = 0 in A1 . Since {[zi1 ]}i=1,...,a1 is a
basis of A1 , we get αi ∈ m. It follows that for each i there exists πi1 ∈ K1 such that
αi = ∂1K (πi1 ). (3.6)
Hence we have

a1 8
a1 9
0 = ∂2K (π 2 ) + zi1 ∧ ∂1K (πi1 ) = ∂2K π 2 − zi1 ∧ πi1 .
i=1 i=1
In particular, there exist π 3 ∈ K3 , pi1 ∈ Ker ∂1K , and β

∈ R such that

a1
a1 −q11
a2
π − 2
zi1 ∧ πi1 = ∂3K (π 3 ) + zi1 ∧ pi1 + z
2 ∧ β
, which implies
i=1 i=1
=1

a1 −q11
a2
π 2 = ∂3K (π 3 ) + zi1 ∧ (πi1 + pi1 ) + z
2 ∧ β
. (3.7)
i=1
=1
Combining (3.6) and (3.7) we get

⎛ ⎞
π3
π2 ⎜ ⎟
Ker ∂2F 3 = ∂3F ⎝(πi1 + pi1 )i=1,...,a1 ⎠ .
(αi )i=1,...,a1
−(β
)
=1,...,a2 −q11
Exactness at Degree Three: Im ∂4F ⊆ Ker ∂3F : We show that both components of
the element
⎛ ⎞
π4
⎜ ⎟
⎜ (πi2 )i=1,...,a1 ⎟
⎜ ⎟
⎜ ⎟
∂3F ◦ ∂4F ⎜(π
1 )
=1,...,a2 −q11 ⎟ are zero. The first component is:
⎜ ⎟
⎜ (αt )t=1,...,a −q ⎟
⎝ 3 12 ⎠
(βs )s=1,...,a 2 −q11

1
8 −q11 −q12 a12 −q11 9

a1 a2 a3
∂3K ∂4K (π 4 ) + zi1 ∧ πi2 + z
2 ∧ π
1 + zt3 ∧ αt −
πs3 ∧ βs
i=1
=1 t=1 s=1
8 a12 −q11 9 a2

−q11

a1
a1
+ zi1 ∧ ∂2K (πi2 ) + zi1 ∧ si
p 1
∧ βs − z
2 ∧ ∂1K (π
1 )
i=1 i=1 s=1
=1
−q11 a12 −q11

a1 a2
=− zi1 ∧ ∂2K (πi2 ) + z
2 ∧ ∂1K (π
1 ) − πs3 ) ∧ βs
∂3K (
i=1
=1 s=1
a12 −q11

a1 8
a1 9 −q11
a2
+ zi1 ∧ ∂2K (πi2 ) + zi1 ∧p
si
1
∧ βs − z
2 ∧ ∂1K (π
1 ) = 0.
i=1 s=1 i=1
=1
The last equality follows from the definition of si

πs3 in (2.6). As p 1 ∈ Ker ∂ K for all
8 a1 −q11 1
2 9 1
i and s, we have ∂1 ∂2 (πi ) + s=1 p
K K 2 si ∧ βs = 0, for each 1 i a1 , so the
second component is zero.
⎛ ⎞
π3
⎜ ⎟
Im ∂4F ⊇ Ker ∂3F : If ⎝ (πi1 )i=1,...,a1 ⎠ ∈ Ker ∂3F , then
(α
)
=1,...,a2 −q11

a1 −q11
a2
∂3K (π 3 ) + zi1 ∧ πi1 − z
2 ∧ α
= 0 in K2 , and (3.8)
i=1
=1
∂1K (πi1 ) = 0 for all 1 i a1 . (3.9)
It follows that for each i, there exist πi2 ∈ K2 and βij ∈ R such that

a1
πi1 = ∂2K (πi2 ) + zj1 ∧ βij . (3.10)
j =1
Thus (3.8) becomes:

8
a1 9
a1 8
a1 9 −q11
a2
∂3K π −
3
zi1 ∧ πi2 + zi1 ∧ zj1 ∧ βij − z
2 ∧ α
= 0. (3.11)
i=1 i=1 j =1
=1
a 1 a2 −q11 2
i,j =1 [zi ]
∧ [zj1 ] ∧ [βij ] −
=1 [z
] ∧ [α
] = 0. By Propo-
In A2 we obtain 1
a 1 a12 −q11 1
sition 2.2, we get [α
] = 0 for all
and j =1 [zj1 ] ∧ [βij ] = s=1 psi ] ∧ [βs ]
[
for some βs in R. In particular, there exist π
1 ∈ K1 and πi2 ∈ K2 such that
α
= ∂1K (π
), and
1
(3.12)
a12 −q11

a1
zj1 ∧ βij = ∂2K (πi2 ) + si
p 1
∧ βs .
j =1 s=1
Equation (3.10) becomes:
a12 −q11

πi1 = ∂2K (πi2 + πi2 ) + si
p 1
∧ βs , (3.13)
s=1
and Eq. (3.11) now becomes:
8 9 −q11
a1 a1
2

a1
a1
0= ∂3K π −
3
zi1 ∧ πi2 + zi1 ∧ ∂2K (πi2 ) + zi1 ∧ p
si
1
∧ βs
i=1 i=1 i=1 s=1
−q11
a2
z
2 ∧ ∂1K (π
)
1
−

=1
8 −q11 a12 −q11 9

a1 a2
zi1 ∧ (πi2 + πi2 ) − z
2 ∧ π
πs3 ∧ βs ,
1
= ∂3K π 3 − +
i=1
=1 s=1
where the second equality uses the definition of

πs3 from (2.6). Hence, there exist
4 K 1 K
π ∈ K4 , pi ∈ Ker ∂2 , p
∈ Ker ∂1 , and αt ∈ R such that
2
−q11 a12 −q11

a1 a2
+ πi2 ) − ∧ π
πs3 ∧ βs
1
π − 3
zi1 ∧ (πi2 z
2 +
i=1
=1 s=1

a1 −q11
a2 −q12
a3
= ∂4K (π 4 ) + zi1 ∧ pi2 + z
2 ∧ p
1 + zt3 ∧ αt .
i=1
=1 t=1
Thus,

a1 −q11
a2
zi1 ∧ (πi2 + πi2 + pi2 ) + z
2 ∧ (π
+ p
1 )
1
π 3 = ∂4K (π 4 ) +
i=1
=1
−q12
a3 a12 −q11

+ zt3 ∧ αt − πs3 ∧ βs .
(3.14)
t=1 s=1
Equations (3.12), (3.13), and (3.14), now yield

