paper_id
stringlengths 9
16
| version
stringclasses 26
values | yymm
stringclasses 311
values | created
timestamp[s] | title
stringlengths 6
335
| secondary_subfield
sequencelengths 1
8
| abstract
stringlengths 25
3.93k
| primary_subfield
stringclasses 124
values | field
stringclasses 20
values | fulltext
stringlengths 0
2.84M
|
|---|---|---|---|---|---|---|---|---|---|
1301.0761 | 1 | 1301 | 2013-01-04T16:46:24 | Pseudo-multiplications and their properties | [
"math.RA"
] | We examine some properties of pseudo-multiplications, which are a special kind of associative binary relations defined on $\bar{\mathbb{R}}_+ \times \bar{\mathbb{R}}_+$. | math.RA | math |
PSEUDO-MULTIPLICATIONS AND THEIR PROPERTIES
PAUL PONCET
ABSTRACT. We examine some properties of pseudo-multiplications, which
are a special kind of associative binary relations defined on R+ × R+.
1. INTRODUCTION AND MOTIVATIONS
In [4, Chapter I], we considered the idempotent Radon -- Nikodym theo-
rem due to Sugeno and Murofushi [7], and we proved a converse statement
in the special case of the Shilkret integral [5, 3]. We would like to generalize
this converse statement to the idempotent ⊙-integral
Z ∞
B
f ⊙ dν,
where ⊙ is a pseudo-multiplication (see e.g. Sugeno and Murofushi [7],
Benvenuti and Mesiar [1]), i.e. a binary relation satisfying a series of natural
properties. In this note we examine the properties of pseudo-multiplications
we shall need to reach this goal.
2. PSEUDO-MULTIPLICATIONS
We consider a binary relation ⊙ defined on R+ × R+ with the following
properties:
• associativity;
• continuity on (0, ∞) × [0, ∞];
• continuity of the map s 7→ s ⊙ t on (0, ∞], for all t;
• monotonicity in both components;
• existence of a left identity element 1⊙, i.e. 1⊙ ⊙ t = t for all t;
• absence of zero divisors, i.e. s ⊙ t = 0 ⇒ 0 ∈ {s, t}, for all s, t;
• 0 is an annihilator, i.e. 0 ⊙ t = t ⊙ 0 = 0, for all t.
We call such a ⊙ a pseudo-multiplication. Note that the axioms above are
stronger than in [7], where associativity is not assumed.
The map O : R+ → R+ defined by O(t) = inf s>0 s ⊙ t is a kernel in the
sense that, for all s, t:
Date: November 7, 2018.
2010 Mathematics Subject Classification. Primary 03E72; Secondary 49J52.
Key words and phrases. pseudo-arithmetical operations; pseudo-multiplications.
1
• O(s) 6 O(t) whenever s 6 t;
• O(t) 6 t;
• O(O(t)) = O(t).
An element t of R+ is ⊙-finite if O(t) = 0 (and t is ⊙-infinite otherwise).
We conventionally write t ≪⊙ ∞ for a ⊙-finite element t.
Example 2.1. The usual multiplication × is a pseudo-multiplication with
1× = 1 and every element but ∞ is ×-finite; in this case, the idempotent ⊙-
integral specializes to the Shilkret integral [5]. The infimum ∧ is a pseudo-
multiplication with 1∧ = ∞ and every element is ∧-finite; in this case, the
idempotent ⊙-integral specializes to the Sugeno integral [6].
Lemma 2.2. Given a pseudo-multiplication ⊙, there exists some positive
⊙-finite element if and only if 1⊙ is ⊙-finite.
Proof. Assume that O(t) = 0 for some t > 0. Then O(1⊙) ⊙ t 6 s ⊙ 1⊙ ⊙
t = s ⊙ t for all s > 0, so that O(1⊙) ⊙ t 6 O(t) = 0. Since ⊙ has no zero
divisors, this implies that O(1⊙) = 0, i.e. 1⊙ is ⊙-finite.
(cid:3)
If O(1⊙) = 0, we say that the pseudo-multiplication ⊙ is non-degenerate.
This amounts to say that the set of ⊙-finite elements differs from {0}.
Example 2.3. The multiplication × and the infimum ∧ are non-degenerate
pseudo-multiplications. The binary relation ⊗ defined by s ⊗ t = t if s > 0
and 0 ⊗ t = 0 is an example of degenerate pseudo-multiplication.
Proposition 2.4. Given a non-degenerate pseudo-multiplication ⊙, the fol-
lowing conditions are equivalent for an element t ∈ R+:
• t is ⊙-finite;
• s ⊙ t ≪⊙ ∞ for some s > 0;
• s ⊙ t 6 1⊙ for some s > 0;
• t ⊙ s′ 6 1⊙ for some s′ > 0;
• t ⊙ s′ ≪⊙ ∞ for some s′ > 0.
Proof. Assume that t is ⊙-finite. Then s ⊙ t ≪⊙ ∞ with s = 1⊙.
Assume that s ⊙ t ≪⊙ ∞ for some s > 0. Then lims′→0+ s′ ⊙ s ⊙ t = 0,
so there exists s′ > 0 such that (s′ ⊙ s) ⊙ t 6 1⊙.
Assume that s ⊙ t 6 1⊙ for some s > 0. Then s ⊙ t ⊙ s′ 6 1⊙ ⊙ s′ = s′
for all s′ > 0, so that lims′→0+ s ⊙ t ⊙ s′ = 0. By continuity and the fact
that ⊙ has no zero divisors we get lims′→0+ t ⊙ s′ = 0. This implies that
t ⊙ s′ 6 1⊙ for some s′ > 0.
Assume that t ⊙ s′ ≪⊙ ∞ for some s′ > 0. Then lims→0+ s ⊙ t ⊙ s′ = 0,
so that O(t) ⊙ s′ = 0. This shows that O(t) = 0, i.e. t is ⊙-finite.
(cid:3)
Lemma 2.5. Given a pseudo-multiplication ⊙, the set of ⊙-finite elements
is either {0}, or [0, ∞], or of the form [0, φ) for some φ ∈ (1⊙, ∞] such that
O(φ) = φ.
2
Proof. Denote by F⊙ the set of ⊙-finite elements, and assume that F⊙ 6=
{0}. Then, by Lemma 2.2, F⊙ is an interval containing 0 and 1⊙, so it
satisfies [0, φ) ⊂ F⊙ ⊂ [0, φ] for some φ > 1⊙.
If φ = ∞ we have
F⊙ = [0, ∞) or F⊙ = [0, ∞]. In the case where F⊙ = [0, ∞) the lemma
asserts also that O(φ) = φ. We have O(φ) 6 ∞ = φ. Suppose that
O(φ) < φ. Then O(φ) ∈ [0, φ) ⊂ F⊙, so O(φ) is ⊙-finite. This means that
O(O(φ)) = 0, hence O(φ) = 0. So φ is ⊙-finite, a contradiction.
n) = t′
n, so that s ⊙ t′
n = t′
n > 0, t′
n) = O(O(tn)) = O(tn) = t′
n 6 tn for all n. This implies that t′
Now suppose that φ < ∞. Since φ is the supremum of F⊙, there exists
a decreasing sequence (tn) of ⊙-infinite elements that tends to φ. Let t′
n =
n is ⊙-infinite, hence
O(tn). Since O(t′
n → φ when n → ∞. Again,
φ 6 t′
n for all 0 < s < 1⊙ and all n. With n → ∞,
O(t′
we obtain s ⊙ φ = φ for all 0 < s < 1⊙. This shows that O(φ) = φ > 0, so
φ is ⊙-infinite, and F⊙ = [0, φ).
(cid:3)
Example 2.6. We have F× = [0, ∞), F∧ = [0, ∞], and F⊗ = {0} (see the
definition of ⊗ in Example 2.3). Given 0 < φ < ∞ we can also build an
example of pseudo-multiplication ⊙φ with F⊙φ = [0, φ) as follows. If tanh
denotes the hyperbolic tangent, we define s ⊙φ t = max(s, t) if s > φ or
t > φ, and s ⊙φ t = φ tanh(tanh−1(s/φ) tanh−1(t/φ)) otherwise.
Lemma 2.7. A pseudo-multiplication ⊙ is non-degenerate if and only if the
monoid ([0, 1⊙], ⊙) is commutative.
Proof. Assume that ([0, 1⊙], ⊙) is commutative. Then
O(1⊙) = inf
s∈(0,1⊙]
s ⊙ 1⊙ = inf
s∈(0,1⊙]
1⊙ ⊙ s
= inf
s∈(0,1⊙]
s = 0.
This shows that ⊙ is non-degenerate.
Conversely, assume that ⊙ is non-degenerate. We show that, in [0, 1⊙],
the binary relation ⊙ is continuous; this will imply, by Faucett's theorem
[2, Lemma 5], that ([0, 1⊙], ⊙) is commutative. So let sn → s and tn → t
in [0, 1⊙]. We prove that sn ⊙ tn → s ⊙ t. If s > 0 this is a consequence
of the properties of the pseudo-multiplication ⊙, so suppose that s = 0,
and let us show that sn ⊙ tn → 0. For this purpose, let ε > 0, and let
n = sn ⊕ ε. Since s′
n ⊙ tn tends to ε ⊙ t.
s′
But s′
n ⊙ tn = (sn ⊕ ε) ⊙ tn = (sn ⊙ tn) ⊕ (ε ⊙ tn), and taking the limit
superior gives ε ⊙ t = ℓ ⊕ (ε ⊙ t), where ℓ := lim sup(sn ⊙ tn). This shows
that ℓ 6 ε ⊙ t, for all ε > 0. In other words, ℓ 6 O(t). Moreover, we are
in the non-degenerate case, and t 6 1⊙, so that O(t) = 0. Hence, ℓ = 0, so
that lim(sn ⊙ tn) = 0, which is the desired result.
(cid:3)
n tends to ε > 0, the sequence s′
We summarize the previous lemmata in a single theorem:
3
Theorem 2.8. Given a pseudo-multiplication ⊙, the following conditions
are equivalent:
• ⊙ is non-degenerate, i.e. 1⊙ is ⊙-finite;
• there exists some positive ⊙-finite element;
• the monoid ([0, 1⊙], ⊙) is commutative;
• the set F⊙ of ⊙-finite elements is either [0, ∞] or of the form [0, φ)
for some φ ∈ (1⊙, ∞].
Moreover, in the case where F⊙ = [0, φ), then φ satisfies O(φ) = φ and
t ⊙ φ = φ ⊙ t = φ, for all 0 < t 6 φ. In particular, φ is idempotent.
Proof. Let 0 < t < φ. Since t is ⊙-finite, there is some s′ > 0 such that
t ⊙ s′ 6 1⊙ by Proposition 2.4. Thus, φ = O(φ) 6 t ⊙ φ = t ⊙ O(φ) 6
t⊙(s′ ⊙φ) = (t⊙s′)⊙φ 6 1⊙ ⊙φ = φ. We obtain that t⊙φ = φ, and when
t → φ we get φ ⊙ φ = φ, i.e. φ is idempotent. Using again Proposition 2.4,
the fact that φ be ⊙-infinite implies φ ⊙ s > φ for all s > 0. Hence, if
0 < t < φ, we get φ = φ ⊙ φ > φ ⊙ t > φ, i.e. φ ⊙ t = φ.
(cid:3)
Corollary 2.9. Given a pseudo-multiplication ⊙, it is not possible to find
t < φ and t′ > φ such that t ⊙ t′ = φ, if φ denotes the supremum of the set
of ⊙-finite elements.
Proof. Again, let F⊙ be the set of ⊙-finite elements. The result is clear if
F⊙ = {0} or F⊙ = [0, ∞]. Now if F⊙ = [0, φ) for some φ ∈ (1⊙, ∞], we
write φ = t ⊙ t′ with t < φ. Then φ = φ ⊙ φ = φ ⊙ (t ⊙ t′) = (φ ⊙ t) ⊙ t′ =
φ ⊙ t′ by Theorem 2.8. Since φ > 1⊙, this implies φ > 1⊙ ⊙ t′ = t′.
(cid:3)
REFERENCES
[1] Pietro Benvenuti and Radko Mesiar. Pseudo-arithmetical operations as a basis for the
general measure and integration theory. Inform. Sci., 160(1-4):1 -- 11, 2004.
[2] W. M. Faucett. Compact semigroups irreducibly connected between two idempotents.
Proc. Amer. Math. Soc., 6:741 -- 747, 1955.
[3] Victor P. Maslov. Méthodes opératorielles. Éditions Mir, Moscow, 1987. Translated
from the Russian by Djilali Embarek.
[4] Paul Poncet. Infinite-dimensional idempotent analysis: the role of continuous posets.
PhD thesis, École Polytechnique, Palaiseau, France, 2011.
[5] Niel Shilkret. Maxitive measure and integration. Nederl. Akad. Wetensch. Proc. Ser.
A 74 = Indag. Math., 33:109 -- 116, 1971.
[6] Michio Sugeno. Theory of fuzzy integrals and its applications. PhD thesis, Tokyo
Institute of Technology, Japan, 1974.
[7] Michio Sugeno and Toshiaki Murofushi. Pseudo-additive measures and integrals. J.
Math. Anal. Appl., 122(1):197 -- 222, 1987.
CMAP, ÉCOLE POLYTECHNIQUE, ROUTE DE SACLAY, 91128 PALAISEAU CEDEX,
FRANCE, AND INRIA, SACLAY -- ÎLE-DE-FRANCE
E-mail address: [email protected]
4
|
1310.8399 | 1 | 1310 | 2013-10-31T06:39:53 | Absolute Algebra III - The saturated spectrum | [
"math.RA"
] | Let B1 denote the set {0,1} with the usual operations except that $1+1=1$, in other words, the smallest characteristic 1 semifield . We compare two possible analogues of the notion of prime ideal for B1--algebras. We then consider the relations between these notions and Deitmar's theory of F1--schemes. | math.RA | math | Absolute algebra
III -- The saturated spectrum
Paul Lescot
LMRS
CNRS UMR 6085
UFR des Sciences et Techniques
Universit´e de Rouen
Avenue de l'Universit´e BP12
76801 Saint-Etienne du Rouvray (FRANCE)
3
1
0
2
t
c
O
1
3
]
.
A
R
h
t
a
m
[
1
v
9
9
3
8
.
0
1
3
1
:
v
i
X
r
a
Abstract
We investigate the algebraic and topological preliminaries to a geometry in
characteristic 1.
Keywords:
Characteristic one, spectra, Zariski topologies
2000 MSC: 06F05, 20M12, 08C99
Email address: [email protected] ()
URL: http://www.univ-rouen.fr/LMRS/Persopage/Lescot/ ()
Preprint submitted to Journal of Pure and Applied Algebra
April 6, 2021
1. Introduction
The theory of characteristic 1 semirings (i.e. semirings with 1 + 1 = 1)
originated in many different contexts : pure algebra (see e.g. LaGrassa's
PhD thesis [8]), idempotent analysis and the study of Rmax
([1, 3]), and
Zhu's theory ([12]), itself inspired by considerations of Hopf algebras (see
[11]). Its main motivation is now the Riemann Hypothesis, via adeles and
the theory of hyperrings (cf. [2, 3, 4], notably §6 from [4]).
+
For example, it has by now become clear (see [4],Theorem 3.11) that the
classification of finite hyperfield extensions of the Krasner hyperring K is
one of the main problems of the theory. If H denotes an hyperring extension
of K, B1 the smallest characteristic one semifield and S the sign hyperring,
then there are canonical mappings B1 → S → K → H, whence mappings
Spec(H) → Spec(K) → Spec(S) → Spec(B1) ,
thus Spec(H) "lies over"Spec(B1) (see [4], §6, notably diagram (43), where
B1 is denoted by B).
The ultimate goal of our investigations is to provide a proper algebraic
geometry in characteristic one. The natural procedure is to construct "affine
B1 -- schemes"and endow them with an appropriate topology and a sheaf of
semirings ; a suitable glueing procedure will then produce general "B1 --
schemes". This program is not yet completed ; in this paper, we deal with
a natural first step : the extension to B1 -- algebras of the notions of spec-
trum and Zariski topology, and the fundamental topological properties of
these objects. In order to construct a structure sheaf over the spectrum of a
B1 -- algebra, Castella's localization procedure ([1]) will probably be useful.
As in our two previous papers, we work in the context of B1 -- algebras,
characteristic one semirings. For such an A, one may define prime
i.e.
ideals by analogy to classical commutative algebra. In order to define the
spectrum of a B1 -- algebra A, two candidates readily suggest themselves : the
set Spec(A) of prime (in a suitable sense) congruences, and the set P r(A) of
prime ideals ; in contrast to the classical situation, these two approaches are
not equivalent. In fact both sets may be equipped with a natural topology
of Zariski type (see [10], Theorem 2.4 and Proposition 3.15), but they do not
in general correspond bijectively to one another ; nevertheless, the subset
P rs(A) ⊆ P r(A) of saturated prime ideals is in natural bijection with the set
of excellent prime congruences (see below) on A.
2
It turns out (§3) that there is another, far less obvious, bijection between
P rs(A) and the maximal spectrum MaxSpec(A) ⊆ Spec(A) of A. This map-
ping is actually an homeomorphism for the natural (Zariski -- type) topologies
mentioned above. As a by -- product, we find a new point of view on the
descrption of the maximal spectrum of the polynomial algebra B1[x1, ..., xn]
found in [9] and [12]. The homeomorphism in question is actually functorial
in A (§4).
In §5, we show that the theory of the nilradical and of the root of an
ideal carry over, with some precautions, to our setting ; the situation is even
better when one restricts oneself to saturated ideals. This allows us, in §6, to
establish some nice topological properties of
MaxSpec(A) ≃ P rs(A) ;
namely, it is T0 and quasi -- compact (Theorem 6.1), and the open quasi --
compact sets constitute a basis stable under finite intersections. Furthermore
this space is sober , i.e. each irreducible closed set has a (necessarily unique)
generic point. In other words, P rs(A) satisfies the usual properties of a ring
spectrum that are used in algebraic geometry (see e.g. the canonical reference
[6]): P rs(A) is a spectral space in the sense of Hochster([7]).
In the last paragraph, we discuss the particular case of a monogenic B1 --
algebra, that is, a quotient of the polynomial algebra B1[x] ; in [9], we had
listed the smallest finite such algebras.
In a subsequent work I shall investigate how higher concepts and meth-
ods of commutative algebra (minimal prime ideals, zero divisors, primary
decomposition) carry over to characteristic one semirings.
3
2. Definitions and notation
We shall review some the definitions and notation of our previous two
papers ([9], [10]).
B1 = {0, 1} denotes the smallest characteristic one semifield ; the oper-
ations of addition and multiplication are the obvious ones, with the slight
change that
1 + 1 = 1 .
A B1 -- module M is a nonempty set equipped with an action
B1 × M → M
satisfying the usual axioms (see [9], Definition 2.3); as first seen in [12],
Proposition 1 (see also [9], Theorem 2.5), B1 -- modules can be canonically
identified with ordered sets having a smallest element (0) and in which any
two elements a and b have a least upper bound (a + b). In particular, one
may identify finite B1 -- modules and nonempty finite lattices.
A (commutative) B1 -- algebra is a B1 -- module equipped with an associa-
tive multiplication that has a neutral element and satisfies the usual axioms
relative to addition (see [9], Definition 4.1). In the sequel, except when oth-
erwise indicated, A will denote a B1 -- algebra.
An ideal I of A is by definition a subset containing 0, stable under addi-
tion, and having the property that
∀x ∈ A ∀y ∈ I xy ∈ I ;
I is termed prime if I 6= A and
ab ∈ I =⇒ a ∈ I or b ∈ I .
By a congruence on A, we mean an equivalence relation on A compatible
with the operations of addition and multiplication. The trivial congruence
C0(A) is characterized by the fact that any two elements of A are equivalent
under it ; the congruences are naturally ordered by inclusion, and
MaxSpec(A)
will denote the set of maximal nontrivial congruences on A.
For R a congruence on A, we set
I(R) := {x ∈ Ax R 0} ;
4
it is an ideal of A
A nontrivial congruence R is termed prime if
ab R 0 =⇒ a R 0 or b R 0 ;
the set of prime congruences on A is denoted by Spec(A). It turns out that
(see [10], Proposition 2.3)
MaxSpec(A) ⊆ Spec(A) .
For J an ideal of A, there is a unique smallest congruence RJ such that
J ⊆ I(R) ; it is denoted by RJ . Such congruences are termed excellent .
An ideal J of A is termed saturated if it is of the form I(R) for some
congruence R ; this is the case if and only if J = J, where
J := I(RJ ) .
We shall denote the set of prime ideals of A by P r(A), and the set of
saturated prime ideals by P rs(A).
For S ⊆ A, let us set
W (S) := {P ∈ P r(A)S ⊆ P} ,
and
V (S) := {R ∈ Spec(A)S ⊆ I(R)} .
As seen in [10], Theorem 2.4 and Proposition 3.4, the family (W (S))S⊆A is
the family of closed sets for a topology on P r(A), and the family (V (S))S⊆A is
the family of closed sets for a topology on Spec(A). We shall always consider
Spec(A) and P r(A) as equipped with these topologies, and their subsets with
the induced topologies.
For M a commutative monoid, we define the Deitmar spectrum SpecD(M)
as the set of prime ideals (including ∅) of M (in [5], this is denoted by
Spec FM ). We define F (M) = B1[M] as the "monoid algebra of M over
B1 "; the functor F is adjoint to the forgetful functor from the category of
B1 -- algebras to the category of monoids (for the details, see [9], §5). Fur-
thermore, there is an explicit canonical bijection between SpecD(M) and a
certain subset of Spec(F (M)) (see [10], Theorem 4.2).
For S a subset of A, let < S > denote the intersection of all the ideals
of A containing S (there is always at least one such ideal : A itself). It is
5
clear that < S > is an ideal of A, and therefore is the smallest ideal of A
containing S. As in ring theory, one may see that
< S >= {
nXj=1
ajsjn ∈ N, (a1, ..., an) ∈ An, (s1, ..., sn) ∈ Sn}.
We shall denote by SP the category whose objects are spectra of B1 --
algebras and whose morphisms are the continuous maps between them.
6
3. A new description of maximal congruences
Let A denote a B1 -- algebra.
For P a saturated prime ideal of A, let us define a relation SP on A by :
xSPy ≡ (x ∈ P and y ∈ P) or(x /∈ P and y /∈ P) .
Then SP is a congruence on A : if xSP y and x′SP y′, then one and only one
of the following holds :
(i) x ∈ P, y ∈ P, x′ ∈ P and y ′ ∈ P ,
(ii) x ∈ P, y ∈ P, x′ /∈ P and y ′ /∈ P ,
(iii) x /∈ P, y /∈ P, x′ ∈ P and y ′ ∈ P ,
(iv) x /∈ P, y /∈ P, x′ /∈ P and y ′ /∈ P .
In case (i), x + x′ ∈ P and y + y ′ ∈ P, whence x + x′SPy + y ′ ; in
cases (ii) and (iv), x + x′ /∈ P and y + y ′ /∈ P (as P is saturated), whence
x + x′SPy + y ′. Case (iii) is symmetrical relatively to case (ii), therefore, in
all cases, x + x′SPy + y ′ : SP is compatible with addition.
In cases (i), (ii) and (iii), xx′ ∈ P and yy ′ ∈ P, whence xx′SP yy ′ ; in
case (iv) xx′ /∈ P and yy ′ /∈ P (as P is prime), whence also xx′SPyy ′ : SP is
compatible with multiplication, hence is a congruence on A.
As 0 ∈ P and 1 /∈ P, 0 6 SP 1, therefore SP is nontrivial ; but each x ∈ A
is either in P (whence xSP 0) or not (whence xSP 1). It follows that
A
SP
= {¯0, ¯1} ≃ B1 ;
in particular, SP is maximal : SP ∈ MaxSpec(A).
Obviously, I(SP) = P.
Furthermore, let (x, y) ∈ A2 be such that xRP y ; then there is z ∈ P
such that x + z = y + z. If x ∈ P then y + z = x + z ∈ P, whence y ∈ P
(as y + (y + z) = y + z and P is saturated) ; symmetrically, y ∈ P implies
x ∈ P, whence the assertions (x ∈ P) and (y ∈ P) are equivalent, and xSPy.
We have shown that
We shall denote by αA the mapping
RP ≤ SP .
αA : P rs(A) → MaxSpec(A)
P 7→ SP .
7
Let R ∈ MaxSpec(A) ; then R ∈ Spec(A), whence I(R) is prime ; by
Theorem 3.8 of [10], it is saturated, i.e. I(R) ∈ P rs(A). Let us set
βA(R) := I(R) .
Theorem 3.1. The mappings
and
αA : P rs(A) 7→ MaxSpec(A)
βA : MaxSpec(A) 7→ P rs(A)
are bijections, inverse of one another. They are continuous for the topologies
on P rs(A) and MaxSpec(A) induced by the topologies on P r(A) and Spec(A)
mentioned above, whence P rs(A) and MaxSpec(A) are homeomorphic.
Proof. Let R ∈ MaxSpec(A) ; then
αA(βA(R)) = αA(I(R)) = SI(R) .
Let us assume xRy ; then, if x ∈ I(R) one has xR0, whence yR0 and
y ∈ I(R); by symmetry, y ∈ I(R) implies x ∈ I(R), thus (x ∈ I(R)) and
(y ∈ I(R)) are equivalent, i.e. xSI(R)y. We have proved that R ≤ SI(R). As
R is maximal, we have R = SI(R), whence
and
αA(βA(R)) = SI(R) = R ,
αA ◦ βA = IdM axSpec(A) .
Let now P ∈ P rs(A) ; then
(βA ◦ αA)(P) = βA(αA(P))
= βA(SP)
= I(SP)
= P ,
βA ◦ αA = IdP rs(A) ,
8
whence
and the first statement follows.
Let now F denote a closed subset of P rs(A) ; then F = G ∩ P rs(A) for
G a closed subset of P r(A) and G = W (S) := {P ∈ P r(A)S ⊆ P} for some
subset S of A. But then, for R ∈ MaxSpec(A), R ∈ β−1
A (F ) if and only if
βA(R) ∈ F , i.e. I(R) ∈ G ∩ P rs(A), that is I(R) ∈ G, or S ⊆ I(R), which
means R ∈ V (S). Thus
β−1
A (F ) = V (S) ∩ MaxSpec(A)
is closed in MaxSpec(A). We have shown the continuity of βA.
Let now H ⊆ MaxSpec(A) be closed ; then H = MaxSpec(A) ∩ L for
some closed subset L of Spec(A), and L = V (T ) for some subset T of A. Then
a saturated prime ideal P of A belongs to α−1
A (H) if and only if αA(P) ∈ H,
that is
i.e.
SP ∈ MaxSpec(A) ∩ L ,
SP ∈ V (T )
or T ⊆ I(SP). But I(SP) = P whence P belongs to α−1
T ⊆ P, that is
A (H) if and only if
α−1
A (H) = W (T ) ∩ P rs(A) ,
which is closed in P rs(A).
Let us consider the special case in which A is in the image of F : A =
F (M), for M a commutative monoid. Let P be a prime ideal of M ; as seen
in [10], Theorem 4.2, P is a saturated prime ideal in A, and one obtains in
this way a bijection between SpecD(M) and P rs(A). The following is now
obvious :
Theorem 3.2. The mapping
ψM : SpecD(M) → MaxSpec(F (M))
P 7→ αF (M )( P )
is a bijection.
Two particular cases are of special interest :
1. M is a group ; then SpecD(M) = {∅}, whence MaxSpec(F (G)) has
exactly one element.
9
2. M = Cn :=< x1, ..., xn > is the free monoid on n variables x1, ..., xn.
Then the elements of SpecD(M) are the (PJ)J⊆{1,...,n}, where
PJ := [j∈J
xjCn
(a fact that was already used in [10], Example 4.3). Then
ψM (PJ ) = αF (M )( PJ ) = S PJ
whence xψM (PJ)y if and only if either (x ∈ PJ and y ∈ PJ ) or (x /∈ PJ
and y /∈ PJ ). But we have seen in [9], Theorem 4.5, that
F (M) = B1[x1, ..., xn]
could be identified with the set of finite formal sums of elements of M.
Obviously, an element x of F (M) belongs to PJ if and only if at least
one of its components involves at least one factor xj(j ∈ J). It is now
clear that, using the notation of [9], Definition 4.6 and Theorem 4.7,
We hereby recover the description of MaxSpec(B1[x1, ..., xn]) given in
[9](Theorems 4.7, 4.8 and 4.10).
ψM (PJ ) =eJ .
The following result will be useful
Theorem 3.3. Any proper saturated ideal of a B1 -- algebra A is contained in
a saturated prime ideal of A.
Proof. Let J be a proper saturated ideal of A ; as I(RJ ) = J = J 6= A, RJ 6=
C0(A). By Zorn's Lemma, one has RJ ≤ R for some R ∈ MaxSpec(A).
According to Theorem 2.1, R = αA(P) = SP for a saturated prime ideal P
of A, therefore RJ ≤ SP and
J = J = I(RJ ) ⊆ I(SP) = P .
10
4. Functorial properties of spectra
Let ϕ : A → C denote a morphism of B1 -- algebras, and let R ∈ Spec(C).
We define a binary relation ϕ(R) on A by :
′
∀(a, a
) ∈ A2 a ϕ(R)a
′
≡ ϕ(a)Rϕ(a
′
) .
It is clear that ϕ(R) is a congruence on A, and that
I( ϕ(R)) = ϕ−1(I(R)) .
In particular I( ϕ(R)) is a prime ideal of A, hence ϕ(R) ∈ Spec(A) : ϕ
maps Spec(C) into Spec(A). Let F := V (S) be a closed subset of Spec(A),
and let R ∈ Spec(C) ; then R ∈ ϕ−1(F ) if and only if ϕ(R) ∈ F , that
is S ⊆ I( ϕ(R)), or S ⊆ ϕ−1(I(R)), i.e. ϕ(S) ⊆ I(R), or R ∈ V (ϕ(S)).
Therefore ϕ−1(F ) = V (ϕ(S)) is closed in Spec(C) : ϕ is continuous.
Furthermore, for ϕ : A → C and ψ : C → D one has
]ψ ◦ ϕ = ϕ ◦ ψ : Spec(D) → Spec(A) .
It follows that the equations H(A) = Spec(A) and H(ϕ) = ϕ define a con-
travariant functor H from Za to SP.
Let J denote an ideal in C, and let us assume aRϕ−1(J)a′ ; then there is
an x ∈ ϕ−1(J) with a + x = a′ + x. Now ϕ(x) ∈ J and
ϕ(a) + ϕ(x) = ϕ(a + x)
+ x)
) + ϕ(x) ,
= ϕ(a
= ϕ(a
′
′
whence ϕ(a)RJ ϕ(a′) and a ϕ(RJ )a′. We have established
Proposition 4.1. Let A and C denote B1 -- algebras, ϕ : A → C a morphism
and J an ideal of C : then
Rϕ−1(J) ≤ ϕ(RJ ) .
Theorem 4.2. Let A and C denote two B1 -- algebras, and ϕ : A → C a mor-
phism. Then ϕ : Spec(C) → Spec(A) maps MaxSpec(C) into MaxSpec(A),
and the diagram
P rs(C)
↓αC
ϕ−1
→
P rs(A)
↓αA
MaxSpec(C)
ϕ→ MaxSpec(A)
commutes.
11
Proof. Let P ∈ P rs(C), then, for all (a, a′) ∈ A2
a ϕ(SP)a
′
or (ϕ(a) /∈ P and ϕ(a
′
) /∈ P)
or (a /∈ ϕ−1(P) and a
′
/∈ ϕ−1(P))
Therefore
′
⇐⇒ ϕ(a)SP ϕ(a
⇐⇒ (ϕ(a) ∈ P and ϕ(a
)
′
) ∈ P))
⇐⇒ (a ∈ ϕ−1(P) and a
′
∈ ϕ−1(P))
⇐⇒ aSϕ−1(P)a
′
.
( ϕ ◦ αC)(P) = ϕ(αC(P))
= ϕ(SP )
= Sϕ−1(P)
= αA(ϕ−1(P))
= (αA ◦ ϕ−1)(P)
whence ϕ ◦ αC = αA ◦ ϕ−1 .
Incidentally we have proved that ϕ maps MaxSpec(C) = αC(P rs(C))
into αA(P rs(A)) = MaxSpec(A), i.e. the first assertion.
12
5. Nilpotent radicals and prime ideals
The usual theory generalizes without major problem to B1 -- algebras.
Theorem 5.1. In the B1 -- algebra A,let us define
Nil(A) := {x ∈ A(∃n ≥ 1)xn = 0} .
Then Nil(A) is a saturated ideal of A, and one has
\P∈P r(A)
P = \P∈P rs(A)
P = Nil(A) .
Proof. Let M := TP∈P r(A) P and N = TP∈P rs(A) P.
If x ∈ Nil(A) and
P ∈ P r(A), then, for some n ≥ 1, xn = 0 ∈ P, whence (as P is prime) x ∈ P
: Nil(A) ⊆ M.
As P rs(A) ⊆ P r(A), we have M ⊆ N.
Let now x /∈ Nil(A) ; then
(∀n ∈ N) xn 6= 0 .
Define
E := {J ∈ Ids(A)(∀n ≥ 0)xn /∈ J}.
This set is nonempty ({0} ∈ E) and inductive for ⊆, therefore, by Zorn's
Lemma, there exists a maximal element P of E. As 1 = x0 /∈ P, P 6= A.
Let us assume ab ∈ P, a /∈ P and b /∈ P ; then P + Aa and P + Ab are
saturated ideals of A strictly containing P, whence there exists two integers m
and n with xm ∈ P + Aa and xn ∈ P + Ab. By definition of the closure of an
ideal,
there
are u = p1 + λa ∈ P + Aa and v = p2 + µb ∈ P + Ab such that xm + u = u
and xn + v = v. Then
and
ub = p1b + λ(ab) ∈ P
xmb + ub = (xm + u)b = ub ,
whence, as P is saturated, xmb ∈ P.
Then
xmv = xmp2 + µxmb ∈ P ;
13
as
xm+n + xmv = xm(xn + v)
= xmv ,
we obtain xm+n ∈ P, a contradiction.
Therefore P is prime and saturated and x = x1 /∈ P, whence x /∈ N. We
have proved that N ⊆ Nil(A), whence M = N = Nil(A).
Corollary 5.2.
Proof.
Nil(A) = \P∈P r(A)
P .
P (by Theorem 5.1)
Nil(A) = \P∈P r(A)
⊆ \P∈P r(A)
⊆ \P∈P rs(A)
= \P∈P rs(A)
P
P
P
= Nil(A) (also by Theorem 5.1).
Definition 5.3. For I an ideal of A, we define the root r(I) of I by
r(I) := {x ∈ A(∃n ≥ 1)xn ∈ I}.
Lemma 5.4.
(i) r(I) is an ideal of A.
(ii) r(I) ⊆ r(I) ; in particular, if I is saturated then so is r(I).
(iii) r({0}) = Nil(A).
Proof.
(i) Obviously, 0 ∈ r(I).
14
If x ∈ r(I) and y ∈ r(I), then xm ∈ I for some m ≥ 1 and yn ∈ I for
some n ≥ 1, whence
(cid:18)m + n − 1
j
(cid:19)xjym+n−1−j
(x + y)m+n−1 =
(
m+n−1Xj=0
m+n−1Xj=0
=
xjym+n−1−j)
∈ I ,
as xj ∈ I for j ≥ m and ym+n−1−j ∈ I for j ≤ m − 1 (as, then,
m + n − 1 − j ≥ n). Thus x + y ∈ r(I).
For a ∈ A, (ax)m = amxm ∈ I, whence ax ∈ r(I). Therefore r(I) is an
ideal of A.
(ii) Let x ∈ r(I) then there is u ∈ r(I) such that x + u = u, and there is
n ≥ 1 such that un ∈ I. Let us show by induction on j ∈ {0, ..., n}
that un−jxj ∈ I. This is clear for j = 0. Let then j ∈ {0, ..., n − 1},
and assume that un−jxj ∈ I ; then
un−j−1xj+1 + un−jxj = un−j−1xj(x + u)
= un−j−1xju
= un−jxj ,
whence un−j−1xj+1 ∈ I = I . Thus, for j = n, we obtain
xn = un−nxn ∈ I ,
whence x ∈ r(I).
If now I is saturated, then
r(I) ⊆ r(I)
⊆ r(I) (by the above)
= r(I) ,
whence r(I) = r(I) is saturated.
(iii) That assertion is obvious.
15
Proposition 5.5. For each saturated ideal I of the B1 -- algebra A , one has
r(I) = \P∈P rs(A);I⊆P
P .
Remark 5.6. For I = {0}, this is part of Theorem 5.1.
Proof. Let x ∈ r(I), and let P ∈ P rs(A) with I ⊆ P ; then, for some n ≥ 1
xn ∈ I, whence xn ∈ P and x ∈ P :
r(I) ⊆ \P∈P rs(A);I⊆P
P .
Let now y ∈ A, y /∈ r(I), and denote by π the canonical projection
π : A ։ A0 :=
A
RI
.
∀n ≥ 1 yn /∈ I ,
∀n ≥ 1yn 6 RI 0 ,
As I is saturated, one has
whence
or
∀n ≥ 1 π(y)n = π(yn) 6= 0 .
Therefore π(y) /∈ Nil(A0), whence, according to Theorem 5.1, there exists a
saturated prime ideal P0 of A0 such that π(y) /∈ P0. But then P := π−1(P0)
is a saturated prime ideal of A containing I with y /∈ P, whence
y /∈ \P∈P rs(A);I⊆P
P .
16
6. Topology of spectra
We can now establish the basic topological properties of the spectra
P rs(A) (analogous, in our setting, to Corollary 1.1.8 and Proposition 1.1.10(ii)
of [6]).
Theorem 6.1. P rs(A) and MaxSpec(A) are T0 and quasi -- compact.
Proof. According to Theorem 3.1, P rs(A) and MaxSpec(A) are homeomor-
phic, therefore it is enough to establish the result for P rs(A).
Let P and Q denote two different points of P rs(A) ; then either P * Q
or Q * P. Let us for instance assume that P * Q ; then Q /∈ W (P) ; set
O := P rs(A) ∩ (P r(A) \ W (P)) .
Then O is an open set in P rs(A), Q ∈ O and, obviously, P /∈ O. Therefore
P rs(A) is T0.
Let (Ui)i∈I denote an open cover of P rs(A) :
P rs(A) =[i∈I
Ui ;
each P rs(A)\Ui is closed, whence P rs(A)\Ui = P rs(A)∩W (Si) for some sub-
set Si of A. Therefore P rs(A)∩(Ti∈I W (Si)) = ∅, i.e. P rs(A)∩W (Si∈I Si) =
∅. Therefore P rs(A) ∩ W (<Si∈I Si >) = ∅, whence, according to Theorem
3.3, <Si∈I Si > = A. Let J =< Si∈I Si > ; then 1 ∈ J, hence there is
x ∈ J such that 1 + x = x. Furthermore, there exist n ∈ N, (i1, ..., in) ∈ I n
, xik ∈ Sik and (a1, ..., an) ∈ An such that x = a1xi1 + ... + anxin. But then
1 + a1xi1 + ... + anxin = a1xi1 + ... + anxin
whence
and
It follows that
1 ∈ < {xi1, ..., xin} > ⊆
n[j=1
Sij
n[j=1
Sij = A .
P rs(A) ∩ W (
n[j=1
Sij ) = ∅ ,
17
that is
or
P rs(A) ∩
n\j=1
W (Sij ) = ∅ ,
P rs(A) =
n[j=1
Uij
:
P rs(A) is quasi -- compact.
For f ∈ A, let
D(f )
:= P rs(A) \ (P rs(A) ∩ W ({f }))
= {P ∈ P rs(A)f /∈ P}.
Proposition 6.2.
1. Each D(f )(f ∈ A) is open and quasi -- compact in
P rs(A) (see [6], Proposition 1.1.10 (ii)).
2. The family (D(f ))f ∈A is an open basis for P rs(A) (see [6], Proposition
1.1.10(i)); in particular, the open quasi -- compact sets constitute an open
basis.
3. A subset O of P rs(A) is open and quasi -- compact if and only if it is of
the form P rs(A) ∩ W (I) for I an ideal of finite type in A.
4. The family of open quasi -- compact subsets of P rs(A) is stable under
finite intersections.
5. Each irreducible closed set in P rs(A) has a unique generic point (see
[6], Corollary 1.1.14(ii)).
Proof.
1. The openness of D(f ) is obvious.
Let us assume D(f ) = Si∈I Ui, where the Ui's are open sets in D(f ).
Each Ui can be written as
Ui = D(f ) ∩ Vi ,
for Vi an open set in P rs(A), i.e. P rs(A) \ Vi = W (Si) for Si a subset
of A. Then
whence
Vi = P rs(A) \ (\i∈I
D(f ) ⊆[i∈I
P rs(A) ∩ W ([i∈I
18
W (Si)) ,
Si) ⊆ W ({f }) ,
that is, setting
f ∈
S :=[i∈I
Si ,
\P∈W (S)∩P rs(A)
P =
\P∈P rs(A);S⊆P
P .
Therefore, by Proposition 5.5, f ∈ r(< S >) : there is n ≥ 1 such that
f n ∈ < S >. Thus, there is g ∈< S > such that f n + g = g ; one has
j=1 ajsj for aj ∈ A, sj ∈ S ; for each j ∈ {1, ..., m}, sj ∈ Sij
j=1 Sij >, whence
j=1 Sij >, and reading the above argument in reverse order
g = Pm
for some ij ∈ I. Let S0 = {s1, ..., sm} ; then g ∈< Sn
f n ∈ <Sm
with S replaced by Sm
j=1 Sij yields that
D(f ) =
Uij ,
m[j=1
whence the quasi -- compactness of D(f ).
2. Let U be an open set in P rs(A), and P ∈ U. We have P rs(A) \ U =
P rs(A) ∩ W (S) for some subset S of A. As P /∈ W (S), S * P, whence
there is an s ∈ S with s /∈ P. It is now clear that P ∈ D(s) and
D(s) ⊆ P rs(A) \ W (S) = U .
3. Let O ⊆ P rs(A) be open and quasi -- compact ; according to (2), one
may write O = Sj∈J D(fj) with fj ∈ A. But then, there is a finite
subset J0 of J such that O =Sj∈J0 D(fj). Now
P rs(A) \ O = \j∈J0
D(fj)
= P rs(A) ∩ W (< fjj ∈ J0 >)
is of the required type.
Conversely, if P rs(A) \ O = P rs(A) ∩ W (I) with I =< g1, ..., gn >,
i=1 D(gi); as a finite union of quasi -- compact
it is clear that O = Sn
subspaces of P rs(A), O is therefore quasi -- compact.
4. Let O1, ..., On denote quasi -- compact open subsets of P rs(A) ; then,
according to (iii), we may write
P rs(A) \ Oj = P rs(A) ∩ W (Ij)
19
for some finitely generated ideal Ij of A. Thus
P rs(A) \ (O1 ∩ ... ∩ Om) =
=
m[j=1
m[j=1
(P rs(A) \ Oj)
(P rs(A) ∩ W (Ij))
= P rs(A) ∩
m[j=1
W (Ij)
= P rs(A) ∩ W (
mYj=1
Ij)
= P rs(A) ∩ W (I1...Im) ,
whence, according to (iii), O1 ∩ ... ∩ Om is quasi -- compact, as I1...Im is
finitely generated.
5. Let F denote an irreducible closed set in P rs(A) ; then F = P rs(A) ∩
W (S) for S a subset of A. We have seen above that, setting I := < S >,
one has F = P rs(A) ∩ W (I). As F is not empty, I 6= A. Let us assume
ab ∈ I ; then, for each P ∈ F , one has ab ∈ I ⊆ P , whence a ∈ P or
b ∈ P, i.e. P ∈ F ∩ W ({a}) or P ∈ F ∩ W ({b}) :
F = (F ∩ W ({a})) ∪ (F ∩ W ({b})) .
As F is irreducible, it follows that either F = F ∩ W ({a}) or F =
F ∩ W ({b}). In the first case we get F ⊆ W ({a}), i.e.
a ∈ \P∈P rs(A);I⊆P
P = I(Proposition 5.5) ;
similarly, in the second case, b ∈ I : I is prime. But then
{I} = P rs(A) ∩ W (I)
= F
and I is a generic point for F .
It is unique as, in a T0 -- space, an (irreducible) closed set admits at
most one generic point (see [6], (0.2.1.3)).
20
Corollary 6.3. P rs(A) and MaxSpec(A) are spectral spaces in the sense of
Hochster ([7],p. 43).
[6], Corollary 1.1.14) Let F = P rs(A) ∩ W (S) be a
Theorem 6.4. (cf.
nonempty closed set in P rs(A) ; then F is homeomorphic to P rs(B), where
B :=
with I := < S >.
A
RI
Proof. As seen above, one has F = P rs(A)∩W (I), whence, as F 6= ∅, I 6= A.
Let A0 :=
, and let π : A → A0 denote the canonical projection.
A
RI
Let us now define
ψ : P rs(A0) → F
Q 7→ π−1(Q) .
Then ψ is well -- defined (as π−1(Q) is a saturated prime ideal of A that
contains I), and injective (as, for each Q ∈ P rs(A0), π(ψ(Q)) = Q).
Let P ∈ F ; then π(P) is an ideal of A0. Let us assume π(v) ∈ π(P) ;
then
for some a ∈ P, that is
But then
for some i ∈ I, whence
π(v) + π(a) = π(a)
π(a + v) = π(a) .
a + v + i = a + i
v + (a + i) = a + i
As a + i ∈ P and P is saturated, it follows that v ∈ P : π(P) is saturated.
Furthermore , if π(1) ∈ π(P), one has π(1) + π(v) = π(v) for some v ∈ P,
whence there is w ∈ I such that 1 + v + w = v + w, whence 1 + v + w ∈ P and
(as P is saturated) 1 ∈ P and P = A, a contradiction. Therefore π(P) 6= A0.
Let us assume π(x)π(y) ∈ π(P) : then xy + i = q + i for some i ∈ I,
whence
(x + i)(y + i) = xy + xi + iy + i2 ∈ P ,
and x + i ∈ P or y + i ∈ P ; as P is saturated, it follows that x ∈ P or y ∈ P,
whence π(x) ∈ π(P) or π(y) ∈ π(P) : π(P) is prime.
21
As P is saturated, one sees in the same way that ψ(π(P)) = π−1(π(P)) =
P, whence ψ is surjective.
Let G := F ∩ W (S0) be closed in F ; then P ∈ ψ−1(G) if and only if
ψ(P) ∈ F ∩ W (S0), that is S ⊆ π−1(P) and S0 ⊆ π−1(P), i.e. π(S ∪ S0) ⊆ P
:
ψ−1(G) = P rs(A0) ∩ W (π(S ∪ S0))
is closed in F , and ψ is continuous.
Let now H := P rs(A0)∩W ( ¯G) be closed in P rs(A0), and let Q ∈ P rs(A0)
; as π is surjective, ¯G ⊆ Q if and only if π−1( ¯G) ⊆ π−1(Q) = ψ(Q), and it
follows that
ψ(H) = F ∩ W (π−1( ¯G))
is closed in F . Therefore ψ is an homeomorphism.
22
7. Remarks on the one -- generator case
Let us now consider the case of a nontrivial monogenic B1 -- algebra con-
B1[x]
taining strictly B1, i.e. A =
is a quotient of the free algebra B1[x]
with x ≁ 0, x ≁ 1. Denote by α the image of x in A ; then α /∈ {0, 1}, and
α generates A as a B1 -- algebra.
∼
Let us suppose that, for some (u, v) ∈ A2, αu = 1 + αv ; then α is not
nilpotent, as from αn = 0 would follow
0 = αnv = αn−1(αv) = αn−1(1 + αu) = αn−1 + αnu = αn−1 ,
whence αn−1 = 0 and, by induction on n, 1 = α0 = 0, a contradiction.
Therefore three cases may appear
(i) α is nilpotent.
(ii) α is not nilpotent and there does not exist (u, v) ∈ A2 such that αu =
1 + αv.
(iii) (α is not nilpotent) and there exists (u, v) ∈ A2 such that αu = 1 + αv.
In case (i), any prime ideal of A must contain α, hence contain αA; the
ideal αA is, according to the above remark, saturated, and is not contained
in a strictly bigger saturated ideal other than A itself (in both cases, as any
element of A not in αA is of the shape 1 + αx). Therefore P rs(A) = {αA},
whence Nil(A) = αA. In this case we see that
A
RN il(A)
≃ B1 .
In cases (ii) and (iii), no power of α belongs to Nil(A) ; as Nil(A) is
saturated, it follows that Nil(A) = {0}.
