2,1982,4

Journal of Algebra
EDITOR-IN-CHIEF
Michel Broué
Journal of Algebra
Institut Henri Poincaré, 11, rue Pierre et Marie Curie,
F-75005 Paris, France
E-mail: jalgebra@ihp.jussieu.fr
EDITORIAL BOARD
Nicolás Andruskiewitsch Kazuhiko Kurano Aner Shalev

Universidad de Córdoba Meiji University Hebrew University
Facultad de Matemática, kurano@math.meiji.ac.jp Institute of Mathematics
Astronomía y Física shalev@math.huji.ac.il
andrus@famaf.unc.edu.ar Gus I. Lehrer
University of Sydney Ronald Solomon
Luchezar L. Avramov School of Mathematics and Statistics Ohio State University*
University of Nebraska g.lehrer@maths.usyd.edu.au solomon@math.ohio-state.edu
avramov@math.unl.edu
Martin Liebeck J.T. Stafford
Eva Bayer-Fluckiger Imperial College, London* University of Michigan*
École Polytechnique Federale m.liebeck@ic.ac.uk jts@umich.edu
de Lausanne*
eva.bayer@epfl.ch Peter Littelmann Gernot Stroth
Universität zu Köln Universität Halle–Wittenberg
Steven Dale Cutkosky Mathematisches Institut stroth@mathematik.uni-halle.de
University of Missouri littelma@math.uni-koeln.de
cutkoskys@missouri.edu
Michel Van den Bergh
Dihua Jiang Laurent Moret-Bailly Limbergs Universitair Centrum
University of Minnesota Université de Rennes 1 Department WNI
School of Mathematics laurent.moret-bailly@univ-rennes1.fr vdbergh@luc.ac.be
dhjiang@math.umn.edu
Leonard L. Scott, Jr. Changchang Xi
Masaki Kashiwara University of Virginia* Beijing Normal University
University of Kyoto lls2l@virginia.edu School of Mathematical Sciences
masaki@kurims.kyoto-u.ac.jp xicc@bnu.edu.cn
Vera Serganova
E.I. Khukhro UC Berkeley Efim Zelmanov
Institute of Mathematics, SO RAN Department of Mathematics University of California, San Diego*
khukhro@yahoo.co.uk serganova@math.berkeley.edu ezelmano@euclid.ucsd.edu
EDITOR OF THE SECTION ON COMPUTATIONAL ALGEBRA

Gerhard Hiss
Lehrstuhl D für Mathematik
RWTH Aachen
E-mail: gerhard.hiss@math.rwth-aachen.de
SECTION ON COMPUTATIONAL ALGEBRA
Jon Carlson Meinolf Geck Gunter Malle

University of Georgia* Aberdeen University Universität Kaiserslautern
jfc@math.uga.edu King’s College Fachbereich Mathematik
Department of Mathematical Sciences malle@mathematik.uni-kl.de
John Cremona m.geck@maths.abdn.ac.uk
University of Warwick Eamonn O’Brien
Derek Holt The University of Auckland*
Mathematics Institute University of Warwick
j.e.cremona@warwick.ac.uk e.obrien@auckland.ac.nz
Mathematics Institute
dfh@maths.warwick.ac.uk Bruno Salvy
Patrick Dehornoy INRIA Rocquencourt
Université de Caen William M. Kantor
University of Oregon* Algorithms Project
Laboratoire de Mathématiques bruno.salvy@inria.fr
patrick.dehornoy@math.unicaen.fr kantor@darkwing.uoregon.edu
Reinhard Laubenbacher Jean-Yves Thibon
Harm Derksen Virginia Tech Université de Marne-la-Vallée
University of Michigan* Virginia Bioinformatics Institute Institut Gaspard Monge
hderksen@umich.edu reinhard@vbi.vt.edu jyt@univ-mlv.fr
FOUNDING EDITOR: Graham Higman, 1964–1984

EDITOR-IN-CHIEF: Walter Feit, 1985–2000
*Department of Mathematics
Journal of Algebra 321 (2009) 743–757
Contents lists available at ScienceDirect
Journal of Algebra
www.elsevier.com/locate/jalgebra
Congruences on monoids of transformations preserving

the orientation on a finite chain ✩
Vítor H. Fernandes a,b,∗ , Gracinda M.S. Gomes a,c , Manuel M. Jesus a,b
a
Centro de Álgebra da Universidade de Lisboa, Av. Prof. Gama Pinto, 2, 1649-003 Lisboa, Portugal
b
Departamento de Matemática, Faculdade de Ciências e Tecnologia, Universidade Nova de Lisboa, Monte da Caparica,
2829-516 Caparica, Portugal
c
Departamento de Matemática, Faculdade de Ciências, Universidade de Lisboa, 1749-016 Lisboa, Portugal
a r t i c l e i n f o a b s t r a c t
Article history: The main subject of this paper is the description of the congruences
Received 12 April 2006 on certain monoids of transformations on a finite chain Xn with n
Available online 5 December 2008 elements. Namely, we consider the monoids ORn and PORn of
Communicated by Michel Broue
all full, respectively partial, transformations on Xn that preserve
Keywords:
or reverse the orientation, as well as their respective submonoids
Congruences OP n and POP n of all orientation-preserving elements. The
Orientation-preserving inverse monoid PORI n of all injective elements of PORn is also
Orientation-reversing considered.
Transformations © 2008 Elsevier Inc. All rights reserved.
Introduction and preliminaries
In the 1987 “Szeged International Semigroup Colloquium” J.-E. Pin asked for an effective description
of the pseudovariety of semigroups O generated by all semigroups of order-preserving full transfor-
mations on a finite chain, i.e. an algorithm to decide whether or not a finite semigroup belongs to O.
This problem only had essential progresses after 1995. First, Higgins [17] proved that O is self-dual
and does not contain all R-trivial semigroups (and so O is properly contained in A, the pseudo-
variety of all finite aperiodic semigroups), although every finite band belongs to O. Next, Vernitskii
and Volkov [23] generalized Higgins’s result by showing that every finite semigroup whose idempo-
tents form an ideal is in O and in [8] the first author proved that the pseudovariety of semigroups
POI generated by all semigroups of injective order-preserving partial transformations on a finite chain
is a (proper) subpseudovariety of O. On the other hand, Almeida and Volkov [2] showed that the
✩
This work was developed within the activities of Centro de Álgebra da Universidade de Lisboa, supported by FCT and FEDER,
within project POCTI-ISFL-1-143.
*
Corresponding author at: Centro de Álgebra da Universidade de Lisboa, Av. Prof. Gama Pinto, 2, 1649-003 Lisboa, Portugal.
E-mail addresses: vhf@fct.unl.pt (V.H. Fernandes), ggomes@cii.fc.ul.pt (G.M.S. Gomes), mrj@fct.unl.pt (M.M. Jesus).
0021-8693/$ – see front matter © 2008 Elsevier Inc. All rights reserved.
doi:10.1016/j.jalgebra.2008.11.005
744 V.H. Fernandes et al. / Journal of Algebra 321 (2009) 743–757
interval [O, A] of the lattice of all pseudovarieties of semigroups has the cardinality of the continuum
and Repnitskiı̆ and Volkov [22] proved that O is not finitely based. Moreover, Repnitskiı̆ and Volkov
proved that any pseudovariety of semigroups V such that POI ⊆ V ⊆ O ∨ R ∨ L is not finitely based,
where R and L are the pseudovarieties of semigroups of all R-trivial semigroups and of all L-trivial
semigroups, respectively.
Pin’s question, which is still unanswered, has motivated the study of various extensions of monoids
of order-preserving transformations, namely monoids of order-reversing, orientation-preserving or
orientation-reversing transformations.
Clearly, intrinsically related with the description of a pseudovariety it is the knowledge of the
congruences and the Green relations on its members.
For n ∈ N, let X n be a finite chain with n elements, say X n = {1 < 2 < · · · < n}. As usual, we denote
by PT n the monoid (under composition) of all partial transformations of X n . The submonoid of PT n
of all full transformations of X n and the (inverse) submonoid of all injective partial transformations
of X n are denoted by Tn and In , respectively.
Let a = (a1 , a2 , . . . , at ) be a sequence of t (t 0) elements from the chain X n . We say that
a is cyclic [anti-cyclic] if there exists no more than one index i ∈ {1, . . . , t } such that ai > ai +1
[ai < ai +1 ], where at +1 denotes a1 . Notice that, the sequence a is cyclic [anti-cyclic] if and only
if a is empty or there exists i ∈ {0, 1, . . . , t − 1} such that ai +1 ai +2 · · · at a1 · · · ai
[ai +1 ai +2 · · · at a1 · · · ai ] (the index i ∈ {0, 1, . . . , t − 1} is unique unless a is constant
and t 2). Let s ∈ PT n and suppose that Dom(s) = {a1 , . . . , at }, with t 0 and a1 < · · · < at . We
say that s is an orientation-preserving [orientation-reversing] transformation if the sequence of its
images (a1 s, . . . , at s) is cyclic [anti-cyclic]. It is easy to show that the product of two orientation-
preserving or of two orientation-reversing transformations is orientation-preserving and the prod-
uct of an orientation-preserving transformation by an orientation-reversing transformation is clearly
orientation-reversing (see [5]).
Denote by POP n the submonoid of PT n of all partial orientation-preserving transformations
of X n . As usual, OP n denotes the monoid POP n ∩ Tn of all full transformations of X n that preserve
the orientation. This monoid was considered by McAlister in [21], by Catarino and Higgins in [5], by
Catarino in [4] and by Arthur and Ruškuc in [3]. The injective counterpart of OP n , i.e. the inverse
monoid POPI n = POP n ∩ In , was studied by the first author in [10,12].
More comprehensive classes of monoids are obtained when we take transformations that ei-
ther preserve or reverse the orientation. In this way we get PORn , the submonoid of PT n of
all partial transformations that preserve or reverse the orientation. Within Tn sits the submonoid
ORn = PORn ∩ Tn and inside In is PORI n = PORn ∩ In .
The following diagram, with respect to the inclusion relation and where Cn denotes the cyclic
group of order n, clarifies the relationship between these semigroups:
The study of transformations that respect the orientation is intrinsically associated to the knowl-
edge of the ones that respect the order. A transformation s in PT n is called order-preserving if
x y implies xs ys, for all x, y ∈ Dom(s). Denote by POn the submonoid of PT n of all partial
order-preserving transformations of X n . The monoid POn ∩ Tn of all full transformations of X n that
preserve the order is denoted by On . This monoid has been largely studied by several authors (e.g. see
V.H. Fernandes et al. / Journal of Algebra 321 (2009) 743–757 745
[1,16,18,20]). The injective counterpart of On is the inverse monoid POI n = POn ∩ In , which is con-
sidered, for example, in [6,8,9,11,13].
In this paper, on the one hand we aim to describe the Green’s relations on some of the monoids
mentioned above and to use the descriptions obtained to calculate their sizes and ranks. This type
of questions were considered by Catarino and Higgins [5] for OP n and for ORn ; by Fernandes [10]
for POPI n and by the authors [14] for PORI n . In Section 1 we study the monoids POP n and
PORn . On the other hand, we want to describe the congruences of the monoids OP n , POP n , ORn ,
PORn and PORI n . It was proved by Aı̆zenštat [1], and later by Lavers and Solomon [20], that the
congruences of On are exactly the Rees congruences. A similar result was proved by the first author
[11] for the monoid POI n and by the authors [15] for the monoid POn . Fernandes [10] proved
that the congruences on POPI n are associated with its maximal subgroups. In Section 2, under
certain conditions, on an arbitrary finite semigroup we define a class of congruences associated to
its maximal subgroups. In Section 3, we show that, as in the case of POPI n , all congruences in the
monoids referred above are of this type.
Next, for completeness, we recall some notions and fix the notation.
Let M be a monoid. We denote by E ( M ) its set of idempotents. Let J be the quasi-order on M
defined by u J v if and only if MuM ⊆ M v M, for all u , v ∈ M. Denote by J u the J-class of an
element u ∈ M. As usual, a partial order relation J is defined on the set M /J by setting J u J J v if
and only if u J v, for all u , v ∈ M. For u , v ∈ M, we write u <J v and also J u <J J v if and only if
u J v and (u , v ) ∈
/ J.
The Rees congruence ρ I on M associated to an ideal I of M is defined by (u , v ) ∈ ρ I if and only
if u = v or u , v ∈ I , for all u , v ∈ M. For convenience, we admit the empty set as an ideal. In what
follows the identity congruence will be denoted by 1 and the universal congruence by ω . The rank
of M is, by definition, the minimum of the set {| X |: X ⊆ M and X generates M }. For more details,
see e.g. [19].
A subset C of the chain X m is said to be convex if x, y ∈ C and x z y imply that z ∈ C . An
equivalence ρ on X m is convex if its classes are convex. We say that ρ is of weight k if | X m /ρ | = k.
m−1
Clearly, the number of convex equivalences of weight k on X m is k−1 .
Now let G be a cyclic group of order n. It is well known that there exists a one-to-one correspon-
dence between the subgroups of G and the (positive) divisors of n. Since G is abelian, all subgroups
are normal, and so there is a one-to-one correspondence between the congruences of G and the
(positive) divisors of n. These correspondences are, in fact, lattice isomorphisms.
The dihedral group D n of order 2n (n 3) can be defined by the group presentation

x, y xn = 1, y 2 = 1, yx = x−1 y
and its proper normal subgroups are:

n
(1) x2 , y , x2 , xy and x p , with p a divisor of n, when n is even;
n
(2) x , with p a divisor of n, when n is odd.
p
See [7] for more details.

The following concept will be used in Section 3. Let ( P 1 , 1 ) and ( P 2 , 2 ) be two disjoint posets.
The ordinal sum of P 1 and P 2 (in this order) is the poset P 1 ⊕ P 2 with universe P 1 ∪ P 2 and partial
order defined by: for all x, y ∈ P 1 ∪ P 2 , we have x y if and only if x ∈ P 1 and y ∈ P 2 ; or x, y ∈ P 1
and x 1 y; or x, y ∈ P 2 and x 2 y. Observe that this operator on posets is associative but not
commutative.
1. The monoids POP n and PORn
In this section we describe the Green’s relations and calculate the sizes and the ranks of the
monoids POP n and PORn . We show that their structure is similar to the one of the monoids
POPI n , PORI n , PT n and In . In particular, in all of them, the J-classes are the sets of all elements
with the same rank and form a chain, with respect to the partial order J . Notice also that all these
monoids are regular.
In what follows, we must have in mind that an element of PORn is either in POP n or it re-
verses the orientation. Denote by PRn the set of all orientation-reversing partial transformations
of X n . Clearly, PORn = POP n ∪ PRn . In view of the next lemma, we have POP n ∩ PRn =
{s ∈ POP n : | Im(s)| 2}.
Lemma 1.1. (See [5].) Let a be a cyclic [anti-cyclic] sequence. Then a is (also) anti-cyclic [cyclic] if and only if a
has no more than two distinct values.
It is easy to show that E (PORn ) = E (POP n ).

Let us consider the permutation of order two

1 ··· n − 1 n
2
h= .
n n − 1 ··· 2 1
Clearly, h2 = 1 and h is an orientation-reversing full transformation. We showed in [14] that POPI n

together with h form a set of generators of PORI n . Similarly, by just noticing that, given an
orientation-reversing transformation s, the product sh is an orientation-preserving transformation,
it follows that PORn is generated by POP n ∪ {h}.
We prove that PORn is regular, using the fact that POP n is already known to be regular [10]. It
remains to show that all the elements of PRn are regular. Let s be an orientation-reversing transfor-
mation. Then sh ∈ POP n and so there exists s ∈ POP n such that (sh)s (sh) = sh. Multiplying on the
right by h, we obtain s(hs )s = s, whence s is a regular element of PORn .
Next consider the following permutation

1 2 ··· n − 1 n
g= ,
2 3 ··· n 1
which is an element of POP n such that g n = 1. As in [10, Proposition 3.1], it is a routine matter to
prove the following (non-unique) factorization of an element of POP n :
Proposition 1.2. Let s ∈ POP n . Then there exist i ∈ {0, 1, . . . , n − 1} and u ∈ POn such that s = g i u.
As an immediate consequence of this proposition, we have
Corollary 1.3. The monoid POP n is generated by POn ∪ { g }.
Corollary 1.4. Let s ∈ PORn . Then there exist i ∈ {0, 1, . . . , n − 1}, j ∈ {0, 1} and u ∈ POn such that
s = g i uh j .
Notice that, with the notation of the last corollary, we can always take:
(1) j = 0, if s ∈ POP n ;
(2) u ∈ On , if s ∈ ORn .
Therefore, wherever in this paper we take such a factorization of an element s of PORn , we will
consider j and u as above.
Denote by M n either the monoid PORn or the monoid POP n .
Proposition 1.5. Let s and t be elements of M n . Then
(1) sRt if and only if Ker(s) = Ker(t );

(2) sLt if and only if Im(s) = Im(t );
(3) s J t if and only if | Im(s)| | Im(t )|.
Proof. Since M n is a regular submonoid of PT n , conditions (1) and (2) follow immediately from well
known results on regular semigroups (e.g. see [19]).
Next we prove condition (3). First, suppose that s J t. Then there exist x, y ∈ M n such
that s = xt y. Since Im(s) ⊆ Im(t y ) and | Im(t y )| = | Im(t ) y | | Im(t )|, then | Im(s)| | Im(t )|. Con-
versely, let s, t ∈ M n be such that | Im(s)| | Im(t )|. By Corollary 1.4, there exist i 1 , i 2 ∈ {0, . . . , n − 1},
j 1 , j 2 ∈ {0, 1} and u , v ∈ POn such that s = g i 1 uh j 1 and t = g i 2 vh j 2 . Thus | Im(s)| = | Im(u )| and
| Im(t )| = | Im( v )|, since g i 1 , g i 2 , h j 1 , h j 2 are permutations. Hence u J v in POn (see [16]) and so
there exist x, y ∈ POn such that u = xv y. Then

s = g i 1 uh j 1 = g i 1 xv yh j 1 = g i 1 xg n−i 2 g i 2 vh j 2 h2− j 2 yh j 1 = g i 1 xg n−i 2 t h2− j 2 yh j 1 ,
with g i 1 xg n−i 2 , h2− j 2 yh j 1 ∈ M n , and so s J t in M n , as required. 2
It follows, from condition (3), that M n /J = { J 0 <J J 1 <J · · · <J J n }, where J k denotes the set
{s ∈ Mn | | Im(s)| = k}, for all 0 k n.
On the other hand, given an element s ∈ M n ∩ In , from conditions (1) and (2) above and from
the corresponding descriptions for the monoids POPI n and PORI n ([10, Proposition 2.4] and
[14, Proposition 5.3], respectively), it follows that the H-class of s in M n ∩ In coincides with its
H-class in Mn . Thus, as for the monoid POPI n [10, Proposition 2.6], we have
Proposition 1.6. Let s ∈ POP n be such that 1 | Im(s)| = k n. Then | H s | = k. Moreover, if s is an idempo-
tent, H s is a cyclic group of order k.
Since a transformation s ∈ PORn is both orientation-preserving and orientation-reversing if and

only if | Im(s)| 2, we obtain the following:
Corollary 1.7. Let s ∈ PORn be such that 1 | Im(s)| = k 2. Then | H s | = k. Moreover, if s is an idempotent,
H s is a cyclic group of order k.
Also, as for the monoid PORI n [14, Proposition 5.3], we get
Proposition 1.8. Let s ∈ PORn be such that 3 | Im(s)| = k n. Then | H s | = 2k. Moreover, if s is an idem-
potent, H s is a dihedral group of order 2k.
Let s be an element of PORn with rank k, 0 k n. Suppose that Im(s) = {b1 , . . . , bk }. Consider-
ing the kernel classes of s, we obtain two types of partitions of the domain of s into intervals:
P 1 ... P k
(a) Dom(s) = ˙ i =1 P i with s =
k
; or
b1 ... bk
k+1 P ... P P
(b) Dom(s) = ˙ i =1 P i with s = b 1 ... b k bk+1 .
1 k 1
Notice that, in the first case, P 1 , . . . , P k are precisely the kernel classes of s whereas, in the second
one, the kernel classes are P 1 ∪ P k+1 , P 2 , . . . , P k .
Now let k ∈ {2, . . . , n} and suppose that s is an element of POP n with rank k. If s satisfies (a)
then Ker(s) is a convex equivalence on Dom(s) of weight k. On the other hand, if s verifies (b), we can
associate to s a convex relation of weight k + 1 (with classes P 1 , . . . , P k , P k+1 ). Therefore the number
of R-classes of rank k with the same domain as s is given by

|Dom(s)| − 1 |Dom(s)| − 1 |Dom(s)|
+ = ,
k−1 k k

n n j n j nn−k
whence the total number of R-classes of rank k is equal to j =k j k
. As j k
= k j −k
, we have
n n j n
n n−k n
n−k n−k n n−k
j =k j k
= k j =k j −k
= k j =0 j
= k
2 . Since the number of L-classes of rank k is,
n
clearly, equal to and, by Proposition 1.6, each H-class of rank k has k elements, the monoid
k nn
POP n has precisely k k k
2n−k elements of rank k. Furthermore, by noticing that the number of
transformations of rank 1 of PT n (and so of POP n ) is equal to (2n − 1)n, we conclude the following
result:

n n2 n−k
Proposition 1.9. |POP n | = 1 + (2n − 1)n + k=2 k k 2 .
As there is a natural bijection between POP n and PRn (obtained by simply reversing the se-
quence of the images), |PORn | = 2|POP n | − |{s ∈ POP n | | Im(s)| 2}|, whence:
n2
n n2 n−k
Proposition 1.10. |PORn | = 1 + (2n − 1)n + 2 2
2n−2 + k=3 2k k 2 .
Naturally, at this point, we would like to compute the rank of these monoids.
Let us consider the following elements s0 , s1 , . . . , sn−1 of POI n :

2 ··· n − 1 n
s0 =
1 ··· n − 2 n − 1
and

1 ··· n − i − 1 n−i n − i + 2 ··· n
si = ,
1 ··· n − i − 1 n − i + 1 n − i + 2 ··· n
for i ∈ {1, 2, . . . , n − 1}. Consider also the elements u 1 , . . . , un−1 of On defined by

1 ··· i−1 i i + 1 ··· n
ui = ,
1 ··· i − 1 i + 1 i + 1 ··· n
for 1 i n − 1. Since POn = s0 , . . . , sn−1 , u 1 , . . . , un−1 (see [16]), it follows from Corollary 1.3 that
Corollary 1.11. POP n = s0 , . . . , sn−1 , u 1 , . . . , un−1 , g .
Also, as g n−1 u i g = u i +1 , for 1 i n − 2, s0 = g n−1 (s1 g )n−1 and si = g i −1 s1 g n−i +1 , for

1 i n − 1, we get
Corollary 1.12. POP n = s1 , u 1 , g .
Finally, since any generating set of POP n must clearly contain a permutation, a non-permutation
full transformation and a non-full transformation, we must have
Theorem 1.13. POP n has rank 3.
Next we observe that, given an orientation-reversing partial transformation s, we obtain

sh ∈ POP n , whence sh = x1 x2 · · · xk , for some x1 , x2 , . . . , xk ∈ {s1 , u 1 , g } and k ∈ N. Thus s = sh2 =
x1 x2 · · · xk h and so we may conclude the following:
Corollary 1.14. PORn = s1 , u 1 , g , h.
Let A be a set of generators of PORn . As for POP n , the set A must contain at least one non-
permutation full transformation and one non-full transformation. On the other hand, for n 3, the
group of units of PORn is the dihedral group D n , which has rank two. Hence we must also have two
permutations in A. This proves the next result.
Theorem 1.15. For n 3 the monoid PORn has rank 4.

2. Congruences associated to maximal subgroups
In this section we construct a family of congruences associated to maximal subgroups of a J-class

that satisfies certain conditions. As we will show in Section 3, this family provides a description for
the congruences of the monoids we want to consider.
We start with a simple technical lemma.
Lemma 2.1. Let S be a semigroup and let s, t , u ∈ S be such that s is regular and sHt. Then there exist
v 1 , v 2 ∈ S such that v 1 s = us, v 1 t = ut, sv 2 = su, t v 2 = tu, v 1 sR v 1 R v 1 t and sv 2 L v 2 Lt v 2 .
Proof. It is well known (e.g. see [19]) that ss = tt and s s = t t, for some inverses s of s and t
of t. Let v 1 = uss = utt and v 2 = s su = t tu. Then v 1 s = us, v 1 t = ut, sv 2 = su and t v 2 = tu. On
the other hand, as ss Rs, tt Rt, s sLs and t t Lt, we obtain v 1 = uss Rus = v 1 s, v 1 = utt Rut = v 1 t,
v 2 = s su Lsu = sv 2 and v 2 = t tu Ltu = t v 2 , as required. 2
Let S be a finite semigroup and let J be a J-class of S. Denote by B ( J ) the set of all elements
s ∈ S such that J J J s . It is clear that B ( J ) is an ideal of S. We associate to J a relation π J on S
defined by: for all s, t ∈ S, we have sπ J t if and only if
(a) s = t; or
(b) s, t ∈ B ( J ); or
(c) s, t ∈ J and sHt.
Lemma 2.2. (See [10].) The relation π J is a congruence on S.
Assume that J is regular and take a group H-class H 0 of J . Also, suppose that there exists a
mapping
ε : J −→ H 0 ,
s −→ s̃,
which satisfies the following property: given s, t ∈ J such that st ∈ J , there exist x, y ∈ H 0 such that
bHt implies = xs̃b̃,

sb (1)
aH s implies at = ãt̃ y . (2)
The existence of such a map for the monoid In (and for some of its submonoids) was showed by
the first author in [10].
To each congruence π on H 0 , we associate a relation ρπ on S defined by: given s, t ∈ S, we have
s ρπ t if and only if sπ J t and s, t ∈ J implies s̃π t̃ .
Theorem 2.3. The relation ρπ is a congruence on S.
Proof. First, observe that ρπ is an equivalence relation, since H and π are equivalence relations and
B ( J ) ∩ J = ∅. So, it remains to prove that ρπ is compatible with the multiplication.
Let s, t ∈ S be such that sρπ t and assume that s = t. Let u ∈ S. As sπ J t and π J is a congruence, we
have usπ J ut and su π J tu. In order to prove that usρπ ut, suppose that us, ut ∈ J . Then us, ut ∈ / B( J )
and, as B ( J ) is an ideal, s, t ∈
/ B ( J ). Since s = t, we must therefore have s, t ∈ J and sHt. Also we
get s̃π t̃. Now, by Lemma 2.1, there exists v 1 ∈ S such that v 1 s = us, v 1 t = ut and v 1 sR v 1 R v 1 t. Hence
v 1 s = x v˜1 s̃ and v
we have s, v 1 , v 1 s ∈ J . As t H s, it follows from condition (1) that 1 t = x v˜1 t̃, for some
x ∈ H 0 . Thus, as π is a congruence,
v 1 s = x v˜1 s̃π x v˜1 t̃ = v

=
us t
1t = u
and so usρπ ut. Similarly, we prove that su ρπ tu, as required. 2
3. On the congruences of OP n , POP n , ORn , PORI n and PORn
The goal of this section is to describe the congruences of the monoids OP n , POP n , ORn ,
PORI n and PORn . We will use a method that generalizes the process developed by the first au-
thor to describe the congruences of the monoid POPI n [10]. In fact, this new technique will also
comprise that case.
Although there are details that differ from one case to the other, we will present the proof in a
way that solves the problem simultaneously for all these monoids.
To prove our main result, Theorem 3.3, we need to fix some notation and recall some properties
of the monoids OP n , ORn , POPI n , PORI n , POP n and PORn presented in [4,5,10,14] or in this
paper.
First remember that OP n = On , g , POPI n = POI n , g , POP n = POn , g , ORn = On , g , h,
PORI n = POI n , g , h and PORn = POn , g , h.
Let us fix T ∈ {On , POI n , POn } and let M be either the monoid T , g or the monoid T , g , h.
Both T and M are regular monoids (moreover, if T = POI n then M and T are inverse monoids) and,
for the partial order J , the quotients T /J and M /J are chains. More precisely, for S ∈ { T , M }, we
have

S /J = J 0S <J J 1S <J · · · <J J nS , if T ∈ {POI n , POn },
and

S /J = J 1S <J · · · <J J nS , if T = On ,
where J kS denotes the J-class of S of the elements of rank k, for k suitably defined. Since S /J is a
chain, the sets I kS = {s ∈ S | | Im(s)| k}, with 0 k n, together with the empty set (if necessary),
constitute all the ideals of S (see [11]). Observe also that T is an aperiodic monoid (i.e. T has only
trivial subgroups); the H-classes of rank k of T , g have precisely k elements, for 1 k n; and the
H-classes of rank k of T , g , h have precisely 2k elements, for 3 k n, and k elements, for k = 1, 2.
For a J-class J kM of M (necessarily regular, since M is regular), with 1 k n, we want to find
a particular group H-class H k in J kM and a mapping ε : J kM → H k satisfying the conditions of Theo-
rem 2.3. Notice that B ( J kM ) = J 0M ∪ J 1M ∪ · · · ∪ J kM−1 or B ( J kM ) = J 1M ∪ · · · ∪ J kM−1 .
Given s ∈ PT n with Dom(s) = {i 1 < · · · < ik }, where 1 k n, define s ∈ Tn by, for every x ∈ X n ,
⎧
⎨ (i 1 )s, if 1 x i 1 ,
(x)s = (i j )s, if i j −1 < x i j and 2 j k,
⎩
(ik )s, if ik < x n.
It is clear that s and s have the same rank. Moreover:
(a) If s ∈ POn then s ∈ On ;

(b) If s ∈ POP n then s ∈ OP n ; and
(c) If s ∈ PORn then s ∈ ORn .
Fix 1 k n and consider the following elements of In (which are permutations of {1, . . . , k}):

1 ··· k 1 2 ··· k − 1 k 1 2 ··· k − 1 k
= , γ= and η= .
1 ··· k 2 3 ··· k 1 k k − 1 ··· 2 1
Let J k be the J-class of M of the elements of rank k. If T ∈ {POI n , POn }, take the elements:
ek = , gk = γ and hk = η.
When T = On , consider the following full transformations of X n :
ek = , gk = γ and hk = η.
Notice that hk gki = gkk−i hk , for 1 i k.

Denote by H k the H-class of M of the idempotent ek and observe that:
(a) If M = T , g , then H k is the cyclic group of order k generated by gk ; and

(b) If k 3 and M = T , g , h, then H k is the dihedral group of order 2k generated by gk and hk .
Let s ∈ J k . Suppose that {a1 < · · · < ak } is the transversal of the kernel of s formed by the minimum
element of each kernel class. Let Im(s) = {b1 < · · · < bk } and take the injective partial order-preserving
transformations

1 ··· k b1 · · · bk
σL = , σR =
a1 · · · ak 1 ··· k
and

a1 · · · ak 1 ··· k
σL = , σ R = .
1 ··· k b1 · · · bk
Define s L , s R , sL , sR ∈ T by:
(a) s L = σ L , s R = σ R , sL = σ L and sR = σ R , if T ∈ {POI n , POn };

(b) s L = σ L , s R = σ R , sL = σ L and sR = σ R , if T = On .
Clearly, s L Rek Ls R and s L sL = ek = sR s R .

Now let b p 1 , . . . , b pk ∈ {b1 , . . . , bk } be such that b p
= a
s, for 1
k. There exists i ∈ {0, . . . , k − 1}
such that b p i+1 < · · · < b pk < b p 1 < · · · < b p i if s is orientation-preserving, or b p i+1 > · · · > b pk >
b p 1 > · · · > b p i if s is orientation-reversing. It can be proved in a routine manner that s L ss R = gkk−i
if s is orientation-preserving, or s L ss R = gkk−i hk if s is orientation-reversing.
Let aq1 , . . . , aqk ∈ {a1 , . . . , ak } be such that b

= aq
s, for 1
k, and consider the following
b1 ··· bk
injective partial transformation: σ = aq1 ··· aqk . Define ŝ ∈ M by:
(a) ŝ = σ , if T ∈ {POI n , POn };

(b) ŝ = σ , if T = On .
Clearly, ŝ is an inverse of s and it is easy to show that s = s ŝsL s L ss R sR and s L s ŝsL = ek .

Next consider the mapping
ε : J k −→ H k ,
s −→ s̃ = s L ss R .
Observe that, given s, t ∈ J k such that sHt, we have s L = t L , s R = t R , sL = t L and sR = t R . Moreover,

since s ŝ and t t̂ are idempotents of J k with the same kernel and the same image, s ŝ = t t̂.
Lemma 3.1. Let s, t ∈ M be such that s, t , st ∈ J k . Then there exist

1 ,
2 ∈ {0, 1, k − 1} such that, for all
a, b ∈ M,
= g 1 s̃b̃;
(1) bRt implies sb

k
(2) aLs implies at = ãt̃ gk 2 .

Proof. Suppose that s reverses the orientation and t preserves the orientation. Let

P 1 · · · P i P i +1 · · · P k P k+1 Q 1 · · · Q j Q j +1 · · · Q k Q k+1
s= and t = ,
s1 · · · s i si +1 · · · sk s1 t1 ··· tj t j +1 · · · tk t1
possibly with P k+1 = ∅ or Q k+1 = ∅. Then Im(s) = {si +1 > · · · > sk > s1 > · · · > si } and Im(t ) =
k− j
{t j +1 < · · · < tk < t 1 < · · · < t j }, for some 0 i , j k − 1. Hence s̃ = gkk−i hk and t̃ = gk . As
s, t , st ∈ J k , then Im(s) is a transversal of Ker(t ) and we have two possible situations:
(a) If si ∈ Q 1 , . . . , s1 ∈ Q i , sk ∈ Q i +1 , . . . , si +1 ∈ Q k , then

P 1 · · · P i P i +1 · · · Pk P k+1
st = .
ti · · · t1 tk · · · t i +1 ti
Since ( P i − j )st = {t j +1 } if 0 j i − 1 and ( P i − j +k )st = {t j +1 } if i j k − 1, it follows that

k −i + j

st = gk hk , whence st = s̃t̃.
(b) If si ∈ Q 2 , . . . , s1 ∈ Q i +1 , sk ∈ Q i +2 , . . . , si +1 ∈ Q k+1 , then

P1 · · · P i +1 P i +2 · · · Pk P k+1
st =
t i +1 · · · t1 tk · · · t i +2 t i +1
and so ( P i − j +1 )st = {t j +1 } if 0 j i and ( P k+i − j +1 )st = {t j +1 } if i + 1 j k − 1.

k+ j −i −1
Therefore, in both cases, st = gk st = gkk−1 s̃t̃ = s̃t̃ gk .
hk and so
If s preserves the orientation or t reverses the orientation, it is a routine matter to show that, in
the situation analogous to (a), we always have st = s̃t̃. On the other hand, the situation analogous
to (b) can be summarized by Table 1:
Table 1
s t
1
2
orientation-preserving orientation-preserving 1 1
orientation-reversing orientation-preserving k−1 1
orientation-preserving orientation-reversing 1 k−1
orientation-reversing orientation-reversing k−1 k−1
Finally, as R-related elements have the same kernel and L-related elements have the same image,
it is clear that
1 does not depend of the element of the R-class of t (s fixed) and
2 does not depend
of the element of the L-class of s (t fixed), as required. 2
The next proposition follows from this lemma and Theorem 2.3.
Proposition 3.2. Let k ∈ {1, . . . , n} and let π be a congruence on H k . Then ρπ is a congruence on M.
Notice that, if π is the universal congruence of H k , then the relation ρπ is the congruence π J k
of M. On the other hand, if π is the identity congruence of H k , then the relation ρπ is the Rees con-
gruence of M associated to the ideal I kM−1 . Thus, for k = 1, the relation ρπ is the identity congruence
of M and, for 2 k n, there exist s ∈ B ( J k ) = I kM−1 and t ∈ J k , whence (s, t ) ∈
/ ρπ and so ρπ is not
the universal congruence of M.
At this point, we can state our main result.
Theorem 3.3. The congruences of M ∈ {OP n , POPI n , POP n , ORn , PORI n , PORn } are exactly the
congruences ρπ , where π is a congruence on H k , for k ∈ {1, . . . , n}, and the universal congruence.
Denote by Con( S ) the lattice of the congruences on a semigroup S.

Recall that Con( T ) is formed only by the Rees congruences of T ∈ {On , POI n , POn }
(see [1,11,15]).
On the other hand, it is clear that Con(OP 1 ) = Con(OR1 ) = {1} and Con( M ) = {1, ω} if
M ∈ {POPI 1 , PORI 1 , POP 1 , POR1 }.
To prove Theorem 3.3 we start by establishing some auxiliary results.
Let c 1 , . . . , cn ∈ Tn be the constant mappings such that Im(c i ) = {i }, for all 1 i n. Observe that,
if s, t ∈ Tn are such that c i s = c i t, for all 1 i n, then s = t.
The version of this property for partial transformation is the following: let c 1 , . . . , cn ∈ PT n be
the n partial identities of rank one such that Dom(c i ) = Im(c i ) = {i }, for all 1 i n. Then, given
s, t ∈ PT n such that c i s = c i t, for all 1 i n, we also have s = t.
In what follows, c 1 , . . . , cn denote the constant mappings of Tn if T = On , and the partial identities
of rank one of PT n if T ∈ {POn , POI n }. In both cases c 1 , . . . , cn ∈ T . Moreover, for all 1 i n and
s ∈ M, we have c i s ∈ T . In fact, c i s is either a constant map of image {(i )s} or the empty map.
Let ρ be a congruence on M and consider ρ = ρ ∩ ( T × T ). Then ρ is a Rees congruence of T and
so ρ = ρ I T , for some 1 k n + 1.
k−1
This notation will be used in the sequel.
Lemma 3.4. If k = 1 then ρ = 1.
Proof. First notice that k = 1 if and only if ρ = 1. Let s, t ∈ M be such that sρ t. Then c i sρ c i t and,
since c i s, c i t ∈ T , we have c i sρ c i t, whence c i s = c i t, for all 1 i n. Thus s = t and so ρ = 1, as
required. 2
From now on, consider k 2.

Lemma 3.5. ρ I M ⊆ ρ .
k−1
Proof. It suffices to show that sρ t, for all s, t ∈ I kM−1 . Let s, t ∈ I kM−1 .

(1) If s, t ∈ T then s, t ∈ I kT−1 and so sρ t, whence sρ t.
(2) If s ∈ M \ T and t ∈ T then, by Corollary 1.4, there exist i ∈ {0, 1, . . . , n − 1}, j ∈ {0, 1} and u ∈ T
such that s = g i uh j . Since sJu, we get g n−i sh2− j = u ∈ I kT−1 . As c 1 ∈ I kT−1 , we have u ρ c 1 , whence
u ρ c 1 . Then s = g i uh j ρ g i c 1 h j . On the other hand, since g i c 1 h j ∈ I kT−1 (in fact, g i c 1 h j is a constant
map), we also obtain g i c 1 h j ρ t. Hence g i c 1 h j ρ t and so sρ t.
(3) Finally, suppose that s, t ∈ M \ T . By Corollary 1.4, there exist i 1 , i 2 ∈ {0, 1, . . . , n − 1},
j 1 , j 2 ∈ {0, 1} and u , v ∈ T such that s = g i 1 uh j 1 and t = g i 2 vh j 2 . Since sρ I M t, it follows that
k−1
u ρ I M g n−i 1 +i 2 vh2− j 1 + j 2 . If g n−i 1 +i 2 vh2− j 1 + j 2 ∈ M \ T then, by (2), we have u ρ g n−i 1 +i 2 vh2− j 1 + j 2 ,

k−1
otherwise the same holds by (1). Hence s = g i 1 uh j 1 ρ g i 2 vh j 2 = t, as required. 2
Lemma 3.6. Let s, t ∈ M be such that sρ t. Then | Im(s)| k if and only if | Im(t )| k.
Proof. It suffices to show that | Im(s)| k implies | Im(t )| k. So, suppose that | Im(s)| k.
(1) If s, t ∈ T then sρ t. Since s ∈
/ I kT−1 , we have s = t, whence | Im(t )| k.
(2) Next consider s ∈ T and t ∈ M \ T . If t ∈ I kM−1 then t ρ I M c 1 and so t ρ c 1 , by Lemma 3.5. Hence
k−1
sρ c 1 . By (1), we obtain s = c 1 and this is a contradiction, for c 1 has rank one. Therefore | Im(t )| k.
(3) Finally, if s ∈ M \ T , by Corollary 1.4, there exist i ∈ {0, 1, . . . , n − 1}, j ∈ {0, 1} and u ∈ T such
that s = g i uh j . Then g n−i sh2− j = u ∈ T and u ρ g n−i th2− j . Since u Js and | Im(s)| k, by (1) or (2), we
deduce that | Im(t )| = | Im( g n−i th2− j )| k, as required. 2
Lemma 3.7. Let s ∈ M. Then there exists an inverse s ∈ M of s such that s s ∈ E ( T ).
Proof. Let s ∈ M. By Corollary 1.4, there exist i ∈ {0, 1, . . . , n − 1}, j ∈ {0, 1} and u ∈ T such that
s = g i uh j . Let u ∈ T be an inverse of u and consider s = h2− j u g n−i . Then s is an inverse of s and
s s = h2− j u g n−i g i uh j = h2− j u uh j ∈ E ( T ), as required. 2
Lemma 3.8. Let t ∈ M and let D be a transversal of Ker(t ). Then there exists an inverse t ∈ M of t such that
Im(t ) = D.
Proof. Consider the injective partial transformation ξ defined by Dom(ξ ) = Im(t ) and, for all x ∈
Dom(ξ ), (x)ξ is the unique element in D ∩ (x)t −1 . Define t by:
(a) t = ξ , if T ∈ {POI n , POn };

(b) t = ξ , if T = On .
It is a routine matter to show that t ∈ M and t is an inverse of t with image D, as required. 2
Lemma 3.9. Let s, t ∈ M be such that sρ t and | Im(s)| k. Then sHt.
Proof. Let s and t be inverses of s and t, respectively, such that s s, t t ∈ E ( T ), by Lemma 3.7. As sρ t,
then s st t ρ s tt t = s t ρ s s. Since s s, s st t ∈ T , we obtain s sρ s st t. Now, as | Im(s s)| = | Im(s)| k,
it follows that s s = s st t and so s = (st )t. Similarly, as | Im(t )| k, by Lemma 3.6, we also have
t = (ts )s, whence sLt.
Next let D be a transversal of Ker(t ). By Lemma 3.8, there exists an inverse t of t such that
Im(t ) = D. As | Im(t t )| = | Im(t )| k and t t ρ t s, by the argument above, it follows that t t Lt s. Since
t t Lt Ls, we get t sLs and so D = Im(t ) contains a transversal of Ker(s). As s and t are L-related,
the transversals of Ker(s) and Ker(t ) have the same number of elements, whence D must also be a
transversal of Ker(s). We conclude that any transversal of Ker(t ) is a transversal of Ker(s) and vice-
versa. Thus Ker(s) = Ker(t ) and so sRt. Therefore sHt, as required. 2
Lemma 3.10. Let s, t ∈ M be such that s = t and sHt. Then there exists z ∈ T such that | Im( zs)| = | Im( zt )| =
| Im(s)| − 1 and (zs, zt ) ∈
/ L.
Proof. First, notice that, s and t have the same image and the same kernel. Next let D be a transversal
of Ker(s). As s = t, there exists i ∈ D such that (i )s = (i )t. Let τ be the partial identity with domain
D \ {i } and define z by:
(a) z = τ , if T ∈ {POI n , POn };

(b) z = τ , if T = On .
Clearly, z ∈ T and | Im( zs)| = | Im(s)| − 1 = | Im(t )| − 1 = | Im( zt )|. On the other hand, as Im( zs) =
Im(s) \ {(i )s} and Im( zt ) = Im(t ) \ {(i )t } = Im(s) \ {(i )t } and (i )s = (i )t, we get Im( zs) = Im( zt ) and so
(zs, zt ) ∈
/ L, as required. 2
Finally, we can prove Theorem 3.3.
Proof of Theorem 3.3. Let ρ be a congruence of M. Let 1 k n + 1 be such that ρ = ρ ∩ ( T × T ) =

ρ I T . By Lemma 3.4, we have ρ = 1, for k = 1. Thus we can consider k 2. On the other hand,
k−1
ρ I M ⊆ ρ , by Lemma 3.5, and so if k = n + 1 the relation ρ is the universal congruence on M. Hence,
k−1
in what follows, we can also assume k n.
Let s, t ∈ M be such that sρ t and | Im(s)| > k. By Lemma 3.9, we have sHt. Suppose that s = t
and let m = | Im(s)|. By Lemma 3.10, there exists z ∈ T such that | Im( zs)| = | Im( zt )| = m − 1 and
(zs, zt ) ∈
/ L. On the other hand, as m − 1 k and zsρ zt, by Lemma 3.9, we obtain (zs, zt ) ∈ H, which
is a contradiction. Thus s = t.
Now let π be the congruence of H k induced by ρ , i.e. π = ρ ∩ ( H k × H k ). As ρ I M ⊆ ρ , to prove
k −1
that ρ = ρπ it remains to show that, for all s, t ∈ J kM , we have sρ t if and only if sρπ t. Take s, t ∈ J kM .
First, suppose that sρ t. By Lemma 3.9, we conclude that sHt and so sπ J k t. Moreover, s L = t L and
s R = t R , whence s̃ = s L ss R ρ s L ts R = t L tt R = t̃ and so s̃π t̃. Thus sρπ t. Conversely, assume that sρπ t.
Then sHt and s̃π t̃. Hence s L = t L , s R = t R , sL = t L , sR = t R and s̃ρ t̃. Next consider the inverses ŝ and
t̂ of s and t, respectively. Then s ŝ = t t̂ and it follows
s = s ŝsL (s L ss R )sR ρ s ŝsL (s L ts R )sR = t t̂t L (t L tt R )t R = t ,
as required. 2
Let k ∈ {1, . . . , n}. Let π1 , π2 ∈ Con( H k ). If π1 ⊂ π2 , it is easy to show that ρπ1 ⊂ ρπ2 . On the other
hand, it is clear that given k1 , k2 ∈ {1, . . . , n} such that k1 < k2 , π1 ∈ Con( H k1 ) and π2 ∈ Con( H k2 ), we
have ρπ1 ⊂ ρπ2 .
Denote by Dk the lattice of the congruences of the group H k , for 1 k n. By Theorem 3.3, we
deduce the following description of Con( M ).
Theorem 3.11. The lattice of the congruences of the monoid M is isomorphic to the ordinal sum of lattices
D1 ⊕ D2 ⊕ · · · ⊕ Dn ⊕ D1 .
Example 3.12. Consider the monoid POR6 . Applying the last result, we get the following Hasse
diagram for Con(POR6 ):
References
[1] A.Ya. Aı̆zenštat, Homomorphisms of semigroups of endomorphisms of ordered sets, Uch. Zap. Leningr. Gos. Pedagog.
Inst. 238 (1962) 38–48 (in Russian).
[2] J. Almeida, M.V. Volkov, The gap between partial and full, Internat. J. Algebra Comput. 8 (3) (1998) 399–430.
[3] R.E. Arthur, N. Ruškuc, Presentations for two extensions of the monoid of order-preserving mappings on a finite chain,
Southeast Asian Bull. Math. 24 (2000) 1–7.
[4] P.M. Catarino, Monoids of orientation-preserving transformations of a finite chain and their presentations, in: J.M. Howie,
N. Ruškuc (Eds.), Semigroups and Applications, World Scientific, 1998, pp. 39–46.
[5] P.M. Catarino, P.M. Higgins, The monoid of orientation-preserving mappings on a chain, Semigroup Forum 58 (1999) 190–
206.
[6] D.F. Cowan, N.R. Reilly, Partial cross-sections of symmetric inverse semigroups, Internat. J. Algebra Comput. 5 (1995) 259–
287.
[7] D. Dummit, R. Foote, Abstract Algebra, second ed., John Wiley & Sons, Inc., 1999.
[8] V.H. Fernandes, Semigroups of order-preserving mappings on a finite chain: A new class of divisors, Semigroup Forum 54
(1997) 230–236.
[9] V.H. Fernandes, Normally ordered inverse semigoups, Semigroup Forum 58 (1998) 418–433.
[10] V.H. Fernandes, The monoid of all injective orientation preserving partial transformations on a finite chain, Comm. Alge-
bra 28 (2000) 3401–3426.
[11] V.H. Fernandes, The monoid of all injective order preserving partial transformations on a finite chain, Semigroup Forum 62
(2001) 178–204.
[12] V.H. Fernandes, A division theorem for the pseudovariety generated by semigroups of orientation preserving transforma-
tions on a finite chain, Comm. Algebra 29 (2001) 451–456.
[13] V.H. Fernandes, Semigroups of order-preserving mappings on a finite chain: Another class of divisors, Izv. Vyssh. Uchebn.
Zaved. Mat. 3 (478) (2002) 51–59 (in Russian).
[14] V.H. Fernandes, G.M.S. Gomes, M.M. Jesus, Presentations for some monoids of injective partial transformations on a finite
chain, Southeast Asian Bull. Math. 28 (2004) 903–918.
[15] V.H. Fernandes, G.M.S. Gomes, M.M. Jesus, Congruences on monoids of order-preserving or order-reversing transformations
on a finite chain, Glasg. Math. J. 47 (2005) 413–424.
[16] G.M.S. Gomes, J.M. Howie, On the ranks of certain semigroups of order-preserving transformations, Semigroup Forum 45
(1992) 272–282.
[17] P.M. Higgins, Divisors of semigroups of order-preserving mappings on a finite chain, Internat. J. Algebra Comput. 5 (1995)
725–742.
[18] J.M. Howie, Product of idempotents in certain semigroups of transformations, Proc. Edinb. Math. Soc. 17 (1971) 223–236.
[19] J.M. Howie, Fundamentals of Semigroup Theory, Oxford University Press, 1995.
[20] T. Lavers, A. Solomon, The endomorphisms of a finite chain form a rees congruence semigroup, Semigroup Forum 59 (2)
(1999) 167–170.
[21] D. McAlister, Semigroups generated by a group and an idempotent, Comm. Algebra 26 (1998) 515–547.
[22] V.B. Repnitskiı̆, M.V. Volkov, The finite basis problem for the pseudovariety O, Proc. Roy. Soc. Edinburgh Sect. A 128 (3)
(1998) 661–669.
[23] A. Vernitskii, M.V. Volkov, A proof and generalisation of Higgins’ division theorem for semigroups of order-preserving
mappings, Izv. Vuzov. Mat. (1) (1995) 38–44.
Journal of Algebra
Cohomology over Fiber products of local rings ✩

W. Frank Moore
Cornell University, Mathematics, 587 Malott Hall, 14850, Ithaca, NY, United States
Article history: Let S and T be local rings with common residue field k, let R be
Received 26 April 2007 the fiber product S ×k T , and let M be an S-module. The Poincaré
Communicated by Michel Van Den Bergh series P M R
of M has been expressed in terms of P M S
, P kS and P kT
by Kostrikin and Shafarevich, and by Dress and Krämer. Here, an
Keywords:
explicit minimal resolution, as well as theorems on the structure of
Yoneda algebra
Fiber product Ext R (k, k) and Ext R ( M , k) are given that illuminate these equalities.
Minimal free resolution Structure theorems for the cohomology modules of fiber products
Poincaré series of modules are also given. As an application of these results, we
compute the depth of cohomology modules over a fiber product.
© 2008 Elsevier Inc. All rights reserved.
0. Introduction
In homological investigations one often has information on properties of a module over a certain
ring, and wants to extract information on its properties over a different ring. In this paper we consider
the following situation: S → k ← T are surjective homomorphisms of rings, k is a field, R is the fiber
product S ×k T , and M an S-module. We further assume that S and T are either local rings with
common residue field k, or connected graded k-algebras.
The starting point of this paper is the construction of an explicit minimal free resolution of M,
viewed as an R-module, from minimal resolutions of M and k over S and k over T . This is carried
out in Section 1. The structure of the R-free resolution allows us to obtain precise information on the
multiplicative structure of cohomology over R, given by composition products. Some of the results
obtained in this work have been proved in the graded case by use of standard resolutions. However,
no similar approach can be used in the local case.
The symbol denotes a coproduct, also known as a free product, of k-algebras.
✩
This research was partially supported through NSF grant # DMS-0201904, and is part of the author’s PhD thesis.
E-mail address: frankmoore@math.cornell.edu.
doi:10.1016/j.jalgebra.2008.10.015
W.F. Moore / Journal of Algebra 321 (2009) 758–773 759
Theorem A. The canonical homomorphism of graded k-algebras
Ext S (k, k) Ext T (k, k) → Ext S ×k T (k, k)
defined by the universal property of coproducts of k-algebras is bijective. For every S-module M, the canonical
homomorphisms of graded left Ext R (k, k)-modules
Ext R (k, k) ⊗Ext S (k,k) Ext S ( M , k) → Ext R ( M , k)
defined by the multiplication map, is bijective.
S
The isomorphisms imply relations between the Poincaré series P M R
(t ) of M over R to P M (t ), P kS (t )
and P k (t ); this relationship was proved for M = k by Kostrikin and Shafarevich [6], and by Dress and
T
Krämer [3, Theorem 1] in the present setting. In [7], Polishchuk and Positselski proved the preceding
theorem, when S and T are connected k-algebras by using cobar constructions.
By combining Theorem A with an observation of Dress and Krämer concerning second syzygy
modules over fiber products, we obtain the following corollary.
Corollary B. Let L be an R-module. Then the second syzygy Ω 2 L decomposes as Ω 2 L = M ⊕ N where M and
N are S and T -modules, respectively. Furthermore, one has an exact sequence of graded left R-modules

0 → Σ −2 R ⊗S Ext S ( M , k) ⊕ Σ −2 R ⊗T Ext T ( N , k) → L → L/L2 → 0,
where R = Ext R (k, k), S = Ext S (k, k), T = Ext T (k, k) and L = Ext R ( L , k).
Theorem A shows that Ext− (k, k), as a functor in the ring argument, transforms products into co-
products. We prove that Ext R (−, k), as a functor from R-modules to Ext R (k, k)-modules, has a similar
property. In the graded setting, this was shown by Polishchuk and Positselski [7]. Their methods do
not extend to the local case, where even the equality of Poincaré series given is new. To give the
result, let M, N, and V be S, T , and k-modules respectively, that satisfy M /p M ∼=V ∼= N /q M, where
p, q are the kernels of the surjections S → k, T → k, respectively. Define an R-module M × V N by the
exact sequence
ι μ−ν
0 → M ×V N −
→ M × N −−−→ V → 0. (1)
Theorem C. In the notation above, the short exact sequence (1) induces a short exact sequence of graded left
Ext R (k, k)-modules
μ∗
−ν ∗ ι∗
0 → Ext R ( V , k) −−−−→ Ext R ( M , k) ⊕ Ext R ( N , k) −
→ Ext R ( M × V N , k) → 0.
In Section 4 we study the depth of Ext R ( M , k) over Ext R (k, k) for an R-module M. The notion of
depth was used in [4] to study the homotopy Lie algebras of simply connected CW complexes, and of
local rings. More recently, Avramov and Veliche [1] have shown that small depth of Ext R ( M , k) over
Ext R (k, k) is responsible for significant complications in the structure of the stable cohomology of M
over R. For cohomology modules of R-modules, large depth is impossible:
Theorem D. Let M be an R-module. Then one has an inequality
depthExt R (k,k) Ext R ( M , k) 1,
with equality if M is an S or T -module. In particular, one has depth Ext R (k, k) = 1.

760 W.F. Moore / Journal of Algebra 321 (2009) 758–773
1. Resolutions over a fiber product
In this section, we set notation and define a complex that will be used throughout the article.
Let k be a field.

1.1. A k-algebra A is graded if there is a decomposition of A = i ∈Z A i as k-vector spaces, and for all
i , j ∈ Z, one has A i A j ⊆ A i + j . We use both upper and lower indexed graded objects and adopt the
notation A i = A −i . One says that A is connected if A 0 = k and A i = 0 for i < 0 (or equivalently, A i = 0
for i > 0).

1.2. A left module M over a graded k-algebra A is graded if there is a decomposition M = i ∈Z M i
as k-vector spaces, and for all i , j ∈ Z, one has A i M j ⊆ M i + j . For a homogeneous element m ∈ M i , we
denote its degree by |m|.
For graded left A-modules M and N and an integer j, the degree j homomorphisms from M to N
form the abelian group

Homgr A ( M , N ) j = φ : M → N φ( M i ) ⊆ N i − j for each i ⊆ Hom A ( M , N ).
The collection of graded homomorphisms from M to N form the graded group

Homgr A ( M , N ) = Homgr A ( M , N ) j .
j ∈Z
1.3. In the sequel we assume that we are in one of the following situations:
(1) A is a commutative noetherian local ring, m denotes its unique maximal ideal, and k = A /m its
residue field.
(2) A is a non-negatively graded, connected, degree-wise finite k-algebra, and m denotes the unique
graded maximal ideal A + ; in this case, the differentials in a complex of graded R-modules pre-
serve degrees.
We say that an A-module M is convenient provided it is finitely generated in case (1), and if it is
bounded below, degree-wise finite, and graded in case (2).
One says that a complex of free A-modules F is minimal if each module in the complex is con-
venient, and ∂( F ) ⊆ m F . When such a resolution exists, it is unique up to isomorphism. If M is a
convenient A-module, then M has a minimal free resolution, and such a resolution is unique.
1.4. In case 1.3(2), the A-module ExtiA ( M , k) can be computed by taking a minimal graded free reso-
lution F of M over A, and computing H i (Homgr A ( M , k)). This provides ExtiA ( M , k) the structure of a
graded A-module for each i; we denote the jth graded piece of ExtiA ( M , k) by ExtiA ( M , k) j .

Definition 1.5. Let A be a ring and X a graded set, X = n0 Xn . We let A
X denote the graded free
left A-module with basis Xn in degree n, and set Xn = 0 when Xn = ∅. We call A X a graded based
A
module over A with basis X. Homomorphisms of based modules are identified with their matrices in
the chosen bases.
For a based module A X, we identify A ⊗ A A X and A X by means of the canonical isomorphism.
We
use A [XY] to denote the graded based A-module A X ⊗ A A Y with graded basis XY = n [XY]n , where
[XY]n is the set of symbols
{xy | x ∈ Xi , y ∈ Y j , i + j = n}.
Notation 1.6. We consider a diagram of homomorphisms of rings,
τ
S ×k T T
σ πT (1.6.1)
S k
πS
where π S and πT are surjective, and R := S ×k T is the fiber product:

S ×k T = (s, t ) ∈ S × T : π S (s) = πT (t ) .
If S and T are as in either case of 1.3, then so is R, and its maximal ideal is m = p ⊕ q; we thus
identify p and q with subsets of R. Every S-module is considered an R-module via σ , and similarly
for T -modules.
Construction 1.7. Let M be an S-module. Let P → M and E → k be minimal free resolutions of M,

respectively k, over S, and let F → k be a free resolution of k over T such that E 0 = S and F 0 = T .
Choose bases P, E, and F of the graded modules P , E, and F over S, S and T , respectively, so that
E0 = {1 S } and F0 = {1 T }. Consider the elements of P, E1 and F1 as letters of an alphabet. The
degree of a word in this alphabet is defined to be the sum of the degrees of its letters. The length of
a word w is defined to be the number of letters in w.
Let G be the set of all words of the form
{ f 1 e 2 f 3 · · · e 2l−2 f 2l−1 p 2l } and {e 1 f 2 e 3 · · · e 2l−1 f 2l p 2l+1 },
where e i , f i and p i range over E1 , F1 and P respectively, and l 0. Form the free graded R-module
G = R G.
Every word w ∈ G has the form xw for some letter x and a (possibly empty) word w . Assume
that one has ∂( E ) ⊆ p E, ∂( P ) ⊆ p P , ∂( F ) ⊆ q F , and set
⎧
⎪ P
⎨ ∂ (x) for x ∈ P,
∂ ( w ) = ∂ (x) w for x ∈ E,
G E
⎪
⎩ F
∂ (x) w for x ∈ F,
and extend ∂ G to an endomorphism of G by R-linearity. Set ∂iG = ∂ G |G i . We remark that a matrix ϕ

with entries in p defines a homomorphism S ϕ of free S-modules, as well as a homomorphism R ϕ of
free R-modules.
Theorem 1.8. The maps of free modules ∂iG defined in Construction 1.7 give a minimal free resolution
∂iG ∂1G
G : · · · → G i −→ G i −1 → · · · → G 1 −→ G 0 → 0,
of the R-module M.
Remark 1.9. The first few degrees of the complex G in Construction 1.7 look as follows. Note that
each map in the diagram acts on the leftmost letter of a word.
R
P3
R
P2
R
[F1 P2 ]
R
P1
R
[E1 F1 P1 ]
R
[F1 P1 ]
R
[F2 P1 ]
R
P0
R
[E2 F1 P0 ]
R
[E1 F1 P0 ]
R
[F1 E1 F1 P0 ]
R
[F1 P0 ]
R
[E1 F2 P0 ]
R
[F2 P0 ]
R
[F3 P0 ]
To explain the notation used, let R [E2 G] denote the R-linear span of words whose first letter is
in Ei for some i 2, see Definition 1.5. Let R [F1 E1 G] denote the span of words starting with a letter
from F1 , followed by a letter from E. Symbols such as R [F2 G], R [E1 F1 G], etc. are defined similarly.
Proof of Theorem 1.8. To show that G is a complex, let w be a word of degree i, with i 2. Suppose
w = xy w where w is a word, x is a letter of degree 1 and y is an arbitrary letter. For x ∈ E1 and
y ∈ F one has

∂ 2 ( w ) = ∂ ∂ E (x) y w ∈ ∂(p y w ) = p∂( y w ) = p∂ F ( y ) w ⊆ pq w = 0.
The cases with x ∈ F1 and y ∈ P, and with x ∈ F1 and y ∈ E are similar. If w = xw where x is a
letter of degree greater than or equal to 2, then ∂ 2 ( w ) = 0, since R P, R E, and R F are complexes of
R-modules, and hence one has ∂ 2 ( w ) = 0.
The proof that H (G ) = M, proceeds in several steps. First, we describe a decomposition of an
arbitrary cycle into a sum of cycles of a special type. Then we show that each summand is in fact an
element of mG. Finally, we prove that each summand is a boundary through a computation that uses
the special form of these cycles and the exactness of the complexes E, F and P
Let a be an element of G i , with i 1. It has a unique expression (cf. Remark 1.9 for notation)
a = (x + x ) + ( y + y ) + ( z + z ) where
x ∈ R [E2 G], y ∈ R [F2 G], z ∈ R P,
x ∈ R [F1 E1 G], y ∈ R [E1 F1 G], z ∈ R [F1 P].
Notice that one has
∂(x + x ) ∈ R [E1 G], ∂( y + y ) ∈ R [F1 G], ∂(z + z ) ∈ R P.
Thus, ∂(a) = 0 implies that each one of x + x , y + y and z + z is a cycle. Next, we show each is a
boundary, by giving details for x + x ; the other cases are similar.
Since one has p ∩ q = 0 in R, and G i −1 is a free R-module, it follows that p(G i −1 ) ∩ q(G i −1 ) = 0.
Therefore, ∂(x + x ) = 0 implies that x and x are cycles as well. Let l( w ) denote the leftmost letter in
the word w. We may express x according to the decomposition

R R
[E2 G]i = [Ej w ].
2 ji
w ∈ Gi − j
l( w ) ∈ F
If w is basis element of degree i − j with l( w ) ∈ F and 2 j i, then one has ∂( R [Ej w ]) ⊆ R [Ej−1 w ].
Hence, each component of x in the decomposition above is a cycle. For similar reasons the compo-
nents of x in R [F1 E1 G]

R R
[F1 E1 G] = [F1 Ej−1 w ]
2 ji
w ∈ Gi − j
l( w ) ∈ F
are cycles. Therefore, it is enough to show that every cycle of the form

x= re e w ∈ R [Ej w ] or
e ∈E j

x = r f e f e w ∈ R [F1 Ej−1 w ]
f ∈F1 e ∈E j −1
where w is a fixed word, and re , r f e are in R, is a boundary.

We give details for x, the other case is similar. We first show that re ∈ m for each e ∈ E j . Indeed,
there is a commutative diagram of R-modules
··· R
[E3 w ] R
[E2 w ] R
[E1 w ] Rw 0

σ3
σ2
σ1
σ0 (1.9.1)
··· E3 E2 E1 E0 0

where the vertical maps send e ∈E j r e e w to e ∈E j σ (re )e. The image of x in E is a cycle, and hence
a boundary, of E. As E is minimal, the claim follows.
Suppose t = ∂( f ) for f ∈ F1 . Then te w = ∂( f e w ) is a boundary of G. As F is a resolution of
k over T , the images of F1 form a minimal generating set for q. Hence qe w consists entirely of
boundaries.
The claims above show it suffices to prove the theorem when the coefficients are in p. In dia-
gram (1.9.1), we may also define a morphism of complexes γ : p E → p R [E w ] by sending e∈E j se e to

viewing se ∈ p ⊂ R. Note that γ
e ∈E j se e w, and σ |p R [E w ] are inverses of one another.
Suppose that x ∈ p [E w ] is a cycle. Then
R
σ (x) is a cycle in E. Hence there exists u so that ∂ E (u ) =

σ (x). One then has

∂ γ(u ) = γ
∂ E (u ) = γ σ (x) = x. 2
Two special cases of the theorem are used in Section 3.
Example 1.10. When M = k, we can take P = E. Let D be the resolution given by Theorem 1.8. Since
P0 = {1}, we can also replace all basis elements of the form w1 with a basis element w of the same
degree, and set ∂ D (1 R ) = 0. Therefore, in low degrees, D has the form
R
E3
R
E2
R
[F1 E2 ]
R
E1
R
[E1 F1 E1 ]
R
[F1 E1 ]
R
[F2 E1 ]
R
R
[E2 F1 ]
R
[E1 F1 ]
R
[F1 E1 F1 ]
R
F1
R
[E1 F2 ]
R
F2
R
F3
Example 1.11. When M = T , applying the theorem with the roles of S and T reversed, one has P i = 0
for i = 0, and P0 = {1}. Let C be the resolution given by Theorem 1.8. Letting w denote the basis
element w1 as above, we see that C is given by the top half of the diagram in Example 1.10.
2. Cohomology of coproducts of graded connected algebras
In this section, we remind the reader of the construction of coproducts of connected algebras, and
collect some facts regarding their depth and Hilbert functions. The results in this section are applied
to the Ext algebras of rings as in 1.3 in Section 4.
2.1. If B and C are graded connected k-algebras, the coproduct of B and C in this category is the free
product of B and C , denoted B C , and can be described as follows. A k-basis for ( B C )n consists of
all elements of the form x1 ⊗ · · · ⊗ x p with |x1 | + · · · + |x p | = n, and the factors xi alternate between
elements of the homogeneous bases of B + and C + . Multiplication in B C is given by

v ⊗ · · · ⊗ xy ⊗ · · · ⊗ w for x, y ∈ B or x, y ∈ C ;
( v ⊗ · · · ⊗ x)( y ⊗ · · · ⊗ w ) =
v ⊗ ··· ⊗ x ⊗ y ⊗ ··· ⊗ w otherwise.
We say that a tensor product of the form x1 ⊗ x2 ⊗ · · · ⊗ x p has length p, and that it starts
in B i (respectively C i ) if x1 is in B i (respectively C i ); the expression that the product ends in B i
(respectively C i ) has a similar meaning. In the sequel, we set A = B C , and we consider B and C as
subalgebras of A. Thus, A is free as a right graded B-module with basis given by those elements in
the basis of A described above that end in C .
2.2. For a graded left B-module M, A ⊗ B M is a graded left A-module. A k-basis for ( A ⊗ B M )n consists
of all elements of the form x1 ⊗ · · · ⊗ x p −1 ⊗ m p such that |x1 | + · · · + |x p −1 | + |m p | = n, p 1, and
the xi alternate between being elements of the homogeneous bases of B + and C + , with x p −1 in C .
The action of A on A ⊗ B M is given by
⎧
⎨ v ⊗ · · · ⊗ xy ⊗ · · · ⊗ w for x, y ∈ B or x, y ∈ C ;
( v ⊗ · · · ⊗ x)( y ⊗ · · · ⊗ w ) = v ⊗ · · · ⊗ xw for x ∈ B and y ∈ M;
⎩
v ⊗ · · · ⊗ x ⊗ y ⊗ · · · ⊗ w otherwise.
We say that a tensor of the form x1 ⊗ x2 ⊗ · · · ⊗ x p has length p, and starts in B i (respectively C i ) if
x1 is in B i (respectively C i ). We identify the span of those tensors of length one with M, and hence
M is a left B-submodule of A ⊗ B M.
Put L = A ⊗ B M. One then sees that every homogeneous element l ∈ L can be written in the form
l = l B + lC + l M (2.2.1)
where the terms in l B and l C start in B and C , respectively, and l M is in M.
2.3. Let B and C be graded connected k-algebras, and choose minimal homogeneous sets of generators
X of B and Y of C as k-algebras. One then has resolutions of k of the form
∂2B ∂1B
··· B
U B
X B 0
∂2C ∂1C
··· C
V C
Y C 0
where ∂1B in 2.3 is given by ∂1B (x) = x, and one may similarly take ∂1C ( y ) = y. One can form a free
resolution of k over A, which starts as
∂2 ∂1
F= ··· A
U ⊕ AV A
X ⊕ AY A 0, (2.3.1)
where the maps are given by the following formulas, see [5, Example 21(c)]:

A ⊗ B ∂2B 0
∂1 = A ⊗ B ∂1B + A ⊗C ∂1C , ∂2 =
0 A ⊗C ∂2C
Proposition 2.4. Let B and C be graded connected k-algebras with B 1 = 0 = C 1 , and let M be a graded left
B-module with M i = 0 for i < 0 and M 0 = 0. If either of the conditions
(i) C + C 1 = 0 = C 2 or B + B 1 = 0 = B 2 ;
(ii) C + = C 1 and dimk C 1 2, or B + = B 1 and dimk B 1 2
are satisfied, then one has
Ext1A (k, A ⊗ B M )2 = 0.
Proof. Put L = A ⊗ B M. With F as in (2.3.1), standard isomorphisms identify Homgr A ( F , L ) with the
complex
∂1∗
0 → L −−→ L X ⊕ L Y → L U ⊕ L V → · · · ,
where ∂1∗ is given by

∂1∗ (l) = (xl)x∈X , ( yl) y∈Y .
Fix elements β, γ of the same degree and set

φβ,γ = (xβ)x∈X , ( y γ ) y∈Y . (2.4.1)
One checks directly that this is a cocycle. We first prove that depth A L = 1 holds when C satisfies
C + C 1 = 0 = C 2 by choosing β and γ so that φβ,γ represents a nonzero cohomology class.
If C satisfies hypothesis (i), then one can choose elements c , c ∈ Y with cc = 0 and nonzero
elements b ∈ B 1 and m ∈ M 0 . Set
β = cc m and γ = bcm.
Suppose that φ = φβ,γ is a coboundary. Then there exists l ∈ L so that φ = ((xl)x∈X , ( yl) y ∈Y ). Write
l in the form (2.2.1) and fix y ∈ Ya for some a 1. Using the decomposition in (2.2.1), one has
ybcm = yl B + yl C + yl M . (2.4.2)
Note that ybcm starts in C a , while yl B starts in C >a . Also, the length of ybcm is 4 and the length
of yl M is 2. Therefore, it follows that yl C = yl M = 0 for all y ∈ Y, and 2.2 implies l M = 0. One can
similarly deduce Xl B = 0 by considering the equation xcc m = xl B + xl C .
Returning to (2.4.2), one now has ybcm = yl B . From 2.2, one concludes l B = bcm. One can similarly
deduce that l C = cc m. For each x ∈ X and y ∈ Y, one then has
xβ + y γ = φ(x + y ) = (x + y )(β + γ )
= xβ + xγ + y β + y γ .
Therefore, xγ + y β = 0 holds. Arguing as before, we obtain xγ = y β = 0 for each x ∈ X and y ∈ Y,

and hence yc = 0 for each y ∈ Y, contradicting our choice of c.
If B satisfies hypotheses (i), choose b, b ∈ X such that bb = 0. For any c ∈ Y, one can argue as
above, taking β = cbcm and γ = bb cm to deduce that φβ,γ is not a coboundary.
For case (ii), assume that C + = C 1 and dimk C 1 2, so that we may choose c , c ∈ Y. In the notation
from (2.4.1), we set φ = (0, ( y γ y )) ∈ L X ⊕ L Y where

bcm, y = c,
γy =
bc m, y = c .
One checks directly that φ is a cocycle.

If φ is a coboundary, then there exists an l ∈ L so that φ = ((xl)x∈X , ( yl) y ∈Y ). Arguing as before,
one gets ybcm = yl B . Since the elements of Y are left non-zerodivisors on elements of L starting
in B, one has l B = bcm. However, one also gets by a similar argument that l B = bc m, a contradiction.
Therefore, φ is not a coboundary. One may give a similar argument if the hypothesis on B in case (ii)
is satisfied. 2
Definition 2.5. For a graded connected k-algebra B and a graded left B-module M, one defines the
depth of M over B by means of the formula

depth B M = inf n ∈ N ExtnA (k, M ) = 0 .
The following proposition appears in [5, Example 36(e)2] when B and C are universal enveloping
algebras of graded Lie algebras. Our argument is an adaptation of what appears there.
Proposition 2.6. Let B and C be graded connected k-algebras with B 1 = 0 = C 1 , and let M be a graded left
B-module with M i = 0 for i < 0 and M 0 = 0. One then has
depth A ( A ⊗ B M ) 1.
In particular, equality holds if any of the conditions of Proposition 2.4 are satisfied.
Proof. We show that for any b ∈ B 1 \ {0}, c ∈ C 1 \ {0}, one of bl or cl is nonzero. If l M = 0, then one
has cl M = 0, and hence every term in cl M starts in C and has length 2. But every term in cl B has
length at least 3, and each term in cl C starts in C 2 . Therefore, none of the terms in cl B or cl C can
cancel the terms in cl M , and hence cl = 0. One can similarly argue that if one of l B or l C is nonzero,
then cl B or bl C , respectively, is nonzero. 2
Lemma 2.7. If B ∼
= k[x]/(x2 ) and C ∼
= k[ y ]/( y 2 ), then for each A-module L, one has
depth A L 1.
In particular, for each B-module M, one has depth A ( A ⊗ B M ) = 1.
Proof. Consider the k-subalgebra D of A generated by xy + yx. The algebra D is a central polynomial
subalgebra such that A is a free graded left and right D-module with basis {1, x, y , xy }. In such a
situation, the depth of A-modules may be instead computed over D, see [1, Corollary A.7]. Therefore,
the depth of A-modules are bounded above by the global dimension of D, which is one. One may
now appeal to Proposition 2.6 for the final claim. 2
We conclude this section by collecting some numerical data on modules over coproducts of alge-
bras.
2.8. Let M be a graded free k-vector space with M i = 0 for i 0 and dimk M i finite for all i. The
Hilbert series of M is the formal Laurent series

M (t ) = dimk M i t i .
i
If M (t ) is defined, we say that M has a Hilbert series.
2.9. Let M and N be graded vector spaces with Hilbert series. One then has
( M ⊗k N )(t ) = M (t ) N (t ). (2.9.1)
Let Tk ( M ) denote the tensor algebra of M over k. If M i = 0 for i 0, then one also has
1
Tk ( M )(t ) = . (2.9.2)
1 − M (t )
The first claim is clear. For the second, note that since M i = 0 for i 0, Tk ( M ) has a Hilbert series.
Furthermore, there is an isomorphism of graded vector spaces Tk ( M ) ∼= k ⊕ ( M ⊗k Tk ( M )). The desired
result follows.
Lemma 2.10. Let B and C be graded connected k-algebras, M a graded right B-module, and suppose that B,
C , and M have Hilbert series. One then has
M (t )C (t )
( A ⊗ B M )(t ) = .
B (t ) + C (t ) − B (t )C (t )
Proof. The basis and multiplication table of A given in 2.1 shows there is an isomorphism of right
B-modules
A∼
= C ⊗k Tk ( B + ⊗k C + ) ⊗k B .
Tensoring over B with M on the right gives

A ⊗B M ∼
= C ⊗k Tk ( B + ⊗k C + ) ⊗k M .
A computation of Hilbert series using (2.9.1) and (2.9.2) gives the desired equality. 2
3. The Yoneda algebra
The notation and assumptions from Theorem 1.8 are in force throughout the section.
3.1. For a ring A and an A-module L, we say L has a Poincaré series if there exists a free resolution
of L such that each free module is finitely generated. In this case, one then defines the Poincaré series
of L over A to be the formal power series

P LA (t ) := dimk ExtiA ( L , k)t i .
i
For example, if S and M are as in 1.3(1) with S noetherian and M finitely generated, M has a Poincaré
series.
In addition to the setup given in Theorem 1.8, we assume for the remainder of the section that M
and k have Poincaré series over S, and that k has a Poincaré series over T .
Theorem 3.2. One has an equality of formal power series

1 P kS (t ) 1 1
R
= S
+ −1 .
PM (t ) PM (t ) P kS (t ) P kT (t )
Proof. One may describe the basis G in Theorem 1.8 as a basis of the k-vector space k
F ⊗k
Tk (k E1 ⊗k k F1 ) ⊗k k P. Therefore, (2.9.1) and (2.9.2) give
P (t ) F (t )
G (t ) = .
1 − ( E (t ) − 1)( F (t ) − 1)
Since G is a minimal resolution of M over R, the formula above gives the desired equality. 2
The equality of Poincaré series given in Theorem 3.2 was first obtained in [6] for M = k and in [3,
Theorem 1] in general.
3.3. Let R = Ext R (k, k), S = Ext S (k, k) and T = Ext T (k, k) denote the Ext algebras of R, S, and T ,
respectively. The functor Ext− (k, k) applied to the diagram of homomorphisms of rings
τ τ∗
R T R T
induces a diagram πT∗
σ πT σ∗
of graded algebras
S k S k
πS π S∗
and hence defines a unique homomorphism of graded k-algebras
φ : S T → R.
Theorem 3.4. The homomorphism of graded k-algebras φ is bijective.
For an S-module M, we let M S be the graded left S -module Ext S ( M , k), and let M R be the
graded left R-module Ext R ( M , k). The homomorphism σ : R → S also induces a homomorphism of
graded left S -modules θ ∗ = Extσ ( M , k): M S → M R . Since θ ∗ is left σ ∗ -equivariant, the formula ξ ⊗
μ → ξ · θ ∗ (μ) defines a homomorphism
λ : R ⊗S M S → M R .
Theorem 3.5. The homomorphism of graded left R-modules λ is bijective.
In order to prove Theorems 3.4 and 3.5, we set up notation and describe the multiplication tables
for the action of R on M R .
Notation 3.6. In the notation of Construction 1.7, one has isomorphisms
MR ∼
= Hom R (G , k) ∼
= Hom R (G /mG , k) ∼
= Homk (G /mG , k)
of k-vector spaces. Let {ξ wR | w ∈ G} ⊆ Hom R (G , k) be the graded basis dual to the image of G in
G /m(G ). Also, let {ξeS | e ∈ E} ⊆ Hom R (G , k) be the graded basis dual to the basis given by the image
of E in E /p E. We abuse language and say that ξ wR starts with (respectively ends in) E if the first
(respectively last) letter of w is in E. Also, we say that ξ wR has length n if w has length n.
Our first lemma concerns the image of words of length one.
Lemma 3.7. For e ∈ E, f ∈ F, p ∈ P, and m ∈ M, one has

σ ∗ ξeS = ξeR , τ ∗ ξ fT = ξ fR , and θ ∗ ξ pS = ξ pR .
Proof. Set
D = R
(D \ E) + q E ⊆ D .
The definition of ∂ D shows that D is a subcomplex of D. Also, since R /q ∼ = S one has D / D = E as

complexes of R-modules. Let R : D → k and S : E → k be the augmentation maps, and let ψ be the
canonical surjection ψ : D → D / D = E. One then has R = S ψ , and hence σ ∗ (ξeS ) := ξeS ψ = ξeR , as
desired. The other cases are similar. 2
Next we provide a partial multiplication table for the left action of R on M R .

Lemma 3.8. Let w be a word D ∪ G, x be a letter in E ∪ F, and let m ∈ M. Let l( w ) denote the first letter of w
and let r ( w ) denote the last letter of w. Then the left action of R on M R satisfies

ξ fRw if l( w ) ∈ E ∪ P and x = f ∈ F,
ξxR · ξ wR =
ξeRw if l( w ) ∈ F and x = e ∈ E.
Proof. Let w = e w ∈ Gi , with e ∈ E j . We define a chain map ψ w ∈ Hom R (G , D )−i such that R ψw =
ξ wR as follows. Set

Gw = v ∈ G v ∈
/ G w ∪ G(E j +1 ) w ,
where G w denotes elements of G that end in w (including w), and G(E j +1 ) w denotes elements of
G ending in a letter of E j +1 , followed by w . Let R [G w ] be the free R-module generated by G w .
The definition of ∂ G shows R [G w ] is a subcomplex of G. As graded R-modules, G / R [G w ] is iso-
morphic to R [G w ] ⊕ R [G(Ej+1 ) w ]. For v w ∈ G w, set α ( v w ) = v , and extend α by R-linearity to
a homomorphism α : R [G w ] → D. One then has

α ∂( v w ) = α ∂( v ) w = ∂( v ) = ∂ α ( v w ) .
Let B be the subcomplex of G / R [G w ] spanned by G(E j +1 ) w ∪ { w }, and let C be the resolution

of T as an R-module given in Example 1.11. As C is acyclic, and G(E j +1 ) w ∪ { w } is a basis of B,
there exists a chain map β : B → C of degree −i satisfying β( w ) = 1 ∈ C 0 .
Now one can extend α to all of R [G w ] ⊕ R [G(Ej+1 ) w ] by defining it on B to be the composition
β
B −→ C → D. Note that the words in the image of B under α end in letters from E. Let ψ w denote
α
the composition G G / R [G w ] −→ D. One then has R ψ w = ξ wR , and hence ξ fR · ξ wR = ξ fR ψ w for each
element f of F.
By construction, one has ξ fR ψ w ( f w ) = 1. We show that ξ fR ψ w ( v ) is zero for all other basis ele-
ments v. If v ∈ G w , then ψ w ( v ) = 0. If v ∈ G w, then write v = v w. Then

1 if v = f ,
ξ fR ψ w ( v ) = ξ fR ( v ) =
0 otherwise.
If v ∈ G(E j +1 ) w , then ψ w ( v ) is in the span of words ending in E. Hence ξ fR ψ w ( v ) = 0. The other

cases of the left action are similar, and often easier. 2
Proof of Theorem 3.4. Under the hypothesis of the section, the k-algebras R, S , and T are degree-
wise finite. Theorem 3.2 together with Lemma 2.10 show that R(t ) = (S T )(t ). As φ is a homoge-
neous k-linear map, it suffices to prove that it is surjective. Lemma 3.8 shows that

ξeR e ∈ E ∪ ξ fR f ∈ F
generates R as a k-algebra, and Lemma 3.7 shows that these generators are in the image of φ . 2
Proof of Theorem 3.5. Under the hypothesis of the section, the graded R-modules M R and R ⊗S M S
are degree-wise finite. Lemma 2.10 and Theorem 3.2 show that the Hilbert series of R ⊗S M S and
M R are equal, so it is enough to prove that λ is surjective. By Lemma 3.8, M R is generated as a left
R-module by {ξ pR | p ∈ P}. By Lemma 3.7, λ(1 ⊗ ξ pS ) = ξ pR , and hence λ is surjective. 2
The functor Ext R (−, k) has a property similar to the one given in Theorem 3.4. We use below the
non-standard notation of μ∗ = λ(1 ⊗ Ext S (μ, k)) and similarly define ν ∗ .
Theorem 3.9. Let M be an S-module, N be a T -module, and V be a k-vector space for which there exists
μ ν N with ker μ = p M and ker ν = q N. The
exist surjective π S and π T -equivariant homomorphisms M −
→V ←
−
exact sequence of R-modules
ι μ−ν
0 → M ×V N −
→ M × N −−−→ V → 0
induces an exact sequence of graded left R-modules
(μ∗ ,−ν ∗ ) ι∗
0 → R ⊗k V ∗ −−−−−→ M R × N R −
→L→0
where L = Ext R ( M × V N , k), and V ∗ = Homk ( V , k). In particular, one has
R R R R
PM × V N (t ) = P M (t ) + P N (t ) − (rankk V ) P k (t ).
Proof. The sequence of R-modules defining M × V N induces an exact sequence of graded R-modules
(μ∗ ,−ν ∗ ) ∗
Σ −1 L → R ⊗k V ∗ −−−−−−→ M R × N R −ι→ L → Σ R ⊗k V ∗ .
Thus, we need to show that (μ∗ , −ν ∗ ) is injective. Set n = rankk V , and let F be a free S-module
μ
of rank n. One then has S-linear maps F → M −→ V that induce homomorphisms of graded left S -
modules Ext S ( V , k) → M S → Ext S ( F , k). Tensoring with R over S on the left and using Theorem 3.5,
one obtains the top row of the diagram:
μ∗
R ⊗S Ext S ( V , k) MR Ext R ( F , k)
∼
= ∼
=
n
R ⊗k V ∗ R/RS +
Note that R ⊗k V ∗ ∼ = Rn . The kernel of the bottom arrow is clearly (RS + )n , and therefore the kernel
of μ is contained in the image of (RS + )n under the left vertical isomorphism. Similarly, the kernel
∗
of ν ∗ is contained in the image of RT + . Hence one has

Ker μ∗ , −ν ∗ = Ker μ∗ ∩ Ker ν ∗ ⊆ RS + ∩ RT + = 0. 2
3.10. In the situation of 1.3(2), the maps λ and φ from Theorems 3.4 and 3.5 are homomorphisms of
bigraded algebras and modules, respectively. We therefore have the following proposition.
Recall that a graded module M over a graded connected algebra A is said to be Koszul when
ExtiA ( M , k) j = 0 if j = i. A graded connected algebra A is Koszul if k is Koszul as an A-module. The
equivalence of the first two conditions was proved in [2, Theorem 1.c].
Proposition 3.11. The following conditions are equivalent.
(i) The algebra R is Koszul.

(ii) The algebras S and T are Koszul.
(iii) There exists an S-module M that is Koszul as an R-module.
4. Depth of cohomology modules
The notation and conventions from 3.3 are still in force.

In this section, we compute the depth of the cohomology module of an R-module, where R is the
fiber product of commutative noetherian local rings ( S , p, k) and ( T , q, k). For uses of this invariant,
see [4] or [1].
Theorem 4.1. Let ( S , p, k) and ( T , q, k) be commutative noetherian local rings, set R = S ×k T , and let L be a
nonzero finitely generated R-module. If p = 0 and q = 0, then one has
depthR L 1.
Equality holds if L is an S-module or a T -module.
In order to prove Theorem 4.1, we use an observation of Dress and Krämer in [3, Remark 3]. Recall
that the syzygy Ω1R L of an R-module L is the kernel of a (graded) free cover F → L; it is defined
uniquely up to isomorphism; for n 2 one sets ΩnR L = Ω1R ΩnR−1 L. Every finitely generated R-module
L has a free cover ϕ : Q → L such that ker ϕ ⊆ m Q .
Proposition 4.2. Let L be a left R-module. Then Ω2R ( L ) ∼

= M ⊕ N where M is an S-module and N is a T -
module.
Proof. Recall that the maximal ideal of R is m = p ⊕ q. Let ϕ : Q → Q be a minimal free presentation
of L over R. One then has
Ω2R ( L ) = Ker ϕ = Ker ϕ ∩ m A
= Ker ϕ ∩ (p Q ⊕ q Q )
= (Ker ϕ ∩ p Q ) ⊕ (Ker ϕ ∩ q Q ).
To see the last equality, suppose that (x1 , x2 ) in p Q ⊕ q Q satisfies ϕ ((x1 , x2 )) = 0. Note that ϕ ((x1 , 0))
is in ϕ (p Q ) ⊆ p Q and ϕ ((0, x2 )) is in ϕ (q Q ) ⊆ q Q . Also, p Q ∩ q Q = 0, hence ϕ ((x1 , 0)) = 0 =
ϕ ((0, x2 )).
Take M = Ker ϕ ∩ p Q and N = Ker ϕ ∩ q Q to get the desired result. 2
By putting together Theorem 3.5 and Proposition 4.2, we obtain a nearly complete description of
the cohomology of R-modules.
Corollary 4.3. Set L = Ext R ( L , k), and let N T = Ext T ( N , k). There is then an exact sequence of graded left
R-modules

0 → Σ −2 R ⊗S M S ⊕ Σ −2 R ⊗T N T → L → L/L2 → 0.
We can now prove Theorem 4.1.
Proof of Theorem 4.1. If pd R L is finite, then rankk L is finite, hence depthR L = 0. If pd R L = ∞,

then Ω R2 ( L ) = 0, and by Proposition 4.2, we have Ω R2 ( L ) = M ⊕ N for some S-module M and some
T -module N, with M or N nonzero.
Set X := Ext R ( M ⊕ N , k). Corollary 4.3 provides an exact sequence
ψ
HomR k, L/L2 −→ Ext1R k, Σ −2 X −→ Ext1R (k, L).
ϑ
If S k[x]/(x2 ) or T k[ y ]/( y 2 ), then Proposition 2.4 shows that the module Ext1R (k, Σ −2 X )2
is nonzero. However, HomR (k, L/L2 ) is nonzero only in internal degree zero and one. Therefore,
ϑ is not surjective, which provides the desired inequality. If L is an S or T -module, then Proposi-
tion 2.6 gives equality.
Otherwise, one has R ∼ = k[x]/(x2 ) k[ y ]/( y 2 ), and both the inequality and equality follow from
Lemma 2.7. 2
Acknowledgments
The author would like to thank his PhD advisor Luchezar Avramov, as well as Greg Piepmeyer,
Srikanth Iyengar, and Anders Frankild, for many useful discussions and comments throughout the
evolution of this paper. I would also like to thank the referee for many useful suggestions.
References
[1] Luchezar L. Avramov, Oana Veliche, Stable cohomology over local rings, Adv. Math. 213 (1) (2007) 93–139.
[2] Jörgen Backelin, Ralf Fröberg, Koszul algebras, Veronese subrings and rings with linear resolutions, Rev. Roumaine Math.
Pures Appl. 30 (1985) 85–97.
[3] Andreas Dress, Helmut Krämer, Bettireihen von faserprodukten lokaler ringe, Math. Ann. 215 (1975) 79–82.
[4] Yves Felix, Stephen Halperin, C. Jacobsson, C. Löfwall, Jean-Claude Thomas, The radical of the homotopy Lie algebra, Amer. J.
Math. 110 (1988) 301–322.
[5] Yves Felix, Stephen Halperin, Jean-Claude Thomas, Rational Homotopy Theory, Springer-Verlag, Berlin, 2001.
[6] Alexei I. Kostrikin, Igor R. Shafarevich, Groups of homologies of nilpotent algebras, Dokl. Akad. Nauk. SSSR 115 (1957) 1066–
1069 (in Russian).
[7] Alexander Polishchuk, Leonid Positselski, Quadratic Algebras, Univ. Lecture Ser., vol. 37, American Mathematical Society,
2005, p. 57.
Journal of Algebra
Calabi–Yau Frobenius algebras

Ching-Hwa Eu, Travis Schedler ∗
MIT, Department of Mathematics, Room 2-332, 77 Massachusetts Ave, 02139 Cambridge, MA, United States
Article history: We define Calabi–Yau Frobenius algebras over arbitrary base com-
Received 17 October 2007 mutative rings. We define a Hochschild analogue of Tate cohomol-
Available online 4 December 2008 ogy, and show that this “stable Hochschild cohomology” of periodic
Communicated by J.T. Stafford
CY Frobenius algebras has a Batalin–Vilkovisky and Frobenius alge-
Keywords:
bra structure. Such algebras include (centrally extended) preprojec-
Hochschild cohomology tive algebras of (generalized) Dynkin quivers, and group algebras
Calabi–Yau of classical periodic groups. We use this theory to compute (for
Frobenius algebras the first time) the Hochschild cohomology of many algebras re-
Batalin–Vilkovisky lated to quivers, and to simplify the description of known results.
Periodic algebras Furthermore, we compute the maps on cohomology from extended
Dynkin preprojective algebras to the Dynkin ones, which relates
our CY property (for Frobenius algebras) to that of Ginzburg (for
algebras of finite Hochschild dimension).
Contents
1. Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 775
1.1. Definitions and notation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 776
2. General theory . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 778
2.1. Stable Hochschild (co)homology . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 778
2.2. Relative Serre duality for the stable module category . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 783
2.3. Calabi–Yau and periodic Frobenius algebras . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 784
2.4. Hochschild cohomology of symmetric algebras is BV . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 795
3. Hochschild (co)homology of ADE preprojective algebras over any base . . . . . . . . . . . . . . . . . . . . . . 796
3.1. Reminder of characteristic zero results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 797
3.2. Extension of results to Z and arbitrary characteristic . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 799
3.3. The maps between the extended Dynkin and Dynkin preprojective algebras . . . . . . . . . . . . . 804
4. Hochschild (co)homology of centrally extended preprojective algebras . . . . . . . . . . . . . . . . . . . . . . 805
* Corresponding author.
E-mail address: trasched@math.mit.edu (T. Schedler).
doi:10.1016/j.jalgebra.2008.11.003
C.-H. Eu, T. Schedler / Journal of Algebra 321 (2009) 774–815 775
5. Periodic group algebras of finite groups . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 808

Acknowledgments . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 810
Appendix A. Finite groups acting freely on spheres . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 810
Appendix B. Frobenius algebras over general commutative base rings . . . . . . . . . . . . . . . . . . . . . . . . . . 811
References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 814
1. Introduction
Frobenius algebras have wide-ranging applications to geometry, e.g., to TQFTs in [Moo01,Laz01,

Cos07], and are closely related to the string topology Batalin–Vilkovisky (BV) algebra [CS99,CS04], as
described in e.g. [Cos07,HL04,CV06,TZ06].
One motivation for this work is to explain the following algebraic phenomenon: the Hochschild
cohomology of many interesting Frobenius algebras has a BV structure. For example, this is true for
symmetric Frobenius algebras, and for preprojective algebras of Dynkin quivers (which are not sym-
metric).
To explain this, we define Calabi–Yau (CY) Frobenius algebras, whose Hochschild cohomology has not
only the usual Gerstenhaber algebra structure, but a BV structure (at least when the algebra is also
periodic, as we will explain). The CY Frobenius property is similar to the CY property of [Gin06], but
not the same: any Frobenius algebra over C which is CY in the sense of [Gin06] must be a direct sum
of matrix algebras.
The prototypical example of a CY Frobenius algebra A of dimension m (CY (m)) over a field k is one
that has a resolution (for A e := A ⊗k A op )

A ∨ → P m → P m−1 → · · · → P 0 A , A ∨ := Hom A e A , A e , (1.0.1)
with each P i a projective A-bimodule (the sequence must be exact). Our definition is a weakened ver-
sion of this: A ∨ Ω m+1 A in the stable A-bimodule category (this notion is recalled in Appendix B).
Any symmetric Frobenius algebra has A ∼ = A ∨ , and so we say it is CY Frobenius of dimension −1.
Our methods give a very short, simple proof of the fact that the Hochschild cohomology of such alge-
bras is BV (Section 2.4), for which many (generally more complicated) proofs are given in e.g. [Cos07,
Tra02,Men04,TZ06,Kau07b,Kau07a].
In [ES98b,ES98a,EE07], it was also noticed that the Hochschild cohomology of preprojective alge-
bras of Dynkin quivers has a self-duality property: one has H H i ( A ) ∼ = H H 5+6 j−i ( A )∗ for such algebras,
with i , 5 + 6 j − i 1, and moreover one has the periodicity H H i ( A ) ∼
= H H i +6 j when i , 6 j + i 1 (here,
j is an integer).
We explain this by introducing stable Hochschild cohomology HH• ( A ), of Frobenius algebras A. This
is a Hochschild analogue of Tate cohomology, which coincides with the usual Hochschild cohomology
in positive degrees, and is defined using the stable module category by HH• ( A ) = Hom(Ω • A , A ) (in
the latter form, this was studied in various papers, e.g., [ES06]). We show that this is a Z-graded ring,
and prove it is graded-commutative. To our knowledge, this is the first time Hom(Ω • A , A ) has been
studied in this way. We also define the notion stable Hochschild homology HH• ( A ), which coincides
with the usual notion in positive degrees, and prove that HH• ( A ) is a graded module over HH• ( A )
using a natural contraction action, which generalizes contraction in the Z0 -graded case. These results
are in Theorem 2.1.15.
We then show that, in general, HH• ( A ) ∼ = HH−1−• ( A )∗ (which makes the contraction operations
graded self-adjoint), and that the CY (m)-Frobenius property produces dualities HH• ( A ) ∼ = HHm−• ( A ).
•
Put together, we obtain a graded Frobenius algebra structure on HH ( A ) (Theorem 2.3.27), explaining
the aforementioned results of [ES98b,ES98a,EE07]. In particular, HH• ( A ) ∼ = HH2m+1−• ( A )∗ , as modules
over HH0 ( A ); in the preprojective algebra cases, m = 2, which explains the aforementioned duality.
Call a (Frobenius) algebra periodic if it has a periodic A-bimodule resolution (or, more generally,
A Ω n A in the stable bimodule category, for some n). We prove that the stable Hochschild cohomology
of periodic CY Frobenius algebras has a BV structure (Theorem 2.3.64). More generally, for any periodic
776 C.-H. Eu, T. Schedler / Journal of Algebra 321 (2009) 774–815
(not necessarily CY) Frobenius algebra, the structure of calculus [TT05,GDT89] on the pair (HH• , HH• )
extends to the Z-graded setting (HH• , HH• ) (Theorem 2.3.47). Moreover, we show that, in the (cen-
trally extended) preprojective cases, the BV structure and Frobenius algebra structure are compatible:
the BV differential is graded self-adjoint—we call such an algebra a BV Frobenius algebra.
Additionally, all of the above work is done in the context not only of Frobenius algebras over a
field, but over an arbitrary base commutative ring. Precisely, we use the notion of Frobenius extensions
of the first kind [NT60,Kas61], which says that the algebra is projective over a base commutative ring k
and that the duality is nondegenerate over this base. (Perhaps this could be generalized further to an
A ∞ -Frobenius property, but we do not do this here.)
New computational results (extending [ES98b,ES98a,EE07,Eu08,Eu07a]) concerning preprojective
algebras include:
• The computation of Hochschild (co)homology of preprojective algebras of Dynkin type over k = Z

(in types D , E, the results [EE07,Eu08,Eu07a] are over characteristic-zero fields), and the proof
that HH• is BV Frobenius over any field (Theorem 3.2.7);
• The computation of cup product and BV structure on Hochschild (co)homology of centrally ex-
tended preprojective algebras [ER06] (the groups were computed over C in [Eu06]), which we show
is BV Frobenius (Theorem 4.0.12);
• The computation of the induced maps from the Hochschild (co)homology in the extended Dynkin
case to the Dynkin case by cutting off the extending vertex (Theorem 3.3.4). This explains the
structure in the Dynkin case and elucidates the relationship between the usual Calabi–Yau and
Calabi–Yau Frobenius properties (which is analogous to Euclidean vs. spherical geometry).
We also explain and simplify the cited known results.

Our other main example is the case of group algebras of finite groups, which are automatically
Calabi–Yau Frobenius (since they are symmetric). We are interested in when these are periodic (and
hence HH• is BV and Frobenius, by Theorem 2.3.64). We show (Theorem 5.0.5) that the periodic
(CY) Frobenius algebras are just the classical periodic algebras, i.e., those whose group cohomology
is periodic (using classical results). In Appendices A and B, we also give an elementary topological
proof that groups that act freely and simplicially on a sphere (such as finite subgroups of SO(n)) are
periodic (CY) Frobenius.
1.1. Definitions and notation
Here we recall some standard definitions and state the notation we will use throughout.
All complexes will be assumed to have decreasing degree (i.e., they are chain complexes), unless
otherwise specified.
Let us fix, once and for all, a commutative ring k. When we say “algebra over k,” we mean an
algebra A over k → A such that the image of k is central in A. Bimodules over an algebra A over k
will be assumed to be symmetric as k-bimodules (i.e., A-bimodules mean A e := A ⊗k A op -modules).
Notation 1.1.1. The category A-mod means finitely-generated A-modules, for any ring A. The category
A-modk means finitely-generated A-modules which are finitely-generated projective as k-modules.
Notation 1.1.2. We will abbreviate “finitely-generated” as “fg.”
By a “Frobenius algebra over a commutative ring k,” we will mean what is also known as a “Frobe-
nius extension of the first kind” in the literature [NT60,Kas61], namely:
Definition 1.1.3. A Frobenius algebra A over a commutative ring k is a k-algebra which is a fg projec-
tive k-module, and which is equipped with a nondegenerate invariant inner product ( , ), i.e.:
(ab, c ) = (a, bc ), ∀a, b, c ∈ A ;
(−, a) : A → Homk ( A , k) is an isomorphism of k-modules. (1.1.4)

Example 1.1.5. Any group algebra k[G ] is Frobenius if G is finite, using the pairing ( g , h) = δ g ,h−1 for
g , h ∈ G. In fact, this algebra is symmetric (meaning A ∼
= A ∗ as A-bimodules).
Example 1.1.6. The preprojective algebra Π Q is known to be Frobenius if Q is Dynkin (cf. e.g., [ES98a,
ER06]). It is not difficult to see (e.g., through explicit bases as in [Eu08]) that these are in fact Frobe-
nius over Z.
Example 1.1.7. Centrally extended preprojective algebras of Dynkin quivers were defined and proved
to be Frobenius in [ER06] (working over C).
Example 1.1.8. For any two Frobenius algebras A , B over k, the algebra A ⊗k B is also Frobenius. In
particular, so is A e .
We will mainly be interested in the cases where k is a field or k = Z, and A is a fg free k-module.
(These are generally known as “free Frobenius extensions of the first kind,” cf. [NT60].)
We refer to Appendix B for some general results about Frobenius algebras and the stable module
category over arbitrary commutative rings, which are direct generalizations of standard results in the
case where k is a field. In particular, the results there justify the following definitions:
Definition 1.1.9. Let the projectively stable module category A-mod be the category whose objects are
fg A-modules, and whose morphisms Hom A are given by

Hom A ( M , N ) := f ∈ Hom A ( M , N )
/{morphisms that factor through a projective A-module}. (1.1.10)
Let A-modk ⊂ A-mod be the full subcategory of modules which are projective as k-modules.
Definition 1.1.11. For any algebra A over k, define the functor ∨ : A-mod → A op -mod by
M ∨ := Hom A ( M , A ), (1.1.12)
with the natural induced maps on morphisms.
Definition 1.1.13. For any Frobenius algebra A, let η : A −∼→ A be the Nakayama automorphism de-
fined by

(a, b) = η−1 (b), a , ∀a, b ∈ A . (1.1.14)
∼
Definition 1.1.15. For any (k-linear) automorphism φ : A −→ A, and any A-module M, let φ M de-
note M with the twisted action given by precomposing A → Endk ( M ) by φ . Similarly, for any bimod-
ule N, φ N ψ denotes twisting the left action by φ and the right action by ψ . If either of φ, ψ is the
identity, we may omit it from the notation.
We have A η ∼
= A ∗ as A-bimodules (conversely, such an isomorphism is equivalent to (−,−) with
automorphism η ).
Definition 1.1.16. Let A be a Frobenius algebra over k and M , N fg A-modules which are projective as
k-modules. For any integer i ∈ Z, let us denote

ExtiA ( M , N ) := Hom A Ω i M , N . (1.1.17)
Remark 1.1.18. When G is a finite group, and k is a field (of any characteristic), the cohomology
groups Extki [G ] (k, M ) are the Tate cohomology groups over k with coefficients in M. Indeed, we may
compute Extki [G ] (k, M ) by the complex Hom( P • , M ) where P • is any two-sided projective resolution
of k (i.e., an exact (unbounded) complex of projectives such that the cokernel of P 1 → P 0 is k). This
is one of the standard definitions of Tate cohomology (cf. e.g. [AM04, Definition 7.1]).
Remark 1.1.19. By the same token, it makes sense to define the Tate cohomology of any Hopf alge-
bra H which is Frobenius over k by ExtiH (k, M ), where k is the augmentation module. Note that,
if k is a PID, then a Hopf algebra H over k is automatically Frobenius if it is fg projective as a
k-module [LS69].
Next, we recall the definition of a Gerstenhaber algebra.
Definition 1.1.20. A Gerstenhaber algebra (V • , ∧, [ , ]) over k is a Z-graded supercommutative algebra

(V , ∧), together with a bracket [ , ] : V ⊗ V → V of degree −1, such that the induced bracket of degree
zero on the shifted graded k-module V •+1 is a Lie superbracket, satisfying the Leibniz identity,
[a ∧ b, c ] = a ∧ [b, c ] + (−1)mn b ∧ [a, c ], a ∈ Vm, b ∈ Vn, c ∈ V . (1.1.21)
Finally, we recall the definition of a BV algebra.
Definition 1.1.22. A Batalin–Vilkovisky (BV) algebra (V • , ∧, ) is a Z-graded supercommutative alge-

bra (V • , ∧) equipped with an operator : V → V of degree −1 such that 2 = 0, and such that the
bracket [ , ] defined by
(−1)m+1 [a, b] = (a ∧ b) − (a) ∧ b − (−1)m a ∧ (b) + (−1)m+n a ∧ b ∧ (1), a ∈ Vm, b ∈ Vn,
(1.1.23)
endows (V • , ∧, [ , ]) with a Gerstenhaber algebra structure. Here, 1 ∈ V 0 is the algebra unit.
2. General theory
The goal of this section is to prove that periodic Calabi–Yau Frobenius algebras have BV and Frobe-
nius structures on their Hochschild cohomology (Theorems 2.3.27, 2.3.64). Along the way, we will
prove several general results about the Hochschild (co)homology of Frobenius algebras over an arbi-
trary base commutative ring. We also explain why our definition of Calabi–Yau Frobenius implies the
CY condition of [ES06] (Theorem 2.3.21), and give a new proof that the Hochschild cohomology of
symmetric algebras is BV (Section 2.4).
We will need a straightforward generalization of the stable module category to a version relative
to k, which we relegated to Appendix B.
2.1. Stable Hochschild (co)homology
In this section, we define and begin the study of the stable Hochschild (co)homology, by replacing
Ext by Ext in the definition. By Remark 1.1.18, this is the Hochschild version of Tate cohomology of
finite groups.
For Hochschild homology, we will first need the notion of stable tensor product:
Definition 2.1.1. Let A be any algebra over k (projective as a symmetric k-module). For any A op -
module M and any A-module N, such that M , N are fg projective as k-modules, define

M ⊗ A N := f ∈ M ⊗ A N f ∈ Ker( M ⊗ A N → M ⊗ A I ),

for any k-split injection N → I, with I relatively injective . (2.1.2)
The condition that f ∈ Ker( M ⊗ A N → M ⊗ A I ) for any particular k-split injection N → I as above
is equivalent to the condition holding for all such k-split injections (cf. Appendix B).
Proposition 2.1.3. If A is a Frobenius algebra, then the definition of ⊗ A is symmetric in the following sense:

= f ∈ M ⊗ A N f ∈ Ker( M ⊗ A N → J ⊗ A N ),
M ⊗A N ∼

for any k-split injection M → J , with J a relatively injective A op -module . (2.1.4)
Thus, one has
M ⊗A N ∼
= N ⊗ A op M . (2.1.5)
Proof. Fix k-split injections N → I , M → J , for I , J relatively injective (= fg projective) A- and A op -

modules, respectively. Since I , J are projective, the maps M ⊗ A I → J ⊗ A I and J ⊗ A N → J ⊗ A I are
injective. Hence, the kernel of M ⊗ A N → J ⊗ A N is the same as the kernel of M ⊗ A N → M ⊗ A I
(both are the kernel of M ⊗ A N → J ⊗ A I ). 2
Definition 2.1.6. Let A be a Frobenius algebra over k. For any A op -module M, and any A-module N,
both which are fg projective over k, we define the ith stable Tor groups by
ToriA ( M , N ) := M ⊗ A Ω i N . (2.1.7)
Proposition 2.1.8. If A is Frobenius over k, then
ToriA ( M , N ) ∼
= ToriA ( M , N ), i 1, (2.1.9)
and moreover, the definition (2.1.7) is symmetric in the following sense:
M ⊗A Ωi N ∼
= Ωi M ⊗A N, ∀i ∈ Z. (2.1.10)
Proof. The first statement follows similarly to Corollary B.0.14. Similarly, we find that Ω i M ⊗ A N ∼ =
Tori ( M , N ) for i 1, which yields (2.1.10) for i 1. The statement is tautological for i = 0. To extend
to negative i, we may use the trick (Ω −i M ) ⊗ A Ω i (Ω −i N ) ∼ = Ω i (Ω −i M ) ⊗ A Ω −i N. 2
Definition 2.1.11. Suppose that A is Frobenius over k, and let M be any A-bimodule which is fg
projective as a k-module. The ith stable Hochschild (co)homology groups (which only depend on the
stable equivalence class of M) are defined by
e
HHi ( A , M ) := ExtiA e ( A , M ), HHi ( A , M ) := ToriA ( A , M ). (2.1.12)
Corollary 2.1.13. With A , M as in the definition, HHi ( A , M ) ∼

= HHi ( A , M ) and HHi ( A , M ) ∼
= HHi ( A , M ) for
i 1.
Remark 2.1.14. One could pose the definitions of stable Hochschild (co)homology and stable Ext
and Tor when A is not Frobenius, but one probably wants Ω to be an autoequivalence to have a
reasonable notion (e.g., if A is “relatively self-injective”; see Appendix B).
One has many algebraic structures attached to Hochschild cohomology and homology: put to-
gether, these form the structure of calculus (cf. e.g., [TT05]; see Definition 2.3.42 for the definition).
This includes cup products ∪ for Hochschild cohomology, and contraction maps H H j ( A ) ⊗ H H
( A ) →
H H
− j ( A ) for j
. For f ∈ H H j ( A ), we denote by i f : H H
( A ) → H H
− j ( A ) the corresponding con-
traction. We now show that the cup and contraction structures extend to the stable, Z-graded setting.
Theorem 2.1.15.
(i) Let A be any Frobenius algebra over k. Then one has a well-defined cup product on HH• ( A , A ), giving the
structure of an associative algebra, and extending the cup product on H H 1 ( A , A ).
(ii) One has a well-defined contraction operation HH j ( A , A ) ⊗ HHk ( A , M ) → HHk− j ( A , M ), which extends
the usual contraction operation, and satisfies the relation
i f i g (x) = i f ∪ g (x), (2.1.16)
where i f (x) is the contraction of f ∈ HH• ( A , A ) with x ∈ HH• ( A , M ). (M is any fg A-bimodule.)

(iii) The algebra HH• ( A , A ) is graded-commutative.
Proof. (i) The cup product is easy to define: for f ∈ HH j ( A , A ) and g ∈ HHk ( A , A ), we have f ∈
Hom(Ω j A , A ) and Ω j g ∈ Hom(Ω j +k A , Ω j A ), so we may consider the composition
f ∪ g := f ◦ Ω j g = f ◦ Ω | f | g , (2.1.17)
where | f | denotes the Hochschild cohomology degree. It follows immediately that the cup product is
associative.
(ii) To define the contraction operation, note that f ∈ HH j ( A , A ) = Hom(Ω j A , A ) induces a map
Ω j A ⊗ Ae Ω k A → A ⊗ Ae Ω k A , (2.1.18)
for all k. Applying the equivalence Ω , we obtain a map
A ⊗ A e Ω j +k A → A ⊗ A e Ω k A , (2.1.19)
for all k ∈ Z. This is the desired map. We automatically get the intertwining property (2.1.16).
(iii) We have isomorphisms in the stable module category,
Ω j A ⊗ A Ω k A −∼→ Ω j +k A , (2.1.20)
which follow from the fact that Ω j A may be considered as a projective left A-module for all j (using
the sequence Ω 1 A → A ⊗ A A, which is split as left A-modules). We will need the
Claim 2.1.21. Let us use Ω i A := (Ω 1 A )⊗ A i for i 1. Given f ∈ Hom(Ω j A , Ω k A ), we may form a represen-
tative of Ω f ∈ Hom(Ω j +1 A , Ω k+1 A ) by either Id ⊗ A f or by (−1) j −k f ⊗ A Id.
Proof. This follows from the fact that the stable module category is a suspended category as in [SA04]
(since it is a full monoidal subquotient of the derived category which is closed under suspension).
However, we give an explicit argument. The terms of the normalized bar resolution may be written
as A ⊗ Ω • A, and if we construct this by splicing together sequences Ω n+1 A → A ⊗ Ω n A Ω n A, it
is easy to see that f lifts to f ⊗ A Id. Writing the terms of the normalized bar resolution as Ω • A ⊗ A,
we obtain the desired sign corrections of (−1) j −k . 2
As a consequence, if we have f ∈ HH j ( A , A ), g ∈ HHk ( A , A ), then we may compute f ∪ g in

two ways. First, if j , k 0, then letting f ∈ Hom(Ω j A , A ), g ∈ Hom(Ω k A , A ), we may use ei-
ther the formula f ⊗ A g or (−1)| f || g | g ⊗ A f . Similarly, if j , k are arbitrary, we take instead
f ∈ Hom(Ω a+ j A , Ω a A ), g ∈ Hom(Ω b+k A , Ω b A ) for a + j , a, b + k, b 0. Then the same argument
for the composition Ω b f ◦ Ω a+ j g yields either (−1) jb f ⊗ A g or (−1)(a+ j )k g ⊗ A f . So, we obtain
(−1) jb f ∪ g = (−1)(b+k) j g ∪ f , as desired. 2
Finally, we present a duality property for stable Hochschild homology, which will induce a Frobe-
nius algebra structure on Hochschild cohomology for “Calabi–Yau Frobenius algebras.” To do this, we
first need to explain how to write standard complexes computing HH• and HH• . More generally, we
define these computing stable Ext and Tor.
Definition 2.1.22. For any Frobenius algebra A over k and any fg A-module M which is projective
over k, call a two-sided resolution of M, an exact k-split complex of fg projective A-modules,
· · · → P 2 → P 1 → P 0 → P −1 → P −2 → · · · , (2.1.23)
such that M is the cokernel of P 1 → P 0 (and the kernel of P −1 → P −2 ):
P 1 → P 0 M → P −1 → P −2 . (2.1.24)
Definition 2.1.25. For any Frobenius algebra A, any fg left A-module M which is projective over k,
any fg right A-module N which is projective over k, and any two-sided resolution P • of M, define
the associated (“standard”) complex computing stable Tor,
C •A ( N , M ) := N ⊗ A P • . (2.1.26)
Similarly, if now M , N are both fg left A-modules which are k-projective, we define the associated
(“standard”) complex computing stable Ext,
C •A ( M , N ) := Hom A ( P • , N ). (2.1.27)
We call the classes of the complexes C • , C • in the (unbounded) derived category of fg k-modules, the
stable M ⊗ LA N and RHom( M , N ).
Definition 2.1.28. For any two-sided A e -resolution of A, we call C • ( A , A ) := C •A ( A , A ) and C • ( A , A ) :=

e
C •A e ( A , A ) the associated “standard” complexes computing stable Hochschild homology and cohomol-
ogy, respectively.
Remark 2.1.29. Note that we could have chosen to reserve the words “standard” for complexes re-
sulting from the bar resolution of A (which can be completed to a two-sided resolution, as we will
explain).
Theorem 2.1.30. Let A be a Frobenius algebra over k. We have the following duality:
∼
Dk : HHi ( A , A ) −
→ HH−1−i ( A , A )∗ , if k is a field, (2.1.31)
q.i.
e ∼ e
∗
Dk : C •A ( A , A ) −
→ A
C− 1−• ( A , A ) , in general. (2.1.32)
Moreover, using (2.1.31), the contraction maps become graded self-adjoint:
Dk (i f x) = (−1)| f ||x| i ∗f Dk (x). (2.1.33)
An easy extension of the theorem to coefficients in any bimodule M which is fg projective over k
yields
∗ q.i. ∗
Dk : HHi ( A , M ) −∼→ HH−1−i A , M ∗ ⊗ A A ∨ , ∼ ∗ ∨
e e
C •A ( A , A ) −→ C −
A
1−• A , M ⊗ A A . (2.1.34)
To prove the theorem, the following easy identifications will be useful:
Lemma 2.1.35. Let A be a Frobenius algebra over k.
(i) For any left A-module M which is fg projective over k,

(φ M )∗ ∼
= M∗ φ , (φ M )∨ ∼
= M∨ φ , φA ⊗A M ∼
= φ M. (2.1.36)
Furthermore,

M ∨ ⊗ A A∗ ∼
= M∗ ∼
= M∨ η. (2.1.37)
(ii) As A-bimodules, we have
φ Aψ
∼
= A φ −1 ◦ψ , A∨ ∼
= A η−1 , A∨ ⊗ A A∗ ∼
=A∼
= A∗ ⊗ A A∨. (2.1.38)
Proof. (i) The first set of identities is immediate. For the second, we use (B.0.10):
M ∨ ⊗ A A∗ ∼
= M η∗ −1 ⊗ A A ∗ ∼
= M∗. (2.1.39)
(ii) The first identity is clear. Applying (2.1.37) to the case of bimodules, we have
∗
A∗ ⊗ A A∨ ⊗ A A∗ ∼
= A ∨ ⊗ Ae A e ∼
= A∗, (2.1.40)
which immediately gives the last identity, and hence the second. 2
Proof of Theorem 2.1.30. Given any two-sided resolution P • of A, the lemma shows that the exact
∨ ∗
complex P − 1−• ⊗ A A must be another two-sided resolution of A. Furthermore, we have
∗
∨
P− 1−• ⊗ A A
∗
⊗ Ae A ∼ ∨
= P− ∗∼
1−• ⊗ A A = Hom A P −1−• , A
e e
∗ ∼
= A ⊗ A e P −1−• , (2.1.41)
where the last isomorphism uses the standard adjunction.

To show the graded self-adjoint property (2.1.33), we first note the following naturality: apply-
ing ( f̃ ∨ ⊗ A Id A ∗ ) ⊗ A e Id on the LHS of (2.1.41) (where f̃ is the lift of f to P • ) is the same as
applying (Id ⊗ A e f̃ )∗ on the RHS. Now, the LHS can be replaced by the complex Hom A e ( P −1−• ,
A ∗ ⊗ A P • ) (which is the total complex, summing the • degrees). Applying ( f̃ ∨ ⊗ A Id A ∗ ) ⊗ A e Id be-
comes right composition with f̃ . Using the same argument as in the proof of Theorem 2.1.15(iii),
this is chain-homotopic to applying left composition with (−1)| f |·|−1−•| (Id A ∗ ⊗ A f̃ ). Now, via
the quasi-isomorphism A ⊗ A e P • ( P − ∨ ∗ ∗
1−• ⊗ A A ) ⊗ A P • Hom A ( P −1−• , A ⊗ A P • ), applying
e e
(−1)| f |·|−1−•| Id A ∗ ⊗ A f̃ gets carried to this map, which we showed is chain-homotopic to the map
obtained from applying (Id ⊗ A e f̃ )∗ on the RHS of (2.1.41). Now, passing to homology, we get the de-
sired graded self-adjointness. (In the Calabi–Yau Frobenius case which we will define, we can deduce
this result more simply as in the proof of Theorem 2.3.27). 2
2.2. Relative Serre duality for the stable module category
From now on, let A be a Frobenius algebra over k. As was noticed in [ES06], when k = a field, the
Auslander–Reiten homomorphisms give a Serre duality for the stable module category. We recall this
and produce a relative version.
Notation 2.2.1. Let ν : A-mod → A-mod be the Nakayama functor, ν = ∗ ◦ ∨.
Note that, when k is a field, ν sends projectives to injectives, and vice-versa, and in fact induces
an equivalence of categories {projective A-modules} ↔ {injective A-modules}. Also, by (B.0.10), one
has the following simple formula for ν M (which will be useful later):
νM ∼
= A∗ ⊗ A M ∼
= η−1 M . (2.2.2)
Proposition 2.2.3. (See [ES06].) When k is a field, we have functorial isomorphisms
Hom A ( M , N ) ∼
= Hom A ( N , Ω ν M )∗ . (2.2.4)
The proof is based on the Auslander–Reiten formulas (cf., e.g. [ARS97,ASS06]).