⎛ ⎞
π4
⎛ ⎜ 2 ⎞ ⎟
π3 ⎜(πi + πi2 + pi2 )i=1,...,a1 ⎟
⎜ ⎟
⎜ ⎟ ⎜ ⎟
Ker ∂3F 3 ⎝ (πi1 )i=1,...,a1 ⎠ = ∂4F ⎜ (π
+ p
1 )
=1,...,a2 −q11 ⎟ .
1
⎜ ⎟
(α
)
=1,...,a2 −q11 ⎜ (αt )t=1,··· ,a3 −q12 ⎟
⎝ ⎠
(βs )s=1,...,a 2 −q11
1
Exactness at Degree Four: Im ∂5F ⊆ Ker ∂4F : We show that all three components
of the element
⎛ ⎞
π5
⎜ ⎟
⎜ (πi3 )i=1,...,a1 ⎟
⎜ ⎟
⎜ 2
(π
)
=1,...,a2 −q11 ⎟
⎜ ⎟
⎜ ⎟
⎜ 1
(πt )t=1,...,a3 −q12 ⎟
∂4F ◦ ∂5F ⎜
⎜
⎟ are zero. The first component is:
⎟
⎜ (αr )r=1,...,a4 −a ⎟
⎜ ⎟
⎜ (π 1 ) ⎟
⎜ s s=1,...,a12 −q11 ⎟
⎜ ⎟
⎝(βu )u=1,...,a1 a2 −a1 q11 −q12 +b ⎠
(γ
i )
=1....,a2 −q11 ; i=1,...,a1
8
a1 −q11
a2 −q12
a3 4 −a
a
∂4K ∂5K (π 5 ) + zi1 ∧ πi3 − z
2 ∧ π
2 + zt3 ∧ πt1 + zr4 ∧ αr
i=1
=1 t=1 r=1
a1 a2 − a1 q11
a12 −q11 −q12 + b 9

− πs3 ∧ πs1 −

πu4 ∧ βu
s=1 u=1
a1 a2 − a1 q11
8 a1 −q11 2
−q12 + b −q11 9

a1 a2
+ zi ∧ ∂3 (πi ) +
1 K 3
si
p 1
∧ πs1 + ui
p 2
∧ βu − z
2 ∧ γ
i
i=1 s=1 u=1
=1
−q11
a2 8 9 a3
−q12 a1 −q11 2

a1
+ z
2 ∧ ∂2K (π
2 ) + zi1 ∧ γi
+ zt3 ∧ ∂1K (πt1 ) − πs3 ∧ ∂1K (πs )

=1 i=1 t=1 s=1

a1 −q11
a2 −q12
a3
=− zi1 ∧ ∂3K (πi3 ) − z
2 ∧ ∂2K (π
2 ) − zt3 ∧ ∂1K (πt1 )
i=1
=1 t=1
a1 a2 − a1 q11
a12 −q11 a12 −q11 −q12 + b

− πs3 ) ∧ πs1 +
∂3K ( πs3 ∧ ∂1K (πs1 ) −
πu4 ) ∧ βu
∂4K (
s=1 s=1 u=1
a1 a2 − a1 q11
a12 −q11 8 a1 9 −q12 + b 8 9

a1
a1
+ zi1 ∧ ∂3K (πi3 ) + zi1 ∧ p
si
1
∧ πs1 + zi1 ∧ p
ui
2
∧ βu
i=1 s=1 i=1 u=1 i=1
−q11
a2 a1 −q11
a2 −q11
a2 a1
− zi1 ∧ z
2 ∧ γ
i + z
2 ∧ ∂2K (π
2 ) + z
2 ∧ zi1 ∧ γ
i

=1 i=1
=1
=1 i=1
−q12
a3 a12 −q11

+ zt3 ∧ ∂1K (πt1 ) − πs3 ∧ ∂1K (πs1 ) = 0,

t=1 s=1
by definitions of si
πs3 , p 1, ui
πu4 , and p 2 . For each 1 i a , as p
1 si
1 is in Ker ∂ K and
1
ui is in Ker ∂2 , we have
p 2 K
a1 a2 − a1 q11
8 a1 −q11 2
−q12 + b −q11 9
a2
∂2K ∂3K (πi3 ) + si
p 1
∧ πs1 + ui
p 2
∧ βu − z
2 ∧ γ
i
s=1 u=1
=1
a12 −q11

+ si
p 1
∧ ∂1K (πs1 )
s=1
a12 −q11 a12 −q11

=− si
p 1
∧ ∂1K (πs1 ) + si
p 1
∧ ∂1K (πs1 ) = 0.
s=1 s=1
Therefore, the second component is zero. For each 1

a2 − q11 ,

a1
!
∂1K ∂2K (π
2 ) + zi1 ∧ γ
i = 0.
i=1
Thus, the third component is zero.

⎛ ⎞
π4
⎜ ⎟
⎜ (πi2 )i=1,...,a1 ⎟
⎜ ⎟
⎜ ⎟
Im ∂5F ⊇ Ker ∂4F : If ⎜(π
1 )
=1,...,a2 −q11 ⎟ ∈ Ker ∂4F , then
⎜ ⎟
⎜ (αt )t=1,...,a −q ⎟
⎝ 3 12 ⎠
(βs )s=1,...,a 2 −q11

1
−q11 −q12 a12 −q11

a1 a2 a3
∂4K (π 4 ) + zi1 ∧ πi2 + z
2 ∧ π
1 + zt3 ∧ αt −
πs3 ∧ βs = 0,
i=1
=1 t=1 s=1
(3.15)
a12 −q11

∂2K (πi2 ) + si
p 1
∧ βs = 0 for all 1 i a1 , and (3.16)
s=1
∂1K (π
1 ) =0 for all 1
a2 − q11 . (3.17)
The equality (3.17) implies that there exist π

2 ∈ K2 and γi
∈ R such that

a1
π
1 = ∂2K (π
2 ) + zi1 ∧ γi
. (3.18)
i=1
a12 −q11 1
In Aa11 the equality (3.16) becomes s=1 ps1 , . . . , [
([ 1 ])∧[β ] = 0. Applying
psa 1 s
Lemma 2.1 we obtain [βs ] = 0 for all s, thus
βs = ∂1K (πs1 ) for some πs1 ∈ K1 . (3.19)
The equality (3.16) now becomes
a12 −q11 8 a1 −q11 2

9

∂2K (πi2 ) + si
p 1
∧ ∂1K (πs1 ) = ∂2K πi −
2
si
p 1
∧ πs1 = 0.
s=1 s=1
Therefore, there exist πi3 in K3 , δij k and γi

in R such that for all i we have:
−q12 a12 −q11

a1 a2
πi2 = ∂3K (πi3 ) + zj1 ∧ zk1 ∧ δij k + z
2 ∧ γi
+ si
p 1
∧ πs1 . (3.20)
j,k=1
=1 s=1
Putting together (3.18), (3.19), and (3.20), into the equality (3.15) we get:
0 = ∂4K (π 4 )
2 −q11
8 a1 −q12 9

a1
a1 a2
+ zi1 ∧ si
p 1
∧ πs1 + ∂3K (πi3 ) + zj1 ∧ zk1 ∧ δij k + z
2 ∧ γi

i=1 s=1 j,k=1
=1
−q11
a2 8 9 a3
−q12 a12 −q11

z
2 ∧ ∂2K (π
) + πs3 ∧ ∂1K (πs1 )
2
+ zi1 ∧ γi
+ zt3 ∧ αt −

=1 i=1 t=1 s=1
8 a12 −q11 −q11 9

a1 a2
πs3 ∧ πs1 − z
2 ∧ π

2
= ∂4K π 4 + zi1 ∧ πi3 +
s=1 i=1
=1
8
a1 −q11
a1 a2 −q12
a3 9
+ zi1 ∧ zj1 ∧ zk1 ∧ δij k + zi1 ∧ z
2 ∧ (γi
+ γi
) + zt3 ∧ αt .
i,j,k=1 i=1
=1 t=1
In A3 this reduces to:

a1 −q11
a1 a2 −q12
a3
[zi1 ]∧[zj1 ]∧[zk1 ]∧[δij k ]+ [zi1 ]∧[z
2 ]∧[γi
+γi
]+ [zt3 ]∧[αt ] = 0.
i,j,k=1 i=1
=1 t=1
By Proposition 2.6, for each t there exists πt1 ∈ K1 , and for each i there exist
elements δu , εj s in R, and πi3 ∈ K3 such that
αt = ∂1K (πt )
1
(3.21)
and

a1 −q11
a2
zj1 ∧ zk1 ∧ δij k + z
2 ∧ (γi
+ γi
)
j,k=1
=1
a1 a2 − a1 q11
−q12 + b −q11
a1 a1
2