In fact, A is integral, whence
{0} ∈ P rs(A). If P ∈ P rs(A) and P 6= {0}, then P contains some power
of α, hence contains α, hence contains αA. As above we see that P = αA ;
but, in case (iii), αA is not saturated. In case (ii) it is easy to see that αA is
prime and saturated. Therefore
1. In case (ii), P rs(A) = {{0}, αA} ; {0} is a generic point, that is
{{0}} = P rs(A) ,
and αA a "closed point"({αA} is closed) ;
23
2. In case (iii), P rs(A) = {{0}}.
One may remark that B1[x] itself falls into case (ii).
In [9], pp. 75 -- 79, we have enumerated (up to isomorphism) monogenic
B1 -- algebras of cardinality ≤ 5. It is easy to see where these algebras fall
in the above classification ; we keep the numbering used in [9]. Let then
3 ≤ A ≤ 5. We have the following repartition
Case (i) : (6),(8),(12),(15),(18),(24)
Case (ii): (7),(10),(11),(16),(19),(25),(26)
Case (iii): (5),(9),(13),(14),(17),(20),(21),(22),(23),(27),(28)
24
8. Acknowledgement
This paper was written during a stay at I.H.E.S. (December 2010-March
2011) ; a preliminary version appeared as prepublication IHES/M/11/06. I
am grateful both to the staff and to the colleagues who managed to make
this stay pleasant and stimulating.
I am also deeply indebted to Profesor
Alain Connes for his constant moral support.
I am very grateful for the invitations to lecture upon this work at the
JAMI conference Noncommutative geometry and arithmetic (Johns Hopkins
University, March 22-25 2011), and at the AGATA Seminar (Montpellier,
April 7, 2011) ; for these, I thank respectively Professors Alain Connes and
Caterina Consani, and Professor Vladimir Vershinin.
References
25
9. Bibliography
References
[1] D. Castella L'alg`ebre tropicale comme alg`ebre de la caract´eristique 1
: polynomes rationnels et fonctions polynomiales, preprint ; arXiv :
0809.0231 v1
[2] A. Connes and C. Consani Schemes over F1 and zeta functions, Com-
positio Mathematica 146 (6)(2010), pp.1383 -- 1415.
[3] A. Connes and C. Consani Characteristic one, entropy and the absolute
point, to appear on the Proceedings of the JAMI Conference 2009, JHUP
(2011).
[4] A. Connes and C. Consani The hyperring of adele classes, Journal of
Number Theory 131(2011), pp. 159 -- 194.
[5] A.Deitmar Schemes over F1, in Number Fields and Function Fields -
two parallel worlds, pp. 87-100, Birkhauser, Boston, 2005.
[6] A. Grothendieck, J.A. Dieudonn´e El´ements de G´eom´etrie Alg´ebrique I,
Publ. Math. IHES, No.4, 1960.
[7] M. Hochster Prime ideal structure in commutative rings, Trans. Amer.
Math. Soc. 142(1969), pp. 43-60 .
[8] S. LaGrassa Semirings : ideals and polynomials, PhD thesis, University
of Iowa, 1995.
[9] P. Lescot Alg`ebre Absolue, Ann. Sci. Math. Qu´ebec 33(2009), no 1, pp.
63-82.
[10] P. Lescot Absolute Algebra II-Ideals and Spectra, Journal of Pure and
Applied Algebra 215(7), 2011, pp. 1782 -- 1790.
[11] A. Zelevinsky Representations of Finite Classical Groups. A Hopf Al-
gebra Approach, Lecture Notes in Mathematics, 869, Springer-Verlag,
Berlin-New York, 1981, 184 pp.
[12] Y. Zhu Combinatorics and characteristic one algebra, preprint, 2000.
26
|
0802.1085 | 3 | 0802 | 2011-09-28T05:14:44 | Auslander Bounds and Homological Conjectures | [
"math.RA",
"math.AC",
"math.KT",
"math.RT"
] | Inspired by recent works on rings satisfying Auslander's conjecture, we study invariants, which we call Auslander bounds, and prove that they have strong relations to some homological conjectures. | math.RA | math |
Auslander Bounds and Homological Conjectures
Jiaqun Wei∗
Abstract
Inspired by recent works on rings satisfying Auslander's conjecture, we study invariants,
which we call Auslander bounds, and prove that they have strong relations to some homo-
logical conjectures.
Keywords: Auslander bound Homological conjecture
MSC 2000: 16E10 16G10 16E30 16E65.
tilting module
1
Introduction
Throughout this paper, rings are associative with nonzero identities and modules are left
modules unless otherwise specified. Let R be a ring, we denote by ModR the category of all
left R-modules, and we denote by modR the full subcategory of all R-modules having finitely
generated projective resolutions.
Auslander posed the following conjecture (cf. [8]).
(Auslander Conjecture): Let R be an artin algebra. Then for every M ∈ modR,
R(M, N ) = 0 for all i > bM and for all
there exists an integer bM such that Exti
R-modules N ∈ modR satisfying Extj
R(M, N ) = 0 for all sufficiently large j.
It is known that if the Auslander conjecture holds for all finite dimensional algebras then
the finitistic dimension conjecture is true for all finite dimensional algebras [8]. However, the
Auslander conjecture fails in general by counterexamples in [11, 17]. Rings satisfying the as-
sertion in the Auslander conjecture are studied in [4, 9, 16]. In [4] the authors investigate in
detail the relationship between such rings and some homological conjectures, for instance, the
Auslander-Reiten conjecture, which we recall as follows.
Auslander-Reiten Conjecture: Let R be a ring. If M ∈ modR and Exti
R(M, M ⊕R) =
0 for all i > 0, then M is projective.
In this paper, we continue the study and focus on the number bM , which we called the left
Auslander bound of M and denote by lAbM , for the fixed module M . Note that the Auslander
bound can be ∞. We prove the following result.
Theorem If M ∈ modR and Exti
R(M, M ⊕ R) = 0 for all sufficiently large i, then
lAbM coincides with the projective dimension of M .
∗Supported by the National Natural Science Foundation of China (Grant No. 10971099)
The result has clear relation with the Auslander-Reiten conjecture. More generally, it relates
Auslander bounds with an equivalent version of the Wakamatsu-tilting conjecture (EWTC for
short, see Section 3 for details), which asserts that an R-module T ∈ modR is tilting if and only if
(1) Exti
R(T, T ) = 0 for all i > 0 and (2) there is an exact sequence 0 → R → T0 → · · · → Tn → 0
for some n, where each Ti ∈ addT . We refer to [1] for the history and development of tilting
theory. In fact, we obtain that if R satisfies the condition that lAbM < ∞ for every M ∈ modR,
then R satisfies the conjecture (EWTC). This extends [4, Theorem A]. As we see, all rings with
finite global repetition index, in particular, the finite dimensional algebra O/πO, where O is
a classical order of finite global dimension over a discrete valuation ring D with uniformizing
parameter π and residue class field K, satisfies the condition in the last result.
The above theorem is proved in Section 2, after some investigations on basic properties
of Auslander bounds. Relations between homological conjectures and Auslander bounds are
presented in Section 3, where we also formulate some new homological conjectures.
We introduce some notions in the following.
Let R be a ring and C, D ∈ ModR. Let t be a non-negative integer. By Ext>t
R (C, D) = 0 we mean that Ext>t
R (C, D) = 0
we mean that Exti
R (C, D) = 0
for some t. Given an R-module M and an integer t ≥ 0, we denote by M >t (>tM , resp.)
the subcategory of all modules N such that Ext>t
R (N, M ) = 0, resp.). The
notions M ≫ and ≫M are defined similarly.
R(C, D) = 0 for all i > t. By Ext≫
R (M, N ) = 0 (Ext>t
For an R-module M , we denote by addM the class of all modules isomorphic to direct
to denote
summands of finite direct sums of copies of T . We use pdM (idM , fdM , resp.)
projective (injective, flat, resp.) dimension of M .
We denote by Ro the opposite ring of R. Thus ModRo is the category of all right R-modules.
In case that R is an artin algebra, we denote by D the usual dual functor between modR and
modRo.
2 Auslander bounds
Throughout this section, we fix R a ring. If a class C consists of a single R-module, say C,
then we use C instead of C.
We introduce the following notion.
Definition 2.1 Let C, D be two classes of R-modules. The Auslander bound of the pair (C, D),
denoted by Ab(C, D), is defined in the following way:
If there are no C ∈ C and D ∈ D such that Ext≫
Otherwise, Ab(C, D) is the minimal non-negative integer m such that Ext>m
R (C, D) = 0, then Ab(C, D) = −1;
R (C, D) =
R (C, D) = 0, or ∞ if no such minimal
0 for any C ∈ C and D ∈ D with Ext≫
integer exists.
It is easy to see that Ab(C, D) < ∞ for any two R-modules C, D.
Let M ∈ ModR.
It is also easy to see that Ab(M, C) is just the minimal non-negative
integer m such that M ≫ T C = M >m T C, or ∞ if no such integer exists, or −1 if M ≫ T C = Ø.
Similarly Ab(C, M ) is just the minimal non-negative integer m such that C T ≫M = C T >mM ,
or ∞ if no such integer exists, or −1 if C T ≫M = Ø.
We use the following simple notions for M ∈ ModR.
• LAbM := Ab(M, ModR)
•
lAbM := Ab(M, modR)
• RAbM := Ab(ModR, M )
•
rAbM := Ab(modR, M )
(called the big left Auslander bound of M );
(the small left Auslander bound of M );
(the big right Auslander bound of M );
(the small right Auslander bound of M ).
It is easy to see that the above Auslander bounds are non-negative.
Remark 2.2 (1) LAbM is just the minimal bound on the vanishing of ExtR(M, −) in [5]. In
[9, 16], lAbM is also denoted by eR(M, −) in case M ∈ modR.
(2) One can similarly define the Tor-Auslander bound of a pair (C, D), denoted by tAb(C, D),
i>m(−, −). Then tAb(M, ModR) for a right R-module M is just the minimal
by the bifunctor TorR
bound on the vanishing of TorR(M, −) in [13].
(3) Obviously, LAbM ≤ pdM and RAbM ≤ idM with the equality holds if the latter is
finite. If M ∈ modR, then lAbM = pdM provided that pdM < ∞.
We say that C has the two-out-of-three property provided that any two terms in a short
exact sequence are in C implies the third term is also in C. The proof of the following lemma is
easy.
Lemma 2.3 Let M ∈ ModR.
(1) All subcategories modR, M ≫, and ≫M have the two-out-of-three property.
(2) All subcategories modR, M ≫,and ≫M are closed under direct summands and finite direct
sums. Moreover, M ≫ (≫M , resp.) is closed under arbitrary direct products (direct sums, resp.).
(3) M ≫ = (ΩiM )≫, where ΩiM denotes an i-th syzygy of M .
(4) ≫M = ≫(Ω−iM ), where Ω−iM denotes an i-th cosyzygy of M .
(5) If R is an artin algebra and M ∈ modR, then D(M ≫) = ≫(D(M )).
Lemma 2.4 Let M, N ∈ ModR and C be a class of R-modules.
(1) Ab(M ⊕ N, C) ≤ max{Ab(M, C), Ab(N, C)}.
(2) rAb(C, M ⊕ N ) ≤ max{Ab(C, M ), Ab(C, N )}.
(3) If R is an artin algebra and M ∈ modR, then lAbM = rAb(D(M )).
Proof.
(1) Clearly we can assume that k := max{Ab(M, C), Ab(N, C)} < ∞. Now note
that (M ⊕ N )≫ T C = M ≫ T N ≫ T C = M >k T N >k T C = (M ⊕ N )>k T C, so we have that
Ab(M ⊕ N, C) ≤ k.
The proof of (2) is dual to that of (1). The proof of (3) follows from Lemma 2.3(5).
✷
Proposition 2.5 Let C, D, E, F be four classes of R-modules.
Ab(C, E) ≤ Ab(C, F) ≤ Ab(D, F).
If C ⊆ D and E ⊆ F , then
Proof. We may assume that 0 ≤ Ab(C, F) = t < ∞. Then Ext>t
and F ∈ F with Ext≫
R (C, F ) = 0 for any C ∈ C
R (C, F ) = 0, by the definition. In particular, since E ⊆ F, we have that
R (C, E) = 0 for any C ∈ C and E ∈ E with Ext≫
Ext>t
The remaining part is proved similarly.
R (C, E) = 0. This shows that Ab(C, E) ≤ t.
✷
Let C, D be two classes of R-modules. We have the following result.
Lemma 2.6 Ab(C, D) = sup {Ab(M, D) M ∈ C}
= sup {Ab(C, N ) N ∈ D}
= sup {Ab(C, D) C ∈ C, D ∈ D}.
Proof. By Proposition 2.5, we have that
sup {Ab(C, D) C ∈ C, D ∈ D} ≤ sup {Ab(M, D) M ∈ C} ≤ Ab(C, D).
To prove the other part, we may assume that 0 ≤ sup {Ab(C, D) C ∈ C, D ∈ D} = t < ∞.
Take any C ∈ C and D ∈ D with Ext≫
R (C, D) = 0.
Hence, we have that Ab(C, D) ≤ t.
✷
R (C, D) = 0, then we easily see that Ext>t
We call Ab(C, C) the global Auslander bound of the class C. We denote by GAbR the global
Auslander bound of ModR and by gAbR the global Auslander bound of modR. Note that gAbR
is just the Ext-index of R in [9, 16].
If R is an artin algebra, then there is a duality D between modR and modRo. Hence we can
easily obtain that gAbR = gAbRo in this case.
To calculate the Auslander bound of a module, it is enough to calculate the Auslander bound
of its syzygies, as the follow result shows.
Lemma 2.7 Let M ∈ ModR and C be a class of R-modules.
(1) 0 ≤ Ab(M, C) ≤ m if and only if Ab(ΩmM, C) = 0, where ΩmM denotes an m-th syzygy
of M .
(2) 0 ≤ Ab(C, M ) ≤ m if and only if Ab(C, Ω−mM ) = 0, where Ω−mM denotes an m-th
cosyzygy of M .
Proof. (1) Since M ≫ = (ΩmM )≫ by Lemma 2.3 (3) and M >m = (ΩmM )≥1 by dimension
shifting, we see that 0 ≤ Ab(M, C) ≤ m ⇐⇒ Ø 6= M ≫ T C ⊆ M >m T C
⇐⇒ Ø 6= (ΩmM )≫ T C ⊆ (ΩmM )≥1 T C
⇐⇒ Ab(ΩmM, C) = 0.
The proof of (2) is dual to that of (1).
✷
By the above lemma, we easily obtain the following result.
Proposition 2.8 The following are equivalent for a class C such that Ab(C, C) 6= −1.
(1) The global Auslander bound of C is not more than n.
(2) Ab(ΩnC, C) = 0, where ΩnC denotes the class of all n-th syzygies of R-modules in C.
(3) Ab(C, Ω−nC) = 0, where Ω−nC denotes the class of all n-th cosyzygies of R-modules in
C.
In some cases, the Auslander bound of a module can be tested by special modules, as the
following theorem shows.
Theorem 2.9 Let M ∈ ModR.
(1) If lAbM < ∞ and R ∈ M ≫, then lAbM = min{tR ∈ M >t}.
(2) If LAbM < ∞ and R(κ) ∈ M ≫ for all cardinals κ, then LAbM = min{tR(κ) ∈ M >t for
all cardinals κ}.
(3) If RAbM < ∞ and I ∈ ≫M for all injective R-modules I, then RAbM = min{tI ∈ >tM
for all injective R-modules I}.
Proof. (1) Assume that lAbM = m < ∞. Let t = min{tR ∈ M >t}. Then t < ∞, since
R ∈ M ≫ by the assumption.
Note that R ∈ modR, so we have that R ∈ M >t T modR ⊆ M ≫ T modR. Now take
any N ∈ M ≫ T modR and any projective resolution of N :
· · · → P1 → P0 → N → 0 with
each Pi finitely generated projective. Then we have all Pi ∈ M >t T modR, and hence all
ΩiN ∈ M ≫ T modR, by Lemma 2.3 (1). Therefore, for all i > t, we obtain that Exti
R(M, N ) ≃
Exti+1
R (M, ΩmN ) = 0, by dimension shifting and the definition of
lAbM . It follows that N ∈ M >t T modR. Consequently, M >t T modR = M ≫ T modR, that is,
lAbM ≤ t.
R (M, ΩN ) ≃ · · · ≃ Exti+m
On the other hand, since R ∈ M ≫ T modR by assumptions, we have that R ∈ M >m by the
definition of lAbM . It follows that t ≤ m too. Hence the conclusion follows.
The proof of (2) is similar as (1) and the proof of (3) is dual to (2).
✷
Immediately, we obtain the following corollary [16, Corollary 3.3].
Corollary 2.10 Assume that idR < ∞.
(1) If lAbM < ∞ for every M ∈ modR, then gAbR = idR.
(2) If R is left noetherian and LAbM < ∞ for every M ∈ ModR, then GAbR = idR.
We note that assumptions in Theorem 2.9 (1) can not be removed. For example, let R be an
artin algebra of finite representation type with idR = ∞. Then it is easy to see that gAbR < ∞.
However, it is obvious that there are modules M ∈ modR such that min{tR ∈ M >t} = ∞.
Thus the condition R ∈ M ≫ is needed. Now let R be a Gorenstein ring with gAbR = ∞ (such
rings exist by [11]). Then there are modules M ∈ modR such that lAbM = ∞ by Corollary
2.10 (1). However, it is easy to see that min{tR ∈ M >t} ≤ idR < ∞ for any M ∈ modR. So
the condition that lAbM < ∞ is also needed.
We remark that it is an openh question whether GAbR = gAbR if R is left noetherian.
The following theorem is our main result which relates Auslander bounds to Auslander-
Reiten conjecture as claimed in the introduction.
Theorem 2.11 Let M ∈ ModR.
(1) Assume that M ⊕ R ∈ M ≫ T modR, then lAbM = pdM .
(2) Assume that M ⊕ R(κ) ∈ M ≫ for any cardinal κ, then LAbM = pdM .
(3) Assume that M, I ∈ ≫M for any injective R-module I, then RAbM = idM .
Proof. (1) Clearly we need only prove that pdM ≤ lAbM .
We can assume that lAbM = m < ∞. Since M ∈ modR, we can take a projective resolution
of M : · · · → P1 → P0 → M → 0 with each Pi finitely generated projective. Then each ΩiM ∈
M ≫ T modR by Lemma 2.3 (1), as M ⊕ R ∈ M ≫ T modR. It follows that ΩiM ∈ M >m for
each i, by the definition of lAbM . Now by applying the functor HomR(−, Ωm+1M ) to the exact
sequence 0 → ΩmM → Pm−1 → · · · → P0 → M → 0, we obtain that Ext1
R(ΩmM, Ωm+1M ) ≃
Ext2
R (M, Ωm+1M ) by dimension shifting. The latter is 0
R(Ωm−1M, Ωm+1M ) ≃ · · · ≃ Extm+1
since Ωm+1M ∈ M >m by the above argument. It follows that the exact sequence 0 → Ωm+1M →
Pm → ΩmM → 0 splits, and consequently, pdM ≤ m.
The proof of (2) is similar as (1) and the proof of (3) is dual to (2).
✷
3 Homological conjectures
As pointed out in [11, 17], Auslander's conjecture fails for artin algebras in general. However,
we can consider a finitistic version of Auslander's conjecture. Let R be a ring. We set
• FLAb(R) := sup {LAbM LAbM < ∞}, and
• flAb(R) := sup {lAbM M ∈ modR and lAbM < ∞}.
Similarly, we have notions FRAb(R) and frAb(R) defined by right Auslander bounds. Note
that flAb(R) = frAb(Ro) in case that R is an artin algebra.
Now we formulate the following conjecture.
Finitistic Auslander Conjecture (FAC for short):
• (lFAC): flAb(R) < ∞ for every artin algebra R, or dually
• (rFAC):
frAb(R) < ∞ for every artin algebra R.
It is easy to see that the finitistic Auslander conjecture implies the finitistic dimension
conjecture for artin algebras by Remark 2.2 (3).
It is also clear that flAbR = frAbR < ∞ if gAbR < ∞, by Lemma 2.6. Similarly, FLAbR =
FRAbR < ∞ if GAbR < ∞. For example, every group algebra kG with k a field and G finite
has the property GAb(kG) < ∞, see [3, Theorem 2.4] and [4, Appendix A].
As the finitistic dimension conjecture fails for commutative noetherian rings in general,
the conjecture (FAC) fails in the case, too. Moreover, it is pointed out in [4] that there is a
commutative notherian ring R with infinite Krull dimension such that lAbM < ∞ for every
M ∈ modR but gAbR = ∞.
It is unknown whether gAbR < ∞ if R is an artin algebra such that lAbM < ∞ for every
M ∈ modR.
The following result gives a partial answer to the conjecture (FAC).
Proposition 3.1 Let R be a ring.
(1) If idR < ∞, then flAb(R) < ∞.
(2) If R is left noetherian and idR < ∞, then FLAb(R) < ∞.
Proof. (1) Indeed, we have that lAbM = min{tR ∈ M >t} ≤ idR provided lAbM < ∞ and
M ∈ modR, by Theorem 2.9.
(2) If R is left noetherian and idR < ∞, then idR(κ) < ∞ for any cardinal κ. Now the
conclusion follows from Theorem 2.9 again.
✷
Now we turn to other related homological conjectures.
Let R be a ring and T ∈ ModR with S = EndRT . Recall from [18] that T is Wakamatsu-
tilting if it satisfies
(1) T ∈ modR and TS ∈ modSo,
(2) R ≃ End(TS), and
(3) Exti
R(T, T ) = 0 = Exti
S o(T, T ) for all i > 0.
Equivalently, as shown in [18], T is Wakamatsu-tilting if
(W1) T ∈ modR,
(W2) Exti
(W3) there is an exact sequence 0 → R →f0 T0 →f1 T1 →f2 · · · with each Ti ∈ addT
R(T, T ) = 0 for all i > 0, and
and each Imfi ∈ >0T , for all i ≥ 0.
It is clear that T is Wakamatsu-tilting if and only if TS is Wakamatsu-tilting.
Recall also that T is tilting [2, 15] if it satisfies
(T1) T ∈ modR and pdT < ∞,
(T2) Exti
(T3) there is an exact sequence 0 → R → T0 → · · · → Tn → 0 with each Ti ∈ addT ,
R(T, T ) = 0 for all i > 0, and
for some integer n.
We note that T is tilting if and only if TS is tilting, where S = EndRT [15]. The following
conjecture is cited from [14].
Wakamatsu Tilting Conjecture (WTC for short): Every Wakamatsu-tilting mod-
ule of finite projective dimension is tilting.
It is pointed out in [14] that, if the finitistic dimension conjecture holds for a ring R, then
the conjecture (WTC) holds for R. We have an equivalent version of the conjecture (WTC)
(and so we denote this conjecture by EWTC, where E means equivalent).
Proposition 3.2 The conjecture (WTC) holds for all rings if and only if the following conjec-
ture (EWTC) holds for all rings R.
(EWTC) : An R-module T ∈ modR is tilting if it satisfies conditions (T2) and (T3)
in the definition of tilting modules.
Proof. Let R be a ring and T ∈ modR with S = EndRT .
(EWTC) ⇒ (WTC): Assume that T is a Wakamatsu-tilting R-module with pdT < ∞.
Then we have an exact sequence 0 → Pn → · · · → P0 → T → 0 for some n. Applying the
functor HomR(−, T ), we obtain an induced exact sequence 0 → S → T0 → · · · → Tn → 0 in
modSo, since Exti
R(T, T ) = 0 for all i > 0. Since T is a Wakamatsu-tilting R-module, T is also
a Wakamatsu-tilting So-module and so Exti
S o(T, T ) = 0 for all i > 0. Hence we get that TS is
tilting provided that (EWTC) holds for So. Consequently, T is also a tilting R-module.
(WTC) ⇒ (EWTC): If (T2) and (T3) in the definition of tilting modules holds for T ∈
modR, then T is Wakamatsu-tilting. Moreover, by applying the functor HomR(−, T ) to the exact
sequence in (T3), we easily see that TS is Wakamatsu-tilting with finite projective dimension.
It follows that TS is tilting provided that (WTC) holds for So. Now by the left-right symmetry,
we get that T is tilting.
✷
If we specify n = 0 in the condition (T3), then R ∈ addT .
In this case, the conjecture
(EWTC) is just the Auslander-Reiten conjecture.
The following result gives a partial answer to the conjecture (EWTC), which extends [4,
Theorem A].
Proposition 3.3 Let R be a ring. If lAbM < ∞ for every M ∈ modR, then the conjecture
(EWTC) holds for R.
Proof. Assume that T ∈ modR satisfies the conditions (T2) and (T3). Then we easily obtain
that T ⊕ R ∈ T ≫. Now by the assumption and Theorem 2.11, we get that pdT = lAbT < ∞.
It follows that T is tilting.
✷
We now consider another homological conjecture.
Gorenstein Symmetry Conjecture : Let R be an artin algebra. Then idR < ∞
if and only if id(RR) < ∞.
Gorenstein Symmetry conjecture clearly makes sense for any ring. It was proved in [4] that
if R is a two-sided noetherian ring such that lAbM < ∞ for every M ∈ modR and lAbN < ∞
for every N ∈ modRo, and (1) R is an artin algebra, or (2) R has a dualizing complex, then
idR < ∞ if and only if idRoR < ∞ (whence, idR = idRoR by [10]). The following result also
gives a similar answer to the Gorenstein Symmetry conjecture. Note that we do not know if
LAbM < ∞ for every M ∈ ModR provided that lAbM < ∞ for every M ∈ modR, even when
R is an artin algebra. We do not know whether LAbM < ∞ for every M ∈ ModR implies that
RAbM < ∞ for every M ∈ ModR.
Proposition 3.4 Let R be a two-sided noetherian ring. Assume that
(1) LAbM < ∞ for every M ∈ ModR and every M ∈ ModRo, or
(2) R has a dualizing complex and rAbM < ∞ for every M ∈ modR and every
M ∈ modRo.
Then idR < ∞ if and only if idRo < ∞.
Proof. Assume (1) holds. If idR < ∞, then GAbR < ∞ by assumptions and Corollary 2.10
(2). It follows that FRAb(R) < ∞. Using the fact that RAbM = idM for any R-module M
with idM < ∞ in Remark 2.2 (3), we further obtain that the injective version of the finitistic
dimension conjecture holds for R. This shows that idRo < ∞ by [6, Proposition 7]. Similarly,
we can prove that if idRo < ∞ then idR < ∞.
(idR)+1(N, M ) ≃ HomRo(Ext(idR)+1
Assume now (2) holds. If idR < ∞, then for any M ∈ modR and any injective Ro-module
N , it holds that TorR
(M, R), N ) = 0. Hence fdRoN < ∞. By
the definition of dualizing complex [4, Section 3.4], Ro has a dualizing complex if so is R. In
this case, we have that all Ro-modules of finite flat dimension have finite projective dimension,
RoR for any injective Ro-module N . Now applying
by [12, Theorem].
Theorem 2.11 to the Ro-module R, we obtain that idRoR = rAbR and the latter is finite by
assumptions. Thus, we have that idRo < ∞. The proof of the other part is also similar.
✷
R
It follows that N ∈ ≫
In the remaining part, we discuss a class of rings with finite global Auslander bound.
Let R be a ring and M an R-module. Assume that n is a nonnegative integer. Following
Goodearl and Zimmermann-Huisgen [7], we say that a projective resolution of M is repetitive
at degree n if there exists a decomposition Ωn(M ) = P ⊕ Ai such that P is projective and each
Ai occurs as a direct summand of infinitely many Ωj(M ). The repetition index of M , denoted
rep(M ), is the least nonnegative integer k such that there is a projective resolution of M which is
repetitive at degree k (if such a k exists), or ∞ (otherwise). The corresponding global repetition
index is Grep(R) = sup {rep(M )M ∈ ModR}.
We have the following result which relates the repetition index to the Auslander bound.
Lemma 3.5 Let R be a ring and M ∈ ModR. If rep(M ) = m < ∞, then LAbM ≤ m.
Proof. Since rep(M ) = m < ∞, M has a projective resolution such that Ωm(M ) = P ⊕ Ai with
P projective and that each Ai occurs as a direct summand of infinitely many Ωj(M ). So we have
that ΩmM ∈ add(P ⊕(⊕i>tΩiM )), for any t. Now take any N ∈ M ≫ and assume that N ∈ M >t
for some t. It follows that Ext1
R (M, N ) = 0 by
dimension shifting and the definition of M >t. Note that (ΩmM )≫ = M ≫ by Lemma 2.3, so
LAb(ΩmM ) = 0. It follows that LAbM ≤ m by Lemma 2.7.
R(⊕i>tΩiM, N ) ≃ Qi>t Exti+1
R(ΩmM, N ) ≤ Ext1
✷
Consequently, we obtain the following result.
Proposition 3.6 If R is a ring with GrepR < ∞, then GAbR < ∞. In this case, the conjecture
(EWTC) holds. In particular, the Auslander-Reiten conjecture holds in this case.
Proof. By Lemma 3.5 and Proposition 3.3.
✷
Let R be an artin ring. Recall that M ∈ modR has a ultimately closed projective resolution
if there is some m such that the m-th syzygy ΩmM ≃ ⊕Mi with each Mi ∈ add(ΩmiM ) for
some mi < m. In this case, we have that rep(M ) ≤ m, see [7]. One defines artin rings such that
every finitely generated module has an ultimately closed projective resolution to be of projective
ultimately closed type. It was proved in [2] that the Auslander-Reiten conjecture holds for artin
algebras of projective ultimately closed type. Lemma 3.5 and Proposition 3.3 together also imply
that the conjecture (EWTC) and Auslander's conjecture hold for such artin algebras.
In [7], the authors studied finite dimensional algebra R = O/πO, where O is a classical order
over a discrete valuation ring D with uniformizing parameter π and residue class field K. The
homological properties of O are to a great extent determined by those of R while the latter
algebra is substantially easier to handle. In their paper, it was shown that, if gdO = d < 1, then
GrepR = d − 1, in particular, the finitistic dimension of R is finite. Combining these with results
in this paper and [4], we also know that in case O has finite global dimension, the algebra O/πO
also satisfies the Auslander conjecture, the conjecture (FAC), the Auslander-Reiten conjecture,
Gorenstein Symmetry conjecture and the conjecture (EWTC).
Results in this section suggest the following conjecture which generalizes Auslander-Reiten
conjecture.
Generalized Auslander-Reiten Conjecture : Let R be a ring and M ∈ modR.
If M ⊕ R ∈ M ≫, then pdM < ∞.
By Lemma 3.5, the conjecture holds for artin algebras of projective ultimately closed type.
ACKNOWLEDGEMENTS
It is a pleasure to thank the referee for his/her carefully reading and excellent suggestions.
References
[1] L. Angeleri-Hugel, D. Happel and H. Krause (eds), Handbook of tilting theory, London Math. Soc.
Lect. Note Ser. 332 (2007).
[2] M. Auslander and I. Reiten, Applications of contravariantly finite subcategories, Adv. Math. 86 (1)
(1991), 111-152.
[3] D. J. Benson, J. F. Carlson and G. R. Robinson, On the vanishing of group cohomology, J. Algebra
131 (1) (1990), 40-73.
[4] L. W.Christensen and H. Holm, Algebras that satisfy auslanders condition on vanishing of cohomol-
ogy, Math. Z., DOI: 10.1007/s00209-009-0500-4.
[5] K. M. Cowley, One-Sided Bounds and the Vanishing of Ext, J. Algebra 190 (1997), 361-371.
[6] K.R. Fuller and Y. Wang, Redundancy in resolutions and finitistic dimensions of noetherian rings,
Comm. Algebra 21 (8) (1993), 2983-2994.
[7] K.R. Goodearl and B. Zimmermann-Huisgen, Repetitive resolutions over classical orders and finite
dimensional algebras, in Algebras and Modules II (I. Reiten, S.O. Smalø and O. Solberg, Eds.)
Canad. Math. Soc. Conf. Proc. Series 24 (1998) 205-225
[8] D. Happel, Homological
conjectures
in representation theory of finite-dimensional alge-
from http://www.math.ntnu.no/
available
bras, Sherbrook Lecture Notes Series
∼oyvinso/Nordfjordeid/Program/references.html.
(1991),
[9] C. Huneke and D.A. Jorgensoen, Symmetry in the vanishing of Ext over Gorenstein rings, Math.
Scand 93 (8) (2003), 161-184.
[10] Y. Iwanaga, On rings with finite self-injective dimension II, Tsukuba J. Math. 4 (1) (1980), 107-113.
[11] D.A. Jorgensen and L.M. Sega., Nonvanishing cohomology and classes of Gorenstein rings, Adv.
Math. 188 (2) (2004), 470-490.
[12] P. Jorgensen, Finite flat and projective dimensions, Comm. Algebra 33 (2005), 2275-2279.
[13] E. Kirkman and J. Kuzmanovich, On the finitistic dimension of fixed subrings, Comm. Algebra 22
(1994), 4621-4635.
[14] F. Mantese and I. Reiten, Wakamatsu Tilting modules, J. Algebra 278 (2004), 532-552.
[15] Y. Miyashita, Tilting modules of finite projective dimension, Math. Z. 193 (1986), 113-146.
[16] I. Mori, Symmetry in the vanishing of Ext over stably symmetric algebras, J. Algebra 310 (2) (2007),
708-729.
[17] S.O. Smalø, Local limitations of the Ext functor do not exist, Bull. London Math. Soc. 38 (1) (2006),
97-98.
[18] T. Wakamatsu, On modules with trivial self-extensions, J. Algebra 114 (1988) 106-114.
Jiaqun Wei
School of Mathematics Science, Nanjing Normal University, Nanjing 210046, China
Email: [email protected]
|
1611.00410 | 1 | 1611 | 2016-11-01T22:21:07 | Drinfeld Orbifold Algebras for Symmetric Groups | [
"math.RA",
"math.CO"
] | Drinfeld orbifold algebras are a type of deformation of skew group algebras generalizing graded Hecke algebras of interest in representation theory, algebraic combinatorics, and noncommutative geometry. In this article, we classify all Drinfeld orbifold algebras for symmetric groups acting by the natural permutation representation. This provides, for nonabelian groups, infinite families of examples of Drinfeld orbifold algebras that are not graded Hecke algebras. We include explicit descriptions of the maps recording commutator relations and show there is a one-parameter family of such maps supported only on the identity and a three-parameter family of maps supported only on 3-cycles and 5-cycles. Each commutator map must satisfy properties arising from a Poincar\'{e}-Birkhoff-Witt condition on the algebra, and our analysis of the properties illustrates reduction techniques using orbits of group element factorizations and intersections of fixed point spaces. | math.RA | math |
DRINFELD ORBIFOLD ALGEBRAS FOR SYMMETRIC GROUPS
B. FOSTER-GREENWOOD AND C. KRILOFF
Abstract. Drinfeld orbifold algebras are a type of deformation of skew group alge-
bras generalizing graded Hecke algebras of interest in representation theory, algebraic
combinatorics, and noncommutative geometry. In this article, we classify all Drinfeld
orbifold algebras for symmetric groups acting by the natural permutation representa-
tion. This provides, for nonabelian groups, infinite families of examples of Drinfeld
orbifold algebras that are not graded Hecke algebras. We include explicit descriptions
of the maps recording commutator relations and show there is a one-parameter family
of such maps supported only on the identity and a three-parameter family of maps sup-
ported only on 3-cycles and 5-cycles. Each commutator map must satisfy properties
arising from a Poincar´e-Birkhoff-Witt condition on the algebra, and our analysis of the
properties illustrates reduction techniques using orbits of group element factorizations
and intersections of fixed point spaces.
1. Introduction
Numerous algebras of intense recent study and interest arise as deformations of skew
group algebras S(V )#G, where G is a finite group acting linearly on a finite-dimensional
vector space V and S(V ) is the symmetric algebra. A grading on the skew group algebra
is determined by assigning degree one to vectors in V and degree zero to elements
of the group algebra. Drinfeld graded Hecke algebras are constructed by identifying
commutators of elements of V with carefully chosen elements of degree zero (i.e., from
the group algebra) to yield a deformation of the skew group algebra.
In [SW12a],
Drinfeld orbifold algebras are similarly defined but additionally allow for degree-one
terms in the commutator relations. The resulting algebras are also deformations of the
skew group algebra.
Besides capturing a new realm of deformations of skew group algebras, Drinfeld orb-
ifold algebras encompass many known algebras of interest in representation theory, non-
commutative geometry, and mathematical physics. The term "Drinfeld orbifold alge-
bras" alludes to the subject's origins in [Dri86], where Drinfeld introduced a broad class
of algebras to serve as noncommutative coordinate rings for singular orbifolds. When
the group is a Coxeter group acting by its reflection representation, Drinfeld's algebras
are isomorphic (see [RS03]) to the graded Hecke algebras from [Lus88], which arise from
a filtration of an affine Hecke algebra when the group is crystallographic (see [Lus89]).
The representation theory of these algebras is useful in understanding representations
and geometric structure of reductive p-adic groups.
2010 Mathematics Subject Classification. 16S80 (Primary) 16E40, 16S35, 20B30 (Secondary).
Key words and phrases. skew group algebra, deformations, Drinfeld orbifold algebra, Hochschild
cohomology, Poincar´e-Birkhoff-Witt conditions, symmetric group.
1
2
B. FOSTER-GREENWOOD AND C. KRILOFF
A recent focus on symplectic reflection algebras, which are Drinfeld Hecke algebras
for symplectic reflection groups acting on a symplectic vector space, began with [EG02].
The importance of these algebras lies in the fact that the center of the skew group
algebra is the ring of invariants, C[V ]G = Spec(V /G), and in the philosophy that the
center of a deformation of the skew group algebra may then deform C[V ]G (see the
surveys [Gor08, Bel16]). As a special case, rational Cherednik algebras arise by pairing
a reflection representation with its dual and are related to integrable Calogero-Moser
systems in physics and deep results in combinatorics (see for instance the surveys [Gor10,
Eti14]).
Drinfeld orbifold algebras afford two advantageous views: as quotient algebras satis-
fying a Poincar´e-Birkhoff-Witt (PBW) condition and as formal algebraic deformations
of skew group algebras. While PBW conditions relate an algebra to homogeneous shad-
ows of itself that have well-behaved bases, algebraic deformation theory (`a la Gersten-
haber [GS88]) focuses on how the multiplicative structure varies with a deformation
parameter and provides a framework of understanding via Hochschild cohomology. In
particular, every formal deformation arises from a Hochschild 2-cocycle.
Fruitful techniques arise from a melding of the PBW perspective with the defor-
mation theory perspective (see the survey [SW15]). Braverman and Gaitsgory [BG96]
and also Polishchuk and Positelski [PP05] initiated the use of homological methods to
study PBW conditions in the context of quadratic algebras of Koszul type. Etingof and
Ginzburg applied some of these ideas in an expanded setting in their seminal paper on
symplectic reflection algebras. The study of Drinfeld orbifold algebras also benefits from
relating PBW conditions to formal deformations. Shepler and Witherspoon prove two
characterizations of Drinfeld orbifold algebras: a concrete ring theoretic version [SW12a,
Theorem 3.1] (proved using Composition-Diamond Lemmas and Groebner basis theory)
and a cohomological version [SW12a, Theorem 7.2].
In the present case study, we classify Drinfeld orbifold algebras for symmetric groups
acting by the natural permutation representation. In Section 4, we apply [SW12a, Theo-
rem 7.2] and use Hochschild cohomology to find possible degree-one terms of the commu-
tator relations for a Drinfeld orbifold algebra. In Section 7, we then work with [SW12a,
Theorem 3.1] to determine compatible degree-zero terms (if they exist). Our main re-
sult, stated in Theorems 6.1 and 6.2, is an explicit description of the parameter maps
that define Drinfeld orbifold algebras for symmetric groups.
Parameter maps of Drinfeld orbifold algebras record commutators of elements of the
vector space V and can be categorized based on their support, i.e., which group elements
appear in the image. Drinfeld orbifold algebra maps (see Definition 2.1) with their
linear part supported only on the identity give rise to Lie orbifold algebras, as defined
in [SW12a]. Lie orbifold algebras generalize universal enveloping algebras of Lie algebras,
just as symplectic reflection algebras generalize Weyl algebras. We summarize our results
classifying Lie and Drinfeld orbifold algebras.
Theorem. For the symmetric group Sn (n ≥ 3) acting on V ∼= Cn by the natural
permutation representation, there is a one-parameter family of Lie orbifold algebras.
The remaining algebras have commutator relations supported only on 3-cycles and
5-cycles.
ORBIFOLD ALGEBRAS FOR Sn
3
Theorem. For the symmetric group Sn (n ≥ 4) acting on V ∼= Cn by the natural per-
mutation representation, there is a three-parameter family of Drinfeld orbifold algebras
supported on 3-cycles and 5-cycles. For n = 3, the family involves only two parameters.
In Section 3, we present the algebras via generators and relations. The exam-
ples contribute to an expanding medley of "degree-one deformations". For instance,
Shakalli [Sha12] uses actions of Hopf algebras to construct examples of deformations of
quantum skew group algebras involving degree-one terms in the commutator relations.
Shepler and Witherspoon [SW12a] consider Drinfeld orbifold algebras for groups acting
diagonally. The algebras we construct are among the first examples of Drinfeld orbifold
(but not Hecke) algebras for nonabelian groups.
A fundamental problem in deformation theory is to determine which Hochschild 2-
cocycles actually lift to deformations. The results of Section 6 provide a family of
2-cocycles that lift to define Drinfeld orbifold algebras for symmetric groups. However,
we also show, in Proposition 6.3, that for symmetric groups, degree-one Hochschild 2-
cocycles simultaneously supported on and off the identity do not lift to yield Drinfeld
orbifold algebras (and in fact do not even define Poisson structures). This contrasts
with the Drinfeld Hecke algebra case in which every polynomial degree-zero Hochschild
2-cocycle determines a deformation of the skew group algebra.
Reduction techniques in Section 5 and simplifications in Section 7 may prove helpful
in predicting for which group actions and spaces candidate cocycles will lift to yield
Drinfeld orbifold algebras.
In particular, a variation of Lemma 5.4 may be effective
for other groups with centralizers acting by monomial matrices, and the pattern to the
values in Lemmas 7.1 and 7.6 might generalize to other group representations through
analysis of intersections of fixed point spaces. As further exploration, one could consider
Drinfeld orbifold algebras for symmetric groups in the twisted or quantum settings, as
has been done for Drinfeld Hecke algebras for symmetric groups in [Wit07, Example 2.17]
and [NW16, Theorem 6.9].
2. Preliminaries
Throughout, we let G be a finite group acting linearly on a vector space V ∼= Cn. All
tensors will be over C.
Skew group algebras. Let G be a finite group that acts on a C-algebra R by algebra
automorphisms, and write gs for the result of acting by g ∈ G on s ∈ R. The skew
group algebra R#G is the semi-direct product algebra R ⋊ CG with underlying vector
space R ⊗ CG and multiplication of simple tensors defined by
(r ⊗ g)(s ⊗ h) = r(gs) ⊗ gh
for all r, s ∈ R and g, h ∈ G. The skew group algebra becomes a G-module by letting G
act diagonally on R ⊗ CG, with conjugation on the group algebra factor:
g(s ⊗ h) = (gs) ⊗ (gh) = (gs) ⊗ ghg−1.
In working with elements of skew group algebras, we commonly omit tensor symbols
unless the tensor factors are lengthy expressions.
4
B. FOSTER-GREENWOOD AND C. KRILOFF
If G acts linearly on a vector space V ∼= Cn, then G also acts on the tensor algebra
T (V ) and symmetric algebra S(V ) by algebra automorphisms. The skew group algebras
T (V )#G and S(V )#G become graded algebras when elements of V are assigned degree
one and elements of G are assigned degree zero.
Cochains. A k-cochain is a G-graded linear map α = Pg∈G αgg with components
αg :Vk V → S(V ). (Details in Section 4 motivate the use of cohomological terminology.)
We regard a map α on Vk V as a multilinear alternating map on V k and write
If each αg maps into V , then α is called a linear cochain, and if each αg maps into C,
then α is called a constant cochain.
α(v1, . . . , vk) in place of α(v1 ∧ · · · ∧ vk). Of course, if α(v1, . . . , vk) = 0, then α is zero
on any permutation of v1, . . . , vk. Also, if α is zero on all k-tuples of basis vectors, then
α is zero on any k-tuple of vectors. We exploit these facts often in the computations in
Section 7.
The support of a cochain α is the set of group elements for which the component
αg is not the zero map. The kernel of a cochain α is the set of vectors v0 such that
α(v0, v1, . . . , vk−1) = 0 for all v1, . . . , vk−1 ∈ V .
and component αg, the map hαg is defined by (hαg)(v1, . . . , vk) = h(αg(h−1
The group G acts on the components of a cochain. Specifically, for a group element h
v1, . . . , h−1
vk)).
hαg ⊗ hgh−1.
In turn, the group acts on the space of cochains by letting hα = Pg∈G
Thus α is a G-invariant cochain if and only if hαg = αhgh−1 for all g, h ∈ G.
Drinfeld orbifold algebras. For a parameter map κ = κL + κC , where κL is a linear
2-cochain and κC is a constant 2-cochain, the quotient algebra
Hκ = T (V )#G/hvw − wv − κL(v, w) − κC(v, w) v, w ∈ V i
is called a Drinfeld orbifold algebra if the associated graded algebra gr Hκ is isomor-
phic to the skew group algebra S(V )#G. The condition gr Hκ ∼= S(V )#G is called a
Poincar´e-Birkhoff-Witt (PBW) condition, in analogy with the PBW Theorem for
universal enveloping algebras.
Further, if Hκ is a Drinfeld orbifold algebra and t is a complex parameter, then
Hκ,t := T (V )#G[t]/hvw − wv − κL(v, w)t − κC (v, w)t2 v, w ∈ V i
is called a Drinfeld orbifold algebra over C[t]. In [SW12a, Theorem 2.1], Shepler
and Witherspoon make an explicit connection between the PBW condition and de-
formations in the sense of Gerstenhaber [GS88] by showing how to interpret Drinfeld
orbifold algebras over C[t] as formal deformations of the skew group algebra S(V )#G.
We summarize the broader context of formal deformations in Section 4.
Drinfeld orbifold algebra maps. Though the defining PBW condition for a Drin-
feld orbifold algebra Hκ involves an isomorphism of algebras, Shepler and Witherspoon
proved an equivalent characterization [SW12a, Theorem 3.1] in terms of properties of
the parameter map κ.
Definition 2.1. Let κ = κL + κC where κL is a linear 2-cochain and κC is a constant
2-cochain, and let Alt3 denote the alternating group on three elements. We say κ is a
ORBIFOLD ALGEBRAS FOR Sn
5
Drinfeld orbifold algebra map if the following conditions are satisfied for all g ∈ G
and v1, v2, v3 ∈ V :
(2.0)
(2.1)
(2.2)
(2.3)Xσ∈Alt3 Xxy=g
(2.4)
Xσ∈Alt3 Xxy=g
im κL
g ⊆ V g,
the map κ is G-invariant,
Xσ∈Alt3
κL
g (vσ(2), vσ(3))(gvσ(1) − vσ(1)) = 0 in S(V ),
y (vσ(2), vσ(3))) = 2 Xσ∈Alt3
κC
x (vσ(1) + yvσ(1), κL
y (vσ(2), vσ(3))) = 0.
κL
x (vσ(1) + yvσ(1), κL
κC
g (vσ(2), vσ(3))(gvσ(1) − vσ(1)),
As a special case, if the linear component κL of a Drinfeld orbifold algebra map is
supported only on the identity, then we also call κ a Lie orbifold algebra map.
Remark 2.2. If Hκ is a Drinfeld orbifold algebra, then κ must satisfy conditions (2.1)-
(2.4), but not necessarily the image constraint (2.0). However, [SW12a, Theorem 7.2
∼= Hκ as
(ii)] guarantees there will exist a Drinfeld orbifold algebra Heκ such that Heκ
g ⊆ V g for each g in G. Thus,
in classifying Drinfeld orbifold algebras, it suffices to only consider Drinfeld orbifold
algebra maps.
filtered algebras andeκ satisfies the image constraint imeκL
Theorem 2.3 ([SW12a, Theorem 3.1 and Theorem 7.2 (ii)]). A quotient algebra Hκ
satisfies the PBW condition gr Hκ ∼= S(V )#G if and only if there exists a Drinfeld
orbifold algebra map eκ such that Hκ ∼= Heκ.