In order to make proper sense of the Serre duality for arbitrary k, it is necessary to replace the
groups above by complexes and dual complexes. First, observe that the above isomorphism can be
rewritten, by replacing M by Ω i M, as
ExtiA ( M , N ) ∼
= Ext−
A
1− i
( N , ν M )∗ . (2.2.5)
The following then gives a version of the above for general k:
Theorem 2.2.6. Let M , N be A-modules which are fg projective as k-modules. Then, one has a functorial
quasi-isomorphism in the derived category,
C •A ( M , ν N ) C −
A
1−•
( N , M )∗ . (2.2.7)
Proof. Fix a two-sided projective resolution P • of N. Applying ν , we obtain a resolution of ν ( N ). By

Proposition B.0.4, Hom A ( M , ν P −1−• ) may be used to compute Ext•A ( M , ν N ), since ν P • is a two-sided
resolution of ν N consisting of relatively injectives. Furthermore, for any two-sided resolution Q • of M,
we may obtain quasi-isomorphisms
∼
C •A ( M , ν N ) = Hom A ( Q • , ν N ) ←∼− Hom A ( Q • , ν P −1−• ) −→ Hom A ( M , ν P −1−• ). (2.2.8)
Next, we show that Hom A ( M , ν P −1−• ) Hom A ( P −1−• , M )∗ . For any module L, there is a functorial
map L ∨ ⊗ A M → Hom A ( L , M ), which is an isomorphism if L is projective. Applying this map and its
dual, we obtain
∗ adj.
∼
Hom A ( P −1−• , M )∗ −→ P −
∨ ∼
1−• ⊗ A M = Hom A ( M , ν P −1−• ), (2.2.9)
the last map using adjunction. 2

2.3. Calabi–Yau and periodic Frobenius algebras
We will modify the definition of Calabi–Yau algebra [Gin06] to suit Frobenius algebras. First, we
recall this definition.
Definition 2.3.1. (See [Gin06].) An associative algebra A over a commutative ring k is called Calabi–
Yau if it has finite Hochschild dimension, and one has a quasi-isomorphism in the derived category
of A e -modules,
∼
f : A [d] −→ RHom A e ( A , A ⊗ A ) (2.3.2)
which is self-dual:
f ! ◦ ι = f [−d], (2.3.3)
where, for any map g : M → N of A-bimodules, g ! : RHom A e ( N , A e ) → RHom A e ( M , A e ) is the natural

map, and ι : A → RHom A e ( A , RHom A e ( A , A ⊗ A )) is the natural map. Here [−] denotes the shift in
the derived category.
However, if A is Frobenius and of finite Hochschild dimension (part of being (usual) CY), then
Ω i A = 0 for large enough i, and hence A 0 in the stable module category. Hence, H H i ( A , M ) = 0
for all i 1, so A has Hochschild dimension zero. That is, A is a projective A-bimodule (i.e., A is
separable).
For a separable Frobenius algebra to additionally be Calabi–Yau of dimension zero, we require
exactly that A ∼
= A ∨ as A-bimodules. By Lemma 2.1.35, this is equivalent to A ∗ ∼
= A, i.e., A must be a
symmetric, separable Frobenius algebra.
This is not general enough, so we replace this notion. Let us first restate the usual Calabi–Yau
property in terms of a quasi-isomorphism of complexes:
0 Pm P m−1 ··· P0 A
(2.3.4)
0 P 0∨ P 1∨ ··· ∨
Pm A.
Here and from now on, the functor ∨ will be in the category of A-bimodules, i.e., M ∨ = Hom A e ( M ,
A ⊗ A ).
If now A becomes a Frobenius algebra, as explained earlier, we cannot have such a quasi-
isomorphism (for m 1): in fact, when dualizing the top sequence, we get something that begins
with A ∨ → P 0∨ → P 1∨ . However, it still makes sense to ask for a commutative diagram, with exact
rows, as follows:
A∨ Pm P m−1 ··· P0 A
(2.3.5)
A∨ P 0∨ P 1∨ ··· ∨
Pm A.
In fact, since the dual is automatically exact in the Frobenius case, by a standard result of homological
algebra, such a diagram must automatically exist given a resolution as in the top row. So it is enough
to ask for such a resolution.
On the level of the stable module category, having such a resolution implies the following condi-
tion:
Definition 2.3.6. A Frobenius algebra A over k is called Calabi–Yau Frobenius of dimension m if one
has isomorphisms in the stable module category Stabk ( A e ), for some m ∈ Z:1
f : A ∨ Ω m+1 A . (2.3.7)
If there is more than one such m, then we pick the smallest nonnegative value of m (which exists
because such algebras must be periodic as in the subsequent definition).
If additionally A has a grading, such that the above isomorphism is a graded isomorphism when
composed with some shift (considering the stable module category to now be graded), then we say
that A is a graded Calabi–Yau Frobenius algebra. More precisely, if f : A ∨ (m ) Ω m+1 A is a graded
isomorphism, where (
) denotes the shift by
with the new grading, then one says that A is graded
Calabi–Yau Frobenius with dimension m of shift m .
We remark that the above definition of Calabi–Yau Frobenius is (apparently) stronger than the
notion of Calabi–Yau for self-injective algebras discussed in [ES06] for the case k is a field: see Theo-
rem 2.3.21 and the comments thereafter.
Definition 2.3.8. A Frobenius algebra over k is called periodic Frobenius of period n for some n > 0 if
one has (in Stabk ( A e ))
g : A Ωn A, (2.3.9)
and that is the smallest positive n for which one has such an isomorphism. If A has a grading, we
define graded periodic Frobenius of period n and shift n as before (if g : A (n ) Ω n A is a graded
isomorphism in the stable module category).
Note that it makes sense to be Calabi–Yau Frobenius of negative dimension. In particular, any
symmetric Frobenius algebra is either Calabi–Yau Frobenius of dimension −1, or periodic Calabi–Yau Frobenius
of dimension n − 1 and period n for some n 1.
Also, any periodic Frobenius algebra must have even period, unless 2 · Id 0 in the stable module
category, e.g., char k = 2, as we will see in Theorem 2.3.47. In particular, the CY dimension must be
odd for symmetric Frobenius algebras.
Example 2.3.10. The preprojective algebras of ADE Dynkin quivers are periodic Calabi–Yau Frobenius
(using [EE07]) of dimension 2 and shift 2, and of period 6 and shift 2h (twice the Coxeter number).
The essential ingredient is the Schofield resolution [RS] (cf., e.g., [EE07]), with R = k I , where I is the
vertex set:
A ∨ (2) → A ⊗ R A (2) → A ⊗ R V ⊗ R A → A ⊗ R A A . (2.3.11)
Here, V is the free k-module spanned by the edges of the quiver. This completes to a periodic
projective resolution of length 6, since the Nakayama automorphism has order 2 (for more de-
tails, see Proposition 2.3.15). The fact that the shift is 2h follows from the fact that A ∨ (2) ∼
= A η (h),
which amounts to the fact that the degree of the image of Id ∈ A ⊗ R A ∗ under the isomorphism
1
The self-duality property required for Calabi–Yau algebras (2.3.3) can still be imposed here: this would be the condition
that f : A ∨ Ω m+1 A satisfies f ∨ = Ω −m−1 f . We do not need this.
A ⊗R A∗ ∼ = A ⊗ R A from the pairing, is h − 2 (i.e., h − 2 is the degree of the product of any basis
element with its dual basis element).
We note that an important part of the above is showing that Π Q is free (equivalently, projective)
over Z; this follows from explicit Z-bases (such as those in [Eu08], which one may verify is integral;
types A , D over Z are also in [Sch07, §4.2.2]).
Example 2.3.12. Similarly, the centrally extended preprojective algebras [ER06], over k = C, are peri-
odic Calabi–Yau Frobenius with dimension 3 (of shift 4) and period 4 (of shift 2h), using [Eu06]. These
are symmetric, i.e., have trivial Nakayama automorphism. Note that these are not, in general, torsion-
free over Z (cf. Section 4), hence not Frobenius over Z, although the definition may be modified to
correct this.
Example 2.3.13. Similarly to the Dynkin case, one may consider preprojective algebras of generalized
Dynkin type: this refers to preprojective algebras of type T n (otherwise known as L n ) which can be
obtained from Π A 2n by passing to fixed points under the Nakayama automorphism, Π T n := (Π A 2n )η .
In other words, this is associated to a graph of T n type.
In [Eu07b] (cf. [BES07]), using a variant of the Schofield resolution [RS], it is proved that Π T n is
periodic Calabi–Yau Frobenius of dimension 5 (and shift h + 2) and period 6 (and shift 2h). Also,
in [Eu07b], a correction to results of [ES98a] in type A is given.
Example 2.3.14. (See [BBK02].) The trivial extension algebras of path algebras of Dynkin quivers are
periodic Calabi–Yau Frobenius (in fact, symmetric) of dimension 2h − 3 and period 2h − 2, where h is
the Coxeter number. These are “almost-Koszul dual” to the preprojective algebras; see [BBK02].
For Calabi–Yau Frobenius algebras, being periodic Frobenius is closely related to having finite-order
Nakayama automorphism:
Proposition 2.3.15. If A is Calabi–Yau Frobenius of dimension = −1, then the following are equivalent:
(i) A is periodic Frobenius;

(ii) For some p > 0, one has
A A η p in the stable bimodule category2 (2.3.16)
(e.g., if η p is inner).
In the situation that the above are satisfied, then the Calabi–Yau dimension m, the period n, and the smallest
p > 0 such that (2.3.16) holds, are related by
n = p · gcd(n, m + 1), (2.3.17)
and r = gcd(n, m + 1) is the smallest positive integer such that Ω r A A ηk for some integer k.
In particular, a Calabi–Yau Frobenius algebra of dimension −2 must have infinite-order

Nakayama automorphism (since if it were periodic, the CY dimension is defined to be nonnega-
tive). Being dimension −1 is a special case, consisting of stably symmetric Frobenius algebras that are
not periodic.
2
Note that, if we had defined Calabi–Yau and periodic Frobenius algebras using complexes rather than the stable module
category, then this would be replaced by an honest isomorphism of bimodules, A ∼
= A η p , and hence η p would have to be inner.
Proof. (ii) implies (i). We have Ω m+1 A A ∨ ∼

= A η−1 . For any p such that (2.3.16) holds, we have
p ·(m+1)
A Aη Ω
− p A, yielding (i).
(i) implies (ii). As in the previous paragraph, we deduce that Ω k·(m+1) A A η−k . If Ω n A A, then
we would deduce that A Ω −n·(m+1) A A ηn , yielding (ii).
To obtain (2.3.17), let r > 0 be the smallest positive integer such that Ω r A A ηk for some inte-
ger k. It follows that r | (m + 1) and r | n. Since r | (m + 1), it must be that k is relatively prime to p,
otherwise A k· m+1 Ω m+1 A A η−1 would contradict the minimality of p. Similarly, we deduce that
η r
n = p · r, and r = gcd(n, m + 1). 2
Also, we have the following growth criterion:
Notation 2.3.18. For any fg k-module M, let g ( M ) = g k ( M ) 0 denote its minimal number of gener-
ators.
Proposition 2.3.19. Suppose that k has finite Krull dimension (or just that its maximal ideal spectrum has
finite dimension). If A is Calabi–Yau Frobenius of dimension = −1, or if A is periodic Frobenius, then for any
fg A-modules M , N which are projective as k-modules, there is a positive integer p 1 such that, for all i ∈ Z,
Exti ( M , N ) and Tori ( M ∗ , N ) are generated by at most p generators over k. We may take p = ag ( M ) · g ( N ) + b
for some a, b 0 depending only on A and k.
Moreover, for any fg A-bimodule L which is projective as a k-module, there is a positive integer q 1
such that HHi ( A , L ), HHi ( A , L ) are generated by at most q generators, for all i ∈ Z. Again, we may take q =
a · g ( L ) + b for some a , b 0 depending only on A and k.
Proof. Without loss of generality, assume that the tautological map k → A is injective. Under either
assumption of the proposition ( A is CY Frobenius of dim = −1 or A is periodic Frobenius), we have
Ω r A A φ for some r 1 and some automorphism φ of A. Thus, for all integers s, we may write

s = r · k + r for some 0 r < r, and then Ω As e A Ω rA e A φk . Since Ω As M Ω As e A ⊗ A M, we have

= Hom Ω As e A ⊗ A M , N ∼
Exts ( M , N ) ∼ = Hom φ −k Ω rA e A ⊗ A M , N . (2.3.20)
Furthermore, using the normalized bar resolution of A (twisted by automorphisms of A), we may take
Ω As e A to be isomorphic, as a right A-module, to ( A /k)⊗s ⊗ A. Then, localizing k at any prime ideal,

Hom(φ −k Ω rA e A ⊗ A M , N ) can have rank at most ( g ( A ) − 1)r · g ( M ) · g ( N ). By Theorem 1 of [Swa67],
r
Hom(φ −k Ω A e A ⊗ A M , N ) can then be generated by at most ( g ( A ) − 1)r · g ( M ) · g ( N ) + d elements,
where d is the dimension of the maximal ideal spectrum of k.
We can apply the same idea for Tors ( M , N ). For HH j ( A , L ) and HH j ( A , L ) where L is now an A-
bimodule, we now use that these are given by Hom A e (Ω j A , L ) and Ω j A ⊗ A e L, and apply the same
argument as above. 2
In particular, the above proposition rules out tensor products of periodic Frobenius algebras over
a field (which are not separable) from being periodic or Calabi–Yau Frobenius of dimension = −1, by
the Künneth theorem.
The following theorems also have graded versions (by incorporating the shifts in the definitions),
but we omit them for simplicity.
Theorem 2.3.21.
(i) For any Calabi–Yau Frobenius algebra A of dimension m, and any fg A-modules M , N which are projective
over k, one has isomorphisms ( functorial in M , N)
∗
Hom A ( M , N ) ∼
= Hom A N , Ω −m M , if k is a field, (2.3.22)
−•
RHom•A ( M , N ) RHomm
A ( N , M )∗ , generally. (2.3.23)
That is, Ω −m is a (right) Serre functor for the stable module category relative to k.
(ii) For any periodic Frobenius algebra A of period n, and any fg A-modules M , N, one has isomorphisms

Hom A ( M , N ) ∼
= Hom A Ω n M , N , RHom•A ( M , N ) RHom•+n
A ( M , N ). (2.3.24)
As a corollary, we see that any Calabi–Yau Frobenius algebra over a field is also Calabi–Yau in the
sense of [ES06]. We do not know if the reverse implication holds.
Proof. (i) By (2.2.2) (which holds for arbitrary k) and (2.1.38), we have a stable equivalence
M Ω m+1 ν ( M ), i.e., ν M Ω −m−1 M . (2.3.25)
Since Ω ν already provides a right Serre functor in the senses needed for (i) (using Theorem 2.2.6), we
now know that Ω −m also provides such a right Serre functor (note that the above stable equivalences
are clearly functorial).
(ii) This is easy. 2
Remark 2.3.26. For general Frobenius algebras, one can still say that Hom( M , N ) ∼ = Hom( N , A ∗ ⊗ A
∗
Ω M ) (by (2.2.2) and Theorem 2.2.6), so that the Serre functor involves shifting and twisting by
an invertible (under ⊗ A ) bimodule. Similarly, the following results on Hochschild (co)homology have
analogues for arbitrary Frobenius algebras using twisting (by powers of η ) as well as shifting. For
this, one considers the bigraded algebra Hom•A e ( A , A η• ) and its bigraded module Ω • A ⊗ A e A η• . We
do not need this for our examples. However, it might be interesting to try to apply this formalism to
finite-dimensional Hopf algebras (analogously to [BZ08]).
We now present results on stable Hochschild cohomology of periodic and CY Frobenius algebras:
Theorem 2.3.27. Let A be a Calabi–Yau Frobenius algebra of dimension m, and M any A-bimodule.
(i) One has isomorphisms and quasi-isomorphisms

D : HH• ( A , M ) ∼
= HHm−• ( A , M ), RHom•A e ( A , M ) A ⊗ LA e M m−•
. (2.3.28)
For M = A, these isomorphisms intertwine cup product with contraction:
D ◦ i f (x) = f ∪ D(x), ∀x ∈ HH• ( A , A ), f ∈ HH• ( A , A ). (2.3.29)
(ii) Let M be fg projective over k. One has
∗
HH• ( A , M ) ∼
= HHm−• A , M ∗ , if k is a field, (2.3.30)
• −•
∗ ∗
RHom A e ( A , M ) RHomm
Ae A, M , generally. (2.3.31)
In the case M = A, we may rewrite this, respectively, as
(2m+1)−•
HH• ( A , A ) ∼
= HH(2m+1)−• ( A , A )∗ , RHom•A e ( A , A ) RHom A e ( A , A )∗ . (2.3.32)
(iii) The induced pairing
HH• ( A , A ) ⊗ HH(2m+1)−• ( A , A ) → k (2.3.33)
is invariant with respect to cup product:
( f , g ∪ h) = ( f ∪ g , h), | f | + | g | + |h| = 2m + 1, (2.3.34)
and is nondegenerate if k is a field. Moreover, for all k, one has a nondegenerate invariant pairing in the
derived category (of degree −(2m + 1)),3
RHom A e ( A , A ) ⊗ RHom A e ( A , A ) → k, (2.3.35)
inducing (2.3.32).
In other words, C • is a Frobenius algebra in the derived category of k-modules, and if k is a field, HH• is a
graded Frobenius algebra over k (using a definition that only requires finite-generation in each degree).
We also remark that (2.3.34) and (2.3.29), together with the graded commutativity of Theo-
rem 2.1.15(iii), give another proof of the graded self-adjointness of i ∗− (2.1.33) in this case.
Proof. (i) Let us pick a two-sided resolution of A:
· · · → P 2 → P 1 → P 0 A → P −1 → P −2 → · · · , (2.3.36)
so that, removing the A, the complex P • is an exact complex of projectives. Since Ω m+1 A A ∨ (in
the stable module category), we may form chain maps between the two obtained resolutions,
∨
P •+(m+1) ↔ P − 1−• , (2.3.37)
such that their composition on either side induces identity maps on the level of Hom(Ω i A , Ω i A ) and
Hom(Ω i A ∨ , Ω i A ∨ ) for all i ∈ Z. As a result, upon applying the functor A ⊗ A e −, we obtain quasi-
isomorphisms of the resulting complexes,
∼ ∨
P •+(m+1) ⊗ A e M −→ P − 1−• ⊗ A M .
e (2.3.38)
However, it is clear that these are standard complexes computing HHm+1+• ( A , M ) and HH−1−• ( A , M ),
which is what we needed.
To prove the intertwining property (2.3.29), we note that, for f ∈ HH j ( A , A ), applying i f in the
LHS of (2.3.29) is the same as applying the corresponding element of Hom( A , Ω − j A ) to the M = A
in (2.3.38). Similarly, applying f ∪ − to the RHS is post-composing with f , which is applying the same
element of Hom( A , Ω − j A ) to the A in the RHS of (2.3.38).
(ii) This follows from part (i) and Theorem 2.1.30. To fix the signs (so as to obtain the Frobenius
property in part (iii)), we use for the duality x → (−1)|x|·m D∗ ◦ Dk ◦ D(x), rather than only D∗ ◦
Dk ◦ D. Alternatively, we can use ν A = A ∨∗ A η2 Ω −2(m+1) A and Theorem 2.2.6, and similarly for
arbitrary M, which will give the desired property.
(iii) For this, we will use functoriality of the isomorphisms in (ii). (The result also follows from
Theorem 2.1.30(iii).)
3
Here, a pairing X ⊗ Y → k is nondegenerate if it induces a (quasi-)isomorphism X Y ∗ , or equivalently, Y X ∗ .
In the case that k is a field, the first isomorphism of (2.3.32) comes from the functorial isomor-
∼
phisms of (2.2.7) (in the case k is a field, we can rewrite this as Hom( M , N ) −→ Hom( N , ν Ω M )∗ ).
Thus, we have the commutative square (for any g ∈ HH ( A , A ))
j
∼
Hom A e ( A , Ω −i A ) Hom A e (Ω −i A , Ω −(2m+1) A )∗
Ω −i − j g ◦ (◦ Ω −i g )∗ (2.3.39)
∼
Hom A e ( A , Ω −i − j A ) Hom A e (Ω −i − j A , Ω −(2m+1) A )∗
which gives exactly the invariance property needed. On the level of complexes, we may make the
desired statement as follows. Let P • , as in (2.3.36), be a two-sided resolution of A. Then, we have the
following sequence, where the isomorphisms mean quasi-isomorphisms, and all complexes are total
complexes graded by •:
∼ ∼ ∼ ∗
C • ←∼− Hom A e ( P • , P −• ) −→ Hom A e ( P −• , ν P 1+• )∗ −→ Hom A e ( P −• , P •−(2m+1) )∗ −→ C (2m+1)−• .
(2.3.40)
Next, for any j ∈ Z and any g ∈ Hom A e ( P •+ j , P • ), we have the commutative square
∼
Hom A e ( P • , P −• ) Hom A e ( P −• , P •−(2m+1) A )∗
g◦ (◦ g )∗ (2.3.41)
∼
Hom A e ( P • , P −•− j ) Hom A e ( P −•− j , P •−(2m+1) )∗
which gives the desired result. The pairing C • ⊗ C • → k in the derived category is given by replac-
ing C • with Hom A e ( P • , P −• ) and using composition. 2
We now need to recall the definition of calculus:
Definition 2.3.42. (See [Tsy04, p. 93].) A precalculus is a pair of a Gerstenhaber algebra (V • , ∧, [ , ])

and a graded vector space W • together with:
(1) A module structure ι− : V • ⊗ W −• → W −• of the graded commutative algebra V • on W −• ;

(2) An action L− of the graded Lie algebra V •+1 on W −• , which satisfies the following compatibility
conditions:
ιa Lb − (−1)|a|(|b|+1) Lb ιa = ι[a,b] , (2.3.43)
La∧b = La ιb + (−1)|a| ιa Lb . (2.3.44)
A calculus is a precalculus (V • , W • , [ , ], ∧, ι− , L− ) together with a differential d of degree 1 on W •

satisfying the Cartan identity:
La = dιa − (−1)|a| ιa d. (2.3.45)
It is a result of [GDT89] that, for any associative algebra A, the collection

HH• ( A , A ), HH• ( A , A ), { , }, ∪, i − , L− , B (2.3.46)
is a calculus, where B is the Connes differential, { , } the Gerstenhaber bracket, and L− the Lie deriva-
tive operation.
Theorem 2.3.47. Let A be any periodic Frobenius algebra of period n. Then, n must be even if 2 · HH• ( A , A ) = 0,
and:
(i) One has isomorphisms and quasi-isomorphisms
HH• ( A , A ) ∼
= HH•+n ( A , A ), HH• ( A , A ) ∼
= HH•+n ( A , A ), (2.3.48)
∼
→ C •+
C •A e ( A , A ) − n
A e ( A , A ), (2.3.49)
and similarly a quasi-isomorphism between the standard complex computing stable Hochschild homology
and its shift by n. Moreover, the isomorphisms may be induced by cup product on the left with the element
∼
1 ∈ HHn ( A , A ) representing the given isomorphism Ω n A −
→ A, and by the contraction i 1 .
(ii) The stable Hochschild cohomology is a Gerstenhaber algebra, which extends the Gerstenhaber structure
on usual Hochschild cohomology.
(iii) The stable Hochschild cohomology and stable Hochschild homology form the structure of a calculus, ex-
tending the usual calculus structure.
Proof. (i) The isomorphisms follow as in previous proofs from the stable module isomorphism 1 :
A Ω n A. To show that they are induced by cup product or contraction with 1 ∈ HHn ( A , A ), let us
construct a projective resolution4 of A such that
⊗ A j
ΩjA ∼
= Ω1 A . (2.3.50)
Following the proof of Theorem 2.1.30(iii), we construct this from any exact sequence Ω 1 A → P A
such that P is a fg projective A-bimodule and Ω 1 A is an A-bimodule which is projective as a left and
right A-module (separately), by splicing together (Ω 1 A )⊗ A j → P ⊗ A (Ω 1 A )⊗ A ( j −1) (Ω 1 A )⊗ A ( j −1)
for all j 1. If we construct these inductively by tensoring on the left by Ω 1 A, then we see that
the sequences are all exact since Ω 1 A is a projective right A-module; also, P ⊗ A (Ω 1 A )⊗ A ( j −1) is a
projective A e -module because the result is obvious in the case that P is a free A e -module.
Now, given f ∈ Hom A e (Ω j A , Ω k A ), we may construct Ω f ∈ Hom A e (Ω j +1 A , Ω k+1 A ) by applying
Ω 1 A ⊗ A −, by construction of the above resolution. On the other hand, the isomorphism HHi ( A , A ) ∼ =
HHi +n ( A , A ) is given by the stable module isomorphism 1 : Ω n A A. That is, we use the stable
module isomorphism Ω n+i A Ω i A, which by the above is 1 ⊗ A Id : Ω n A ⊗ A Ω i A → A ⊗ A Ω i A, so
HHi ( A , A ) ∼
= HHi +m ( A , A ) is given by cup product on the left with 1 .
Now, since cup product with 1 induces an isomorphism, we must have a right inverse (1 )−1 such
that 1 ∪ (1 )−1 = Id, and hence it is a two-sided inverse by Theorem 2.1.15(i), (iii):

Id = 1 ∪ (1 )−1 ∪ 1 ∪ (1 )−1 = 1 ∪ (1 )−1 ∪ 1 ∪ (1 )−1

= (1 )−1 ∪ 1 ∪ 1 ∪ (1 )−1 = (1 )−1 ∪ 1 . (2.3.51)
Thus, we also deduce the statement at the beginning of the theorem, that either |1 | is even, or
2 · Id = 0. (This can also be deduced from the fact that 2 · (1 ∪ 1 ) = 0 if |1 | is odd.)
(ii), (iii) To deduce this, we use that the desired structures exist in positive degree and satisfy the
necessary axioms. Using the element 1 from part (i) (and Theorem 2.1.15), this result follows from
the following general proposition (see also the comments after the statement). 2
4
This is isomorphic to the normalized bar resolution.
Proposition 2.3.52. Let (V • , W • ) be a (pre)calculus and z ∈ V a homogeneous element. Then, there is a

unique extension of the calculus structure to the localization (V [ z−1 ]• , V [ z−1 ]• ⊗V • W • ), where by conven-
tion, z−1 ∧ z = 1.
Now, if A is a periodic Frobenius algebra, with a homogeneous element 1 ∈ HH1 inducing

the periodicity, then we claim that (HH• ( A ), HH• ( A )) = (HH0 ( A )[(1 )−1 ], HH0 ( A )[(1 )−1 ]). There
is clearly a map (HH0 ( A ))[(1 )−1 ] → HH• ( A ), which is an isomorphism in nonnegative degrees,
and must therefore be an isomorphism. As a result, we deduce that the calculus structure on
(HH0 ( A ), HH0 ( A )) extends uniquely to a calculus structure on (HH• ( A ), HH• ( A )).
We remark that this calculus is not periodic in a trivial way: it is not true that the Lie derivatives
or B must commute with i 1 . However, one can write formulas for all the operations in terms of
operations on degrees 0, 1, 2, . . . , |1 | − 1 and involving 1 .
Proof of Proposition 2.3.52. We use the notation of Definition 2.3.42, since we are discussing general
calculi and not only the Hochschild calculus. By definition, V [ z−1 ] is graded commutative. Further-
more, note that, since z ∧ z = (−1)|z| z ∧ z, either | z| is even, or V [ z−1 ] is an algebra over Z/2. Either
way, z is central in V [ z−1 ] (not merely graded-central), and we can omit any mention of (−1)|z| .
Let φ : V → V [ z−1 ] denote the localization map. Note that, if f ∈ ker(φ), i.e., f ∧ zk = 0 for
some k 0, then { f , g } ∧ zk+1 = 0 by the Leibniz rule, so { f , g } ∈ ker(φ). Let ψ : W → W [ z−1 ] :=
V [ z−1 ] ⊗V W denote the base-change map. Then, we have x ∈ ker(ψ) iff ιzk (x) = 0 for some k 1,
and similarly to the above, we deduce that La (x), d(x) ∈ ker(ψ) for all a ∈ V using the calculus iden-
tities. Similarly, for any y ∈ W , and any f ∈ ker(φ), we have L f ( y ) ∈ ker(ψ). Thus, it makes sense to
speak about the calculus structures as being defined on (φ(V ), ψ(W )), and our goal is to extend the
structure to (V [ z−1 ], W [ z−1 ]) and verify that the calculus identities are still satisfied.
For operators, we will use [−,−] to denote the graded commutator: [α , β] := α ◦ β − (−1)|α ||β| β ◦ α .
For example, [La , ιb ] := La ◦ ιb − (−1)(|a|−1)|b| ιb ◦ La .
We extend the Gerstenhaber bracket from φ(V ) as follows:

z−1 , g := − z−2 ∧ [ z, g ], (2.3.53)
together with skew-symmetry and the Leibniz rule. We must check that this yields a well-defined
bracket, which amounts to the computation

( f ∧ z) ∧ z−1 , g := z−1 ∧ [ f ∧ z, g ] + ( f ∧ z) ∧ z−1 , g

= [ f , g ] + z−1 ∧ f ∧ [ z, g ] − f ∧ z−1 ∧ [ z, g ] = [ f , g ]. (2.3.54)
It is easy to check that this yields a Gerstenhaber bracket, and we omit this.
We extend the Lie derivative L− to (V [ z−1 ]• , W [ z−1 ]• ) as follows. For f ∈ φ(V ), we extend the
operation L f to W [ z−1 ] using (2.3.43), with a := f and b := z, using the same procedure as above.
To define the derivative L f ∧z−1 we use (2.3.44) with a := f and b := z. It is straightforward that this
is well defined.
We must verify that the above gives a precalculus. We know that the identities are satisfied when
everything is in (φ(V ), ψ(W )). Thus, to verify that (2.3.43) holds, we need to show that the LHS
(graded)-commutes with ιz . This follows because ιb (graded)-commutes with ιz , and [La , ιz ] = ι[a,z] ,
which graded-commutes with ιb . We may then inductively show that (2.3.43) holds: first, if it holds
for any (a, b), it must hold replacing a by a ∧ z∧ j for any j ∈ Z by construction. Then, inductively, if
the identity holds for (a, b), we may deduce that it holds for (a, b ∧ z−1 ) using our definition of Lb∧z−1
(which is based on (2.3.44)).
To verify that (2.3.44) holds, it is enough to show that the identity for (a, b ∧ z) implies the identity
for (a, b). We may first prove this for a ∈ φ(V ), and then prove it for all a using the identity for ( z, a)
and hence for (a, z). That is, it suffices to prove that the identity for (a, b ∧ z) and (a, z) implies the
identity for (a, b). We have
La∧b ιz + (−1)|a|+|b| ιa∧b Lz = La∧b∧z = La∧z ιb + (−1)|a| ιa∧z Lb
= La ιb ιz + (−1)|a| ιa Lz ιb + (−1)|a| ιa∧z Lb
= La ιb ιz + (−1)|a|+|b| (ιa∧b Lz − ιa ι[b,z] ) + (−1)|a| (ιa ι[z,b] + ιa Lb ιz ),
(2.3.55)
which gives the identity upon cancelling the two inner terms and multiplying on the right by ι(z)−1 .
We can similarly verify that L− gives a Lie action. As before, it suffices to show that the identity
L[x, y] = [Lx , L y ] for (a, b ∧ z) and (a, z) implies the identity for (a, b). Since z is invertible, it suffices
to verify that [La , Lb ]ιz = L[a,b] ιz . Since (using the Leibniz rule and the fact that z is central)
L[a,b∧z] + (−1)(|a|+1)(|b|+1) Lb∧[z,a] = L[a,b]∧z = L[a,b] ιz + (−1)|a|+|b|+1 ι[a,b] Lz , (2.3.56)
it suffices to verify that the LHS equals the RHS after substituting the desired identity L[a,b] ιz =
[La , Lb ]ιz . That is, it suffices to prove that
L[a,b∧z] + (−1)(|a|+1)(|b|+1) Lb∧[z,a] = [La , Lb ]ιz + (−1)|a|+|b|+1 ι[a,b] Lz . (2.3.57)
We have

LHS = La , Lb ιz + (−1)|b| ιb Lz + (−1)(|a|+1)(|b|+1) Lb ι[z,a] − (−1)|a|(|b|+1) ιb [Lz , La ], (2.3.58)
RHS = [La , Lb ]ιz + (−1)|b| La ιb Lz − (−1)|a||b| ιb La Lz , (2.3.59)
from which (2.3.57) follows by expanding ι[z,a] = ιz La − La ιz in the first line, and making two pairwise
cancellations.
Next, we have to extend differential d. For this, we use the Cartan identity (2.3.45), with a := z (or
a power of z). We need to check that, with this definition, (2.3.45) holds, and that d2 = 0. We will
show that (2.3.45) holds applied to any element b ∈ W [ z−1 ]. First, we show that (2.3.45) must hold
when b ∈ ψ(W ). This amounts to the statement that

ι[z,b] = ιz , [d, ιb ] . (2.3.60)
We simplify the RHS as
def.
ιz , [d, ιb ] = [ιz , d], ιb = −(−1)|z| [Lz , ιb ] = (−1)|z|+|b|(|z|+1) ι[b,z] = ι[z,b] , (2.3.61)
as desired (the first step used [ιz , ιb ] = 0 by graded-commutativity). Next, we show that (2.3.45) must
hold for all b. This amounts to the statement that
[d, ιa∧z ] = [d, ιa ]ιz + (−1)|a| ιa [d, ιz ], (2.3.62)
which follows immediately using ιa∧z = ιa ιz .

Finally, to show that d2 = 0, it suffices to show that [d2 , ιz ] = 0, i.e., dLz + Lz d = 0. Actually, we
know that this identity holds when applied to ψ(W ), so it is enough to show that [dLz + Lz d, ιz ] = 0.
We have
[dLz + Lz d, ιz ] = −dι[z,z] + L2z + L2z − ι[z,z] d = −L[z,z] + 2L2z = 0, (2.3.63)
using at the end the fact that L− is an action. 2

Adding the CY Frobenius condition, we obtain the
Theorem 2.3.64.
(i) Let A be any periodic Calabi–Yau Frobenius algebra. Then, the stable Hochschild cohomology is a BV
algebra, with BV differential obtained by the duality (2.3.28) from the Connes differential. That is, the
differential := D ◦ B ◦ D, where D is the duality (2.3.28) and B the Connes differential, satisfies (1.1.23);
(ii) If A is only CY (m) Frobenius (and not necessarily periodic), and m 1, then we may still define :=
D ◦ B ◦ D in degrees m, and (1.1.23) is satisfied when |a|, |b| 0 and 1 |a| + |b| m.
Proof. (i) The proof is similar to the proof of Theorem 3.4.3 in [Gin06]. Namely, using (2.3.29)
and (2.3.43), we have
{a, b} = Di {a,b} D(1) = a ∪ DLb D(1) − (−1)|a|(|b|+1) DLb D(a). (2.3.65)
Now, using (2.3.45) and (2.3.44), we have
RHS = a ∪ (b) − (−1)|b| a ∪ b ∪ (1) − (−1)|a|(|b|+1) (b ∪ a) + (−1)|a||b| b ∪ (a). (2.3.66)
Using graded-commutativity, this immediately gives (1.1.23).

(ii) The above proof goes through in the general Calabi–Yau Frobenius case in the degrees indi-
cated. Note that we needed |a|, |b| 0 because we used (2.3.43) applied to a and b. 2
Remark 2.3.67. In fact, for any graded-commutative algebra V • , giving the structure of calculus using
W • := V m−• which satisfies the intertwining property (2.3.29) (for D the tautological isomorphism)
is equivalent to giving V • a BV algebra structure. The above theorem showed that calculus + duality
gives BV; the other direction is as follows: The intertwining property (2.3.29) uniquely specifies what
the module structure of V • on W • is, and the differential then gives L. One may then deduce the
remaining identities from 2 = 0 and the BV identity (1.1.23). The identity L[x, y ] = [Lx , L y ] says
precisely that 2 is a derivation5 ; (2.3.43) says is a differential operator of order 2, together
with the BV identity (i.e., that a ⊗ b → (−1)|a|+1 [a, b] is the principal symbol of ). Then, (2.3.44) is
a consequence of (2.3.45) (and it is a proof that being an operator of order 2 yields the Leibniz
rule for its principal symbol).
Put together, any periodic Calabi–Yau Frobenius algebra has Hochschild cohomology which is a BV
algebra and a Frobenius algebra (in the derived category), and together with Hochschild homology
forms a periodic calculus (together with an isomorphism between the two that intertwines cup prod-
uct with contraction). Moreover, the shift functor Ω −m is a (right) Serre functor for the category of fg
left modules.
In the case of (generalized, centrally extended) preprojective algebras (Examples 2.3.10, 2.3.12,
2.3.13) over C (or any field for the ordinary preprojective algebras), the formula for the extension
of B is quite simple, as we will prove
B • = (−1)• Dk ◦ B ∗−2−• ◦ Dk , (2.3.68)
B −1 = 0. (2.3.69)
This says that B is graded self-adjoint. As a consequence, so is the BV differential . This motivates
the
5
It is worth remarking, by comparison, that, for a graded-commutative algebra with an odd differential operator of or-
der 2, the Jacobi identity for its principal symbol ±[ , ] says that 2 is a differential operator of order 2 (so being a
derivation is between this and 2 = 0). Skew-symmetry of [ , ] is automatic.
Definition 2.3.70. A BV Frobenius algebra is a Z-graded Frobenius algebra H (= a Z-graded algebra

whose graded components are fg projective over k, and with a homogeneous invariant inner product
of some fixed degree) together with a graded self-adjoint differential • : H • → H •−1 of degree −1,
i.e.,
• = (−1)• †• , (2.3.71)
†
with • the adjoint of • .
This leads to the
Question 2.3.72. Are the formulas (2.3.68), (2.3.69) valid for any periodic Calabi–Yau Frobenius algebra?
More generally:
Question 2.3.73. For which Frobenius algebras A does (2.3.68), (2.3.69) define an extension of the usual
calculus to (HH• , HH• )?
If the answer to Question 2.3.72 is positive, then the stable Hochschild cohomology of any periodic
CY Frobenius algebra is BV Frobenius. If the answer to 2.3.73 includes CY Frobenius algebras, then one
does not need the periodicity assumption.
To answer the above questions, we suggest to work on the level of standard Hochschild chains.
Consider the two-sided resolution N̂ • of A given by N • A → ( N •∨ ⊗ A A ∗ ), where N • is the normal-
ized bar resolution. Then, the chain complex N̂ • ⊗ A e A has the form
· · · → A ⊗ Ā ⊗ Ā → A ⊗ Ā → A → A ∗ → A ∗ ⊗ ( Ā )∗ → A ∗ ⊗ ( Ā )∗ ⊗ ( Ā )∗ → · · · . (2.3.74)
Then, the two-sided Connes differential B̂ should be given, on chains, by
⎧
⎨ B, i 0,
B̂ i := 0, i = −1, (2.3.75)
⎩
(−1)i B ∗ , i −2.
We hope to address this in a future paper.
2.4. Hochschild cohomology of symmetric algebras is BV
In this subsection, we give a simple proof that a symmetric Frobenius algebra over an arbitrary
base ring k has ordinary Hochschild cohomology which is BV. This is based on using the formulas
(2.3.68), (2.3.69) to extend B (since the algebra need not be periodic, we cannot use Theorem 2.3.64).
Actually, we show this more generally for “stably symmetric” algebras, i.e., A A ∨ .
First, let us extend the Lie derivative operation to all of (HH• , HH• ), in the spirit of (2.3.68),
(2.3.69):
Lb := B̂i b − (−1)|b| i b B̂ . (2.4.1)
We claim that (i a Lb − (−1)|a|(|b|+1) Lb i a )(x) = i {a,b} (x), when a, b ∈ HH0 ( A , A ) and x ∈ HH−1 ( A , A ).
We have (using graded commutators)

Lb (x) = (−1)|x|·(|b|+1) Dk (−1)|b| B ∗ i b∗ − i b∗ B ∗ Dk (x) = −(−1)|x|·(|b|+1) Dk i b∗ , B ∗ Dk (x). (2.4.2)
The signs above use the identity c · |x| + d · (|x| + c ) = d · |x| + c · (|x| + d) = (c + d) · |x| + c · d (setting
c = 1, d = |b| or vice-versa). We then obtain
∗
[ia , Lb ](x) = (−1)|x|·(|a|+|b|+1) Dk i b∗ , B ∗ , ia∗ Dk (x) = (−1)|x|·(|a|+|b|+1) Dk ia , [ B , i b ] Dk (x)
= (−1)|x|·(|a|+|b|+1) Dk i ∗{a,b} Dk (x) = i {a,b} (x). (2.4.3)
But, as in the proof of Theorem 2.3.64, this identity immediately gives the BV identity (1.1.23), letting
x = D(1) ∈ HH−1 ( A , A ). Note also that, by definition, B̂ D(1) = 0, so that (1) = 0.
3. Hochschild (co)homology of ADE preprojective algebras over any base
As mentioned in Example 2.3.10, the preprojective algebra of a quiver of type ADE is a peri-
odic Calabi–Yau Frobenius algebra. In this section, we explicitly describe its Hochschild (co)homology
over an arbitrary base (including positive-characteristic fields). In characteristic zero, this has al-
ready been done in [EE07,Eu06]. We also review and simplify the algebraic structures on Hochschild
(co)homology, and prove that HH• is BV Frobenius (verifying Question 2.3.72 in this case). Finally, in
Section 3.3, we explicitly describe the maps relating the Hochschild (co)homology of Dynkin prepro-
jective algebras with the extended Dynkin case; in the latter case, the Hochschild cohomology groups
were described over k = C in [CBEG07] and later over Z in [Sch07].
The new theorems proved here are: Theorem 3.2.7, which computes the Hochschild (co)homology
of the Dynkin preprojective algebras over Z and proves that it is a BV Frobenius algebra, and The-
orem 3.3.4, which compares the Hochschild cohomology of the Dynkin and extended Dynkin pre-
projective algebras over Z. We also restate in a simplified form two theorems from [EE07,Eu07a]
(Theorems 3.1.2 and 3.1.13).
We first recall the definition of the preprojective algebras Π Q of ADE type.
Notation 3.0.1. We will use [n] to denote the degree-n-component of a graded vector space (discarding
other degrees), and (n) to denote shifting a graded vector space by degree n. In particular, A [m] is
a graded vector space concentrated in degree m, and ( A (n))[m] = ( A [m − n])(n). The vector space
A [m](−m) is concentrated in degree zero.
Notation 3.0.2. We will use deg( z) to denote the degree of an element z in a graded algebra or
module. This is to distinguish with Hochschild degree, where we denote |a| = m if a ∈ HHm ( A , M ) (so
deg(a) would denote the degree with respect to the grading on A and M). The notation [m], (m) refer
to the deg(−) grading, and never to Hochschild degree.
Let k be a commutative ring. Let Q be a quiver of ADE type with vertex set I . By convention,
Q also denotes the edges of the quiver. Let Q := Q Q ∗ be the double quiver, where Q ∗ := {a∗ | a ∈
Q } is the quiver obtained by reversing all arrows (a∗ is the reverse of a).
Let P Q , P Q be the associated path algebras over k, and let Π Q := P Q /((r )) with r := a∈ Q [a, a∗ ].
Let e i denote the image of the vertex i for any i ∈ I .
Recall from Example 2.3.10 that Π Q is a Frobenius algebra over k (in fact, periodic Calabi–Yau).
Let ( , ) denote an invariant inner product, and let η be the Nakayama automorphism of Π Q , so that
(x, y ) = (η−1 ( y ), x). Recall [RS,ES98a,Eu06] that we may choose ( , ), η such that
η(e i ) = e η̄(i) , defining η̄ : I → I by (3.0.3)
αη̄(i) = − w 0 (αi ); (3.0.4)
here, w 0 is the longest element of the Weyl group of the root system attached to Q , and αi , i ∈ I are
the roots. Furthermore, η may be uniquely chosen to act on Q ⊂ P Q so that

− Q , if Q is of type D , E,
η Q∗ ⊂ Q and η( Q ) ⊂ (3.0.5)
Q, if Q is of type A ,
since Q is a tree. As a consequence of these formulas, we see that η and η̄ are involutions. We
remark that η is always nontrivial, even though, for D 2n+1 , E 7 and E 8 , η̄ becomes trivial. (Except,
over characteristic 2, η is trivial for D 2n+1 , E 7 , and E 8 .)
Let m1 , m2 , . . . , m| I | be the exponents of the root system attached to Q , in increasing order. Let
h := m| I | + 1 be the Coxeter number.
Recall that the Hilbert series of a Z0 -graded vector space M is defined to be h( M ; t ) :=
m0 dim M [m ] t m
. If E is a Z0 -graded k I -module for any field k, then we define the matrix-

valued Hilbert series of E, h( E ; t ), by h( E ; t )i j := m0 dim E [m]i , j t
d
, where E [m]i , j := e i E [m]e j ,
where e i , e j ∈ k are the idempotents corresponding to the vertices i , j ∈ I .
I
3.1. Reminder of characteristic zero results
Let A := Π Q , and assume that k is a characteristic-zero field. We may then describe the Hochschild
homology, H H ∗ ( A ), and cohomology, H H ∗ ( A ), as follows:
Definition 3.1.1. (See [EE07].) Let U := ( H H 0 ( A )[< h − 2])(2), L := H H 0 ( A )[h − 2](−(h − 2)), K :=
H H 2 ( A )(2), and Y := H H 6 ( A )[−h − 2](h + 2). Also, let P ∈ End(k I ) be the permutation matrix corre-
sponding to the involution η̄ . Let I η̄ be the set of vertices fixed by η̄ .
Theorem 3.1.2 (Restated). (See [EE07].)
(i) U has Hilbert series

h (U ; t ) = t 2mi . (3.1.3)
i: mi < h2
(ii) We have natural isomorphisms K ∼ = ker( P + 1) and L ∼

η̄
= kI .
(iii) As graded vector spaces, one has
HH0 ( A , A ) ∼
= U (−2) ⊕ Y (h − 2), HH1 ( A , A ) ∼
= U (−2), HH2 ( A , A ) ∼
= K (−2), (3.1.4)
∗
6+ i
HH ( A, A) ∼ i
= HH ( A , A )(−2h), HH ( A , A )(2) ∼
i
= HH 5− i
( A , A )(2) , (3.1.5)
HHi ( A , A ) ∼
= HH2−i ( A , A )(2), (3.1.6)
H H ( A) ∼
0
= U (−2) ⊕ L (h − 2), H H 0( A) ∼
=k . I
(3.1.7)
3.1.1. The cup product

Let us summarize also the cup product structure, which was computed in [ES98a,Eu08]. We explain
it using our language and results. In view of the first isomorphism of (3.1.5) and Theorem 2.3.47, it is
enough to consider cup products among elements of Hochschild degrees between 0 and 5.
Since the Calabi–Yau Frobenius dimension is 2 of shift 2, by Theorem 2.3.27, the Hochschild coho-
mology is a Frobenius algebra with pairing of Hochschild degree −5 (meaning, ( f , g ) = 0 implies that
| f | + | g | = 5 in Hochschild degree), and of graded degree 4 (meaning, in graded degree, ( f , g ) = 0
implies deg( f ) + deg( g ) = −4). In particular, for all i ∈ Z, the composition
∪ ( ,Id)
HHi ( A , A ) ⊗ HH5−i ( A , A ) −→ HH5 ( A , A ) −−−→ k (3.1.8)
is a perfect pairing of graded degree 4, the same as the second isomorphism of (3.1.5).
Moreover, if | f | + | g | + |h| = 5 (in Hochschild degree), using the graded-commutativity of cup

product, we have
( f ∪ g , h) = (−1)| g |·|h| ( f ∪ h, g ) = (−1)| f |·(| g |+|h|) ( g ∪ h, f ), (3.1.9)
and since the pairing is perfect, we see that knowing the cup product HH| f | ⊗ HH| g | → HH5−|h| deter-
mines the cup product in the other two pairs of Hochschild degrees, (| f |, |h|) and (| g |, |h|). That is,
we may divide the cup products into the unordered triples summing to 5 modulo 6:
(0, 0, 5), (0, 1, 4), (0, 2, 3), (1, 1, 3), (1, 2, 2), (1, 5, 5), (2, 4, 5), (3, 3, 5), (3, 4, 4), (3.1.10)
and the cup product in any fixed two Hochschild degrees of a triple determines the other two pairs
of cup products.
The first triple above corresponds to multiplication in the center Z ( A ), via the quotient Z ( A )
HH0 ( A , A ) which performs U (−2) ⊕ L (h + 2) U (−2) ⊕ Y (h + 2) (see [ES98a,Eu08] for an explicit
computation of this multiplication). Then, the next two triples describe HH1 ( A , A ) and HH2 ( A , A )
as HH0 ( A , A )-modules. As explained in [ES98a,Eu08], HH1 ( A , A ) is cyclic as an HH0 ( A , A )-module
(generated by the Euler vector field), and since K is concentrated in graded degree zero, the structure
on HH2 ( A , A ) is the obvious (trivial) one: it is a k-vector space.
The cup products between Hochschild degrees (1, 1) and (3, 3) are trivial for graded degree rea-
sons. For types D , E , A 2n+1 , the cup product is also trivial in Hochschild degrees (3, 4) for degree
reasons, and in degrees (2, 4) by an argument using the BV identity (see [Eu08]—the argument only
shows that the cup product is h-torsion, and in fact it appears to be nontrivial for type D 2n+1 in
characteristic two). In type A 2n , the cup product between degrees (3, 4) and (2, 4) is nontrivial and
can be explicitly computed (see [ES98a]; see also the similar type T case in [Eu07b]). When nontriv-
ial, the products between degrees (3, 4) and (2, 4) are only between the lowest possible degrees (so,
it reduces to a pairing of vector spaces concentrated in bottom degree, which in fact has rank one
since the bottom-degree part in Hochschild degree 4 has dimension one).
This leaves only the cup products (1, 2, 2) and (1, 5, 5). These are best described as cup products
HH2 ( A , A ) ⊗ HH2 ( A , A ) → HH4 ( A , A ) ∼

= HH1 ( A , A )∗ , (3.1.11)
HH5 ( A , A ) ⊗ HH5 ( A , A ) → HH4 ( A , A ) ∼

= HH1 ( A , A )∗ . (3.1.12)
Here, we obtain a nondegenerate symmetric bilinear pairing α on K , and a symplectic pairing β

on Y , respectively (one must obtain some symmetric and skew-symmetric bilinear pairings on K
and Y , respectively, since K and Y are concentrated in degree zero, and U has k-dimension equal to
one in each graded degree; nondegeneracy is then a result of explicit computations in [ES98a,Eu08]
([Eu08, Theorem 4.0.8] for types D , E; throughout [ES98a, part II] for type A)).
3.1.2. The Connes and BV differentials

Using the dualities and intertwining properties, one immediately obtains the contraction maps.
It remains only to compute the Connes differential, which yields the BV differential by duality, and
then using the Cartan and BV identities, one immediately computes the Lie derivatives and Gersten-
haber bracket.
We reprint the Connes differential from [Eu07a]. Let ( zk ), (ωk ) be homogeneous bases for
U (−2), Y (h − 2) ⊂ HH0 ( A , A ), respectively, with deg( zk ) = k. Let (θk ) ⊂ U (−2) ⊂ HH1 ( A , A ) be a ho-
mogeneous basis for HH1 ( A , A ) with deg(θk ) = k, and let ( f k ) ⊂ K ⊂ HH2 ( A , A ) be a basis. We will
write ( f k∗ ), (θk∗ ), ( zk∗ ), (ωk∗ ) for the dual bases of HH3 ( A , A ), HH4 ( A , A ), and HH5 ( A , A ). (Using dual
notation is where we diverge from [Eu08,Eu07a]). Let us abusively identify these elements with their
images under periodicity and D. So, for example, θk may denote the corresponding element in any
group HH1+6s or HH1+6s for any s ∈ Z. (This also differs from the notation of [Eu08,Eu07a].)
Theorem 3.1.13. (See [Eu07a, Theorem 5.0.10].) The Connes differential B • : HH• ( A , A ) → HH•+1 ( A , A ) is
given by

0 = B 6s = B 4+6s = B 2+6s (U ) = B 3+6s Y ∗ , (3.1.14)

k k
B 1+6s (θk ) = 1 + + sh zk , B 3+6s zk∗ = (s + 1)h − 1 − θk∗ , (3.1.15)
2 2

1
B 2+6s (ωk ) = + s hβ −1 (ωk ), (3.1.16)
2

B 5+6s f k∗ = (s + 1)hα −1 f k∗ , (3.1.17)
where in the first line, B 2+6s (U ) means the image of the summand of U (2hs) under B 2+6s , and similarly for
B 3+6s (Y ∗ ).
Hence, the same formulas are valid for HH• where now s ∈ Z is arbitrary. As a consequence of
writing it this way, using the symmetry and skew-symmetry of α , β , respectively, it is easy to verify
the (new)
Corollary 3.1.18. The Connes differential B is graded self-adjoint with respect to the duality Dk . Hence, the
same is true for the BV differential .
Each line of (3.1.14)–(3.1.17) verifies B i = (−1)i B ∗−2−i for the concerned summands of H H i .
Remark 3.1.19. In the generalized Dynkin case of type T n (Example 2.3.13), a similar observation to
the above, together with the computation of B found in [Eu07b], shows that B is graded self-adjoint
in the T n case, and hence so is , i.e., HH• is BV Frobenius.
3.2. Extension of results to Z and arbitrary characteristic
Now, we explain the general Z-structure of Hochschild (co)homology. Note that, by the Universal
Coefficient Theorem, one may immediately deduce the k-module structure from this for any k; we
explain it for fields F p (with p prime) to see the duality. We will also see that the stable Hochschild
cohomology HH• is BV Frobenius over any base field, in Theorem 3.2.7(v) below, and give the com-
plete structure of over any field.
Definition 3.2.1. We define (and redefine) the vector spaces T , U , K , K , Y , Y , L by

K := HH0 ( A , A )[0], K := Torsion HH−1 ( A , A )[0] , T := HH0 ( A , A )[> 0], (3.2.2)

Y := HH−1 ( A , A )[h − 2] −(h − 2) , Y := Torsion HH0 ( A , A )[h − 2] −(h − 2) , (3.2.3)

U (−2) := HH0 ( A , A ) < (h − 2) , L := HH0 ( A , A )[h − 2] −(h − 2) . (3.2.4)
Let T ∗ (abusively) denote the graded Z-module

T ∗ := H 1 T L ∗ ∼
= HomZ ( T , Q/Z), (3.2.5)
where L ∗ denotes the derived dual, and H 1 the first cohomology.

Note that, after tensoring by C, the vector spaces U , K , Y , L above become (isomorphic to) the
ones defined previously. Also, K , K , Y , Y , L are all concentrated in degree zero (by construction),
hence the shifts in the definition. On the other hand, U , T , T ∗ live in multiple degrees. Also, note
that, by e.g. the Universal Coefficient Theorem, when torsion appears in the cohomology over Z, then
if we work instead over F p , the torsion is tensored by F p and replicated in degree one lower (one
higher for homology). The free part carries over (tensored by F p of course).
Notation 3.2.6. Let the bad primes consist of 2 for Q = D n , 2, 3 for Q = E 6 or E 7 , and 2, 3, 5 for
Q = E 8 , and no others.
Theorem 3.2.7.
(i) The modules T , T ∗ are finite with torsion elements of order equal to bad primes, and K , Y are finite with
torsion of order dividing the Coxeter number h. The modules K , Y , L , U are free over Z.
(ii) We have
HH0 ( A , A ) ∼
= U (−2) ⊕ Y (h − 2) ⊕ Y (h − 2), HH1 ( A , A ) ∼
= U (−2), (3.2.8)
HH ( A , A ) ∼
2
= K (−2) ⊕ T (−2), HH ( A , A ) ∼
3
= K (−2) ⊕ K (−2),
(3.2.9)
HH4 ( A , A ) ∼
= U ∗ (−2) ⊕ T ∗ (−2), HH5 ( A , A ) ∼
= U ∗ (−2) ⊕ Y (−h − 2), (3.2.10)
HHi ( A , A ) ∼
= HH 2− i
( A , A )(2), HH 6+ i ∼ i
= HH (−2h), (3.2.11)
HH0 ( A , A ) ∼
= U (−2) ⊕ L (h − 2), HH0 ( A , A ) ∼
= ZI ⊕ T . (3.2.12)
(iii) For any prime p, letting M p := M ⊗ F p for all M, we have
HH0 ( A p , A p ) ∼
= U p (−2) ⊕ Y p (h − 2) ⊕ Y p (h − 2), HH1 ( A p , A p ) ∼
= U p (−2) ⊕ T p (−2), (3.2.13)
∗
HH2 ( A p , A p ) ∼
= K p (−2) ⊕ K p (−2) ⊕ T p (−2), HH5−i ( A p , A p )(2) ∼
= HHi ( A p , A p )(2) , (3.2.14)
HHi ( A p , A p ) ∼
= HH2−i ( A p , A p )(2), HH6+i ∼
= HHi (−2h), (3.2.15)
HH0 ( A p , A p ) ∼
= U p (−2) ⊕ L p (h − 2), HH0 ( A p , A p ) ∼
= F Ip ⊕ T p . (3.2.16)
(iv) For any (bad) prime p, T p∗ is cyclic as an HH0 ( A p , A p )-module.

(v) The BV differential (and hence the Connes differential) is graded self-adjoint over any field k. In partic-
ular, for k = F p , is zero restricted to any summand K p , Y p , T p , T p∗ .
The remainder of this subsection is devoted to the proof of the theorem.
Lemma 3.2.17. The center of A does not increase over positive characteristic. That is, the map Z ( A ) ⊗ F p →
Z ( A ⊗ F p ) is an isomorphism for all ADE quivers.
We omit the proof of the above lemma, which was done using MAGMA for type E, and a straight-
forward explicit computation in the A n , D n cases, using bases in terms of paths in the quiver. Note
that the lemma is actually true for all quivers, since the non-Dynkin case is proved in [CBEG07,Sch07].
Corollary 3.2.18. The groups HH1 ( A , A ), HH5 ( A , A ), HH1 ( A , A ), and HH3 ( A , A ) are all free Z-modules.
Proof. It is enough to show that they are torsion-free. For HH1 , this follows from the fact that the
differential d0 in C • ( A , A ) must have saturated image (otherwise HH0 ( A , A ) would increase in some
positive characteristic); alternatively, this is a consequence of the universal coefficient theorem. For
∼
HH5 ( A , A ), we use the derived duality Dk : C • ( A , A ) −→ C 5−• ( A , A )∗ , so that the differential d0
corresponds to d∗4 in the latter (again, we could also use the universal coefficient theorem). Then,
the duality D of dimension 2 and the periodicity 6 gives the results for Hochschild homology:
HH j ∼
= HH6n+2− j whenever j , 6n + 2 − j are both positive. 2
Using the duality D and the periodicity by period 6, it remains only to compute the torsion
of HH2 ( A , A ), HH3 ( A , A ), HH4 ( A , A ), and HH6 ( A , A ). Using the duality Dk , HH2 and HH4 must have
dual torsion, so it is really enough to compute in degrees 2, 3, and 6. We will see that the torsion
in these degrees will be nontrivial, but only in bad primes (2 for D n , 2, 3 for E 6 and E 7 , and 2, 3, 5
for E 8 ), and primes dividing the Coxeter number h.
First, by the duality D, the torsion of HH0 ( A , A ) and HH2 ( A , A ) are isomorphic. Since HH0 ( A , A )
has no torsion in degree zero, and the inclusion HH0 ( A , A ) ⊂ HH0 ( A , A ) is full in nonzero degrees,
the torsion of HH0 ( A , A ) (and hence of HH2 ( A , A )) is the same as that of HH0 ( A , A ). The latter was
computed in [Sch07], and we collect results for convenience:
Proposition 3.2.19. (See [Sch07, Theorem 4.2.60].) The module HH0 ( A , A ) ∼

= Z I ⊕ T , where T := HH0 ( A , A )+
is finite and given as follows:
• For Q = An , T = 0,
• For Q = D n ,

T∼
= Z/2(m). (3.2.20)
4|m, 0<m2(n−2)
• For Q = E n , T is a ( finite) direct sum of shifted copies of Z/2 and Z/3, and in the case n = 8, also of Z/5.
In particular:
T E6 ∼
= Z/2(4) ⊕ Z/3(6), (3.2.21)

T E7 ∼
= Z/2(4) ⊕ Z/2(8) ⊕ Z/2(16) ⊕ Z/3(6), (3.2.22)

T E8 ∼
= Z/2(4) ⊕ Z/2(8) ⊕ Z/2(16) ⊕ Z/2(28) ⊕ Z/3(6) ⊕ Z/3(18) ⊕ Z/5(10). (3.2.23)
Moreover, for any Q and any bad prime p, there exists a top-degree torsion element, r p ,top , such that all
homogeneous p-torsion elements [x] have the property that [x · z] = r p ,top for some homogeneous central
element z ∈ HH0 (Π Q , Π Q ).
We deduce immediately that T ∗ ⊗ F p is cyclic under dual contraction, and hence (by (2.1.33)) also
under contraction. By the intertwining property (2.3.29), we deduce part (iv) of Theorem 3.2.7.
It remains only to compute the torsion of HH3 ( A , A ) and HH6 ( A , A ), i.e., to compute K and Y
and verify that there is no other torsion. We will use some results of [Eu08] for this, but let us explain
them using our language. Using the formulation of Definition 3.2.1 and the (normalized) bar complex,
it suffices to compute the torsion of the cokernels of the maps
d−1
= A −−→ A ∗ ∼
C0( A, A) ∼ = A η−1 −−−→ A ∼
C −1 ( A , A ) ∼
d0
= C −1 ( A , A ), = C 0 ( A , A ). (3.2.24)

To express these maps, write Id = i x∗i ⊗ xi ∈ A ∗ ⊗ A. Then, the map A ⊗ A → A ∗ ⊗ A in the nor-
is given by (x ⊗ y ) → xy · Id. So, the “conorm” differential d0 in (3.2.24) must
malized bar resolution
be given by y → i xi yx∗i ∈ A ∗ . Since we may assume deg(xi ) = − deg(x∗i ), the image can only be

nonzero if deg( y ) = 0. Similarly, the “norm” differential d−1 is given by d−1 ( y ) = i x∗i yxi ∈ A, now
∗
viewing xi as an element of A via ( , ). Here also, only deg( y ) = 0 need be considered. We deduce
Proposition 3.2.25.
(i) The conorm, d0 , and norm, d−1 , maps are given by

d0 ( y ) = xi yx∗i ∈ A ∗ , d−1 ( y ) = x∗i yxi ∈ A , (3.2.26)
i i
where

Id = x∗i ⊗ xi ∈ A ∗ ⊗ A . (3.2.27)
i
The image of the conorm and norm maps d0 , d−1 must lie in the top degree of A ∗ , A, respectively.
(ii) Identifying A ∗ with A using ( , ), and letting ωi ∈ e i Ae η̄(i ) denote the image of e ∗i ∈ A ∗ , we have

d0 (e i ) = tr(Id|e j Ae i )ω j , d−1 (e i ) = δ j ,η̄( j) δi ,η̄(i ) tr(η|ei Ae j )ω j . (3.2.28)
j∈ I j∈ I
Proof. Part (i) has already been proved and is more generally true for any (graded) Frobenius alge-
bra A. We show part (ii), using the fact that A [0] ∼ = k I as a subalgebra, with η acting by the permu-
tation η̄ . Let f : A → k be the function such that (a, b) = f (ab). We have that f∗(e i , ωi ) = f (ωi ) = 1
compute d∗0 (e i ), it is enough to find f (e j d0 (e i )) = f (
(e j x
e i x
)) for all j ∈ I . This
for all i. Thus, to
is the same as
(e j x
e i , x
), which is the trace of the projection A e j Ae i . Similarly, we have

e j x∗
e i , x
= η(e i x
)e j , x∗
= tr x → e η̄(i) η(x)e j = tr(η|ei Ae j )δ j,η̄( j) δi,η̄(i) . 2 (3.2.29)

We note that in [Eu08], sums such as (3.2.26) (with xi a basis) are used to describe K and Y ; the
proposition above explains their origin through norm and conorm maps. In particular, bases are not
needed, and under the connectivity assumption A [0] ∼ = k I of (ii), one can re-express the sum as a
trace. We believe that the necessity of using such formulas to describe usual Hochschild cohomology
gives further justification for studying stable Hochschild cohomology.
Using the proposition, to compute K and Y , it suffices to compute two matrices: an I × I -
η
matrix, H Q , whose entries are ( H Q )i j = tr(Ide i Ae j ), and an I η̄ × I η̄ -matrix, H Q , whose entries are
η
( H Q )i j = tr(ηei Ae j ), where I η̄ := {i ∈ I: η̄(i ) = i }. These matrices were computed in the D , E cases
in [Eu08]. In fact, H Q itself was originally computed for all Dynkin cases in [MOV06]: since A is
free over k, H Q = h( A ; 1) is the Hilbert series matrix h( A ; t ) evaluated at t = 1. In [MOV06] is the
following formula for h( A ; t ):
−1
h( A ; t ) = 1 + P t h 1 − Ct + t 2 . (3.2.30)
η
So H Q = (1 + P )(2 − C )−1 . These are then easy to compute. It is also not difficult to compute H Q ,
which we omit, since the D , E cases are already in [Eu08], and the A case is easy. We obtain the
following (for k = Z):
Proposition 3.2.31. K and Y are zero if Q = A n , and otherwise are given by
⎧ ⊕2 n −2
⎪ Z/2) 2
⎪
⎪
, Q = Dn ,
⎨ ⊕2 ,
(Z/ 2) Q = E6,
K ∼
= (3.2.32)
⎪
⎪ (Z/2)⊕6 , Q = E7,
⎪
⎩
(Z/2)⊕8 , Q = E8,
⎧
⎪ Z/2, Q = D n , n even,
⎪
⎪
⎨ Z/(n − 1) = Z/(h/2), Q = D , n odd,
Y ∼
n
= (3.2.33)
⎪ Z/3, Q = E 7 ,
⎪
⎪
⎩
0, otherwise.
This proves (i) of Theorem 3.2.7. At this point, (ii) and (iii) are immediate from the dualities and
the Universal Coefficient Theorem.
It remains to prove part (v). Since we already know (by Corollary 3.1.18) the result for characteristic
zero, it suffices to take k = F p for some prime p. As before, let 1 ∈ HH0 be the identity, and 1 ∈
HH6 induce the periodicity. First, we note that (1) = 0 and (1 ) = 0 because this is true over Z
(by [Eu07a]), and moreover that {(1 )i , (1 ) j } = 0 for all i , j for the same reason. Hence, L 1 is also
graded self-adjoint, and it suffices to verify that ( (a), b) = (−1)|a| (a, (b)), when 0 |a|, |b| 5,
and hence either |a| + |b| = 6 or |a| = |b| = 0. To do this, we will show that kills summands of the
form K p , Y p , T p and T p∗ (the second statement of (v)).
It is clear, for graded degree reasons, that kills summands of the form Y , K , T , T ∗ over k = Z.
It remains to show that the new summands appearing over F p are also killed. For K in HH2 , this is
true for degree reasons and the fact that the kernel of on HH1 [0] is zero (from the characteristic-
zero case), using that 2 = 0. For degree reasons, the summand of T p∗ (−2) in HH3 and the summand
of Y (−h − 2) in HH5 must be killed.
It remains only to show that the summand of T p (−2) in HH1 is killed. Note first that there is
an element H Eu ∈ HH1 (Π Q ), the “half-Euler vector field,” whose action on closed paths (which must
have even length since Q is a tree) is to multiply by half the path-length. It follows from [Eu08]
(cf. [Sch05], §10 for the extended Dynkin case and the following subsection) that H Eu generates
HH1 (Π Q ) as an HH0 (Π Q )-module in the case that k = Z. Now, over k = F p , U (−2)⊗F p is isomorphic
to a direct summand U p (−2) of HH1 (Π Q ). From the argument in [Eu07a], we know that the operator
|HH1 ( A , A ) acts on U p (−2) by (zH Eu ) = ( deg2(z) + 1)z, for all z ∈ HH0 (Π Q )[< h − 2].
We claim that the Lie derivative L H Eu acts on HH• by multiplication by half the graded degree.
For HH• , this follows from the explicit formula for the Lie derivative (as argued in [Eu07a]), and then
this extends easily to HH• by taking the unique extension guaranteed by Proposition 2.3.52 (using the
construction given in the proof). It follows from this that, under the Gerstenhaber bracket, ad( H Eu )
acts on HH• by multiplication by half the graded degree.
From the identity (1.1.23) and the fact that (1) = 0, we have
deg(x)
( H Eu ∪ x) − ( H Eu ) ∪ x + H Eu ∪ (x) = { H Eu , x} = x, (3.2.34)
2
for any homogeneous x ∈ HH1 . Since ( H Eu ) = 1 and |HH2 = 0, we obtain

deg(x)
H Eu ∪ (x) = + 1 x. (3.2.35)
2
For any choice of splitting of T p in HH1 , the structure of T as given in Proposition 3.2.19 shows that
deg(x)
2
+ 1 = 0 as an element of F p , for all x ∈ T p . Hence, we have
H Eu ∪ (x) = 0. (3.2.36)
Since, for degree reasons, (x) is in the U p (−2)-summand of HH0 , we deduce that (x) = 0. This
completes the proof of Theorem 3.2.7.
3.3. The maps between the extended Dynkin and Dynkin preprojective algebras
In this subsection, we will interpret K , Y , and T in terms of the preprojective algebras of the cor-
responding extended Dynkin quivers. For this, we use the projection π : Π Q̃ Π Q and functoriality
of Hochschild homology. We need to recall a few results from [Sch07] first.
Let Q be an ADE quiver and let Q̃ be the corresponding extended Dynkin quiver. By [CBEG07,
Sch07], we know that Π Q̃ is (ordinary) Calabi–Yau of dimension 2, and in particular has Hochschild
dimension 2. Let Z Q̃ := HH0 (Π Q̃ , Π Q̃ ) and Z Q̃ ,+ := Z Q̃ [ 1]. The ring Z Q̃ is closely related to the
Kleinian singularity ring: one has Z Q̃ ⊗ C ∼
= C[x, y ]Γ where Γ is the group corresponding to Q̃ under
2π i
the McKay correspondence, and one can even replace C with Z[ |Γ1 | , e |Γ | ]. There is a standard integral
presentation of C[x, y ]Γ which actually describes Z Q̃ over Z (see, e.g., [Sch07, Propositions 6.4.2, 7.4.1,
and 8.4.1]).
By [Sch07, §10.1], we know that, as graded Z-modules,
HH0 (Π Q̃ , Π Q̃ ) ∼
= Z I ⊕ Z Q̃ ,+ ⊕ T , (3.3.1)
HH1 (Π Q̃ , Π Q̃ ) ∼
= Z Q̃ (2) ⊕ Z Q̃ ,+ , (3.3.2)
HH2 (Π Q̃ , Π Q̃ ) ∼
= Z Q̃ (2), HH ∼i
= HH2−i . (3.3.3)
Theorem 3.3.4. The induced maps π∗,i : HHi (Π Q̃ , Π Q̃ ) → HHi (Π Q , Π Q ) are given as follows:
(0) π∗,0 is an isomorphism on Z I ⊕ T , and kills Z Q̃ ,+ ;

(1) π∗,1 is a surjection Z Q̃ (2) U , with kernel the elements of degree h (and killing the second fac-
tor, Z Q̃ ,+ );
(2) π∗,2 : Z Q̃ (2) → U ⊕ Y (h) ⊕ Y (h) is a surjection onto U ⊕ Y (h), killing Z Q̃ [> h − 2], and sending
Z Q̃ [h − 2] onto torsion.
Moreover, these maps give rise to maps HHi (Π Q̃ , Π Q̃ ) → HHi (Π Q , Π Q ) for i ∈ {0, 1}, which describe the
image of central elements, and describe descent of outer derivations, related to the above by D. On HH0 , the
map Z Q̃ [h − 2] → L (h − 2) maps isomorphically to the saturation of the kernel of L (h − 2) (Y (h − 2) ⊕
Y (h − 2)) (i.e., the kernel of L (h − 2) Y (h − 2)).
Proof. Part (0) is a consequence of [Sch07, Theorem 4.2.60].

(1) Let us prove this by computing instead the map on HH1 ; there are duality isomorphisms
∼
HH1 −→ HH1 which must commute with π∗ because they are realized by explicit maps of chain com-
plexes expressed in terms of the quiver. Recall also that HH1 ( A , A ) is the space of outer derivations
(derivations of A modulo inner derivations).
We claim that the outer derivations descend from Π Q̃ to Π Q . To see this, we use the explicit
description of them for Π Q̃ from [Sch07, §10.2]: the outer derivations are realized by certain Q-
linear combinations of the half-Euler vector field (multiplying by the degree in Q , setting degree
in Q ∗ to be zero), and maps φx : y → {x, y }, using the Poisson bracket { , } induced by the necklace
Lie bracket on HH0 (or the Poisson structure on C[x, y ]Γ ). The latter was shown to make sense as
a map HH0 (Π Q , Π Q ) ⊗ Π Q → Π Q in [Sch07, §5.2], for any quiver Q , and the half-Euler vector
field obviously makes sense. Also, although HH1 consists of some fractions of sums of these outer
derivations, clearly an outer derivation is a multiple of some integer on Π Q̃ only if the same is true
in Π Q .
Next, we claim that all outer derivations on Π Q are obtained in this way. This is an immediate
consequence of the fact that HH0 (Π Q̃ , Π Q̃ ) surjects to U (−2), since HH1 (Π Q , Π Q ) is U (−2) times
the half-Euler vector field mentioned above (cf. [Eu08, Proposition 8.0.4] and [ES98a, II]). We thus
deduce the desired statement, and (1).
Next, we prove (2). We note that this is equivalent to the desired statement on the level of HH0 ,
∼
i.e., for the map Z Q̃ → Z Q , by virtue of the duality maps D : HH0 (Π Q̃ , Π Q̃ ) −→ HH2 (Π Q̃ , Π Q̃ ) and
D : HH0 (Π Q , Π Q ) HH2 (Π Q , Π Q ), where the latter is the quotient L (Y ⊕ Y ) on L, and the
identity on U .
To prove (2), we use the fact that the Connes differential, and hence the BV differential, are func-
torial. For the extended Dynkin side, by [Sch07, Theorem 10.3.1], the BV differential : HH1 → HH0 is
the map sending z · H Eu to ( H Eu + 1) z, where H Eu is the half-Euler vector field, and z ∈ H H 1 ; kills
the derivations related to the Poisson bracket as above. In other words, the map B : HH1 → H H 2 sends
(z, w ) ∈ Z Q̃ (2) ⊕ ( Z Q̃ )+ to ( H Eu + 1)(z) ∈ Z Q̃ (here we ignored the shift by two in applying H Eu).
On the Dynkin side, the Connes differential is also given by z → ( H Eu + 1)( z), for z ∈ U . Hence, by
functoriality of the Connes differential, we deduce that π∗,2 is as described in degrees < h, and in
degree h, has to at least map to Y (there can be h/2-torsion on account of the ( H Eu + 1)).
To complete the argument, it suffices to prove the surjectivity to Y : in terms of HH0 , we have to
show that the map Z Q̃ [h − 2] → L (h − 2) maps surjectively to the kernel of L (h − 2) Y (h − 2). For
this, we can perform a relatively easy explicit computation, showing that the elements from [Sch07]
η
map to the saturation of the column span of H Q . For A n , D n this is straightforward; for E 6 , E 8 , there
is nothing to show; and for E 7 , where h − 2 = 16, this alternatively follows from Proposition 7.3.3
of [Eu06] (which computes the square of an element z8 ∈ Z Q̃ [8]: this turns out to be the needed
∼
element which spans the rank-one kernel of L Y . But, we already know that Z Q̃ [8] −→ Z Q [8] by
the above.) 2
Remark 3.3.5. The above gives an alternative (integral) computation of the algebra structure on
HH0 (Π Q , Π Q ) given in [Eu08, §7]: this must be obtained from truncating the “Kleinian singulari-
ty” algebra HH0 (Π Q̃ , Π Q̃ ) at degrees h − 2, and composing with the kernel map ker( L Y ) → L.
The asserted relation to the Kleinian singularity C2 /Γ associated to Q is that Z Q̃ ∼
= e i 0 Π Q̃ e i 0 [Sch05,
Theorem 10.1.1], where i 0 is the extending vertex of Q̃ , and that e i 0 Π Q̃ e i 0 ⊗ C ∼
= C[x, y ]Γ (alterna-
tively, instead of by C, one can tensor by any ring containing |Γ1 | and |Γ |th roots of unity).
Remark 3.3.6. In [Sch07], the exact structure of T turned out to be mandated by the requirement
that, for non-Dynkin, non-extended Dynkin quivers Q̂ Q̃ Q , the torsion of HH0 (Π Q̂ , Π Q̂ ) is Z/ p
in degrees 2p
for all primes p and all
1, and these are generated by elements of the form

[r ] (where r = a∈ Q aa∗ − a∗ a is the relation). The specific structure of the torsion in the Dynkin
1 p
p
and extended Dynkin cases compensates for the fact that Z Q̃ is missing some degrees that would
otherwise be necessary to produce the torsion of HH0 (Π Q̂ , Π Q̂ ) (using the description in [Sch07, The-
orem 4.2.30], of torsion elements of HH0 (Π Q̂ , Π Q̂ ) not coming from HH0 (Π Q , Π Q ) as cyclic products
of elements r Q̃ with elements of Z Q̃ .)
4. Hochschild (co)homology of centrally extended preprojective algebras
In this section, we compute the BV structure on the Hochschild cohomology of centrally ex-
tended preprojective algebras A over k = C, and verify that the BV differential is graded self-adjoint
(hence, the Hochschild cohomology is a BV Frobenius algebra). From this, the structure of calculus
on (HH• ( A ), HH• ( A )) easily follows as in Remark 2.3.67, using the duality D (we omit the explicit
formulas).
As before, let Q be a quiver of A D E type with vertex set I . In [ER06], the centrally extended
μ
preprojective algebra Π Q is defined as a central extension of Π Q , in terms of a parameter μ ∈ k I .

We assume that μ is a regular weight, i.e., if μ = i ∈ I μi · e i for {e i } ⊂ k I the idempotents corre-
sponding to I , then ( i μi ωi , α ) = 0 for anyroot α of the root system attached to Q , with ωi the
weights. Explicitly, for all α = i ∈ I αi e i ∈ Z such that i ∈ I αi − a∈ Q αh(a) αt (a) = 1,
I 2
fundamental
we have i αi μi = 0.
Let P Q [ z] be the algebra of polynomials in the central parameter z with coefficients in P Q . Then
we define

a, a∗ .
μ
Π Q := P Q [ z] z·μ− (4.0.1)
a∈ Q
μ
This is a graded algebra with deg(e ) = deg(e ∗ ) = 1, deg( z) = 2, for all e ∈ Q . Now, let A := Π Q , and
let Z be the center of A. Let h denote the Coxeter number of Q . Let A + := A [ 1] denote the part of
positive degree, and let A top := A [2h − 2] denote the part of A of top degree.
In [ER06], it is proved that Π μ is Frobenius over k = C. We note that, over k = Z, this is not, in
general, true. For example, for Q = A 2 and μ = ρ = i ∈ I e i , we obtain that Π A ∼
ρ
= P Q /2P Q [ 3], 2
which is not fg or projective over Z. However, certain parameters μ ∈ Z I should yield a Frobenius
algebra, and
we hope to explore this in a future paper. Namely, these parameters should be those
μ
such that ( i ∈ I μi ωi , α ) = ±1 for all roots α ; more generally, for Π Q ⊗ F p to be Frobenius over p,

the condition should be that ( i ∈ I μi ωi , α ) is not a multiple of p. We hope to explore this in a future
paper.
For the rest of this section, let us take k := C and assume that μ is regular (( i μi ωi , α ) = 0).
There is a periodic resolution of A of period 4 [Eu06, §3], and A is a symmetric algebra, so we
immediately deduce (as stated in Example 2.3.12) that A is a periodic Calabi–Yau Frobenius algebra
of dimension 3 (of shift 4) and period 4 (of shift 2h).
Theorem 4.0.2. (See [Eu06].) The Hochschild cohomology groups of A over k := C are given by ( for n 0):

H H 4n+1 ( A ) ∼
= Z ∩ μ−1 [ A , A ] (−2nh − 2) ∼
= z Z (−2nh − 2), (4.0.3)

H H 4n+2 ( A ) ∼
= A / [ A , A ] + μ Z (−2nh − 2), (4.0.4)
HH 4n+3
( A) ∼
= A + /[ A , A ](−2nh − 4), (4.0.5)
H H 4n+4 ( A ) ∼
= Z / A top (−2(n + 1)h). (4.0.6)
From the periodicity, we immediately deduce the groups HH• , by allowing n to be an arbitrary integer
in the above. The fact that HHi ∼
= (HH3−i )∗ says that the nondegenerate trace pairing [ELR07] induces
nondegenerate pairings

Z ∩ μ−1 [ A , A ] (−2) ⊗ A / [ A , A ] + μ Z → k, (4.0.7)
A + /[ A , A ](−4) ⊗ Z / A top → k. (4.0.8)
To describe the cup products, as before, it suffices to describe the product between two degrees for
every triple of integers between 0 and 3, which sums to 3 modulo 4:
(0, 0, 3), (0, 1, 2), (1, 1, 1), (1, 3, 3), (2, 2, 3). (4.0.9)
Recall from [ELR07,Eu06] the Hilbert series for these graded vector spaces, using again m1 < . . . <
m| I | = h to denote the exponents of the root system (note that the sets {h − mi } = {mi } are identical):

r

h HH0 ( A ); t = h HH1 ( A ); t = t 2mi −2 + t 2mi + · · · + t 2h−6 , (4.0.10)
i =1

r

h HH2 ( A ); t = h HH3 ( A ); t = t −2 + 1 + · · · + t 2mi −6 . (4.0.11)
i =1
Theorem 4.0.12.
(i) As modules over HH0 ( A ), we have HH1 ( A ) ∼ = HH0 ( A ) and HH2 ( A ) ∼= HH3 ( A ) ∼
= (HH0 ( A ))∗ .
(ii) All of the cup products HHi ( A ) ⊗ HH j ( A ) → HHi + j ( A ) for 1 i j 3 are zero except for HH1 ( A ) ⊗
HH2 ( A ) → HH3 ( A ), which, using the identifications of (i), is the canonical map HH0 ( A ) ⊗ (HH0 ( A ))∗ →
(HH0 ( A ))∗ .
Proof. (i) From [Eu06, p. 10] it follows easily that HH1 ( A ) is of the desired form. Since there must
∼
be a unique (up to scaling) element in degree zero, we can use the explicit isomorphism HH0 ( A ) −→
HH ( A ), z → z · Eu where Eu is the Euler vector field. Then, the statements about HH ( A ), HH ( A )
1 2 3
follow immediately from Theorem 2.3.27. (We may even show compatibility with the duality pairings
defined in [Eu06] using the trace map of [ELR07], by a simple computation along the lines of [Eu08].)
(ii) For graded degree reasons, using (4.0.10), (4.0.11), and the fact that HH4 ( A ) ∼ = HH0 ( A )(−2h),
the triples (1, 3, 3) and (2, 2, 3) of multiplications (4.0.9) are zero. Then, HH1 ( A ) ∪ HH1 ( A ) = 0 since
Eu ∪ Eu = 0, by graded-commutativity. The final statement then follows from Theorem 2.3.27. 2
We now describe explicitly the Connes and BV differentials. For this, we fix the isomorphism
D : HH• ( A ) ∼
= HH3−• ( A ) of Theorem 2.3.27, and use the elements Eu ∈ HH4m+1 ( A ), Eu ∗ ∈ HH4m+2 ( A ),
and 1 ∈ HH4m+3 ( A ). Here, the notation is a bit abusive, since really Eu ∗ ∈ HH6 ( A ), 1∗ ∈ HH7 ( A ) using
∗
Theorem 2.3.27, but we identify these elements with their images under the periodicity. We describe
all elements of HH4m+2 ( A ), HH4m+3 ( A ) by Eu ∗ / z, 1∗ / z for z ∈ HH0 ( A ), which refers to the unique
elements so that z ∪ Eu ∗ / z = Eu ∗ and z ∪ 1∗ / z = 1∗ .
Theorem 4.0.13. With the above identifications, the BV differential is given by, for all m ∈ Z,
2m = 0, (4.0.14)

4m+1 (zEu ) = deg(z) + 4 − 2hm z · (1 )∪m , (4.0.15)

4m+3 1∗ / z = 2h(1 − m) − 4 − deg(z) Eu ∗ / z. (4.0.16)
In particular, is graded self-adjoint, i.e., HH• is BV Frobenius.
Proof. We use the Cartan identity (2.3.45) in the case a = Eu: Bi Eu + i Eu B = L Eu . Also, it is easy
to check (as in e.g. [Eu07a]) that L Eu ( f ) = deg( f ) · f for all f ∈ HH• ( A ) (and hence for HH• ( A ) as
well). From this (using (4.0.10), (4.0.11), the fact that the Calabi–Yau shift is 4, and the vanishing
of HH1 ( A ) ∪ HH3 ( A )) we compute B |HH0 ( A ) :

D Eu ∪ 1∗ / z = ( Bi Eu + i Eu B ) D 1∗ / z = 2h − 4 − deg(z) · 1∗ / z, (4.0.17)
which implies that 3 (1∗ / z) = (2h − 2 − deg( z)) Eu ∗ / z, using Theorem 4.0.12(ii). Then, 2 = 0 implies
that 2 = 0.
Inductively, we claim that 2m = 0 and 2m+1 is an isomorphism, for all m 1. Assume that
2m = 0 and 2m+1 is an isomorphism. We need only show that 2m−1 is also an isomorphism.
This follows from L Eu = B 3−2m i Eu + i Eu B 4−2m = i Eu B 4−2m , together with the fact that L Eu and i Eu
are isomorphisms (here, we use that the degrees of HH0 ( A ) are between 0 and 2h − 6). At the same
time, we may deduce the desired formulas (since L Eu multiplies by degree).
It remains only to prove that the formulas still hold in Hochschild degrees > 3. To show this, we
consider the formula
L1 = Bi 1 − i 1 B , (4.0.18)
for 1 ∈ HH4 ( A ) the periodicity element. From this, we may compute that, applied to degrees 3,
L1 kills even degrees, and on odd degrees acts by L1 (zEu ) = 2hz(1 )m (where | zEu | = 4m + 1), and
L1 (1∗ / z) = 2hEu ∗ / z. Since L1 is a derivation and L1 (1 ) = 0, we deduce the desired result. 2
5. Periodic group algebras of finite groups
As mentioned already, for any finite group G, the group algebra k[G ] is Frobenius, and in fact
symmetric (hence, Calabi–Yau Frobenius). It is natural to ask when such group algebras are periodic.
Certainly, if k[G ] is periodic, then its Hochschild cohomology is periodic. It is well known that one
has the following formula for Hochschild cohomology, as an abstract graded k-module:

HH• k[G ], k[G ] = H • Z G (c i ), k , (5.0.1)
conjugacy classes C i
with representative c i ∈ C i
where Z G (c i ) is the centralizer of c i in G, and the H • ( H , k) denotes the group cohomology of H with
coefficients in k. (For an explanation, see Proposition 5.0.6, where we give a refined version.)
Hence, in order for the Hochschild cohomology to be periodic, it must be that the numbers of
generators of the cohomology groups of G are bounded. Let us now set k := Z. Then, the classical
Suzuki–Zassenhaus theorem classifies all such groups. These groups are those such that all abelian
subgroups are cyclic, and they fall into six explicit families (cf. [AM04, p. 150]). Moreover, these all
have periodic group cohomology. Since this property is preserved under taking subgroups, we deduce
the (probably well known)
Proposition 5.0.2. The group algebra Z[G ] of a finite group G has periodic Hochschild cohomology iff all
abelian subgroups of G are cyclic. For such groups, k[G ] has periodic Hochschild cohomology (relative to k),
for all commutative rings k.
We would like to know if such group algebras are in fact periodic Calabi–Yau Frobenius algebras
(since they are symmetric, it is enough to check if they are periodic Frobenius). This is stronger than
having periodic Hochschild cohomology, since we actually need Ω n k[G ] k[G ] for some n 1. This
would be satisfied if we could show that such k[G ] have periodic resolutions.
Fortunately, there is a very similar classical result of Swan (which also used the (mod-p) classifi-
cation of periodic groups):
Theorem 5.0.3. (See [Swa60].) Let R := Z[ S −1 ] for some set S of primes. Let G be a finite group. Then, there
is a periodic resolution of R as an R [G ] module, i.e.,
0 → R → P n−1 → P n−2 → · · · → P 0 R → 0, (5.0.4)
iff G has periodic group cohomology with coefficients in R.
We deduce the following:
Theorem 5.0.5. The group algebra Z[G ] is periodic Calabi–Yau Frobenius iff G has the property that all abelian
subgroups are cyclic. Equivalently, Z[G ] is periodic CY Frobenius of period n iff its Hochschild cohomology (or
the group cohomology of G) is periodic of period n.
Proof. Using the above results, it is enough to show how to explicitly pass between a projective
resolution of k as a k[G ]-module, and a projective resolution of k[G ] as a k[G ]e -module, in such a way
as to preserve periodicity of a given period. More generally, we prove the following proposition. 2
Proposition 5.0.6. Let H = ( H , μ, , η, , S ) be any Hopf algebra over k which is a projective k-module. Let
e
H := (1 ⊗ S ) ◦ ( H ) ⊂ H ⊗ H op . Then, given any projective resolution P • of k as an H -module, Ind H
H P •
is a projective resolution of H as an H -bimodule, which is split as a complex of left H -modules.
Conversely, if Q • is a resolution of H as an H -bimodule which is split as a complex of right H -modules,
then Q • ⊗ H k is a projective resolution of k as an H -module.
To remove the left–right asymmetry, note that a sequence of (projective) H -bimodules which are
split as left H -modules can have the H -bimodule action twisted to a m b := S (b)mS (a) to make
them split as right H -modules rather than as left H -modules. If we apply ⊗ H k to the twisted version
of H , we still obtain k.
Proof. We only prove the first statement, since the last one is easy. The proof is essentially a refine-
⊗ H op ∼
ment of the usual proof of (5.0.1). We claim that (1) Ind HH k = H as H -bimodules, and (2) with
the left ( H ⊗ 1) and right H actions, H e is isomorphic as an H -bimodule to H ⊗ H with the usual
outer H -bimodule structure. By part (2) of the claim, and the fact that H is projective over k, we
⊗ H op
deduce that H ⊗ H op is a projective H module. Since Ind HH is the functor H e ⊗ H −, we obtain
the desired result.
To prove the claim, consider the k-linear maps
φ : H ⊗ H → H ⊗ H op , φ( g ⊗ h) = g · (1 ⊗ S ) ◦ (h), (5.0.7)

ψ:H⊗H op
→ H ⊗ H, ψ( g ⊗ h) = g · ( S ⊗ 1) ◦ S −1 h . (5.0.8)
By the antipode identity, coassociativity, and the counit condition, φ ◦ ψ = Id = ψ ◦ φ . On the other
hand, φ intertwines the right H -module structure on H ⊗ H with the right H = H -module structure
on H ⊗ H op , and ψ intertwines in the opposite direction; also, both intertwine the standard left
H -module structure. So, we obtain part (2). Part (1) then follows by explicit (easy) computation. 2
Corollary 5.0.9. A Hopf algebra has a periodic bimodule resolution which is split as a complex of right modules
iff its augmentation module has a periodic left module resolution. A Hopf algebra which is a Frobenius algebra
is periodic Frobenius iff its augmentation module k satisfies Ω n k k in the stable left-module category.
As remarked earlier, if k is a PID, any Hopf algebra which is fg as a k-module is automatically

Frobenius [LS69], so we can remove the Frobenius assumption from the corollary in this case (replac-
ing with fg projective over k).
Proof. The first assertion follows immediately from the proposition. For the second, we show that
H He
one has a stable module equivalence Ω n k k iff one has a stable bimodule equivalence Ω n H H .
We can use the functors – ⊗ H k, H e ⊗ H – to achieve this. 2

The periodic groups described by the theorem include all groups which act freely on spheres:
Theorem 5.0.10. Let G be any finite group which acts freely and orientation-preserving on a sphere S
with
1. Then, for some r 1, the group algebra k[G ] is a periodic Calabi–Yau Frobenius algebra, of dimension
+1
r
− 1 and period
+r 1 .
This theorem follows from e.g. Lemma 6.2 of [AM04] (a spectral sequence argument showing that
group cohomology is periodic in this case), by using Theorem 5.0.5. We give, however, a simple topo-
logical proof in Appendix A, in the case when G acts cellularly on a finite CW complex homeomorphic
to S
.
Corollary 5.0.11. For any finite subgroup G < SO(

+ 1) := SO(
+ 1, R) for any
1, k[G ] is periodic
Calabi–Yau Frobenius.
We note that this corollary may also be deduced from the version of the theorem proved in Ap-
pendix A, where G acts cellularly. To do this, we form a CW decomposition of S
+1 by geodesic
codimension-one slices, fixed under the orbit of G, which separate a given point x from all of its
orbits under G.
Acknowledgments
We thank P. Etingof and V. Ginzburg for useful discussions and advice. The second author was
partially supported by an NSF GRF.
Appendix A. Finite groups acting freely on spheres
In this appendix, we provide an elementary proof of Theorem 5.0.10 in the case that G acts cel-
lularly on a finite CW complex homeomorphic to S
. Our proof avoids the use of the classification of
periodic groups, and is purely topological.
The main idea of the proof is to construct a bimodule resolution of k[G ] by constructing a CW
complex which is homotopic to G, and which admits a free action of G × G op , such that the induced
G × G op -module structure on the cellular homology, k[G ], is the standard bimodule structure. The CW
complex will have finitely many cells of each dimension, and the resulting cellular chain complex will
be periodic. Thus, this complex yields a periodic bimodule resolution of k[G ].
To do this, we need the following simple topological lemma:
Notation A.0.1. Let D n , S n denote the n-dimensional disc and sphere, respectively.
Lemma A.0.2. Let m, n 0 be any integers. Consider the topological space X := D m+1 × S n , and let
f : ∂ X → S m be the attaching map which is the first projection of S m × S n to S m . Then, the glued topolog-
ical space X ∪ f S m is homeomorphic to S n+m+1 .
In the special case that n = 0, the above construction is the standard way to build S m+1 out of S m :
we attach two hemispheres D m+1 to S m placed at the equator.
Proof. Let us view S n+m+1 ⊆ Rn+m+2 as the unit sphere. Let Y ◦ ⊂ S n+m+1 be the subset

m
◦ n+m+1 n+m+2 2
Y := (x0 , . . . , xn+m+1 ) ∈ S ⊂R : | xi | < 1 . (A.0.3)
i =0
Now, Y ◦ is an open subset of S n+m+1 homeomorphic to D m+1 × S n , under the map

1
D m+1 × S n (x, y) → x, y . (A.0.4)
1 − x2
The complement of Y ◦ is the subset S m × {0} ⊂ S n+m+1 . That is, ∂(Y ◦ ) ∼

= S m , and the attaching map
S m+1 × S n → ∂(Y ◦ ) is f . 2
In fact, we will not only use the statement of the lemma, but (a cellular version of) the explicit
homeomorphism given in the proof.
Proof of the cellular version of Theorem 5.0.10. We will construct the topological space X :=
S ∞ × S ∞ /G S ∞ as an explicit CW complex with finite n-skeleta, whose associated cellular chain com-
plex is (
+ 1)-periodic. Moreover, the group G × G will act freely and cellularly (since G is finite, this
means that the group sends each d-dimensional cell onto another d-dimensional cell). This gives the
desired result by the remarks at the beginning of the subsection.
Let Y denote the finite CW complex with Y ∼ = S
that we are given. First, we construct from this
a topological space Z ∼ = D
+1 that extends the action, viewing Y as the boundary of Z . To do this,
let Z := (Y × [0, 1])/(Y × {0}), and let G act in the obvious way (preserving the second component).
We will think of Z as the solid unit disc in R
+1 and of Y as its boundary. We will also view Z as
a mere topological space isomorphic to D
+1 , i.e., a single
+ 1-cell with a G-action, for the purpose
of constructing complexes.
Now, set W
:= Y . We inductively construct a copy W of S ∞ by attaching (viewing Z k on the LHS
as a single k(
+ 1)-cell):
Z ×Y ∼
att.
= D
+1 × Y −−→ Y = W
, (A.0.5)
Z2 × Y ∼
att.
= D 2(
+1) × Y −−→ W (
+1)+
, (A.0.6)
Z ×Y ∼
3 3(
+1) att.
=D × Y −−→ W 2(
+1)+
, . . . , (A.0.7)
where “att.” means an attaching map (so NOT a map of topological spaces). We define these at-
taching maps to be the composition of the first projection S k(
+1)−1 × Y S k(
+1)−1 with the
∼
homeomorphism S k(
+1)−1 −→ W (k−1)(
+1)+
= W k(
+1)−1 , which exists inductively by Lemma A.0.2.
∼
Constructed this way, each homeomorphism S k(
+1)+
−→ W k(
+1)+
has image in the same sum of
∼
top cells of Z × Y : it is the sum of the
cells of Y which make up S
−→ Y .
k
Thus, on the level of chains, if we label the cells of W ∞ in degree k(

+ 1) + p by ck, p = Z k × c 0, p ,
where c 0, p are the cells of Y for 0 p
, the complex C • ( W ∞ ) is a periodic free complex.
By construction, W ∞ has a free action of G. Now, finally, set X := W ∞ × G, and let us view X
as the homeomorphic space W ∞ × W ∞ /G W ∞ , and equip X with the resulting free cellular action
of G × G. We then have that X is a topological space whose homology is k[G ] with the usual bimodule
action. We deduce that C • ( X ) is a free periodic resolution of k[G ]. 2
Appendix B. Frobenius algebras over general commutative base rings
In this appendix, we will extend some results known for Frobenius algebras over fields to a relative
context, using e.g. [ARS97] as a reference for the usual results.
We first deduce a relative self-injectivity for A.
Notation B.0.1. For any k-algebra A which is fg projective as a k-module, let ∗ denote the functor
∗ : Homk (−, k) : A-mod → A op -mod.
Definition B.0.2. Call an A-module I which is fg projective over k “injective relative to k” or “relatively
injective” if, for any A-modules M, N which are projective over k, and any injection M → N whose
cokernel is a projective k-module (i.e., the injection is k-split), there exists a unique dotted arrow
completing any diagram of the following form:
(B.0.3)
M N.
The following helps explain the meaning of relative injectivity:

Proposition B.0.4. Let A be any algebra over a commutative ring k. If M is any module over A which is fg
projective as a k-module, then any relatively injective A-module is acyclic for the functor Hom A ( M , −). (I.e.,
such I satisfy ExtiA ( M , I ) = 0 for all i 1.)
Proof. Let us take a projective resolution P • of M, i.e., · · · → P 1 → P 0 M. Since M is projective

over k, the resolution is k-split. Thus, the relative injectivity property will guarantee that there is no
positive homology of Hom( P • , I ). 2
The main use of “relative to k” is to make the dualization ∗ : M → Homk ( M , k) behave well:
Lemma B.0.5. Assume A is a k-algebra which is a fg projective k-module. Then:
(i) The dualization ∗ : A-mod → A op -mod is a contravariant functor which sends projective modules to
modules which are injective relative to k, and vice-versa.
(ii) There is a functorial isomorphism M ∗∗ ∼ = M, for M fg projective over k. That is, the restriction of ∗ to
fg A-modules which are projective over k (in both the domain and codomain) is an anti-equivalence of
categories, and ∗ ◦ ∗ Id.
The proof is just as in the case where k is a field, so we omit it.
Corollary B.0.6. If A is as in the lemma, then A-mod has enough relatively injectives in the following sense:
for every fg A-module M which is projective over k, there exists a relatively injective module I and a k-split
injection M → I .
Proof. For any k-projective M ∈ A-mod, pick a surjection P M ∗ in the category A op -mod. This is
k-split. Then, dualizing, one obtains a k-split injection M → P ∗ , and P ∗ is relatively injective. 2
As a corollary, we also deduce the relative self-injectivity for Frobenius algebras:
Corollary B.0.7. If A is a Frobenius algebra over k, then A is relatively injective as an A-module. Moreover, all
projectives are relatively injective and vice-versa.
Proof. We know that A ∗ is isomorphic to A as an A-module (using the invariance and nondegeneracy
of the pairing). Now, as an A op -module, A is projective; hence A ∗ is relatively injective as an A-
module.
For the last statement, we use the fact that all projectives are direct summands of free modules,
and hence are direct summands of relatively injective modules, and are hence relatively injective. For
the converse, for any relatively injective module M, M ∗ is a projective A op -module, hence a relatively
injective A op -module, and hence a projective A-module. 2
Next we show that the other duality ∨ also behaves well for Frobenius algebras in the relative
context. All of the following results are more generally true for relatively self-injective algebras, which
we define as algebras A satisfying the conclusion of Corollary B.0.6: they are fg projective over k
and relatively injective (equivalently, A ∗ is projective, i.e., all projectives are relatively injective and
vice-versa). The same proofs apply.
Proposition B.0.8. Suppose A is Frobenius over k. Then:
(i) The functor ∨ restricts to a functor on full subcategories:

{fg A-modules which are projective over k} ↔ fg A op -modules which are projective over k , (B.0.9)
and ∨ ◦ ∨ Id on these subcategories.

(ii) The functor ∨ preserves exact k-split complexes of fg projective k-modules.
Proof. (i) First, let us show that, if M ∈ A-mod is fg projective over k, then so is M ∨ :
∗
M ∨ = Hom A ( M , A ) ∼
= Hom A M , A ∗ = Hom A M , Homk A ∗ , k
adj. ∗
∼
= Homk A ∗ ⊗ A M , k = A ∗ ⊗ A M ∼ = (η−1 M )∗ , (B.0.10)
which is fg projective over k. Moreover, this is functorial in M, and we deduce that ∨ ◦ ∨ Id.
(ii) For bounded-below complexes, this follows from the fact (Proposition B.0.4) that
ExtiA ( M , A ) = 0 for all i 1. Then, for any unbounded k-split exact complex of projectives, we trun-
cate at an arbitrary place. 2
Corollary B.0.11. Let A be a Frobenius algebra over k. We have mutually inverse autoequivalences
Ω, Ω −1 : A-modk → A-modk , which yield exact sequences
0 → Ω M → P M → 0, (B.0.12)
with P a fg projective A module, for all fg A-modules M which are projective over k.
Proof. Choose, for every module M, a sequence (B.0.12), and similarly a sequence of the form
0 → M → I Ω −1 ( M ) → 0, (B.0.13)
with I satisfying the relative injectivity property. (To obtain such a sequence, form a sequence of
the form (B.0.12) for M ∗ in the category A op -mod, and then dualize.) Then, using the fact that a
map M → N factors through a specific injection M → I for I relatively injective iff it factors through
any other injection into a relatively injective (both are true iff M factors through all injections), it is
straightforward to finish using the same arguments as in the case when k is a field (see, e.g., [ARS97,
ASS06]). 2
Corollary B.0.14. Let A be any Frobenius algebra over k. For any degree i 1, and any fg A-modules M , N
which are projective over k,

ExtiA ( M , N ) ∼
= Hom A Ω i M , N . (B.0.15)
Proof. ExtiA ( M , N ) can be computed using a projective resolution of M,
di +1 di d1
P i +1 −−−→ P i −→ · · · −−→ P 0 M , (B.0.16)
by taking the quotient

f ∈ Hom A e ( P i , N ) f ◦ di +1 = 0 /{ g ◦ di } g ∈Hom Ae ( P i−1 , N ) . (B.0.17)
Now, factor P i → P i −1 as P i Ω i M → P i −1 (for Ω i M = im di ∼= coker di +1 ), so that we may now

write ExtiA ( M , N ) as Hom A (Ω i M , N )/{morphisms factoring through Ω i M → P i −1 }. Then, by the ob-
servation from the proof of Corollary B.0.11 that morphisms factor through Ω i M → P i −1 iff they
factor through any other injection to an injective A-bimodule relative to k (and injective A-bimodules
relative to k are the same as fg projective A-bimodules by Corollary B.0.7), we obtain the desired
result. 2
Remark B.0.18. When k is a field, the above actually endows the stable module category with the
structure of a triangulated category, which is the quotient of the derived category of fg A-modules by
finite complexes of projective A-modules. However, this is NOT true for general k (we had to restrict
to the non-abelian subcategory of k-projectives before doing anything).
References
[AM04] A. Adem, R.J. Milgram, Cohomology of Finite Groups, second ed., Grundlehren Math. Wiss. (Fundamental Principles of
Mathematical Sciences), vol. 309, Springer-Verlag, Berlin, 2004.
[ARS97] M. Auslander, I. Reiten, S.O. Smalø, Representation Theory of Artin Algebras, Cambridge Stud. Adv. Math., vol. 36,
Cambridge Univ. Press, Cambridge, 1997, corrected reprinting of the 1995 original.
[ASS06] I. Assem, D. Simson, A. Skowroński, Elements of the Representation Theory of Associative Algebras. vol. 1. Techniques
of Representation Theory, London Math. Soc. Stud. Texts, vol. 65, Cambridge Univ. Press, Cambridge, 2006.
[BBK02] S. Brenner, M.C.R. Butler, A.D. King, Periodic algebras which are almost Koszul, Algebr. Represent. Theory 5 (4) (2002)
331–367.
[BES07] J. Białkowski, K. Erdmann, A. Skowroński, Deformed preprojective algebras of generalized Dynkin type, Trans. Amer.
Math. Soc. 359 (6) (2007) 2625–2650, electronically published January 2007.
[BZ08] K.A. Brown, J.J. Zhang, Dualising complexes and twisted Hochschild (co)homology for noetherian Hopf algebras, J. Al-
gebra 320 (5) (2008) 1814–1850, arXiv: math/0603732.
[CBEG07] W. Crawley-Boevey, P. Etingof, V. Ginzburg, Noncommutative geometry and quiver algebras, Adv. Math. 209 (1) (2007)
274–336, math.AG/0502301.
[Cos07] K. Costello, Topological conformal field theories and Calabi–Yau categories, Adv. Math. 210 (1) (2007) 165–214, arXiv:
math/0412149v7.
[CS99] M. Chas, D. Sullivan, String topology, Ann. of Math., in press, arXiv: math/9911159.
[CS04] M. Chas, D. Sullivan, Closed string operators in topology leading to Lie bialgebras and higher string algebra, in:
The Legacy of Niels Henrik Abel, Springer-Verlag, Berlin, 2004, pp. 771–784.
[CV06] R.L. Cohen, A.V. Voronov, Notes on string topology, in: String Topology and Cyclic Homology, in: Adv. Courses Math.
CRM Barcelona, Birkhäuser, Basel, 2006, pp. 1–95, arXiv: math/0503625.
[EE07] P. Etingof, C.-H. Eu, Hochschild and cyclic homology of preprojective algebras of ADE quivers, Mosc. Math. J. 7 (4)
(2007) 601–612, arXiv: math/0609006.
[ELR07] P. Etingof, F. Latour, E. Rains, On central extensions of preprojective algebras, J. Algebra 313 (1) (2007) 165–175, arXiv:
math/0606403.
[ER06] P. Etingof, E. Rains, Central extensions of preprojective algebras, the quantum Heisenberg algebra, and 2-dimensional
complex reflection groups, J. Algebra 299 (2) (2006) 570–588, arXiv: math/0503393.
[ES98a] K. Erdmann, N. Snashall, On Hochschild cohomology of preprojective algebras I, II, J. Algebra 205 (2) (1998) 391–412,
413–434.
[ES98b] K. Erdmann, N. Snashall, Preprojective algebras of Dynkin type, periodicity and the second Hochschild cohomology,
in: Algebras and Modules, II, Geiranger, 1996, in: CMS Conf. Proc., vol. 24, Amer. Math. Soc., Providence, RI, 1998,
pp. 183–193.
[ES06] K. Erdmann, A. Skowroński, The stable Calabi–Yau dimension of tame symmetric algebras, J. Math. Soc. Japan 58 (1)
(2006) 97–128.
[Eu06] C.-H. Eu, Hochschild and cyclic homology of central extensions of preprojective algebras of ADE quivers, arXiv:
math/0606412, 2006.
[Eu07a] C.-H. Eu, The calculus structure of the Hochschild homology/cohomology of preprojective algebras of Dynkin quivers,
arXiv: math.RT/0703568, 2007.
[Eu07b] C.-H. Eu, Hochschild and cyclic (co)homology of preprojective algebras of quivers of type T, arXiv: 0710.4176, 2007.
[Eu08] C.-H. Eu, The product in the Hochschild cohomology ring of preprojective algebras of Dynkin quivers, J. Algebra 320 (4)
(2008) 1477–1530, arXiv: math.RT/0703568.
[GDT89] I.M. Gelfand, Yu.L. Daletskii, B.L. Tsygan, On a variant of noncommutative differential geometry, Dokl. Akad. Nauk
SSSR 308 (6) (1989) 1239–1297 (in Russian), translation in Soviet Math. Dokl. 40 (2) (1990) 422–426.
[Gin06] V. Ginzburg, Calabi–Yau algebras, arXiv: math.AG/0612139, 2006.
[HL04] A. Hamilton, A. Lazarev, Homotopy algebras and noncommutative geometry, arXiv: math/0410621, 2004.
[Kas61] F. Kasch, Projektive Frobenius–Erweiterungen, S.-B. Heidelberger Akad. Wiss. Math.-Nat. Kl. (1960/1961) 87–109
(in German).
[Kau07a] R. Kaufmann, Moduli space actions on the Hochschild co-chains of a Frobenius algebra. I. Cell operads, J. Noncommut.
Geom. 1 (3) (2007) 333–384, arXiv: math/0606064.
[Kau07b] R.M. Kaufmann, On spineless cacti, Deligne’s conjecture and Connes–Kreimer’s Hopf algebra, Topology 46 (1) (2007)
39–88, arXiv: math/0403340.
[Laz01] C.I. Lazaroiu, On the structure of open–closed topological field theory in two dimensions, Nuclear Phys. B 605 (1–3)
(2001) 159–191.
[LS69] R.G. Larson, M.E. Sweedler, An associative orthogonal bilinear form for Hopf algebras, Amer. J. Math. 91 (1969) 75–93.
[Men04] L. Menichi, Batalin–Vilkovisky algebras and cyclic cohomology of Hopf algebras, K-Theory 3 (2004) 231–251, arXiv:
math/0311276.
[Moo01] G. Moore, Some comments on branes, G-flux, and K -theory, in: Strings 2000, Proceedings of the International Super-
strings Conference, Ann Arbor, MI, Internat. J. Modern Phys. A 16 (5) (2001) 936–944.
[MOV06] A. Malkin, V. Ostrik, M. Vybornov, Quiver varieties and Lusztig’s algebra, Adv. Math. 203 (2) (2006) 514–536, arXiv:
math/0403222.
[NT60] T. Nakayama, T. Tsuzuku, On Frobenius extensions. I, Nagoya Math. J. 17 (1960) 89–110.
[RS] C.M. Ringel, A. Schofield, Wild algebras with periodic Auslander–Reiten translate, unpublished manuscript.
[SA04] M. Suarez-Alvarez, The Hilton–Heckmann argument for the anti-commutativity of cup products, Proc. Amer. Math.
Soc. 132 (8) (2004) 2241–2246, arXiv: math/0209029.
[Sch05] T. Schedler, A Hopf algebra quantizing a necklace Lie algebra canonically associated to a quiver, Int. Math. Res.
Not. 2005 (12) (2005) 725–760, IMRN/14217.
[Sch07] T. Schedler, Hochschild homology of preprojective algebras over the integers, arXiv: 0704.3278, 2007.
[Swa60] R.G. Swan, Periodic resolutions for finite groups, Ann. of Math. 72 (2) (1960) 267–291.
[Swa67] R.G. Swan, The number of generators of a module, Math. Z. 102 (4) (1967) 318–322.
[Tra02] T. Tradler, The BV algebra on Hochschild cohomology induced by infinity inner products, arXiv: math/0210150, 2002.
[Tsy04] B. Tsygan, Cyclic homology, in: J. Cuntz, B. Tsygan, G. Skandalis (Eds.), Cyclic Homology in Non-Commutative Geom-
etry. Operator Algebras and Non-Commutative Geometry, II., in: Encyclopaedia Math. Sci., vol. 121, Springer-Verlag,
Berlin, 2004.
[TT05] D. Tamarkin, B. Tsygan, The ring of differential operators on forms in noncommutative calculus, in: Graphs and Pat-
terns in Mathematics and Theoretical Physics, in: Proc. Sympos. Pure Math., vol. 73, Amer. Math. Soc., Providence, RI,
2005, pp. 105–131.
[TZ06] T. Tradler, M. Zeinalian, On the cyclic Deligne conjecture, J. Pure Appl. Algebra 204 (2006) 280–299, arXiv:
math/0404218.
Journal of Algebra
Mutations vs. Seiberg duality ✩

Jorge Vitória
Mathematics Institute, University of Warwick, Coventry, CV4 7AL, United Kingdom
Article history: For a quiver with potential, Derksen, Weyman and Zelevinsky de-
Received 12 November 2007 fined in [H. Derksen, J. Weyman, A. Zelevinsky, Quivers with po-
Communicated by J.T. Stafford tentials and their representations I: Mutations, arXiv: 0704.0649v2
[math.RA]] a combinatorial transformation, mutations. Mukhopad-
Keywords:
hyay and Ray, on the other hand, tell us how to compute Seiberg
Mutations
Seiberg duality dual quivers for some quivers with potentials through a tilting pro-
Derived equivalences cedure, thus obtaining derived equivalent algebras. In this text,
Tilting complexes we compare mutations with the concept of Seiberg duality given
Quivers with potentials by [S. Mukhopadhyay, K. Ray, Seiberg duality as derived equiva-
lence for some quiver gauge theories, arXiv: hep-th/0309191v2],
concluding that for a certain class of potentials (the good ones)
mutations coincide with Seiberg duality, therefore giving derived
equivalences.
1. Preliminaires
In this section we introduce the material from [8] that will be used and recall some definitions.
We will use the following notation: K is a field; K Q is the path algebra of the quiver Q over K
(concatenation of paths is written as composition of functions); Proj( R ) is the full subcategory of pro-
jective right modules over a K-algebra R; P ( R ) is the full subcategory of finitely generated projective
right modules over R; K b ( Q ) and D b ( Q ) are, respectively, the bounded homotopy category and the
bounded derived category of right modules over K Q .
Definition 1.1. A potential on a quiver is an element of the vector space spanned by the cycles of the
quiver (denote it by K Q cyc ).
✩
Supported by FCT, Portugal, research grant SFRH/BD/28268/2006.
E-mail address: j.n.s.vitoria@warwick.ac.uk.
doi:10.1016/j.jalgebra.2008.11.012
J. Vitória / Journal of Algebra 321 (2009) 816–828 817
Definition 1.2. Let A = Q 1 , i.e., the vector space spanned by all arrows. For each ξ ∈ A ∗ (the dual
of A), define a cyclic derivative
∂/∂ξ : K Q cyc → K Q

n
a1 . . . an → ξ(ak )ak+1 . . . an a1 . . . ak−1 . (1.1)
k=1
If x ∈ Q 1 , we will denote by ∂/∂ x the cyclic derivative correspondent to the element of A ∗ which
is the dual of x in the dual basis of A.
Definition 1.3. Two potentials are cyclically equivalent if S − S lies in the span of elements of the form
a1 . . . an−1 an − a2 . . . an a1 . A pair ( Q , S ) is said to be a quiver with potential if Q has no loops and no
two cyclically equivalent paths appear on S. Two quivers with potentials ( Q , S ) and ( Q̃ , S̃ ) are said
to be right equivalent if there is isomorphism φ between K Q and K Q̃ such that φ( S ) is cyclically
equivalent to S̃.
Definition 1.4. Given a quiver with potential ( Q , S ), define the jacobian algebra of ( Q , S ) as J ( Q , S ) =
K Q / J ( S ), where J ( S ) = (∂ S /∂ x)x∈ Q 1 .
Remark 1.5. Two right equivalent quivers with potentials have isomoprhic jacobian algebras (see [8]).
Definition 1.6. Define the trivial part of a quiver with potential ( Q triv , S triv ) by taking S triv as the de-
gree two homogeneous component of S and Q triv as the subquiver of Q consisting only in the arrows
appearing in S triv . The reduced part ( Q red , S red ) is formed by the non-trivial part of the potential S
and by the quiver obtained by taking the quotient of A by the arrows appearing in S triv .
The following theorem will allow us to define mutation on a quiver with potential.
Theorem 1.7. (See [8].) For a quiver with potential ( Q , S ), there exist a trivial quiver with potential
( Q triv , S triv ) and a reduced quiver with potential ( Q red , S red ) such that ( Q , S ) is right equivalent to
( Q triv ⊕ Q red , S triv + S red )— Q triv ⊕ Q red stands for the quiver obtained by taking the direct sum of the arrow
spans.
Let us now describe the procedure of mutation of a quiver with potential ( Q , S ) on a vertex k
(denote it by μk ( Q , S )).
(1) Suppose k does not belong to any 2-cycle and that S does not have any cycle starting and finish-
ing on k (if it does, substitute it by a cyclically equivalent potential that does not).
(2) Change the quiver in the following way:
• Reflect arrows starting or ending at k. Denote reflected arrows by (.)∗ ;
α β
• Create one new arrow for each path of the form •i •k • j and denote it by [β α ].
We denote the resulting quiver by Q̃ .
(3) Change the potential in the following way:
• Substitute
factors appearing in S of the form β α by the new arrow [β α ] and denote it by [ S ];
• Add Δk = [β α ]α ∗ β ∗ to [ S ]. We denote the resulting potential by S̃.
α β
•i •k •j
(4) The mutation at k of ( Q , S ) is μk ( Q , S ) = ( Q̄ , S̄ ) := ( Q̃ red , S̃ red )
Note that these mutations generalize reflection functors on quivers with no relations in the sense
that if you do a mutation on either a sink or a source, this procedure reduces to reflect arrows on
that vertex.
818 J. Vitória / Journal of Algebra 321 (2009) 816–828
Let us recall Rickard’s theorem, starting by defining tilting complex [10].
Definition 1.8. A tilting complex over a ring R is an object T of K b ( P ( R )), such that:
(1) ∀i = 0, Hom K b ( P ( R )) ( T , T [i ]) = 0;
(2) T generates K b ( P ( R )) as a triangulated category.
Theorem 1.9 (Rickard). Let R and S be two rings. Then D b ( R ) is equivalent to D b ( S ) iff there is a tilting
complex T over R such that S ∼
= End K b ( R ) ( T )op .
2. Seiberg duality
We will define Seiberg duality on quivers as a tilting procedure and therefore as an equivalence
of derived categories. This approach to Seiberg duality is provided by [9]. For more background see
[1,2,6]. To check if a complex is tilting we will have to compute homomorphisms in the derived
category between (finitely generated) projective modules.
Remark 2.1. Note that K b ( P ( R )) is a full subcategory of K b ( R ) and therefore, for an object T in
K b ( P ( R )) (in particular for a tilting complex) we have End K b ( R ) ( T )op = End K b ( P ( R )) ( T )op .
Let ( Q , S ) be a quiver with potential with n vertices such that every vertex is contained in some
cycle (which we shall assume from now on) and, for each vertex k, consider the following complex

n
Tk = T ik ,
i =1
where
T ik = 0 → P i → 0, if i = k
and
(α j ) j
Pj
T kk = 0 Pk 0.
j →k
Lemma 2.2. T k is a tilting complex over the jacobian algebra of ( Q , S ) iff Hom K ( P ( Q )) ( T kk , T sk [−1]) = 0, ∀s.
Proof. (1) Hom K b ( P ( Q )) ( T k , T k [i ]) = 0 ∀i = 0. It is clear that if r , s = k, then Hom K ( P ( Q )) ( T rk , T sk [i ]) = 0,

∀i = 0. Now, suppose s = k and r = k. Then we only have to check that the set Hom K ( P ( Q )) ( T rk , T kk [1])
reduces to zero. Note that, since a homomorphism between P r to P k is identified with an element of
the path algebra with each term being a path from r to k, every such homomorphism factors through

j →k P j .
0 Pi 0

0 j →k Pj Pk 0
Such factorization implies that these maps of complexes are homotopic to zero, thus zero in the
homotopy category.
If r = k then we also have such a homotopy just by taking identity maps.

0 j →k Pj Pk 0

0 j →k Pj Pk 0
(2) add(T k ) generates K b ( P ( Q )) as a triangulated category. It is enough to prove that the stalk
complexes of indecomposable projective modules are generated by the direct summands of T k .
Consider the direct summands of T k and take the cone of the map T kk to k
j →k T j defined by:

0 j →k Pj Pk 0
id

0 j →k Pj 0
That cone is just the following complex (the underlined term is in degree zero):
((α j ) j ,0)
0 j →k Pj Pk ⊕ ( j →k P j) 0 (2.1)
Consider the map from the complex (2.1) to the stalk complex of P k in degree zero defined by identity
in the first component and −(α j ) j in the second component and consider the map from this same
stalk complex to (2.1) defined by the inclusion of P k . We will prove that the composition of these
maps is homotopic to the identity map, hence proving that these complexes are isomorphic in the
derived category. In fact, that follows from the following diagram:
((α j ) j ,0)
0 j →k Pj Pk ⊕ ( j →k P j) 0
(id,−(α j ) j )
0 Pk 0
(0,id)
(id,0)
((α j ) j ,0)
0 j →k Pj Pk ⊕ ( j →k P j) 0
Similarly we can see it for the reverse composition and therefore (2.1) is isomorphic to the stalk
complex P k in degree zero.
Hence, the complex is tilting iff we have Hom K ( P ( Q )) ( T kk , T sk [−1]) = 0, ∀s. 2
Definition 2.3. Given a quiver with potential ( Q , S ), define δ( Q , S ) as the set of vertices for which
the complex above is tilting over J ( Q , S ), i.e.,

δ( Q , S ) = k ∈ Q 0 : Hom K ( P ( Q )) T kk , T sk [−1] = 0, ∀s .
If δ( Q , S ) = ∅, then we say that ( Q , S ) is locally dualisable in δ( Q , S ). Furthermore, if δ( Q , S ) = Q 0

then we say that ( Q , S ) is globally dualisable.
Remark 2.4. Note that to check whether the complex is tilting we just need to check that there is no
element f in the path algebra such that
(α j ) j
j →k Pj Pk
f
(2.2)
0 Ps
commutes. The existence of such an f implies that the set of relations must contain the set
{ f α j : j → k}, which is easy to see once we differentiate the potential in order to the arrows.
The above remark allows us, given a potential S for Q , to determine δ( Q , S ).