∂3K (πi ) + ∧ δu +
3
= ui
p 2
si
p 1
∧ zj1 ∧ εj s
u=1 j =1 s=1
Thus, (3.15) further becomes:
8 a12 −q11 −q11 9

a1 a2
πs3 ∧ πs1 − z
2 ∧ π

2
0= ∂4K π +
4
zi1 ∧ πi3 +
s=1 i=1
=1
a1 a2 − a1 q11
8 −q12 + b −q11
a1 a1
2
9

a1
∧ ∂3K (πi ) + ∧ δu
3
+ zi1 ui
p 2
+ si
p 1
∧ zj1 ∧ εj s
i=1 u=1 j =1 s=1
−q12
a3
zt3 ∧ ∂1K (πt )
1
+
t=1
8 a12 −q11 −q11

a1 a2
a1
πs3 ∧ πs1 − z
2 ∧ π
2 − zi1 ∧ πi
3
= ∂4K π +
4
zi1 ∧ πi3 +
s=1 i=1
=1 i=1
a1 a2 − a1 q11
−q12 + b −q11
a1 a1
2
−q12 9
a3
∧ δu zt3 ∧ πt
1
+
πu4 +
πs3 ∧ zj1 ∧ εj s − .
u=1 j =1 s=1 t=1
The second equality above follows from definitions of

πs3 and
πu4 in (2.6) and
(2.11). Therefore, there exist π ∈ K5 , δr ∈ R, pi ∈ Ker ∂3 , p
∈ Ker ∂2K and
5 3 K 2
ps1 ∈ Ker ∂1K with ps1 as in Proposition 2.8, such that

a1 −q11
a2 −q12
a3
zi1 ∧ (πi3 + πi ) + z
2 ∧ π
− zt3 ∧ πt
3 2 1
π4 −
i=1
=1 t=1
a1 a2 − a1 q11
a12 −q11 8 9 −q12 + b

a1
+ πs3 ∧ πs1 +
zj1 ∧ εj s + πu4 ∧ δu

s=1 j =1 u=1
−q11 a12 −q11 4 −a

a1 a2 a
= ∂5K (π 5 ) + zi1 ∧ pi3 + z
2 ∧ p
2 +
πs3 ∧ ps1 + zr4 ∧ δr .
i=1
=1 s=1 r=1
This implies:

a1 −q11
a2
zi1 ∧ (πi3 + πi + pi3 ) − z
2 ∧ (π
− p
2 )
3 2
π 4 = ∂5K (π 5 ) +
i=1
=1
−q12
a3 4 −a
a a12 −q11 8 9

a1
∧ πt πs3 ∧ πs1 +
1
+ zt3 + zr4 ∧ δr − zj1 ∧ εj s − ps1
t=1 r=1 s=1 j =1
a1 a2 − a1 q11
−q12 + b

− πu4 ∧ δu .

u=1
(3.22)
By (3.21), the expression (3.20) becomes:
−q12 a12 −q11

a1 a2
πi2 = ∂3K (πi3 ) + zj1 ∧ zk1 ∧ δij k + z
2 ∧ γi
+ si
p 1
∧ πs1
j,k=1
=1 s=1
(3.23)
a1 a2 − a1 q11
a12 −q11 8 9 −q12 + b

a1
= ∂3K (πi3 + πi ) + ∧ πs1 + ∧ δu
3
si
p 1
zj1 ∧ εj s + ui
p 2
s=1 j =1 u=1
−q11
a2
− z
2 ∧ γi
.

=1
We conclude that by using (3.18), (3.19), (3.21), (3.22), (3.23), and Proposition 2.8,
the chosen kernel element is in the image of ∂5F :
⎛ ⎞
π5
⎜ ⎟
⎛ ⎜ ⎞ (πi3 + πi 3 + pi3 )i=1,...,a1 ⎟
π4 ⎜ ⎟
⎜ 2
(π
− p
)
=1,...,a2 −q11
2 ⎟
⎜ ⎟ ⎜ ⎟
⎜ (πi2 )i=1,...,a1 ⎟ ⎜ ⎟
⎜ ⎟ ⎜ 1
(πt )t=1,...,a3 −q12 ⎟
⎜(π 1 ) ⎟ F ⎜ ⎟
Ker ∂4 3 ⎜

=1,...,a2 −q11 ⎟ = ∂5 ⎜
F
⎟.
⎜ ⎟ ⎜ (δr )r=1,...,a4 −a ⎟
⎜ (αt )t=1,...,a −q ⎟ ⎜8 9 ⎟
⎝ 3 12 ⎠ ⎜ 1 a1 1 ⎟
⎜ πs + j =1 zj ∧ εj s − ps 1
⎟
(βs )s=1,...,a 2 −q11 ⎜ s=1,...,a12 −q11 ⎟
⎜ ⎟
(δu )u=1,...,a1 a2 −a1 q11 −q12 +b
1
⎝ ⎠
(γ
i )
=1,...,a2 −q11 ; i=1,...,a1
Therefore, the complex F in Construction 3.1 is a minimal free resolution of k

up to degree five.

Remark 3.2 If all multiplications on the algebra A are trivial, that is, qij = 0 for
all i, j 1, and all Massey operations are zero, Golod’s construction in [8] gives a
minimal free resolution of the residue field k of a local ring R in terms of the Koszul
complex K of R. Without these assumptions, our complex F in Construction 3.1
generalizes Golod’s resolution up to degree five.
4 Applications of the Construction
For a local ring R of embedding dimension n and codepthc, set βi :=

∞
rankk TorRi (k, k) for i 0 for the Betti numbers of k over R. Let
i
i=0 bi t denote
the series on the right hand side of the inequality (1). The sequence P := {Pi }i0
defined by:
Pi := bi − βi , for all i 0, (4.1)
gives the coefficients of the series P(t) defined in the Introduction. The ring
R is Golod if and only if Pi = 0 for all i 0. Our goal in this section
is to give a description of the sequence {Pi }0i5 in terms of the invariants
of the multiplicative structure of the algebra A = H(K R ) and discuss various
consequences of Theorem 3.1. First, using Theorem 3.1, we explicitly describe the
Betti numbers βi in terms of those invariants, up to degree five.
Corollary 4.1 Let R be a local ring of embedding dimension n. Let ai , qij , a and
b be as in Summary 2.5. Then the following equalities hold:
n!
β0 = 1, β1 = n, β2 = 2 + a1 ,
n! n! n!
β3 = 3 + na1 + a2 − q11 , β4 = 4 + 2 a1 + na2 + a3 + a1 − (n + 1)q11 − q12 ,
2
n! n! n! n+1! !
β5 = 5 + 3 a1 + 2 a2 +na3 +a4 +na1 +2a1 a2 −
2
2 +2a1 q11 −(n+1)q12 +b−a.
Proof This is a direct consequence of Theorem 3.1. The formulas for β0 , . . . , β3

are clear, and we simplify the following expressions for β4 and β5 :
n! n!
β4 = 4 + 2 a1 + a1 − q11 + n(a2 − q11 ) + a3 − q12
2
n! n!
β5 = 5 + 3 a1 + n(a1 − q11 ) + a1 a2 − a1 q11 − q12 + b
2
!
+ n2 (a2 − q11 ) + a1 a2 − a1 q11 + n(a3 − q12 ) + a4 − a.
We now obtain the desired expressions from the statement.