The process of determining the set of all Drinfeld orbifold algebra maps consists of
two phases. For reasons discussed in Section 4, we use language from cohomology and
deformation theory to describe each phase. First, one finds all pre-Drinfeld orbifold
algebra maps, i.e., all G-invariant linear 2-cochains κL satisfying Properties (2.0) and
(2.2). A bijection between pre-Drinfeld orbifold algebra maps and a particular set of
representatives of Hochschild cohomology classes facilitates this step (see Lemma 4.1).
Second, we determine for which pre-Drinfeld orbifold algebra maps κL there exists a
compatible G-invariant constant 2-cochain κC such that Properties (2.3) and (2.4) hold.
We say κC clears the first obstruction if Property (2.3) holds and clears the second
obstruction if Property (2.4) holds. If a G-invariant constant 2-cochain κC clears both
obstructions, then we say κL lifts to the Drinfeld orbifold algebra map κ = κL + κC.
3. Orbifold algebras for symmetric groups
Let e1, . . . , en be the standard basis of V ∼= Cn. Let the symmetric group Sn act on
V by its natural permutation representation, so σei = eσ(i) for σ in Sn. The main effort
of this paper is in proving Theorems 6.1 and 6.2, which describe all Drinfeld orbifold
algebra maps for Sn acting by the natural permutation representation. As corollaries of
6
B. FOSTER-GREENWOOD AND C. KRILOFF
the theorems in Section 6, we present here the resulting PBW deformations of the skew
group algebra S(V )#Sn via generators and relations.
First, the one-dimensional space of Lie orbifold algebra maps classified in Theorem 6.1
yields a family of Lie orbifold algebras arising as deformations of S(V )#Sn.
Theorem 3.1 (Lie Orbifold Algebras over C[t]). Let the symmetric group Sn (n ≥ 3)
act on V ∼= Cn by its natural permutation representation. Then for a ∈ C,
Hκ,t = T (V )#Sn[t]/heiej − ejei − a(ei − ej)t 1 ≤ i < j ≤ ni
is a Lie orbifold algebra over C[t]. Further, the algebras Hκ,1 are precisely the Drinfeld
orbifold algebras such that κL is supported only on the identity.
Second, in Theorem 6.2, we determine (for Sn) all Drinfeld orbifold algebra maps
such that the linear component κL is supported only off the identity. The relations
in the consequent PBW deformations of S(V )#Sn involve sums of basis vectors over
I denote the
certain subsets of [n] := {1, . . . , n}. For I ⊆ [n], let eI =Pi∈I ei, and let e⊥
complementary vector e[n] − eI .
Theorem 3.2. Let the symmetric group Sn (n ≥ 3) act on V ∼= Cn by its natural
permutation representation. For a, b, c ∈ C and 1 ≤ i < j ≤ n let
{i,j,k}) ⊗ ((ijk) − (kji)),
(ae{i,j,k} + be⊥
κL(ei, ej) = Xk6=i,j
((ijk) − (kji)) + (a − b)2 Xσ a 5-cycle
σ2(i)=j
and let
κC (ei, ej ) = c Xk6=i,j
Then
2(σ − σ−1) − Xσ a 5-cycle
σ(i)=j
(σ − σ−1) .
Hκ,t = T (V )#Sn[t]/heiej − ejei − κL(ei, ej)t − κC(ei, ej)t2 1 ≤ i < j ≤ ni
is a Drinfeld orbifold algebra over C[t]. Further, the algebras Hκ,1 are precisely the
Drinfeld orbifold algebras such that im κL
g ⊆ V g for each g ∈ Sn and κL is supported
only off the identity.
We illustrate Theorem 3.2 for some small values of n. Note that the parameter b is
irrelevant when n = 3, and the sums over 5-cycles are absent in the cases n = 3 and
n = 4.
Example 3.3. For the symmetric group S3 acting on V ∼= C3 by the natural permuta-
tion representation, the Drinfeld orbifold algebras such that im κL
g ⊆ V g for each g ∈ S3
and κL
1 = 0 are the algebras of the form
Hκ = T (V )#S3/heiej − ejei − κ(ei, ej) 1 ≤ i < j ≤ 3}i,
where for some a, c ∈ C
κ(e1, e2) = κ(e2, e3) = κ(e3, e1) = (a(e1 + e2 + e3) + c) ⊗ ((123) − (321)).
This example coincides with [SW12a, Example 3.4] with a change of basis.
ORBIFOLD ALGEBRAS FOR Sn
7
Example 3.4. For the symmetric group S4 acting on V ∼= C4 by the natural permuta-
g ⊆ V g for each g ∈ S4
tion representation, the Drinfeld orbifold algebras such that im κL
and κL
1 = 0 are the algebras of the form
Hκ = T (V )#S4/heiej − ejei − κ(ei, ej) 1 ≤ i < j ≤ 4}i,
where
κ(e1, e2) = (a(e1 + e2 + e3) + be4 + c) ⊗ ((123) − (321))
+(a(e1 + e2 + e4) + be3 + c) ⊗ ((124) − (421)),
and κ(eσ(1), eσ(2)) = σ(κ(e1, e2)) for σ in S4. (In acting by σ, recall that σei = eσ(i) and
στ = στ σ−1 for σ, τ ∈ Sn.)
Example 3.5. For the symmetric group S5 acting on V ∼= C5 by the natural permuta-
g ⊆ V g for each g ∈ S5
tion representation, the Drinfeld orbifold algebras such that im κL
and κL
1 = 0 are the algebras of the form
Hκ = T (V )#S5/heiej − ejei − κ(ei, ej) 1 ≤ i < j ≤ 5}i,
where
κ(e1, e2) =
(a(e1 + e2 + e3) + b(e4 + e5) + c) ⊗ ((123) − (321))
+ (a(e1 + e2 + e4) + b(e5 + e3) + c) ⊗ ((124) − (421))
+ (a(e1 + e2 + e5) + b(e3 + e4) + c) ⊗ ((125) − (521))
− (a − b)2 ⊗ ((12345) + (12543) + (12453) + (12354) + (12534) + (12435))
+ (a − b)2 ⊗ ((21345) + (21543) + (21453) + (21354) + (21534) + (21435))
− 2(a − b)2 ⊗ ((23145) + (25143) + (24153) + (23154) + (25134) + (24135))
+ 2(a − b)2 ⊗ ((13245) + (15243) + (14253) + (13254) + (15234) + (14235)),
and κ(eσ(1), eσ(2)) = σ(κ(e1, e2)) for σ in S5.
Remark 3.6. If we specialize to t = 1 and let a = b = 0 in Theorem 3.2, then the
linear component κL is identically zero, thus recovering Drinfeld graded Hecke algebras
for Sn.
4. Deformation algebras and Hochschild cohomology
Our goal in this section is to describe linear and constant 2-cochains κ that are G-
invariant and satisfy the mixed Jacobi identity
[v1, κ(v2, v3)] + [v2, κ(v3, v1)] + [v3, κ(v1, v2)] = 0
in S(V )#G.
When κ is expanded asPg∈G κgg, it becomes clear that the mixed Jacobi identity for
κL is equivalent to Property (2.2) of a Drinfeld orbifold algebra map, and the mixed
Jacobi identity for κC is equivalent to Property (2.3) in the special case that the left side
of (2.3) is zero. In light of the relation between Drinfeld orbifold algebras and formal
deformations, Hochschild cohomology becomes a tool to facilitate finding Drinfeld orb-
ifold algebra maps, as summarized in Lemma 4.1. We first review some background on
8
B. FOSTER-GREENWOOD AND C. KRILOFF
deformation theory and cohomology before turning to the specific case of the symmetric
groups.
Deformations and Hochschild cohomology. Let A be an algebra over C. For a
complex parameter t, a deformation over C[t] of A is the vector space A[t] with an
associative multiplication ∗, which is C[t]-bilinear and for a, b in A is recorded in the
form
a ∗ b = ab + µ1(a ⊗ b)t + µ2(a ⊗ b)t2 + · · ·
for some maps µi : A ⊗ A → A with the sum finite for each pair a, b.
Identifying
coefficients on ti in the expressions a ∗ (b ∗ c) and (a ∗ b) ∗ c yields a cohomological relation
involving the maps µ1, . . . , µi. For example, identifying coefficients of t shows that µ1
is a Hochschild 2-cocycle, and identifying coefficients of t2 shows the (Hochschild)
coboundary of µ2 must be half of the Gerstenhaber bracket of µ1 with itself.
Generally, for an A-bimodule M the Hochschild cohomology of A with coefficients
A⊗Aop(A, M ), and if M = A, we simply write HH•(A).
in M is HH•(A, M ) := Ext•
Hochschild cohomology may be computed using various resolutions, each with their own
advantages. The maps µi defining the multiplication of a formal deformation algebra
are most easily regarded as cochains on a bar resolution. However, when A is a skew
group algebra, advantageous formulations of Hochschild cohomology arise from a Koszul
resolution and frame cohomology in terms of invariant theory. Conversions between the
bar complex and Koszul complex are key to the proof of [SW12a, Theorem 2.1] that
shows how to interpret a Drinfeld orbifold algebra over C[t] as a formal deformation of a
skew group algebra. The parameter map κ of a Drinfeld orbifold algebra Hκ,t over C[t]
may be identified with a cochain on the Koszul complex, and the linear part κL relates
to the first multiplication map µ1, while the constant part κC relates to the second
multiplication map µ2 (see [SW12a, Remark 2.5]).
Cohomological relations involving the maps µi have implications for the components
of the parameter map κ. Indeed, the conditions on κ given in [SW12a, Theorem 3.1] have
a parallel statement [SW12a, Theorem 7.2] in terms of cohomological spaces and opera-
tions. While Properties (2.1) and (2.2) of a Drinfeld orbifold algebra map are stated with
minimal machinery, the cohomological interpretations aid in organizing computations
and also reveal some hidden implications emphasized in Remark 4.2.
We now record descriptions of the cohomological spaces we use in our computations
and refer the reader to [SW12a], for example, for more details on the bar and Koszul
resolutions, chain maps, and isomorphisms that lead to these spaces. Let G be a finite
group acting linearly on a vector space V ∼= Cn. Let H • be the G-graded vector space
p−codim(V g)^ (V g)∗ ⊗
codim(V g)^ (cid:0)(V g)∗(cid:1)⊥ ⊗ Cg,
where V g is the fixed point space of g. Thus H • is tri-graded by cohomological degree
p, homogeneous polynomial degree d, and group element g. For any set R carrying a
G-action, we write RG for the set of elements fixed by every g in G. With the group
G acting diagonally on the tensor product (and with conjugation on the group algebra
H • =Lg∈G H •
g with components
H p,d
g = Sd(V g) ⊗
ORBIFOLD ALGEBRAS FOR Sn
9
factor), the Hochschild cohomology of S(V )#G can be computed using the series of
isomorphisms
HH•(S(V )#G) ∼= HH•(S(V ), S(V )#G)G ∼= (H •)G.
The first isomorphism follows from S¸tefan [S¸te95] (for example), and the description of
H • was first given independently by Farinati [Far05] and by Ginzburg-Kaledin [GK04].
Note that, together, the exterior factors of H p,d
identify with a subspace of Vp V ∗,
and then, since Sd(V g) ⊗Vp V ∗ ⊗ Cg ∼= Hom(Vp V, Sd(V g)g), the space H • may be
identified with a subspace of the cochains introduced in Section 2. The next lemma
records the relationship between Properties (2.1) and (2.2) of a Drinfeld orbifold algebra
map and Hochschild cohomology. When d = 1, the lemma is a restatement of [SW12a,
Theorem 7.2 (i) and (ii)]. When d = 0, the lemma is a restatement of [SW08, Corollary
8.17(ii)]. Despite its cohomological heritage, it is also possible to give a linear algebraic
proof of Lemma 4.1 in the spirit of [RS03, Lemma 1.8].
g
Lemma 4.1. For a 2-cochain κ =Pg∈G κgg with im κg ⊆ Sd(V g) for each g ∈ G, the
following are equivalent:
(a) The map κ is G-invariant and satisfies the mixed Jacobi identity, i.e., for all
v1, v2, v3 ∈ V
[v1, κ(v2, v3)] + [v2, κ(v3, v1)] + [v3, κ(v1, v2)] = 0
in S(V )#G,
where [·, ·] denotes the commutator in S(V )#G.
(b) For all g, h ∈ G and v1, v2, v3 ∈ V :
(i) h(κg(v1, v2)) = κhgh−1(hv1, hv2) and
(ii) κg(v1, v2)(gv3 − v3) + κg(v2, v3)(gv1 − v1) + κg(v3, v1)(gv2 − v2) = 0.
(c) The map κ is an element of
(H 2,d)G =Mg∈G(cid:16) Sd(V g)g ⊗
2−codim(V g)^ (V g)∗ ⊗
codim(V g )^ (cid:0)(V g)∗(cid:1)⊥(cid:17)
G
.
Remark 4.2. Part (c) of Lemma 4.1 illuminates some hidden implications of parts (a)
and (b). For instance, κ can only be supported on elements g with codim V g ∈ {0, 2},
which is readily seen from part (c) by noting that negative exterior powers are zero and
that an element g with codimension one acts nontrivially on H 2,d
.
g
In practice, one may simplify the computation of (H •)G by computing centralizer in-
variants for a set of conjugacy class representatives and then expanding into G-invariants.
g )Z(g), where C is a set of conjugacy class representatives,
and Z(g) is the centralizer of g. We review the explicit passage from a Z(g)-invariant to
a G-invariant, which will be especially relevant in translating the results of Lemma 4.5
Formally, (H •)G ∼=Lg∈C (H •
into the maps in Definition 4.6. Recall that a cochain α = Pg∈G αg is G-invariant
if and only if hαg = αhgh−1 for all g, h ∈ G. Thus, if α is a G-invariant cochain,
then αg is Z(g)-invariant for each g, and α is determined by its components for a
set of conjugacy class representatives. In particular, a centralizer invariant αg extends
uniquely to a G-invariant element, supported on the conjugacy class of g, via the map
10
B. FOSTER-GREENWOOD AND C. KRILOFF
αg 7→Ph∈[G/Z(g)]
have the following commutative diagram:
hαg, where [G/Z(g)] is a set of left coset representatives. Further, we
αg ∈ (H 2,d
g )Z(g)
α ∈ (H 2,d)G
κg :V2 V → Sd(V g)
κ :V2 V → Mh∈[G/Z(g)]
Sd(V hgh−1
)hgh−1.
The vertical arrows are via the isomorphism Sd(V g)g ⊗V2 V ∗ ∼= Hom(V2 V, Sd(V g)g).
The horizontal arrows are via the orbit-sum maps
αg 7→ Xh∈[G/Z(g)]
hαg
and
κg 7→ Xh∈[G/Z(g)]
hκg
hg.
Hochschild cohomology for symmetric groups. We now turn to the specific ex-
ample of cohomology of skew group algebras of symmetric groups with the natural
permutation representation. Much of the Hochschild cohomology of S(V )#Sn may be
extracted as subcases of Hochschild cohomology for skew group algebras of complex
reflection groups G(r, p, n) found in [SW08]. However, we provide computations for
Sn ∼= G(1, 1, n) here for purposes of self-containment and notational consistency.
For the remainder of the section, let e1, . . . , en be the standard basis of V ∼= Cn, and
let the symmetric group Sn act on V by its natural permutation representation. Thus
for σ in Sn, we have σei = eσ(i).
We first show that for the symmetric group acting by its natural permutation repre-
sentation, elements of Hochschild 2-cohomology with polynomial degree zero or one can
only be supported on the identity or 3-cycles.
Lemma 4.3. Let Sn (n ≥ 3) act on V ∼= Cn by its natural permutation representation,
αg be an element of (H 2,1 ⊕ H 2,0)Sn. If g is not the identity and not
a 3-cycle, then αg = 0.
and let α =Pg∈Sn
Proof. Let α =Pg∈Sn
αg be an element of the cohomology space (H 2,1 ⊕ H 2,0)Sn. If
codim(V g) 6∈ {0, 2}, then αg = 0 by Remark 4.2. Under the permutation representation,
the only elements with codim V g ∈ {0, 2} are the identity, the 3-cycles, and the double-
transpositions, so it remains to show that αg = 0 if g is a double-transposition. In fact,
since α is determined by its components for a set of conjugacy class representatives, it
suffices to show (H 2,1
g )Z(g) = 0 for g = (12)(34). The vectors
g ⊕ H 2,0
v1 = e1 − e2 − e3 + e4,
v2 = e1 − e2 + e3 − e4,
v3 = e1 + e2 − e3 − e4,
v4 = e1 + e2 + e3 + e4,
vk = ek
for 5 ≤ k ≤ n
ORBIFOLD ALGEBRAS FOR Sn
11
form a g-eigenvector basis of V with (V g)⊥ = Span{v1, v2} and V g = Span{v3, . . . , vn}.
The wedge product v∗
1 ∧ v∗
vol⊥
2 is a scalar multiple of the volume form
4 ∧ e∗
g := e∗
1,
3 + e∗
2 + e∗
4 + e∗
3 ∧ e∗
2 ∧ e∗
1 ∧ e∗
so vol⊥
elements of H 2,1
g is a basis forV2((V g)⊥)∗. The transposition (12) commutes with g but scales
g by negative one, so (H 2,1
g )Z(g) = 0.
g ⊕ H 2,0
g ⊕ H 2,0
(cid:3)
The cohomology elements in the next lemma give rise to the parameter maps of the
Lie orbifold algebras exhibited in Theorem 3.1.
Lemma 4.4. Let G = Sn (n ≥ 3) act on V ∼= Cn by its natural permutation represen-
tation. The subspace of (H 2,1 ⊕ H 2,0)Sn consisting of elements supported only on the
identity is one-dimensional with basisP1≤i<j≤n(ei − ej) ⊗ e∗
Proof. Since the permutation representation of Sn is a self-dual reflection representation,
generators of (H •
rem from the invariant theory of reflection groups (see [Sol63], or the expository [Kan01,
Chapter 22]). Specifically, the power sums fk = ek
n with 1 ≤ k ≤ n form a set
of algebraically independent invariant polynomials, and the differential forms
1 )G ∼= (S(V ) ⊗V V ∗)G ∼= (S(V ) ⊗V V )G are given by Solomon's Theo-
1 + · · · + ek
i ∧ e∗
j .
αk :=
1
k + 1
∂fk+1
∂ei
⊗ e∗
i = ek
1 ⊗ e∗
1 + · · · + ek
n ⊗ e∗
n
for 0 ≤ k ≤ n − 1
nXi=1
generate (S(V ) ⊗V V ∗)G as an exterior algebra over S(V )G ∼= C[f1, . . . , fn]. Thus,
(S(V ) ⊗V2 V ∗)G is freely generated as an S(V )G-module by {αkαl 0 ≤ k < l ≤ n − 1}.
Since αkαl has polynomial degree k + l, every element of (H 2,1
multiple of α1α0.
1 )G is a scalar
1 ⊕ H 2,0
(cid:3)
The cohomology in the next lemma serves two purposes. The polynomial degree one
elements give rise to pre-Drinfeld orbifold algebra maps supported only on 3-cycles,
while the polynomial degree zero elements are needed in constructing multiple liftings
of a pre-Drinfeld orbifold algebra map.
Lemma 4.5. Let Sn (n ≥ 3) act on V ∼= Cn by its natural permutation representation.
The subspace of (H 2,1 ⊕ H 2,0)Sn consisting of elements supported only on 3-cycles is
two-dimensional if n = 3 and three-dimensional if n ≥ 4.
Proof. Recall that a cohomology element is determined by its components for a set of
conjugacy class representatives. Thus, if α is supported only on 3-cycles, it suffices to
choose a representative, say g = (123), and find a basis of
(H 2,1
g ⊕ H 2,0
g )Z(g) ∼= ((V g ⊕ C) ⊗V2(cid:0)(V g)⊥(cid:1)∗ ⊗ Cg)Z(g).
Let ω = e2πi/3. Then the vectors
v1 = e1 + ω2e2 + ωe3,
v2 = e1 + ωe2 + ω2e3,
v3 = e1 + e2 + e3,
vk = ek
for 4 ≤ k ≤ n
12
B. FOSTER-GREENWOOD AND C. KRILOFF
form a g-eigenvector basis of V with (V g)⊥ = Span{v1, v2} and V g = Span{v3, . . . , vn}.
The wedge product v∗
1 ∧ v∗
2 is a scalar multiple of the volume form
vol⊥
g := e∗
3 ∧ e∗
1,
2 + e∗
3 + e∗
1 ∧ e∗
2 ∧ e∗
so vol⊥
g is a basis forV2((V g)⊥)∗. Each element of the centralizer
Z(123) = h(123)i × Sym{4,...,n}
acts trivially on vol⊥
g , so
(H 2,1
g ⊕ H 2,0
g )Z(g) ∼= ((V g)Z(g) ⊕ C) ⊗V2(cid:0)(V g)⊥(cid:1)∗ ⊗ Cg.
The vectors e1 + e2 + e3 and e4 + · · · + en form a basis of (V g)Z(g), so the cohomology
elements
α(123) = (e1 + e2 + e3) ⊗ vol⊥
and β(123) = (e4 + · · · + en) ⊗ vol⊥
(123) ⊗ (123)
(123) ⊗ (123)
span (H 2,1
(123))Z(g), while γ(123) = vol⊥
g ⊗ (123) spans (H 2,0
(123))Z(g).
(cid:3)
The following definition arises from expanding the centralizer invariants determined
in the proof of Lemma 4.5 into Sn-invariants and applying the isomorphism Sd(V g)g ⊗
Definition 4.6. For parameters a, b ∈ C, let κL
linear 2-cochain with component maps κL
V2 V ∗ ∼= Hom(V2 V, Sd(V g)g). (See the discussion following Remark 4.2.)
(ijk) :V2 V → V (ijk) defined by
(ijk)(ek, ei) = a(ei + ej + ek) + b Xl6=i,j,k
3-cyc = P(ijk)∈Sn
κL
(ijk)(ei, ej ) = κL
(ijk)(ej , ek) = κL
el
κL
(ijk) ⊗ (ijk) be the
and κL
3-cyc(el, em) = 0 if {el, em} ∩ V (ijk) 6= ∅.
For a parameter c ∈ C, let κC
with component maps κC
κC
(ijk) ⊗ (ijk) be the constant cochain
3-cyc =P(ijk)∈Sn
(ijk) :V2 V → C defined by
κC
(ijk)(ei, ej) = κC
(ijk)(ej , ek) = κC
(ijk)(ek, ei) = c
and κC
3-cyc(el, em) = 0 if {el, em} ∩ V (ijk) 6= ∅.
Also let κ3-cyc = κL
3-cyc + κC
3-cyc.
Notice that if a and b are not both zero, then ker κL
(ijk) = V (ijk), and if c 6= 0, then
ker κC
(ijk) = V (ijk).
In view of the equivalences in Lemma 4.1, the polynomial degree one elements of
Hochschild 2-cohomology computed in Lemmas 4.3, 4.4, and 4.5 yield a description of
all pre-Drinfeld orbifold algebra maps.
Corollary 4.7. The pre-Drinfeld orbifold algebra maps for Sn (n ≥ 3) acting by its
natural permutation representation are the linear 2-cochains κL = κL
3-cyc, with
3-cyc as in Definition 4.6 and κL
κL
1 (ei, ej) = a1(ei − ej) for some a1 ∈ C.
1 + κL
ORBIFOLD ALGEBRAS FOR Sn
13
In Theorems 6.1 and 6.2, we will show that the maps κL
3-cyc lift (separately, but
not in combination) to Drinfeld orbifold algebra maps. Any two liftings of a particular
pre-Drinfeld orbifold algebra map must differ by a constant 2-cochain that satisfies
the mixed Jacobi identity. Recalling Lemma 4.1, the desired constant 2-cochains are
revealed by the polynomial degree zero elements of Hochschild 2-cohomology computed
in Lemmas 4.3, 4.4, and 4.5.
1 and κL
Corollary 4.8. The Sn-invariant constant 2-cochains satisfying the mixed Jacobi iden-
tity are the maps κC
3-cyc as in Definition 4.6.
5. Notation and reduction techniques
In this section, we gather notation and reduction techniques to facilitate the process
of lifting pre-Drinfeld orbifold algebra maps κL to Drinfeld orbifold algebra maps κ =
κL + κC. We first introduce operators on cochains to make it easier to refer to the
properties of a Drinfeld orbifold algebra map (Definition 2.1) for a group G acting
linearly on a vector space V ∼= Cn. Along the way, we indicate how our notation relates
to the cohomological interpretation (see [SW12a, Theorem 7.2]) of each property. We
then record symmetries that reduce the computations involved in clearing obstructions.
We use these symmetries heavily in Section 7.
A variation on the coboundary. First, we define a map ψ to compactly describe the
left-hand side of Property (2.2) and the right-hand side of Property (2.3) of a Drinfeld
orbifold algebra map. For a linear or constant 2-cochain α, let ψ(α) =Pg∈G ψgg be the
3-cochain with components ψg :V3 V → S(V ) given by
ψg(v1, v2, v3) = αg(v1, v2)(gv3 − v3) + αg(v2, v3)(gv1 − v1) + αg(v3, v1)(gv2 − v2).
The map ψ is the negation of the coboundary operator on cochains arising from the
Koszul resolution. In particular, ψ(κL) = −d∗
3 is the
coboundary operator that takes two-cochains to three-cochains (see the proof of [SW12a,
Lemma 7.1]).
3κL and ψ(κC ) = −d∗
3κC, where d∗
A variation on the cochain bracket. Next, we define a map φ to compactly describe
the left-hand sides of Properties (2.3) and (2.4) of a Drinfeld orbifold algebra map. For
α a linear or constant 2-cochain and β a linear 2-cochain, let φ(α, β) = Pg∈G φgg be
the 3-cochain with components φg =Pxy=g φx,y, where φx,y :V3 V → V ⊕ C is given
by
φx,y(v1, v2, v3) = αx(v1+yv1, βy(v2, v3))+αx(v2+yv2, βy(v3, v1))+αx(v3+yv3, βy(v1, v2)).
Thus φ(α, β) is G-graded, with components φg, and also (G × G)-graded, with compo-
nents φx,y. The map φ(α, β) is closely related to the cochain bracket [α, β] in [SW12a,
Definition 5.6, Corollary 6.7]. Indeed, φ(κL, κL) = − 1
2 [κL, κL] and φ(κC , κL) = −[κC, κL],
as explained in [SW12a, proof of Lemma 7.1].
To the extent possible, and especially in Section 7, we make remarks or calculations
g and
that apply to both φ(κL, κL) and φ(κC , κL), and in these instances, we write φ∗
φ∗
x,y for the corresponding components of φ(κ∗, κL) where ∗ denotes either L or C.
14
B. FOSTER-GREENWOOD AND C. KRILOFF
Drinfeld orbifold algebra maps (condensed definition). Equipped with the defi-
nitions of ψ and φ, the properties of a Drinfeld orbifold map κ = κL +κC (Definition 2.1)
may be expressed succinctly:
(2.0) im κL
g ⊆ V g for each g in G,
(2.1) the map κ is G-invariant,
(2.2) ψ(κL) = 0,
(2.3) φ(κL, κL) = 2ψ(κC ),
(2.4) φ(κC , κL) = 0.
Invariance relations. Recall that a cochain α = Pg∈G αgg with components αg :
Vk V → S(V ) is G-invariant if and only if hαg = αhgh−1 for all g, h ∈ G. Equivalently,
h(αg(v1, . . . , vk)) = αhgh−1(hv1, . . . , hvk)
for all g, h ∈ G and v1, . . . , vk ∈ V . Thus a G-invariant cochain is determined by its
components for a set of conjugacy class representatives.
In the following lemma, one can let α = κL or α = κC and let β = κL to see that
if κL and κC are G-invariant, then φ(κ∗, κL) and ψ(κ∗) are also G-invariant. This is
helpful because, for instance, if φg = 2ψg for some g ∈ G, then acting by h ∈ G on both
sides shows φhgh−1 = 2ψhgh−1 also. Thus if φg = 2ψg for all g in a set of conjugacy class
representatives, then φ(κL, κL) = 2ψ(κC ). Similar reasoning applies to Properties (2.2)
and (2.4) of a Drinfeld orbifold algebra map.
Lemma 5.1. Let G be a finite group acting linearly on V ∼= Cn.
If α and β are
G-invariant 2-cochains with β linear and α linear or constant, then φ(α, β) and ψ(α)
are G-invariant. Specifically, at the component level, we have for all g, h ∈ G and
v1, v2, v3 ∈ V
h(φx,y(v1, v2, v3)) = φhxh−1,hyh−1(hv1, hv2, hv3),
and
Proof. To see that
h(φg(v1, v2, v3)) = φhgh−1(hv1, hv2, hv3),
h(ψg(v1, v2, v3)) = ψhgh−1(hv1, hv2, hv3).
h(φx,y(v1, v2, v3)) = φhxh−1,hyh−1(hv1, hv2, hv3)
for all x, y, h ∈ G and v1, v2, v3 ∈ V , note that, using invariance of α and β,
h(αx(vi + yvi, βy(vj, vk))) = αhxh−1(hvi + hyvi, βhyh−1 (hvj, hvk))
= αhxh−1(hvi + hyh−1
(hvi), βhyh−1(hvj, hvk)).
Then also
where the last equality holds since the correspondence (x, y) ↔ (hxh−1, hyh−1) is a
bijection between the set of factor pairs of g and the set of factor pairs of hgh−1.
hφg = Xxy=g
hφx,y = Xxy=g
φhxh−1,hyh−1 = φhgh−1,
ORBIFOLD ALGEBRAS FOR Sn
15
To see that h(ψg(v1, v2, v3)) = ψhgh−1(hv1, hv2, hv3), again use that h(αg(vi, vj)) =
(cid:3)
αhgh−1(hvi, hvj) and hgvk = hgh−1
Orbits of factorizations. The next observations involve the action of G on G × G by
diagonal (componentwise) conjugation and provide a method for narrowing the number
of terms and basis triples we must consider in evaluating φ(κ∗, κL) in Section 7.
(hvk).
If expressions φx,y(u, v, w) are organized in an array with rows indexed by factoriza-
tions xy of g and columns indexed by basis triples {u, v, w}, then φg(u, v, w) corresponds
to a column sum. Our goal is to use invariance relations to show how to use column
sums in a carefully chosen subarray to determine the column sums for the full array.
We first consider the effect of acting on a column sum for a subarray with rows indexed
by factorizations in the same orbit under a subgroup.
Lemma 5.2. Let G be a finite group acting linearly on V ∼= Cn, and let α and β be
G-invariant 2-cochains with β linear and α linear or constant. Recall that φ(α, β) has
components φg = Pxy=g φx,y. Fix g in G, and let H be a subgroup of the centralizer
Z(g). If g = xy, then for all z in Z(g) and u, v, w in V ,
φzx′,zy′(zu, zv, zw)
φx′,y′(zu, zv, zw).
z X(x′,y′)∈H (x,y)
φx′,y′(u, v, w) = X(x′,y′)∈H (x,y)
= X(x′,y′)∈zH (x,y)
Proof. The first equality is an application of Lemma 5.1, and the second equality holds
because the elements of H(x, y) and zH(x, y) = {z(x′, y′) (x′, y′) ∈ H (x, y)} are in
bijection via the map (x′, y′) 7→ z(x′, y′).
(cid:3)
We use the following characterization of when "coset orbits" of factorizations coincide
to ensure there is no double-counting in the proof of Lemma 5.4.
Lemma 5.3. Let G be a finite group, and let g be an element of G with factorization
g = xy. Let K = Z(x) ∩ Z(y), the stabilizer of (x, y) under componentwise conjugation,
and let H be a subgroup of Z(g) normalized by K. Then the orbits z1H(x, y) and z2H (x, y)
are disjoint or equal, with equality if and only if z1HK = z2HK.
Proof. Let z1, z2 ∈ Z(g) and suppose the orbits z1H(x, y) and z2H (x, y) intersect non-
trivially so that z1h1(x, y) = z2h2(x, y) for some h1, h2 in H. Then h−1
1 z2h2 ∈ K, so
z−1
1 z2 ∈ HKH = HK (since K normalizes H), and hence z1HK = z2HK. Then
1 z−1
z1H(x, y) = z1HK (x, y) = z2HK (x, y) = z2H (x, y).
(cid:3)
The next lemma, stated in the specific case of the symmetric group, provides a method
for using subgroups to reduce the number of expressions φx,y(ei, ej, ek) that must be
evaluated when verifying φ(α, β) = 0. Choosing the subgroup is a balancing act -- using
a small subgroup decreases the number of factorizations to consider but typically in-
creases the number of basis triples, while using a large subgroup increases the number
of factorizations to consider but decreases the number of basis triples.
16
B. FOSTER-GREENWOOD AND C. KRILOFF
Lemma 5.4. Let Sn act on V ∼= Cn by its natural permutation representation, and
let α and β be Sn-invariant 2-cochains with β linear and α linear or constant. Recall
that φ(α, β) has components φg =Pxy=g φx,y. Suppose g is in Sn and has factorization
g = xy. Let K = Z(x) ∩ Z(y), the stabilizer of (x, y) under componentwise conjugation,
and let H be a subgroup of Z(g) normalized by K. Let B be the set of all three element
subsets of {e1, . . . , en}, and let BH be a set of H-orbit representatives of B. If
φx′,y′(ei, ej, ek) = 0
for all {ei, ej, ek} ∈ BH,
then
φx′,y′(ei, ej, ek) = 0
for all {ei, ej, ek} ∈ B.
X(x′,y′)∈H (x,y)
X(x′,y′)∈Z(g)(x,y)
X(x′,y′)∈H (x,y)
X(x′,y′)∈zH (x,y)
Proof. Use Lemma 5.2 with z ranging over the elements of H to show that
φx′,y′(ei, ej, ek) = 0 for all {ei, ej, ek} ∈ B.
Then use Lemma 5.2 again to show for each z in Z(g),
φx′,y′(ei, ej , ek) = 0 for all {ei, ej , ek} ∈ B.
Let [Z(g)/HK] be a set of left coset representatives of HK, and use Lemma 5.3 to
conclude,
φx′,y′(ei, ej, ek)
X(x′,y′)∈Z(g)(x,y)
= Xz∈[Z(g)/HK]
X(x′,y′)∈zH (x,y)
φx′,y′(ei, ej , ek) = 0 for all {ei, ej, ek} ∈ B.
(cid:3)
Remark 5.5. Though Lemma 5.4 is stated for the symmetric group, a similar idea
might be useful in other groups where each centralizer acts by monomial matrices with
respect to some basis.
6. Lifting to deformations
In Section 4 we determined all pre-Drinfeld orbifold algebra maps for the symmetric
group acting by its natural permutation representation. When n ≥ 4, the space of such
maps is three-dimensional with one dimension of maps supported only on the identity
and two dimensions of maps supported only on 3-cycles. We now determine for which
candidate maps κL there exists a constant 2-cochain κC so that κ = κL +κC also satisfies
Properties (2.3) and (2.4) of a Drinfeld orbifold algebra map. We consider three cases, κL
supported only on the identity (Theorem 6.1), supported only on 3-cycles (Theorem 6.2),
and finally a combination supported on both the identity and 3-cycles (Proposition 6.3).
ORBIFOLD ALGEBRAS FOR Sn
17
Lie orbifold algebra maps. The next theorem shows that for the symmetric group
acting by its permutation representation, every pre-Drinfeld orbifold algebra map sup-
ported only on the identity lifts uniquely to a Lie orbifold algebra map. The correspond-
ing Lie orbifold algebras are described in Theorem 3.1.
Theorem 6.1. The Lie orbifold algebra maps for the symmetric group Sn (n ≥ 3)
acting on V ∼= Cn by the natural permutation representation form a one-dimensional
vector space generated by the map κ : V2 V → V ⊗ CSn with κ(ei, ej) = ei − ej for
1 ≤ i < j ≤ n.
Proof. Let κL be a pre-Drinfeld orbifold algebra map supported only on the identity.
By Corollary 4.7, we have κL(ei, ej) = a(ei − ej) for some a ∈ C. It is straightforward to
show that φ(κL, κL) = 0, so now Property (2.3), φ(κL, κL) = 2ψ(κC ), holds if and only if
ψ(κC ) = 0. By Corollary 4.8, the G-invariant constant 2-cochains such that ψ(κC ) = 0
(i.e., satisfying the mixed Jacobi identity) are supported only on 3-cycles and have
κC
(ijk)(ei, ej) = κC
(ijk)(ej , ek) = κC
(ijk)(ek, ei) = c
for some scalar c.
Turning to Property (2.4), if c = 0, then κC ≡ 0, so φ(κC , κL) = 0, and κ = κL is a
Lie orbifold algebra map. If c 6= 0, then κ = κL + κC is not a Lie orbifold algebra map
since φ(κC , κL) 6= 0. In particular, the component φ(123) of φ(κC , κL) is nonzero on the
basis triple e1, e2, e3:
2[κC
(123)(e1, e2 − e3) + κC
(123)(e2, e3 − e1) + κC
(123)(e3, e1 − e2)] = 12c 6= 0.
In general, a lifting need not be unique. See [SW12a, Example 4.3] for an example of
a cyclic group having a Lie orbifold algebra map with κL and κC both nonzero.
(cid:3)
Other Drinfeld orbifold algebra maps. Next we describe all possible Drinfeld orb-
ifold algebra maps supported only off of the identity. We outline the proof here but
relegate the details of clearing the obstructions to Section 7. The corresponding Drin-
feld orbifold algebras are described in Theorem 3.2.
Theorem 6.2. For Sn (n ≥ 3) acting on V ∼= Cn by its natural permutation represen-
tation, the Drinfeld orbifold algebra maps supported only off the identity are precisely
the maps of the form κ = κL
3-cyc, with κ3-cyc as in Definition 4.6 and
κC
5-cyc as in Definition 7.4.
Proof. Suppose κL is a pre-Drinfeld orbifold algebra map supported only off of the
identity. By Corollary 4.7, we must have κL = κL
3-cyc for some parameters a, b ∈ C as
in Definition 4.6. Now, the goal is to find all G-invariant maps κC such that Properties
(2.3) and (2.4) of a Drinfeld orbifold algebra map also hold.
3-cyc + κC
5-cyc + κC
First, we find a particular lifting.
• First obstruction. In Propositions 7.2 and 7.3, we evaluate φ(κL
3-cyc, κL
3-cyc). The
results suggest Definition 7.4 of an Sn-invariant map κC
5-cyc so that
φ(κL
3-cyc, κL
3-cyc) = 2ψ(κC
5-cyc),
18
B. FOSTER-GREENWOOD AND C. KRILOFF
as verified in Proposition 7.5.
• Second obstruction. In Proposition 7.7, we show φ(κC
5-cyc, κL
3-cyc) = 0.
Thus κ = κL
3-cyc + κC
5-cyc is a Drinfeld orbifold algebra map.
Next, we see how the particular lifting can be modified to produce all other possible
liftings. Let κC be any G-invariant constant 2-cochain.
3-cyc, κL
• First obstruction.
Given that φ(κL
3-cyc, κL
φ(κL
ψ(κC − κC
tion 4.6 for some parameter c ∈ C.
3-cyc) = 2ψ(κC ) if and only if ψ(κC − κC
5-cyc) = 0 if and only if κC − κC
5-cyc = κC
3-cyc) = 2ψ(κC
5-cyc), we have that
5-cyc) = 0. By Corollary 4.8,
3-cyc, with κC
3-cyc as in Defini-
• Second obstruction.
In Propositions 7.2 and 7.3, we show φ(κC
3-cyc, κL
3-cyc) = 0,
and in Proposition 7.7, we show φ(κC
5-cyc, κL
3-cyc) = 0, so
φ(κC
3-cyc + κC
5-cyc, κL
3-cyc) = φ(κC
3-cyc, κL
3-cyc) + φ(κC
5-cyc, κL
3-cyc) = 0.
Thus the liftings of κL
3-cyc + κC
κ = κL
3-cyc + κC
3-cyc to a Drinfeld orbifold algebra map are the maps of the form
5-cyc.
(cid:3)
Finally, we show for Sn that there are no Drinfeld orbifold algebra maps that are
supported both on and off the identity.
Proposition 6.3. Let κL be a pre-Drinfeld orbifold algebra map for the natural permu-
tation representation of the symmetric group Sn (n ≥ 3). If κL is supported both on the
identity and off the identity, then κL does not lift to a Drinfeld orbifold algebra map.
1 + κL
3-cyc, where κL
Proof. Let κL 6≡ 0 be a pre-Drinfeld orbifold algebra map. By Corollary 4.7, we have
κL = κL
3-cyc is as in Definition 4.6 with parameters a, b ∈ C, and
κL
1 (ei, ej) = a1(ei − ej) for some parameter a1 ∈ C. Suppose κC is an Sn-invariant
constant 2-cochain. We show that if κC clears the first obstruction, i.e., φ(κL, κL) =
2ψ(κC ), then a = b = 0 or a1 = 0, so κL is supported only on the identity or only on
the 3-cycles.
We let g = (123) and compare the g-components of 2ψ(κC ) and φ(κL, κL). First,
note that since κC is Sn-invariant,
κC
(123)(e1, e2) = κC
(123)(e2, e3) = κC
(123)(e3, e1),
and so
ψ(123)(e1, e2, e3) = κC
On the other hand,
(123)(e1, e2)(e1−e3)+κC
(123)(e2, e3)(e2−e1)+κC
(123)(e3, e1)(e3−e2) = 0.
φ(κL, κL) = φ(κL
1 , κL
The (123)-component of φ(κL
component of φ(κL
3-cyc, κL
1 ) + φ(κL
1 , κL
1 ) is certainly zero, and by Proposition 7.2, the (123)-
3-cyc) is also zero. Only the cross terms remain, so
3-cyc, κL
1 ) + φ(κL
1 , κL
3-cyc) + φ(κL
3-cyc, κL
3-cyc).
φ(123) = φ(123),1 + φ1,(123).
Recall
φx,y(ei, ej, ek) = κL
x [ei + yei, κL
y (ej , ek)] + κL
x [ej + yej, κL
y (ek, ei)] + κL
x [ek + yek, κL
y (ei, ej)].
ORBIFOLD ALGEBRAS FOR Sn
19
The cross terms of φ(123) evaluated on the basis triple e1, e2, e3 are
φ(123),1(e1, e2, e3) = 2a1(κL
= 4a1(κL
= 12a1[a(e1 + e2 + e3) + b(e4 + · · · + en)]
(123)[e1, e2 − e3] + κL
(123)[e1, e2] + κL
(123)[e2, e3 − e1] + κL
(123)[e3, e1])
(123)[e2, e3] + κL
(123)[e3, e1 − e2])
and
φ1,(123)(e1, e2, e3) = 2κL
1 [e1 + e2 + e3, κL
(123)(e1, e2)]
= 2a1b[(n − 3)(e1 + e2 + e3) − 3(e4 + · · · + en)],
which sum to
φ(123)(e1, e2, e3) = 2a1[(6a + (n − 3)b)(e1 + e2 + e3) + 3b(e4 + · · · + en)].
Thus, if n ≥ 4, then φ(123)(e1, e2, e3) = 2ψ(123)(e1, e2, e3) if and only if a1 = 0 or
3-cyc has only one parameter, a, and φ(123)(e1, e2, e3) =
b = 0 = a.
12a1a(e1 + e2 + e3), which is zero if and only if a1 = 0 or a = 0.
(cid:3)
If n = 3, then κL
7. Clearing the obstructions
In Section 4, we defined a pre-Drinfeld orbifold algebra map κL
3-cyc. This section is
devoted to lifting κL
3-cyc to a Drinfeld orbifold algebra map and provides the details of the
proof of Theorem 6.2 outlined in Section 6. In view of Definition 2.1, we first evaluate
φ(κL
such that φ(κL
Theorem 9.2]. We then "clear the second obstruction" by showing φ(κC
These computations show that κ = κL
5-cyc
5-cyc). Existence of such a map is predicted by [SW12b,
3-cyc) = 0.
3-cyc) and "clear the first obstruction" by defining a G-invariant map κC
5-cyc is a Drinfeld orbifold algebra map.
3-cyc) = 2ψ(κC
3-cyc + κC
3-cyc, κL
5-cyc, κL
3-cyc, κL
Clearing the First Obstruction. We begin by recording simplifications of φ∗
x,y, a
summand of the component φ∗
3-cyc), where ∗ stands for L or C. Simplifi-
cation of φ∗
x,y(ei, ej, ek) depends on the location of the basis vectors relative to the fixed
spaces V x and V y, so we use the following indicator function. For y ∈ Sn and v ∈ V ,
let
g of φ(κ∗
3-cyc, κL
δy(v) =(1 if v ∈ V y
0 otherwise.
Lemma 7.1. Let κ∗
a term of the component φ∗
and let 1 ≤ i, j, k ≤ n.
3-cyc with ∗ = L or ∗ = C be as in Definition 4.6 and let φ∗
x,y denote
3-cyc). Let x and y be 3-cycles such that xy = g,
g of φ(κ∗
3-cyc, κL
(1) If ei, ej ∈ V y, then φ∗
(2) If ei ∈ V y ∩ V x, then φ∗
(3) If ei ∈ V y\V x and ej 6∈ V y, then
x,y(ei, ej , ek) = 0.
x,y(ei, ej , ek) = 0.
x,y(ei, ej, yej) = 2(b − a)[δy(xei) − δy(x−1
φ∗
ei)]κ∗
x[ei, xei].
20
B. FOSTER-GREENWOOD AND C. KRILOFF
(4) If ei /∈ V y, then φ∗
x,y(ei, yei, y2
ei) = 0.
Note that φ∗
along with these cases.
x,y can be evaluated on any basis triple by using the alternating property
Proof. Consider
φ∗
x,y(ei, ej, ek) = κ∗
Recall that if z is a 3-cycle then V z ⊆ ker κ∗
z.
y (ej, ek)] + κ∗
x[ei + yei, κL
x[ej + yej, κL
y (ek, ei)] + κ∗
x[ek + yek, κL
y (ei, ej)].
(1) If ei, ej ∈ V y, then φ∗
(2) If ei ∈ V y ∩ V x, then the first term of φ∗
x,y(ei, ej, ek) = 0 since V y ⊆ ker κL
y .
x,y vanishes because ei ∈ V x ⊆ ker κ∗
x,
and the second two terms of φ∗
x,y vanish because ei ∈ V y ⊆ ker κL
y .
(3) If ei ∈ V y\V x and ej 6∈ V y, then φ∗
bilinearity and V x ⊆ ker κ∗
sions κ∗
y (ej, yej)]. Using
x, the right hand side is a linear combination of expres-
x[ei, hei] for h ∈ hxi. The appropriate coefficients, a or b, can be extracted
x. Thus,
x,y(ei, ej, yej) = 2κ∗
hei ∈ V x ⊆ ker κ∗
x[ei, κL
φ∗
x,y(ei, ej, yej) = 2κ∗
using the fixed space indicator function. Also,Ph∈hxi
x[ei, hei] + 2b Xh∈hxi
y (ej, yej)]
(1 − δy(hei))κ∗
x[ei, κL
δy(hei)κ∗
x[ei, hei]
= 2a Xh∈hxi
= 2(b − a) Xh∈hxi
δy(hei)κ∗
x[ei, hei]
= 2(b − a)[δy(xei) − δy(x−1
ei)]κ∗
(4) Lastly, if ei /∈ V y, note that φ∗
x,y(ei, yei, y2
y (yei, y2
ei) = κL
y (ei, yei) = κL
since κL
combination of the vector u = ei + yei + y2
(which is in the kernel of κ∗
x,y(ei, yei, y2
φ∗
x), to see that
ei) = 2κ∗
y (y2
x(u, (a − b)u + bu0) = 0.
x[ei, xei].
ei) = 2κ∗
ei, ei). Express κL
x[ei + yei + y2
y (ei, yei)]
y (ei, yei) as a linear
ei and the vector u0 = e1 + · · · + en
ei, κL
As mentioned in the outline of the proof of Theorem 6.2, the next two propositions
(cid:3)
are used to evaluate both φ(κL
3-cyc, κL
3-cyc) and φ(κC
3-cyc, κL
3-cyc).
Proposition 7.2. Let κ∗
let φ∗
g be the g-component of φ(κ∗
3-cyc with ∗ = L or ∗ = C be as in Definition 4.6. For g ∈ Sn,
3-cyc, κL
3-cyc). If g is not a 5-cycle then φ∗
g ≡ 0.
Proof. Since κ∗
3-cyc is supported only on 3-cycles, we first determine the cycle types that
arise as a product xy with x and y both 3-cycles. Since σxσy = σ(xy), it suffices to
examine representatives of orbits of factor pairs (x, y) under the action of Sn by diagonal
conjugation. Orbit representatives and their products are
(123)(456),
(123)(324) = (124),
(123)(345) = (12345),
(123)(123) = (132),
(123)(234) = (12)(34),
(123)(321) = 1.