From now on we will drop the superscript on T whenever the vertex with respect to which we
are considering the tilting complex is fixed.
Definition 2.5. The Seiberg dual algebra of a quiver Q with potential S (or of its jacobian algebra) at
the vertex k ∈ δ( Q , S ) is the endomorphisms algebra of T k as defined above.
Rickard’s theorem then tells that Seiberg dual algebras have derived equivalent categories of mod-
ules.
3. Seiberg duality for good potentials
Let us consider the following class of potentials.
Definition 3.1. A potential is said to be a good potential if each arrow appears at least twice and no
subpath of length two appears twice.
Note that, in particular, a quiver with a good potential has the property that every arrow is con-
tained in at least two distinct cycles.
Proposition 3.2. A quiver with good potential is globally dualisable.
Proof. This is an immediate consequence of the definition of good potential since all the relations we
d
get from these kind of potentials are of the form ∂ S /∂ a = i =1 λi v i = 0 where λi ∈ K, d 2 and the
v i ’s are paths starting with different arrows thanks to the requirement that no subpath of length two
should be shared between two terms of the potential. Thus, there cannot occur any relations of the
type u α j = 0 and therefore δ( Q , S ) = Q 0 . 2
Let ( Q , S ) be a quiver with good potential. We want to give a presentation of its Seiberg dual
algebra at a fixed vertex k. We will see that this algebra is in fact the jacobian algebra of a quiver
with potential. We will call this quiver the Seiberg dual quiver.
First we should compute the quiver. It has the same number of vertices as the initial quiver (since
we will have that number of indecomposable projectives in End D b ( Q ) ( T ) corresponds to the number
of direct summands of T ) and, for each irreducible homomorphism between the T i ’s, draw an arrow
between the correspondent vertices. As we will see in the next theorem, those irreducible homomor-
phisms are of three types (this terminology, used for simplicity of language, is inspired by [9]):
• arrows of the form a, where a is also an arrow in Q , will be called internal arrows;
• arrows of the form α ∗ will be called dual arrows;
• arrows of the form [β α ] will be called mesonic arrows.
The theorem below shows that this choice of notation is an adequate one since the procedure to get
the Seiberg dual quiver is the same as the one that allow us to mutate the initial quiver. Also as in
mutations we will do this in two essential steps: obtain a quiver Q̃ that may contain more arrows
than the irreducible homomorphisms and then, looking at relations, eliminate the appropriate arrows
that do not correspond to irreducible ones (those will be the arrows lying in 2-cycles).
It turns out that relations on the Seiberg dual quiver can also be encoded in a potential (see
Proposition 3.4) and it will be determined as follows:
n
(1) Determine S̃ := [ S ] + i =1 [βi α j ]α ∗j βi∗ (eventually containing some arrows representing non-
irreducible homomorphisms);
(2) For every arrow a in a two cycle ab, take the relation ∂ S̃ /∂ a = 0 and substitute b in S̃ using this
equality (and thus eliminate b from the quiver, since b is not irreducible as it can be written as a
compostion of arrows). Call S̄ to the potential thus obtained.
Remark 3.3. Again, for language simplicity, arrows appearing in two cycles will be called massive
arrows and the process described on item 2 of the algorithm above will be called integration over
massive arrows.
Let us start by comparing the mutated quiver and the Seiberg dual quiver (no relations on them,
yet).
Theorem 3.4. Let Q be a quiver with a good potential S. The underlying quiver of the mutation of Q coincides
with the underlying quiver of the Seiberg dual of Q .
Proof. (1) First we prove that Seiberg duality at k inverts incoming arrows to k. The complex T k has
in degree zero one copy of P j for every arrow from j to k, therefore for each such arrow you get one
projection map from the direct sum to P i and therefore an irreducible homomorphism from T k to T j ,
hence getting an arrow from k to j in the dual quiver. For each arrow α j from j to k, denote the
correspondent homomorphism from T k to T j by α ∗j . There are no more irreducible homomorphims:
any other homomorphism factors through some factor of the direct sum first.
(2) Now we prove that Seiberg duality at k inverts outgoing arrows from k. This requires the
commutativity of a diagram like the following:
0 Pi 0
(α j ) j
0 j →k Pj Pk 0
The commutativity of the diagram requires that (α j ) j of = 0 and so we have to check the relations
on the quiver to obtain such a condition. Fix an arrow β from i to k and take the (cyclic) derivative
d
of the potential in order to beta. Since S is a good potential, ∂ S /∂β = t =1 λt v t where the v t ’s are
paths
from i to k (since β v t is a cycle for all t) and d 2. To give a homomorphism from P i to
j →k P j we just need to give a homomorphism from P i to each P j by the universal property of the
direct sum. Call jt to the index of the projective factor in j →k P j such that α jt is on the path v t .
Observe that v t = α jt ṽ t , where ṽ t is a path from i to jt as in the picture.
•k
α jt β
• jt •i
ṽ jt
Set the homomorphism from P i to each P j as follows:
• zero if j = jt for some t;

• λt ṽ t if j = jt for some t;
and set β ∗ to be the homomorphism induced by this set of homomorphisms to the direct sum and
therefore to the complex T k . Clearly this map makes the diagram above commute. Now we need to
prove that this is irreducible. If not, then it factors through other T r with the homomorphism from T i
to T r being irreducible and therefore coming from an arrow γ : i → r (r = k since the quiver has no
two cycles). But the existence of such a factorization would imply that all terms β v t in the potential
share a subpath of length two γ β which is a contradiction since, by assumption, the potential is a
good one and d 2. Hence β ∗ is irreducible. By construction, these homomorphisms are the only
irreducible ones.
αj βi
(3) For each path of length two in the initial quiver of the form • j •k •i we get
a homomorphism βi α j from T j to T i . It will be irreducible iff there is not a homomorphism in
the opposite direction. In fact this follows from the fact that S is a good potential and therefore
every arrow appears in S. Thus, if a is the arrow going in the opposite direction, ∂ S̃ /∂ a gives an
explicit factorization of the mesonic arrow. On the other hand, if it is not contained in a 2-cycle, then
it is irreducible, since it could only factor through the stalk complex of P k which is not, however,
projective in End D b ( Q ) ( T ). Denote this homomorphism by [βi α j ].
(4) Finally, if none of the previous cases apply, then homomorphisms between T j and T i are just
arrows from j to i. Again, these homomorphisms are irreducible iff they are not contained in a 2-cycle
and a similar argument to the one above applies to this case.
Let Q̃ be the quiver obtained by taking all the homomorphisms considered in the cases above,
even if they are not irreducible. Determining this quiver Q̃ is, therefore, clearly the same procedure
via mutations or via Seiberg duality. Now, since both mutation and Seiberg duality require the elim-
ination of 2-cycles after this step (in the later case to get only the irreducible homomorphisms), the
quiver obtained by mutation at k and the quiver obtained by Seiberg duality on k are the same. 2
At this point, we shall prove that the algorithm above allows us to obtain the Seiberg dual potential
of a quiver with potential ( Q , S ) on a fixed vertex k.
Proposition 3.5. The algorithm described above computes a potential for the Seiberg dual quiver such that its
jacobian algebra is End D b ( Q ) ( T ).
Proof. Let the homomorphisms represented by dual arrows of outgoing arrows be as it is described
on the proof of (3.4). We will first prove that the relations induced by the potential S̃ obtained
through the algorithm above are satisfied by End D b ( Q ) ( T ). Let τ (i , j ) be the coefficient of [βi α j ]
in [ S ].
• Relations coming from differentiating on βi∗ (dual of an outgoing arrow):

n
∂ S̃ /∂βi∗ = [βi α j ]α ∗j = βi (α j ) j = 0,
j =1
since the map in question is homotopic to zero in the complex category.

• Relations coming from differentiating on α ∗j (dual of an incoming arrow):

∂ S̃ /∂ α ∗j = βi∗ [βi α j ] = βi∗ βi α j .
i i
∗
Let us check that i βi βi = 0. In fact, let us compute the mth entry of this vector. For that we
look to the appearances of αm in [ S ]. So, if we have in [ S ] some subexpression of the form

d
τ (it , m)[βit αm ] ṽ it ,
t =1

then we have the mth entry of i βi∗ βi given by

d
τ (it , m) ṽ it βit ,
t =1
which is zero since ∂ S /∂ αm = 0.

• Relations coming from differentiating on a:
∂ S̃ /∂ a = ∂[ S ]/∂ a = 0,
since this is essentially the same as ∂ S /∂ a (eventually with some extra square brackets).
• Relations coming from differentiating on [βi α j ] (mesonic arrow).
We just need to check that
∂[ S ]/∂[βi α j ] = α ∗j βi∗
but this follows by definition of α ∗j and βi∗ as homomorphisms (see proof of (3.3)).
Observe now that integration over massive arrows does not change the relations induced by the
potential since the expressions obtained by differentiating in order to a massive arrows are zero in
the jacobian algebra, according to the proof above.
The last thing we need to check is that this potential S̄ gives all the relations. If not, suppose first
that the potential of Q̄ is of the form S̄ + W . Then, for any arrow a in W
∂ S̄ /∂ a + ∂ W /∂ a = 0
implies that ∂ W /∂ a = 0 and hence W = 0. Suppose now that there is one non-zero relation r such
that r is not of the form ∂ S̄ /∂ a for all a in the quiver. This relation is a linear combination of ho-
momorphisms such that each term of the linear combination is a map from some fixed T j to some
fixed T i . If this relation does not involve dual arrows, then these homomorphisms can be expressed
as linear combinations of elements of the path algebra from j to i and therefore this is a relation iff
there is some internal arrow a such that ∂ S /∂ a is equal to r up to square brackets. Thus we get a
contradiction and therefore r has to involve dual arrows. However, the construction of dual arrows as
homomorphisms makes it easy to see that all possible relations involving them are contemplated in
the cases above and thus proving that in fact all the relations are encoded in the potential S̄. 2
Definition 3.6. If one massive arrow appears in two different 2-cycles of S̃, that is, we get an expres-
sion of the form:

d
l
S̃ = λi abi + au j + W ,
i =1 j =1
where λi ∈ K; a and b i ’s are arrows; d 2; the u i ’s are paths of length 2 and a does not appear
in W , then we say that the b i ’s are related arrows.
Given Q a quiver with good potential S, suppose that S̃ can be written as follows:

S̃ = λi a i b i + σi , j a i u i , j + b i v i + W , (3.1)
i =1 j
where σi , j , λi , μi ∈ K, the ai b i ’s are 2-cycles (i.e., the ai ’s and the b i ’s are massive arrows), the b i ’s
are mesonic (thus the coefficient of b i v i is 1), W does not have any term involving massive arrows
and i = j implies ai = a j (that is, no related arrows occur). Note that b i = b j because of the fact that
S, being good, does not have repeated subpaths of length two.
Theorem 3.7. Let Q be a quiver with a good potential S. If k is a vertex such that no related arrows occur in
the mutation, there is a right equivalence φ from ( Q̃ , S̃ ) to ( Q̃ , S + S̄ ), where S is trivial and S̄ is obtained
by Seiberg duality.
Proof. Since there are no related arrows, let us assume that S̃ is of the form (3.1). Take the automor-
phisms given by:
φi : K Q̃ → K Q̃ ,
1
ai → ai − vi,
λi
1
b i → b i − σi , j u i , j ,
λi
j
z → z if z = ai , b i , z ∈ Q 1 .
Computing φ( S̃ ), being φ the composition of all φi ’s, we get

1
φ( S̃ ) = λi a i b i − σi , j u i , j v i + W
λi
i =1 j
which reduced part is exactly

1
− σi , j u i , j v i + W .
λi
i =1 j
Now, if we do integration over massive arrows in (3.1), taking in account that:

∂ S̃ /∂ ai = λi bi + σi , j u i , j ∂ S̃ /∂ bi = λi ai + v i
j
and using the relations ∂ S̃ /∂ ai = 0 and ∂ S̃ /∂ b i = 0 in S̃ we get:

1
− σi , j u i , j v i + W
λi
i =1 j
which is the same as φ( S̃ )red . 2
Corollary 3.8. If Q is a quiver with a good potential S and if k is a vertex such that no related arrows arise in
the mutation procedure, then mutation at k and Seiberg duality at k are isomorphic. In particular mutation of
good potentials give derived equivalent algebras.
Proof. We have that J ( Q triv ⊕ Q red , S triv + S red ) ∼

= J ( Q red , S red ). Conjugating this fact with Remark 1.5
and Theorems 3.4 and 3.7, we get the result. 2
4. An example
Given a Del Pezzo surface S, we can realise its derived category of coherent sheaves as a path
algebra with relations. This can be done using strongly exceptional collections. For the purpose of
what follows, let us recall some results and definitions.
Definition 4.1. An exceptional collection on a projective surface is a collection of coherent sheaves

{ E 1 , . . . , E n } such that:
• Extk ( E i , E i ) = 0, ∀k > 0 and Hom( E i , E i ) = K,

• Extk ( E i , E j ) = 0, ∀1 j < i n, ∀k > 0,
• the stalk complexes of these sheaves generate D b (Coh( X )) as a triangulated category.
It is strongly exceptional if, furthermore, Extk ( E i , E j ) = 0, ∀1 i , j n, ∀k > 0.
Theorem 4.2. If S is a Del Pezzo surface, we have a strongly exceptional collections of sheaves given by:
• { O , O (1), O (2)} if S = P 2 ,
• { O , O (1, 0), O (0, 1), O (1, 1)} if S = P 1 × P 1 ,
• { O , O ( E 1 ), . . . , O ( E r ), O (1), O (2)} if S is d P r with r 8, where each E i is an exceptional curve of the
blow up and d P r is the Del Pezzo obtained by blowing up 1 r 8 points in P 2 .
Definition 4.3. Let X be a non-singular projective variety. A coherent sheaf T is said to be tilting if:
• Extk ( T , T ) = 0, ∀k > 0,
• T generates D b (Coh( X )) as triangulated category,
• B = End( T ) has finite global dimension.
Theorem 4.4. Let X be a non-singular projective variety, T a coherent sheaf on X and B = End( T ). Then the
following are equivalent:
(1) T is tilting;
(2) There is an equivalence Φ : D b (Coh( X )) → D b (mod( B )) of triangulated categories with Φ( T ) = B,
where mod( B ) is the category of finitely generated right modules over B.
Now, if we take the direct some of a strongly exceptional collection over S, we get a tilting sheaf
and, therefore, a derived equivalence between Coh( S ) and K Q / I for some quiver Q and some ideal
of relations I . These are determined looking at the irreducible homomorphisms between the sheaves
in the collection and taking relations between those homomorphisms. See [4,5,7] for further details
on exceptional collections and tilting sheaves.
Example 4.5. d P 1 with exceptional collection { O , O ( E 1 ), O (1), O (2)}
a
•1 •2
b
c1 c2
d1
•4 d2
•3
d3
with relations:
d3 c 1 = d1 c 2 ,
d2 c 1 a = d1 b ,
d3 b = d2 c 2 a.
This example, however, does not fit in our setting of quivers with potentials. In fact what we
ought to consider is not S itself but X = ω S —the total space of the canonical bundle of S—instead.
Calabi–Yau three-fold and if we let π : X → S be the natural projection, we get
This is a local
B̃ = End X ( i π ∗ E i ) is derived equivalent to Coh( X ), where ( E i )i is an exceptional collection over S.
This algebra B̃ can also be seen as a path algebra of a quiver which can be obtained from the cor-
respondent Del Pezzo quiver adding one arrow for each relation in the opposite direction of the
composition of arrows in that relation. These will be quivers with potentials. This procedure is de-
scribed in a general situation in [3].
Example 4.6. The completed quiver for d P 1 with exceptional collection { O , O ( E 1 ), O (1), O (2)} is:
a
Q = •1 •2
R3 b
R2 R1 c1 c2
d1
•4 d2
•3
d3
with potential:
S = R 3 (d3 c 1 − d1 c 2 ) + R 1 (d1 b − d2 c 1 a) + R 2 (d2 c 2 a − d3 b).
Note that this is a good potential. Therefore, using our results, mutations on any vertex of this
quiver will give us derived equivalent path algebras. Since the one above is derived equivalent to
Coh( X ), so will be μk ( Q , S ). Let us finish by presenting μ1 ( Q , S ).
a∗
Q̃ = •1 •2
[aR 1 ]
R3 b∗
R ∗1 R ∗2 c1 c2
[aR 2 ]
d1
• 4 d2
•3
d3
[bR 1 ]
[bR 2 ]
We take a cyclically equivalent potential since there are terms on it starting and ending at 1. Then
let us substitute paths of length two passing through 1 by new arrows and add Δ1 .
S̃ = R 3 d3 c 1 − R 3 d1 c 2 − d2 c 1 [aR 1 ] + d1 [bR 1 ] − d3 [bR 2 ] + d2 c 2 [aR 2 ]
+ [aR 1 ] R ∗1 a∗ + [aR 2 ] R ∗2 a∗ + [bR 1 ] R ∗1 b∗ + [bR 2 ] R ∗2 b∗ .
Clearly this potential is not reduced. Let us consider the following right equivalence
φ: K Q̃ → K Q̃ ,
d1 → d1 − R ∗1 b∗ ,
d3 → −d3 + R ∗2 b∗ ,
[bR 1 ] →
[bR 1 ] + c 2 R 3 ,
[bR 2 ] →
[bR 2 ] + c 1 R 3 ,
u → u if u = d1 , d3 , [bR 1 ], [bR 2 ], u ∈ Q 1 .
If we compute φ( S̃ ), it is of the form S + S̄ and thus we can take the reduced part or integrate over
massive arrows. In any case, as proved in Theorem 3.7, we get the same result which is:
a∗
Q̄ = •1 •2
[aR 1 ]
R3 b∗
R1 R2 c1 c2
[aR 2 ]
d2
•4 •3
with potential
S̄ = c 2 R 3 R ∗1 b∗ + c 1 R 3 R ∗2 b∗ + d2 c 2 [aR 2 ] − d2 c 1 [aR 1 ] + [aR 1 ] R ∗1 a∗ + [aR 2 ] R ∗2 a∗ .
Since this new jacobian algebra is derived equivalent to J ( Q , S ), it is also derived equivalent to
Coh( X ).
References
[1] P. Aspinwall, I. Melkinov, D-Branes on vanishing del Pezzo surfaces, J. High Energy Phys. 0412 (2004) 42.
[2] P.S. Aspinwall, L.M. Fidkowski, Superpotentials for quiver gauge theories, arXiv: hep-th/0506041.
[3] I. Assem, T. Brüstle, R. Schiffler, Cluster-tilted algebras as trivial extensions, Bull. Lond. Math. Soc. 40 (2008).
[4] D. Baer, Tilting sheaves in representation theory of algebras, Manuscripta Math. 60 (1988) 3.
[5] A. Bondal, Representation of associative algebras and coherent sheaves, Math. USSR Izv. 34 (1) (1990) 23–42.
[6] V. Braun, On Berenstein–Douglas–Seiberg duality, arXiv: hep-th/0211173v1.
[7] T. Bridgeland, T-structures on some local Calabi–Yau varieties, J. Algebra 289 (2) (2005) 453–483.
[8] H. Derksen, J. Weyman, A. Zelevinsky, Quivers with potentials and their representations I: Mutations, arXiv: 0704.0649v2
[math.RA].
[9] S. Mukhopadhyay, K. Ray, Seiberg duality as derived equivalence for some quiver gauge theories, arXiv: hep-th/0309191v2.
[10] J. Rickard, Morita theory for derived categories, J. London Math. Soc. 39 (1989).
Journal of Algebra
Growth of primary decompositions of Frobenius powers

of ideals
Trung T. Dinh
Department of Mathematics, 155 S 1400 E, University of Utah, Salt Lake City, UT 84112, USA
Article history: It was previously known, by work of Smith–Swanson and of Sharp–

Received 2 April 2008 Nossem, that the linear growth property of primary decompositions
Available online 5 December 2008 of Frobenius powers of ideals in rings of prime characteristic has
Communicated by Luchezar L. Avramov
strong connections to the localization problem in tight closure
Keywords:
theory. The localization problem has recently been settled in
Commutative rings negative by Brenner and Monsky, but the linear growth question
Frobenius powers is still open. We study growth of primary decompositions of
Linear growth of primary decompositions Frobenius powers of dimension one homogeneous ideals in graded
Associated primes rings over fields. If the ring is positively graded we prove that the
linear growth property holds. For non-negatively graded rings we
are able to show that there is a “polynomial growth”. We present
explicit primary decompositions of Frobenius powers of an ideal,
which were known to have infinitely many associated primes,
having this linear growth property. We also discuss some other
interesting examples.
1. Introduction
Let R be a Noetherian commutative ring of prime characteristic p. Let I be an ideal of R. For each
q = p e the qth Frobenius power of I , denoted by I [q] , is the ideal of R generated by qth powers of
generators of I . We say that the Frobenius powers of I have linear growth of primary decompositions if
there exists a number c such that, for each q = p e , there is a primary decomposition
I [q] = Q 1 ∩ Q 2 ∩ · · · ∩ Q r
√
with the property that ( Q i )c [q] ⊆ Q i for all i = 1, . . . , r. (This number r depends on q.)
E-mail address: dinh@math.utah.edu.
doi:10.1016/j.jalgebra.2008.10.022
830 T.T. Dinh / Journal of Algebra 321 (2009) 829–846
The linear growth property of primary decompositions of Frobenius powers of ideals has strong
connections to the localization problem in tight closure theory. This connection was first discovered
by Smith and Swanson in [8]. In that paper they proved that if the linear growth property holds for an
ideal, then tight closure commutes with the localization of that ideal at one element. Later, Sharp and
Nossem
[6] showed that, under the additional hypothesis that the ring R has a test element, and the
set q= p e Ass R / I [q] is finite, tight closure of I commutes with localization at arbitrary multiplicative
subsets as long as the linear growth property holds for the Frobenius powers of I .
Beside its importance in tight closure theory, the problem of investigating the linear growth prop-
erty of primary decompositions of Frobenius powers is interesting in its own right. One of the reasons
is that the ordinary powers of ideals were shown to have the linear growth property. (The definition
of linear growth property for ordinary powers is obtained by replacing Frobenius powers in the state-
ment by the corresponding ordinary powers.) This fact was first proved by Swanson [9], and Sharp [5]
and Yao [12] later independently gave shorter proofs. Though these proofs use different methods, they
rely on the technique of passing to Rees algebras to reduce the problem to the case of principal ide-
als
generatedn by nonzero divisors, and the fact that for any ideal I in a Noetherian ring R, the set
n>0 Ass R / I is finite. One cannot hope to use similar ideas to attack the linear growth question
of Frobenius powers of ideals, for there is no analogue
of the concept of Rees algebras associated to
Frobenius powers of ideals. Moreover, the set [q] may be infinite due to examples of
q= p e Ass R / I
Katzman [3] and of Singh and Swanson [7].
It is an open question whether every ideal in Noetherian rings of prime characteristic has the
linear growth property. There are no known examples of ideals for which the linear growth property
does not hold. Due to the lack of techniques available in the literature to attack the problem, there
has been a very limited effort to show that certain (classes of) ideals have the linear growth property.
To the author’s knowledge the only work toward this question, besides the very general statement of
Sharp and Nossem mentioned above, was the work of Smith and Swanson in [8]. They proved that
monomial ideals in a monomial ring have the linear growth property. Part of that paper was also
devoted to analyzing primary decompositions of the Frobenius powers of the ideal I = (x, y ) in the
hypersurface
k[t , x, y ]
R= ,
(xy (x − y )(x − t y ))
in which k is
a field of positive characteristic p. This is the well-known example with the property
that the set q= p e Ass R / I [q] is infinite, due to Katzman [3]. Smith and Swanson proved that this ideal
has linear growth of primary decompositions. In [10, §3], Swanson raised the question whether for
each concrete ideal I available in the literature with the property that q= p e Ass R / I [q] is infinite, we
can find “good” primary decompositions, namely decompositions growing linearly.
This paper was inspired by the work of Smith–Swanson and was partially motivated by the above
question of Swanson. It is thus mainly concerned with homogeneous ideals of dimension one in
graded rings over fields of prime characteristics. In Section 2 we give a short proof of the fact that
for dimension one homogeneous ideals in a finitely generated, positively graded ring over a field of
positive characteristic, the linear growth property holds.
Section 3 studies dimension one homogeneous ideals in non-negatively graded, finitely generated
rings over fields of positive characteristics. We are not able to show that the Frobenius powers of these
ideals have linear growth of primary decompositions, but are able to prove that there are primary
decompositions having polynomial growth. A consequence of this result, however, produces a class of
ideals having the linear growth property, which includes Katzman’s ideal as a special case.
Section 4 analyzes primary decompositions of the Frobenius powers of the ideal I = (u , v , x, y ) in
the hypersurface
k[t , u , v , x, y ]
R= .
(u 2 x2 + tuvxy + v 2 y 2 )
T.T. Dinh / Journal of Algebra 321 (2009) 829–846 831
This hypersurface was studied

by Singh and Swanson in [7]. By a slight modification of their proofs,
we show that the set q= p e Ass R / I [q] is infinite. We prove that this ideal has the linear growth
property.
Sections 5 and 6 are reserved for discussions of some other examples. In particular, we would like
to pose a question on the finiteness of the set of associated primes of the Frobenius powers of the
ideal studied by Monsky in [4], which was used recently by Brenner and Monsky in [1] to disprove
the localization conjecture for tight closure. This is an interesting question since its answer will likely
give us some information on the validity of the linear growth question.
2. Dimension one homogeneous ideals in positively graded rings
It was shown independently by Huneke [2] and Vraciu [11] that, for dimension one homogeneous
ideals in a finitely generated, positively graded ring over a field of positive characteristic, tight closure
commutes with localization. Using some of the main results in these papers, we show in this section
that such ideals have the linear growth property. In view of the result of Sharp and Nossem men-
tioned in the introduction, this gives an alternative proof of the commutativity of tight closure and
localization for this class of ideals.
We first, however, start with a remark that for the linear growth question in general, we need only
find bounds for the embedded components.
Proposition 2.1. Let R be a Noetherian ring of prime characteristic p and I an ideal of R. Then there is a num-
ber c such that for each q = p e and for each primary decomposition
I [q] = Q 1 ∩ Q 2 ∩ · · · ∩ Q r ,
√
we have ( Q i )c [q] ⊆ Q i for all isolated components Q i .
Proof. If P is a minimal√prime over I , then I R P ∩ R is the unique isolated component of I whose

radical is P , hence P = I R P ∩ R. There are only a finite number of minimal primes over I , thus
we can find a number c such that P c ⊆ I R P ∩ R for all P minimal over I . Now let q = p e . For each
prime ideal P minimal over I we have Q = I [q] R P ∩ R is the unique isolated component of I [q] whose
radical is P , and

( Q )c[q] = P c[q] ⊆ ( I R P ∩ R )[q] ⊆ I [q] R P ∩ R = Q .
This gives us the conclusion of the proposition. 2
Theorem 2.2. Let R be a finitely generated, positively graded ring over a field of prime characteristic p. Let I
be a homogeneous ideal of height d − 1, where d is the dimension of R. Then the Frobenius powers of I have
linear growth of primary decompositions.
Proof. Denote by m the maximal homogeneous ideal of R. Let z ∈ m be such that it is not contained
in any minimal prime of I . We claim that there is a constant c such that
I [q] : z∞ = I [q] : zcq

for all q = p e . Here we define I [q] : J ∞ = n>0 I [q] : J n for each ideal J of R.
To prove the claim, we note that it was proved in [11], or [2], that there is a constant c such that
I [q] : m∞ = I [q] : mcq

for all q = p e . Let a ∈ I [q] : z∞ be an arbitrary element. Then azn ∈ I [q] for some n. Since z is not in
any minimal prime of I , the ideal ( zn , I [q] ) contains some power of m. Hence
a ∈ I [q] : m∞ = I [q] : mcq ⊆ I [q] : zcq .
Now the claim is proved, and it follows that

I [q] = I [q] + R zcq ∩ I [q] : z∞ .
Since z is not contained in any minimal prime of I , the ideal I [q] : z∞ is precisely the intersection of
all isolated components of any primary decomposition of I [q] , and the ideal I [q] + R zcq is primary to

the homogeneous maximal ideal m. Pick a number c such that I + R zc ⊇ mc . Then

I [q] + R zcq ⊇ mc [q] .
Therefore, combining with Proposition 2.1, from the decomposition

I [q] = I [q] + R zcq ∩ I [q] : z∞ ,
we obtain primary decompositions with the desired property. 2
3. Dimension one homogeneous ideals in non-negatively graded rings
Let k be a field of positive characteristic p. We define an N-grading on the polynomial ring

k[t , x1 , . . . , xn ] as follows: deg t = 0 and deg xi = 1 for all i = 1, . . . , n. In this section we study primary
decompositions of the Frobenius powers of the dimension one ideal I = (x1 , . . . , xn ) in the graded ring
k[t , x1 , . . . , xn ]
R=
( f1, f2, . . . , fr )
in which f 1 , f 2 , . . . , f r are homogeneous polynomials. We are not able to prove that the linear growth
property holds in this case, but are able to establish polynomial growth. We will need some prepara-
tion.
The following lemma was extracted from the proof of Lemma 2.1 in [8].
Lemma 3.1. Let k be a field of prime characteristic p and
k[t , x1 , . . . , xn ]
R= ,
( f1, . . . , fr )
where t , x1 , . . . , xn are variables and f 1 , . . . , f r are homogeneous polynomials (with respect to the grading
defined earlier) of positive degrees. Let I = (x1 , . . . , xn ). Suppose that for each q = p e there is a polynomial
hq (t ) ∈ k[t ] such that the ideal I [q] : hq (t ) is primary to the ideal (x1 , . . . , xn ).
s s s
Let hq (t ) = τ1 1 τ2 2 · · · τl l be the decomposition of hq (t ) into a product of powers of distinct irreducible
polynomials. (This number l may depend on q.) Then there is a constant c such that for each q = p e the
ideal I [q] has a primary decomposition
I [q] = Q ∩ Q 1 ∩ · · · ∩ Q l ,
in which Q is primary to (x1 , . . . , xn ), and Q i = I [q] + R τi i is primary to (x1 , . . . , xn , τi ) for each i = 1, . . . , l.

s
Furthermore,

( Q i )cq+si ⊆ Q i
for all i = 1, . . . , l.
Proof. Saying that I [q] : hq (t ) is primary to (x1 , . . . , xn ) is equivalent to saying that

I [q] : hq (t ) = I [q] : hq (t ) : g (t )
for all g (t ) ∈ k[t ] \ {0}. In particular when applied to g (t ) = hq (t ) we obtain that

I [q] = I [q] + Rhq (t ) ∩ I [q] : hq (t ) .
The ideal Q = I [q] : hq (t ) is primary to (x1 , . . . , xn ) as defined. We want to analyze primary decompo-
sitions of the ideal I [q] + Rhq (t ).
The ideal I [q] + Rhq (t ) has dimension 0, thus every minimal prime must be maximal. Each maximal
ideal containing I [q] + Rhq (t ) must be of the form (x1 , . . . , xn , τi ) for some i. We have the following
obvious inclusion

I [q] + Rhq (t ) ⊆ I [q] + R τ1 1 ∩ I [q] + R τ2 2 ∩ · · · ∩ I [q] + R τl l .
s s s
For the reverse inclusion, let Q i be the unique (x1 , . . . , xn , τi )-primary component of I [q] + Rhq (t ).
Then

Q i = I [q] + Rhq (t ) R (x1 ,...,xn ,τi ) ∩ R ⊇ I [q] + R τi i .
s
Hence

I [q] + Rhq (t ) = I [q] + R τ1 1 ∩ I [q] + R τ2 2 ∩ · · · ∩ I [q] + R τl l .
s s s
Our discussion so far leads to the following primary decomposition

I [q] = I [q] : hq (t ) ∩ I [q] + R τ1 1 ∩ I [q] + R τ2 2 ∩ · · · ∩ I [q] + R τl l
s s s
= Q ∩ Q 1 ∩ Q 2 ∩ · · · ∩ Q l,
√
in which Q i = (x1 , . . . , xn , τi ), and clearly
q s
( Q i )nq+si = (x1 , . . . , xn , τi )nq+si ⊂ x1 , . . . , xn , τi i = I [q] + R τi i ,
q s
as desired. 2
The lemma above in some sense says that there are primary decompositions of the Frobenius
powers of the ideal I which grow at most as fast as the “primary decompositions” of the polynomi-
als hq (t ). The proposition below shows that there are such polynomials whose “primary decomposi-
tions” grow polynomially. We first need a lemma.
Lemma 3.2. Let R be a domain and (ai j : i = 1, . . . , k, j = 1, . . . , l) a matrix with entries in R. Let b1 , . . . , bl
be elements in the quotient field of R such that for each i = 1, . . . , k, the sum ai1 b1 + · · · + ail bl is an element
of R and is divisible by every nonzero minor of the matrix (ai j ). Then there exist b1 , b2 , . . . , bl in R such that
ai1 b1 + · · · + ail bl = ai1 b1 + · · · + ail bl for all i = 1, . . . , k.
Proof. We may assume that k l and that the matrix (ai j ) has maximal rank. We may then assume
that the submatrix (ai j : 1 i , j k) has nonzero determinant. Now let (b1 , . . . , bl ) be the unique
solution in the quotient field of R of the linear system
⎡ ⎤⎡ ⎤ ⎡ ⎤
a11 a12 ··· · · · a1l x1 a11 b1 + · · · + a1l bl
⎢ a21 a22 ··· · · · a2l ⎥ ⎢ x2 ⎥ ⎢ a21 b2 + · · · + a2l bl ⎥
⎢ ⎥⎢ ⎥ ⎢ ⎥
⎢ · · ··· ··· · ⎥⎢ · ⎥ ⎢ · ⎥
⎢ ⎥⎢ ⎥ ⎢ ⎥
⎢ ak1 ak2 ··· · · · akl ⎥ ⎢ · ⎥ = ⎢ ak1 bl + · · · + akl bl ⎥ .
⎢ ⎥⎢ ⎥ ⎢ ⎥
⎢ 0 ··· 1 ··· 0 ⎥⎢ · ⎥ ⎢ 0 ⎥
⎣ · · ··· ··· · ⎦ ⎣ · ⎦ ⎣ · ⎦
0 ··· 0 ··· 1 xl 0
Using Cramer’s rule and the fact that the determinant of the coefficient matrix divides every entry
of the column vector on the right-hand side, we can conclude that this solution is in R. 2
Proposition 3.3. Let S = k[t , x1 , . . . , xn ] be a polynomial ring in n + 1 variables over a field k. Define the grad-
ing as in the beginning of the section. Let f 1 , . . . , f r be homogeneous polynomials of positive degrees. Then for
q q q
each positive integer q there is a polynomial hq (t ) in k[t ] such that the ideal (x1 , x2 , . . . , xn , f 1 , . . . , f r ) : S hq (t )
is primary to (x1 , . . . , xn ).
Moreover these polynomials hq can be chosen to have the following property: there is a constant c not
s s s
depending on q such that for each decomposition hq (t ) = τ1 1 τ2 2 · · · τl l into a product of powers of distinct
irreducible polynomials, the numbers s j are bounded above by cqn−1 .
Proof. To simplify the notation let x denote x1 , . . . , xn , and for each u = (u 1 , . . . , un ) ∈ Nn , where N
u u
denotes the set of non-negative integers, let xu stand for x1 1 · · · xn n , |u| stand for the integer
u 1 + u 2 + · · · + un and
u
denote max{u i : i = 1, . . . , n}.
Let deg f i = di for each i = 1, . . . , r and write

fi = A i , v xv ,
|v|=di
where A i ,v ∈ k[t ] for all i and v. We call the xv the monomial terms of f i .
For each positive integer d we define the matrix M d as follows. The rows of this matrix are indexed
by the vectors u ∈ Nn with |u| = d and
u
< q. Its columns are indexed by the vectors wi ∈ Nn with
|wi | = d − di and i = 1, . . . , r. For each such u and such wi the (u, wi )-entry is A i ,v if wi + v = u and
is 0 otherwise. Now we define hq (t ) to be the least common multiple of all the nonzero minors of all
matrices M d with 1 d n(q − 1).
It is clear that there is a number α such that the sizes of the matrices M d with d = 1, . . . , n(q − 1),
are bounded above by α qn−1 , for all q. Hence the nonzero minors of these matrices are polynomials
in t of degrees bounded above by β qn−1 for some β not depending on q. The polynomial hq (t ) was
defined to be the least common multiple of all these polynomials, thus there is a positive integer c
s s s
such that for every decomposition hq (t ) = τ1 1 τ2 2 · · · τl l into a product of powers of distinct irreducible
n−1
polynomials, the numbers s j are bounded by cq , for all q.
The main task now is to show that for any g (t ) ∈ k[t ] \ {0} we have
q q q q
x1 , . . . , xn , f 1 , . . . , f r : hq (t ) = x1 , . . . , xn , f 1 , . . . , f r : hq (t ) g (t ).
q q
Let f ∈ (x1 , . . . , xn , f 1 , . . . , f r ) : hq (t ) g (t ). That means
q q
f g (t )hq (t ) ∈ x1 , . . . , xn , f 1 , . . . , f r .
q q
We need to show that f hq (t ) ∈ (x1 , . . . , xn , f 1 , . . . , f r ).
q q
Since (x1 , . . . , xn , f 1 , . . . , f r ) is homogeneous, we may assume that f is homogeneous as well. We
q
may further assume that none of the monomial terms of f is divisible by any of xi and that the
degree d of f is at most n(q − 1). We have a presentation

q
f g (t )hq (t ) = b i xi + c 1, w xw f1 + · · · + c r , w xw fr c j ,w ∈ k[t ], b i ∈ S
|w|=d−d1 |w|=d−dr
q

= b i xi + c 1, w A 1, v xu + · · · + c r , w A r , v xu
|u|=d w+v=u |u|=d w+v=u
q

= b i xi + c 1, w A 1, v + c 2, w A 2, v + · · · + c r , w A r , v xu
|u|=d w+v=u w+v=u w+v=u

= c 1, w A 1, v + c 2, w A 2, v + · · · + c r , w A r , v xu .
|u|=d,
u
<q w+v=u w+v=u w+v=u
The last equality follows from the assumption that none of the monomial terms of f is divisible by
q
any of xi for all i. It follows from this equality that for each u with |u| = d and
u
< q we have

c 1, w A 1, v + c 2, w A 2, v + · · · + cr ,w A r ,v ≡ 0 mod g (t )hq (t ) ,
w+v=u w+v=u w+v=u
which implies that

c 1, w c 2, w c r ,w
A 1, v + A 2, v + · · · + A r ,v
g (t ) g (t ) g (t )
w+v=u w+v=u w+v=u
is an element of k[t ] and is divisible by hq (t ). Apply Lemma 3.2 in this situation, with the matrix (ai j )
c j ,w
taken as the matrix M d defined earlier, and the b j taken as the g (t )
. Note that by the definition
of hq (t ), the above sum is divisible by each nonzero minor of the matrix (ai j ). Therefore by Lemma 3.2
we can find {c 1 ,w : |w| = d − d1 }, . . . , {cr ,w : |w| = d − dr } in k[t ] such that for each u with |u| = d and
u
< q,
c 1, w c r ,w
A 1, v + · · · + A r ,v = c 1 ,w A 1,v + · · · + cr ,w A r ,v .
g (t ) g (t )
w+v=u w+v=u w+v=u w+v=u
This in turn implies that

f hq (t ) = c 1 ,w A 1,v + c 2 ,w A 2,v + · · · + cr ,w A r ,v xu
|u|=d,
u
<q w+v=u w+v=u w+v=u

bi xi + c 1 ,w A 1,v + c 2 ,w A 2,v + · · · + cr ,w A r ,v xu
q
=
|u|=d w+v=u w+v=u w+v=u

q u
= b i xi + c 1, w A 1, v x + · · · + c r , w A r , v xu
|u|=d w+v=u |u|=d w+v=u

bi xi + c 1 ,w xw cr ,w xw
q
= f1 + · · · + fr ,
|w|=d−d1 |w|=d−dr
where the bi are polynomials in S. Hence f ∈ (x1 , . . . , xn , f 1 , . . . , f r ) : hq (t ).

q q
2
The main result of this section is the following
Theorem 3.4. Let k be a field of prime characteristic p. Fix n 2 and set
k[t , x1 , . . . , xn ]
R= ,
( f1, . . . , fr )
where t , x1 , . . . , xn are variables and f 1 , . . . , f r are homogeneous polynomials with respect to the grading in
which deg t = 0 and deg xi = 1 for i = 1, . . . , n. Let I = (x1 , . . . , xn ). Then there is a constant c such that for
each q = p e there is a primary decomposition
I [q] = Q 1 ∩ Q 2 ∩ · · · ∩ Q m
√ n −1
with the property that ( Q i )c [q ] ⊆ Q i for all i = 1, . . . , m.
Proof. We may assume that the f i are of positive degrees. Proposition 3.3 guarantees the existence
of polynomials hq (t ) ∈ k[t ] such that for each q = p e the ideal I [q] : hq (t ) is primary to the ideal
(x1 , . . . , xn ). Moreover, there is a constant c with the property that for every irreducible decomposition
s s
hq (t ) = τ1 1 · · · τl l ,
the exponents s j are bounded above by cqn−1 . Now the conclusion of the theorem follows from
Lemma 3.1 and Proposition 2.1. 2
Corollary 3.5. Let k be a field of prime characteristic p and consider
k[t , x, y ]
R= ,
( f1, . . . , fr )
where f 1 , . . . , f r are homogeneous polynomials with respect to the grading in which deg t = 0 and deg x =
deg y = 1. Then the Frobenius powers of the ideal I = (x, y ) have linear growth of primary decompositions.
4. Explicit primary decompositions of an example of Singh and Swanson
Let k be a field of prime characteristic p. It was proved in [7] that, for the integral domain
k[t , u , v , x, y ]
R= ,
(u 2 x2 + tuxv y + v 2 y 2 )
the set
R
Ass
(xq , yq )
q= p e
is infinite. By a slight modification of their proofs we show in this section that the same conclusion
holds for the Frobenius powers of the ideal (u , v , x, y ), namely that the set
R
Ass
(uq , v q , xq , yq )
q= p e
is also infinite. Therefore, as mentioned in the introduction of this paper, we want to analyze primary
decompositions of the Frobenius powers of the dimension one ideal I = (u , v , x, y ) in this hypersur-
face. We show that they have linear growth property.
To prove the infiniteness of the set of associated primes mentioned above, we need a modification
of Proposition 2.2 in [7].
Proposition 4.1. Let A be an N-graded ring which is generated, as an A 0 -algebra, by nonzero divisors t 1 , . . . , tn
of degree 1. Let R be the extension ring
A [u 1 , . . . , un , x1 , . . . , xn ]
R= .
(u 1 x1 − t 1 , . . . , un xn − tn )
Let m1 , . . . , mn be positive integers and f ∈ A a homogeneous element of degree r. Then whenever k1 , . . . , kn

are positive integers such that ki + mi > r for all i, we have
m +k1 n +kn 1 +k1 n +kn

m m
x1 1 , . . . , xm
n , um
1 , . . . , um
n
k k
R : A 0 f x11 · · · xnn = t 1 1 , . . . , tn n A : A 0 f . 2
Proof. The inclusion ⊇ is obvious. Define a Zn+1 -grading in R as follows: deg xi = e i , deg u i =
en+1 − e i and deg h = ren+1 if h ∈ A r , where e i = (0, . . . , 1, . . . , 0) is the ith standard basis element
of Zn+1 . Now let h ∈ A 0 be such that
k m +k1 n +kn 1 +k1 n +kn

k
hf x11 · · · xnn ∈ x1 1 , . . . , xm
n , um
1 , . . . , um
n R.
Then there are homogeneous elements c 1 , . . . , cn , b1 , . . . , bn in R satisfying
k m +k1 n +kn 1 +k1 n +kn

k
hf x11 · · · xnn = c 1 x1 1 + · · · + cn xm
n + b1 um
1 + · · · + bn u m
n .
The degree of the element on the left is ren+1 + k1 e 1 + · · · + kn en = (k1 , . . . , kn , r ). By the assump-
tion that ki + mi > r for all i we can easily see that b1 = · · · = bn = 0. The proposition now follows
from [7, Proposition 2.2]. 2
We will also make use of the following result from [7].
Lemma 4.2. (See [7, Lemmas 4.4(3) and 3.3].) Consider the polynomial ring k[t , x, y ]. Define the sequence of
polynomials P 0 = 1, P 1 = t, P n+1 = t P n − P n−1 for n = 1, 2, . . . .
(a) For any positive integer n, one has

xn , yn , x2 + txy + y 2 :k[t ] xyn−1 = ( P n−1 ).
(b) If char k = p > 0 then the set { P p e −2 : e ∈ N} has infinitely many irreducible factors.
Theorem 4.3. Let k be a field of prime characteristic p and
k[t , u , v , x, y ]
R= .
(u 2 x2 + tuxv y + v 2 y 2 )
Then the set
R
Ass
(uq , v q , xq , yq )
q= p e
is infinite.
Proof. For each q = p e define the ideal

J q = u q , v q , xq , y q R : R (ux)( v y )q−2 uv .
k[t ][a,b]
Set A 0 = k[t ] and A = a2 +tab+b2
. With this setting then the ring R can be presented as
A [u , v , x, y ]
R= .
(ux − a, v y − b)
Thus applying Proposition 4.1 with f = abq−2 ∈ A q−1 , we obtain

J q ∩ k[t ] = u q , v q , xq , y q R :k[t ] (ux)( v y )q−2 uv

= aq−1 , bq−1 A :k[t ] abq−2
= ( P q−2 ) by Lemma 4.2(a).
This implies that

u q , v q , xq , y q , P q−2 ⊆ J q ⊆ (u , v , x, y , P q−2 ).

Hence Min( J q ) = Min(u , v , x, y , P q−2 ). The set e∈N+ Min(u , v , x, y , P p e −2 ) is infinite since the set of
irreducible factors of the members of { P p e −2 : e ∈ N+ } is infinite by Lemma 4.2(b). Thus the set
R
Ass
(uq , v q , xq , yq )
q= p e
is infinite as well. 2
We now try to analyze primary decompositions of the Frobenius powers of the ideal I = (u , v , x, y )
in the ring of Theorem 4.3. We will actually work in a more general context, namely in the hypersur-
faces
k[t , u , v , x, y ]
R=
(r0 u 2 x2 + r1 uxv y + r2 v 2 y 2 )
where r0 , r1 , r2 are polynomials in k[t ]. We will show that the Frobenius powers of the ideal
I = (u , v , x, y ) have linear growth of primary decompositions for a large class of polynomials r0 ,
r1 , r2 , which includes the example above.
For each set of polynomials r0 , r1 , r2 in k[t ] we define a generalization of the sequence of polyno-
mials defined in Lemma 4.2 as follows
P 0 = 1,
P 1 = r1 ,
P n+1 = r1 P n − r0 r2 P n−1 for all n 1.

Note that P n is just the determinant of the n × n matrix

⎛ ⎞
r1 r0
⎜ r2 r1 r0 ⎟
⎜ ⎟
Mn = ⎜ · · · ⎟.
⎝ r2 r1 r0
⎠
r2 r1
The following lemma generalizes Lemma 4.4(1) from [7].
Lemma 4.4. Consider the polynomial ring k[t , x, y ] and an integer n 1. Then

xyn−1 P n−1 ∈ xn , yn , r0 x2 + r1 xy + r2 y 2 .
Proof. With the grading deg t = 0, deg x = deg y = 1, the module
k[t , x, y ]
W =
(xn , yn , r0 x2 + r1 xy + r2 y 2 )
is N-graded. Its nth graded component W n , considered as module over k[t ], is generated by
xn−1 y , xn−2 y 2 , . . . , xyn−1 .
The relations among these generators come from the equations

xn−2−i y i r0 x2 + r1 xy + r2 y 2 = 0, i = 0, . . . , n − 2.
This implies that the relation matrix for W n , as a k[t ]-module, is precisely the matrix M n−1 defined
above. Therefore the determinant of this matrix, which is P n−1 , annihilates W n , and that implies the
conclusion of the lemma. 2
We use this lemma to prove the following
Lemma 4.5. Let k be any field and S = k[t , u , v , x, y ]. For each integer n 0 define

I n = un , v n , xn , yn , r0 u 2 x2 + r1 uxv y + r2 v 2 y 2 .
Assume furthermore that 2 deg r1 > deg r0 + deg r2 . Then
(a) The polynomial P n is nonzero for each n 0.

(b) Denote by L n the least common multiple of all (nonzero) polynomials P 1 , P 2 , . . . , P n−1 . We have
(r0 r2 )a+b Ln (ux)a ( v y )b xc yd ∈ I n
for all a, b, c , d 0 and 2a + 2b + c + d = 2n.

(c) I n : r0n r2n L n ⊇ I n + (u , v , x, y )2n .
Proof. (a) It is easy to show, by induction, that the degree of the polynomial P n is n deg r1 , provided
2 deg r1 > deg r0 + deg r2 . Hence P n is a nonzero polynomial for each n 0.
(b) We shall prove that for c d we have
r0a L n (ux)a ( v y )b xc yd ∈ I n ,
and for c d,
r2b L n (ux)a ( v y )b xc yd ∈ I n .
It is clear that we may assume that c d without loss of generality. We shall proceed by induction
on b. If b = 0 then a + c n, hence (ux)a xc ∈ I n . The next step in the induction procedure is when
b = 1. We consider two cases, according as c > d and c = d.
If c > d then from 2a + c + d = 2n − 2 we get a + c > n − 1, hence (ux)a xc ∈ I n . Now suppose c = d.
Put m = n − c. Hence a + 1 = m and it suffices to show that

L n (ux)a ( v y ) ∈ (ux)m , ( v y )m , r0 (ux)2 + r1 uxv y + r2 ( v y )2 .
But this follows from Lemma 4.4 above and the definition of L n .
Now let b 2 and suppose the conclusion is true for integers smaller than b. Modulo r0 (ux)2 +
r1 uxv y + r2 ( v y )2 , we have

r2b L n (ux)a ( v y )b xc yd = −r2b−1 L n (ux)a ( v y )b−2 r0 (ux)2 + r1 uxv y xc yd
= −r0 r2b−1 Ln (ux)a+2 ( v y )b−2 xc yd − r1 r2b−1 Ln (ux)a+1 ( v y )b−1 xc yd ,
and by the induction hypothesis this sum belongs to I n .

(c) This follows easily from (b). Indeed, consider a, b, c, d with a + b + c + d = 2n, we want to show
that r0n r2n L n xa y b u c v d ∈ I n . We may assume that a c and b d. Then c + d n and
r0n r2n L n xa y b u c v d = r0n r2n L n (ux)c ( v y )d xa−c y b−d ∈ I n
by (b). 2
Proposition 4.6. With the same notations as in Lemma 4.5, for any g (t ) ∈ k[t ] \ {0} and for any positive
integer n we have

I n + (u , v , x, y )2n : r02n r22n g (t ) = I n + (u , v , x, y )2n : r02n r22n .
Proof. We may assume that r0 r2 is nonzero. Let f (t , u , v , x, y ) be a homogeneous polynomial in u, v,

x, y of degree d < 2n. We need to show that if
f (t , u , v , x, y ) g (t )r02n r22n ∈ I n
for some g (t ) ∈ k[t ] \ {0}, then
f (t , u , v , x, y )r02n r22n ∈ I n .
We may assume that none of the monomial terms of f (t , u , v , x, y ) is divisible by any of un , v n ,

xn , yn . We have a presentation
f (t , u , v , x, y ) g (t )r02n r22n

= Aun + B v n + C xn + D yn + E irjs u i v j xr y s r0 u 2 x2 + r1 uxv y + r2 v 2 y 2 ,
i , j ,r ,s
where A, B, C , D are polynomials in t, u, v, x, y which are homogeneous in u, v, x, y, E irjs are

polynomials in k[t ] and the sum is taken over all i , j , r , s 0 with i + j + r + s = d − 4. In the sequel
when the conditions on the indices are omitted we always mean that the sums are taken so that the
terms are of correct degrees. We may write the right-hand side as
Aun + B v n + C xn + D yn

+ r0 E irjs u i +2 v j xr +2 y s + r1 E irjs u i +1 v j +1 xr +1 y s+1 + r2 E irjs u i v j +2 xr y s+2
= Aun + B v n + C xn + D yn
i −2,r −2 i j r s
i −1,r −1

+ r0 E js u v x y + r1 E j −1,s−1 u i v j xr y s + r2 E irj−2,s−2 u i v j xr y s .
Since we assumed that none of the monomial terms of f (t , u , v , x, y ) is divisible by any of un , v n ,

xn , yn , the terms that are divisible by any of un , v n , xn , yn in the above sum cancel. We finally get
i −2,r −2
f (t , u , v , x, y ) g (t )r02n r22n = r0 E js + r1 E ij−−11,,rs−−11 + r2 E irj−2,s−2 u i v j xr y s , (1)
i + j +r +s=d
0i , j ,r ,sn−1
in which, by convention, E irjs = 0 if min{i , j , r , s} < 0. It follows from (1) that
i −2,r −2
r0 E js + r1 E ij−−11,,rs−−11 + r2 E irj−2,s−2 ≡ 0 mod r02n r22n g (t ) (2)
for all i + j + r + s = d, 0 i , j , r , s n − 1. We claim that in fact g (t ) divides each E irjs that appears
in the above sums.
We first prove the proposition assuming the claim. If g (t ) divides each E irjs , then there are Ē irjs
in k[t ] such that E irjs = g (t ) Ē irjs for all i, j, r, s. It then follows from (1) that
i −2,r −2
f (t , u , v , x, y )r02n r22n = r0 Ē js + r1 Ē ij−−11,,rs−−11 + r2 Ē irj−2,s−2 u i v j xr y s .
i + j +r +s=d
0i , j ,r ,sn−1
Now by tracing back the calculations made in the previous paragraphs and using the same regrouping
arguments as in the end of the proof of Proposition 3.3 we get

f (t , u , v , x, y )r02n r22n = Āun + B̄ v n + C̄ xn + D̄ yn + Ē irjs u i v j xr y s r0 u 2 x2 + r1 uxv y + r2 v 2 y 2 ,
which of course gives us that
f (t , u , v , x, y )r02n r22n ∈ I n .
We now prove the claim. Define the following sets

A 0 = E irjs i + j + r + s = d − 4, −2 i , r < n − 2, 0 j , s < n ,

A 1 = E irjs i + j + r + s = d − 4, −1 i , j , r , s < n − 1 ,

A 2 = E irjs i + j + r + s = d − 4, 0 i , r < n, −2 j , s < n − 2 .
Note that A 0 , A 1 , A 2 are just the sets of all E irjs that appear as coefficients of r0 , r1 , r2 in (1), corre-
spondingly. We have

A 1 \ A 0 = E irjs i + j + r + s = d − 4, n − 2 ∈ {i , r } or − 1 ∈ { j , s} ,

A 1 \ A 2 = E irjs i + j + r + s = d − 4, −1 ∈ {i , r } or n − 2 ∈ { j , s} .
Since i + j + r + s = d − 4 2n − 5, A 1 \ A 0 ∪ A 2 is either empty or consists of only E irjs with

min{i , j , r , s} < 0 which are 0. Hence we can conclude that for each nonzero E irjs that appears in (1),
either E irjs ∈ A 0 or E irjs ∈ A 2 .
We shall need the following subsets

A 0 \ A 2 = E irjs i + j + r + s = d − 4, −2 i , r < n − 2, 0 j , s < n,

{−2, −1} ∩ {i , r } = ∅ or {n − 2, n − 1} ∩ { j , s} = ∅ ,

A 2 \ A 0 = E irjs i + j + r + s = d − 4, 0 i , r < n, −2 j , s < n − 2,

{n − 2, n − 1} ∩ {i , r } = ∅ or {−2, −1} ∩ { j , s} = ∅ .
Fix a nonzero E irjs , we would like to show that it is divisible by g (t ) as claimed. We will consider
three cases, according as E irjs ∈ A 0 \ A 2 , E irjs ∈ A 2 \ A 0 and E irjs ∈ A 0 ∩ A 2 .
First suppose E irjs ∈ A 0 \ A 2 . Then {n − 2, n − 1} ∩ { j , s} =
∅. We claim that, for each integer l such
i +l,r +l i +l,r +l i +l,r +l
that E j −l,s−l exists and has non-negative indices, we have E j −l,s−l ∈ A 0 . If not, suppose E j −l,s−l ∈
A 2 \ A 0 and has non-negative indices. Then, by the description above, we have {n − 2, n − 1} ∩ {i + l,
r + l} = ∅. Thus
i + j + r + s = (i + r ) + ( j + s) (n − 2 − l) + (n − 2 + l) = 2n − 4 > d − 4,
a contradiction. The relation in (2) can be rewritten as
i +1,r +1 i +2,r +2
r0 E irjs ≡ −r1 E j −1,s−1 − r2 E j −2,s−2 mod r02n r22n g (t ).
i +1,r +1 i +2,r +2
By what we just discussed, the elements E j −1,s−1 and E j −2,s−2 either have some negative indices or
belong to A 0 . If any of them belong to A 0 then by multiplying both sides by r0 we can, using (2),
replace it by combinations of some E with all the subscripts decreased by at least 1. Keep doing this
process until we end up with elements with some negative indices, which are 0. Since each time the
subscripts decreased by at least 1, we need at most min{ j , s} steps, hence the power of r0 on the
left-hand side will be at most min{ j , s} + 1. We obtain that
min{ j ,s}+1
r0 E irjs ≡ 0 mod r02n r22n g (t ).
It follows that E irjs ≡ 0 mod g (t ) since min{ j , s} + 1 < 2n.

By a symmetric argument, if E irjs ∈ A 2 \ A 0 then
min{i ,r }+1 ir
r2 E js ≡ 0 mod r02n r22n g (t ),
which results that E irjs ≡ 0 mod g (t ).

Now we consider the last case, namely when E irjs ∈ A 0 ∩ A 2 . We use again the relation
i +1,r +1 i +2,r +2
r0 E irjs ≡ r1 E j −1,s−1 − r2 E j −2,s−2 mod r02n r22n g (t ).
Each nonzero E on the right-hand side is either in A 0 ∩ A 2 or in A 0 \ A 2 or in A 2 \ A 0 . If any of them

is in A 0 ∩ A 2 , then by multiplying both sides by r0 we can express it as a combination of other E
with subscripts decreased by at least 1. Iterating this process, we can conclude that there is a relation

h
i +τ ,r +τ
r0l E irjs ≡ c τ E j −τ ,s−τ mod r02n r22n g (t ),
τ =1
i +τ ,r +τ
in which l, h min{ j , s} + 1 and each E j −τ ,s−τ is either in A 0 \ A 2 or in A 2 \ A 0 . It follows from the
discussion above that for each τ ,
1+min{ j −τ ,s−τ } 1+min{i +τ ,r +τ } i +τ ,r +τ

r0 r2 E j −τ ,s−τ ≡ 0 mod r02n r22n g (t ).
Hence
1+l+min{ j −1,s−1} 1+min{i +h,r +h} ir

r0 r2 E js ≡ 0 mod r02n r22n g (t ),
and this gives us that E irjs ≡ 0 mod g (t ) since
1 + min{i + h, r + h} 2n,
1 + l + min{ j − 1, s − 1} < 2n.
This finishes the proof of the claim, hence of the proposition. 2
Lemma 4.7. Use the same notation and hypothesis as in Lemma 4.5(c). We then have

I n : r03n r23n L n = I n + (u , v , x, y )2n : r02n r22n
and this ideal is primary to the ideal (u , v , x, y ) if r0 r2 is nonzero.
Proof. The inclusion

I n : r03n r23n L n ⊇ I n + (u , v , x, y )2n : r02n r22n
follows from Lemma 4.5. For the converse,

I n : L n r03n r23n ⊆ I n + (u , v , x, y )2n : r02n r22n : r0n r2n L n

= I n + (u , v , x, y )2n : r02n r22n by Proposition 4.6.
The conclusion that ( I n + (u , v , x, y )2n ) : r02n r22n is primary to the ideal (u , v , x, y ) is another immediate
consequence of Proposition 4.6. 2
We now prove the main theorem of this section.
Theorem 4.8. Let k be a field of prime characteristic p. Consider the hypersurface
k[t , u , v , x, y ]
R= ,
(r0 u 2 x2 + r1 uxv y + r2 v 2 y 2 )
where r0 , r1 , r2 are polynomials in k[t ]. Let I = (u , v , x, y ). Suppose that r0 r2 is nonzero and that 2 deg r1 >
deg r0 + deg r2 . Then there is a constant c such that for each q = p e , there is a primary decomposition
I [q] = Q 1 ∩ Q 2 ∩ · · · ∩ Q s
√
satisfying ( Q i )c [q] ⊆ Q i for all i = 1, . . . , s.
Proof. It follows from Lemma 4.7 that the ideal I [q] : r0 r2 L q is primary to the ideal (u , v , x, y ).
3q 3q
Recall that the polynomial L q is defined to be the least common multiple of the polynomials
P 1 , P 2 , . . . , P q−1 . It is readily seen that the degrees of the polynomials P n are bounded above by a lin-
3q 3q
ear function in n. Thus there is a constant c such that for each q = p e the polynomial hq (t ) = r0 r2 L q
has an irreducible decomposition
s s
hq (t ) = τ1 1 · · · τl l
in which si cq for all i = 1, . . . , l. Therefore the Frobenius powers of the ideal I = (u , v , x, y ) have
linear growth of primary decompositions by Lemma 3.1 and Proposition 2.1. 2
5. Another example of Singh and Swanson
Consider the hypersurface
k[t , u , v , w , x, y , z]
R=
(u 2 x2 + v 2 y 2 + tuxv y + t w 2 z2 )
with k an arbitrary field. This is a UFD and is F-regular if char k = p > 0. It was proved in [7, Theo-
rem 6.6] that if char k = p > 0 then the set
R
Ass
(xq , yq , zq )
q= p e
has infinitely many maximal elements. Again, we can modify their proof to obtain the following
Theorem 5.1. Consider the hypersurface
k[t , u , v , w , x, y , z]
R= ,
(u 2 x2 + v 2 y 2 + tuxv y + t w 2 z2 )
where k is a field of prime characteristic p > 0. Then the set
R
Ass
(uq , v q , w q , xq , yq , zq )
q= p e
has infinitely many (maximal) elements.
Proof. Consider the ideal

J q = u q , v q , w q , xq , y q , zq R : R (ux)( v y )q−2 uv w q−1 .
Set A 0 = k[t ] and A = (a2 +kb[t2][+atab

,b,c ]
+tc 2 )
. With this setting then the ring R can be represented as
A [u , v , w , x, y , z]
R= .
(ux − a, v y − b, w z − c )
Applying Proposition 4.1 with f = abq−2 ∈ A q−1 , we obtain

J q ∩ k[t ] = u q , v q , w q , xq , y q , zq R :k[t ] (ux)( v y )q−2 uv w q−1

= uq , v q , w q , xq , yq , zq R :k[t ] f uv w q−1

= aq−1 , bq−1 , c A :k[t ] abq−2
= ( P q−2 ),
where P q−2 is the polynomial defined in Lemma 4.2. Now we can now use the same argument as in
the last paragraph of Theorem 4.3. 2
However we do not know if the Frobenius powers of the ideal (u , v , w , x, y , z) have linear growth
of primary decomposition. Thus we would like to ask:
Question. Let k be a field of prime characteristic p. Do the Frobenius powers of the ideal I = (u , v ,
w , x, y , z) in the hypersurface
k[t , u , v , w , x, y , z]
R=
(u 2 x2 + v 2 y 2 + tuxv y + t w 2 z2 )
have linear growth of primary decompositions?
6. The example of Brenner and Monsky
Consider the hypersurface
k[t , x, y , z]
R= ,
(z4 + z2 xy + zx3 + zy 3 + tx2 y 2 )
where k is an algebraically closed field of characteristic 2. This hypersurface was first studied by Mon-
sky in [4] and recently it was used again by Brenner–Monsky to disprove the localization conjecture in
tight closure theory. More specifically, Brenner and Monsky showed that I ∗S = ( I S )∗ for I = (x4 , y 4 , z4 )
and S = k[t ] \ {0}, where (−)∗ denotes the tight closure operation.
It is reasonable to ask whether there is any example of ideals that do not have the linear growth
property. We restate the theorem of Sharp and Nossem in [6] which relates the linear growth property
and the localization problem.
Theorem 6.1. (See [6, Theorem 7.14].) Let R be a ring of positive characteristic p and I an ideal of R. Suppose

that R has a pm0 -weak test element, and that the set q= p e Ass R / I [q] is finite.
Then the tight closure of I commutes with the localization at any multiplicative subset of R if the Frobenius
powers of I have linear growth of primary decompositions.
Consider the ring R and the ideal I in the example of Brenner and Monsky mentioned above. The
ring R is an affine domain over a field, hence it possesses a test element. Thus, in view of the above
theorem, and the result of Brenner and Monsky, the linear growth property would fail for I if the set
[q] is finite. Thus it would be interesting to know the answer to the following question.
q=2e Ass R / I
Question. Let k be an algebraically closed field of characteristic 2. Set
k[t , x, y , z]
R= .
( z4 + z2 xy + zx3 + zy 3 + tx2 y 2 )
Is the set
R
Ass
(xq , yq , zq )
q=2e
finite?
Acknowledgments
The author would like to thank Irena Swanson for suggesting the problem and for many useful
discussions. He would also like to thank Paul C. Roberts for his financial support. The author is grateful
to Anurag Singh and the referee for several valuable comments.
References
[1] H. Brenner, P. Monsky, Tight closure does not commute with localization, arXiv: 0710.2913.
[2] C. Huneke, The saturation of Frobenius powers of ideals, Special issue in honor of Robin Hartshorne, Comm. Algebra 28 (12)
(2000) 5563–5572.

[3] M. Katzman, Finiteness of e Ass F e ( M ) and its connection to tight closure, Illinois J. Math. 40 (2) (1996) 330–337.
[4] P. Monsky, Hilbert–Kunz functions in a family: Point-S 4 quartics, J. Algebra 208 (1998) 343–358.
[5] R. Sharp, Injective modules and linear growth of primary decompositions, Proc. Amer. Math. Soc. 128 (3) (2000) 717–722.
[6] R. Sharp, N. Nossem, Ideals in a perfect closure, linear growth of primary decompositions, and tight closure, Trans. Amer.
Math. Soc. 356 (9) (2004) 3687–3720.
[7] A. Singh, I. Swanson, Associated primes of local cohomology and of Frobenius powers, Int. Math. Res. Not. 33 (2004) 1703–
1733.
[8] K. Smith, I. Swanson, Linear bounds on growth of associated primes for monomial ideals, Comm. Algebra 25 (1997) 3071–
3079.
[9] I. Swanson, Powers of ideals. Primary decompositions, Artin–Rees lemma and regularity, Math. Ann. 307 (2) (1997) 299–
313.
[10] I. Swanson, Ten Lectures on Tight Closures, IPM Lecture Notes Series, vol. 3, IPM, Tehran, 2002.
[11] A. Vraciu, Local cohomology of Frobenius images over graded affine algebras, J. Algebra 228 (1) (2000) 347–356.
[12] Y. Yao, Primary decomposition: Compatibility, independence and linear growth, Proc. Amer. Math. Soc. 130 (6) (2002)
1629–1637.
Journal of Algebra
Braided bialgebras of Hecke-type ✩

A. Ardizzoni a,∗ , C. Menini a , D. Ştefan b
a
University of Ferrara, Department of Mathematics, Via Machiavelli 35, I-44100 Ferrara, FE, Italy
b
University of Bucharest, Faculty of Mathematics, Academiei 14, RO-010014 Bucharest, Romania
Article history: The paper is devoted to prove a version of Milnor–Moore Theorem

Received 9 April 2008 for connected braided bialgebras that are infinitesimally cocom-
Available online 5 December 2008 mutative. Namely in characteristic different from 2, we prove that,
Communicated by Nicolás Andruskiewitsch
for a given connected braided bialgebra ( A , c A ) which is infinites-
imally λ-cocommutative for some element λ = 0 that is not a root
Keywords:
Braided bialgebras of one in the base field, then the infinitesimal braiding of A is of
Braided enveloping algebras Hecke-type of mark λ and A is isomorphic as a braided bialgebra
Milnor–Moore Theorem to the symmetric algebra of the braided subspace of its primitive
elements.
Introduction
The structure of cocommutative connected bialgebras is well understood in characteristic zero. By

Milnor–Moore Theorem [MM] such a bialgebra is the enveloping algebra of its primitive part, regarded
as a Lie algebra in a canonical way. This result is one of the most important ingredients in the proof
of Cartier–Gabriel–Kostant Theorem [Di,Sw], that characterizes cocommutative pointed bialgebras in
characteristic zero as “products” between enveloping algebras and group algebras.
Versions of Milnor–Moore Theorem and Cartier–Gabriel–Kostant Theorem for graded bialgebras
(over Z and Z2 ) can be also found in the work of Kostant, Leray and Milnor–Moore. For other more
recent results, analogous to Milnor–Moore Theorem, see [Go,Kh,LR,Ma2,R1,R2,St].
The Z2 -graded bialgebras, nowadays called superbialgebras, appeared independently in the work
of Milnor–Moore and MacLane. From a modern point of view, superalgebras can be seen as bialgebras
in a braided monoidal category. These structures play an increasing role not only in algebra (classi-
✩
The paper was written while A. Ardizzoni and C. Menini were members of GNSAGA with partial financial support from
MIUR. D. Ştefan was partially supported by INDAM, while he was visiting professor at University of Ferrara, and by CERES
Project 4-147/2004.
*Corresponding author.
E-mail addresses: alessandro.ardizzoni@unife.it (A. Ardizzoni), men@unife.it (C. Menini), dragos.stefan@fmi.unibuc.ro
(D. Ştefan).
doi:10.1016/j.jalgebra.2008.11.011
848 A. Ardizzoni et al. / Journal of Algebra 321 (2009) 847–865
fication of finite dimensional Hopf algebras or theory of quantum groups), but also in other fields of
mathematics (algebraic topology, algebraic groups, Lie algebras, etc.) and physics. Braided monoidal
categories were formally defined by Joyal and Street in the seminal paper [JS], while (bi)algebras in
a braided category were introduced in [Ma1]. By definition, ( A , ∇, u , , ε ) is a bialgebra in a braided
category M, if ∇ is an associative multiplication on A with unit u and is a coassociative comulti-
plication on A with counit ε such that ε and are morphism of algebras where the multiplication
on A ⊗ A is defined via the braiding. In other words, the last property of rereads as follows
∇ = (∇ ⊗ ∇)( A ⊗ c A , A ⊗ A )( ⊗ ), (1)
where c X ,Y : X ⊗ Y → Y ⊗ X denotes the braiding in M. The relation above had already appeared in
a natural way in [Ra], where Hopf algebras with a projection are characterized. More precisely, let A
be a Hopf algebra and let p : A → A be a morphism of Hopf algebras such that p 2 = p. To these data,
Radford associates an ordinary Hopf algebra B := Im( p ) and a Hopf algebra R = {a ∈ A | ( A ⊗ p )(a) =
a ⊗ 1} in the braided category BB YD of Yetter–Drinfeld modules. Then he shows that A R ⊗ B, where
on R ⊗ B one puts the tensor product algebra and tensor product coalgebra of R and B in the category
B YD (the braiding of B YD is used to twist the elements of R and B). It is worth to notice that the
B B
above Hopf algebra structure on R ⊗ B can be constructed for an arbitrary Hopf algebra R in BB YD .
It is called the bosonization of R and it is denoted by R # B. Notably, the “product” that appears in
Cartier–Gabriel–Kostant Theorem is precisely the bosonization of an enveloping Lie algebra, regarded
as a bialgebra in the category of Yetter–Drinfeld modules over its coradical. The result of Radford was
extended for more general classes of bialgebras with a projection in [AMS1,AMS2].
Bosonization also plays a very important role in the “lifting” method for the classification of finite
dimensional pointed Hopf algebras, introduced by N. Andruskiewitsch and H.-J. Schneider. Roughly
speaking, the lifting method requires two steps. If A is a pointed Hopf algebra, then gr A, the graded
associated of A with respect to the coradical filtration, is a Hopf algebra with projection onto the
homogeneous component of degree 0. Hence, by Radford’s result, gr A is the bosonization of a
graded connected Hopf algebra R in BB YD , where B is the coradical of A. Accordingly to the lift-
ing method, first one has to classify all connected and graded Hopf algebras R in BB YD such that
dim R = dim A / dim B. Then one has to find all Hopf algebras A such that gr A R # B, with R as in
the first step.
Therefore, in many cases, for proving a certain property of Hopf algebras, it suffices to do it in the
connected case. The price that one has to pay is that we have to work with Hopf algebras in a braided
category (usually BB YD ), and not with ordinary Hopf algebras. Motivated by this observation, in this
paper we will investigate connected and cocommutative bialgebras in a braided category. Actually
nowadays people recognize that it is more appropriate to work with braided bialgebras, that were
introduced in [Ta] (see e.g. [Gu3] and [Kh]).
To define a braided bialgebra we first need a braided vector space, that is a pair ( A , c), where A is
a vector space and c is a solution of the braid equation (2). Then we need an algebra ( A , ∇, u ) and
a coalgebra ( A , , ) which are compatible with the braiding (see Definitions 1.2 and 1.4). Now, for
defining braided bialgebras, one can proceed as in the classical case; see Definition 1.6.
The prototype braided bialgebra is T := T ( V , c), the tensor algebra of a braided K -vector space
( V , c). The braiding c lifts uniquely to an operator c T : T ⊗ T → T ⊗ T . The usual algebra structure on
T ( V ) is compatible with c T , so T ⊗ T is an algebra (for the multiplication, of course, we use c T and
not the usual flip map). Therefore, there is a unique coalgebra structure on T , so that V is included
in the space of primitive elements of T and the comultiplication is an algebra map.
For constructing other examples of braided bialgebras, we focus on the case when ( V , c) is a
braided vector space such that c is a braiding of Hecke-type of mark λ ∈ K ∗ , that is c is a root
of ( X + 1)( X − λ). Then, to every K -linear map b : V ⊗ V → V which is compatible with c (i.e.
a so-called braided bracket, see Definition 2.1) we associate a new braided bialgebra U ( V , c, b), called
enveloping algebra, as follows. The set

X c,b = c( z) − λ z − b( z) z ∈ V ⊗ V = T 2 ( V )
A. Ardizzoni et al. / Journal of Algebra 321 (2009) 847–865 849
contains only primitive elements in T ( V , c), so the ideal I c,b generated by X c,b in T ( V , c) is a coideal
too. Hence the quotient U ( V , c, b) of T ( V , c) through I c,b is a braided bialgebra. As a particular case
we obtain the braided symmetric algebra S ( V , c) := U ( V , c, 0).
It is worthwhile noticing that the braided subspace ( P , c P ) of primitive elements of a connected
braided bialgebra ( A , c A ) can always be endowed with a braided bracket b P : P ⊗ P → P whenever
c P is a braiding of Hecke-type. In this case one can consider the corresponding enveloping algebra
(see Proposition 2.4).
The main result of the paper is Theorem 5.5. In fact, we prove that, for a given field K with
char K = 2, if ( A , c A ) is a connected braided bialgebra which is infinitesimally λ-cocommutative for
some regular element λ = 0 in K , then
• the infinitesimal braiding c P of A is of Hecke-type of mark λ, and

• A is isomorphic as a braided bialgebra to the symmetric algebra S ( P , c P ) of ( P , c P ) whenever
λ = 1.
We would like to stress that, since the symmetric algebra is indeed a universal enveloping algebra
with trivial bracket, the foregoing result could be thought as a strong version of Milnor–Moore Theo-
rem for connected braided bialgebras that are infinitesimally cocommutative.
We also point out that, when λ = 1, A is isomorphic to the symmetric algebra S ( P , c P ) just as
coalgebra by means of Kharchenko’s results (see Remark 5.7 and [Kh, Theorem 7.2]).
To achieve our result we characterize bialgebras of type one with infinitesimal braiding of Hecke
type (see Theorem 2.15). Moreover, for a given braided vector space ( V , c) of Hecke-type of mark
λ = 0, 1, we show that b = 0 is the unique c-bracket on ( V , c) for which the K -linear canonical map
ιc,b : V → U ( V , c, b) is injective whenever (3)!λ = 0 (see Theorem 4.3).
It is worth noticing that, as a consequence of Theorem 2.15, we can prove (see Remark 5.6) that
for a connected braided Hopf algebra ( H , c H ) and λ ∈ K ∗ , the following assertions are equivalent:
• H is cosymmetric in the sense of [Kh, Definition 3.1] and its infinitesimal braiding is of Hecke-
type of mark λ,
• H is infinitesimally λ-cocommutative.
On the other hand, as a consequence of Theorem 4.3, we get (see Remark 4.6) that a braided Lie
algebra introduced by Gurevich in [Gu2] has trivial bracket in the Hecke case.
1. Braided bialgebras
Throughout this paper K will denote a field. All vector spaces will be defined over K and the
tensor product of two vector spaces will be denoted by ⊗.
In this section we define the main notion that we will deal with, namely braided bialgebras. We
also introduce one of the basic examples, namely the tensor algebra of a braided vector space.
Definition 1.1. A pair ( V , c) is called braided vector space if c : V ⊗ V → V ⊗ V is a solution of the

braid equation
c1 c2 c1 = c2 c1 c2 (2)
where c1 = c ⊗ V and c2 = V ⊗ c. A morphism of braided vector spaces ( V , c V ) and ( W , c W ) is a

K -linear map f : V → W such that c W ( f ⊗ f ) = ( f ⊗ f )c V .
Note that, for every braided vector space ( V , c) and every λ ∈ K , the pair ( V , λc) is a braided vec-
tor space too. A general method for producing braided vector spaces is to take an arbitrary braided
category (M, ⊗, K , a, l, r , c ), which is a monoidal subcategory of the category of K -vector spaces.
Hence any object V ∈ M can be regarded as a braided vector space with respect to c := c V , V . Here,
c X ,Y : X ⊗ Y → Y ⊗ X denotes the braiding in M. The category of comodules over a coquasitriangu-

lar Hopf algebra and the category of Yetter–Drinfeld modules are examples of such categories. More
particularly, every bicharacter of a group G induces a braiding on the category of G-graded vector
spaces.
Definition 1.2. (See Baez [Ba].) A braided algebra, or c-algebra is a quadruple ( A , ∇, 1, c) where ( A , c)
is a braided vector space and ( A , ∇, 1) is an associative unital algebra such that ∇ and u commute
with c, that is the following conditions hold:
c(∇ ⊗ A ) = ( A ⊗ ∇)(c ⊗ A )( A ⊗ c), (3)
c( A ⊗ ∇) = (∇ ⊗ A )( A ⊗ c)(c ⊗ A ), (4)
c(1 ⊗ a) = a ⊗ 1, c(a ⊗ 1) = 1 ⊗ a for all a ∈ A . (5)
A morphism of braided algebras is, by definition, a morphism of ordinary algebras which, in addition,
is a morphism of braided vector spaces.
Remark 1.3. 1) Let ( A , ∇, u , c) is a braided algebra. Then A ⊗ A is an associative algebra with mul-
tiplication ∇ A ⊗ A := (∇ ⊗ ∇)( A ⊗ c ⊗ A ) and unit 1 ⊗ 1. Moreover, A ⊗ A is a c A ⊗ A -algebra, where
c A ⊗ A = ( A ⊗ c ⊗ A )(c ⊗ c)( A ⊗ c ⊗ A ). This algebra structure on A ⊗ A will be denoted by A ⊗c A.
2) If A is an object in a braided monoidal category M and c := c A , A then the above four compati-
bility relations hold automatically, as the braiding c is a natural morphism.
Definition 1.4. A braided coalgebra (or c-coalgebra) is a quadruple (C , , ε , c) where (C , c) is a braided

vector space and (C , , ε ) is a coassociative counital coalgebra such that the comultiplication and
the counit ε commute with c, that is the following relations hold:
(C ⊗ )c = (c ⊗ C )(C ⊗ c)( ⊗ C ), (6)
( ⊗ C )c = (C ⊗ c)(c ⊗ C )(C ⊗ ), (7)
(ε ⊗ C )c(c ⊗ d) = ε (d)c = (C ⊗ ε )c(d ⊗ c ) for all c , d ∈ C . (8)
A morphism of braided coalgebras is, by definition, a morphism of ordinary coalgebras which, in

addition, is a morphism of braided vector spaces.
1.5. Recall that a coalgebra C is called connected if the coradical C 0 of C is one-dimensional. In this
case there is a unique group-like element g ∈ C such that C 0 = K g. Sometimes, we will write (C , g ),
to emphasize the group-like element g. We also ask that f ( g C ) = g D , for any morphism f : (C , g C ) →
( D , g D ) of connected coalgebras.
By definition, a c-coalgebra C is connected if C 0 = K g and, for any x ∈ C ,
c(x ⊗ g ) = g ⊗ x, c( g ⊗ x) = x ⊗ g . (9)
Definition 1.6. (See Takeuchi [Ta].) A braided bialgebra is a sextuple ( A , ∇, 1, , ε , c) where
• ( A , ∇, 1, c) is a braided algebra,
• ( A , , ε , c) is a braided coalgebra,
• and ε are morphisms of algebras (on the vector space A ⊗ A we take the algebra structure
A ⊗c A).
Remark 1.7. Note that : A → A ⊗c A is multiplicative if and only if
∇ = (∇ ⊗ ∇)( A ⊗ c ⊗ A )( ⊗ ). (10)
1.8. We will need graded versions of braided algebras, coalgebras and bialgebras. By definition,
n+m
a braided algebra ( A , ∇, 1, c) is graded if A = n∈N A and ∇( A ⊗ A ) ⊆ A
n n m
. The braiding c
is assumed to satisfy c( A n ⊗ A m ) ⊆ A m ⊗ A n . In this case, it is easy to see that 1 ∈ A 0 .
Therefore a graded braided algebra can be defined by means of maps ∇ n,m : A n ⊗ A m → A n+m and
c : An ⊗ Am → Am ⊗ An , and an element 1 ∈ A 0 such that:
n,m

∇ n+m, p ∇ n,m ⊗ A p = ∇ n,m+ p An ⊗ ∇ m, p for all n, m, p ∈ N, (11)
0,n n ,0 n
∇ (1 ⊗ a) = a = ∇ (a ⊗ 1) for all a ∈ A , n ∈ N, (12)

cn+m, p ∇ n,m ⊗ A p = A p ⊗ ∇ n,m cn, p ⊗ Am An ⊗ cm, p for all n, m, p ∈ N, (13)

cn,m+ p An ⊗ ∇ m, p = ∇ m, p ⊗ An Am ⊗ cn, p cn,m ⊗ A p for all n, m, p ∈ N, (14)
c0,n (1 ⊗ a) = a ⊗ 1 and cn,0 (a ⊗ 1) = 1 ⊗ a for all a ∈ An , n ∈ N. (15)
The multiplication ∇ can be recovered from (∇ n,m )n,m∈N as the unique K -linear map such that
∇(x ⊗ y ) = ∇ p ,q (x ⊗ y ) for all p , q ∈ N, x ∈ A p , y ∈ A q . Analogously, the braiding c is uniquely defined
by c(x ⊗ y ) = c p ,q (x ⊗ y ) for all p , q ∈ N, x ∈ A p , y ∈ A q . We will say that ∇ n,m and cn,m are the
(n, m)-homogeneous components of ∇ and c, respectively.
Graded braided coalgebras can be describedin a similar way. By definition a braided coalgebra
(C , , ε , c) is graded if C = n∈N C n , (C n ) ⊆ p +q=n C p ⊗ C q , c(C n ⊗ C m ) ⊆ C m ⊗ C n and ε |C n = 0,
for n > 0 . If π p denotes the projection onto C p then the comultiplication is uniquely defined
by the maps p ,q : C p +q → C p ⊗ C q , where p ,q := (π p ⊗ π q )|C p+q . The counit is given by a map
ε0 : C 0 → K , while the braiding c is uniquely determined by a family (cn,m )n,m∈N , as for braided
algebras. The families (n,m )n,m∈N , (cn,m )n,m∈N and ε 0 have to satisfy the relations that are dual
to (11)–(15), namely:

n,m ⊗ C p n+m, p = C n ⊗ m, p n,m+ p for all n, m, p ∈ N, (16)
0
ε ⊗ C n 0,n (c ) = c = C n ⊗ ε0 n,0 (c ) for all c ∈ C n , n ∈ N, (17)
p
C ⊗ n,m cn+m, p = cn, p ⊗ C m C n ⊗ cm, p n,m ⊗ C p for all n, m, p ∈ N, (18)
m, p
n n,m+ p
m n, p
n,m p
n m, p

⊗C c = C ⊗c c ⊗C C ⊗ for all n, m, p ∈ N, (19)
0
ε ⊗ C c(c ⊗ d) = ε0 (d)c = C ⊗ ε0 c(d ⊗ c ) for all c ∈ C n , d ∈ C 0 . (20)
We will say that n,m is the (n, m)-homogeneous component of .

A graded braided bialgebra is a braided bialgebra which is graded both as an algebra and as a
coalgebra.

Remark 1.9. Let C = n∈N C n be a graded braided coalgebra. By [Sw, Proposition 11.1.1], if (C n )n∈N is

the coradical filtration, then C n ⊆ mn C m . Therefore, if C 0 is one-dimensional then C is connected.
Definition 1.10. A graded braided coalgebra will be called 0-connected if its homogeneous component
of degree 0 is of dimension one.
Lemma 1.11. Let (C , c) be a connected braided coalgebra. Then c induces a canonical braiding cgr C on gr C
such that (gr C , cgr C ) is a 0-connected graded braided coalgebra, where gr C is constructed with respect to the
coradical filtration on C .
Proof. Let (C n )n∈N be the coradical filtration. Since C is connected, we have C 0 = K g, where g is the
unique group-like element of C . We claim that c(C n ⊗ C m ) ⊆ C m ⊗ C n . For n = 0 this relation holds
true as, by definition, c(x ⊗ g ) = g ⊗ x for all x ∈ C . We choose a basis { y i | i ∈ I } on C m and we
assume that the above inclusion is true for n. Let x ∈ C n+1 and y ∈ C m . Since C n+1 = C n+1 + ⊕ K g ,
by [Sw, Proposition 10.0.2], (x) = x ⊗ g + g ⊗ x + k=1 xk ⊗ xk , where xk , xk ∈ C n . Moreover,
q

c(x ⊗ y ) = y i ⊗ xi ,
i∈ I
with xi ∈ C , and the set {i ∈ I | xi = 0} finite. Since (C ⊗ )c = (c ⊗ C )(C ⊗ c)( ⊗ C ) we get

p

p

p

q

y i ⊗ (xi ) = y i ⊗ xi ⊗ g + y i ⊗ g ⊗ xi + (c ⊗ C )(C ⊗ c) xk ⊗ xk ⊗ y .
i =1 i =1 i =1 k=1

q
hypothesis, (c ⊗ C )(C ⊗ c)( k=1 xk ⊗ xk ⊗ y ) ∈ C m ⊗ C n ⊗ C n , so this element can be
By induction
written as i ∈ I y i ⊗ z i , with z i ∈ C n ⊗ C n and the set {i ∈ I | z i = 0} is finite. Hence, for all i ∈ I , we
have
(xi ) = xi ⊗ g + g ⊗ xi + zi .
Thus xi ∈ C n+1 , so c(x ⊗ y ) ∈ C m ⊗ C n+1 . Hence, by induction, c(C n ⊗ C m ) ⊆ C m ⊗ C n , so
c(C n−1 ⊗ C m + C n ⊗ C m−1 ) ⊆ C m ⊗ C n−1 + C m−1 ⊗ C n .

n,m
Therefore c induces a unique K -linear map cgr C : grn C ⊗ grm C → grm C ⊗ grn C . We define cgr C :=
n,m
n,m cgr C . Now it is easy to see that (gr C , cgr C ) is a graded braided coalgebra. 2
Remark 1.12. If (C , c, g ) is a connected braided coalgebra then c induces a canonical braiding c P

on the space P (C ) = {c ∈ C | (c ) = c ⊗ g + g ⊗ c }, of primitives elements in C . Indeed, by [Sw,
Proposition 10.0.1] we have P (C ) = (Ker ε ) ∩ C 1 . Thus c maps P (C ) ⊗ P (C ) to itself, see the proof of
the preceding lemma.
Lemma 1.13. Let ( A , ∇, u , c) be a c-algebra and let : A → A ⊗c A be a morphism of algebras. We fix x, a ∈ A

such that (a) = a ⊗ 1 + 1 ⊗ a. Then

(xa) = (∇ ⊗ A )( A ⊗ c) (x) ⊗ a + ( A ⊗ ∇) (x) ⊗ a . (21)
p
Proof. Let (x) = x ⊗ xi . Then
i =1 i

p

(xa) = (x) · (a ⊗ 1 + 1 ⊗ a) = xi ⊗ xi · (a ⊗ 1 + 1 ⊗ a)
i =1
p
p
= xi c xi ⊗ a + xi ⊗ xi a
i =1 i =1

= (∇ ⊗ A )( A ⊗ c) (x) ⊗ a + ( A ⊗ ∇) (x) ⊗ a . 2
Proposition 1.14. To any braided vector space ( V , c) we can associate a 0-connected graded braided bialgebra
( T = T ( V , c), ∇T , 1T , T , εT , c T ) where
• ( T = T ( V , c), ∇T , 1T ) is the tensor algebra T ( V ) i.e. the free algebra generated by V .

• c T is constructed iteratively from c.
• T : T → T ⊗cT T is the unique algebra homomorphism defined by setting T ( v ) = 1T ⊗ v + v ⊗ 1T
for every v ∈ V .
• εT : T → K is the unique algebra homomorphism defined by setting ε T ( v ) = 0 for every v ∈ V .
Remark 1.15. Note that T is the dual construction of the quantum shuffle product, introduced by
n,m
Rosso in [Ro]. The K -linear map c T is given in the figure below:
where each crossing represents a copy of c.

Remark 1.16. If A = n∈N A n is a 0-connected graded c-bialgebra then n,0 = 0,n = Id An . Indeed,
the proof given in the case A = T ( V ) works for an arbitrary connected graded c-bialgebra.
Theorem 1.17. Let ( V , c) be a braided vector space. Then i V : V → T ( V ) is a morphism of braided vector
spaces. If ( A , ∇ A , 1 A , c A ) is a braided algebra and f : V → A is a morphism of braided vector spaces then
there is a unique morphism f : T ( V , c) → A of braided algebras that lifts f . If, in addition, A is a braided
bialgebra and f ( V ) is contained in P ( A ), the set of primitive elements of A , then f is a morphism of braided
bialgebras.
n
1.18. Recall [Ka, p. 74] that the X -binomial coefficients k X
are defined as follows. We set (0) X =
(0)! X = 1 and, for n > 0, define (n) X := 1 + X + · · · + X n−1 and (n)! X = (1) X (2) X · · · (n) X . Then:

n (n)! X
= . (22)
k X (k)! X (n − k)! X
n
It is well known that is a polynomial. Therefore we may specialize X at an arbitrary element
k X
λ ∈ K . In this way we get an element nk λ ∈ K . Note that, if λ is a root of unity, then nk λ may be
n
zero. If char( K ) = 0 and λ = 1, the formula shows that k 1 is the classical binomial coefficient.
Definition 1.19. We say that a braided vector space ( V , c) is of Hecke-type (or that c is of Hecke-type) of
mark m(c) = λ if
(c + Id V ⊗2 )(c − λ Id V ⊗2 ) = 0.
Remark 1.20. For every Hecke-type braiding c of mark λ, the operator č := −λ−1 c is also of Hecke-
type. We have m(č) = λ−1 . Note that, for d = č, one has ď = c.
2. Braided enveloping algebras
In this section we introduce the main example of c-bialgebras that we will deal with, namely the
enveloping algebra of a c-Lie algebra.
Given a vector spaces V , W and a K -linear map α : V ⊗ V → W , we will denote α ⊗ V and V ⊗ α
by α1 and α2 , respectively.
Definition 2.1. Let ( V , c) be a braided vector space. We say that a K -linear map b : V ⊗ V → V is a
c-bracket, or braided bracket, if the following compatibility conditions hold true:
cb1 = b2 c1 c2 and cb2 = b1 c2 c1 . (23)
Let b be a c-bracket on ( V , c) and let b be a c -bracket on ( V , c ). We will say that a morphism of

braided vector spaces from V to V is a morphism of braided brackets if f b = b ( f ⊗ f ).
Let b be a bracket on a braided vector space ( V , c) of Hecke-type with mark λ. Let I c,b is the
two-sided ideal generated by the set

X c,b = c( z) − λ z − b( z) z ∈ V ⊗ V = T 2 ( V ) .
The enveloping algebra of ( V , c, b) is by definition the algebra
T ( V , c)
U = U ( V , c, b) := .
I c,b
We will denote by πc,b : T ( V , c) → U ( V , c, b) the canonical projection.

When b = 0, the enveloping algebra of ( V , c, 0) is called the c-symmetric algebra, or braided sym-
metric algebra, if there is no danger of confusion. It will be denoted by S ( V , c).
Proposition 2.2. Let b be a bracket on a braided vector space ( V , c) of Hecke-type with mark λ. Then I c,b is a
coideal in T ( V , c). Moreover, on the quotient algebra (and coalgebra) there is a natural braiding cU such that
(U ( V , c, b), cU ) is a braided bialgebra.
Proof. We denote T ( V , c) and T n ( V , c) by T and T n , respectively. We first prove that c T maps I c,b ⊗ T
into T ⊗ I c,b and T ⊗ I c,b into I c,b ⊗ T . Let x ∈ T n , y ∈ T m , t ∈ T p and z ∈ T 2 . Since c verifies the
braid equation and (23) we get

cn+m+2, p T n ⊗ c ⊗ T m ⊗ T p = T p ⊗ T n ⊗ c ⊗ T m cn+m+2, p ,

cn+m+1, p T n ⊗ b ⊗ T m ⊗ T p = T p ⊗ T n ⊗ b ⊗ T m cn+m+2, p .
Then

cT T n ⊗ (c − λ Id T 2 − b) ⊗ T m ⊗ T p = T p ⊗ T n ⊗ (c − λ Id T 2 − b) ⊗ T m c T ,
relation that shows us that c T maps I c,b ⊗ T into T ⊗ I c,b . The other property can be proved similarly.
We claim that X c,b is a coideal in T . In fact we will prove that X c,b contains only primitive
elements in T . Let z ∈ T 2 . By Proposition 1.14 we have
T (z) = z ⊗ 1 + z + c(z) + 1 ⊗ z,

T c(z) = c( z) ⊗ 1 + c(z) + c2 (z) + 1 ⊗ c(z).
Thus T (c( z) − λ z) = (c( z) − λ z) ⊗ 1 + 1 ⊗ (c( z) − λ z), as c is a Hecke braiding of mark λ. It follows

that c( z) − λ z ∈ P ( T ). Since b( z) ∈ V we deduce that c( z) − λ z − b( z) ∈ P ( T ), so X c,b ⊆ P ( T ).
Now, by (21), it easily follows that I c,b , the ideal generated by X c,b , is a coideal. It remains to
show that c factors through a braiding cU of U := U ( V , c, b), that makes U a braided bialgebra. Let
πc,b : T → U be the canonical projection. By the foregoing, cT maps the kernel of πc,b ⊗ πc,b into
itself, so there is a K -linear morphism cU : U ⊗ U → U ⊗ U such that
cU (πc,b ⊗ πc,b ) = (πc,b ⊗ πc,b )c T .
Since T is a c T -bialgebra, this relation entails that U is a cU -bialgebra and that the canonical projec-
tion πc,b becomes a morphism of braided bialgebras. 2
Letn K := Ker(Sn ) where Sn denotes the

n
Remark 2.3. Let ( V , c) be an arbitrary braided vector space.
quantum symmetrizer [AS, 2.3]. It is well known that nN K is an ideal and a coideal in T ( V , c),
see [AS, Section 3]. Since

cnT,m T n ⊗ K m + K n ⊗ T m ⊆ K m ⊗ T n + T m ⊗ K n

it follows that B ( V , c) = T ( V , c)/( n∈N K n ) is a quotient graded braided bialgebra of T ( V , c), that is
called the Nichols algebra of ( V , c). Let us now assume that c is a braiding of Hecke-type of mark λ.
By the definition of Hecke operators we have

Im(c − λ Id T 2 ) ⊆ Ker(Id T 2 + c) = K 2 ⊆ K n.
n∈N
Therefore, there is a morphism of braided bialgebras ϕ : S ( V , c) → B ( V , c) such that ϕ | V = Id V . Ob-

viously ϕ is surjective, since B ( V , c) is generated by V . Later (see Theorem 2.17) we will see that the
space of primitive elements in S ( V , c) and the homogeneous component S 1 ( V , c) = V are identical
whenever λ is “regular.” In this case, by [Mo, Theorem 5.3.1], it follows that ϕ is injective too. Thus,
S ( V , c) and B ( V , c) are isomorphic braided bialgebras.
We are going to investigate some basic properties of these objects.
Proposition 2.4. Let ( A , ∇, 1, , ε , c A ) be a connected braided bialgebra. Let P be the space of primitive
elements of A . Assume that there is λ ∈ K ∗ such that c P := c A | P ⊗ P is a braiding of Hecke-type on P of mark λ.
Then:
(a) ∇(c P − λ Id P ⊗2 )( P ⊗ P ) ⊆ P , so we can define b P : P ⊗ P → P by b P = ∇(c P − λ Id P ⊗2 )| P ⊗ P .

(b) The map b P is a braided bracket on the braided vector space ( P , c P ).
(c) Let f : ( V , c, b) → ( P , c P , b P ) be a morphism of braided brackets and assume that c is a braiding of
Hecke-type with mark λ. Then there is a unique morphism of braided bialgebras f : U ( V , c, b) → A that
lifts f .
Proof. First, observe that, by Remark 1.12, c A ( P ⊗ P ) ⊆ P ⊗ P so that it makes sense to consider
cP : P ⊗ P → P ⊗ P .
(a) By assumption, c P is a Hecke operator on P of mark λ. For z ∈ P ⊗ P , by (21) we get
∇(z) = ∇( z) ⊗ 1 + z + c P (z) + 1 ⊗ ∇( z),
∇c P (z) = ∇c P (z) ⊗ 1 + c P (z) + c2P (z) + 1 ⊗ ∇c P (z).
Therefore ∇(c P − λ Id P ⊗2 )( z) ∈ P . This shows that ∇(c P − λ Id P ⊗2 )( P ⊗ P ) ⊆ P .

(b) We have to prove the compatibility relation between c and b, that is we have (23). But these
relations follows immediately by the braid relation and the fact that A is a c A -algebra.
(c) Apply the universal property of T ( V , c) (see Theorem 1.17) to get a morphism f : T ( V , c) → A
of braided bialgebras that lifts f . Since f is a morphism of braided brackets, by the definition of b P , it
results that f maps c( z) − λ z − b( z) to 0. Therefore f factors through a morphism
f : U ( V , c, b) → A,
which lifts f and is compatible with the braidings (note that U ( V , c, b) is a braided bialgebra in view
of Proposition 2.2). 2
Proposition 2.5. Let ( A , ∇, 1, , ε , c A ) be a connected braided bialgebra. Let P denote the primitive part
of A. Assume that there is λ ∈ K ∗ such that ∇c A = λ∇ on P ⊗ P . If c P = c A | P ⊗ P then c P is of Hecke-type
on P of mark λ. Moreover, if ( V , c) is a braided vector space such that c is a Hecke operator of mark λ and
f : ( V , c) → ( P , c P ) is a morphism of braided vector spaces then there is a unique morphism of braided bial-
gebras f : S ( V , c) → A that lifts f . If A is graded then
f respects the gradings on S ( V , c) and A .
Proof. For z ∈ P ⊗ P , by (21) we get
∇(z) = ∇( z) ⊗ 1 + z + c A (z) + 1 ⊗ ∇( z),
∇c A (z) = ∇c A (z) ⊗ 1 + c A (z) + c2A (z) + 1 ⊗ ∇c A (z).
By assumption, ∇c A ( z) = λ∇( z) whence
0 = c A ( z) + c2A ( z) − λ z − λc A ( z) = (c A + Id A ⊗2 )(c A − λ Id A ⊗2 )( z).
Thus c P is a Hecke operator of mark λ. By taking b = 0 in Proposition 2.4, it results that there is
f
that lifts f . 2
Remark 2.6. The above proposition still works under the slighter assumption ∇c A = λ∇ on
Im f ⊗ Im f .

2.7. Let T n := T n ( V ) and let T n := 0mn T m . By construction πc,b is a morphism of algebras
and coalgebras from T ( V , c) to U ( V , c, b). Thus U n
:= πc,b ( T ) defines a braided bialgebra filtration
n
on U ( V , c, b), i.e. (U n )n∈N is an algebra and coalgebra filtration on U ( V , c, b) which is compatible

with cU . It will be called the standard filtration on U ( V , c, b). In general, this filtration and the corad-
ical filtration (U n )n∈N are not identical, but we always have U n ⊆ U n , for any n ∈ N.
If b = 0 then S ( V , c) := U ( V , c, 0) is a graded c S -bialgebra, S ( V , c) = n∈N S n ( V , c). The standard
filtration on S ( V , c) is the filtration associated to this grading.
Proposition 2.8. Let b be a c-bracket on a braided vector space ( V , c) of Hecke-type. Then U ( V , c, b) is a

connected coalgebra. Moreover, for every braided vector space ( V , c), S ( V , c) is a 0-connected graded braided
coalgebra.
Proof. We know that the tensor algebra of an arbitrary braided vector space is a 0-connected coalge-
bra. By definition, U ( V , c, b) is a quotient coalgebra of T ( V , c), where λ = m(c). Then, in view of [Mo,
Corollary 5.3.5], U ( V , c, b) is connected. In particular, braided symmetric algebras are connected coal-
gebras. They are also 0-connected since they are graded quotients of T ( V , c). 2
Remark 2.9. Let b be a c-bracket on a braided vector space ( V , c) of Hecke-type. The composition of
the inclusion V → T ( V , c) with the canonical projection πc,b gives a map
ιc,b : V → U ( V , c, b).
Its image is included in the space of primitive elements of U ( V , c, b). In general ιc,b is neither injec-
tive nor onto. Our purpose now is to investigate when ιc,b is injective (see Theorem 4.3).
Definition 2.10. Let ( B , ∇ B , 1 B , B , ε B , c B ) be a graded braided bialgebra. For every a, b, n ∈ N, set
ΓaB,b := ∇ Ba,b aB,b .
Lemma 2.11. Let ( B , ∇ B , 1 B , B , ε B , c B ) be a 0-connected graded braided bialgebra. Then
nB,0 (z) = z ⊗ 1 B 0,n

and B = 1 B ⊗ z for every z ∈ B n .
Moreover

nB,1 ∇ Bn,1 = Id B n ⊗ B 1 + ∇ Bn−1,1 ⊗ B 1 B n−1 ⊗ c1B,1 nB−1,1 ⊗ B 1 . (24)
Proof. The first assertion follows by Remark 1.16. Since B is a graded bialgebra we have

nB,1 ∇ Bn,1 = ∇ Bn,0 ⊗ ∇ B0,1 B n ⊗ c0B,0 ⊗ B 1 nB,0 ⊗ 0B,1

+ ∇ Bn−1,1 ⊗ ∇ B1,0 B n−1 ⊗ c1B,1 ⊗ B 0 nB−1,1 ⊗ 1B,0

= Id B n ⊗ B 1 + ∇ Bn−1,1 ⊗ B 1 B n−1 ⊗ c1B,1 nB−1,1 ⊗ B 1 ,
so that (24) holds. 2
Definition 2.12. Let ( A , ∇, 1 A ) be a graded algebra. We say that A is a strongly N-graded algebra
whenever ∇ i , j : A i ⊗ A j → A i + j is an epimorphism for every i , j ∈ N (equivalently ∇ n,1 : A n ⊗ A 1 →
A n+1 is an epimorphism for every n ∈ N).
Dually, let (C , , ε ) be a graded coalgebra. We say that C is a strongly N-graded coalgebra whenever
i , j : C i + j → C i ⊗ C j is a monomorphism for every i , j ∈ N (equivalently n,1 : C n+1 → C n ⊗ C 1 is a
monomorphism for every n ∈ N).
For more details on these (co)algebras see e.g. [AM1].
Definition 2.13. An element λ ∈ K ∗ is called n-regular whenever (k)λ = 0, for any 1 k n. If λ is

n-regular for any n > 0, we will simply say that λ is regular.
Remarks 2.14. 1) Note that λ = 1 is regular if and only if λ is not a root of one, while 1 is regular if
and only if char( K ) = 0.
2) If λ is n-regular (respectively regular) then λ−1 is also n-regular (respectively regular).
Theorem 2.15. Let ( B , ∇ B , 1 B , B , ε B , c B ) be a 0-connected graded braided bialgebra and let λ ∈ K ∗ be reg-
ular. The following are equivalent.
1, 1
(1) B is a bialgebra of type one and c B is a braiding of Hecke-type of mark λ.
1, 1 1, 1
(2) B is strongly N-graded as a coalgebra, ∇ B is surjective and c B is a braiding of Hecke-type of mark λ.
(3) B is strongly N-graded as a coalgebra and (c1B,1
− λ Id B 2 )1B,1 = 0.
1, 1 1, 1
(4) B is strongly N-graded as an algebra, B is injective and c B is a braiding of Hecke-type of mark λ.
1, 1 1, 1
(5) B is strongly N-graded as an algebra and ∇ B (c B − λ Id B 2 ) = 0.
Proof. (1) ⇒ (2). By definition, B is strongly N-graded both as a coalgebra and as an algebra.
(2) ⇒ (3). We have
1, 1 (24) 1,1
0 = c B − λ Id B 2 c1B,1 + Id B 2 = c B − λ Id B 2 1B,1 ∇ B1,1 .
1, 1 1, 1 1, 1
Since ∇ B is an epimorphism, we get (c B − λ Id B 2 ) B = 0.
(3) ⇒ (1). We have

nB,1 ΓnB,1 = nB,1 ∇ Bn,1 nB,1 = nB,1 + ∇ Bn−1,1 ⊗ B 1 B n−1 ⊗ c1B,1 nB−1,1 ⊗ B 1 nB,1
(24)

= nB,1 + ∇ Bn−1,1 ⊗ B 1 B n−1 ⊗ c1B,1 1B,1 nB−1,2

= nB,1 + ∇ Bn−1,1 ⊗ B 1 B n−1 ⊗ λ1B,1 nB−1,2

= nB,1 + λ ∇ Bn−1,1 nB−1,1 ⊗ B 1 nB,1 = nB,1 + λ ΓnB−1,1 ⊗ B 1 nB,1 ,
so that

nB,1 ΓnB,1 = nB,1 + λ ΓnB−1,1 ⊗ B 1 nB,1 . (25)
Let us prove by induction that
ΓnB,1 = (n + 1)λ Id B n+1 for every n 1. (26)
(n = 1) We have
(25)
1B,1 Γ1B,1 = 1B,1 + λ1B,1 = (2)λ 1B,1 .
1, 1
Since, by hypothesis, B is injective, we obtain Γ1B,1 = (2)q Id B 2 .
(n − 1 ⇒ n). We have
(25)
nB,1 ΓnB,1 = nB,1 + λ ΓnB−1,1 ⊗ B 1 nB,1 = nB,1 + λ(n)λ nB,1 = (n + 1)λ nB,1 .
n ,1
Since, by hypothesis, B is injective, we obtain ΓnB,1 = (n + 1)λ Id B n+1 .
We have so proved (26).
Since λ is regular, we have (n + 1)λ = 0 so that
(26)
∇ Bn,1 nB,1 = ΓnB,1 = (n + 1)λ Id B n+1
n ,1
is bijective for every n 1. Therefore ∇ B is surjective for every n 1. Equivalently B is strongly
N-graded as an algebra and hence of type one. Moreover
(24)
c1B,1 − λ Id B 2 c1B,1 + Id B 2 = c1B,1 − λ Id B 2 1B,1 ∇ B1,1 = 0.
(1) ⇔ (4) ⇔ (5). It follows by dual arguments. 2

Definition 2.16. A graded coalgebra C = n∈N C n is called strict if it is 0-connected and P (C ) = C 1 .
Theorem 2.17. (Cf. [AS, Proposition 3.4].) Let ( V , c) be a braided vector space of Hecke-type with regular
mark λ. Then S ( V , c) is a bialgebra of type one. In particular S ( V , c) is a strict coalgebra.
Proof. By definition, we have
T ( V , c)
S := S ( V , c) = U ( V , c, 0) = .
(c(z) − λ z | z ∈ V ⊗ V )
Thus, since S is a graded quotient of the graded braided bialgebra T ( V , c), we get that S is strongly
N-graded as an algebra. Moreover ∇ S1,1 (c1S,1 − λ Id S 2 ) = 0. By Theorem 2.15, we conclude. 2
3. Categorical subspaces
Definition 3.1. (See [Kh, 2.2].) A subspace L of a braided vector space ( V , c) is said to be categorical
if
c( L ⊗ V ) ⊆ V ⊗ L and c( V ⊗ L ) ⊆ L ⊗ V . (27)
Theorem 3.2. Let ( V , c) be a braided vector space of Hecke-type of mark λ. Assume λ = 0, 1. Let L be a
categorical subspace of V . Then L = 0 or L = V .
Proof. From (c + Id V ⊗ V )(c − λ Id V ⊗ V ) = 0 and λ = 1, we get c = (λ − 1)−1 (c2 − λ Id V ⊗ V ) so that
1 (27)
c( L ⊗ V ) = c2 − λ Id V ⊗ V ( L ⊗ V ) ⊆ L ⊗ V .
λ−1
Then
c( L ⊗ V ) ⊆ ( V ⊗ L ) ∩ ( L ⊗ V ) = L ⊗ L . (28)
From (c + Id V ⊗ V )(c − λ Id V ⊗ V ) = 0 and λ = 0, we get Id V ⊗ V = λ−1 (c2 + (1 − λ)c) so that
1 (28), (27)
L⊗V = c2 + (1 − λ)c ( L ⊗ V ) ⊆ L ⊗ L.
λ
Since L ⊆ V , we deduce that L = 0 or L = V . 2
Proposition 3.3. Let b be a c-bracket on a braided vector space ( V , c) of Hecke-type of mark λ. Assume
λ = 0, 1. Then b is zero or surjective.
Proof. Let L = Im(b) ⊆ V . We have that
cb1 = b2 c1 c2 ⇒ c( L ⊗ V ) ⊆ V ⊗ L ,
cb2 = b1 c2 c1 ⇒ c( V ⊗ L ) ⊆ L ⊗ V .
Thus L is a categorical subspace of V . By Theorem 3.2, we get that L = 0 or L = V . 2
Proposition 3.4. Let V be an object in the monoidal category H H YD of Yetter–Drinfeld modules over some
Hopf algebra H . Assume that c V , V is a braiding of Hecke type of mark λ and that λ = 0, 1. Then V is simple
H YD .
in H
H YD 2
H
Proof. Any subspace of V in is categorical. We conclude by Theorem 3.2.
4. Trivial braided brackets
Let b be a braided bracket on ( V , c). Our aim now is to answer the following natural question:
when is ιc,b : V → U ( V , c, b) injective?
Proposition 4.1. Let b be a c-bracket on a braided vector space ( V , c) of Hecke type. If gr U ( V , c, b) is the
graded associated to the standard filtration on U ( V , c, b), then gr U ( V , c, b) is a graded braided bialgebra
and there is a canonical morphism θ : S ( V , c) → gr U ( V , c, b) of graded braided bialgebras. Moreover θ is
surjective.

Proof. Let T n := 0mn T
m
and let (U n )n∈N be the standard filtration on U := U ( V , c, b). Let
∇U and cU be the multiplication and the braiding of U := U ( V , c, b), respectively. If T := T ( V , c)
and c T is the braiding of T then the canonical projection πU : T → U is a morphism of braided
bialgebras. Since c T ( T n ⊗ T m ) ⊆ T m ⊗ T n we deduce that cU (U n ⊗ U m ) ⊆ U ⊗ U , for any
m n
n, m ∈ N. Hence cU induces a braiding cgr U : gr U ⊗ gr U → gr U ⊗ gr U . The standard filtration is a
coalgebra filtration, as πU is a morphism of coalgebras, so gr U is a coalgebra. One can prove easily
that, with respect to this coalgebra structure, gr U becomes a graded braided bialgebra.
Let λ = m(c). We define θ 1 : V → U 1 /U 0 by θ 1 = p ιc,b , where p : U 1 → U 1 /U 0 is the canon-
ical projection. The image of θ 1 is included in the component of degree 1 of gr U , so Im θ 1 ⊆
P (gr U ). Clearly θ 1 is a map of braided vector spaces. One can check that ∇gr U cgr U = λ∇gr U on
U 1 /U 0 ⊗ U 1 /U 0 . By Proposition 2.5 (see also Remark 2.6) there is a unique morphism of graded
braided bialgebras θ : S ( V , c) → gr U that lifts θ 1 . On the other hand, gr U ( V , c, b) is generated as an
algebra by U 1 /U 0 . Since U 1 /U 0 is included into the image of θ, we conclude that θ is surjective. 2
Remark 4.2. The second statement in the following theorem is well known, see [Gu1]. Nevertheless
we include it for sake of completeness.
Theorem 4.3. Let K be a field with char K = 2. Let b be a c-bracket on a braided vector space ( V , c) of Hecke-
type of mark λ = 0 such that (3)!λ = 0.
Assume that the K -linear map ιc,b : V → U ( V , c, b) is injective.
• If λ = 1, then b = 0.
• If λ = 1, then b fulfills
bc = −b and bb1 (Id V ⊗3 − c2 + c2 c1 ) = 0 (29)
i.e. ( V , c, b) is a generalized Lie algebra in the sense of [Gu1].