Proposition 4.2 Let R be a local ring of embedding dimension n and codepth c.
Let ai , qij , a, and b be as in Summary 2.5 and P as in (4.1). Then, the following
hold:
P0 = P1 = P2 = 0, P3 = q11 , P4 = (n + 1)q11 + q12 , and

n+1! !
P5 = 2 + 2a1 q11 + (n + 1)q12 + a − b.
∞ i =
Proof Recall that the right hand side of the inequality (1) is i=1 bi t
(1+t) n
. Comparing the coefficients on both sides of (1) we have the following
1− ci=1 ai t i+1
recursive formulas for bi :

i−1
n!
b0 = 1, b1 = n, bi = aj bi−j −1 + i , for all i 2.
j =1
In particular, we obtain:
n! n! n!
b2 = 2 + a1 , b4 = 4 + 2 a1 + na2 + a3 + a12 ,
n! n! n! !
b3 = 3 + na1 + a2 , b5 = 5 + 3 a1 + n2 a2 + na3 + a4 + na12 + 2a1 a2 .
Now, the expression for Pi = bi − βi for 0 i 5 follows from Corollary 4.1.

Alternatively, assume qij = 0 for all 1 i +j 4, and Massey products are also
trivial, we have a = b = 0. In Construction 3.1 we obtain the maximum possible
values of the Betti numbers, so the differences between the actual Betti numbers and
these maximum values are exactly the Pi ’s in the proposition.

Corollary 4.3 Let R be a local ring, ai , qij , a, and b be as in Summary 2.5 and P
as in (4.1).
(a) If q11 = q12 = 0, then Pi = 0 for all 0 i 4 and P5 = a. !
(b) If q11 = q12 = q13 = q22 = 0, then P5 = rankk Spank A1 , A1 , A1 .
(c) If codepth R = 4, then R is Golod if and only if Pi = 0 for all 1 i 5.
Proof (a) and (b) are straightforward from Proposition 4.2. For (c), result in [8]
showed that R is Golod if and only if all products on A and all ternary Massey
products are trivial. If codepth R = 4, then the r-ary Massey products are trivial for
all r 4. Thus, part (c) follows from Proposition 4.2.

Remark 4.4 A result of Burke [6, Corollary 6.10] implies that for a local ring R of
codepth c the following implication holds:
If Pi = 0 for all 0 i c + 1, then Pi = 0 for all i 0.
Thus, Corollary 4.3(c) is a consequence of this result as well.

We now examine some local rings with rational Poincaré series of certain forms
and describe PR
k (t) in terms of the algebraic invariants of A.
Proposition 4.5 Let (R, m, k) be a local ring of embedding dimension n and
codepth c. Let ai , qij , a, and b be as in Summary 2.5. If the Poincaré series of
(1+t)n
R is rational of the form PR k (t) = d(t) , then
8
c 9
d(t) = 1 − ai t i+1 + q11 t 3 + (q11 + q12 )t 4 + (q12 − b + a)t 5 + f (t)t 6 ,
i=1
for some f (t) ∈ Z[t].

Proof Let Pi be as in (4.1). Set
∞

c
P(t) = Pi t i , α(t) = 1 − ai t i+1 and γ (t) = d(t) − α(t).
i=0 i=1
Then
(1 + t)n (1 + t)n (1 + t)n · γ (t)

P(t) = − = ⇐⇒
α(t) d(t) α(t) · (α(t) + γ (t))
γ (t) P(t) · α(t)
= ⇐⇒
α(t) + γ (t) (1 + t)n
γ (t) P(t) · α(t)
= ⇐⇒
α(t) (1 + t)n − P(t) · α(t)
P(t) · (α(t))2
γ (t) = .
(1 + t)n − P(t) · α(t)
We compare the coefficients of t i for all 0 i 5 on both sides of the equality:

8 9
γ (t) · (1 + t)n − P(t) · α(t) = P(t) · (α(t))2 .
By Proposition 4.2, P0 = P1 = P2 = 0, and thus the left and right hand sides of
the above equation become:
LHS = (γ0 + γ1 t + γ2 t 2 + γ3 t 3 + γ4 t 4 + γ5 t 5 + · · · )
! ! ! ! ! n! ! 5 !
· 1+ nt+ n2 t 2 + n3 − P3 t 3 + n4 − P4 t 4 + 5 + P3 a1 − P5 t + · · · ,
RHS = (P3 t 3 + P4 t 4 + P5 t 5 + . . . )
!
· 1 − 2a1 t 2 − 2a2 t 3 + (a12 − 2a4 )t 4 + (a1 a2 − 2a5 )t 5 + · · · .
It is clear that γ0 = γ1 = γ2 = 0 and comparing the coefficients of t 3 we get:

γ3 = P3 = q11 . Comparing the coefficients of t 4 and by Proposition 4.2 we get:
γ4 + nγ3 = P4 ⇐⇒ γ4 = P4 − nγ3 = (n + 1)q11 + q12 − nq11 = q11 + q12 .
Finally, comparing the coefficients of t 5 and by Proposition 4.2 we get:

n!
γ5 + nγ4 + = P5 − 2a1 P3 ⇐⇒
2 γ3
!
γ5 = P5 − 2a1 P3 − nγ4 − n2 γ3
! ! n!
= n+12 + 2a1 q11 + (n + 1)q12 − b + a − 2a1 q11 − n(q11 + q12 ) − 2 q11
= q12 − b + a.
Therefore, the expression for d(t) = α(t) + γ (t) in the statement holds.

There are many classes of local rings for which the Poincaré series is rational of
k (t) = (1+t) /d(t). In light of Proposition 4.5, we write the coefficients
the form PR n
of the polynomial d(t) in terms of the invariants ai , qij , b and a for some special
cases and provide a uniform expression of the Poincaré series in these cases.
Corollary 4.6 If (R, m, k) is a local ring of embedding dimension n and codepth

at most 3, and ai , qij , b are as in Summary 2.5, then
(1 + t)n−1
k (t) =
PR .
1 − t − (a1 − 1)t 2 − (a3 − q11 )t 3 + q12 t 4 − bt 5
Proof By [3, Theorem 3.5] the Poincaré series of the ring R is rational given by
k (t) = (1 + t) /d(t) with deg d(t) 6 and d(−1) = 0. Since codepth R 3 we
PR n
have a4 = 0 and a = 0. Thus, Proposition 4.5 gives:
(1 + t)n
PkR (t) = .
1 − a1 t2 − (a2 − q11 )t 3 − (a3 − q11 − q12 )t 4 − (b − q12 )t 5 − bt 6
Simplifying the fraction by the common factor (1+t), we get the desired conclusion.

Remark 4.7 In [5, Lemma 3.6] Avramov defined the invariant τ for non-Gorenstein
ring R of codepth 3 as follows: τ = 1 if R is of class T, and τ = 0 otherwise.
By comparing [5, (3.6.2)] and the Poincaré expression in Corollary 4.6 we see that
τ is our b = rankk (Coker ψ) in (2.9), which is a multiplicative invariant of the
homology algebra A. It follows from [3, Theorem 3.5] that b = 1 for a ring of class
C(3) or T, and b = 0 for rings of class C(1), C(2), S, B, G(r), and H(q11 , q12 ).
Corollary 4.8 If (R, m, k) is a Gorenstein local ring, not complete intersection, of
embedding dimension n and codepth 4, and ai , qij , a, b are as in Summary 2.5, then
(1 + t)n−2
k (t) =
PR .
1 − 2t − (a1 − 3)t 2 + (q11 − 2)t 3 + (−q11 + q12 − 1)t 4 − (q11 − q12 − b + a − 1)t 5
Proof Since R is Gorenstein we have a2 = 2a1 − 2, a3 = a1 , and a4 = 1. The

statement follows from [3, Theorem 3.5] and Proposition 4.5.