ORBIFOLD ALGEBRAS FOR Sn
21
If the cycle type of g does not appear in this list, then certainly φ∗
case g = (12345) to Proposition 7.3 and show here that φ∗
g = (12)(34), and g = (123), and hence also for their conjugates.
g ≡ 0. We leave the
g ≡ 0 for g = 1, g = (123)(456),
Besides narrowing the set of representative elements g to consider, the list of orbit
representatives reveals a way to organize the terms φ∗
g. Specifically, if the cycle
type of g occurs with multiplicity m in the list, then the factor pairs with product g are in
m orbits under the diagonal conjugation action of Z(g), and we can use a representative
from each orbit to generate all the terms φ∗
x,y needed to evaluate φ∗
g.
x,y of φ∗
Case 1 (g = 1). The identity component φ∗
terms φ∗
im κL
that φ∗
3-cyc) reduces to the sum of
x,x−1, where x ranges over the set of 3-cycles in Sn. For each 3-cycle x, we have
x−1 (v, w)] = 0 for all u, v, w ∈ V . It follows
x,x−1 ≡ 0 for each 3-cycle x ∈ Sn, and hence, φ∗
x−1 ⊆ V x−1
= V x ⊆ ker κ∗
1 of φ(κ∗
3-cyc, κL
x, so κ∗
x[u, κL
1 ≡ 0.
Case 2 (g = (123)(456)). If g = (123)(456), then the component φ∗
reduces to the sum of two terms, φ∗
im κL
and y are exchanged. It follows that both terms φ∗
vectors, and hence φ∗
3-cyc)
y,x, where x = (123) and y = (456). Note that
y (v, w)] = 0 for all u, v, w ∈ V , and the same holds if x
y,x are zero on any triple of
y ⊆ V x ⊆ ker κ∗
x,y and φ∗
g of φ(κ∗
3-cyc, κL
x, so κ∗
x,y + φ∗
x[u, κL
g ≡ 0.
Case 3 (g = (12)(34)). Note that Z((12)(34)) = h(1324), (12)i × Sym{5,...,n}, and
φ∗
g =
X
(x,y)∈Z(g)((123),(234))
φ∗
x,y = φ∗
(123),(234) + φ∗
(342),(421) + φ∗
(214),(143) + φ∗
(431),(312)
+ φ∗
(213),(134) + φ∗
(432),(321) + φ∗
(124),(243) + φ∗
(341),(412).
Applying Lemma 5.4 with H = h(1324)i, which is a normal subgroup of Z(g), yields
that to show φ∗
g ≡ 0, it is sufficient to prove
X(x,y)∈H ((123),(234))
φ∗
x,y(ei, ej, ek) = [φ∗
(123),(234)+φ∗
(342),(421)+φ∗
(214),(143)+φ∗
(431),(312)](ei, ej, ek) = 0
for H-orbit representatives {ei, ej, ek} of the basis triples. Since V x ∩ V y ⊆ ker φx,y
holds by Lemma 7.1 (2), we only consider {ei, ej, ek} ⊆ {e1, e2, e3, e4}, and since the
three element subsets of {e1, e2, e3, e4} are in the same H-orbit, it suffices to show
[φ∗
(123),(234) + φ∗
(342),(421) + φ∗
(214),(143) + φ∗
(431),(312)](e1, e2, e3) = 0.
The first three terms are zero by Lemma 7.1 (3) since xei, x−1
and the fourth term is zero by Lemma 7.1 (4). Hence φ∗
(12)(34) ≡ 0.
ei, /∈ V y in all three cases
Case 4 (g = (123)). If g = (123), then Z((123)) = h(123)i × Sym{4,...,n} and
g = φ∗
φ∗
(132),(132) + Xr∈[n]\[3]
φ∗
(12r),(r23) + φ∗
(23r),(r31) + φ∗
(31r),(r12).
We first consider the term φ∗
Z(g)-orbit under diagonal conjugation. Note that im κL
(132)(v, w)) = 0 for all u, v, w ∈ V , and thus, φ∗
κ∗
(132)(u, κL
(132) ⊆ V (132) ⊆ ker κ∗
(132),(132) ≡ 0.
(132),(132) since the factorization (132)(132) is in its own
(132), so
22
B. FOSTER-GREENWOOD AND C. KRILOFF
Applying Lemma 5.4 with H = h(123)i, which is a normal subgroup of Z(g), yields
(31r),(r12) = 0, it is sufficient to
(12r),(r23) + φ∗
φ∗
(23r),(r31) + φ∗
prove
that in order to show Pr∈[n]\[3]
X(x,y)∈H ((124),(423))
φ∗
x,y(ei, ej, ek) = [φ∗
(124),(423) + φ∗
(234),(431) + φ∗
(314),(412)](ei, ej, ek) = 0
for H-representatives {ei, ej, ek} of the basis triples. By Lemma 7.1 (2) V x ∩ V y ⊆
ker φ∗
x,y, and hence we only consider {ei, ej , ek} ⊆ {e1, e2, e3, e4}. The three element
subsets of {e1, e2, e3, e4} form two H-orbits, with representatives {e1, e2, ek} for k = 3
and k = 4. Note that
[φ∗
(314),(412)](e1, e2, e3) = 0
(124),(423) + φ∗
(234),(431) + φ∗
by Lemma 7.1 (3) applied to all three terms, noting that xei, x−1
Also
ei, /∈ V y in all cases.
[φ∗
(124),(423) + φ∗
(234),(431) + φ∗
(314),(412)](e1, e2, e4) = 0
by Lemma 7.1 (3) with xei, x−1
applied to the third term. This verifies that φ∗
ei, /∈ V y applied to the first two terms and Lemma 7.1 (4)
(123) ≡ 0 and completes the proof.
(cid:3)
Proposition 7.3. Let κ3-cyc = κL
a, b, c ∈ C, and let φ∗
and g is a 5-cycle. Then φC
1 ≤ i ≤ n,
g ≡ 0. For φL
g denote the g-component of φ(κ∗
3-cyc + κC
3-cyc be as in Definition 4.6, with parameters
3-cyc), where ∗ = L or ∗ = C
g (ei, ej , ek) = 0, and for
g , if ei ∈ V g, then φL
3-cyc, κL
g (ei, gei, g2
φL
g (ei, gei, g3
φL
ei) = 2(a − b)2(ei + gei − g2
ei) = 2(a − b)2(−2ei + 2 g2
ei − g3
ei + g3
ei) and
ei − g4
ei).
Proof. It suffices to evaluate φ∗
g for the conjugacy class representative g = (12345), since
the results for any conjugate of g can be obtained by the orbit property described in
Lemma 5.1. Note that Z(g) = h(12345)i × Sym{6,...,n}, and as seen in the proof of
Proposition 7.2, the factorizations of g as a product of 3-cycles are all in the same
Z(g)-orbit under diagonal conjugation, so
(234),(451) + φ∗
(123),(345) + φ∗
(345),(512) + φ∗
(451),(123) + φ∗
(512),(234).
g = φ∗
φ∗
Note that for each pair of 3-cycles x and y with xy = g = (12345) we have V g ⊆
x,y by Lemma 7.1 (2), so if any of the vectors in a basis
g is zero on that triple. This leaves for further consideration only
V x ∩ V y, and V x ∩ V y ⊆ ker φ∗
triple lie in V g then φ∗
the cases where {ei, ej, ek} ⊆ {e1, e2, e3, e4, e5}.
First, consider φ∗
φ∗
(234),(451)(e1, e2, e3) are both zero. By Lemma 7.1 (4), the term φ∗
also zero. Applying Lemma 7.1 (3) to the remaining terms yields
g(e1, e2, e3). By Lemma 7.1 (1), the terms φ∗
(123),(345)(e1, e2, e3) and
(451),(123)(e1, e2, e3) is
φ∗
g(e1, e2, e3) = 2(a − b)(κ∗
(512)(e1, e2) − κ∗
(345)(e3, e4)).
ORBIFOLD ALGEBRAS FOR Sn
23
(ijk)(ei, ej) = (a − b)(ei + ej + ek) + b(e1 + · · · + en), and
Recall that κC
(ijk)(ei, ej) = c, κL
that eI =Pi∈I ei for I ⊆ {1, . . . , n}. Then
φ∗
g(e1, e2, e3) =(2(a − b)2(e{5,1,2} − e{3,4,5})
=(2(a − b)2(e1 + e2 − e3 − e4)
0
0
if ∗ = L,
if ∗ = C
if ∗ = L,
if ∗ = C.
Next, consider φ∗
g(e1, e2, e4). By Lemma 7.1 (1), the term φ∗
(123),(345)(e1, e2, e4) is zero.
Using the alternating property to apply Lemma 7.1 (3) to the remaining terms yields
φ∗
g(e1, e2, e4) = 2(a − b)(κ∗
(234)(e2, e3) + κ∗
(345)(e4, e5) − κ∗
(512)(e1, e2))
0
=(2(a − b)2(e{2,3,4} + e{3,4,5} − e{4,5,1} − e{5,1,2})
=(2(a − b)2(−2e1 + 2e3 + e4 − e5)
if ∗ = L,
if ∗ = C.
0
(451)(e4, e5) − κ∗
if ∗ = L,
if ∗ = C
Finally, note that for any ei 6∈ V g, the values of φ∗
are obtained from the cases φ∗
power of g and using the orbit property in Lemma 5.1.
g(e1, e2, e3) and φ∗
g(ei, gei, g2
ei)
g(e1, e2, e4) by acting by an appropriate
(cid:3)
ei) and φ∗
g(ei, gei, g3
The next definition of a map κC
5-cyc supported only on 5-cycles is motivated by the
3-cyc) = 2ψ(κC ). When G is S3 or S4 there are no 5-cycles and
requirement φ(κL
κC
5-cyc is the zero map.
3-cyc, κL
Definition 7.4. For parameters a, b ∈ C, define an Sn-invariant map κC
with component maps κC
define κC
g :V2 V → C. If g is not a 5-cycle, let κC
g by the skew-symmetric matrix
κC
g g
g ≡ 0. If g is a 5-cycle,
5-cyc =Pg∈Sn
[κC
g ] = (a − b)2([g] − [g]T − 2[g2] + 2[g2]T ),
where [g] denotes the matrix of g with respect to the basis e1, . . . , en, and the (i, j)-entry
of [κC
g ] records κC
g (ei, ej).
In practice we use the consequences that V g ⊆ ker κC
g , and if ei 6∈ V g, then
κC
g (ei, gei) = −(a − b)2
and κC
g (ei, g2
ei) = 2(a − b)2.
Also, κC
1 ≤ i, j ≤ n.
5-cyc is G-invariant, i.e., κC
hgh−1(hei, hej) = κC
g (ei, ej) for all h, g ∈ Sn and all
Proposition 7.5. Let κL
parameters a, b ∈ C. Then φ(κL
3-cyc and κC
3-cyc, κL
3-cyc) = 2ψ(κC
5-cyc).
5-cyc be as in Definitions 4.6 and 7.4, with common
24
B. FOSTER-GREENWOOD AND C. KRILOFF
5-cyc). If g is not a 5-cycle, then φg ≡ 0 by Proposition 7.2; and κC
Proof. We compare the component φg of φ(κL
3-cyc) with the component 2ψg of
2ψ(κC
5-cyc is not sup-
ported on g, so 2ψg ≡ 0 also. If g is a 5-cycle, then note that the results of Proposition 7.3
can be written in the form
3-cyc, κL
φg(ei, gei, g2
φg(ei, gei, g3
ei) = −2(a − b)2((gei − ei) + 2(g2
ei) = 2(a − b)2(2(gei − ei) + 2(g2
ei − gei) + (g3
ei − gei) − (g4
ei − g2
ei − g3
ei)),
ei)),
while
ψg(ei, gei, g2
ψg(ei, gei, g3
ei) = κC
ei) = κC
g (g2
g (g3
Finally, if ei ∈ V g, then φg(ei, ej, ek) = 0 and
ei)(gei − ei) + κC
ei)(gei − ei) + κC
g (gei, g2
g (gei, g3
ei, ei)(g2
ei, ei)(g2
ei − gei) + κC
ei − gei) + κC
g (ei, gei)(g3
g (ei, gei)(g4
ei − g2
ei − g3
ei),
ei).
ψg(ei, ej , ek) = κC
g (ej, ek)(gei − ei) + κC
g (ek, ei)(gej − ej) + κC
g (ei, ej )(gek − ek) = 0,
where the first term vanishes because gei − ei = 0 and the second two terms vanish
because V g ⊆ ker κC
g . Since ψ is alternating, we see that ψg(ei, ej, ek) = 0 whenever
{ei, ej , ek} ∩ V g 6= ∅.
(cid:3)
Clearing the Second Obstruction. The final step in showing κ = κL
Drinfeld orbifold algebra map is to verify φ(κC
5-cyc, κL
3-cyc) = 0.
3-cyc + κC
5-cyc is a
We begin with a lemma that describes simplifications of the summands φx,y of the
components φg of φ(κC
3-cyc). As in the analogous Lemma 7.1, simplification of
φx,y(ei, ej, ek) depends on where the vectors in the basis triple lie relative to the fixed
spaces V x and V y. Recall that for σ ∈ Sn and v ∈ V , δσ(v) = 1 if v ∈ V σ and δσ(v) = 0
otherwise.
5-cyc, κL
Lemma 7.6. Let κC
rameters a, b ∈ C. Let φx,y denote a summand of the component φg of φ(κC
Let x be a 5-cycle and y be a 3-cycle. Let ei, ej, ek be basis vectors.
3-cyc be as in Definitions 7.4 and 4.6, with common pa-
3-cyc).
5-cyc and κL
5-cyc, κL
(1) If ei, ej ∈ V y, then φx,y(ei, ej , ek) = 0.
(2) If ei ∈ V y ∩ V x, then φx,y(ei, ej , ek) = 0.
(3) If ei ∈ V y\V x and ej 6∈ V y, then
φx,y(ei, ej , yej) = 2(a − b)3hδy(xei) − 2δy(x2
ei) + 2δy(x−2
ei) − δy(x−1
ei)i .
(4) If ei 6∈ V y, then φx,y(ei, yei, y2
ei) = 0.
Note that φx,y can be evaluated on any basis triple by using the alternating property
along with these cases.
Proof. The proofs of parts (1), (2), and (4) are the same as in the proof of Lemma 7.1
since V y ⊆ ker κL
5-cyc. The proof of part (3) is the
x is also true for κC
y , and V x ⊆ ker κC
same up until the last step, so if ei ∈ V y\V x and ej 6∈ V y, then
ORBIFOLD ALGEBRAS FOR Sn
φx,y(ei, ej , yej) = 2(b − a) Xh∈hxi
δy(hei)κC
x [ei, hei]
= 2(a − b)3hδy(xei) − 2δy(x2
ei) + 2δy(x−2
ei) − δy(x−1
ei)i .
25
(cid:3)
The proof of the next proposition uses these simplifications to verify that indeed
5-cyc, κL
3-cyc) = 0, as mentioned in the outline of the proof of Theorem 3.2. This
φ(κC
clears the second obstruction.
Proposition 7.7. Let κC
parameters a, b ∈ C. For every g ∈ Sn, the component φg of φ(κC
zero.
3-cyc be as in Definitions 7.4 and 4.6, with common
3-cyc) is identically
5-cyc and κL
5-cyc, κL
5-cyc is supported only on 5-cycles and κL
Proof. Since κC
3-cyc is supported only on 3-cycles,
we first determine the cycle types that arise as a product xy with x a 5-cycle and y a
3-cycle. Since σxσy = σ(xy), it suffices to examine representatives of orbits of factor
pairs (x, y) under the action of Sn by diagonal conjugation. Orbit representatives and
their products are
(12345)(678),
(12345)(567) = (1234567),
(12345)(456) = (1234)(56),
(12345)(546) = (12346),
(12345)(356) = (123)(456),
(12345)(536) = (1236)(45),
(12345)(345) = (12354),
(12345)(543) = (123),
(12345)(245) = (12534),
(12345)(542) = (12)(34).
If the cycle type of g does not appear in this list, then certainly φg ≡ 0. We show further
that φg ≡ 0 for g = (12345)(678), g = (1234567), g = (1234)(56), g = (123)(456),
g = (12345), g = (12)(34), and g = (123), and hence also for their conjugates.
Besides narrowing the set of representative elements g to consider, the list of orbit
representatives reveals a way to organize the terms φx,y of φg. Specifically, if the cycle
type of g occurs with multiplicity m in the list, then the factor pairs with product g are in
m orbits under the diagonal conjugation action of Z(g), and we can use a representative
from each orbit to generate all the terms φx,y needed to evaluate φg.
Case 1 (g = (12345)(678)). Note that g has a unique factorization as a product of
a 5-cycle and a 3-cycle. To show φg = φ(12345),(678) ≡ 0, we apply Lemma 5.4 with
H = Z(g) = h(12345), (678)i × Sym{9,...,n}. In particular, it suffices to evaluate φg on
Z(g)-orbit representatives of the basis triples. Since V x ∩V y ⊆ ker φx,y, we only consider
{ei, ej , ek} ⊆ {el 1 ≤ l ≤ 8}. If {ei, ej, ek} does not contain at least two elements from
{e6, e7, e8}, then φx,y(ei, ej, ek) = 0 by Lemma 7.6 (1). The remaining triples partition
into Z(g)-orbits
Z(g){e1, e6, e7} and Z(g){e6, e7, e8}.
Note that φx,y(e1, e6, e7) = 0 by Lemma 7.6 (3) and φx,y(e6, e7, e8) = 0 by Lemma 7.6 (4).
26
B. FOSTER-GREENWOOD AND C. KRILOFF
Case 2 (g = (1234567)). Note that Z(g) = h(1234567)i × Sym{8,...,n}. The factoriza-
tions of g as a product of a 5-cycle and a 3-cycle are in a single Z(g)-orbit. To show
φg ≡ 0, it suffices to verify
X
(x,y)∈Z(g)((12345),(567))
φx,y(ei, ej, ek) = 0
for Z(g)-orbit representatives {ei, ej , ek} of the basis triples. Since V x ∩ V y ⊆ ker φx,y,
we only consider {ei, ej, ek} ⊆ {el 1 ≤ l ≤ 7}. The remaining triples partition into
Z(g)-orbits
Z(g){e1, e2, e3},
Z(g){e1, e2, e6},
Z(g){e1, e2, e4},
Z(g){e1, e2, e5},
and Z(g){e1, e3, e5}.
(The sum of orbit sizes is 7 + 7 + 7 + 7 + 7 =(cid:0)7
φx,y(ei, ej, ek) of the sum and record the results in a table with rows indexed by the
elements of Z(g)((12345), (567)). Notice φ(45671),(123)(e1, e2, e3) = 0 by Lemma 7.6 (4),
all other zero entries result from Lemma 7.6 (1), and all remaining entries are found
using Lemma 7.6 (3). Each column sum is zero.
3(cid:1).) We simplify each remaining term
x, y
φx,y(e1, e2, e3) φx,y(e1, e2, e4) φx,y(e1, e2, e5) φx,y(e1, e2, e6) φx,y(e1, e3, e5)
(12345), (567)
(23456), (671)
0
0
0
0
0
0
0
−2(a − b)3
(34567), (712)
2(a − b)3
−4(a − b)3
4(a − b)3
−2(a − b)3
0
0
0
(45671), (123)
0
(56712), (234)
−2(a − b)3
2(a − b)3
2(a − b)3
(67123), (345)
(71234), (456)
0
0
0
0
−4(a − b)3
4(a − b)3
4(a − b)3
0
0
0
0
0
0
0
−4(a − b)3
0
Case 3 (g = (1234)(56)). Note that Z(g) = h(1234), (56)i × Sym{7,...,n}. The factor-
izations of g as a product of a 5-cycle and a 3-cycle are in two Z(g)-orbits
Z(g)((12345), (564))
and Z(g)((12356), (634)).
To show φg ≡ 0, it suffices to verify
X
φx,y(ei, ej, ek) +
X
(x,y)∈Z(g)((12356),(634))
φx,y(ei, ej, ek) = 0
(x,y)∈Z(g)((12345),(564))
for Z(g)-orbit representatives {ei, ej , ek} of the basis triples. Since V x ∩ V y ⊆ ker φx,y,
we only consider {ei, ej, ek} ⊆ {el 1 ≤ l ≤ 6}. The remaining triples partition into
Z(g)-orbits
Z(g){e1, e2, e3},
Z(g){e1, e2, e5},
Z(g){e1, e3, e5},
and Z(g){e1, e5, e6}.
(The sum of orbit sizes is 4 + 8 + 4 + 4 = (cid:0)6
3(cid:1).) We simplify each term φx,y(ei, ej, ek)
of the sum and record the results in a table with rows indexed by the elements of
Z(g)((12345), (564)) ∪ Z(g)((12356), (634)). Each zero entry is color-coded and tagged by
ORBIFOLD ALGEBRAS FOR Sn
27
the applicable part of Lemma 7.6, all nonzero entries are found using Lemma 7.6 (3),
and each column sum is zero.
x, y
φx,y(e1, e2, e3) φx,y(e1, e2, e5) φx,y(e1, e3, e5) φx,y(e1, e5, e6)
(12345), (456)
(23415), (156)
(34125), (256)
(41235), (356)
(12346), (465)
(23416), (165)
(34126), (265)
(41236), (365)
(12356), (346)
(23456), (416)
0
0
0
0
0
0
0
0
0
0
(34156), (126)
6(a − b)3
(41256), (236)
−6(a − b)3
(12365), (345)
(23465), (415)
0
0
0
2(a − b)3
2(a − b)3
0
0
−2(a − b)3
−2(a − b)3
0
0
0
0
0
0
−6(a − b)3
(34165), (125)
6(a − b)3
0
0
0
0
0
0
0
0
0
0
0
0
0
−6(a − b)3
6(a − b)3
6(a − b)3
(41265), (235)
−6(a − b)3
6(a − b)3
−6(a − b)3
−2(a − b)3
0
2(a − b)3
0
2(a − b)3
0
−2(a − b)3
0
0
0
0
0
0
0
0
0
Case 4 (g = (123)(456)). Note that Z(g) = h(123), (456), (14)(25)(36)i × Sym{7,...,n},
and the subgroup
H = h(123), (456)i
is normal in Z(g). The factorizations of g as a product of a 5-cycle and a 3-cycle are in
a single Z(g)-orbit (of size 18), which partitions into two H-orbits (of size 9)
Z(g)((12345), (563)) = H((12345), (563)) ∪ H ((45612), (236)).
To show φg ≡ 0, it suffices to verify
X
(x,y)∈H ((12345),(563))
φx,y(ei, ej, ek) = 0
for H-orbit representatives {ei, ej , ek} of the basis triples. Since V x ∩ V y ⊆ ker φx,y,
we only consider {ei, ej, ek} ⊆ {el 1 ≤ l ≤ 6}. The remaining triples partition into
H-orbits
H{e1, e2, e4}, H{e1, e4, e5}, H {e1, e2, e3},
and H {e4, e5, e6}.
28
B. FOSTER-GREENWOOD AND C. KRILOFF
(The sum of orbit sizes is 9 + 9 + 1 + 1 = (cid:0)6
3(cid:1).) We simplify each term φx,y(ei, ej, ek)
of the sum and record the results in a table with rows indexed by the elements of
H ((12345), (563)). Each zero entry is color-coded and tagged by the applicable part of
Lemma 7.6, all nonzero entries are found using Lemma 7.6 (3), and each column sum is
zero.
x, y
φx,y(e1, e2, e3) φx,y(e4, e5, e6) φx,y(e1, e2, e4) φx,y(e1, e4, e5)
(12345), (563)
(12356), (643)
(12364), (453)
(23145), (561)
(23156), (641)
(23164), (451)
(31245), (562)
(31256), (642)
(31264), (452)
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
6(a − b)3
−6(a − b)3
0
6(a − b)3
0
0
6(a − b)3
0
0
0
0
0
−6(a − b)3
−6(a − b)3
Case 5 (g = (12345)). Note that Z(g) = H × Sym{6,...,n}, where H is the normal
subgroup
H = h(12345)i.
The factorizations of g as a product of a 5-cycle and a 3-cycle are in three Z(g)-orbits
Z(g)((12354), (354)),
Z(g)((12534), (254)),
and Z(g)((12346), (645)),
which partition into H-orbits
Z(g)((12354), (354)) = H((12354), (354)),
Z(g)((12534), (254)) = H((12534), (254)),
and
Z(g)((12346), (645)) = [r≥6
H((1234r), (r45)).
To show φg ≡ 0, it suffices to verify
X
X
X
φx,y(ei, ej , ek)
φx,y(ei, ej , ek)
φx,y(ei, ej , ek) = 0
(x,y)∈H ((12354),(354))
+
+
(x,y)∈H ((12534),(254))
(x,y)∈H ((12346),(645))
for H-orbit representatives {ei, ej , ek} of the basis triples. Since V x ∩ V y ⊆ ker φx,y,
we only consider {ei, ej, ek} ⊆ {el 1 ≤ l ≤ 6}. The remaining triples partition into
H-orbits
ORBIFOLD ALGEBRAS FOR Sn
29
H{e1, e2, e3}, H{e1, e2, e4}, H {e1, e2, e6},
and H {e1, e3, e6}.
(The sum of orbit sizes is 5 + 5 + 5 + 5 = (cid:0)6
3(cid:1).) We simplify each term φx,y(ei, ej, ek)
of the sum and record the results in a table with rows indexed by the elements of
Z(g)((12354), (354))∪ Z(g) ((12534), (254))∪ Z(g) ((12346), (645)). Each zero entry is color-
coded and tagged by the applicable part of Lemma 7.6, all nonzero entries are found
using Lemma 7.6 (3), and each column sum is zero.
x, y
φx,y(e1, e2, e3) φx,y(e1, e2, e4) φx,y(e1, e2, e6) φx,y(e1, e3, e6)
(12354), (354)
(23415), (415)
0
0
(34521), (521)
−2(a − b)3
(45132), (132)
0
0
2(a − b)3
2(a − b)3
−2(a − b)3
(51243), (243)
2(a − b)3
−2(a − b)3
(12534), (254)
0
−4(a − b)3
(23145), (315)
4(a − b)3
(34251), (421)
−4(a − b)3
(45312), (532)
4(a − b)3
0
0
0
(51423), (143)
−4(a − b)3
4(a − b)3
(12346), (645)
(23456), (651)
0
0
(34516), (612)
−2(a − b)3
(45126), (623)
2(a − b)3
(51236), (634)
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
2(a − b)3
0
0
0
0
0
0
0
0
0
0
0
0
0
−2(a − b)3
−2(a − b)3
2(a − b)3
0
0
Case 6 (g = (12)(34)). Note that Z(g) = H × Sym{5,...,n}, where H is the dihedral
group
H = h(1324), (12)i = {1, (1324), (12)(34), (4231), (12), (13)(24), (34), (14)(23)}.
The factorizations of g as a product of a 5-cycle and a 3-cycle are in a single Z(g)-orbit,
which we partition into H-orbits
To show φg ≡ 0, it suffices to verify
Z(g)((12345), (542)) = [r≥5
H((1234r), (r42)).
X
(x,y)∈H ((12345),(542))
φx,y(ei, ej, ek) = 0
30
B. FOSTER-GREENWOOD AND C. KRILOFF
for H-orbit representatives {ei, ej , ek} of the basis triples. Since V x ∩ V y ⊆ ker φx,y,
we only consider {ei, ej, ek} ⊆ {el 1 ≤ l ≤ 5}. The remaining triples partition into
H-orbits
H{e1, e2, e3}, H{e1, e2, e5},
and H {e1, e3, e5}.
(The sum of the orbit sizes is 4 + 2 + 4 =(cid:0)5
3(cid:1).) We simplify each term φx,y(ei, ej, ek)
of the sum and record the results in a table with rows indexed by the elements of
H ((12345), (542)). Notice that φ(21435),(531)(e1, e3, e5) = φ(43215),(513)(e1, e3, e5) = 0 fol-
lows from Lemma 7.6 (4), all other zero entries result from Lemma 7.6 (1), and all
nonzero entries are found using Lemma 7.6 (3). Each column sum is zero.
x, y
φx,y(e1, e2, e3) φx,y(e1, e2, e5) φx,y(e1, e3, e5)
(12345), (542)
0
−4(a − b)3
0
(12435), (532)
4(a − b)3
−4(a − b)3
4(a − b)3
(21345), (541)
0
(21435), (531)
−4(a − b)3
(34125), (524)
(34215), (514)
0
0
4(a − b)3
4(a − b)3
−4(a − b)3
−4(a − b)3
0
0
4(a − b)3
−4(a − b)3
(43125), (523)
4(a − b)3
−4(a − b)3
4(a − b)3
(43215), (513)
−4(a − b)3
4(a − b)3
0
Case 7 (g = (123)). Note that Z(g) = h(123)i × Sym{4,...,n}. The factorizations of g
as a product of a 5-cycle and a 3-cycle are in a single Z(g)-orbit
Z(g)((12345), (543)) = {((123rs), (sr3)) {r, s} ⊆ {4, . . . , n}}.
Consider the subgroup H = h(123), (45)i of Z(g). To show φg ≡ 0, it suffices to verify
X
(x,y)∈H ((12345),(543))
φx,y(ei, ej, ek) = 0
for H-orbit representatives {ei, ej , ek} of the basis triples. Since V x ∩ V y ⊆ ker φx,y,
we only consider {ei, ej, ek} ⊆ {el 1 ≤ l ≤ 5}. The remaining triples partition into
H-orbits
H{e1, e2, e3}, H{e1, e2, e4},
and H {e1, e4, e5}.
(The sum of the orbit sizes is 1 + 6 + 3 =(cid:0)5
3(cid:1).) We simplify each term φx,y(ei, ej, ek)
of the sum and record the results in a table with rows indexed by the elements of
H ((12345), (543)). Notice that φ(23154),(451)(e1, e4, e5) = φ(23145),(541)(e1, e4, e5) = 0 fol-
lows from Lemma 7.6 (4), all other zero entries result from Lemma 7.6 (1), and all
ORBIFOLD ALGEBRAS FOR Sn
31
nonzero entries are found using Lemma 7.6 (3). Each column sum is zero.
x, y
φx,y(e1, e2, e3) φx,y(e1, e2, e4) φx,y(e1, e4, e5)
(12354), (453)
(12345), (543)
(23154), (451)
(23145), (541)
(31254), (452)
(31245), (542)
0
0
0
0
0
0
0
0
2(a − b)3
−2(a − b)3
−2(a − b)3
2(a − b)3
0
0
−2(a − b)3
−2(a − b)3
2(a − b)3
2(a − b)3
References
(cid:3)
[Bel16] Gwyn Bellamy. Symplectic reflection algebras. In Noncommutative Algebraic Geometry, Math-
ematical Sciences Research Institute Publications, pages 167 -- 224. Cambridge University Press,
2016.
[BG96] Alexander Braverman and Dennis Gaitsgory. Poincar´e-Birkhoff-Witt theorem for quadratic
algebras of Koszul type. J. Algebra, 181(2):315 -- 328, 1996.
[Dri86] V. G. Drinfel′d. Degenerate affine Hecke algebras and Yangians. Funktsional. Anal. i Prilozhen.,
20(1):69 -- 70, 1986.
[EG02] Pavel Etingof and Victor Ginzburg. Symplectic reflection algebras, Calogero-Moser space, and
deformed Harish-Chandra homomorphism. Invent. Math., 147(2):243 -- 348, 2002.
[Eti14] Pavel Etingof. Exploring noncommutative algebras via deformation theory. In Trends in Con-
temporary Mathematics, INdAM Series, Volume 8, pages 59 -- 72. Springer, 2014.
[Far05] Marco Farinati. Hochschild duality, localization, and smash products. J. Algebra, 284(1):415 --
434, 2005.
[GK04] Victor Ginzburg and Dmitry Kaledin. Poisson deformations of symplectic quotient singulari-
[Gor08]
[Gor10]
ties. Adv. Math., 186(1):1 -- 57, 2004.
Iain G. Gordon. Symplectic reflection algebras. In Trends in representation theory of algebras
and related topics, EMS Ser. Congr. Rep., pages 285 -- 347. Eur. Math. Soc., Zurich, 2008.
Iain G. Gordon. Rational Cherednik algebras. In Proceedings of the International Congress of
Mathematicians. Volume III, pages 1209 -- 1225. Hindustan Book Agency, New Delhi, 2010.
[GS88] Murray Gerstenhaber and Samuel D. Schack. Algebraic cohomology and deformation theory.
In Deformation theory of algebras and structures and applications (Il Ciocco, 1986), volume
247 of NATO Adv. Sci. Inst. Ser. C Math. Phys. Sci., pages 11 -- 264. Kluwer Acad. Publ.,
Dordrecht, 1988.
[Kan01] Richard Kane. Reflection groups and invariant theory. CMS Books in Mathematics/Ouvrages
de Math´ematiques de la SMC, 5. Springer-Verlag, New York, 2001.
[Lus88] George Lusztig. Cuspidal local systems and graded Hecke algebras. I. Inst. Hautes ´Etudes Sci.
Publ. Math., (67):145 -- 202, 1988.
[Lus89] George Lusztig. Affine Hecke algebras and their graded version. J. Amer. Math. Soc., 2(3):599 --
635, 1989.
[NW16] Deepak Naidu and Sarah Witherspoon. Hochschild cohomology and quantum Drinfeld Hecke
algebras. Selecta Math. (N.S.), 22(3):1537 -- 1561, 2016.
[PP05] Alexander Polishchuk and Leonid Positselski. Quadratic algebras, volume 37 of University
Lecture Series. American Mathematical Society, Providence, RI, 2005.
[RS03] Arun Ram and Anne V. Shepler. Classification of graded Hecke algebras for complex reflection
groups. Comment. Math. Helv., 78(2):308 -- 334, 2003.
32
[Sha12]
B. FOSTER-GREENWOOD AND C. KRILOFF
Jeanette Shakalli. Deformations of quantum symmetric algebras extended by groups. J. Alge-
bra, 370:79 -- 99, 2012.
Louis Solomon. Invariants of finite reflection groups. Nagoya Math. J., 22:57 -- 64, 1963.
[Sol63]
[S¸te95] Drago¸s S¸tefan. Hochschild cohomology on Hopf Galois extensions. J. Pure Appl. Algebra,
103(2):221 -- 233, 1995.
[SW08] Anne V. Shepler and Sarah Witherspoon. Hochschild cohomology and graded Hecke algebras.
Trans. Amer. Math. Soc., 360(8):3975 -- 4005, 2008.
[SW12a] Anne V. Shepler and Sarah Witherspoon. Drinfeld orbifold algebras. Pacific J. Math.,
259(1):161 -- 193, 2012.
[SW12b] Anne V. Shepler and Sarah Witherspoon. Group actions on algebras and the graded Lie struc-
ture of Hochschild cohomology. J. Algebra, 351:350 -- 381, 2012.
[SW15] Anne V. Shepler and Sarah Witherspoon. Poincar´e-Birkhoff-Witt theorems. In Commutative
Algebra and Noncommutative Algebraic Geometry, Mathematical Sciences Research Institute
Publications, pages 259 -- 290. Cambridge University Press, 2015.
[Wit07] Sarah Witherspoon. Twisted graded Hecke algebras. J. Algebra, 317(1):30 -- 42, 2007.
Department of Mathematics and Statistics, California State Polytechnic University,
Pomona, California 91768, USA
E-mail address: [email protected]
Department of Mathematics and Statistics, Idaho State University, Pocatello, Idaho
83209, USA
E-mail address: [email protected]
|
1902.06033 | 1 | 1902 | 2019-02-16T03:38:13 | Liftable derived equivalences and objective categories | [
"math.RA",
"math.AG",
"math.RT"
] | We give two proofs to the following theorem and its generalization: if a finite dimensional algebra $A$ is derived equivalent to a smooth projective scheme, then any derived equivalence between $A$ and another algebra $B$ is standard, that is, isomorphic to the derived tensor functor by a two-sided tilting complex. The main ingredients of the proofs are as follows: (1) between the derived categories of two module categories, liftable functors coincide with standard functors; (2) any derived equivalence between a module category and an abelian category is uniquely factorized as the composition of a pseudo-identity and a liftable derived equivalence; (3) the derived category of coherent sheaves on a certain projective scheme is triangle-objective, that is, any triangle autoequivalence on it, which preserves the the isomorphism classes of complexes, is necessarily isomorphic to the identity functor. | math.RA | math |
LIFTABLE DERIVED EQUIVALENCES AND OBJECTIVE
CATEGORIES
XIAOFA CHEN, XIAO-WU CHEN∗
Abstract. We give two proofs to the following theorem and its generalization:
if a finite dimensional algebra A is derived equivalent to a smooth projective
scheme, then any derived equivalence between A and another algebra B is
standard, that is, isomorphic to the derived tensor functor by a two-sided tilt-
ing complex. The main ingredients of the proofs are as follows: (1) between
the derived categories of two module categories, liftable functors coincide with
standard functors; (2) any derived equivalence between a module category
and an abelian category is uniquely factorized as the composition of a pseudo-
identity and a liftable derived equivalence; (3) the derived category of coherent
sheaves on a certain projective scheme is triangle-objective, that is, any tri-
angle autoequivalence on it, which preserves the the isomorphism classes of
complexes, is necessarily isomorphic to the identity functor.
1. Introduction
Let k be a field. For a finite dimensional k-algebra A, we denote by A-mod the
abelian category of finitely generated A-modules and by Db(A-mod) its bounded
derived category. By a derived equivalence between two algebras A and B, we mean
a k-linear triangle equivalence F : Db(A-mod) → Db(B-mod). It is a well-known
open question [13] whether any derived equivalence is standard, that is, isomorphic
to the derived tensor functor by a two-sided tilting complex. We refer to the
introduction of [6] for known cases where the question is answered affirmatively.
The geometric analogue of standard functors are Fourier-Mukai functors, where
two-sided tilting complexes are replaced by Fourier-Mukai kernels. The famous
theorem in [12] states that any derived equivalence between smooth projective
schemes is a Fourier-Mukai functor.
We are inspired by the following theorem, which seems to be folklore. It provides
a large class of algebras, for which the above open question is answered affirmatively.
Theorem. Let A be a finite dimensional algebra. Assume that there is a derived
equivalence between A and a smooth projective scheme. Then any derived equiva-
lence F : Db(A-mod) → Db(B-mod) is standard.
The goal is to give a proof to this theorem and its generalization. Indeed, we
give two proofs. The first proof uses the homotopy category of small dg categories
and dg lifts of triangle functors, while the second one is more elementary and uses
the notion of triangle-objective categories.
Let us describe the content of this paper.
In Section 2, we recall basic facts
about dg categories and dg enhancements. In Section 3, we recall the homotopy
category of small dg categories and the notion of liftable functors. In Section 4,
we prove that between the bounded derived categories of two module categories,
Date: February 19, 2019.
2010 Mathematics Subject Classification. 18E30, 16G10, 16D90.
Key words and phrases. derived equivalence, dg category, enhancement, liftable, objective.
∗ The corresponding author.
E-mail: [email protected], [email protected].
1
2
XIAOFA CHEN, XIAO-WU CHEN
liftable functors coincide with standard functors; see Theorem 4.2. We mention
that this result also seems to be folklore.
In Section 5, we prove the following factorization theorem: any derived equiv-
alence between a module category and an abelian category is uniquely factorized
as the composition of a pseudo-identity in the sense of [5] and a liftable derived
equivalence; see Theorem 5.3. Then we give the first proof to the above theorem.
In Section 6, we introduce the following notion of triangle-objective categories:
a triangulated category is triangle-objective, if any triangle autoequivalence on it,
which preserves the isomorphism classes of objects, is isomorphic to the identity
functor. We prove that the bounded derived category of coherent sheaves on a
certain projective scheme is triangle-objective; see Proposition 6.6. It implies the
above theorem, when the field k is algebraically closed.
Throughout, we work over a fixed field k. All algebras, categories and functors
are required to be k-linear. The word dg stands for "differential graded". In the
dg setting, all morphisms and elements are by default homogeneous. Modules are
by default left modules.
2. DG categories and enhancements
In this section, we recall basic facts and notation for dg categories and enhance-
ments. The standard references for dg categories are [8, 7].
Let C be a dg category. For two objects X and Y , the Hom complex is denoted
by C(X, Y ) = (Lp∈Z C(X, Y )p, d = dX,Y ), where d is the differential of degree one
satisfying the graded Leibniz rule. An element in the subspace C(X, Y )p will be
called a homogeneous morphism of degree p with the notation f = p.
We denote by H 0(C) the homotopy category of C, which has the same objects
as C and whose Hom spaces are given by the zeroth cohomologies H 0(C(X, Y )).
Similarly, one has the category Z 0(C), whose Hom spaces are given by the zeroth
cocycles Z 0(C(X, Y )).
The opposite dg category Cop has the same objects and Hom complexes as C,
whose composition f ′ ◦op f of morphisms f ′ and f is given by (−1)f ·f ′f ◦ f ′. For
two dg categories C and D, we have their tensor dg category C ⊗ D, whose objects
are the pairs (C, D) with C ∈ C and D ∈ D, and whose Hom complexes are the
tensor product of the corresponding Hom complexes in C and D.
In the following examples, we fix the notation for our concerned dg categories.
Let A be a finite dimensional algebra. Denote by A-Mod the category of left A-
modules. In particular, k-Mod denotes the category of k-vector spaces.
Example 2.1. Let A be an additive category. A complex in A is denoted by X =
(Lp∈Z X p, dX ), where the differentials dp
X = 0. We
denote by Cdg(A) the dg category formed by complexes in A. The p-th component
of the Hom complex Cdg(A)(X, Y ) is given by
X : X p → X p+1 satisfy dp+1
X ◦ dp
Cdg(A)(X, Y )p = Y
n∈Z
HomA(X n, Y n+p),
◦ f n − (−1)pf n+1 ◦ dn
Y
whose elements will be denoted by f = {f n}n∈Z. The differential d acts on f such
that d(f )n = dn+p
X for each n ∈ Z. We are also interested in
the full dg subcategory C b
dg(A) formed by bounded complexes.
We observe that its homotopy category H 0(Cdg(A)) coincides with the classical
dg(A)) corresponds to the
homotopy category K(A) of complexes in A, where H 0(C b
bounded homotopy category Kb(A).
For two complexes X and Y of A-modules, the traditional notation of the Hom
complex Cdg(A-Mod)(X, Y ) is HomA(X, Y ).
LIFTABLE DERIVED EQUIVALENCES AND OBJECTIVE CATEGORIES
3
Example 2.2. The dg category Cdg(k-Mod) is usually denoted by Cdg(k). Let C
be a dg category. By a left dg C-module, we mean a dg functor M : C → Cdg(k).
The following notation will be convenient:
for a morphism f : X → Y in C and
m ∈ M (X), the resulting element M (f )(m) ∈ M (Y ) is written as f.m. Here the
dot indicates the left C-action on M . We denote by C-DGMod the dg category
formed by left dg C-modules, whose Hom complexes are defined similarly as in
Example 2.1.
Denote by C-DGProj the full dg subcategory of C-DGMod formed by dg-projective
C-modules. Here, we recall that a dg C-module is dg-projective if and only if it is
isomorphic to a direct summand of a semi-free dg C-module in Z 0(C-DGMod);
compare [8, 3.1] and [7, Appendix B.1].
We identify a left Cop-modules with a right dg C-module. Then we obtain the dg
category DGMod-C of right dg C-modules. For a right dg C-module N , a morphism
f : X → Y in C and m ∈ N (Y ), the right C-action on N is given such that m.f =
(−1)f ·mN (f )(m) ∈ N (X).
We identify a dg C-D-bimodule M : Dop ⊗ C → Cdg(k) with a left dg C ⊗ Dop-
module. We observe that for each object C ∈ C, M (−, C) is a right dg D-module.
Given a dg functor F : C → D, we have a dg C-D-bimodule MF defined such that
MF (D, C) = D(D, F (C)).
Example 2.3. Let C be a dg category. Denote by B the bar resolution of the
C-C-bimodule C; see [8, 6.6]. Then we have the following dg functor
pC = B ⊗C − : C-DGMod −→ C-DGProj.
For each left dg C-module M , pC(M ) is a semi-free C-modules, and there is a
canonical surjective quasi-isomorphism pC(M ) → M . We call pC the dg-projective
resolution functor of C.
Let C be a dg category. Recall that both H 0(C-DGMod) and H 0(C-DGProj) have
natural triangulated structures. The derived category D(C) is the Verdier quotient
of H 0(C-DGMod) by the triangulated subcategory of acyclic modules. It is well
known that the canonical functor H 0(C-DGProj) → D(C) is a triangle equivalence;
see [8, Theorem 3.1].
The Yoneda functor
YC : C −→ DGMod-C, X 7→ C(−, X)
is a fully-faithful dg functor. In particular, it induces a full embedding
H 0(YC) : H 0(C) −→ H 0(DGMod-C).
Recall that H 0(DGMod-C) has a natural triangulated structure. The dg category
C is said to be pretriangulated, provided that the essential image of H 0(YC) is a
triangulated subcategory. The terminology is justified by the evident fact: the
homotopy category H 0(C) of a pretriangulated dg category C has a canonical tri-
angulated structure.
Let T be a triangulated category. By an enhancement of T , we mean a triangle
equivalence E : T → H 0(C) with C a pretriangulated dg category. In general, the
enhancement is not necessarily unique. We refer to [11, 3] for more details.
Let A be an abelian category. We are mainly concerned with the bounded derived
category Db(A). As we have seen in Example 2.1, Cb
dg(A) provides a canonical en-
hancement for Kb(A). Following [9, 9.8], we now recall the canonical enhancement
of Db(A).
Example 2.4. Consider the dg category C b
full dg subcategory C b,ac
dg(A) of bounded complexes, and its
dg (A) formed by acyclic complexes. The bounded dg derived
4
XIAOFA CHEN, XIAO-WU CHEN
category of A is defined to be the Drinfeld dg quotient
Db
dg(A) = C b
dg(A)/C b,ac
dg (A).
Recall that the dg category Db
dg(A) by freely adding
new morphisms εX : X → X of degree −1 for each acyclic complex X, such that
d(εX ) = 1X ; see [7, 3.1]. By [11, Lemma 1.5], Db
dg(A) is pretriangulated. By [7,
Theorem 3.4], there is a canonical isomorphism of triangulated categories
dg(A) is obtained from C b
canA : Db(A) −→ H 0(Db
dg(A)),
which acts on objects by the identity. We will call canA the canonical enhancement
of Db(A).
3. The homotopy category and liftable functors
In this section, we recall the notion of liftable triangle functors between bounded
derived categories, and the homotopy category of small dg categories.
Recall that a dg functor F : C → D is a quasi-equivalence, provided that the
induced chain maps C(C, C′) → D(F (C), F (C′)) are all quasi-isomorphisms, and
that H 0(F ) : H 0(C) → H 0(D) is dense. In this situation, H 0(F ) is an equivalence.
Lemma 3.1. Let F : C → D be a dg functor between two pretriangulated dg cate-
gories. Assume that H 0(F ) is an equivalence. Then F is a quasi-equivalence.
Proof. It suffices to show that the induced chain map C(C, C′) → D(F (C), F (C′)) is
a quasi-isomorphism. Recall that H i(C(C, C′)) is isomorphic to HomH0(C)(C, Σi(C′)),
where Σ denotes the translation functor on the triangulated category H 0(C). Sim-
ilarly, we identify H i(D(F (C), F (C′))) with HomH0(D)(F (C), Σi(F (C′))). By as-
sumption, H 0(F ) is a triangle equivalence between triangulated categories H 0(C)
and H 0(D); compare [3, Remark 1.8(i)]. Then H 0(F ) induces an isomorphism
HomH0(C)(C, Σi(C′)) −→ HomH0(D)(F (C), Σi(F (C′))).
We infer the required quasi-isomorphism.
(cid:3)
In the following examples, we fix the notation for some quasi-equivalences, which
will be used in the next section.
Example 3.2. Recall that a right dg D-module M is quasi-representable, provided
that it is isomorphic to D(−, D) in D(Dop) for some object D in D. Denote by ¯D
the full dg subcategory of DGProj-D formed by dg-projective quasi-representable
modules. Then the Yoneda embedding induces a quasi-equivalence YD : D → ¯D.
We identify a dg algebra B with a dg category with one object. We denote
by B-DGModfd the full dg subcategory of B-DGMod consisting of those left dg
B-modules with finite dimensional total cohomologies. Similarly, we have the dg
category B-DGProjfd.
The following example is implicitly contained in [8, 7.2].
Example 3.3. Let θ : C → B be a quasi-isomorphism between dg algebras. Then
there is a quasi-equivalence
B ⊗C − : C-DGProj −→ B-DGProj.
Using infinite devissage, one infers that the natural map P → B ⊗C P is a quasi-
isomorphism for any dg-projective C-module P .