Proof. Denote by T n the nth graded component of T = T ( V , c ) and set T n := 0mn T m and

T n := mn T . m
Let γ := c − λ Id T 2 − b, let F := Im(γ ) and R := Im(λ Id V ⊗2 − c). We denote the component of

degree 1 of the map θ of Proposition 4.1 by θ 1 : V → U 1 /U 0 .
Let x ∈ Ker θ 1 . It follows that ιc,b (x) ∈ U 0 , so there is a ∈ K such that x − a1 ∈ ( F ). Since ( F ) ⊆ T 1
we get a = 0. It results that ιc,b (x) = 0, so x = 0. In conclusion, θ 1 is injective. But, in view of Theo-
rem 2.17, S ( V , c) is a strict coalgebra so that P ( S ( V , c)) = S 1 ( V , c) = V . Thus, since θ 1 , the restriction
of θ to V , is injective, by [Mo, Lemma 5.3.3] it follows that θ is injective too. Since θ is always sur-
jective, we conclude that θ is an isomorphism of braided bialgebras.
The algebras S ( V , c) and U ( V , c, b) are the quotients of T ( V , c) through the two-sided ideals gen-
erated by R and F , respectively. Set
(2)
ζ = (λ Id V ⊗3 − c1 ) λ2 Id V ⊗3 − λc2 + c2 c1 = (λ Id V ⊗3 − c2 ) λ2 Id V ⊗3 − λc1 + c1 c2 .
Since (2)λ = 0, one can easily see that R = {x ∈ T 2 | c(x) = −x} (note that if x is in this set, then
(λ Id V ⊗2 − c )(x) = (2)λ x) and since (3)!λ = 0 one gets that

( R ⊗ V ) ∩ ( V ⊗ R ) = x ⊗ T 3 ( V ) c1 (x) = c2 (x) = −x = Im ζ. (30)
Since the canonical map θ : S ( V , b) → gr U ( V , c, b) is an isomorphism, by [BG, Lemma 0.4], it follows

that the following conditions are satisfied:
F ∩ T 1 = 0 (31)

T 1 · F · T 1
∩ T 2 = F . (32)
We claim (31) implies
bc = −b. (33)
In fact we have
γ (c + IdT 2 ) = (c − λ IdT 2 − b)(c + IdT 2 ) = −b(c + IdT 2 ) (34)
so that Im[b(c + Id T 2 )] = Im[γ (c + Id T 2 )]. We will prove that Im[γ (c + Id T 2 )] = F ∩ T 1 , from which
the conclusion will follow.
(⊆) It follows by (34).
(⊇) Let y ∈ F ∩ T 1 . Then there is x such that y = γ (x) = c(x) − λx − b(x). Since y , b(x) ∈ T 1
and c(x), x ∈ V ⊗ V , it results c(x) = λx and y = −b(x). Thus γ (c + Id T 2 )(x) = (2)λ γ (x) = (2)λ y . Since
(2)λ = 0, we get y ∈ Im[γ (c + Id T 2 )].
Now, we define α := −(2)− 1
λ b| R . From (33), we deduce that
1
(Id T 2 − α )(c − λ Id T 2 ) = c − λ Id T 2 + b(c − λ Id T 2 ) = γ
(2)λ
so that F = Im γ = Im[(Id T 2 − α )(c − λ Id T 2 )] = {x − α (x) | x ∈ R }. By [BG, Lemma 3.3] (32) implies

that α satisfies the following two conditions:
(α ⊗ V )(x) − ( V ⊗ α )(x) ∈ R for all x ∈ ( R ⊗ V ) ∩ ( V ⊗ R ),
α (α ⊗ V − V ⊗ α )(x) = 0 for all x ∈ ( R ⊗ V ) ∩ ( V ⊗ R ).
The second property is equivalent to the fact that b(b1 − b2 ) = 0 on Im ζ, which at its turn is equiva-
lent to
bb1 ζ = bb2 ζ. (35)
Let us prove that b satisfies the following conditions:

bb1 λ2 Id V ⊗3 − λc2 + c2 c1 = 0, (36)

bb2 λ2 Id V ⊗3 − λc1 + c1 c2 = 0. (37)
We have
(30) (23) (33)

bb2 ζ = bb2 c1 c2 ζ = bcb1 ζ = −bb1 ζ.
In view of (35), we obtain bb2 ζ = 0 = bb1 ζ as char( K ) = 2. We have
(33)
0 = bb1 ζ = bb1 (λ Id V ⊗3 − c1 ) λ2 Id V ⊗3 − λc2 + c2 c1 = (2)λ bb1 λ2 Id V ⊗3 − λc2 + c2 c1
and
(33)
0 = bb2 ζ = bb2 (λ Id V ⊗3 − c2 ) λ2 Id V ⊗3 − λc1 + c1 c2 = (2)λ bb2 λ2 Id V ⊗3 − λc1 + c1 c2
so that b satisfies (36) and (37). We have
(23) (33) (36)

0 = b[b1 c2 c1 − cb2 ] = bb1 c2 c1 − bcb2 = bb1 c2 c1 + bb2 = bb1 −λ2 Id V ⊗3 + λc2 + bb2
so that bb2 = λbb1 (λ Id V ⊗3 − c2 ). Similarly using cb1 = b2 c1 c2 in (23), (33) and (37) we get bb1 =
λbb2 (λ Id V ⊗3 − c1 ). Using these formulas we obtain
bb2 = λbb1 (λ Id V ⊗3 − c2 ) = λλbb2 (λ Id V ⊗3 − c1 )(λ Id V ⊗3 − c2 )

(37) (33)
= λ2 bb2 λ2 Id V ⊗3 − λc1 + c1 c2 − λc2 = −λ3 bb2 c2 = λ3 bb2
so that (λ3 − 1)bb2 = 0. Assume b is not zero. By Proposition 3.3, if λ = 0, 1, we get that b is surjective.
Thus bb2 = b( V ⊗ b) is surjective too and hence λ3 = 1, contradicting (3)λ = 0. 2
Remarks 4.4. Concerning the converse of Theorem 4.3, let us note that if b = 0 then the map
ιc,b : V → U ( V , c, b) = S ( V , c) is clearly injective. On the other hand, let K be a field of character-
istic zero. Given a c-bracket b on a braided vector space ( V , c) where c2 = Id V ⊗ V (i.e. ( V , c) is of
Hecke-type of regular mark 1), then the canonical map ιc,b : V → U ( V , c, b) is injective whenever
b fulfills (29) (see [Kh, Theorem 5.2]).
Example 4.5. (See [Mas].) Let K be a field with char( K ) = 2. Let V = K x and let c : V ⊗ V → V ⊗ V ,
c = Id V ⊗ V . Define b : V ⊗ V → V by b(x ⊗ x) = ax for some a ∈ K \ {0}.
Assume there exists λ ∈ K \ {0, 1} such that (3)!λ = 0 i.e. such that λ is not a primitive third root
of unity. Clearly (c + Id V ⊗ V )(c − λ Id V ⊗ V ) = 0 so that c is of Hecke-type of mark λ. Moreover b is a
c-bracket on the braided vector space ( V , c). Thus we can consider the universal enveloping algebra
T ( V , c) K [X]
U = U ( V , c, b) := .
(c(z) − λ z − b(z) | z ∈ V ⊗ V ) ((1 − λ) X 2 − a X )
The canonical map ιc,b : V → U ( V , c, b) is clearly injective. Nevertheless b = 0.
Remark 4.6. Let ( V , I ⊕ I ∗ = V ⊗2 , [ , ]) be a braided Lie algebra in the sense of [Gu2, Definition 1]
such that I := I − = Im(q Id V ⊗2 − S ) and I ∗ := I + = Im(Id V ⊗2 + S ) where I ± ⊆ V ⊗2 are as in [Gu2,
Section 1]. Here S : V ⊗2 → V ⊗2 is a braiding of Hecke-type of mark q. Let b := −(2)q [ , ]. Then b
is a morphism in the braided monoidal category A defined as before [Gu2, Definition 1]. Thus b is
compatible with the braiding in A. This entails that b : V ⊗ V → V is an S-bracket. In view of [Gu2,
Proposition 5], we have that the map θ : S ( V , c) → gr U ( V , c, b) of Proposition 4.1 is an isomorphism.
This implies that the canonical map ι S ,b : V → U ( V , S , b) is injective. In fact let x ∈ V be in the kernel
of ι S ,b . Then πU (x) = ι S ,b (x) = 0 so that θ(x) = πU (x) + U 0 = 0 + U 0 . Since θ is injective, we get that
x = 0. Therefore, in view of Theorem 4.3, if q = −1 is not a cubic root of one, we deduce that b = 0
and hence [ , ] = 0 (note that in [Gu2] the characteristic of K is assumed to be zero).
Anyway, we outline that the case considered in [Gu2] does not require, in general, neither that
I := I − = Im(q Id V ⊗2 − S ) nor that I ∗ := I + = Im(Id V ⊗2 + S ). In fact the aim of [Gu2] is to introduce
a braided counterpart of the notion of S-Lie algebra (for S involutive) such that the corresponding
enveloping algebra is a quadratic algebra. Here S needs not to be of Hecke-type.
5. A Milnor–Moore type theorem for braided bialgebras
In this section we prove the main result of this paper, Theorem 5.5, which represents a variant of
Milnor–Moore Theorem for braided bialgebras. Then we deduce some consequences of this theorem,
including applications to certain classes of bialgebras in braided categories.
Definition 5.1. Let ( A , c A ) be a connected braided bialgebra. Let P := P ( A ). The braiding c P = c A | P ⊗ P

will be called the infinitesimal braiding of A.
Remarks 5.2. Let ( A , c A ) be a connected braided bialgebra. Let P := P ( A ) and let c P be the infinitesi-
mal braiding of A . If gr A denotes the graded associated with respect to the coradical filtration, then
gr A is strictly graded. Thus
P (gr A ) = gr1 A P ( A ) = P .
1, 1
Through this identification, cgr A is equal to c P . In conclusion the infinitesimal braiding of gr A is the
infinitesimal braiding c P of A.
Definition 5.3. Let ( A , c A ) be a connected braided bialgebra and let P := P ( A ). The component
1, 1
gr A : A 2 / A 1 → A 1 / A 0 ⊗ A 1 / A 0 = P ⊗ P is called the infinitesimal comultiplication of A.
Let c P = c A | P ⊗ P and let λ ∈ K ∗ . We will say that gr A is λ-cocommutative (or that ( A , c A ) is
1, 1
1, 1 1, 1
infinitesimally λ-cocommutative) if c P ◦ gr A = λgr A, that is we have
1, 1 1, 1 1, 1
cgr A ◦ gr A = λgr A . (38)
Proposition 5.4. Let K be a field with char K = 2. Let A be a connected braided bialgebra and assume that
its infinitesimal braiding is of Hecke-type of mark λ = 0, 1 such that (3)!λ = 0. Let P be the space of primitive
elements of A and let b P = ∇(c P − λ Id P ⊗2 )| P ⊗ P be the c P -bracket on the braided vector space ( P , c P ) defined
in Proposition 2.4. Then b P = 0.
Let f : ( V , c, b) → ( P , c P , 0) be a morphism of braided brackets and assume that c is a braiding of Hecke-
type with mark λ. Then there is a unique morphism of braided bialgebras f : U ( V , c, b) → A that lifts f .
Proof. By Proposition 2.4(b) it follows that b P is a c P -bracket on ( P , c P ), hence we can apply the
universal property of U := U ( P , c P , b P ). There is a unique morphism of braided bialgebras φ A : U → A
that lifts Id P . Observe that the canonical map ιc,b : P → U is injective, as φ A ιc,b is the inclusion of P
into A. Now apply Theorem 4.3 to obtain that b P = 0. The last part follows by 2.4. 2
Theorem 5.5. Let K be a field with char K = 2. Let ( A , c A ) be a connected braided bialgebra which is infinites-
imally λ-cocommutative for some regular element λ = 0 in K . Let P = P ( A ). Then
• the infinitesimal braiding c P of A is of Hecke-type of mark λ, and

• A is isomorphic as a braided bialgebra to the symmetric algebra S ( P , c P ) of ( P , c P ) whenever λ = 1.
Proof. Let B := gr A. Clearly B is strongly N-graded as a coalgebra (see e.g. [AM2, Theorem 2.10]).
By assumption ( A , c A ) is infinitesimally λ-cocommutative and hence the same holds for ( B , c B ) i.e.
(c1B,1 − λ Id B 2 )1B,1 = 0. Since B is also 0-connected, by Theorem 2.15, B is a bialgebra of type one
1, 1
and c B is a braiding of Hecke-type of mark λ. In particular the infinitesimal braiding of A is of
Hecke-type of mark λ and B is generated as an algebra by B 1 so that A is generated as a K -algebra
by P = P ( A ) = B 1 . Therefore, the canonical braided bialgebra homomorphism f : U ( P , c P , b P ) → A,
arising by the universal property of the universal enveloping algebra, is surjective. Assume λ = 1. By
Proposition 5.4, b P = 0 hence U ( P , c P , b P ) = S ( P , c P ). On the other hand, by Theorem 2.17, P is the
primitive part of S ( P , c P ) and the restriction of f to P is injective so that f is injective by [Mo,
Lemma 5.3.3]. In conclusion f is an isomorphism. 2
Remark 5.6. Let ( H , c H ) be a connected braided Hopf algebra and let B := gr H be the graded coal-
gebra associated to the coradical filtration of H . Let λ ∈ K ∗ . Since B is always strongly N-graded
as a coalgebra (see e.g. [AM2, Theorem 2.10]), in view of Theorem 2.15, the following assertions are
equivalent:
• H is cosymmetric in the sense of [Kh, Definition 3.1] (see also [Kh, Theorem 3.5]) and c1B,1 is a
braiding of Hecke-type of mark λ,
• c1B,1 ◦ 1B,1 = λ1B,1 i.e. H is infinitesimally λ-cocommutative.
Remark 5.7. With hypothesis of Theorem 5.5, if λ = 1 then A is isomorphic as a braided bialgebra to
the universal enveloping algebra U ( P , c P , b P ) of ( P , c P , b P ). In fact regularity of λ in this case means
char( K ) = 0. By Theorem 4.3 bc = −b and bb1 (Id V ⊗3 − c2 + c2 c1 ) = 0 so that [Kh, Theorem 6.1] applies.
Corollary 5.8. Let K be a field with char K = 2. Let ( V , c) be a braided vector space such that c is a braiding
of Hecke-type of regular mark λ = 0, 1. Let A be a braided bialgebra such that gr A is isomorphic as a braided
bialgebra to S ( V , c) then A is isomorphic to the symmetric algebra S ( V , c) of ( V , c).
Proof. Obviously the infinitesimal comultiplication of S ( V , c) is λ-cocommutative and, by Proposi-

tion 2.8, S ( V , c) is connected. Thus gr A has the same properties. Hence A itself is connected and by
Remark 5.2, ( A , c A ) is infinitesimally λ-cocommutative. We conclude by applying Theorem 5.5. 2
Acknowledgment
We would like to thank the referee for helpful comments and suggestions.
References
[AM1] A. Ardizzoni, C. Menini, Braided bialgebras of type one, Comm. Algebra, in press, arXiv: math.CT/0702604.
[AM2] A. Ardizzoni, C. Menini, Associated graded algebras and coalgebras, submitted for publication, arXiv: 0704.2106v2.
[AMS1] A. Ardizzoni, C. Menini, D. Ştefan, A monoidal approach to splitting morphisms of bialgebras, Trans. Amer. Math.
Soc. 359 (2007) 991–1044.
[AMS2] A. Ardizzoni, C. Menini, D. Ştefan, Weak projections onto a braided Hopf algebra, J. Algebra 318 (1) (2007) 180–201.
[AS] N. Andruskiewitsch, H.-J. Schneider, Pointed Hopf algebras, in: S. Montgomery, et al. (Eds.), New Directions in Hopf
Algebras, in: Math. Sci. Res. Inst. Publ., vol. 43, Cambridge Univ. Press, 2002, pp. 1–68.
[Ba] J.C. Baez, Hochschild homology in a braided tensor category, Trans. Amer. Math. Soc. 334 (1994) 885–906.
[BG] A. Braverman, D. Gaitsgory, Poincaré–Birkhoff–Witt theorem for quadratic algebras of Koszul type, J. Algebra 181 (1996)
315–328.
[Di] J. Dieudonné, Introduction to the Theory of Formal Groups, Marcel Dekker, New York, 1973.
[Go] F. Goichot, Un théorème de Milnor–Moore pour les algèbres de Leibnitz, in: J.-L. Loday (Ed.), Dialgebras and Related
Operads, in: Lecture Notes in Math., vol. 1763, Springer, 2001.
[Gu1] D.I. Gurevich, Generalized translation operators in Lie groups, Soviet J. Contemp. Math. Anal. 18 (1993) 57–70.
[Gu2] D. Gurevich, Hecke symmetries and braided Lie algebras, in: Spinors, Twistors, Clifford Algebras and Quantum Defor-
mations, Sobótka Castle, 1992, in: Fundam. Theor. Phys., vol. 52, Kluwer Acad. Publ., Dordrecht, 1993, pp. 317–326.
[Gu3] D.I. Gurevich, The Yang–Baxter equation and the generalization of formal Lie theory, Soviet Math. Dokl. 33 (3) (1986)
758–762.
[JS] A. Joyal, R. Street, Braided monoidal categories, Adv. Math. 102 (1993) 20–78. Formerly Macquarie Math. Reports
No. 850087 (1985) and No. 860081 (1986).
[Kh] V.K. Kharchenko, Connected braided Hopf algebras, J. Algebra 307 (2007) 24–48.
[Ka] C. Kassel, Quantum Groups, Grad. Texts in Math., vol. 155, Springer, 1995.
[LR] J.-L. Loday, M. Ronco, On the structure of cofree Hopf algebras, J. Reine Angew. Math. 592 (2006) 123–155.
[Ma1] S. Majid, Examples of braided groups and braided matrices, J. Math. Phys. 34 (1993) 3246–3253.
[Ma2] A. Masuoka, Formal groups and unipotent affine groups in non-categorical symmetry, J. Algebra 317 (1) (2007) 226–249.
[Mas] A. Masuoka, private communication to the authors.
[MM] J.W. Milnor, J.C. Moore, On the structure of Hopf algebras, Ann. of Math. 81 (1965) 211–264.
[Mo] S. Montgomery, Hopf Algebras and Their Actions on Rings, CBMS Reg. Conf. Ser. Math., vol. 82, 1993.
[Ra] D.E. Radford, Hopf algebras with projection, J. Algebra 92 (1985) 322–347.
[R1] M. Ronco, Eulerian idempotents and Milnor–Moore Theorem for certain non-cocommutative Hopf algebras, J. Alge-
bra 254 (1) (2002) 152–172.
[R2] M. Ronco, A Milnor–Moore Theorem for Dendriform Algebra, in: Contemp. Math., vol. 267, 2000, pp. 245–263.
[Ro] M. Rosso, Quantum groups and quantum shuffles, Invent. Math. 133 (1998) 399–416.
[St] C.R. Stover, The equivalence of certain categories of twisted Lie and Hopf algebras over a commutative ring, J. Pure Appl.
Algebra 86 (3) (1993) 289–326.
[Sw] M. Sweedler, Hopf Algebras, Benjamin, New York, 1969.
[Ta] M. Takeuchi, Survey of Braided Hopf Algebras, in: Contemp. Math., vol. 267, 2000, pp. 301–324.
Journal of Algebra
A new integral basis for the centre of the Iwahori–Hecke

algebra of type A
Andrew Francis a,∗ , Lenny Jones b
a
School of Computing and Mathematics, University of Western Sydney, NSW 1797, Australia
b
Department of Mathematics, Shippensburg University, Pennsylvania, USA
Article history: We describe a recursive algorithm that produces an integral basis

Received 20 May 2008 for the centre of the Iwahori–Hecke algebra of type A consisting of
Available online 29 November 2008 linear combinations of monomial symmetric polynomials of Jucys–
Communicated by Leonard L. Scott, Jr.
Murphy elements. We also discuss the existence of integral bases
Keywords:
for the centre of the Iwahori–Hecke algebra that consist solely
Iwahori–Hecke algebra of monomial symmetric polynomials of Jucys–Murphy elements.
Symmetric group Finally, for n = 3, we show that only one such basis exists for
Center the centre of the Iwahori–Hecke algebra, by proving that there are
Centre exactly four bases for the centre of the corresponding symmetric
Integral basis group algebra which consist solely of monomial symmetric
Jucys–Murphy element polynomials of Jucys–Murphy elements.
Symmetric polynomial
Monomial
1. Introduction
The centre of the Iwahori–Hecke algebra of type A n−1 shares a number of characteristics of the
centre of the corresponding group algebra Z S n . For instance, it has the same dimension (the number
of conjugacy classes in S n ), it has an integral basis specialising to class sums [6], and like the centre
of the group algebra, it has long been known to be equal to the set of symmetric polynomials in
Jucys–Murphy elements when considered over the rational function field [1,9]. This last result has
been extended to the integral Iwahori–Hecke algebra [3], mirroring G. Murphy’s result for the centre
of the integral group algebra precisely. A natural question is then whether the basis given by Murphy
in [9] generalizes to the Iwahori–Hecke algebra. Unfortunately, this generalization fails [1, p. 64]. See
Section 4 for examples and comments for n = 3, 4, 5.
E-mail addresses: a.francis@uws.edu.au (A. Francis), lkjone@ship.edu (L. Jones).
doi:10.1016/j.jalgebra.2008.10.018
A. Francis, L. Jones / Journal of Algebra 321 (2009) 866–878 867
In this paper, we discuss the possibility that some sort of variation of Murphy’s monomial symmet-
ric polynomial basis might exist for the Iwahori–Hecke algebra, and we provide evidence to indicate
that integral bases of the centre of the Iwahori–Hecke algebra consisting solely of monomial symmet-
ric polynomials in Jucys–Murphy elements are rare (Section 5). In particular, in Section 6 we show
that there is exactly one basis of monomial symmetric polynomials in Jucys–Murphy elements for the
centre of the Iwahori–Hecke algebra of S 3 , by proving that there are exactly four such bases for
the centre of the corresponding group algebra. Our main theorem, however, shows that there always
exists an integral basis for the centre of the Iwahori–Hecke algebra consisting of integral linear com-
binations of monomial symmetric polynomials in Jucys–Murphy elements (Theorem 4.1). This is the
first such basis for the centre of the Iwahori–Hecke algebra given in terms of symmetric polynomials,
and the second description of an integral basis after [2,6].
We begin with definitions and preliminary results (Sections 2 and 3).
2. Definitions
2.1. Partitions and compositions
A composition λ is a finite ordered set of positive integers. If λ = (λ1 , . . . , λr ), the λi are called
the components of λ, and the number of components in λ, denoted (λ), is called its length. If λ is
r
a composition we write |λ| = i =1 λi . If |λ| = n we say λ is a composition of n, and we write λ n.
We denote the empty composition (the unique composition of zero) as ∅, and we define |∅| := 0.
A partition of n is a composition whose components are weakly decreasing from left to right. If λ
is a partition of n we write λ n. For any composition λ there is a unique partition with the same
components as λ, denoted λ̂. Repeated components of a composition will sometimes be denoted by
exponents; for instance (13 , 2) is shorthand for (1, 1, 1, 2). If λ = (λ1 , . . . , λr ) n then we define λ − 1
to be the composition (λ1 − 1, . . . , λr − 1) of n − (λ), excluding zero components. For example, if
λ = (3, 1, 4) 8, then λ − 1 = (2, 3) 5. For a given n, if a composition λ satisfies (λ) + |λ| n,
we define λ to be the partition of n obtained from the parts of λ by adding 1 to all parts, and
n − (λ) − |λ| copies of 1. Note that the construction of λ is relative to n: for λ = (3, 1, 4), we have
λ = (5, 4, 2, 1), if n = 12; while λ = (5, 4, 2, 1, 1, 1), if n = 14. In particular, if λ n, then λ − 1 = λ.
If λ = (λ1 , λ2 , . . . , λt ) is any nonempty composition, then we define

(λ1 , λ2 , . . . , λt −1 ) if (λ) 2,
λ :=
∅ if (λ) = 1.
Let Λ be the set of all compositions of all positive integers together with ∅. For the sake of conve-
nience in the sequel, we define recursively the following ordering on Λ. For any two compositions λ
and μ in Λ, we have

|λ| < |μ|, or
λ<μ ⇐⇒ (2.1)
0 < |λ| = |μ| and λ < μ .
2.2. The symmetric group
The symmetric group S n is the group of permutations of {1, . . . , n}. We will use the Coxeter pre-
sentation for S n given by the generators S = {si = (i i + 1) | 1 i n − 1}, and subject to the relations
s2i = 1 for each i; si s j = s j si if |i − j | 2; and si si +1 si = si +1 si si +1 for 1 i n − 2. An expression
w = si 1 . . . sik for w ∈ S n is said to be reduced if k is minimal. In this case, we say the length of w,
denoted ( w ), is k.
An element w of S n is called increasing if it can be written in the form si 1 si 2 · · · sik with 1 i 1 <
i 2 < · · · < ik < n. For such an element w, we say w is of shape λ, if λ is the composition of k defined
as follows. Without any rearrangement of the si j , we rewrite w = si 1 si 2 · · · sik as α1 α2 · · · αt , where
the αi are disjoint cycles of maximal length. Then λ = ((α1 ), (α2 ), . . . , (αt )). For example, s1 s3 s4 s6
868 A. Francis, L. Jones / Journal of Algebra 321 (2009) 866–878
has shape (1, 2, 1). Note that there exists an increasing element w in S n of shape λ if and only if
|λ| + (λ) = k + t n.
The conjugacy classes of S n are indexed by partitions of n. Denote the conjugacy class of cycle
type λ n by C λ . Note that C λ is the unique conjugacy class that contains a minimal element of
shape λ − 1. From the alternative perspective, if w is an increasing element of shape λ in S n , then
(λ) + |λ| n and w is a minimal element in the conjugacy class C λ .
2.3. Symmetric polynomials
A polynomial in the variables { X 1 , . . . , X n } is symmetric if it is fixed by the action of S n on the

indices of the variables. Each partition λ = (λ1 , . . . , λr ) with r n determines a monomial symmetric
polynomial,
λ λ
mλ = mλ ( X 1 , . . . , X n ) := X σ1(1) . . . X σr(r ) ,
σ ∈ Sn
where m∅ := 1. Note that we also write mλ1 ,λ2 ,...,λr for mλ .

For example,
m2,1 ( X 1 , X 2 , X 3 ) := X 12 X 2 + X 12 X 3 + X 22 X 1 + X 22 X 3 + X 32 X 1 + X 32 X 2 .
Given a composition λ = (λ1 , λ2 , . . . , λr ) with r n, we define the monomial quasi-symmetric poly-

nomial p λ as
λ λ λ
p λ ( X 1 , X 2 , . . . , X n ) := X j1 X j2 ··· X jr,
1 2 r
where the sum is taken over all ordered r-tuples ( j 1 , j 2 , . . . , j r ) with 1 j 1 < j 2 < · · · < j r n. We
define p ∅ := 1, and we also write p λ1 ,λ2 ,...,λr in lieu of p λ when λ = (λ1 , λ2 , . . . , λr ).
For example, p 2,1 = X 12 X 2 + X 12 X 3 + X 22 X 3 and p 1,2 = X 1 X 22 + X 1 X 32 + X 2 X 32 so that m2,1 =
p + p 1, 2 .
2, 1
2.4. The Iwahori–Hecke algebra
The Iwahori–Hecke algebra Hn of S n is the unital associative algebra generated over R = Z[ξ ] by
the set { T̃ s | s ∈ S }, with relations

T̃ ws + ξ T̃ w if ( ws) < ( w ),
T̃ w T̃ s =
T̃ ws otherwise,
where T̃ w := T̃ si . . . T̃ sir if w = si 1 . . . sir is reduced. The Iwahori–Hecke algebra Hn is an R-module

1
with basis { T̃ w | w ∈ S n reduced}, and specializes at ξ = 0 to the symmetric group algebra Z S n . We
let Z (Hn ), respectively Z (Z S n ), denote the centre of Hn , respectively Z S n . Specialization of Z (Hn ) at
ξ = 0 gives Z (Z S n ). It is known that Z (Hn ) has an integral R-basis {Γλ | λ n} characterized by the
following properties:
• Γλ specializes at ξ = 0 to the sum of elements in the conjugacy class C λ in Z (Z S n ), and

• the only shortest elements from any conjugacy class appearing in Γλ are those from C λ , and they
appear with coefficient 1.
These results can be found in [2,6]. We refer to this basis in the sequel as the Geck–Rouquier basis. If
λ = (λ1 , λ2 , . . . , λt ), then we write Γλ as Γλ1 ,λ2 ,...,λt .
The Iwahori–Hecke algebra is equipped with an inner product defined via the standard trace func-
tion tr( w ∈ S n r w T̃ w ) = r1 as follows,

T̃ u , T̃ v
:= tr( T̃ v T̃ u ) = 1 if vu = 1,
0 otherwise.
If h is central in Hn then T̃ w , h
is the coefficient of T̃ w in h. This inner product gives rise, via
specialization at ξ = 0, to an inner product on the group algebra.
The Jucys–Murphy elements [7,9] L i are defined by setting L 1 = 0 and for 2 i n,

L i := T̃ ( j i ) .
1 j i −1
For example, L 2 = T̃ s1 and L 3 = T̃ s1 s2 s1 + T̃ s2 . Specialization at ξ = 0 gives corresponding Jucys–Murphy

elements in Z S n .
Let M ξ = M ξ (n) = {mλ ( L 1 , . . . , L n ) | (λ) n}, the set of monomial symmetric polynomials in
Jucys–Murphy elements in Z (Hn ), and let M 0 denote the corresponding set in Z (Z S n ).
3. Preliminaries
In this section we recall some of the machinery from [3] necessary for the first main result of this
paper, Theorem 4.1.
For k a positive integer, define
k

k+m−1
a(k) := ξ 2m .
2m − 1
m=1
This polynomial was introduced in [8], but can also be defined using q-integers, as in [3] (noting the
r
different definition of ξ used there). For a composition λ = (λ1 , . . . , λr ), define a(λ) = i =1 a(λi ). For
compositions λ and μ with |λ|, |μ| < n, define the polynomial A λ,μ ∈ R to be the coefficient of p λ in

|γ |=|λ|−|μ| a(γ ) p p , setting A λ,μ = 0 if |λ| |μ|. Define the matrix A
γ μ (k) by labelling its rows with
compositions λ, where |λ| < k, according to the order defined in Section 2.1, and with (λ, μ)-entry
equal to A λ,μ . Define the matrix Z (k) recursively by setting Z (1) = (1) and

Z (k) I
Z (k+1) := .
0 I
Define

Ξ (k) 0
Ξ (k+1) := Z (k+1) A (k+1)
0 Ξ (k)
and Ξ (1) = (1). The matrix Ξ (k) is the matrix of coefficients of minimal length conjugacy class ele-
ments in quasi-symmetric polynomials of Jucys–Murphy elements [3, Theorem 6.7]. Specifically,

(k)
T̃ w , p μ ( L 1 , . . . , L n ) = Ξλ ,μ ,
for λ, μ k and w of shape λ. For the sake of convenience, we use the bijection λ → λ , that maps
compositions of a fixed positive integer k to compositions of nonnegative integers less than k, to
(k) (k)
define the re-indexed matrix X λ,μ := Ξλ ,μ . To obtain coefficients in symmetric polynomials (as op-
posed to quasi-symmetric polynomials) of Jucys–Murphy elements, define the matrix T (k) indexed by
compositions λ, μ of k by setting

1 if λ = μ,
(k)
T λ,μ = 1 if λ̂ = μ,
0 otherwise.
Finally, M (k) is the matrix indexed by partitions of k obtained from ( T (k) )−1 X (k) T (k) by deleting
those rows and columns labelled by non-partitions.
For positive integers n and k n/2, the (λ, μ)-entry of M (k) is T̃ w , mμ
, where w ∈ S n is an
increasing element of shape λ [3, Theorem 7.1] (there is always such a w since k n/2). Also, by the
nature of the construction of M (k) , its entries are independent of n [3, Lemma 5.2]. One of the main
results of [3] is that M (k) is invertible over R. This fact was originally conjectured by G. James [8]. We
give a brief summary of the construction of the inverse of M (k) , needed in Section 4.
Let

Υ 0
Υ (k+1) := Z (k+1) A (k+1) with Υ (1) = (1).
0 Υ
Then Ξ (k) and K (k

) Υ (k) K (k) are inverse [3, Lemma 5.5], where K (k) is defined recursively
K (k) − K (k)
by K (k+1) = and K (1) = (1). As with Ξ (k) above, we define the re-indexed matrix
0 − K (k)
(k) (k)
Y λ,μ := Υλ ,μ .
The matrices ( T (k) )−1 X (k) T (k) and ( T (k) )−1 K (k) Y (k) K (k) T (k) are inverse. Let N (k) be the matrix ob-
tained from ( T (k) )−1 K (k) Y (k) K (k) T (k) by deleting rows and columns not labelled by partitions. It is
shown in [3] that M (k) N (k) = I , and the entries of N (k) are from R.
4. A new integral basis for Z (Hn ) using symmetric polynomials in Jucys–Murphy elements
In the case of the group algebra of the symmetric group, Murphy has shown that the set of mono-
mial symmetric polynomials mμ in Jucys–Murphy elements, such that an increasing element of shape
μ is in S n , is a Z-basis for Z (Z S n ) [9]. We have verified that the generalization of this particular set
is not an R-basis for Z (Hn ) when n = 3, 4, 5, and we conjecture that this set is an R-basis for Z (Hn )
only when n = 1, 2. (While a statement generalizing Murphy’s result to Z (Hn ) for all n appears in
[8, Theorem 3.6], the proof contains an error.)
We present the details here of the counterexample when n = 3. In this case, the direct gen-
eralization of Murphy’s symmetric group result states that the set of monomials {m∅ , m1 , m2 } in
Jucys–Murphy elements is an R-basis for Z (H3 ). If this is true, then the transition matrix from this
monomial “basis” to the Geck–Rouquier basis {Γ1,1,1 , Γ2,1 , Γ3 } must be invertible in R. This transition
matrix is:

1 0 3
0 1 2ξ .
0 0 1 + ξ2
However, its determinant is 1 + ξ 2 , which is not invertible in R. Observe that this set of monomials
does specialize at ξ = 0 to a Z-basis for Z (Z S 3 ).
While a more general statement—that there is an integral basis for Z (Hn ) consisting solely of
elements from M ξ —also seems unlikely in general (though true for n 4, see Section 5), we show
here that there always exists an integral basis for Z (Hn ) which consists of R-linear combinations of
elements from M ξ . This is the first such basis for Z (Hn ) given in terms of symmetric polynomials,
and the second description of an integral basis after [2,6].
For a partition λ, let
(|λ|)
Mλ = N μ,λ mμ ,
μ|λ|
(|λ|)
where N μ,λ is the entry in the matrix N (|λ|) whose row is labelled by μ and whose column is labelled
by λ.
Theorem 4.1. The set B = {Mλ | (λ) + |λ| n} is an R-basis for Z (Hn ).
Proof. We will show that Mλ is the class element Γλ of the Geck–Rouquier basis, where λ is the
unique partition of n whose corresponding conjugacy class has a minimal length element of shape λ,
plus a linear combination of class elements whose shortest elements are strictly shorter than |λ|.
Firstly, the coefficient of Γλ in any central element h is T̃ w , h
for w an increasing element of
shape λ − 1, as a direct result of the characterization given in Section 2.
For the moment, consider n 2k—large enough so that all w of shape λ k are in S n . The ele-
ments of M ξ , indexed by partitions ordered according to (2.1), are related to the class elements Γλ
via a block upper-triangular matrix, thanks to [8, Theorem 2.7]. That is, for monomial symmetric
polynomials of degree less than or equal to k, we have
(m∅ , m1 , . . . , m1k ) = (Γ∅ , . . . , Γ1k ) M ,
where M is a block upper triangular matrix with diagonal blocks M (0) , . . . , M (k) . As noted in the
previous section, the diagonal blocks are independent of n.
Restricting to monomials of degree k we have

(mk , . . . , m1k ) = (Γk , . . . , Γ1k ) M (k) + linear combinations of Γν for |ν − 1| < k .
Therefore
(M(k) , . . . , M(1k ) ) = (mk , . . . , m1k ) N (k)

= (Γk , . . . , Γ1k ) + linear combinations of Γν for |ν − 1| < k .
Because the entries of the matrices M (k) and N (k) are independent of n, we have

Mλ = Γλ + a linear combination of Γν |ν − 1| < |λ| .
The theorem follows by induction on |λ|. 2
Remark. The bases given in Theorem 4.1 embed in each other. That is, if |λ| + (λ) m n, then the
linear combination of elements from M ξ represented by Mλ is in the integral basis for both Z (Hm )
and Z (Hn ).
Theorem 4.1 also provides an effective algorithm for computing an integral basis for Z (Hn ). The
existing algorithm for computing class elements, given in [2], requires the computation of each class
element independently, with terms T̃ w added individually. That is, many computations are required
for every conjugacy class in the symmetric group. However, the algorithm from Theorem 4.1 requires
only the computation of the n matrices N (0) , . . . , N (n−1) . Once these matrices are computed, basis
elements corresponding to minimal elements of length less than n can be obtained immediately in
terms of elements from M ξ . The problem of obtaining the matrices N (k) reduces to calculating the
matrix A (k) , defined in Section 3, whose entries are determined by structure constants in the ring of
quasi-symmetric polynomials. But most importantly, because of the recursive nature of the algorithm
described in Theorem 4.1, once a basis of Z (Hn−1 ) is obtained, only one additional matrix, N (n−1) , is
required to complete the basis for Z (Hn ).
To illustrate Theorem 4.1, we provide the bases for Z (Hn ) produced therein when n = 3, 4 and 5.
The matrices M (k) and N (k) , when k 4, are used in the calculations. For k 3 they are:
M (0) = N (0) = M (1) = N (1) = (1),

1 + ξ2 1 1 −1
M (2) = , N (2) = ,
ξ 2
1 −ξ 2
1 + ξ2

1 + 5ξ 2 + 5ξ 4 + ξ 6 3 + 5ξ 2 + ξ 4 1
(3)
M = 2ξ 2 + 4ξ 4 + ξ 6 1 + 4ξ 2 + ξ 4 1
3ξ 4 + ξ 6 3ξ 2 + ξ 4 1
and

1 + ξ2 −2ξ 2 − 3 2 + ξ2
(3)
N = −2ξ 2 − ξ 4 1 + 5ξ 2 + 2ξ 4 −1 − 3ξ 2 − ξ 4 .
3ξ 4 + ξ 6 −3ξ 2 − 7ξ 4 − 2ξ 6 1 + 3ξ 2 + 4ξ 4 + ξ 6
Example 4.2. n = 3

B = m ∅ , m 1 , m 2 − ξ 2 m 1, 1 .
Note that the last element in this basis is obtained from first column of N (2) , corresponding to the
partition (2), because the element of shape (2), s1 s2 , is in S 3 . The other column of N (2) is labelled
by (1, 1), but S 3 contains no element of shape (1, 1).
Example 4.3. n = 4

B = m∅ , m1 , m2 − ξ 2 m1,1 , −m2 + 1 + ξ 2 m1,1 ,

1 + ξ 2 m3 + −2ξ 2 − ξ 4 m2,1 + 3ξ 4 + ξ 6 m1,1,1 .
Here we have both partitions of 2 giving an element in S 4 , and so both columns of N (2) are
represented. We also have the first column of N (3) represented, labelled by the partition (3).
Example 4.4. n = 5

B = m∅ , m1 , m2 − ξ 2 m1,1 , −m2 + 1 + ξ 2 m1,1 ,

1 + ξ 2 m3 − 2ξ 2 + ξ 4 m2,1 + 3ξ 4 + ξ 6 m1,1,1 ,

−2ξ 2 − 3 m3 + 1 + 5ξ 2 + 2ξ 4 m2,1 + −3ξ 2 − 7ξ 4 − 2ξ 6 m1,1,1 ,

1 + 5ξ 2 + 5ξ 4 + ξ 6 m4 − ξ 8 + 6ξ 6 + 9ξ 4 + 4ξ 2 m2,2 − ξ 8 + 6ξ 6 + 9ξ 4 + 3ξ 2 m3,1

+ ξ 10 + 7ξ 8 + 14ξ 6 + 8ξ 4 m2,1,1 − ξ 12 + 8ξ 10 + 20ξ 8 + 16ξ 6 m1,1,1,1 .
In this case, the first two columns of N (3) , labelled by partitions (3) and (2, 1), are represented, but
not the third column, because S 5 contains no element of shape (1, 1, 1). The final element of the basis
is obtained from the first column of N (4) , which corresponds to the partition (4) and the increasing
element s1 s2 s3 s4 in S 5 (N (4) not shown).
5. Bases for Z (Hn ) that are subsets of M ξ
While Theorem 4.1 gives a basis for Z (Hn ) in terms of elements from M ξ , it is clearly not a basis
consisting solely of elements from M ξ when n 3. The question of whether such a basis exists in
general is open.
Open Question 5.1. For any n, does there exist an integral basis for Z (Hn ) consisting solely of elements
from M ξ ?
When n 4, there is such a basis. For example, Z (H3 ) has the basis {m∅ , m1 , m1,1 }, and Z (H4 )
has bases
{m∅ , m1 , m2 , m1,1 , m1,1,1 }, {m∅ , m1 , m1,1 , m1,1,1 , m2,1,1 } and {m∅ , m1 , m1,1 , m1,1,1 , m2,2,2 }.
See [4] for details of the n = 4 case, including a characterization of monomial bases of Z (Z S 4 ).
We do not know of any such basis of monomials for Z (H5 ). However, since we have a basis of
monomials for Z (H4 ), we would have an affirmative answer to Open Question 5.1 for Z (H5 ) if we
could find three monomials indexed by partitions of n 4 that span the three class elements whose
shortest terms are of length three or four (s1 s2 s3 , s1 s2 s4 and s1 s2 s3 s4 ), namely Γ4,1 , Γ3,2 and Γ5 ,
modulo shorter class elements. One of these could be m1,1,1,1 , which is equal to Γ5 . Thus, the problem
reduces to finding two monomials that span Γ4,1 and Γ3,2 modulo other class elements. Since there
is no bound on the size of the partition for a monomial, beyond requiring the partition to have fewer
than 5 parts (because the monomial is in L 1 , . . . , L 5 and L 1 = 0), there are infinitely many candidates
for such monomials.
6. There is only one integral basis for Z (H3 ) that is a subset of M ξ
Any subset of M ξ that is a basis for Z (Hn ) specializes to a subset of M 0 that is a basis for Z (Z S n ).
Unfortunately, not every integral basis for Z (Z S n ) which consists of elements of M 0 is a specialization
at ξ = 0 of an integral basis for Z (Hn ) which consists of elements of M ξ . See the counterexample in
Section 4. However, determining the subsets of M 0 that are bases for Z (Z S n ) allows us to restrict
our attention to the corresponding subsets of M ξ in Z (Hn ). In this section, we show that there are
exactly four subsets of M 0 which are bases for Z (Z S 3 ), only one of which corresponds to an integral
basis for Z (H3 ).
6.1. Coefficients of class sums in elements of M 0 for Z S 3
Let C be the matrix with three rows and infinitely many columns whose entries are the coefficients
of the class elements Γλ (rows) in the monomial symmetric polynomials mμ (columns) in Jucys–
Murphy elements for Z S 3 , where μ has fewer than three parts. In other words, the entries of C are
w λ−1 , mμ
, where w λ−1 is an increasing element of shape λ − 1, and λ = (1, 1, 1), (2, 1), or (3).
For example, from Table 1, which gives the first twenty columns of C , the coefficient of Γ2,1 in
m4,3 is w 1 , m4,3
= s1 , m4,3
= 8. The rows of C are indexed by class elements ordered by the
length of a shortest term in the class sum. The columns of C are indexed by monomial symmetric
polynomials ordered from left to right according to the ordering on compositions given in Section 2.1.
The monomials corresponding to partitions with more than two parts are not listed since in those
situations the monomials are either zero (exactly three parts) or nonexistent (more than three parts).
For any integer k, we define the k-block of C in the following way. If k 2, then the k-block is
the submatrix of C consisting of those columns labelled by mλ for λ k. The 0, or empty block, has
the single column headed by m∅ , while the 1-block has the single column headed by m1 . A block is
called even or odd according to whether k is even or odd.
Certain rows of zeros appearing in the k-blocks “alternate” according to whether k is even or odd.
More precisely, we have the following:
Table 1
Coefficients of class sums in monomial symmetric polynomials of Jucys–Murphy elements in Z S 3 . This table was generated
using GAP [10] with CHEVIE [5].
m∅ m1 m2 m1,1 m3 m2,1 m4 m2,2 m3,1

Γ1,1,1 1 0 3 0 0 0 7 2 2
Γ2,1 0 1 0 0 3 2 0 0 0
Γ3 0 0 1 1 0 0 5 1 4
m5 m3,2 m4,1 m6 m3,3 m4,2 m5,1 m7 m4,3 m5,2 m6,1

Γ1,1,1 0 0 0 23 2 8 10 0 0 0 0
Γ2,1 11 4 6 0 0 0 0 43 8 12 22
Γ3 0 0 0 21 3 6 12 0 0 0 0
Lemma 6.1. Let k be a nonnegative integer, and let μ k. Let w λ−1 be an element of shortest length appearing
in Γλ . If ( w λ−1 ) ≡ k (mod 2), then w λ−1 , mμ
= 0.
Proof. A column from an even block is a sum of products of Jucys–Murphy elements; the products
being of even total degree. Each group element w in a Jucys–Murphy element is of odd length, be-
cause it is a transposition. A product of two elements of odd length is even, of even length is even,
and of mixed lengths is odd. Thus, the length of a product of an even number of odd-length elements
will be even. Consequently, the coefficient of any element of odd length (s1 and s1 s2 s3 ) in such a
product will be zero.
Similarly, a column from an odd block is a sum of products of an odd number of Jucys–Murphy el-
ements. Consequently, the coefficient of even length elements (the identity and s1 s2 ) will be zero. 2
Lemma 6.2. The following relations hold for monomial symmetric polynomials in two variables:
(1) mi = m2 mi −2 − m2,2mi −4 for i 5;

(2) mi ,i = m1,1mi −1,i −1 for i 2;
(3) mi + j ,i = mi ,i m j for i , j 1.
Proof. Straightforward. 2
Because there are only two non-zero Jucys–Murphy elements in Z S 3 , namely the specializations
of L 2 and L 3 , Lemma 6.2 gives relations among elements of M 0 for Z S 3 .
Lemma 6.3. The following monomial recursions hold in Z (Z S 3 ):
(1) mi = 5mi −2 − 4mi −4 for i 5;

(2) mi ,i = mi −1,i −1 + 2mi −2,i −2 for i 3;
(3) mi + j ,i = 5mi + j −2,i − 4mi + j −4,i for i 1 and j 5.
Proof. To prove (1), we first verify the recurrence for i = 5, 6, 7, 8. For instance, from Table 1, we have

0 0 0
m5 = 11 =5 3 −4 1 = 5m3 − 4m1
0 0 0
and

23 7 3
m6 = 0 =5 0 −4 0 = 5m4 − 4m2 ,
21 5 1
where each monomial is written as a vector with respect to the basis of class sums for the centre.
If i 9, we have by Lemma 6.2(1) and induction,
mi = m2 mi −2 − m2,2 mi −4
= m2 (5mi −4 − 4mi −6 ) − m2,2 (5mi −6 − 4mi −8 )
= 5(m2 mi −4 − m2,2mi −6 ) − 4(m2 mi −6 − m2,2mi −8 )
= 5mi −2 − 4mi −4 .
The proof of (2) is similar. To prove (3), we use Lemma 6.2(3) and recurrence (1) of this lemma:
m i + j ,i = m i ,i m j
= mi ,i (5m j −2 − 4m j −4 )
= 5mi ,i m j −2 − 4mi ,i m j −4
= 5mi + j −2,i − 4mi + j −4,i . 2
Lemma 6.4. If i 1, then
1
1, mi
= 1 + (−1)i 2i + 5 ,
6
1
s1 , mi
= 1 − (−1)i 2i + 1 ,
6
1
s1 s2 , mi
= 1 + (−1)i 2i − 1 .
6
Proof. The lemma is verified easily for 1 i 5 using Table 1. For i > 5, we use the recurrences of
Lemma 6.3, and proceed by induction.
For instance, Table 1 gives, respectively for i = 1, . . . , 5, the values 0, 3, 0, 7, 0 of the inner prod-
uct 1, mi
, and this agrees with the claim in each case. If i > 5, then mi = 5mi −2 − 4mi −4 by
Lemma 6.3, and so
1, mi
= 5 1, mi −2
− 4 1, mi −4

1 1
=5 1 + (−1)i −2 2i −2 + 5 − 4 1 + (−1)i −4 2i −4 + 5
6 6
1
= 1 + (−1)i 5 2i −2 + 5 − 4 2i −4 + 5
6
1
= 1 + (−1)i 5.22 − 4 2i −4 + 5
6
1
= 1 + (−1)i 2i + 5 .
6
The other inductions are similar. 2

Lemma 6.5. If i , j 1, then
1 i
1, mi ,i
= 2 + 2(−1)i ,
3
s1 , mi ,i
= 0,
1 i
s 1 s 2 , m i ,i
= 2 − (−1)i ,
3
1
1, mi + j ,i
= 1 + (−1) j 2i + j + 2i + 4(−1)i ,
6
1
s 1 , m i + j ,i
= 1 − (−1) j 2i + j + 2i ,
6
1
s 1 s 2 , m i + j ,i
= 1 + (−1) j 2i + j + 2i − 2(−1)i .
6
Proof. The first three equalities are easily verified using Table 1 when 1 i 2. For i 3, use
Lemma 6.3(2) and proceed by induction.
For the second three equalities, we first verify the eight cases: 1 i 2, 1 j 4, all but one of
which is contained in Table 1. Then, for fixed i 1, the closed forms follow by induction on j, using
Lemma 6.3(3). Similarly, for fixed j 1, we use Lemma 6.2(3) and Lemma 6.3(2) to write
mi + j ,i = mi ,i m j = (mi −1,i −1 + 2mi −2,i −2 )m j = mi −1+ j ,i −1 + 2mi −2+ j ,i −2 ,
and proceed by induction on i. 2
Corollary 6.6. All coefficients of class elements in the monomials mi + j ,i , for i , j 1, are even.
6.2. Monomial bases for Z (Z S 3 ) and Z (H3 )
Lemma 6.7. The only pair of non-trivial monomials that integrally span {Γ1,1,1 , Γ3 } is {m2 , m2,2 }.
Proof. To begin with, Lemma 6.1 tells us that any two-element integral spanning set for {Γ1,1,1 , Γ3 }
cannot contain monomials from odd blocks. Secondly, we are restricted to monomials of form m2 j and
mi ,i in light of Corollary 6.6 (for i , j 1). Lemmas 6.4 and 6.5 imply that the matrix of coefficients
for the rows labelled by Γ1,1,1 and Γ3 for the two monomials m2 j and mi ,i is

(1 + (−1)2 j )(22 j + 5)/6 (2i + 2(−1)i )/3
,
(1 + (−1)2 j )(22 j − 1)/6 (2i − (−1)i )/3
with determinant
(−1)i +1 (4 j + 1) + 2i +1
d= .
3
If d = ±1, then {m2 j , mi ,i } is an integral spanning set for {Γ1,1,1 , Γ3 }. We determine all values for i
and j such that d = ±1.
If i is odd, then d > 0, and so we only need to find values of i and j for d = 1. But, since i , j 1, it
follows that d ≡ 3 (mod 4), and so there are no solutions in this case. If i is even, then d ≡ 1 (mod 4).
Thus, d = 1, and we must solve the exponential Diophantine equation
2i +1 = 4 j + 4.
If j 2, then reduction modulo 8 shows that this equation has no solutions. Therefore, the only
solution is i = 2 and j = 1, which gives the set of monomials {m2 , m2,2 }. 2
Allowing the trivial monomial m∅ produces only three more pairs of monomials that integrally
span {Γ1,1,1 , Γ3 }:
Proposition 6.8. The only pairs of monomials that integrally span {Γ1,1,1 , Γ3 } are {m∅ , m2 }, {m∅ , m1,1 },
{m∅ , m2,2 } and {m2 , m2,2 }.
Proof. If the monomial m∅ is included in a spanning set, the other monomial included to span
{Γ1,1,1 , Γ3 } must have coefficient one on Γ3 . The only monomials satisfying this condition are m2 ,
m1,1 and m2,2 (Lemmas 6.4 and 6.5). 2
Proposition 6.9. The only monomial that integrally spans {Γ2,1 } is m1 .
Proof. To span this one-element set integrally, a monomial must clearly have coefficient 1 or −1
on Γ2,1 . From Lemma 6.4, the only monomial of form mi that has coefficient 1 or −1 on Γ2,1 is m1 .
Monomials from even blocks are excluded because they have coefficient zero on Γ2,1 (Lemma 6.1). At
the same time, other monomials in odd blocks are all of form mi , j for i = j, and so always have even
coefficients by Corollary 6.6. This proves the claim. 2
Theorem 6.10. There are exactly four bases for Z (Z S 3 ) consisting solely of elements from M 0 . They are
{m∅ , m1 , m2 }, {m∅ , m1 , m1,1 }, {m∅ , m1 , m2,2 } and {m1 , m2 , m2,2 }.
Proof. This follows immediately from Propositions 6.8 and 6.9. 2
Theorem 6.11. The only subset of M ξ that is an integral basis for Z (H3 ) is {m∅ , m1 , m1,1 }.
Proof. The proof reduces to checking the columns corresponding to the monomials that form bases
for Z (Z S 3 ) (Theorem 6.10). These are:
m∅ m1 m2 m 1, 1 m 2, 2
Γ 1, 1, 1 1 0 3 0 2+ ξ 2
Γ 2, 1 0 1 2ξ 0 ξ 3 + ξ2
Γ3 0 0 1 + ξ2 1 1 + 4ξ + ξ 2
The only three columns that together give an integrally invertible matrix are m∅ , m1 and m1,1 . 2
References
[1] Richard Dipper, Gordon James, Blocks and idempotents of Hecke algebras of general linear groups, Proc. London Math. Soc.
(3) 54 (1) (1987) 57–82.
[2] Andrew Francis, The minimal basis for the centre of an Iwahori–Hecke algebra, J. Algebra 221 (1) (1999) 1–28.
[3] Andrew Francis, John J. Graham, Centres of Hecke algebras: The Dipper–James conjecture, J. Algebra 306 (2006) 244–
267.
[4] Andrew Francis, Lenny Jones, Monomial bases for the centres of the group algebra and Iwahori–Hecke algebra of S 4 , arXiv:
0709.0326.
[5] M. Geck, G. Hiss, F. Lübeck, G. Malle, G. Pfeiffer, CHEVIE—a system for computing and processing generic character tables,
in: Computational Methods in Lie Theory, Essen 1994, Appl. Algebra Engrg. Comm. Comput. 7 (3) (1996) 175–210.
[6] Meinolf Geck, Raphaël Rouquier, Centers and simple modules for Iwahori–Hecke algebras, in: Finite Reductive Groups,
Luminy, 1994, Birkhäuser Boston, Boston, MA, 1997, pp. 251–272.
[7] A.A.A. Jucys, Symmetric polynomials and the center of the symmetric group ring, Rep. Math. Phys. 5 (1) (1974) 107–112.
[8] Andrew Mathas, Murphy operators and the centre of Iwahori–Hecke algebras of type A, J. Algebraic Combin. 9 (1999)
295–313.
[9] G.E. Murphy, The idempotents of the symmetric group and Nakayama’s conjecture, J. Algebra 81 (1983) 258–265.
[10] Martin Schönert, et al., GAP—Groups, Algorithms, and Programming, fifth ed., Lehrstuhl D für Mathematik, Rheinisch West-
fälische Technische Hochschule, Aachen, Germany, 1995.
Journal of Algebra
The Nagata automorphism is shifted linearizable ✩

Stefan Maubach a,∗ , Pierre-Marie Poloni b
a
Radboud University Nijmegen, Postbus 9010, 6500 GL, Nijmegen, The Netherlands
b
Institut Mathématiques de Bourgogne, 9, avenue Alain Savary, BP 47870, 21078 Dijon Cedex, France
Article history: A polynomial automorphism F is called shifted linearizable if there

Received 3 June 2008 exists a linear map L such that L F is linearizable. We prove that
Available online 4 December 2008 the Nagata automorphism N := ( X − 2Y Δ− Z Δ2 , Y + Z Δ, Z ) where
Communicated by Michel Broué
Δ = X Z + Y 2 is shifted linearizable. More precisely, defining L (a,b,c)
as the diagonal linear map having a, b, c on its diagonal, we prove
Keywords:
Nagata automorphism that if ac = b2 , then L (a,b,c ) N is linearizable if and only if bc = 1.
Linearization problem We do this as part of a significantly larger theory: for example,
Tame automorphism any exponent of a homogeneous locally finite derivation is shifted
Locally nilpotent derivation linearizable. We pose the conjecture that the group generated
by the linearizable automorphisms may generate the group of
automorphisms, and explain why this is a natural question.
1. Preliminaries
1.1. Introduction
One of the main problems in affine algebraic geometry is to understand the polynomial automor-
phism group of affine spaces. In particular, it would be very useful to find some generators of these
groups. The case of dimension one is easy: every automorphism of the affine line is indeed affine.
(For a polynomial map, to be affine means to be of degree 1.)
In dimension two, the situation is well-known too. The Jung–van der Kulk theorem asserts that the
automorphism group of the affine plane is generated by the affine and the de Joncquière subgroups
[13,14]. Therefore, every automorphism of A2 is called tame.
✩
Funded by Veni-grant of council for the physical sciences, Netherlands Organisation for scientific research (NWO). Partially
funded by the Mathematisches Forschungsinstitut Oberwolfach as an Oberwolfach–Leibniz–Fellow.
*
Corresponding author.
E-mail addresses: s.maubach@math.ru.nl (S. Maubach), ppoloni@u-bourgogne.fr (P.-M. Poloni).
doi:10.1016/j.jalgebra.2008.11.008
880 S. Maubach, P.-M. Poloni / Journal of Algebra 321 (2009) 879–889
The case of dimension 3 is still open. Recently, Umirbaev and Shestakov solved in [18,19], the thirty
years old tame generators problem by proving that some automorphisms of C3 are non-tame and in
particular that the famous Nagata map is non-tame.
Actually, there are several candidate generator sets for the automorphism group of An (see Sec-
tion 4).
Nevertheless, from a “geometric point of view”, it is important to find generators which do not
depend on the choice of coordinates. Notice that, since a non-tame automorphism may be conjugate
to a tame one (Theorem 3.3 gives such an example), the notion of tame automorphism is not a
relevant geometric notion.
Therefore, it seems natural to define tamizable automorphisms, i.e. automorphisms which are con-
jugate to a tame one. In particular, it leads us to the following questions:
1. Is the Nagata automorphism tamizable?

2. Are all automorphisms of C3 tamizable?
Note that if the answer to the first question is negative, then it will be very difficult to prove it.
(The concept of degree is not invariant under conjugation, and so, the proof of Umirbaev–Shestakov
does not give ideas for this.)
In this paper, we will investigate the second question and study what consequences a positive an-
swer will give. It will lead us to consider the subgroup GLINn (C) ⊆ GAn (C) generated by linearizable
automorphisms. It turns out that this group contains all tame automorphisms, and, more surprising,
that the Nagata automorphism belongs to GLIN3 (C).
More precisely, we will show that “twice Nagata” is even linearizable! “Twice Nagata” stands for
the map (2I ) ◦ N, i.e. each component of the Nagata automorphism multiplied by 2. Then
4 −4
N 3 (2N ) N 3 = 2I
as explained in Theorem 3.3. In fact, we will prove that if D is a homogeneous locally finite derivation
on C[n] , then there exists s ∈ C∗ such that s exp( D ) = (sI ) ◦ exp( D ) is linearizable. We say that exp( D )
is shifted linearizable.
In the analytic realm, this is a known local fact, due to the Poincaré–Siegel theorem (see [2],
Chapter 5, or 8.3.1 of [6]). Roughly, this theorem states that for almost all s ∈ C∗ , and analytic map F
satisfying F (0) = 0, sF is holomorphically linearizable locally around 0. This theorem was the starting
point of a very interesting story1 about the (negative) solution of the Markus–Yamabe conjecture
and its link to the Jacobian conjecture, see [4,7,8]. One of the conjectures which was posed and killed
“along the way” of this story was Meister’s Linearization conjecture (see p. 186 of [6] or [5]). However,
the current article can be seen as a partial positive answer to a generalized Meister’s conjecture—in
fact, to such an extent that we revive a reformulated Meister’s conjecture:
Meister’s Linearization Problem. For which F ∈ GAn (C) does there exist some s ∈ C∗ such that sF is
linearizable?
This article is organized as follows. In Section 1 we define notations and mention well-known facts
on derivations. In Section 2 we show how to shift-linearize homogeneous derivations. In Section 3 we
use the previous section on Nagata’s map as an example, and explain exactly for which shifts it is
linearizable and when it is not. (We will prove that sN is linearizable if and only if s = 1, −1.) In
the last Section 4 we will discuss how the results of this article influence the current conjectures on
generators of GAn (C).
1
To save space we have to refer to [6, p. 185] and beyond, or the review [1].
S. Maubach, P.-M. Poloni / Journal of Algebra 321 (2009) 879–889 881
1.2. Notations and definitions
Let R be a commutative ring with one. (In this article, R will be C almost exclusively.) R [n] will
denote the polynomial ring in n variables over R. GAn ( R ) will denote the group of polynomial auto-
morphisms on R [n] .
We will identify an automorphism F ∈ GAn ( R ) with the n-tuple ( F 1 , . . . , F n ) of elements of
R [ X 1 , . . . , X n ], where F i = F ( X i ) for each i. A polynomial automorphism F = ( F 1 , . . . , F n ) ∈ GAn ( R ) be
given, we will denote by F ∗ the map F ∗ : R n → R n , (x1 , . . . , xn ) → ( F 1 (x1 , . . . , xn ), . . . , F n (x1 , . . . , xn )).
We will denote I for the identity map. Furthermore ∂ X (∂Y , ∂ Z , . . .) will denote the derivative with
respect to the variable X (Y , Z , . . .).
An R-derivation (or simply derivation if no confusion is possible) on an R-algebra A is an R-linear
map D : A → A that satisfies the Leibniz rule D (ab) = aD (b) + bD (a) for each a, b ∈ A. The set of
R-derivations (or derivations) on A is denoted by DER R ( A ) (or DER( A )). The set of R-derivations on
R [n] is denoted by DERn ( R ). DER( A ) forms a Lie algebra, as for any two derivations D, E the map
[ D , E ] := D E − E D is again a derivation, as can be easily checked. A locally nilpotent derivation is a
derivation D such that for each a ∈ A one finds an m ∈ N such that D m (a) = 0. For example: D = ∂ X
on C[ X ]. If R = k, a field, we define a locally finite derivation as a derivation D such that for each a ∈ A
the k-span of a, D (a), D 2 (a), . . . is finite dimensional. For example: D = ( X + 1)∂ X on C[ X ]. We use
LNDn (k), LFDn (k) for the sets of locally nilpotent respectively locally finite derivations on k[n] .
If D is a derivation on a ring A containing Q, then one can define the map exp( T D ) : A [[ T ]] →
T
∞ i
A [[ T ]] as the map sending f to i =0 i ! D ( f ). It is an automorphism of A [[ T ]], and its inverse
i
is exp(− T D ). In case D is locally nilpotent, the map exp( D ) : A → A is well-defined and again an
automorphism (with inverse exp(− D )). In case D is locally finite, one cannot always define the expo-
∞ a
nential map. For one, the field k must satisfy “a ∈ k then i =0 i ! ∈ k.” We will only take exponents of
locally finite derivations in case k = C.
We define the derivation δ on C[ X , Y , Z ] and the polynomial Δ ∈ C[ X , Y , Z ] by δ := −2Y ∂ X + Z ∂Y ,
and Δ := X Z + Y 2 . We will also denote by N the Nagata automorphism:

N = exp(Δδ) = X − 2Y Δ − Z Δ2 , Y + Z Δ, Z .
If λ ∈ C, we denote by N λ the following automorphism of C[ X , Y , Z ]:

N λ := exp(λΔδ) = X − λ2Y Δ − λ2 Z Δ2 , Y + λ Z Δ, Z .
Note that one can also use this formula to define N λ as an automorphism of k[ X , Y , Z ] for any
field k and any λ ∈ k.
1.3. A basic result
Lemma 1.1. Let D ∈ LND(C[n] ), and p ∈ C[n] , p = 0. If exp( D )( p ) = λ p, then λ = 1, and D ( p ) = 0.
Proof. Let q ∈ N be such that D q ( p ) = 0, D q+1 ( p ) = 0. Then D q ( p ) = D q (exp( D ))( p ) = D q (λ p ) =

) hence λ = 1. Assume q 1. Now 0 = D q−1 (0) = D q−1 (exp( D )( p ) − p ) =
λ D q ( p
q−1
( i =1 (i !)−1 D i ( p )) = D q ( p ). Contradiction, hence q = 0. 2
q
D
2. Shifted linearizability
2.1. Definition
We will define F ∈ GAn (C) to be shifted linearizable if there exists a linear map L ∈ GLn (C) such
that L F is linearizable, i.e. there exist G ∈ GAn (C) and L
∈ GLn (C) such that G −1 L F G = L
.
A special case is if sF is linearizable, where s ∈ C∗ . In this case L = sI .
2.2. Previous lemmas
Let us recall two well-known lemmas about linearizable polynomial maps.
Lemma 2.1. Let F , G ∈ GAn (C) be such that F ∗ (0) = G ∗ (0) = 0. Then, the linear part of F G is the composition
of the linear part of F and the linear part of G.
In particular, if L is the linear part of some F ∈ GAn (C) such that F ∗ (0) = 0, then L ∈ GLn (C) and L −1 is
the linear part of F −1 .
Lemma 2.2. Let F ∈ GAn (C) be a linearizable polynomial map such that F ∗ (0) = 0. Then, F is conjugate to its
linear part.
Proof. Let φ ∈ GAn (C) and L ∈ GLn (C) be such that φ −1 F φ = L.

Consider an affine map α such that α∗ (0) = φ∗−1 (0). One can easily check that the map φ α is such
that (φ α )∗ (0) = 0 and (φ α )−1 F (φ α ) = α −1 L α ∈ GLn (C).
Now, accordingly with the previous lemma, let L 1 , L 2 ∈ GLn (C) be the linear parts of F and φ α
respectively. We have L 2−1 L 1 L 2 = α −1 L α . Thus, L 1 and L are conjugate, and so F and L 1 are conjugate
too. 2
2.3. Noncommuting derivations forming a Lie algebra
It is well known that any two-dimensional Lie algebra over C which is non-abelian is essentially
the Lie algebra C X + CY where [ X , Y ] = X . This Lie algebra turns up in this section as the sub Lie
algebra of DERn (C) generated by two derivations D , E satisfying [ E , D ] = D.
Lemma 2.3. Let D , E be derivations, E ∈ LFDn (C), such that [ E , D ] = α D where α ∈ C. Then
exp(β E ) D = e α β D exp(β E )
for any β ∈ C.
The assumption E ∈ LFDn (C) is only here to make sure that exp(β E ) is well-defined. However, if
one interprets β as a variable in the ring C[n] [[β]], this assumption is not necessary.
Proof. One can either apply the supposedly well-known formulae

exp( A ) B exp(− A ) = exp [ A , −] ◦ B
where A , B are elements of a Lie algebra, or give a direct proof. We do the latter:
By hypothesis, we have E D = D (α I + E ). Thus, we obtain by induction that E k D = D (α I + E )k for
each natural number k.
Now, we can compute directly:
+∞
βk
k
exp(β E ) D = E D
k!
k=0
+∞
βk
=D (α I + E )k
k!
k=0
+∞ k

βk k i k −i
=D (α I ) E
k! i
k=0 i =0
+∞ k
(α β I )i (β E )k−i
=D
i! (k − i )!
k=0 i =0

= D exp(α β I ) exp(β E )
= e α β D exp(β E ). 2
Corollary 2.4. Let D , E ∈ LFDn (C) and suppose [ D , E ] = α D where α ∈ C. Then for any β, λ ∈ C we have

exp(β E ) exp(λ D ) = exp e α β λ D exp(β E ).
In particular, if α β ∈ 2π i Z then exp(β E ) and exp(λ D ) commute for each λ ∈ C.
Proof. Follows from Lemma 2.3, which one can use to show that
i
exp(β E ) D i = e α β D i exp(β E ). 2
Corollary 2.5. Let D , E ∈ LFDn (C) and suppose [ D , E ] = α D where α ∈ C. Then for any β, λ ∈ C,
exp(β E ) exp(λ D ) is conjugate to exp(β E ) as long as α β ∈
/ 2π i Z. In particular,

exp(−μ D ) exp(β E ) exp(λ D ) exp(μ D ) = exp(β E )
where μ = λ(e −α β − 1)−1 .
Proof. By Corollary 2.4, replacing λ by −e −α β μ, we get

exp(β E ) exp −e −α β μ D = exp(−μ D ) exp(β E ).
This means that

exp(−μ D ) exp(β E ) exp(λ D ) exp(μ D ) = exp(β E ) exp −e −α β μ + λ + μ D = exp(β E )
since −e −α β μ + λ + μ = 0. 2
2.4. Linearizing exponents of homogeneous derivations
As an application of the previous section we will show how to shift-linearize exponents of homo-
geneous derivations.
A grading deg on C[n] is called monomial if each monomial (or equivalently, each variable X i ) is
homogeneous. It is the typical grading one puts on C[n] : one assigns weights to the variables X i . In
fact, let us state
w i := deg( X i )
for this article. A homogeneous derivation is a derivation that sends homogeneous elements to homo-
geneous elements—in this article, homogeneous w.r.t. some monomial grading. It is not too difficult to
check that there exists a unique k such that a homogeneous element of degree d is sent to a homo-
geneous element of degree d + k or to the zero element. We say that D is homogeneous of degree k.
(Above, we did not specify in which set w i , d, k are. Typical is to have them in N, Z, or even R,
and that is what we think of in this article. It is however possible to choose a grading which takes
values in a group, i.e. a group grading. The above explanation makes sense for this.)
n
For this section, define the derivation associated to deg as E := E deg := i =1 w i X i ∂ X i . (E stands
for Euler derivation.) The goal of this section is to prove the following theorem.
Theorem 2.6. If D ∈ LFDn (C) is homogeneous of degree k = 0 w.r.t. a monomial grading, then exp( D ) is
shifted linearizable.
Proof. Follows immediately from Corollary 2.5 and Lemma 2.7 below, and the observation that exp( E )
is a linear map: the diagonal map (e w 1 X 1 , . . . , e w n X n ). 2
Lemma 2.7. Let D be a homogeneous derivation of degree k with respect to a monomial grading deg. Then
[ E deg , D ] = kD. In particular, if k = 0, then D and E deg commute.
v v
Proof. Let M := X 1 1 · · · Xn n (v i ∈ N) be an arbitrary monomial of degree d. Then deg( D ( M )) = d + k,
n
and E ( M ) = i =1 v i w i M = dM. Similarly E ( D M ) = (d + k) D ( M ). Now one can see that

[ E , D ]( M ) = E D ( M ) − D E ( M ) = (d + k) D M − D (dM ) = kD ( M ).
Thus, [ E , D ] = kD. 2
3. When is the Nagata automorphism shifted linearizable?
3.1. Using gradings
For the rest of this section, we let D := Δδ be the derivation which defines the Nagata auto-
morphism: N = exp( D ). This derivation is homogeneous to several monomial gradings. The set of
monomial gradings form a vector space (for if deg1 , deg2 are the associated degree functions, then
deg1 + deg2 and c deg1 where c ∈ C are degree functions associated to a grading too).
Let us explain how we find all monomial gradings for which D is homogeneous. More details on
such procedure one can find in [16] and pp. 228–234 of [6], where it is explained how to do this
to prove that Robert’s derivation is a counterexample to Hilbert’s 14th problem. First, notice that the
variables X , Y , Z are homogeneous, lets say of degree s, t, u, respectively. These values determine the
degree function deg completely. Now we need to satisfy the following two requirements:
(1) D ( X ), D (Y ), D ( Z ) all are homogeneous,

(2) deg( D ( X )) − deg( X ) = deg( D (Y )) − deg(Y ).
(This condition comes from the fact that there should be a constant d, the degree of D, for which we
have: if H is homogeneous of degree n, then D ( H ) is homogeneous of degree n + d or D ( H ) = 0.)
From D ( X ), D (Y ) homogeneous we derive that Δ = X Z + Y 2 is homogeneous, and thus that
s + u = 2t. Now (1) is satisfied. From (2) we get that s − (t + 2t ) = t − (u + 2t ) yields exactly the same
equation s + u = 2t. Thus [deg( X ), deg(Y ), deg( Z )] = [s, t , 2t − s] and the derivation is of degree 3t − s.
The degree function is associated with the (semisimple) derivation E := s X ∂ X + tY ∂Y + (2t − s) Z ∂ Z
and the diagonal linear map exp( E ) := (e s X , et Y , e 2t −s Z ).
Thus the set of gradings for which the derivation D = Δδ is homogeneous, is two-dimensional.
A possible basis is {deg1 , deg2 } where
deg1 (( X , Y , Z )) = (1, 0, −1),
deg2 (( X , Y , Z )) = (0, 1, 2).
The degree function deg1 corresponds to the (semisimple) derivation E 1 := X ∂ X − Z ∂ Z , where deg2
corresponds to E 2 := Y ∂Y + 2Z ∂ Z . Any degree function deg = s deg1 +t deg2 (s, t ∈ C) which is a linear
combination of deg1 , deg2 corresponds to E := sE 1 + t E 2 . The linear map corresponding to the linear
combination sE 1 + t E 2 is L s,t := (e s X , et Y , e 2t −s Z ). The set of these maps is exactly the set

L := L (a,b,c) = (a X , bY , c Z ) ac = b2 , abc = 0 .
Thus we have proven the following lemma.
Lemma 3.1. D is of degree 0 with respect to deg = s deg1 +t deg2 if and only if s = 3t.
Definition 3.2. Let us define L b := (b3 X , bY , b−1 Z ), and L0 as the set { L b | b = 0}. Note that

L0 := (a X , bY , c Z ) ac = b2 , bc = 1 .
3.2. Explicit formulae for shifted linearizableness of the Nagata map
One can use Corollary 2.5, Theorem 2.6, and results of the previous section, to immediately get
formulas for many linear maps L (a,b,c ) ∈ L which satisfy L (a,b,c ) N is linearizable. However, let us give
the following formulas, which are slightly more elegant, and can be easily checked directly. Moreover,
they work for any field k.
The following formulas can be easily checked:
• L− 1 −1
(a,b,c ) D L (a,b,c ) = (bc ) D, which implies
• L− 1 −1 −1
(a,b,c ) exp(λ D ) L (a,b,c ) = exp(b c λ D ), which implies
−1 c −1 λ
• N λ L (a,b,c) = L (a,b,c) N b .
The first two equations are not true over all fields, but the last one is. (Since N is defined and
invertible over Z, it makes sense over any field.) Using the last equation, the following is easy:
Theorem 3.3. Let a, b, c ∈ k∗ , ac = b2 , and bc = 1. Then L (a,b,c ) N λ is conjugate to L (a,b,c ) . More precisely,
choosing μ = bc λ(1 − bc )−1 , we have

N −μ L (a,b,c ) N λ N μ = L (a,b,c ) .
The particular case that L (a,b,c ) is a multiple of the identity, gives the formulae for s ∈ k∗ :
− s2 λ s2 λ
N 1− s 2 sN λ N 1−s2 = sI .
This gives the formula for s = 2, λ = 1 from the introduction. In the same introduction it was an-
nounced that we can linearize the maps sN for any s = 1, −1, which indeed follows from this.
Remark 3.4. The maps L (a,b,c ) N as in Theorem 3.3, are non-tame (provided char(k) = 0) but lineariz-
able (and in particular, tamizable).
3.3. The non-linearizable case
We will now consider what happens if the grading of the previous section is such that D is
homogeneous of degree 0. By Lemma 2.7 this means that E commutes with D, and hence also
exp( E ) commutes with exp( D ). By Lemma 3.1 and Definition 3.2, we can say exp( E ) ∈ L0 , i.e.
exp( E ) = L b = (b3 X , bY , b−1 Z ) for some b ∈ C∗ . Now there are several ways of showing that L b N λ
is non-linearizable, we will use invariants.
Definition 3.5. Let ϕ ∈ GAn (C), μ ∈ C. Then Eμ (ϕ ) := { p ∈ C[ X 1 , . . . , X n ] | ϕ ( p ) = μ p } is defined as

the eigenspace of ϕ with respect to μ.
If L b N is linearizable, it will be linearizable to its linear part which is L b (see Lemma 2.2). We will
show that E1 ( L b N λ ) and E1 ( L b ) are so different that they contradict the following property.
Lemma 3.6. If ϕ , ϕ̃ ∈ GAn (C) are conjugate (i.e. there exists σ ∈ GAn (C) such that ϕ̃ = σ −1 ϕσ ) then Eμ (ϕ )
and Eμ (ϕ̃ ) are isomorphic (in fact, Eμ (ϕ̃ ) = σ −1 (Eμ (ϕ ))).
Proof.
p ∈ Eμ (ϕ ) ⇐⇒ ϕ ( p) = μ p
⇐⇒ ϕσ σ −1 ( p ) = μ p
⇐⇒ σ −1 ϕσ σ −1 ( p ) = μσ −1 ( p )

⇐⇒ σ −1 ( p ) ∈ Eμ σ −1 ϕσ . 2
Lemma 3.7. Let b ∈ C∗ be no root of unity, λ ∈ C∗ . Then L b N λ ( p ) = p for some p ∈ C[ X , Y , Z ] if and only if
p ∈ C[ Z 2 Δ].
Corollary 3.8. L b N λ is non-linearizable for any b, λ ∈ C∗ .
Proof of Lemma 3.7. Give weights w ( X ) = 3, w (Y ) = 1, w ( Z ) = −1. This makes A := C[ X , Y , Z ] into

a graded algebra ⊕n∈Z A n . D and L b are homogeneous: L b ( A n ) = A n and D ( A n ) ⊆ A n . Because of
the latter, N λ is homogeneous too: N λ ( A n ) ⊆ A n (actually “=” since it is an automorphism). Hence
L b N λ ( A n ) = A n . For L b we have L b ( p ) = bn p if p ∈ A n .
We want to find all p such that L b N λ ( p ) = p. It suffices to classify all such p which are homoge-
neous. Let n = deg( p ). It now must hold that N λ ( p ) = L b−1 ( p ) = b−n p. Because of Lemma 1.1, we have
b−n = 1 and p ∈ ker Δδ . Hence, since b is no root of unity we get n = 0, and so p ∈ ker Δδ ∩ A 0 =
C[Δ, Z ] ∩ A 0 . Since Δ ∈ A 2 and Z ∈ A −1 , we get that p ∈ C[Δ Z 2 ]. It is easy to check that such p
indeed satisfy L b N λ ( p ) = p. 2
Proof of Corollary 3.8. Assume L b N λ is linearizable. We split the proof in two cases.
First assume that bm = 1 for some m ∈ N∗ . There exists some ϕ ∈ GA3 (C) such that
ϕ −1 L b N λ ϕ = L b . Thus I = ( L b )m = (ϕ −1 L b N λ ϕ )m = ϕ −1 N mλ ϕ , since L b = exp( E ) commutes with
N λ = exp(λ D ). So N mλ must be the identity, which implies that m = 0 and leads to a contradic-
tion.
So b is not a root of unity. Then, by Lemma 3.7, E1 ( L b N λ ) is isomorphic to C[Δ Z 2 ]. By Lemma 3.6,
we must have that E1 ( L b N λ ) is isomorphic to E1 ( L b ). However, their transcendence degrees differ,
since X Z 3 , Y Z ∈ E1 ( L b ). 2
4. Generators of GAn (C) and conjectures
4.1. Tamizable automorphisms
Recall that a polynomial automorphism F ∈ GAn (k) is called tame if it belongs to the subgroup
Tn (k) ⊆ GAn (k) generated by the affine subgroup and by En (k), where En (k) denotes the subgroup
generated by the elementary maps, i.e. the maps of the form
F = ( X 1 , . . . , X i −1 , X i + a, X i +1 , . . . , X n )
for some polynomial a ∈ k[ X 1 , . . . , X̂ i , . . . , X n ] and some 1 i n.

The following definition and the Problems 1 and 2 were given to us by A. Dubouloz.
Definition 4.1. A polynomial automorphism ϕ is called tamizable if there exists a polynomial auto-
morphism σ such that σ −1 ϕσ is tame.
Now let us repeat the conjectures from the introduction.
Problem 1. Is N tamizable? (Is every automorphism of C[3] tamizable?)
Problem 2. Is N tamizable by conjugation of an element of GA2 (C[ Z ])?
Connected to this, we also mention the following problem, which we took from [9, p. 120]:
Problem 3. Is every tame Ga -action on C3 conjugate to a triangular action?
Note that the Problems 1 and 3 cannot both be true, since Bass has proved in [3] that N cannot
be conjugate to a triangular automorphism of C[3] .
4.2. Known conjectures
Since the “tame generators conjecture” (which hardly anyone believed because of the automor-
phism N) was disproved by Umirbaev–Shestakov in [18,19] (and also before this feat was accom-
plished), several new conjectures have been made of “understandable” sets which could generate all
of GAn (C) for any n. We will mention several of them.
Conjecture 1. Let k be a field of characteristic zero. Then GAn (k) = GLNDn (k), which is defined as e LNDn (k) ,
GLn (k).
Conjecture 2. GAn (C) = GLFDn (C), which is defined as e LFDn (C) .
For k = C, Conjecture 2 is different than Conjecture 1, as GLNDn (C) ⊆ GLFDn (C) but it is not clear
if all exponents of for example semisimple derivations are in the previous set. In fact, in our opinion,
Conjecture 2 is more natural, as it is obvious that GLFDn (C) is a normal subgroup, but we do not
know if the subgroup GLNDn (C) is normal.
Another one is the following, from [11] (where it is stated only for k = C).
Conjecture 3. Let k be a field. GAn (k) = GLFn (k), where GLFn (k) is the group generated by all locally finite
polynomial automorphisms (which are polynomial automorphisms F for which the sequence {deg( F n )}n∈N is
bounded).
The subgroup GLFn (k) is normal for any field k: If F ∈ GLFn (C), then the sequence
{deg(ϕ −1 F m ϕ )}m∈N is bounded by the bounded sequence {deg(ϕ −1 ) deg( F m ) deg(ϕ )}m∈N .
Then there is the following conjecture, which to our knowledge originates from Shpilrain in [12,
Problem 2, p. 16] (there stated for k = C):
Conjecture 4. GAn (k) = GSHPn (k), where GSHPn (k) = GAn−1 (k[ X n ]), Affn (k), interpreting GAn−1 (k[ X n ])
as the automorphisms in GAn (k) which fix the last variable.
He suggests immediately that this conjecture may have counterexamples in dimension 3 of the
form exp( D ) where D ∈ LND3 (C) which does not have coordinates in its kernel, as constructed by
G. Freudenburg in [10]. Also, it is not clear if GSHPn (k) is a normal subgroup of GA3 (k).
4.3. The group GLINn (k)
Let us denote by Linn (k) the set of linearizable polynomial automorphisms. We define GLINn (k) :=
Linn (k) as the group generated by the linearizable automorphisms. This is by construction the small-
est normal subgroup of GAn (k) containing GLn (k).
Lemma 4.2. If k = F2 , then GLINn (k) contains Tn (k).
Proof. It suffices to show the lemma for an elementary map E f := ( X 1 + f , X 2 , . . . , X n ) where f ∈

k[ X 2 , . . . , X n ]. Let a ∈ k, a = 0, 1 (hence the requirement!). Define L := (a X 1 , X 2 , . . . , X n ), and b :=
(1 − a)−1 . The result follows since E f = L −1 ( E −bf L E bf ). 2
Remark 4.3. The first author will show in a future preprint that ( X + Y 3 , Y ) ∈ T2 (F2 ) is not in
GLIN2 (F2 ).
Corollary 4.4. If k = F2 , then GLINn (k) is the smallest normal subgroup of GAn (k) containing Tn (k).
In light of this lemma, and the result of Theorem 3.3 (being N ∈ GLINn (C)), it is natural to pose
the following (as far as we know, new) conjecture.
Conjecture 5. GLINn (k) = GAn (k) (if k = F2 ).
For k = F2 , one might replace GLINn (k) by the smallest normal subgroup of GAn (k) containing
Tn (k). We remark that for k = C we have the following chain of inclusions:
GLINn (C)
⊆
Tn (C) GLFDn (C) ⊆ GLFn (C) ⊆ GAn (C).
⊆
GLNDn (C)
Any inequality or equality in this chain would be very interesting. (The set GSHPn (C) is sort of sepa-
rate.) Remark that GLFDn , GLFn and GLINn are all normal, only the latter two can be defined over any
field.
Let us recall the following conjecture from [15,17].
Conjecture 6. Let F ∈ GAn (Fq ). If q is even and q = 2, then only half (the even ones) of the bijections of
(Fq )n → (Fq )n are given by maps in GAn (Fq ).
Here, we say that a bijection of (Fq )n is even, if it is even if seen as an element of the permutation
group on qn elements. In [15, Theorem 2.3], it is concluded that the tame automorphisms over Fq give
all bijections in case q is odd or q = 2, and only the even bijections in case q = 4, 8, 16, . . . .
Remark 4.5. If Conjecture 6 would not be true for some q = 2m , m 2, this would give a ridiculously
simple counterexample to the (already rejected) “tame generators problem” for Fq . Also, it will imply
that Conjecture 4 is not true, and the smallest normal subgroup of GAn (k) containing Tn (k) does not
equals GAn (k) (and en passant Conjecture 5 is not true for k = Fq ).
The remark follows from the fact that any conjugate of an even bijection is again even, and from
the fact that any F = ( F 1 ( X , Y , Z ), F 2 ( X , Y , Z ), Z ) ∈ GA2 (F2m [ Z ]), m 2, is even: fix Z = a ∈ F2m , and
the map F a := ( F 1 ( X , Y , a), F 2 ( X , Y , a), a) is a tame map on F22m × {a} by Jung–van der Kulk theorem
(and hence even because of Theorem 2.3 in [15]).
Acknowledgment
The authors would like to thank the staff of the Mathematisches Forschungsinstitut Oberwolfach
for their hospitality.
References
[1] AMS featured article review by G.H. Meisters on the paper [4], http://www.ams.org/mathscinet-getitem?mr=1483974.
[2] V.I. Arnold, Geometrical Methods in the Theory of Ordinary Differential Equations, Grundlehren Math. Wiss. (Fundamental
Principles of Mathematical Sciences), vol. 250, Springer-Verlag, New York, 1988.
[3] H. Bass, A nontriangular action of G a on A 3 , J. Pure Appl. Algebra 33 (1) (1984) 1–5.
[4] A. Cima, A. van den Essen, A. Gasull, E. Hubbers, F. Mañosas, A polynomial counterexample to the Markus–Yamabe conjec-
ture, Adv. Math. 131 (2) (1997) 453–457.
[5] B. Deng, G.H. Meisters, G. Zampieri, Conjugation for polynomial mappings, Z. Angew. Math. Phys. 46 (6) (1995) 872–882.
[6] A. van den Essen, Polynomial Automorphisms and the Jacobian Conjecture, Progr. Math., vol. 190, Birkhäuser, 2000.
[7] A. van den Essen, A counterexample to a conjecture of Meisters, in: Automorphisms of Affine Spaces, Curaçao, 1994, Kluwer
Academic, Dordrecht, 1995, pp. 231–233.
[8] A. van den Essen, E. Hubbers, Polynomial maps with strongly nilpotent Jacobian matrix and the Jacobian conjecture, Linear
Algebra Appl. 247 (1996) 121–132.
[9] G. Freudenburg, Algebraic Theory of Locally Nilpotent Derivations. Invariant Theory and Algebraic Transformation Groups,
VII, Encyclopaedia Math. Sci., vol. 136, Springer-Verlag, Berlin, 2006.
[10] G. Freudenburg, Local slice construction in k[ X , Y , Z ], Osaka J. Math. 34 (1997) 757–767.
[11] J.-Ph. Furter, S. Maubach, Locally finite polynomial endomorphisms, J. Pure Appl. Algebra 211 (2) (2007).
[12] J. Guitierrez, V. Shpilrain, J.-T. Yu, Affine algebraic geometry, in: Contemp. Math., vol. 369, 2004, pp. 1–30.
[13] H. Jung, Über ganze birationale Transformationen der Ebene, J. Reine Angew. Math. 184 (1942) 161–174 (in German).
[14] W. van der Kulk, On polynomial rings in two variables, Nieuw Arch. Wiskd. (3) 1 (1953) 33–41.
[15] S. Maubach, Polynomial automorphisms over finite fields, Serdica Math. J. 27 (4) (2001) 343–350.
[16] S. Maubach, An algorithm to compute the kernel of a derivation up to a certain degree, J. Symbolic Comput. 29 (6) (2000)
959–970.
[17] S. Maubach, A problem on polynomial maps over finite fields, arXiv: 0802.0630v1, unpublished.
[18] I. Shestakov, U. Umirbaev, The tame and the wild automorphisms of polynomial rings in three variables, J. Amer. Math.
Soc. 17 (1) (2004) 197–227.
[19] I. Shestakov, U. Umirbaev, Poisson brackets and two-generated subalgebras of rings of polynomials, J. Amer. Math.
Soc. 17 (1) (2004) 181–196.
Journal of Algebra
Lie properties of symmetric elements in group rings

A. Giambruno a,1 , C. Polcino Milies b,2 , Sudarshan K. Sehgal c,∗,3
a
Dipartimento di Matematica, Università di Palermo, Via Archirafi 34, 90123 Palermo, Italy
b
Instituto de Matemática e Estatística, Universidade de São Paulo, Caixa Postal 66281, CEP-05315-970, São Paulo, Brazil
c
Department of Mathematical and Statistical Sciences, University of Alberta, Edmonton, Canada T6G 2G1
Article history: Let ∗ be an involution of a group G extended linearly to the group

Received 3 June 2008 algebra K G. We prove that if G contains no 2-elements and K
Available online 29 November 2008 is a field of characteristic p = 2, then the ∗-symmetric elements
Communicated by Michel Van den Bergh
of K G are Lie nilpotent (Lie n-Engel) if and only if K G is Lie
Keywords:
nilpotent (Lie n-Engel).
Group algebra © 2008 Elsevier Inc. All rights reserved.
Lie nilpotent
Lie n-Engel
Symmetric element
1. Introduction
Let R be a ring with an involution ∗ . Let R + = {r ∈ R | r ∗ = r } be the set of symmetric elements

of R under ∗ and R − = {r ∈ R | r ∗ = −r } the set of skew symmetrics. A general question of interest
is which properties of R + or R − can be lifted to R (see [10]). For example, a classical result of
Amitsur [1] states that if R + (or R − ) satisfies a polynomial identity, then so does R.
Group rings are naturally endowed with an involution; the one obtained as a linear extension
of the involution of G given by g → g −1 . We shall refer to this as the classical involution. For this
particular involution, Giambruno and Sehgal [5] classified group algebras K G of groups with no 2-
elements such that ( K G )+ is Lie nilpotent and G. Lee completed this work [12]. The implications of
commutativity of ( K G )+ and ( K G )− have also been investigated [2,3].
Recently, there has been a surge of activity in studying more general involutions of K G; namely,
the maps obtained from arbitrary involutions of G, extended linearly to K G. Properties of ( K G )+ and
*
E-mail addresses: agiambr@unipa.it (A. Giambruno), polcino@ime.usp.br (C. Polcino Milies), s.sehgal@ualberta.ca
(S.K. Sehgal).
1
Partially supported by MIUR of Italy.
2
Partially supported by FAPESP, Proc. 2005/60411-8 and CNPq., Proc. 300243/79-0(RN) of Brazil.
3
Supported by NSERC of Canada.
doi:10.1016/j.jalgebra.2008.09.041
A. Giambruno et al. / Journal of Algebra 321 (2009) 890–902 891
( K G )− were considered in [4,11] and, recently, Gonçalves and Passman considered the existence of
bicyclic units u in the integral group rings such that the group u , u ∗ is free [8]. Marciniak and Sehgal
had proved that, with respect to the classical involution, u , u ∗ is always free if u = 1 [14]. Also,
Gonçalves and Passman constructed free pairs of unitary units in group algebras [7]. Still another type
of involution has been of interest, the oriented involutions, which are linear extensions of involutions
of G, twisted by a homomorphism G → {±1}. The latter were introduced by Novikov [15], in the
context of K -theory.
Throughout this paper ∗ will denote an involution of K G obtained as a linear extension of an
involution of G. We prove the following.
Theorem A. Let G be a group with no 2-elements and K a field of characteristic p = 2. Then, ( K G )+ is Lie
n-Engel if and only if K G is Lie m-Engel.
Theorem B. Let G be a group with no 2-elements and K a field of characteristic p = 2. Then, ( K G )+ is Lie
nilpotent if and only if K G is Lie nilpotent.
2. Some basic facts and notations
We collect important facts for use in later sections and set up some notation. For a given prime
integer p, an element x ∈ G will be called a p-element if its order is a power of p. We write
P = {x ∈ G | x is a p-element},

Q = x ∈ G xq = 1, for some integer q, (2p , q) = 1 ,

G+ = g ∈ G g∗ = g .
The following are some basic results. We recall that a group G is said to be p-abelian if G , the
commutator group of G, is a finite p-group.
Theorem 2.1. (See [19, Theorem V.6.1].) Let K be a field of characteristic p > 0. Then K G is Lie n-Engel if and
only if G is nilpotent and contains a normal p-abelian subgroup A such that G / A is a finite p-group.
Theorem 2.2. (See Passi, Passman and Sehgal [16].) The group algebra K G is Lie nilpotent if and only if G is
nilpotent and p-abelian.
Theorem 2.3. (See Giambruno and Sehgal [5].) Assume char( K ) = 2 and that G contains no 2-elements. If,
with respect to the classical involution, ( K G )+ is Lie nilpotent then K G is Lie nilpotent.
Lemma 2.4. For any semiprime ring with involution R which is Lie n-Engel, with 2R = R, we have
[ R + , R + ] = 0 and R satisfies St4 , the standard identity in four variables.
Proof. We first remark that R + , being Lie n-Engel, satisfies a polynomial identity. Hence by a result
of Amitsur [1], R is a PI-ring i.e., satisfies an ordinary polynomial identity.
Consider a prime ideal P of R. If P ∗ = P , then S = ( P + P ∗ )/ P is a nonzero ideal of R / P and
a + a∗ + P = a∗ + P , for any a ∈ P . Since R + is Lie n-Engel, for any a, b ∈ P , [a∗ + P , b∗ + P , . . . ,
b∗ + P ] = [a + a∗ , b + b∗ , . . . , b + b∗ ] + P = P . Hence the prime ring S is Lie n-Engel. Since S is also a
prime PI-ring, by Posner’s theorem [6, Theorem 1.11.13] its central localization A = S ⊗ Z F is a finite
dimensional simple algebra over F , where Z = 0 is the center of S and F is the field of fractions
of Z . Moreover S and A satisfy the same polynomial identities, hence also A is Lie n-Engel. If we now
apply Wedderburn theorem and then tensor with a splitting field F of A, we obtain m × m matrices
over F being still Lie n-Engel. A direct inspection shows that m = 1 in this case. Hence A, and so S is
892 A. Giambruno et al. / Journal of Algebra 321 (2009) 890–902
commutative. Recalling that S is an ideal of the prime ring R / P we also get that R / P is commutative
and, so, [ R , R ] ⊆ P .
In case P ∗ = P , the ring R̄ = R / P is a prime PI-ring with induced involution. Moreover since
2R = R, char R̄ = 2 and the symmetric elements of R̄ are Lie n-Engel. By Posner’s theorem the cen-
tral localization A of R̄ is a finite dimensional simple algebra with induced involution and A + is
Lie n-Engel. After tensoring with the algebraic closure F of the center of A, we obtain M m ( F ),
the algebra of m × m matrices over F with induced involution. Moreover M m ( F )+ is Lie n-Engel.
By [6, Theorem 3.6.8] we may assume that the involution on M m ( F ) is either the transpose or the
symplectic involution and a direct inspection on M m ( F ) shows that the Lie n-Engel property forces
[ Mm ( F )+ , Mm ( F )+ ] = 0 and m 2. It follows that R̄ satisfies
St4 and [ R̄ + , R̄ + ] = 0.

The outcome of the above is that St4 (r1 , . . . , r4 ) ∈ P , for all r1 , . . . , r4 in R, and [ R + , R + ] ⊂ P

where the intersection runs over all prime ideals of R. Since R is semiprime, P = 0 and the proof
is complete. 2
We remind the reader that a group G is called LC if it is not commutative and for every pair of
elements g , h ∈ G we have that gh = hg if and only if either g or h or gh is central in G. The LC
groups with a unique nonidentity commutator are precisely those groups G with center Z (G ) such
that G /Z (G ) ∼
= C 2 × C 2 , a direct product of two cyclic groups of order 2 (see [9, Proposition III.3.6]).
We shall need the following.
Theorem 2.5. (See Jespers and Ruiz [11].) Let K be a field of characteristic different from 2. Then ( K G )+ is
commutative if and only if G is either abelian or an LC group with a unique nonidentity commutator, which is
of order 2.
The next lemma has also been proved by G. Lee [13] in a different manner.
Lemma 2.6. Let K be a field of characteristic p > 2, G a finite group and J the Jacobson radical of K G. Suppose
that K G / J is isomorphic to a direct sum of simple algebras of dimension at most four over their center. Then
P is a subgroup.
Proof. First we observe that if we extend the base field K to its algebraic closure K̄ , then K̄ G / J̄
still satisfies the hypothesis. Actually K̄ G / J̄ is isomorphic to a direct sum of copies of K̄ and 2 × 2
matrices over K̄ . Hence we may assume that K is algebraically closed.
Next we claim that the hypothesis is inherited by subgroups and homomorphic images of G.
In fact, by the Amitsur–Levitski theorem, K G / J satisfies St4 , the standard identity of degree four.
Moreover, as K G is a finite dimensional algebra, J is nilpotent, say J k = 0. But then St4k is a poly-
nomial identity of K G. Let now H be a subgroup of G. As a subalgebra of K G, K H and also
K H / J ( K H ) still satisfies the polynomial identity. Now, K H / J ( K H ), being semisimple decomposes
as K H / J ( K H ) = M n1 ( K ) ⊕ · · · ⊕ M nt ( K ) and each M ni ( K ) satisfies St4k . Being a prime algebra, it fol-
lows that each M ni ( K ) satisfies St4 . By the Amitsur–Levitski theorem this implies that ni 2. Hence
K H / J ( K H ) has the desired decomposition. A similar proof holds for homomorphic images of G.
Let g , h ∈ G be p-elements and let H be the subgroup they generate. Our aim is to show that gh
is a p-element. Since the hypothesis is inherited by subgroups, without loss of generality we may
assume that G = H .
In case every irreducible representation of G is of degree one, i.e., K G / J is isomorphic to copies
of K , then K G / J is commutative. Hence (G , G ) ⊆ J where (G , G ) is the kernel of the natural
projection K G → K (G /G ). Since J is nilpotent, (G ) is nilpotent and, so, G is a p-group. This says
that the p-elements of G form a subgroup and gh is a p-element, as desired.
Therefore we may assume that G = g , h has at least one irreducible representation of degree
two. Under these hypotheses we shall reach a contradiction.
Let V be a 2-dimensional vector space over K on which G acts irreducibly. By taking the quotient
with the kernel T of the corresponding representation, we may also assume that G acts faithfully
on V (notice that G / T still satisfies the hypotheses of the lemma). Define the subspaces

V g = v ∈ V ( g − 1) v = 0 and V h = v ∈ V (h − 1) v = 0 .
Thus g − 1 ∈ End( V ) is such that Im( g − 1) = ( g − 1) V and ker( g − 1) = V g . This says that
n n
dim V − dim V g = dim( g − 1) V . Now, since g p = 1, for some n, ( g − 1) p = 0; since dim V = 2,
we get ( g − 1) = 0 and, so, ( g − 1) V = 0. It follows that ( g − 1) V ⊆ V g and by the above,
2 2
2 − dim V g dim V g . Thus dim V g 1. Similarly dim V h 1.

Suppose that V g ∩ V h = 0. Since ( g − 1) V g = 0, g acts trivially on V g and, so on V g ∩ V h . Simi-
larly also h acts trivially on V g ∩ V h . This implies that G = g , h acts trivially on V g ∩ V h and this
contradicts the fact that G acts faithfully on V . Thus V g ∩ V h = 0 and V = V g ⊕ V h .
Notice that g − 1 : V h → V g maps V h into V g . Also, if v ∈ ker( g − 1), then (h − 1) v = 0 and
( g − 1) v = 0 implies that v ∈ V g ∩ V h = 0. Thus g − 1 is an isomorphism. Similarly h − 1 : V g → V h is
an isomorphism.
Choose v ∈ V h , v = 0. Then ( g − 1) v ∈ V g and is nonzero. This says that {( g − 1) v , v } is a basis
1 1 1 0
of V . In this basis g and h have matrices A g = and A h = , respectively, where α ∈ K is
01 α1
nonzero.
Now, the matrices A g and A h generate the group SL(2, q), for some q a power of p. It follows
that the hypotheses of the lemma hold for SL(2, q) and also for its subgroup SL(2, p ). But it is known
that SL(2, p ) has irreducible representations of degree 1, 2, . . . , p and this is a contradiction since
char K > 2. 2
Lemma 2.7. Let G be a finite group, K a field of characteristic p > 2 and ∗ an involution on G. If ( K G )+ is Lie
n-Engel then P is a subgroup.
Proof. Let J be the Jacobson radical of K G. Then R = K G / J is a semisimple algebra with induced
involution and R + is Lie n-Engel. By Lemma 2.4 R satisfies St4 . Hence if we write K G / J as a sum
of simple algebras A i , each A i satisfies St4 . Since any simple algebra of dimension m2 over its center
does not satisfy any identity of degree less than 2m, we deduce that R is isomorphic to a direct sum
of simple algebras of dimension at most four over their center. Therefore the group G satisfies the
hypotheses of the previous lemma and P is a subgroup. 2
Lemma 2.8. Let K be a field of characteristic p > 0. If ( K G )+ is Lie n-Engel, then for every symmetric element g
n
of G, g p is central.
Proof. Let g ∈ G + . If x ∈ G + then [x, g , . . . , g ] = 0 implies xg p = g p x. So assume that x = x∗ . Then

n n
[x + x∗ , g p ] = 0 and
n

x + x∗ g p = g p x + x∗ .
n n
Hence, either xg p = g p x or xg p = g p x∗ .
n n n n
Suppose xg p = g p x∗ . Then xg p ∈ G + and by the first part (xg p ) g p = g p (xg p ). Then, we can
n n n n n n n
pn pn pn
cancel g on the right and obtain again that xg = g x, as desired. 2
Lemma 2.9. Assume A is an abelian group with no 2-elements and let ∗ : A → A be an automorphism of
order 2. Then
A2 ⊂ A1 × A2,
where A 1 = {a ∈ A | a∗ = a} and A 2 = {a ∈ A | a∗ = a−1 }.

Proof. Given b = a2 ∈ A 2 , write

−1
b = aa∗ a a∗ .
This gives the required decomposition. 2
Corollary 2.10. If A is an abelian torsion group with no 2-elements and ∗ : A → A is an automorphism of

order 2 then
A = A1 × A2,
where A 1 = {a ∈ A | a∗ = a} and A 2 = {a ∈ A | a∗ = a−1 }.
Lemma 2.11. Let G be any group, K a field of characteristic p > 2 and ∗ an involution on G. Let A be a torsion
abelian normal subgroup of G, without elements of order 2, and let x ∈ G \ A be an element such that x∗ = x−1 c
with c ∈ A. Then, there exists a symmetric element b ∈ A such that (xb)∗ = (xb)−1 .
Proof. Write A = A 1 × A 2 , with A 1 = {a ∈ A | a∗ = a} and A 2 = {a ∈ A | a∗ = a−1 } as in Corollary 2.10.

Notice that xx∗ = c is in A and is symmetric, so actually c ∈ A 1 . Also x−1 cx ∈ A 1 .
As A 1 has no elements of order 2, we can find b ∈ A 1 such that b2 = x−1 c −1 x. This means that
b−1 x−1 = bx−1 c and thus (xb)−1 = bx−1 c = (xb)∗ , as desired. 2
Theorem 2.12. Let G be a finite group of odd order, K a field of characteristic p > 2 and ∗ an involution on G.
If ( K G )+ is Lie n-Engel, then K G is Lie nilpotent.
Proof. By Lemma 2.7, P is a subgroup of G. Since (|G / P |, | P |) = 1 by the theorem of Schur–

Zassenhaus we can write G = P X with X a p -group. Since ( K G )+ is Lie n-Engel, by Lemma 2.4
and Theorem 2.5 X is abelian. It follows that G is a p-abelian group and by Theorem 2.2, in order to
complete the proof it is enough to show that G is nilpotent. Now, since P is nilpotent, it is actually
enough to prove that G / P is nilpotent.
If P = 1 we are done, by induction. Hence, we may assume that P = 1 and, thus, that P is abelian.
Suppose that there exists an invariant subgroup H = H ∗ = 1 contained in ζ , the center of G. Since
( K (G / H ))+ and ( K H )+ are both n-Engel, by induction we get that G / H and H are nilpotent. Hence,
G is nilpotent and we are done. Therefore, without loss of generality, we may assume that G contains
no central element z = 1, as z, z∗ would then give a central subgroup invariant under ∗ .
Since | X ∗ | = | X |, it follows that X ∗ is another complement to P so, by Schur–Zassenhaus, X ∗ is
conjugate to X ; i.e. X ∗ = X y for some y ∈ P . For an element x ∈ X , let x1 ∈ X be such that
x∗ = y −1 x1 y .
Then xx∗ = xy −1 x1 y = xx1 (x1 , y ) and (x1 , y ) ∈ P . It follows that (xx∗ ) p = (xx1 ) p d for some ele-
n n
ment d ∈ P . Thus, by Lemma 2.8, [(xx1 ) p , P ] = 0. Hence xx1 ∈ ζ . This implies xx1 = 1; i.e. x1 = x−1 .
n
Thus, x = y x y, for all x ∈ X . So, we can write x∗ = x−1 xy −1 x−1 y = x−1 c with c ∈ P .
∗ − 1 −1
We shall prove that, for any fixed element x ∈ X , we have that (x, P ) = 1. As P is of odd order, by
Lemma 2.9, we have P = A 1 × A 2 where A 1 = {a ∈ P | a∗ = a} and A 2 = {a ∈ P | a∗ = a−1 }.
By Lemma 2.11 there exists an element b ∈ A 1 such that (xb)∗ = (xb)−1 .
Since (xb, P ) = 1 implies (x, P ) = 1, we may assume hereafter, that x∗ = x−1 . Since ( K G )+ is Lie
n-Engel, [a + a∗ , (x + x−1 ) p ] = 0 for all a ∈ P . Hence [a + a∗ , x p + (x−1 ) p ] = 0. Notice that x p =
n n n n
− 1 pn
(x ) (mod P ) implies x 2pn
= 1 (mod P ) and thus also x = 1 (mod P ). So, x and x −1
belong
to different cosets of P in G, and we get that [a + a∗ , x p ] = 0. Since x is a p -element, we get
n
[a + a∗ , x] = 0. If xa = ax we must have a∗ x = xa and ax = xa∗ . From this we have xax−1 = a∗ and

xa∗ x−1 = a. Combining, we conclude that x2 commutes with a. Thus xa = ax for all a ∈ P .
This says that X is a normal subgroup of G, so G = P × X and G is nilpotent. 2
The following is an easy but useful observation.
Lemma 2.13. Let K be any field of characteristic different from 2 and G any group. For any element c ∈ G not
of even order, the element 1 + c ∈ K G is not a zero divisor.
Proof. We first remark that if 1 + c is a zero divisor in K G, then it is also a zero divisor in K c . If
o(c ) = ∞ then K c contains no zero divisors. So, assume that o(c ) = pk q with ( p , q) = 1, where p =
k
char( K ) > 2. Then c p has order q, (q, 2) = 1 so, in each component of the Wedderburn decomposition
k
of K c it maps to a root of unity different from −1. Consequently, 1 + c p is a unit in K c . Thus
pk pk
(1 + c ) =1+c is a unit and so 1 + c is also a unit. A similar argument holds if char( K ) = 0. 2
Remark 2.14. Let G be a finite group and K a field of characteristic p > 2 such that ( K G )+ is Lie
n-Engel. Then it follows from Lemma 2.7 that P is a subgroup, so Lemma 2.4 gives that ( K G / P )+ is
commutative and then Theorem 2.5 shows that G / P is either abelian or an LC group with a unique
nonidentity commutator, which is an element of order 2.
We shall need the following generalization of Theorem 2.12 where we do not require that |G | is
odd.
Lemma 2.15. Let G be a finite group and K of characteristic p > 2 such that ( K G )+ is Lie n-Engel. If G / P is
abelian, then G is nilpotent.
Proof. Notice that, if |G | is odd, we already know, from Theorems 2.12 and 2.2 that G is nilpotent.
We shall give a proof for arbitrary finite groups G, by induction on |G |.
If G has a central element z = 1, as before we have that z, z∗ is a central subgroup invariant
under ∗ so, by induction, G / z, z∗ is nilpotent and we are done. We show that this is always the case
by proving that, if ζ = 1 then G = 1.
Since G / P is abelian, it can be written in the form G / P = ( M / P ) × ( N / P ) where M / P is of odd
order and N / P is a 2-group. Clearly, M is invariant under ∗ , so ( K M )+ is Lie n-Engel and M is
nilpotent. Thus, we can write M = P × Q , with (| Q |, p ) = 1.
We claim that the elements of Q also commute with 2-elements. In fact, take q ∈ Q and let t be
a 2-element. Then (q, t ) ∈ Q and, since G / P is commutative, G ⊂ P so actually (q, t ) ∈ Q ∩ P = 1, as
desired. This shows that Q is central, so Q = 1. Consequently, we can write G = P T , where T is a
2-group.
Pick x ∈ T such that x2 = 1 (and thus, also (x∗ )2 = 1). Then (xx∗ ) p ∈ ζ so (xx∗ ) p = 1 and xx∗ ∈ P .
n n
Consequently, x = xc for some element c ∈ P and, x = x = c xc, c = xc x . As c = xx∗ is sym-

∗ ∗∗ ∗ ∗ − 1 −1
metric, we have cx = xc −1 . We compute
pn pn
0 = c , x + x∗ = c , x(1 + c ) ,
n n
thus c (x(1 + c )) p = (x(1 + c )) p c. As
pn
= x p (1 + c ) 1 + c −1 (1 + c ) · · · 1 + c −1 (1 + c ),
n
x(1 + c )
we get

cx p − x p c (1 + c ) 1 + c −1 (1 + c ) · · · 1 + c −1 (1 + c ) = 0
n n
and Lemma 2.13 shows that x p c = cx p , so also xc = cx. Hence c 2 = 1 so c = 1 and x∗ = x is symmet-
n n
n n
ric. By Lemma 2.8, x p ∈ ζ so x p = 1 and, as x2 = 1 we get x = 1. This shows that G = P is abelian
and G = 1 2
3. ( K G )+ Lie n-Engel
In this section, we shall assume throughout that char( K ) = p > 2 and that G has no 2-elements.
We shall also assume that ( K G )+ is Lie n-Engel. This implies that K G satisfies a ∗ -polynomial identity,
so it also satisfies a polynomial identity by a theorem of Amitsur [1]. It then follows from a theorem
of Passman [17, p. 197] that G has a normal p-abelian subgroup A of finite index. We can assume
A is ∗ invariant by replacing it by A ∩ A ∗ ; i.e. we have the following.
Remark 3.1. If K G + is Lie n-Engel, then there exists a normal subgroup A of G, which is ∗ -invariant
and such that G / A is finite and A is a finite p-group.
Proposition 3.2. P is a subgroup and G / P is abelian.
Proof. We wish to prove that if x, y ∈ P then xy ∈ P . We can assume, without loss of generality,
that G = x, y , x∗ , y ∗ . Since G is finitely generated and A is of finite index in G, we have that
A is also finitely generated. Since A is invariant under ∗ , we may factor by A and assume that
A is abelian. Then, we have A = F × T , where F is finitely generated free abelian and T is finite. We
write | T | = pm s, with ( p , s) = 1.
m m
Set A 1 = A p s = F p s and consider G / A 1 . Then G / A 1 is of finite order, say p t with ( p , t ) = 1.

Now (xy ) p is an element of order dividing t (mod A 1 ) and by Lemma 2.7 it is a p-element, so

(xy ) p = 1 (mod A 1 ).
A similar argument shows that, for any positive integer r such that (r , p ) = 1, replacing A 1 by A r1

we get that (xy ) p = 1 (mod A r1 ). Hence (xy ) p ∈ r A r1 = 1. Consequently, we have (xy ) p = 1, as
claimed.
Since ( K G / P )+ is Lie n-Engel, it follows from Lemma 2.4 that ( K G / P )+ is commutative. By Theo-
rem 2.5 then G / P is either abelian or an LC group with a unique nonidentity commutator, which is
an element of order 2. Since G / P has no 2 elements, it follows that it is abelian. 2
Corollary 3.3. T = {x ∈ G | o(x) is finite} is a subgroup and T = P × Q where Q is central in G.
Proof. By Proposition 3.2, G / P is abelian and, thus, T is a subgroup of G. By using Remark 3.1 we
see that T is locally finite.
Hence, by the finite case, it follows that T = P × Q . Further, for x ∈ G we have that (x, Q ) ⊂ P ∩ Q
as Q G and G / P is abelian. Hence (x, Q ) = 1, as claimed. 2
Proposition 3.4. Suppose that there exists a normal subgroup A of G, invariant under ∗ , such that A is abelian
and G / A is of odd order. Let x ∈ G be an element whose order, modulo A is q = 2, relatively prime to p. Then
(x, A ) = 1.
Proof. We know, from Proposition 3.2 that G / P is abelian, so G ⊂ P . Take a ∈ A; we want to show
that (x, a) = 1. Thus, we can assume that G = x, a, x∗ , a∗ is finitely generated. In this case, A is a
normal subgroup of finite index in a finitely generated group, so it is finitely generated and we can
write
A = F × T, where F is torsion free and T is finite.
m m m
Suppose | T | = pm , with (2p , ) = 1 and set A 1 = A p = F p . Since xq ∈ A we see that xqp ∈ A 1
m
so the order of x p , modulo A 1 , divides q. As G / A 1 has no 2-elements, Corollary 3.3 shows that
m
(x p , a) = 1 (mod A 1 ). Also, as xq ∈ A, we have (xq , a) = 1.
Thus, (x, a) = 1 (mod A 1 ) and, since G ⊆ P , (x, a) = 1 (mod P ∩ A 1 ). Since ( P ∩ A 1 ) =
m
( P ∩ F p ) = 1, the proposition is proved. 2
Proposition 3.5. Assume that there exists a normal subgroup A of G, invariant under ∗ , such that G / A and A
are both finite p-groups and A is abelian. Then G acts as a finite p-group of automorphisms on A .
Proof. As A is finite, normal, G does act as a finite group of automorphisms, by conjugation. We

need to show that every element x ∈ G acts as a p-element. Since G / A is a finite p-group, we may
assume that x ∈ A. It will suffice to show that

x p , A = 1 implies (x, A ) = 1 for any prime p = p . (1)
It should be mentioned that it follows from Theorems 2.12 and 2.2 that if G is finite then
G is nilpotent and thus the proposition is true in this case.

If (x p , A ) = 1 then ((x∗ ) p , A ) = 1 and thus also ((xx∗ ) p , A ) = 1. We know, by Lemma 2.8 that
∗ pn ∗
(xx ) is central. This implies that (xx , A ) = 1.
We shall handle separately the cases when p = 2 and when p = 2.

(a) Let p = q = 2.
Define B = { y ∈ A | ( y q , A ) = 1}.
It is easy to see that B is a ∗ -invariant normal subgroup containing x and x∗ and clearly B / A
is abelian. By Lemma 2.9, for any x ∈ B we can write x2 = x1 x2 where x∗1 = x1 (mod A ) and x∗2 =
x2−1 (mod A ). We shall prove separately that (x1 , A ) = 1 and (x2 , A ) = 1.
Since, x∗1 = x1 c, with c ∈ A , the group H = x1 , A is ∗ -invariant. Set ζ = ζ ( H ). The group H /ζ is
finite as x1 ∈ ζ . Thus, in this factor group we get (x1 , A ) = 1 and therefore, for any a ∈ A we have
q
−1 −q q
x1 ax1 = az, with z ∈ ζ . Consequently, a = x1 ax1 = azq , so zq = 1 and thus also z = 1. It follows that
(x1 , A ) = 1.
Now, consider x∗2 = x2−1 c, with c ∈ A . Then, by Lemma 2.11, we can modify x2 by an element of A
and assume that x∗2 = x2−1 . From the hypothesis, for any a ∈ A we have
pn pn − pn
0= x2 + x2−1 , a + a∗ = x2 + x2 , a + a∗ .
pn − pn pn
If x2 A = x2 A then x2 , A is a finite, ∗ -invariant group, and we are done. So, let x2 A =
−p n
x2 A . Consequently
pn
x2 , a + a∗ = 0.
pn pn pn pn pn pn
and, thus x2 a = ax2 (as x2 a = ax2 ), or x2 a = a∗ x2 and x2 a∗ = ax2
q q
Hence, either x2 a = ax2
n
which implies (x2 , a) = 1. Since also (x2 , A ) = 1 we conclude that (x2 , A ) = 1.
2p q
For any x ∈ B, we have shown that (x2 , A ) = 1. Since also (xq , A ) = 1, we can conclude again that
(x, A ) = 1, proving (a).
Now, let us prove case (b) when p = 2. Using Corollary 2.10, we can write A = B 1 × B 2 where
b1 = b1 and b2x = b2−1 for all b1 ∈ B 1 , b2 ∈ B 2 . We claim that B 2 = 1. In fact, take b ∈ B 2 . We have
x
xx∗ b = bxx∗ = xb−1 x∗ .
Thus
−1 −1
x∗ b x∗ = b−1 , x−1 b∗ x = b∗ .
This implies that b∗ ∈ B 2 and x∗ b∗ (x∗ )−1 = (b∗ )−1 . As ( K G )+ is Lie n-Engel, we have
pn pn pn
bb∗ , x + x∗ = 0 and so bb∗ x + x∗ = x + x∗ bb∗ .
Since x and x∗ both invert b and b∗ and because p is odd, we get
pn −1 pn
x + x∗ bb∗ = x + x∗ bb∗ .
Since x + x∗ is not a zero divisor by Lemma 2.13, we have (bb∗ )−1 = bb∗ and, as G contains no
2-elements, we conclude that bb∗ = 1, thus b∗ = b−1 .
We need still another calculation:

xb + (xb)∗ = xb + b∗ x∗ = xb + b−1 x∗ = xb + x∗ b = x + x∗ b.
Thus [(x + x∗ )b, (x + x∗ ) p ] = 0. Consequently

n
pn pn+1
x + x∗ b x + x∗ = x + x∗ b
and
p n +1 pn+1
x + x∗ b−1 = x + x∗ b.
As above, b2 = 1 and also b = 1. This proves (1) and the proposition. 2
Lemma 3.6. Assume that there exists a normal subgroup A of G, invariant under ∗ , such that A is abelian and
n
G / A is a finite p-group. Then, A p is central.
Proof. Take x ∈ G, a ∈ A. Then
pn
0 = x + x∗ , a + a∗ = x + x∗ , b + b∗ ,
where b = a p . If x A = x∗ A, then [x, b + b∗ ] = 0. Otherwise, x∗ = xc, with c ∈ A so x + x∗ = x(1 + c )

n
and [x(1 + c ), b + b∗ ] = 0. Since 1 + c is not a zero divisor, we obtain again that [x, b + b∗ ] = 0.
We claim that xb = bx. If not, xb = b∗ x and xb∗ = bx. Thus xbx−1 = b∗ , xb∗ x−1 = b and x2 bx−2 = b.
k
We have that (x2 , b) = 1. Also x p ∈ A, for some integer k; therefore (x, b) = 1. This proves our state-
ment. 2
Lemma 3.7. Assume that there exists a normal subgroup A of G, invariant under ∗ , such that A is abelian and
G / A is finite. If there exists an element x ∈ G such that x2 ∈ A and x A = x∗ A, then (x, A ) = 1.
Proof. Consider A 1 = {a ∈ A | ax = a} and A 2 = {a ∈ A | ax = a−1 }. From the hypothesis, we have that

x∗ = xc, with c ∈ A. For an element a ∈ A 1 we compute xa∗ x−1 = x∗ a∗ (x−1 )∗ = (ax )∗ = a∗ , which
shows that A 1 is ∗ -invariant. We set Ḡ = G / A 1 .
Since Lemma 2.9 shows that A 2 ⊂ A 1 × A 2 we have Ā 2 ⊂ A 2 and thus āx̄ = ā−1 , for all a ∈ A 2 .
We claim that Ā 2 = 1. Take a ∈ A 2 . As Lemma 2.9 also applies to ∗ , we shall consider separately
the cases when a∗ = a and a∗ = a−1 .
If a∗ = a we have that [a, (x + x∗ ) p ] = 0 and so a(x + x∗ ) p = (x + x∗ ) p a. This implies (x +
n n n
∗ pn −1 ∗ pn ∗
x ) a = (x + x ) a and, as Lemma 2.13 shows that (x + x ) is not a zero divisor, we get a2 = 1,
and thus also a = 1, as desired.
If a∗ = a−1 , then xa + (xa)∗ = xa + a∗ x∗ = xa + x∗ (a∗ )−1 = xa + x∗ a = (x + x∗ )a. Then, by hypothesis,
[(x + x∗ )a, (x + x∗ ) p ] = 0. Therefore,
n
p n +1 pn
x + x∗ a = x + x∗ a x + x∗ .
n +1 n +1
Since a(x + x∗ ) = (x + x∗ )a−1 we get (x + x∗ ) p a = (x + x∗ ) p a−1 .
Again, (x + x∗ ) is not a zero divisor, by Lemma 2.13, so a = a−1 which implies a = 1.

This proves our claim. Hence, we see that A 2 ⊂ A 1 so (x, a2 ) = 1 for all a ∈ A; since A is abelian,
this implies that (x, a)2 = 1 and since G contains no 2-elements, we get (x, a) = 1 for all a ∈ A. 2
Proof of Theorem A. By Remark 3.1, if char( K ) = p > 2, G has no 2-elements and ( K G )+ is Lie n-
Engel, then G contains a normal p-abelian subgroup A of finite index which is ∗ -invariant. Choose A
maximal among the groups with these properties. We shall show that G / A is a finite p-group.
First we note that, as A is a finite p-group, we may factor by A and assume that A is abelian. Set
H = G / A and let P denote the p-Sylow subgroup of H . By Remark 2.14, H / P is either abelian or an
LC group with a unique nonidentity commutator of order 2.
We shall first handle the case when H / P is abelian. Then, by Corollary 3.3, H = P × Q × T where
T is a 2-group, | Q | = q is odd and relatively prime to p and both Q and T are abelian. Thus, there
exists a subgroup L such that A L G, L / A ∼ = Q , G /L ∼ = P × T and L is ∗ -invariant.
For any x ∈ L we have that (xq , A ) = 1 and, by Proposition 3.4, also (x, A ) = 1. Thus, A is central
in L and the index of L over its center is a divisor of q. It follows, by Schur’s theorem [19, Theo-
rem I.4.2] that | L | is a divisor of q. Also L ⊂ G , so L is a p-group; thus L = 1 and L is abelian,
which implies L = A; hence Q = 1.
Now we have H = P × T so we can find a subgroup N such that A N G, G / N ∼ = P and
N/A ∼ = T.
Let N 1 = {b ∈ N | b2 ∈ A }. Take x ∈ N 1 , x = 1. Then x2 = 1 (mod A ), (x∗ )2 = 1 (mod A ) and, as
N / A is abelian, also (xx∗ )2 = 1 (mod A ). Moreover, by Lemma 2.8, (xx∗ ) p is central. It follows that
n
(xx∗ , A ) = 1. Then xx∗ , A is abelian and, as A ⊂ xx∗ , A , from the maximality of A we get xx∗ ∈ A.
Thus x∗ A = x−1 A = x A. Then, by Lemma 3.7, (x, A ) = 1 and again, from the maximality of A, it
follows that x ∈ A. We conclude that N 1 = 1, so also N = 1 and H = P , as desired.
Now we consider the remaining case, namely when H / P is LC with a unique nonidentity com-
mutator, which is of order 2. Let M be the subgroup of G containing A such that M / A = P , the
p-Sylow subgroup of G / A = H . Let z ∈ G be an element such that zM is the unique commutator of
order 2 in G / M. Then, z∗ = z (mod M ). Consider L = M , z. By the abelian case z ∈ A. Hence, we
have A M G with G / M abelian. Again from the previous case, it follows that G / A is a p-group.
To complete the proof of necessity, after Theorem 2.1, we need to prove that G is nilpotent. Since
A is a finite p-group it is nilpotent, so it will suffice to prove that G / A is nilpotent. Hence, we may
assume that A is an abelian finite p-group.
It follows from Proposition 3.5 that G acts as a finite p-groups of automorphisms on A . Hence,
by [19, Lemma V.4.1], we have that A ⊂ ζr (G ), for some positive integer r. Thus, we can assume that
A = 1 and that A is abelian.
n n
By Lemma 3.6, A p is central, so we may factor by A p and assume that A is of bounded p-power
exponent. As G / A is a finite p-group, G acts as a finite p-group of automorphisms on A, so we can
use again [19, Lemma V.4.1] to obtain that A ⊂ ζs (G ) for some positive integer s. Since G / A is a finite
p-group, it is nilpotent, and hence G is nilpotent, as desired.
The case when char( K ) = 0 is easy to see. The hypothesis implies that (Z/ p Z)G is Lie n-Engel, for
any prime integer p, so G is a p-group. Since p is arbitrary, it follows that G = 1.
The converse is trivial. 2
4. Lie nilpotency
In this section, we shall prove our second main result, namely that if G is a group with no 2-
elements, K is a field of characteristic p = 2, ( K G )+ is Lie nilpotent, then K G is Lie nilpotent.
Since Lie nilpotency implies n-Engel, for some positive integer n, we can apply Theorem A. Thus,
we can assume that p > 2 and we know that G is nilpotent and that it contains a p-abelian, nor-
mal, subgroup A, which is invariant under ∗ , such that G / A is a finite p-group. Hence, according
to Theorem 2.2, we are left only to prove that G is a finite p-group and we already know, from
Proposition 3.2 that G is a p-group.
We shall now prove a crucial special case of Theorem B.
Theorem 4.1. Let G be a group with no 2-elements and K a field of characteristic p > 2, such that ( K G )+ is Lie
nilpotent. Suppose that G contains a normal subgroup A, invariant under ∗ , such that G / A is a cyclic p-group.
Then G is a finite p-group.
Proof. Since the theorem is clearly true for finite groups, we shall assume that A is infinite.
Let x ∈ G be such that G / A = x A . It suffices to prove that the group (x, A ) ⊂ A is finite.
Since G contains no 2-elements, it will be enough to prove that (x, A 2 ) is finite. We recall that,
by Lemma 2.9, we have A 2 ⊂ A 1 × A 2 , where A 1 = {a ∈ A | a∗ = a} and A 2 = {a ∈ A | a∗ = a−1 }.
Our first aim is to reduce to the case when A 2 = 1.
So, let A 2 = 1. If A 2 ∩ ζ is infinite then, by the arguments in [5, p. 4257], K G is Lie nilpotent and
G is finite. Thus, we may assume that A 2 ∩ ζ is finite. Since A is bounded modulo the center by
Lemma 3.6, we conclude that A 2 is bounded. Write

B = (x, A 2 ) = (x, a2 ) a2 ∈ A 2 .
pm m
As 1 = (x, a2 ) = (x, a2 ) p , we see
that B is a group of bounded p-exponent. Suppose B is infinite.
Then, by [18, Theorem 4.3.5], B = i B i , an infinite direct product of cyclic groups. For an arbitrary
positive integer s, we are going to produce elements ai ∈ A 2 such that, after a possible renumbering
of the indices, (x, ai ) ∈ B i , 1 i s, so that

e = x + x∗ , a1 + a∗1 , a2 + a∗2 , . . . , as + a∗s = 0.
This will be a contradiction proving B is finite. Notice that

e = x, a1 + a∗1 , a2 + a∗2 , . . . , as + a∗s + x∗ , a1 + a∗1 , a2 + a∗2 , . . . , as + a∗s
vanishes if and only if each of the two summands vanishes as can be seen by considering the two
cases x A = x∗ A and x A = x∗ A. It will therefore suffice to find a1 , a2 , . . . , as so that

x, a1 + a∗1 , a2 + a∗2 , . . . , as + a∗s = 0.
For s = 1 we pick a1 ∈ A 2 such that 1 = (x, a1 ) ∈ B 1 . Then, it can be checked directly that [x, a1 +
a∗1 ] = 0.
Let us suppose that we already have a1 , . . . , as−1 as stated. Let N be the normal closure of
a1 , . . . , as−1 . Then N is a finite abelian group as A 2 is of bounded exponent and every element
has a finite number of conjugates. Remember that, as (x, ab) = (x, a)(x, b), every element of B is a
commutator. Thus there exists an index s so that B s ∩ N = 1. Choose as ∈ A 2 so that 1 = (x, as ) ∈ B s .
Then also (x, a2s ) = (x, as )2 = 1 and so a2s ∈
/ N.
We know already, by induction that

x, a1 + a∗1 , a2 + a∗2 , . . . , as−1 + a∗s−1 = xα = 0, α ∈ K N.
Therefore

e = x, a1 + a∗1 , a2 + a∗2 , . . . , as + a∗s

= x, as + a−
s
1
α = x a s + a−1 x −x
s − as − as α.
We claim that as N is not equal to a− 1 x −x

s N, a s N or a s N. In the first case a s ∈ N and in the second
2
(x, as ) ∈ N, both contradictions. If as N = as N, then x as x = as (mod N ), so x−2 as x2 = as (mod N ).