Remark 4.9 Using the Avramov’s notation from [3] for the classification of Goren-
stein local rings R of codepth 4 given by Kustin and Miller [11] and comparing
the Poincaré expressions from [3, Theorem 3.5] and Corollary 4.8, we obtain the
following table of algebra invariants of a Gorenstein local ring R:
Class q11 q12 q22 q13 a b

C(4) 6 4 1 1 1 4
GT 3 3 1 1 1 1
GS 0 0 1 1 1 0
GH(p) p p+1 1 1 1 0
We provide more examples of rings for which we can calculate the multiplicative
invariants qij , a, b from the Poincaré series of the ring.
Corollary 4.10 Let k be a field, I be an ideal of k[x, y, z, w] such that

(x, y, z, w)3 ⊆ I ⊆ (x, y, z, w)2 , and set R = k[x, y, z, w]/I . Let ai , qij , a, b
be as in Summary 2.5. If R has a4 = 3 and its Poincaré series is of the form
1
k (t) =
PR , then
(1 − t)(1 − 3t)
q11 = a2 − 8, q12 = 8, a = 3, and b = 0.
Proof By hypothesis,
(1 + t)4 (1 + t)4
k (t) =
PR = .
(1 − t)(1 − 3t)(1 + t)4 1 − 7t 2 − 8t 3 + 3t 4 + 8t 5 + 3t 6
Proposition 4.5 thus gives:
a1 = 7, a2 − q11 = 8, a3 − q11 − q12 = −3, a4 − q12 + b − a = −8.

and
(4.2)
From the second equality we obtain q11 = a2 − 8 and from the third we get q12 =
a3 − a2 + 11. Since a4 = 3 and a1 = 7, we get a2 = a3 + 3 and thus q12 = 8.
The last equality in (4.2) becomes a − b = 3. By the definitions of a and b we have
0 a a4 = 3 and b 0, thus a = 3 and b = 0.

Example 4.11 Rings satisfying the hypotheses of Corollary 4.10 are discussed by
Christensen and Veliche in [7] and Yoshino in [17]. For examples, consider the
following ideals in Q = Q[x, y, z, w]:
I1 = (yw, xw + zw + w 2 , z2 + w 2 , xz + zw + w 2 , y 2 + yz, xy + zw, x 2 + zw),
I2 = (zw + w 2 , yw, z2 + w 2 , yz + xw + w 2 , xz + w 2 , xy + y 2 + xw + w 2 , x 2 + xw + w 2 ),
I3 = (zw, yw, xw − w 2 , yz, xz, xy − z2 , x 2 − y 2 ),
I4 = (w 2 , yw + zw, xw, yz + z2 , y 2 + zw, xy + xz, x 2 + zw).
The rings Q/Ii satisfy the hypotheses of Corollary 4.10 and have a2 = 10 + i, for
1 i 4.
Example 4.12 In an unpublished note, Roos [16], inspired by a paper of Katthän
[10], constructed several examples of non-Golod rings R of codepth 4 with trivial
algebra multiplications on A. We provide here two of them. For each one, there
exists a Golod homomorphism from a complete intersection ring, hence it has
rational Poincaré series. The algebra multiplication on A was checked using the
DGAlgebras package [15] of Macaulay2 [13]. Proposition 4.5 confirms that
indeed q11 = q12 = 0, and moreover it gives us the exact size of the space generated
by the triple Massey products A1 , A1 , A1 , known to be nonzero.
Consider the following ideals in Q = Q[x, y, z, w]:
J1 = (w 3 , xy 2 , xz2 + yz2 , x 2 w, x 2 y + y 2 w, y 2 z + z2 w),

J2 = (w 3 , xy 2 , xz2 + yz2 , x 2 w + zw 2 , y 2 w + xzw, y 2 z + yz2 ).
The rings Q/Ji with i = 1, 2 are non-Artinian of codepth 4 with
PQ
Q/Ji (t) = 1 + 6t + (10 + i)t + (7 + i)t + 2t ,
2 3 4
Q/Ji (1 + t)4
PQ (t) = .
1 − 6t 2 − (10 + i)t 3 − (7 + i)t 4 − t 5 + t 6
For both rings Q/Ji , q11 = 0 = q12 by Proposition 4.5, hence b = 0 and a = 1.
Since q22 = q13 = 0, the space spanned by the Massey products A1 , A1 , A1 has
rank one.
By [1, Example 7.1], not all local rings have rational Poincaré series. However,
for every local ring R with residue field k, there exists a unique sequence of integers
{εi }i0 such that the Poincaré series of R can be expressed as
3∞
i=1 (1 + t
2i−1 )ε2i−1
PkR (t) = 3 ∞ ,
i=1 (1 − t )
2i ε2i
and εi is called the i-th deviation of R; see e.g., [4, Remark 7.1.1]. Theorem 3.1
allows us to describe the first five deviations in terms of the algebraic invariants of A.
Corollary 4.13 Let R be a local ring of embedding dimension n and ai , qij be as
in Summary 2.5. Then the first five deviations of R are:
ε1 = n, ε2 = a1 , ε3 = a2 − q11 ,
!
ε4 = a3 − q12 + a21 − q11 ,
ε5 = a4 + a1 a2 − a1 q11 − q12 + b − a.
Proof By comparing the coefficients on the left and right sides of the equality
∞
∞

(1 + t 2i−1 )ε2i−1 = (1 − t 2i )ε2i · (β0 + β1 t + β2 t 2 + β3 t 3 + β4 t 4 + β5 t 5 + · · · )
i=1 i=1
one obtains the following relations between the Betti numbers {βi }1i5 and the
deviations {εi }1i5 :
! ε1 !
β1 = ε1 , β2 = ε2 + ε21 , β3 = ε3 + ε2 ε1 + 3 ,
! ! ε1 !
β4 = ε4 + ε3 ε1 + 1+ε
2
2
+ ε2 ε1
2 + 4 ,
! ! ε1 ! ε1 !
β5 = ε5 + ε4 ε1 + ε3 ε2 + ε3 ε21 + ε22 ε1 − ε1 ε22 + ε2 3 + 5 .
The first four relations were also given by Avramov [4, page 62] and Gulliksen
and Levin [9, Proposition 3.3.4, Theorem 4.4.3]. Note that we have corrected the
expression for β4 , compared to that given in [4, page 62]. The expressions for εi
described in the statement follow from these relations and Corollary 4.1.