In particular, the above quasi-
equivalence restricts to a quasi-equivalence
B ⊗C − : C-DGProjfd −→ B-DGProjfd.
LIFTABLE DERIVED EQUIVALENCES AND OBJECTIVE CATEGORIES
5
We identify a usual algebra as a dg algebra concentrated in degree zero. Then
dg modules are just complexes of usual modules. For a finite dimensional algebra
A, we denote by A-mod the abelian category of finite dimensional left A-modules,
and by A-proj its full subcategory formed by finitely generated projective modules.
Example 3.4. Let A be a finite dimensional algebra. Then A-DGMod is identi-
fied with Cdg(A-Mod). The dg-projective resolution functor pA : Cdg(A-Mod) →
A-DGProj restricts to
pA : C b
dg(A-mod) → A-DGProjfd.
Since pA sends each acyclic complex X to a contractible complex pA(X), it induces
a dg functor
p′
A : Db
For the construction, we put p′
where εX is the new generator in defining Db
Indeed, taking H 0(p′
observe that p′
well-known triangle equivalence Db(A-mod) ≃ H 0(A-DGProjfd).
dg(A-mod) −→ A-DGProjfd.
A(εX ) to be any contracting homotopy on pA(X),
dg(A-mod); see Example 2.4. We
A), we obtain the
A is a quasi-equivalence.
Denote by C−,b
dg (A-proj) the dg category formed by bounded-above complexes of
finitely generated projective A-modules, which have bounded cohomologies. Since
bounded-above complexes of projective modules are dg-projective, we have the
inclusion
incA : : C−,b
dg (A-proj) −→ A-DGProjfd.
This is a quasi-equivalence. We just recall that there is a well-know triangle equiv-
alence between the homotopy category K−,b(A-proj) and Db(A-mod) .
We denote by dgcat the category of small dg categories, whose morphisms are
dg functors. The homotopy category Hodgcat is the localization of dgcat with
respect to all the quasi-equivalences. In other words, Hodgcat is obtained from
dgcat by formally inverting quasi-equivalences. For two dg categories C and D,
we denote by [C, D] the corresponding Hom set in Hodgcat, whose elements are
usually denoted by C 99K D. We mention that any such morphism can be realised
as a roof C F←− C′ F ′
−→ D of dg functors, where F is a quasi-equivalence; moreover,
F can be taken as a semi-free resolution of C; see [7, Appendix B.5]. For details,
we refer to [16].
For the set-theoretical consideration relevant to us, we use the following remark.
Remark 3.5. We call a dg category C quasi-small, provided that the homotopy
category H 0(C) is essentially small. We choose for each isomorphism class in H 0(C)
a representative in C. These objects form a small dg full subcategory C′. By the
construction, the inclusion C′ ֒→ C is a quasi-equivalence. So, we identify C with
C′, and view C as an object in Hodgcat.
Denote by [cat] the category of small categories, whose morphisms are the iso-
morphism classes of functors. In particular, equivalences of categories are isomor-
phisms in [cat]. Therefore, the homotopy functor H 0 : dgcat → [cat] inverts quasi-
equivalences. By the universal property of the localization, we have the induced
functor
H 0 : Hodgcat −→ [cat],
C 7→ H 0(C).
Following [8, 7.1], a quasi-functor from C to D is a dg C-D-bimodule X such
that for each object C ∈ C, the right D-module X(−, C) is quasi-representable. We
denote by rep(C, D) the full subcategory of D(C ⊗ Dop) formed by quasi-functors; it
is a triangulated subcategory; see [7, Appendix E.2]. We denote by Iso(rep(C, D))
the set of isomorphism classes of quasi-functors.
6
XIAOFA CHEN, XIAO-WU CHEN
For each quasi-functor M , we take its dg-projective resolution p(M ). The quasi-
functor p(M ) defines a dg functor pM : C → ¯D sending C to (pM )(−, C). The
following diagram
pM
−−→ ¯D
YD←−− D
C
defines a morphism ΦM : C 99K D in Hodgcat. Here, we recall the quasi-equivalence
YD in Example 3.2.
The following bijection is fundamental; see [15, Sublemmas 4.4 and 4.5] and [16,
p.279, Corollary 1]. For an elementary proof, we refer to [2]. As mentioned in [3,
Remark 6.6], the morphism ΦM might be viewed as the generalized Fourier-Mukai
transform with M being its kernel.
Lemma 3.6. Keep the notation as above. Then there is a bijection
Iso(rep(C, D)) −→ [C, D], M 7→ ΦM .
The following notion is modified from [3, Definition 6.7]. Here, since the unique-
ness of the enhancement is not known, we have to fix the canonical one.
Definition 3.7. Let F : Db(A) → Db(B) be a triangle functor. We say that F is
liftable, provided that there is a morphism F : Db
dg(B) in Hodgcat,
called a dg lift of F , such that F is isomorphic to can−1
dg(A) 99K Db
B H 0( F )canA.
We observe that the composition of liftable functors is still liftable. Using the
following well-known lemma, we infer that a quasi-inverse of a liftable equivalence
is also liftable. We point out that liftable functors are called standard in [9, 9.8].
However, we reserve the terminology "standard functors" for the classical ones, that
is, derived tensor functors by complexes.
Lemma 3.8. Let F : Db(A) → Db(B) be a triangle equivalence. Then any dg lift
F of F is an isomorphism in Hodgcat.
dg(B) of F , where F1 is
Proof. We use the roof presentation Db
a quasi-equivalence. It follows that the dg category C is also pretriangulated. By
assumption, we infer that H 0(F2) is an equivalence. By Lemma 3.1, the dg functor
F2 is a quasi-equivalence, which implies that F is an isomorphism.
(cid:3)
dg(A)
F1←− C
F2−→ Db
4. Liftable and standard functors
In this section, we prove that the category of quasi-functors between the bounded
dg derived categories of two module categories is triangle equivalent to a certain
derived category of bimodules over the given algebras. Consequently, between the
bounded derived categories of two module categories, liftable functors coincide with
standard functors.
Let A and B be two finite dimensional algebras. Recall from [13, Definition 3.4]
that a triangle functor F : Db(A-mod) → Db(B-mod) is standard, provided that
F ≃ X ⊗L
A −
sends bounded complexes to bounded complexes, we infer that the underlying
complex XA of right A-modules is perfect, that is, isomorphic to some object in
Kb(Aop-proj).
A − for some bounded complex X of B-A-bimodules. Since X ⊗L
We will identify Db(A-mod) with K−,b(A-proj); compare Example 3.4. For
each complex P ∈ K−,b(A-proj) and N ≥ 0, we consider the brutal truncation
σ≥−N (P ), which is a subcomplex of P consisting of P n for n ≥ −N . The inclusion
incN : σ≥−N (P ) → P fits into a canonical exact triangle
(4.1)
σ≥−N (P )
incN−−−→ P −→ σ<−N (P ) −→ Σσ≥−N (P ),
where σ<−N P = P/σ≥−N (P ) is the quotient complex.
The following result is standard.
LIFTABLE DERIVED EQUIVALENCES AND OBJECTIVE CATEGORIES
7
Lemma 4.1. Let F : K−,b(A-proj) → K−,b(B-proj) be a triangle functor. Then
there is a natural number N0 such that H 0(F (incN )) is an isomorphism for each
complex P and N ≥ N0.
A the full subcategory of K−,b(A-proj)
Proof. For each interval I, we denote by DI
formed by those complexes X satisfying H i(X) = 0 for i /∈ I. Similarly, we
B of K−,b(B-proj) . These subcategories are closed under
have the subcategories DI
extensions.
The subcategory D[0,0]
is equivalent to A-mod, which has only finitely many
A ) ⊆ D[−N0+1,N0−1]
simple A-modules up to isomorphism. It follows that F (D[0,0]
A ) ⊆ D[a−N0+1,b+N0−1]
for N0 > 0 large enough. More generally, we have F (D[a,b]
.
It follows that F (σ<−N (P )) ∈ D(−∞,−2]
for each N ≥ N0. Applying the triangle
functor F to (4.1) and taking cohomologies, we infer the required isomorphism. (cid:3)
B
B
B
A
We denote by D(B ⊗ Aop) the derived category of complexes of B-A-bimodules.
Theorem 4.2. There is a triangle equivalence
rep(Db
dg(A-mod), Db
dg(B-mod)) ∼−→ {M ∈ D(B ⊗ Aop) MA is perfect},
sending a dg Db
dg(A-mod)-Db
dg(B-mod)-bimodule X to X(B, A).
Consequently, a triangle functor F : Db(A-mod) → Db(B-mod) is liftable if and
only if it is standard.
Proof. We use the sequence of quasi-equivalences in Example 3.4
Db
dg(A-mod)
p′
A−−→ A-DGProjfd incA←−−− C−,b
dg (A-proj).
In this proof, we identify Db
D := C−,b
dg (B-proj).
dg(A-mod) with C := C−,b
dg (A-proj), Db
dg(B-mod) with
We will actually prove that there is a triangle equivalence
rep(C, D) ∼−→ {M ∈ D(B ⊗ Aop) MA is perfect},
sending X to X(B, A), whose quasi-inverse sends M to the dg C-D-bimodule XM
defined by XM (Q, P ) = HomB(Q, M ⊗A P ) for P ∈ C and Q ∈ D.
Take X ∈ rep(C, D) and fix an isomorphism
ξ−,P : X(−, P ) ∼−→ D(−, F (P ))
in D(Dop) for each P ∈ C. Therefore, the dg bimodule X induces a triangle functor
F : K−,b(A-proj) −→ K−,b(B-proj).
In particular, X(B, P ) is isomorphic to F (P ) in D(B), which has bounded coho-
mologies. We claim that the following natural map
(4.2)
θP : X(B, A) ⊗A P −→ X(B, P ), m ⊗ p 7→ (−1)m·pp.m
is a quasi-isomorphism. Here, we view p ∈ P as an element in C(A, P ), and then
p.m denotes the left C-action on X. By the claim, X(B, A) ⊗A P has bounded
cohomologies for each P ∈ C. Therefore, the underlying complex X(B, A)A of right
A-modules is perfect.
We observe that θP is an isomorphism in the case that P ≃ Σi(A). It follows that
θP is an isomorphism for any bounded complex P in C. In general, we will show
that H i(θP ) is an isomorphism. By translation, we will only show that H 0(θP ) is
an isomorphism. We consider the brutal truncation σ≥−N (P ), which is a bounded
8
XIAOFA CHEN, XIAO-WU CHEN
subcomplex of P . The inclusion incN : σ≥−N (P ) → P induces the vertical maps in
the following commutative diagram.
X(B, A) ⊗A σ≥−N (P )
θσ
≥−N (P )
/ X(B, σ≥−N (P ))
X(B, A) ⊗A P
θP
/ X(B, P )
∼
∼
/ F (σ≥−N (P ))
/ F (P )
Since X(B, A) has bounded cohomologies, the leftmost vertical map induces an
isomorphism on H 0 for sufficiently large N . By Lemma 4.1, a similar remark holds
for the rightmost one. Then the claim follows from the isomorphism θσ≥−N (P ).
For each Q ∈ D, we claim that the following natural map
(4.3)
δ : X(Q, P ) −→ HomB(Q, X(B, P )),
x 7→ (q 7→ x.q)
is a quasi-isomorphism. Here, q ∈ Q is viewed as an element in D(B, Q), and x.q
denotes the right D-action on X.
To see the claim, we use the isomorphism ξB,P : X(B, P ) → F (P ) in D(B).
Then we have the following quasi-isomorphisms of complexes
HomB(Q, X(B, P )) ≃ HomB(Q, F (P )) = D(Q, F (P )) ≃ X(Q, P ).
Then the claim follows, since δ is compatible with the above quasi-isomorphisms.
Combining the quasi-isomorphisms (4.2) and (4.3), we obtain a roof of quasi-
isomorphisms
(4.4)
X(Q, P ) δ−→ HomB(Q, X(B, P ))
HomB (Q,θP )
←−−−−−−−− HomB(Q, X(B, A) ⊗A P ).
This gives rise to an isomorphism of dg C-D-bimodules in D(C ⊗ Dop). Then the
required mutually inverse equivalences follows immediately.
It remains to prove the consequence. The "if" part is well known. Assume that
F ≃ M ⊗L
A − for a bounded complex M of B-A-bimodules with MA perfect. By
using projective resolutions of B-A-bimodules and truncations, we may assume that
MA lies in Kb(Aop-proj). Then the dg functor
M ⊗A − : Db
dg(A-mod) −→ Db
dg(B-mod)
is well-defined, which is a dg lift of F .
For the "only if" part, we assume that F admits a dg lift F : C 99K D. By
Lemma 3.6, we may assume that F = ΦX for some X ∈ rep(C, D). We identify X
with its dg-projective resolution p(X). By definition, H 0(ΦX )(P ) is representing
the right dg D-module X(−, P ). By (4.4), we infer that H 0(ΦX )(P ) is isomorphic
to X(B, A) ⊗A P . More precisely, we might replace X(B, A) by a bounded-above
complex M of finitely generated projective B-A-bmodules. Then H 0(ΦX )(P ) is
isomorphic to M ⊗A P . Consequently, F is identified with
M ⊗A − : K−,b(A-proj) −→ K−,b(B-proj).
This proves that F is standard.
(cid:3)
5. A factorization theorem for derived equivalences
In this section, we prove a factorization theorem for derived equivalences: any
derived equivalence between a module category and an abelian category is a com-
position of a pseudo-identity with a liftable derived equivalence.
The following notions are taken from [5, Definitions 3.8 and 5.1]; compare [5,
Lemma 5.2]. For an abelian category B, we identify B as the full subcategory of
Db(B) formed by stalk complexes concentrated in degree zero. More generally, we
/
/
/
/
LIFTABLE DERIVED EQUIVALENCES AND OBJECTIVE CATEGORIES
9
denote by Σn(B) the full subcategory formed by stalk complexes concentrated in
degree −n.
Definition 5.1. We call a triangle functor F : Db(B) → Db(B) a pseudo-identity,
provided that F (X) = X for each complex X, and that for each integer n, the
restriction F Σn(B) : Σn(B) → Σn(B) is the identity functor.
The abelian category B is called D-standard, provided that any pseudo-identity
on Db(B) is isomorphic, as triangle functors, to the identity functor.
We observe that a pseudo-identity is necessarily an autoequivalence; see [5,
Lemma 3.6]. The main motivation of introducing D-standard categories is the
following result: the module category A-mod for a finite dimensional algebra A is
D-standard if and only if any derived equivalence F : Db(A-mod) → Db(B-mod)
is standard; see [5, Theorem 5.10]. Therefore, the well-known open question [13]
about standard derived equivalences is equivalent to the conjecture that any module
category A-mod is D-standard. On the other hand, there exists a triangle functor
between the bounded derived categories of module categories, which is neither an
equivalence nor standard; see [14, Corollary 1.5].
In what follows, A will be a finite dimensional algebra and A an abelian category.
Lemma 5.2. Let F : Db(A-mod) → Db(A-mod) be a triangle functor. Assume
that there is an isomorphism θ : F (A) → A such that θ ◦ F (a) = a ◦ θ for each
morphism a : A → A. Then F is isomorphic to a pseudo-identity.
Proof. We observe that F induces isomorphisms
HomDb(A-mod)(A, Σn(A)) −→ HomDb(A-mod)(F (A), F Σn(A))
for each integer n. The cases n 6= 0 are trivial, since both sides equal zero. If n = 0,
we just use the assumption F (a) = θ−1 ◦ a ◦ θ for each endomorphism a on A.
We identify Kb(A-proj) with the smallest triangulated subcategory of Db(A-mod)
containing A and closed under direct summands. We observe that F (Kb(A-proj)) ⊆
Kb(A-proj). By Beilinson's Lemma, the restriction F Kb(A-proj) is an equivalence.
Then F is an autoequivalence by applying the last statement in [4, Proposition 3.4]
or, alternatively by the equivalence in [10, Theorem 6.2].
Recall that a complex X lies in A-mod if and only if HomDb(A-mod)(A, Σn(X)) =
0 for n 6= 0.
It follows from the equivalence F that F (X) lies in A-mod for
X ∈ A-mod. So, we have the restriction F A-mod : A-mod → A-mod. By the
isomorphism θ, it is standard to see that F A-mod is isomorphic to the identity
functor. Then we are done by [5, Corollary 3.9].
(cid:3)
The following factorization theorem extends [5, Proposition 5.8], which is essen-
tially due to [13, Corollary 3.5].
Theorem 5.3. Let F : Db(A-mod) → Db(A) be a triangle equivalence. Then
there is a factorization F ≃ F2F1 of triangle functors, where F1 : Db(A-mod) →
Db(A-mod) is a pseudo-identity and F2 : Db(A-mod) → Db(A) is a liftable equiv-
alence.
Moreover, such a factorization is unique. More precisely, for another factoriza-
2 a liftable equiva-
1 a pseudo-identity on Db(A-mod) and F ′
1 and F2 ≃ F ′
2.
tion F ≃ F ′
1 with F ′
lence, we have F1 ≃ F ′
2F ′
dg(A)(T )op to be the opposite dg endomor-
Proof. Set T = F (A), and Γ = EndDb
phism algebra of T in Db
dg(A).
Recall that H n(Γ) is isomorphic to HomDb(A)(T, Σn(T )) for each integer n. By
the equivalence F , we infer that H n(Γ) = 0 for n 6= 0 and that H 0(Γ) is isomorphic
to A. Denote by τ≤0(Γ) the good truncation of Γ, that is, τ≤0(Γ) = Li<0 Γi ⊕
10
XIAOFA CHEN, XIAO-WU CHEN
Γ. Then τ≤0(Γ) is a dg subalgebra of Γ and H 0(Γ) is a quotient algebra of
Kerd0
τ≤0(Γ). Therefore, we have quasi-isomorphisms of dg algebras Γ ←֓ τ≤0(Γ) ։ A.
dg(A), Db
For each object X ∈ Db
dg(A)(T, X) is naturally a left dg Γ-module.
We observe that H n(Db
dg(A)(T, X)) is isomorphic to HomDb(A)(T, Σn(X)), which
is further isomorphic to HomDb(A-mod)(A, ΣnF −1(X)) by the equivalence F . It
follows that Db
dg(A)(T, X) lies in Γ-DGModfd.
We define a morphism F : Db
dg(A) 99K Db
dg(A-mod) by the following diagram.
Db
dg(A)
F
Db
dg(A)(T,−)
/ Γ-DGModfd
pΓ
/ Γ-DGProjfd
Γ⊗τ≤0 (Γ)−
Db
dg(A-mod)
p′
A
/ A-DGProjfd
A⊗τ≤0 (Γ)−
τ≤0(Γ)-DGProjfd
For the dg-projective resolution functor pΓ, we refer to Example 2.3. For the
quasi-equivalence p′
A, we refer to Example 3.4. The other two quasi-equivalences
are induced by quasi-isomorphisms between dg algebras; see Example 3.3. The
middle four dg categories are quasi-small. So we have to apply Remark 3.5 and
view them in Hodgcat.
By chasing the diagram, we observe that H 0( F )(T ) is isomorphic to A. Consider
the following composition
F1 : Db(A-mod) F−→ Db(A)
canA-modH0( F )canA
−−−−−−−−−−−−−→ Db(A-mod).
We have an isomorphism θ : F1(A) → A. Recall that any morphism a : A → A in
Db(A-mod) is given by the right multiplication by some element in A. Then by
chasing the diagram for F , we observe that θ ◦ F1(a) = a ◦ θ. By Lemma 5.2, F1 is
isomorphic to a pseudo-identity. In particular, F1 is an autoequivalence. Therefore,
H 0( F ) is also an equivalence, and thus by Lemma 3.1 F is an isomorphism in
Hodgcat. We observe that F2 = F (F1)−1 is liftable, whose dg lift is given by
( F )−1.
For the uniqueness of factorizations, we just observe that F ′
is liftable. By Theorem 4.2, F ′
1F1
Lemma 5.9], we infer that F ′
1F1
are done.
2)−1F2
−1 is standard, which is a pseudo-identity. By [5,
−1 is isomorphic to the identity functor. Then we
(cid:3)
1(F1)−1 ≃ (F ′
Corollary 5.4. Assume that there are triangle equivalences among Db(A-mod),
Db(A) and Db(B). Then the following statements are equivalent:
(1) The category A-mod is D-standard;
(2) Any triangle equivalence Db(B) → Db(A) is liftable;
(3) Any triangle autoequivalence on Db(A) is liftable.
Proof. For "(1) ⇒ (2)", we apply Theorem 5.3 to infer that all derived equivalences
Db(A-mod) → Db(A) and Db(A-mod) → Db(B) are liftable. Then (2) follows. The
implication "(2) ⇒ (3)" is trivial.
For "(3) ⇒ (1)", we take a pseudo-identity F1 on Db(A-mod). By Theorem 5.3,
there is a liftable equivalence F2 : Db(A-mod) → Db(A). Then F2F1(F2)−1 is a
triangle autoequivalence on Db(A), which is necessarily liftable. It follows that F1
is also liftable. Recall that a liftable pseudo-identity is isomorphic to the identity
functor; compare the last paragraph in the proof of Theorem 5.3. This proves
(1).
(cid:3)
✤
✤
✤
/
/
/
o
o
O
O
LIFTABLE DERIVED EQUIVALENCES AND OBJECTIVE CATEGORIES
11
Remark 5.5. Keep the assumptions as above. We do not know the relation be-
tween these equivalent statements and the D-standardness of the abelian categories
A and B.
We are in a position to give the first proof to the theorem in the introduction.
For a noetherian scheme X, we denote by coh-X the abelian category of coherent
sheaves on X.
Theorem 5.6. Assume that there is a triangle equivalence between Db(A-mod)
and Db(coh-X) with X a smooth projective scheme. Then A-mod is D-standard, or
equivalently, any triangle equivalence F : Db(A-mod) → Db(B-mod) is standard.
Proof. Recall from [12] that any triangle autoequivalence on Db(coh-X) is a Fourier-
Mukai functor, and thus liftable by [3, Proposition 6.11]. Applying Corollary 5.4
and [5, Theorem 5.10], we are done.
(cid:3)
6. The objective categories
In this section, we introduce the notions of objective categories and triangle-
objective triangulated categories. The basic examples of triangle-objective trian-
gulated categories are the bounded derived categories of coherent sheaves over pro-
jective varieties over an algebraically closed field.
We say that an endofunctor F on a category A is object-preserving, if F (X) ≃ X
for each object X ∈ A.
Definition 6.1. An additive category A is called objective, provided that any
object-preserving autoequivalence on A is isomorphic to the identity functor IdA.
Similarly, a triangulated category T is called triangle-objective, provided that
any object-preserving triangle autoequivalence on T is isomorphic, as a triangle
functor, to the identity functor IdT .
The following observation motivates the above notions.
Lemma 6.2. Let A be an abelian category. Consider the following statements:
(1) The abelian category A is D-standard and objective;
(2) The bounded derived category Db(A) is triangle-objective;
(3) The abelian category A is D-standard.
Then we have the implications "(1) ⇒ (2) ⇒ (3)".
Proof. To see "(1) ⇒ (2)", we take an object-preserving triangle autoequivalence
F on Db(A). The restriction F A is object-preserving. By the assumptions in (1),
we infer that F A is isomorphic to the identity functor IdA. By [5, Corollary 3.9],
F is isomorphic to a pseudo-identity on Db(A). Since A is D-standard, we infer
that F is isomorphic to the identity functor. The implication "(2) ⇒ (3)" is clear,
since any pseudo-identity is object-preserving.
(cid:3)
Let R be a commutative noetherian k-algebra. Denote by R-mod the abelian
category of finitely generated R-modules.
Given a k-algebra automorphism σ : R → R and an R-module M , we denote by
σ(M ) the twisted module: the new R-action is given by a◦m = σ−1(a).m, where
the dot "." denotes the R-action on M . This gives rise to the twist automorphism
σ(−) : R-mod −→ R-mod.
Example 6.3. Denote by k[ǫ] the algebra of dual numbers. By [5, Theorem 7.1],
k[ǫ]-mod is D-standard. However, k[ǫ]-mod is not objective and Db(k[ǫ]-mod) is
not triangle-objective, provided that the field k contains at least three elements.
12
XIAOFA CHEN, XIAO-WU CHEN
Fix a ∈ k satisfying a 6= 0, 1. Consider the automorphism σ on k[ǫ] such that
σ(ǫ) = aǫ. The twist automorphisms σ(−), defined on k[ǫ]-mod and Db(k[ǫ]-mod),
are both object-preserving, but neither is isomorphic to the identity functor.
The following condition arises naturally.
Condition (Obj): any k-algebra automorphism σ : R → R satisfying σ(I) = I
for each ideal I, necessarily equals IdR.
Lemma 6.4. Let R be a commutative noetherian ring satisfying Condition (Obj).
Then R-mod is objective.
Proof. Assume that F : R-mod → R-mod is an object-preserving autoequivalence.
Since F (R) ≃ R, it follows that F is isomorphic to the twist automorphism σ(−)
for some automorphism σ. We observe that σ(R/I) ≃ R/σ(I) for each ideal I. By
the isomorphism σ(R/I) ≃ R/I and taking their annihilator ideals, we infer that
σ(I) = I. By Condition (Obj), we have σ = IdR. Consequently, F is isomorphic to
the identity functor.
(cid:3)
Here are some examples of rings satisfying Condition (Obj).
Example 6.5. (1) The polynomial algebras satisfy Condition (Obj). More gener-
ally, we assume that R is an integral domain such that any invertible element is a
scalar. Then R satisfies Condition (Obj).
To verify the condition, we take an automorphism σ : R → R satisfying σ(I) = I.
For any non-scalar a ∈ R, we have σ(Ra) = Rσ(a) = Ra. It follows that σ(a) = λa
for some λ ∈ k. Similarly, σ(1 + a) = λ′(1 + a) for some λ′ ∈ k. By comparing
these two identities, we infer that λ = 1 = λ′.
(2) Any reduced affine algebra over an algebraically closed field satisfies Con-
dition (Obj). More generally, we assume that the Jacobson radical of R is zero
and that for each maximal ideal m, the natural homomorphism k → R/m is an
isomorphism. Then R satisfies Condition (Obj).
For the verification, we claim that a − σ(a) is contained in any maximal ideal
m. By assumption, there is some λ ∈ k satisfying a − λ ∈ m. Then we have
σ(a) − λ ∈ σ(m) = m. The claim follows immediately.
The following result shows that objective categories are ubiquitous in algebraic
geometry. For a sheaf F , we denote by supp(F ) its support.
Proposition 6.6. Let (X, O) be a noetherian scheme such that there is a finite
affine open covering X = S Ui, where Ui = Spec(Ri) with each Ri satisfying Con-
dition (Obj). Then coh-X is objective.
Assume further that X is projective such that the maximal torsion subsheaf
T0(O) ⊆ O of dimension zero is trivial. Then Db(coh-X) is triangle-objective.
Proof. Let F : coh-X → coh-X be an object-preserving auto-equivalence. In par-
ticular, F fixes the structure sheaf O.
It is well-known that there is a unique
automorphism θ on X such that F ≃ θ∗, the pullback functor; see [1, Theorem 5.4].
For each closed subset Z ⊆ X, we have an ideal sheaf I with supp(O/I) = Z.
Then we have supp(θ∗(O/I)) = θ−1(Z). By the isomorphism θ∗(O/I) ≃ O/I, we
infer that θ−1(Z) = Z. In particular, for the given affine open subsets Ui, we have
θ−1(Ui) = Ui. Therefore, the restriction θUi : Ui → Ui corresponds to an k-algebra
automorphism σi on Ri, that is, θUi = Spec(σi).
We have the following commutative diagram
coh-X
res
/ coh-Ui
Ri-mod
θ∗
(θUi )∗
σi (−)
coh-X
res
/ coh-Ui
Ri-mod,
/
/
LIFTABLE DERIVED EQUIVALENCES AND OBJECTIVE CATEGORIES
13
where "res" is the restriction functor, and we identify coh-Ui with Ri-mod. The
restriction functor "res" induces the well-known equivalence between coh-Ui and the
Serre quotient category of coh-X by those sheaves supported on the complement
of Ui; compare [1, Example 4.3]. It follows that (θUi )∗ and thus σi (−) are object-
preserving. By the assumption on Ri, it follows that σi = IdRi and thus θUi = IdUi
for each i. Therefore, θ = IdX, proving the first statement.
For the last statement, we apply [11, Lemma 9.2] to infer that coh-X has an
ample sequence in the sense of [12]. By [5, Proposition 5.7], we deduce that coh-X
is D-standard. Using the proved statement and Lemma 6.2, we are done.
(cid:3)
By Example 6.5(2), a reduced projective scheme over an algebraically closed
field satisfies the above conditions. Hence, the following immediate consequence
of Proposition 6.6 and Lemma 6.2 gives the second proof to the theorem in the
introduction, when the field k is algebraically closed. As a consequence, the smooth
hypothesis of the scheme can be relaxed.
Corollary 6.7. Let A be a finite dimensional algebra. Assume that there is a
triangle equivalence between Db(A-mod) and Db(coh-X) for a projective scheme X
satisfying the conditions in Proposition 6.6. Then Db(A-mod) is triangle-objective,
and thus A-mod is D-standard.
(cid:3)
Acknowledgements. This work is supported by the National Natural Science
Foundation of China (Nos. 11522113 and 11671245), the Fundamental Research
Funds for the Central Universities, and Anhui Initiative in Quantum Information
Technologies (AHY150200).
References
[1] M. Brandenburg, Rosenberg's reconstruction theorem, Expo. Math. 36 (2018), 98 -- 117.
[2] A. Canonaco, and P. Stellari, Internal hom via extensions of dg functors, Adv. Math.
277 (2015), 100 -- 123.
[3] A. Canonaco, and P. Stellari, A tour about existence and uniqueness of dg enhance-
ments and lifts, J. Geom. Phys. 122 (2017), 28 -- 52.
[4] X.W. Chen, Representablity and autoequivalence groups, arXiv:1810.00332v2, 2018.
[5] X.W. Chen, and Y. Ye, The D-standard and K-standard categories, Adv. Math. 333
(2018), 159 -- 193.
[6] X.W. Chen, and C. Zhang, The derived-discrete algebras and standard equivalences, J.
Algebra 525 (2019), 259 -- 283.
[7] V. Drinfeld, DG quotients of DG categories, J. Algebra 272 (2004), 643 -- 691.
[8] B. Keller, Deriving DG categories, Ann. Sci. ´Ecole Norm Sup. (4) 27(1) (1994), 63 -- 102.
[9] B. Keller, On triangulated orbit categories, Doc. Math. 10 (2005), 551 -- 581.
[10] H. Krause, Completing perfect complexes, arXiv:1805.10751v3, 2018.
[11] V. Lunts, and D. Orlov, Uniqueness of enhancements for triangulated categories, J.
Amer. Math. Soc. 23 (2010), 853 -- 908.
[12] D. Orlov, Equivalences of derived categories and K3 surfaces, J. Math. Sci. 84 (1997),
1361 -- 1381.
[13] J. Rickard, Derived equivalences as derived functors, J. London Math. Soc. 43(2) (1991),
37 -- 48.
[14] A. Rizzardo, and M. Van den Bergh, An example of a non-Fourier-Mukai functor
between derived categories of coherent sheaves, arXiv:1410.4039v2, 2015.
[15] B. Toen, The homotopy category of dg-categories and derived Morita theory, Invent. Math.
167 (2007), 615 -- 667.
[16] B. Toen, Lectures on dg-categories, in: Topics in Algebraic and Topological K-Theory,
Lect. Notes Math. 2008, 243 -- 302, Springer Berlin Heidelberg, 2011.
Xiaofa Chen, Xiao-Wu Chen
Key Laboratory of Wu Wen-Tsun Mathematics, Chinese Academy of Sciences,
School of Mathematical Sciences, University of Science and Technology of China, Hefei 230026,
Anhui, PR China
|
0707.3754 | 3 | 0707 | 2012-02-06T16:38:53 | A hermitian analogue of the Broecker-Prestel theorem | [
"math.RA"
] | The Broecker-Prestel local-global principle characterizes weak isotropy of quadratic forms over a formally real field in terms of weak isotropy over the henselizations and isotropy over the real closures of that field. A hermitian analogue of this principle is presented for algebras of index at most two. An improved result is also presented for algebras with a decomposable involution, algebras of pythagorean index at most two, and algebras over SAP and ED fields. | math.RA | math |
A HERMITIAN ANALOGUE OF THE BR OCKER -- PRESTEL THEOREM
VINCENT ASTIER AND THOMAS UNGER
ABSTRACT. The Br ocker -- Prestel local-global principle characterizes weak iso-
tropy of quadratic forms over a formally real field in terms of weak isotropy
over the henselizations and isotropy over the real closures of that field. A
hermitian analogue of this principle is presented for algebras of index at most
two. An improved result is also presented for algebras with a decomposable
involution, algebras of pythagorean index at most two, and algebras over SAP
and ED fields.
1. INTRODUCTION
In the algebraic theory of quadratic forms over fields the problem of deter-
mining whether a form is isotropic (i.e., has a non-trivial zero) has led to the
development of several powerful local-global principles. They allow one to
test the isotropy of a form over the original field ("global" situation) by test-
ing it over a collection of other fields where the original problem is potentially
easier to solve ("local" situation).
The most celebrated local-global principle is of course the Hasse -- Minkowski
theorem which gives a test for isotropy over the rational numbers Q in terms
of isotropy over the p-adic numbers Qp for each prime p and the real numbers
R. More generally, Q may be replaced by any global field F and the collection
(cid:8){Qp}p prime, R(cid:9) may be replaced by the collection of local fields {Fv} (com-
pletions of F with respect to all valuations v on F). (See, for example, [Lam,
p. 170].)
Another landmark theorem in the theory of quadratic forms is Pfister's local-
global principle. Here a formally real field F plays the role of "global" field,
whereas the corresponding "local" fields are given by the collection of real
closures {FP} where P runs through the orderings of F. Pfister's local-global
principle then says that a quadratic form q is weakly hyperbolic over F if and
2000 Mathematics Subject Classification. 16K20, 11E39.
Key words and phrases. Central simple algebras, involutions, quadratic and hermitian forms,
local-global principles.
1
2
VINCENT ASTIER AND THOMAS UNGER
only if it is hyperbolic over FP for each ordering P on F or, in more familiar ter-
minology, q is a torsion form in the Witt ring W(F) if and only if the signature
of q is zero for each ordering P on F. (See, for example, [Lam, p. 253].)
In this paper we are interested in a local-global principle for weak isotropy,
due to Br ocker [Br o] and Prestel [Pr2, Theorem 8.12]. The role of "global"
field is again played by a formally real field F, but this time the "local" fields
are made up of the collection {FH
v } of henselizations of F with respect to real
valuations v on F (i.e., valuations with formally real residue field) together
with the collection {FP} of real closures of F at orderings P of F:
Theorem 1.1 (Br ocker -- Prestel). Let F be a formally real field and let q be a non-
singular quadratic form over F. Then q is weakly isotropic over F if and only if the
following conditions are satisfied:
(i) q ⊗ FH
(ii) q ⊗ FP is isotropic for every ordering P on F.
v is weakly isotropic for every real valuation v on F;
It is possible to state this theorem with weaker local conditions. For instance,
it suffices to consider only archimedean orderings, cf. [Pr2, Theorem 8.13] or
[EP, Theorem 6.3.1].
Quadratic form theory has proved very useful in providing tools for study-
ing central simple algebras with an involution and hermitian forms over divi-
sion algebras, which can be thought of as "twisted" versions of quadratic form
theory. One can find ample examples of this phenomenon in the book [KMRT]
and in the recent literature.
Given the usefulness of local-global principles in quadratic form theory it
seems natural to try to develop them in more generality. The second author
has been involved in several such projects [LSU, LU, LUV]. In this paper we
extend the Br ocker -- Prestel local-global principle to algebras of index at most
2 with an involution of the first kind over formally real fields.
The structure of this paper is as follows: in Section 2 we recall the basic facts
about quadratic and hermitian forms and algebras with involution that will
be needed. In Section 3 we show that, for central simple algebras of arbitrary
index with decomposable involution of the first kind, the problem at hand
reduces to the generalized version of Pfister's local-global principle in [LU].
In Section 4 we prove the hermitian version of the Br ocker -- Prestel theorem
for algebras of index at most 2. In Section 5 we enlarge the class of algebras
for which the theorem holds to those that become of index at most 2 over the
A HERMITIAN ANALOGUE OF THE BR OCKER -- PRESTEL THEOREM
3
pythagorean closure of their centre. In Section 6 we show that the theorem
holds for arbitrary central simple algebras with an involution of the first kind
over a SAP field. Finally, in Section 7 we show that over an ED field the theo-
rem reduces to the local-global principle studied in [LSU].
We thank Karim Becher for many discussions on this and related topics dur-
ing Unger's visit to the University of Konstanz in March 2007. Unger also
wishes to thank Alex Prestel for encouraging him to look at the Br ocker -- Prestel
theorem in a hermitian setting during his UCD visit in April 2004.
2. PRELIMINARIES
We assume that the reader is familiar with the basic notions from the theo-
ries of quadratic forms, hermitian forms and central simple algebras with an
involution. We refer to the standard references [Lam, Sch, KMRT] for details.
Let F be a field of characteristic different from 2 and let D be a central divi-
sion algebra over F equipped with an involution ϑ of the first kind. Let ϕ be a
quadratic form over F or, more generally, a hermitian or skew-hermitian form
over (D, ϑ). The forms ϕ appearing in this paper are assumed to be nonsingu-
lar. In addition we exclude alternating forms over F since they are hyperbolic
and thus of no relevance to the problem at hand. Thus, our forms ϕ can always
be diagonalized.
For n ∈ N we denote the n-fold orthogonal sum
ϕ ⊥ ϕ ⊥ . . . ⊥ ϕ
simply by n × ϕ. We call ϕ weakly isotropic or weakly hyperbolic when there
exists an n ∈ N such that n × ϕ is isotropic or hyperbolic, respectively. A
weakly hyperbolic form is also called a torsion form. A form which is not
weakly isotropic is called strongly anisotropic.
Let A be a central simple F-algebra, equipped with an involution σ of the
first kind. We call (A, σ) isotropic if there exists a nonzero x ∈ A such that
σ(x)x = 0. The reader can verify that this is equivalent with the definition
given in [KMRT, 6.A]. In particular, (A, σ) is isotropic if there exists an idem-
potent e 6= 0 in A such that σ(e)e = 0.
We call (A, σ) weakly isotropic if there exist nonzero x1, . . . , xn ∈ A such that
(1)
σ(x1)x1 + · · · + σ(xn)xn = 0
and strongly anisotropic otherwise (see [LSU, Definition 2.2]).
4
VINCENT ASTIER AND THOMAS UNGER
Assume now that the division algebra D is Brauer equivalent to A. It follows
from [KMRT, 4.A] that σ is the adjoint involution adh of some hermitian or skew-
hermitian form h over (D, ϑ). We then call (A, σ) hyperbolic if h is a hyperbolic
form. This is equivalent with the existence of a nonzero idempotent e ∈ A
such that σ(e) = 1 − e, cf. [KMRT, 6.B].
We call (A, σ) weakly hyperbolic if there exists an n ∈ N such that
n × (A, σ) := (Mn(F), t) ⊗F (A, σ)
is hyperbolic, where t denotes the transpose involution, see [LU, Definition 3.1].
The reader can now easily verify that for σ = adh and for any n ∈ N,
n × (A, adh) ∼= (Mn(A), adn×h),
(2)
n × (A, adh) is isotropic ⇔ n × h is isotropic ⇔ (1) holds,
n × (A, adh) is hyperbolic ⇔ n × h is hyperbolic
(see also [LSU, Lemma 2.4]).
Now let F be a formally real field and let q be a quadratic form over F. For
an ordering P on F we denote the signature of q at P by sigP q.
Let (A, σ) be a central simple F-algebra with involution of the first kind. The
signature of σ at P is defined by
sigP σ := qsigP Tσ,
where Tσ is the involution trace form of (A, σ), which is a quadratic form on A
defined by
Tσ(x) := TrdA(σ(x)x), ∀x ∈ A,
where TrdA denotes the reduced trace of A, see [KMRT, 11.10].
We assume from now on that F is a formally real field.
3. THE DECOMPOSABLE CASE
Let A be a central simple F-algebra equipped with an involution σ of the
first kind. We call (A, σ) decomposable if we can write
(3)
(A, σ) ∼= (Q1, σ1) ⊗F · · · ⊗F (Qr, σr) ⊗F (Ms(F), t),
for certain quaternion division algebras Qi with involution of the first kind σi,
1 ≤ i ≤ r and s ≥ 1. Here t denotes again the transpose involution.
A HERMITIAN ANALOGUE OF THE BR OCKER -- PRESTEL THEOREM
5
The following proposition shows that for such algebras the generalized ver-
sion of Pfister's local-global principle [LU, Theorem 3.2] suffices to character-
ize weak isotropy.
Proposition 3.1. Let (A, σ) be a decomposable algebra with involution over F. The
following statements are equivalent:
(i) (A, σ) is weakly isotropic;
(ii) (A, σ) is weakly hyperbolic;
(iii) Tσ is weakly isotropic;
(iv) Tσ is weakly hyperbolic;
(v) sigP σ = 0 for all orderings P on F;
(vi) sigP Tσ = 0 for all orderings P on F.
trace forms Tσi are all 2-fold Pfister forms [KMRT, 11.6]. Since Tσ = Nr
Proof. Write (A, σ) as in (3). Since we are interested in when (A, σ) is weakly
isotropic or weakly hyperbolic, we may assume that s = 1. The involution
i=1 Tσi
(this follows from the formula for the reduced trace of a tensor product), Tσ
is a Pfister form, which proves the equivalence of (iii) and (iv). Obviously
(ii) ⇒ (i) ⇒ (iii). By the definition of signature of an involution, (v) and (vi)
are equivalent. Finally, (iv) and (vi) are equivalent by Pfister's local-global
principle and (v) and (ii) are equivalent by the generalized version of Pfister's
local-global principle [LU, Theorem 3.2].
4. ALGEBRAS OF INDEX AT MOST 2
In this section we assume that the index of A is at most 2. This means that
either A is split, i.e., A is isomorphic to a full matrix algebra over F, or A is
isomorphic to a full matrix algebra over a quaternion division F-algebra D.
Remark 4.1. Let a, b ∈ F× and let D = (a, b)F be a quaternion division algebra
with quaternion conjugation γ. The norm form of D is h1, −a, −b, abi. Let h be
a hermitian form over (D, γ). Then h ≃ hα1, . . . , αni and all αi ∈ F×. To h we
can associate its trace form qh(x) := h(x, x) which is a quadratic form over F.
It is clear that h is (weakly) isotropic if and only if qh is (weakly) isotropic. A
simple computation shows that
qh ≃ h1, −a, −b, abi ⊗ hα1, . . . , αni.
6
VINCENT ASTIER AND THOMAS UNGER
Furthermore, a well-known theorem of Jacobson [Jac] says that, if D is a divi-
sion algebra, then the trace map
tr : S(D, γ) → S(F), h 7→ qh
is an injective morphism of Witt semi-groups of isometry classes of hermitian
forms over (D, γ) and quadratic forms over F, respectively.
When D is split, D ∼= M2(F) ∼= (1, 1)F, its norm form is hyperbolic and thus
qh is always hyperbolic.
Theorem 4.2. Let A be a central simple F-algebra of index ≤ 2, equipped with an
involution σ of the first kind. Then (A, σ) is weakly isotropic over F if and only if the
following conditions are satisfied:
(i) (A ⊗F FH
(ii) (A ⊗F FP, σ ⊗ idFP) is isotropic for all orderings P on F.
v , σ ⊗ idFH
) is weakly isotropic for all real valuations v on F;
v
Proof. We treat the split case first: assume that A is a full matrix algebra over
F, A = Mℓ(F). If σ is symplectic, then (ℓ is even and) σ is adjoint to a skew-
symmetric bilinear form over F and is thus necessarily hyperbolic. Hence the
statement is trivially true in this case. If σ is orthogonal, then σ is adjoint to a
quadratic form over F and the result follows immediately from Theorem 1.1.
Next, assume that ind(A) = 2, i.e. that A = Mn(D), with D a quaternion di-
vision algebra over F. We will prove the nontrivial direction by contraposition.
Thus, assume that (A, σ) is strongly anisotropic.
Let σ be orthogonal. Then σ = adh, where h is a skew-hermitian form over
(D, γ). Let K be a generic splitting field of D. It follows from [Dej, Corollaire]
or [PSS, Corollary 3.4] that (A, σ) remains strongly anisotropic after scalar ex-
tension to K. But AK := A ⊗F K is split. Thus, by the split case, there exists a
real valuation w on K such that
(AK ⊗K KH
w , σK ⊗ idK H
w
) ∼= (A ⊗F KH
w , σ ⊗ idK H
w
)
is strongly anisotropic, or there exists an ordering P on K such that (AK ⊗K
KP, σK ⊗ idKP) is anisotropic.
If there is a real valuation w on K such that (A ⊗F KH
w , σ ⊗ idK H
w
anisotropic, let v be the restriction of w to F and note that FH
) is strongly
w . Consider
v ⊂ KH
A HERMITIAN ANALOGUE OF THE BR OCKER -- PRESTEL THEOREM
7
the commutative diagram:
(A, σ)
⊗FH
v
⊗K
(A ⊗F K, σ ⊗ idK)
⊗K H
w
(A ⊗F FH
v , σ ⊗ idFH
v
⊗K H
w /
)
/ (A ⊗F KH
w , σ ⊗ idK H
)
w
Now (A ⊗F FH
would get a contradiction after scalar extension to KH
w .
v , σ ⊗ idFH
v
) has to be strongly anisotropic, for otherwise we
This argument can be reproduced in the ordering case by considering real
closures instead of henselizations.
Finally, suppose that σ is symplectic. Then σ = adh, where h is a her-
mitian form over (D, γ). By assumption (A, σ) = (Mn(D), adh) is strongly
anisotropic. By (2) this will be the case if and only if h is strongly anisotropic
over (D, γ). By Remark 4.1 this will be the case if and only if the quadratic
form qh is strongly anisotropic over F. Thus, by Theorem 1.1, there exists a real
valuation v on F such that qh ⊗ FH
v , or there
v
exists an ordering P on F such that q ⊗ FP is anisotropic over FP.
is strongly anisotropic over FH
If there exists a real valuation v on F such that qh ⊗ FH
v , consider the following commutative diagram:
over FH
v is strongly anisotropic
tr
S(D, γ)
⊗FH
v
S(F)
⊗FH
v
S(D ⊗F FH
v , γ ⊗ idFH
v
tr
)
/ S(FH
v )
v
v
v , γ ⊗ idFH
v = qh⊗FH
, we obtain that h ⊗ FH
v
). We observe that D ⊗F FH
v
Since qh ⊗ FH
(D ⊗F FH
since otherwise qh ⊗ FH
tradict the strong anisotropy of qh ⊗ FH
FH
v , adh⊗FH
This argument can be repeated when there exists an ordering P on F such
is strongly anisotropic over
remains a division algebra,
v would be hyperbolic by Remark 4.1, which would con-
v . By (2) we can conclude that (Mn(D) ⊗F
) is strongly anisotropic.
v
that q ⊗ FP is anisotropic upon replacing FH
v by FP.
/
/
/
/
/
8
VINCENT ASTIER AND THOMAS UNGER
Let D be a division algebra, finite-dimensional over its centre F. Morandi
showed that a valuation v on F extends to D if and only if D ⊗F FH
v is a divi-
sion algebra [Mor, Theorem 2]. When this happens, we call the valuation D-
admissible. In a similar vein we call and ordering P on F D-admissible if D ⊗F FP
is a division algebra.
When A = Mn(D) with D a quaternion division algebra, the notion of D-
admissibility can be used to restrict the set of local conditions needed to test
for weak isotropy in Theorem 4.2:
In case σ is symplectic, only D-admissible orderings and valuations are
needed (as observed in the proof of the theorem).
In case σ is orthogonal we do not know whether a set of valuations smaller
than the one provided by our use of the classical Br ocker -- Prestel theorem
would suffice. The set of orderings however is allowed to only consist of non-
D-admissible ones. This can be seen as follows: since K splits D, the norm
form N of D becomes isotropic and thus hyperbolic over K. Consequently,
sigP N ⊗ K = 0 for all orderings P on K and hence sigP′ N = 0 where P′ de-
notes the restriction of P to F (since signatures do not change under scalar
extension). Therefore, N ⊗ FP′ is hyperbolic and D ⊗F FP′ is split.