− x − 1 −1
Also a p-power of x commutes with as , consequently (x, as ) ∈ N, again a contradiction. Thus xas α = 0
as α = 0, and e = 0 as desired.
We have proved that B = (x, A 2 ) is finite. Let B̄ be the finite group which is the normal and
∗ closure of B. Then ( K G / B̄ )+ is Lie nilpotent. Also, the image of A 2 in G / B̄ is central. Thus, A 2 can
be assumed to be finite modulo B̄ (otherwise (G / B̄ ) is finite and so is G ). Consequently, A 2 is finite.
Factoring by the normal and ∗ closure of A 2 we can assume that A is infinite and that A = A 1 ; i.e.
a∗ = a for all a ∈ A as we wanted.
Since G is nilpotent, by induction on its class of nilpotency, we can assume that (G /ζ ) is finite.
Thus G ζ /ζ ∼
= G /ζ ∩ G is finite. If ζ ∩ G is finite, we are done. Suppose B = ( A , x) ∩ ζ is infinite. We
shall show that this leads to a contradiction, completing the proof.
m m
Again B is of bounded exponent, as 1 = (a p , x) = (a, x) p and we can write B = i B i , an infinite
direct product of cyclic groups.
We observe

x + x∗ , a = [x, a] + x∗ , a = xa 1 − (a, x) + x∗ a 1 − a, x∗ .
Let us choose ai ∈ A such that 1 = (ai , x) ∈ B i . Then

x + x∗ , a1 , . . . , as = xa1 · · · as 1 − (a1 , x) · · · 1 − (as , x) + x∗ a1 · · · as 1 − a1 , x∗ · · · 1 − as , x∗ .
This expression is not zero for any s, as can be seen by considering the two cases x A = x∗ A or
x A = x∗ A. This contradiction proves the theorem. 2
Proof of Theorem B. Since ( K G )+ is Lie nilpotent, it is also Lie n-Engel, for some positive integer n.
Thus, by Theorem A, also K G is Lie n-Engel and, by Theorem 2.1, G is nilpotent and there exists a
normal subgroup A of G such that both G / A and A are finite p-groups. By replacing A by A ∩ A ∗ ,
we can assume that A is normal and ∗ -invariant.
We wish to prove that G is finite. It is a p-group, as P is a subgroup and G / P is abelian by
Proposition 3.2.
By induction on the class of nilpotency of G we may assume that G is metabelian. Therefore,
G / A is metabelian.
If G / A is cyclic, we are done by Theorem 4.1. Let us first assume that G / A is abelian. Then G / A =
G 1 / A × G 2 / A where g 1∗ = g 1 (mod A ) and g 2∗ = g 2−1 (mod A ), for all g 1 ∈ G 1 , g 2 ∈ G 2 . Then ( g 1 , A )
is finite and so (G 1 , A ) is finite. Similarly, (G 2 , A ) is finite. Thus (G , A ) is finite. Factoring by (G , A )
we conclude that A is central and G, being central by finite, has a finite derived group.
Let us now consider the general case, namely, assume there exists a normal subgroup L of G
containing A such that both G / L and L / A are abelian and L can be assumed ∗ -invariant. Then, by the
abelian case, L is finite. Factoring by L , we may assume that L is abelian, and we are done by the
special case. Again, the converse is trivial. 2
5. Note added in proof
The classification in Theorems A and B has now been completed for all groups by G. Lee, E. Spinelli
and S.K. Sehgal in “Lie properties of symmetric elements in group rings II” which is to appear in the
Journal of Pure and Applied Algebra.
References
[1] S.A. Amitsur, Identities in rings with involution, Israel J. Math. 7 (1968) 63–68.
[2] O. Broche Cristo, Commutativity of symmetric elements in group rings, J. Group Theory 9 (2006) 673–683.
[3] O. Broche Cristo, C. Polcino Milies, Commutativity of skew symmetric elements in group rings, Proc. Edinb. Math.
Soc. (2) 50 (1) (2007) 37–47.
[4] O. Broche, E. Jespers, C. Polcino Milies, M. Ruiz, Antisymmetric elements in group rings II, J. Algebra Appl., in press.
[5] A. Giambruno, S.K. Sehgal, Lie nilpotency of group rings, Comm. Algebra 21 (1993) 4253–4261.
[6] A. Giambruno, M. Zaicev, Polynomial Identities and Asymptotic Methods, Math. Surveys Monogr., vol. 122, Amer. Math.
Soc., Providence, RI, 2005.
[7] J.Z. Gonçalves, D.S. Passman, Unitary units in group algebras, Israel J. Math. 125 (2001) 131–155.
[8] J.Z. Gonçalves, D.S. Passman, Involutions and free pairs of bicyclic units in integral group rings, preprint.
[9] E.G. Goodaire, E. Jespers, C. Polcino Milies, Alternative Loop Rings, North-Holland Math. Stud., vol. 184, Elsevier, Amsterdam,
1996.
[10] I.N. Herstein, Rings with Involution, Univ. of Chicago Press, Chicago, 1976.
[11] E. Jespers, M. Ruiz, On symmetric elements and symmetric units in group rings, Comm. Algebra 34 (2006) 727–736.
[12] G. Lee, Group rings whose symmetric elements are Lie nilpotent, Proc. Amer. Math. Soc. 127 (1999) 3153–3159.
[13] G. Lee, Groups whose irreducible representations have degree at most 2, J. Pure Appl. Algebra 199 (2005) 183–195.
[14] Z.S. Marciniak, S.K. Sehgal, Constructing free subgroups of integral group ring units, Proc. Amer. Math. Soc. 125 (1997)
1005–1009.
[15] S.P. Novikov, Algebraic construction and properties of hermitian analogues of K -theory over rings with involution from the
viewpoint of Hamiltonian formalism. Applications to differential topology and the theory of characteristic classes II, Izv.
Akad. Nauk SSSR Ser. Mat. 34 (3) (1970) 475–500; English transl. in Math. USSR Izv. 4 (3) (1970).
[16] I.B.S. Passi, D. Passman, S.K. Sehgal, Lie solvable group rings, Canad. J. Math. 25 (1973) 748–757.
[17] D.S. Passman, The Algebraic Structure of Group Rings, Wiley, New York, 1977.
[18] D.J.S. Robinson, A Course in the Theory of Groups, Springer, New York, 1982.
[19] S.K. Sehgal, Topics in Group Rings, Marcel Dekker, New York, 1978.
Journal of Algebra
A class of projectively full ideals in two-dimensional Muhly

local domains
Raymond Debremaeker
Department of Mathematics, K.U. Leuven, Celestijnenlaan 200B, 3001 Leuven, Belgium
Article history: Let ( R , M) be a regular local domain of dimension d 2 and

Received 5 June 2008 let x1 , . . . , xd be a regular system of parameters. Then, Ciuperca,
Available online 17 November 2008 Heinzer, Ratliff Jr., and Rush have proved that every ideal of the
Communicated by Luchezar L. Avramov
form I = (xn1 , x2 , . . . , xd ) R with n ∈ N+ , is projectively full.
Keywords:
If ( R , M) is a two-dimensional Muhly local domain (i.e., an inte-
Integrally closed ideal grally closed Noetherian local domain with algebraically closed
Projectively equivalent ideals residue field and the associated graded ring an integrally closed
Projectively full ideal domain), then we are able to prove a similar result for every
Rees valuation of an ideal minimal ideal basis x1 , . . . , xd of M such that x1 ∈ / rad(x2 , . . . , xd ).
1. Introduction
Let I be a regular proper ideal of a Noetherian ring R (i.e., I contains an element with zero
annihilator and I = R). An ideal J of R is projectively equivalent to I if there exist positive integers n
and m such that I n and J m have the same integral closure, i.e., I n = J m . Here the overbar “-” is used
to denote the integral closure of an ideal.
In [2], Ciuperca, Heinzer, Ratliff Jr., and Rush, introduced the notion of projectively full ideal. An
ideal I as above is called projectively full if the set P( I ) of integrally closed ideals projectively equiva-
lent to I is the set { I n | n ∈ N+ }.
In Proposition 3.6 of [3] Ciuperca, Heinzer, Ratliff Jr., and Rush prove the following:
Let ( R , M) be a local ring and I a normal ideal in R with I M2 . If both the associated graded ring
Mn In
G (M) = n0 Mn+1 and the fiber cone ring F ( I ) = n0 M I n are reduced, then I is projectively full.
This proposition implies the following result [3, Corollary 3.7]:
E-mail address: raymond.debremaeker@wis.kuleuven.be.
doi:10.1016/j.jalgebra.2008.10.016
904 R. Debremaeker / Journal of Algebra 321 (2009) 903–911
Let ( R , M) be a regular local domain of altitude d 2 and let x1 , . . . , xd be a regular system of parameters.
Then every ideal of the form I = (xn1 , x2 , . . . , xd ) R is projectively full.
The purpose of this paper is to prove a similar result in case ( R , M) is a two-dimensional Muhly
local domain, i.e., a two-dimensional normal Noetherian local domain with algebraically closed residue
field and the associated graded ring an integrally closed domain.
More precisely we will prove the following
Proposition. Let ( R , M) be a two-dimensional Muhly local domain and let x1 , . . . , xd be a minimal ideal
basis of M such that x1 ∈
/ rad(x2 , . . . , xd ). Then every ideal of the form I = (xn1 , x2 , . . . , xd ) with n ∈ N+ , is
projectively full.
In Section 2 we will prove that the ideals of the form I = (xn1 , x2 , . . . , xd ) are contracted from
M
R[ x1
], i.e.,

M
IR ∩ R = I.
x1
Using this, it will follow that these ideals and their transforms (for explanation of terminology, see
Section 2) are integrally closed (or complete) for all n 1. These facts together with the observation
that the blowup BlM ( R ) of R at M is a desingularization of R, will enable us to exploit Zariski’s
theory of complete ideals in two-dimensional regular local rings to prove our main result in Section 3.
For the definition and properties of the blowup of a ring at an ideal, the reader is referred to [6].
Background information on Zariski’s theory of complete ideals in 2-dimensional regular local rings
can be obtained from [7], [11, Chapter 14] and [12]. Good references for briefly stated definitions and
theorems from Zariski’s theory of complete ideals are [9] and [10].
2. Preliminaries
Let ( R , M) be a two-dimensional Muhly local domain and let x1 , x2 , . . . , xd be a minimal ideal

basis of M such that x1 ∈ / rad(x2 , . . . , xd ). Note that the condition “x1 ∈/ rad(x2 , . . . , xd )” is equivalent
R
to “(x2 , . . . , xd ) is a prime ideal of R” and also equivalent to “ (x ,..., x )
is a one-dimensional regular
2 d
local ring,” and it implies that ( xx2 , . . . ,
1
xd
x1
)R[ M
x1
]∩ R = (x2 , . . . , xd ). It follows that

x2 xd
M 1 := x1 , ,...,
x1 x1
is a maximal ideal of R [ M
x
] lying over M (i.e., M 1 ∩ R = M) and the ring
1

M
R := R
x1 M1
is a two-dimensional regular local ring. To see this, note that the associated graded ring of R be-
ing a two-dimensional integrally closed domain, implies that the closed fibre of the blowup of R
at M is a non-singular curve. It follows that every local ring of BlM ( R ) is regular (see [5, p. 403]
and [8, p. 259]). The local ring R is called an immediate (or a first) quadratic transform of R. Let
I := (xn1 , x2 , . . . , xd ) with n 2. Then in the ring R [ M
x
], one has 1

M x2 xd
IR = x1 xn1−1 , , . . . , .
x1 x1 x1
R. Debremaeker / Journal of Algebra 321 (2009) 903–911 905
The ideal J := (xn1−1 , x2

x1
, . . . , xxd1 ) is called the transform of I in R [ M
x
]. It is clear that M 1 is the only
1
prime ideal of R [ xM ] containing J . Localizing at M 1 yields in R
1
I R = x1 J M 1
and the ideal I := J M 1 is called the transform of I in ( R , M ), where M denotes the maximal ideal
of R . Since the transform I = R , the immediate quadratic transform ( R , M ) of ( R , M) is called an
immediate base point of I . For i ∈ {2, . . . , d}, we have

M M
IR = xi R ,
xi xi
hence there are no immediate base points of I on the chart R [ Mxi

]. This shows that ( R , M ) is in
fact the only immediate base point of I .
We are now ready to prove a few results about the ideals I = (xn1 , x2 , . . . , xd ) which will be used
in the proof of our main result.
Lemma 2.1. Every ideal of the form I = (xn1 , x2 , . . . , xd ) with n 1 is contracted from R [ M
x
]. 1
Proof. We proceed by induction on n and we first consider the case n = 1. Then it is clear
that I R [ M x1
] ∩ R = I because I = M in that case. Now suppose that (xn1 , x2 , . . . , xd ) R [ M
x1
]∩R=
n
(x1 , x2 , . . . , xd ). Then we have to prove that

M
xn1+1 , x2 , . . . , xd R ∩ R = xn1+1 , x2 , . . . , xd .
x1
To this end we observe that

M
xn1+1 , x2 , . . . , xd ⊆ xn1+1 , x2 , . . . , xd R ∩ R ⊆ xn1 , x2 , . . . , xd .
x1
Now (xn1+1 , x2 , . . . , xd ) ⊂ (xn1 , x2 , . . . , xd ) are adjacent ideals, i.e., their lengths differ by one. To see
/ (xn1+1 , x2 , . . . , xd ) because x1 ∈
this it suffices to observe that xn1 ∈ / rad(x2 , . . . , xd ). It follows from the
adjacentness that

M
xn1+1 , x2 , . . . , xd R ∩ R = xn1+1 , x2 , . . . , xd or xn1 , x2 , . . . , xd .
x1
So it remains to prove that (xn1+1 , x2 , . . . , xd ) R [ M

x1
] ∩ R = (xn1 , x2 , . . . , xd ) is not possible. For if this
would be the case, then one would have

M M
xn1 , x2 , . . . , xd R = xn1+1 , x2 , . . . , xd R ,
x1 x1
hence

x2 xd x2 xd
x1 xn1−1 , ,..., = x1 xn1 , , . . . ,
x1 x1 x1 x1
and thus

x2 xd x2 xd
xn1−1 , ,..., = xn1 , , . . . , .
x1 x1 x1 x1
R
This would imply that in the one-dimensional regular local ring S := (x ,..., xd )
one would have
2
n−1
N = N , where N denotes the maximal ideal of S. This is impossible and hence
n

M
xn1+1 , x2 , . . . , xd R ∩ R = xn1+1 , x2 , . . . , xd . 2
x1
This result will be used to prove in the next lemma that the ideals I = (xn1 , x2 , . . . , xd ) are complete.
Lemma 2.2. With the previous assumptions and notations, we have:
(i) Every ideal (xn1 , xx2 , . . . , xd ) of R [ M

x
1 1 x1
] is complete.
(ii) Every ideal (xn1 , x2 , . . . , xd ) of R is complete.
R[ M ]
Proof. First let us recall that
x 1 ∼
= R
( xx21 , . . . , xxd1 ) ∩ R = (x2 , . . . , xd ), and the ring
x x
( x2 ,..., xd ) (x2 ,...,xd ) since
1 1
R
S = (x ,..., xd )
is a one-dimensional regular local domain with maximal ideal N = (x1 ) S, where x1
2
denotes the natural image of x1 in S.
x J
Let J := (xn1 , xx2 , . . . , xd ), then the ideal (x ,...,x ) of S is equal to N n and hence integrally closed.
1 1 2 d
This implies that J is also integrally closed, which proves (i).
To prove (ii), we may suppose n 2, the case n = 1 being trivial. In R [ M
x
] we have that 1

M x2 xd
xn1 , x2 , . . . , xd R = x1 xn1−1 , ,..., .
x1 x1 x1
Since ( R , M) is a two-dimensional Muhly local domain, the ring R [ M

x
] is an integrally closed do- 1
main and since (xn1−1 , x2
x1
, . . . , xxd1 ) is a complete ideal of R [ M
x
], it follows that x1 (xn1−1 , x2
x1
, . . . , xxd1 ) is
1
complete too. But

M
xn1 , x2 , . . . , xd R ∩ R = xn1 , x2 , . . . , xd
x1
by Lemma 2.1 and thus (xn1 , x2 , . . . , xd ) is complete. 2
Corollary 2.3. Every ideal of the form I = (xn1 , x2 , . . . , xd ), n 2, has just one immediate base point ( R , M )
where R = R [ M
x
] M 1 with M 1 := (x1 , x2
x1
, . . . , xxd1 ) and the transform I of I in R is a simple complete ideal.
1
Proof. It is clear that I R [ M

x
] = ( xi ) R [ M
x
] for i = 2, . . . , d, hence the immediate base points of I can
i i
only show up on the chart R [ M
x
]. In the ring R [ M
x
] we have
1 1

M x2 xd
IR = x1 xn1−1 , ,...,
x1 x1 x1
where J := (xn1−1 , x2
x1
, . . . , xxd1 ) is the transform of I in R [ M
x
] and M 1 is the only prime ideal of R [ M
x
]
1 1
containing J . It follows that R [ M
x
] M 1 is the only immediate base point of I .
1
The transform J of I in R [ M
x
] is complete because of Lemma 2.2(i). This implies that the transform
1
I = J M 1 of I in R is also complete, and I is simple (i.e., not a product of two proper ideals) because
ord R ( I ) = 1. 2
In order to state our next result we recall that ( R , M) being a two-dimensional Muhly local do-
main implies that the M-adic order function ord R is a valuation which will be denoted by v M . If
I = (xn1 , x2 , . . . , xd ) with n 2, then I has only one immediate base point ( R , M ) (see Corollary 2.3),
and R is a two-dimensional regular local ring. Since the transform I of I in the two-dimensional
regular local ring R is a simple complete ideal, we know by Zariski’s theory of complete ideals that
there corresponds to I a unique prime divisor of R , namely the unique Rees valuation of I , denoted
by w (see e.g. [4, Corollaries 3.6 and 3.7]). Thus T ( I ) = { w }.
Throughout this paper the set of Rees valuations of an ideal A in a local ring will be denoted
by T ( A ).
Lemma 2.4. For every ideal of the form I = (xn1 , x2 , . . . , xd ) with n 2, we have that T ( I ) ⊆ { v M , w } and
w ∈ T ( I ).
Proof. Since ( R , M ) is the only immediate base point of I , it follows that the blowup Bl I M ( R )
is obtained by blowing up R at M and then blowing up the local ring ( R , M ) ∈ BlM ( R ) at the
transform I of I in R (see [6, Lemma 1.11]). This implies that T ( I M) = { v M , w } and hence
T (I ) ⊆ {v M, w }
because T ( I M) = T ( I ) ∪ T (M) (cf. [11, Proposition 10.4.8]). Finally we have to show that w ∈ T ( I ).
Suppose not, then one would have that T ( I ) = { v M ). This would imply that I has no immediate base
points, contradicting Corollary 2.3. 2
We close this section by recalling information on the transform I of I = (xn1 , x2 , . . . , xd ) (n 2)

in its unique immediate base point ( R , M ) and on the unique Rees valuation w of I . Since w is
a prime divisor of R (i.e., the valuation ring (W , MW ) of w dominates ( R , M ) and the transcen-
R
dence degree of the residue field of W over M is one), we know by Abhyankar [1, Proposition 3]
that there exists a unique finite quadratic sequence starting from R and dominated by (W , MW ):
( R , M ) = ( R 1 , M1 ) < ( R 2 , M2 ) < · · · < ( R s , Ms ) < (W , MW ),
i.e., ( R i , Mi ) is an immediate quadratic transform of ( R i −1 , Mi −1 ) for i = 2, . . . , s and W is the

ord R s -valuation ring. Thus the unique Rees valuation w of I is the ord R s -valuation.
Moreover the transform of I in R s , denoted I R s , is the maximal ideal of R s , i.e.,
I R s = Ms ,
and conversely, the inverse transform of Ms in R is I (see e.g. [9, p. 608]).
3. Proof of the main result
Let ( R , M) be a two-dimensional Muhly local domain and let x1 , . . . , xd be a minimal ideal basis
of M such that x1 ∈ / rad(x2 , . . . , xd ). Then we have to prove that the ideal I := (xn1 , x2 , . . . , xd ) is
projectively full for all integers n 1. The case n = 1 is clear since M has only one Rees valuation,
namely v M = ord R -valuation.
So we may assume in the rest of the proof that n 2. Let J be a complete M-primary ideal of R
that is projectively equivalent to I (i.e., I i = J j with i, j positive integers). Then we have to prove
that J = I ν for some positive integer ν . As J is projectively equivalent to I , J has the same Rees
valuations as I , hence T ( J ) = T ( I ).
Since T ( I ) ⊆ { v M , w } and w ∈ T ( I ) (see Lemma 2.4), it follows that J , just like I , has precisely
one immediate base point, namely the center of w on BlM ( R ), i.e., ( R , M ).
Indeed, it is clear that J has at least one immediate base point since otherwise one would have
that Bl J ( R ) is dominated by BlM ( R ) and thus T ( J ) ⊆ T (M) = { v M } by Proposition 2.2 in [5]. This
would imply that T ( I ) = { v M } contradicting Lemma 2.4. Any immediate base point of J is dominated
by a Rees valuation ring of J (one can prove this using an argument similar to that used by Göhner
in his proof of Proposition 2.2 in [5]), hence by the valuation ring (W , MW ) of w since w is the
only element = v M in T ( J ). This proves that J has only one immediate base point, namely the
unique local ring of BlM ( R ) dominated by (W , MW ) (which is called the center of w on BlM ( R )).
It follows from the discussion at the end of Section 2 that the center of w on BlM ( R ) is ( R , M ).
Hence, ( R , M ) is the unique immediate base point of J .
Next, we prove that w is the only Rees valuation of the integral closure J of the transform J of J
in its unique immediate base point ( R , M ). Let r := ord R ( J ), then we have in R that
J R = xr1 J
and the ideal J is the transform of J in ( R , M ). (Note, we do not know whether J is complete or
not.) It follows that
J R = xr1 J
where J is a complete M -primary ideal in the two-dimensional regular local ring ( R , M ) and we
claim that
T ( J ) = { w }.
In order to prove the claim, let us suppose that there exists a Rees valuation w of J with w = w.
We now show this leads to a contradiction.
Since w is a Rees valuation of J , we have
W ∈ Bl J ( R )
where W denotes the valuation ring of w . Because J R = xr1 J , we have Bl J ( R ) = Bl J R ( R ) (see [6]).
From the fact that ( R , M ) is the only immediate base point of J , it follows that the blowup
Bl J M ( R ) is obtained by blowing up R at M and then blowing up the local ring ( R , M ) ∈ BlM ( R )
at J R (see [6, Lemma 1.11]). This implies that W ∈ Bl J M ( R ) and it follows that
w ∈ T ( J M) = T (M) ∪ T ( J )
where T (M) = { v M } and T ( J ) ⊆ { v M , w }.

As w = w, we have w = v M , which implies that W ∈ BlM ( R ). Since W dominates the local
ring R ∈ BlM ( R ), we have the desired contradiction. Thus J cannot have Rees valuations = w and
this proves that T ( J ) = { w }, since every ideal in a Noetherian ring has a set of Rees valuations
(see [11, Theorem 10.2.2]).
From the theory of complete ideals in a two-dimensional regular local ring (see for exam-
ple [4,9,10]), we know that J is some power of the simple complete M -primary ideal that cor-
responds to w via Zariski’s one-to-one correspondence. This unique simple complete M -primary
ideal is the inverse transform of Ms in R , where Ms is as in the discussion at the end of Section 2
(cf. [11, Theorem 10.2.2]). Since we know that Ms is the transform of I in R s , we have conversely
that I = (xn1−1 , xx2 , . . . , xd ) is the inverse transform of Ms in R (see the end of Section 2). Thus
x
1 1
ν
x2 xd
J = I ν = xn1−1 , ,..., for some ν ∈ N+ .
x1 x1
The fact that J is projectively equivalent to I will enable us to prove that
ν = r,
where r = ord R ( J ).
Since I i = J j for some i , j ∈ N+ , ord R ( I ) = 1 and ord R ( J ) = r, it follows that i = r j and hence
J j R = Ir j R.
Recalling that I R = x1 (xn1−1 , x2

x1
, . . . , xxd1 ), we have that
r j
x2 xd
xn1−1 ,
rj
J j R = x1 ,..., . (1)
x1 x1
On the other hand, J R = xr1 J implies
ν j
x2 xd
xn1−1 ,
rj rj
J j R = ( J R ) j = x1 ( J ) j = x1 ,..., . (2)
x1 x1
Comparing (1) and (2) yields
r j ν j
x2 xd x2 xd
xn1−1 , ,..., = xn1−1 , , . . . , .
x1 x1 x1 x1
It follows that r j = ν j, hence r = ν .

Thus
r
x2 xd
J =I = r
xn1−1 , ,...,
x1 x1
and hence
J R = xr1 J = xr1 I r = (x1 I )r = I r R .
From J R = I r R , it follows that
J = Ir R ∩ R.
To see this, consider the following sequence of inclusions
J ⊆ J R ∩ R ⊆ J VM ∩ J W ∩ R ,
where VM is the valuation ring of v M = ord R and W is the valuation ring of w.

Since T ( J ) ⊆ { v M , w } and J is complete, we have
J VM ∩ J W ∩ R = J
and thus
J = J R ∩ R.
Since J R = I r R , we see that J = I r R ∩ R (in particular I r R ∩ R is a complete ideal of R).

Finally we prove that I r R ∩ R = I r . Since I r ⊆ I r R ∩ R and I r R ∩ R is complete, we have that
I ⊆ I r R ∩ R. Further we have the following inclusions
r
I r ⊆ I r R ∩ R ⊆ I r VM ∩ I r W ∩ R ,
and
I r VM ∩ I r W ∩ R = I r
because T ( I r ) = T ( I ) ⊆ { v M , w } and w ∈ T ( I r ), hence
Ir R ∩ R = Ir .
Thus J = I r , i.e., I = (xn1 , x2 , . . . , xd ) is projectively full.
Final Remark. In the special case where ( R , M) is a two-dimensional regular local ring, any minimal
ideal basis x1 , x2 of M satisfies the condition x1 ∈/ rad(x2 ). However, this is not necessarily true if
( R , M) is a two-dimensional Muhly local domain that is not regular, as the example below shows.
Let
k [ X , Y , Z ]( X , Y , Z )
R := ,
( X 2 − Y Z )( X ,Y , Z )
with k an algebraically closed field and X , Y , Z indeterminates over k.

It follows that R = k[x, y , z](x, y ,z) with x2 = yz, where x, y, z denote the natural images of X , Y , Z .
Then, y, x, z is a minimal ideal basis of M such that y ∈ / rad(x, z), while the minimal ideal basis x,
y, z of M clearly does not satisfy the condition x ∈ / rad( y , z).
More generally, let ( R , M) be any two-dimensional Muhly local domain of the form
k[ X 1 , . . . , X d ]( X 1 ,..., Xd )
R := ,
N ( X 1 ,..., Xd )
where k is an algebraically closed field, X 1 , . . . , X d are indeterminates over k, and N is a homogeneous

prime ideal of height d − 2 in k[ X 1 , . . . , X d ]. Using the projective Nullstellensatz, it can be shown that
there exists a minimal ideal basis
x1 , x2 , . . . , xd
of M, such that (x2 , . . . , xd ) is a prime ideal. Hence

x1 ∈
/ rad x2 , . . . , xd .
References
[1] S. Abhyankar, On the valuations centered in a local domain, Amer. J. Math. 78 (1956) 321–348.
[2] C. Ciuperca, W.J. Heinzer, L.J. Ratliff Jr., D.E. Rush, Projectively equivalent ideals and Rees valuations, J. Algebra 282 (2004)
140–156.
[3] C. Ciuperca, W.J. Heinzer, L.J. Ratliff Jr., D.E. Rush, Projectively full ideals in Noetherian rings, J. Algebra 304 (2006) 73–93.
[4] R. Debremaeker, V. Van Lierde, The effect of quadratic transformations on degree functions, Beitrage Algebra Geom. 47
(2006) 121–135.
[5] H. Göhner, Semifactoriality and Muhly’s condition (N) in two dimensional local rings, J. Algebra 34 (1975) 403–429.
[6] W. Heinzer, B. Johnston, D. Lantz, K. Shah, Coefficient ideals in and blowups of a commutative Noetherian domain, J. Alge-
bra 162 (1993) 355–391.
[7] C. Huneke, Complete ideals in two-dimensional regular local rings, in: Commutative Algebra, in: Math. Sci. Res. Inst. Publ.,
vol. 15, 1989, pp. 325–337.
[8] J. Lipman, Rational singularities with applications to algebraic surfaces and unique factorization, Publ. Math. Inst. Hautes
Études Sci. 36 (1969) 195–279.
[9] S. Noh, The value semigroups of prime divisors of the second kind in 2-dimensional regular local rings, Trans. Amer. Math.
Soc. 336 (1993) 607–619.
[10] S. Noh, Simple complete ideals in two-dimensional regular local rings, Comm. Algebra 25 (1997) 1563–1572.
[11] I. Swanson, C. Huneke, Integral Closure of Ideals, Rings, and Modules, London Math. Soc. Lecture Note Ser., vol. 336, Cam-
bridge Univ. Press, Cambridge, 2006.
[12] O. Zariski, P. Samuel, Commutative Algebra, vol. II, reprint of the 1960 edition, Grad. Texts in Math., vol. 29, Springer-Verlag,
New York, 1975.
Journal of Algebra
Some reducible Specht modules for Iwahori–Hecke algebras

of type A with q = −1
Matthew Fayers a,∗ , Sinéad Lyle b
a
School of Mathematical Sciences, Queen Mary, University of London, Mile End Road, London E1 4NS, UK
b
School of Mathematics, University of East Anglia, Norwich NR4 7TJ, UK
Article history: The reducibility of the Specht modules for the Iwahori–Hecke
Received 10 June 2008 algebras in type A is still open in the case where the defining
Available online 2 December 2008 parameter q equals −1. We prove the reducibility of a large class
of Specht modules for these algebras.
Keywords:
Specht modules
Iwahori–Hecke algebras
1. Introduction
Let n be a non-negative integer, F a field, and q an element of F. The Iwahori–Hecke algebra

H = HF,q (Sn ) is a finite-dimensional F-algebra which arises in various mathematical contexts. Its
representation theory bears a close resemblance to the representation theory of the group algebra
FSn (which arises in the special case q = 1). A particularly important class of modules for H is the
class of Specht modules; these arise as cell modules for a certain cellular basis of H, and in cases
where H is semi-simple the Specht modules are irreducible and afford all the irreducible representa-
tions of H.
In the non-semi-simple case (where q is a root of unity in F), it is still interesting to know
which Specht modules are irreducible; for the case where q = 1 and F has prime characteristic p,
this amounts to asking which ordinary irreducible representations of the symmetric group remain
irreducible in characteristic p. The classification of irreducible Specht modules has been studied by
several authors, and is almost complete. This paper is a contribution towards completing the remain-
ing open case, namely the case where q = −1 ∈ F and the characteristic of F is not 2. Our main result
is Theorem 2.1, where we prove the reducibility of a large class of Specht modules. We hope to be
able to extend our results in future.
E-mail address: m.fayers@qmul.ac.uk (M. Fayers).
doi:10.1016/j.jalgebra.2008.11.006
M. Fayers, S. Lyle / Journal of Algebra 321 (2009) 912–933 913
We now give an indication of the layout of this paper. In Section 2, we give some very basic
definitions and state our main result; we also present a conjectured classification of the irreducible
Specht modules in the case where F has infinite characteristic. In Section 3, we recall the additional
definitions and background theory that we shall need. In Section 4, we state some fundamental results
on homomorphisms between various modules for H; we use these to prove further results which aid
us in proving reducibility of Specht modules. In Section 5, we recall the Fock space representation of
the quantum group U v ( sl2 ) and its applications to representation theory of Iwahori–Hecke algebras,
and we use these results to show reducibility of certain Specht modules. Finally in Section 6, we
combine our results to complete the proof of the main theorem.
2. The main result and a conjecture
In this section, we give our main theorem, and also present a conjectured classification of irre-
ducible Specht modules in the case where char(F) = ∞. First we review the background required to
enable us to state our results.
2.1. Iwahori–Hecke algebras, partitions and Specht modules
Throughout this paper F denotes a fixed field and q a non-zero element of F. We define e to be
the multiplicative order of q in F if q = 1, or the characteristic of F if q = 1; we adopt the convention
that a field whose prime subfield is Q has infinite characteristic. In this paper we shall be primarily
concerned with the case where e = 2 (that is, q = −1 ∈ F), though we shall state results for arbitrary
values of q as long as it is convenient. More general results than those quoted can easily be found
elsewhere, especially in the book by Mathas [21], which is our main reference. Note, however, that
we do not always follow Mathas’s conventions; in particular, we use the Specht modules defined by
Dipper and James [4] rather than those of Mathas.
Given any integer n 1, the Iwahori–Hecke algebra of the symmetric group Sn is defined to be the
unital associative F-algebra HF,q (Sn ) with generators T 1 , . . . , T n−1 and relations
( T i − q)( T i + 1) = 0 (1 i n − 1)
T i T i +1 T i = T i +1 T i T i +1 (1 i n − 2)
Ti T j = T j Ti (1 i < j − 1 n − 2).
The combinatorics describing the representation theory of HF,q (Sn ) is based on compositions and
partitions. A composition of n is defined to be a sequence λ = (λ1 , λ2 , . . .) of non-negative integers
such that the sum |λ| = λ1 + λ2 + · · · equals n; if in addition we have λ1 λ2 · · · , we say that λ is
a partition of n. When writing compositions and partitions, we often omit trailing zeroes and group
together equal non-zero parts, and we write ∅ for the unique partition of 0. If λ is a partition, we
write λ for the conjugate partition to λ; this is the partition in which

λi = { j ∈ N | λ j i }.
We say that a partition λ is e-regular if there does not exist i 1 such that λi = λi +e−1 > 0, and we
say that λ is e-restricted if there is no i with λi − λi +1 e.
With a partition [λ] is associated its Young diagram, which is the set

[λ] = (i , j ) ∈ N2 j λi .
We refer to elements of N2 as nodes, and to elements of [λ] as nodes of λ.

Now we can describe some modules for HF,q (Sn ). For any composition of n, one defines a module
λ known as the permutation module. If λ is a partition, then M λ has a distinguished submodule
M F,q F,q
914 M. Fayers, S. Lyle / Journal of Algebra 321 (2009) 912–933
λ called the Specht module, which is the main object of study in this paper. We retain the subscript
S F,q
F, q in our notation, to enable us to make statements about modules without reference to the under-
λ is irreducible, we mean that it is
lying Iwahori–Hecke algebra; for example, when we say that S F,q
irreducible as an HF,q (Sn )-module, where n = |λ|.
2.2. The main result
The purpose of this paper is to consider the question of which Specht modules for HF,q (Sn ) are
irreducible. As with many statements about the representation theory of HF,q (Sn ), the classification
of irreducible Specht modules for HF,q (Sn ) involves the parameter e defined above, which is often
called the ‘quantum characteristic’. If e = ∞, then all the Specht modules are irreducible, and they
afford all irreducible representations of HF,q (Sn ) as λ varies over the set of partitions of n. So we
assume from now on that e is finite.
In the case where e > 2, the classification of the irreducible Specht modules for HF,q (Sn ) is com-
plete; this result was proved by the authors and others over the course of several papers [5,6,14,16,17,
19,20]. To describe the classification, we begin with the case of Specht modules labelled by e-regular
partitions; the irreducibility of these Specht modules is determined by a theorem known as the Carter
Criterion [21, Corollary 5.43], of which a special case appears in Theorem 3.8 below. Applying a the-
orem concerning conjugate partitions (Corollary 3.3 below), one obtains a corresponding result for
Specht modules labelled by e-restricted partitions. The general case is a natural combination of the
e-regular case and the e-restricted case; a partition labelling an irreducible Specht module (called a
JM-partition) consists of an e-regular partition and an e-restricted partition, each labelling an irre-
ducible Specht module, joined together in a simple way.
When e = 2, the Carter Criterion is still valid, so the classification of irreducible Specht modules
labelled by 2-regular or 2-restricted partitions is known. However, for the case of partitions which
are neither 2-regular nor 2-restricted, things are different; one cannot just take the definition of
JM-partitions and set e = 2. In fact, the partitions of n which are neither 2-regular nor 2-restricted and
label irreducible Specht modules for HF,−1 (Sn ) seem to take a very different form from JM-partitions.
Roughly speaking, JM-partitions tend to be ‘thin’, by which we mean that a JM-partition has very few
diagonal nodes (i , i ) relative to its size; conversely, the partitions labelling irreducible Specht modules
when e = 2 tend to be closer to rectangular partitions (indeed, the rectangular partitions (ab ) all
label irreducible Specht modules when e = 2 and char(F) = ∞). Our main theorem illustrates this
difference, since it implies in particular that when e = 2 a Specht module labelled by a (neither
2-regular nor 2-restricted) JM-partition is reducible.
To give our main result, we need to define ladders: for l 1, the lth ladder in N2 is defined to be
the set

Ll = (i , j ) ∈ N2 i + j = l + 1 .
Given a partition λ, we define the lth ladder of λ to be the intersection Ll (λ) = Ll ∩ [λ]. We say that
Ll (λ) is disconnected if the nodes in Ll (λ) do not form a consecutive subset of Ll ; that is, there exist
1 a < b < c l such that (a, l + 1 − a) and (c , l + 1 − c ) lie in [λ], but (b, l + 1 − b) does not. Now
we can state our main theorem.
Theorem 2.1. Suppose F is any field, and λ is a partition. If there is some l such that the lth ladder of λ is
λ
disconnected, then the Specht module S F,−1 is reducible.
The reader may prefer a statement of Theorem 2.1 that does not involve ladders: it is a simple
exercise to show that a partition λ has a disconnected ladder if and only if there exist 1 a < b such
that λa − λa+1 2 and λb = λb+1 > 0.
Theorem 2.1 will be proved in the subsequent sections. For the rest of this section, we consider
how to extend it to give a complete classification of irreducible Specht modules.
2.3. A conjecture in infinite characteristic
In the case where char(F) = ∞, the decomposition numbers for HF,q (Sn ) may be computed using
the LLT algorithm [18]; so for any λ, there is a finite algorithm to determine whether S F, λ is reducible.
q
For the case e = 2, Andrew Mathas and the first author have carried out these computations for
partitions of size at most 45, and on the basis of this have made a conjecture.
In order to state this conjecture, we need to introduce some more terminology concerning Young
diagrams. If λ is a partition, then we say that a node (i , j ) of λ is removable if [λ] \ {(i , j )} is the
Young diagram of a partition (i.e. if j = λi > λi +1 ), while a node (i , j ) not in [λ] is an addable node of
λ if [λ] ∪ {(i , j )} is the Young diagram of a partition. For any node (i , j ) in N2 , we define its residue
to be the residue modulo 2 of the integer j − i.
Now suppose λ is neither 2-regular nor 2-restricted. Let a be maximal such that λa − λa+1 2, let
b be maximal such that λb = λb+1 > 0, and let c be maximal such that λa+c > 0. We say that λ is an
FM-partition if the following conditions hold:
• λi − λi +1 1 for all i = a;
• λb a − 1 b;
• λ1 > · · · > λc ;
• if c = 0, then all the addable nodes of λ, except possibly those in the first row and first column,
have the same residue;
• if c > 0, then all addable nodes of λ have the same residue.
Now we can give our conjecture; note that the case where a partition is 2-regular or 2-restricted
is covered by the discussion in Section 2.2, so we can restrict attention to partitions which are neither
2-regular nor 2-restricted.
Conjecture 2.2. Suppose char(F) = ∞ and that λ is neither 2-regular nor 2-restricted. Then the Specht mod-
λ
ule S F,−
1 is irreducible if and only if either λ or λ is an FM-partition.
This still leaves open the case where λ has prime characteristic. Thanks to the theory of decompo-
sition maps (see Theorem 3.4 below), we know that the set of partitions labelling irreducible Specht
modules for q = −1 in characteristic p is a subset of the set of partitions labelling irreducible Specht
modules in infinite characteristic. Experimental evidence suggests that it is a rather small subset; in
fact, it seems likely that for any prime p there are only finitely many partitions which are neither
2-regular nor 2-restricted and label irreducible Specht modules. This statement has been proved in
the case p = 2 by James and Mathas [17]; here the only such partition is (22 ). We hope to be able to
make a more precise statement in the future.
3. Useful background results
In this section we summarise some simple background results which we shall need in order to
prove our Theorem 2.1.
3.1. Irreducible modules for HF,q (Sn ) and the dominance order
In order to examine the reducibility of Specht modules, it will be helpful to understand the clas-
sification of irreducible HF,q (Sn )-modules. Let e be as defined in Section 2.2, and suppose that λ is
λ
a partition of n. If λ is e-regular, then the Specht module S F,q has an irreducible cosocle which is
labelled D λF,q ; the modules D λF,q give all the irreducible HF,q (Sn ) modules, as λ ranges over the set
of e-regular partitions of n.
The classification of irreducible Specht modules is a special case of the decomposition number prob-
λ : D μ ], as λ and μ vary. The most basic results
lem, which asks for the composition multiplicities [ S F, q F,q
on this problem concern the dominance order. If λ and μ are partitions, we say that λ dominates μ
(and write λ Q μ) if for each i 1 we have
λ1 + · · · + λ i μ1 + · · · + μ i .
Now we have the following.
Proposition 3.1. (See [4, Corollaries 4.12 and 4.14].) Suppose λ and μ are partitions of n, with μ e-regular.
Suppose M is either the permutation module M F, λ or the Specht module S λ . If [ M : D μ ] > 0, then μ Q λ.
q F,q F,q
μ
If μ = λ, then [ M : D F,q ] = 1.
3.2. Conjugate partitions and duality
Let T 1 , . . . , T n−1 be the standard generators of HF,q (Sn ). Let : HF,q (Sn ) → HF,q (Sn ) be the
involutory automorphism sending T i to q − 1 − T i , and let ∗ : HF,q (Sn ) → HF,q (Sn ) be the anti-
automorphism sending T i to T i . Given a module M for HF,q (Sn ), define M to be the module with
the same underlying vector space and with action
h · m = h m,
and define M ∗ to be the module with underlying vector space dual to M and with HF,q (Sn )-action
given by

h · f (m) = f h∗ m .
We can describe the effect of these functors on Specht modules, using conjugate partitions.
Proposition 3.2. Suppose λ is a partition, and let λ denote the conjugate partition. Then
λ ∗
λ
S F,q ≡ S F,q .
Proof. This is the result of [21, Exercise 3.14(iii)]. Although Mathas’s definition of Specht modules is
different from ours, his description of the relationship between the two definitions [21, p. 38] ensures
that the result holds for our Specht modules also. 2
This has the following immediate corollary, which will be very useful.
λ is irreducible if and only if S λ is.

Corollary 3.3. Suppose λ is a partition. Then S F,q F,q
3.3. Decomposition maps and adjustment matrices
In this section we quote a result which will allow us to assume that our underlying field F is the
field of rational numbers. The theorem we shall state was proved by Geck in [10], and arises from
a consideration of decomposition maps between Iwahori–Hecke algebras defined over different rings.
(For an introduction to decomposition maps, see Geck’s article [11].) The theorem is most conveniently
stated in terms of the decomposition matrix of HF,q (Sn ); this is the matrix with rows indexed by the
partitions of n and columns indexed by the e-regular partitions, and with the (λ, μ)-entry being the
decomposition number [ S F,λ : D μ ].
q F,q
Theorem 3.4. Let ζ be a primitive eth root of unity in C. Let D (C, ζ ) and D (F, q) denote the decomposition
matrices of HC,ζ (Sn ) and HF,q (Sn ) respectively. Then there is a square matrix A with non-negative integer
entries and with 1s on the diagonal, such that
D (F, q) = D (C, ζ ) A .
The important consequence of this theorem from our point of view is that for any field F and any
λ has at least as many composition factors as S λ ; in particular, if
partition λ, the Specht module S F,q C,ζ
λ
S C,ζ λ .
is reducible, then so is S F,q
3.4. Cores and blocks
Here we give the classification of the blocks of HF,−1 (Sn ); this is based on the combinatorics of
dominoes. Define a domino to be a pair of horizontally or vertically adjacent nodes in N2 . A removable
domino of a partition λ is a domino contained in [λ] such that the removal of this domino leaves
the Young diagram of a partition. The core of λ is the partition obtained by repeatedly removing
removable dominoes until there are no more. This partition has the form (l, l − 1, . . . , 2, 1) for some
l 0, and is independent of the way is which the removable dominoes are chosen at each stage. The
weight of λ is the number of dominoes removed to obtain the core.
Example. Let λ = (62 , 5, 2, 1). Then λ has core (3, 2, 1) and weight 7, as we can see from the following
diagram.
Now we can address the blocks of HF,−1 (Sn ). Because the Specht modules are the cell modules
arising from a particular cellular basis of HF,−1 (Sn ), it follows that each Specht module lies entirely
within one block of HF,−1 (Sn ) [21, Corollary 2.22]; we abuse notation by saying that a partition λ
λ
lies in a block B if S F,−1 lies in B. Each block must contain at least one Specht module (because every
simple module occurs as a composition factor of some Specht module), and so a classification of the
blocks of HF,−1 (Sn ) may be described by giving the appropriate partition of the set of partitions of n.
This is done as follows.
Theorem 3.5. (See [21, Corollary 5.38].) Suppose λ and μ are two partitions of n. Then λ and μ lie in the same
μ
block of HF,−1 (Sn ) if and only if λ and μ have the same core. Hence if μ is 2-regular, then [ S F,−
λ
1 : D F,−1 ] = 0
unless λ and μ have the same core.
As a consequence of this theorem, we can define the weight and core of a block B, meaning the
weight and core of any partition labelling a Specht module lying in B.
3.5. Rouquier blocks
Suppose B is a block of HF,−1 (Sn ), with core ν = (l, l − 1, . . . , 1) and weight w. We say that B
is Rouquier if w l + 1. Rouquier blocks are very useful because we have an explicit formula for
their decomposition numbers in the case where the underlying field has infinite characteristic. We
now describe these results, following [15] (where a Rouquier block is referred to as a ‘block with an
enormous 2-core’).
Suppose B , ν , w are as above, with w l + 1. If λ is a partition in B, then there is a unique way

to partition the set [λ] \ [ν ] into dominoes. Given this partition, we define λhi to be the number of
horizontal dominoes in row i of [λ] \ [ν ], and we define λiv to be the number of vertical dominoes
in column i of [λ] \ [ν ], for i 1. Then λh and λ v are partitions, with |λh | + |λ v | = w. Moreover, λ is
uniquely specified by ν , λh , λ v , and in fact given any pair σ , τ of partitions with |σ | + |τ | = w, there
is a partition μ in B with μh = σ , μ v = τ .
Example. Suppose λ = (13, 8, 7, 4, 3, 2, 15 ). Then λ has weight 7 and core (7, 6, 5, 4, 3, 2, 1), so lies
in a Rouquier block. The Young diagram [λ] may be drawn as follows, and we see that (λh , λ v ) =
((3, 12 ), (2)):
So we may label a partition λ in a Rouquier block B by the pair (λh , λ v ); we remark that the pair
(λh , (λ v ) ) is often referred to as the 2-quotient of λ. It is easy to see that λ is 2-regular if and only if
λ v = ∅, while λ is 2-restricted if and only if λh = ∅. Now we can describe the decomposition numbers
for a Rouquier block in infinite characteristic; given any partitions α , β, γ with |α | = |β| + |γ |, let c βαγ
be the corresponding Littlewood–Richardson coefficient (see [9]).
Theorem 3.6. (See [15, Theorem 2.5].) Suppose char(F) = ∞ and B is a Rouquier block of HF,−1 (Sn ). Suppose
λ and μ are partitions in B, with μ 2-regular, and let (λh , λ v ) and (μh , ∅) be the corresponding pairs of
partitions. Then
μ μh
1 : D F,−1 = c
λ
S F,− .
λh λ v
The only corollary we need from this is the following.
Corollary 3.7. Suppose B is a Rouquier block of HF,−1 (Sn ), and λ is a partition in B which is neither 2-regular
λ
nor 2-restricted. Then S F,−1 is reducible.
Proof. By the results of Section 3.3, we may assume that char(F) = ∞. Since λ is neither 2-regular
nor 2-restricted, the labelling partitions λh and λ v are both non-empty. It is well known and easy
to prove from the definition of Littlewood–Richardson coefficients that if β and γ are non-empty
partitions, then there are at least two partitions α for which c βαγ > 0, and now the result follows
from Theorem 3.6. 2
3.6. Alternating partitions
The question of irreducibility of Specht modules labelled by 2-regular partitions has been settled
for some time. We shall need this result later, so we quote it here, concentrating for simplicity on the
case where char(F) = ∞. Say that a partition is alternating if for every i either λi + λi +1 is odd or
λi +1 = 0.
Theorem 3.8. (See [21, Theorem 5.42].) Suppose char(F) = ∞, and λ is a 2-regular partition. Then S F,−
λ
1 is
irreducible if and only if λ is alternating.
We combine this with the following simple lemma.
Lemma 3.9. Suppose λ is a partition with core (l − 1, l − 2, . . . , 1) for some l 1, and that λl+1 = 0. Then λ
is alternating.
Proof. We use induction to prove a stronger statement, namely that λi ≡ i + l (mod 2) for 1 i l.
The induction is on the weight w of λ; if w = 0 then λ is the core (l − 1, l − 2, . . . , 1), which certainly
has the required property. For the inductive step, suppose that w > 0 and let ν be a partition obtained
by removing a domino from [λ]. Then ν satisfies the hypotheses of the lemma, so by induction we
have νi ≡ i + l (mod 2) for i = 1, . . . , l. In particular, we have νi = νi +1 if 1 i l − 1. This means
that the domino added to [ν ] to obtain [λ] must be horizontal; for if it were vertical, consisting of
the nodes (i , j ) and (i + 1, j ) say, then we would have νi = νi +1 (= j − 1), and hence i l. But this
would give λl+1 > 0, contradicting our assumptions. So the added domino is horizontal, and hence
λi ≡ νi (mod 2) for all i, which gives the required conclusion. 2
When we combine Lemma 3.9 with Proposition 3.1, the following result is immediate.
Corollary 3.10. Suppose char(F) = ∞, and μ is a 2-regular partition with core (l − 1, l − 2, . . . , 1). If λ is a
μ
partition such that [ S F,−
λ
1 : D F,−1 ] > 0 and λl+1 = 0, then λ = μ.
3.7. Ladders and James’s regularisation theorem
We saw above that when λ is 2-regular, the simple module D λF,−1 occurs as a composition factor
λ
of S F,− 1 with multiplicity 1. James has given an extension of this result to the case where λ is
not 2-regular, giving an explicit simple module which occurs exactly once as a composition factor
λ
of S F,−1 . This was done in [12] for symmetric groups, and extended to the general case in [13].
This is very useful from the point of view of classifying irreducible Specht modules, since if S F,− λ
1 is
μ λ
irreducible, the theorem tells us which irreducible module D F,−1 is isomorphic to S F,−1 .
Recall the definition of ladders from Section 2.2. Given a partition λ, it is easily seen that λ
is 2-regular if and only if for each l the nodes in the lth ladder of λ are as high as possible,
i.e. Ll (λ) = {(1, l), (2, l − 1), . . . , (s, l + 1 − s)} for some s. Furthermore, for any λ we may obtain a
2-regular partition by moving the nodes in each ladder of λ to the topmost positions in that ladder;
this 2-regular partition is called the regularisation of λ, written λreg .
Example. Let λ = (4, 23 ). Then we have
Ll (λ) = Ll for l 3,

L4 (λ) = (1, 4), (3, 2), (4, 1) ,

L5 (λ) = (4, 2) ,
Ll (λ) = ∅ for l 6.
By replacing L4 (λ) with {(1, 4), (2, 3), (3, 2)} and L5 (λ) with {(1, 5)}, we obtain λreg = (5, 3, 2):
λ= ; λreg = .
Theorem 3.11. (See [13, Theorem 6.21].) Suppose λ, μ are partitions of n, with μ 2-regular. If [ S F,−
λ
1 :
μ reg
D F,−1 ] > 0, then μ Q λ reg
. Furthermore, [ S F,−
λ
1 : D λF,−1 ] = 1.
We note the following corollary concerning homomorphisms.
Corollary 3.12. Suppose λ, μ are partitions of n. Suppose that either:
μ
1. λreg S μreg , and there exists a non-zero homomorphism from S F,−1 to S F,−
λ
1 ; or
μ
2. λreg S μ, and there exists a non-zero homomorphism from M F,−1 to S F,−
λ
1.
λ
Then S F,−1 is reducible.
Proof. This follows immediately from Theorem 3.11 and Proposition 3.1. 2
3.8. A useful lemma
In this section we recall a very useful result; the general version of this result was the main tool
used in [19] for proving the reducibility of Specht modules.
Given a partition λ and i ∈ {0, 1}, define the partition λ−i by removing all removable nodes of
residue i. Then we have the following, which is a corollary of [2, Lemma 2.13].
Lemma 3.13. Suppose λ is a partition and i ∈ {0, 1}. Then S F,− λ

1 has at least as many composition factors
λ− i
λ− i
λ
as S F,− 1 . In particular, if S F,−1 is reducible, then so is S F,−1 .
We also need a ‘dual’ result to this. Given λ and i ∈ {0, 1}, define λ+i by adding all addable nodes
of residue i. Then the following result may be proved in exactly the same way as Lemma 3.13.
Lemma 3.14. Suppose λ is a partition and i ∈ {0, 1}. Then S F,− λ

1 has at least as many composition factors
λ+i λ+i λ
as S F,−1 . In particular, if S F,−1 is reducible, then so is S F,−1 .
To use the latter result in inductive proofs, it will be helpful to have the following lemma, in which
we write w (λ) for the weight of a partition λ.
Lemma 3.15. Suppose λ is a partition and i ∈ {0, 1}. Then w (λ+i ) w (λ).
Proof. A much more general result is proved in [7, Lemma 3.6]. 2
4. Homomorphisms
In this section we recall some results concerning the existence of homomorphisms between per-
mutation modules and Specht modules, and we use these to construct homomorphisms in particular
cases; we also recall the second author’s analogue of a special case of the Carter–Payne theorem for
homomorphisms between Specht modules. In conjunction with Corollary 3.12, these results will be
useful for proving reducibility of Specht modules.
4.1. Homomorphisms from permutation modules to Specht modules
Suppose μ is a composition of n. A μ-tableau is a function T from [μ] to Z0 ; given a tableau T ,

we write T (i , j ) instead of T ((i , j )), and we usually depict T by drawing the Young diagram [μ], and
filling each node with its image under T . Given i , j 1, we write T i , j for the number of entries equal
to j in row i of T . If λ is another composition of n, then we say that a μ-tableau T has content λ if
for every i there are exactly λi nodes mapped to i by T .
Example. Let μ = (5, 3, 1) and λ = (4, 3, 2). Then the tableau
T=
is a μ-tableau of content λ. The values T i, j are given by the following matrix:
1 2 3
1 3 0 2
.
2 1 2 0
3 0 1 0
μ
For each μ-tableau T of content λ, Dipper and James define a homomorphism Θ T : M F,q → M F, λ ,
q
for any F, q. The homomorphisms Θ T and ΘU are equal if T and U are row equivalent; that is, U i , j =
T i , j for each i , j. We say that T is row-standard if the entries in T are weakly increasing along rows,
and we write T (μ, λ) for the set of row-standard μ-tableaux of content λ. Dipper and James prove
that the set

Θ T T ∈ T (μ, λ)
λ . μ
is a basis for the space of homomorphisms from M F,q to M F,q
A particular set of homomorphisms Θ T can be used to give a convenient characterisation of the
Specht module. Suppose λ is a partition, and suppose d, t 1 are such that t λd+1 . Define the
composition λd,t by
λd + t (i = d),
λdi ,t = λd+1 − t (i = d + 1),
λi (otherwise).
Then there is a unique tableau A ∈ T (λ, λd,t ) with the property that A (i , j ) = i for all (i , j ) ∈ [λ] with
d,t
i = d + 1. We write ψ d,t for the homomorphism Θ A : M F,
λ → M λ . (Warning: in [20] and elsewhere,
q F,q
the map ψ d,t is written as ψd,λd+1 −t . Our convention is more convenient here.)
Now the Specht module can be characterised as follows.
Theorem 4.1. (See [4, Theorem 7.5].) Suppose λ is a partition. Then

λ
S F,q = ker ψ d,t .
d1 1t λd+1
This theorem is known as the kernel intersection theorem, and is very useful for constructing and
classifying homomorphisms M → S F,
λ , when M is a module for which one knows all homomorphisms
q
M → M F,λ ; specifically, the homomorphisms M → S λ are precisely the homomorphisms Θ : M →
q F,q
M F,q such that ψ d,t ◦ Θ = 0 for all d, t. This approach has been particularly exploited by the second
λ
author, using an explicit description of the composition ψ d,t ◦ Θ T , when T ∈ T (μ, λ). Before we can
give this result, we need to give a very brief account of quantum binomial coefficients. For any non-
negative integer a, we define the quantum integer [a] = 1 + q + q2 +· · · + qa−1 , and the quantum factorial
[a]! = [1][2] . . . [a]. Now for 0 b a the quantum binomial coefficient is defined to be

a [a]!
= .
b [b]![a − b]!
a
Of course, if q = 1 then this coincides with the usual binomial coefficient b .
The only property of quantum binomial coefficients we need is the following, which is well known.
Lemma 4.2. Suppose q = −1, and 0 b a. Then

a/2
a (if a is odd or b is even),
= b/2
b 0 (if a is even and b is odd).
Suppose μ and λ are two partitions, and d, t are chosen as above. Given T ∈ T (μ, λ), let V T ⊆
T (μ, λd,t ) be the set of row-standard tableaux V with the property that for each (i , j ) ∈ [μ] either
V (i , j ) = T (i , j ) or V (i , j ) = d = T (i , j ) − 1. (In other words, V is a row-standard tableau obtained
from T by replacing t of the entries equal to d + 1 with ds.)
Given V ∈ V T , define

x= ( V i ,d − T i ,d ) T k,d ,
i 1 k >i
and set
(q)
V
x i ,d
bT V = q ,
T i ,d
i 1
considered as an element of the field F. Now we have the following statement.
Proposition 4.3. (See [20, Proposition 2.14].) Suppose λ, μ, T , d, t are as above. Then
(q)
ψ d,t ◦ ΘT = b T V ΘV .
V ∈V T
We shall use this to prove the following proposition.
Proposition 4.4. Suppose ν and ξ are partitions; put l = ν1 , and suppose that ξl−1 l. Define partitions λ, μ
by
λi = ξi + 2νi ,
μi = ξi + 2νi
μ
for all i 1. Then there is a non-zero H-homomorphism from M F,−1 to S F,−
λ
1.
Example. Put ξ = (24 ) and ν = (12 ). Then we have l = 2, so that ξl−1 l, and so when q = −1
μ
1 , where λ = (4 , 2 ) and μ = (6 ). We have
λ 2 2 2
there is a non-zero homomorphism from M F,−1 to S F,−
2 2
(4 ,2 )
λreg = (5, 4, 2, 1), which does not dominate μ, and so by Corollary 3.12 we deduce that S F,− 1 is
reducible.
Proof of Proposition 4.4. Using the kernel intersection theorem, we need to construct a linear com-
bination

θ= c T ΘT
T ∈T (μ,λ)
such that ψ d,t ◦ θ = 0 for all d, t. To do this, we begin with the semi-simple Iwahori–Hecke algebra
HQ = HQ,1 (Sm ), i.e. the group algebra QSm , where m = |ν |. The module M Q, ν
1 contains the Specht
module S Q,1 as a submodule; since HQ is semi-simple, S Q,1 is also a quotient of M Q,
ν ν ν , so there
1
is a non-zero HQ -homomorphism φ : M Q,1 → S Q,1 . Regarding this as a homomorphism from M Q,
ν ν ν
1
to itself, and using the fact that the homomorphisms Θ T for T ∈ T (ν , ν ) span the space of such
homomorphisms, we may write

φ= a T ΘT
T ∈T (ν ,ν )
with a T ∈ Q. By re-scaling, we may assume that the a T are coprime integers.

ν implies that
The fact that the image of φ lies in the Specht module S Q, 1

a T ψ d ,τ ◦ Θ T = 0
T ∈T (ν ,ν )
for all d, τ with 1 τ νd+1 . By Proposition 4.3, this means that for each pair d, τ we have

(1)
aT b T V ΘV = 0.
T ∈T (ν ,ν ) V ∈V T
Since the set {Θ V | V ∈ T (ν , ν d,τ )} is linearly independent, this says that for each V ∈ T (ν , ν d,τ ), the
sum
(1)
aT bT V
T ∈T (ν ,ν )| V ∈V ( T )
vanishes.
μ
Now we construct a homomorphism θ : M F,−1 → M F,− λ λ
1 , whose image we claim lies in S F,−1 . For
each T ∈ U (ν , ν ), let T̂ be the row-standard μ-tableau given by

2T i , j + 1 ( j ξi ),
T̂ i , j =
2T i , j ( j > ξi )
for each i , j. Then we have T̂ ∈ T (μ, λ), and we define

θ= a T Θ T̂ .
T ∈T (ν ,ν )
(We are committing a minor abuse of notation here: by a T , we really mean the image of a T in F,
which is well defined since we are assuming that each a T is an integer.) The tableau T is easily
recovered from T̂ , so the tableaux T̂ are distinct, and therefore the homomorphisms Θ T̂ are linearly
independent; since we assume that the integers a T are coprime, this implies that θ is non-zero.
(Alternatively, one can use the results of Section 3.3 and assume throughout that F = Q.)
Fix a pair d, t with 1 t λd+1 . Using the kernel intersection theorem and Proposition 4.3, our
task is to show that the sum
(−1)
aT b ΘV
T̂ V
T ∈T (ν ,ν ) V ∈V T̂
equals zero.
Claim. Suppose T ∈ T (ν , ν ) and V ∈ V T̂ , and define

βi = j V (i , j ) = T̂ (i , j )
(−1)
as above. If βi is odd for any i, then b = 0.
T̂ V
Proof. Fix i such that βi is odd. Then we claim that T̂ i ,d is odd. This will then imply that the integer
yi (−1)
y i = V i ,d is even, so that β equals zero, and hence b is zero.
i T̂ V
The fact that βi > 0 means that T̂ i ,d+1 > 0. Write l = ν1 as above, and suppose first that i > l. Then
by construction the entries in row i of T̂ are 1, 2, . . . , ξi each occurring once; since d + 1 occurs, we
have d + 1 ξi , so that T̂ i ,d = 1.
Alternatively, suppose that i l. Then we claim that ξi d, which will mean that
T̂ i ,d = 1 + 2T i ,d ,
which is odd. We are given that ξl−1 l, i.e. ξl l − 1, and hence ξi l − 1. So if d l − 1 we are
done. If d > l − 1 then d + 1 > l, so T i ,d+1 = 0; but by assumption there is an entry equal to d + 1 in
row i of T̂ , and hence we must have ξi d + 1. 2
As a consequence of the claim, we need only consider those pairs ( T ∈ T (ν , ν ), V ∈ V T̂ ) for which
βi is even for each i. Given such a pair, this condition implies that t = i βi is even and there is a
unique W ∈ T (ν , ν d,t /2 ) such that V = Ŵ (where we define Ŵ analogously to T̂ ). Furthermore, for
each such V and each T ∈ T (ν , ν ) we have V ∈ V T̂ if and only if W ∈ V T , and if this is the case then
(−1) (1)
it is easy to calculate that b = b T V , using Lemma 4.2. Now the result follows. 2
T̂ V
Example. Let ξ, ν be as in the previous example. The two tableaux in T (ν , ν ) are
1 2
T1 = , T2 = ,
2 1
μ μ
and a non-zero homomorphism from M Q,1 to S Q,1 is given by
φ = ΘT 1 − ΘT 2 .
(62 ) (42 ,22 )

So a homomorphism from M F,−1 to S F,−1 is given by
θ = ΘT̂ 1 − ΘT̂ 2 ,
where
T̂ 1 = , T̂ 2 = .
4.2. Homomorphisms between Specht modules
We now quote a result due to the second author which gives the existence of non-zero homo-
morphisms between Specht modules under certain circumstances. This is a generalisation to Iwahori–
Hecke algebras of a special case of the Carter–Payne theorem [3]. Recall the notion of the residue of
a node from Section 3.8.
Theorem 4.5. (See [20, Theorem 4.1.1].) Suppose λ is a partition, and that λ has an addable node (i , λi + 1) and
a removable node ( j , λ j ) of the same residue, with i < j. Let μ be the partition obtained by adding the node
μ
(i , λi + 1) and removing the node ( j , λ j ). Then there exists a non-zero homomorphism from S F,−1 to S F,−
λ
1.
This result will be helpful in conjunction with Corollary 3.12. A particular application is the fol-
lowing.
Proposition 4.6. Suppose λ is a partition. Suppose λ has
• an addable node (i , λi + 1) lying in ladder Lm , and

• a removable node ( j , λ j ) lying in ladder Ll ,
where m > l and l ≡ m (mod 2). Then S F,−

λ
1 is reducible.
Proof. By replacing λ with its conjugate if necessary and appealing to Corollary 3.3, we may assume
that i < j. Since l ≡ m (mod 2) and the nodes in ladder Ll all have residue (l + 1) (mod 2), the
addable node (i , λi + 1) and the removable node ( j , λ j ) both have the same residue. So if we define
μ λ
μ as in Theorem 4.5, then there is a non-zero homomorphism from S F,− 1 to S F,−1 . By Corollary 3.12,
it suffices to show that λreg S μreg ; this follows from [5, Lemma 2.1], given the assumption that
m > l. 2
5. The Fock space and canonical bases
Now we introduce the Fock space, which is our most powerful tool. In fact, via Ariki’s Theorem,
this theory provides an algorithm for computing the decomposition matrix of HF,−1 (Sn ) completely
when char(F) = ∞. However, it does not seem easy to use this algorithm to decide the reducibility of
Specht modules, and our application of the Fock space will be less direct.
Let v be an indeterminate over Q, and let h = Qh0 ⊕ Qh1 ⊕ Q D be a three-dimensional vector
space. In this section we work with the quantum group U = U v ( sl2 ), which may be realised as the
associative algebra over Q with generators e 0 , e 1 , f 0 , f 1 and v h (h ∈ h), subject to well-known rela-
tions; these may be found in [18], which is an excellent background reference for this section.
Define the Fock space to be the Q( v )-vector space F with a basis {s(λ)} indexed by the set of
all partitions. Let , be the inner product on F for which the basis {s(λ)} is orthonormal. The Fock
space has the structure of a U -module, and has important connections to the representation theory of
Iwahori–Hecke algebras. It will suffice for our purposes to describe the action on F of the ‘negative’
generators f 0 , f 1 and their ‘quantum divided powers’
(a) f ia
fi = .
v a−1 + v a−3 + · · · + v 3−a + v 1−a
a :i
Fix i ∈ {0, 1} and a 1, and suppose λ and μ are partitions. Write λ −→ μ if the Young diagram for
μ may be obtained from the Young diagram for λ by adding a addable nodes of residue i. If this is
the case, then for each j such that λ j = μ j define

+1 (res( j , λ + 1) = i ),
j = −1 (res( j , λ j + 1) = 1 − i ),
j
and set

N (λ, μ) = j × no. of nodes of [μ] \ [λ] below row j .
j |λ j =μ j
(a)
Now the action on F of the quantum divided power f i is given by Q( v )-linear extension of
(a)

fi s(λ) = v N (λ,μ) s(μ).
a :i
μ|λ−→μ
Of particular importance is the submodule of F generated by the vector s(∅). This submodule
is isomorphic to (and therefore often identified with) the irreducible highest-weight representation
M (Λ0 ) of U . This representation is equipped with a Q( v + v −1 )-linear map called the bar involution,
which can be specified by the conditions
s(∅) = s(∅)
and
f i (m) = f i (m)
for i ∈ {0, 1} and m ∈ M (Λ0 ).

The bar involution allows us to define the canonical basis of M (Λ0 ), via the following theorem.
Theorem 5.1. For each 2-regular partition μ there is a unique element G (μ) of M (Λ0 ) with the properties
• G (μ) = G (μ), and

• G (μ) = λ dλμ ( v )s(λ), where dλμ ( v ) is a polynomial in v, with dμμ ( v ) = 1, and dλμ ( v ) divisible by v
for λ = μ.
The set

G (μ) μ a 2-regular partition
is a Q( v )-basis of M (Λ0 ).
Now we can state (a special case of) Ariki’s Theorem, which gives the connection to the represen-
tation theory of Iwahori–Hecke algebras.
Theorem 5.2. (See [1, Theorem 4.4].) Suppose λ and μ are partitions of n, with μ 2-regular, and let dλμ ( v ) =
G (μ), s(λ) as in Theorem 5.1. Then
μ
1 : D Q,−1 = dλμ (1).
λ
S Q,−
In view of this theorem, the polynomials dλμ ( v ) are known as ‘v-decomposition numbers’. It is
known [21, Theorem 6.28] that dλμ ( v ) has non-negative integer coefficients, and is zero unless λ and
μ have the same core and weight (and therefore the same size). The non-negativity of the coefficients
has the following obvious consequence, in conjunction with Theorems 3.11 and 5.2.
Lemma 5.3. Suppose λ and μ are partitions of n, with μ 2-regular.
μ
1. If [ S F,−
λ
1 : D F,−1 ] = 0, then dλμ ( v ) = 0.
μ
2. If [ S F,−
λ
1 : D F,−1 ] = 1 (in particular, if μ = λ
reg
), then dλμ ( v ) = v s for some s.
Remark. In the case where μ = λreg , the integer s in Lemma 5.3 has been computed explicitly by the
first author in [8]; however, we shall not need this result in the present paper.
Next we prove a crucial lemma, which enables us to use Fock space computations to prove re-
ducibility of Specht modules.
Lemma 5.4. Suppose λ is a partition, and suppose X and Y are bar-invariant elements of M (Λ0 ), such that

X , s(λ) = v x , Y , s(λ) = v y
for some x = y. Then the Specht module S Q,−

λ
1 is reducible.
Proof. Suppose not, and write ν = λreg . Then by Theorem 3.11 and Lemma 5.3 we have G (μ), s(λ) =
0 for any μ = ν . We write X and Y as linear combinations of canonical basis vectors

X= αμ ( v )G (μ), Y= βμ ( v )G (μ);
μ μ
since X and Y are bar-invariant, αμ ( v ) and βμ ( v ) lie in Q( v + v −1 ) for each μ. Taking inner products
with s(λ) yields
αν ( v )dλν ( v ) = v x , βν ( v )dλν ( v ) = v y .
This gives
v y αν ( v ) = v x βν ( v )
with αν ( v ) and βν ( v ) non-zero, but this is impossible if x = y and αν ( v ) and βν ( v ) lie in Q( v +

v −1 ). 2
Remark. In fact, Lemma 5.4 shows something rather stronger than the reducibility of S F,− λ
1 . Let us say
that λ is homogeneous if there is some x such that every v-decomposition number dλμ ( v ) is either
zero or a monomial of degree x. According to popular conjectures relating v-decomposition numbers
to the Jantzen filtration of the Specht module, the homogeneity of λ ought to imply that the Specht
λ
module S Q,− 1 is completely reducible.
A weaker condition we might impose on λ is that it is quasi-homogeneous, meaning that there is
some x such that every dλμ ( v ) lies in v x .Q( v + v −1 ). As a representation-theoretic interpretation, we
would speculate that quasi-homogeneity corresponds to the Specht module S F,− λ
1 being self-dual.
What we have actually shown is that if the hypotheses of Lemma 5.4 are satisfied, then λ is not
homogeneous or even quasi-homogeneous. It would be very interesting to classify homogeneous and
quasi-homogeneous partitions, and the authors hope to be able to say something more about this in
the future.
Our next step is to prove a result in which we phrase the action of a certain composition of powers
of f 0 , f 1 in a convenient form, in certain special cases. Fix partitions μ and λ with [μ] ⊆ [λ], and fix
x ∈ {0, 1}. Suppose that the following condition holds:
for each i with λi > 0, the node (i , μi + 1) has residue x. (∗)
In other words, μi ≡ i + x (mod 2) whenever λi > 0. Note that this implies in particular that λ1 −
μ1 1.
Define a sequence of partitions μ = μ0 , μ1 , μ2 , . . . as follows: for j 0, μ j +1 is obtained from
μ j by adding all addable nodes that are contained in [λ]. We define a j = |μ j | − |μ j−1 | for j 1, and
then set
(a ) (a3 ) (a2 ) (a1 )

f = · · · f 1−4x f x f 1− x f x ∈ U.
(For j sufficiently large we have μ j = λ, and so a j = 0 for large j, so this definition makes sense.)
Our objective is to compute the coefficient of s(λ) in f s(μ). To do this, we construct a λ-tableau T
by filling each node of [μ] with a 0, and then for j 1, filling each node of [μ j ] \ [μ j −1 ] with a j.
Given a node (k, h) ∈ [λ], let j = T (k, h), and set

N (k, h) = m < k T (m, λm ) < j , T (m, λm ) ≡ j (mod 2)

− m < k T (m, λm ) < j , T (m, λm ) ≡ j (mod 2) .

Finally, set N = (k,h)∈[λ] N (k, h ). Now we have the following.
Lemma 5.5. With the above definitions, we have

f s(μ), s(λ) = v N .
Proof. The hypothesis on μ means that for any i, the nodes (i , μi + 1), (i , μi + 2), . . . , (i , λi ) are filled
with the integers 1, 2, . . . , λi − μi in T . In particular, for each j 1, the nodes (k, h) with T (k, h) = j
all have residue x + j (mod 2). Now the lemma is straightforward to prove, given the above formula
(a) (a)
for the actions of f 0 , f 1 on F . 2
Example. Set λ = (72 , 52 , 4), μ = (7, 6, 3, 2, 1), x = 1. Then we have
μ0 = μ,

μ1 = 72 , 4, 3, 2 ,

μ2 = 72 , 5, 4, 3 ,
μ j = λ for j 3,
and
T= .
The values of N (k, l) are given by
so that
(2) (3) (4)

f1 f0 f 1 s (7, 6, 3, 2, 1) , s 72 , 52 , 4 = v6.
We now use the last two results to prove the following proposition, which covers the bulk of cases
in our main theorem.
Proposition 5.6. Suppose λ is a partition satisfying the hypothesis of Theorem 2.1. Let l = λ1 , and suppose
that for i = 1, . . . , l we have λi l − i + 2. Suppose also that for 1 k λ1 the ladder Lk (λ) is connected.
λ
Then S F,− 1 is reducible.
Proof. By the results of Section 3.3, we may assume that F = Q. The hypothesis of Theorem 2.1
is that some ladder Lm (λ) is disconnected; take the smallest such m, and choose i such that
(i − 1, m + 2 − i ) ∈ [λ]
/ (i , m + 1 − i ) and ( j , m + 1 − j ) ∈ [λ] for some j > i. Now define
μ = (l + 1, l, l − 1, . . . , 3, 2),
μ̂ = (l + 1, l, l − 1, . . . , l − i + 3, l − i , l − i − 1, . . . , 2, 1).
Setting x = l + 1 (mod 2), we find that μ and μ̂ both satisfy (∗); we define the operator f and the
tableau T corresponding to (λ, μ, x) as above, we define f̂ and T̂ corresponding to (λ, μ̂, x) in the
same way. Our aim is to show that the hypotheses of Lemma 5.4 are satisfied, with
X = f G (μ), Y = f̂ G (μ̂).
Certainly X and Y are bar-invariant elements of the Fock space. We now claim that we can ignore all
terms in G (μ), G (μ̂) except the leading terms, i.e.

X , s(λ) = f s(μ), s(λ) , Y , s(λ) = f̂ s(μ̂), s(λ) .
Note that μ is an alternating partition with core (l − 1, l − 2, . . . , 1); so by Corollary 3.10 and The-
orem 5.2, any ν = μ which gives a non-zero term dνμ ( v )s(ν ) in G (μ) satisfies νl+1 > 0. But this
means that [ν ] [λ], which obviously implies that f s(ν ), s(λ) = 0. The same argument applies to
G (μ̂), and so we can concentrate on f s(μ) and f s(μ̂). If we define the integer N corresponding to T
as above, and define N̂ from T̂ analogously, then by Lemma 5.5 we have

f s(μ), s(λ) = v N , f̂ s(μ̂), s(λ) = v N̂ ,
and it remains to prove the purely combinatorial statement that N̂ = N.

In fact, we shall estimate N̂ − N, and show that it is strictly positive. To do this, we compare
N (k, h) with N̂ (k, h) for the various nodes (k, h) ∈ [λ]. It will help to introduce some notation: for any
j 0, we let a j be the number of k ∈ {1, . . . , i − 1} such that T (k, λk ) = j; that is, the number of rows
of T above row i ending in j . We also define b j to be the number of nodes (k, h) with k i such
that T (k, h) = j.
First we note that T and T̂ agree on rows 1, . . . , i − 1, so certainly for any node (k, h) with k < i
we have N̂ (k, h) = N (k, h). So we concentrate on nodes in rows i and below. Consider first the nodes
lying in [μ] \ [μ̂]. There are two of these in each row from i to l; these are labelled 0 0 in T , and
1 2 in T̂ . By the definitions of N and N̂, each such pair of nodes contributes nothing to N, and
contributes a1 to N̂ (the node 1 contributes a0 , while the node 2 contributes −a0 + a1 ).
Next let (k, h) be a node of [λ] \ [μ] with k i, and let j = T (k, h) = k + h − l − 2. Then T̂ (k, h) =
j + 2, and using the definitions we can compute
N̂ (k, h) − N (k, h) = a j +1 − a j
(note that the rows from i to k − 1 make no contribution). Summing, we find that
N̂ − N = (l − i + 1)a1 + (a2 − a1 )b1 + (a3 − a2 )b2 + · · ·
= (l − i + 1 − b1 )a1 + (b1 − b2 )a2 + (b2 − b3 )a3 + · · · .
We claim that each term of the latter sum is non-negative, and that some term is positive. Certainly
b1 l − i + 1, since there can be at most one node labelled 1 in any of rows i , . . . , l. Also, b j b j −1
for j 2, since each node labelled j must have a node labelled j − 1 immediately to its left. So each
term of the sum is non-negative. Now let j = T (i − 1, m − i + 2); that is, j = m − l − 1. If j 2, we
claim that (b j −1 − b j )a j > 0; otherwise we claim (l − i + 1 − b1 )a1 > 0. Suppose j 2. To see that
the factor b j −1 − b j is positive, note that there is a node labelled j − 1 at the end of row i, but no
node labelled j in this row; in any subsequent row, if there is a node labelled j then there is a node
labelled j − 1 immediately to its left. Similarly if j = 1, there is no node in row i labelled 1, and so
l − i + 1 > b1 .
Now we show that a j > 0. The last hypothesis of the proposition implies that ladder m does not
meet row 1, so that T (1, λ1 ) < j; on the other hand, there is a node labelled j in row i − 1, so
T (i − 1, λi −1 ) j. For any 1 < k i − 1 it is easy to see that T (k, λk ) T (k − 1, λk−1 ) + 1 so every
value from T (1, λ1 ) to T (i − 1, λi −1 ) occurs as some T (k, λk ) for 1 k i − 1. In particular, the value
j occurs, and we are done. 2
6. Proof of Theorem 2.1
Proposition 6.1. Suppose λ is a partition and l 1. Suppose that (1, l) and (l, 1) both lie in Ll (λ), but that
Ll (λ) is disconnected. Then S F,−
λ
1 is reducible.
Proof. Suppose λ has weight w and core (r , r − 1, . . . , 1); we proceed by induction on w, and for
fixed w by reverse induction on r. The starting case for this induction is where r w − 1, so that λ
lies in a Rouquier block. Since λ is certainly neither 2-regular nor 2-restricted, Corollary 3.7 gives the
result in this case.
For the case where r < w − 1, recall the definition of the partition λ+i for i ∈ {0, 1} from Sec-
tion 3.8. By Lemma 3.15, we have w (λ+i ) w (λ), and obviously if w (λ+i ) = w (λ) but λ+i = λ then
the core of λ+i is larger than the core of λ. So to complete the inductive step is suffices to show that
for either i = 0 or i = 1 we have λ+i = λ with λ+i also satisfying the hypotheses of the lemma.
Let j = l (mod 2). Consider two cases.
• Suppose λ has at least one addable node of residue j. Then we have λ+ j = λ, and (since all of
the nodes in Ll have residue 1 − j) we have Ll (λ+ j ) = Ll (λ), so λ+ j satisfies the hypotheses of
the lemma.
• Alternatively, suppose λ has no addable nodes of residue j. Then the nodes (1, l + 1) and (l + 1, 1)
must lie in [λ] (since these nodes have residue j, and if either of them were not contained in [λ]
then it would be an addable node). On the other hand, [λ] cannot contain all the nodes in Ll+1
since it does not contain all the nodes in Ll ; so ladder Ll+1 is disconnected. Now we can replace
l with l + 1 (and j with 1 − j), and appeal to the previous case. 2
Proposition 6.2. Suppose λ is a partition and l 1. Suppose the ladder Ll (λ) is disconnected, and that (1, l) ∈
[λ] λ
/ (l, 1). Then S F,−1 is reducible.
Proof. Divide Ll (λ) into ‘segments’ of consecutive nodes; the condition that Ll (λ) is disconnected
is precisely the statement that there are at least two segments. Let s be the length of the shortest
segment other than the segment containing the node (1, l), and proceed by induction on s. Let j =
l + 1 (mod 2) be the common residue of the nodes in Ll , and let λ+ = λ+(1− j ) be the partition defined
in Section 3.8.
Suppose s = 1. This means that there is some i ∈ {3, . . . , l − 1} such that (i , l + 1 − i ) is a node
of λ but neither (i − 1, l + 2 − i ) nor (i + 1, l − i ) is. In particular, this implies that (i , l + 1 − i ) is
a removable node of λ. We claim that this node is also a removable node of λ+ : since neither of
the nodes (i − 1, l + 2 − i ), (i + 1, l − i ) lies in [λ] by assumption, neither of the nodes (i , l + 2 − i ),
(i + 1, l + 1 − i ) can lie in [λ+(1− j ) ]; so (i , l + 1 − i ) is a removable node of λ+ , as claimed. Now
consider the addable node a = (1, λ+ +
1 + 1) of λ . Since λ
+ cannot have addable nodes of residue
1 − j, a must have residue j. Moreover, since (1, l) ∈ [λ], a lies in ladder Lm for some m > l. Hence
λ+
by Proposition 4.6 S F,− λ
1 is reducible, and so by Lemma 3.14 S F,−1 is reducible.
Now we consider the inductive step; suppose s > 1, and that

(i , l + 1 − i ), (i + 1, l − i ), . . . , (i + s − 1, l + 2 − i − s)
is a segment of length s, with 3 i l − s. That is, the nodes listed above are nodes of λ, but the
nodes (i − 1, l + 2 − i ), (i + s, l + 1 − i − s) are not. This means that the nodes
(i + 1, l + 1 − i ), (i + 2, l − i ), . . . , (i + s − 1, l + 3 − i − s)
are either nodes or addable nodes of λ, and hence (since they have residue 1 − j) are nodes of λ+ .
On the other hand, neither of the nodes (i , l + 2 − i ), (i + s, l + 2 − i − s) is a node or an addable
node of λ, so neither of these nodes is a node of λ+ . So Ll+1 (λ+ ) includes a segment of length s − 1;
it also contains the node (1, l + 1) but not the node (l + 1, 1), and so we may apply the inductive
λ+
hypothesis, replacing λ with λ+ and l with l + 1, to deduce that S F,− 1 is reducible. Now we can apply
Lemma 3.14. 2
Example. Let λ = (5, 32 , 2). Then λ satisfies the hypotheses of Proposition 6.2, with l = 5. We have
s = 2 and j = 0, and we examine the partition λ+ = λ+1 = (6, 33 ). This partition also satisfies the
hypotheses of Proposition 6.2, with l = 6 and s = 1. We construct the partition λ++ = (λ+ )+0 =
(7, 4, 32 , 1); this satisfies the hypotheses of Proposition 4.6, since it has an addable node (1, 8) ∈ L8
λ++ is reducible, and hence so is S λ+ , and hence so is S λ
and a removable node (4, 3) ∈ L6 . So S F,− 1 F,−1 F,−1 .
The Young diagrams of the partitions, with the residues of their nodes marked, are given below.
λ λ+ λ++
Proposition 6.3. Suppose λ satisfies the hypothesis of Theorem 2.1, that λ1 = λ1 , and that λi λ1 + 1 − i for
i = 1, . . . , λ1 . Then S F,−
λ
1 is reducible.
Proof. Using Proposition 4.4 and Corollary 3.12, our task is to show that we can find partitions ν , ξ
such that ξν −1 ν1 , λi = ξi + 2νi for all i, and λreg S μ, where μ is the partition given by μi =
1
ξi + 2νi for all i.
Let m be minimal such that Lm (λ) is disconnected, and let i , k be such that k i + 2, (i , m +
1 − i ), (k, m + 1 − k) lie in [λ] and none of the nodes (i + 1, m − i ), . . . , (k − 1, m + 2 − k) lies in [λ].
We would like to assume that i + 1 m − i, i.e. the node (i + 1, m − i ) lies on or above the main
diagonal of the Young diagram. If this is not the case, then we can replace λ with λ (appealing to
Corollary 3.3), and replace (m, i , k) with (m, ı̄, k̄), where ı̄ = m + 1 − k, k̄ = m + 1 − i; it is then easy
to check that ı̄ + 1 m − ı̄ .
So we shall assume that i + 1 m − i. Since (i , m + 1 − i ) ∈ [λ]
/ (i + 1, m − i ), we have λi − λi +1 2.
So we if we define ν = (1i ) and ξ = (λ1 − 2, λ2 − 2, . . . , λi − 2, λi +1 , λi +2 , . . .), then ξ is a partition.
Furthermore, we have
ξν1 −1 = ξi −1 = λi −1 − 2 λi − 2 m − 1 − i i = ν1 ,
and it remains to show that λreg S μ. In fact, we shall show that μ1 < (λreg )1 , which is certainly good
enough. By assumption Lλ1 (λ) = Lλ1 , and hence Lλ1 (λreg ) = Lλ1 , and this means that (λreg )1 = λ1 .
On the other hand,

μ1 = max ξ1 , ν1 = max{λ1 − 2, i } < λ1 ,
and we are done. 2
Proof of Theorem 2.1. Suppose λ satisfies the hypothesis of Theorem 2.1. By replacing λ with λ if
necessary and appealing to Corollary 3.3, we may assume that λ1 λ1 .
Let m be minimal such that ladder Lm (λ) is disconnected. If m λ1 , then Lm (λ) meets the top
row of [λ] (i.e. (1, m) ∈ [λ]), and so we may appeal to Proposition 6.1 or Proposition 6.2. So we can
assume that m > λ1 , and in particular no disconnected ladder of λ meets the top row.
If λ1 = λ1 , then we may appeal to Proposition 6.3, so instead we suppose that λ1 > λ1 . Let l = λ1 ,
and suppose that λl 2. Then we have (l, 2) ∈ [λ]. Since λ1 > l, we also have (1, l + 1) ∈ [λ], and now
the assumption that Ll+1 (λ) is connected means that the nodes (2, l), (3, l − 1), . . . , (l − 1, 3) all lie
in λ. So λi l − i + 2 for i = 1, . . . , l, and we may appeal to Proposition 5.6.
We are left with the case where λl = 1. In this case, λ has a removable node (l, 1), of residue
j = l + 1 (mod 2). We now consider two cases.
• Suppose the nodes in ladder Lm have residue j. We claim that there is an addable node of λ in
this ladder. Indeed, let i , k be such that k i + 2, (i , m + 1 − i ), (k, m + 1 − k) lie in [λ] and none of
the nodes (i + 1, m − i ), . . . , (k − 1, m + 2 − k) lies in [λ]. Then (i , m − i ) and (k − 1, m + 1 − k) lie
in [λ], and since Lm (λ) is the first disconnected ladder of λ, the nodes (i + 1, m − 1 − i ), . . . , (k −
2, m + 2 − k) lie in [λ]. So λ has an addable node (i + 1, m − i ) ∈ Lm . Now we may appeal to
Proposition 4.6.
• Alternatively, suppose the nodes in ladder Lm have residue 1 − j, and consider the partition λ− j .
This is strictly smaller than λ since it does not contain the node (l, 1), but also has a disconnected
ladder, i.e. Lm (λ− j ) = Lm (λ). So by induction on |λ| and Lemma 3.13, S F,−
λ
1 is reducible. 2
Acknowledgments
The first author was financially supported by a Research Fellowship from the Royal Commission
for the Exhibition of 1851; he is very grateful to the Commission. This work was carried out while
the first author was a visiting Postdoctoral Fellow at the Massachusetts Institute of Technology; he is
grateful to Richard Stanley for the invitation, and to M.I.T. for its hospitality.
Conjecture 2.2 in this paper was made while the first author was working with Andrew Mathas at
the University of Sydney in 2004; the first author is very grateful for the invitation.
Some of this work was carried out at MSRI Berkeley in March 2008, during the concurrent pro-
grammes ‘Combinatorial representation theory’ and ‘Representation theory of finite groups and related
topics’. Both authors acknowledge generous financial support from MSRI, and wish to thank the or-
ganisers of these excellent programmes.
References
[1] S. Ariki, On the decomposition numbers of the Hecke algebra of G (m, 1, n), J. Math. Kyoto Univ. 36 (1996) 789–808.
[2] J. Brundan, A. Kleshchev, Representation theory of the symmetric groups and their double covers, in: Groups, Combinatorics
& Geometry, Durham, 2001, World Sci. Publishing, River Edge, NJ, 2003, pp. 31–53.
[3] R. Carter, M. Payne, On homomorphisms between Weyl modules and Specht modules, Math. Proc. Cambridge Philos. Soc. 87
(1980) 419–425.
[4] R. Dipper, G. James, Representations of Hecke algebras of general linear groups, Proc. London Math. Soc. (3) 52 (1986)
20–52.
[5] M. Fayers, Reducible Specht modules, J. Algebra 280 (2004) 500–504.
[6] M. Fayers, Irreducible Specht modules for Hecke algebras of type A, Adv. Math. 193 (2005) 438–452.
[7] M. Fayers, Weights of multipartitions and representations of Ariki–Koike algebras, Adv. Math. 206 (2006) 112–144 (Cor-
rected version: www.maths.qmul.ac.uk/~mf/papers/weight.pdf).
[8] M. Fayers, q-analogues of regularisation theorems for linear and projective representations of the symmetric group, J. Al-
gebra 316 (2007) 346–367.
[9] W. Fulton, Young Tableaux, with Applications to Representation Theory and Geometry, London Math. Soc. Stud. Texts,
vol. 35, Cambridge Univ. Press, 1997.
[10] M. Geck, Brauer trees of Hecke algebras, Comm. Algebra 20 (1992) 2937–2973.
[11] M. Geck, Representations of Hecke algebras at roots of unity, Astérisque 252 (1998) 33–55.
[12] G. James, On the decomposition matrices of the symmetric groups II, J. Algebra 43 (1976) 45–54.
[13] G. James, The decomposition matrices of GLn (q) for n 10, Proc. London Math. Soc. (3) 60 (1990) 225–265.
[14] G. James, S. Lyle, A. Mathas, Rouquier blocks, Math. Z. 252 (2006) 511–531.
[15] G. James, A. Mathas, Hecke algebras of type A with q = −1, J. Algebra 184 (1995) 102–158.
[16] G. James, A. Mathas, A q-analogue of the Jantzen–Schaper theorem, Proc. London Math. Soc. (3) 74 (1997) 241–274.
[17] G. James, A. Mathas, The irreducible Specht modules in characteristic 2, Bull. London Math. Soc. 31 (1999) 457–462.
[18] A. Lascoux, B. Leclerc, J.-Y. Thibon, Hecke algebras at roots of unity and crystal bases of quantum affine algebras, Comm.
Math. Phys. 181 (1996) 205–263.
[19] S. Lyle, Some reducible Specht modules, J. Algebra 269 (2003) 536–543.
[20] S. Lyle, Some q-analogues of the Carter–Payne theorem, J. Reine Angew. Math. 608 (2007) 93–121.
[21] A. Mathas, Iwahori–Hecke Algebras and Schur Algebras of the Symmetric Group, Univ. Lecture Ser., vol. 15, Amer. Math.
Soc., Providence, RI, 1999.
Journal of Algebra
Anti-affine algebraic groups

Michel Brion
Université de Grenoble I, Département de Mathématiques, Institut Fourier, UMR 5582 du CNRS, 38402 Saint-Martin d’Hères Cedex, France
Article history: We say that an algebraic group G over a field is anti-affine if

Received 2 July 2008 every regular function on G is constant. We obtain a classification
Available online 1 November 2008 of these groups, with applications to the structure of algebraic
Communicated by Peter Littelmann
groups in positive characteristics, and to the construction of many
Keywords:
counterexamples to Hilbert’s fourteenth problem.
Group scheme © 2008 Elsevier Inc. All rights reserved.
Algebraic group
Albanese morphism
Hilbert’s fourteenth problem
0. Introduction
We say that a group scheme G of finite type over a field k is anti-affine if O (G ) = k; then G is
known to be smooth, connected and commutative. Examples include abelian varieties, their universal
vector extensions (in characteristic 0 only) and certain semi-abelian varieties.
The classes of anti-affine groups and of affine (or, equivalently, linear) group schemes play comple-
mentary roles in the structure of group schemes over fields. Indeed, any connected group scheme G,
of finite type over k, has a smallest normal subgroup scheme G ant such that the quotient G /G ant is
affine. Moreover, G ant is anti-affine and central in G (see [DG70]). Also, G has a smallest normal con-
nected affine subgroup scheme G aff such that G /G aff is an abelian variety (as follows from Chevalley’s
structure theorem, see [BLR90]). This yields the Rosenlicht decomposition: G = G aff G ant and G aff ∩ G ant
contains (G ant )aff as an algebraic subgroup of finite index (see [Ro56]).
Affine group schemes have been extensively investigated, but little seems to be known about their
anti-affine counterparts; they only appear implicitly in work of Rosenlicht and Serre (see [Ro58,Ro61,
Se58a]). Here we obtain some fundamental properties of anti-affine groups, and reduce their structure
to that of abelian varieties.
In Theorem 2.7, we classify anti-affine algebraic groups G over an arbitrary field k with separable
closure k s . In positive characteristics, G is a semi-abelian variety, parametrized by a pair ( A , Λ) where
A is an abelian variety over k, and Λ is a sublattice of A (k s ), stable under the action of the Galois
E-mail address: michel.brion@ujf-grenoble.fr.
doi:10.1016/j.jalgebra.2008.09.034
M. Brion / Journal of Algebra 321 (2009) 934–952 935
group. The classification is a bit more complicated in characteristic 0: the parameters are then triples
( A , Λ, V ) where A and Λ are as above, and V is a subspace of the Lie algebra of A. In both cases, A
is the dual of the abelian variety G /G aff .
We illustrate this classification by describing the universal morphisms from anti-affine varieties to
commutative algebraic groups, as introduced by Serre (see [Se58a,Se58b]).
Together with the Rosenlicht decomposition, our classification yields structure results for several
classes of group schemes. As a first consequence, any connected commutative group scheme over a
perfect field k is the almost direct product of an anti-affine group, a torus, and a connected unipotent
group scheme (see Theorem 3.4 for a precise statement).
In another direction, if the ground field k is finite, then any anti-affine group is an abelian variety.
This gives back a remarkable result of Arima: any connected group scheme over a finite field has the
decomposition G = G aff G ab where G ab is the largest abelian subvariety of G; moreover, G aff ∩ G ab is
finite (see [Ar60,Ro61]).
Arima’s result does not extend to (say) uncountable and algebraically closed fields, as there ex-
ist semi-abelian varieties that are anti-affine but non-complete. Yet we obtain a structure result for
connected algebraic groups over perfect fields of positive characteristics, namely, the decomposition
G = H S where H ⊂ G aff denotes the smallest normal connected subgroup such that G aff / H is a torus,
and S ⊂ G is a semi-abelian subvariety; moreover, H ∩ S is finite (Theorem 3.7).
Our classification also has rather unexpected applications to Hilbert’s fourteenth problem. In its
algebro-geometric formulation, it asks if the coordinate ring of every quasi-affine variety is finitely
generated (see [Za54], and [Win03] for the equivalence with the invariant-theoretic formulation). The
answer is known to be negative, the first counterexample being due to Rees (see [Re58]). Here we
obtain many counterexamples, namely, all Gm -torsors associated to ample line bundles over anti-
affine, non-complete algebraic groups (Theorem 3.9).
Some of the preceding statements bear a close analogy to known results on complex Lie groups.
Specifically, any connected complex Lie group G has a smallest closed normal subgroup G tor such that
the quotient G /G tor is Stein. Moreover, G tor is connected and central in G, and every holomorphic
function on G tor is constant. The latter property defines the class of toroidal complex Lie groups,
also known as Cousin groups, or quasi-tori, or (H.C) groups. Toroidal groups may be parametrized
by pairs ( T , Λ) where T is a complex torus, and Λ is a sublattice of the dual torus. Any connected
commutative complex Lie group admits a unique decomposition G = G tor ×(C∗ )m ×Cn (see the survey
[AK01] for these results). Yet this analogy is incomplete, as Chevalley’s structure theorem admits no
direct generalization to the setting of complex Lie groups. In fact, the maximal closed connected Stein
subgroups of a connected Lie group need not be pairwise isomorphic (see [AK01] again), or normal,
or co-compact.
Returning to the algebraic setting, our structure results have applications to homogeneous spaces,
which will be developed elsewhere. A natural question asks for their generalizations to group schemes
over (say) local artinian rings, or discrete valuation rings.
Notation and conventions. Throughout this article, we denote by k a field with separable closure k s
and algebraic closure k̄. The Galois group of k s over k is denoted by Γk . A Γk -lattice is a free abelian
group of finite rank equipped with an action of Γk .
By a scheme, we mean a scheme of finite type over k, unless otherwise specified; a point of a
scheme will always mean a closed point. Morphisms of schemes are understood to be k-morphisms,
and products are taken over k. A variety is a separated, geometrically integral scheme.
We use [SGA3] as a general reference for group schemes. However, according to our conventions,
any group scheme G is assumed to be of finite type over k. The group law of G is denoted multiplica-
tively, and e G stands for the neutral element of G (k), except for commutative groups where we use an
additive notation. By an algebraic group, we mean a smooth group scheme G, possibly non-connected.
An abelian variety is a connected and complete algebraic group. For these, we refer to [Mu70], and to
[Bo91] for affine algebraic groups.
Given a connected group scheme G, we denote by G aff the smallest normal connected affine sub-
group scheme of G such that the quotient G /G aff is an abelian variety, and by
αG : G → G /G aff =: A (G ) (0.1)
936 M. Brion / Journal of Algebra 321 (2009) 934–952
the quotient homomorphism. The existence of G aff is due to Chevalley in the setting of algebraic
groups over algebraically closed fields; then G aff is an algebraic group as well, see [Ro56,Ch60].
Chevalley’s result implies the existence of G aff for any connected group scheme G, see [Ra70,
Lem. IX.2.7] or [BLR90, Thm. 9.2.1].
Also, we denote by
ϕG : G → Spec O(G ) (0.2)
the canonical morphism, known as the affinization of G. Then ϕG is the quotient homomorphism by
G ant , the largest anti-affine subgroup scheme of G. Moreover, G ant is a connected algebraic subgroup
of the center of G (see [DG70, Sec. III.3.8] for these results).
1. Basic properties
1.1. Characterizations of anti-affine groups
Recall that a group scheme G over k is affine if and only if G admits a faithful linear representation
in a finite-dimensional vector space; this is also equivalent to the affineness of the K -group scheme
G K := G ⊗k K for some field extension K /k. We now obtain analogous criteria for anti-affineness:
Lemma 1.1. The following conditions are equivalent for a k-group scheme G:
(i) G is anti-affine.
(ii) G K is anti-affine for some field extension K /k.
(iii) Every finite-dimensional linear representation of G is trivial.
(iv) Every action of G on a variety X containing a fixed point is trivial.
Proof. (i) ⇔ (ii) follows from the isomorphism O (G K ) O (G ) ⊗k K .

(i) ⇔ (iii) follows from the fact that every linear representation of G factors through the affine
quotient group scheme G /G ant .
Since (iv) ⇒ (iii) is obvious, it remains to show (iii) ⇒ (iv). Let x be a G-fixed point in X with local
ring Ox and maximal ideal mx . Then each quotient Oxn/mx is a finite-dimensional k-vector space on
n
which G acts linearly, and hence trivially. Since m

n x = {0}, it follows that G fixes Ox pointwise.
Thus, G acts trivially on X . 2
Remark 1.2. The preceding argument yields another criterion for affineness of a group scheme;
namely, the existence of a faithful action on a variety having a fixed point. This was first observed by
Matsumura (see [Ma63]).
Next, we show that the class of anti-affine groups is stable under products, extensions and quo-
tients:
Lemma 1.3. Let G 1 , G 2 be connected group schemes. Then:
(i) G 1 × G 2 is anti-affine if and only if G 1 and G 2 are both anti-affine.

(ii) Given an exact sequence of group schemes
1 −→ G 1 −→ G −→ G 2 −→ 1,
if G is anti-affine, then so is G 2 . Conversely, if G 1 and G 2 are both anti-affine, then so is G.

Proof. (i) follows from the isomorphism O (G 1 × G 2 ) O (G 1 ) ⊗k O (G 2 ).

(ii) The isomorphism O (G 2 ) O (G )G 1 (the algebra of invariants under the action of G 1 on O (G )
via left multiplication) yields the first assertion.
If G 1 is anti-affine, then its action on O (G ) is trivial (as O (G ) is a union of finite-dimensional k-
G 1 -submodules, and G 1 acts trivially on any such module by Lemma 1.1). Thus, O (G 2 ) O (G ) which
implies the second assertion. 2
Anti-affineness is also stable under isogenies:
Lemma 1.4. Let f : G → H be an isogeny of connected commutative algebraic groups. Then G is anti-affine if
and only if so is H .
Proof. If G is anti-affine, then so is H by Lemma 1.3(ii). For the converse, note that f induces an
isogeny G ant → I , where I is a subgroup scheme of H , and in turn an isogeny G /G ant → H / I . As
G /G ant is affine, so is H / I . But H / I is also anti-affine, and hence is trivial. Thus, f restricts to an
isogeny G ant → H . In particular, dim(G ant ) = dim( H ) = dim(G ), whence G ant = G. 2
1.2. Rigidity
In this subsection, we generalize some classical properties of abelian varieties to the setting of
anti-affine groups. Our results are implicit in [Ro56,Se58a]; we give full proofs for the sake of com-
pleteness.
Lemma 1.5. Let G be an anti-affine algebraic group, and H a connected group scheme.
(i) Any morphism (of schemes) f : G → H such that f (e G ) = e H is a homomorphism (of group schemes),
and factors through H ant ; in particular, through the center of H .
(ii) The abelian group (for pointwise multiplication) of homomorphisms f : G → H is free of finite rank.
Proof. (i) Consider the quotient homomorphism (0.1)
α H : H → H / H aff =: A ( H ).
By rigidity of abelian varieties (see e.g. [Co02, Lem. 2.2]), the composition α H f : G → A ( H ) is a ho-
momorphism. Equivalently, the morphism
F : G × G −→ H , (x, y )
−→ f (xy ) f (x)−1 f ( y )−1
factors through the affine subgroup scheme H aff . As G × G is anti-affine, and F (e G , e G ) = e H , it follows
that F factors through e H ; thus, f is a homomorphism.
The composition of f with the homomorphism (0.2)
ϕ H : H → H / H ant
is a homomorphism from G to an affine group scheme. Hence ϕ H f factors through e H / H ant , that is, f
factors through H ant .
(ii) We may assume that k is algebraically closed; then G aff is a connected affine algebraic group.
By [Co02, Lem. 2.3], it follows that any homomorphism f : G → H fits into a commutative square
f
G H
αG αH
α( f )
A (G ) A(H )
where α ( f ) is a homomorphism. This yields a homomorphism

α : Hom(G , H ) −→ Hom A (G ), A ( H ) , f
−→ α ( f ).
If α ( f ) = 0, then f factors through H aff , and hence is trivial. Thus, Hom(G , H ) is identified to a
subgroup of Hom( A (G ), A ( H )); the latter is free of finite rank by [Mu70, p. 176]. 2
Next, we show that anti-affine groups are “divisible” (this property is the main ingredient of the
classification of anti-affine groups in positive characteristics):
Lemma 1.6. Let G be an anti-affine algebraic group, and n a non-zero integer. Then the multiplication map
n G : G → G, x
→ nx is an isogeny.
Proof. Let H denote the cokernel of n G ; then n H is trivial. Hence the abelian variety H / H aff is trivial,
i.e., H is affine. But H is anti-affine as a quotient of G, so that H is trivial. 2
2. Structure
2.1. Semi-abelian varieties
Throughout this section, we consider connected group schemes G equipped with an isomorphism

α : G /G aff A
where A is a prescribed abelian variety. We then say that G is a group scheme over A.
Our aim is to classify anti-affine groups over A, up to isomorphism of group schemes over A (in
an obvious sense). We begin with the case where G aff is a torus, i.e., G is a semi-abelian variety. Then
G is obtained as an extension of algebraic groups
α
1 T G A 0 (2.1)
where T is a torus. Moreover, as T ks := T ⊗k k s is split, we have a decomposition of quasi-coherent

sheaves on A ks :

α∗ (OG ks ) = Lλ (2.2)
λ∈Λ
where Λ denotes the character group of T (so that Λ is a Γk -lattice), and Lλ consists of all eigenvec-
tors of T ks in α∗ (OG ks ) with weight λ. Each Lλ is an invertible sheaf on A ks , algebraically equivalent
to 0. Thus, Lλ yields a k s -rational point c (λ) of the dual abelian variety A ∨ . Moreover, the map
c : Λ → A ∨ (k s ), λ
→ c (λ) (2.3)
is a Γk -equivariant homomorphism, which classifies the extension (2.1) up to isomorphism of exten-

sions (as follows e.g. from [Se59, VII.3.16]). In other words, the extensions (2.1) are classified by the
homomorphisms T ∨ → A ∨ , where T ∨ denotes the Cartier dual of T .
Proposition 2.1.
(i) With the preceding notation, G is anti-affine if and only if c is injective.

(ii) The isomorphism classes of anti-affine semi-abelian varieties over A correspond bijectively to the sub-Γk -
lattices of A ∨ (k s ).
Proof. (i) By the decomposition (2.2), we have
0
O(G ks ) = H 0 A ks , α∗ (OG ks ) = H ( A k s , Lλ )
λ∈Λ
and of course H 0 ( A ks , L0 ) = O ( A ks ) = k s . Thus, G is anti-affine if and only if H 0 ( A ks , Lλ ) = 0 for all

λ = 0.
On the other hand, H 0 ( A ks , L) = 0 for any invertible sheaf L on A ks which is algebraically trivial
but non-trivial. (Otherwise, Lk̄ = O Ak̄ ( D ) for some non-zero effective divisor D on A k̄ . We may find
an integral curve C in A k̄ that meets properly Supp( D ). Then the pull-back of L to C has positive
degree, contradicting the algebraic triviality of L.)
Thus, G is anti-affine if and only if Lλ is non-trivial for any λ = 0.
(ii) Given two injective and Γk -equivariant homomorphisms
c 1 , c 2 : Λ −→ A ∨ (k s ),
the corresponding anti-affine groups are isomorphic over A if and only if the corresponding exten-
sions differ by an automorphism of T , i.e., there exists a Γk -equivariant automorphism f of Λ such
that c 2 = c 1 f . This amounts to the equality c 1 (Λ) = c 2 (Λ). 2
In positive characteristics, the preceding construction yields all anti-affine groups:
Proposition 2.2. Any anti-affine algebraic group over a field of characteristic p > 0 (resp. over a finite field) is
a semi-abelian variety (resp. an abelian variety).
Proof. The multiplication map p G is an isogeny by Lemma 1.6. In particular, the group G (k̄) con-
tains only finitely many points of order p. Thus, every unipotent subgroup of G k̄ is trivial. By [SGA3,
Exp. XVII, Thm. 7.2.1], it follows that (G k̄ )aff is a torus, i.e., G k̄ is a semi-abelian variety. Hence G is a
semi-abelian variety as well, see [BLR90, p. 178].
If k is finite (so that k s = k̄), then the group A ∨ (k s ) is the union of its subgroups A ∨ ( K ), where K
ranges over all finite subfields of k s that contain k. As a consequence, every point of A ∨ (k s ) has finite
order. Hence any sublattice of A ∨ (k s ) is trivial. 2
2.2. Vector extensions of abelian varieties
In this subsection, we assume that k has characteristic 0. Recall that every abelian variety A has
a universal vector extension E ( A ) by the k-vector space H 1 ( A , O A )∗ regarded as an additive group. In
other words, any extension G of A by a vector group U fits into a unique commutative diagram
0 H 1 ( A , O A )∗ E ( A) A 0
γ id (2.4)
0 U G A 0
(see [Ro58,Se59,MM74]). Note that E ( A )aff = H 1 ( A , O A )∗ .
Proposition 2.3.
(i) E ( A ) is anti-affine.
(ii) With the notation of the diagram (2.4), G is anti-affine if and only if the classifying map γ :
H 1 ( A , O A )∗ → U is surjective.
(iii) The anti-affine groups over A obtained as vector extensions are classified by the subspaces of the k-vector
space H 1 ( A , O A ).
Proof. (i) The affinization epimorphism (0.2)
ϕ = ϕ E ( A) : E ( A ) → V
induces epimorphisms
H 1 ( A , O A )∗ → W , A = E ( A )/ H 1 ( A , O A )∗ → V / W
where W is a subspace of V . The latter epimorphism must be trivial, and hence ϕ restricts to an
epimorphism
δ : H 1 ( A , O A )∗ → V .
Moreover, V is a vector group, and δ is k-linear. The extension given by the commutative diagram
0 H 1 ( A , O A )∗ E ( A) A 0
δ id
0 V H A 0
is split, as the map −ϕ + id : E ( A ) × V → V factors through a retraction of H onto V . Since E ( A ) is

the universal extension, it follows that δ = 0, i.e., V = 0.
(ii) The group G is the quotient of E ( A ) × U by the diagonal image of H 1 ( A , O A )∗ . Since
O( E ( A )) = k, it follows that O(G ) is the algebra of invariants of O(U ) under H 1 ( A , O A )∗ acting
by translations via γ . This implies the assertion.
(iii) follows from (ii) by assigning to γ the image of the transpose map γ t : U ∗ → H 1 ( A , O A ). 2
Remark 2.4. In the preceding statement, the assumption of characteristic 0 cannot be omitted in view
of Proposition 2.2. This may also be seen directly as follows. If k has characteristic p > 0, any vector
extension 0 → U → G → A → 0 splits after pull-back under the multiplication map p A : A → A (since
p A is an isogeny, and p U = 0). This yields an isogeny U × A → G. Thus, G cannot be anti-affine in
view of Lemma 1.4.
2.3. Classification of anti-affine groups
To complete this classification, we may assume that k has characteristic 0, in view of Proposi-
tion 2.2.
Let G be an anti-affine algebraic group. Then G aff is a connected commutative algebraic group, and
hence admits a unique decomposition
G aff = T × U (2.5)
where T is a torus, and U is connected and unipotent; U has then a unique structure of k-vector
space. Thus, G /U is a semi-abelian variety (extension of A by T ) and G / T is a vector extension
of A by U . Moreover, the quotient homomorphisms p U : G → G /U , p T : G → G / T fit into a cartesian
square
pU
G G /U
pT α G /U (2.6)
αG / T
G /T A
where αG /U (resp. αG / T ) is the quotient by T (resp. U ). Moreover, α = αG /T p T = αG /U p U . This yields

a canonical isomorphism of algebraic groups over A:

G G /U × A G / T . (2.7)
Proposition 2.5. With the preceding notation, G is anti-affine if and only if G /U and G / T are both anti-affine.
Proof. If G is anti-affine, then so are its quotient groups G /U and G / T .

For the converse, we may assume that k is algebraically closed in view of Lemma 1.1. Note that
the diagram (2.6) yields an isomorphism of quasi-coherent sheaves on A:
α∗ (OG ) αG /U ,∗ (OG /U ) ⊗O A αG /T ,∗ (OG /T ). (2.8)
Moreover, we have a decomposition

αG /U ,∗ (OG /U ) = Lλ
λ∈Λ
as in (2.2), where L0 = O A while H 0 ( A , Lλ ) = 0 for any λ = 0. On the other hand, the quasi-coherent
sheaf αG / T ,∗ (OG / T ) admits an increasing filtration with subquotients isomorphic to O A , by the next
lemma applied to the U -torsor αG / T : G / T → A. It follows that

H 0 A , Lλ ⊗O A αG /T ,∗ (OG /T ) = 0
for any λ = 0. Thus,
0
O(G ) = H 0 A , α∗ (OG ) H A , Lλ ⊗O A αG / T ,∗ (OG / T )
λ∈Λ
0

=H A , αG / T ,∗ (OG / T ) = O (G / T ) = k. 2
Lemma 2.6. Let π : X → Y be a torsor under a non-trivial connected unipotent algebraic group U . Then the
quasi-coherent sheaf π∗ (O X ) admits an infinite increasing filtration with subquotients isomorphic to OY .
Proof. We claim that there is an isomorphism of quasi-coherent sheaves over Y :

u : π∗ (O X ) π∗ (O X ⊗k O(U ))U .
Here the right-hand side denotes the subsheaf of U -invariants in the quasi-coherent sheaf π∗ (O X ⊗k
O(U )), where U acts via its natural action on O X and its action on O(U ) by left multiplication.
The assertion of the lemma follows from that claim, as the U -module O (U ) admits an infinite
increasing filtration with trivial subquotients.
To prove the claim, we first construct a natural isomorphism

uM : M ( M ⊗k O(U ))U
for any U -module M. Indeed, the right-hand side may be regarded as the space of U -equivariant
morphisms f : U → M. Any such morphism is of the form f m : u → u · m for a unique m ∈ M, namely,
m = f (e U ). We then set u M (m) := f m .
Next, if the U -module M is also a k-algebra where U acts by algebra automorphisms, then u M
is an isomorphism of M U -algebras, where the algebra of invariants M U acts on ( M ⊗k O (U ))U via
multiplication on M. Moreover, u M commutes with localization by elements of M U . Thus, the isomor-
phisms u O(π −1 (Y i )) , where (Y i )i ∈ I is an affine open covering of Y , may be glued to yield the desired
isomorphism. 2
Combining the results of Propositions 2.2, 2.3 and 2.5, we obtain the following classification:
Theorem 2.7.
(i) In positive characteristics, the isomorphism classes of anti-affine groups over an abelian variety A corre-
spond bijectively to the sub-Γk -lattices Λ ⊂ A ∨ (k s ).
(ii) In characteristic 0, the isomorphism classes of anti-affine groups over A correspond bijectively to the pairs
(Λ, V ), where Λ is as in (i), and V is a subspace of H 1 ( A , O A ).
Remark 2.8.
(i) The preceding classification may be formulated in terms of the dual variety A ∨ only, as
H 1 ( A , O A ) is naturally isomorphic to the tangent space T 0 ( A ∨ ) (the Lie algebra of A ∨ ); see e.g.
[Mu70, p. 130].
(ii) To classify the anti-affine groups G without prescribing an isomorphism G /G aff A, it suffices
to replace the sublattices Λ (resp. the pairs (Λ, V )) with their isomorphism classes under the
natural action of the automorphism group Aut( A ) of the abelian variety A (resp. of the natural
action of Aut( A ) × Aut( A ) on pairs).
2.4. Universal morphisms
Throughout this subsection, we assume that the ground field k is perfect. We investigate mor-
phisms from a prescribed variety to anti-affine algebraic groups, by adapting results and methods of
Serre (see [Se58a,Se58b]).
Consider a variety X equipped with a k-rational point x. Then there exists a universal morphism
to a semi-abelian variety
σ X ,x : X −→ S , x
−→ e S .
Indeed, this is a special case of [Se58a, Thm. 7] when k is algebraically closed, and the case of a perfect
field follows by Galois descent as in [Se59, Sec. V.4] (see [Wit06, Thm. A.1] for a generalization to an
arbitrary field).
We say that σ X ,x is the generalized Albanese morphism of the pointed variety ( X , x), and S = S X is
the generalized Albanese variety, which indeed depends only on X . The formation of σ X ,x commutes
with base change to perfect field extensions.
Recalling the extension of algebraic groups (2.1)
αS
1 T S A 0,
the composite morphism
α X ,x := α S σ X ,x : X −→ A = A X
is the Albanese morphism of X , i.e., the universal morphism to an abelian variety that maps x to the
origin.
We note that the pull-back map
α ∗X ,x : A ∨ (k) ⊂ Pic( A ) −→ Pic( X )
is independent of the choice of x ∈ X (k) (indeed, any two Albanese morphisms differ by a translation
by a k-rational point of A, and the translation action of A on A ∨ is trivial); we denote that map
by α ∗X . Likewise, the analogous map H 1 ( A , O A ) → H 1 ( X , O X ) is independent of x.
We also record the following observation:
Lemma 2.9. Let X be a complete variety equipped with a k-rational point. Then the pull-back maps A ∨ (k) →
Pic( X ) and H 1 ( A , O A ) → H 1 ( X , O X ) are both injective.
Proof. This follows from general results on the Picard functor (see [BLR90, Chap. 8]); we provide a
direct argument. We may assume that k is algebraically closed. Let L ∈ A ∨ (k) such that α ∗X ( L ) = 0.
Consider the corresponding extension
α
1 Gm G A 0 (2.9)
as a Gm -torsor over A. Then the pull-back of this torsor under α X ,x is trivial, that is, the projection
X × A G → X has a section. Thus, α X ,x lifts to a morphism γ : X → G and hence to a morphism X → H
where H ⊂ G denotes the algebraic subgroup generated by the image of γ . Since X is complete, H
is an abelian variety (as follows e.g. from [SGA3, Exp. VIB, Prop. 7.1]). Thus, α restricts to an isogeny
β : H → A. By the universal property of the Albanese morphism, it follows that β is an isomorphism.
Thus, the extension (2.9) is split; in other words, L is trivial.
Likewise, given u ∈ H 1 ( A , O A ) such that α ∗X (u ) = 0, one checks that u = 0 by considering the
associated extension
0 −→ Ga −→ G −→ A −→ 0. 2
We now obtain a criterion for anti-affineness of the generalized Albanese variety:

Proposition 2.10. Given a pointed variety ( X , x), the associated semi-abelian variety S = S X is anti-affine if
and only if O ( X k̄ )∗ = k̄∗ .
Under that assumption, S is classified by the pair ( A , Λ), where A is the Albanese variety of X , and Λ is the
kernel of the pull-back map
α ∗X : A ∨ (k̄) −→ Pic( Xk̄ ).
In particular, this kernel is a Γk -lattice.
Proof. Denote by ϕ : S → S / S ant the affinization morphism (0.2). Then S / S ant is affine and semi-
abelian, hence a torus. Clearly, the composite ϕσ X ,x is the universal morphism from X to a torus, that
maps x to the neutral element.
Given a Γk -lattice M, the morphisms from X to the dual torus M ∨ correspond bijectively to the
Γk -equivariant homomorphisms M → O( X k̄ )∗ . Moreover, the exact sequence of Γk -modules
1 −→ k̄∗ −→ O ( X k̄ )∗ −→ O ( X k̄ )∗ /k̄∗ −→ 1
is split by the evaluation map at x ∈ X (k), and O ( X k̄ )∗ /k̄∗ is a Γk -lattice. Thus, the morphisms of
pointed varieties

( X , x) −→ M ∨ , e M ∨
correspond bijectively to the Γk -equivariant homomorphisms
M −→ O ( X k̄ )∗ /k̄∗ .
In particular, there is a universal such morphism, to the dual torus of the lattice O ( X k̄ )∗ /k̄∗ . This
yields the first assertion.
Assuming that O ( X k̄ )∗ = k̄∗ , consider a sub-Γk -lattice M ⊂ A ∨ (k̄) and the corresponding extension
1 M∨ G A 0.
We regard G as an M ∨ -torsor over A. Then the morphisms γ : X → G that lift the Albanese morphism
α X ,x : X → A are identified to the sections of the pull-back M ∨ -torsor X × A S → X , as in the proof of
Lemma 2.9. Such sections exist if and only if the pull-back map M → Pic( X k̄ ) is trivial; moreover, any
two sections differ by a morphism X → T , i.e. by a k-rational point of T . Thus, there exists a unique
section such that the associated morphism γ maps x to e S .
As a consequence, the liftings of α X ,x to semi-abelian varieties over A are classified by the sub-
Γk -lattices of Λ := ker(α ∗X ). We now show that Λ is a Γk -lattice, thereby completing the proof. For
this, we may assume that k is algebraically closed.
If X is complete, then Λ is trivial by Lemma 2.9. In the general case, let i : X → X be an open
immersion into a complete variety. We may assume that α X ,x extends to a morphism α X ,x : X → A;
then α X ,x is the Albanese morphism of ( X , i (x)). Since α ∗X is injective, Λ is identified to a subgroup
of the kernel of
i ∗ : Pic( X ) → Pic( X ).
But ker(i ∗ ) is the group of Cartier divisors with support in X \ X , as O ( X )∗ = k∗ . In particular, the
abelian group ker(i ∗ ) is free of finite rank, and hence so is Λ. 2
By another result of Serre (see [Se58a, Thm. 8]), a pointed variety ( X , x) admits a universal mor-
phism to a commutative algebraic group,
γ X ,x : X −→ G , x
−→ e G
if and only if O ( X ) = k, that is, X is anti-affine. Then G is also anti-affine, as this group is generated
over k̄ by the image of X . In positive characteristics, the universal group G is just the generalized
Albanese variety, by Proposition 2.2. In characteristic 0, this group may be described as follows:
Proposition 2.11. Let ( X , x) be a pointed anti-affine variety over a field k of characteristic 0 and let G be the
associated anti-affine group. Then G is classified by the triple ( A , Λ, V ) where A and Λ are as in the preceding
proposition, and V is the kernel of the pull-back map α ∗X : H 1 ( A , O A ) −→ H 1 ( X , O X ).
The proof is analogous to that of Proposition 2.10, taking into account the isomorphism (2.7) and
the structure of anti-affine vector extensions of A.
Remarks 2.12. (i) The associated data A , Λ, V may be described explicitly in terms of completions, for
smooth varieties in characteristic 0. Namely, given such a variety X , there exists an open immersion
i : X → X where X is a smooth complete variety. Then Pic0 ( X ) is an abelian variety with dual the
Albanese variety A X = A X .
If O ( X k̄ )∗ = k̄∗ , then the Γk -lattice Λ of Proposition 2.10 is the group of divisors supported in X \ X
and algebraically equivalent to 0, by the arguments in [Se58b, Sec. 1].
Under the (stronger) assumption that O ( X ) = k, the subspace V ⊂ H 1 ( A , O A ) of Proposition 2.11
equals

H1 ( X , O X ) = H 0 X , i ∗ (O X )/O X ,
X\X
as follows from similar arguments.

(ii) Dually, one may also consider morphisms from varieties, or schemes, to a prescribed anti-
affine group G. In fact, such a group admits a modular interpretation, which generalizes the duality
of abelian varieties.
To state it, recall that any abelian variety A classifies the invertible sheaves on A ∨ , algebraically
equivalent to 0 and equipped with a rigidification along the zero section.
The universal extension E ( A ) has also a modular interpretation: it classifies the algebraically trivial
invertible sheaves on A ∨ , equipped with a rigidification along the first infinitesimal neighborhood
T 0 ( A ∨ ) (see [MM74, Prop. 2.6.7]).
It follows that the algebraically trivial invertible sheaves on A ∨ , equipped with rigidifications along
a basis of the lattice Λ and along the subspace V ⊂ T 0 ( A ∨ ), are classified by an anti-affine algebraic
group over A with data (Λ, V ).
3. Some consequences
3.1. The Rosenlicht decomposition
We first obtain a variant of a structure theorem for algebraic groups due to Rosenlicht (see [Ro56,
Cor. 5, p. 440]), in the setting of group schemes.
Proposition 3.1. Let G be a connected group scheme over a field k. Then:
(i) The group law of G yields an exact sequence of group schemes
1 −→ G aff ∩ G ant −→ G aff × G ant −→ G −→ 1. (3.1)

In other words, we have the decomposition G = G aff G ant .

(ii) The connected subgroup scheme (G ant )aff ⊂ G ant is an algebraic group, contained in G aff ; moreover, the
quotient (G aff ∩ G ant )/(G ant )aff is finite.
(iii) The quotient group scheme G := G /(G ant )aff has the decomposition G = G ab G aff where G ab =
G ant /(G ant )aff is the largest abelian subvariety of G , and G aff = G aff /(G ant )aff .
(iv) Any subgroup scheme H ⊂ G such that G = G aff H contains G ant .
Proof. (i) Since G aff is a normal subgroup scheme of G, and G ant is contained in the center of G,
we see that the multiplication map G aff × G ant → G is a homomorphism with kernel isomorphic to
G aff ∩ G ant ; the image G aff G ant is a normal subgroup scheme of G by [SGA3, Exp. VIA 5.3, 5.4]. The
quotient G /(G aff G ant ) is affine, as a quotient of G /G ant ; but it is also an abelian variety, as a quotient
of G /G aff . Thus, this quotient is trivial.
(ii) The smoothness of (G ant )aff follows from Proposition 2.2. By rigidity (see e.g. [Co02, Lem. 2.2]),
every homomorphism from (G ant )aff to an abelian variety is trivial. As a consequence, (G ant )aff ⊂ G aff .
The scheme (G aff ∩ G ant )/(G ant )aff is affine (as a quotient of a subgroup scheme of G aff ) and proper
(as a subgroup scheme of the abelian variety G ant /(G ant )aff ). Hence this scheme is finite.
(iii) follows readily from (i) and (ii).
(iv) Note that G = G aff H 0 , as G is connected. Thus,
0 0
G = G aff H aff H ant
0
and H ant ⊂ G ant ; in particular, H ant
0 0
is contained in the center of G. On the other hand, G aff H aff is
affine, so that
0
0
0
0

G / H ant G aff H aff H ant ∩ G aff H aff
is affine as well. Since the quotient homomorphism G → G /G ant is the affinization, it follows that
0
H ant contains G ant . 2
Next, we consider the functorial properties of the Rosenlicht decomposition. By the results of
[DG70, III.3.8], the formation of G ant commutes with base change to arbitrary field extensions,
and with homomorphisms of group schemes. Also, note that the homomorphism (0.2) ϕG : G →
Spec O (G ) = G /G ant depends only on G regarded as a scheme. In particular, G ant depends only on
the pointed scheme (G , e G ).
These properties are also satisfied by G aff under additional assumptions. Specifically, if G is a
connected algebraic group over a perfect field k, then G aff is the largest connected affine algebraic
subgroup of G; the formation of G aff commutes with base change to any perfect field extension of k
and with homomorphisms of algebraic groups (see [Co02] for these results). The quotient homomor-
phism αG : G → G /G aff is the Albanese morphism of the pair (G , e G ). In particular, G aff depends only
on the pointed variety (G , e G ).
The assumption that k is perfect cannot be omitted in view of the following example, obtained by
a construction of Raynaud (see [SGA3, Exp. XVII, App. III, Prop. 5.1]):
Example 3.2. Let k be a non-perfect field of characteristic p > 0 and choose a finite, non-trivial field
extension K /k such that K p ⊂ k. Given a non-trivial abelian variety A over k, let A K := A ⊗k K (a non-
trivial abelian variety over K ) and
G := Π K /k ( A K )
where Π K /k denotes the Weil restriction; in other words, G is the unique k-scheme such that
G ( R ) = A K ( R ⊗k K ) (3.2)
for any k-algebra R. Then G is a commutative connected algebraic k-group, as follows e.g. from the
results of [Oe84, A.2] that we shall use freely.
We claim that
GK = U × AK (3.3)
where U is a connected unipotent algebraic K -group; in particular, (G K )aff = U and (G K )ant = A K .