Remark 4.14 The formulas for ε2 , ε3 , ε4 in Corollary 4.13 were previously given
in [2, Corollary 6.2] and [9, Proposition 3.3.4]. The expression for ε5 obtained by
Avramov in [2, Corollary 6.2] is, in our notations:
ε5 = a4 + a1 a2 + a1 q11 − q12 − a13 + b − a,
where b := rankk (Ker ) with
:A1 ⊗ A1 ⊗ A1 −
→ A2 ⊗ A1 ⊕ A1 ⊗ A2 , defined by
([x], [y], [z]) = ([x] ∧ [y], [y] ∧ [z]), for all [x], [y], [z] ∈ A1 .
The map ψ in (2.8), that defines our invariant b = rankk (Coker ψ), differs from
just by its codomain. Thus, one can relate b in [2] and our b by:
b = rankk (Ker ψ) = a13 − 2a1 q11 + b.
Note that in both references [2] and [9] there is a shift in the indexing of the
deviations, their εi is our εi+1 .
5 An Example Illustrating the Construction
In this section, we consider the Artinian local ring of codepth 4 from [2, Section 7]:
R = Q[x, y, z, w]/(x 3 , y 3 , z3 − xy 2 , x 2 z2 , xyz2 , y 2 w, w 2 ).
This ring is of a particular interest to us, since its Koszul homology algebra A =
H(K R ) has nontrivial multiplication and a nontrivial ternary Massey product that
does not come from this multiplication. The free modules {Fi }i=0,...,5 are given
in terms of the Koszul algebra components {Ki }i=0,...,5 , the ranks ai of Ai , and
the multiplicative invariants qij , a and b. The differential maps of the complex F
si
in Construction 3.1 are given in terms of elements zi1 , z
2 , zt3 , zr4 , p 1 , ui
πs3 , p 2 , and
4
πu , see Summary 2.5. We explicitly describe them all for this ring. The bases of
Ai = Hi (K) are computed with Macaulay2 [13]. We use our results from Sect. 2
to obtain the other elements needed in Construction 3.1.
5.1 The ring R is graded Artinian with Ri = 0 for all i 6. The other graded
components have the following basis elements:
R0 : 1
R1 : x, y, z, w
R2 : x 2 , xy, xz, xw, y 2 , yz, yw, z2 , zw
R3 : x 2 y, x 2 z, x 2 w, xyz, xyw, xz2 , xzw, y 2 z, yz2 , yzw, z3 , z2 w
R4 : x 2 yz, x 2 yw, x 2 zw, xyzw, xz3 , xz2 w, y 2 z2 , yz2 w, z4
R5 : x 2 yzw, xz4 .
<
5.2 The Koszul algebra of R has the form K = (R 4 ) ∼ = R⊕R 4 ⊕R 6 ⊕R 4 ⊕R. Let
{T1 , T2 , T3 , T4 } be the standard ordered vector basis of the free module K1 = R 4 .
Set Tij = Ti ∧ Tj and consider the ordered basis {T12 , T13 , T23 , T14 , T24 , T34 } for
<
the free module K2 = 2 (R 4 ) ∼ = R 6 . Set Tij k = Ti ∧ Tj ∧ Tk and consider the
<
ordered basis {T123 , T124 , T134 , T234 } for the free module K3 = 3 (R 4 ) ∼ = R4.
< = T1 ∧ T2 ∧ T3 ∧ T4 and consider the basis {T1234 } for the free module
Set T1234
K4 = 4 (R 4 ) ∼ = R. By Macaulay2, we obtain bases of Ai = Hi (K) and ranks
a1 = 7, a2 = 15, a3 = 14, a4 = 5.
5.3 A basis of A1 is {[zi1 ]}i=1,...,7 where:
z11 = wT4 , z21 = x 2 T1 , z31 = ywT2 , z41 = y 2 T2 ,

z51 = y 2 T1 − z2 T3 , z61 = yz2 T1 , z71 = xz2 T1 .
5.4 Using Koszul and ring relations, we obtain that the space A1 · A1 has rank
q11 = 7 and its basis is given by the classes of
z11 ∧ z21 = −x 2 wT14 , z11 ∧ z51 = z2 wT34 , z11 ∧ z61 = −yz2 wT14 ,
z11 ∧ z71 = −xz2 wT14 , z21 ∧ z31 = x 2 ywT12 , z31 ∧ z51 = −yz2 wT23 ,
z41 ∧ z51 = −y 2 z2 T23 .
Therefore, a basis of A2 is {[z

2 ]}
=1,...,8 , where
z12 = ywT24 , z22 = y 2 T24 , z32 = xz2 T12 , z42 = yz2 T13 ,
z52 = x 2 yT12 − xz2 T13 , z62 = x 2 zT13 , z72 = yz2 wT12 , z82 = z4 T23 .
5.5 All other products among the basis elements of A1 are zero in A2 , except for
the Koszul relations of the nonzero products A1 · A1 above. It follows that the kernel
of the multiplication map φ1 : A1 ⊗ A1 → A2 is generated by a12 − q11 = 42

elements. We record the indices of basis elements of A1 · A1 by the set
S = {(1, 2), (1, 5), (1, 6), (1, 7), (2, 3), (3, 5), (4, 5)}.
A basis of Ker φ1 , as defined in (2.5), is given by elements of two types:

(1) [zi1 ] ⊗ [zj1 ], for 1 i, j 7, (i, j ) ∈ S and (j, i) ∈ S;
(2) [zi1 ] ⊗ [zj1 ] + [zj1 ] ⊗ [zi1 ], for (i, j ) ∈ S.
Therefore, according to the types above, one defines p s1(i,j ) = (

ps1(i,j ) k )k=1,...,7 as:

z1 if k = i
s(i,j ) k = j
(1) p 1 for all 1 i, j 7, (i, j ) ∈ S and (j, i) ∈ S;
0 if k = i,
⎧
⎪
⎨zj if k = i
⎪ 1
s1(i,j ) k = zi1 if k = j
(2) p for all (i, j ) ∈ S.
⎪
⎪
⎩0 if k = i, j,
5.6 Next, we find the nonzero elements

πs3 ∈ K3 defined in (2.6) as follows. For the
elements coming from Koszul relations in K, we choose πs3 = 0. The only nonzero
products in K2 come from elements of type (1) and they are
z21 ∧ z41 = x 2 y 2 T12 = −z41 ∧ z21 and z51 ∧ z71 = xz4 T13 = −z71 ∧ z51 .
Therefore, we choose the following nontrivial liftings in K3 :

πs3(2,4) = xz2 T123 = −
πs3(4,2) and
πs3(5,7) = −x 2 yzT123 = −
πs3(7,5) .
5.7 We describe next the elements of A3 .

Claim 1 All the products in A1 · A1 · A1 are zero.
Proof First, all the products in A1 · A1 · A1 not involving [z11 ] have the coefficients
of [Tij k ] of degree six, and R6 = 0. Thus, all such products are zero. Second, the
only pairs (i, j ) with 1 < i < j 7 such that [zi1 ] ∧ [zj1 ] is nonzero in A1 · A1 are
in the set {(2, 3), (3, 5), (4, 5)}, but [z11 ∧ z31 ] = [z11 ∧ z41 ] = 0. Hence, all products
involving [z11 ] at least once are zero. The claim now follows.