Remark 4.3. Some authors prefer to describe the local conditions involving val-
uations in Theorem 1.1 in terms of residue forms over the residue fields, rather
than extended forms over the henselizations. The reason for this is that the
residue fields are in general much easier to work with than the henselizations.
When A is not split, Theorem 4.2 can be reformulated as a local-global prin-
ciple for weak isotropy of hermitian forms over (D, γ) (when σ is symplectic)
or skew-hermitian forms over (D, γ) (when σ is orthogonal). In this way it
resembles the statement for quadratic forms (Theorem 1.1) more closely. In
addition, each local condition involving a D-admissible valuation can also be
stated in terms of residue (skew-)hermitian forms over the residue division
algebras using [Lar, Theorem 3.4].
Remark 4.4. The key ingredient used in the proof of Theorem 4.2 is the index-
two version of what is sometimes called the "anisotropic splitting conjecture",
which says: Let (A, σ) be a central simple algebra with orthogonal involution
over a field F, and let K be a generic splitting field of A. If (A, σ) is anisotropic
over F, then (A, σ) remains anisotropic over K.
A HERMITIAN ANALOGUE OF THE BR OCKER -- PRESTEL THEOREM
9
This conjecture is known to be true when the index of A is one (this is trivial
since K/F is purely transcendental) or two (by [PSS] or [Dej]), and when A is
a division algebra (by [Kar]).
5. ALGEBRAS OF PYTHAGOREAN INDEX AT MOST 2
Let A be a central simple F-algebra and let eF denote the pythagorean closure
of F. The pythagorean index of A, pind(A), is defined by Becher [Bec, §4] as
follows:
pind(A) := ind(A ⊗F eF),
where ind denotes the (Schur-) index. For example, let A = (a, b)F ⊗F (c, d)F
be a biquaternion division algebra over F, where a is a sum of squares in F, but
not a square. Then ind(A) = 4, but pind(A) ≤ 2.
The following lemma can be deduced immediately from [LSU, Lemma 3.9]:
Lemma 5.1. Let (A, σ) be a central simple algebra with involution of the first kind
over F. Then (A ⊗F eF, σ ⊗ ideF) is weakly isotropic if and only if (A, σ) is weakly
isotropic.
Proposition 5.2. Let A be a central simple F-algebra of pythagorean index pind(A) ≤
2, equipped with an involution σ of the first kind. Then (A, σ) is weakly isotropic over
F if and only if the following conditions are satisfied:
(i) (A ⊗F FH
(ii) (A ⊗F FP, σ ⊗ idFP) is isotropic for all orderings P on F.
v , σ ⊗ idFH
) is weakly isotropic for all real valuations v on F;
v
Proof. The non-trivial direction can be obtained as follows: assume that the
local conditions (i) and (ii) are satisfied. Then they are satisfied for (A ⊗F
w for every valuation w on eF,
eF, σ ⊗ ideF) since we can assume that FH
where v is the restriction of w to F, and that eFP = FP0 where P0 = P ∩ F for
every ordering P on eF.
Since by assumption ind(A ⊗F eF, σ ⊗ ideF) ≤ 2 it follows from Theorem 4.2
that (A ⊗F eF, σ ⊗ ideF) is weakly isotropic. We may conclude by Lemma 5.1.
v ⊆ eFH
6. ALGEBRAS OVER SAP FIELDS
Let F be a field, satisfying the Strong Approximation Property, or SAP field
for short. There are many ways of characterizing SAP fields (see [LSU, Defi-
nition 3.1] for an overview of all the equivalent definitions). For example, F is
10
VINCENT ASTIER AND THOMAS UNGER
SAP if and only if for all a, b ∈ F× the quadratic form h1, a, b, −abi is weakly
isotropic.
Proposition 6.1. Let A be a central simple algebra equipped with an involution σ of
the first kind over a SAP field F. Then (A, σ) is weakly isotropic over F if and only if
the following conditions are satisfied:
v , σ ⊗ idFH
v
) is weakly isotropic for all real valuations v on F;
(i) (A ⊗F FH
(ii) (A ⊗F FP, σ ⊗ idFP) is isotropic for all orderings P on F.
Proof. Let eF be the pythagorean closure of F, as before. Then A ⊗F eF is Brauer
equivalent to a quaternion division algebra (−1, b)eF for some b ∈ eF×, by [LSU,
Lemma 3.10]. Hence, pind(A) ≤ 2 and we can conclude by Proposition 5.2.
Remark 6.2. In the proof above we use the fact that every algebra of exponent 2
over a SAP field F has pythagorean index at most two. Karim Becher has
kindly communicated to us that the converse also holds, resulting in yet an-
other characterization of SAP fields: F is SAP if and only if pind(A) ≤ 2 for
any algebra A of exponent 2 over F.
7. ALGEBRAS OVER ED FIELDS.
Let F be a field, satisfying the Effective Diagonalization Property, or ED field
for short. Again, there are several different ways of characterizing ED fields,
but we will just use the original definition due to Ware [War, §2]: a field F
is ED if and only if every quadratic form over F is effectively diagonalizable,
i.e., every quadratic form ϕ over F is isometric to a form hb1, . . . , bni satisfying
bi ∈ P ⇒ bi+1 ∈ P for all 1 ≤ i < n and all orderings P on F.
For algebras with involution of the first kind over ED fields, Theorem 4.2
reduces to the following simpler statement, which is a reformulation of the
main theorem in [LSU].
Proposition 7.1. Let F be an ED field and let (A, σ) be a central simple algebra
with involution of the first kind over F. Then (A, σ) is weakly isotropic if and only if
(A ⊗F FP, σ ⊗ idFP) is weakly isotropic for every ordering P on F.
Proof. We show the non-trivial direction by contraposition. Assume that (A, σ)
is strongly anisotropic. By [LSU, Theorem 3.8] there exists an ordering P of F
such that (AP, σP) := (A ⊗F FP, σ ⊗ idFP) is definite.
A HERMITIAN ANALOGUE OF THE BR OCKER -- PRESTEL THEOREM
11
Since sigP σP := psigP TσP, TσP is definite and thus strongly anisotropic over
FP. To show that (AP, σP) is strongly anisotropic, assume the opposite for the
sake of contradiction. Then we can find nonzero x1, . . . , xℓ ∈ AP such that
∑ℓ
i=1 σP(xi)xi = 0. Taking the reduced trace of both sides gives TσP weakly
isotropic, a contradiction.
For quadratic forms a similar phenomenon happens in the presence of SAP
fields: the Br ocker -- Prestel theorem, Theorem 1.1, collapses to a theorem of
Prestel stating that a quadratic form over a SAP field F is weakly isotropic if
and only if it is isotropic over all real closures of F, cf. [Pr2, Theorem 9.1], [Pr1]
and [ELP].
REFERENCES
[Bec]
[Br o]
[Dej]
[EP]
[ELP]
[Jac]
[Kar]
K.J. Becher, Totally positive extensions and weakly isotropic forms, manuscripta math.
120 (2006), 83 -- 90.
L. Br ocker, Zur Theorie der quadratischen Formen uber formal reellen K orpern,
Math. Ann. 210 (1974), 233 -- 256.
I. Dejaiffe, Formes antihermitiennes devenant hyperboliques sur un corps de
d´eploiement, C. R. Acad. Sci. Paris, S´erie I 332 (2001), 105 -- 108.
A.J. Engler, A. Prestel, Valued fields, Springer Monographs in Mathematics, Springer-
Verlag, Berlin (2005).
R. Elman, T.Y. Lam, A. Prestel, On some Hasse principles over formally real fields,
Math. Z. 134 (1973), 291 -- 301.
N. Jacobson, A note on hermitian forms, Bull. Amer. Math. Soc. 46 (1940), 264 -- 268.
N. Karpenko, On anisotropy of orthogonal involutions, J. Ramanujan Math. Soc. 15
(2000), no. 1, 1 -- 22.
[KMRT] M.-A. Knus, A.S. Merkurjev, M. Rost, J.-P. Tignol, The Book of Involutions, Colloquium
[Lam]
[Lar]
Publications 44, American Mathematical Society, Providence, RI (1998).
T.Y. Lam, Introduction to quadratic forms over fields, Graduate Studies in Mathematics
67, American Mathematical Society, Providence, RI (2005).
D.W. Larmour, A Springer Theorem for Hermitian forms, Math. Z. 252 (2006), 459 --
472.
[LSU] D.W. Lewis, C. Scheiderer, T. Unger, A weak Hasse principle for central simple al-
gebras with an involution, Proceedings of the Conference on Quadratic Forms and
Related Topics (Baton Rouge, LA, 2001), Doc. Math. 2001, Extra Vol., 241 -- 251.
D.W. Lewis, T. Unger, A local-global principle for algebras with involution and her-
mitian forms, Math. Z. 244 (2003), 469 -- 477.
[LU]
[LUV] D.W. Lewis, T. Unger, J. Van Geel, The Hasse principle for similarity of Hermitian
forms, J. Algebra 285 (2005), 196 -- 212.
12
[Mor]
[PSS]
[Pr1]
[Pr2]
VINCENT ASTIER AND THOMAS UNGER
P. Morandi, The Henselization of a valued division algebra, J. Algebra 122 (1989),
232 -- 243.
R. Parimala, R. Sridharan, V. Suresh, Hermitian analogue of a theorem of Springer,
J. Algebra 243 (2001), 780 -- 789.
A. Prestel, Quadratische Semi-Ordnungen und quadratische Formen, Math. Z. 133
(1973), 319 -- 342.
A. Prestel, Lectures on Formally Real Fields, Lecture Notes in Mathematics 1093,
Springer-Verlag, Berlin (1984).
[Sch] W. Scharlau, Quadratic and Hermitian Forms, Grundlehren Math. Wiss. 270, Springer-
[War]
Verlag, Berlin (1985).
R. Ware, Hasse principles and the u-invariant over formally real fields, Nagoya
Math. J. 61 (1976), 117 -- 125.
FACHBEREICH MATHEMATIK UND STATISTIK, UNIVERSIT AT KONSTANZ, D-78457 KON-
STANZ, GERMANY
E-mail address: [email protected]
SCHOOL OF MATHEMATICAL SCIENCES, UNIVERSITY COLLEGE DUBLIN, BELFIELD, DUB-
LIN 4, IRELAND
E-mail address: [email protected]
|
1707.09389 | 1 | 1707 | 2017-07-28T19:39:22 | Generalized Hirano inverses in rings | [
"math.RA"
] | We introduce and study a new class of generalized inverse in rings. An element $a$ in a ring $R$ has generalized Hirano inverse if there exists some $b\in R$ such that $bab=b, b\in comm^2(a), a^2-ab \in R^{qnil}$ | math.RA | math |
GENERALIZED HIRANO INVERSES IN RINGS
MARJAN SHEIBANI ABDOLYOUSEFI AND HUANYIN CHEN
Abstract. We introduce and study a new class of general-
ized inverses in rings. An element a in a ring R has general-
ized Hirano inverse if there exists b ∈ R such that bab = b, b ∈
comm2(a), a2 − ab ∈ Rqnil. We prove that the generalized Hi-
rano inverse of an element is its generalized Drazin inverse.
An element a ∈ R has generalized Hirano inverse if and only if
there exists p = p2 ∈ comm2(a) such that a2 − p ∈ Rqnil. We
then completely determine when a 2×2 matrix over projective-
free rings has generalized Hirano inverse. Cline's formula and
additive properties for generalized Hirano inverses are thereby
obtained.
1. Introduction
Let R be an associative ring with an identity. The commutant
of a ∈ R is defined by comm(a) = {x ∈ R xa = ax}. Set
Rqnil = {a ∈ R 1 + ax ∈ U(R) for every x ∈ comm(a)}. We say
a ∈ R is quasinilpotent if a ∈ Rqnil. For a Banach algebra A it is
well known that (see [4, page 251])
a ∈ Aqnil ⇔ lim
n→∞
k an k
1
n = 0.
The double commutant of a ∈ R is defined by comm2(a) = {x ∈
R xy = yx for all y ∈ comm(a)}. An element a in R is said to
have generalized Drazin inverse if there exists b ∈ R such that
b = bab, b ∈ comm2(a), a − a2b ∈ Rqnil.
The preceding b is unique, if such element exists. As usual, it will
be denoted by ad, and called the generalized Drazin inverse of a.
2010 Mathematics Subject Classification. 15A09, 32A65, 16E50.
Key words and phrases. generalized Drazin inverse, matrix ring; Cline's for-
mula; additive property; Banach algebra.
1
2
MARJAN SHEIBANI ABDOLYOUSEFI AND HUANYIN CHEN
Generalized Drazin inverse is extensively studied in both matrix
theory and Banach algebra (see [3, 7, 8, 9] and [10]).
The goal of this paper is to introduce and study a new class of
generalized inverses in a ring. An element a ∈ R has generalized
Hirano inverse if there exists b ∈ R such that
b = bab, b ∈ comm2(a), a2 − ab ∈ Rqnil.
We shall prove that the preceding b is unique, if such element exists.
It will be denoted by ah, and called the generalized Hirano inverse
of a.
In Section 2 we prove that the generalized Hirano inverse of an
element is its generalized Drazin inverse. An element a ∈ R has
generalized Hirano inverse if and only if there exists p = p2 ∈
comm2(a) such that a2 − p ∈ Rqnil.
Recall that a commutative ring R is projective-free provided that
every finitely generated projective R-module is free. The class of
projective-free rings is very large. It includes local rings, principle
ideal domains, polynomial rings over every principal ideal domain,
etc. In Section 3, we completely determine when a 2×2 matrix over
projective-free rings has Hirano inverse. Let R be a projective-free
ring. We prove that A ∈ M2(R) has generalized Hirano inverse if
and only if det(A), tr(A2) ∈ J(R); or det2(A) ∈ 1 + J(R), tr(A2) ∈
2 + J(R); or det(A) ∈ J(R), tr(A2) ∈ 1 + J(R) and x2 − tr(A2)x +
det2(A) = 0 is solvable.
Let a, b ∈ R. Then ab has generalized Drazin inverse if and
only if ba has generalized Drazin inverse, and (ba)d = b((ab)d)2a.
This was known as Cline's formula for generalized Drazin inverses.
It plays an important role in revealing the relationship between
the generalized Drazin inverse of a sum of two elements and the
generalized Drazin inverse of a block triangular matrix of the form.
In Section 4, we establish Cline's formula for generalized Hirano
inverses. Let R be a ring, and let a, b, c ∈ R.
If aba = aca, we
prove that ac has generalized Hirano inverse if and only if ba has
generalized Hirano inverse, and that (ba)h = b((ac)h)2a.
Finally, we consider additive properties of such generalized in-
verses in a Banach algebra. Let A be a Banach algebra and a, b ∈ A.
GENERALIZED HIRANO INVERSES IN RINGS
3
If a, b ∈ A have generalized Hirano inverses and satisfy
a = abπ, bπbaπ = bπb, bπaπba = bπaπab,
we prove that a+b has generalized Hirano inverse. Here aπ = 1−aah
and bπ = 1 − bbh.
Throughout the paper, all rings are associative with an identity
and all Banach algebras are complex. We use N(R) to denote the
set of all nilpotent elements in R and J(R) to denote the Jacobson
radical of R. χ(A) = det(tIn − A) is the characteristic polynomial
of n × n matrix A. N stands for the set of all natural numbers.
2. Generalized Drazin inverses
The goal of this section is to explore the relations between gen-
eralized Drazin inverses and generalized Hirano inverses. We begin
with
Lemma 2.1. Let R be a ring and a ∈ Rqnil. If e2 = e ∈ comm(a),
then ae ∈ Rqnil.
Proof. Let x ∈ comm(ae). Then xae = aex, and so (exe)a =
ex(ae) = eaex = aex = a(exe), i.e., exe ∈ comm(a). Hence,
1 − a(exe) ∈ U(R), and so 1 − (ae)x ∈ U(R). This completes the
proof.
(cid:3)
Theorem 2.2. Let R be a ring and a ∈ R. If a has generalized
Hirano inverse, then it has generalized Drazin inverse.
Proof. Suppose that a has generalized Hirano inverse. Then we
have x ∈ R such that a2 − ax ∈ Rqnil, x ∈ comm2(a) and xax = x.
As a ∈ comm(a), we see that ax = xa. Hence, a2 − a2x2 = a2 −
a(xax) = a2 − ax ∈ Rqnil, and so
a2(1 − a2x2) = (a2 − a2x2)(1 − a2x2) ∈ Rqnil,
by Lemma 2.1. As 1 − a2x2 = 1 − ax ∈ R is an idempotent and
1 − ax ∈ comm(a2 − a2x2), it follows by Lemma 2.1 that
a(a − a2x) = a(a − a2(ax2)) = a2(1 − a2x2) ∈ Rqnil.
By using Lemma 2.1 again, (a − a2x)2 = a(a − a2x)(1 − ax) ∈ Rqnil.
Let y ∈ comm(a−a2x). Then y2 ∈ comm(a−a2x)2. Thus, 1 −(a−
4
MARJAN SHEIBANI ABDOLYOUSEFI AND HUANYIN CHEN
a2x)2y2 ∈ U(R). We infer that (1 − (a − a2x)y)(1 + (a − a2x)y) ∈
U(R). Therefore 1 − (a − a2x)y ∈ U(R), and so a − a2x ∈ Rqnil. As
ax = xa and xax = x, we conclude that a has generalized Drazin
inverse x.
(cid:3)
Corollary 2.3. Let R be a ring and a ∈ R. Then a has at most
one generalized Hirano inverse in R, and if the generalized Hirano
inverse of a exists, it is exactly the generalized Drazin inverse of a,
i.e., ah = ad.
Proof. In view of [8, Theorem 4.2], every element has at most one
generalized Drazin inverse, we obtain the result by Theorem 2.1.
(cid:3)
Lemma 2.4. Let R be a ring and a ∈ R. Then the following are
equivalent:
(1) a has generalized Hirano inverse.
(2) There exists b ∈ R such that
b = ba2b, b ∈ comm2(a), a2 − a2b ∈ Rqnil.
Proof. =⇒ By hypothesis, there exists b ∈ R such that
bab = b, b ∈ comm2(a), a2 − ab ∈ Rqnil.
As a ∈ comm(a), we see that ab = ba. Hence,
a2 − a2b2 = a2 − a(bab) = a2 − ab ∈ Rqnil, b2a2b2 = (bab)2 = b2.
Let xa = ax. Then x ∈ comm(a); hence, bx = xb. Thus, b2x = xb2,
and so b2 ∈ comm2(a), as required.
⇐= By hypothesis, there exists b ∈ R such that
ba2b = b, b ∈ comm2(a), a2 − a2b ∈ Rqnil.
Thus,
a2 − a(ab) = a2 − a2b ∈ Rqnil, (ab)a(ab) = a(ba2b) = ab.
Let xa = ax. Then xb = bx, and so (ab)x = a(bx) = a(xb) =
(ax)b = x(ab). Hence, ab ∈ comm2(a). Accordingly, a ∈ R has
generalized Hirano inverse.
(cid:3)
Theorem 2.5. Let R be a ring and a ∈ R. Then the following are
equivalent:
GENERALIZED HIRANO INVERSES IN RINGS
5
(1) a has generalized Hirano inverse.
(2) There exists p2 = p ∈ comm2(a) such that a2 − p ∈ Rqnil.
Proof. (1) ⇒ (2) By Lemma 2.4, there exists b ∈ R such that
b = ba2b, b ∈ comm2(a), a2 − a2b ∈ Rqnil.
Set p = a2b. Then p2 = p ∈ R. Let xa = ax. Then xb = bx,
and so xp = x(a2b) = (a2b)x. hence, p ∈ comm2(a). Furthermore,
a2 − p ∈ Rqnil, as required.
(2) ⇒ (1) Since a2 − p ∈ Rqnil, we see that a2 + 1 − p ∈ U(R).
Set e = 1 − p and b = (a2 + 1 − p)−1(1 − e). Let xa = ax. Then
px = xp and xa2 = a2x, and so x(a2 + 1 − p) = (a2 + 1 − p)x. This
implies that x(ab) = (ab)x, i.e., ab ∈ comm2(a). We check that
a2 − a(ab) = a2 − a2(a2 + 1 − p)−1(1 − e)
= a2 − (a2 + e)(a2 + 1 − p)−1(1 − e)
= a2 − p
∈ Rqnil.
Clearly, a2 + e ∈ U(R), and so
(ab)a(ab) = a3(a2 + 1 − p)−2(1 − e)
= a(a2 + e)(a2 + 1 − p)−2(1 − e)
= a(a2 + e)−1(1 − e)
= ab.
Therefore a has generalized Hirano inverse ab, as asserted.
(cid:3)
Corollary 2.6. Let R be a ring and a ∈ R. Then the following are
equivalent:
(1) a has generalized Hirano inverse.
(2) There exists b ∈ R such that
ab = (ab)2, b ∈ comm2(a), a2 − ab ∈ Rqnil.
Proof. (1) ⇒ (2) This is obvious.
(2) ⇒ (1) Set p = ab. If ax = xa, then bx = xb; hence, px =
(ab)x = axb = x(ab) = xp. Thus, p ∈ comm2(a). Since a2 − p ∈
Rqnil, we complete the proof by Theorem 2.5.
(cid:3)
6
MARJAN SHEIBANI ABDOLYOUSEFI AND HUANYIN CHEN
In light of Theorem 2.5 and [2, Theorem 3.3], we easily prove
that every element in a ring R is the sum of a tripotent e (i.e.,
e3 = e) and a nilpotent that commute if and only if every element
in R has generalized Drazin inverse and Rqnil = N(R). This also
gives the link between generalized Drazin inverses and tripotents in
a ring (see [6])
Proposition 2.7. Let R be a ring. If a ∈ R has generalized Hirano
inverse, then there exists a unique idempotent p ∈ R such that
pa = ap, a2 − p ∈ Rqnil.
Proof. In view of Theorem 2.5, there exists an idempotent p ∈ R
such that
p ∈ comm2(a), a2 − p ∈ Rqnil.
Suppose that there exists an idempotent q ∈ R such that
qa = aq, a2 − q ∈ Rqnil.
Let e = 1 − p and f = 1 − q. Then a2 − e, a2 − f ∈ U(R), ef = f e
and a2e, a2f ∈ Rqnil. Thus,
1 − (1 − e)f = 1 − (1 − e)(a2 − e)−1(a2 − e)f
= 1 − (1 − e)(a2 − e)−1a2f
= 1 − ca2f,
where c = (1 − e)(a2 − e)−1. As p ∈ comm2(a), we have c ∈
comm(a2f ).
It follows from a2f ∈ Rqnil that 1 − ca2f ∈ U(R).
Therefore
1 − (1 − e)f = 1 − (1 − e)2f 2 = (cid:0)1 − (1 − e)f(cid:1)(cid:0)1 + (1 − e)f(cid:1),
and so 1+(1−e)f = 1. This implies that f = ef . Likewise, e = f e.
Accordingly, e = ef = f e = f , and then p = 1 − e = 1 − f = q,
hence the result.
(cid:3)
If R is not only a ring, but a Banach algebra, the converse of
precious proposition can be guaranteed.
Corollary 2.8. Let A be a Banach algebra and a ∈ A. Then the
following are equivalent:
(1) a has generalized Hirano inverse.
GENERALIZED HIRANO INVERSES IN RINGS
7
(2) There exists an idempotent p ∈ A such that
pa = ap, a2 − p ∈ Rqnil.
Proof. (1) ⇒ (2) This is obvious by Proposition 2.7.
(2) ⇒ (1) As a2 − p ∈ Aqnil, a2 + 1 − p ∈ U(A). Set e = 1 − p and
b = (a2 + 1 − p)−1(1 − e). As in the proof of Theorem 2.5, there
exists b ∈ A such that
b = bab, ba = ab, a2 − ab ∈ Aqnil.
Similarly to Theorem 2.1, we see that a − a2b ∈ Aqnil.
In light
of [4, Theorem 7.5.3], a ∈ R has generalized Drazin inverse, and so
b ∈ comm2(a). This completes the proof.
(cid:3)
3. Matrices over projective-free rings
For any commutative ring R, we note that
M2(R)qnil = {A ∈ M2(R) A2 ∈ M2(cid:0)J(R)(cid:1)}(see [11, Example 4.3]).
The aim of this section is to completely determine when a 2 × 2
matrices over projective-free rings has generalized Hirano inverse.
Theorem 3.1. Let R be a projective-free ring, and let A ∈ M2(R).
If A has generalized Hirano inverse, then
(1) A2 ∈ M2(J(R)); or
(2) (I2 − A2)2 ∈ M2(J(R)); or
(3) χ(A2) has a root in J(R) and a root in 1 + J(R).
Proof. By virtue of Theorem 2.5, we may write A2 = E + W , where
E2 = E, W ∈ M2(R)qnil and E ∈ comm2(A). In light of [1, Propo-
sition 11.4.9], there exists U ∈ GL2(R) such that UEU −1 = 0, I2 or
(cid:18) 1 0
0 0 (cid:19). Then A2 ∈ M2(R)qnil, I2−A2 ∈ M2(R)qnil, or (cid:18) 1 0
0 0 (cid:19).
Case 1. A2 ∈ M2(R)qnil. Then A ∈ M2(R)qnil, and so A2 ∈
M2(J(R)).
Case 2. I2 − A2 ∈ M2(R)qnil. Then (I2 − A2)2 ∈ M2(J(R)).
Case 3. There exist U ∈ GL2(R) such that UEU −1 = (cid:18) 1 0
0 0 (cid:19).
Hence, UA2U −1 = UEU −1 + UW U −1. It follows from (UEU −1)
8
MARJAN SHEIBANI ABDOLYOUSEFI AND HUANYIN CHEN
(UW U −1) = (UW U −1)(UEU −1) that UW U −1 = (cid:18) λ 0
some λ, µ ∈ J(R). Hence, UA2U −1 = (cid:18) 1 + λ 0
0 µ (cid:19) for
µ (cid:19). Therefore
0
χ(A2) has a root µ in J(R) and a root 1 + λ in 1 + J(R), as
desired.
(cid:3)
Corollary 3.2. Let A ∈ M2(Z). Then A has generalized Hirano
inverse if and only if
(1) A2 = 0, or
(2) (I2 − A2)2 = 0, or
(3) A2 = A4.
Proof. ⇐= This is clear by Theorem 2.5.
=⇒ Since J(Z) = 0, by Theorem 3.1, we have
Case 1. A2 = 0.
Case 2. (I2 − A2)2 = 0.
Case 3. χ(A2) has a root 0 and a root 1. By Cayley-Hamilton
Theorem, A2(A2 − I2) = 0; hence, A2 = A4. In light of Theorem
2.5, A ∈ M2(Z) has generalized Hirano inverse.
Therefore we complete the proof.
(cid:3)
Example 3.3. Let Z(2) = { m
let A = (cid:18) 1 2
n m, n ∈ Z, (m, n) = 1, 2 ∤ n}, and
3 4 (cid:19). Then A ∈ M2(Z(2)) has no generalized Hirano
inverse.
Proof. Clearly, Z(2) is a commutative local ring with J(Z(2)) =
2Z(2). Since A2 = (cid:18) 7
15 22 (cid:19), we see that A2, (I2 − A2)2 6∈
10
M2(J(R)). In addition, tr(A2) = 29 and det(A2) = 4. Taking σ :
Z(2) → Q to be the natural map, and p(x) = x2 − 29x + 4 ∈ Z(2)[x].
Clearly, p∗(x) = x2 −29x+4 ∈ Q[x] is irreducible, and so x2 −29x+
4 = 0 is not solvable in Z(2). Hence, x2 − tr(A2)x + det(A2) = 0 is
not solvable in Z(2). In light of Theorem 3.1, A ∈ M2(Z(2)) has no
generalized Hirano inverse.
(cid:3)
Theorem 3.4. Let R be a projective-free ring. Then A ∈ M2(R)
has generalized Hirano inverse if and only if
GENERALIZED HIRANO INVERSES IN RINGS
9
(1) det(A), tr(A2) ∈ J(R); or
(2) det2(A) ∈ 1 + J(R), tr(A2) ∈ 2 + J(R); or
(3) det(A) ∈ J(R), tr(A2) ∈ 1 + J(R), and x2 − tr(A2)x +
det2(A) = 0 is solvable.
Proof. ⇐= Case 1. det(A), tr(A2) ∈ J(R). By Cayley-Hamilton
Theorem, we have A2 − tr(A2)A2 + det(A2)I2 = 0; hence, A2 =
tr(A2)A2 − det(A2)I2 ∈ J(M2(R)). Thus, A4 ∈ M2(J(R)), and
so A2 ∈ M2(R)qnil. Thus, A has generalized Hirano inverse, by
Theorem 2.5.
Case 2. det2(A) ∈ 1 + J(R), tr(A2) ∈ 2 + J(R). Then tr(I2 −
A2) = 2 − tr(A2) ∈ J(A) and det(I2 − A2) = 1 − tr(A2) + det(A2) ∈
J(R). As in Case 1, I2 − A2 ∈ M2(R)qnil. In light of Theorem 2.5,
A has generalized Hirano inverse.
Case 3. det(A) ∈ J(R), tr(A2) ∈ 1 + J(R) and x2 − tr(A2)x +
det(A2) = 0 is solvable. Write A2 = (cid:18) a b
c d (cid:19). Then a + d =
tr(A2) ∈ U(R). Hence, a or d are invertible. Say a ∈ U(R).
Suppose x2 −tr(A2)x+ det(A2) = 0 is solvable. Say x2
1 −tr(A2)x1 +
2 − tr(A2)x2 + det(A2) =
det(A2) = 0. Set x2 = tr(A2) − x1. Then x2
0. As x1x2 = det(A2) ∈ J(R), we see that x1 ∈ J(R), x2 ∈ U(R)
or x1 ∈ U(R), x2 ∈ J(R). Say x1 ∈ J(R), x2 ∈ U(R). Then x2
2 =
a − x1
tr2(A2)+(x1−2tr(A2))x1 ∈ 1+J(R). Let U = (cid:18)
(cid:19).
1 − 2ax1 − det(A2)) ∈ U(R)
x1 − a
It follows from det(U) = atr(A2) + (x2
that U ∈ GL2(R). One easily checks that
b
c
U −1A2U = (cid:18) x1
x2 (cid:19) ,
and so
U −1A2U = (cid:18) x1
x2 (cid:19) = (cid:18) 0
1 (cid:19) +(cid:18) x1
x2 − 1 (cid:19) .
If C = (cij) ∈ comm(cid:18) x1
fore C(cid:18) 0
1 (cid:19) = (cid:18) 0
x2 (cid:19), then c12(x1 − x2) = c21(x1 −
1 (cid:19) C. This shows that (cid:18) 0
1 (cid:19) ∈
x2) = 0. As x1 − x2 ∈ U(R), we get c12 = c21 = 0. There-
10
MARJAN SHEIBANI ABDOLYOUSEFI AND HUANYIN CHEN
comm2(cid:18) x1
x2 (cid:19). In light of Theorem 2.5, A ∈ M2(R) has gen-
eralized Hirano inverse.
=⇒ In view of Theorem 3.1, A2 ∈ M2(J(R)); or (I2 − A2)2 ∈
M2(J(R)); or χ(A2) has a root in J(R) and a root in 1 + J(R).
Case 1. A2 ∈ M2(J(R)). Hence, det(A), tr(A2) ∈ J(R).
Case 2. (I2−A2)2 ∈ M2(J(R)). Then I2 − A2 ∈ N(M2(R/J(R)).
In light of [?, ???], χ(I2 − A2) ≡ t2(cid:0)mod N(R/J(R)(cid:1). It follows
from N(R/J(R)) = 0 that χ(I2−A2)−t2 ∈ J(R)[t]. As χ(I2−A2) =
t2 − tr(I2 − A2)t + det(I2 − A2), we have tr(I2 − A2), det(I2 − A2) ∈
J(R). Hence, tr(A2) ∈ 2 + J(R). Since det(I2 − A2) = 1 − tr(A2) +
det2(A), we get det2(A) ∈ 1 + J(R), as desired.
Case 3. χ(A2) has a root α in J(R) and a root β in 1+J(R). Since
χ(A2) = x2 − tr(A2)x + det2(A), we see that det2(A) = αβ ∈ J(R),
and so det(A) ∈ J(R). Moreover, tr(A2) = α + β ∈ 1 + J(R).
Therefore we complete the proof.
(cid:3)
Corollary 3.5. Let R be a projective-free ring, and let A ∈ M2(R).
If χ(A2) = (t − α)(t − β), where α, β ∈ J(R)S(1 + J(R)), then A
has generalized Hirano inverse.
Proof. Since χ(A2) = (t − α)(t − β), det(A2) = αβ and tr(A2) =
α + β.
Case 1. α, β ∈ J(R). Then det(A), tr(A2) ∈ J(R).
Case 2. α, β ∈ 1 + J(R). Then det2(A) ∈ 1 + J(R), tr(A2) ∈
2 + J(R).
Case 3. α ∈ J(R), β ∈ 1 + J(R). Then det(A) ∈ J(R), tr(A2) ∈
1 + J(R).
Case 4. α ∈ 1 + J(R), β ∈ J(R). Then det(A) ∈ J(R), tr(A2) ∈
1 + J(R).
Therefore A has generalized Hirano inverse, by Theorem 3.4. (cid:3)
Corollary 3.6. Let R be a projective-free ring.
A ∈ M2(R) has Hrano inverse if and only if
If 1
2 ∈ R, then
(1) det(A), tr(A2) ∈ J(R); or
(2) det2(A) ∈ 1 + J(R), tr(A2) ∈ 2 + J(R); or
(3) det(A) ∈ J(R), tr(A2) ∈ 1+J(R) and there exists u ∈ U(R)
such that tr2(A2) − 4det2(A) = u2.
GENERALIZED HIRANO INVERSES IN RINGS
11
If x2 −
Proof. Suppose that det(A) ∈ J(R), tr(A2) ∈ 1 + J(R).
tr(A2)x + det2(A) = 0 is solvable, then this equation has a root x1.
Set x2 = tr(A2) − x1. Then det(A2) = x1(tr(A2) − x1) = x1x2 ∈
J(R). Set u = x1 − x2. Then u ∈ U(R). Therefore tr2(A2) −
4det2(A) = (x1 + x2)2 − 4x1x2 = (x1 − x2)2 = u2. Conversely,
assume that tr2(A2) − 4det2(A) = u2 for a u ∈ U(R). Then tr(A2)+u
is a root of x2 − tr(A2)x + det2(A) = 0. In light of Theorem 3.4,
we complete the proof.
(cid:3)
2
Example 3.7. Let Z(2) = { m
let A = (cid:18) 5 6
n m, n ∈ Z, (m, n) = 1, 2 ∤ n}, and
3 2 (cid:19) ∈ M2(Z(2)). Then A ∈ M2(Z(2)) has generalized
Hirano inverse.
Proof. Clearly, Z(2) is a commutative local ring with J(Z(2)) =
2Z(2). Clearly, A2 = (cid:18) 43 42
21 22 (cid:19). Then det(A2) = 64 ∈ J(Z(2)),
tr(A2) = 65 ∈ 1 + J(Z(2)). Thus, det(A) ∈ J(Z(2)), and that
x2 − tr(A2)x + det2(A) = 0 has a root 1. Therefore we are done by
Theorem 3.4.
(cid:3)
4. Cline's formula
In [9, Theorem 2.1], Liao et al. proved Cline's formula for gen-
eralized Drazin inverse. Lian and Zeng extended this result and
proved that if aba = aca then ac has generalized Drazin inverse
if and only if ba has generalized Drazin inverse (see [10, Theorem
2.3]). The aim of this section is to generalize Clines formula from
generalized Drazin inverses to generalized Hirano inverses.
Theorem 4.1. Let R be a ring, and let a, b, c ∈ R.
If aba =
aca, Then ac has generalized Hirano inverse if and only if ba has
generalized Hirano inverse. In this case, (ba)h = b((ac)h)2a.
Proof. Suppose that ac has generalized Hirano inverse and (ac)h =
d. Let e = bd2a and f ∈ comm(ba). Then
f e = f b(dacd)2a = f bacacd4a = babaf cd4a = b(acaf c)d4a.
12
MARJAN SHEIBANI ABDOLYOUSEFI AND HUANYIN CHEN
Also we have
ac(acaf c) = ababaf c = af babac = af bacac
= abaf cac = (acaf c)ac.
As d ∈ comm2(ac), we see that (acaf c)d = d(acaf c). Thus, we
conclude that
f e = b(acaf c)d4a = bd4(acaf c)a
= bd4abaf ca = bd4af baca
= bd4af baba = bd4ababaf
= bd4acacaf = bd2af = ef.
This implies that e ∈ comm2(ba). We have
e(ba)e = bd2a(ba)bd2a = bd2ababacd3a
= bd2(ac)3d3a = bd2a = e.
Let p = ac − acd2 then,
pac = acac − acd2ac = acac − dac = (ac)2 − acd
that is contained in Rqnil. Also we have
(ba)2 − bae = baba − babd2a = baba − babacd2da
= baca − bacacd2da = b(ac − acd2)a = bpa.
Since
abpa = ab(ac − acd2)a = ac(ac − acd2)a = acpa,
we deduce that
(pa)b(pa) = (pa)c(pa).
Then by [10, Lemma 2.2], bpa ∈ Rqnil. Hence e is generalized
Hirano inverse of ba and as the generalized Hirano inverse is unique
we deduce that e = bd2a = (ba)h.
(cid:3)
Corollary 4.2. Let R be a ring and a, b ∈ R. If ab has generalized
Hirano inverse, then so has ba, and (ba)h = b((ab)h)2a.
Proof. This follows directly from Theorem 4.1 as aba = aba.
(cid:3)
Corollary 4.3. Let R be a ring, and let a, b ∈ R.
generalized Hirano inverse, then so is (ba)k.
If (ab)k has
GENERALIZED HIRANO INVERSES IN RINGS
13
Proof. Let k = 1, then by Corollary 4.2, ab has generalized Hirano
inverse if and only if ba has generalized Hirano inverse. Now let
k ≥ 2, as (ab)k = a(b(ab)k−1), then by Corollary 4.2, (b(ab)k−1)a has
generalized Hirano inverse, so (ba)k has generalized Hirano inverse.
(cid:3)
Lemma 4.4. Let A be a Banach algebra, a, b ∈ A and ab = ba.
(1) If a, b ∈ Rqnil, then a + b ∈ Rqnil.
(2) If a or b ∈ Rqnil, then ab ∈ Rqnil.
Proof. Seebb [13, Lemma 2.10 and Lemma 2.11].
(cid:3)
Theorem 4.5. Let A be a Banach algebra , and let a, b ∈ A. If
a, b have generalized Hirano inverses and ab = ba. Then ab has
generalized Hirano inverse and
(ab)h = ahbh.
Proof. In order to prove that (ab)h = ahbh, we need to prove
ahbhabahbh = ahbh,
ahbh ∈ comm2(ab) and (ab)2 − abahbh ∈ Rqnil.
As ahaah = ah and bhbbh = bh, also a can be commuted with
ah, b, bh and b can be commuted with a, ah, bh, we have
ahbhabahbh = ahbh.
Let e ∈ comm2(ab), then ahbhe = ahebh = eahbh, as ah, bh ∈
com(ab). Note that
(ab)2 − abahbh
= a2(b2 − bbh) + b2(a2 − aah) − (a2 − aah)(b2 − bbh)
that is in Rqnil by Lemma 4.4, which implies that (ab)2 − abahbh ∈
Rqnil.
(cid:3)
Corollary 4.6. Let A be a Banach algebra. If a ∈ A has generalized
Hirano inverse, then an has generalized Hirano inverse and (an)h =
(ah)n for all n ∈ N.
Proof. It easily follows from Theorem 4.5 and using induction on
n.
(cid:3)
14
MARJAN SHEIBANI ABDOLYOUSEFI AND HUANYIN CHEN
We note that the converse of the previous corollary is not true.
Example 4.7. Let Z5 = Z/5Z be the ring of integers modulo 5.
Then −2 ∈ Z5 has no generalized Hirano inverse. If b is the gen-
eralized Hirano inverse of −2, then −2b2 = b which implies that
b = 0 or b = 2, in both cases (−2)2 − (−2)b is not quasinilpotent, a
contradiction. But (−2)2 = −1 ∈ Z has generalized Hirano inverse
1.
5. Additive properties
In this section, we are concern on additive properties of gen-
eralized Hirano inverses in a Banach algebra. Let p ∈ R be an
idempotent, and let x ∈ R. Then we write
x = pxp + px(1 − p) + (1 − p)xp + (1 − p)x(1 − p),
and induce a representation given by the matrix
x = (cid:18)
pxp
(1 − p)xp (1 − p)x(1 − p) (cid:19)p
px(1 − p)
,
and so we may regard such matrix as an element in R. The following
lemma is crucial.
Lemma 5.1. Let R be a Banach algebra, let a ∈ A and let
x = (cid:18) a c
0 b (cid:19)p
,
relative to p2 = p ∈ A. If a ∈ pAp and b ∈ (1 − p)A(1 − p) have
generalized Hirao inverses, then so is x in A.
Proof. In view of Corollary 2.3, a and b have generalized Drazin
inverses, and that a2 − aad, b2 − bbd ∈ Aqnil. Hence, In view of [13,
Lemma 2.1], we can find u ∈ pA(1 − p) such that
e = (cid:18) ad u
bd (cid:19)p
0
∈ comm2(x), e = e2x and x − x2e ∈ Aqnil.
Thus,
x2 − xe = (cid:18) a2 − aad
0
v
b2 − bbd (cid:19)p
GENERALIZED HIRANO INVERSES IN RINGS
15
for some v ∈ A. Clearly, v(a2 − aad) = 0 and v2 = 0. It follows
by [3, Lemma 2.1] that (a2 − aad) + v ∈ Aqnil. Furthermore, (b2 −
bbd)(cid:0)(a2−aad)+v(cid:1) = 0. By using [3, Lemma 2.1] again, (a2−aad)+
v + (b2 − bbd) ∈ Aqnil. Therefore x2 − xe ∈ Aqnil. Accordingly, x
has generalized Hirao inverses, as asserted.
(cid:3)
Lemma 5.2. Let A be a Banach algebra. Suppose that a ∈ Aqnil
and b ∈ A has generalized Hirano inverses. If
a = abπ, bπba = bπab,
then a + b has generalized Hirano inverse.
Proof. Case 1. b ∈ Aqnil. Then bπ = 1, and so it follows from
bπba = bπab that ab = ba. In light of Lemma 4.4, a + b ∈ Aqnil.
By using Lemma 4.4 again, (a + b)2 ∈ Aqnil. Therefore a + b has
generalized Hirano inverse by Theorem 2.5.
Case 2. b 6∈ Aqnil.
In view of Theorem 2.1, b has generalized
Drazin inverse, and so we have
b = (cid:18) b1
0
0
b2 (cid:19)p
, a = (cid:18) a11 a1
a21 a2 (cid:19)p
,
where b1 ∈ U(pAp), b2 ∈ ((1−p)A(1−p))qnil ⊆ Aqnil. From a = abπ,
we see that a11 = a21 = 0, and so
a + b = (cid:18) b1
a1
0 a2 + b2 (cid:19)p
.
It follows from bπba = bπab that a2b2 = b2a2. By Lemma 4.4,
a2 + b2 ∈ Aqnil, and then (a2 + b2)2 ∈ Aqnil. In light of Theorem
2.5, a2 + b2 has generalized Hirano inverse. Since b has generalized
Hirano inverse in A, one easily checks that b1 = pbp has generalized
inverse. According to Lemma 5.1, a + b has generalized Hirano
inverse, thus completing the proof.
(cid:3)
Theorem 5.3. Let A be a Banach algebra. If a, b ∈ A have gener-
alized Hirano inverses and satisfy
a = abπ, bπbaπ = bπb, bπaπba = bπaπab,
16
MARJAN SHEIBANI ABDOLYOUSEFI AND HUANYIN CHEN
then a + b has generalized Hirano inverse and
∞
(bh)n+2a(a + b)n(cid:1)aπ
(bh)n+2a(a + b)n(ah)k+2b(a + b)k+1
∞
∞
(a + b)h = (cid:0)bh +
Pn=0
Pn=0
Pk=0
Pn=0
−
+
∞
(bh)n+2a(a + b)nahb −
bha(ah)n+2b(a + b)n.
∞
Pn=0
Proof. Case 1. b ∈ Aqnil. Then bπ = 1, and so we obtain the result
by applying Lemma 5.2.
Case 2. b 6∈ Aqnil. We have
b = (cid:18) b1
0
0
b2 (cid:19)p
, a = (cid:18) a11 a1
a21 a2 (cid:19)p
,
where b1 ∈ U(pAp), b2 ∈ ((1 − p)A(1 − p))qnil ⊆ Aqnil. It follows
from a = abπ that a11 = a21 = 0, and so
a1
a + b = (cid:18) b1
0 a2 + b2 (cid:19)p
.
Since bπbaπ = bπb, bπaπba = bπaπab, we see that aπ
2 b2a2.
Now, it follows that a2 +b2 ∈ (1−p)A(1−p) has generalized Hirano
inverse. As b1 = pbp has generalized Hirano inverse, it follows by
Lemma 5.1 that a + b has generalized Hirano inverse. According to
Corollary 2.2, we see that (a + b)h = (a + b)d, ah = ad and bh = bd.
Therefore we complete the proof, by [3, Theorem 2.2].
(cid:3)
2 a2b2 = aπ
Corollary 5.4. Let A be a Banach algebra and a, b ∈ A. If a, b ∈ A
have generalized Hirano inverses and satisfy
ab = ba, a = abπ, bπ = baπ = bπb,
then a + b has generalized Hirano inverse and (a + b)h = bh.
Proof. Since a = abπ, we see that abh = 0. Thus, (bh)n+2a(a +
b)n = a(bh)n+2(a+b)n = 0. Therefore the result follows by Theorem
5.3.
(cid:3)
Corollary 5.5. Let A be a Banach algebra and a, b ∈ A. If a, b
have generalized Hirano inverses and ab = ba = 0, then a + b has
generalized Hirano inverse and (a + b)h = ah + bh.
GENERALIZED HIRANO INVERSES IN RINGS
Proof. This is obvious by Theorem 5.3.
17
(cid:3)
References
[1] H. Chen, Rings Related Stable Range Conditions, Series in Algebra 11,
World Scientific, Hackensack, NJ, 2011.
[2] H. Chen and M. Sheibani, Strongly 2-nil-clean rings, J. Algebra Appl.,
16(2017) 1750178 (12 pages), DOI: 10.1142/S021949881750178X.
[3] D.S. Cvetkovic-Ilica; D.S. Djordjevic and Y. Wei, Additive results for
the generalized Drazin inverse in a Banach algebra, Linear Algebra Appl.
418(2006), 53 -- 61.
[4] R. Hart, Invertiblity and Singularity for Bounded Linear Operators, Cam-
bridge University Press, New York, NJ, 1988.
[5] R.E. Harte, On quasinilpotents in rings, Panam. Math. J., 1(1991), 10 -- 16.
[6] Y. Hirano and H. Tominaga, Rings in which every element is a sum of two
idempotents, Bull. Austral. Math. Soc., 37(1988), 161 -- 164.
[7] J.J. Koliha, A generalized Drazin inverse, Glasgow Math. J., 38(1996),
367 -- 381.
[8] J.J. Koliha and P. Patrick, Elements of rings with equal spectral idempo-
tents, J. Austral. Math. Soc., 72(2002), 137 -- 152.
[9] Y. Liao; J. Chen and J. Cui, Cline's formula for the generalized Drazin
inverse, Bull. Malays. Math. Sci. Soc., 37(2014), 37 -- 42.
[10] H. Lian and Q. Zeng, An extension of Cline's formula for a generalized
Drazin inverse, Turk. J. Math., 40(2016), 161 -- 165.
[11] Z. Wang and J. Chen, Pseudo Drazin inverses in associative rings and
Banach algebras, Linear Algebra Appl., 437(2012), 1332 -- 1345.
[12] K. Yang; X.C. Fang, Common properties of the operator products in local
spectral theorey, Acta Math. Sinica, English Series, 31(2015), 1715 -- 1724.
[13] H. Zhu; D. Mosic and J. Chen, Generalized Drazin invertibility of the
product and sum of two elements in a Banach algebra and its applications,
Turk. J. Math., 41(2017), 548 -- 563.
Women's University of Semnan (Farzanegan), Semnan, Iran
E-mail address: <[email protected]>
Department of Mathematics, Hangzhou Normal University, Hang
-zhou, China
E-mail address: <[email protected]>
|
1701.00026 | 1 | 1701 | 2016-12-30T22:30:51 | Almost projective and almost injective modules | [
"math.RA"
] | We describe rings over which every right module is almost injective. We give a description of rings over which every simple module is a almost projective. | math.RA | math |
ALMOST PROJECTIVE AND ALMOST INJECTIVE MODULES
ABYZOV A. N.