Moreover, G ant = A but G aff is not smooth, and (G aff ) K = (G K )aff .
Indeed, for any K -algebra R, we have

G K ( R ) = G ( R ) = A K R ⊗ K ( K ⊗k K )
and K ⊗k K is a finite-dimensional K -algebra. The multiplication map μ : K ⊗k K → K yields an exact

sequence
0 → m → K ⊗k K → K → 0
and the ideal m is nilpotent, as (x ⊗ 1 − 1 ⊗ x) p = 0 for any x ∈ K . This yields a functorial morphism
G K ( R ) → A K ( R ) and, in turn, an extension of K -group schemes
α
1 U GK AK 0 (3.4)
where U has a filtration with subquotients isomorphic to the Lie algebra of A K . In particular, U is
smooth, connected and unipotent. Moreover, α is the Albanese morphism of (G K , e G K ).
For any k-scheme S, the map G ( S ) → A K ( S K ) that sends any f : S → G to α f K : S K → A K is
bijective. This yields a morphism β : A → G such that α β K is the identity map of A K . It follows that
β is a closed immersion of group schemes; we shall identify A with β( A ), and likewise A K with
β K ( A K ). As β K splits the extension (3.4), this yields the decomposition (3.3).
As a consequence, G ant = A and hence G = G aff A. Also, G aff is not smooth; indeed, any morphism
from a connected affine algebraic group to G is constant, as follows from the equality (3.2) together
with [Co02, Lem. 2.3]. Thus, the finite group scheme G aff ∩ A is non-trivial: otherwise, G G aff × A,
so that G aff would be smooth. In particular, (G aff ) K = U = (G aff ) K .
In particular, G /G aff is the quotient of A by a non-trivial subgroup scheme. On the other hand,
the quotient map αG : G → G /G aff is easily seen to be the Albanese morphism of (G , e G ) considered
in [Wit06]. Thus, the formation of the Albanese morphism does not commute with arbitrary field
extensions.
3.2. Structure of connected commutative algebraic groups
We first obtain a simple characterization of non-affine group schemes that are minimal for this
property:
Proposition 3.3. The following conditions are equivalent for a non-trivial group scheme G:
(i) G is non-affine and every subgroup scheme H ⊂ G, H = G is affine.

(ii) G is anti-affine and has no non-trivial anti-affine subgroup.
(iii) G is anti-affine and the abelian variety A (G ) = G /G aff is simple.
If one of these conditions holds, then either G is an abelian variety or G contains no complete subvariety of
positive dimension.
Proof. (i) ⇔ (ii) follows easily from the fact that a group scheme H is affine if and only if H ant 0
is
trivial.
(ii) ⇒ (iii) Assume that A (G ) contains a non-trivial abelian variety B, and denote by H the pull-
back of B in G. Then H ant is a non-trivial subgroup of G, a contradiction.
(iii) ⇒ (ii) Let H be an anti-affine subgroup of G. Then H aff ⊂ G aff , as G aff is the largest connected
affine subgroup of G; hence A ( H ) is identified with a subgroup of A (G ). Thus, either A ( H ) is trivial
so that H is affine, or A ( H ) = A (G ) so that G aff H = G. In the latter case, H = G by Proposition 3.1.
Under one of these conditions, consider the algebraic subgroup H ⊂ G generated by a complete
subvariety of G. Then H is complete as well (see e.g. [SGA3, Exp. VIB, Prop. 7.1]); thus, either H = G
or H is trivial. 2
Next, we obtain a decomposition of connected commutative group schemes over perfect fields:
Theorem 3.4. Let G be a connected commutative group scheme over a perfect field k. Then there exist a
subtorus T ⊂ G and a connected unipotent subgroup scheme U ⊂ G such that the group law of G induces
an isogeny
f : G ant × T × U −→ G . (3.5)
Moreover, T is unique up to isogeny, and U is unique up to isomorphism; if G is an algebraic group, then so

is U .
Proof. The Rosenlicht decomposition yields an exact sequence of group schemes
ψ
1 G aff ∩ G ant G aff G /G ant 1.
Moreover, we have unique decompositions G aff = T × U and G /G ant = T × U , where T , T are

tori and U , U are connected unipotent group schemes. This yields epimorphisms ψs : T → T ,
ψu : U → U . Thus, we may find a subtorus T ⊂ T such that ψs restricts to an isogeny T → T .
If k has characteristic 0, we may also find a (connected) unipotent subgroup U ⊂ U such that ψu
restricts to an isomorphism U → U , as U and U are vector groups. Then the homomorphism f
induces an isogeny T × U → G /G ant . Thus, f is an isogeny, and T , U are unique up to isogeny; hence
the vector group U is uniquely determined.
In positive characteristics, G aff ∩ G ant contains the torus (G ant )aff and the quotient is finite; hence
ψu is an isogeny. Thus, our statement holds with U = U , but for no other choice of U . 2
The assumption that k is perfect cannot be omitted in the preceding result, as shown by Exam-
ple 3.2.
3.3. Further decompositions in positive characteristics
In this subsection, we combine the Rosenlicht decomposition with the particularly simple structure
of anti-affine algebraic groups in positive characteristics, to obtain information on general algebraic
groups.
We begin with the case where the field k is finite. Then Propositions 2.2 and 3.1 immediately
imply the following result, due to Arima in the setting of algebraic groups (see [Ar60, Thm. 1] and
also [Ro61, Thm. 4]):
Proposition 3.5. Let G be a connected group scheme over a finite field k. Then G = G aff G ab where G ab denotes
the largest abelian subvariety of G. Moreover, G aff ∩ G ab is finite.
In particular, the Albanese morphism of (G , e G ) is trivialized by the finite cover G aff × G ab → G

(possibly non-étale).
Returning to a possibly infinite field k, we record the following preliminary result:
Lemma 3.6. Let G be a connected algebraic group over a perfect field k. Then:
(i) There exists a smallest normal connected algebraic group H ⊂ G such that G / H is a semi-abelian variety.
The quotient homomorphism G → G / H is the generalized Albanese morphism of the pointed variety
( G , e G ).
(ii) We have
H = R u (G aff )[G , G ] = R u (G aff )[G aff , G aff ] (3.6)
where R u (G aff ) denotes the unipotent radical of G aff , and [G , G ] the derived group.
(iii) The formation of H commutes with perfect field extensions.
(iv) The group H k̄ is generated by all connected unipotent subgroups of G k̄ .
Proof. By the Rosenlicht decomposition, we have [G , G ] = [G aff , G aff ]. Define H by the equality (3.6);
then H is a connected normal subgroup of G. Moreover, the quotient G aff / H is a connected commu-
tative reductive group, i.e., a torus. Thus, G / H is a semi-abelian variety.
Consider a morphism f : G → S, where S is a semi-abelian variety, and f (e S ) = e G . Then f is a
homomorphism by [Ro61, Thm. 3]. Hence f factors through G / R u (G aff ) (as every unipotent subgroup
of S is trivial) and also through G /[G , G ] (as S is commutative). Thus, f factors through G / H . This
proves (i) and (ii), while (iii) and (iv) are obtained by similar arguments. 2
Under the assumptions of the preceding lemma, we say that H is geometrically unipotently gener-
ated, and write H := G gug .
Also, given a group scheme G, a normal subgroup scheme H ⊂ G and a subgroup scheme S ⊂ G,
we say that S is a quasi-complement to H in G if G = H S and H ∩ S is finite; equivalently, the natural
map S → G / H is an isogeny. We may now state our structure result:
Theorem 3.7. Let G be a connected algebraic group over a perfect field k of positive characteristic and let T be
a maximal torus of the radical R (G aff ). Then:
(i) T is a quasi-complement to G gug in G aff .

(ii) S := T G ant is a quasi-complement to G gug in G, and is a semi-abelian subvariety of G with maximal
torus T .
(iii) The generalized Albanese morphism of (G , e G ) is trivialized by the finite cover G gug × S → G.
Proof. (i) By the structure of affine algebraic groups (see [Bo91]) and the equality (3.6), we have
G aff = R (G aff )[G aff , G aff ] = R u (G aff ) T [G aff , G aff ]
= R u (G aff )[G aff , G aff ] T = G gug T .
We now show the finiteness of G gug ∩ T . For this, we may assume that G is affine. Since the homo-
morphism

G gug ∩ T → G gug ∩ R (G ) / R u (G ) ⊂ G / R u (G )
is finite, we may also assume G to be reductive. Then G gug = [G , G ] is semi-simple and T is the
largest central torus, so that their intersection is indeed finite.
(ii) By the Rosenlicht decomposition and (i), G = G gug S. Also, the quotient (G gug ∩ S )/(G gug ∩ T ) is
finite, since G gug ∩ S is affine and T = S aff . Thus, G gug ∩ S is finite, i.e., S is a quasi-complement to
G gug in G.
We know that G ant is a semi-abelian variety contained in the center of G. Thus, S is a semi-abelian
variety as well. Moreover, the maximal torus (G ant )aff of G ant is a central subtorus of G aff , and hence
is contained in T . Thus, T is the maximal torus of S.
(iii) follows readily from (ii). 2
Remarks 3.8. (i) The quasi-complements constructed in the preceding theorem are all conjugate under
R u (G aff ). But G gug may admit other quasi-complements in G; namely, all subgroups T G ant where T
is a quasi-complement of G gug in G aff . Such a subtorus T need not be contained in R (G aff ), e.g., when
G aff is reductive and non-commutative.
(ii) With the notation and assumptions of the preceding theorem, G ant also admits quasi-
complements in G, namely, the subgroups T G gug where T is a quasi-complement to (G ant )aff in T .
In contrast, G aff may admit no quasi-complement in G. Indeed, such a quasi-complement S comes
with a finite surjective morphism to G /G aff , and hence is an abelian variety. Thus, S exists if and only
if G ant is an abelian variety, and then S = G ant . Equivalently, G = G aff G ab as in Proposition 3.5.
(iii) In characteristic 0, the group G gug still admits quasi-complements in G aff , but may admit no
quasi-complement in G.
For example, let C be an elliptic curve, E (C ) its universal extension, H the Heisenberg group of
upper triangular 3 × 3 matrices with diagonal entries 1, and G = ( H × E (C ))/Ga where the additive
group Ga is embedded in H as the center, and in E (C ) as E (C )aff . Then G is a connected algebraic
group; moreover, G gug = G aff ∼ = H and G ant ∼= E (C ). Since G ant is non-complete, there exist no quasi-
complement to G aff in G.
The same example shows that G ant may admit no quasi-complement in G. Yet such a quasi-
complement does exist when G is commutative, by Theorem 3.4.
3.4. Counterexamples to Hilbert’s fourteenth problem
In this subsection, we construct a class of smooth quasi-affine varieties having a non-noetherian

coordinate ring.
Recall that every connected algebraic group G is quasi-projective, i.e., G admits an ample invertible
sheaf L (see e.g. [Ra70, Cor. V 3.14]). Clearly, the associated Gm -torsor over G (that is, the complement
of the zero section in the total space of the associated line bundle V(L)) is a smooth quasi-affine
variety. This simple construction yields our examples:
Theorem 3.9. Let π : X → G denote the Gm -torsor associated to an ample invertible sheaf L on a non-
complete anti-affine algebraic group. Then the ring O ( X ) is not noetherian.

Proof. As X = SpecOG ( n∈Z Ln ), we have O ( X ) = n∈Z H 0 (G , Ln ). Moreover, H 0 (G , OG ) = k by as-
sumption, and the k-vector space H 0 (G , Ln ) is infinite-dimensional for any n > 0 by the next lemma.
Since O ( X ) is a domain, it follows that H 0 (G , Ln ) = 0 for any n < 0, i.e., the algebra O ( X ) is posi-
tively graded. Clearly, this algebra is not finitely generated, and hence non-noetherian by the graded
version of Nakayama’s lemma. 2
Lemma 3.10. Let L be an ample invertible sheaf on an anti-affine algebraic group G. If G is non-complete, then
the k-vector space H 0 (G , L) is infinite-dimensional.
Proof. We may assume that k is algebraically closed. The quotient homomorphism α = αG : G →

A (G ) =: A is a torsor under the connected commutative affine algebraic group G aff . Since the Picard
group of G aff is trivial, it follows that L = α ∗ (M) for some invertible sheaf M on A. Moreover, M is
ample by the ampleness of L together with [Ra70, Lem. XI 1.11.1]. We have

H 0 (G , L) H 0 A , M ⊗ α∗ (OG ) . (3.7)
In the case where G is a semi-abelian variety, Eqs. (2.2) and (3.7) yield the decomposition

H 0 ( G , L) H 0 ( A , M ⊗ Lλ ).
λ∈Λ
As each Lλ is algebraically trivial, M ⊗ Lλ is ample, and hence admits non-zero global sections (see
[Mu70, p. 163]); this yields our statement in this case.
In the general case, we may assume in view of Proposition 2.2 and the isomorphism (2.8) that
k has characteristic 0, and G aff is a non-zero vector space U . Then M ⊗ α∗ (OG ) admits an infinite
increasing filtration with subquotients isomorphic to M, by Lemma 2.6. Since H 1 ( A , M) = 0 (see
[Mu70, p. 150]), it follows that H 0 (G , L) admits an infinite increasing filtration with subquotients
isomorphic to H 0 ( A , M), a non-zero vector space. 2
Example 3.11. The smallest examples arising from the preceding construction are threefolds; they may
be described as follows.
Consider an invertible sheaf L of positive degree on an elliptic curve C . If k has characteristic 0,
let π : G → C denote the Ga -torsor associated to the canonical generator of H 1 (C , OC ) H 0 (C , OC )∗ .
Then G is the universal extension E (C ), and the Gm -torsor on G associated to the ample invertible
sheaf π ∗ (L) yields the desired example X .
When k = C, the analytic manifolds associated to G and X are both Stein; see [Ne88] which also
contains an analytic proof of the fact that O ( X ) is not finitely generated. More generally, the universal
extension E ( A ) of a complex abelian variety of dimension g is analytically isomorphic to (C∗ )2g , see
e.g. [Ne88, Rem. 7.7]. In particular, the complex manifold associated to E ( A ) is Stein.
Returning to a field k of arbitrary characteristics, assume that the elliptic curve C has a k-rational
point x of infinite order (such curves exist if k contains either Q or F p (t ), see [ST67]). Denote by M
the invertible sheaf on C associated to the divisor (x) − (0). Then M is algebraically trivial and has
infinite order. Thus, G := SpecOC ( n∈Z Mn ) is an anti-affine semi-abelian variety, and

X := SpecOC Lm ⊗O A Mn
(m,n)∈Z2
is the desired example.

It should be noted that O ( X ) is finitely generated for any smooth surface X , as shown by Zariski
(see [Za54]). Also, Kuroda has constructed counterexamples to Hilbert’s original problem, in dimen-
sion 3 and characteristic 0 (see [Ku05]).
Another consequence of Lemma 3.10 is the following:
Proposition 3.12. For any completion G of a connected algebraic group G, the boundary G \ G is either empty
or of codimension 1.
Proof. We argue by contradiction, and assume that G \ G is non-empty of codimension 2. We may

further assume that G is normal; then the map i ∗ : O (G ) → O (G ) is an isomorphism, where i : G → G
denotes the inclusion. It follows that G is anti-affine and non-complete.
Choose an ample invertible sheaf L on G. Then i ∗ (L) is the sheaf of sections of some Weil divisor
on G; in particular, this sheaf is coherent. Thus, the k-vector space H 0 (G , i ∗ (L)) = H 0 (G , L) is finite-
dimensional, contradicting Lemma 3.10. 2
Remark 3.13. With the assumptions of the preceding proposition, one may show (by completely dif-
ferent methods) that the boundary has pure codimension 1. For a G-equivariant completion G (that is,
the action of G on itself by left multiplication extends to G), this follows easily from [Br07, Thm. 3].
Namely, we may assume that k is algebraically closed and G is normal; then G G ×G aff G aff , and
G aff \ G aff has pure codimension 1 in G aff , as G aff is affine.
Acknowledgments
Most of the results of this article were first presented at the 2007 Algebra Summer School in
Edmonton. I thank the organizers and the participants for their stimulating interest. Also, many thanks
are due to Stéphane Druel, Adrien Dubouloz, David Harari and Gaël Rémond for fruitful discussions.
After a first version of this text was posted on arXiv, I was informed by Carlos Sancho de Salas of
his much earlier book [Sa01], where the classification of anti-affine groups over algebraically closed
fields is obtained. Subsequently, he extended this classification to arbitrary fields in [SS08], jointly
with Fernando Sancho de Salas. The approach of [Sa01,SS08] differs from the present one, their key
ingredient being the classification of certain torsors over anti-affine varieties. The terminology is also
different: the variedades cuasi-abelianas of [Sa01], or quasi-abelian varieties of [SS08], are called anti-
affine groups here. I warmly thank Carlos Sancho de Salas for making me aware of his work, and for
many interesting exchanges.
References
[AK01] Y. Abe, K. Kopfermann, Toroidal Groups. Line Bundles, Cohomology and Quasi-Abelian Varieties, Lecture Notes in Math.,
vol. 1759, Springer-Verlag, New York, 2001.
[Ar60] S. Arima, Commutative group varieties, J. Math. Soc. Japan 12 (1960) 227–237.
[Bo91] A. Borel, Linear Algebraic Groups, second ed., Grad. Texts in Math., vol. 126, Springer-Verlag, New York, 1991.
[BLR90] S. Bosch, W. Lütkebohmert, M. Raynaud, Néron Models, Ergeb. Math., vol. 21, Springer-Verlag, New York, 1990.
[Br07] M. Brion, Some basic results on actions of non-affine algebraic groups, arXiv: math.AG/0702518.
[Ch60] C. Chevalley, Une démonstration d’un théorème sur les groupes algébriques, J. Math. Pures Appl. (9) 39 (1960) 307–317.
[Co02] B. Conrad, A modern proof of Chevalley’s theorem on algebraic groups, J. Ramanujan Math. Soc. 17 (2002) 1–18.
[DG70] M. Demazure, P. Gabriel, Groupes algébriques, Masson, Paris, 1970.
[Ku05] S. Kuroda, A counterexample to the fourteenth problem of Hilbert in dimension three, Michigan Math. J. 53 (2005)
123–132.
[Ma63] H. Matsumura, On algebraic groups of birational transformations, Atti Accad. Naz. Lincei Rend. Cl. Sci. Fis. Mat. Natur.
(8) 34 (1963) 151–155.
[MM74] B. Mazur, W. Messing, Universal Extensions and One Dimensional Crystalline Cohomology, Lecture Notes in Math.,
vol. 370, Springer-Verlag, New York, 1974.
[Mu70] D. Mumford, Abelian Varieties, Oxford University Press, Oxford, 1970.
[Ne88] A. Neeman, Steins, affines and Hilbert’s fourteenth problem, Ann. of Math. 127 (1988) 229–244.
[Oe84] J. Oesterlé, Nombres de Tamagawa et groupes unipotents en caractéristique p, Invent. Math. 78 (1984) 13–88.
[Ra70] M. Raynaud, Faisceaux amples sur les schémas en groupes et les espaces homogènes, Lecture Notes in Math., vol. 119,
Springer-Verlag, New York, 1970.
[Re58] D. Rees, On a problem of Zariski, Illinois J. Math. 2 (1958) 145–149.
[Ro56] M. Rosenlicht, Some basic theorems on algebraic groups, Amer. J. Math. 78 (1956) 401–443.
[Ro58] M. Rosenlicht, Extensions of vector groups by abelian varieties, Amer. J. Math. 80 (1958) 685–714.
[Ro61] M. Rosenlicht, Toroidal algebraic groups, Proc. Amer. Math. Soc. 12 (1961) 984–988.
[Sa01] C. Sancho de Salas, Grupos algebraicos y teoría de invariantes, Aportaciones Mat. Textos, vol. 16, Sociedad Matemática
Mexicana, México, 2001.
[SS08] C. Sancho de Salas, F. Sancho de Salas, Principal bundles, quasi-abelian varieties and structure of algebraic groups,
arXiv: 0806.3712.
[SGA3] M. Demazure, A. Grothendieck (Eds.), Schémas en groupes I, II, III, Séminaire de Géométrie Algébrique du Bois Marie
1962/64 (SGA 3), Lecture Notes in Math., vol. 151, 152, 153, Springer-Verlag, New York, 1970.
[Se58a] J.-P. Serre, Morphismes universels et variété d’Albanese, in: Séminaire Chevalley (1958–1959), Exposé No. 10, Docu-
ments Mathématiques 1, Soc. Math. France, Paris, 2001.
[Se58b] J.-P. Serre, Morphismes universels et différentielles de troisième espèce, in: Séminaire Chevalley (1958–1959), Exposé
No. 11, Documents Mathématiques 1, Soc. Math. France, Paris, 2001.
[Se59] J.-P. Serre, Groupes algébriques et corps de classes, Hermann, Paris, 1959.
[ST67] I.R. Shafarevich, J. Tate, The rank of elliptic curves, Sov. Math. Dokl. 8 (1967) 917–920.
[Win03] J. Winkelmann, Invariant rings and quasiaffine quotients, Math. Z. 244 (2003) 163–174.
[Wit06] O. Wittenberg, On Albanese torsors and the elementary obstruction, Math. Ann. 340 (2008) 805–838.
[Za54] O. Zariski, Interprétations algébro-géométriques du quatorzième problème de Hilbert, Bull. Sci. Math. 78 (1954) 155–
168.
Journal of Algebra
Embeddings of root systems II: Permutation root systems

Matthew J. Dyer
Department of Mathematics, 255 Hurley Building, University of Notre Dame, Notre Dame, Indiana 46556-4618, USA
Article history: This paper studies a class of embeddings of “permutation” root

Received 17 July 2008 systems, with properties abstracting those of examples of embed-
Available online 29 November 2008 dings of linear root systems over commutative rings, as considered
in the first paper in this series. The root system of one Coxeter
Keywords:
group W (regarded as a permutation representation) is realized
Coxeter groups with its roots as sets of roots of another Coxeter group W , with
Root systems action induced by an embedding of W in (a completion of) W .
By a standard based permutation root system of a Coxeter system ( W , S ), we mean an abstract

realization of the standard (real, reduced) root system Φ of ( W , S ) (as in [13], for example) regarded
just as W -set, with additional structure given by the action of the sign group {±1} and by specifying
the standard subset of positive roots. We call the elements of Φ abstract roots.
In this paper, we consider abstract “embeddings” of standard based permutation root systems, in
which the abstract roots of one standard based permutation root system are realized as pairwise
disjoint sets of abstract roots of a second standard based permutation root system, with group action
coming from an embedding of the Coxeter group of the first root system in (a completion of) the
second Coxeter group. This abstracts some of the main formal properties of examples of embeddings
of linear root systems over commutative rings which were considered in [6].
In more detail, the content of this paper is as follows. Section 1 contains preparatory remarks on
actions of G × {±1}, where G is a group and {±1} is the sign group, on a set of the form R × {±1},
where the action of {±1} is the natural one. Such an action is equivalent to an action of G on R
together with a cocycle N: G → P ( R ) of G in the power set P ( R ) (with symmetric difference as ad-
dition and natural G-action). An expression g 1 · · · gn in G is defined to be compatible if N ( g 1 · · · gn ) is
the disjoint union (rather than just symmetric difference) of the sets g 1 · · · g i −1 N ( g i ). In Section 2, we
discuss the standard based permutation root system Φ ∼ = R × {±1}, where R is the set of reflections,
of a Coxeter system ( W , S ); for group action, we consider the action on Φ of the completion G = W
of W consisting of permutations of Φ which act locally (i.e. on the roots of each finitely generated
E-mail address: dyer.1@nd.edu.
doi:10.1016/j.jalgebra.2008.11.001
954 M.J. Dyer / Journal of Algebra 321 (2009) 953–981
reflection subgroup) as some element of W . In this case, compatibility of expressions g 1 · · · gn with

gi ∈ W generalizes the length compatibility condition l( w 1 · · · w n ) = l( w 1 ) + · · · + l( w n ) for elements
w i of W . The restriction of the corresponding cocycle N : W → P ( R ) to W is the reflection cocycle
of ( W , S ) in the sense of [9] i.e. N ( w ) is the set of reflections t of W such that l(t w ) < l( w ), and so
we call N the reflection cocycle of W.
In Section 3, we introduce a notion of a based permutation root system of a Coxeter system ( W , S ).
Roughly, W × {±1} acts suitably on a set X , with free action of {±1} and a distinguished set X +
(of “positive” elements) of {±1}-orbit representatives on X ; the standard based permutation root
system is realized with its (positive resp., negative) abstract roots as certain pairwise disjoint subsets
of the set of (positive resp., negative) elements of X . A cocycle N : W → P ( R ) of ( W , S ) comes from
a based permutation root system iff any expression xy with l(xy ) = l(x) + l( y ) is compatible for N;
we call such a cocycle a root cocycle. The notion of a root cocycle is a simple abstraction of a “cocycle
compatibility condition” which appeared as [6, 4.5(g)]. In Section 4, we study root cocycles which are
coboundaries; these correspond to a based permutation root system in two different ways. The results
show how “degenerate” permutation root systems (with some possibly empty abstract roots and on
which W may not act faithfully) may arise naturally in the study of non-degenerate ones.
In Section 5, we consider a cocycle N : G → P ( R ). We define, by some rank-two conditions, a
notion of an admissible subset S of G with respect to N. The first main result of the paper is Theo-
rem 5.3, which states that for an admissible subset S generating a subgroup W of G, the pair ( W , S )
is a Coxeter system and the restriction of N to W is a non-degenerate root cocycle for ( W , S ); further,
for w i ∈ W , an expression w = w 1 · · · w n is compatible for N iff l( w ) = l( w 1 ) + · · · + l( w n ), where l
is the length function of ( W , S ). Given the results of Section 3, the proof of this is almost identical to
the proof of [17, Theorem 1.2]. We obtain as Corollary 5.9 a simultaneous abstract characterization up
to isomorphism of Coxeter systems together with their reflection cocycles, by rank two conditions.
In Section 6, we apply Theorem 5.3 in the following situation: R is a subset of the set of re-
flections T of a Coxeter system ( W , S ), and G is a subgroup of W acting on R by conjugation,

with cocycle N : G → P ( R ) given by N ( g ) = N ( g ) ∩ R where N is the reflection cocycle of W . The

theorem provides embeddings of W in W (or in favorable situations cases, in W itself) for which
there are corresponding “abstract embeddings” of standard based permutation root systems; the ab-
stract roots (resp., abstract positive roots) of ( W , S ) are realized as pairwise disjoint sets of abstract
roots (resp., abstract positive roots) of ( W , S ). Under additional hypotheses, the reflection in each
abstract root of ( W , S ) is a (possibly infinite, convergent, suitably ordered) product of the (possibly
non-commuting) reflections in the abstract roots of ( W , S ) which it contains. The Coxeter group
embeddings W → W so arising include both the inclusions of reflection subgroups [9] (taking G a
reflection subgroup of W and R as its set of reflections) and the embeddings W → W defined by
Mühlherr in [17] (as a special case of the situation where R = T and G = W ).
In Section 7, we consider a class of the above embeddings W → W which naturally generalizes
Mühlherr’s embeddings W → W . For embeddings in this class, each simple reflection of W is a

(possibly infinite, convergent in W ) product
i w J i of longest elements in W of finite standard
parabolic subgroups W J of ( W , S ), for pairwise disjoint subsets J i of S such that the w J i commute
i
pairwise. In Section 8, we consider the special case of the embeddings in Section 7 in which each
simple reflection of W is a convergent, possibly infinite product of commuting simple reflections
of W . Our second main result (Theorem 8.4) completely classifies all embeddings of this special
type, essentially as those induced by “strong” admissible maps (roughly, admissible maps in the sense
of [16] with certain divisibility conditions on Coxeter numbers from [16] replaced by equalities). As
an application, the cocycle compatibility condition [6, 4.5(g)] proved there for certain embeddings of
“Z-fusion” root systems over commutative rings is extended to the corresponding embeddings of Z -
fusion root systems for Z an arbitrary subring of R, just by inspection of the corresponding Coxeter
graphs as described in [6].
Despite the complete solution of the special case in Section 8, it seems to be a subtle problem,
for general Coxeter systems ( W , S ), to concretely describe the admissible subsets S of W or W ,
with respect to their reflection cocycles N . Although by results of Section 5, one may largely confine
attention to the rank two case | S | = 2, we leave this problem open even for W of type A. In a sub-
M.J. Dyer / Journal of Algebra 321 (2009) 953–981 955
sequent note, we shall provide an algorithm, involving the elementary roots of Brink and Howlett [4],
for determining in general whether a finite subset S of W is admissible.
The ideas in this paper clarify formal properties (e.g. the above-mentioned cocycle compatibility)
of embeddings of linear root systems (over commutative rings, say, as in [6]). This is accomplished by
an extension from the standard based permutation root system (in which reflections correspond to
abstract roots, identified with singleton sets) to general based permutation root systems (for which re-
flections correspond to general sets). However, the main observation of the paper is that the resulting
embeddings of based permutation root systems apparently exist far more generally than the (motivat-
ing) embeddings of linear root systems. To accommodate the additional embeddings of permutation
root systems in the linear setting, a corresponding extension of the notion of linear root system over a
ring is naturally suggested, in which reflections are no longer associated just to “rank one” objects (the
span of a root) but to submodules of higher rank. This idea in turn suggests the question of whether
there is some natural notion of a “based root system object” in a (suitable) category, such that the
root system objects in the category of sets (or “signed sets”) are essentially the based permutation
root systems, and the root system objects of a (suitable) category of modules provide essentially the
above-mentioned extension of the notion of linear root system. Though these questions seem very
natural and interesting in light of our results here, we do not discuss them further in this paper.
1. Cocycles and compatibility
1.1. Throughout this paper, we denote the group of units Z∗ of the ring of integers as {±1} and call it
the sign group. Let X be a set with a free (equivalently, fixed-point free) action of the sign group, and
X + be a set of orbit representatives for {±1} on X . Let X /{±1} denote the set of {±1}-orbits on X
and π : X → X /{±1} denote the natural surjection x → {x, −x}. The composite map X + → X X /±1
is a bijection, and the map X + × {±1} → X given by (x, ) → x is a {±1}-equivariant bijection, where
{±1} acts on X + × {±1} by multiplication on the second factor. For this reason, we may (and often
will) assume without loss of generality that X is a set of the form R × {±1} with natural ±1-action
and that X + = R × {1}, and we often identify X /{±1} with R.
1.2. For a set R, we frequently consider the power set P ( R ) as an abelian group under symmetric
difference. For a group G acting (from the left) on a set R, there is a G-action on P ( R ) defined by
g ( A ) = { g (a) | a ∈ A } for A ⊆ R; this makes P ( R ) a G-module i.e. an Z[G ]-module, where Z[G ] is the
integral group ring of G.
A cocycle (of G on P ( R )) is a function N : G → P ( R ) such that N (xy ) = N (x) + xN ( y ) for x, y ∈ G.
The cocycle N is called a coboundary if there is some subset A ⊆ R with N ( g ) = A + g A for all g ∈ G.
1.3. Let G be a group and X be a set with a left action of the group G × {±1} such that the induced
action of {±1} is free. There is a unique G-action on R := X /{±1} making the projection π : X →
X /{±1} = R a G-equivariant map; we write the G-action on R as ( g , x) → g · x to distinguish it from
the action ( g , x) → g (x) : G × X → X . Let X + be a set of orbit representatives for {±1} on X , and
identify R = X + as in 1.1. One readily checks (cf. [8, Proposition 1.5]) that the map N : G → P ( R )
defined by N ( g ) = {x ∈ X + | g −1 (x) ∈ / X + } is a cocycle. This gives a pair ( R , N ) where R is a G-set and
N : G → P ( R ) is a cocycle. To avoid possible confusion, we systematically write the induced action
of G × {±1} on P ( X ) as (h, A ) → h( A ) and the induced action of G on P ( R ) as ( g , A ) → g · A. We
abbreviate (−1)( A ) as − A for A ⊆ X .
On the other hand, given a set R with a left action ( g , r ) → g · r by a group G and a cocycle
N : G → P ( R ), one defines a left G × {±1}-action on the set X = R × {±1} by the formula
( g , )(r , ) = ( g · r , ν ) for g ∈ G, r ∈ R, , ∈ {±1} where ν = −1 if r ∈ N ( g −1 ) and ν = 1 other-
wise. Setting X + = X × {1}, one sees that ( R , N ) is the pair associated to ( X , X + ) as in the paragraph
above.
The above sets up a correspondence between pairs ( X , X + ) where X is set with a {±1}-free
G × {±1}-action and X + is a set of orbit representatives of {±1} on X , and pairs ( R , N ) consisting of
a G-set R and a cocycle N : G → P ( R ).
Remarks. The cocycle N depends on the choice of X + , but its first cohomology class is independent
of the choice i.e. different choices of the set X + of orbit representatives give corresponding cocycles N
which differ by a coboundary. Similar results to those above are well known to hold in much greater
generality; see e.g. [8, Appendix A].
1.4. Let s ∈ G with s2 = 1. Then ∅ = N (1G ) = N (s2 ) = N (s) + s · N (s) so N (s) = s · N (s). Hence N (s) and
R \ N (s) are s-invariant subsets of R. Let αs := {x ∈ X + | s(x) ∈ − X + }. From s · N (s) = N (s), s2 αs = αs
and s(αs ) ⊆ − X + , it follows that s(αs ) = −αs .
Remarks. The set {αs , −αs } may be regarded as a prototype for a permutation root system (it actu-
ally is one attached to a cyclic group of order two). In a similar way as one defines Coxeter groups
by imposing rank two relations on the generating involutions (simple reflections), we shall define
permutation root systems in general by imposing rank two conditions on a cocycle.
1.5. The following Lemma generalizes the above-noted s-invariance of N (s) for involutions s of G.

Lemma. Let N : G → P ( R ) be a cocycle. Then P := g ∈G N ( g ) is a G-invariant subset of R.
n n−1 −1
Proof. Let g ∈ G. Induction on n shows that i =0 N(gi ) = i =0 g i · N ( g ) and i =−n N(gi ) =
−1
i =−n g · N ( g ) for n ∈ N, from which N(gi ) = g i · N ( g ). This implies that if x ∈ G and
i
i ∈Z i ∈Z
m ∈ Z, then

N gm x ∪ N g n = g m · N (x) ∪ g n · N ( g ).
n∈Z n∈Z
Hence

N gn ∪ N gn x = g n · N ( g ) ∪ N (x) .
n∈Z n∈Z
So for g ∈ G,

N (x) = N gn ∪ N gn x = g n · N ( g ) ∪ N (x) .
x∈G x∈G x∈G
n∈Z n∈Z

Hence x∈G N (x) is g-invariant for each g ∈ G, and therefore it is G-invariant. 2
Remarks. Essentially
the same argument shows that if H is a subgroup of G and X ⊆ G with H ⊆
X = H X , then x∈ X N (x) is H -invariant.
1.6. Consider a group G, G-set R and cocycle N : G → P ( R ) as above. For any g 1 , . . . , gn ∈ G, the
cocycle condition gives
N ( g 1 · · · gn ) = N ( g 1 ) + g 1 · N ( g 2 ) + · · · + g 1 · · · gn−1 · N ( gn ) (1)
i.e. N ( g 1 · · · gn ) is the set of elements of R which are contained in an odd number of the sets
g 1 · · · g i −1 · N ( g i ) for i = 1, . . . , n. We shall say that ( g 1 , . . . , gn ) is a compatible sequence (in G,
with respect to N) with value g if g = g 1 · · · gn and (1) gives a partition of N ( g ) i.e. the sets
g 1 · · · g i −1 · N ( g i ) for i = 1, . . . , n are pairwise disjoint. More informally, we say that g = g 1 · · · gn
is a compatible expression for g (in G) or, even more loosely, just that g 1 · · · gn is compatible.
Equivalently, g 1 · · · gn is compatible iff N ( g 1 ) ⊆ N ( g 1 g 2 ) ⊆ · · · ⊆ N ( g 1 · · · gn ). If g 1 · · · gn is com-

patible, then gn−1 · · · g 1−1 is compatible. For any g , g 1 , . . . , gn in G and expressions g = g 1 · · · gn ,
g i = h i ,1 · · · h i ,mi for i = 1, . . . , n, the expression
g = h1,1 · · · h1,m1 h2,1 · · · h2,m2 · · · hn,1 · · · hn,mn
is compatible iff g = g 1 · · · gn is compatible and each expression g i = h i ,1 · · · h i ,ni is compatible.

For g , h ∈ G, the expression gh is compatible iff N ( g ) ⊆ N ( gh) iff N ( g −1 ) ∩ N (h) = ∅ iff N ( gh) =
N ( g ) ∪ g · N (h). We shall frequently use the above simple facts about compatibility without special
mention.
1.7. We shall make no use of the following observations about compatibility in this paper. To G, R,
N as in the previous subsection, one may attach an associative, unital ring A as follows: A is a free
Z-module with a basis of symbols t g for g ∈ G, and with multiplication determined by the formula

t gh , if gh is compatible,
t g th =
0, otherwise.
We also define an A-module M which is free as Z-module, with Z-basis of symbols m B for B ⊆ R,
and with A-module structure determined by the formula

m N ( g )+ g · B , if N ( g ) ∩ g · B = ∅,
t gmB =
0, otherwise.
If N is the trivial cocycle (i.e. N ( g ) = ∅ for all g ∈ G) then A identifies naturally with the group
algebra Z[G ] of G and M identifies with the permutation module (over Z) associated to the permu-
tation representation of G on P ( R ).
2. Coxeter group completions
2.1. The special cases of the formalism of Section 1 of interest in this paper are associated to Coxeter
groups. Let ( W , S ) be any Coxeter system and let T = { wsw −1 | w ∈ W , s ∈ S } denote its set of
reflections. There is (see [3, Chapitre IV]) a faithful permutation action of W on the set Ψ := T × {±1},
such that for any w ∈ W , t ∈ T and ∈ {±1},

1, if l( wt ) > l( w ),
w (t , ) = wt w −1 , w ,t , w ,t :=
−1, if l( wt ) < l( w ),
where l : W → N is the length function of ( W , S ). We regard W as a group of permutations

of T × {±1}. The formalism of Section 1 applies with G = W acting on Ψ .
Remarks. It is well known, and easily shown (see e.g. [8]) that as W -set, Ψ is isomorphic to the
standard (real, reduced) root system Φ of ( W , S ) as in [13]; indeed the map Φ → Ψ defined by
α → (sα , ) for ∈ {±1} and α ∈ Φ+ gives an isomorphism where sα ∈ T is the reflection in α .
2.2. Maintain the notation of 2.1. For α ∈ Ψ , sα ∈ T denotes the corresponding reflection, defined by
s(t , ) = t for (t , ) ∈ T × {±1}. Define W to be the set of all permutations θ of Ψ such that for any
finitely generated reflection subgroup W of W , there is some element w ∈ W (possibly depending
on W ) such that θ(t , ) = w (t , ) for all t ∈ W ∩ T and ∈ {±1}. An equivalent definition would
be obtained if we replaced “finitely generated reflection subgroup” by “finitely generated standard
parabolic subgroup” or by “finitely generated parabolic subgroup.” Generally, we denote the standard
parabolic subgroup generated by a subset K of S as W K . It is easy to see that W is a group of
permutations of Ψ , containing W as a subgroup. The action of W on Ψ clearly commutes elemen-

twise with the action of {±1}. Take Ψ+ := T × {1} as a set of orbit representatives for {±1} on Ψ .
Identify R := Ψ/{±1} = T via {(t , 1), (t , −1)} ↔ t for t ∈ T ; then the induced W -action on R = T is
the action of W on T by conjugation i.e. g · t = gt g −1 for g ∈ G, t ∈ T . As in 1.3, define the cocycle
N : W → P ( T ); explicitly, one has N ( g ) = {t ∈ T | g −1 (t , 1) = ( g −1 t g , −1)}.
We call the W × {±1}-set Ψ = T × {±1} with the set Ψ+ = T × {1} of {±1}-orbit representatives,
the standard based permutation root system of ( W , S ). We shall often regard it as W × {±1}-set.
It is immediate form the definitions, that for w ∈ W , N ( w ) = {t ∈ T | l(t w ) < l( w )}. The cocy-
cle w → N ( w ): W → P ( T ) is called the reflection cocycle of ( W , S ), [9], and we therefore call
N itself the reflection cocycle of W . Since | N ( w )| = l( w ) for w ∈ W , as is well known, it fol-
lows that an expression w 1 · · · w n with w i ∈ W is compatible (with respect to N) iff l( w 1 · · · w n ) =
l ( w 1 ) + · · · + l ( w n ).
Remarks. The group W is easily seen to be isomorphic to the group (denoted the same way) attached
to (certain reflection representations of) W in [6] and also plays a basic role in [8], where it is
generally denoted as G. From [8], there are examples of W for which W has non-trivial intersection
with the group of automorphisms of Ψ induced by a diagram automorphism of ( W , S ); in particular,
one may have non-trivial σ ∈ W with N (σ ) = ∅, which is impossible with σ ∈ W . In general, one
may view W as a “completion” of the group W including additional elements σ possibly of zero or
not necessarily finite “length” | N (σ )|. Compatibility of an expression g = g 1 · · · gn is a generalization
to
W of the length compatibility condition l( w 1 · · · w n ) = l( w 1 ) + · · · + l( w n ) for w i in W .
2.3. We collect for later use some basic facts about reflection subgroups of Coxeter systems from [9]
and [7].
Lemma. Let ( W , S ) be a Coxeter system with reflections T and standard length function l. Let W be a reflec-
tion subgroup of W i.e. W = T where T := W ∩ T . Define N : W → P ( T ) by N ( w ) = N ( w ) ∩ T .
(a) Let χ ( W ) = S := {t ∈ W | N (t ) = {t }}. Then ( W , S ) is a Coxeter system. We call S the canonical set
of Coxeter generators for W (with respect to ( W , S )).
(b) The set of reflections of ( W , S ) is T .
(c) N is the reflection cocycle of ( W , S ). Equivalently, for t ∈ T and w ∈ W , one has l (t w ) < l ( w ) iff
l(t w ) < l( w ) where l denotes the length function of ( W , S ).
(d) If w ∈ W , there is a unique element x ∈ W w such that N (x) = ∅. Moreover, χ ( w −1 W w ) =
x−1 χ ( W )x and N ( yx) ∩ W = N ( y ) for all y ∈ W . We say that x is the distinguished coset repre-
sentative of the coset W w.
(e) If W is generated by a set X ⊆ T , then |χ ( W )| | X |. In particular, if W is generated by two distinct
reflections of T , then |χ ( W )| = 2.
2.4. More generally than above, for a reflection subgroup W of W , we say that y ∈ W is a distin-
guished representative in yW if N ( y −1 ) ∩ W = ∅. Contrary to the case of cosets yW with y ∈ W ,
not every coset yW with y ∈
W need have a distinguished coset representative.
Example. Take ( W , S ) so that W is infinite and each of its irreducible components W i is a cyclic
group of order two, generated by y i , say. Then W is the group of all permutations of T × {±1} =
∈ {±1}, i ∈ I } which fix each subset { y i } × {±1} for i ∈ I . Hence
{( y i , ) | W identifies with the
product
i W i and the reflection cocycle N : W → P ( T ) is a bijection. Let y ∈ W with N ( y ) = T
( y corresponds to ( y i )i ∈ I in the product). Then for W = W , there is no distinguished representative
in W for the coset yW .
2.5. The next Lemma describes basic properties of

W and its action on Ψ .
Lemma. Let W be a reflection subgroup of W . Set T := W ∩ T and S := χ ( W ).
(a) For g ∈
W and α ∈ Ψ , s g (α ) = gsα g −1 .
(b) W W.
(c) For any g ∈ W , g W g −1 is a reflection subgroup of W .

(d) If g , h ∈ W and Γ ⊆ Ψ with g (α ) = h(α ) for all α ∈ Γ , then g (α ) = h(α ) for all
α ∈ ( sα | α ∈ Γ ∩ T ) × {±1}.
(e) If | S | is finite and g ∈ W , then N ( g ) ∩ T = N ( g ) ∩ T for some g ∈ W .
(f) If g ∈ W and N ( g ) ∩ T = ∅, then N ( g ) ∩ S = ∅.
Proof. It is well known (and an immediate consequence of the definitions) that wsα w −1 = s w (α ) for
w ∈ W and α ∈ Ψ . Now let g ∈ W and α , β ∈ Ψ . Choose a finite subset K ⊆ S with sα , sβ ∈ W K
and an element w ∈ W such that the restrictions of g and w to ( W K × {±1}) are equal. Since
sα (β) ∈ ( W K ∩ T ) × {±1}, we have g (sα (β)) = w (sα (β)) = s w (α ) ( w (β)) = s g (α ) ( g (β)). Since β is arbi-
trary, this gives that gsα = s g (α ) g as required for the first part of (a), Parts (b)–(c) follow immediately
from (a).
To prove (d), we may suppose without loss of generality that h = 1 W . Then it will suffice to show
that if α , β ∈ Ψ with g (α ) = α and g (β) = β , then g (sα (β)) = sα (β). But by (a) and the assumptions,
g (sα (β)) = s g (α ) ( g (β)) = sα (β).
Now we prove (e). There is some g ∈ W such that g (α ) = g (α ) for all α ∈ T × {±1}. Then
clearly N ( g ) ∩ T = N ( g ) ∩ T . Write g = g x where g ∈ W and N (x) ∩ T = ∅. Then N ( g ) ∩ T =
N ( g ) ∩ T = N ( g ) ∩ T .
Finally, we prove (f). We have N ( g ) ∩ W = ∅ for some finitely generated standard parabolic sub-
group W of W (i.e. a reflection subgroup W of W with χ ( W ) a finite subgroup of S ). Replacing
W by W , we may assume without loss of generality that S is finite. By (e), there is then some
g ∈ W with N ( g ) ∩ T = N ( g ) ∩ T = ∅. Then N ( g ) ∩ S = N ( g ) ∩ S = ∅ since g = 1. 2
2.6. We conclude this section by recording the following useful fact (the case s ∈ S of which was used
at the end of the proof of [6, Proposition 2.4]).
Lemma. Let w ∈ W and t ∈ T . Suppose that t = xsx−1 where s ∈ T , x ∈ W and l(xs) > l(x). Then l( wt ) >
l( w ) iff l( wxs) > l( wx).
Proof. Let N be the reflection cocycle of ( W , S ). Then l( wxs) < l( wx) iff s ∈ N (x−1 w −1 ) =
N (x−1 ) + x−1 N ( w −1 )x iff s ∈ x−1 N ( w −1 )x (since s ∈
/ N (x−1 )) iff t ∈ N ( w −1 ) iff l( wt ) < l( w ). The
result follows. 2
3. Permutation root systems
3.1. Let ( W , S ) be a Coxeter system with standard length function l and reflections T :=
{ wsw −1 | w ∈ W , s ∈ S }. We equip T × {±1} with the standard W × {±1}-action from 2.1.
Consider a pair ( X , X + ) where X is a set equipped with a given action of the group W × {±1} such
that the induced action of the subgroup {±1} on X is free, and X + is a set of orbit representatives
for {±1} acting on X . Letting R := X /{±1}, we have the bijection i : X + → X → X /{±1} = R.
As in 1.3, we have a corresponding action of W × {±1} on P ( X ), W -actions on R and P ( R ), and a
cocycle N : W → P ( R ) (depending on X + ). To avoid confusion, we write the actions on X and P ( X )
as ( g , A ) → g ( A ) for g ∈ G × {±1} and the W -actions on R and P ( R ) as ( w , A ) → w · A for w ∈ W .
Let αr := i −1 ( N (r )) ∈ P ( X ) for r ∈ S, Π = {αr | r ∈ S }, Φ = { w (α ) | w ∈ W , α ∈ Π}, Φ+ = Φ ∩ P ( X + )
and Φ− := {−α | α ∈ Φ+ }.
Proposition. The following conditions (i)–(iii) on the pair ( X , X + ) as above are equivalent:
(i) For any x, y in W with l(xy ) = l(x) + l( y ), the expression xy is compatible i.e. N (x) ⊆ N (xy ).
(ii) (1) There exist subsets R t of R, for t ∈ T , such that R r = N (r ) for r ∈ S and R wt w −1 = w · R t for t ∈ T ,
w ∈ W . (2) If t , t ∈ T with R t ∩ R t = ∅, then t = t . (3) For w ∈ W , N ( w ) = t ∈T Rt .
l(t w )<l( w )
(iii) (1) There are well-defined subsets α(t , ) of X for (t , ) ∈ T × {±1} such that α(r ,1) = αr for r ∈ S, and
( w , )(α(t , ) ) = α( w , )(t , ) for all ( w , ) ∈ W × {±1} and (t , ) ∈ T × {±1}. (2) If α(t , ) ∩ α(t , ) = ∅,
then (t , ) = (t , ). (3) One has Φ = {α(t , ) | (t , ) ∈ T × {±1}} with α(t ,1) ∈ Φ+ and α(t ,−1) ∈ Φ− for
t ∈ T.
Proof. For the proof, identify R with X + via the bijection i, so αr = N (r ). We first show that (i)
implies (ii). Assume (i) holds. To prove that the sets R t are well-defined by the conditions in (ii)(1),
it suffices to show that if r , s ∈ S and w ∈ W with wr w −1 = s, then w · N (r ) = N (s). By 1.4, re-
placing w if necessary by an element of minimal length in the double coset {1, s} w {1, r }, we may
suppose without loss of generality that l(sw ) > l( w ) and l( wr ) > l( w ). We have wr = sw, so by (i),
N (s) ∩ N ( w ) = ∅ and N (s) ⊆ N (sw ) = N ( wr ) = N ( w ) ∪ w · N (r ). This implies N (s) ⊆ w · N (r ). The
same argument applied to w −1 s = r w −1 shows that N (r ) ⊆ w −1 · N (s). Hence N (s) = w · N (r ) as
required. To prove that (ii)(2) holds, it will suffice to show that if s ∈ S and t ∈ T with R s ∩ R t = ∅,
then t = s. Write t = wr w −1 with w ∈ W and r ∈ S. We have to show that if N (s) ∩ w · N (r ) = ∅,
then s = wr w −1 . We may again assume without loss of generality that l(sw ) > l( w ) and l( wr ) > l( w ).
Suppose first that l(swr ) = l( w ) + 2. The condition (i) implies that the expression swr is compatible.
Hence the sets N (s), s · N ( w ) and sw · N (r ) are pairwise disjoint, which gives that s · N (s) = N (s) and
s · sw · N (r ) = w · N (r ) are disjoint. This contradiction implies that we must have l(swr ) = l( w ) + 2,
and then the exchange condition for ( W , S ) implies that sw = wr and s = wr w −1 as required. Finally,
we show that (ii)(3) holds. Choose a reduced expression w = r1 · · · rn for w. By (i), the expression
w = r1 · · · rn is compatible, and so
N ( w ) = N (r1 ) ∪ r1 · N (r2 ) ∪ · · · ∪ r1 · · · rn−1 · N (rn )
(disjoint union). Let t i := r1 · · · r i −1 r i r i −1 · · · r1 for i = 1, . . . , n. The above shows that N ( w ) =

R t1 ∪ R t2 ∪ · · · ∪ R tn . But it is well known consequence of 2.1–2.2 that {t 1 , . . . , tn } = {t ∈ T | l(t w ) <
l( w )}, so (ii)(3) holds. This completes the proof that (i) implies (ii).
Next, we show that (ii) implies (iii). Set α(t ,1) = R t and α(t ,−1) = − R t for t ∈ T . Clearly, (−1)α(t , ) =
α(t ,− ) so to prove (iii)(1) it suffices to show that A := w (α(t ,1) ) = α w (t ,1) i.e. A = α( wt w −1 , ) where
= −1 if l( wt ) < l( w ) and = −1 otherwise. We have A ∩ − A = ∅ (since α(t ,1) ∩ −α(t ,1) = ∅). Let
π : X → R = X + be the natural projection. Since π is W -equivariant, we have π ( A ) = w · π (α(t ,1) ) =
w · R t = R wt w −1 , and so A ⊆ R wt w −1 ∪ − R wt w −1 . Now if l( wt ) < l( w ), then l(t w −1 ) < l( w −1 )
so R t ⊆ N ( w −1 ) by (ii)(3), w ( R t ) ⊆ − X + by definition of N and A = − R wt w −1 = α( wt w −1 ,−1) =
α( wt w −1 , ) . Similarly, if l( wt ) > l( w ), then l(t w −1 ) > l( w −1 ) so R t ∩ N ( w −1 ) = ∅ by (ii)(3), w ( R t ) ⊆ X +
by definition of N and A = R wt w −1 = α( wt w −1 ,1) = α( wt w −1 , ) . Hence (iii)(1) holds. Next, (iii)(2) fol-
lows immediately from (ii)(2). Finally, the first part of (iii)(3) follows from (iii)(1) and the definition
of Φ , and the remainder follows since α(t ,1) ⊆ X + and α(t ,−1) ⊆ − X + for all t ∈ T . This proves that (ii)
implies (iii).
we prove that (iii) implies (i). Assume that (iii) holds. We first claim that for w ∈ W ,
Finally,
N(w) = t ∈T α(t ,1) . For the inclusion “⊇,” observe that if z ∈ α(t ,1) where l(t w ) < l( w ), then
l(t w )<l( w )
w −1 ( z) ∈ w −1 (α(t ,1) ) = α( w −1 t w ,−1) so z ∈ X + ∩ w (− X + ) = N ( w ). For the reverse inclusion, let w :=
r1 · · · rn be a reduced expression for w, and set t i = r1 · · · r i −1 r i r i −1 · · · r1 for i = 1, . . . , n. Let z ∈ N ( w )
i.e. z ∈ X + with rn · · · r1 ( z) ∈ − X + . There is some i with 1 i n such that r i −1 · · · r1 ( z) ∈ X + and
r i r i −1 · · · r1 ( z) ∈ − X + . Then r i −1 · · · r1 (x) ∈ αri so z ∈ r1 · · · r i −1 (α(ri ,1) ) = α(t i ,1) with t i ∈ N (t ). This
proves the claim. Now (i) follows from the claim since if x, y ∈ W , then

t ∈ T
l(tx) < l(x) ⊆ t ∈ T
l(txy ) < l(xy ) if l(xy ) = l(x) + l( y ). (3.1.1)
Hence (iii) implies (i), and the Proposition is proved. 2
3.2. When the conditions of Proposition 3.1 hold, we say that Φ is a based permutation root system
of ( W , S ) in X with respect to X + , with simple roots Π , positive roots Φ+ and negative roots Φ− .
In terms of N and R, we shall say that N : W → P ( R ) is a root cocycle. Observe that Φ , R and N
are completely determined by ( W , S ), X as W -set, and the subset X + of X ; in fact, even the indexed
families of subsets {α(t , ) }(t , )∈ T ×{±1} and { R t }t ∈ T are uniquely
determined by this data.

When 3.1(i)–(iii) hold, W acts naturally on { R t | t ∈ T }, on t ∈ T R t , on Φ and on α ∈Φ α ⊆ X . In
general, the action of W on Φ (or even on X ) need not be faithful and the sets R t , α(t , ) could be
empty (e.g. if X = ∅ and W is non-trivial).
3.3. The following Corollary records some simple facts which are immediate consequences of Propo-
sition 3.1 and its proof.
Corollary. Let Φ be a based permutation root system of ( W , S ) in X with respect to X + , associated to the root
cocycle N on P ( R ).
(a) Φ = Φ+ ∪ Φ− (disjoint union).

(b) For r ∈ S, r (αr ) = −αr and r · N (r ) = N (r ).
(c) Φ+ = { w (αr ) | r ∈ S , w ∈ W , l( wr ) > l( w )}.
(d) Suppose that N (r ) = ∅ for all r ∈ S. Then all sets R t and α(t , ) are non-empty. The map (t , ) →
α(t , )
is an isomorphism of W × {±1}-sets T × {±1} ∼ = Φ , and W acts faithfully on Φ , faithfully on α ∈Φ α
and faithfully on X . In this case, we say that Φ (and N) are non-degenerate; otherwise they are said to be
degenerate.
3.4. To illustrate the above notions, we give a trivial first construction of new based permutation root
systems from existing ones, replacing each abstract root for the original one by its product with some
set (in a consistent way).
Example. Let Φ be a based permutation root system of ( W , S ) in X with respect to X + , with notation
as in 3.2. Let X i , for i in an index set I , denote the set of W × {±1}-orbits on X .Fix a set A i
for each i ∈ I ; one could for instance take all A i equal to a fixed set A. Let Z := i ∈ I ( A i × X i )
with W × {±1}-action given by g (ai , xi ) = (ai , g (xi )). The induced {±1}-action on Z is free, with
Z
+ := i ∈ I A i × ( X i ∩ X + ) as set of orbit representatives. For any (t , ) ∈ T × {±1}, define β(t , ) :=
i ∈ I A i × (α(t , ) ∩ X i ). It is easy to see that Ψ := {β(t , ) | (t , ) ∈ T × {±1}} defines a based permutation
root system in Z with respect to Z + , with positive roots {β(t ,1) | t ∈ T } and simple roots {β(s,1) | s ∈ S }.
Remarks. One has a natural W × {±1}-equivariant map π : Z → X given by (ai , xi ) → xi for i ∈ I ,

ai ∈ A i , xi ∈ X . This may be viewed as an analogue (see [8, Appendix A]) in the category of W × {±1}-
sets of a fiber bundle in the category of topological spaces. In the example above, the bundle is
“trivial.” In general, any based permutation root system of ( W , S ) may be similarly regarded as such
a bundle over the standard based permutation root system, and for many classes of examples arising
later in this paper, or in other natural ways, this bundle is non-trivial. We leave the interested reader
to examine the generalization of the above construction using possibly non-trivial bundles.
3.5. One motivation for defining based permutation root systems is to abstract some features, espe-
cially in relation to their embeddings, of root systems over commutative rings (as considered in [6])
and over non-commutative rings, as we consider elsewhere. We record here how a root basis datum
over a commutative ring, in the sense of [6], gives rise to a based permutation root system.
Let B = ( R , R + , M , M ∨ , ?, ?, Π, Π ∨ , ι) be a root basis datum over the commutative ring R, as
defined in [6, 2.3]. Let ( W , S ) be the associated Coxeter system, with reflections T , and let Π ⊆ Φ+ ⊆
Φ be the simple roots, positive roots and roots respectively; they are subsets of the R-module M. For
each α ∈ Φ , we have the corresponding reflection sα ∈ T .
Define X = Φ and X + := Φ+ . Then X has a natural structure of W × {±1}-set (by restriction
of the R-linear W -action on M), and the group {±1} acts freely on X with X + as a set of orbit
representatives. For each (t , ) ∈ T × {±1}, define α(t , ) := {β ∈ Φ+ | sβ = t }. Let Ψ = {α(t , ) | (t , ) ∈
T × {±1}}, Ψ+ = {α(t ,1) | t ∈ T }, and Δ := {α(s,1) | s ∈ S }.
The following is immediate from the properties of root basis data listed in [6, §2] and the defini-
tion 3.2 of based permutation root system.
Proposition. The set Ψ defined above is a non-degenerate based permutation root system of ( W , S ) in X with
respect to X + , with positive roots Ψ+ and simple roots Δ.
3.6. Somewhat more generally, let (Φ, F ) be an abstract root system, and Φ+ be a positive system
of Φ with simple roots Δ and associated Coxeter system ( W , S ), in the sense of [8]. We will not
repeat the definition here in full, but recall from loc cit that (in particular) Φ is a set with a fixed-
point free action of the sign group, Φ+ is a set of orbit representatives for {±1} on Φ , and F is a
map from Φ to the group of permutations of Φ . We identify W with the group of permutations of Φ
generated by sα := F (α ) for α ∈ Φ , and let T = {sα | α ∈ Φ} be the set of reflections of ( W , S ).
Proposition. Taking X = Φ , X + = Φ+ , α(t , ) = {β ∈ Φ+ | sβ = t } and Ψ , Ψ+ , Δ defined by the same

equations as in 3.5 gives a non-degenerate based permutation root system Ψ of ( W , S ) in X with respect
to X + , with positive roots Ψ+ and simple roots Δ.

Remarks. Observe that in 3.5 and 3.6, α ∈Ψ α = X and that for each β ∈ X , one has a permu-
tation sβ ∈ T of X (in fact, sβ = t if β ∈ α(t , ) ). This extra structure given by the sβ for β ∈ X is
not necessarily present for arbitrary permutation root systems and so the permutation root systems
arising as above from abstract root systems (or root basis data) are very special. For examples of
based permutation root systems associated to root system embeddings in Section 6 of this paper,
we will have a permutation sβ of X defined for each β ∈ X , but often X = α ∈Ψ α , and often for
(t , ) ∈ T × {±1}, we will have t = β∈α(t , ) sβ (in suitable order), with sβ ∈
/ W instead of sβ = t ∈ W
for β ∈ α(t , ) .
3.7. As another example, let X be the set of all roots (real and imaginary) of a Kac–Moody Lie al-
gebra [14], with positive roots X + , real roots denoted Φ and associated Coxeter group ( W , S ) with
reflections T . The set of positive real roots is Φ+ = X + ∩ Φ . For each α ∈ Φ , we have the correspond-
ing reflection sα ∈ T . Regard W as a group of permutations of X .
Proposition. Taking α(t , ) = {β ∈ Φ+ | sβ = t } (a singleton set) for (t , ) ∈ T × {±1} and Ψ , Ψ+ , Δ defined

by the same equations as in 3.5 gives a non-degenerate based permutation root system Ψ of ( W , S ) in X with
respect to X + , with positive roots Ψ+ and simple roots Δ.

Remarks. For any based permutation root system, X + \ α ∈Φ+ α is W -stable; in the above example,
this set is the set of positive imaginary roots.
3.8. We now illustrate the notions of root cocycles and of based permutation root systems by some
concrete examples of types arising later in the paper.
Example. Consider the symmetric group W on {1, 2, . . . , 5} as a Coxeter group (of type A 4 ) with
simple reflections S consisting of the adjacent transpositions (i , i + 1) for i = 1, . . . , 4. Let T denote
the set of reflections T := {(i , j ) | 1 i < j 5} of G, and regard T × {±1} as the standard based
permutation root system of ( W , S ). We have the corresponding action of W on T by conjugation
and the reflection cocycle N : W → P ( T ). We take R := T with W -action by conjugation.
(1) Let S = {(1, 2)(3, 4), (2, 3)} and W = S . It is easy to see that ( W , S ) is a Coxeter system of
type B 2 , with reflections

T := a := (1, 2)(3, 4), b := (2, 3), aba = (1, 4), bab = (1, 3)(2, 4) .
The cocycle w → N ( w ) : W → P ( R ) is easily seen to be a (non-degenerate) root cocycle of ( W , S )

with
R a = {(1, 2), (3, 4)}, R aba = {(1, 4)}, R bab = {(1, 3), (2, 4)}, R b = {(2, 3)}.
The roots of the corresponding based permutation root system of ( W , S ) in X := T × {±1} with
respect to X + := T × {1} are the subsets α(t , ) = {(x, ) | x ∈ R t } of T × {±1}, for (t , ) ∈ T × {±1}.
(2) Let Z := {(1, 5), (4, 5), (2, 5), (3, 5)} which we observe is a W -stable subset of T . We introduce
the action of {±1} on Z such that −(1, 5) = (4, 5) and −(2, 5) = (3, 5). This makes Z a W × {±1}-
set. Fix the set Z + = {(3, 5), (4, 5)} of {±1}-orbit representatives on Z . Define subsets β(t , ) of Z for
(t , ) ∈ T × {±1} by
β(b,1) = {(3, 5)}, β(b,−1) = {(2, 5)}, β(aba,1) = {(4, 5)}, β(aba,−1) = {(1, 5)}
and β(a, ) = β(bab, ) = ∅ for ∈ {±1}. Then Ψ := {β(t , ) | (t , ) ∈ T × {±1}} is a (degenerate) based
permutation root system of ( W , S ) in Z with respect to Z + , with Ψ+ = {β(t ,1) | t ∈ T }.
3.9. Suppose that ( W , S ) is a Coxeter system and N : W → P ( R ) is a cocycle. The following Proposi-
tion gives weaker conditions than those of Proposition 3.1(ii) for N to be a root cocycle. We use here
and later the notation

(xy )k/2 , if k is even,
(xyx · · ·)k =
(xy )k/2 x, if k is odd
for k ∈ N and elements x, y of any group.
Proposition.
(a) There is at most one family of subsets { R t }t ∈ T of W satisfying the following two conditions (i)–(ii):
(i) R s = N (s) for each s ∈ S,
(ii) R wt w −1 = w · R t for w ∈ W , t ∈ T .
(b) If N is a coboundary of G, say N ( g ) = g · A + A for all g ∈ G for some A ⊆ R, then the family { R t }t ∈ T in
(a) exists iff the set A + w r ,s · A is r , s-invariant for all distinct r , s ∈ S with mr ,s finite, where w r ,s =
(rsr · · ·)mr,s = (srs · · ·)mr,s .
(c) If the family { R t }t ∈ T exists in (a), then N is a root cocycle iff R t ∩ R t = ∅ implies t = t .
Proof. Part (a) is obvious. It also is clear that the R t are well-defined iff for each r , s ∈ S and w ∈ W
with wr w −1 = s, one has w ( N (r )) = N (s). A well-known fact about conjugacy of simple reflections
(see [8, Lemma 1.29]) then implies, along with 1.4, that { R t }t ∈ T exists in (a) iff for each r , s ∈ S
with r = s and the order mr ,s of rs finite, one has w r ,s · N (r ) = N (s ) for each r , s ∈ {r , s} with
r w r ,s = w r ,s s . Part (b) follows by a simple computation from this.
“Only if” in (c) follows from Proposition 3.1. For “if” in (c), suppose that { R t }t ∈ T exists. Let w =
r1 · · · rn be a reduced expression for w in ( W , S ), and set t i = r1 · · · r i −1 r i r i −1 · · · r1 ∈ T for i = 1, . . . , n,
so N (t ) = {t 1 , . . . , tn } with the t i pairwise distinct. Since R t ∩ R t = ∅ for t = t , we have R t i ∩ R t j = ∅
for i = j, so the cocycle condition gives

N ( w ) = R t 1 + · · · + R tn = R t 1 ∪ · · · ∪ R tn = Rt .
t ∈T
l(t w )<l( w )
The condition 3.1(i) follows using (3.1.1), completing the proof of (c). 2
3.10. We conclude this section with an observation about restrictions of root cocycles to Coxeter
subgroups. Let ( W , S ) be a Coxeter system and N : W → P ( T ) be its reflection cocycle. Also, let
N : W → P ( R ) be a root cocycle of ( W , S ), for some set R.
Proposition. Fix a Coxeter system ( W , S ) with W a subgroup of W , and let i : W → W be the inclusion
map.
(a) If N i : W → P ( T ) is a (necessarily non-degenerate) root cocycle of ( W , S ), then Ni : W → P ( R ) is a

root cocycle of ( W , S ).
(b) If N is non-degenerate and Ni is a root cocycle of ( W , S ) then N i is a root cocycle of ( W , S ); in that case,
both Ni and N i are non-degenerate.
Proof. Let l, l be the length function of ( W , S ) and ( W , S ), respectively. Then for w ∈ W , N ( w ) =

{t ∈ T | l (t w ) < l ( w )} and N ( w ) = t ∈ N ( w ) R t where the R t are pairwise disjoint. Consider
x, y ∈ W with l(xy ) = l(x) + l( y ). If N i is a root cocycle, then N (x) ⊆ N (xy ), so l (xy ) = l (x) + l ( y )
and N (x) ⊆ N (xy ), which shows that Ni is a root cocycle. On the other hand, if N is non-degenerate
and Ni is a root cocycle, then

R t = N (x) ⊆ N (xy ) = Rt
t ∈ N (x) t ∈ N (xy )
with the R t non-empty and pairwise disjoint implies that N (x) ⊆ N (xy ), which shows that N i is a
root cocycle. Note N i is necessarily non-degenerate, because if s ∈ S, then N (s) = ∅ since s = 1 W .
Finally, if N is non-degenerate, then N ( w ) = ∅ for all 1 W = w ∈ W , so N (s) = ∅ for s ∈ S and Ni is
non-degenerate. This completes the proof. 2
4. Root systems and coboundary root cocycles
4.1. Let Φ bea based permutation

root system of ( W , S ) in X , as in Section 3, with positive roots
Φ+ . Let A := α ∈Φ+ α = t ∈ T α(t ,1) . For w ∈ W ,

A + w ( A) = α( wt w −1 ,1) + α( wt w −1 ,−1) + α( wt w −1 ,1)
t ∈T t ∈T t ∈T
l( wt )<l( w ) l( wt )>l( w )

= (α(t ,1) ∪ α(t ,−1) )
t ∈T
l(t w )<l( w )
Hence if x, y ∈ W with l(xy ) = l(x) + l( y ), we have A + x( A ) ⊆ A + xy ( A ) by (3.1.1) i.e. the coboundary

w → A + w ( A ) : W → P ( X ) is a root cocycle. The sets X t analogous to R t in Proposition 3.1(ii) (i.e.
the pairwise disjoint subsets of X satisfying X s = A + s( A ) for s ∈ S and w ( X t ) = X wt w −1 for t ∈ T ,
w ∈ W ) are X t := α(t ,1) ∪ α(t ,−1) for t ∈ T .
4.2. By 4.1, a based permutation root system Φ in X gives rise to a coboundary root cocycle on X (as
well as to the root cocycle on R := X /{±1} as in Proposition 3.1, which is generally not a coboundary).
Here we establish a partial converse by showing that a coboundary root cocycle on a set X gives rise
to a based permutation root system on a subset of X (as well as to the based permutation root system
on X × {±1} given by Proposition 3.1).
Let N : W → P ( R ) be a root cocycle for ( W , S ). Assume that N is a coboundary, say N ( w ) =
A + w · A for all w ∈ W , where A ⊆ R. Define sets β(t ,1) := R t ∩ A and β(t ,−1) := R t \ A for t ∈ T ,
where { R t }t ∈ T is associated to N as in Proposition 3.1. Set Z := t ∈ T R t (disjoint union) with W -
action ( w , z) → w · z. We define an action of {±1} on Z , with the sets R t {±1}-stable, by setting
1z := z and (−1) z := t · z for all z ∈ R t (note t · R t = R t ).
Proposition.
(a) The above-defined

W action and {±1}-action on Z commute, making Z a W × {±1}-set.
(b) Z + := t ∈ T β(t ,1) ⊆ Z is a set of orbit representatives for {±1} on Z .
(c) The family Ψ := {β(t , ) | (t , ) ∈ T × {±1}} is a based permutation root system for ( W , S ) on Z with
respect to Z + , with positive roots Ψ+ := {β(t ,1) | t ∈ T } and simple roots Δ := {β(s,1) | s ∈ S }.
Proof. Since
R t ⊆ N (t ) = A + t · A = ( A \ t · A ) ∪ (t · A \ A ),
and t ( R t ) = R t , we actually have β(t ,1) = R t ∩ ( A \ t · A ) and β(t ,−1) = R t ∩ (t · A \ A ). This implies that
t (β(t , ) ) = β(t ,− ) , that β(t , ) ∩ β(t , ) = ∅ implies that (t , ) = (t , ) (by 3.1(ii)) and that (b) holds. For
t ∈ T , z ∈ R t and w ∈ W , we have w · z ∈ R wt w −1 ; since w · (t · z) = ( wt w −1 ) · ( w · z), the action of w
and −1 on Z commute and we see that (a) holds.
The main part of the proof is to show that for w ∈ W and (t , ) ∈ T × {±1},

−1, if l( wt ) < l( w ),
w · β(t , ) = β( wt w −1 , ) where = (4.2.1)
1, if l( wt ) > l( w ).
We first prove this for t = s ∈ S. Suppose that l( ws) > l( w ). Then N ( ws) = N ( w ) + w · N (s) =
N ( w ) ∪ R wsw −1 (disjoint union). That is,
A + ( ws) · A = ( A + w · A ) ∪ R wsw −1
(disjoint union). Let z ∈ β(s,1) = A \ s · A ⊆ R s . Hence w · z ∈ w · A \ ( ws) · A and in particular,

w·z∈ / ( ws) · A. But

w · z ∈ R wsw −1 ⊆ A + ( ws) · A = A \ ( ws) · A ∪ ( ws) · A \ A )

so in fact w · z ∈ R wsw −1 ∩ A \ ( ws) · A ⊆ A ∩ R wsw −1 = β( wsw −1 ,1) . This shows that
w (β(s,1) ) ⊆ β( wsw −1 ,1) . Hence
w (β(s,−1) ) = w (−β(s,1) ) ⊆ − w (β(s,1) ) = β( wsw −1 ,−1)
also. Since β(s,1) ∪ β(s,−1) = R s (disjoint union) and w · R s = R wsw −1 , it follows that
w (β(s,1) ) ∪ w (β(s,−1) ) = R wsw −1 = β( wsw −1 ,1) ∪ β( wsw −1 ,−1)
(disjoint unions). Hence w (β(s, ) ) = β( wsw −1 , ) if l( ws) > l( w ). Since s(β(s, ) ) = β(s,− ) , this implies
w (β(s, ) ) = β( wsw −1 ,− ) if l( ws) < l( w ) proving (4.2.1) for t = s ∈ S. To prove (4.2.1) in general, write
t = xsx−1 where s ∈ S, x ∈ W and l(xs) > l(x). By what has already been shown,

w · β(t , ) = w · x · β(s, ) = ( wx) · β(s, ) = β( wxsx−1 w −1 , ) = β( wt w −1 , )
where = 1 if l( wxs) > l( wx) and = −1 if l( wxs) < l( wx). Using Lemma 2.6, one has = ,
completing the proof of the Proposition. 2
4.3. We state a very simple Lemma for use in the next subsection.
Lemma. Assume that N i : W → P ( R ) is a cocycle for i = 1, 2. Let N := N 1 + N 2 be their sum (also a cocycle
W → P ( R )). Let P i = w ∈ W N i ( w ) ⊆ R for i = 1, 2.
(a) If N 1 is an arbitrary cocycle and N 2 is the coboundary defined by N 2 ( w ) = A + w · A for some A ⊆ R \ P 1 ,

then P 1 ∩ P 2 = ∅.
(b) If P 1 ∩ P 2 = ∅, then N is a root cocycle of ( W , S ) iff N 1 and N 2 are root cocycles of ( W , S ).
Proof.For (a), note that ifA ∩ P 1 = ∅ , then w · A ∩ P 1 = ∅ for w ∈ W by Lemma 1.5 and hence
P 2 = w ∈ W ( A + w · A ) ⊆ w ∈ W w · A is disjoint from P 1 as required.
For (b), observe that if w ∈ W , then N ( w ) = N 1 ( w ) ∪ N 2 ( w ) with N i ( w ) ⊆ P i where P 1 ∩ P 2 = ∅.
Hence for x, y ∈ W with l(xy ) = l(x) + l( y ), we have N (x) ⊆ N (xy ) iff N i (x) ⊆ N i (xy ) for i = 1, 2. This
proves (b). 2
4.4. Let N : G → P ( R ) be a cocycle, and ( W , S ) be a Coxeter system with W a subgroup of G. Then

for any g ∈ G, the map w → g −1 · N ( g w g −1 ) : W → P ( R ) is a cocycle of W ; indeed, for x ∈ W , the
defining property of a cocycle implies that

g −1 · N gxg −1 = N (x) + ( A + x · A )
where A := N ( g −1 ), so w → g −1 · N ( g w g −1 ) : W → P ( R ) is the sum of the cocycle

x → N (x) : W → P ( R ) and the coboundary x → A + x · A : W → P ( R ).
Corollary. Assume that g is G such that each expression g w with w ∈ W is compatible with respect to N
(we call such an element g a distinguished coset representative in g W ; this is consistent with our existing
terminology from Section 2 in case W is a reflection subgroup). Let A := N ( g −1 ). Then the following conditions
(i)–(iii) are equivalent:
(i) The cocycle x → N (x) : g W g −1 → P ( R ) of g W g −1 is a root cocycle of the Coxeter system

( g W g −1 , g S g −1 ).
(ii) The cocycle x → g −1 · N ( gxg −1 ) : W → P ( R ) of W is a root cocycle of ( W , S ).
(iii) Both the cocycle w → N ( w ) : W → P ( R ) of W and the coboundary w → A + w · A : W → P ( R ) of W
are root cocycles of ( W , S ).
Proof. Note that (i) is equivalent to the statement that for x, y ∈ W with l(x) + l( y ) = l(xy ), we
have N ( gxg −1 ) ⊆ N ( gxyg −1
) or equivalently that g −1 · N ( gxg −1 ) ⊆ g −1 · N ( gxyg −1 ). Hence (i) is
equivalent to (ii). Let P := w ∈ W N ( w ). Since g is distinguished in g W , we have P ∩ A = ∅. Hence
the equivalence of (ii)–(iii) follows from Lemma 4.3 by the comments preceding the statement of the
Corollary. 2
4.5. The following Example is a continuation of Example 3.8, and we maintain the notation in use
there.
Example. Let g denote the 3-cycle (3, 4, 5) in W . It is easy to see that g is a distinguished coset
representative in its coset g W (not every coset of W in W has such a distinguished representative;
for example, the coset (1, 2) W does not. Some general information on the structure of the set of
distinguished coset representatives can be obtained as a special case of the results of [2]).
One can check that the cocycle w → N ( w ) : g W g −1 → P ( R ) is a root cocycle of ( g W g −1 , g S g −1 ).
(This fails with g replaced by (4, 5), which is distinguished in (4, 5) W .) Let A := N ( g −1 ) =
{(3, 5), (4, 5)}. From Corollary 4.4, we have that w → N ( w ) : W → P ( T ) is a root cocycle of ( W , S )
(as already noted in Example 3.8(1)) and that w → A + w · A is a coboundary root cocycle of ( W , S ).
The based permutation root system root system Ψ defined in Example 3.8(2) is that attached by 4.2
to the (coboundary) root cocycle w → A + w · A.
There are examples in which a based permutation root system Ψ arising in a similar way to that
above is non-degenerate.
5. Rank two characterization of root cocycles
5.1. Throughout this section, G is a group acting on a set R and N : G → P ( R ) is a cocycle.
Definition. A subset S of G will be called an admissible subset of G (with respect to N), or said to be
admissible in G, if the following conditions hold.
(i) For any s ∈ S, s2 = 1G and N (s) = ∅ (so s is an involution).

(ii) Let r , s be distinct elements of S and mr ,s = | rs| ∈ N2 ∪ {∞}. Then the expression (rsr · · ·)i is
compatible (for N) for any i ∈ N with i mr ,s .
(iii) If r , s ∈ S are distinct, g ∈ G and N ( g ) ∩ N (x) = ∅ for some x ∈ r , s, then N ( g ) ∩ ( N (r ) ∪ N (s)) = ∅.
5.2. In general, a subset S of G is admissible iff all subsets of S of cardinality at most two are
admissible. For a subgroup G of G, we say that a subset S ⊆ G is admissible in G if it is admissible
with respect to the cocycle of G given by restriction of N.
Example.
(1) The empty set is admissible. If r is any involution in G with N (r ) = ∅, then {r } is admissible.
(2) If r, s are commuting involutions in G such that N (r ) ∩ N (s) = ∅ but N (r ), N (s) = ∅, then {r , s} is
an admissible subset of G (and r · N (s) = N (s), s · N (r ) = N (r )), as one sees on noting that both
unions in
N (r ) ∪ r · N (s) = N (rs) = N (sr ) = N (s) ∪ s · N (r )
are unions of disjoint sets.

(3) Let ( W , S ) be a dihedral Coxeter system with S = {r , s}, and let m denote the order of rs.
Suppose G = W and that N : W → P ( T ) is the reflection cocycle of ( W , S ). The admissible
subsets of W have at most two elements. Those with at most one element are described by (1)
above (note the involutions of W are its reflections, together with the longest element if W is
finite). The admissible subsets of W with two elements are the sets {(rsr · · ·)k , (srs · · ·)l } where
k and l are odd natural numbers such that k + l m and either k + l divides m, or k = l and k
divides m (here and elsewhere in the paper, we use the natural arithmetic and order conventions
on N ∪ {∞}, which imply in particular that ∞ is divisible by every positive integer; similarly for
cardinal numbers).
5.3. Assume that S ⊆ G consists of involutions and let W = S denote the subgroup of G generated
by S. Let l : W → N be the length function of the pair ( W , S ) i.e. for w ∈ W ,
l( w ) := min{n ∈ N | w = r1 · · · rn for some r i ∈ S }.
We say that an expression w = r1 · · · rn is a reduced expression of w ∈ W if all r i ∈ S and n = l( w ).

Set T := { wsw −1 | w ∈ W , s ∈ S }.
Theorem. Assume that S is admissible in G (or even just in W ). Then
(a) The pair ( W , S ) is a Coxeter system with l as its standard length function and T as its set of reflections.
(b) The cocycle w → N ( w ) : W → P ( R ) is a non-degenerate root cocycle of ( W , S ).
Remarks. It would be interesting to have a version of the theorem for possibly degenerate root cocy-
cles.
5.4. Theorem 5.3 will be proved in 5.4–5.8, throughout which notation and assumptions are as in the
Theorem. The following terminology extends that in [17]. We say that g ∈ G is admissible at s ∈ S
if either N (s) ⊆ N ( g ) or N (s) ∩ N ( g ) = ∅. We say that g ∈ G is admissible at S if it is admissible at
all s ∈ S.
Note that the assumption N (s) = ∅ for s ∈ S implies that we cannot have both N (s) ⊆ N ( g ) and
N (s) ∩ N ( g ) = ∅ for g ∈ G; we use this frequently, without special mention.
5.5. Let r , s ∈ S with r = s. Set S r ,s = {r , s} and W r ,s = S r ,s .Then ( W r ,s , S r ,s ) is a dihedral Coxeter

system of order 2mr ,s where mr ,s := | rs|. Let T r ,s = { waw −1 | w ∈ W r ,s , a ∈ S r ,s } denote the set of all
reflections of this dihedral Coxeter group, lr ,s denote its length function and R r ,s := x∈ W r ,s x · N (a).
a∈ S r ,s
Lemma.
(a) There are well-defined, pairwise disjoint subsets R r ,s;t of R r ,s for t ∈ T r ,s such that R r ,s;t = w · N (a)
−1
whenever t = waw with w ∈ W r ,s and a ∈ S r ,s .
(b) R r ,s = w ∈ W r ,s N ( w ).

(c) R r ,s = t ∈ T r ,s R r ,s;t (disjoint union).

(d) For w ∈ W r ,s , N ( w ) = t ∈ T r ,s R r ,s;t (disjoint union).
lr ,s (t w )<lr ,s ( w )
(e) Any w ∈ W r .s is admissible at both r and s.
Proof. From the assumptions 5.1(i)–(ii), the map w → N ( w ) : W r ,s → P ( R ) is a non-degenerate root

cocycle of ( W r ,s , S r ,s ). Then all parts of the Lemma follow from the results on root cocycles of general
Coxeter systems given in Section 3. 2
5.6. The next lemma records some consequences of 5.5 which will be used frequently below.
Lemma. Let r , s ∈ S with r = s. Suppose that g ∈ G with R r ,s ∩ N ( g ) = ∅. Let y ∈ W r ,s .
(a) N ( yg ) ∩ R r ,s = N ( y ).
(b) The expression yg is compatible.
(c) N (r ) ⊆ N ( yg ) iff N (r ) ⊆ N ( y ),
(d) N (r ) ∩ N ( yg ) = ∅ iff N (r ) ∩ N ( y ) = ∅.
(e) yg is admissible at r.
(f) If N ( yg ) ∩ R r ,s = ∅, then y = 1 W .
(g) If ( N (r ) ∪ N (s)) ⊆ N ( yg ), then m := mr ,s is finite and y = (rsr · · ·)m = (srs · · ·)m .
Proof. We have N ( yg ) = N ( y ) + y · N ( g ). Now N ( y ) ⊆ R r ,s by 5.5(b) and

y · N ( g ) ∩ R r ,s = y · N ( g ) ∩ y −1 · R r ,s = y · N ( g ) ∩ R r ,s = y · ∅ = ∅.
proving both (a) and (b). Then (c)–(d) follow from (a) since N (r ) ⊆ R r ,s . By 5.5(e), y is admissible at r,
so (e) follows from (a) and the definition of admissibility at r. In the situation of (f), we have N ( y ) = ∅.
Since w → N ( w ) is a non-degenerate root cocycle of ( W r ,s , S r ,s ), we have {t ∈ T r ,s | lr ,s (t y ) <
lr ,s ( y )} = ∅ by 3.1(ii)(3) and so y = 1 W . Similarly, in (g) we have S r ,s ⊆ {t ∈ T r ,s | lr ,s (t y ) < lr ,s ( y )}
and the conclusion of (g) follows. 2
5.7. The following is the main technical Lemma for the proof of Theorem 5.3. The terminology we
have introduced makes the statement and proof the same, mutatis mutandis, as that of [17, Propo-
sition 3.5], with steps here justified by the assumptions 5.1(i)–(iii) and their consequences 5.5–5.6
instead of by the Lemmas on Coxeter groups in [17]. Nonetheless, we shall supply details for com-
pleteness.
Lemma. Let w ∈ W .
(a) An expression w = r1 · · · rn with r i ∈ S is compatible iff it is reduced.

(b) Let r , s ∈ S be distinct. Then there exists v ∈ W such that w v −1 ∈ W r ,s , N ( v ) ∩ R r ,s = ∅ and l( w ) =
l( w v −1 ) + l( v ).
(c) w is admissible at S.
(d) For s ∈ S, l(sw ) = l( w ) − 1 iff N (s) ⊆ N ( w ).
(e) For s ∈ S, l(sw ) = l( w ) + 1 iff N (s) ∩ N ( w ) = ∅.
(f) If r , s ∈ S with r = s and N (r )∪ N (s) ⊆ N ( w ), then m = mr ,s is finite and there exists a reduced expression
of w beginning with (rsr · · ·)m .
Proof. The proof is by induction on l( w ). If l( w ) = 0, then w = 1 W and (a)–(f) hold.

Now suppose n = l( w ) > 0 and that (a)–(f) hold for elements of length at most n − 1. Choose a
reduced expression w = r1 · · · rn for w. Let r = r1 and w = r2 · · · rn . Then l( w ) = n − 1. Hence l(r w ) =
l( w ) = 1 + l( w ) so (by (e) for w ), N (r ) ∩ N ( w ) = ∅. Hence w = r w is a compatible expression
for w. By (a) for w , the reduced expression w = r2 · · · rn for w is compatible. Hence w = r1 · · · rn is
compatible. This proves:
(I) Every reduced expression for w is compatible. Further, if s ∈ S and l(sw ) = l( w ) − 1, then
N ( s ) ⊆ N ( w ).
Now fix s ∈ S with s = r. By (b) for w , there exists v ∈ W such that w v −1 ∈ W r ,s ,
N ( v ) ∩ R r ,s = ∅ and l( w ) = l( w v −1 ) + l( v ). Let x := w v −1 ∈ W r ,s . Then w = xv where x :=
rx ∈ W r ,s . Since N ( v ) ∩ R r ,s = ∅, it follows that w is admissible at r and at s, by Lemma 5.6(e).
We have l( w v −1 ) = l(x) = l(rx ) l(x ) + 1 so

l( w ) l w v −1 + l( v ) = l(x) + l( v ) l(x ) + 1 + l( v ) = l( w ) + 1 = l( w )
which implies l( w ) = l( w v −1 ) + l( v ). This shows:

(II) w is admissible at S (i.e. (c) holds). Moreover, (b) holds for pairs (r , s) such that there is a
reduced expression of w beginning with r or s.
Next, fix t ∈ S with N (t ) ⊆ N ( w ) and t = r. By (II), there exists z ∈ W such that w z−1 ∈ W r ,t ,
N ( z) ∩ R r ,t = ∅ and l( w ) = l( w z−1 ) + l( z). Let x = w z−1 ∈ W r ,t . Since N (r ) ⊆ N ( w ) by (I), while
N (r ) ∩ N ( z) = ∅, we have z = w so x = 1 W , l(x) > 0 and l( z) < l( w ). Choose a reduced expression z =
s1 · · · sk for z. By (c) for z, z = s1 · · · sk is a compatible expression.
Since N (r ) ∪ N (t ) ⊆ N ( w ), Lemma 5.6(g) gives that m = mr ,s is finite and that x = (rtr · · ·)m . By
5.1(ii), (trt · · ·)m−1 is a compatible expression for rx. By 5.6(b), (trt · · ·)m−1 s1 · · · sk is a compatible
expression for r w = w , of length m − 1 + k. By (a) for w , l( w ) = m − 1 + k and so l( w ) = 1 + l( w ) =
m + k. Hence w = (rtr · · ·)m s1 · · · sk is a reduced expression for w. Also, t w = (rtr · · ·)m−1 s1 · · · sk so
l(t w ) m − 1 + k. But m + k = l( w ) l(t w ) + 1 m + k so l(t w ) = m − 1 + k = l( w ) − 1. Since also
N (r ) ⊆ N ( w ) and l(r w ) = l( w ) − 1, this establishes:
(III) For t ∈ S, if N (t ) ⊆ N ( w ) then l(t w ) = l( w ) − 1. Parts (d) and (f) hold for w.
Now let w = t 1 · · · tk be a compatible expression for w with all t i ∈ S. Then N (t 1 ) ⊆ N ( w ), so
l(t 1 w ) = l( w ) − 1 by (III). Since t 2 · · · tk is a compatible representation of t 1 w, (a) for t 1 w implies that
t 2 · · · tk is reduced and therefore l(t 1 w ) = k − 1, l( w ) = k and w = t 1 · · · tk is a reduced expression.
Recalling (I),
(IV) Part (a) holds for w.
Now let r , s ∈ S be arbitrary with r = s. If ( N (r ) ∪ N (s)) ∩ N ( w ) = ∅, we can take v = w to sat-
isfy (b), by the assumption 5.1(iii) and Lemma 5.5(b). If ( N (r ) ∪ N (s)) ∩ N ( w ) = ∅, then by (c) for w
(see (II)) and symmetry, we may assume without loss of generality that N (r ) ⊆ N ( w ). Then by (IV),
l(r w ) = l( w ) − 1 and by (II)
(V) Part (b) holds for w.
Now let s ∈ S be arbitrary with l(sw ) = l( w ) + 1. From (d) for w (see (III)), N (s) ⊆ N ( w ). By (c)
for W (see (II)) we have N (s) ∩ N ( w ) = ∅. Hence:
(VI) For s ∈ S, if l(sw ) = l( w ) + 1 then N (s) ∩ N ( w ) = ∅.
Now let t ∈ S with N (t ) ∩ N ( w ) = ∅. Assume l(t w ) = l( w ) + 1. By (d) for w (see (III)), l(t w ) =
l( w ) − 1 (or else N (t ) ⊆ N ( w ), a contradiction). So l(t w ) = l( w ). From (I)–(VI) above, parts (a)–(d) hold
both for w and w := t w. Note that l(t w ) = l( w ) = l( w ) − 1, so by (d) applied to w , N (t ) ⊆ N ( w ).
By (c) for w , N (t ) ∩ N ( w ) = ∅. Since N (t ) ∩ N ( w ) = ∅ and N (t ) ∩ N ( w ) = ∅, both expressions
w = t w and w = t w are compatible. But this implies that w = tt w is a compatible expression
for w, which is impossible since tt is obviously not a compatible expression. Hence (recalling (VI)).
(VII) For t ∈ S, if N (t ) ∩ N ( w ) = ∅ then l(t w ) = l( w ) + 1. Part (e) holds for w.
This completes the proof of the Lemma. 2
5.8. Proof of Theorem 5.3. Observe that ( W , S ) is a pair consisting of a group G and set S of involu-
tory generators of G, satisfying the following conditions:
(i) for w ∈ W and s ∈ S, either l(sw ) = l( w ) + 1 or l(sw ) = l( w ) − 1;

(ii) if w ∈ W and r , s ∈ S are such that r = s, l(sw ) < l( w ) and l(r w ) < l( w ), then m = mr ,s is finite
and w has a reduced expression of the form w = (rsr · · ·)m .
This implies that ( W , S ) is a Coxeter system (see [17, §2] for a sketch of a proof). This gives
5.3(a). For 5.3(b), let x, y ∈ W with l(xy ) = l(x) + l( y ). Then xy has a reduced expression xy = r1 · · · rn
such that x = r1 · · · r i for some i. Since r1 · · · rn is reduced, it is compatible by Lemma 5.7(a), and this
implies that xy is compatible. This completes the proof of 5.3(b) and hence the proof of 5.3.
5.9. Theorem 5.3 has the following immediate Corollary, which may be regarded as a simultaneous
abstract characterization up to isomorphism of Coxeter systems with their reflection cocycles. In the
formulation, we use an obvious notion of isomorphism of pairs ( R , N ) consisting of a G-set R and a
cocycle N : G → P ( R ).
Corollary. Let N : G → P ( R ) be a cocycle and S ⊆ G. Then the following conditions are equivalent.
(i) S is a set of Coxeter generators of G and the pair ( R , N ) is isomorphic to the pair ( T , N ) where N : G →
P ( T ) is the reflection cocycle of (G , S ).
(ii) S is the set of all elements s of G with | N (s)| = 1, R = s∈ S w · N (s) and S is an admissible subset of G
w ∈G
which generates G.
(iii) | N (s)| = 1 for all s ∈ S, R = s∈ S w · N (s), and S is an admissible subset of G which generates G.
w ∈G
6. Embeddings of permutation root systems
We shall work in 6.1–8.4 with a fixed “ambient” Coxeter system ( W , S ). The following notation
concerning ( W , S ) will be standard. Denote the set of reflections of ( W , S ) as T and the standard
length function as l . Set Ψ = T × {±1}, the standard based permutation root system of ( W , S ).
Define the group G = W , the action of G × {±1} on Ψ , the G-action on T = Ψ/{±1}, and the cocycle
N : G → P ( T ) as in Section 2. We denote the G-action on T = Ψ/{±1} (by conjugation) now as
and its subgroup W as groups
( g , t ) → gt g −1 instead of ( g , t ) → g · t as earlier. We usually regard W
of permutations of Ψ . For α = (t , ) ∈ Ψ , let sα := t ∈ T ⊆ W denote the corresponding reflection
in W (as a permutation of Ψ ).
6.1. The following Theorem describes a class of Coxeter subgroups of W for which the corresponding
standard based permutation root system Φ is “abstractly embedded” in Ψ , in the sense that positive
(resp., negative) roots of Φ are realized as subsets of Ψ+ (resp., −Ψ+ ) with W -action as the restriction
of the natural W -action on P (Ψ ) induced by the W -action on Ψ .
Theorem. Let S be a subset of the set of involutions of W . Let W := S and let l be the length func-
tion of ( W , S ). Fix a W –stable subset R of T (e.g. R = T ). Then the map N : W → P ( R ) defined by
w → N ( w ) ∩ R is a cocycle of W . We assume that S is an admissible subset of W with respect to N.
(a) ( W , S ) is a Coxeter system. Denote its set of reflections as T .

(b) N is a non-degenerate root cocycle i.e. there is a family { R t }t ∈ T of pairwise disjoint, non-empty subsets
of R such that R t = w N (r ) w −1 whenever t = wr w −1 with w ∈ W , r ∈ S, and for any w ∈ W , N ( w ) =

{t ∈ T |l(t w )<l( w )} R t (disjoint union).
(c) Define Φ = { R t × { } | t ∈ T , ∈ {±1}} ⊆ P (Ψ ). The W -action on P (Ψ ) induces by restriction a W -
action on P (Ψ ) under which Φ is W -stable. The W -action on Φ satisfies w ( R t , ) = ( R wt w −1 , )
where = 1 if l( wt ) > l( w ) and = −1 if l( wt ) < l( w ), for any w ∈ W , t ∈ T and ∈ {±1}; in par-
ticular, Φ is isomorphic to the standard based permutation root system of ( W , S ). For α = R t × { } ∈ Φ ,
we denote the corresponding reflection in W as r α := t.
Proof. For x, y ∈ W , we have

N (xy ) = N (xy ) ∩ R = N (x) + xN ( y )x−1 ∩ R

= N (x) ∩ R ) + xN ( y )x−1 ∩ R = N (x) + xN ( y )x−1 ∩ xRx−1

= N (x) + x N ( y ) ∩ R x−1 = N (x) + xN ( y )x−1 .
This shows that N is a cocycle, and then the rest of the theorem follows from Theorem 5.3 and
Proposition 3.1. 2
6.2. Much of the rest of this paper will be concerned with explaining the connection of Theorem 6.1
with known results, and describing special cases where the assumption that S is admissible can be
simplified or in which stronger conclusions can be drawn.
The following Corollary describes a special case of 6.1 in which the reflection rα ∈ W in a root
α ∈ Φ is a (suitably ordered) product of the reflections sβ ∈ W for β ∈ α . Other conditions under
which a similar conclusion holds are given in 6.3, 6.5 and 7.4.
Corollary. Suppose in Theorem 6.1 that S ⊆ W (so W ⊆ W ) and R = T . Then for each α ∈ Φ , there
is some total order α of α ⊆ Ψ such that rα = β∈(α ,α ) sβ := sβ1 · · · sβn where α = {β1 , . . . , βn } with
β1 <α · · · <α βn .
Proof. If the Corollary is true for α ∈ Φ , it is true for w (α ) for any w ∈ W , using 2.5(a). Hence we
may assume that α = R s × {1} = N (s) × {1} where s ∈ S. Choose a reduced expression s = sn · · · s1 for s
as an element of W . Then N (s) = {t 1 , . . . , tn }, | N (s)| = n, where t i = sn · · · s2 s1 s2 · · · sn and s = t 1 · · · tn .
Then α = {β1 , . . . , βn } where βi = (t i , 1) and the ordering β1 <α · · · <α βn of α has the required
property. 2
6.3. The following Corollary shows how reflection subgroups arise in the setting of Theorem 6.1.
Corollary. Let S be a subset of the set T of reflections of W . Let W := S , R = T := W ∩ T and let l be

the length function of ( W , S ). Define the cocycle N : W → P ( R ) by w → N ( w ) ∩ R. Then the following
conditions are equivalent:
(i) S is admissible for W with respect to N.

(ii) The condition 5.1(ii) holds.
(iii) χ ( r , s) = {r , s} for any distinct r , s ∈ S.
(iv) S = χ ( W ).
(v) Let W ⊇ W be the setwise stabilizer of T in W . Then S is admissible for W with respect to the cocycle

N : w → N ( w ) ∩ R of W .
If these conditions hold, then R t = {t } for all t ∈ T .

Proof. Let t be an element of S. Since t ∈ T , we have t ∈ N (t ). Hence t ∈ N (t ) ∩ W = N (t ). Since

t 2 = 1, we see that 5.1(i) holds without additional assumptions.
By the definitions, (i) implies (ii). Assume that (ii) holds. Let r , s ∈ S with r = s. Set S = χ ( W ),
so that N is the reflection cocycle of ( W , S ). Let l be the length function of ( W , S ). Also, set U :=
χ ( r , s) and let l be the length function of the Coxeter system ( r , s, U ). For any i ∈ N with i mr ,s ,
we have by compatibility of (rsr · · ·)i with respect to N that

0 = l(1) < l(r ) < l(rs) < · · · < l (rsr · · ·)i .
By 2.3(c) applied to ( W , S ) in ( W , S ),

0 = l (1) < l (r ) < l (rs) < · · · < l (rsr · · ·)i .
By 2.3(c) applied to ( r , s, U ) in ( W , S ), it follows in turn that

0 = l (1) < l (r ) < l (rs) < · · · < l (rsr · · ·)i
and similarly with r and s interchanged, by symmetry. Since the elements of U are in r , s and
have l -length 1, and since |U | 2 we must have U = {r , s}. This shows that (ii) implies (iii). The
equivalence of (iii) and (iv) follows from [9, Proposition 3.5]. We next prove that (iv) implies (v).
Assume (iv) holds. We use frequently below that W is a subgroup of W and N extends N from
W to W . This implies first, from above, that 5.1(i) holds for N and S, and second, that if r, s
are distinct elements of S, then (rsr · · ·)i is compatible in W with respect to N for i ∈ N with
i mr ,s , since (rsr · · ·)i is a reduced expression in ( W , S ) and N is the reflection cocycle of ( W , S ).
To prove 5.1(iii), let w ∈ W with N ( w ) ∩ N (x) = ∅ where x ∈ W J := J with J ⊆ S. It suffices
to show that (if | J | = 2) N ( w ) ∩ N (r ) = ∅ for some r ∈ J . But we have N (x) = N (x) ⊆ W J , so
N ( w ) ∩ W J = N ( w ) ∩ W J = ∅. This implies by Lemma 2.5(iii) that there is some r ∈ N ( w ) ∩ J . Then
r ∈ N ( w ) ∩ N (r ) as required to complete the proof that (iv) implies (v). Finally, that (v) implies (i)
is trivial. 2
6.4. As well as inclusions of reflection subgroups, the embeddings 6.1 include also the Coxeter group
embeddings considered by Mühlherr in [17]. One of the special features of Mühlherr’s embeddings,
and their generalizations studied in Section 7, is conveniently described in terms of the notion of a
reflection order of a Coxeter system, which we recall below.
Let ( W , S ) be a Coxeter system with standard real root system Φ (e.g. as in [13, Chapter 5]).
A total order of the reflections T of ( W , S ) is called a reflection order if for any α , β, γ ∈ Φ+ with
γ ∈ R0 α + R0 β and sα sβ , one has sα sγ sβ . It is trivial that if the irreducible component
of W are ( W i , S i ) for i ∈ I , a total order of T is a reflection order iff its restriction to the reflections
of ( W i , S i ) is a reflection order of ( W i , S i ) for each i ∈ I . By [5, 2.13], if W is finite, a total order
of the reflections T of ( W , S ) is a reflection order iff there is a reduced expression w S = r1 · · · rn
i of the longest element w S of ( W , S ) such that is of the form t 1 t 2 · · · tn where t i =
r1 · · · r i −1 r i r i −1 · · · r1 (note that for any such reduced expression, T = {t 1 , t 2 , . . . , tn } with | T | = n, and
that w S = w − 1
S = t 1 · · · tn ). In this way, one obtains an explicit description of the reflection orders of
an arbitrary Coxeter system all of the irreducible components of which are finite Coxeter systems.
6.5. Let P be a set of non-empty, pairwise disjoint subsets of S such that for each J ∈ P , W J := J
is a finite (standard parabolic) subgroup of W with longest element w J . Set S := { w J } J ∈ P .
Theorem. The following conditions (i)–(ii) are equivalent:
(i) S is an admissible subset of W with respect to N ;

(ii) for any distinct r , s ∈ S, the expression (rsrs · · ·)i is compatible with respect to N for all i ∈ N with i
mr ,s , where mr ,s is the order of rs.
Assume henceforward that (i)–(ii) hold and define l, T , { R t }t ∈ T and Φ etc as in Theorem 6.1. Let α ∈ Φ , say
α = R t × { } with t ∈ T , and let y ∈ W . Set W α := sβ | β ∈ α , a reflection subgroup of W
(a) W y (α ) = yW α y −1 .
(b) R t is the set of all refections {sβ | β ∈ α } of W α .
(c) Let be a reflection order of the reflections R t of ( W α , χ ( W α ))and let α denote the corresponding
total order of α from the bijection β → sβ : α → R t . Then rα = β∈(α ,α ) sβ := sβ1 · · · sβn where α =
{β1 , . . . , βn } with β1 <α · · · <α βn and rα := t is the reflection of W corresponding to α .
(d) If α ∈ Φ+ and l( ysα ) > l( y ), then N ( y −1 ) ∩ W α = ∅ i.e. as an element of W , y is the (unique) distin-
guished coset representative in yW α (see 2.3(d)).
(e) The Chevalley–Bruhat order on ( W , S ) is the restriction of the Chevalley–Bruhat order on ( W , S ).
Proof. Clearly (i) implies (ii). We show (ii) implies (i). Assume that (ii) holds. Then 5.1(ii) holds,
obviously. Clearly, 5.1(i) holds since each w J is of order 2 and N ( w J ) = W J ∩ T = ∅ since w J ∈ P and
J = ∅ by assumption. To show 5.1(iii) holds, suppose that w J , w K ∈ S are distinct and that g ∈ W
with N ( g ) ∩ N (x) = ∅ for some x ∈ w J , w K . By Lemma 2.5(f), this implies that N ( g ) ∩ ( J ∪ K ) = ∅.
Since J ⊆ N ( w J ) and K ⊆ N ( w K ), we have N ( g ) ∩ ( N ( w J ) ∪ N ( w K )) = ∅ as required.
Now assume that (i)–(ii) hold. Part (a) follows immediately from Lemma 2.5(a). We now prove (b).
If α is simple for Φ+ , then α = R s × {1} for some s ∈ S, say s = w J with J ∈ P . Then {sβ | β ∈ α } =
R s = N (s) = N ( w J ) = W J ∩ T as is well known. Clearly, W α = W J , so (b) holds for simple α . For
arbitrary α ∈ Φ , write α = w (γ ) where γ is simple. Write γ = R s × {1}, where s ∈ S, so R s is the set
of all reflections of W γ from above. Note
{sβ | β ∈ α } = {s w (β) | β ∈ γ } = w {sβ | β ∈ γ } w −1 = w R s w −1 .
This is clearly the set of all reflections of the reflection subgroup W α = w W γ w −1 it generates, prov-
ing (b) in general.
Next, we prove (d). We have y (α ) ∈ Φ+ i.e for β ∈ α , y (β) ∈ Ψ+ . But from (b), α is the set of all
positive roots in Ψ of W α , so N ( y −1 ) ∩ W α = ∅ as required.
It is clear that (c) holds for simple α , from (b) and the proof of Corollary 6.2. In general, write
α = w (γ ) where γ is simple and w ∈ W . Using (a), (d) and 2.3(d), χ ( W α ) = w χ ( W γ ) w −1 so
∼
=
z → w zw −1 induces an isomorphism of Coxeter systems ( W γ , χ ( W γ )) −→ ( W α , χ ( W α )). The set
of reflections W γ is R s , where s = w −1 t w ∈ S. The total order of R s defined by t 1 t 2 iff
wt 1 w −1 wt 2 w −1 is therefore a reflection order of ( W γ , χ ( W γ )). Let γ denote the total order
of γ induced by the bijection β → sβ : γ → R s . We then have γ = { w −1 (β1 ), . . . , w −1 (βn )} where
w −1 (β1 ) <γ · · · <γ w −1 (βn ). Since γ ∈ Π , we have from above that rγ = s w −1 (β1 ) · · · s w −1 (βn ) since γ
is simple and it follows that rα = wrγ w −1 = sβ1 · · · sβn as required to complete the proof of (d).
Finally, we prove (e) (cf. [12]). Recall [1, Chapter 2, Exercise 14] that is the unique partial order
on W such that the following Z -property (lifting property) holds: If x, y ∈ W with l (sx) > l (x) and
l (sy ) < l( y ), then x y iff x sy iff sx y. It will suffice to show that the restriction of to W
has the analogous properties with respect to the length function l of ( W , S ).
So let x, y ∈ W and r ∈ S with l(rx) > l(x) and l(r y ) < l( y ). Write r as the longest element of a
finite parabolic subgroup, say r = w J where J ∈ P , and choose a reduced expression r = s1 · · · sm for r
in ( W , S ). Then l (si x) > l (x) and l (si y ) < l ( y ) for all i, since x and y are compatible at S. It follows
that l (x) < l (sn x) < · · · < l (s1 · · · sn x) and l ( y ) > l (sn y ) > · · · > l (s1 · · · sn y ). Hence x y if x sn y
iff x sn−1 sn y iff . . . iff x s1 · · · sn y = r y. Similarly, x y iff sn x y iff . . . iff rx = s1 · · · sn x y. This
completes the proof of (e). 2
7. Root system embeddings using Coxeter group completions
Retain the standard notation concerning ( W , S ) from Section 6. This section describes an exten-
sion (Theorem 7.4) of Theorem 6.5 to the situation in which S is a suitable set of “longest elements”
corresponding to certain infinite parabolic subgroups W of W , where all irreducible com-
w J in W J
ponents of each W J are finite. The only essential difference is that one has to interpret the product
in Theorem 6.5(c) as a “convergent” infinite product of permutations. The special case considered in
Section 8 arises naturally in relation to embeddings of root systems over commutative rings as in [6].

7.1. We shall introduce a notation ∈W for “convergent infinite products” of certain special
l∈( L ,) al
families of possibly non-commuting permutations al ∈ W , the order of terms in the product being
specified by a total order of L.
More generally, let ( L , ) be some totally ordered index set and {al }l∈ L be a family of permutations
of a set P . We assume that for any α ∈ P , there is a finite subset L α of L and some βα , βα ∈ P such
that (al1 · · · aln )(α ) = βα and (al−1 · · · al−1 )(α ) = βα for any finite subset {li 1 < · · · < lin } ⊇ L α of L. Then
n 1
we define a permutation g = i ∈( L ,) ai of Ψ by g (α ) = βα for all α ∈ P (we have g −1 (α ) = βα ).

Note g −1 = l∈( L ,op ) bl where bl := al−1 and op is the opposite total order to on L. For any

permutation h of P , we have h( i ∈( L ,) al )h−1 = i ∈( L ,) (hal h−1 ).
If the permutations al commute pairwise, then their product is independent of and will be
denoted simply l∈ L a L . Slightly more generally, if { L k }k∈ K is a partition of L such that al and al
commute whenever l ∈ L k and l ∈ L k with k = k , then

al = al
l∈( L ,) k∈ K l∈( L k ,k )
where k is the restriction of to a total order on L k .

, then
from Lemma 2.5(d) that, in case P = Ψ , if all al are in W
It follows immediately

i ∈( L ,) ai ∈ W .
7.2. Consider a reflection subgroup W of ( W , S ), with (possibly infinitely many)irreducible com-

ponents ( W i , S i ) for i ∈ I . So S i = χ ( W i ), ( W i , S i ) is irreducible, and χ ( W ) = i ∈ I S i where the
S i are pairwise disjoint and all elements of S i commute with all elements of S j if i = j. We shall say
that W is sparse (in W ) if it satisfies the following condition: for any r ∈ S , there are only finitely
many i ∈ I such that there is some element of W i (or equivalently, some element of S i ) which does
not fix (r , 1) ∈ Ψ .
From Lemma 2.5(d), we see that W is sparse iff for every finitely-generated reflection subgroup
(resp., finitely-generated parabolic subgroup, resp., finitely-generated standard parabolic subgroup)
W of W , there are only finitely many i ∈ I such that W i (or equivalently, S i ) does not fix
, then g W g −1
( W ∩ T ) × {±1} elementwise. This implies further that if W is sparse and g ∈ W
is sparse.
Suppose W is sparse. Let ( L , ) be a totally ordered index set and {al }l∈ L be a family of elements
of i ∈ I W i such that for each i ∈ I , there are only finitely many l ∈ L such that al is a non-identity
element of W i . Given α ∈ Ψ , let L α denote the (finite) set of all l ∈ L such that al is a non-identity
element of W i for one of the (finitely many) i ∈ I such that some element of W
i does not fix α . It
is clear that the L α for α ∈ Ψ satisfy the condition on L α in 7.1 and hence that l∈( L ,) al is defined.
Lemma. Let W be a sparse reflection subgroup of ( W , S ) with irreducible components W i for i ∈ I . Let
w i ∈ W i for each i ∈ I .

(a) w := i ∈ I w i is defined and in W . Set L := N ( w ) ∩ W .

(b) Set L i = N ( w i ) ∩ W i . Then L = i ∈ I L i (disjoint union).
(c) For each i, choose a reduced expression w i = r i ,1 · · · r i ,ni of w i in W i (with respect to χ ( W i )).
Then L i = {t ∈ T ∩ W | l (t w i ) < l ( w i )} = {t i ,1 , . . . , t i ,ni } where t i , j = r i ,1 · · · r i , j −1 r i , j r i , j −1 · · · r i ,1
(and | L i | = ni ).
(d) Choose
any total order of L with the following property: for each i, t i ,1 ≺ · · · ≺ t i ,ni . Then i ∈ I w i =
t ∈( L ,op ) t.

(e)
The set of all elements i ∈ I w i with w i ∈ W i forms a subgroup of W isomorphic to the product group
W under the map w →
( w ) .
i∈ I i i∈ I i i i ∈ I
Remarks. The orders of L constructed in (d) are the restrictions to L of the reflection orders
of W ∩ T (with respect to Coxeter generators χ ( W )) such that r ≺ s for all r ∈ L, s ∈ ( W ∩ T ) \ L. In
particular, if each Wi is finite and wi is its longest element, then
W ∩ T = N ( w ) ∩ W and the orders on W ∩ T as in (d) are precisely the reflection orders
of W ∩ T .

Proof. The remarks preceding the statement of the Lemma show that w := i w i is defined and
(since L i ⊆ N ( w i ) is finite) that t ∈( L ,op ) t is defined for any total order of L. By 7.1, both products

in W . By the definition of w, for any α ∈ Ψ , there is some finite I α ⊆ I such that w (α ) =
are
( i ∈ I w i )(α ) for all finite I with I ⊇ I ⊇ I α . But for any finite subset I of I ,

N wi ∩ W = N ( w i ) ∩ W = N ( w i ) ∩ W i
i∈ I i∈ I i∈ I
(disjoint union), using that the W i are the irreducible components of W , and (b) follows. Part (c) is
well-known using that x → N (x) ∩ W gives the reflection cocycle of ( W , χ ( W ).
Let be a total order as in (d) and i denote its restriction to L i . We have t ∈( L , ) t =
op
i i
t i ,ni · · · t i ,1 = w i . By 7.1, t ∈( L ,)op t= i∈ I ( t ∈( L i ,i ) t )
op = i∈ I w i = w as required for (d). The proof
of (e) is omitted. 2
7.3. A subset J of S will be said to be good if W J is a sparse reflection subgroup, the irreducible
components of which are finite Coxeter systems ( W J , J i ) for I in some index set. The longest element
i
of ( W J , J i ) will be denoted as w J i for i ∈ I , We define w J =
i ∈ I w J i ∈ W . We have w J = 1 W
2
and
i
N(w J) = W
J ∩ T = i∈ I (W Ji ∩ T ). Hence w J ∈
W iff I is finite iff J is finite; in that case, w J is
the longest element of W J , so the notation is consistent with standard notation.
7.4. We now state our generalization of Theorem 6.5. Let P be a set of non-empty, pairwise disjoint
good subsets of S , and set S := { w J } J ∈ P (where the elements w J and the notion of good subsets
of S are defined in 7.3).
Theorem. With S as above, the equivalence of the conditions 6.5(i)–(ii) holds. If these equivalent conditions
hold, the statements (a)–(c) and the part of (d) before “i.e.” of Theorem 6.5 hold with the product in 6.5(c)
defined as in 7.1.
, 6.5(e) does not make sense in this

Remarks. Since Chevalley–Bruhat order is not defined on W
generality.
Proof. The proof is the same, mutatis mutandis, as that of the corresponding parts of Theorem 6.5,
and its elements w J , and the convergent infinite
using the previously established facts about W
products of 7.1. We omit further details. 2
8. Embeddings from strong admissible maps
We maintain until 8.4 the general notation concerning a Coxeter system ( W , S ) from 6.1.
8.1. If ( W , S ) is of finite rank, an element c = s1 · · · sn where S = {s1 , . . . , sn }, | S | = n is called

a standard Coxeter element of ( W , S ). Conjugates of standard Coxeter elements are called Coxeter
elements of ( W , S ).
If W is finite and irreducible, all Coxeter elements are conjugate in W and so have the same finite
order, which is called the Coxeter number of c. If W is infinite and irreducible, but of finite rank,
any Coxeter element c has infinite order (see [11]), and we define the Coxeter number of ( W , S ) to
be ∞. Finally, if ( W , S ) is irreducible of infinite rank, then Coxeter elements of W are not defined
but we still define the Coxeter number of ( W , S ) to be ∞. It follows that any infinite irreducible
Coxeter system has Coxeter number ∞. We define the Coxeter number of an arbitrary Coxeter system
to be the least common multiple of the Coxeter numbers of its irreducible components.
Henceforward in this paper, by “Coxeter element” we shall always mean “standard Coxeter ele-
ment.”
Lemma.
(a) Let ( W , S ) be an irreducible Coxeter system of finite rank and (possibly infinite) Coxeter number h. As-
sume that there is a p artition S = S 1 ∪ S 2 such that each (non-empty) S i consists of pairwise commuting

reflections. Define the involutions c i := i ∈ S s. Then c := c 1 c 2 is a Coxeter element (so has order h) and
i
the expression (c 1 c 2 c 1 · · ·)k is compatible with respect to N if k h.
(b) Let c be a Coxeter element of an infinite, irreducible, finite rank Coxeter system ( W , S ) of Coxeter number
h. Then l (cn ) = nl (c ) for all n ∈ N. In particular, c has infinite order
Proof. For finite W , (a) is a well-known result (see [19]). A proof of (a) for infinite W associated to
a symmetrizable generalized Cartan matrix can be found in [10, Corollary 9.6]; the same proof works
mutatis mutandis for arbitrary Coxeter groups using their standard reflection representation. In fact,
[10] itself involves an extension of Steinberg’s arguments for the finite case and a uniform proof of
(a) for both finite and infinite W can be given along similar lines.
We indicate another proof of (a), via (b). Note in general that c 1 c 2 is compatible and c 1 c 2 is a
Coxeter element of W . Hence (a) (for infinite W ) follows directly from (b). A proof of (b) for general
Coxeter systems is given in [18], extending ideas introduced in a proof of (b) for Coxeter systems
attached to symmetric generalized Cartan matrices in [15]. 2
Remarks. At the end of this section, we also show how (a)–(b) in general may be deduced from the
corresponding facts in the special case ( W , S ) is simply laced (i.e. all entries of its Coxeter matrix
are in {1, 2, 3}).
8.2. The following technical fact extends (and will be deduced from) Lemma 8.1(a).
Lemma. Let J be a subset of S such that ( W J , J ) is irreducible, of Coxeter number h. Suppose that J = J 1 ∪ J 2
is a partition of J such that each J i is a (non-empty) set of pairwise commuting elements of J and W J is
i
a sparse reflection subgroup of ( W , S ). Let c i = w J i ∈ W for i = 1, 2. Then c := c 1 c 2 has order h and

(c 1 c 2 c 1 · · ·)n is a compatible expression with respect to N for any n ∈ N with n h.
Proof. Let c := c 1 c 2 . Since N (c i ) = N ( w J i ) = W J i ∩ T from 7.3, we have N (c 1 ) ∩ N (c 2 ) = ∅ and hence

the expression c = c 1 c 2 is compatible.
If J is finite, the Lemma follows from Lemma 8.1. The case in which J is infinite will be treated by
reduction to the case of finite J . We need only show that the expression ck = (ccc · · ·)k is compatible
for any k ∈ N. Fix a total order of J such that a < b for all a ∈ J 1 and b ∈ J 2 . For any L ⊆ J , let
L denote the restriction of to a total order of L. Clearly, c = s∈( J ,) s. Note that the product

s is convergent in
s∈( L , L ) W for any L ⊆ J , to c L := w J 1 ∩ L w J 2 ∩ L .
For K ⊆ S, define K ∗ ⊆ S as the set of all elements of S which appear in some sequence
(sn , sn−1 , . . . , s1 , s0 ) with s0 ∈ K , si ∈ J for i = 1, . . . , n, si joined to si −1 in the Coxeter graph
of ( W , S ) for i = 1, . . . , n, and
sn < sn−1 < · · · < s1 . Then K ⊆ K ∗ ⊆ K ∪ J . By the sparseness of W J i
for i = 1, 2, it follows that K ∗ = s∈ K {s}∗ is finite for each finite K ⊆ J . It is easy to see that for s ∈ S,
c L (s, ) = c s∗ (s, ) ∈ ( W s ∗ ∩ T ) × {±1}
for any L ⊆ J with L ⊇ s∗ .

The definitions imply that N (ck ) ⊆ W J ∩ T for all k ∈ N. Hence it is sufficient to show that for any
k ∈ N and any finite L ⊆ J ,

N (c ) ∩ W L ⊆ N c 2 ∩ W L ⊆ · · · ⊆ N ck ∩ W L . (8.2.1)
Recursively define a sequence L i of finite subsets of J , for i ∈ N, as follows. First, let L 0 = L. If L i

is defined, let L i +1 := ( L i )∗ ⊇ L i . From above and Lemma 2.5(d), we see that if α ∈ ( W L ∩ T ) ×
i
{±1}, then and c (α ) = c K (α ) for any finite K ⊆ J with K ⊇ L i +1 , and c (α ) ∈ ( W L i+1 ∩ T ) × {±1}.
Using irreducibility of ( W J , J ) and the fact that J is infinite, choose a finite subset K of J such that
K ⊇ L k , | K | 2k and ( W K , K ) is irreducible. Note that the Coxeter number of any irreducible finitely-
generated Coxeter system exceeds its rank, so W K has Coxeter number h K > 2k. Applying the results
already proved to (the finite set) K instead of J , using | K | 2k, we deduce that the expression
ckK = (c K c K c K · · ·)k is compatible, and hence N (c K ) ⊆ N (c 2K ) ⊆ · · · ⊆ N (ckK ). But since K ⊇ L k , it
follows that c iK (α ) = c i (α ) for any α ∈ ( W L ∩ T ) × {±1} and i = 0, . . . , k. Hence by definition of the
reflection cocycle N , N (c i ) ∩ W L = N (c iK ) ∩ W L . Thus (8.2.1) above holds, as required to complete
the proof. 2
8.3. The following Corollary leads to simple necessary and sufficient conditions for admissibility of a
set S := { w J } J ∈ P where P is a set of non-empty, pairwise disjoint good (e.g. finite) subsets of S , each
consisting of pairwise commuting elements of S (see Theorem 7.4 and Theorem 6.5).
Corollary. Suppose that J , K are pairwise disjoint, non-empty good subsets of S , each consisting of pairwise
commuting elements of S . Then
(a) The order of w J w K is equal to the Coxeter number h of ( W J ∪ K , J ∪ K ).

(b) The expressions ( w J w K w J · · ·)n and ( w K w J w K · · ·)n are compatible for all n ∈ N with n h iff each
irreducible component of ( W J ∪ K , J ∪ K ) has Coxeter number h.
Proof. Let ( W L , L i ) for i ∈ I be the irreducible components of ( W J ∪ K , J ∪ K ). Then it is easy to

i
see that w J = i ∈ I w J ∩ L i , w K = i ∈ I w K ∩ L i and w J w K = i ∈ I w J ∩ L i w K ∩ L i where the elements
w J ∩ L i w K ∩ L i for i ∈ I commute pairwise. Further, ( w J w K )n = i ∈ I ( w J ∩ L i w K ∩ L i )n from which one
readily sees that the order of w J w K is the least common multiple of the orders of the elements
w J ∩ L i w K ∩ L i for i ∈ I . Since Lemma 8.2 implies that the order of w J ∩ L i w K ∩ L i is the Coxeter number
h i of ( W L , L i ), this proves (a). Similarly, for n ∈ N,
i

N ( w J w K w J · · ·)n = N ( w J ∩ L i w K ∩ L i w J ∩ L i · · ·)n
i∈ I
(disjoint union) so ( w J w k w J · · ·)n is compatible iff each ( w J ∩ L i w K ∩ L i w J ∩ L i · · ·)n is compatible. By 8.2

again, this holds iff n h i . So all ( w J w K w J · · ·)n and ( w K w J w K · · ·)n with n ∈ N and n h are
compatible iff each n ∈ N with n h satisfies n h i for all i ∈ I . Since each h i h, this holds iff
h i = h for all i ∈ I . 2
8.4. We say that a Coxeter graph (or the corresponding Coxeter system) is locally finite if each of
its vertices is joined to only finitely many other vertices. A Coxeter system is locally finite if each of
its irreducible components is locally finite. Hence any Coxeter system with finite Coxeter number is
locally finite.
The following gives very simple sufficient conditions for admissibility of certain subsets of W of
the type considered in Theorems 6.5 and 7.4, and determines the Coxeter matrix of the corresponding
Coxeter system ( W , S ) as in those theorems.
Theorem. Let P be a partition of S into non-empty subsets, each consisting of pairwise commuting simple
reflections. Then the following conditions are equivalent:
(i) Each J ∈ P is good and S := { w J } J ∈ P is an admissible subset of W with respect to N .