Therefore, A1 · A2 = A1 · A2 , it has rank q12 = 10, and its basis is given by the
classes of
z11 ∧ z32 = xz2 wT124 , z11 ∧ z42 = yz2 wT134 ,

z11 ∧ z52 = x 2 ywT124 − xz2 wT134 , z11 ∧ z62 = x 2 zwT134 ,
z21 ∧ z12 = x 2 ywT124 , z21 ∧ z22 = x 2 y 2 T124 ,
z31 ∧ z62 = −x 2 yzwT123 , z51 ∧ z12 = yz2 wT234 ,

z51 ∧ z22 = y 2 z2 T234 , z51 ∧ z32 = −xz4 T123 = z41 ∧ z62 .
A basis of A3 is {[zt3 ]}t=1,...,4 , where
z13 = yz2 wT123 , z23 = y 2 z2 T123 , z33 = yz2 wT124 , z43 = z4 T234 .
5.8 Using Proposition 2.4, we describe the elements p ui

2 , for 1 u a a −
1 2
a1 q11 − q12 + b = 46 + b and 1 i 7, as defined in (2.10).
Claim 2 Let A, B be as in Proposition 2.4 and let b be as in Summary 2.5. Then
Spank A = A1 ⊗ (A1 · A1 ) and b = 0.
Proof It is clear that Spank A ⊆ A1 ⊗ (A1 · A1 ). For 1 i, j, h 7, any nonzero

element [zi1 ] ⊗ [zj1 ] ∧ [zh1 ] in A1 ⊗ (A1 · A1 ) has (j, h) ∈ S or (h, j ) ∈ S. We show
that each such element is in Spank A.
In the case {(i, j ), (i, h)} ⊆ S we have

7
[zi1 ] ⊗ [zj1 ] ∧ [zh1 ] = [zk1 ] ⊗ [
ps1(i,j ) k ] ∧ [zh1 ],
k=1
as in case (1) of 5.5. By Koszul relation, the case (i, h) ∈ S reduces to the case
(i, j ) ∈ S in which we have

7
[zi1 ]⊗[zj1 ]∧[zh1 ] = [zi1 ]⊗[zj1 ]∧[zh1 ]+[zj1 ]⊗[zi1 ]∧[zh1 ] = [zk1 ]⊗[
ps1(i,j ) k ]∧[zh1 ],
k=1
as in case (2) of 5.5. The first equality above follows from the proof of Claim 1.
This implies SpanQ A = A1 ⊗ (A1 · A1 ), so rankQ (Spank A) = a1 q11 . By
Proposition 2.4(b) we get b = 0.

Remark that Claim 2 implies Claim 1, since SpanQ A ⊆ Ker φ2 .
By definition of A2 as in Summary 2.5 and Claim 2, the following holds:
!
A1 ⊗ A2 = A1 ⊗ (A1 · A1 ) ⊕ (A1 ⊗ A2 ) = (SpanQ A) ⊕ (A1 ⊗ A2 ).
By Proposition 2.4, Ker φ2 = (SpanQ A) ⊕ B, where B = (Ker φ2 ) ∩ (A1 ⊗ A2 ).

In order to find a basis for B, we record the indices of basis elements of A1 · A2 by
the set
U = {(1, 3), (1, 4), (1, 5), (1, 6), (2, 1), (2, 2), (3, 6), (5, 1), (5, 2), (5, 3), (4, 6)}.
A basis of B is given by elements of two types:

(1 ) [zi1 ] ⊗ [z
2 ] for all 1 i 7 and 1
8 such that (i,
) ∈ U ;
(2 ) [z41 ] ⊗ [z62 ] − [z51 ] ⊗ [z32 ].
Therefore, according to the types above, one defines p u2(i,
) = (
pu2(i,
) k )k=1,...,7 as:

z2 if k = i
u(i,
) k =
(1 ) p 2 for all 1 i 7 and 1

8 such that (i,
) ∈ U ;
0 if k = i,
⎧
⎪
⎪ 2 if k = 4
⎨z6
u(4,6) k = −z32 if k = 5
(2 ) p 2
⎪
⎪
⎩0 if k = 4, 5.
5.9 It is easy to check that in K3 we have the equalities:
(1 ) zi1 ∧ z
2 = 0 for all 1 i 7 and 1
8 such that (i,
) ∈ U ;
(2 ) z41 ∧ z62 − z51 ∧ z32 = 0.
Therefore, we may choose
πu4 = 0 for all 1 u 46.
5.10 A similar argument as in the proof of Claim 1 gives A1 · A3 = 0, and hence
q13 = 0. Moreover, A2 · A2 has rank q22 = 2 and its basis is given by the classes of
z12 ∧ z62 = −x 2 yzwT1234 and z22 ∧ z62 = −xz4 T1234 .
As A1 · A3 = 0, any element in A1 , A1 , A1 ⊆ A4 has a representative given by

42
42

πs3 ∧ ps1 such that [1
psk ∧ ps1 ] = 0, for all 1 k 7, (5.1)
s=1 s=1
where ps1 ∈ Ker ∂1K is as in Proposition 2.8, and p si

1 and πs3 are defined in (2.5) and
(2.6) respectively. As described above, the only nontrivial lifting elements πs3 ∈ K3
are
πs(2,4) = −
3 πs(4,2) and
3 πs(5,7) = −
3 3
πs(7,5) . Since all of them contain T123 , only the
component of ps1 that contains z11 = wT4 contributes to a nonzero Massey product.
Thus, there exist α, β ∈ R such that

42
[
πs3 ∧ ps1 ] = [
πs3(2,4) ∧ ps1(2,4) ] + [
πs3(5,7) ∧ ps1(5,7) ]
s=1
πs3(2,4) ∧ z11 ] ∧ [α] + [

= [ πs3(5,7) ∧ z11 ] ∧ [β]
= [xz2 wT1234 ] ∧ [α] + [x 2 yzwT1234 ] ∧ [β]

= [z21 ], [z41 ], [z11 ] ∧ [α] − [z12 ∧ z62 ] ∧ [β].
The last equality follows from the following computation:
[z21 ], [z41 ], [z11 ] = {[(∂3K )−1 (z21 ∧ z41 ) ∧ z11 + z21 ∧ (∂3K )−1 (z41 ∧ z11 )]}
= {[
πs3(2,4) ∧ z11 + z21 ∧
πs3(4,1) ]}
= {[xz2 wT1234 ]}.
By abuse of notation, we write [z21 ], [z41 ], [z11 ] = [xz2 wT1234 ], which is not in
(A1 · A3 + A2 · A2 ), as showed in [2, Section 7]. It is clear now that the rank of the
Q-vector space
A1 · A3 + A2 · A2 + SpanQ A1 · A1 · A1
is a = 3 and its basis is given by
[z12 ] ∧ [z62 ], [z22 ] ∧ [z62 ], and [z21 ], [z41 ], [z11 ].
Thus, a basis of A4 is {[zr4 ]}r=1,2 where
z14 = yz2 wT1234 and z24 = y 2 z2 T1234 .
5.11 By Proposition 4.2, we conclude that the ring R has the following invariants:
a1 = 7, a2 = 15, a3 = 14, a4 = 5,
q11 = 7, q12 = 10, q13 = 0, q22 = 2,
a = 3, b = 0,
P3 = 7, P4 = 45, P5 = 221.
Acknowledgments The authors thank Hailong Dao for inspiring them to work on this project
and for very fruitful discussions. They also thank Frank Moore for helping them with the
DGAlgebras package [15] to check the computations, and thank the referee for helpful
suggestions. The first author was supported by the Naval Academy Research Council in Summer
2020.
References
1. D. J. Anick, A counterexample to a conjecture of Serre, Ann. of Math. (2) 115 (1982), no. 1,
1–33.
2. L. L. Avramov, On the Hopf algebra of a local ring, Izv. Akad. Nauk SSSR Ser. Mat. 38 (1974),
253–277.
3. L. L. Avramov, Homological asymptotics of modules over local rings, Commutative algebra
(Berkeley, CA, 1987), Math. Sci. Res. Inst. Publ., Springer, New York, vol. 15, (1989), 33–62.
4. L. L. Avramov, Infinite free resolutions, Six Lectures on Commutative Algebra (Bellaterra,