Abstract. We describe rings over which every right module is almost injec-
tive. We give a description of rings over which every simple module is a almost
projective.
Let M, N be right R-modules. A module M is called almost N- injective, if for
any submodule N ′ of N and any homomorphism f : N ′ → M, either there exists a
homomorphism g : N → M such that f = gι or there exists a nonzero idempotent
π ∈ EndR(N) and a homomorphism h : M → π(N) such that hf = πι, where
ι : N ′ → N is the natural embedding. A module M is called almost injective if it
is almost N-injective for every right R-module N. Dually, we define the concept of
almost projective modules. A module M is called almost N-projective, if for any
natural homomorphism g : N → N/K and any homomorphism f : M → N/K,
either there exists a homomorphism h : M → N such that f = gh or there exists
a non-zero direct summand N ′ of N and a homomorphism h′ : N ′ → M such
that gι = f h′, where ι : N ′ → N is the natural embedding. A module M is called
almost projective if it is almost N-projective for every right R-module N.
The concepts of almost injective module and almost projective module were
studied in the works [1]-[7] by Harada and his colleagues. Note that, in [7] an
almost projective right R-module is defined as a module which is almost N-
projective to every finitely generated right R-module N. In recent years, almost
injective modules were considered in [8]-[12]. The problem of the description of
the rings over which all modules are almost injective was studied in [10].
In
some special cases, this problem was solved in [10]. In particular, in the case of
semiperfect rings. In this article, we study the structure of the rings over which
every module is almost injective, in general. We also give the characterization of
2010 Mathematics Subject Classification. 16D40, 16S50, 16S90.
Key words and phrases. almost projective, almost injective modules, semiartinian rings, V-
rings.
1
2
ABYZOV A. N.
the module M such that every simple module is almost projective (respectively,
almost injective) in the category σ(M).
Let M, N be right R-modules. We denote by σ(M) the full subcategory of
Mod-R whose objects are all R-modules subgenerated by M. If N ∈ σ(M) then
the injective hull of the module N in σ(M) will be denoted by EM (N). The
Jacobson radical of the module M is denoted by J(M).
The Loewy series of a module M is the ascending chain of submodules
0 = Soc0(M) ⊂ Soc1(M) = Soc(M) ⊂ . . . ⊂ Socα(M) ⊂ Socα+1(M) ⊂ . . .,
where Socα+1(M)/ Socα(M) = Soc(M/ Socα(M)) for all ordinal numbers α and
Socβ(M) for a limit ordinal number α. Denote by L(M) the sub-
Socα(M) = Sβ<α
module Socξ(M), where ξ is the smallest ordinal such that Socξ(M) = Socξ+1(M).
The module M is semiartinian if and only if M = L(M). In this case ξ is called
the Loewy length of module M and is denoted by Loewy(M). The ring R is called
right semiartinian if the module RR is semiartinian.
The present paper uses standard concepts and notations of ring theory (see,
for example [13]-[15] ).
1. Almost projective modules
A module M is called an I0-module if every its nonsmall submodule contains
nonzero direct summand of the module M.
Theorem 1.1. For a module M, the following assertions are equivalent:
1) Every simple module in the category σ(M) is almost projective.
2) Every module in the category σ(M) is either a semisimple module or con-
tains a nonzero M-injective submodule.
3) Every module in the category σ(M) is an I0-module.
Proof. 1)⇒2) Let xR ∈ σ(M) be a non-semisimple cyclic module. Then the
module xR contains an essential maximal submodule N. Let f : EM (xR) →
EM (xR)/N be the natural homomorphism and ι : xR/N → EM (xR)/N be the
embedding. Assume that there exists a homomorphism g : xR/N → EM (xR)
such that f g = ι. Since g(xR/N) ⊂ f −1(xR/N) = xR and N is an essential
submodule of xR, then g(xR/N) ⊂ N. Consequently f g = 0, which is impossible.
Since the module xR/N is almost projective, for some nonzero direct summand
N ′ of EM (xR) and homomorphism h : N ′ → xR/N we get ιh = f ι′, where
ι′ : N ′ → EM (xR) is the embedding. Consequently f (N ′) ⊂ xR/N, i.e. N ′ ⊂
f −1(xR/N) = xR.
2)⇒3) The implication follows from [16, Theorem 3.4].
ALMOST PROJECTIVE AND ALMOST INJECTIVE MODULES
3
3)⇒1) Let S be a simple right R-module, f : A → B be an epimorphism right
R-modules and g : S → B be a homomorphism. Without loss of generality, as-
sume that g 6= 0. If Ker(f ) is not an essential submodule of f −1(g(S)), then there
exists a simple submodule S′ of f −1(g(S)) such that f (S′) = g(S). In this case, ob-
viously, there is a homomorphism h : S → A such that f h = g. Assume Ker(f ) is
an essential submodule of f −1(g(S)). Then f −1(g(S)) is a non-semisimple module
and by [16, Theorem 3.4], f −1(g(S)) contains a nonzero injective submodule A′.
There exists a homomorphism g′ : g(S) → S such that gg′(s) = s for all s ∈ g(S).
Then g(g′fA′) = f ι, where ι : A′ → A is the embedding and fA′ : A′ → g(S) is
the restriction of the homomorphism f to A′.
(cid:3)
Corollary 1.1. Every right R-module is an I0-module if and only if every simple
right R-module is almost projective.
A right R-module M is called a V -module (or cosemisimple) if every proper
submodule of M is an intersection of maximal submodules of M. A ring R is
called a right V -ring if RR is a V -module. It is known that a right R-module M
is a V -module if and only if every simple right R-module is M-injective. A ring
R is called a right SV -ring if R is a right semiartinian right V -ring.
Theorem 1.2. For a regular ring R, the following assertions are equivalent:
1) Every right R-module is an I0-module.
2) R is a right SV -ring.
3) Every right R-module is almost projective.
4) Every simple right R-module is almost projective.
Proof. The equivalence 1)⇔2) follows from [16, theorem 3.7]. The implication
3)⇒4) is obvious. The implication 4)⇒1) follows from Theorem 1.1.
2)⇒3) Let S be a simple right R-module. We claim that the module S is almost
projective. Let f : A → B be an epimorphism right R-modules and g : S → B
be a homomorphism. Without loss of generality, assume that Ker(f ) 6= 0. Then
Ker(f ) contains a simple injective submodule S′ and for the homomorphism h =
0 ∈ Hom(S′, S) we get f ι = gh, where ι : S → A is the natural embedding. (cid:3)
A ring R is called a I-finite (or orthogonally finite) if it does not contain an
infinite set of orthogonal nonzero idempotents.
Theorem 1.3. For a I-finite ring R, the following assertions are equivalent:
1) Every right R-module is almost projective.
2) Every simple right R-module is almost projective.
3) R is an artinian serial ring and J 2(R) = 0.
4
ABYZOV A. N.
Proof. The implicatio 1)⇒2) is obvious.
2)⇒3) By Theorem 1.1 and [14, 13.58], R is a semiperfect ring. Then by [16,
Theorem 3.2], R is an artinian serial ring and J 2(R) = 0.
3)⇒1) Let M be a right R-module. We claim that the module M is almost
projective. Let f : A → B be an epimorphism of right R-modules and g : M → B
be a homomorphism. If f −1(g(M)) is a semisimple module, then it is obvious
that there is a homomorphism h such that g = f h. Assume f −1(g(M)) is a
non-semisimple module. Then the module f −1(g(M)) contains an injective and
projective local submodule L of length two. Since L is a projective module, then
there is a homomorphism h′ : L → g−1(f (L)) such that f ι = gg−1(f (L))h′, where
ι : L → A is the natural embedding.
(cid:3)
2. Almost V -modules
A right R-module M is called an almost V -module if every simple right R-
module is almost N-injective for every module N ∈ σ(M). A ring R is called
a right almost V -ring if every simple right R-module is almost injective. Right
almost V -rings have been studied in [11].
Lemma 2.1. For a module M, the following assertions are equivalent:
1) M is not a V -module.
2) There exists a submodule N of the module M such that the factor mod-
ule M/N is an uniform, Soc(M/N) is a simple module and M/N 6=
Soc(M/N).
Proof. The implicatio 2)⇒1) is obvious.
1)⇒2) Since M is not a V -module, there is a submodule M0 such that J(M/M0) 6=
0. Without loss of generality, assume that J(M/M0) contains a simple submod-
ule S. Let S′ be a complement of submodule S in M/M0. Then (M/M0)/S′
is an uniform module, Soc((M/M0)/S′) is a simple module and (M/M0)/S′
6=
Soc((M/M0)/S′).
(cid:3)
Proposition 2.1. Let M be an almost V -module. Then:
1) The Jacobson radical J(N) of every module N ∈ σ(M) is semisimple.
2) The factor module N/J(N) of every module N ∈ σ(M) is a V -module.
3) The injective hull EM (S) of every simple module S ∈ σ(M) is either a
simple module or a local M-projective module of length two.
Proof. 1) Assume that in the category σ(M) there exists a module whose Ja-
cobson radical is not semisimple. Then there exists a module N ∈ σ(M) and a
ALMOST PROJECTIVE AND ALMOST INJECTIVE MODULES
5
non-zero element x ∈ J(N) such that the module xR contains an essential max-
imal submodule A. Let B be a complement of submodule A in N. Consider the
homomorphism f : xR ⊕ B → xR/A is defined by f (xr + b) = xr + A, where
r ∈ R, b ∈ B. Assume that there exists a homomorphism g : N → xR/A such
that gι = f, where ι : xR ⊕ B → E is the natural embedding. Since x ∈ J(N),
gι(x) = 0. On the other hand, f (x) 6= 0. This is a contradiction. If there is a
nonzero idempotent π ∈ EndR(N) and a homomorphism h : xR/A → π(N) such
that πι = hf, then hf (π(N) ∩ (A ⊕ B)) = 0 and πι(π(N) ∩ (A ⊕ B)) 6= 0 for a
nonzero submodule π(N) ∩ (A ⊕ B), that is impossible. Thus a Jacobson radical
J(N) of every module N ∈ σ(M) is semisimple.
2) Let N ∈ σ(M) be a module and S ∈ σ(M) be a simple module, N0 be a
submodule of N ′ = N/J(N) and f : N0 → S be a homomorphism. We show
that there exists a homomorphism g such that f = gι, where ι : N0 → N ′ is the
natural embedding. Without loss of generality, assume that N0 is essential in N ′
and f 6= 0. Assume Ker(f ) is an essential submodule of N0. If there exists a non-
zero idempotent π ∈ EndR(N ′) and a homomorphism h : S → π(N ′) such that
πι = hf, then hf (π(N ′) ∩ Ker(f )) = 0 and πι(π(N ′) ∩ Ker(f )) 6= 0 for a nonzero
submodule π(N ′)∩Ker(f ), that is impossible. Thus there exists a homomorphism
g such that f = gι. Assume Ker(f ) is not an essential submodule of N0. Then
there exists a simple module S′ such that N0 = Ker(f ) ⊕ S′. Assume that there
exists a non-zero idempotent π ∈ EndR(N ′) and a homomorphism h : S → π(N ′)
such that πι = hf. Since Ker(f ) ⊕ S′ is essential in N ′, Ker(f ) ⊂ (1 − π)N ′
and (1 − π)N ′ ⊕ π(S′) = (1 − π)N ′ ⊕ S′, then π(S′) is essential in π(N ′). Since
J(N ′) = 0, we get π(S′) = π(N ′), and consequently N ′ = (1 − π)N ′ ⊕ S′. Then
there exists a g : (1−π)N ′ ⊕S′ → S homomorphism is defined by g(n+s) = f (s),
where n ∈ (1 − π)N ′, s ∈ S′ such that f = gι. Hence N ′ is a V -module.
3) Let S ∈ σ(M) be a simple module and EM (S) 6= S. By 2), J(EM (S)) = S.
Let A1, A2 be maximal submodules of EM (S). From the proof of [17, 13.1(a)], we
see that EndR(A1), EndR(A2) are local rings. Assume that A = Ai ⊕ Aj is a CS-
module, where i, j ∈ {1, 2}. Let B is a closed submodule of A and A 6= B. Then
B is complement of some simple submodule S′ in A. Consider the homomorphism
f : S′ ⊕ B → S′ is defined by the formula f (s + b) = s, where s ∈ S′, b ∈ B. Since
S′ ∈ J(A) and S′ is an almost A-injective module, there is a non-zero idempotent
π ∈ EndR(A) and a homomorphism g ∈ HomR(S, π(A)) such that gf = πι,
where ι : S′ ⊕ B → A is the natural embedding. It's clear that B ⊂ (1 − π)A and
S ∩(1 −π)A = 0. Consequently B = (1 −π)A. Thus A is a CS-module. From [17,
7.3(ii)] and the fact that every monomorphism φ : Ai → Aj is an isomorphism we
deduce that Ai is an Aj-injective module. If A1 6= A2 then by [15, 16.2], A1 is an
6
ABYZOV A. N.
A1 + A2-injective, which is impossible. Thus the module EM (S) has an unique
maximal submodule, and consequently EM (S) is a local module of length two. We
claim that EM (S) is projective in the category σ(M). Let N be a submodule of
EM (S)⊕M such that N +M = EM (S)⊕M and π : EM (S)⊕M → EM (S) be the
natural projection. Assume that J(N) ⊂ N ∩ M. Since N/J(N) is a V -module,
π(N) = EM (S) is a V -module, which is impossible. Thus there exists a simple
submodule S′ of J(N) such that S′∩M = 0. Let A be a complement of submodule
S′ in N such that M ∩ N ⊂ A. Consider the homomorphism f : S′ ⊕ A → S′
is defined by f (s + a) = s, where s ∈ S′, a ∈ A. Since S′ ⊂ J(N) and S′ is an
almost N-injective module, there is a non-zero idempotent π′ ∈ EndR(N) and a
homomorphism g ∈ HomR(S′, π′(N)) such that gf = π′ι, where ι : S′ ⊕ A → N
is the natural embedding. Since A ⊂ (1 − π′)(N), S′ ⊂ J(N) and A ⊕ S′ is
an essential submodule of N, we deduce that π′(S) is essential in π′(N) and
π(S′) 6= π′(N). Since
π′(N) ∩ A = π′(N) ∩ N ∩ M = π′(N) ∩ M = 0
and lg(EM (S)) = lg(π′(N)) = 2, we have π(π′(N)) = EM (S). Then π′(N) ⊕M =
EM (S) ⊕ M. By [15, 41.14], the module EM (S) is projective in the category
σ(M).
(cid:3)
Theorem 2.1. For a module M, the following assertions are equivalent:
1) M is an almost V -module.
2) Every module in the category σ(M) is either a V -module or contains a
nonzero direct summand which is a projective object in the category σ(M).
3) There exist an independent set of local submodules {Ai}i∈I of the module
M such that:
a) Ai is both an M-injective and an M-projective module of length two
for all i ∈ I;
b) J(M) = ⊕i∈IJ(Ai);
c) M/J(M) is a V -module.
Proof. 1)⇒2) Let N be a module in the category σ(M) which is not a V -module.
Then by Lemma 2.1 and Proposition 2.1, there is a submodule N ′ of N such
that the factor module N/N ′ is nonzero and projective in the category σ(M).
Consequently the natural epimorphism f : N → N/N ′ splits and the module N
contains a nonzero direct summand which is a projective in the category σ(M).
2)⇒1) Let M be a right R-module and S be a simple right R-module. We
claim that S is an almost M-injective module. Let M0 be a submodule of M and
f : M0 → S be a homomorphism. Without loss of generality, assume that f 6= 0,
ALMOST PROJECTIVE AND ALMOST INJECTIVE MODULES
7
M0 is an essential submodule of M and EM (S) 6= S. There is a homomorphism
g : M → EM (S) such that gι = ι′f, where ι : M0 → M and ι′ : S → EM (S) the
natural embeddings. Assume that S 6= g(M). Then by the condition 2), g(M) is
a projective module. Consequently M = Ker(g) ⊕ M ′. Since M0 is an essential
submodule of M, then M0 ∩ M ′ is a simple module and fM0∩M ′ : M0 ∩ M ′ → S
is an isomorphism. Then M0 = (M0 ∩ M ′) ⊕ Ker(f ). Let π : Ker(g) ⊕ M ′ → M ′
be the natural projection. Then πι = f −1
M0∩M ′f.
1)⇒3) By Zorn's Lemma, there is a maximal independent set of submodules
{Ai}i∈I of the module M such that Ai is a local module of length two for all
i ∈ I. According to Proposition 2.1, M/J(M) is a V -module and Ai is both an
M-injective and an M-projective module for all i ∈ I. Assume that J(M) 6=
⊕i∈IJ(Ai). Then by the condition 1), there is a simple submodule S of M such
that S ⊂ J(M) and S ∩ ⊕i∈I J(Ai) = 0. Let S′ be a complement of submodule S
in M such that it contains ⊕i∈I J(Ai). Then M/S′ is not a simple module, which
is an essential extension of the simple module (S + S′)/S′. By Proposition 2.1,
M/S′ is an M-projective module of length two. Consequently, there is a local
submodule of length two L of M such that M = L ⊕ S′. This contradicts with
the choice of the set {Ai}i∈I. Thus J(M) = ⊕i∈IJ(Ai).
3)⇒2) Let S ∈ σ(M) be a simple module and EM (S) 6= S. By [15, 16.3], there
exists an epimorphism f : ⊕i∈I ′Mi → EM (S), where Mi = M for all i ∈ I ′. Since
EM (S) is not a V -module, by [15, 23.4], f ǫi(J(M)) 6= 0 for some i ∈ I ′, where
ǫi : Mi → ⊕i∈I ′Mi is a natural embedding. Then, by the conditions a) and b)
of 3), EM (S) ∼= Ai for some i ∈ I. Thus every essential extension of a simple
module in the category σ(M) is either a simple or a local M-projective module
of length two. Then the implication follows directly from Lemma 2.1.
(cid:3)
Corollary 2.1. For a ring R, the following assertions are equivalent:
1) R is a right almost V -ring.
2) Each right R-module is either a V -module or contains a nonzero direct
summand which is a projective module.
3) There exist a set of orthogonal idempotents {ei}i∈I of the ring R such that:
a) eiR is a local injective right R-module of length two for every i ∈ I;
b) J(P ) = ⊕i∈IJ(eiR);
c) R/J(R) is a right V -ring.
Theorem 2.2. For a right noetherian ring R, the following assertions are equiv-
alent:
1) Every right R-module is a direct sum of an injective module and a V -
module.
8
ABYZOV A. N.
2) Every right R-moduleis a direct sum of a projective module and a V -
module.
3) R is a right almost V -ring.
Proof. 3)⇒1), 2) By Zorn's Lemma, there is a maximal independent set of local
submodules of length two {Li}i∈I of the module M. Since R is a right noetherian
ring, by [13, 6.5.1], there exists a submodule N of M such that M = ⊕i∈ILi ⊕ N.
By Proposition 2.1 3), ⊕i∈I Li is both injective and projective. We claim that
N is a V -module. Assume that N is not a V -module. Then by the Proposition
2.1 3) and Lemma 2.1, there exists a factor module N/N0 of N which is a local
projective module of length two. Consequently, the module N/N0 is isomorphic
to a submodule of N, which contradicts the choice of the set {Li}i∈I. Thus N is
a V -module.
2)⇒3) Let S be a right simple module. Assume that E(S) 6= S. By the
condition 2), E(S) is a projective module and by [13, 7.2.8], EndR(E(S)) is a
local ring. Then by [13, 11.4.1], E(S) is a local module.
If J(E(S)) is not a
simple module, then by the condition 2), the module E(S)/S is projective, and
consequently S is a direct summand of E(S), which is impossible. Thus the
injective hull of a every simple right R-module is either a simple or a projective
module of length two. Consequently R is a right almost V -ring by [11, Theorem
3.1].
1)⇒3) Since RR is a noetherian module then by the condition 1), RR = M ⊕N,
where M is a finite direct sum of uniform injective modules and N is a V -module.
By [13, 7.2.8, 11.4.1], M = L1 ⊕ . . . ⊕ Ln, where Li is a local module for every
1 ≤ i ≤ n. Assume that J(Li0) is nonzero and is not a simple module for some
i0. Then there is a non-zero element r ∈ J(Li0) such that rR 6= J(Li0). Let T be
maximal submodule of rR. By the condition 1), the injective hull of every simple
right R-module is either a simple module or a module of length two. Then the
local module Li0/T is not an injective module and it is not a V -module, which
contradicts to condition 1). From these considerations, it follows that there exists
a family of orthogonal idempotents e1, . . . en of ring R satisfying the condition a)
and b) of Corollary 2.2, and RR/J(R) is the direct sum of a semisimple module
and a V -module. By [15, 23.4], R/J(R) is a right V -ring. Then, by Corollary 2,
R is an almost right V -ring.
(cid:3)
Theorem 2.3. For a regular ring R, the following assertions are equivalent:
1) R is a right V -ring.
2) Every right R-module is a direct sum of an injective module and a V -
module.
ALMOST PROJECTIVE AND ALMOST INJECTIVE MODULES
9
3) Every right R-module is a direct sum of a projective module and a V -
module.
4) R is a right almost V -ring.
Proof. The implications 1)⇒2), 1)⇒3), 1)⇒4) are obvious.
i=1Li, then J(⊕∞
2)⇒1) Assume that the ring R is not a right V -ring. Then E(S) 6= S for some
simple right R-module S. By the condition 2) we have ⊕∞
i=1Li = M ⊕ N, where
∼= E(S) for every i, M is an injective module and N is a V -module. Since
Li
J(⊕∞
i=1Li) is essential in ⊕∞
i=1Li) ∩ N = J(N) is essential in N,
and consequently N = 0. Let I = {r ∈ R E(S)r = 0}. We can conside the
module ⊕∞
i=1Li as a right module over the ring R/I. Assume that R/I is not a
semisimple artinian ring. Then the ring R/I contains a countable set of non-zero
orthogonal idempotents {ei}∞
i=1. For every i ∈ N, there is an element li ∈ Li
such that liei 6= 0. Since the right R/I-module ⊕∞
i=1Li is injective, there exists
a homomorphism f : R/IR/I → ⊕∞
i=1Li, such that f (ei) = liei for all i. Since
f (R/IR/I) ⊂ ⊕n
i=1Li for some n ∈ N, we obtain a contradiction with the fact
that liei 6= 0 for all i ∈ N. Thus R/I is a semisimple artinian ring. Consequently
E(S) = S. This contradiction shows that R is a right V -ring.
3)⇒1) Assume that the ring R is not a right V -ring. Then by Lemma 2.1, there
exists a right ideal I of R such that the right R-module R/I is an uniform, is not a
simple module and Soc(R/IR) is a simple module. Then, by the condition 3), the
module R/I is projective, and consequently R/IR is isomorphic to a submodule
of RR, which is impossible. This contradiction shows that R is a right V -ring.
The implication 4)⇒1) follows directly from Corollary 2.1.
(cid:3)
3. Rings Over Which Every Module Is Almost Injective
Let M be a right R-module. Denote by SI(M) the sum of all simple injective
submodules of the module M. Clearly, SI(RR) is ideal of ring R.
Lemma 3.1. Let R be a ring with the following properties:
a) in the ring R there exists a finite set of orthogonal idempotents {ei}i∈I such
that eiR is local injective right R-module of length two, for each i ∈ I and
J(R) = ⊕i∈IJ(eiR);
b) R/J(R) is a right SV -ring and Loewy(RR) ≤ 2;
c) R/SI(RR) is a right artinian ring.
Then we have the following statement:
1) the injective hull of every simple right R-module is either a simple module
or a local projective module of length two;
10
ABYZOV A. N.
2) every right R-module is a direct sum of a injective module and a V -module;
3) every right R-module is a direct sum of a projective module and a V -
module;
4) if S a simple submodule of the right R-module N, S ⊂ J(N) and S∩N ′ = 0
for some submodule N ′ of N, then there are submodules L, N ′′ of N such
that L is a local module of length two, S ⊂ L, N ′ ⊂ N ′′ and N = N ′′ ⊕ L.
Proof. 1) Let S be a simple right R-module and E(S) 6= S. Since R/J(R) is a
right V -ring and J(R) is a semisimple right R-module, then E(S)S′ 6= 0 for some
simple submodule of S′ of right R-module J(R)R. From condition a) it follows
that S′ is essential in some injective local submodule of the module ⊕i∈IeiR.
Therefore, E(S) ∼= ei0R for some i0 ∈ I. Thus injective hull of every simple right
R- module is either a simple module or a local projective module of length two.
2), 3) Let M be a right R-module. By Lemma of Zorn there is a maximal
independent set of submodules of {Li}i∈I of a module M such that Li is a local
injective module of length two, for each i ∈ I. Clearly, E(⊕i∈ILi)SI(R) = 0.
Then from the condition c) it follows that E(⊕i∈I Li) = ⊕i∈I Li. Therefore M =
⊕i∈ILi ⊕ N for some submodule N of a module M. It is clear that module ⊕i∈ILi
is injective and projective. If N is V -module, then from Lemma 2.1 and condition
1) follows that for some submodule N0 of the module N factor module N/N0 is
a local projective module of length two. Therefore N = N0 ⊕ L where L is a
injective local module of length two, which impossible. Thus N is a V -module.
4) From conditions 1) and 2), it follows that S ⊂ L where L is a local injective
submodule of a module N of length two. Let L′ is a complement of L in N which
contains the submodule N ′. Then (S + L′)/L′ is a essential submodule of N/L′
and N/L′ 6= (S + L′)/L′. From condition 1), it follows that N/L′ is a local module
of length two. Therefore, the natural homomorphism f : N → N/L′ induces an
isomorphism fL : L → N/L′. Then N = L ⊕ L′.
(cid:3)
(cid:3)
Lemma 3.2. Let M be a right R-module and N be a injective submodule of M.
If N ′ is submodule of M and N ′ ∩ N = 0, then N ′ ⊂ N ′′ and M = N ′′ ⊕ N for
some submodule N ′′ of M
Proof. Let M ′ is a complement of N in M which contains the submodule N ′.
Then E(M) = E(N ′) ⊕ N and M = (E(N ′) ∩ M) ⊕ N.
(cid:3)
(cid:3)
Theorem 3.1. For a ring R the following conditions are equivalent:
1) Every right R-module is almost injective.
2) R is a right semiartinian ring, Loewy(RR) ≤ 2 and every right R-module
is a direct sum of an injective module and a V -module.
ALMOST PROJECTIVE AND ALMOST INJECTIVE MODULES
11
3) R is a right semiartinian ring, Loewy(RR) ≤ 2 and every right R-module
is a direct sum of a projective module and a V -module.
4) The ring R satisfies the following conditions:
a) in the ring R there exists a finite set of orthogonal idempotents {ei}i∈I
such that eiR is a local injective right R-module of length two, for each
i ∈ I and J(R) = ⊕i∈I J(eiR);
b) R/J(R) is a right SV -ring and Loewy(RR) ≤ 2;
c) R/SI(RR) is a right artinian ring.
5) The ring R is isomorphic to the ring of formal matrix
(cid:18) T T MS
S (cid:19), where
0
a) S is a right SV -ring and Loewy(S) ≤ 2;
b) for some ideal I of a ring S the equality MI = 0 holds and the ring
(cid:18) T T MS/I
S/I (cid:19) is an artinian serial, with the square of the Jacobson
0
radical equal to zero.
Proof. the Implication 4)⇒2) and 4)⇒3) follow from Lemma 2.
1)⇒4) From corollary 2.1 it follows that R/J(R) is a right V -ring. According to
[10, proposition 2.6] Loewy(RR) ≤ 2. Then RR/Soc(RR) is a semisimple module
of finite length, and from corollary 2.1 follows that the ring R contains a finite
set of orthogonal idempotents {ei}i∈I satisfying the condition a) of 4). Therefore,
RR = ⊕i∈IeiR ⊕ A, where A is a semiartinian right R-module and Loewy(A) ≤ 2.
As AJ(R) = 0, then, by corollary 2.1, A is a V -module. Suppose that Soc(A)
contains an infinite family of primitive orthogonal idempotents {fi}i∈I ′ such that
fiR 6= E(fiR) for each i ∈ I ′. Let B is a complement of ⊕i∈I ′fiR in RR, which
contains the J(R). Consider the homomorphism f : ⊕i∈I ′fiR⊕B → ⊕i∈I ′E(fiR),
defined by f (r + b) = r, where r ∈ ⊕i∈I ′fiR, b ∈ B. Assume that ι : ⊕i∈I ′fiR ⊕
B → RR is a natural embedding.
If there exists a homomorphism g : RR →
⊕i∈I ′E(fiR) such that f = gι then f (⊕i∈I ′fiR) ⊂ g(RR) ⊂ ⊕i∈I ′′E(fiR), where
I ′′ ⊂ I ′.Therefore I ′′ < ∞, which is impossible. Since the module ⊕i∈I ′E(fiR) is
a almost RR-injective, then there exists non-zero idempotent π ∈ EndR(RR) and a
homomorphism h : ⊕i∈I ′E(fiR) → π(RR) such that πι = hf. Since ⊕i∈I ′fiR⊕B is
essential in RR, then πι 6= 0. Therefore, h 6= 0. Then h(E(fi0R)) 6= 0 for some i0 ∈
I ′. Since πι(J(R)) = hf (J(R)) = 0, then J(π(RR)) = 0. From proposition 2.1 it
follows that E(fi0R) is a local projective module of length two. Since J(π(RR)) =
0, then Ker(hE(fi0 ,R)) and Im(hE(fi0 ,R)) is a simple modules. Then Im(hE(fi0 ,R))
is a direct summand of the module RR. Therefore, Ker(hE(fi0 R)) is a direct
summand of the module E(fi0R), which is impossible. Thus, Soc(A) = SI(RR)⊕
12
ABYZOV A. N.
B where B is a module of finite length. Since A/Soc(A), Soc(A)/SI(RR) is a
modules of finite length, then A/SI(RR) is a module of finite length. Therefore,
R/SI(RR) is right artinian ring.
4)⇒1) Suppose that the ring R satisfy the condition 4) and M, N are right
R-modules. We claim that M is an almost N-injective module. Let N0 is a
submodule of N, and ι : N0 → N be the natural embedding and f : N0 → M
is a homomorphism. Without loss of generality, we can assume that N0 is an
essential submodule of N. In this case Soc(N) = Soc(N0).
Consider the following three cases.
Case f (J(N) ∩ Soc(N)) = 0, f (SI(N)) = 0. There exists a homomorphism
g : N → E(M), such that the equality holds f = gι. If g(N)SI(RR) 6= 0, then
exists a primitive idempotent e ∈ R such that eR is a simple injective module and
g(N)e 6= 0. Then neR is a simple injective module and f (neR) = g(neR) 6= 0 for
some n ∈ N, which contradicts the equality f (SI(N)) = 0. Thus g(N)SI(RR) =
0. Since R/SI(RR) is a right Artinian ring and by Corollary 2.1, R/SI(RR) is
an almost right V -ring, then by [10, Corollary 3.2], RR/SI(RR) is an Artinian
serial ring and J 2(RR/SI(RR)) = 0. Then by [14, 13.67], g(N) = N1 ⊕ N2, where
N1 is a semisimple module and N2 is a direct sum of local modules of length
two. If N2 6= 0 then there exists an epimorphism h : N2 → L, where L is a local
module of length two. Since L is a projective module, hπg is a split epimorphism,
where π : N1 ⊕ N2 → N2 is the natural projection. Consequently, hπgL′ is an
isomorphism for some local submodule L′ of the module N and f (Soc(L′)) =
g(Soc(L′)) 6= 0, which contradicts the equality f (J(N) ∩ Soc(N)) = 0. Then
g(N) ⊂ Soc(E(M)) ⊂ M. Hence, we can conside the homomorphism g as an
element of the Abelian group HomR(N, M).
Case f (J(N) ∩ Soc(N)) 6= 0. If f (N0) is not a V -module, then by Lemma 2.1,
there exists an epimorphism h : f (N0) → L, where L is an uniform but is not a
simple module, whose socle is a simple module. By lemma 3.1, L is a projective
and injective module. Since L is a projective module, N0 = f −1(Ker(h)) ⊕
L′, f (N0) = Ker(h)⊕f (L′), where L′ is a submodule of N0 and L ∼= L′. By Lemma
3.2, the following conditions are satisfied for some direct summands M ′, N ′ of
modules M and N, respectively:
M = M ′ ⊕ f (L′), Ker(h) ⊂ M ′, N = N ′ ⊕ L′, f −1(Ker(h)) ⊂ N ′.
Let π1 : M ′ ⊕ f (L′) → f (L′), π2 : N ′ ⊕ L′ → L′ be natural projections. There
exists an isomorphism h′ : f (L′) → L′, such that f h′ = 1f (L′). Then we have the
equality (h′π1)f = π2ι.
ALMOST PROJECTIVE AND ALMOST INJECTIVE MODULES
13
If f (N0) is a V -module, then for some simple submodule S of J(N)∩Soc(N) we
have the equality f (N0) = f (S) ⊕ M ′, where M ′ is a submodule of the module M
and f (S) 6= 0. Let π : f (S) ⊕ M ′ → f (S) be the natural projection. We can con-
sider the homomorphism f as an element of the Abelian group HomR(N0, f (N0)).
Then N0 = Ker(πf ) ⊕ S. By Lemma 3.1, the following conditions are satisfied for
some submodules N ′ and L′ of the module N:
N = N ′ ⊕ L′, Ker(πf ) ⊂ N ′, lg(L′) = 2, Soc(L′) = S.
By Corollary 2.1, R is a right almost V -ring. Then by [11, 2.9], there exists
a decomposition M = M1 ⊕ M2 of module M, such that M1 is a complement
for f (S) in M and M ′ ⊂ M1. Easy to see that π2(f (S)) is a simple essential
submodule of M2, where π2 : M1 ⊕ M2 → M2 is the natural projection. Let
h : S → π2f (S) be the isomorphism is defined by h(s) = π2f (s) for every
s ∈ S. We can consider the homomorphism h−1 as an element of the Abelian
group HomR(π2f (S), L′). If M2 is a simple module, then we have the equality
(h−1π2)f = π′ι, wher π′ : N ′ ⊕ L′ → L′ is the natural projection. If M2 is not
a simple module, then since M2 is an injective module, there is an isomorphism
π2f (S) = h−1. Then we have the equality (h′π2)f = π′ι.
h′ : M2 → L′ such that h′
Case f (SI(N)) 6= 0. In this case, for some simple injective submodule S
of the module N we have f (S) 6= 0. Since f (S) is an injective module, M =
f (S) ⊕ M0, where M0 is a submodule of M. Let π : f (S) ⊕ M0 → f (S) be the
natural projection. Then N0 = Ker(πf ) ⊕ S. By Lemma 3.2, there exists a direct
summand N ′ of N such that:
N = N ′ ⊕ S, Ker(πf ) ⊂ N ′.
Let π′ : N ′ ⊕ S → S be the natural projection. There is an isomorphism h :
f (S) → S, such that f h = 1f (S). Then we have the equality (hπ)f = π′ι.
2)⇒4) Suppose that the ring R satisfy the condition 2). According to the
condition 2), we have that R/J(R)R = A ⊕ B, where A is an injective module
and B is a V -module. By [18, Theorem 3.2], A has finite Goldie dimension.
Since A is a semiartinian module and J(A) = 0, it follows that A is a semisimple
module. Therefore, by [15, 23.4], R/J(R) is a V -ring.
By the condition 2), this implies RR = A′ ⊕ B′, where A′ is an injective module
and B′ is a V -module. It is easy to see, according to the condition 2), the injective
hull of every simple R-module has the length at most 2. Then by [18, Theorem
3.2], A′ is a finite direct sum of modules of length at most 2.
Let M be a right injective R/SI(R)-module and {Li}i∈I be a maximal inde-
pendent set of submodules of M with lg(Li) = 2 for all i. By the condition 2),
14
ABYZOV A. N.
⊕i∈ILi is an injective R/SI(R)-module. Consequently, there exists a submodule
N of M such that M = ⊕i∈I Li ⊕ N. If N(Soc(R)/SI(R)) 6= 0, then N contains
a simple submodule S, such that S is not injective as right R-module. Then the
injective hull E(S) of the right R-module S has the length two and obviously
E(S)SI(R) = 0. Consequently, S is not a injective right R/SI(R)-module and
there exists a local injective submodule L of N of length two such that S ⊂ L. This
contradicts the choice of the set {Li}i∈I. Consequently, N(Soc(R)/SI(R)) = 0
and since R/ Soc(R) is a Artinian semisimple ring, we have N is a semisimple
module. Thus, every injective right R/SI(R)-module is a direct sum of injective
hulls of simple modules and since [13, 6.6.4], we have that R/SI(RR) is a right
Artinian ring.
3)⇒4) Suppose that the ring R satisfy the condition 3). If R′ = R/J(R) is not
a right V -ring, then by Lemma 2.1, there is a right ideal T of the ring R′ such
that the right R′-module R′/T is an uniform but is not a simple module, whose
socle is a simple module. Consequently, by the condition 3) the module R′/T is
projective and isomorphic to a submodule of R′
R′, which is impossible. Hence,
R/J(R) is a V -ring.
Let S be a simple right R-module and E(S) 6= S. By condition 3), E(S)
is a projective module. By [13, 7.2.8, 11.4.1], E(S) is a local module. Since
Loewy(R) ≤ 2, it follows that E(S)/S is a semisimple module. Consequently,
J(E(S)) = S and lg(E(S)) = 2.
By Zorn's Lemma there is a maximal independent set of submodules {Li}i∈I
of RR such that Li is a local injective module of length two for all i ∈ I. Since
Loewy(RR) ≤ 2, it follows that I is a finite set and I < lg(RR/Soc(RR)). Then
RR = ⊕i∈ILi ⊕ eR, where e2 = e ∈ R. By Lemma 2.1 and the condition 3), eR is
a V -module. Consequently, J(R) = ⊕i∈I Li.
i=1fiR. Since E(⊕∞
Now assume that Soc(eR) contains an infinite set of orthogonal primitive idem-
potents {fi}∞
i=1 with E(fiR) 6= fiR for all i. There exists a subset I ′ of I, such
that Z(Li) 6= 0 for all i ∈ I ′ and f R = ⊕i∈I\I ′Li ⊕ eR is a nonsingular module,
where f 2 = f ∈ R. There exists a homomorphism f : RR → E(⊕∞
i=1fiR) such
that f (r) = r for all r ∈ ⊕∞
i=1fiR) is a nonsingular module, it
is generated by the module ⊕i∈I\I ′Li ⊕ eR. From the condition 3), implies that
E(⊕∞
i=1fiR) can be considered
∼= f R for all i ∈ I ′′. There exists a
as a direct summand of ⊕i∈I ′′Mi, where Mi
finite subset {i1, . . . , ik} of I ′′ such that the following inclusion holds f (RR) ⊂
Mi1 ⊕ . . . ⊕ Mik . Let π : Mi1 ⊕ . . . ⊕ Mik ⊕ (⊕i∈I ′′\{i1,...,ik}Mi) → ⊕i∈I ′′\{i1,...,ik}Mi
be the natural projection. Since ⊕i∈I ′′Mi is nonsingular and f (RR) is an essen-
tial submodule of E(⊕∞
i=1fiR) is
i=1fiR) is a projective module. Consequently, E(⊕∞
i=1fiR), then π(E(⊕∞
i=1fiR)) = 0. Then E(⊕∞
ALMOST PROJECTIVE AND ALMOST INJECTIVE MODULES
15
a direct summand of Mi1 ⊕ . . . ⊕ Mik , and consequently, E(⊕∞
i=1fiR) is finitely
generated. By [18, Theorem 3.2], E(⊕∞
i=1fiR) has finite Goldie dimension, which
is impossible. Thus, Soc(eR) = SI(R) ⊕ S, where S is a semisimple module of
finite length. Consequently, R/SI(RR) is a right Artinian ring.
4)⇒5) Suppose that the ring R satisfy the condition 4). There is an idem-
potent e ∈ R such that eR = ⊕i∈IeiR. It is clear that eRSI(R) = 0 and
SI(R) ⊂ (1 −e)R. By the condition 4), (1 −e)R is a semiartinian V -module, then
(1 − e)Re = 0. Easy to see that (1 − e)R/J(R)(1 − e) ∼= (1 − e)R(1 − e), where e =
e + J(R). By [19, Theorem 2.9], (1 − e)R/J(R)(1 − e) ∼= EndR/J(R)(1 − e)R/J(R)
is a right SV -ring and Loewy((1−e)R/J(R)(1−e)) ≤ 2. Thus, the Peirce decom-
position (cid:18) eRe eReeR(1 − e)(1−e)R(1−e)
(1 − e)R(1 − e)
0
(cid:19) of the ring R satisfies the conditions
a) and b) of 4). By Lemma 3.1, every right module over the ring R/SI(R)
is a direct sum of an injective module and a V -module.
It is clear that ev-
ery V -module over a right Artinian ring is semisimple, then, by [14, 13.67],
R/SI(R) ∼= (cid:18) eRe eReeR(1 − e)(1−e)R(1−e)/SI(R)
(1 − e)R(1 − e)/SI(R) (cid:19) is an Artinian serial ring
0
whose the square of the Jacobson radical is zero.
5)⇒4) Put
R′ = (cid:18) T T MS
S (cid:19) , I ′ = (cid:18) 0 0
0 I (cid:19) , e = (cid:18) 1 0
0 0 (cid:19) , f = (cid:18) 0 0
0 1 (cid:19) .
0
R′ such that eR′
Since eR′I ′ = 0 and R′/I ′ is an Artinian serial ring whose the square of the
Jacobson radical is zero, there exists a finite set of orthogonal idempotents {ei}i∈I
R′ = ⊕i∈I eiR′ ⊕ A, and for
and a semisimple submodule A of R′
every i eiR′ is a local right R′-module of length two and eiR′ as right R/I ′-
module is injective. We claim that eiR′ is an injective R′-module for every i.
Suppose that E(eiR′)I ′ 6= 0. Then, there exists an elements r ∈ I ′, m ∈ E(eiR′)
such that mrR′ = Soc(eiR′). Since eiR′I ′ = 0 and S is a regular ring, then
Soc(eiR′) = mrR′ = mrR′rR′ = 0. This is a contradiction. Thus, eiR′ is an
injective module for every i. Since S ∼= R′/eR′ is a right V -ring and f R′eR′ = 0,
we have f R′ is a V -module. Since
R′
R′ = ⊕i∈I eiR′ ⊕ A ⊕ f R′
and A⊕f R′ is a V -module, we have that J(R′) = ⊕i∈IJ(eiR′) and Loewy(R′
2. Since R′/J(R′) ∼= T /J(T ) × S, it follows that R′/J(R′) is a right SV -ring.
R′) ≤
There exists a right ideal I ′′ of R′ such that
Soc(f R′
R′) = I ′′ ⊕ (Soc(f R′
R′) ∩ I ′).
16
ABYZOV A. N.
R′/I ′) <
R′/I ′ and lg(f R′
Since the right R′-module I ′′ isomorphic to the submodule f R′
R′) ∩ I ′.
∞, we have that lg(I ′′) < ∞. Let N is a simple submodule of Soc(f R′
We show that N is an injective module. Assume that E(N)e 6= 0. Then, there
exists elements r ∈ R′, n ∈ E(N) such that nerR′ = N. Since eR′I ′ = 0 and S
is a regular ring, we have NI ′ = N, and consequently N = NI ′ = nerR′I ′ = 0,
which is impossible. Thus NeR = 0. Consequently, we can consider N as a
module over the ring R′/eR′. Since R′/eR′ ∼= S is a right V -ring, it follows that
E(N) = N. Thus Soc(f R′
R′) ∩ I ′ = Soc(I ′) ⊂ SI(R′). Since R′/ Soc(R′)R′ is a
semisimple module and I ′/ Soc(I ′)R′ isomorphic to a submodule of R′/ Soc(R′)R′,
R′, I ′/ Soc(I ′)R′
we have that I ′/ Soc(I ′)R′ is a module of finite length. Since R/I ′
are modules of finite length, we have R′/ Soc(I ′)R′ is a module of finite length.
Consequently, R′/SI(R′) is a right Artinian ring.
(cid:3)
Theorem 3.2. For a ring R the following conditions are equivalent:
1) Every R-module is almost injective.
2) The ring R is a direct product of the SV -ring whith Loewy(RR) ≤ 2, and
an artinian serial ring, with the square of the Jacobson radical equal to
zero.
Proof. The implication 2)⇒1) follows from the previous theorem.
1)⇒2) According to Theorem 3.1, the ring R isomorphic to the formal upper
triangular matrix ring R′ = (cid:18) T M
0 S (cid:19), satisfying the conditions of Theorem 3.1.
5). Since every left R′-module is almost injective, from the analogue of Theorem
7 on the left-hand side, it implies that J(R′) contained in a finite direct sum of
left local injective R′-modules of length two. Since M ′ = (cid:18) 0 M
it follows that M ′ = J(Pn
thogonal primitive idempotents and R′e′
two for every 1 ≤ i ≤ n. For every 1 ≤ i ≤ n, the idempotent e′
n are or-
i is a local injective module of length
i has
0 (cid:19) ⊂ J(R′),
1, . . . , e′
0
i, where e′
i) = Pn
i=1 J(R′)e′
i=1 R′e′
the form (cid:18) fi mi
S. Since J(R′)(cid:18) fi mi
ei (cid:19), where fi, ei are idempotents respectively rings T and
0 (cid:19) is a simple submodule of the
ei (cid:19) = (cid:18) J(T )fi Mei
0
0
0
left R-module M ′, it follows that Mei
6= 0. Since
i + J(R′) is a primitive idempotent of the ring R′/J(R′), we have fi = 0. Thus,
e′
ei (cid:19), where ei is a primitive idempotent of the ring S and miei = mi.
i = (cid:18) 0 mi
e′
6= 0, and consequently ei
0
ALMOST PROJECTIVE AND ALMOST INJECTIVE MODULES
17
0
i=1 ei) = 0.
i=1 J(R′)e′
i, then M = ⊕n
i=1Mei is a decomposi-
0 (cid:19) = Pn
Since M ′ = (cid:18) 0 M
tion of the semisimple left T -module into a direct sum of simple submodules
and M(1 − Pn
If there exists a primitive idempotent e of the ring
S such that eS ∼= eiS, where 1 ≤ i ≤ n, then Me 6= 0. Then the right ideals
(Pn
i=1 ei)S of the ring S do not contain isomorphic simple
right R-submodules. Consequently, e = Pn
i=1 ei is a central idempotent of the
ring S and the ring R is isomorphic to the direct product of the SV -ring (1 − e)S
and the Artinian serial ring (cid:18) T M
0 eS (cid:19) whose the square of the Jacobson radical
i=1 ei)S and (1 − Pn
is zero.
(cid:3)
The following theorem follows from the previous theorem and [20, theorem 1.7].
Theorem 3.3. For commutative rings R the following conditions are equivalent:
1) Every R-module is almost injective;
2) Every R-module is an extension of the semisimple module by an injective
one.
References
[1] Y. Baba, Note on almost M-injectives, Osaka J. Math. 26(1989) 687698
[2] M. Harada, T. Mabuchi, On almost M-projectives, Osaka J. Math. l 26(1989) 837848
[3] Y. Baba, M. Harada, On almost M-projectives and almost M-injectives, Tsukuba J. Math.
14(1990) 5369
[4] M. Harada, On almost relative injectives on Artinian modules, Osaka J. Math. 27(1990)
963971
[5] M. Harada, Direct sums of almost relative injective modules, Osaka J. Math. 28(1991)
751758
[6] M. Harada, Note on almost relative projectives and almost relative injectives, Osaka J.
Math. 29(1992) 435446
[7] M. Harada, Almost projective modules, J. Algebra 159(1993) 150157
[8] A. Alahmadi, S. K. Jain, A note on almost injective modules, Math. J. Okayam, 51(2009)
101-109
[9] A. Alahmadi, S. K. Jain, S. Singh, Characterizations of Almost Injective Modules, Con-
temp. Math. 634(2015) 11-17
[10] M. Arabi-Kakavand, S. Asgari, Y. Tolooe, Rings Over Which Every Module Is Almost
Injective, Communications in Algebra 44(7)(2016) 2908-2918
[11] M. Arabi-Kakavand, S. Asgari, H. Khabazian, Rings for which every simple module is
almost injective, Bull. Iranian Math. Soc. 42(1)(2016) 113-127
[12] S. Singh, Almost relative injective modules, Osaka J. Math. 53( 2016) 425438
[13] Kasch, F. Modules and Rings, Academic Press 1982.
18
ABYZOV A. N.
[14] A.A. Tuganbaev, Ring Theory. Arithmetical Modules and Rings, MCCME, Moscow, 2009,
472 .