(ii) For any two distinct J , K ∈ P , the Coxeter system ( W J ∪ K , J ∪ K ) is locally finite and all of its irreducible
components have the same Coxeter number h J , K .
If these conditions hold, the order in

W of w J w K is h J , K for any distinct J , K ∈ P .
Proof. By 8.3, it is enough to prove that (i)–(ii) are equivalent. Suppose first that (i) holds. Let h J , K
be the order of w J w K in G. By definition of admissibility and 8.3, all irreducible components of
( W J ∪ K , J ∪ K ) have Coxeter number h J , K . We also have to show also that any s ∈ J ∪ K is joined in
the Coxeter graph of ( W J ∪ K , J ∪ K ) to only finitely many vertices r ∈ J ∪ K . By symmetry between
J and K , it suffices to show that s ∈ J is joined to only finitely many r ∈ K , which follows from the
definitions since K is a good subset consisting of commuting simple reflections. This shows that (i)
implies (ii).
Conversely, assume that (ii) holds. Let J ∈ P . Obviously, the irreducible components of W J are
finite Coxeter systems since J consists of pairwise commuting involutions. To show W J is sparse, it
suffices to show that if s ∈ S , then s is joined (in the Coxeter graph of ( W , S )) to at most finitely
many vertices of J . For this, let K ∈ P with s ∈ K . If K = J , then s is joined to no vertex of J since
J consists of pairwise commuting reflections. If K = J , then since ( W K ∪ J , K ∪ J ) has locally finite
Coxeter graph, s is joined to only finitely many vertices of K ∪ J . This shows that J is good, so
wJ ∈W is defined. Now 8.3 and 7.4 imply that S is compatible. 2
8.5. Let Z be a subring of R. We introduced in [6, 3.7] a notion of a Z fusion root basis da-
tum B = ( R , R + , M , M ∨ , ?, ?, Π, Π ∨ , , ι) with associated Coxeter system ( W , S ). Let R = R J be
the subring of R defined as in subsection 3.6 of [6], and consider a root basis datum B =
( R , R + , M , M ∨ , ?, ? Δ, Δ∨ , ι ) related to B and R as in (5.1) and (4.2) of [6] (so the root system
Φ ⊆ M of B is naturally a subset of the root system Ψ ⊆ M of B ). The associated Coxeter systems
of B and B are denoted as ( W , S ) and ( W , S ), respectively. The group G (with W and W as sub-
groups) associated to B in (4.3) of [6] is immediately seen to identify with the group W of this
paper, and the cocycle N : G → P ( T ) of (4.5) of [6], where T is the set of reflections of ( W , S ),
identifies with N : W → P ( T ) of this paper.
The following is an immediate consequence of Theorem 8.4 and the relation between the Coxeter
graphs of ( W , S ) and ( W , S ) described in [6, 5.3].
with respect to N .
Corollary. The set S is an admissible subset of W
Remarks. In case Z = Z, the “cocycle compatibility condition” proved as [6, 4.5(g)] follows also from
3.1(ii)(3); we observe that the arguments of this paper imply validity of this cocycle compatibility
condition more generally for arbitrary subrings Z of R.
8.6. We next reformulate 8.3–8.4 using a variant of the notion of admissible map as defined in [16].
Let ( W , S ) be a Coxeter system with Coxeter matrix (mr ,s )r ,s∈ S , and let ( W , S ) be any Coxeter
system. A strong admissible map from ( W , S ) to ( W , S ) is defined to be a function f : S → S such
that the following conditions hold:
(i) for each s ∈ S, f −1 (s) is a set of commuting simple reflections of S .

(ii) for distinct r , s ∈ S, each connected component of f −1 ({r , s}) is a Coxeter system with Coxeter
number mr ,s .
(iii) For each r ∈ S and s ∈ S, r is joined in the Coxeter graph of ( W , S ) to only finitely many
elements of f −1 (s).
Remarks. As in the proof of 8.4, one sees that if f is surjective, the condition (iii) may be replaced
by the condition that for all r , s ∈ S with mr ,s = ∞, ( W J , J ) is locally finite where J = f −1 ({r , s}).
We define “admissible maps” exactly as strong admissible maps but with “mr ,s ” in (ii) replaced by
“dividing mr ,s .” This is the same as the definition of “admissible map” in Lusztig [16] except that
there the condition (iii) is also omitted; however, (iii) is required for the facts about admissible maps
in loc cit to apply to Coxeter systems which are not locally finite.
8.7. The proofs of the following simple properties of admissible maps are left to the reader.
Lemma.
(a) Any admissible map (and, a fortiori, any strongly admissible map) from ( W , S ) to ( W , S ) induces a
morphism between the unlabeled graphs underlying the corresponding Coxeter graphs.
(b) Let ( W i , S i ) for i ∈ I (resp., ( W j , S j ) for j ∈ J ) denote the irreducible components of a Coxeter system
( W , S ) (resp. ( W , S )). Then a function f : S → S is a strongly admissible map ( W , S ) → ( W , S ) iff
there is a map σ : I → J with the following property: for each i ∈ I , f ( S i ) = S σ (i ) and the restriction of f
to a map S i → S σ (i ) is strongly admissible from ( W i , S i ) to ( W σ (i ) , S σ (i ) ).
(c) If f : S → S is a (strongly) admissible map from ( W , S ) to ( W , S ), and L ⊆ S, then the restriction of f
to a map L := f −1 ( L ) → L is a (strongly) admissible map from ( W L , L ) to ( W L , L ).
8.8. Now we describe the relationship between the Coxeter systems associated to the domain and
codomain of an admissible map.
Proposition. Let f : S → S be a strongly admissible map as above.
(a) The sets f −1 (s) for

s ∈ S are pairwise disjoint good sets of pairwise commuting simple reflections of S .

(b) The elements s := r ∈ f −1 (s) r for s ∈ S are well-defined elements of W and S := {s | s ∈ S , f −1 (s) =
.
∅} is an admissible subset of W
(c) Let J := f ( S ) and K := S \ J . Then ( W , S ) ∼

= ( W J , J ) × ( W K , K ) as Coxeter systems i.e. J (resp., K ) is
the union of the sets of simple reflections of certain irreducible components of ( W , S ).
(d) Define the subgroup W := S of W . Then ( W , S ) is a Coxeter system.
(e) The map s → s for s ∈ S extends to an epimorphism θ : W → W with kernel W K . Hence θ induces
an isomorphism ( W J , J ) ∼= ( W , S ) of Coxeter systems. In particular, if f is surjective, then θ is an
∼
=
isomorphism ( W , S ) −→ ( W , S ).
Proof. This follows directly from Corollary 8.3, Theorem 7.4 and Lemma 8.7(a). 2
8.9. Before proceeding, we note the following.
Lemma. Let f : S → S be a surjective strongly admissible map from ( W , S ) to ( W , S ) with S finite.
(a) The corresponding group embedding W → W maps Coxeter elements of ( W , S ) to Coxeter elements
of ( W , S ).
(b) If ( W , S ) is irreducible, then ( W , S ) is irreducible, and ( W , S ) and ( W , S ) have the same Coxeter
number.
(c) If ( W , S ) is irreducible, all irreducible components of ( W , S ) have the same Coxeter number as ( W , S ).
(d) The Coxeter systems form a category with morphisms given by strongly admissible maps, and composition
of morphisms given by composition of underlying functions.
Proof. Let
c = s1 · · · sn be a Coxeter element of ( W , S ). Its image in ( W , S ) is
−1 (s )
−
r ∈ f (s1 )
1 r r ∈ f −1 (s2 ) r · · · r ∈ f −1 (sn ) r which is a Coxeter element of ( W , S ) since the sets f i
form a partition of S . This proves (a). Then (b) follows from (a) and Lemma 8.7(a), and (c) follows
from (b) and 8.7(b). The proof of (d) uses (c) and Lemma 8.7; we leave details to the reader. 2
8.10. For any non-trivial Coxeter system ( W , S ), there are many admissible maps S → S, with re-
spect to suitably chosen ( W , S ). The Proposition and Remark below, which are elaborations of [17,
Theorem 5.1], single out particularly useful or natural classes of such admissible maps, with ( W , S )
satisfying stringent additional conditions, including being simply laced. Many variants could be given
(see the following remark).
Proposition. Let ( W , S ) be a Coxeter system and kr ,s ∈ N2 for distinct r , s ∈ S be integers such that kr ,s =
mr ,s − 1 if mr ,s is finite and kr ,s 2 if mr ,s is infinite. Let κ be a non-zero cardinal number such that κ is
divisible by kr ,s for all r = s in S. Then there is a simply laced Coxeter system ( W , S ) and a surjective strongly
admissible map f : S → S such that
(i) | f −1 (s)| = κ for all s ∈ S.

(ii) If r , s ∈ S are distinct and J := f −1 ({r , s}) ⊆ S , then all irreducible components of ( W J , J ) are Coxeter
systems of type A kr ,s (resp., Ã 2kr ,s −1 ) if mr ,s is finite (resp., infinite).
If from Proposition 8.8 maps W inside W and is length

κ is finite, the corresponding embedding W → W
κ -tupling.
Proof. Choose a set S and a surjective map f : S → S satisfying (i). Using the fact that the Coxeter
graphs of type A n and Ã 2n−1 are bipartite graphs, and the divisibility assumptions about κ , it is
easy to see that one may construct a Coxeter graph on vertex set S such that, for the corresponding
Coxeter system ( W , S ), f is admissible and (ii) holds. 2
Remarks. If ( W , S ) has only finitely many distinct labels mr ,s in its Coxeter graph, one may clearly
choose the kr ,s and κ so that κ is finite. As a variant in general, one could take κ to be a cardinal
such that κ is infinite if some mr ,s is infinite, and otherwise κ is divisible by mr ,s − 1 for all r = s with
mr ,s finite, and require instead of (ii) that, for all distinct r , s ∈ S and J := f −1 ({r , s}), all irreducible
components of ( W J , J ) are of type A mr ,s −1 (see [14] or [8] for the definition of A ∞ ). The universal
symmetric based root system of ( W , S ) over commutative rings from [6] gives rise to a surjective
strongly admissible map of this type, canonically associated to ( W , S ). However, it has the technical
disadvantage compared to the embeddings in the Corollary, that even if S is finite, one may have to
take κ infinite (if some mr ,s = ∞).
8.11. Proposition 8.10 and Lemma 8.9 have the following Corollary.
Corollary. Let ( W , S ) be a Coxeter system with only finitely many distinct entries in its Coxeter matrix.
(a) There exists a simply laced Coxeter system ( W , S ) and a positive integer κ such that there is a surjective
admissible map f : S → S from ( W , S ) to ( W , S ) with | f −1 (s)| = κ for all s ∈ S. The group embedding
ι : W → W induced by the admissible map f is length κ -tupling.
(b) If S is finite, then so is S and ι maps Coxeter elements of ( W , S ) to Coxeter elements of ( W , S ); moreover,
if in addition ( W , S ) is irreducible, then the Coxeter number of every irreducible component of ( W , S ) is
equal to the Coxeter number of ( W , S ).
8.12. For simply laced ( W , S ), a proof of the various parts of the Lemma 8.1 can be found in [10,
15,19]; in general, proofs can be given from [19] and [18]. To finish this section, we indicate as an
application of the preceding results another proof of Lemma 8.1(a)–(b), by reduction to the simply
laced case.
One first observes that the proofs of Proposition 8.10 and Corollary 8.11 in this Section only require
the special case of 8.1(a) in which ( W , S ) is simply laced (in fact, of type A n for n 1 or Ã n
for n 3).
We now prove 8.1(b) as follows. Let ( W , S ) be an irreducible infinite finitely generated Coxeter
system, and ( W , S ) be a Coxeter system given as in 8.9, so there is an admissible map f : S → S.
We regard the embedding W → W given by this map as an inclusion, so l ( w ) = κ l( w ) for any

w ∈ W . Let c be a Coxeter element of ( W , S ). As an element of W , c = d1 · · · dn where d1 , . . . , dn
are the Coxeter elements of the irreducible components of ( W , S ), which all have the same Coxeter
number h = ∞ as ( W , S ) by 8.10. By the simply laced case of 8.1(b), one has l (dm i
) = ml (di ) for all
i = 1, . . . , n and all m ∈ N. Hence
m
l cm = l cm m m
1 · · · cn = l c 1 + · · · + l cn = m l (c 1 ) + · · · l (cn ) = ml (c )
and l(cm ) = l (cm )/κ = ml (c )/κ = ml(c ) as required. Similarly, 8.1(a) is reduced to its simply laced
case.
We remark that, at the expense of only slightly longer arguments, the reductions of 8.1(a)–(b)
to the corresponding results for simply laced Coxeter systems can be accomplished using the (not-
necessary length κ -tupling) embeddings of [17, Theorem 5.1], and that also [18, Theorem 2], which is
stronger than 8.1(b), can be similarly reduced to its special case for simply laced Coxeter systems.
References
[1] Anders Björner, Francesco Brenti, Combinatorics of Coxeter Groups, Grad. Texts in Math., vol. 231, Springer, New York,
2005.
[2] Anders Björner, Michelle L. Wachs, Generalized quotients in Coxeter groups, Trans. Amer. Math. Soc. 308 (1) (1988) 1–37.
[3] N. Bourbaki, Éléments de mathématique. Fasc. XXXIV. Groupes et algèbres de Lie. Chapitre IV: Groupes de Coxeter et sys-
tèmes de Tits. Chapitre V: Groupes engendrés par des réflexions. Chapitre VI: systèmes de racines, Actualités Scientifiques
et Industrielles, vol. 1337, Hermann, Paris, 1968.
[4] Brigitte Brink, Robert B. Howlett, A finiteness property and an automatic structure for Coxeter groups, Math. Ann. 296 (1)
(1993) 179–190.
[5] M.J. Dyer, Hecke algebras and shellings of Bruhat intervals, Compos. Math. 89 (1) (1993) 91–115.
[6] M.J. Dyer, Embeddings of root systems I: Root systems over commutative rings, J. Algebra (2009), doi: 10.1016/j.jalgebra.
2008.10.008.
[7] M.J. Dyer, On parabolic closures in Coxeter groups, preprint, 2007.
[8] M.J. Dyer, On rigidity of abstract root systems of Coxeter groups, preprint, 2007.
[9] Matthew Dyer, Reflection subgroups of Coxeter systems, J. Algebra 135 (1) (1990) 57–73.
[10] S. Fomin, A. Zelevinsky, Cluster algebras IV, Compos. Math. 143 (2007) 112–164.
[11] Robert B. Howlett, Coxeter groups and M-matrices, Bull. London Math. Soc. 14 (2) (1982) 137–141.
[12] Axel Hultman, Fixed points of involutive automorphisms of the Bruhat order, Adv. Math. 195 (1) (2005) 283–296.
[13] James E. Humphreys, Reflection Groups and Coxeter Groups, Cambridge Stud. Adv. Math., vol. 29, Cambridge Univ. Press,
Cambridge, 1990.
[14] Victor G. Kac, Infinite-Dimensional Lie Algebras, Cambridge Univ. Press, Cambridge, 1990.
[15] M. Kleiner, A. Pelley, Admissible sequences, preprojective modules and reduced words in the Weyl group of a quiver,
preprint, arXiv: Math.RT/0607001.
[16] George Lusztig, Some examples of square integrable representations of semisimple p-adic groups, Trans. Amer. Math.
Soc. 277 (2) (1983) 623–653.
[17] B. Mühlherr, Coxeter groups in Coxeter groups, in: Finite Geometry and Combinatorics, Deinze, 1992, in: London Math. Soc.
Lecture Note Ser., vol. 191, Cambridge Univ. Press, Cambridge, 1993, pp. 277–287.
[18] David E. Speyer, Powers of Coxeter elements in infinite groups are reduced, preprint, 2007, arXiv: 0710.3188v1.
[19] Robert Steinberg, Finite reflection groups, Trans. Amer. Math. Soc. 91 (1959) 493–504.
Journal of Algebra
Secant varieties to osculating varieties of Veronese

embeddings of Pn
A. Bernardi a , M.V. Catalisano b , A. Gimigliano c,∗ , M. Idà a
a
Dip. di Matematica, Univ. di Bologna, Italy
b
DIPTEM, Univ. di Genova, Italy
c
Dip. di Matematica and C.I.R.A.M., Univ. di Bologna, Italy
Article history: A well-known theorem by Alexander–Hirschowitz states that all

Received 24 July 2008 the higher secant varieties of V n,d (the d-uple embedding of Pn )
Communicated by Luchezar L. Avramov have the expected dimension, with few known exceptions. We
study here the same problem for T n,d , the tangential variety to
Keywords:
Algebraic geometry
V n,d , and prove a conjecture, which is the analogous of Alexander–
Secant varieties Hirschowitz theorem, for n 9. Moreover, we prove that it holds
Tensor decomposition for any n, d if it holds for d = 3. Then we generalize to the
case of O k,n,d , the k-osculating variety to V n,d , proving, for n = 2,
a conjecture that relates the defectivity of σs ( O k,n,d ) to the Hilbert
function of certain sets of fat points in Pn .
0. Introduction
A well-known theorem by Alexander and Hirschowitz (see [AH1]) states:
Theorem 0.1 (Alexander–Hirschowitz). Let X be a generic collection of s 2-fat points in Pnκ . If ( I X )d ⊂

κ [x0 , . . . , xn ] is thevector space of forms of degree d which are singular at the points of X , then dim( I X )d =
n+d
min{(n + 1)d, n }, as expected, unless:
– d = 2, 2 s n;
– n = 2, d = 4, s = 5;
– n = 3, d = 4, s = 9;
*
E-mail addresses: abernardi@dm.unibo.it (A. Bernardi), catalisano@diptem.unige.it (M.V. Catalisano), gimiglia@dm.unibo.lt
(A. Gimigliano), ida@dm.unibo.it (M. Idà).
doi:10.1016/j.jalgebra.2008.10.020
A. Bernardi et al. / Journal of Algebra 321 (2009) 982–1004 983
– n = 4, d = 3, s = 7;
– n = 4, d = 4, s = 14.
Notice that with “m-fat point at P ∈ Pn ” we mean the scheme defined by the ideal I m
P ⊂
κ [x0 , . . . , xn ].
An equivalent reformulation of the theorem is in the language of higher secant varieties; let
n+d
V n,d ⊂ P N , with N = n − 1, be the d-ple (Veronese) embedding of Pn , and let σs ( V n,d ) be its
(s − 1)th higher secant variety, that is, the closure of the union of the Ps−1 ’s which are s-secant to
V n,d . Then Theorem 0.1 is equivalent to:
Theorem 0.2. All the higher secant varieties σs ( V n,d ) have the expected dimension min{s(n + 1) − 1,
n+d
n
− 1}, unless s, n, d are as in the exceptions of Theorem 0.1.
An application of the theorem is in terms of the Waring problem for forms (or of the decomposi-
tion of a supersymmetric tensor), in fact Theorem 0.1 gives that the general form of degree d in n + 1

1 n+d
variables can be written as the sum of n+ 1 d
dth powers of linear forms, with the same list of
exceptions (e.g. see [Ge] or [IK]).
In [CGG1] a similar problem has been studied, namely whether the dimension of σs ( T n,d ) is the
expected one or not, where T n,d is the tangential variety of the Veronese variety V n,d . This too trans-
lates into a problem of representation of forms: the generic form parameterized by σs ( T n,d ) is a form
F of degree d which can be written as F = L d1−1 M 1 + · · · + L ds −1 M s , where the L i , M i ’s are linear forms.
The following conjecture was stated in [CGG1]:
n+d
Conjecture 1. The secant variety σs ( T n,d ) has the expected dimension, min{2sn + s − 1, n
− 1}, unless:
(i) d = 2, 2 2s < n;
(ii) d = 3, s = n = 2, 3, 4.
In the same paper the conjecture was proved for d = 2 (any s, n) and for s 5 (any d, n), while
in [B] it is proved for n = 2, 3 (any s, d).
In [CGG1] (via inverse systems) it is shown that σs ( T n,d ) is defective if and only if a certain 0-
dimensional scheme Y ⊂ Pn does not impose independent conditions to forms of degree d in R :=
κ [x0 , . . . , xn ]. The scheme Y = Z 1 ∪ · · · ∪ Z s is supported at s generic points P 1 , . . . , P s ∈ Pn , and at
each of them the scheme Z i lies between the 2-fat point and the 3-fat point on P i (we will call Z i a
(2, 3, n)-scheme, for details see Section 1 below).
Hence Conjecture 1 can be reformulated in term of ( I Y )d having the expected dimension, with the
same exceptions, in analogy with the statement of Theorem 0.1.
Theorem 0.1 has been proved thanks to the Horace differential lemma ([AH2], Proposition 9.1; see
also here Proposition 1.5) and an induction procedure which has a delicate beginning step for d = 3;
different proofs for this case are in [Ch1,Ch2] and in the more recent [BO], where an excellent history
of the question can be found.
Also the proof of Conjecture 1 presents the case of d = 3 as a crucial one; the first main result in
this paper (Corollary 2.5) is to prove that if Conjecture 1 holds for d = 3, then it holds also for d 4
(and any n, s). The procedure we use is based on Horace differential lemma too.
We also prove Conjecture 1 for all n 9, since with that hypothesis we can check the case d = 3
by making use of COCOA (see Corollary 2.4).
A more general problem can be considered (see also [BCGI]): let O k,n,d be the k-osculating variety
to V n,d ⊂ P N , and study its (s − 1)th higher secant variety σs ( O k,n,d ). Again, we are interested in the
problem of determining all s for which σs ( O k,n,d ) is defective, i.e. for which its dimension is strictly
less than its expected dimension (for precise definitions and setting of the problem, see Section 1 of
the present paper and in particular Question Q (k, n, d)).
Also in this general case we found in [BCGI] (via inverse systems) that σs ( O k,n,d ) is defective
if and only if a certain 0-dimensional scheme Y ⊂ Pn does not impose independent conditions to
984 A. Bernardi et al. / Journal of Algebra 321 (2009) 982–1004
forms of degree d in R = κ [x0 , . . . , xn ]. The scheme Y = Z 1 ∪ · · · ∪ Z s is supported at s generic points

P 1 , . . . , P s ∈ Pn , and at each of them the ideal of the scheme Z i is such that I kP+2 ⊂ I Z i ⊂ I kP+1 (for
i i
details see Lemma 1.2 below).
The following (quite immediate) lemma ([BCGI], 3.1) describes what can be deduced about the
postulation of the scheme Y from information on fat points:
Lemma 0.3. Let P 1 , . . . , P s be generic points in Pn , and set X := (k + 1) P 1 ∪ · · · ∪ (k + 1) P s , T := (k + 2) P 1 ∪

· · · ∪ (k + 2) P s . Now let Z i be a 0-dimensional scheme supported at P i , (k + 1) P i ⊂ Z i ⊂ (k + 2) P i , and set
Y := Z 1 ∪ · · · ∪ Z s . Then, Y is regular in degree d if h1 (I T (d)) = 0 or if h0 (I X (d)) = 0.
Moreover, Y is not regular in degree d if
d+n
(i) h1 (I X (d)) > max{0, deg(Y ) − n
},
or if
d+n
(ii) h0 (I T (d)) > max{0, n
− deg(Y )}.
All cases studied in [BCGI] lead us to state the following:
Conjecture 2a. The secant variety σs ( O k,n,d ) is defective if and only if Y is as in case (i) or (ii) of the lemma
above.
The conjecture amounts to saying that I Y does not have the expected Hilbert function in degree d
only when “forced” by the Hilbert function of one of the fat point schemes X , T .
Notice that (i), respectively (ii), obviously implies that X , respectively T , is defective. Hence, if
Conjecture 2a holds and Y is defective in degree d, then either T or X are defective in degree d too,
and the defectivity of Y is either given by the defectivity of X or forced by the high defectivity of T .
Thus if the conjecture holds, we have another occurrence of the “ubiquity” of fat points: the prob-
lem of σs ( O k,n,d ) having the right dimension reduces to a problem of computing the Hilbert function
in degree d of two schemes of s generic fat points in Pn , all of them having multiplicity k + 1, respec-
tively k + 2.
In [BC] and [BF] the conjecture is proved in P2 for s 9.
Notice that the Conjecture 2a implies the following one, more geometric, which relates the defec-
tivity of σs ( O k,n,d ) to the dimensions of the kth and the (k + 1)th osculating space at a generic point
of the (s − 1)th higher secant variety of the Veronese variety σs ( V n,d ):
Conjecture 2b. If the secant variety σs ( O k,n,d ) is defective then at a generic point P ∈ σs ( V n,d ), either the kth
k+n
osculating space O k,σs ( V n,d ), P does not have dimension min{s − 1, d+n − 1}, or the (k + 1)th osculating
k+n+1 n
d+n n
space O k+1,σs ( V n,d ), P does not have dimension min{s n
− 1, n − 1}.
The implication follows from the fact that (see [BBCF]) for P ∈ P 1 , . . . , P s :
O k,σs ( V n,d ), P = O k, V n,d , P 1 , O k, V n,d , P 2 , . . . , O k, V n,d , P s .
The other main result in this paper is Theorem 3.5, which proves Conjecture 2a for n = 2.
1. Preliminaries and notations
In this paper we will always work over a field κ such that κ = κ and char κ = 0.
1.1. Notations
(i) If P ∈ Pn is a point and I P is the ideal of P in Pn , we denote by m P the fat point of multiplicity
m supported at P , i.e. the scheme defined by the ideal I m P.
(ii) Let X ⊆ P N be a closed irreducible projective variety; the (s − 1)th higher secant variety of X is
the closure of the union of all linear spaces spanned by s points of X , and it will be denoted by σs ( X ).
(iii) Let X ⊂ P N be a variety, and let P ∈ X be a smooth point; we define the kth osculating space to
X at P as the linear space generated by (k + 1) P ∩ X (i.e. by the kth infinitesimal neighbourhood of P
in X ) and we denote it by O k, X , P ; hence O 0, X , P = { P }, and O 1, X , P = T X , P , the projectivised tangent
space to X at P .
Let U ⊂ X be the dense set of the smooth points where O k, X , P has maximal dimension. The kth
osculating variety to X is defined as:

O k, X = O k, X , P .
P ∈U
(iv) We denote by V n,d the d-uple Veronese embedding of Pn , i.e. the image of the map defined
n+d
by the linear system of all forms of degree d on Pn : νd : Pn → P N , where N = n − 1.
(v) We denote the kth osculating variety to the Veronese variety by O k,n,d := O k, V n,d . When k = 1,
the osculating variety is called tangential variety and it is denoted by T n,d .
Hence, the (s − 1)th higher secant variety of the kth osculating variety to the Veronese variety V n,d
will be denoted by σs ( O k,n,d ).
Since the case d k is trivial, and the description for k = 1 given in [CGG1], together with [BCGI,
Proposition 4.4] describe the case d = k + 1 completely, from now on we make the general assumption,
which will be implicit in the rest of the paper, that d k + 2.
It is easy to see ([BCGI], 2.3) that the dimension of O k,n,d is always the expected one, that is,
k+n
dim O k,n,d = min{ N , n + n
− 1}. The expected dimension for σs ( O k,n,d ) is:

k+n
expdim σs ( O k,n,d ) = min N , s(n + − 1) + s − 1
n
(there are ∞s(dimO k,n,d ) choices of s points on O k,n,d , plus ∞s−1 choices of a point on the Ps−1
spanned by the s points; when this number is too big, we expect that σs ( O k,n,d ) = P N ).
When dim σs ( O k,n,d ) < expdim σs ( O k,n,d ), the osculating variety is said to be defective.
In [BCGI], taking into account that the cases with n = 1 can be easily described, while if n 2 and
d = k one has dim σs ( O k,n,d ) = N, we raised the following question:
Question Q (k, n, d). For all k, n, d such that d k + 1, n 2, describe all s for which σs ( O k,n,d ) is
defective, i.e.

k+n d+n k+n
dim σs ( O k,n,d ) < min N , s(n + − 1) + s − 1 = min − 1, s + sn − 1 .
n n n
We were able to answer the question for s, n, d, k in several ranges, thanks to the following lemma
(see [BCGI], 2.11 and results of Section 2):
Lemma 1.2. For any k, n, d ∈ N such that n 2, d k + 1, there exists a 0-dimensional subscheme Z =
Z (k, n) ∈ Pn depending only from k and n and not from d, such that:
(a) Z is supported on a point P , and one has:

k+n
(k + 1) P ⊂ Z (k, n) ⊂ (k + 2) P , with l( Z ) = + n;
n
(b) denoting by Y = Y (k, n, s) the generic union in Pn of Z 1 , . . . , Z s where Z i ∼

= Z for i = 1, . . . , s, then

d+n
dim σs ( O k,n,d ) = expdim σs ( O k,n,d ) − h0 IY (d) + max 0, − l(Y ) .
n
In particular, σs ( O k,n,d ) is not defective if and only if Y is regular in degree d, i.e. h 0 (IY (d)) · h1 (IY (d)) =
0.
The homogeneous ideal of this 0-dimensional scheme Z is defined in [BCGI], 2.5 through inverse
systems, so we do not have an explicit geometric description of it in the general case. Anyway, for
k = 1 it is possible to describe it geometrically as follows (see [CGG1], Section 2):
Definition 1.3. Let P be a point in Pn , and L a line through P ; we say that a 0-dimensional scheme
X ⊂ Pn is a (2, 3, n)-scheme supported on P with direction L if I X = I 3P + I 2L . Hence, the length of a
(2, 3, n)-point is 2n + 1. The scheme Z (1, n) of Lemma 1.2 is a (2, 3, n)-scheme.
We say that a subscheme of Pn is a generic union of s (2, 3, n)-schemes if it is the union of
X 1 , . . . , X s where X i is a (2, 3, n)-scheme supported on P i with direction L i , with P 1 , . . . , P s generic
points and L 1 , . . . , L s generic lines through P 1 , . . . , P s .
We are going to use these schemes in Section 2, so we need to know more about them; but first
we recall the Horace differential lemma of [AH2], writing it in the context where we shall use it.
Definition 1.4. In the algebra of formal functions κ [[x, y ]], where x = (x1 , . . . , xn−1 ), a vertically graded
(with respect to y) ideal is an ideal of the form:

I = I 0 ⊕ I 1 y ⊕ · · · ⊕ I m−1 ym−1 ⊕ ym
where for i = 0, . . . , m − 1, I i ⊂ κ [[x]] is an ideal.

Let Q be a smooth n-dimensional integral scheme, let K be a smooth irreducible divisor on Q .
We say that Z ⊂ Q is a vertically graded subscheme of Q with base K and support z ∈ K , if Z is a
0-dimensional scheme with support at the point z such that there is a regular system of parameters
Q ,z ∼
(x, y ) at z such that y = 0 is a local equation for K and the ideal of Z in O = κ [[x, y ]] is vertically
graded.
Let Z ⊂ Q be a vertically graded subscheme with base K , and p 0 be a fixed integer; we denote
p p
by Res K ( Z ) ⊂ Q and Tr K ( Z ) ⊂ K the closed subschemes defined, respectively, by the ideals:
p +1 p p
IRes p ( Z ) := I Z + I Z : I K I K , ITr p ( Z ), K := I Z : I K ⊗ O K .
K K
p p
In Res K ( Z ) we take away from Z the ( p + 1)th “slice;” in Tr K ( Z ) we consider only the ( p + 1)th
“slice.” Notice that for p = 0 we get the usual trace and residual schemes: Tr K ( Z ) and Res K ( Z ).
Finally, let Z 1 , . . . , Z r ⊂ Q be vertically graded subschemes with base K and support zi , Z = Z 1 ∪
· · · ∪ Z r , and p = ( p 1 , . . . , p r ) ∈ Nr .
We set:
p p p p p p
Tr K ( Z ) := Tr K1 ( Z 1 ) ∪ · · · ∪ Tr Kr ( Z r ), Res K ( Z ) := Res K1 ( Z 1 ) ∪ · · · ∪ Res Kr ( Z r ).
Proposition 1.5 (Horace differential lemma). (See [AH2], Proposition 9.1.) Let H be a hyperplane in Pn and let
W ⊂ Pn be a 0-dimensional closed subscheme.
Let S 1 , . . . , S r , Z 1 , . . . , Z r be 0-dimensional irreducible subschemes of Pn such that S i ∼= Z i , i = 1, . . . , r,
Z i has support on H and is vertically graded with base H , and the supports of S = S 1 ∪ · · · ∪ S r and Z =
Z 1 ∪ · · · ∪ Z r are generic in their respective Hilbert schemes. Let p = ( p 1 , . . . , p r ) ∈ Nr . Assume:
(a) H 0 (ITr W ∪Tr H ( Z ), H (n))

p = 0, and
H
(b) H (IRes
0
W ∪Res H ( Z ) (n
p − 1)) = 0,
H
then

H 0 I W ∪ S (n) = 0.
Definition 1.6. A 2-jet is a 0-dimensional scheme J ⊂ Pn with support at a point P ∈ Pn and degree 2;
namely the ideal of J is of type: I 2P + I L , where L ⊂ Pn is a line containing P . We will say that
J 1 , . . . , J s are generic in Pn , if the points P 1 , . . . , P s are generic in Pn and L 1 , . . . , L s are generic lines
through P 1 , . . . , P s .
Remark 1.7. Let X ⊂ Pn be a (2, 3, n)-scheme supported at P with direction L and ( y 1 , . . . , yn ) be local
coordinates around P , such that L becomes the yn -axis; then, I X = ( y 1 yn2 , . . . , yn−1 yn2 , yn3 , y 21 , y 1 y 2 ,
. . . , yn2−1 ) ( yn appears only in the first n generators). Let H , respectively K , be a hyperplane through
L, respectively transversal to L; then, we can assume I H = ( yn−1 ), respectively I K = ( yn ). We now
p p
compute Res H ( X ) and Tr H ( X ). One has:
(a) Res H X = Res0H ( X ), I Res H ( X ) = ( I X : yn−1 ) = ( y 1 , . . . , yn−1 , yn2 ), hence Res H X is a 2-jet lying on L;
(b) Tr H ( X ) = Tr0H ( X ), I Tr H ( X ) = I X +( yn−1 ) = ( y 1 yn2 , . . . , yn−2 yn2 , yn3 , y 21 , y 1 y 2 , . . . , yn2−2 ), hence Tr H ( X )
is a (2, 3, n − 1)-scheme of H .
Hence the scheme X as a vertically graded scheme with base H has only two layers; in other
p
words, Tr H ( X ) is empty for p > 1, and Res1H ( X ) is a (2, 3, n − 1)-scheme of H , while Tr1H ( X ) is a 2-jet
lying on L.
p p
Now we want to compute Res K ( X ) and Tr K ( X ). Consider first:
(b) I Tr K ( X ) = I X + ( yn ) = ( yn , y 21 , y 1 y 2 , . . . , yn2−1 ), hence Tr H ( X ) is a 2-fat point of K ∼

= Pn−1 ;
(a) I Res K X = ( I X : yn ) = ( y 1 yn , . . . , yn−1 yn , yn2 , y 21 , y 1 y 2 , . . . , yn2−1 ), hence Res K X is a 2-fat point
of Pn .
So the scheme X , as a vertically graded scheme with base K , has only three layers; the 0-layer
is Tr K ( X ) = Tr0K ( X ), the 1-layer is the 0-layer of Res K X = Res0K ( X ), hence it is again a 2-fat point of
K∼ = Pn−1 , and the 2-layer is the 1-layer of Res K X , hence it is a point of Pn . In other words, Tr H ( X )
p
is empty for p > 2, Res K ( X ) is a 2-fat point of P , while Tr K ( X ) is a 2-fat point of K ; Res K ( X ) is a
1 n 1 2
2-fat point of K doubled in a direction transversal to K (i.e., I Res2 ( X ) = ( yn2 , y 21 , y 1 y 2 , . . . , yn2−1 )), while
K
Tr2K ( X ) is a point of Pn .
We will use in the sequel the fact that by adding s generic 2-jets to any 0-dimensional scheme
Z ⊂ Pn we impose a maximal number of independent conditions to forms in I Z (d), for all d. This is
probably classically known, but we write a proof here for lack of a reference:
Lemma 1.8. Let Z ⊆ Pn be a scheme, and let J ⊂ Pn be a generic 2-jet. Then:

h0 I Z ∪ J (d) = max h0 I Z (d) − 2, 0 .
Proof. Let P be the support of J ; then we know that h0 (I Z ∪ P (d)) = max{h0 (I Z (d)) − 1, 0}, so if
h0 (I Z (d)) 1 there is nothing to prove. Let h0 (I Z (d)) 2, then h0 (I Z ∪ P (d)) = h0 (I Z (d)) − 1 1.
Since J is generic, if h0 (I Z ∪ J (d)) = h0 (I Z ∪ P (d)), then every form of degree d containing Z ∪ P should
have double intersection with almost every line containing P , hence it should be singular at P . This
means that when we force a form in the linear system | H 0 (I Z (d))| to vanish at P , then we are
automatically imposing to the form to be singular at P , and this holds for P in a dense open set
of Pn , say U . If the form f is generic in | H 0 (I Z (d))|, its zero set V meets U in a non-empty subset
of V , so f is singular at whatever point P we choose in V ∩ U , and this means that the hypersurface
V is not reduced. Since the dimension of the linear system | H 0 (I Z (d))| is at least 2, this is impossible
by Bertini Theorem (e.g. see [J], Theorem 6.3). 2
Let Z ⊆ Pn be a zero-dimensional scheme; the following simple lemma gives a criterion for adding
to Z a scheme D which lies on a smooth hypersurface F ⊆ Pn and is made of s generic 2-jets on F ,
in such a way that D imposes independent conditions to forms of a given degree in the ideal of Z
(see Lemma 4 in [Ch1] and Lemma 1.9 in [CGG2] for the case of simple points on a hypersurface).
Lemma 1.9. Let Z ⊆ Pn be a zero dimensional scheme. Let F ⊆ Pn be a smooth hypersurface of degree d
and let Z = ResF Z . Let P 1 , . . . , P s be generic points on F , let L 1 , . . . , L s lines with P i ∈ L i , and such that
each line L i is generic in T P i (F ); let J i be the 2-jet with support at P i and contained in L i . We denote by
D s = J 1 ∪ · · · ∪ J s the union of these s 2-jets generic in F .
(i) If dim( I Z + D s−1 )t dim( I Z )t −d + 2, then dim( I Z + D s )t = dim( I Z )t − 2s.

(ii) If dim( I Z )t −d = 0 and dim( I Z )t 2s, then dim( I Z + D s )t = 0.
Proof. (i) By induction on s. If s = 1, by assumption dim( I Z )t dim( I Z )t −d + 2, hence in the exact

φ
sequence 0 → H 0 (I Z (t − d)) → H 0 (I Z (t − d)) → H 0 (I Z ∩F ,F (t )) → · · · the cokernel of the map φ
has dimension at least 2 and so ( I Z )t cuts on F a linear system (i.e. | H 0 (I Z ∩F ,F (t ))|) of (projective)
dimension 1. We have dim( I Z + P 1 )t = dim( I Z )t − 1, since otherwise each hypersurface in |( I Z )t |
would contain the generic point P 1 of F , that is, would contain F .
Assume dim( I Z + J 1 )t = dim( I Z + P 1 )t = dim( I Z )t − 1; this means that if we impose to S ∈ |( I Z )t | the
passage through P 1 automatically we impose to S to be tangent to L 1 at P 1 , and L 1 being generic in
T P 1 (F ), this means that each S passing through P 1 is tangent to F at P 1 . Let’s say that this holds
for P 1 in the open not empty subset U of F ; for S generic in |( I Z )t |, U = S ∩ F ∩ U is not empty,
hence the generic S is tangent to F at each P ∈ U . This means that |( I Z )t | cuts on F a linear system
of positive dimension whose generic element is generically non-reduced, and this is impossible, by
Bertini Theorem (e.g. see [J], Theorem 6.3).
Now let s > 1. Since dim( I Z + D s−2 )t dim( I Z + D s−1 )t > dim( I Z )t −d by assumption, and ResF ( Z +
D s−1 ) = Z , the case s = 1 gives dim( I Z + D s )t = dim( I Z + D s−1 )t − 2. So, by the induction hypothesis, we
get

dim( I Z + D s )t = dim( I Z )t − 2(s − 1) − 2 = dim( I Z )t − 2s.
(ii) Assume first dim( I Z )t 2; it is enough to prove dim( I Z + J 1 )t = 0 since then also
dim( I Z + D s )t = 0. If dim( I Z )t = 2 this follows by (i) and if dim( I Z )t = 0 this is trivial. If dim( I Z )t = 1,
then if dim( I Z + P 1 )t = 0 we are done. If dim( I Z + P 1 )t = 1, then by the genericity of P 1 we have that the
unique S in the system contains F , i.e. S = F ∪ G, but then Z ⊆ G, which contradicts dim( I Z )t −d = 0.
Otherwise, let dim( I Z )t = 2v + δ 3, δ = 0, 1. If δ = 0, then dim( I Z + D v −1 )t 2 = dim( I Z )t −d + 2,
and by (i) we get dim( I Z + D v )t = dim( I Z )t − 2v = 0, and, since s v, it follows that dim( I Z + D s )t = 0.
If δ = 1, then dim( I Z + D v −1 )t 3 dim( I Z )t −d + 2, and, by (i), dim( I Z + D v −1 )t = 3 and
dim( I Z + D v )t = dim( I Z )t − 2v = 1. Notice that the only element in ( I Z + D v )t cannot have F as a fixed
component, otherwise we would have dim( I Z )t −d = 1 and not = 0; hence dim( I Z + D v + P v +1 )t = 0 and
so, since 2s 2v + 1 and D v ∪ P v +1 ⊂ D s , dim( I D s )t = 0. 2
Now we give a lemma which will be of use in the proof of Theorem 2.2.
Lemma 1.10. Let R ⊆ Pn be a zero dimensional scheme contained in a (2, 3, n)-scheme with r = deg Y 2n;
assume moreover that, if r n + 1, then R is a flat limit of the union of a 2-fat point of Pn and of a scheme
(eventually empty) contained in a 2-fat point of a Pn−1 , and that, if r n, then R is contained in a 2-fat point
of a Pn−1 . Then, there exists a flat family for which R is a special fiber and the generic fiber is the generic union
in Pn of δ 2-fat points, h 2-jets and simple points, where r = (n + 1)δ + 2h + , 0 δ 1, 0 1, and
2h + n.
Proof. In the following we denote by 2t P a 2-fat point of a linear variety K ⊆ Pn , K ∼ = Pt . We first

notice that if A is a subscheme of 2n P with deg A = n then A is a scheme of type 2n−1 P . The proof is
by induction on n: if n = 2, the statement is trivial since the only scheme of degree 2 in P2 is a 2-jet,
i.e. a 21 P . Now assume the assertion true for n − 1, let A be a subscheme of 2n P with deg A = n and
let H be a hyperplane through the support of A. Since deg 2n P ∩ H = n, we have n − 1 deg A ∩ H n.
If deg A ∩ H = n then A = 2n−1 P and we are done. If deg A ∩ H = n − 1 then Res H A is a simple point,
and by induction A ∩ H = 2n−2 P . Hence there is a hyperplane K such that A ∩ H is a 2-fat point of
H ∩ K , and working for example in affine coordinates, it is easy to see that A is a 2-fat point of the
Pn−1 generated by H ∩ K and a normal direction to H .
In order to prove the lemma, it is enough to prove that the generic union in Pn of h 2-jets and
simple points, with 0 1 and 2h + n, specializes to any possible subscheme M of a scheme
of type 2n−1 P : in fact, if r n we are done, if r n + 1, the collision of a 2n P with M gives R.
By induction on n: if n = 2, the statement is trivial. Let us now consider the generic union of h
2-jets and simple points in Pn , with 0 1 and 2h + n. We have two cases.
Case 1. If 2h + n − 1, we specialize everything inside a hyperplane H where, by induction as-

sumption, this scheme specializes to any possible subscheme of a scheme of type 2n−2 P , i.e., to any
possible subscheme of degree n − 1 of a scheme of type 2n−1 P .
Case 2. If 2h + = n, we have to show that the generic union of h 2-jets and simple points special-
izes to a scheme 2n−1 P .
If n is odd, then h = n− 2
1
and = 1; by induction assumption, n− 2
1
2-jets specialize to a scheme
of type 2n−2 P , and the generic union of the last one with a simple point specializes to a scheme of
type 2n−1 P .
If n is even, then h = n2 and = 0; by induction assumption, n2 − 1 2-jets specialize to a scheme
of degree n − 2 contained in a scheme of type 2n−2 P , which is a 2n−3 P , so it is enough to prove that
the generic union of the last one with a 2-jet specializes to a scheme of type 2n−1 P .
In affine coordinates x1 , . . . , xn , let xn−2 = xn−1 = xn = 0 be the linear subspace containing 2n−3 P ,
so that I 2n−3 P = (x1 , . . . , xn−3 )2 ∩ (xn−2 , xn−1 , xn ), and let (x1 , . . . , xn−3 , xn−2 − a, xn2−1 , xn ) be the ideal
of a 2-jet moving along the xn−2 -axis; then it is immediate to see that the limit for a → 0 of
(x1 , . . . , xn−3 )2 ∩ (xn−2 , xn−1 , xn ) ∩ (x1 , . . . , xn−3 , xn−2 − a, xn2−1 , xn ) is (x1 , . . . , xn−1 )2 ∩ (xn ), which is
the ideal of a 2n−1 P . 2
2. On Conjecture 1
We want to study σs ( T n,d ), and we have seen that its dimension is given by the Hilbert function
of s generic (2, 3, n)-points in Pn .
Definition 2.0. For each n and d we define sn,d , rn,d ∈ N as the two positive integers such that

d+n
= (2n + 1)sn,d + rn,d , 0 rn,d < 2n + 1.
n
In the following we denote by X s,n ⊂ Pn the zero dimensional scheme union of s generic (2, 3, n)-
schemes A 1 , . . . , A s . We also denote by X sn,d the scheme X s,n , with s = sn,d . Hence X sn,d is the union
of the maximum number of generic (2, 3, n)-points that we expect to impose independent conditions
to forms od degree d. We will also use X sn,d +1 to indicate X s+1,n when s = sn,d .
With Y n,d ⊂ Pn we denote a scheme generic union of X sn,d and R n,d , where R n,d is a zero dimen-
sional scheme contained in a (2, 3, n)-point, with deg( R n,d ) = rn,d .
A 0-dimensional subscheme A of Pn is said to be “OPn (d)-numerically settled” if deg A =

h0 (OPn (d)); in this case, h0 (I A (d)) = 0 if and only if h1 (I A (d)) = 0. The scheme Y n,d is OPn (d)-
numerically settled for all n, d.
Remark 2.1. Let A be a 0-dimensional OPn (d)-numerically settled subscheme of Pn , and assume
h0 (I A (d)) = 0. Let B ⊆ A and C ⊇ A be 0-dimensional subschemes of Pn ; then, h0 (IC (d)) = 0, and
h1 (I B (d)) = 0, or equivalently, h0 (I B (d)) = deg A − deg B.
Hence if we prove h0 (IY n,d (d)) = 0 then we know that h1 (IY n,d (d)) = 0, and

h0 I X s,n (d) = 0 for all s > sn,d ,

h1 I X s,n (d) = 0 for all s sn,d .
Moreover, if h0 (IY n,d (d)) = 0 then also h0 (I D (d)) = 0, where D denotes a generic union of X sn,d , of
rn,d r
2-jets and of rn,d − 2 n2,d simple points. In fact, we have h0 (I X s (d)) = deg( R n,d ) = rn,d and
2
n,d
we conclude by Lemma 1.8.
The same conclusion (i.e. h0 (I D (d)) = 0) holds in the weaker assumption that h1 (I X s (d)) = 0,
d+n n,d
since in this case h0 (I X s (d)) = n

− deg( X sn,d ) = rn,d and we get h0 (I D (d)) = 0 by Lemma 1.8.
n,d
Theorem 2.2. Suppose that for all n 5, we have h1 (I X sn,3 (3)) = 0 and h0 (I X s (3)) = 0; then
n,3 +1
h (IY n,d (d)) = h (IY n,d (d)) = 0, for all d 4, n 4.
0 1
Proof. Let us consider a hyperplane H ⊂ Pn ; we want a scheme Z with support on H , made of

p
(2, 3, n)-schemes, and an integer vector p, such that the “differential trace” Tr H ( Z ) ⊂ H is OPn−1 (d)-
numerically settled.
Let us consider n 5 first. Since 0 rn−1,d 2n − 2, we write rn−1,d = nδ + 2h + , with 0 1,
0 δ 1 and 2h + n.
We denote by Z the zero dimensional scheme union of sn−1,d + h + + δ (hence δ = 0 if 0
rn−1,d n, while δ = 1 if n + 1 rn−1,d 2n − 2), (2, 3, n)-schemes Z 1 , . . . , Z sn−1,d +h+ +δ , where each
Z i is supported at P i with direction L i , and:
– the P i ’s are generic on H , i = 1, . . . , sn−1,d + h + + δ ;

– L i ⊂ H for i = 1, . . . , sn−1,d + h;
– if ( , δ) =
(0, 0), the corresponding lines L sn−1,d +h+1 , L sn−1,d +h+2 have generic directions in Pn
(hence not contained in H ).
In case n = 4, instead, we write r3,d = 2h + , with 0 1, and Z is given as before. Notice that
in this case 0 h 3, and it can appear only one line L s3,d +h+1 , not contained in H .
We want to use the Horace differential Lemma 1.5, where the role of the schemes H and Z appear-
ing in the statement of the lemma are played by our hyperplane H and the scheme Z just defined,
and with:
W = A sn−1,d +h+ +1 ∪ · · · ∪ A sn,d ∪ R n,d ,
S = A 1 ∪ · · · ∪ A sn−1,d +h+ +δ ,
p = (0, . . . , 0, 1, . . . , 1,
2 ,
0 ),

sn−1,d h δ
so that Tr H W = ∅ and Res H W = W , and Y n,d = W ∪ S.

Notice that this construction is possible, since sn−1,d + h + 2 sn,d (and even more than that): see
Appendix A, A.1.
In order to simplify notations, we set:
j j j j
T i := Tr H ( Z i ), R i := Res H ( Z i ), j = 0, 1, 2, i = 1, . . . , sn−1,d + h + + δ,
p
T := Tr H W ∪ Tr H ( Z ) = T 10 ∪ ··· ∪ T s0n−1,d ∪ T s1n−1,d +1 ∪ · · · ∪ T s1n−1,d +h ∪ T s2n−1,d +h+ ∪ T s0n−1,d +h+ +δ ,
p
R := Res H W ∪ Res H ( Z ) = W ∪ R 01 ∪ · · · ∪ R 0sn−1,d ∪ R 1sn−1,d +1 ∪ · · ·
∪ R 1sn−1,d +h ∪ R 2sn−1,d +h+ ∪ R 0sn−1,d +h+ +δ .
Observe that, by Remark 1.7:
T 10 , . . . , T s0n−1,d are (2, 3, n − 1)-points in H ∼

= Pn−1 , and R 01 , . . . , R 0sn−1,d are 2-jets in H ;
T s1 , . . . , T s1 are 2-jets in H and R 1s , . . . , R 1s are (2, 3, n − 1)-points in H ;
n−1,d +1 n−1,d +h n−1,d +1 n−1,d +h
T s2 is, when appearing, a simple point of H , and R 2s is a 2-fat point of H doubled

n−1,d +h + n−1,d +h + +δ
in a direction transversal to H ;
T s0 0
+h+ +δ is, when appearing, a 2-fat point on H , and R s is a 2-fat point in Pn with
n−1,d n−1,d +h +
support on H .
We will also make use of the scheme:
B := W ∪ R 1sn−1,d +1 ∪ · · · ∪ R 1s ∪ R 2sn−1,d +h+ .

n−1,d +h
Let us consider the following four statements:

Prop(n, d) : h0 IY n,d (d) = 0; Reg(n, d) : h1 I X s,n (d) = 0 and h0 I X s,n +1 (d) = 0,

Degue(n, d) : h0 I R (d − 1) = 0; Dime(n, d) : h0 I T , H (d) = 0.
If Degue(n, d) and Dime(n, d) are true, we know that Prop(n, d) is true too, by Proposition 1.5.
For the first values of n, d, we will need an “ad hoc” construction, which is given by the following:
Lemma 2.3. Let d = 4 and n ∈ {4, 5, 6}, then Prop(n, d) holds.
Proof. Case n = 4. Here we use the construction of R and T described above, hence we need to show
that Degue(4, 4) and Dime(4, 4) hold. Since s3,4 = 5, and r3,4 = 0, T is made of five generic (2, 3, 3)-
points in H ∼ = P3 , so Dime(4, 4) holds (i.e. h0 (P3 , I T , H (4)) = h0 (P3 , I X 5,3 (4) = 0), e.g. see [CGG1].
In order to prove Degue(4, 4) we want to apply Lemma 1.2, with R made of five 2-jets plus the
scheme B = W ; hence we need to show that h0 (I B (3)) 10, while h0 (IRes H ( B ) (2)) = 0. Since here
s4,4 = 7 = r4,4 , while r3,4 = 0, we have that B = W = Res H ( B ) and it is given by A 6 and A 7 , plus R 4,4 .
Hence we have h1 (I B (3)) = 0, since B is contained in the scheme made of 3 generic (2, 3, 4)-points
(which is known to have maximal Hilbert function, by [CGG1] or [B]); h1 (I B (3)) = 0 is equivalent to
saying that h0 (I B (3)) = 2s3,4 = 10, as required. Moreover h0 (I B (2)) = 0, since there is one only form
of degree two passing through two generic (2, 3, 4)-points in P4 , given by the hyperplane containing
the two double lines, doubled. Since the support of R 4,4 is generic, we get h0 (I B (2)) = 0. So we have
that Degue(4, 4) holds, and Prop(4, 4) holds too.
Case n = 5. Here we need to use a different construction. We have s5,4 = 11, r5,4 = 5, s4,4 = 7 =
r4,4 . We want to use the Horace differential Lemma 1.5 with Z = Z 1 ∪ · · · ∪ Z 8 ∪ R 5,4 , where Z 1 , . . . , Z 8
are (2, 3, 5) schemes supported at generic points of H with direction L 1 , . . . , L 8 ⊂ H , and we specialize
R 5,4 so that R 5,4 ⊂ H , contained in a generic (2, 3, 4)-scheme of H ; with W = A 9 ∪ A 10 ∪ A 11 , and

with p = (0, . . . , 0, 1, 0).

7
p p
Hence T = Tr H W ∪ Tr H ( Z ) = T 10 ∪ T 20 ∪ · · · ∪ T 70 ∪ T 81 ∪ R 5,4 and R = Res H W ∪ Res H ( Z ) = W ∪ R 01 ∪
R 2 ∪ · · · ∪ R 07 ∪ R 18 .
0
We have that the ideal sheaf of T 10 ∪ T 20 ∪ · · · ∪ T 70 ∪ R 5,4 has h1 = 0 and h0 = 2 in degree 4, by

using the previous case and the fact that R 5,4 is contained in a (2, 3, 4)-point, so h0 (I T , H (4)) = 0
by Lemma 1.8, since T 81 is a 2-jet in H ∼ = P4 . We also have h0 (I R (3)) = 0. In fact, let us denote by
U the scheme U = R 8 ∪ W . In order to apply Lemma 1.9 (the R 0i ’s are 2-jets) to get h0 (I R (3)) = 0,
1
we need to show that h0 (IRes H U (2)) = 0 and h1 (IU (3)) = 0. Since U is included in the union of four
(2, 3, 5)-points, which impose independent conditions in degree three (e.g. see [CGG1]), h1 (IU (3)) = 0
follows. Moreover, Res H (U ) is made by three (2, 3, 5)-points, and again h0 (IRes H U (2)) = 0 is known
by [CGG1].
Now, h0 (I T , H (4)) = 0 = h0 (I R (3)) imply Prop(5, 4) by Lemma 1.5, and we are done.
Case n = 6. Here we have s6,4 = 16, r6,4 = 2, while s5,4 = 11, r5,4 = 5. We want to use the Ho-
race differential Lemma 1.5 with Z = Z 1 ∪ · · · ∪ Z 13 ∪ R 6,4 , where Z 1 , . . . , Z 13 are (2, 3, 6) schemes
supported at generic points of H with direction L 1 , . . . , L 12 ⊂ H , while L 13 is not in H , and we spe-
cialize R 6,4 ⊂ H , as a generic 2-jet in H ; with W = A 14 ∪ A 15 ∪ A 16 , and with p = (0, . . . , 0, 1, 2, 0).

11
p p
Hence T = Tr H W ∪ Tr H ( Z ) = T 10 ∪ T 20 ∪ · · · ∪ T 11
0
∪ T 12
1
∪ T 13
2
∪ R 6,4 and R = Res H W ∪ Res H ( Z ) =
¯
W ∪ R 01 ∪ R 02 ∪ · · · ∪ R 011 ∪ R 112 ∪ R 213 .
We have that h0 (I T , H (4)) = 0 by applying Lemma 1.1 and the previous case.
We also have h0 (I R (3)) = 0. In fact, let us denote by U the scheme U = R 112 ∪ R 213 ∪ W . In order to
apply Lemma 1.9 (the R 0i ’s are 2-jets) to get h0 (I R (3)) = 0, we need to show that h0 (IRes H U (2)) = 0
and h1 (IU (3)) = 0.
Since U is included in the union of five (2, 3, 6)-points, which impose independent conditions in
degree three (e.g. see [CGG1]), h1 (IU (3)) = 0 follows. Moreover, Res H (U ) is made by three (2, 3, 6)-
points plus a 2-fat point inside H ∼ = P5 . Since there is only one form of degree two passing through
three generic (2, 3, 6)-points in P , given by the hyperplane containing the three double lines, dou-
6
bled, we get h0 (IRes H U (2)) = 0.

Now, h0 (I T , H (4)) = 0 = h0 (I R (3)) imply Prop(6, 4) by Lemma 1.5, and we are done. 2
Now we come back to the proof of the theorem for the remaining values of n, d; we will work by
induction on both n, d in order to prove statement Prop(n, d) for n 4, d 5 and for n 7, d = 4.
We divide the proof in 7 steps.
Step 1. The induction is as follows: we suppose that Prop(ν , δ) is known for all (ν , δ) such that
4 ν < n and 4 δ d or 4 ν n and 4 δ < d and we prove that Prop(n, d) holds.
The initial cases for the induction are given by Lemma 2.2, and we will also make use of the fact
that Reg(n, 3) with n 4 and Reg(3, d) with d 4 hold respectively by assumption and by [B], while,
by [CGG1], we know everything about the Hilbert function of generic (2, 3, n)-schemes when d = 2.
We will be done if we prove that Degue(n, d) and Dime(n, d) hold for n 4, d 5 and for n 7,
d = 4.
Step 2. Let us prove Dime(n, d). Notice that T is OPn−1 (d)-numerically settled in H ∼ = Pn−1 , hence
Dime(n, d) is equivalent to h (I T , H (d)) = 0.
1
The scheme T is the generic union of X sn−1,d with h 2-jets, of simple points and of δ 2-fat points,
where 2h + + nδ = rn−1,d . Then Dime(n, d) holds for n 5 and d 4 since we are assuming that
Prop(n − 1, d) is true and the union of h 2-jets, simple points and of δ 2-fat points can specialize
to R n−1,d (see Lemma 1.10).
For n = 4 and d 5, Dime(4, d) holds, since we know that h1 (I X s (d)) = 0 by [B] and in this case
3,d
T is the generic union of X s3,d with h 2-jets and simple points so we can apply Lemma 1.8.
Step 3. We are now going to prove Degue(n, d). Since the scheme R is the union of the scheme
B and of sn−1,d 2-jets lying on H (see definitions of R and B above), we can use Lemma 1.9(ii).
Hence, in order to prove that dim( I R )d−1 = 0, i.e. that Degue(n, d) holds, it is enough to prove that
( I Res H ( B ) )d−2 = 0 and that dim( I B )d−1 2sn−1,d .
Step 4. Let us show that ( I Res H ( B ) )d−2 = 0. We set tn,d := sn,d − sn−1,d − h − − δ . The scheme Res H ( B )
is given by W plus, if = 1, one 2-fat point contained in H , plus, if δ = 1, one simple point in H . W is
the generic union of R n,d with tn,d (2, 3, n)-points. Let I denote the ideal of these tn,d (2, 3, n)-points;
if we show that I d−2 = 0, then also ( I Res H ( B ) )d−2 = 0.
The idea is to prove that our (2, 3, n)-points are “too many” to have I d−2 = 0 since they are more
than sn,d−2 + 1; the only problem with this procedure is that there are cases (when d − 2 = 2 or 3)
where I d−2 may not have the expected dimension, so those cases have to be treated in advance.
First let d = 4 (and n 7); if we show that tn,4 > n2 , then we are done, since ( I X s,n )2 = 0 for s > n2 ,
by [CGG1], Proposition 3.3. The inequality tn,4 > n2 is treated in Appendix A, A.2, and proved for n 7,
as required.
Now let d = 5 and n = 4; here we have that s4,3 + 1 = 4, but actually there is one cubic hypersur-
face through four (2, 3, 4)-points in P4 ; nevertheless, since t 4,5 = 14 − 8 − 0 − 0 = 6, and it is known
(see [CGG1] or [B]) that ( I X 6,4 )3 = 0, we are done also in this case.
Eventually, for d = 5, n 5, or in the general case d 6, n 4, if we show that tn,d sn,d−2 + 1,
the problem reduces to the fact that ( I X s
n,d−2
+1 )d−2 = 0. If d = 5, we know that ( I X sn,3 +1 )3 = 0 by
hypothesis, while for d 6 we can suppose that ( I X s +1 )d−2 = 0 by induction on d.
n,d−2
The inequality tn,d sn,d−2 + 1 is discussed in Appendix A, A.1, and proved for all the required
values of n, d.
Thus the condition ( I Res H ( B ) )d−2 = 0 holds.
Step 5. Now we have to check that dim( I B )d−1 2sn−1,d . Since deg Y n,d = h0 (OPn (d)) and deg T =
h0 (OPn−1 (d)), then deg R = h0 (OPn (d − 1)). The scheme R is the union of the scheme B and of sn−1,d
2-jets lying on H , so deg R = deg B + 2sn−1,d . Hence dim( I B )d−1 2sn−1,d is equivalent to h1 (I B (d −
1)) = 0 (and to dim( I B )d−1 = 2sn−1,d ).
Let us consider the case n 5 first. Let Q be the scheme Q = Z sn−1,d +1 ∪ · · · ∪ Z sn,d +h+ +δ ∪
A sn,d +h+ +δ+1 ∪ · · · ∪ A sn,d ∪ A sn,d +1 , where A sn,d +1 is a (2, 3, n) scheme containing R n,d . We have that
B is contained in the scheme Q , which is composed by sn,d − sn−1,d + 1 generic (2, 3, n)-points (notice
that 2h + + δ n + 1, so Z sn−1,d +1 , . . . , Z sn,d +h+ +δ are generic, since only the first h of the lines L i
are in H ).
The generic union of sn,d−1 generic (2, 3, n)-points in Pn is the scheme X sn,d−1 ; by induction, or by
hypothesis if d − 1 = 3, we have h1 (I X s (d − 1)) = 0. Since sn,d − sn−1,d + 1 sn,d−1 (see Step 6),
n,d−1
then B ⊂ Q ⊂ X sn,d−1 and we conclude by Remark 2.1 that h1 (I B (d − 1)) = 0.
Step 6. We now prove the inequality: sn,d − sn−1,d + 1 sn,d−1 (n 5). We have deg Q = deg B + 2h +
+ nδ + (2n + 1 − rn,d ), in fact in order to “go from B to Q ,” we have¯ to add a 2-jet to each of the
R 1i (h in number), a simple point to R 2s if = 1, a 2-fat point of H if δ = 1 and something of
n−1,d +h +
degree (2n + 1 − rn,d ) to R n,d .
Since rn,d 0 and 2h + + nδ = rn−1,d 2n − 2, we have: deg Q = (2n + 1)(sn,d − sn−1,d + 2)
deg( B ) + 2n − 2 + 2n + 1 = deg( B ) + 4n − 1.
Notice that deg(Y n,d−1 ) = deg( B ) + 2sn−1,d , so we have: (2n + 1)(sn,d − sn−1,d + 1) deg(Y n,d−1 ) −
2sn−1,d + 4n − 1.
If we prove that 4n − 1 − 2sn−1,d 0, we obtain: (2n + 1)(sn,d − sn−1,d + 1) deg(Y n,d−1 ) =
(2n + 1)sn,d−1 , and we are done.
The computations to get 4n − 1 − 2sn−1,d 0 can be found in Appendix A.3.
Step 7. We are only left to prove that h1 (I B (d − 1)) = 0 in case n = 4 (d 5).

Recall that now r3,d = 2h + 6, with 0 h 3, 0 1. If r3,d 4, we can apply the same
procedure as in Step 5, since the part of the scheme Q with support on H is generic in P4 . Hence we
only have to deal with r3,d = 5, 6.
The case r3,d = 5 does not actually present itself; this can be checked by considering that

d+3 (d + 3)(d + 2)(d + 1)
= = 7s3,d + r3,d ⇒ (d + 3)(d + 2)(d + 1) = 42s3,d + 6r3,d .
3 6
Hence if r3,d = 5, we get 42s3,d + 30 = 7(6s3,d + 4) + 2, but it is easy to check that (d + 3)(d + 2)(d + 1)
never gives a remainder of 2, modulo 7.
Thus we are only left with the case r3,d = 6, when h = 3 and = 0. In this case we have d ≡ 3
(mod 7), hence d 10; it is also easy to check that r3,d−1 = 3 in this case.
We can add 2s3,d generic simple points to B, in order to get a scheme B which is OP4 (d − 1)-
numerically settled, so now h1 (I B (d − 1)) = 0 is equivalent to h0 (I B (d − 1)) = 0 (by Remark 2.1).
We want to apply Horace differential lemma again in order to prove h0 (I B (d − 1)) = 0; so we
will define appropriate schemes Z B , W B and an integer vector q, such that conditions (a) and (b) of
Proposition 1.5 apply to them, yielding h0 (I B (d − 1)) = 0.
Consider the scheme Z B ⊂ P4 , given by s3,d−1 − 1 (2, 3, 4)-schemes in P4 , such that their support
is at generic points of H , and only for the last one of them the line L i is not in H . Let W B ⊂ P4 be
given by 2s3,d generic simple points, s4,d − s3,d − s3,d−1 − 2 generic (2, 3, 4)-schemes, three generic
(2, 3, 3)-schemes in H ∼= P3 , and the scheme R 4,d . Let also q = (0, . . . , 0,
1 ,
2 ).

s3,d−1 −3 1 1
p q
Let T B = Tr H ( W B ) ∪ Tr H ( Z B ) = X s3,d−1 ∪ E ∪ F , and R B = Res H ( W B ) ∪ Res H ( Z B ).
We have that E and F are, respectively, a 2-jet and a simple point in H (they give the “remainder
scheme” of degree 3, to get that T B is OP3 (d − 1)-numerically settled).
The scheme R B is the union of 2s3,d generic simple points, s4,d − s3,d − s3,d−1 − 2 generic (2, 3, 4)-
schemes, the scheme R 4,d , s3,d−1 2-jets in H , a (2, 3, 3)-scheme in H and a 2-fat point of H doubled
in a direction transversal to H .
If we show that h0 (I R B (d − 2)) = 0 = h0 (I T B , H (d − 1)), then we are done by Proposition 1.5.
We have h0 (I T B , H (d − 1)) = 0, since T B is OP3 (d − 1)-numerically settled, and is given by the
union of X s3,d−1 (whose ideal sheaf has h1 = 0 in degree d − 1 by [B]) with a 2-jet and a simple point,
so we can apply Lemma 1.8.
In order to show that h0 (I R B (d − 2)) = 0 we want to proceed as in Step 5, i.e. by applying
Lemma 1.9, since R B , is made of s3,d−1 − 3 2-jets union the 2s3,d generic simple points and a
scheme that we denote by R B . We will be done if we show that h0 (IRes H ( R B ) (d − 3)) = 0 and
h1 (I R (d − 2)) = 0.
B
The first condition will follow if s4,d − s3,d − s3,d−1 − 2 s4,d−3 , the second condition (since R B is
contained in the union of s4,d − s3,d − s3,d−1 + 1 generic (2, 3, 4)-schemes) if s4,d − s3,d − s3,d−1 + 1
s4,d−2 .
Both inequalities are proved in Appendix A, A.4. 2
Thanks to some “brute force” computation by COCOA, we are able to prove:
Corollary 2.4. For 4 n 9, we have:
(i) h1 (I X sn,3 (3)) = 0 and h0 (I X s +1 (3)) = 0, except for n = 4, in which case we have h0 (I X s,4 (3)) = 0 for
n ,3
s 5.
(ii) h0 (IY n,d (d)) = h1 (IY n,d (d)) = 0, for d 4.
Proof. Part (i) comes from direct computations using CoCoA [CO]. Note that s4,3 = 3 and that
h0 (I X 4,4 (3)) = h1 (I X 4,4 (3)) = 1, see [CGG1].
Part (ii) comes by applying the theorem and part (i). 2
Coming back to the language of secant varieties, Theorem 2.2 and Corollary 2.4 give:
Corollary 2.5. If Conjecture 1 is true for d = 3, then it is true for all d 4. Moreover, for n 9, Conjecture 1
holds.
3. On Conjecture 2a. The case n = 2
In this section we prove Conjecture 2a for n = 2.

We want to use the fact that σs ( O k,n,d ) is defective if at a generic point its tangent space does
not have the expected dimension; actually (see [BCGI]) this is equivalent to the fact that for generic
L i ∈ R 1 , F i ∈ R k , R = κ [x0 , . . . , xn ], i = 1, . . . , s, the vector space L d1−k R k , L d1−k−1 F 1 R 1 , . . . , L ds −k R k ,
L ds −k−1 F s R 1 does not have the expected dimension.
Via inverse systems this reduces to the study of ( I Y )d , where Y = Z 1 ∪ · · · ∪ Z s is a certain 0-
dimensional scheme in Pn . Namely, the scheme Y is supported at s generic points P 1 , . . . , P s ∈ Pn , at
k+n
each of them deg( Z i ) = n + n, and I kP+2 ⊂ I Z i ⊂ I kP+1 (see Lemma 1.2).
i i
When working in P2 , we can specialize the F i ’s to be of the form ki , where i is a generic
linear form through P i . In this way we get a scheme Y = Z 1 ∪ · · · ∪ Z s , and the structure of
each Z i is ((k + 2) P i ∩ L 2i ) ∪ (k + 1) P i , where the line L i is “orthogonal” to i = 0, i.e. if we put
P i = (1, 0, 0), i = x1 and L i = {x2 = 0}, the ideal is of the form: ((x1 , x2 )k+2 + (x2 )2 ) ∩ (x1 , x2 )k+1 =
(xk1+2 , xk1+1 x2 , xk1−1 x22 , . . . , xk2+1 ).
Notice that the forms in I Z i have multiplicity at least k + 1 at P i and they meet L i with multiplicity
at least k + 2; moreover the generic form in I Z i has L i at least as a double component of its tangent
cone at P i .
When F ∈ I Z i and we speak of its “tangent cone” at P i , we mean (with the choice of coordinates
above) either the form in κ [x1 , x2 ] obtained by putting x0 = 1 in F and considering the (homoge-
neous) part of minimum degree thus obtained, or also the scheme (in P2 ) defined by such a form.
When we will say that L i is a “simple tangent” for F , we will mean that L i is a reduced component
for the tangent cone to F at P i .
The strategy we adopt to prove Conjecture 2a is the following: if ( I Y )d does not have the ex-
pected dimension, i.e. h0 (IY (d))h1 (IY (d)) = 0, then the same happens for IY (d); hence Conjecture 2a
would be proved if we show that whenever dim( I Y )d is more than expected, then h1 (I X (d)) >
d+n d+n
max{0, deg(Y ) − n
} or h0 (I T (d)) > max{0, n
− deg(Y )}, where
X := (k + 1) P 1 ∪ · · · ∪ (k + 1) P s ⊂ P2 ; T := (k + 2) P 1 ∪ · · · ∪ (k + 2) P s ⊂ P2 .
The following easy technical Bertini-type lemma and its corollary will be of use in the sequel.
Lemma 3.1. Let F , G be linearly independent polynomials in κ [x]. Then for almost any a ∈ κ , F + aG has at
least one simple root.
Proof. Let M be the greatest common divisor of F and G with F = M P , G = M Q . Let us consider
P Q − Q P , where P and Q are the derivatives of P and Q , respectively. Since P and Q have no
common roots, it easily follows that P Q − Q P cannot be identically zero.
For any β ∈ κ which is neither a root for P Q − Q P , nor for M, nor for Q , let
P (β)
a = a(β) := − ,
Q (β)
so ( F + aG )(β) = M (β)( P + a Q )(β) = 0, and ( F + aG ) (β) = ( M ( P + a Q )+ M ( P + a Q ))(β) = ( M ( P +

a Q ))(β) = ( M ( P − Q (β) Q ))(β) = ( M )(β)( Q P − P Q )(β) = 0, hence β is a simple root for F + aG.
P (β)
Q
Since β assumes almost every value in κ , so does a(β). 2
Corollary 3.2. Let P = (1, 0, 0) ∈ P2 . Let f , g ∈ ( I kP+1 )d , and f , g ∈

/ ( I kP+2 )d . Assume that f , g, have different
tangent cones at P . Then for almost any a ∈ κ , f + ag has at least one simple tangent at P .
Proof. The corollary follows immediately from Lemma 3.1 by de-homogenising the tangent cones to
f , g at P to get two non-zero and non-proportional polynomials F , G ∈ κ [x]. 2
It will be handy to introduce the following definitions.
Definition 3.3. Let P ∈ P2 and L be a line L through P . We say that a scheme supported at one point
is of type Z if its structure is (k + 1) P ∪ ((k + 2) P ∩ L ), and that it is of type Z if its structure is
(k + 1) P ∪ ((k + 2) P ∩ L 2 ).
We will say that a union of schemes of types Z and/or Z is generic if the points of their support
and the relative lines are generic.
The following lemma is the key to prove Conjecture 2a:
Lemma 3.4. Let Y = Z 1 ∪ · · · ∪ Z s ⊂ P2 be a union of s generic schemes of type Z , then either:
(i) ( I Y )d = ( I T )d ;
or
(ii) dim( I Y )d = dim( I X )d − 2s.
Proof. Notice that by the genericity of the points and of the lines, the Hilbert function of a scheme
with support on P 1 , . . . , P s , formed by t schemes of type Z , by t schemes of type Z and by s − t − t
fat points of multiplicity (k + 1) depends only on s, t and t .
Let W t be a scheme formed by t schemes of type Z and by s − t fat points of multiplicity (k + 1).
Let

τ = max t ∈ N dim( I W t )d = dim( I X )d − 2t .
If τ = s, we have W s = Y and dim( I W s )d = dim( I X )d − 2s, hence (ii) holds.

Let τ < s: we will prove that ( I Y )d = ( I T )d . Let W be the scheme
W = W τ = Z 1 ∪ · · · ∪ Z τ ∪ (k + 1) P τ +1 ∪ · · · ∪ (k + 1) P s ,
and let
W ( j ) = Z 1 ∪ · · · ∪ Z τ ∪ (k + 1) P τ +1 ∪ · · · ∪ Z j ∪ · · · ∪ (k + 1) P s , τ + 1 j s,
W (j ) = Z 1 ∪ · · · ∪ Z τ ∪ (k + 1) P τ +1 ∪ · · · ∪ Z j ∪ · · · ∪ (k + 1) P s , τ + 1 j s,
that is W ( j ) , respectively W (j ) , is the scheme obtained from W by substituting the fat point (k + 1) P j
with a scheme of type Z , respectively Z , so
W ⊂ W ( j ) ⊂ W (j ) ,
and deg W ( j ) = deg W + 1, deg W (j ) = deg W + 2 (for τ = s − 1, W (s) = Y ).

If ( I W )d = 0, then trivially ( I Y )d = ( I T )d = 0 and we are done. So assume that ( I W )d = 0.
( j) ( j)
By the definition of τ we have that dim( I W (j) )d > dim( I X )d − 2(τ + 1) = dim( I W )d − 2, hence we
get
0 dim( I W )d − dim( I W )d 1.
( j)
Let us consider the two possible cases.
Case 1. dim( I W )d − dim( I W )d = 0, τ + 1 j s.

( j)
In this case we have ( I W )d = ( I W )d . This means that every form F ∈ ( I W )d meets the line L j
( j)
with multiplicity at least k + 2; but since the line L j is generic through P j , this yields that every line
through P j is met with multiplicity at least k + 2, hence
( I W )d ⊂ ( I kP+j 2 )d , for τ + 1 j s. (1)
In particular, we have that
( I W )d = ( I W )d . (2)
(s)
Now consider the schemes
W (i ,s) = Z 1 ∪ · · · ∪ Z i −1 ∪ (k + 1) P i ∪ Z i +1 ∪ · · · ∪ Z τ ∪ (k + 1) P τ +1 ∪ · · ·
∪ (k + 1) P s−1 ∪ Z s , 1 i τ,
W (i ,s)
= Z 1 ∪ · · · ∪ Z i −1 ∪ Z i ∪ Z i +1 ∪ · · · ∪ Z τ ∪ (k + 1) P τ +1 ∪ · · ·
∪ (k + 1) P s−1 ∪ Z s , 1 i τ,
i.e. W (i ,s) is the scheme obtained from W by substituting the fat point (k + 1) P i to the scheme Z i
and a scheme Z s , of type Z , to the fat point (k + 1) P s , while W (i ,s) is the scheme obtained from
W (i ,s) by substituting a scheme Z i , of type Z , to the fat point (k + 1) P i .
The schemes W (i ,s) and W are made of τ schemes of type Z and s − τ (k + 1)-fat points; the
schemes W (i ,s) and W (s) are made of τ schemes of type Z , s − τ − 1 (k + 1)-fat points and one
scheme of type Z . This yields that:
dim( I W (i,s) )d = dim( I W )d = dim( I W )d = dim( I W )d .

(s) (i , s )
Hence every form F ∈ ( I W (i,s) )d meets the generic line L i with multiplicity at least k + 2, thus we get

( I W (i,s) )d ⊂ I kP+i 2 d , for 1 i τ , (3)
and from this and (2) we have
( I W (i,s) )d = ( I W )d = ( I W )d . (4)
(s)
By (1), (3) and (4) it follows that ( I W )d = ( I T )d , hence, since W ⊂ Y ⊂ T , we get (i).
Case 2. dim( I W )d − dim( I W )d = 1, τ + 1 j s.

( j)
In this case we have
dim( I W )d = dim( I W )d .
( j) ( j)
Let F ∈ ( I W )d = ( I W )d ; hence L j appears with multiplicity two in the tangent cone of F . If F ∈

/
( j) ( j)
( I kP+j 2 )d , then let L j be a generic line not in the tangent cone of F at P j . By substituting the line L j
/ ( I kP+j 2 )d , with the double line
to L j in the construction of W ( j ) , we get another form G ∈ ( I W )d , G ∈

L j in its tangent cone. Then, by Corollary 3.2, the generic form F + aG has a simple tangent at P j ,
and this is a contradiction since a generic choice of the line L j should yield ( I W )d = ( I W )d . Hence
( j) ( j)
F ∈ ( I kP+2 )d , for τ + 1 j s.
j
With an argument like the one we used in Case 1, we also get that F ∈ ( I kP+2 )d for 1 j τ , and
j
(i) easily follows. 2
Now we are ready to prove Conjecture 2a.
Theorem 3.5. The secant variety σs ( O k,2,d ) is defective if and only if one of the following holds:
d+n
(i) h1 (I X (d)) > max{0, deg(Y ) − }, or
d+n n
(ii) h0 (I T (d)) > max{0, n
− deg(Y )}.
Proof. Since if Y is defective in degree d, then Y is, but, by Lemma 3.4, either dim( I Y )d = dim( I X )d −
2s, hence

d+n d+n
h1 I X (d) = h1 IY (d) − 2s > max 0, deg(Y ) − = max 0, deg(Y ) − ,
n n
or ( I Y )d = ( I T )d , hence

d+n d+n
h0 I T (d) = h0 IY (d) > max 0, − deg(Y ) = max 0, − deg(Y ) . 2
n n
Appendix A. Calculations
A.1. We want to prove that (for n 4 and d 6 or for n 5 and d = 5):
sn,d − sn−1,d − h − − δ − 1 sn,d−2 .
Recall:

n+d n+d−1
sn,d (2n + 1) + rn,d = , sn−1,d (2n − 1) + rn−1,d = ,
d d

n+d−2
sn,d−2 (2n + 1) + rn,d−2 = .
d−2
Hence our inequality becomes:

1 n+d 1 n+d−1
− rn,d − − rn−1,d
2n + 1 d 2n − 1 d

1 n+d−2
−h− −δ −1− − rn,d−2 0.
2n + 1 d−2
By using binomial equalities and reordering this is:

1 n+d−1 n+d−2 n+d−2 1 n+d−1 rn−1,d
+ + − +
2n + 1 d d−1 d−2 2n − 1 d 2n − 1

1 n+d−2 1
−h− −δ −1− + (rn,d−2 − rn,d ) 0
2n + 1 d−2 2n + 1
i.e.

1 n+d−2 2 n+d−1 rn−1,d
− +
2n + 1 d−1 (2n + 1)(2n − 1) d 2n − 1
1
−h− −δ −1+ (rn,d−2 − rn,d ) 0.
2n + 1
By using binomial equalities again:

1 n+d−2 2 n+d−2 n+d−2 rn−1,d
− + +
2n + 1 d−1 (2n + 1)(2n − 1) d d−1 2n − 1
1
−h− −δ −1+ (rn,d−2 − rn,d ) 0
2n + 1
i.e.

1 n+d−2 2 2 n+d−2 rn−1,d
1− − +
2n + 1 d−1 2n − 1 (2n + 1)(2n − 1) d 2n − 1
1
−h− −δ −1+ (rn,d−2 − rn,d ) 0
2n + 1
i.e.

n + d − 2 [2n(d − 1) − 3d + 2] rn−1,d 1
+ −h− −δ −1+ (rn,d−2 − rn,d ) 0.
d−1 d(4n2 − 1) 2n − 1 2n + 1
r
n−1,d n 1
Now, 2n −1 0, while h + + δ 2 , and rn,d−2 − rn,d −2n, i.e. (r
2n+1 n,d−2
− rn,d ) − 2n2n+1 −1,
so our inequality holds if:

n + d − 2 [2n(d − 1) − 3d + 2] n
− − 2 0.
d−1 d(4n2 − 1) 2
It is quite immediate to check that the right-hand side is an increasing function in d, e.g. by writing
it as follows:

n+d−2 2n + 2 n
2n − 3 − − + 2 4n2 − 1 0
n−1 d 2
i.e.

n+d−2 2n + 2 n
2n − 3 − − 2n3 − 8n2 + + 2 0.
n−1 d 2
Let us consider the case d = 6 first; our inequality becomes:

n + 4 (10n − 16) n
− 2n3 − 8n2 + +20
5 6 2
i.e.
(n + 4)(n + 3)(n + 2)(n + 1)n(5n − 8) n

− 2n3 − 8n2 + +20
360 2
i.e.
(n + 4)(n + 3)(n + 2)(n + 1)n(5n − 8) − 20n2 (n + 2) n

+ +20
360 2
i.e.
n(n + 2) n
(n + 4)(n + 3)(n + 1)(5n − 8) − 720n + + 2 0.
360 2
Which, for n 4, is easily checked to be true. Hence we are done for n 4, d 6.

Now let us consider the case d = 5; our inequality becomes:

n + 3 (8n − 13) n
− 2n3 − 8n2 + +20
4 5 2
i.e.
(n + 3)(n + 2)(n + 1)n(8n − 13) n

− 2n3 − 8n2 + +20
120 2
i.e.

n4 + 6n3 + 11n2 + 6n (8n − 13) − 240n3 − 960n2 + 60n + 240 0
i.e.
8n5 + 35n4 − 230n3 − 1015n2 − 18n + 240 0
i.e.

1015 18 240
n3 8n2 + 35n − 230 − − + 0.
n n2 n3
Which, for n 6, holds. So we are left to prove our inequality for d = 5 = n; in this case we have:
s5,5 = [ 272
11
] = 24, s4,5 = [ 126
9
] = 14 and r4,5 = 0, hence h = = 0, while s5,3 = [ 56
11
] = 5; so: s5,5 −
s4,5 − 1 s5,3 becomes: 24 − 14 − 1 5, which holds.
A.2. We want to prove that, for all n 7:
n
sn,4 − sn−1,4 − h − − δ >
2
i.e.

n+4 n−1+4 n
/(2n + 1) − rn,4 /(2n + 1) − /(2n − 1) + rn−1,4 /(2n − 1) − h − − δ >
4 4 2
i.e.
(n + 4)(n + 3)(n + 2)(n + 1) (n + 3)(n + 2)(n + 1)n n r n ,4 rn−1,4

− − − + − h − − δ > 0.
24(2n + 1) 24(2n − 1) 2 (2n + 1) (2n − 1)
Now:
r n ,4 2n r n ,4
< 1, hence − > −1;
(2n + 1) (2n + 1) (2n + 1)
rn−1,4 0;
and h + + δ n2 , i.e. −h − − δ − n2 . Therefore we get:

(n + 3)(n + 2)(n + 1) (n + 4) n n r n ,4 rn−1,4
· − − − + −h− −δ
24 (2n + 1) (2n − 1) 2 (2n + 1) (2n − 1)

(n + 3)(n + 2)(n + 1) n+4 n n n
> · − − − −1
24 2n + 1 2n − 1 2 2
(n + 3)(n + 2)(n + 1) [(2n − 1)(n + 4) − n(2n + 1)]
= · −n−1
24 (2n + 1)(2n − 1)

(n + 3)(n + 2)(3n − 2)
= (n + 1) − 1 >0
12(4n2 − 1)
i.e.

(n + 3)(n + 2)(3n − 2) − 12 4n2 − 1 > 0
i.e.
3n3 − 35n2 + 8n > 0
which is true for n 12.

Let us check the cases n = 7, 8, 9, 10, 11.

1 11

If n = 7 we have: s7,4 = [ 15 4
] = 22 (with r7,4 = 0); s6,4 = 16, since 10 4
= 210 = 16 · 13 + 2, so
r6,4 = 2 and h = 1, = δ = 0.
Our inequality becomes: 22− 16 − 1 > 7/2, which holds.
1 12
If n = 8 we have: s8,4 = [ 15 4
] = 33 (with r8,4 = 0); s7,4 = 22, r7,4 = 0 and h = = δ = 0.
Our inequality becomes: 33− 22 > 4, which holds.
1 13
If n = 9 we have: s9,4 = [ 15 4
] = 47 (with r9,4 = 10); s8,4 = 33, and h = = δ = 0.
Our inequality becomes: 47 −33 > 9/2, which holds.
1 14
If n = 10 we have: s10,4 = [ 15 4
] = 66 (with r10,4 = 11) ; s9,4 = 47, and h = 5, = δ = 0.
Our inequality becomes: 66 − 47 − 5 > 5, which holds.
1
15
If n = 11 we have: s10,4 = [ 15 4
] = 91; s10,4 = 66, and h = 5, = 1, δ = 0.
Our inequality becomes: 91 − 66 − 5 − 1 > 11/2, which holds.
A.3. We want to prove that, for d 5, n 4 or d = 4, n 7:
4n − 1 2sn−1,d . (∗)
Since rn−1,d 2n − 2, it is enough to prove that:

2 n−1+d
− 2n + 2 4n − 1
2n − 1 n−1
which is:

n−1+d (4n − 1)(2n − 1)
+ 2n − 2
n−1 2
that is:

n−1+d 3
4n2 − n − (∗∗)
n−1 2
which is surely true if

n−1+d
4n2 − n
n−1
is true. n−1+d
Notice that the function n−1 is an increasing function in d. For d = 4, the inequality becomes:
n(n3 + 6n2 + 11n + 6)

4n2 − n,
24
which can be written:
n3 + 6n2 + 11n + 6 96n − 24
i.e.
n3 + 6n2 − 85n + 30 0
which is surely true if the following is true:
n2 + 6n − 85 0.
The last one is verified for n 8, so we 10are done for d = 4 and n 8.

If (n, d) = (7, 4), sn−1,d = 16 since 4 = 210 = 16 · 13 + 2, and (∗) becomes: 4 · 7 − 1 2 · 16 which
is true. n−1+d
Since the function n−1 is an increasing function in d, we have proved the initial inequality for
d 4 and n 8.
For d = 5 (∗∗) becomes: n5 + 10n4 + 35n3 − 430n2 + 144n + 120 0 which is true for n = 5, 6, 7.
We have hence proved the initial inequality for d 5 and n 5.
8
If (n, d) = (4, 5), sn−1,d = 8 since 3
= 8 · 7, and (∗) becomes: 4 · 4 − 1 2 · 8 which is true.
For d = 6 (∗∗) becomes: n(n + 1)(n + 2)(n + 3)(n + 4)(n + 5) − 120(6)(4n2 − n − 1) 0 which is
true for n = 4. We conclude that the initial inequality is true for d 5 and n 4.
A.4. We want to show that (for d 10): s4,d − s3,d − s3,d−1 − 2 s4,d−3 and s4,d − s3,d − s3,d−1 + 1
s4,d−2 .
The first inequality is equivalent to:

1 d+4 1 d+3 6 1 d+2 3 1 d+1
− + − + −2
9 4 7 3 7 7 3 7 9 4
which follows if:

1 d+4 1 d+1 1 d+3 1 d+2 9
− + − +4
9 4 9 4 7 3 7 3 7
i.e.

d + 1 [(d + 4)(d + 3)(d + 2) − d(d − 1)(d − 2)] 1 (d + 1)(d + 2)(2d + 3) 19
+
9 24 7 6 7
i.e.
d + 1 (12d2 + 24d + 24) 1 19

(d + 1)(d + 2)(2d + 3) +
9 24 42 7
i.e.
(d2 + 2d + 2) 2d2 + 7d + 6 114

+
3 7 7(d + 1)
i.e.
342
d2 − 7d − 4 .
d+1
Which is easily checked to hold for d 10.

Now let us consider the second inequality, which is equivalent to:

1 d+4 1 d+3 6 1 d+2 3 1 d+2
− + − + +1
9 4 7 3 7 7 3 7 9 4
which follows if:

1 d+4 1 d+2 1 d+3 1 d+2 9
− + − −3
9 4 9 4 7 3 7 3 7
i.e.

(d + 1)(d + 2) [(d + 4)(d + 3) − d(d − 1)] 1 (d + 1)(d + 2)(2d + 3) 30
−
9 24 7 6 7
i.e.
(d + 1)(d + 2) (8d + 12) 1 30

(d + 1)(d + 2)(2d + 3) −
9 24 42 7
i.e.
1 1 180
− .
9 7 7(d + 1)(d + 2)(2d + 3)
Which is easily checked to hold for d 10.
References
[AH1] J. Alexander, A. Hirschowitz, Polynomial interpolation in several variables, J. Algebraic Geom. 4 (1995) 201–222.
[AH2] J. Alexander, A. Hirschowitz, An asymptotic vanishing theorem for generic unions of multiple points, Invent. Math. 140
(2000) 303–325.
[B] E. Ballico, On the secant varieties to the tangent developable of a Veronese variety, J. Algebra 288 (2005) 279–286.
[BBCF] E. Ballico, C. Bocci, E. Carlini, C. Fontanari, Osculating spaces to secant varieties, Rend. Circ. Mat. Palermo 53 (2004)
429–436.
[BF] E. Ballico, C. Fontanari, On the secant varieties to the osculating variety of a Veronese surface, Cent. Eur. J. Math. 1
(2003) 315–326.
[BC] A. Bernardi, M.V. Catalisano, Some defective secant varieties to osculating varieties of Veronese surfaces, Collect.
Math. 57 (1) (2006) 43–68.
[BCGI] A. Bernardi, M.V. Catalisano, A. Gimigliano, M. Idà, Osculating varieties of Veronesean and their higher secant varieties,
Canad. J. Math. 59 (2007) 488–502.
[BO] M.C. Brambilla, G. Ottaviani, On the Alexander–Hirschowitz theorem, J. Pure Appl. Algebra 211 (2008) 1229–1251.
[CGG1] M.V. Catalisano, A.V. Geramita, A. Gimigliano, On the secant varieties to the tangential varieties of a Veronesean, Proc.
Amer. Math. Soc. 130 (2001) 975–985.
[CGG2] M.V. Catalisano, A.V. Geramita, A. Gimigliano, Higher secant varieties of the Segre varieties P 1 × · · · × P 1 , J. Pure Appl.
Algebra 201 (2005) 367–380.
[Ch1] K. Chandler, A brief proof of a maximal rank theorem for generic double points in projective space, Trans. Amer. Math.
Soc. 353 (2001) 1907–1920.
[Ch2] K. Chandler, Linear systems of cubics singular at general points of projective space, Compos. Math. 134 (2002) 269–282.
[CO] CoCoATeam, CoCoA: A system for doing computations in commutative algebra, available at http://cocoa.dima.unige.it.
[Ge] A.V. Geramita, Inverse systems of fat points, in: The Curves Seminar at Queens, vol. X, in: Queen’s Papers in Pure and
Appl. Math., vol. 102, 1998.
[IK] A. Iarrobino, V. Kanev, Power Sums, Gorenstein Algebras, and Determinantal Loci, Lecture Notes in Math., vol. 1721,
Springer, Berlin, 1999.
[J] J.P. Jouanoulou, Théoremes de Bertini et applications, Progr. Math., vol. 42, Birkhäuser, Boston, MA, 1983.
Journal of Algebra
Rationality of the vertex algebra V L+ when L is

a non-degenerate even lattice of arbitrary rank
Gaywalee Yamskulna a,b,∗
a
Department of Mathematical Sciences, Illinois State University, Normal, IL 61790, USA
b
Institute of Science, Walailak University, Nakon Si Thammarat, Thailand
Article history: In this paper we prove that the vertex algebra V L+ is rational if L
Received 24 July 2008 is a negative definite even lattice of finite rank, or if L is a non-
Available online 12 November 2008 degenerate even lattice of a finite rank that is neither positive
Communicated by Dihua Jiang
definite nor negative definite. In particular, for such even lattices
L, we show that the Zhu algebras of the vertex algebras V L+ are
Keywords:
Vertex algebra semisimple. This extends the result of Abe from [T. Abe, Rationality
of the vertex operator algebra V L+ for a positive definite even
lattice L, Math. Z. 249 (2) (2005) 455–484] which establishes the
rationality of V L+ when L is a positive definite even lattice of finite
rank.
Published by Elsevier Inc.
1. Introduction
The vertex algebras V L+ are one of the most important classes of vertex algebras along with those
vertex algebras associated with lattices, affine Lie algebras and Virasoro algebras. They were originally
introduced in the Frenkel–Lepowsky–Meurman construction of the moonshine module vertex algebra
(see [FLM]). The study of the representation theory of V L+ began with the case when L is a positive
definite even lattice. In this case, the classification of all irreducible V L+ -modules, and the study of the
complete reducibility property of V L+ -modules was done by Abe, Dong, Jiang, and Nagatomo (see [A1,
A2,AD,DJ,DN1]). When L is a rank one negative definite even lattice, the classification of irreducible
V L+ -modules was completed by Jordan in [J]. Later, in [Y], the author classified all irreducible V L+ -
modules for the case when L is a negative definite even lattice of arbitrary rank, and when L is a
non-degenerate even lattice that is neither positive definite nor negative definite.
* Address for correspondence: Department of Mathematical Sciences, Illinois State University, Normal, IL 61790, USA.
E-mail address: gyamsku@ilstu.edu.
0021-8693/$ – see front matter Published by Elsevier Inc.

doi:10.1016/j.jalgebra.2008.10.009
1006 G. Yamskulna / Journal of Algebra 321 (2009) 1005–1015
In this paper, we continue our study of the representation theory of V L+ when L is a negative
definite even lattice of finite rank and when L is a non-degenerate even lattice that is neither positive
definite nor negative definite. We prove here that V L+ is rational in these cases. The main idea of the
proof is to show that the Zhu algebras A ( V L+ ) are semisimple. Note that a vertex algebra V is called
rational if any V -module is completely reducible.
This paper is organized as follows. In Section 2, we review the necessary background material.
In particular, we recall the construction of irreducible V L+ -modules and recall certain facts about the
Zhu algebras A ( V L+ ) when L is a negative definite even lattice of a finite rank and when L is a
non-degenerate even lattice that is neither positive definite nor negative definite. The proof of the
rationality of V L+ in these cases is contained in Section 3.
2. Preliminaries
First, we discuss relationships between vertex algebras and Zhu algebras. Next, we recall the con-
structions of irreducible V L+ -modules, and review the Zhu algebras A ( V L+ ) when L is a negative
definite even lattice of a finite rank and when L is a non-degenerate even lattice that is neither
positive definite nor negative definite.
2.1. Relationships between vertex algebras and Zhu algebras
Definition 2.1. (See [LLi].) A vertex algebra V is a vector space equipped with a linear map Y (·, z) : V →
(End V )[[ z, z−1 ]], v → Y ( v , z) = n∈Z v n z−n−1 and a distinguished vector 1 ∈ V which satisfies the
following properties:
1. un v = 0 for n 0.
2. Y (1, z) = id V .
3. Y ( v , z)1 ∈ V [[ z]] and limz→0 Y ( v , z)1 = v.
4. (The Jacobi identity)

z1 − z2 z2 − z1
z0−1 δ Y (u , z1 )Y ( v , z2 ) − z0−1 δ Y ( v , z 2 ) Y (u , z 1 )
z0 − z0

z1 − z0
= z2−1 δ Y Y (u , z 0 ) v , z 2 .
z2
We denote the vertex algebra just defined by ( V , Y , 1) or, briefly, by V .
Definition 2.2. A Z-graded vertex algebra is a vertex algebra

V = Vn; for v ∈ V n , n = wt v ,
n∈Z
equipped with a conformal vector ω ∈ V 2 which satisfies the following relations:

1
• [ L (m), L (n)] = (m − n) L (m + n) + 12
(m3 − m)δm+n,0 c V for m, n ∈ Z, where c V ∈ C (the central
charge) and

Y (ω, z) = L (n) z−n−2 = ωm z−m−1 ;
n∈Z m∈Z
• L (0) v = nv = (wt v ) v for n ∈ Z, and v ∈ V n ;

d
• Y ( L (−1) v , z) = dz Y ( v , z ).
G. Yamskulna / Journal of Algebra 321 (2009) 1005–1015 1007
For the rest of this subsection, we assume that V is a Z-graded vertex algebra.
Definition 2.3. (See [DLM].) A weak V -module M is a vector space equipped with a linear map
Y M (·, z) : V → (End M )[[ z]], v → Y M ( v , z) = n∈Z v n z−n−1 which satisfies the following properties:
for v , u ∈ V , and w ∈ M
1. v n w = 0 for n 0.
2. Y M (1, z) = id M .
3. (The Jacobi identity)

z1 − z2 z2 − z1
z0−1 δ Y M (u , z1 )Y M ( v , z2 ) − z0−1 δ Y M ( v , z 2 ) Y M (u , z 1 )
z0 − z0

z1 − z0
= z2−1 δ Y M Y (u , z 0 ) v , z 2 .
z2
Definition 2.4. An irreducible weak V -module is a weak V -module that has no weak V -submodule
except 0 and itself. Here, a weak submodule is defined in the obvious way.
Definition 2.5. (See [DLM].) An (ordinary) V -module is a weak V -module M which carries a C-
grading induced by the spectrum of L (0). Then M = λ∈C M λ where M λ = { w ∈ M | L (0) w = λ w },
and dim M λ < ∞. Moreover, for fixed λ, M n+λ = 0 for all small enough integers n.
Definition 2.6. (See [DLM].) An admissible V -module M is a Z0 -graded weak V -module M =

n∈Z0 M (n) such that v m M (n) ⊂ M (n + wt v − m − 1) for any homogeneous v ∈ V and m, n ∈ Z.
Here, Z0 is the set of non-negative integers.
An admissible V -submodule of M is a weak V -submodule N of M such that N = n∈Z0 N ∩
M (n).
Definition 2.7. An irreducible admissible V -module is an admissible V -module that has no admissible
submodule except 0 and itself.
Definition 2.8. A vertex algebra V is called a rational if every admissible V -module is completely
reducible, i.e., a direct sum of irreducible admissible V -modules.
Proposition 2.9. (See [DLM,Z].)
1. Any ordinary V -module is an admissible V -module.

For any irreducible admissible V -module M, there exists a complex number λ such that M =
2.
∞
n=0 M (λ + n) where M (λ + n) is the L (0)-eigenspace of the eigenvalue λ + n. We call λ the low-
est weight of M.
Next, we will define a Zhu algebra and we will discuss the relationships between the Zhu algebra
and a vertex algebra.
For a homogeneous vector u ∈ V , v ∈ V , we define products u ∗ v, and u ◦ v as follows:

(1 + z)wt u
u ∗ v = Resz Y (u , z ) v ,
z

(1 + z)wt u
u ◦ v = Resz Y (u , z ) v .
z2
Then we extend these products linearly on V . We let O ( V ) be the linear span of u ◦ v for all u , v ∈ V
and we set A ( V ) = V / O ( V ). Also, for v ∈ V , we denote v + O ( V ) by [ v ].
Theorem 2.10. (See [Z].)
1. ( A ( V ), ∗) is an associative algebra with the identity [1]. Moreover, [ω] is a central element of A ( V ).
2. The map M → M (0) gives a bijection between the set of equivalence classes of irreducible admissible
V -modules to the set of equivalence classes of simple A ( V )-modules.
We denote by P ( V ) the set of lowest weights of all irreducible admissible V -modules. The follow-
ing is a key proposition.
Proposition 2.11. (See [A1].) If the Zhu algebra A ( V ) is semisimple, and (λ + Z+ ) ∩ P ( V ) = ∅ for any λ ∈
P ( V ), then V is rational. Here, Z+ is the set of positive integers.
2.2. Vertex algebras V L+ , M (1)+ and their Zhu algebras
First, we will discuss the construction of the vertex algebras M (1)+ , V L+ and irreducible V L+ -
modules. Next, we will review some information about the Zhu algebras A ( M (1)+ ) and A ( V L+ ).
We will follow the setting in [FLM]. Let L be a non-degenerate even lattice of a rank d. We set
L̂ be the canonical central extension of L by the cyclic group κ of order 2. We let e : L → L̂ be
a section such that e 0 = 1 and we let : L × L → κ be the corresponding 2-cocycle. We assume
that is bimultiplicative. Then (α , β)/ (β, α ) = κ α ,β , (α , β) (α + β, γ ) = (β, γ ) (α , β + γ ),
and e α e β = (α , β)e α +β for α , β, γ ∈ L. Let θ denote an automorphism of L̂ defined by θ(e α ) = e −α
and θ(κ ) = κ . Furthermore, we set K = {a−1 θ(a) | a ∈ L̂ }.
We define V L = M (1) ⊗ C[ L ] where M (1) is the Heisenberg vertex operator algebra associated
with h = C ⊗Z L and C[ L ] is the group algebra of L with basis vectors e α for α ∈ L. Note that C[ L ] is
an L̂-module under the action e α e β = (α , β)e α +β . It was shown in [B,FLM,G] that V L is a Z-graded
simple vertex algebra. Moreover, M (1) is its Z-graded vertex subalgebra.
Next, we define a linear automorphism θ : V L → V L by

θ β1 (−n1 )β2 (−n2 ) . . . βk (−nk )e α = (−1)k β1 (−n1 ) . . . βk (−nk )e −α .
Consequently, θ Y ( v , z)u = Y (θ v , z)θ(u ) for u , v ∈ V L . In particular, θ is an automorphism of V L

and M (1).
For any stable θ -subspace U of V L , we denote a ±1 eigenspace of U for θ by U ± .
Proposition 2.12. (See [DM].) M (1)+ and V L+ are simple Z-graded vertex algebras.
Remark 2.13. V L+ is not an admissible module of itself when L is not a positive definite even lattice.
We let h[−1] and M (1)(θ) be the θ -twisted Heisenberg algebra and its unique irreducible module,
respectively. We set χ be a central character of L̂ / K such that χ (κ ) = −1 and we let T χ be the
Tχ T
irreducible L̂ / K -module with central character χ . We define V L = M (1)(θ) ⊗ T χ . Note that V L χ is
an irreducible θ -twisted V L -module (cf. [D,FLM]).
Tχ
We define an action of θ on M (1)(θ) and V L in the following way:

θ β1 (−n1 )β2 (−n2 ) . . . βk (−nk )1 = (−1)k β1 (−n1 ) . . . βk (−nk )1, and

θ β1 (−n1 )β2 (−n2 ) . . . βk (−nk )t = (−1)k β1 (−n1 ) . . . βk (−nk )t
T χ ,±
for βi ∈ h, ni ∈ 1
2
+ Z0 and t ∈ T χ . We denote by M (1)(θ)± and V L the ±1-eigenspace for θ of
Tχ
M (1)(θ) and V L , respectively.
Theorem 2.14. (See [Y].) Let L be an even lattice of a finite rank. Suppose that L is either a negative definite
lattice or a non-degenerate lattice that is neither positive definite nor negative definite. Then the set of all
irreducible admissible V L+ -modules is

T χ ,±
VL T χ is irreducible L̂ / K -module with central character χ such that χ (κ ) = −1 .
Corollary 2.15. Assume that L is a rank d non-degenerate even lattice. If L is either a negative definite lattice
or a non-degenerate lattice that is neither positive definite nor negative definite, then P ( V L+ ) = { 16
d d+8
, 16 }.
Next, we recall the Zhu algebras of M (1)+ and V L+ when L is a negative definite even lattice and
when L is a non-degenerate even lattice that is neither positive definite nor negative definite.
We let L be a rank d even lattice with a non-degenerate symmetric Z-bilinear form ·,· . We set
h = C ⊗Z L and extend ·,· to a C-bilinear form on h. Let {ha | 1 a d} be an orthonormal basis
of h, and set ωa = ωha = 12 ha (−1)2 1 and J a = ha (−1)4 1 − 2ha (−3)ha (−1)1 + 32 ha (−2)2 1. Note that
vectors ωa , and J a generate a vertex operator algebra M (1)+ associated to the one-dimensional vector
space Cha .
Following [AD], we set S ab (m, n) = ha (−m)hb (−n)1, and define E abu t
, E ab , and Λab as follows:
u
E ab = 5S ab (1, 2) + 25S ab (1, 3) + 36S ab (1, 4) + 16S ab (1, 5) (a = b),
u u u
E aa = E ab ∗ E ba ,
t

E ab = −16 3S ab (1, 2) + 14S ab (1, 3) + 19S ab (1, 4) + 8S ab (1, 5) (a = b),
t t
E aa = E ab ∗ E tba ,
Λab = 45S ab (1, 2) + 190S ab (1, 3) + 240S ab (1, 4) + 96S ab (1, 5).
Proposition 2.16. (See [DN2].) For any 1 i , j , k, l d we have [ E ti j ] ∗ [ E kl

t
] = δ jk [ E til ].
Let A u , and A t be the linear subspaces of A ( M (1)+ ) spanned by [ E ab

u
] and [ E ab
t
], respectively. Here,
1 a, b d.
Proposition 2.17. (See [DN2].)

d
1. The spaces A u and A t are two sided ideals of A ( M (1)+ ). Moreover, ideals A u , A t , the unit I u = i =1 [ E ii ]
u
d
of A and the unit I =
u t
i =1 [ E ii ]
t t
of A are independent of the choice of an orthonormal basis.
2. There are algebra isomorphisms between A u and End M (1)− (0), and between A t and End M (1)(θ)− ,
respectively. In particular, under the basis

h1 (−1)1, . . . , hd (−1)1
of M (1)− (0), each [ E ab

u
] corresponds to the matrix element E ab whose (a, b)-entry is 1 and zero elsewhere.
Similarly, under the basis

1 1
h1 − 1, . . . , h d − 1
2 2
of M (1)(θ)− (0), each [ E ab

t
] corresponds to the matrix element E ab whose (a, b)-entry is 1 and zero else-
where.
3. The Zhu algebra A ( M (1)+ ) is generated by [ωa ], [ J a ] for 1 a d, [Λab ] for 1 a = b d and [ E ab
u
],
[ E ab ] for 1 a, b d.
t
4. The quotient algebra A ( M (1)+ )/( A t + A u ) is commutative. Furthermore, it is generated by the images of
[ωa ], [ J a ] for 1 a d and [Λab ] for 1 a = b d.
For any α ∈ L, we set
V L+ [α ] = M (1)+ ⊗ E α ⊕ M (1)− ⊗ F α
and

A V L+ (α ) = V L+ [α ] + O V L+ / O V L+ .
Here, E α = e α + e −α , and F α = e α − e −α . Note that A ( V L+ ) is the sum of A ( V L+ )(α ) for all α ∈ L. For a
Z-graded vertex subalgebra U of V L+ , the identity map induces an algebra homomorphism from A (U )
to A ( V L+ ). For u ∈ U , we use [u ] to denote u + O (U ) and u + O ( V L+ ).
Next, we recall several results in [Y].
Proposition 2.18. (See [Y].) Let L be a negative definite even lattice of a finite rank d. In A ( V L+ ), we have the
following.
1. For any indices a, b, we have [ E ab u

] = 0 and [Λab ] = 0.
2. For α ∈ L − {0}, we let {h1 , . . . , hd } be an orthonormal basis of h such that h1 ∈ Cα . Then
(a) [ E ab
t
] ∗ [ E α ] = [ E α ] ∗ [ E ab
t
] if a = 1 and b = 1.
(b) [ E t1b ] ∗ [ E α ] = − 2α ,1α −1 [ E α ] ∗ [ E t1b ] if b = 1.
(c) [ E tb1 ] ∗ [ E α ] = −(2α , α − 1)[ E α ] ∗ [ E tb1 ] if b = 1.
(d) [ E aa
t
] ∗ [ E α ] = [ E α ] ∗ [ E aa
t
] for a ∈ {1, . . . , d}.
3. Let I t be the unit of the simple algebra A t . Then for any α ∈ L, I t ∗ [ E α ] = [ E α ] ∗ I t .
4. For any α ∈ L, we have

A V L+ (α ) = spanC [u ] ∗ E α u ∈ M (1)+

= spanC E α ∗ [u ] u ∈ M (1)+ .
5. Let α ∈ L − {0}. We set {h1 , . . . , hd } be an orthonormal basis of h such that h1 ∈ Cα . Then
9 t
9
[Ha] = [H 1] − E aa + E t11 for a ∈ {2, . . . , d}.
8 8
Proposition 2.19. (See [Y].) Let L be a non-degenerate rank d even lattice that is neither positive definite nor
negative definite. In A ( V L+ ), we have the following.
1. For any indices a, b, we have [ E ab u

] = 0 and [Λab ] = 0.
2. For α ∈ L such that α , α = 0, we let {h1 , . . . , hd } be an orthonormal basis of h such that h1 ∈ Cα . Then
(a) [ E ab
t
] ∗ [ E α ] = [ E α ] ∗ [ E ab
t
] if a = 1 and b = 1.
(b) [ E t1b ] ∗ [ E α ] = − 2α ,1α −1 [ E α ] ∗ [ E t1b ] if b = 1.
(c) [ E tb1 ] ∗ [ E α ] = −(2α , α − 1)[ E α ] ∗ [ E tb1 ] if b = 1.
(d) [ E aa
t
] ∗ [ E α ] = [ E α ] ∗ [ E aa
t
] for a ∈ {1, . . . , d}.
3. Let I be the unit of the simple algebra A t . Then for any α ∈ L such that α , α = 0, we have I t ∗ [ E α ] =
t
[Eα ] ∗ It .
4. Let α ∈ L such that α , α = 0. We set {h1 , . . . , hd } be an orthonormal basis of h such that h1 ∈ Cα . Then
9 t
9
[Ha] = [H 1] − E aa + E t11 for a ∈ {2, . . . , d}.
8 8
5. For any α ∈ L such that α , α =

0, we have

A V L+ (α ) = spanC [u ] ∗ E α u ∈ M (1)+

= spanC E α ∗ [u ] u ∈ M (1)+ .
6. Let α ∈ L − {0} such that α , α = 0. Then there exists β ∈ L such that β, β < 0, α , β < 0, and
A ( V L+ )(α ) ⊂ A ( V L+ )(α + 2β).
7. A ( V L+ ) is spanned by A ( V L+ )(0), and A ( V L+ )(α ) for all α ∈ L such that α , α =
0.
3. Rationality of the vertex algebra V L+ when L is a non-degenerate even lattice of arbitrary rank
Let L be a non-degenerate even lattice of a finite rank d. If L is positive definite, then it was
shown in [A1,A2] that V L+ is a rational vertex algebra. In this section, we will extend this result to
other cases.
3.1. Case I: L is a negative definite even lattice
In this subsection, we will prove that the vertex algebra V L+ is rational when L is a negative
definite even lattice. The key idea is to show that the Zhu algebra A ( V L+ ) is semisimple.
For the rest of this subsection, we assume that L is a negative definite even lattice. Following [DJ], we
set

α ,α −1
2α , α α
[ B̃ α ] = 2 I ∗ Eα −
t
E t11 ∗ E
2α , α − 1
for α ∈ L −{0}, and [ B̃ 0 ] = I t . Here, E t11 is defined with respect to an orthonormal basis {ha | 1 a d}
of h such that h1 ∈ Cα . Clearly, for i ∈ {1, . . . , d}, we have
1. [ E ti j ] ∗ [ B̃ α ] = 2α ,α −1 [ E ti j ] ∗ [ E α ] when j ∈ {2, . . . , d}, and

α ,α −1
2. [ E ti1 ] ∗ [ B̃ α ] = − 22α ,α −1 [ E ti1 ] ∗ [ E α ].
Lemma 3.1. For 1 a, b d, α ∈ L, [ E ab

t
] ∗ [ B̃ α ] = [ B̃ α ] ∗ [ E ab
t
].
Proof. This follows immediately from Proposition 2.18. 2
Next, we let A tL = SpanC {[ E ab

t
] ∗ [ B̃ α ] | 1 a, b d, α ∈ L }. We then have the following.
Theorem 3.2.
1. A tL is a 2-sided ideal of A ( V L+ ).
2. A tL is a semisimple associative algebra. In particular, A tL is isomorphic to A t ⊗C C[ L̂ / K ]/ J where C[ L̂ / K ]
is the group algebra of L̂ / K and J is the ideal of C[ L̂ / K ] generated by κ K + 1.
Proof. 1. It follows immediately from Propositions 2.17, 2.18 and Lemma 3.1. For 2., we will show that
A tL ∼
= A t ⊗ C[ L̂ / K ]/ J where J is the ideal of C[ L̂ / K ] generated by κ K + 1. Clearly, A tL is an A t -module
and A tL = A t · A tL . Moreover, A tL is a direct sum of M (1)(θ)− . By following the proof of Lemma 3.15
of [Y], we then have that [ B̃ α ] ∗ [ B̃ β ] = (α , β)[ B̃ α +β ] for α , β ∈ L. Notice that a linear map ψ : L̂ → A tL
defined by ψ(e α ) = [ B̃ α ] and ψ(κ ) = − I induced a linear map ψ̄ : L̂ / K → A tL since θ(e α ) = e −α and
t
B̃ α = B̃ −α . Moreover ψ induced an injective algebra homomorphism from C[ L̂ / K ]/ J into A tL and an

algebra isomorphism from A t ⊗ C[ L̂ / K ]/ J onto A tL . Consequently, A tL is semisimple. 2
Corollary 3.3. The set of all irreducible A tL -modules is

h(−1/2) ⊗ T χ T χ are irreducibles L̂ / K -modules with central character χ such that χ ι(κ ) = −1 .
Here, h(−1/2) = {h(−1/2) | h ∈ h}. Note that these h(−1/2) ⊗ T χ are also irreducible A ( V L+ )-modules.
Proof. It follows immediately from Theorem 3.2. 2
Next, we set

A V L+ = A V L+ / A tL .
T χ ,+
For a ∈ A ( V L+ ), we will conveniently denote its image in A ( V L+ ) by a. Clearly, V L
(0) is an A ( V L+ )-
module. Here, T χ is an irreducible L̂ / K -module with central character χ such that χ (ι(κ )) = −1.
Theorem 3.4. A ( V L+ ) is semisimple. In fact, it is isomorphic to C[ L̂ / K ]/ J where J is the ideal of C[ L̂ / K ]

generated by κ K + 1. Consequently,

T χ T χ are irreducibles L̂ / K -modules with central character χ such that χ ι(κ ) = −1
is the set of all irreducible A ( V L+ )-modules. Note that these T χ are also irreducible A ( V L+ )-modules.
Proof. Recall that there are identity maps from A ( M (1)+ ) into A ( V L+ ) and A ( V Z+α ) into A ( V L+ ) for
α ∈ L − {0}. By following the proof of Lemmas 3.8, 3.9 in [Y], we will obtain that every element
in A ( M (1)+ ) is a constant in A ( V L+ ). In particular, [ωa ] = 1
16
and [ H a ] = 9
128
in A ( V L+ ) for all a ∈
{1, . . . , d}. Moreover, for α ∈ L , u ∈ M (1)+ , [ E α ] commutes [u ] in A ( V L+ ), and A ( V L+ ) is a direct sum
of M (1)(θ)+ (0). Since A ( V L+ )(α ) = spanC {[u ] ∗ [ E α ] | u ∈ M (1)+ } for all α ∈ L, we can conclude that

α
A V L+ = spanC E + A tL α ∈ L .
α ∈ L − {0}, we set B α = 2α ,α −1 E α and B 0 = 1. By following the proof of Lemma 3.11 in [Y],

Next, for
we can show that in A ( V L+ ), [ B α ] ∗ [ B β ] = (α , β)[ B α +β ] for α , β ∈ L. Let φ : L̂ → A ( V L+ ) be a linear
map defined by φ(e α ) = [ B α ] and φ(κ ) = −1 + A tL . Then φ induces an algebra isomorphism from
C[ L̂ / K ]/ J onto A ( V L+ ). This implies that A ( V L+ ) is a semisimple algebra. 2
Corollary 3.5. Let M be an A ( V L+ )-module such that A t m = 0 for all m ∈ M. Then M is an A ( V L+ )-module.
Furthermore, M can be rewritten as a direct sum of irreducible A ( V L+ )-modules.
Proof. This follows immediately from Theorem 3.4. 2
Lemma 3.6. The Zhu algebra A ( V L+ ) is semisimple.
Proof. Let M be an A ( V L+ )-module. We will show that M can be rewritten as a direct sum of irre-
ducible A ( V L+ )-modules.
Case 1. A t m = 0 for all m ∈ M.

Then it follows immediately from Corollary 3.5 that M is a direct sum of irreducible A ( V L+ )-
modules.
Case 2. A t m = 0 for some m ∈ M.
First, we will show that M contains a simple A tL -module. Let m ∈ M such that A t m = 0. Clearly,
A tL m is an A tL -module. By Theorem 3.2, we can conclude that A tL m is a direct sum of irreducible A tL -
modules which are also irreducible A ( V L+ )-modules. Consequently, M contains a simple A tL -module.
Next, we will show that M is semisimple as an A ( V L+ )-module. We set N be the direct sum of all
irreducible A tL -submodules of M, and we set

M 0 = m ∈ M At m = 0 .
Note that N is a sum of irreducible A ( V L+ )-modules (cf. Corollary 3.3). Also, M 0 is an A ( V L+ )-

submodule of M since A t is a 2-sided ideal of A ( M (1)+ ) and for m ∈ M 0 , v ∈ A t ,
α
[v ] · E · m = [ v ] ∗ I t · E α · m = [ v ] ∗ E α · I t · m = 0 for all α ∈ L .
Let u ∈ M − N. We will show that u ∈ M 0 . If A t u = 0 then A tL u is direct sum of simple A tL -submodules

of M. This implies that A tL u ⊂ N and u ∈ N. This is impossible. Hence A t u = 0. Furthermore, u ∈ M 0 ,
and M − N ⊂ M 0 . Consequently, M = N + M 0 . By Corollary 3.5, we can conclude that M is a sum of
irreducible A ( V L+ )-modules.
Consequently, A ( V L+ ) is semisimple. 2
Corollary 3.7. Every A ( V L+ )-module is semisimple.
Theorem 3.8. If L is a negative definite even lattice of a finite rank then the vertex algebra V L+ is rational.
Proof. It is a consequence of Proposition 2.11, Corollary 2.15 and Lemma 3.6. 2
3.2. Case II: L is a non-degenerate even lattice that is neither positive definite nor negative definite
Following Section 3.1, we will prove that the vertex algebra V L+ is rational when L is a non-
degenerate even lattice that is neither positive nor negative definite by showing that the Zhu algebra
A ( V L+ ) is semisimple.
For the rest of this subsection, we assume that L is a non-degenerate even lattice that is neither positive
definite nor negative definite. Next, for every α ∈ L, we set [ B α ] in the following way:
1. If α ∈ L such that α , α =
0, we let

2α , α
B α = 2α ,α −1 I t ∗ E α − E t11 ∗ E α .
2α , α − 1
Here, E t11 is defined with respect to an orthonormal basis {ha | 1 a d} such that h1 ∈ Cα .
2. If α ∈ L − {0} such that α , α = 0, we define
1 1
B α = It ∗ Eα + γ , γ E t11 ∗ E α + β, β E t22 ∗ E α
2 1 − 2β, β − 2γ , γ
√
β, β γ , γ t α t α
+ E 12 ∗ E + E 21 ∗ E .
1 − 2β, β − 2γ , γ
Here, γ , β ∈ L Q such that α = γ + β , γ , γ > 0, β, β < 0, and γ , β = 0. Moreover, E t11 ,

E t22 , E t12 , E t21 are defined with respect to an orthonormal basis {ha | 1 a d} so that h1 ∈ Cγ ,
h2 ∈ Cβ .
3. We set [ B 0 ] = [ I t ].
Remark 3.9. Let α ∈ L such that α , α = −2k. It was shown in [J] that E 2α = (1 − 2k)28k+1 +
T χ ,−
k28k+6 [ωα ]. Hence [ B 2α ] = 1 on V L (0) for any χ .
Similar to Section 3.1, we have that for α ∈ L such that α , α =

0, and for i ∈ {1, . . . , d},
1. [ E ti j ] ∗ [ B α ] = 2α ,α −1 [ E ti j ] ∗ [ E α ] when j ∈ {2, . . . , d}, and

α ,α −1
2. [ E ti1 ] ∗ [ B α ] = − 22α ,α −1 [ E ti1 ] ∗ [ E α ].
Furthermore, we have the following.
Lemma 3.10. For 1 a, b d, α ∈ L such that α , α =

0, [ E ab
t
] ∗ [ B α ] = [ B α ] ∗ [ E ab
t
].
Next, we let

Ã tL = SpanC t
E ab ∗ B α 1 a, b d, α∈L .
Notice that Ã tL is an A t -module. Therefore, Ã tL is a direct sum of M (1)(θ)− (0), and for α ∈ L such that
α , α < 0, we have [ B 2α ] = 1 in Ã tL .
Lemma 3.11. Let α ∈ L − {0} such that α , α = 0. Then:
1. There exists β ∈ L so that β, β < 0, α , β < 0 and [ B α ] = [ B α +2β ].