1996), Progr. Math. 166 (1998), Birkhäuser, Basel, 1–118.
5. L. L. Avramov, A cohomological study of local rings of embedding codepth 3, J. Pure
Appl. Algebra 216 (2012), no. 11, 2489–2506.
6. J. Burke, Higher homotopies and Golod rings, preprint, arXiv:1508.03782.
7. L. W. Christensen; O. Veliche, Acyclicity over local rings with radical cube zero, Illinois J.
Math. 51 (2007), no. 4, 1439–1454.
8. E. S. Golod, On the homology of some local rings, Dokl. Akad. Nauk SSSR 144 (1962), 479–
482 (in Russian); English translation: Soviet Math. Dokl. 3 (1962), 745–748.
9. T. H. Gulliksen; G. Levin, Homology of local rings, Queen’s Papers in Pure Appl. Math. 20
(1969), x+192 pp.
10. L. Katthän, A non-Golod ring with a trivial product on its Koszul homology, J. Algebra, 479
(2017), 244–262.
11. A. R. Kustin; M. Miller, Classification of the Tor-algebras of codimension four Gorenstein
local rings, Math. Z. 190 (1985), 341–355.
12. D.-M. Lu; J.H. Palmieri; Q.-S. Wu; and J.J. Zhang, A-infinity structure on Ext-algebras, J. Pure
Appl. Alg. 213 (2009), 2017–2037.
13. D. R. Grayson; M. Stillman, Macaulay2, a software system for research in algebraic geometry,
Available at http://www.math.uiuc.edu/Macaulay2/.
14. J. P. May, Matric Massey products, J. Algebra, 12(4) (1969), 533–568.
15. F. Moore, DGAlgebras: Data type for DG algebras. Version 1.0.1, A Macaulay2 package
available at https://github.com/Macaulay2/M2/tree/master/M2/Macaulay2/packages.
16. J.-E. Roos, On some unexpected rings that are close to Golod rings, (2016), unpublished notes.
17. Y. Yoshino, Modules of G-dimension zero over local rings with the cube of maximal ideal
being zero, Commutative algebra, singularities and computer algebra (Sinaia, 2002), 255–273,
NATO Sci. Ser. II Math. Phys. Chem., 115, Kluwer Acad. Publ., Dordrecht, 2003.

Women in Commutative Algebra

Uploaded by

Document Information

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Women in Commutative Algebra

Uploaded by

Copyright:

Available Formats

Association for Women in Mathematics Series

More information about this series at https://link.springer.com/bookseries/13764

ISSN 2364-5733 ISSN 2364-5741 (electronic)

Mathematics Subject Classification: 13-XX, 14-XX, 18-XX

© The Author(s) and the Association for Women in Mathematics 2021

Syracuse, NY, USA Claudia Miller

On Gerko’s Strongly Tor-independent Modules . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1

A Survey on the Koszul Homology Algebra . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 257

Hannah Altmann and Keri Sather-Wagstaff

Keywords Differential graded algebras · Semidualizing modules · Syzygies ·

We are interested in how existence of certain sequences of modules over a local

© The Author(s), under exclusive license to Springer Nature Switzerland AG 2021 1

Here ecodepth(R) = β0R (mR ) − depth(R) is the embedding codepth of R, where

Let R be a nonzero commutative Noetherian ring with identity. We work with

sup(X) = sup{i ∈ Z | Xi = 0} inf(X) = inf{i ∈ Z | Xi = 0}.

The amplitude of X is amp(X) = sup(X) − inf(X). Frequently we consider these

a semibasis for X is a set of homogeneous elements of X that is a basis for X

3 Perfect DG Modules and Tensor Products

Throughout this section, let A be a positively graded commutative homologically

Another long exact sequence argument gives the desired conclusion.

are strongly Tor-independent, then so is any reordering by the commutativity of

4 Syzygies and Strongly Tor-independent DG Modules

Throughout this section, let (A, mA ) be a local homologically bounded DG algebra,

Proposition 4.4 Let K be a homologically finite DG A-module and set K  =

and argue as in the proof of Lemma 4.3.

Proof Induct on m using Proposition 4.4.

Notice, Im(αi ) ⊆ Ki ⊆ mA Li for i = 1, 2, . . . , j .

where θ = ⊗i=1,...,j −1 αi and β is induced by θ ⊗ αj .

Claim: H (β) is 1-1. Notice that Hi (G ⊗A Gj ) = 0 for all i < r1 + . . . + rj , so it

Proposition 4.4 implies

Thus, the long exact sequence in homology associated to (5) implies Hi (G ⊗A αj )

Thus, the long exact sequence in homology associated to (6) implies Hi (θ ⊗ Lj ) is

have H (Kl ) = 0 for some l. Hence, Kl has a semifree basis concentrated in a

gives rise to a sequence of n strongly Tor-independent non-perfect DG A -modules.

5 Proof of Theorem 1.1

amp(H (K)) = ecodepth(R) = amp(K). (7)

Claim: The sequence K ⊗LR N1 , . . . , K ⊗R Nn is a strongly Tor-independent

From this we get the first equality in the next display.

Tor-independent sequence M1 , . . . , Mn over A. Now, Theorem 4.7 and (7) imply

Use prime avoidance to find an R-regular element x ∈ mR −m2R . Set R = R/xR.

strongly Tor-independent over R. By our induction hypothesis we have

R (M) = min{i 0 | miR M = 0}.

The generalized Loewy length of R is then

R (R/x) | x is a system of parameters of R}.

R/x (R/x). Since R is Cohen–Macaulay,

1. L. L. Avramov, H.-B. Foxby, and S. Halperin, Differential graded homological algebra, in

© The Author(s), under exclusive license to Springer Nature Switzerland AG 2021 11

Xn = x0 , x2 − x3 , x4 − x5 , . . . , x2n − x2n+1 , x2n+2 − x2n+3 , x2n+4

in R(n, t) is a system of parameters for R(n, t).

reg R(n, t) = n/2 + 1.

e(R(n)) = e(R(n − 1)) + e(R(n − 2)).

2 The Family of Toric Rings

2.1 The Family F of Graphs

Definition 2.1 For each n 0 and each t ∈ Fn+1 2 , we construct a ladder-like

• For n = 1, to create Lt1 (regardless of what t is in F22 ), we add another column

We denote the subfamily of graphs coming from t = (1, 1, . . . , 1) by F1 ⊂ F.

x0 x2 x5 x9 x13 x17 x21 · · · x2n+1

s2i := x2i+1 x2i+3 − xj2i x2i+4

coming from the subarray

for some j2i ∈ {0, 2, 3, . . . , 2i − 2} and

s2i+1 := x2i+2 x2i+3 − xj2i+1 x2i+4

coming from the subarray

Then in fact we have a subarray of the form

so that we have the distinguished minors

Proposition 4.4 Let K be a homologically finite DG A-module and set K =

Notice, Im(αi ) ⊆ Ki ⊆ mA Li for i = 1, 2, . . . , j .

have H (Kl ) = 0 for some l. Hence, Kl has a semifree basis concentrated in a

gives rise to a sequence of n strongly Tor-independent non-perfect DG A -modules.

R (R/x) | x is a system of parameters of R}.

R/x (R/x). Since R is Cohen–Macaulay,

reg R(n, t) = n/2 + 1.

:= k[x3 , x5 , . . . , x2n+1 , x2n+3 ] ∼

By the same process (detailed below), we obtain the ideal I ∼