2. For 1 a, b d, we have [ E ab
t
] ∗ [ B α ] = [ B α ] ∗ [ E ab
t
].
Proof. The proof of the first statement is very similar to Lemma 4.7 of [Y]. Let α ∈ L − {0} such
that α , α = 0. By Proposition 2.19, we can conclude that there exist β ∈ L, and u ∈ A t such that
T
β, β < 0, α , β < 0 and [ B α ] = [u ] ∗ [ B α +2β ]. Since [ B α ] = [ B α ] ∗ [ B 2β ] = [ B α +2β ] on V L χ ,− (0) for
any χ , we can conclude that [u ] = [ I t ] and [ B α ] = [ B α +2β ] in Ã tL .
2. It is a consequence of 1. 2
Theorem 3.12. Ã tL is a 2-sided ideal of A ( V L+ ) and Ã tL ∼

= A t ⊗C C[ L̂ / K ]/ J where C[ L̂ / K ] is the group algebra
of L̂ / K and J is the ideal of C[ L̂ / K ] generated by κ K + 1. Furthermore,

h(−1/2) ⊗ T χ T χ are irreducibles L̂ / K -modules with central character χ such that χ ι(κ ) = −1
is the set of all irreducible Ã tL -modules. Note that these h(−1/2) ⊗ T χ are also irreducible A ( V L+ )-modules.
Proof. By Proposition 2.19, Lemmas 3.10 and 3.11, one can easily show that Ã tL is a 2-sided ideal of
A ( V L+ ). Next, by following the proof of Theorem 3.2 step by step, we will obtain that Ã tL ∼
= A t ⊗C
C[ L̂ / K ]/ J . The rest of the theorem is clear. 2
Following Section 3.1, we set

Ã V L+ = A V L+ / Ã tL .
Theorem 3.13. Ã ( V L+ ) is semisimple. In fact, it is isomorphic to C[ L̂ / K ]/ J where J is the ideal of C[ L̂ / K ]

generated by κ K + 1. Consequently,

T χ T χ are irreducibles L̂ / K -modules with central character χ such that χ ι(κ ) = −1
is the set of all irreducible Ã ( V L+ )-modules. Note that these T χ are also irreducible A ( V L+ )-modules.
Proof. The proof is very similar to the proof of Theorem 3.4. 2
Corollary 3.14. Let M be an A ( V L+ )-module such that A t m = 0 for all m ∈ M. Then M is an Ã ( V L+ )-module.
Furthermore, M can be rewritten as a direct sum of irreducible A ( V L+ )-module.
Lemma 3.15. The Zhu algebra A ( V L+ ) is semisimple.
Proof. The proof is very similar to Lemma 3.6. 2
Theorem 3.16. If L is a non-degenerate even lattice of a finite rank that is neither positive definite nor negative
definite then the vertex algebra V L+ is rational.
Proof. It is a consequence of Proposition 2.11, Corollary 2.15 and Lemma 3.15. 2
References
[A1] T. Abe, The charge conjugation orbifold V Z+α is rational when α , α /2 is prime, Int. Math. Res. Not. 12 (2002) 647–665.
[A2] T. Abe, Rationality of the vertex operator algebra V L+ for a positive definite even lattice L, Math. Z. 249 (2) (2005)
455–484.
[AD] T. Abe, C. Dong, Classification of irreducible modules for the vertex operator algebra V L+ : General case, J. Algebra 273 (2)
(2004) 657–685.
[B] R. Borcherds, Vertex algebras, Kac–Moody algebras and the Monster, Proc. Natl. Acad. Sci. USA 83 (1986) 3068–3071.
[D] C. Dong, Twisted modules for vertex algebras associated with even lattices, J. Algebra 165 (1994) 91–112.
[DJ] C. Dong, C. Jiang, Rationality of vertex operator algebras, math.QA/0607679.
[DLM] C. Dong, H.-S. Li, G. Mason, Regularity of rational vertex operator algebras, Adv. Math. 132 (1) (1997) 148–166.
[DM] C. Dong, G. Mason, On quantum Galois theory, Duke Math. J. 86 (1997) 305–321.
[DN1] C. Dong, K. Nagatomo, Representations of vertex operator algebra V L+ for a rank one lattice L, Comm. Math. Phys. 202
(1999) 169–195.
[DN2] C. Dong, K. Nagatomo, Classification of irreducible modules for the vertex operator algebra M (1)+ II: Higher rank, J. Al-
gebra 240 (1) (2001) 289–325.
[FLM] I.B. Frenkel, J. Lepowsky, A. Meurman, Vertex Operator Algebras and the Monster, Pure Appl. Math., vol. 134, Academic
Press, Boston, 1988.
[G] H. Guo, On abelian intertwining algebras and modules, PhD Dissertation, Rutgers University, 1994.
[J] L. Jordan, Classification of irreducible V L+ -modules for a negative definite rank one even lattice L, PhD Dissertation,
University of California at Santa Cruz, 2006.
[LLi] J. Lepowsky, H.-S. Li, Introduction to Vertex Operator Algebras and Their Representations, Progr. Math., vol. 227,
Birkhäuser, Boston, 2003.
[Y] G. Yamskulna, Classification of irreducible modules of the vertex algebra V L+ when L is a nondegenerate even lattice of
an arbitrary rank, J. Algebra 320 (2008) 2455–2480.
[Z] Y. Zhu, Modular invariance of characters of vertexoperator algebras, J. Amer. Math. Soc. 9 (1996) 237–302.
Journal of Algebra
On the decomposition numbers of the Hecke algebra

of type D n when n is even
Jun Hu
Department of Applied Mathematics, School of Science, Beijing Institute of Technology, Beijing 100081, PR China
Article history: Let n 4 be an even integer. Let K be a field with char K = 2

n−1
i =1 (1 + q ) = 0. In
Received 7 August 2008 i
and q an invertible element in K such that
Available online 2 December 2008 this paper, we study the decomposition numbers over K of the
Communicated by Peter Littelmann
Iwahori–Hecke algebra Hq ( D n ) of type D n . We obtain some equal-
Keywords:
ities which relate its decomposition numbers with certain Schur el-
Iwahori–Hecke algebra ements and the decomposition numbers of various Iwahori–Hecke
q-Schur algebra algebras of type A with the same parameter q. When char K = 0,
Dual Specht modules this completely determine all of its decomposition numbers. The
main tools we used are the Morita equivalence theorem estab-
lished in [J. Hu, A Morita equivalence theorem for Hecke algebra
Hq ( D n ) when n is even, Manuscripta Math. 108 (2002) 409–430]
and certain twining character formulae of Weyl modules over a
tensor product of two q-Schur algebras.
1. Introduction
Let n be a natural number. Let K be a field and q, Q two invertible elements in K . Let W n be
the Weyl group of type A n−1 or of type B n . Let H( W n ) be the Iwahori–Hecke algebra of W n with
parameter q if W n = W ( A n−1 ); or with parameters q, Q if W n = W ( B n ). The modular representation
theory of H( W n ) has been well studied in the papers [3,7–9,13] and [32]. In fact, most of the results
of the modular representation theory of these algebras have been generalized to a more general class
of algebras—the cyclotomic Hecke algebras of type G (r , 1, n), where r ∈ N. The latter was now fairly
well understood by the work of [1,2,4,12] and [14].
This paper is concerned with the Iwahori–Hecke algebra Hq ( D n ) of type D n . By definition, Hq ( D n )
is the associative unital K -algebra with generators T u , T 1 , . . . , T n−1 subject to the following relations
E-mail address: junhu303@yahoo.com.cn.
doi:10.1016/j.jalgebra.2008.11.009
J. Hu / Journal of Algebra 321 (2009) 1016–1038 1017
( T u + 1)( T u − q) = 0,
( T i + 1)( T i − q) = 0, for 1 i n − 1,
T u T 2 T u = T 2 T u T 2, Tu T1 = T1Tu,
T i T i +1 T i = T i +1 T i T i +1 , for 1 i n − 2,
Ti T j = T j Ti, for 1 i < j − 1 n − 2,
Tu Ti = Ti Tu, for 2 < i < n.
The algebra Hq ( D n ) can be embedded into a Hecke algebra Hq ( B n ) of type B n with parameters {q, 1}
as a “normal” subalgebra. Namely, let Hq ( B n ) be the associative unital K -algebra with generators
T 0 , T 1 , . . . , T n−1 subject to the following relations
( T 0 + 1)( T 0 − 1) = 0,
T 0 T 1 T 0 T 1 = T 1 T 0 T 1 T 0,
( T i + 1)( T i − q) = 0, for 1 i n − 1,
Ti T j = T j Ti, for 0 i < j − 1 n − 2.
Then the map ι which sends T u to T 0 T 1 T 0 , and T i to T i (for each integer i with 1 i n − 1) can
be uniquely extended to an injection of K -algebras. Throughout this paper, we shall always identify
Hq ( D n ) with the subalgebra ι(Hq ( D n )) using this embedding ι.
Henceforth, we shall assume that the characteristic of the field K (denoted by char K ) is not equal 2. In
this case, if q is not a root of unity, then Hq ( D n ) is semisimple. Since we are only interested in the
modular (i.e., non-semisimple) case, we shall also assume that q is a root of unity in K . Let e be the
smallest positive integer such that 1 + q + q2 + · · · + qe−1 = 0. If q = 1, then e = char K . The modular
representation theory of Hq ( D n ) over K was studied in a number of papers [16,19,20,22,28,33]. The
algebra Hq ( D n ) is a special case of a more general class of algebra—the cyclotomic Hecke algebras of
type G (r , p , n). The latter was studied in [21,23–26] and [27]. The papers [19,20] and [33] studied the
restriction to Hq ( D n ) of simple Hq ( B n )-modules using the combinatorics of Kleshchev bipartitions;
while the papers [16] and [28] studied the simple Hq ( D n )-modules with the aim of constructing the
so-called “canonical basic set.” In both approaches, simple Hq ( D n )-modules have been classified but
using different parameterizations.
One of the major open problems in the modular representation theory of Hecke algebras is the
determination of their decomposition numbers. In the case of type A and type B (or more generally, of
type G (r , 1, n)), thanks to the work of [1] and [30], the decomposition numbers can be computed by
the evaluation at 1 of some alternating sum of certain parabolic affine Kazhdan–Lusztig polynomials
when char K = 0 and q = 1. It is naturaln−1to ask what will happen in the type D case. In [33], it was
proved that if n is odd and f n (q) := i =1 (1 + qi ) = 0 in K , then
Morita
n
Hq ( D n ) ∼ Hq (S(a,n−a) ),
a=(n+1)/2
where S(a,n−a) := S{1,...,a} × S{a+1,...,n} is the parabolic subgroup of the symmetric group Sn on
{1, 2, . . . , n}, Hq (S(a,n−a) ) is the parabolic subalgebra of Hq (Sn ) corresponding to S(a,n−a) . In this
case, if S λ and D μ denote the dual Specht module and simple module of Hq ( B n ) labelled by the
bipartition λ = (λ(1) , λ(2) ) and the e-restricted bipartition μ = (μ(1) , μ(2) ) respectively, then we have
the following equality of decomposition numbers:
1018 J. Hu / Journal of Algebra 321 (2009) 1016–1038

[ S λ ↓Hq ( D n ) : D μ ↓Hq ( D n ) ] = [ S λ(1) : D μ(1) ] S λ(2) :
D μ (2 ) ,
S λ(2) ,
where S λ(1) , D μ(1) (resp., D μ(2) ) denote the dual Specht module and simple module of Hq (Sa )
(resp., of Hq (S{a+1,...,n} )). Hence computing the decomposition numbers of Hq ( D n ) can be reduced
to computing the decomposition numbers of Hq (S(a,n−a) ) where (n + 1)/2 a n.
In [19], it was proved that if n is even and f n (q) = 0 in K , then
Morita
n
Hq ( D n ) ∼ A (n/2) ⊕ Hq (S(a,n−a) ), (1.1)
a=n/2+1
where A (n/2) is the K -subalgebra of Hq (Sn ) generated by Hq (S(n/2,n/2) ) and an invertible element
h(n/2) ∈ Hq (Sn ) (see [19, Definition 1.5] for definition of h(n/2)). Therefore, in this case, computing
the decomposition numbers of Hq ( D n ) can be reduced to computing the decomposition numbers of
Hq (S(a,n−a) ) where n/2 + 1 a n and the decomposition numbers of the algebra A (n/2).
The purpose of this article is to determine the decomposition numbers for the algebra A (n/2).
The main results of this paper provide some explicit formulae which determine these decomposition
numbers of Hq ( D n ) in terms of the decomposition numbers of Hq (Sn/2 ) and certain Schur elements,
see Theorem 2.5, Lemma 2.7 (where it is only assumed char K = 2), Theorem 2.8 (in the case where
char K = 0) and Theorem 2.9. Note that these results are also valid in the case where q = 1.
The paper is organized as follows. In Section 2 we shall first recall some known results about
the modular representation theory of the Hecke algebra of type D n when n is even in the separated
case. Then we shall state our main theorems. In Section 3 we lift the construction of the algebra
A (n/2) to the level of q-Schur algebras and introduce a certain covering A (n/2) of A (n/2). Using
Schur functor, we transfer the original problem of computing the decomposition numbers of A (n/2)
to the corresponding problem for A (n/2). In Section 4, we explicitly compute the Laurent polynomial
f λ ( v ) introduced in [19, Lemma 3.2] in terms of the Schur elements of the Hecke algebra Hq ( B n )
and Hq (Sn/2 ). In Section 5, by computing the twining character formula of some Weyl modules of a
tensor product of two q-Schur algebras, we determine our desired decomposition numbers when K is
of characteristic 0. When K is of odd characteristic, our results only give some equalities about these
decomposition numbers modulo char K .
2. Preliminaries
From now on until the end of this paper, we assume that n = 2m 4 is an even integer, and
−1
n

2 f n (q) := 2 1 + qi = 0. (2.1)
i =1
We shall refer (2.1) as separation condition and say that we are in the separated case.
Let k be a positive integer. A sequence of nonnegative integers λ = (λ1 , λ2 , . . .) is said to be a
composition of k if i 1 λi = k. A composition λ = (λ1 , λ2 , . . .) of k is said to be a partition of k
(denoted by λ k) if λ1 λ2 · · · . If λ is a composition of k, then we write |λ| = k. A bipartition of n
is an ordered pair λ = (λ(1) , λ(2) ) of partitions such that λ(1) is a partition of a and λ(2) is a partition
of n − a for some integer a with 0 a n. In this case, we say that λ is an a-bipartition of n and we
also write λ n.
A partition λ is said to be e-restricted if 0 λi − λi +1 < e for all i. We say that a bipartition
λ = (λ(1) , λ(2) ) is e-restricted if both λ(1) and λ(2) are e-restricted. Recall that for each bipartition
λ = (λ(1) , λ(2) ) of n, there is a dual Specht module1 S λ of Hq ( B n ). If Hq ( B n ) is semisimple, then the
1
Our S λ in this paper was denoted by
S λ in [19].
set { S λ |λ = (λ(1) , λ(2) ) n} forms a complete set of pairwise non-isomorphic simple Hq ( B n )-modules.
In general, if λ is e-restricted, then S λ has a unique simple Hq ( B n )-head D λ , and the set

D λ λ = λ(1) , λ(2) n is e-restricted
forms a complete set of pairwise non-isomorphic simple Hq ( B n )-modules.

Let v be an indeterminate over Z. Let A := Z[ v , v −1 ]. For each bipartition λ = (λ(1) , λ(2) )
of n, a Laurent polynomial f λ ( v ) ∈ A was introduced in [19, Lemma 3.2]. If λ(1) = λ(2) , then it
was proved in [19, Theorem 3.5] that there exist some Laurent polynomials g λ ( v ) ∈ A, such that
f λ ( v ) = ( g λ ( v ))2 . Since char K = 2, we actually have two choices for such g λ ( v ). Henceforth, for each
bipartition λ = (λ(1) , λ(2) ) of n satisfying λ(1) = λ(2) , we fix one choice of such g λ ( v ), and we denote it by

f λ ( v ). Then another choice would be − f λ ( v ). Once this is done, we can canonically define two
Hq ( D n )-submodules S λ+ and S λ− of S λ ↓Hq ( D n ) satisfying S λ ↓Hq ( D n ) = S λ+ ⊕ S λ− (see [21, Theorem 4.6]).
Let Pn denote the set of bipartitions of n. For any λ, μ ∈ Pn , we define λ ∼ μ if λ(1) = μ(2)
and λ(2) = μ(1) . Let τ be the K -algebra automorphism of Hq ( B n ) which is defined on generators by
τ ( T 1 ) = T 0 T 1 T 0 , τ ( T i ) = T i , for any i = 1. Let σ be the K -algebra automorphism of Hq ( B n ) which
is defined on generators by σ ( T 0 ) = − T 0 , σ ( T i ) = T i , for any i = 0. Clearly τ (Hq ( D n )) = Hq ( D n ) and
σ ↓Hq ( D n ) = Id. With our assumption (2.1) in mind, by [21, Corollary 3.7], for each λ = (λ(1) , λ(2) ) ∈ Pn ,

S λ ↓Hq ( D n ) ∼
= Sλ ↓Hq ( D n ) , where
λ := λ(2) , λ(1) .
Note that the inequality (2.1) is only a part of the conditions for the semisimplicity of Hq ( D n ). The
following result depends heavily on our assumption (2.1).
Lemma 2.2. (See [21].) If Hq ( D n ) is semisimple, then the set

+
S λ ↓Hq ( D n ) , S (β,β) −
, S (β,β) λ = λ(1) , λ(2) ∈ Pn /∼, λ(1) = λ(2) , β m
forms a complete set of pairwise non-isomorphic simple Hq ( D n )-modules. In general, if μ n is e-

restricted and μ(1) = μ(2) , then D μ ↓Hq ( D n ) remains irreducible and it is the unique simple Hq ( D n )-head
of S μ ↓Hq ( D n ) ; if α m is e-restricted, then S (+α ,α ) (resp., S (−α ,α ) ) has a unique simple Hq ( D n )-head D +
(α ,α )
(resp., D −
(α ,α ) ). The algebra Hq ( D n ) is split over K and the set
μ = (μ(1) , μ(2) ) ∈ Pn /∼ is e-restricted

D μ ↓Hq ( D n ) , D + −
(α ,α ) , D (α ,α )
μ(1) = μ(2) , α m is e-restricted
forms a complete set of pairwise non-isomorphic simple Hq ( D n )-modules.
Therefore, we can regard the following set

λ = (λ(1) , λ(2) ) ∈ P /∼, λ(1) = λ(2) , and
+ − n
S λ ↓Hq ( D n ) , S (β,β) , S (β,β)
β is a partition of m
as the set of dual Specht modules for Hq ( D n ) in the separated case.

We collect together some facts in the following lemma.
Lemma 2.3. (See [19,21].) Let β be a partition of m and α be an e-restricted partition of m. Let λ be a bipartition
of n and μ be an e-restricted bipartition of n. Then we have
(2.3.1) f λ (q) is an invertible element in K ;

(2.3.2) ( D μ )σ ∼= Dμ ∼ ↓H ( D ) ;
, D μ ↓Hq ( D n ) = D μ q n
(2.3.3) D (α ,α ) ↓Hq ( D n ) ∼
= D+ −
(α ,α ) ⊕ D (α ,α ) ;
+
(2.3.4) ( S (β,β) )τ ∼ −
= S (β,β) , (D+ τ ∼ −
(α ,α ) ) = D (α ,α ) ;
(2.3.5) M τ ∼
= M for any Hq ( B n )-module M.
Note that except for (2.3.5), all the claims in the above lemma depend on the validity of our
assumption (2.1).
Recall the Morita equivalence (1.1) proved in [19]. Let F be the resulting functor from the category
of
n finite dimensional Hq ( D n )-modules to the category of finite dimensional modules over A (m) ⊕
a=m+1 Hq (S(a,n−a) ). For any λ, μ ∈ Pn with μ being e-restricted, we define
S (λ) := S λ(1) ⊗
S λ(2) , D (μ) := D μ(1) ⊗
D μ (2 ) ,
if |λ(1) | =
|λ(2) |, |μ(1) | =
|μ(2) |; and
S (λ) := S λ(1) ⊗ D (μ) := D μ(1) ⊗

A (m) A (m)
S λ(2) ↑H (S , D μ(2) ↑H (S ,
q (m,m) ) q (m,m) )
if |λ(1) | = |λ(2) | = m = |μ(1) | = |μ(2) |.

For each integer i with 1 i m − 1, we define si = (i , i + 1). Then {s1 , s2 , . . . , sm−1 } is the set
of all the simple reflections in Sm . A word w = si 1 · · · sik for w ∈ Sm is a reduced expression if k is
minimal; in this case we say that w has length k and we write ( w ) = k. Given a reduced expression
si 1 · · · sik for w ∈ Sm , we write T w = T i 1 · · · T ik . It is well known that { T w | w ∈ Sm } forms a K -basis of
Hq (Sm ). Let β be a partition of m. Let Sβ be the Young subgroup of Sm corresponding to β . We set

xβ := Tw, y β := (−q)−( w ) T w .
w ∈Sβ w ∈Sβ
Let tβ (resp., tβ ) be the standard β -tableau in which the numbers 1, 2, . . . , m appear in order along
successive rows (resp., columns). Let w β ∈ Sm be such that tβ w β = tβ . If α is an e-restricted partition
of m, then we have the following direct sum decompositions of A (m)-modules:
S (β, β) = S (β, β)+ ⊕ S (β, β)− , D (α , α ) = D (α , α )+ ⊕ D (α , α )− ,
where
S (β, β)± := f (β,β) (q) zβ

⊗ zβ
± zβ
⊗ zβ
h(m) Hq (S(m,m) ),

D (α , α )± :=
⊗ z
± z
⊗ z
h (m) H (S
f (α ,α ) (q) zα α α α q (m,m) ),
and zβ
:= y β
T w β
xβ , β
is the conjugate partition of β , zβ
is the similarly defined generator for the
dual Specht module S β of Hq (S(m+1,...,n) ), zα

and z
are the canonical images of z
and z
α α α in D α

and D α , respectively. Note that by [8, Theorem 3.5] and [32, (5.2), (5.3)], the right ideal of Hq (Sm )
generated by zβ
is isomorphic to the dual Specht module S β , and the dual Specht module S β is also
isomorphic to the right ideal of Hq (Sm ) generated by x# # #
β
T w β
y β . Here “#” denotes the K -algebra
automorphism of Hq (Sn ) which is defined on generators by T i := (−q) T i−1 for each integer i with
#
1 i m − 1. We take this chance to point out that the definition of S (λ)± and D (μ)± given in [19,
p. 428, lines −11, −12] are not correct (although this does not affect any other results in [19]). The
correct definition should be as what we have given above. By direct verification, we see that
± ∼
= S (β, β)± , F D± ∼
F S (β,β) (α ,α ) = D (α , α )± ,
and for any λ, μ ∈ Pn with μ being e-restricted,
F ( S λ ↓Hq ( D n ) ) ∼
= S (λ), F ( D μ ↓Hq ( D n ) ) ∼
= D (μ).
The following lemma is a direct consequence of this Morita equivalence.
Lemma 2.4. Let λ be a bipartition of n and μ an e-restricted bipartition of n. Let β be a partition of m and α
an e-restricted partition of m. Then we have
(2.4.1) if |λ(1) | =
|λ(2) | and μ(1) = μ(2) , then

[ S λ(1) : D μ(1) ][ S λ(2) :
D μ(2) ], if |λ(1) | = |μ(1) | & |λ(2) | = |μ(2) |,
[ S λ ↓Hq ( D n ) : D μ ↓Hq ( D n ) ] =
0, otherwise;
+ −
S λ ↓Hq ( D n ) : D (α ,α ) = 0 = S λ ↓Hq ( D n ) : D (α ,α ) ;
(2.4.2) if |λ(1) | = |λ(2) | = m, λ(1) = λ(2) and |μ(1) | =

|μ(2) |, then
[ S λ ↓Hq ( D n ) : D μ ↓Hq ( D n ) ] = 0;
+ −
S (β,β) : D μ ↓Hq ( D n ) = 0 = S (β,β) : D μ ↓Hq ( D n ) .
Theorem 2.5. Let λ be a bipartition of n with |λ(1) | = |λ(2) | = m. Let μ be an e-restricted bipartition of n with
|μ(1) | = |μ(2) | = m. Let β be a partition of m and α an e-restricted partition of m. Then we have
(2.5.1) if λ(1) = λ(2) and μ(1) = μ(2) , then

[ S λ ↓Hq ( D n ) : D μ ↓Hq ( D n ) ] = [ S λ(1) : D μ(1) ] S λ(2) :
D μ(2) + [ S λ(1) : D μ(2) ] S λ(2) :
D μ (1 ) ,

S λ ↓Hq ( D n ) : D + −
(α ,α ) = S λ ↓Hq ( D n ) : D (α ,α ) = [ S λ(1) : D α ] S λ(2) : D α /2;
(2.5.2) if μ(1) = μ(2) , then

+
S (β,β) : D μ ↓Hq ( D n ) = [ S β : D μ(1) ]
Sβ :
D μ(2) /2 + [ S β : D μ(2) ]
Sβ :
D μ(1) /2,
− +
S (β,β) : D μ ↓Hq ( D n ) = S (β,β) : D μ ↓Hq ( D n ) .
Proof. Using the functor F , it is easy to see that the first equality in (2.5.1) follows from the exactness
A (m)
of the induction functor ↑ H (S ) . The other equalities follow from (2.3.4), (2.3.5) and the following
q (m,m)
Morita equivalence [9]:
Morita
n
Hq ( B n ) ∼ Hq (S(a,n−a) ). 2 (2.6)
a=0
It remains to determine the decomposition numbers
+
S (β,β) : D±
(α ,α ) ,
−
S (β,β) : D±
(α ,α ) .
Lemma 2.7. Let β be a partition of m and α an e-restricted partition of m. Then we have
+
(2.7.1) [ S (β,β) : D+ − −
(α ,α ) ] = [ S (β,β) : D (α ,α ) ];
+
(2.7.2) [ S (β,β) : D− − +
(α ,α ) ] = [ S (β,β) : D (α ,α ) ];
+ + + −
(2.7.3) [ S (β,β) : D (α ,α ) ] + [ S (β,β) : D (α ,α ) ] = [ S β : D α ]2 .
Proof. The first two equalities follow from (2.3.4), while the last equality follows from (2.6) and
(2.3.4) and the fact that
S (β,β) ↓Hq ( D n ) ∼ +
= S (β,β) −
⊕ S (β,β) , D (α ,α ) ↓Hq ( D n ) ∼
= D+ −
(α ,α ) ⊕ D (α ,α ) . 2
+
Therefore, it suffices to determine the decomposition number [ S (β,β) : D+
(α ,α ) ] for each partition
β of m and each e-restricted partition α of m. Note that these decomposition numbers are the 2-
splittable decomposition numbers in the sense of [26].
The purpose of this article is to give an explicit formula for these decomposition numbers. We shall
relate these decomposition numbers with the decomposition numbers of the Hecke algebra Hq (Sm )
and certain Schur elements of Hq ( B n ) and of Hq (Sm ). The main results of this paper are the follow-
ing two theorems:
Theorem 2.8. Let β be a partition of m and α an e-restricted partition of m. Let
dβ,α := [ S β : D α ].
Then we have the following equality in K :

+ f (β,β) (q)
S (β,β) : D+
(α ,α ) = dβ,α + dβ,α /2.
f (α ,α ) (q)
In particular, if char K = 0, then the above equality completely determines the decomposition number.
Note that a priori we do not even know why the right-hand side term in the above equality should
be an element in the prime subfield of K . Note also that we allow q = 1 in the above theorem. The
Laurent polynomials f (β,β) ( v ), f (α ,α ) ( v ) appeared in the above theorem can be computed explicitly
by the next theorem.
Theorem 2.9. Let λ = (λ(1) , λ(2) ) be an arbitrary bipartition of n. Then we have
n(n−1) sλ ( v , 1)
f λ(v ) = v 2 ,
sλ(1) ( v )sλ(2) ( v )
where sλ ( v , ṽ ) (resp., sλ(1) ( v ), sλ(2) ( v )) is the Schur element corresponding to λ (resp., corresponding to λ(1) ,
λ(2) ), and ṽ is another indeterminate over Z.
Note that these Schur elements are some explicit defined Laurent polynomials on v, ṽ. For exam-
ple, if λ is a partition, then

sλ ( v ) = v −( w λ
,0 ) hλi , j v
,
(i , j )∈[λ]
where hλi , j = λi + λ
j − i − j + 1 (the (i , j )th hook length), w λ
,0 is the unique longest element in Sλ
,
λ
denotes the conjugate partition of λ, and for each integer k,
vk − 1
[k] v := ∈ A.
v −1
We refer the reader to [31] for the explicit definitions of Schur elements corresponding to arbitrary
multi-partitions.
Example 2.10. Suppose that K = C, n = 6, β = (2, 1), α = (1, 1, 1). If e = 3, then α is e-restricted
and the assumption (2.1) is satisfied. In this case, let q be a primitive 3th root of unity in C. Then
1 + q + q2 = 0. It is well known that [ S β : D α ] = 1. Applying Theorem 2.9 and the known formulae for
Schur elements, we get that

2
f (β,β) ( v ) = v 4 ( v + 1)4 v 3 + 1 ,

2
2
f (α ,α ) ( v ) = ( v + 1)2 v 2 + 1 v3 + 1 .
Applying Theorem 2.8, we get that

+ 1 q2 (q + 1)2 (q3 + 1)
S (β,β) : D+
(α ,α ) = + 1 = 1.
2 (q + 1)(q2 + 1)(q3 + 1)
If e = 5, then α is also e-restricted and the assumption (2.1) is still satisfied. In that case,
[ S β : D α ] = 0, and hence
+
S (β,β) : D+
(α ,α ) = 0.
3. Lifting to q-Schur algebras
For each integer i ∈ {1, 2, . . . , n − 1} \ {m}, we define

i + m, if 1 i < m,

i=
i − m, if m < i n − 1.
Let be the group automorphism of S(m,m) := Sm × S{m+1,...,n} which is defined on generators by

si = si for each integer i ∈ {1, 2, . . . , n − 1} \ {m}. We use the same notation to denote the algebra
automorphism of the Hecke algebra Hq (S(m,m) ) = Hq (Sm ) ⊗ Hq (S{m+1,...,n} ) which is defined on
generators by Ti = Ti for each integer i ∈ {1, 2, . . . , n − 1} \ {m}.
For each partition λ of n, we set

xλ := Tw,
w ∈Sλ
where Sλ is the Young subgroup of Sn corresponding to λ. We use Λ(n) to denote the set of com-
positions λ = (λ1 , λ2 , . . . , λn ) of n into n parts (each part λi being nonnegative). Let Λ+ (n) be the set
of partitions in Λ(n). Following Dipper and James [10,11], we define the q-Schur algebra S q (n) to be

S q (n) := EndHq (Sn ) xλ Hq (Sn ) .
λ∈Λ(n)
In a similar way, we define the q-Schur algebras S q (m), S q (m) by using the Hecke algebras
Hq (Sm ), Hq (S{m+1,...,n} ), respectively. Note that to define the q-Schur algebra
S q (m), one needs to
use the element xλ for λ ∈ Λ(m).
For any λ, μ ∈ Λ(n), let Dλ,μ be the set of distinguished Sλ -Sμ -double coset representatives
in Sn . Following [11], for each d ∈ Dλ,μ , we define
φλ, μ ∈ HomHq (Sn ) xμ Hq (Sn ), xλ Hq (Sn )
by

d
φλ, μ (xμ h) = T w h, ∀h ∈ Hq (Sn ).
w ∈Sλ dSμ
By [11, Theorem 1.4], the elements in the set

d
φλ, μ λ, μ ∈ Λ(n), d ∈ Dλ,μ
form a K -basis of S q (n) which shall be called standard bases in this paper.
For any λ, μ ∈ Λ(n), by [11, (2.3)],
1 1 ∼
φλ,λ S q (n)φμ ,μ = HomHq (Sn ) xμ Hq (Sn ), xλ Hq (Sn ) .
Let ωn denote the partition (1n ) = (1, 1, . . . , 1) of n. Using the natural isomorphism

n copies
HomHq (Sn ) Hq (Sn ), Hq (Sn ) ∼

= Hq (Sn ),
we can identify Hq (Sn ) with the non-unital K -subalgebra φω 1 1

S (n)φω
n ,ωn q n ,ωn
of S q (n). We use ιn to
denote the resulting injection from Hq (Sn ) into S q (n). In a similar way, we can identify Hq (Sm ) with
the non-unital K -subalgebra φω 1 1
S (m)φω
m ,ωm q m ,ωm
via an injective map ιm , and Hq (S{m+1,...,n} ) with
the non-unital K -subalgebra φ ω1 ,ω
S (m)φ ω1 ,ω via an injective map ιm . Note that here in order not
m m q m m
to confuse with the standard basis elements of S q (m), we denote the standard basis element of S q (m)
by φd .
λ,μ
Let ρ denote the natural injective map from Hq (Sm ) ⊗ Hq (S{m+1,...,n} ) into Hq (Sn ). We are going
from S q (m) ⊗
to lift this map to an injection ρ S q (m) into S q (n). Let D(m,m) be the set of distinguished
right coset representatives of S(m,m) in Sn . Any element h ∈ Hq (Sn ) can be written uniquely as

h= A dw 1 , w 2 T w 1 T w 2 T d , (3.1)
w 1 ∈Sm , w 2 ∈S{m+1,...,n}
d∈D(m,m)
where A dw 1 , w 2 ∈ K for each w 1 , w 2 , d. For any f ∈ S q (m), g ∈ ( f ⊗ g ) ∈ S q (n) as

S q (m), we define ρ
follows: for any μ ∈ Λ(n), and any h ∈ Hq (Sn ) which is given by (3.1), if μ = (μ(1) , μ(2) ), where
μ(1) ∈ Λ(m), then we set

ρ( f ⊗ g )(xμ h) := A dw 1 , w 2 f (xμ(1) T w 1 ) g (xμ
(2) T w 2 ) T d ;
w 1 ∈Sm , w 2 ∈S{m+1,...,n}
d∈D(m,m)
( f ⊗ g )(xμ h) := 0.
otherwise, we set ρ
Lemma 3.2. With the notations as above, we have that
( f ⊗ g ) ∈ S q (n).
ρ
Proof. This is clear by using the fact that

Hq (Sn )
xμ(1) Hq (Sm ) ⊗
xμ(2) Hq (S{m+1,...,n} ) ↑
∼
= xμ Hq (Sn ). 2
Hq S(m,m)
Lemma 3.3. We have the following commutative diagram of algebra maps:
ρ
Hq (Sm ) ⊗ Hq (S{m+1,...,n} ) Hq (Sn )
ιm ⊗
ιm ιn
ρ
S q (m) ⊗
S q (m) S q (n)
Moreover, ρ ω1 ,ω ) = φω1 ,ω and ρ

(φω1 m ,ωm ⊗ φ is an injection.
m m n n
Proof. This follows from direct verification. 2
Note that the unit element of S q (m) ⊗

S q (m), i.e.,

1
φλ,λ ⊗ λ,λ
φ 1
λ∈Λ(m) λ∈Λ(m)
is not mapped by ρ to the unit element of S q (n). Henceforth, we identify S q (m) ⊗ S q (m) with a non-
unital K -subalgebra of S q (n) via the injection ρ . Recall that ([19, Remark 2.4]) the algebra A (m) was
generated by T 1 , . . . , T m−1 , T m+1 , . . . , T n−1 , h(m) and satisfy the following relations
( T i + 1)( T i − q) = 0, for 1 i n − 1, i = m, h(m)2 = zm,m ,
T i T i +1 T i = T i +1 T i T i +1 , for 1 i n − 2, i ∈
/ {m − 1, m}
Ti T j = T j Ti, for 1 i < j − 1 n − 2, i , j = m

h(m) T i +m if 1 i < m,
T i h(m) =
h(m) T i −m if m < i n − 1,
where zm,m is a central element in the Hecke algebra Hq (S(m,m) ) (see [9]). We are going to lift the
elements h(m), zm,m to the corresponding q-Schur algebras.
Lemma 3.4. Let z be an element in the center of Hq (Sn ). We define Z ∈ S q (n) by
Z (xλ h) = xλ hz, ∀λ ∈ Λ(n), h ∈ Hq (Sn ).
Then Z lies in the center of S q (n).

Proof. The condition that z is central in Hq (Sn ) implies that the above-defined map Z is indeed
a right Hq (Sn )-homomorphism. Hence Z ∈ S q (n). It remains to check that φλ, d
μ Z = Z φλ,μ for any
d
λ, μ ∈ Λ(n) and any d ∈ Dλ,μ . Applying [11, (2.1)], and by definition, for any h ∈ Hq (Sn ),
d

d
φλ, μ Z (xμ h) = φλ,μ (xμ hz) = T w hz,
w ∈Sλ dSμ

d
Z φλ, μ (xμ h) = Z T wh = T w hz.
w ∈Sλ dSμ w ∈Sλ dSμ
μ Z = Z φλ,μ , as required. 2
d d
This proves that φλ,
Definition 3.5. We define H (m), Z m,m ∈ S q (n) as follows: for any λ ∈ Λ(n), h ∈ Hq (Sn ),

λ h (m)h , if λ = (λ , λ ) with λ
x (1) (2) (1) ∈ Λ(m),
H (m)(xλ h) :=
0, otherwise,

xλ zm,m h, if λ = (λ(1) , λ(2) ) with λ(1) ∈ Λ(m),
Z m,m (xλ h) :=
0, otherwise.
Lemma 3.6. With the notations as above, we have that
H (m)2 = Z m,m ,
and Z m,m lies in the center of S q (m) ⊗

S q (m). Moreover Z m,m is invertible in S q (m) ⊗
S q (m).
Proof. The equality H (m)2 = Z m,m follows from direct verification and the fact that h(m)2 = zm,m .
Since zm,m lies in Hq (S(m,m) ) and zm,m is invertible in Hq (S(m,m) ) (see [9, (4.12)]), it follows that
Z m,m ∈ S q (m) ⊗
S q (m) and Z m,m is invertible in S q (m) ⊗
S q (m). The inverse of Z m,m is given by
−1 h , if λ = (λ(1) , λ(2) ) with λ(1) ∈ Λ(m),

−1 xλ zm
,m (xλ h ) :=
Zm ,m
0, otherwise,
for any λ ∈ Λ(n), h ∈ Hq (Sn ). The claim that Z m,m lies in the center of S q (m) ⊗
S q (m) also follows
from direct verification. 2
Note that Z m,m is invertible in S q (m) ⊗ S q (m) does not mean that Z m,m is invertible in S q (n). The
point is that the unit element of S q (n) is not the same as the unit element of S q (m) ⊗ S q (m). By the
same reason, we know that H (m) is not an invertible element in S q (n).
We identify Hq (S(m,m) ) with a subalgebra of S q (m) ⊗ S q (m) via the injection ιm ⊗ ιm . For any
λ, μ ∈ Λ(m), let Dλ,μ be the set of distinguished Sλ -Sμ -double coset representatives in Sm . For any
d ∈ Dλ,μ , it is easy to see that
d is a distinguished S λ -S
μ -double coset representative in S{m+1,...,n} .

Here Sλ , Sμ denote the image of Sλ , Sμ under the automorphism .
Lemma 3.7. The automorphism of Hq (S(m,m) ) can be uniquely extended to a K -algebra automorphism
(still denoted by ) of S q (m) ⊗
S q (m) such that for any λ, μ ∈ Λ(m) and any d ∈ Dλ,μ ,
d
λ,μ = φλ,μ .
d
φ
d d
μ = φλ,μ H (m) and φλ,μ H (m) = H (m)φλ,μ .
d d
Furthermore, H (m)φλ,
d d
Proof. For the first claim, it suffices to show that the map which sends φλ, μ to φλ,μ , for any
λ, μ ∈ Λ(m) and any d ∈ Dλ,μ , can be uniquely extends to a K -algebra map. To this end, it is enough
to show that for any λ, μ, ν ∈ Λ(m) and any d ∈ Dλ,μ , d
∈ Dμ,ν ,

d
λ, d
d d
φ μ φμ,ν = φλ,μ φμ,ν . (3.8)
μ φ μ, ν = ν , where A d
∈ K for each d . Then

d d
Suppose that φλ, d
∈Dλ,ν A d
φλ,

d

λ,
d
μ φ μ, ν =
φλ, d A d
φ ν.
d
∈Dλ,ν
By definition, it is easy to verify that for any h ∈ Hq (Sm ),

d
d d
μ xν λ, xμ

d
φ ,ν ( h ) = φμ ,ν (xν h ), φ μ ( h) = φλ, μ (xμ h).
Therefore,

d
A d
φ d

d d
d
λ, xν ) = φλ, d
d
φ μ φμ,ν ( μ φμ,ν (xν ) = λ,ν (xν ) = φλ,μ φμ,ν (

xν ).
d
∈Dλ,ν
This proves (3.8), as required. The proof of the last two equalities are straightforward and will be
omitted. 2
Definition 3.9. Let

A (m) denote the non-unital K -subalgebra of S q (n) generated by S q (m) ⊗
S q (m)
and H (m).
The algebra A (m) can be regarded as a covering of the algebra A (m). Note that the unit element
of
A (m) is the unit element of S q (m) ⊗
S q (m), which is different from the unit element of S q (n).
Note also that, by Lemma 3.6, although H (m) is not invertible in S q (n), H (m) is indeed an invertible
element in the algebra A (m).
Definition 3.10. For each partition λ ∈ Λ+ (m), we define
1
Z λ := y λ
T w λ
φλ, ωm .
By [11], the right ideal of S q (m) generated by Z λ is the Weyl module Δλ of S q (m) with highest
weight λ. In a similar way, we can define the element λ of
Z λ and the Weyl module Δ S q (m).
Lemma 3.11. Let λ, μ ∈ Λ+ (m). Then we have
zλ
⊗
z
zμ

m,m = f (λ,μ) (q ) zλ ⊗ zμ , ( Zλ ⊗
Z μ ) Z m,m = f (λ,μ) (q)( Z λ ⊗
Z μ ).
Proof. Since the dual Specht module S λ is also isomorphic to the right ideal of Hq (Sm ) generated by
x# # #
λ
T w
y λ , the first equality follows from [19, Lemma 3.2]. The second equality follows from the first
λ
equality and the definitions of Z m,m , Z λ and
Zλ. 2
To simplify notations, we shall denote H (m) by θ and S q (m) ⊗

m,m
S q (m) by S q . The set of subspaces
{ Sm is a Z/2Z-Clifford system in A (m) in the sense of [6, (11.12)]. For any λ, μ ∈ Λ+ (m),
,m
q , θ Sm
q
,m
}
we set
μ ,
Δλ,μ := Δλ ⊗ Δ L λ,μ := L λ ⊗
Lμ,
where Δλ , L λ (resp., Δ μ ,L μ ) are the Weyl module, irreducible module of the q-Schur algebra S q (m)
(resp.,
S q (m)), respectively. By definition, L λ (resp.,
L μ ) is the unique simple S q (m)-head (resp., S q (m)-
head) of Δλ (resp., of Δ μ ). Hence L λ,μ is the unique simple S m q
,m
-head of Δλ, μ .
m,m m,m
Note that by Lemma 3.7, the automorphism of S q is induced by θ . For any S q -module M,
m,m m,m
θ
let M be the new S q -module obtained by twisting the action of S q by . We have the following
result.
Lemma 3.12. For any λ, μ ∈ Λ+ (m), we have that
(Δλ,μ )θ ∼
= Δμ,λ , ( L λ,μ )θ ∼
= L μ,λ .
Proof. This follows directly from Lemma 3.7. 2
Definition 3.13. For any λ, μ ∈ Λ+ (m), we set
λ,μ := Δλ,μ ↑ Am(m,m) ,

Δ
A (m)
L λ,μ := L λ,μ ↑ m,m .
S q Sq
σ be the automorphism of
Let A (m) which is defined on generators by
θ j x → (−1) j θ j x, ∀x ∈ S m
q
,m
, j ∈ Z.
σ ↓ S mq ,m = Id. By Lemma 3.6, we can apply [17, (2.2)] and [24, Appendix]. That is, as
Clearly, A (m)-

A (m)-bimodule,

A (m) ∼
A (m) ⊗ S m,m =A (m) ⊕
σ
A (m) , (3.14)
q
A (m)-module structure on (
where the left A (m))
σ was just given by left multiplication, while the
A (m)-module structure on (
right A (m))
σ was given by right multiplication twisted by
σ.
Lemma 3.15. Let λ, μ ∈ Λ+ (m). We have that
L λ,μ ∼
(1) if λ = μ, then =L μ,λ is a simple
A (m)-module;
(2) there is a direct sum decomposition of L λ,λ =
A (m)-module: L+ −
λ,λ ⊕ L λ,λ , where

L+
λ,λ := f (λ,λ) (q) Z λ ⊗
Zλ + Zλ ⊗
Z λθ Sm
q
,m
,

L−

λ,λ := f (λ,λ) (q) Z λ ⊗
Zλ − Zλ ⊗
Z λθ Sm
q
,m
,
where Z λ (resp., Z λ ) is the natural image of Z λ (resp., of

Z λ ) in L λ (resp.,
L λ );

(3) A (m) is split over K , and the set

L λ,μ ,
L+ −
β,β , L β,β λ, μ, β m, λ = μ, (λ, μ) ∈ Pn /∼
forms a complete set of pairwise non-isomorphic absolutely simple

A (m)-modules;
σ ∼
(4) (L λ,μ ) = L λ,μ ∼
=L μ,λ , (
L+
β,β )σ ∼

= L −
β,β .
Proof. We only give the proof of (4), as the other claims follow from (4), (3.14) and Frobenius reci-
procity.
In fact, one can check that the following map gives the isomorphism ( σ ∼
L λ,μ ) = L λ,μ : for any
m,m
h1 , h2 ∈ S q ,
Zλ ⊗
Z μ h1 + Z λ ⊗
Z μ h2 θ → − f (λ,μ) (q) Z λ ⊗
Z μ h2 + Z λ ⊗
Z μ h1 θ;
while the following map gives the isomorphism (

L+ σ ∼ −
m,m
λ,λ ) = L λ,λ : for any h ∈ S q ,

f (λ,λ) (q) Z λ ⊗
Zλ + Zλ ⊗
Z λ θ h → f (λ,λ) (q) Z λ ⊗
Zλ − Zλ ⊗
Z λ θ h. 2
There is also a direct sum decomposition of + ⊕ Δ

λ,λ = Δ
A (m)-module: Δ λ,λ
− , where
λ,λ
+ :=
Δ f (λ,λ) (q) Z λ ⊗
Zλ + Zλ ⊗
Z λθ Sm
q
,m
,
λ,λ

− :=
Δ f (λ,λ) (q) Z λ ⊗
Zλ − Zλ ⊗
Z λθ Sm
q
,m
.
λ,λ
Lemma 3.16. With the same notations as above, we have
(1) if λ = μ, then L λ,μ is the unique simple λ,μ ;

A (m)-head of Δ
+ − + − );
(2) L λ,λ (resp., L λ,λ ) is the unique simple A (m)-head of Δλ,λ (resp., of Δλ,λ
σ ∼Δ
λ,μ )
(3) (Δ = λ,μ ∼
=Δ + )
μ,λ , (Δ
λ,λ
σ ∼Δ
= −
λ,λ .
Proof. These follow from Frobenius reciprocity. 2

Corollary 3.17. Let λ be a partition in Λ+ (m). Let θ acts as the scalar f (λ,λ) (q) (resp., − f (λ,λ) (q)) on the
highest weight vector of Δλ,λ . Then this action can be uniquely extends to a representation of A (m) on Δλ,λ .
The resulting
A (m)-module is isomorphic to Δ+ λ,λ (resp., Δ−
λ,λ ). The same statements hold for L + −
λ,λ and L λ,λ .
Note that the set {φλ,λ1

}λ∈Λ(m) is a set of pairwise orthogonal idempotents in S q (m), and

φ 1
λ∈Λ(m) λ,λ = 1. For any right S q (m)-module M, it is clear that

1
M= M φλ,λ .
λ∈Λ(m)
For each λ ∈ Λ(m), we define M λ := M φλ,λ 1

, and we call M λ the λ-weight space of S q (m)-module
M. Note also that S q (m) is an epimorphic image of the quantum algebra U K (glm ) associated to glm
(cf. [5,15]). Any S q (m)-module M naturally becomes a module over U K (glm ). The definition of weight
space we used here coincides with the usual definition of weight space for U K (glm )-module. In a
similar way, we can define the weight space for any S q (m)-module. Therefore, we have also the
m,m m,m
notion of weight space for any S q -module. The weights of any S q -module are elements in the
set Λ(m) := Λ(m) × Λ(m).
There is a natural additive group structure on Λ(m). Let {e λ }λ∈Λ(m) denote the standard basis of
the group ring Z[Λ(m)] over Z. Then e λ e μ = e λ+μ . For any finite dimensional S q -module M, we
m,m
define the formal character of M as

ch M = dim M λ e λ ∈ Z Λ(m) .
λ∈Λ(m)
For any short exact sequence 0 → M

→ M → M
→ 0 of finite dimensional S q
m,m
-modules, it is clear
that
ch M = ch M
+ ch M
m,m
Therefore, the map ch is a map defined on the Grothendieck group R( S q ) associated to the category
m,m
of finite dimensional S q -modules.
Set e = φω 1
⊗φ ω1 ,ω . Then e is an idempotent in A (m), and e
m,m
A (m)e = A (m), e S q e =
m ,ωm m m

Hq (S(m,m) ). We define a functor F from the category of finite dimensional A (m)-modules to the cat-
egory of finite dimensional A (m)-modules as follows: for any finite dimensional A (m)-module M , N,
and any ϕ ∈ Hom
A (m) ( M , N ), F ( M ) = Me, F ( N ) = Ne, and

F (ϕ )(xe ) := ϕ (x)e , ∀x ∈ M .
m,m
Let F be the Schur functor (induced by e) from the category of finite dimensional S q -modules to
the category of finite dimensional Hq (S(m,m) )-modules. Then we have the following commutative
diagram of functors:
Res
Mod-
m,m
A (m) Mod-S q

F F (3.18)
Res
Mod- A (m) Mod-Hq (S(m,m) )
We set Λ+ (m) := Λ+ (m) × Λ+ (m).
Lemma 3.19. Let (λ, μ) ∈ Λ+ (m) such that λ = μ. Then we have
λ,μ ) = S (λ, μ),

F Δ λ,λ = S (λ, λ)± , F (Δ

±
D (λ, λ)± , if λ is e-restricted;

F L λ,λ =
0, otherwise,

D (λ, μ), if (λ, μ) is e-restricted;

F ( L λ,μ ) =
0, otherwise.
Proof. These follow from (3.18) and direct verification. 2
Lemma 3.20. F is an exact functor. In particular,

F induces a homomorphism from the Grothendieck group
R(A (m)) associated to the category of finite dimensional A (m)-modules to the Grothendieck group R( A (m))
associated to the category of finite dimensional A (m)-modules.
Proof. This follows from the same arguments as in [18, Section 6]. 2
Corollary 3.21. For any λ, μ ∈ Λ+ (m) with μ being e-restricted, we have the following equality of decompo-
sition numbers:
+ +
: + +
Δλ,λ L μ,μ = S (λ, λ)+ : D (μ, μ)+ = S (λ,λ) : D (μ,μ) .
Proof. This follows directly from Lemmas 3.19 and 3.20. 2
+
Therefore, computing the decomposition number [ S (λ,λ) : D+
(μ,μ) ] can be reduced to computing the
decomposition number [Δ :
+
L + ]. The latter will be done in the final section.
μ, μ
λ,λ
4. Computing the Laurent polynomial f λ ( v )
Let a be a fixed integer with 0 a n and λ = (λ(1) , λ(2) ) be a fixed a-bipartition of n. The purpose
of this section is to give a close formula for the Laurent polynomial f λ ( v ), which was introduced in
[19, Lemma 3.2].
Recall that by [19, Lemma 3.2], certain central element (denoted by za,n−a in [19, Lemma 3.2]) of
H v (S(a,n−a) ) acts on the dual Specht module S λ as the scalar f λ ( v ). These central elements za,n−a
arise in the study of certain homomorphism between some (dual) Specht modules over the Hecke
algebra H v ( B n ). Indeed, similar central elements do arise if we consider the more general type B n
Hecke algebra H v , ṽ ( B n ) which has two parameters v , ṽ, where ṽ is another indeterminate over Z. By
some abuse of notations, we will denote the resulting Laurent polynomial by f λ ( v , ṽ ). The relation
with our previous introduced Laurent polynomial f λ ( v ) is given by
f λ ( v ) = f λ ( v , 1).
In this section, we shall give a closed formula for the Laurent polynomial f λ ( v ,
v ).
We fix some notations. Let v , ṽ be two indeterminates over Z. Let A := Z[ v , v −1 , ṽ , ṽ −1 ]. The
Hecke algebra H v , ṽ ( B n ) of type B n over A is the associative unital A
-algebra with generators
T 0 , T 1 , . . . , T n−1 subject to the following relations
( T 0 + 1)( T 0 − ṽ ) = 0,
T 0 T 1 T 0 T 1 = T 1 T 0 T 1 T 0,
( T i + 1)( T i − v ) = 0, for 1 i n − 1,
Ti T j = T j Ti, for 0 i < j − 1 n − 2.
Let tr be the trace form on the Hecke algebras H v , ṽ ( B n ) and H v (Sn ) as defined in [31, p. 697,
line 5]. Note that the trace form is denoted by τ in [31, p. 697, line −5].
Definition 4.1. (See [13, (2.1)], [8, (3.8)].) For any nonnegative integers k, a, b, we set

1, if k = 0,
uk+ = k i −1
i =1 ( v + T i −1 · · · T 1 T 0 T 1 · · · T i −1 ), if 1 k n,

1, if k = 0,
uk− = k i −1
i =1 ( ṽ v − T i −1 · · · T 1 T 0 T 1 · · · T i −1 ), if 1 k n,
ha,b = T w a,b
where

s ···s s ···s ···s 1 · · · sb , if a, b are positive integers,
w a,b =
a 1 a+1 2 a+b−
1, if a or b is zero.
By some abuse of notations, for each integer i ∈ {1, 2, . . . , n − 1} \ {a}, we set

i + n − a, if 1 i < a,

i=
i − a, if a < i n − 1.
Let be the group isomorphism from the Young subgroup S(a,n−a) onto the Young subgroup S(n−a,a)
which is defined on generators by si = si for each integer i ∈ {1, 2, . . . , n − 1} \ {a}. We use the
same notation to denote the algebra isomorphism from the Hecke algebra H v (S(a,n−a) ) onto the
Hecke algebra H v (S(n−a,a) ) which is defined on generators by Ti = Ti for each integer i ∈ {1, 2, . . . ,
n − 1} \ {a}. By abuse of notation, the inverse of will also be denoted by . By [9, (3.23)], we know
that there exists an element za,n−a in the center of H v (S(a,n−a) ) ⊗A A such that
un−−a hn−a,a ua+ ha,n−a un−−a hn−a,a ua+ = un−−a hn−a,a ua+ za,n−a . (4.2)
Let ∗ be the anti-automorphism of H v , ṽ ( B n ) which is defined on generators by T i∗ = T i for each

0 i n − 1.
Lemma 4.3. For any integer a with 1 a n, we have that
(4.3.1) ua+ ha,n−a un−−a hn−a,a ua+ ha,n−a un−−a = ua+ ha,n−a un−−a
za,n−a ,
(4.3.2) ( za,n−a )∗ = za,n−a .
Proof. Using the same argument as [9, (3.23)], we know that there exists an element zn
−a,a in the
such that
center of Hq (S(n−a,a) ) ⊗A A
ua+ ha,n−a un−−a hn−a,a ua+ ha,n−a un−−a = ua+ ha,n−a un−−a zn
−a,a .
Note that
un−−a hn−a,a ua+ ha,n−a un−−a hn−a,a ua+ ha,n−a un−−a hn−a,a ua+
= un−−a hn−a,a ua+ ha,n−a un−−a hn−a,a ua+ za,n−a = un−−a hn−a,a ua+ za2,n−a .
On the other hand, we have that
un−−a hn−a,a ua+ ha,n−a un−−a hn−a,a ua+ ha,n−a un−−a hn−a,a ua+
= un−−a hn−a,a ua+ ha,n−a un−−a zn

−a,a hn−a,a ua+ = un−−a hn−a,a ua+ ha,n−a un−−a zn
−a,a hn−a,a ua+
z
n−a,a
= un−−a hn−a,a ua+ za,n−a

zn
−a,a .
zn
−a,a . Note that (see [9]) za,n−a is invertible in H v , ṽ (Sn ) ⊗A
It follows that za2,n−a = za,n−a Q( v , ṽ ).
Hence we conclude that zn
−a,a = za,n−a , as required. This proves 1).
Applying the anti-automorphism ∗ to both sides of (4.2), we get that
ua+ ha,n−a un−−a

za,n−a = ua+ ha,n−a un−−a hn−a,a ua+ ha,n−a un−−a
= za∗,n−a ua+ ha,n−a un−−a = ua+ ha,n−a un−−a z

∗
a,n−a ,
which implies that ( za,n−a )∗ = za,n−a , as required. 2

For any a-bipartition λ = (λ(1) , λ(2) ) of n, we define zλ = zλ(1) z

λ(2) , where zλ(1) = xλ(1) T w λ(1) y λ(1)
,

( ) (2 )
zλ(2) = xλ(2) T w (2) y λ(2)
. We set λ := (λ , λ ). Recall that λ = (λ(2) , λ(1) ), which is an (n − a)-
1
λ

bipartition of n. We define the Specht modules S λ and the twisted Specht module
S λ to be:

S λ := un−−a hn−a,a ua+ zλ H v , ṽ ( B n ),
S λ := ua+ ha,n−a un−−a z
λ H v , ṽ ( B n ).

Let θλ (resp., δλ ) be the right H v , ṽ ( B n )-module homomorphism from S λ to S λ (resp., from
S λ to S λ )
+ −
given by left multiplication with ua ha,n−a (resp., with un−a hn−a,a ). It is clear that both θλ and δλ are
well defined right H v , ṽ ( B n )-module homomorphisms.
Lemma 4.4. For any a-bipartition λ of n, there exists a Laurent polynomial f λ ( v , ṽ ) ∈ A such that za,n−a zλ =
f λ ( v , ṽ ) zλ ,
zn−a,a z
λ = f λ ( v , ṽ ) zλ . In particular, both θ δ
λ λ and δ θ
λ λ are scalar multiplication by f λ ( v , ṽ ).
Proof. The first part is an easy consequence of [7, (4.1)], by using the same argument as in the proof
of [19, (3.2)]. The second part follows from Lemma 4.3. 2
Let μ be a partition of a. Let zμ := xμ T w μ y μ

. Then S μ := zμ H v (Sa ) is the Specht module
of H v (Sa ) corresponding to μ. By [8, (3.5)] and [32, (5.2), (5.3)], the dual Specht module S μ
of H v (Sa ) is isomorphic to the right ideal generated by y μ
T w μ
xμ . Let θμ (resp., δμ ) be the
right H v (Sa )-module homomorphism from S μ to y μ
T w μ
xμ H v (Sa ) (resp., from y μ
T w μ
xμ H v (Sa )
to S μ ) given by left multiplication with y μ
T w μ
(resp., with xμ T w μ ). The following result is coming
from [29, (5.8)]. But we shall give a different proof here in order to illustrate the technique which will
be used in the proof of the main theorem in this section.
Lemma 4.5. (See [29, (5.8)].) Let μ be a partition of a. Then both θμ δμ and δμ θμ are scale multiplication of
v ( w μ ) sμ ( v ), where sμ ( v ) ∈ A is the Schur element associated to μ. In particular,
zμ T w μ
zμ = v ( w μ ) sμ ( v ) zμ .
Proof. By [7, (4.1)], we have that
xμ ( T w μ y μ
T w μ
xμ T w μ ) y μ
= rμ ( v )xμ T w μ y μ
,
for some rμ ( v ) ∈ A. Therefore, for any h ∈ H v (Sa ),
δμ θμ (zμ h) = xμ T w μ y μ
T w μ
zμ h = zμ T w μ
zμ h
= xμ ( T w μ y μ
T w μ
xμ T w μ ) y μ
h = rμ ( v )xμ T w μ y μ
h = rμ ( v )zμ h,
We claim that rμ ( v ) = 0. In fact, by [31, (5.9)],
tr δμ ( y μ
T w μ
xμ ) = tr(xμ T w μ y μ
T w μ
xμ ) = tr x2μ T w μ y μ
T w μ

= v ( w ) tr(xμ T w μ y μ
T w μ
) = v ( w μ ) v ( w ) = 0.
w ∈Sμ w ∈Sμ
It follows that δμ ( y μ
T w μ
xμ ) = 0, hence (δμ )Q( v ) is an H v (Sa ) ⊗A Q( v )-module isomorphism be-
tween two simple (H v (Sa ) ⊗A Q( v ))-modules. By similar argument, one can show that (θμ )Q( v )
is an (H v (Sa ) ⊗A Q( v ))-module isomorphism between two simple (H v (Sa ) ⊗A Q( v ))-modules. In

particular, this implies that rμ ( v ) = 0, as required.
It remains to show rμ ( v ) = v ( w μ ) sμ ( v ). Since rμ ( v ) = 0, rμ ( v )−1 zμ T w μ
is an idempotent in
H v (Sa ) ⊗A Q( v ). Clearly

S μ ⊗A Q( v ) = rμ ( v )−1 zμ T w μ
H v (Sa ) ⊗A Q( v ) .
Applying [31, (1.6),(5.9)], we get that
1 rμ ( v )
sμ ( v ) = = = v −( w μ ) rμ ( v ).
tr(rμ ( v )−1 zμ T w μ
) tr( zμ T w μ
)
Clearly δμ θμ = v ( w μ ) sμ ( v ) implies that θμ δμ = v ( w μ ) sμ ( v ). This completes the proof of the theo-

rem. 2
Theorem 4.6. With the above notations, we have that
n(n−1) sλ ( v , ṽ )
f λ ( v , ṽ ) = v 2 ṽ n−a ,
∈A
sλ(1) ( v )sλ(2) ( v )
where sλ ( v , ṽ ) (resp., sλ(1) ( v ), sλ(2) ( v )) is the Schur element associated to the bipartition λ (resp., the partitions
λ(1) , λ(2) ). In particular, in the ring A ,
sλ(1) ( v )sλ(2) ( v )|sλ ( v , ṽ ).
Proof. For an a-bipartition λ = (λ(1) , λ(2) ), we define T w λ = T w (1) T

w λ(2) . Then T w
λ
= T w λ(2) T
w λ(1) .
λ
Now applying Lemma 4.5, we get that
z λ = zλ(2) T w
λ T w
z
zλ(2) z λ(1)
λ(1) T w (1)
z
λ λ(2) λ

w λ(2) + w λ(1)
=v sλ(2) ( v )sλ(1) ( v ) zλ(2) z
λ(1)

w λ(2) + w λ(1)
=v sλ(2) ( v )sλ(1) ( v ) z
λ,
where sλ(1) ( v ), sλ(2) ( v ) are the Schur elements associated to the partitions λ(1) , λ(2) , respectively.
Let A ( v , ṽ ) := v −( w λ(2) )−( w λ(1) ) f λ ( v , ṽ )−1 sλ(2) ( v )−1 sλ(1) ( v )−1 . Then we have

2
A ( v , ṽ )un−−a hn−a,a ua+ zλ ha,n−a T w
λ

2
= A ( v , ṽ )un−−a hn−a,a ua+ ha,n−a zλ T w λ
= A ( v , ṽ )2 un−−a hn−a,a ua+ ha,n−a zλ T w λ

un−−a hn−a,a ua+ ha,n−a zλ T w λ
= A ( v , ṽ )2 un−−a hn−a,a ua+ ha,n−a un−−a hn−a,a ua+ ha,n−a zλ T w λ

zλ T w λ
= A ( v , ṽ ) f λ ( v , ṽ )−1 un−−a hn−a,a ua+ ha,n−a un−−a hn−a,a ua+ ha,n−a zλ T w λ
= A ( v , ṽ ) f λ ( v , ṽ )−1 un−−a hn−a,a ua+ za,n−a ha,n−a zλ T w λ
= A ( v , ṽ ) f λ ( v , ṽ )−1 un−−a hn−a,a ua+ za,n−a zλ ha,n−a T w λ
= A ( v , ṽ )un−−a hn−a,a ua+ zλ ha,n−a T w λ

.
In other words, A ( v , ṽ )un−−a hn−a,a ua+ zλ ha,n−a T w

is an idempotent of the Hecke algebra H v , ṽ ( B n ) ⊗A

λ
Q( v , ṽ ). Since
− +

Q( v , ṽ ) = A ( v , ṽ )un−a hn−a,a u a zλ ha,n−a T w

H v , ṽ ( B n ) ⊗A
S λ ⊗A
λ
Q( v , ṽ ) ,
it follows from [31, (1.6)] that
1
sλ ( v , ṽ ) = ,
tr( A ( v , ṽ )un−−a hn−a,a ua+ zλ ha,n−a T w
)
λ
where sλ ( v , ṽ ) is the Schur element associated to the bipartition λ. On the other hand, if we set
q = v , Q 1 = −1, Q 2 = ṽ, and μ = (λ(2) , λ(1) ), then the element mμ in [31] is
n−a

xλ(2) x
λ(1) −v 1−k
un−−a
k=1
in our notation, and the element nμ

in [31] is

a

y λ(1)
y λ(2)

v 1−k )ua+
k=1
in our notation. Note also that the element w μ

in [31, (5.9)] is just
w a,n−a w λ(2)

w λ(1)
in our notation. Therefore, the element zμ T w μ

in the notation of [31, (5.9)] is just
n−a

a
−v 1−k
v 1−k
un−−a hn−a,a ua+ zλ ha,n−a T w
λ
k=1 k=1
in our notations. Applying [31, (5.9)], we have

n−a

a
tr −v 1−k
v 1−k
un−−a hn−a,a ua+ zλ ha,n−a T w
λ
k=1 k=1
= (−1)n v ( w a,n−a )+( w λ(2) )+( w λ(1) ) (−1)a ṽ n−a .
It follows that

n(n−1)
tr un−−a hn−a,a ua+ zλ ha,n−a T w
= v 2
+( w λ(2) )+( w λ(1) )
ṽ n−a .
λ
Hence,
n(n−1) sλ ( v , ṽ )
f λ ( v , ṽ ) = v 2 ṽ n−a ,
∈A
sλ(1) ( v )sλ(2) ( v )
as required. 2
Corollary 4.7. With the above notations, we have that
n(n−1) sλ ( v , 1)
f λ(v ) = v 2 .
sλ(1) ( v )sλ(2) ( v )
5. Twining character formulae
The purpose of this final section is to derive a closed formula of the decomposition number
+ :
[Δ + +
λ,λ L μ,μ ] (where λ, μ ∈ Λ (m)) in terms of the elements f (λ,λ) (q), f (μ,μ) (q), and thus proving
our main result Theorem 2.9.
Let M be an arbitrary finite dimensional
A (m)-module. Then M θ ⊆ M. For each μ ∈ Λ(m), it
follows from the definition that
1
θφμ 1
,μ = φμ,μ θ,
1
φμ 1
, μ θ = θ φ μ, μ .
Let M (μ,μ) be the (μ, μ)-weight space of the A (m)-module M. Note that for any x ∈ M, x ∈ M (μ,μ)
if and only xφμ1 1
,μ = x = xφμ,μ . Then, it is easy to verify that M (μ,μ) θ ⊆ M (μ,μ) . Therefore, it makes
sense to talk about the trace of θ on M (μ,μ) .
We define the twining character of the A (m)-module M as follows:

chθ ( M ) := Tr(θ, M (μ,μ) )e μ ∈ K Λ(m) ,
μ∈Λ(m)
where K [Λ(m)] denotes the group algebra of the additive group Λ(m). It is easy to see that chθ
induces a homomorphism from the Grothendieck group R( A (m)) to K [Λ(m)]. We denote this homo-
morphism again by chθ . We use π to denote the natural map from the group ring Z[Λ(m)] to the
group algebra K [Λ(m)].
Lemma 5.1. For any λ, μ ∈ Λ+ (m) with λ = μ, we have that

+ = − chθ Δ
chθ Δ − = f (λ,λ) (q)π (ch Δλ ),
λ,λ λ,λ

chθ
L+ θ −
λ,λ = − ch L λ,λ = f (λ,λ) (q)π (ch L λ ),
chθ (
L λ,μ ) = 0,
where ch Δλ , ch L λ denote the formal characters of the Weyl module and irreducible module associated to λ of
the q-Schur algebra S q (m), respectively.
Proof. Let ν ∈ Λ(m). By definition,

L λ,μ = ( L λ ⊗ L μ ) ⊕ ( L λ ⊗ L μ )θ , and the (ν , ν )-weight space of
L λ,μ
is just
( L λ )ν ⊗ ( L μ )ν ⊕ ( L λ )ν ⊗ ( L μ )ν θ.
It is easy to see that the trace of θ on this space is 0. This proves chθ ( L λ,μ ) = 0.
Let v λ (resp.,
v λ ) be the highest weight vector of Δλ (resp., of Δ λ ). By definition, we know that
the (ν , ν )-weight space of Δ + is (Δλ )ν ⊗ (Δλ )ν . Let { v λ xi }k be a K -basis of (Δλ )ν , then {
v λ
xi }ki=1
λ,λ i =1
λ )ν . We now apply Corollary 3.17. For any integer 1 i , j k, we have
is a K -basis of (Δ
θ
( v λ xi ⊗
v λ
x j) = vλx j ⊗
v λ
xi .
f (λ,λ) (q)
It follows that

+
Tr θ, Δλ,λ (ν ,ν ) = f (λ,λ) (q) dim(Δλ )ν .

+ ) = f (λ,λ) (q)π (ch Δλ ). The remaining equalities can be proved in a similar way. This
Hence chθ (Δλ,λ
completes the proof of the lemma. 2
For any λ, μ, ν ∈ Λ+ (m) with μ = ν , we set

dλ,μ := [Δλ : L μ ], + :
mλ,μ := Δ + + :
mλμ,ν := Δ
λ,λ L μ,μ , λ,λ L μ,ν .
By (3.18), it is clear that

+ : − 2
Δλ,λ L μ,μ = dλ,μ − mλ,μ .
Therefore, we have the following equality in the Grothendieck group R(

A (m)):

+ =
Δλ,λ mλ,μ
L+ 2 −
μ,μ + dλ,μ − mλ,μ L μ,μ + mλμ,ν [
L μ,ν ].
μ∈Λ+ (m) μ,ν ∈Λ+ (m)
μ=ν
We apply the map chθ to the above equality and using Lemma 5.1. It follows that

f (λ,λ) (q)π (ch Δλ ) = 2mλ,μ − d2λ,μ f (μ,μ) (q)π (ch L μ ). (5.2)
μ∈Λ+ (m)
On the other hand, we have that (in Z[Λ+ (m)])

ch Δλ = dλ,μ ch L μ ,
μ∈Λ+ (m)
and hence (in K [Λ+ (m)])

π (ch Δλ ) = dλ,μ π (ch L μ ). (5.3)
μ∈Λ+ (m)
By the results in [5] and [15], S q (m) is an epimorphic image of the quantum algebra U K (glm ), hence
each irreducible S q (m)-module L μ is a highest weight module over the quantum algebra U K (glm ). It
follows easily that the elements in the set {π (ch L μ )} are K -linear independent. Therefore, we can
compare the coefficients of the equalities (5.2) and (5.3) and deduce that (in K )

f (λ,λ) (q)
mλ,μ = dλ,μ + dλ,μ /2.
f (μ,μ) (q)
If μ is e-restricted, we then apply Corollary 3.21 and this proves Theorem 2.9.
Acknowledgments
The research was supported by Alexander von Humboldt Foundation and partly by Scientific Re-
search Foundation for the Returned Overseas Chinese Scholars, State Education Ministry, the Basic
Research Foundation of BIT and the National Natural Science Foundation of China (Project 10771014).
The author thanks the hospitality of the Mathematical Institute at the University of Cologne during
his visit. He also thanks Professor Peter Littelmann for many helpful discussions.
References
[1] S. Ariki, On the decomposition numbers of the Hecke algebra of G (m, 1, n), J. Math. Kyoto Univ. 36 (1996) 789–808.
[2] S. Ariki, On the classification of simple modules for cyclotomic Hecke algebras of type G (m, 1, n) and Kleshchev multi-
partitions, Osaka J. Math. 38 (4) (2001) 827–837.
[3] S. Ariki, N. Jacon, Dipper–James–Murphy’s conjecture for Hecke algebras of type B, preprint, math.RT/0703447.
[4] S. Ariki, A. Mathas, The number of simple modules of the Hecke algebras of type G (r , 1, n), Math. Z. 233 (3) (2000) 601–
623.
[5] A.A. Beilinson, G. Lusztig, R. Macpherson, A geometric setting for the quantum deformation of GLn , Duke Math. J. 61 (1990)
655–677.
[6] C.W. Curtis, L. Reiner, Methods of Representations Theory, I, Wiley–Interscience, New York, 1981.
[7] R. Dipper, G.D. James, Representations of Hecke algebras of general linear groups, Proc. London Math. Soc. 52 (3) (1986)
20–52.
[8] R. Dipper, G.D. James, Blocks and idempotents of Hecke algebras of general linear groups, Proc. London Math. Soc. 54 (3)
(1987) 57–82.
[9] R. Dipper, G.D. James, Representations of Hecke algebras of type B n , J. Algebra 146 (1992) 454–481.
[10] R. Dipper, G.D. James, The q-Schur algebra, Proc. London Math. Soc. 59 (3) (1989) 23–50.
[11] R. Dipper, G.D. James, q-tensor space and q-Weyl modules, Trans. Amer. Math. Soc. 327 (1) (1991) 251–282.
[12] R. Dipper, G.D. James, A. Mathas, Cyclotomic q-Schur algebras, Math. Z. 229 (1998) 385–416.
[13] R. Dipper, G.D. James, E. Murphy, Hecke algebras of type B n at roots of unity, Proc. London Math. Soc. 70 (3) (1995)
505–528.
[14] R. Dipper, A. Mathas, Morita equivalence of Ariki–Koike algebras, Math. Z. 240 (2002) 579–610.
[15] J. Du, A note on quantized Weyl reciprocity at roots of unity, Algebra Colloq. 2 (1995) 363–372.
[16] M. Geck, On the representation theory of Iwahori–Hecke algebras of extended finite Weyl groups, Represent. Theory 4
(2000) 370–397.
[17] G. Genet, On decomposition matrices for graded algebras, J. Algebra 274 (1) (2004) 523–542.
[18] J.A. Green, Polynomial Representations of GLn , Lecture Notes in Math., vol. 830, Springer-Verlag, 1980.
[19] J. Hu, A Morita equivalence theorem for Hecke algebra Hq ( D n ) when n is even, Manuscripta Math. 108 (2002) 409–430.
[20] J. Hu, Crystal basis and simple modules for Hecke algebra of type D n , J. Algebra 267 (1) (2003) 7–20.
[21] J. Hu, Modular representations of Hecke algebras of type G ( p , p , n), J. Algebra 274 (2) (2004) 446–490.
[22] J. Hu, Branching rules for Hecke algebras of type D n , Math. Nachr. 280 (2007) 93–104.
[23] J. Hu, Crystal basis and simple modules for Hecke algebra of type G ( p , p , n), Represent. Theory 11 (2007) 16–44.
[24] J. Hu, The number of simple modules for the Hecke algebras of type G (r , p , n) (with an appendix by Xiaoyi Cui), J. Algebra
(2009), doi:10.1016/j.jalgebra.2008.01.008, in press.
[25] J. Hu, The representation theory of the cyclotomic Hecke algebras of type G (r , p , n), in: Geometry, Analysis and Topology
of Discrete Groups, in: Adv. Lectures Math., Higher Education Press, 2008, pp. 196–230.
[26] J. Hu, A. Mathas, Morita equivalences of cyclotomic Hecke algebras of type G (r , p , n), J. Reine Angew. Math., in press.
[27] J. Hu, T. Shoji, Schur–Weyl reciprocity between quantum groups and Hecke algebras of type G ( p , p , n), J. Algebra 298 (1)
(2006) 215–237.
[28] N. Jacon, Sur les représentations modulaires des algèbres de Hecke de type D n , J. Algebra 274 (2) (2004) 607–628.
[29] M. Künzer, A. Mathas, Elementary divisors of Specht modules, European J. Combin. 26 (2005) 943–964.
[30] A. Lascoux, B. Leclerc, J.-Y. Thibon, Hecke algebras at roots of unity and crystal bases quantum affine algebras, Comm. Math.
Phys. 181 (1996) 205–263.
[31] A. Mathas, Matrix units and generic degrees for the Ariki–Koike algebras, J. Algebra 281 (2004) 695–730.
[32] E. Murphy, The representations of Hecke algebras of type An , J. Algebra 173 (1995) 97–121.
[33] C. Pallikaros, Representations of Hecke algebras of type D n , J. Algebra 169 (1994) 20–48.
Journal of Algebra
Une démonstration simple de la fidélité de la représentation

de Lawrence–Krammer–Paris
Jean-Yves Hée
LAMFA (CNRS UMR 6140), Université de Picardie Jules Verne, 33, rue Saint-Leu, 80039 Amiens Cedex 1, France
Article history: Le théorème de linéarité des groupes d’Artin–Tits de type sphérique

Reçu le 8 août 2008 et le théorème d’injectivité de tout monoïde d’Artin–Tits dans son
Disponible sur Internet le 1er novembre groupe reposent essentiellement sur la fidélité de la représentation
2008
de Lawrence–Krammer–Paris restreinte au monoïde. Nous démon-
Communiqué par Michel Broué
trons cette fidélité sans utiliser les formes normales des éléments
Mots-clés : du monoïde ni les parties fermées du système de racines associé
Monoïdes et groupes d’Artin–Tits à la matrice de Coxeter en jeu ; nous n’employons que des notions
Homomorphismes initialement injectifs très élémentaires.
Théorèmes de linéarité et d’injectivité © 2008 Elsevier Inc. Tous droits réservés.
The theorem of linearity of the Artin–Tits groups of spherical type
Keywords: and the theorem of injectivity of any Artin–Tits monoid in its
Artin–Tits monoids and groups group are essentially based on the faithfulness of the Lawrence–
Initially injective homomorphisms Krammer–Paris representation restricted to the monoid. We prove
Theorems of linearity and of injectivity
this faithfulness using neither the normal forms of the elements of
the monoid nor the closed subsets of the associated root system;
only very elementary notions are needed.
© 2008 Elsevier Inc. Tous droits réservés.
1. Introduction
Les groupes d’Artin–Tits de type sphérique, c’est-à-dire ceux qui sont associés à un groupe de
Coxeter fini, sont linéaires. Ce théorème de linéarité a d’abord été démontré pour le type A 3 , autre-
ment dit pour le groupe des tresses à quatre brins, par D. Krammer [K1], puis pour tous les types A n
par S. Bigelow par une méthode topologique [Bi] et par D. Krammer par une méthode algébrique [K2].
Il a ensuite été étendu à tous les autres types sphériques par A. Cohen et D. Wales [CW] et indépen-
damment par F. Digne [Di].
Adresse e-mail : jean-yves.hee@u-picardie.fr.
0021-8693/$ – see front matter © 2008 Elsevier Inc. Tous droits réservés.
doi:10.1016/j.jalgebra.2008.09.021
1040 J.-Y. Hée / Journal of Algebra 321 (2009) 1039–1048
Peu de temps après, L. Paris a montré que tous les monoïdes d’Artin–Tits s’injectent dans leurs
groupes [P]. Ce théorème d’injectivité avait été démontré antérieurement pour le type A n par F. Garside
[G] et généralisé à tous les types sphériques par E. Brieskorn et K. Saito [BS] et indépendamment par
P. Deligne [De] ; une démonstration simple dans le cas sphérique se trouve aussi dans l’article [Mi] de
J. Michel. Quelques cas non sphériques avaient également été traités, voir l’introduction de [P].
Les démonstrations de ces deux théorèmes reposent sur l’existence de la représentation dite de
Lawrence–Krammer (nous adoptons ici la dénomination introduite par S. Bigelow dans [Bi] en réfé-
rence aux travaux de R.J. Lawrence [L] et de D. Krammer [K1]) et de ses généralisations construites
dans [CW], [Di] et [P]. Plus précisément, les deux théorèmes résultent d’une propriété essentielle de
ces représentations, la fidélité de leurs restrictions aux monoïdes d’Artin–Tits.
Nous donnons de cette fidélité une démonstration très simple qui n’utilise que des outils élémen-
taires, tel le monoïde des relations binaires sur un ensemble.
Les paragraphes 2 et 3 rassemblent des généralités sur le monoïde d’Artin–Tits B + associé à une
matrice de Coxeter Γ , sur le monoïde Bin(Ω) des relations binaires sur un ensemble Ω et sur les
homomorphismes de monoïdes de B + dans Bin(Ω). On y introduit la notion d’homomorphisme ini-
tialement injectif et on établit deux propositions (1 et 2) qui sont cruciales pour la suite : on les
utilise aux numéros (4.3) et (4.4) et dans la section 5.3.
Au paragraphe 4, nous rappelons la définition de la représentation linéaire de B + construite par
L. Paris [P] dans le cas où Γ est « à liaisons simples et sans triangle » ; nous la notons ρ et l’appelons
représentation de Lawrence–Krammer–Paris. Puis nous démontrons que ρ est fidèle en appliquant la
proposition 1 et un corollaire de la proposition 2.
Le paragraphe 5 est consacré à quelques compléments. En particulier, nous rappelons comment la
fidélité de ρ entraîne les théorèmes de linéarité et d’injectivité évoqués au début de la présente intro-
duction. D’autre part, nous proposons une généralisation de la méthode développée aux paragraphes
2 et 3 à tout monoïde qui, comme B + , peut être défini par une présentation homogène. Enfin, nous
indiquons brièvement comment les propositions 1 et 2 permettent de simplifier la démonstration de
la fidélité des représentations linéaires définies par F. Digne dans [Di] ; cependant, pour ne pas alour-
dir notre texte, nous ne rappelons pas la construction de ces représentations, nous nous concentrons
sur les relations binaires qui servent à en établir la fidélité.
2. Monoïdes d’Artin–Tits et homomorphismes initialement injectifs
(2.1) Dans ce no (2.1), nous fixons des notations et une terminologie qui vont nous servir jusqu’à
la section 5.1 incluse.
(a) Soit Γ = (mi , j )i , j ∈ I une matrice de Coxeter.
(b) Soit M un monoïde multiplicatif dont l’élément neutre est noté 1.
Si x et y sont des éléments de M, on pose (x, y ; 0) = 1 et, si n est un entier 1, on note (x, y ; n)
le produit xyx . . . constitué de n termes alternativement égaux à x et à y, le premier étant égal à x.
(c) Nous disons qu’une famille x = (xi )i ∈ I d’éléments de M est de type Γ dans M si, pour tout
couple (i , j ) d’éléments distincts de I tels que mi , j = ∞, on a (xi , x j ; mi , j ) = (x j , xi ; mi , j ).
(d) Nous notons respectivement W = W (Γ ), B + = B + (Γ ) et B = B (Γ ) le groupe de Coxeter, le
monoïde d’Artin–Tits et le groupe d’Artin–Tits associés à la matrice Γ , c’est-à-dire définis par les
présentations suivantes

W = W (Γ ) = s = (si )i ∈ I s est de type Γ et, pour tout i ∈ I , s2i = 1 ,

B + = B + (Γ ) = b = (b i )i ∈ I b est de type Γ mon
,

B = B (Γ ) = σ = (σi )i∈ I σ est de type Γ .
gr
(e) Nous disons que la matrice de Coxeter Γ est finie (resp. sphérique) si l’ensemble I (resp. le
groupe W ) est fini.
J.-Y. Hée / Journal of Algebra 321 (2009) 1039–1048 1041
(f) Nous désignons par ι = ι(Γ ) : B + → B l’homomorphisme de monoïdes qui, pour tout i ∈ I ,
envoie b i sur σi . L’injectivité de ι est le théorème de L. Paris cité plus haut [P] ; nous revenons sur sa
démonstration dans la section 5.1 ci-dessous.
(g) Nous notons l la fonction longueur sur B + relativement à la famille génératrice (b i )i ∈ I .
Si a et b appartiennent à B + , nous écrivons a ≺ b pour exprimer qu’il existe un élément c de B +
tel que b = ac.
Pour tout élément b de B + , nous posons I (b) = {i ∈ I | b i ≺ b}.
Les ensembles I (b) et les assertions (a) et (b) du no (2.2) ci-dessous vont jouer un rôle déterminant
dans la démonstration de fidélité du no (4.4).
(2.2) Soit ρ : B + → M un homomorphisme de monoïdes. Nous disons que ρ est initialement injectif
si, pour tout couple (b, b ) d’éléments de B + tels que ρ (b) = ρ (b ), on a I (b) = I (b ).
Soient P un ensemble et τ : B + → P P un homomorphisme de monoïdes, la loi du monoïde P P
étant la composition des applications. Pour tout élément b de B + , notons τb : P → P l’image de b
par τ . Soit p un élément de P . Nous disons que p est initialement injectif pour τ si, pour tout couple
(b, b ) d’éléments de B + tels que τb ( p ) = τb ( p ), on a I (b) = I (b ).
De ces définitions, on déduit aussitôt les assertions (a) et (b) ci-dessous.
(a) Si ϕ : M → M est un homomorphisme de monoïdes et si l’homomorphisme composé ρ = ϕ ◦ ρ est
initialement injectif, il en est de même de ρ .
(b) Si P possède un élément p initialement injectif pour τ , l’homomorphisme τ est initialement injectif.
(2.3) La proposition fondamentale qui suit est essentiellement due à D. Krammer, voir [K1, Propo-
sition 5.1] ou [K2, Proposition 2.1].
Proposition 1. Supposons que Im(ρ ) soit un sous-monoïde d’un groupe G. Si l’homomorphisme ρ est initia-
lement injectif, il est injectif.
Démonstration. Soient b, b deux éléments de B + tels que ρ (b) = ρ (b ). Il s’agit de montrer que
b = b . Nous raisonnons par récurrence sur l’entier n = l(b) + l(b ). Si n = 0, on a l(b) = l(b ) = 0, donc
b = b = 1. Supposons que n > 0. On a alors par exemple b = 1, donc il existe i ∈ I (b). D’autre part,
comme ρ est initialement injectif, l’égalité ρ (b) = ρ (b ) entraîne que I (b) = I (b ). On peut donc écrire
b = b i c et b = b i c , où c , c ∈ B + . De l’égalité ρ (b) = ρ (b ), on déduit ρ (b i )ρ (c ) = ρ (b i )ρ (c ). Puisque
G est un groupe, on peut simplifier par ρ (b i ) ; on obtient ρ (c ) = ρ (c ). Or l(c ) + l(c ) = n − 2 < n, donc
c = c d’après l’hypothèse de récurrence, d’où immédiatement b = b . 2
3. Utilisation de relations binaires
(3.1) Soit Ω un ensemble. Nous rassemblons dans ce no (3.1) quelques définitions concernant l’en-
semble Bin(Ω) des relations binaires sur Ω , c’est-à-dire l’ensemble des parties de Ω × Ω .
(a) Si R et S appartiennent à Bin(Ω), nous définissons le produit

R S = (x, z) ∈ Ω × Ω il existe y ∈ Ω tel que xR y et y S z ;
conformément à l’usage, nous écrivons xR y au lieu de (x, y ) ∈ R.

Muni de cette multiplication, l’ensemble Bin(Ω) est un monoïde dont l’élément neutre est la rela-
tion d’égalité.
(b) Nous associons à toute relation binaire R ∈ Bin(Ω) sa matrice α = (αx, y )x, y ∈Ω ∈ MΩ ({0, +}),
définie par : αx, y = + ⇔ xR y. Par exemple, la matrice de la relation binaire R = ∅ est la matrice nulle
α = (0).
Nous obtenons ainsi une bijection de Bin(Ω) sur MΩ ({0, +}), ce qui nous permet, par transport
de structure, de munir MΩ ({0, +}) d’une loi de monoïde que nous allons expliciter et qui peut être
vue comme une multiplication ligne-colonne.
Définissons d’abord le produit de deux éléments de {0, +} : il est égal à 0 si l’un au moins des
deux facteurs est nul, et à + sinon. Définissons aussi la somme d’une famille, finie ou non, d’éléments
de {0, +} : elle est égale à 0 si tous ses termes sont nuls, et à + sinon.
Maintenant, dans le monoïde MΩ ({0, +}), le produit α β de deux matrices α = (αx, y )x, y∈Ω et
β = (βx, y )x, y∈Ω est la matrice γ = (γx, y )x, y∈Ω définie par γx, y = z∈Ω αx,z βz, y (quels que soient
x, y ∈ Ω ).
Lorsque Ω est fini de cardinal m, cette multiplication dans MΩ ({0, +}) est celle qui est définie
dans Mm ({0, 1}) par D. Krammer [K2, p. 133].
(c) Nous notons P (Ω) l’ensemble des parties de Ω et nous appelons homomorphisme canonique
de Bin(Ω) dans P (Ω)P (Ω) l’homomorphisme de monoïdes ϕ : Bin(Ω) → P (Ω)P (Ω) qui, à tout élé-
ment R de Bin(Ω), associe l’application ϕ R : P (Ω) → P (Ω) définie par :
ϕ R (Y ) = {x ∈ Ω | il existe y ∈ Y tel que xR y }.
(d) Si R : B + → Bin(Ω) est un homomorphisme de monoïdes, nous formons l’homomorphisme

τ = ϕ ◦ R : B + → P (Ω)P (Ω) , où ϕ : Bin(Ω) → P (Ω)P (Ω) est l’homomorphisme canonique défini
en (c) ; pour tout élément b de B + , nous convenons de noter R b et τb les images respectives de b par
R et par τ .
(3.2) Soit D = (Ω, R, a) un triplet dans lequel Ω est un ensemble, R = ( R i )i ∈ I est une famille de
type Γ formée d’éléments de Bin(Ω), et a = (ai )i ∈ I est une famille d’éléments de Ω . Nous notons
R : B + → Bin(Ω) l’homomorphisme de monoïdes tel que, pour tout élément i de I , R(b i ) = R i , et
nous adoptons les notations τ , R b et τb (pour b appartenant à B + ) définies en (3.1)(d).
Proposition 2. Supposons remplie la condition
(C) pour tout couple (i , j ) d’éléments distincts de I et pour tout entier n tel que 1 n < mi , j , si l’on pose
c = (b j , b i ; n) = b j b i b j . . . b j et {i , j } = {i , j }, on a ai R c ai .
Alors, pour tout b ∈ B + et tout i ∈ I tels que b i ⊀ b, il existe k ∈ I tel que ai R b ak ; en particulier, on a
ai ∈ τb (Ω).
Démonstration. On raisonne par récurrence sur l(b).

Si l(b) = 0, la relation R b est l’égalité, donc ai R b ai .
Supposons que l(b) 1. Il existe j ∈ I tel que b j ≺ b. On a i = j. Notons n le plus grand en-
tier tel que (b j , b i ; n) ≺ b; on a n 1. Si l’on avait n mi , j , on aurait mi , j = ∞ et (b i , b j ; mi , j ) =
(b j , bi ; mi , j ) ≺ (b j , bi ; n) ≺ b, d’où b i ≺ b, ce qui est exclu. On a donc n < mi , j . Écrivons c = (b j , bi ; n) =
b j b i b j . . . b j et {i , j } = {i , j }. Il existe b ∈ B + tel que b = cb . La maximalité de n entraîne que
b i ⊀ b . Comme l(b ) = l(b) − n < l(b), l’hypothèse de récurrence montre qu’il existe k ∈ I tel que
ai R b ak . Or, d’après (C), on a ai R c ai , donc ai R b ak . 2
Corollaire. Supposons remplies la condition (C) ci-dessus ainsi que la condition
(C ) pour tout élément i de I et tout couple (x, y ) d’éléments de Ω tels que xR i y, on a x = ai .
Alors,
(a) pour tout b ∈ B + , on a, pour tout i ∈ I , ai ∈ τb (Ω) ⇔ b i ⊀ b ; autrement dit, I (b) = {i ∈ I | ai ∈

/ τb (Ω)} ;
(b) l’homomorphisme R : B + → Bin(Ω) est initialement injectif ; plus précisément, l’ensemble Ω est initia-
lement injectif pour τ .
Démonstration. (a) résulte de (C ) et de la proposition précédente.

(b) résulte immédiatement de (a) et de (2.2)(a) et (b). 2
4. Fidélité de la représentation de Lawrence–Krammer–Paris
(4.1) Dans tout le paragraphe 4, nous supposons que la matrice de Coxeter Γ est à liaisons simples
(i.e., si i et j appartiennent à I , on a mi , j 3) et sans triangle (i.e., si i , j , k sont des éléments deux à
deux distincts de I , l’un au moins des trois coefficients mi , j , m j ,k , mk,i est égal à 2).
Sous ces hypothèses, Daan Krammer [K2], dans le cas particulier où Γ est de type A n , et Luis
Paris [P], dans le cas général, ont construit une remarquable représentation ρ : B + → GL( E ), où E est
un espace vectoriel sur le corps K = Q( X , Y ) des fractions rationnelles à deux indéterminées sur Q.
Un résultat important de Krammer et de Paris est que la représentation ρ est fidèle. La représentation
considérée par Krammer se trouve déjà dans [L] ; d’autre part, avant l’étude générale de Paris, le cas
où Γ est de type D n , E 6 , E 7 ou E 8 avait été traité indépendamment dans [CW] et dans [Di].
Nous allons voir que la fidélité de ρ résulte très simplement du corollaire du no (3.2) ci-dessus.
(4.2) Rappelons la construction de ρ donnée par L. Paris dans [P].

Notons encore l la fonction longueur sur W par rapport à la famille génératrice (si )i ∈ I de W .
Soit a = (ai )i ∈ I la base canonique du Z-module V = Z( I ) . Si i et j appartiennent à I , posons
A i , j = 2, 0 ou −1 selon que mi , j = 1, 2 ou 3. On peut identifier le groupe de Coxeter W à un sous-
groupe de GL( V ) au moyen de la représentation fidèle W → GL( V ) définie par : si (a j ) = a j − A i , j ai
quels que soient i et j appartenant à I .
Considérons le système de racines Φ = { w (ai ) | w ∈ W , i ∈ I } et notons Φ + l’ensemble des élé-
ments de Φ dont toutes les coordonnées dans la base a = (ai )i ∈ I sont positives ou nulles. On sait que
Φ est réunion disjointe des ensembles Φ + et Φ − = −Φ + . Pour toute racine a appartenant à Φ + , on
appelle profondeur de a l’entier dp(a) = Min{l( w ) | w ∈ W , w (a) ∈ Φ − }, voir [BH, p. 181].
Posons Ω = Φ + et notons (ea )a∈Ω la base canonique du K -espace vectoriel E = K (Ω) .
Pour tout élément i de I , l’ensemble Φ + \ {ai } est stable par si et toute orbite de cardinal 2 de
Φ \ {ai } sous l’action de si est de la forme {a, a }, où dp(a ) = dp(a) + 1, voir [BH, Lemma 1.7]. On
+
définit un endomorphisme ϕi du K -espace vectoriel E en posant
• ϕi (eai ) = 0,
• ϕi (ea ) = ea , si a ∈ Φ + et si (a) = a,
• ϕi (ea ) = (1 − Y )ea + ea et ϕi (ea ) = Y ea , si a, a ∈ Φ + , a = si (a) et dp(a ) = dp(a) + 1.
On vérifie que, dans le monoïde End( E ) formé des endomorphismes du K -espace vectoriel E, la
famille (ϕi )i ∈ I est de type Γ , voir [P, Proposition 3.1].
Ensuite, en utilisant le fait que la matrice de Coxeter Γ est sans triangle, on construit une famille
particulière ( T (i , a))i ∈ I ,a∈Ω de polynômes appartenant à Z[Y ], voir [P, Theorem 3.2].
Pour tout élément i de I , on définit alors un endomorphisme ψi du K -espace vectoriel E en posant
ψi (ea ) = ϕi (ea ) + X T (i , a)eai .
Grâce aux particularités des polynômes T (i , a) (i ∈ I , a ∈ Ω ), on montre que la famille (ψi )i ∈ I est
de type Γ dans le monoïde End( E ), voir [P, Lemmas 3.6 and 3.7].
Enfin, on vérifie que, pour tout élément i de I , l’endomorphisme ψi appartient à GL( E ), voir
[P, Lemma 3.8].
Nous appelons représentation de Lawrence–Krammer–Paris l’homomorphisme de monoïdes ρ : B + →
GL( E ) tel que, pour tout élément i de I , ρ (b i ) = ψi .
(4.3) Vu la proposition 1, pour montrer que l’homomorphisme ρ est injectif, il suffit de montrer
qu’il est initialement injectif. Dans ce but, nous commençons par déduire de ρ un homomorphisme
de monoïdes R : B + → Bin(Ω).
Pour tout b ∈ B + , notons μ(b) la matrice de ρ (b) dans la base (ea )a∈Ω de E. Pour tout i ∈ I ,
la matrice μ(b i ) de ψi est à coefficients dans Z[ X , Y ] ; il en est donc de même de μ(b) pour tout
b ∈ B + . On obtient ainsi un homomorphisme de monoïdes multiplicatifs μ : B + → MΩf (Z[ X , Y ]), où
f
MΩ (Z[ X , Y ]) désigne l’anneau formé des matrices de type Ω × Ω , à coefficients dans Z[ X , Y ] et
dont chaque colonne ne contient qu’un nombre fini de termes non nuls.
Soient k un corps commutatif totalement ordonné et t un élément de k tel que 0 < t < 1 (par
exemple, k = Q ou R et t = 12 ). L’homomorphisme d’anneaux de Z[ X , Y ] dans k qui envoie X sur 0
et Y sur t induit un homomorphisme d’anneaux ε : MΩf (Z[ X , Y ]) → MΩf (k). L’application composée
ε ◦ μ : B + → MΩf (k) est un homomorphisme de monoïdes. Pour tout b ∈ B + , posons ν (b) = (ε ◦ μ)(b).
Pour tout i ∈ I , les coefficients de la matrice ν (b i ) appartiennent à l’ensemble {0, 1, t , 1 − t } donc à
l’ensemble k+ des éléments 0 de k. Il en résulte que, pour tout b ∈ B + , les coefficients de la matrice
ν (b) appartiennent aussi à k+ . On obtient donc un homomorphisme de monoïdes ν : B + → MΩf (k+ ),
où MΩ (k+ ) désigne le sous-monoïde de MΩ (k) constitué des matrices à coefficients 0.
f f
Soit β : MΩ (k+ ) → Bin(Ω) l’application qui, à toute matrice α = (αx, y )x, y ∈Ω de MΩ (k+ ), associe
f f
la relation binaire dont la matrice α = (α x, y )x, y ∈Ω est définie par : α x, y = + ⇔ αx, y > 0. On vérifie
que β est un homomorphisme de monoïdes ; noter que, dans le cas où Γ est de type A n−1 , l’homo-
morphisme β est essentiellement l’homomorphisme B n+ → Mm ({0, 1}) considéré dans [K2, p. 133].
Considérons enfin l’homomorphisme R = β ◦ ν : B + → Bin(Ω) et posons R = ( R i )i ∈ I , où, pour tout
élément i de I , R i = R (b i ).
(4.4) Proposition 3. L’homomorphisme R : B + → Bin(Ω) est initialement injectif.
Démonstration. D’après le corollaire à la proposition 2, il suffit de vérifier que le triplet D = (Ω =

Φ + , R, a) satisfait aux conditions (C) et (C ) du no (3.2) ci-dessus. Or, un coup d’œil sur la définition
de ϕi montre que la relation binaire R i est la relation telle que, pour x et y appartenant à Φ + ,

xR i y ⇔ x = si ( y ) ou x = y = ai et dp(x) < dp si (x) .
On en déduit aussitôt la condition (C ).

D’autre part, puisque Γ est à liaisons simples, la condition (C) se réduit aux deux propriétés sui-
vantes :
(a) si i et j sont des éléments distincts de I , on a ai R j ai ;

(b) si i et j sont des éléments de I tels que mi , j = 3, on a ai R j R i a j .
On vérifie immédiatement (a) ; quant à la propriété (b), elle résulte des relations ai R j s j (ai ),
s j (ai ) = si (a j ) et si (a j ) R i a j . 2
Conséquence. La représentation de Lawrence–Krammer–Paris est fidèle.
Démonstration. L’homomorphisme ρ : B + → GL( E ) est initialement injectif car, si ρ (b) = ρ (b ), où b

et b appartiennent à B + , alors μ(b) = μ(b ) donc ν (b) = ν (b ) et enfin R b = R b , d’où I (b) = I (b )
d’après la proposition 3.
Il résulte donc de la proposition 1 que la représentation ρ est fidèle. 2
5. Compléments
5.1. Démonstration des théorèmes de linéarité et d’injectivité
(5.1.1) Rappelons les énoncés des théorèmes de linéarité et d’injectivité.
Théorème 1 (Linéarité). Si Γ est sphérique, le groupe d’Artin–Tits B est linéaire : il existe un homomorphisme
injectif B → GL( N , K ), où N est un entier naturel et K un corps commutatif convenables.
Théorème 2 (Injectivité). Le morphisme ι : B + → B est injectif.

Dans les trois numéros suivants, nous rappelons succinctement comment les théorèmes 1 et 2 se
déduisent de la fidélité de la représentation de Lawrence–Krammer–Paris.
(5.1.2) Lemme 1. Soient ρ : B + → G un morphisme de monoïdes de B + dans un groupe G et ρgr : B → G

le morphisme de groupes induit par ρ , c’est-à-dire tel que ρ = ρgr ◦ ι. Supposons que ρ soit injectif. Alors,
(a) le morphisme ι : B + → B est injectif ;

(b) si Γ est sphérique, le morphisme ρgr est injectif.
Démonstration. L’assertion (a) est immédiate.

(b) L’injectivité de ι permet d’identifier tout élément b de B + à son image ι(b) dans B. Soit alors
x ∈ Ker(ρgr ). Puisque Γ est sphérique, il existe des éléments b et b de B + tels que x = b b−1 , cf.
par exemple [Mi], première assertion du corollaire 3.2. On a xb = b , donc ρgr (x)ρgr (b) = ρgr (b ), i.e.
ρ (b) = ρ (b ). Comme ρ est injectif, on a donc b = b , d’où x = 1, ce qui établit l’injectivité de ρgr . 2
Le lemme 2 ci-dessous se démontre aisément. Avant de l’énoncer, nous posons, pour toute partie
J de I , Γ J = (mi , j )i , j ∈ J , B +J = B + (Γ J ), B J = B (Γ J ) et ι J = ι(Γ J ).
Lemme 2. Si le morphisme ι J : B +J → B J est injectif pour tout sous-ensemble fini J de I , il en est de même du
morphisme ι : B + → B.
(5.1.3) La proposition suivante a été démontrée par L. Paris (cf. [P, Theorem 5.1]) en utilisant no-
tamment des notions développées par J. Crisp dans [Cr] ; pour une approche fondée sur les partitions
admissibles définies par B. Mühlherr dans [Mü], voir la thèse d’Anatole Castella (cf. [Ca, section 8.3 et
en particulier le corollaire 8.3.11]).
Proposition 4. Si la matrice de Coxeter Γ est finie, il existe un morphisme de monoïdes injectif B + (Γ ) →

B + (Γ ), où Γ est une matrice de Coxeter finie, à liaisons simples, sans triangle et qui est sphérique si Γ l’est.
(5.1.4) Maintenant, grâce aux lemmes 1 et 2 et à la proposition 4 ci-dessus, il est facile de vérifier
que la fidélité de la représentation de Lawrence–Krammer–Paris (cf. (4.4)) entraîne les théorèmes 1
et 2.
5.2. Généralisation
(5.2.1) Au no (5.2.2) ci-dessous, nous indiquons comment on peut généraliser aux monoïdes définis
par une présentation homogène la proposition 1 du no (2.3) et la proposition 2 du no (3.2) avec son
corollaire.
Fixons d’abord les notations.
Soient I un ensemble et Γ une partie de n∈N ( I n × I n ).

Soient M un monoïde et x = (xi )i ∈ I une famille d’éléments de M. Pour tout élément n de N et
tout élément i = (i 1 , i 2 , . . . , in ) de I n , nous posons xi = xi 1 xi 2 . . . xin . Nous disons que la famille x est
de type Γ dans M si, pour tout couple γ = (i, j) appartenant à Γ , on a xi = xj .
Nous notons B + = B + (Γ ) le monoïde défini par la présentation

B + = B + (Γ ) = b = (b i )i ∈ I b est de type Γ .
Si x = (xi )i ∈ I est une famille de type Γ dans un monoïde M, il existe un unique morphisme de
monoïdes ρ = ρx : B + → M tel que, pour tout i appartenant à I , ρ (b i ) = xi .
En particulier, il existe un unique morphisme de monoïdes l : B + → N tel que, pour tout élément
i de I , l(b i ) = 1. Ce morphisme l n’est autre que la fonction longueur sur B + relativement à la famille
génératrice (b i )i ∈ I .
Si a et b appartiennent à B + , nous écrivons a ≺ b pour exprimer qu’il existe un élément c de B +

tel que b = ac.
Pour tout élément b de B + , nous posons I (b) = {i ∈ I | b i ≺ b}.
(5.2.2) (a) On peut alors recopier mot pour mot les définitions et les énoncés (a) et (b) du no (2.2)
ci-dessus, la proposition 1 du no (2.3) avec sa démonstration, et le no (3.1)(d).
(b) On peut aussi recopier le no (3.2) en entier à condition de remplacer la proposition 2 par la
proposition 2∗ ci-dessous et d’écrire (C∗ ) au lieu de (C) à la première ligne de l’énoncé du corollaire.
Proposition 2∗ . Supposons remplie la condition
(C∗ ) pour tout couple (i , j ) d’éléments distincts de I , il existe un élément m de N2 ∪ {∞} et un couple
γ = (i, j) satisfaisant aux conditions suivantes
• si m = ∞, i = (i 1 , i 2 , . . .) et j = ( j 1 , j 2 , . . .) sont deux suites infinies formées d’éléments de I ,
• si m = ∞, γ = (i, j) = ((i 1 , i 2 , . . . , im ), ( j 1 , j 2 , . . . , jm )) ∈ Γ ,
• i = i 1 , j = j 1 et, pour 1 n < m, ai R c a jn+1 , où c = b j 1 b j 2 . . . b jn .
Alors, pour tout b ∈ B + et tout i ∈ I tels que b i ⊀ b, il existe k ∈ I tel que ai R b ak ; en particulier, on a
ai ∈ τb (Ω).
Démonstration. On raisonne par récurrence sur l(b).

Si l(b) = 0, la relation R b est l’égalité, donc ai R b ai .
Supposons que l(b) 1. Il existe j ∈ I tel que b j ≺ b. On a i = j. Soient m et γ = (i, j) comme dans
la condition (C∗ ). Notons n le plus grand entier m tel que b j 1 b j 2 . . . b jn ≺ b ; on a n 1 car j = j 1 .
Si l’on avait n = m, on aurait m = ∞ et, puisque i = i 1 , b i ≺ bi = bj ≺ b, d’où b i ≺ b, ce qui est exclu.
On a donc n < m. Il existe b ∈ B + tel que b = cb , où c = b j 1 b j 2 . . . b jn . La maximalité de n entraîne
que b jn+1 ⊀ b . Comme l(b ) = l(b) − n < l(b), l’hypothèse de récurrence montre qu’il existe k ∈ I tel
que a jn+1 R b ak . Or, d’après (C∗ ), on a ai R c a jn+1 , donc ai R b ak . 2
5.3. Fidélité des représentations de Digne
(5.3.1) Dans [Di], F. Digne définit certaines représentations linéaires des groupes d’Artin–Tits de
type sphérique cristallographique et démontre qu’elles sont fidèles. Son raisonnement peut être sim-
plifié en appliquant une méthode analogue à celle employée aux numéros (4.3) et (4.4) ci-dessus. Il
est facile en effet de vérifier que les relations binaires en jeu dans les représentations construites par
Digne sont celles que nous définissons au numéro suivant et qu’elles satisfont aux conditions (C) et
(C ) de la proposition 2 et de son corollaire, voir (5.3.4) ci-dessous.
(5.3.2) Définition des relations binaires de Digne.

Soient Φ un système de racines réduit (au sens de Bourbaki [Bo]) dans un R-espace vectoriel V ,
et a = (ai )i ∈ I une base de Φ . Notons Φ + (resp. Φ − ) le système positif (resp. négatif) défini par la
base a. Pour tout élément i de I , désignons par r i la réflexion de vecteur ai qui conserve Φ . Soient
A = ( A i , j )i , j ∈ I la matrice de Cartan attachée à la base a, et Γ = (mi , j )i , j ∈ I la matrice de Coxeter
associée à la famille (r i )i ∈ I ; si i et j appartiennent à I , on a r i (a j ) = a j − A i , j ai , et mi , j est l’ordre du
produit r i r j dans le groupe GL( V ). Nous adoptons les notations W , B + , etc., de (2.1)(d).
Nous allons définir sur Φ + des relations binaires R i (i ∈ I ), que nous appelons relations de Digne ;
elles apparaissent en effet dans [Di, Proposition 3.4, condition (iii), p. 45].
Pour tout élément i de I , nous notons R i la relation binaire sur Φ + définie ainsi : si x et y
appartiennent à Φ + , alors
xR i y ⇔ il existe n ∈ N tel que x = r i ( y ) − nai .

Remarques. (a) Cette relation R i est symétrique ; en effet, comme r i (ai ) = −ai , on a : x = r i ( y )− nai ⇔
y = r i (x) − nai .
(b) Soient i , j deux éléments distincts de I , et k un entier naturel pair. Si la relation ( R i , R j ; k) est
symétrique, alors ( R i , R j ; k) = ( R j , R i ; k) ; cela résulte du fait que R i et R j sont symétriques.
(c) Soient x et y des éléments de Φ + tels que xR i y. Nous allons voir que x et y sont distincts de
ai et qu’ils appartiennent à la même composante indécomposable de Φ ; plus précisément, si x = y,
x et y appartiennent tous deux à la composante indécomposable de Φ qui contient ai .
En effet, notons n l’entier naturel tel que x = r i ( y ) − nai .
D’une part, si y était égal à ai , on aurait x = −ai − nai = −(n + 1)ai , ce qui est absurde puisque x
appartient à Φ + . On a donc y = ai , et aussi x = ai en raison de la symétrie de R i .
D’autre part, supposons par exemple que y appartienne à une composante indécomposable Ψ de
Φ qui ne contient pas ai . On a alors r i ( y ) = y, donc x = y − nai , ce qui, comme x appartient à Φ ,
montre que n = 0, d’où x = y.
(5.3.3) Proposition 5. La famille R = ( R i )i ∈ I est de type Γ dans le monoïde Bin(Φ + ).
Cet énoncé résulte du lien entre les relations R i et les représentations de Digne (voir notamment
dans [Di] le théorème 3.8, la proposition 3.4 et la première ligne de la démonstration de la proposi-
tion 3.6). On peut aussi démontrer cet énoncé directement à partir de la définition des relations R i
en se ramenant au cas où I est de cardinal 3 et en effectuant alors, cas par cas, des vérifications
matricielles élémentaires dans le monoïde MΩ ({0, +}) (voir le no (3.1)(b)).
(5.3.4) Proposition 6. Le triplet D = (Φ + , R, a) satisfait aux conditions (C) et (C ) du no (3.2).
Démonstration. La condition (C ) résulte de la remarque (c) du no (5.3.2).

D’autre part, rappelons l’énoncé de la condition (C) : pour tout couple (i , j ) d’éléments distincts de I
et pour tout entier n tel que 1 n < mi , j , si l’on pose c = (b j , b i ; n) = b j b i b j . . . b j et {i , j } = {i , j }, on a
ai R c ai .
Démontrons maintenant cette propriété.
Si n = 1, alors i = i et on a bien ai R j ai puisque ai = r j (ai ) − (− A j ,i )a j et que − A j ,i appartient
à N. Supposons désormais que n 2.
Si mi , j = 3, on a r j (ai ) = ai + a j = r i (a j ), ce qui entraîne que ai R j (ai + a j ) R i a j , d’où ai ( R j R i )a j .
Si mi , j = 4, on peut supposer que A i , j = −2 et A j ,i = −1, et on doit vérifier les quatre relations
suivantes :
ai ( R j R i )a j , a j ( R i R j )ai , ai ( R j R i R j )ai et a j ( R i R j R i )a j .
Comme R i et R j sont symétriques, la première relation est conséquence de la deuxième ; il suffit

donc de démontrer les trois dernières. Posons b = 2ai + a j et c = ai + a j . On constate que a j R i c R j ai ,
ai R j c R i c R j ai et a j R i bR j bR i a j ; cela établit les trois dernières relations.
Si mi , j = 6, on peut supposer que A i , j = −3 et A j ,i = −1, et il suffit, puisque R i et R j sont symé-
triques, de vérifier les six relations suivantes :
ai ( R j R i )a j , ai ( R j R i R j )ai , a j ( R i R j R i )a j , ai ( R j R i R j R i )a j ,
ai ( R j R i R j R i R j )ai et a j ( R i R j R i R j R i )a j .
Posons b = 3ai + a j , c = 2ai + a j , d = 3ai + 2a j et e = ai + a j . On constate que ai R j e R i a j , ai R j e R i e R j ai ,

a j R i bR j bR i a j (et aussi a j R i c R j c R i a j ), ai R j e R i c R j c R i a j , ai R j e R i c R j c R i e R j ai et a j R i bR j dR i dR j bR i a j ,
ce qui achève la démonstration de la proposition. 2
(5.3.5) Une autre façon de démontrer la fidélité des représentations de Digne dans le cas où Γ est
de type B n , F 4 ou G 2 est d’utiliser un théorème général d’Anatole Castella, voir [Ca, théorème 11.3.4
et corollaire 11.3.5].
Références
[Bi] S. Bigelow, Braid groups are linear, J. Amer. Math. Soc. 14 (2001) 471–486.
[BH] B. Brink, R.B. Howlett, A finiteness property and an automatic structure for Coxeter groups, Math. Ann. 296 (1993) 179–
190.
[Bo] N. Bourbaki, Groupes et algèbres de Lie, Chapitres IV, V, VI, Masson, Paris, 1981.
[BS] E. Brieskorn, K. Saito, Artin-Gruppen und Coxeter-Gruppen, Invent. Math. 17 (1972) 245–271.
[Ca] A. Castella, Automorphismes et admissibilité dans les groupes de Coxeter et les monoïdes d’Artin–Tits, Thèse, Orsay, 2006.
[Cr] J. Crisp, Injective maps between Artin groups, in: J. Cossey, et al. (Eds.), Geometric Group Theory Down Under, Proceedings
of a Special Year in Geometric Group Theory, Canberra, Australia, July 14–19, 1996, de Gruyter, Berlin, 1999, pp. 119–137.
[CW] A.M. Cohen, D.B. Wales, Linearity of Artin groups of finite type, Israel J. Math. 131 (2002) 101–123.
[De] P. Deligne, Les immeubles des groupes de tresses généralisés, Invent. Math. 17 (1972) 273–302.
[Di] F. Digne, On the linearity of Artin braid groups, J. Algebra 268 (2003) 39–57.
[G] F.A. Garside, The braid group and other groups, Q. J. Math. Oxford (2) 20 (1969) 235–254.
[K1] D. Krammer, The braid group B 4 is linear, Invent. Math. 142 (2000) 451–486.
[K2] D. Krammer, Braid groups are linear, Ann. of Math. 155 (2002) 131–156.
[L] R.J. Lawrence, Homological representations of the Hecke algebra, Comm. Math. Phys. 135 (1990) 141–191.
[Mi] J. Michel, A note on words in braid monoids, J. Algebra 215 (1999) 366–377.
[Mü] B. Mühlherr, Coxeter groups in Coxeter groups, in: Finite Geom. and Combinatorics, Cambridge University Press, 1993,
pp. 277–287.
[P] L. Paris, Artin monoids inject in their groups, Comment. Math. Helv. 77 (2002) 609–637.

2,1982,4

Diunggah oleh

Informasi Dokumen

Hak Cipta

Format Tersedia

Bagikan dokumen Ini

Bagikan atau Tanam Dokumen

Opsi Berbagi

Apakah menurut Anda dokumen ini bermanfaat?

Apakah konten ini tidak pantas?

Hak Cipta:

Format Tersedia

2,1982,4

Diunggah oleh

Hak Cipta:

Format Tersedia

Journal of Algebra

Nicolás Andruskiewitsch Kazuhiko Kurano Aner Shalev

EDITOR OF THE SECTION ON COMPUTATIONAL ALGEBRA

SECTION ON COMPUTATIONAL ALGEBRA

Jon Carlson Meinolf Geck Gunter Malle

FOUNDING EDITOR: Graham Higman, 1964–1984

Contents lists available at ScienceDirect

Congruences on monoids of transformations preserving

Introduction and preliminaries

and its proper normal subgroups are:

See [7] for more details.

1. The monoids POP n and PORn

It is easy to show that E (PORn ) = E (POP n ).

Clearly, h2 = 1 and h is an orientation-reversing full transformation. We showed in [14] that POPI n

As an immediate consequence of this proposition, we have

Corollary 1.3. The monoid POP n is generated by POn ∪ { g }.

Proposition 1.5. Let s and t be elements of M n . Then

(1) sRt if and only if Ker(s) = Ker(t );

with g i 1 xg n−i 2 , h2− j 2 yh j 1 ∈ M n , and so s J t in M n , as required. 2

Since a transformation s ∈ PORn is both orientation-preserving and orientation-reversing if and

Also, as for the monoid PORI n [14, Proposition 5.3], we get

for i ∈ {1, 2, . . . , n − 1}. Consider also the elements u 1 , . . . , un−1 of On deﬁned by

Corollary 1.11. POP n = s0 , . . . , sn−1 , u 1 , . . . , un−1 , g .

Also, as g n−1 u i g = u i +1 , for 1 i n − 2, s0 = g n−1 (s1 g )n−1 and si = g i −1 s1 g n−i +1 , for

Corollary 1.12. POP n = s1 , u 1 , g .

Theorem 1.13. POP n has rank 3.

Next we observe that, given an orientation-reversing partial transformation s, we obtain

Corollary 1.14. PORn = s1 , u 1 , g , h.

Theorem 1.15. For n 3 the monoid PORn has rank 4.

2. Congruences associated to maximal subgroups

In this section we construct a family of congruences associated to maximal subgroups of a J-class

Lemma 2.2. (See [10].) The relation π J is a congruence on S.

bHt implies = xs̃b̃,

aH s implies a t = ãt̃ y . (2)

s ρπ t if and only if sπ J t and s, t ∈ J implies s̃π t̃ .

Theorem 2.3. The relation ρπ is a congruence on S.

v 1 s = x v˜1 s̃π x v˜1 t̃ = v

and so usρπ ut. Similarly, we prove that su ρπ tu, as required. 2

3. On the congruences of OP n , POP n , ORn , PORI n and PORn

It is clear that s and s have the same rank. Moreover:

(a) If s ∈ POn then s ∈ On ;

When T = On , consider the following full transformations of X n :

Notice that hk gki = gkk−i hk , for 1 i k.

(a) If M =  T , g , then H k is the cyclic group of order k generated by gk ; and

(a) s L = σ L , s R = σ R , s L = σ L and s R = σ R , if T ∈ {POI n , POn };

Clearly, s L Rek Ls R and s L s L = ek = s R s R .

Let aq1 , . . . , aqk ∈ {a1 , . . . , ak } be such that b

(a) ŝ = σ , if T ∈ {POI n , POn };

Clearly, ŝ is an inverse of s and it is easy to show that s = s ŝs L s L ss R s R and s L s ŝs L = ek .

Observe that, given s, t ∈ J k such that sHt, we have s L = t L , s R = t R , s L = t L and s R = t R . Moreover,

Lemma 3.1. Let s, t ∈ M be such that s, t , st ∈ J k . Then there exist

Since ( P i − j )st = {t j +1 } if 0 j i − 1 and ( P i − j +k )st = {t j +1 } if i j k − 1, it follows that

and so ( P i − j +1 )st = {t j +1 } if 0 j i and ( P k+i − j +1 )st = {t j +1 } if i + 1 j k − 1.

Proposition 3.2. Let k ∈ {1, . . . , n} and let π be a congruence on H k . Then ρπ is a congruence on M.

Denote by Con( S ) the lattice of the congruences on a semigroup S.

Lemma 3.4. If k = 1 then ρ = 1.

From now on, consider k 2.

Proof. It suﬃces to show that sρ t, for all s, t ∈ I kM−1 . Let s, t ∈ I kM−1 .

u ρ I M g n−i 1 +i 2 vh2− j 1 + j 2 . If g n−i 1 +i 2 vh2− j 1 + j 2 ∈ M \ T then, by (2), we have u ρ g n−i 1 +i 2 vh2− j 1 + j 2 ,

Lemma 3.7. Let s ∈ M. Then there exists an inverse s ∈ M of s such that s s ∈ E ( T ).

(a) t = ξ , if T ∈ {POI n , POn };

It is a routine matter to show that t ∈ M and t is an inverse of t with image D, as required. 2

Lemma 3.9. Let s, t ∈ M be such that sρ t and | Im(s)| k. Then sHt.

(a) z = τ , if T ∈ {POI n , POn };

Corollary 1.11. POP n = s0 , . . . , sn−1 , u 1 , . . . , un−1 , g .

Corollary 1.12. POP n = s1 , u 1 , g .

Corollary 1.14. PORn = s1 , u 1 , g , h.

aH s implies at = ãt̃ y . (2)

(a) If M = T , g , then H k is the cyclic group of order k generated by gk ; and

(a) s L = σ L , s R = σ R , sL = σ L and sR = σ R , if T ∈ {POI n , POn };

Clearly, s L Rek Ls R and s L sL = ek = sR s R .

Clearly, ŝ is an inverse of s and it is easy to show that s = s ŝsL s L ss R sR and s L s ŝsL = ek .

Observe that, given s, t ∈ J k such that sHt, we have s L = t L , s R = t R , sL = t L and sR = t R . Moreover,

s = s ŝsL (s L ss R )sR ρ s ŝsL (s L ts R )sR = t t̂t L (t L tt R )t R = t ,

Ext1A (k, A ⊗ B M )2 = 0.