[15] R. Wisbauer, Foundations of Module and Ring Theory Philadelphia: Gordon and Breach
1991
[16] A. N. Abyzov, Weakly regular modules over normal rings, Sibirsk. Mat. Zh., 49:4 (2008),
721738
[17] N. V. Dung, D. V. Huynh, P. F. Smith, R. Wisbauer, Extending Modules, Longman, Harlow
Pitman Research Notes in Mathematics 313 1994
[18] H. Q. Dinh, D. V. Huynh, Some results on self-injective rings and Σ − CS rings, Commun.
Algebra 31(12)(2003) 60636077
[19] G. Baccella, Semi-Artinian V-rings and semi-Artinian von Neumann regular rings, J. Al-
gebra 173(1995) 587612
[20] A. N. Abyzov, Regular semiartinian rings, Russian Mathematics (Izvestiya VUZ. Matem-
atika), 2012, 56:1, 18
[21] P. A. Krylov, A. A. Tuganbaev, Modules over formal matrix rings, Fundament. i prikl.
matem., 15:8 (2009), 145211
Department of Algebra and Mathematical Logic, Kazan (Volga Region) Fed-
eral University, 18 Kremlyovskaya str., Kazan, 420008 Russia
E-mail address: [email protected]
|
1309.2567 | 1 | 1309 | 2013-09-10T16:19:37 | Greedy bases in rank 2 generalized cluster algebras | [
"math.RA",
"math.CO"
] | In this note we extend the notion of greedy bases developed by Lee, Li, and Zelevinsky to rank two generalized cluster algebras, i.e. binomial exchange relations are replaced by polynomial exchange relations. In the process we give a combinatorial construction in terms of a refined notion of compatible pairs on a maximal Dyck path. | math.RA | math |
GREEDY BASES IN RANK 2 GENERALIZED CLUSTER ALGEBRAS
DYLAN RUPEL
To the memory of Andrei Zelevinsky
Abstract. In this note we extend the notion of greedy bases developed by Lee, Li, and Zelevinsky to
rank two generalized cluster algebras, i.e. binomial exchange relations are replaced by polynomial exchange
relations. In the process we give a combinatorial construction in terms of a refined notion of compatible
pairs on a maximal Dyck path.
Contents
Introduction
1.
1.1. Organization
1.2. Related and Future Work
Acknowledgements
2. Rank 2 Generalized Cluster Algebras and their Greedy Bases
2.1. Structure of Rank 2 Generalized Cluster Algebras
2.2. Greedy Elements in Rank 2 Generalized Cluster Algebras
2.3. Combinatorial Construction of Greedy Elements
3. Structure of Maximal Dyck Paths
4. Combinatorics of Graded Compatible Pairs: Shadows
4.1. Unbounded Gradings
4.2. Bounded Gradings
4.3. Magnitudes of Bounded Gradings
5. Appendix: Multinomial Coefficients
References
1
2
2
2
3
3
6
8
13
14
14
19
21
23
25
1. Introduction
At the heart of the theory of cluster algebras is the hope for a combinatorially defined basis of a slightly
complicated inductively defined algebra. This is motivated by the conjectural relationship with the dual
canonical basis: the cluster monomials should be contained in this basis. Any such "good basis" of a cluster
algebra should satisfy two main conditions: it should be independent of the choice of an initial cluster and
it should contain all cluster monomials.
This note grew out of an attempt to understand one such "good basis", namely the greedy basis, of a
rank 2 cluster algebra. This basis consists of indecomposable positive elements with a rich combinatorial
description. Our main goal in this project was to understand these combinatorics and to see in exactly
what generality such a basis should exist. To our surprise this basis is extremely general in the sense that it
exists for all rank 2 generalized cluster algebras defined by arbitrary monic, palindromic polynomial exchange
relations with non-negative integer coefficients.
Our main theorem is the following analogue of the main result from [LLZ], we refer the reader to the
sections below for precise definitions.
Date: September 12, 2018.
1
2
D. RUPEL
Theorem 1.1. Let k denote an ordered field with positive cone Π and suppose P1, P2 ∈ Π[z] are monic,
palindromic polynomials. Denote by k the Z-subalgebra of k generated by the coefficients of P1 and P2. Write
A(P1, P2) for the associated generalized cluster algebra over k.
(a) For each (a1, a2) ∈ Z2 there exists a (unique) greedy element x[a1, a2] ∈ A(P1, P2).
(b) Each greedy element is an indecomposable positive element of A(P1, P2).
(c) The greedy elements x[a1, a2] for (a1, a2) ∈ Z2 form a k-basis of the generalized cluster algebra
A(P1, P2), which we call the greedy basis.
(d) The greedy basis is independent of the choice of initial cluster.
(e) The greedy basis contains all cluster monomials.
1.1. Organization. The paper is organized as follows. Section 2 defines the generalized cluster algebra,
establishes certain analogues of structural results from the classical theory of cluster algebras, and introduces
the greedy basis of a rank 2 generalized cluster algebra. This section also contains the proofs of our main
results, some of them as corollaries of purely combinatorial statements contained in section 4. In section 3
we introduce the underlying structure for the combinatorics we seek to understand, namely maximal Dyck
paths in a lattice rectangle of arbitrary size. Section 4 develops the combinatorics of graded compatible
pairs generalizing the combinatorics extensively studied in [LLZ]. In the appendix, section 5, we collect and
prove auxiliary results related to multinomial coefficients necessary for working with arbitrary powers of
polynomials.
The organization is chosen to allow the section on generalized cluster algebras to be read almost completely
independently of the combinatorial sections. In particular if one is willing to accept the technical results of
section 4, especially Proposition 4.20 and Proposition 4.22, then section 2 and the proof of the main theorem
can be understood on their own.
1.2. Related and Future Work. In [CS] the authors use a special class of generalized cluster algebras to
understand the Teichmuller theory of hyperbolic orbifold surfaces. There they conjecture the positivity for
the initial cluster expansion of all generalized cluster variables. Theorem 1.1, in particular parts (b) and (e),
establishes the positivity of the initial cluster expansion of all generalized cluster variables, verifying their
conjecture in the rank 2 case. Analogous to the inductive proof of positivity for skew-symmetric cluster
algebras of arbitrary rank given in [LS2], it seems likely that one could prove the positivity for generalized
cluster algebras of arbitrary rank inductively from our results presented here.
Chekhov and Shapiro also offer a combinatorial description of their generalized cluster variables by pass-
ing to the universal cover of the orbifold surface and applying the combinatorics of T -paths [S] with an
appropriate coloring. In the case of a disk with two orbifold points (of arbitrary order) and one marked
point on the boundary one obtains a rank 2 generalized cluster algebra. In this case one may describe an
explicit bijection between T -paths in the universal cover and our combinatorial theory of graded compatible
pairs, perhaps this bijection will be expounded upon in a future work.
One of the most powerful tools in the study of classical cluster algebras is the categorification using
the representation theory of valued quivers, for rank 2 cluster algebras this is established in [CZ] and
[R1]. Here the non-initial cluster monomials exactly correspond to rigid representations of the quiver. It
would be interesting to find a generalization of the representation theory of valued quivers for which rigid
representations correspond to non-initial generalized cluster monomials in A(P1, P2). In this case it seems
likely that the Caldero-Chapoton algebras defined in [CLS] would be a useful tool.
Non-commutative analogues of cluster recursions have been studied in [LS1, R2] and a combinatorial
construction was given. Polynomial generalizations have also been studied in [U], though only Laurentness
is established and only for monic, palindromic polynomials P1 = P2. In a forthcoming work [R3] we will
use the combinatorics developed here to establish the Laurentness and positivity of rank 2 non-commutative
generalized cluster variables.
Acknowledgements. The author would like to thank Arkady Berenstein, Kyungyong Lee, Li Li, Thomas
McConville, and Andrei Zelevinsky for useful discussions related to this project.
Many of these ideas were worked out while the author was a postdoctoral research fellow in the Cluster
Algebras program at the Mathematical Sciences Research Institute. The author would like to thank the
MSRI for their hospitality and support. The author would also like to thank the organizers of the Cluster
Algebras program for giving him the opportunity to work in such a stimulating environment.
Rank 2 Generalized Greedy Bases
3
2. Rank 2 Generalized Cluster Algebras and their Greedy Bases
Fix any field k of characteristic zero. Let P1, P2 ∈ k[z] be arbitrary monic palindromic polynomials of
degree d1 and d2 respectively, where a degree d polynomial P (z) is palindromic if P (z) = zdP (z−1). Consider
the ring k(x1, x2) of rational functions in commuting variables x1 and x2. We inductively define rational
functions xk ∈ k(x1, x2) for k ∈ Z by the rule:
(2.1)
xk+1xk−1 =(P1(xk)
P2(xk)
if k is even;
if k is odd.
Write k for the Z-subalgebra of k generated by the coefficients of P1 and P2. Define the generalized cluster
algebra A(P1, P2) to be the k-subalgebra of k(x1, x2) generated by the set {xk}k∈Z of "generalized cluster
variables" which we usually refer to simply as cluster variables. For k ∈ Z the kth clusters in A(P1, P2) is
the pair {xk, xk+1} of neighboring cluster variables.
Example 2.1. We will take P1(z) = 1 + z + z2 and P2(z) = 1 + z + z2 + z3. Then the first few cluster
variables are given by
x3 = x−1
x4 = x−3
x5 = x−5
1 (1 + x2 + x2
1 x−1
1 + x2
1 x−2
1 + x5
+ 3x2
2 [x3
2 [x6
1(1 + x2 + x2
2);
1(1 + x2 + x2
1(2 + x2) + x4
2)3 + 2x1(1 + x2 + x2
2) + x1(1 + x2 + x2
1(3 + 4x2 + 4x2
2)2 + (1 + x2 + x2
2)3];
2 + x3
2) + x3
2)4 + (1 + x2 + x2
1(4 + 9x2 + 14x2
2)5].
2 + 11x3
2 + 6x4
2 + x5
2)
2.1. Structure of Rank 2 Generalized Cluster Algebras. It will be convenient to introduce Laurent
polynomial subalgebras Tk = k[x±1
k+1] ⊂ k(x1, x2) for each k ∈ Z. Then we may formulate our first
structural result on rank 2 generalized cluster algebras.
k , x±1
Theorem 2.2 (Laurent Phenomenon). The generalized cluster algebra A(P1, P2) is contained in Tk for each
k ∈ Z.
Before presenting the proof we need to introduce some additional notation. Write
P1(z) = ρ0 + ρ1z + · · · + ρd1 zd1
and
P2(z) = 0 + 1z + · · · + d2zd2,
since these are monic and palindromic the coefficients satisfy:
ρ0 = ρd1 = 1 = 0 = d2 ;
ρt = ρd1−t
t = d2−t
for 0 ≤ t ≤ d1;
for 0 ≤ t ≤ d2.
We learned the following proof of Laurentness for rank 2 cluster algebras from Arkady Berenstein.
Proof. We will prove for each m ∈ Z the containment xm ∈ Tk. This is accomplished by induction by
considering the element xd1
m+1xm+4 as follows (without loss of generality assume m is odd):
xd1
m+1xm+4 =
xd1
m+1P1(xm+3)
xm+2
=
xd1
m+1P1(xm+3) − P1(xm+1)
xm+2
+
P1(xm+1)
xm+2
d1
Pt=0
=
(ρtxd1
m+1xt
m+3 − ρd1−txd1−t
m+1)
xm+2
d1
Pt=1
+ xm =
ρt(xt
m+1xt
m+3 − 1)xd1−t
m+1
xm+2
+ xm,
where the last equality used that P1 was palindromic. Using that P2 was assumed monic, we see that
xt
m+1xt
2(xm+2) − 1 ∈ xm+2k[xm+2] for each t ≥ 0 and thus
m+3 − 1 = P t
d1
Pt=1
ρt(xt
m+1xt
m+3 − 1)xd1−t
m+1
xm+2
∈ k[xm+1, xm+2].
It follows that xm ∈ k[xm+1, xm+2, xm+3, xm+4] for each m ∈ Z. By a similar calculation one may prove the
membership xm ∈ k[xm−4, xm−3, xm−2, xm−1]. By inductively "shifting the viewing window" we see that
xm ∈ k[xk−1, xk, xk+1, xk+2] ⊂ Tk for each k ∈ Z.
(cid:3)
4
D. RUPEL
Remark 2.3. If k is an ordered field it makes sense to talk about positive elements, we will make this
precise in Section 2.2. In this case, when P1 and P2 have positive coefficients it is not apparent from this
proof of Laurentness that the generalized cluster variables can be expressed as Laurent polynomials with
positive coefficients. Establishing this positivity will be one of the central results of this note.
It follows from the proof of Theorem 2.2 that the rank 2 generalized cluster algebra A(P1, P2) is equal to
each of its lower bound algebras k[xk−1, xk, xk+1, xk+2] for k ∈ Z. This allows us to identify a relatively simple
k-basis of A(P1, P2), indeed for each (a1, a2) ∈ Z2 we define the standard monomial zk[a1, a2] ∈ A(P1, P2)
in the kth cluster by
zk[a1, a2] = x[a2]+
k−1 x[−a1]+
k
k+1 x[a1]+
x[−a2]+
k+2 ,
where we write [a]+ = max(a, 0). We define the cluster monomials to be the set (cid:8)zk[a1, a2](cid:9)a1,a2∈Z≤0,k∈Z
Theorem 2.4. For each k ∈ Z the set of all standard monomials (cid:8)zk[a1, a2](cid:9)a1,a2∈Z forms a k-basis of
A(P1, P2).
Proof. Since A(P1, P2) = k[xk−1, xk, xk+1, xk+2] we know that A(P1, P2) is spanned over k by all monomials
of the form xb1
k+2 where each bi is a nonnegative integer. Using the exchange relations (2.1) we
may eliminate any factors of the form xk−1xk+1 or xkxk+2, in particular we see that the standard monomials
k−1xb2
k+1xb4
k xb3
(cid:8)zk[a1, a2](cid:9)a1,a2∈Z span A(P1, P2).
It only remains to show that the set of standard monomials is linearly independent over k. For this we
k+1, but the set
k+1}a1,a2∈Z is linearly independent over k in Tk. It follows that the standard monomials
(cid:3)
note that the smallest monomial appearing in the kth cluster expansion of zk[a1, a2] is x−a1
of monomials {x−a1
are linearly independent in A(P1, P2) ⊂ Tk.
k x−a2
k x−a2
In fact there is a stronger version of the Laurent phenomenon which states that A(P1, P2) is exactly the
set of all universally Laurent elements of k(x1, x2), i.e. A(P1, P2) consists of all rational functions which
are Laurent when expressed in terms of any given cluster {xk, xk+1}. Moreover, as the next result states,
the check for membership in A(P1, P2) can be restricted to checking Laurentness for any three consecutive
clusters.
Theorem 2.5 (Strong Laurent Phenomenon). For any m ∈ Z we have
(2.2)
A(P1, P2) = \k∈Z
Tk =
m+1
\k=m−1
Tk.
Proof. To prove this we establish an analogue of the equality of upper and lower bounds for acyclic cluster
algebras from [BFZ]. More precisely, the result will follow if we can establish the equality
(2.3)
k[xm−1, xm, xm+1, xm+2] = Tm−1 ∩ Tm ∩ Tm+1.
Indeed, A(P1, P2) has already been identified with the left hand side in the course of proving Theorem 2.2.
The containment of the left hand side of (2.3) into the right hand side follows from the Laurent Phe-
nomenon. Our goal is to establish the opposite containment. We accomplish this by decomposing Tm by
decomposing
and proving the containment upon intersection of each summand with Tm−1 ∩ Tm+1.
Without loss of generality we will assume m is even. We begin by establishing the equality
Tm = k[x−1
m , x−1
m+1] + k[xm, x±1
m+1] + k[x±1
m , xm+1]
(2.4)
Tm−1 ∩ k[x−1
m , x−1
m+1] ∩ Tm+1 =
k
k[xm−1xm+2]
k[xm−1, xd2
k[xm−1xd1
m−1xm+2]
m+2, xm+2]
if 2 ≤ d1d2;
if d1 = d2 = 1;
if d1 = 0;
if d2 = 0.
Notice that each ring on the right hand side of (2.4) is contained in the left hand side. Indeed, by the Laurent
Phenomenon each of these rings is contained in Tm−1 ∩ Tm+1. When d1 = d2 = 1 we have
xm−1xm+2 = (xmx−1
m+1 + x−1
m+1)(x−1
m xm+1 + x−1
m ) = 1 + x−1
m + x−1
m+1 + x−1
m x−1
m+1 ∈ k[x−1
m , x−1
m+1].
Rank 2 Generalized Greedy Bases
5
When d1 = 0 we have xm−1 = x−1
m+1 and
m−1xm+2 = x−d2
xd2
m+1P2(xm+1)x−1
m = P2(x−1
m+1)x−1
m ∈ k[x−1
m , x−1
where we used that P2 is palindromic. Finally when d2 = 0 we have xm+2 = x−1
xm−1xd1
m+2 = x−1
m+1P1(xm)x−d1
m = x−1
m+1P1(x−1
m ) ∈ k[x−1
m , x−1
m+1],
m and
m+1].
Now we check in each case that the intersection on the left hand side of (2.4) is contained in the desired
ring on the right. Consider an element y ∈ k[x−1
m , x−1
m+1], then y may be written in the form
cs,tx−s
m x−t
m+1,
y = Xs,t≥0
(2.5)
where cs,t ∈ k is zero for all but finitely many s and t. Applying the identity xm+1 = P1(xm)
xm−1
we get
It follows from the membership y ∈ Tm−1 that
m x−t
m+1 =Xt≥0
Ct(x−1
m )P t
1(x−1
m )x−t
m+1.
cs,tx−s
m
Ps≥0
P t
1(xm)
y =Xt≥0
xt
m−1 =Xt≥0
cs,tx−s
m
Ps≥0
1(x−1
P t
m )
x−d1t
m xt
m−1,
where the second equality used that P1 is palindromic.
P
s≥0
cs,tzs
1 (z) ∈ k[z] for all t ≥ 0. Denote this polynomial by Ct, i.e.
P t
1(xm)x−d1t
m )x−d1t
Ct(x−1
Ct(x−1
m xt
m )P t
(2.6)
y =Xt≥0
m−1 =Xt≥0
By a similar calculation with (2.5) and the identity xm = P2(xm+1)
we get
(2.7)
y =Xs≥0
C′
s(x−1
m+1)x−d2s
m+1 xs
m+2 =Xs≥0
for some polynomials C′
If d2 = 0, we have x−1
s ∈ k[z].
m = xm+2 and the first equality in (2.6) becomes
xm+2
s(x−1
C′
m+1)P s
2 (x−1
m+1)x−s
m
Ct(xm+2)xd1t
m+2xt
m−1 ∈ k[xm−1xd1
m+2, xm+2].
Similarly, if d1 = 0 we have x−1
m+1 = xm−1 and the first equality in (2.7) becomes
C′
s(xm−1)xd2s
m−1xs
m+2 ∈ k[xm−1, xd2
m−1xm+2].
y =Xt≥0
y =Xs≥0
Now we consider the cases d1d2 6= 0, first we introduce some more notation. Denote by σ(t) the largest
index so that cσ(t),t 6= 0 in the expansion (2.5), then using the last equality in (2.6) we must have σ(t) ≥ d1t
for each t ≥ 0. Similarly denote by τ (s) the largest index so that cs,τ (s) 6= 0 in (2.5), then the last equality
in (2.7) gives τ (s) ≥ d2s for each s ≥ 0.
t = 0, by a similar argument we see that s = 0 when d1d2 ≥ 2. It follows that y ∈ k.
Suppose t is the largest so that cs,t 6= 0 for some s. Then cσ(t),τ (σ(t)) 6= 0 and τ(cid:0)σ(t)(cid:1) ≥ d2σ(t) ≥ d1d2t.
For d1d2 ≥ 2 and t > 0, this implies τ(cid:0)σ(t)(cid:1) > t contradicting the maximality of t. Thus we must have
Now suppose d1d2 = 1. Then we have τ(cid:0)σ(t)(cid:1) ≥ σ(t) ≥ t. Since t was maximal each of these inequalities
must be an equality. We establish the membership y ∈ k[xm−1xm+2] by a simple induction on the maximal
value t such that ct,t 6= 0. Indeed, when t = 0 the claim is clear. When t > 0, the element y − ct,txt
m+2
is contained in k[xm−1xm+2] by the induction hypothesis, it follows that y ∈ k[xm−1xm+2]. This completes
the proof of the equality (2.4).
m−1xt
The final step in the proof of Theorem 2.5 is to show the equality
Tm−1 ∩ k[xm, x±1
m+1] = k[xm−1, xm, xm+1] ⊂ Tm+1.
The inclusion "⊇" is clear, thus we aim to establish the inclusion "⊆". Indeed, any y ∈ k[xm, x±1
written in the form
m+1] can be
N
ctxt
m+1,
y =
Xt=−N
6
D. RUPEL
for some positive integer N where ct ∈ k[xm] for each t. Making the substitution xm+1 = P1(xm)
xm−1
we get
N
N
y =
ctP t
1(xm)x−t
m−1 +
Xt=0
c−t
1(xm)
P t
xt
m−1.
Xt=1
If in addition we assume y ∈ Tm−1, we must have
term 1, and so we must actually have
c−t
c−t
1 (xm) ∈ k[x±1
P t
m ] for 1 ≤ t ≤ N . But P1(xm) has constant
1 (xm) ∈ k[xm] for 1 ≤ t ≤ N . Thus we get the containment
P t
N
N
y =
ctxt
m+1 +
Xt=0
Xt=1
c−t
1(xm)
P t
xt
m−1 ∈ k[xm−1, xm, xm+1]
as desired. By a similar calculation we get k[x±1
m , xm+1] ∩ Tm+1 = k[xm, xm+1, xm+2] ⊂ Tm−1.
We are now ready to complete the proof:
Tm−1 ∩ Tm ∩ Tm+1 = Tm−1 ∩(cid:0)k[x−1
m , x−1
m , x−1
m+1] + k[xm, x±1
m+1] ∩ Tm+1 + k[xm−1, xm, xm+1] + k[xm, xm+1, xm+2]
m , xm+1](cid:1) ∩ Tm+1
m+1] + k[x±1
= Tm−1 ∩ k[x−1
⊂ k[xm−1, xm, xm+1, xm+2],
as desired. This complete the proof of Theorem 2.5.
(cid:3)
2.2. Greedy Elements in Rank 2 Generalized Cluster Algebras. In this section we introduce greedy
elements in a rank 2 generalized cluster algebra. For this we need to work over a field with an inherent
notion of positivity.
Definition 2.6. A prepostive cone Π ⊂ k \ {−1} satisfies the following closure properties:
• For any a, b ∈ Π, we have a + b, ab ∈ Π;
• For any a ∈ k, we have a2 ∈ Π.
A prepositive cone is called positive if k = Π ∪ −Π. A field k together with a positive cone Π is called an
ordered field. We define a total order ≤ on k by declaring for a, b ∈ k that a ≤ b if b − a ∈ Π.
From now on we assume k is ordered with positive cone Π and P1, P2 ∈ Π[z]. Define the semiring
≥0 = k ∩ Π, this is the Z≥0-subsemiring of k generated by the coefficients of P1 and P2. For each k ∈ Z
k
define the positive semiring T ≥0
k+1] ⊂ Tk, the set of positive elements of Tk. We are interested
in those elements of A(P1, P2) which are universally positive in the following sense.
k , x±1
≥0[x±1
k = k
Definition 2.7. An nonzero element x ∈ A(P1, P2) is called universally positive if its expression as a
. A
positive element is called indecomposable if it cannot be written as a sum of two nonzero positive elements.
Laurent polynomial in any cluster {xk, xk+1} only contains nonnegative coefficients, i.e. x ∈ Tk∈Z T ≥0
k
Our aim in this note is to establish the existence of a universally positive indecomposable basis of A(P1, P2)
called the greedy basis. To describe this basis we need to introduce the following notion of a pointed element
of Tk.
Definition 2.8. Let (a1, a2) ∈ Z2. An element x ∈ Tk is said to be pointed at (a1, a2) in the kth cluster if
it can be written in the form
where c(p, q) ∈ k and c(0, 0) = 1.
x = x−a1
k x−a2
k+1 Xp,q≥0
c(p, q)xp
kxq
k+1,
We will simply refer to an element as "pointed" if it is pointed in the initial cluster {x1, x2}.
of pointed elements "bounded" if x[a1, a2] = z1[a1, a2] is a standard monomial for (a1, a2) ∈ Z2 \ Z2
next result claims that any complete bounded collection of pointed elements in A(P1, P2) forms a k-basis.
Consider a collection (cid:8)x[a1, a2](cid:9)a1,a2∈Z where x[a1, a2] is pointed at (a1, a2) ∈ Z2. Call such a collection
Proposition 2.9. Suppose there exists a complete bounded collection (cid:8)x[a1, a2](cid:9)a1,a2∈Z ⊂ A(P1, P2) where
x[a1, a2] is pointed at (a1, a2) ∈ Z2. Then (cid:8)x[a1, a2](cid:9)a1,a2∈Z forms a k-basis of A(P1, P2).
>0. The
Rank 2 Generalized Greedy Bases
7
Proof. By definition the smallest monomial appearing in a pointed element x[a1, a2] is x−a1
monomials {x−a1
is linearly independent in A(P1, P2) ⊂ T1.
, but the
k+1}a1,a2∈Z are linearly independent over k in T1. It follows that the set {x[a1, a2]}a1,a2∈Z
k x−a2
1 x−a2
2
To finish we only need to show that these pointed elements span A(P1, P2). Suppose x ∈ A(P1, P2) and
write x = Pb1,b2∈Z
cb1,b2 x−b1
1 x−b2
2
for some cb1,b2 ∈ Z. Define the "negative support" S(x) of x by
S(x) = {(b1, b2) ∈ Z2
>0 : cb1,b2 6= 0}.
Moreover, write m(x) = max{b1 + b2 : (b1, b2) ∈ S(x)} ∪ {0} and let M (x) ⊂ S(x) denote those pairs
(b1, b2) ∈ S(x) for which b1 + b2 = m(x) is maximal. Notice that the element
x′ = x − X(b1,b1)∈M(x)
cb1,b2 x[b1, b2]
satisfies m(x′) < m(x). Iterating this process of subtracting off maximal points we may produce an element
x′′ ∈ A(P1, P2) with m(x′′) = 0, i.e. S(x′′) = ∅. Then by Theorem 2.4 and using that the collection
(cid:8)x[a1, a2](cid:9)a1,a2∈Z is bounded we may write
x′′ = X(a1,a2)∈Z2\Z2
>0
da1,a2 z1[a1, a2] = X(a1,a2)∈Z2\Z2
>0
da1,a2 x[a1, a2],
(cid:0)da1,a2 ∈ k(cid:1)
where we note that z1[b1, b2] for (b1, b2) ∈ Z2
>0 does not appear by the construction of x′′.
It follows
that x′′, and hence x, is contained in the k-span of (cid:8)x[a1, a2](cid:9)a1,a2∈Z. Thus we see that the collection
(cid:8)x[a1, a2](cid:9)a1,a2∈Z spans A(P1, P2).
In view of Theorem 2.5, it is natural to look at elements that are positive in three consecutive clusters.
The following result gives precise conditions on the coefficients c(p, q) for a pointed element to be positive in
the initial cluster and its two immediate neighbors. The key insight of [LLZ] is that knowing this restricted
positivity can be enough to know an element is universally positive, but more on that later. Though this
result holds in greater generality we restrict attention to elements pointed in the initial cluster.
(cid:3)
Proposition 2.10. Suppose x ∈ T1 is pointed at (a1, a2) ∈ Z2 and x ∈ T ≥0
is positive when
expanded in each of three consecutive clusters. Then the coefficients c(p, q) in the initial cluster expansion
of x satisfy the following inequality in k for all (p, q) ∈ Z2
0 ∩ T ≥0
1 ∩ T ≥0
2
≥0:
c(p, q) ≥ max " p
Xk=1 X(k1,...,kd2 )⊢k
" q
Xℓ=1 X(ℓ1,...,ℓd1 )⊢ℓ
(−1)k−1c(p − k1 − 2k2 − · · · − d2kd2, q)k1
1 · · ·
(−1)ℓ−1c(p, q − ℓ1 − 2ℓ2 − · · · − d1ℓd1)ρℓ1
1 · · · ρ
kd2
a2 − q + k − 1
a2 − q − 1, k1, . . . , kd2(cid:19)#+
d2 (cid:18)
a1 − p − 1, ℓ1, . . . , ℓd1(cid:19)#+!,
d1 (cid:18)
a1 − p + ℓ − 1
ℓd1
,
where ρi and i denote the coefficients of the polynomials P1 and P2 respectively and where [a]+ = max(a, 0).
Before proceeding with the proof we refer the reader to the appendix, section 5, for our notations and
conventions related to multinomial coefficients.
Proof. Consider the initial cluster expansion of x:
(2.8)
x = x−a1
1 x−a2
2 Xp,q≥0
c(p, q)xp
1xq
2.
in (2.8), we leave the substitution
To establish the second inequality we will make the substitution x1 = P1(x2)
x2 = P2(x1)
we may expand xp−a1
as
x3
x0
in (2.8), and thus the verification of the first inequality, to the reader. Applying Corollary 5.6
1
x3 (cid:17)p−a1
=(cid:16) P1(x2)
=Xℓ≥0 X(ℓ1,...,ℓd1 )⊢ℓ
xp−a1
1
(−1)ℓρℓ1
1 · · · ρ
ℓd1
d1 (cid:18)
a1 − p + ℓ − 1
a1 − p − 1, ℓ1, . . . , ℓd1(cid:19)x
ℓ1+2ℓ2+···+d1ℓd1
2
xa1−p
3
,
8
D. RUPEL
where we used that ρ0 = 1 to slightly simplify the expression. Substituting into the initial cluster expansion
(2.8) of x, we see that the coefficient of xq−a2
in the expansion of x as an element of T2 is given by
xa1−p
3
2
(2.9)
Xℓ≥0 X(ℓ1,...,ℓd1 )⊢ℓ
(−1)ℓc(p, q − ℓ1 − 2ℓ2 − · · · − d1ℓd1)ρℓ1
1 · · · ρ
ℓd1
d1 (cid:18)
a1 − p + ℓ − 1
a1 − p − 1, ℓ1, . . . , ℓd1(cid:19) ≥ 0,
which must be non-negative by the membership x ∈ T ≥0
. Note that the summand is zero for any ℓ > q
and is equal to c(p, q) for ℓ = 0. Solving the inequality (2.9) for c(p, q) and remembering that c(p, q) ≥ 0
completes the proof.
(cid:3)
2
We will be interested in certain greedy pointed elements x[a1, a2] ∈ T1 for (a1, a2) ∈ Z2 which are "mini-
mally positive" in the sense that the bound in Proposition 2.10 is sharp.
Definition 2.11. For (a1, a2) ∈ Z2 define the greedy pointed element x[a1, a2] ∈ T1 whose pointed expansion
coefficients c(p, q) are given by the following "greedy recursion":
c(p, q) = max " p
(2.10)
Xk=1 X(k1,...,kd2 )⊢k
" q
Xℓ=1 X(ℓ1,...,ℓd1 )⊢ℓ
(−1)k−1c(p − k1 − 2k2 − · · · − d2kd2, q)k1
1 · · ·
(−1)ℓ−1c(p, q − ℓ1 − 2ℓ2 − · · · − d1ℓd1)ρℓ1
1 · · · ρ
kd2
d2 (cid:18)
a2 − q + k − 1
a2 − q − 1, k1, . . . , kd2(cid:19)#+
,
ℓd1
d1 (cid:18)
a1 − p + ℓ − 1
a1 − p − 1, ℓ1, . . . , ℓd1(cid:19)#+!
for each nonzero (p, q) ∈ Z2
≥0.
Remark 2.12. One might more naturally call the elements x[a1, a2] "frugal" since they spend as little as
possible to be positive, however the originals were coined "greedy" and we will adhere to this terminology.
Clearly, the greedy element x[a1, a2] is unique and indecomposable once it is well-defined, i.e. once
it is known that only finitely many of the coefficients c(p, q) are nonzero. To establish this we will give a
combinatorial construction in the next section. This combinatorial expression will also imply a more pleasant
recursion which eliminates the need for the maximum.
Proposition 2.13. Fix (a1, a2) ∈ Z2 and define c(p, q) by the recursion (2.10). Then for each nonzero
(p, q) ∈ Z2
≥0 the expansion coefficient c(p, q) admits the following closed form:
(2.11)
c(p, q) =
(cid:20) p
Pk=1 P(k1,...,kd2 )⊢k
(cid:20) q
Pℓ=1 P(ℓ1,...,ℓd1 )⊢ℓ
(−1)k−1c(p − k1 − 2k2 − · · · − d2kd2 , q)k1
1 · · ·
(−1)ℓ−1c(p, q − ℓ1 − 2ℓ2 − · · · − d1ℓd1)ρℓ1
1 · · · ρ
kd2
a2−q+k−1
a2−q−1,k1,...,kd2(cid:1)(cid:21)+
d2 (cid:0)
a1−p−1,ℓ1,...,ℓd1(cid:1)(cid:21)+
d1 (cid:0)
a1−p+ℓ−1
ℓd1
if a1q ≤ a2p;
if a1q ≥ a2p.
2.3. Combinatorial Construction of Greedy Elements. The results of this section rely on the nota-
tion and results presented in Section 4 which may be read and understood independently from the results
presented here and above.
Fix (a1, a2) ∈ Z2. Let D = D[a1]+,[a2]+ denote a maximal Dyck path with horizontal edges D1 and
vertical edges D2, see Section 3 for details. Write C = C[a1]+,[a2]+ for the set of compatible pairs in D,
see Definition 4.1. We begin by introducing the coefficients of the combinatorial construction of the greedy
element x[a1, a2].
Definition 2.14. For each compatible pair (S1, S2) ∈ C we define the coefficient
cS1,S2 = ρℓ1
1 · · · ρ
ℓd1
d1
k1
1 · · ·
kd2
d2
,
where ℓr and kr are the cardinalities of the sets {h ∈ D1 : S1(h) = r} and {v ∈ D2 : S2(v) = r} respectively.
We will also write cS1 = ρℓ1
with notation as above, so that cS1,S2 = cS1 cS2.
and cS2 = k1
1 · · · ρ
ℓd1
d1
1 · · ·
kd2
d2
Theorem 2.15. For any (a1, a2) ∈ Z2 the greedy element x[a1, a2] can be computed via
Rank 2 Generalized Greedy Bases
9
(2.12)
x[a1, a2] = x−a1
1 x−a2
cS1,S2xS2
1 xS1
2
.
2 X(S1,S2)∈C
Example 2.16. We continue Example 2.1 and will use the compatible pairs given in Example 4.3. For
P1(z) = 1 + z + z2 and P2(z) = 1 + z + z2 + z3 we may easily match the cluster variable x5 with the
right hand side of (2.12) for (a1, a2) = (5, 2). Indeed, one may associate conveniently factorized expressions
to the compatible pairs presented in Example 4.3, for example the following compatible pairs contribute to
summands containing x4
1:
012 0 1
0
0 2
0
0 3
0
0 3
0
1(1 + x2 + x2
x4
2)
012 0
0 2
1(1 + x2 + x2
x4
2)
0 1
012 01
1(1 + x2 + x2
x4
2)(1 + x2)
,
which clearly agrees with the coefficient of x4
can be found in Example 4.3.
1 in x5. The expressions associated to all other compatible pairs
To establish the combinatorial construction of greedy elements we will need some preliminary results, but
first a definition. Using the symmetry of the exchange relations (2.1) we see that the generalized cluster
algebra A(P1, P2) admits reflection automorphisms σp (p ∈ Z) defined by σp(xk) = x2p−k. One easily
checks that the reflections {σp}p∈Z generate a (possibly infinite) dihedral group and that this group may
be generated by σ1 and σ2 alone. In the next result we take (2.12) as the definition of the greedy element
x[a1, a2].
Proposition 2.17. For any (a1, a2) ∈ Z2 the elements x[a1, a2] defined by (2.12) satisfy the following
symmetry properties:
(2.13)
σ1(x[a1, a2]) = x[a1, d1[a1]+ − a2]
and σ2(x[a1, a2]) = x[d2[a2]+ − a1, a2],
where [a]+ = max(a, 0).
Proof. By symmetry it suffices to establish the second identity in (2.13). We will have a few cases to consider.
• Suppose a1, a2 ≤ 0. Then D[a1]+,[a2]+ = D0,0 consists of a single point and we have
σ2(x[a1, a2]) = σ2(x−a1
this becomes xa1
1 x−a2
2
) = x−a1
3 x−a2
2
.
Using the identity x3 = P1(x2)
. But notice that the maximal Dyck path
D−a1,0 consists of exactly −a1 consecutive horizontal edges and no vertical edges. In particular, in every
compatible pair (S1, S2) ∈ C−a1,0 the vertical grading S2 is trivial and every possible horizontal grading
S1 : D1 → [0, d1] occurs in C−a1,0. Thus we see that x[−a1, a2] = xa1
2 P1(x2)−a1, as claimed.
1 P1(x2)−a1x−a2
1 x−a2
x1
2
• Suppose a1 ≤ 0 < a2. Then D[a1]+,[a2]+ consists of exactly a2 consecutive vertical edges and no horizontal
edges. As in the previous case it is easy to see that x[a1, a2] = x−a1
2 P2(x1)a2 . Applying σ2 and the
identity x3 = P1(x2)
1 x−a2
x1
gives
3 x−a2
1 x−a2
σ2(x[a1, a2]) = x−a1
= xa1
= x−d2a2+a1
1
2 P2(x3)a2
P1(x2)−a1(cid:0)xd2
1 + d2−1xd2−1
where we used that P2 is palindromic to reverse the coefficients. Thus we need to show that
x−a2
1 (cid:1)a2
2 P1(x2)−a1 P2(cid:0)P1(x2)x−1
2 P1(x2)−a1(cid:0)xd2
P1(x2) + · · · + 1x1P1(x2)d2−1 + P1(x2)d2(cid:1)a2 =
1 + d2−1xd2−1
1
1
P1(x2) + · · · + 1x1P1(x2)d2−1 + P1(x2)d2(cid:1)a2 ,
X(S1,S2)∈Cd2a2−a1 ,a2
cS1,S2 xS2
1 xS1
2
.
Fix a partition (k0, k1, . . . , kd2) ⊢ a2. On the left we will expand using (5.1) and take the coefficient of
k1+2k2+···+d2kd2
x
1
in
a2
k0, k1, . . . , kd2(cid:19)(cid:0)P1(x2)d2(cid:1)k0(cid:0)1x1P1(x2)d2−1(cid:1)k1 · · ·(cid:0)d2−1xd2−1
(cid:18)
1
P1(x2)(cid:1)kd2−1(cid:0)xd2
1 (cid:1)kd2 .
and the identity x3 = P1(x2)
x1
to get
σ2(x[a1, a2]) = x−a1
3 x−a2
2
= x−d2a2+a1
1
2
3 xS1
cS1,S2xS2
X(S1,S2)∈Ca1,a2
2 P1(x2)−a1 X(S1,S2)∈Ca1 ,a2
x−a2
cS1,S2xS2
1 xS1
2
. We apply σ2
P(S1,S2)∈Ca1,a2
cS1,S2xd2a2−S2
1
P1(x2)S2xS1
2
.
10
D. RUPEL
On the right we consider the sum P(S1,S2)
of the set {v ∈ D2 : S2(v) = r}, notice that there are exactly (cid:0)
cS1,S2xS2
1 xS1
2
will always have S2 = k1 + 2k2 + · · · + d2kd2. Thus to complete the proof it suffices to show
over all compatible pairs such that kr is the cardinality
a2
k0,k1,...,kd2(cid:1) such vertical gradings S2 and we
XS1∈C(S2)
cS1,S2xS1
2 = P1(x2)−a1(cid:0)P1(x2)d2(cid:1)k0(cid:0)1P1(x2)d2−1(cid:1)k1 · · ·(cid:0)d2−1P1(x2)(cid:1)kd2−1
for each vertical grading S2 as above, where the summation runs over all horizontal gradings S1 such that
kd2 −1
(S1, S2) ∈ Cd2a2−a1,a2. Using that cS1,S2 = cS1cS2 and that in this case we have cS2 = k1
d2−1 , we may
further reduce the problem to showing that
1 · · ·
(2.14)
XS1∈C(S2)
cS1xS1
2 = P1(x2)−a1P1(x2)d2k0 P1(x2)(d2−1)k1 · · · P1(x2)kd2 −1 .
To see this notice that, by Lemma 4.4, any horizontal edge h outside sh(S2) may be assigned any weight
S1(h) without affecting compatibility. At most S2(v) = d2 for every vertical edge v, which leaves −a1 hori-
zontal edges outside the shadow of S2, this accounts for the factor P1(x2)−a1 . Finally for 0 ≤ r ≤ d2 consider
the kr vertical edges v such that S2(v) = r. There will be d2 − r horizontal edges outside sh(S2) for each
such vertical edge, this contributes the factor P1(x2)(d2−r)kr . This completes the proof of (2.14) and thus
the proof of the proposition in this case.
• Suppose 0 < a1, a2. Then by definition x[a1, a2] = x−a1
1 x−a2
2
In the notation of Section 4.2 notice that ϕ∗
d2
of x[d2a2 − a1, a2] we see that the claim will follow if we can establish the following identity:
S2 = d2a2 − S2. Thus comparing with the definition (2.12)
(2.15)
XS1∈C(S2)
cS1,S2xS1
2 = P1(x2)a1−S2 XS′
1∈C(ϕ∗
d2
cS′
1,ϕ∗
d2
S2xS′
1
2
S2)
for each vertical grading S2. According to Lemma 4.4 the number of horizontal edges D1 \ sh(S2) outside
the shadow of S2 is
Similarly the number of horizontal edges D′
1 \ sh(ϕ∗
d2
S2) outside the shadow of ϕ∗
d2
S2 is
a1 − min(a1, S2) = [a1 − S2]+.
d2a2 − a1 − min(d2a2 − a1, ϕ∗
d2 S2) = [d2a2 − a1 − ϕ∗
d2S2]+ = [S2 − a1]+ = [a1 − S2]+ − (a1 − S2).
Since each of these edges contributes a factor of P1(x2) we can rewrite (2.15) as
P1(x2)[a1−S2]+ XS1∈Crs(S2)
cS1,S2xS1
2 = P1(x2)[a1−S2]+ XS′
1∈Crs(ϕ∗
d2
cS′
1,ϕ∗
d2
S2xS′
1
2
,
S2)
where Crs(S2) denotes those horizontal gradings S1 ∈ C(S2) such that supp(S1) ⊂ rsh(S2).
Since P2 is palindromic we have cS2 = cϕ∗
d2
S2. From the definition (4.1) of Ω we have cS1 = cΩ(S1) and
S1 = Ω(S1) for each S1 ∈ Crs(S2). The result then follows by applying the bijection Ω : Crs(S2) →
Crs(ϕ∗
(cid:3)
d2
S2) established in Proposition 4.20.
The next result says that the combinatorially defined greedy elements are actually elements of the gener-
alized cluster algebra.
Corollary 2.18. For any a1, a2 ∈ Z the elements x[a1, a2] ∈ T1 defined by (2.12) are contained in A(P1, P2).
Proof. By Proposition 2.17 we have x[a1, a2] = σ1(x[a1, d1[a1]+ − a2]) ∈ T0 and x[a1, a2] = σ2(x[d2[a2]+ −
a1, a2]) ∈ T2. The result is then a consequence of Theorem 2.5.
(cid:3)
Rank 2 Generalized Greedy Bases
11
We conclude this section by showing that the combinatorially defined elements x[a1, a2] satisfy the recur-
sion (2.10). More precisely, we will show that they satisfy (2.11) and then appealing to Proposition 2.10 we
may conclude that they satisfy (2.10).
Proposition 2.19. For any a1, a2 ≥ 0 the elements x[a1, a2] ∈ T1 defined by (2.12) satisfy (2.11).
Proof. By symmetry it suffices to show the second equality, that is
(2.16) c(p, q) =" q
Xℓ=1 X(ℓ1,...,ℓd1 )⊢ℓ
(−1)ℓ−1c(p, q − ℓ1 − 2ℓ2 − · · · − d1ℓd1)ρℓ1
1 · · · ρ
ℓd1
d1 (cid:18)
a1 − p + ℓ − 1
a1 − p − 1, ℓ1, . . . , ℓd1(cid:19)#+
for a1q ≥ a2p and (p, q) 6= (0, 0). The remainder of the proof splits into two cases.
• Suppose a1 ≤ p. First notice that (−1)ℓ−1(cid:0)
side is zero. Now the inequality a1 ≤ p together with a1q ≥ a2p implies q ≥ a2. But then Lemma 4.21
implies c(p, q) = 0 since we may identify p = S2 and q = S1 for some compatible pair (S1, S2) ∈ Ca1,a2.
a1−p−1,ℓ1,...,ℓd1(cid:1) is always nonpositive and thus the right hand
a1−p+ℓ−1
• Suppose p < a1. Then as in (2.9) we may interpret the quantity
(2.17)
Xℓ≥0 X(ℓ1,...,ℓd1 )⊢ℓ
(−1)ℓc(p, q − ℓ1 − 2ℓ2 − · · · − d1ℓd1)ρℓ1
1 · · · ρ
ℓd1
d1 (cid:18)
a1 − p + ℓ − 1
a1 − p − 1, ℓ1, . . . , ℓd1(cid:19)
as the the coefficient of xq−a2
in the expansion of x[a1, a2] as an element of T2. Applying σ2 we may
interpret this as the coefficient of xa1−p
in σ2(x[a1, a2]) = x[d2a2 − a1, a2]. If we denote the expansion
coefficients of x[d2a2 − a1, a2] by c′(p′, q′) then we may identify the quantity (2.17) with c′(d2a2 − p, q). We
will show that our current assumptions imply this coefficient is zero which implies (2.16).
xa1−p
3
xq−a2
2
1
2
To simplify the notation we will abbreviate a′
2) ∈ C[a′
1, S′
1 = q and S′
1 = d2a2 − a1 and p′ = d2a2 − p. Denote compatible pairs
1]+,a2. Then note that c′(p′, q) is nonzero if and only if there exists
1. Moreover, since we assume (p, q) 6= (0, 0) we have
2 = p′ > a′
1]+,a2 by (S′
in D′ := D[a′
a compatible pair with S′
(S′
1, S′
2) 6= (0, d2a2). Finally, the assumption a1q ≥ a2p translates into the inequality
(2.18)
(d2a2 − a′
1)S′
1 ≥ a2(d2a2 − S′
2).
We now match with the results from Proposition 4.22 by splitting into further subcases. First note that
d2a2 ≤ a′
1 is impossible since we assume a1 > 0 in this case.
-- Suppose a′
(S′
d2a2 ≤ S′
1, S′
1 ≤ 0. Then D′ consists of a2 consecutive vertical edges and no horizontal edges so that
1 = 0, the inequality (2.18) becomes
2) is contained in the segment (cid:2)(0, 0), (0, d2a2)(cid:1). Since S′
2 which is impossible.
-- Suppose 0 < d1a′
1 ≤ a2 and 0 < a′
or below the segment (cid:2)(0, d2a2), (d1a′
1 < d2a2. Then by Proposition 4.22.b the point (S′
1, d2a2 − d1d2a′
1)(cid:3), which is equivalent to the inequality
2 ≤ −d2S′
1 + d2a2.
S′
1, S′
2) must lie on
Combining this with (2.18) gives the inequalities 0 ≥ S′
(S′
1, S′
with (S′
2 ≥ d2a2, which are only satisfied for
2) = (0, d2a2). However we have assumed this is not the case, thus there are no compatible pairs
1, S′
2) = (q, p′) and (2.17) is equal to zero.
1 and S′
-- Suppose 0 < a2 < d1a′
1 and 0 < a′
1 < d2a2. Then Proposition 4.22.c says (S′
below the segment (cid:2)(0, d2a2), (a2, a′
excluded this point), which is equivalent to the inequality
d2a2 − a′
1
1)(cid:3) (actually Proposition 4.22 allows (S′
S′
2 < −
S′
1 + d2a2.
a2
1, S′
2) must lie strictly
2) = (0, d2a2) but we have
1, S′
But notice that this is exactly the complementary inequality to (2.18), i.e.
(d2a2 − a′
1)S′
1 < a2(d2a2 − S′
2),
12
D. RUPEL
thus there are no compatible pairs with (S′
1, S′
2) = (q, p′) and (2.17) is equal to zero.
This completes the proof of Proposition 2.19.
(cid:3)
Proof of Theorem 2.15. Consider (a1, a2) ∈ Z2. Using Proposition 2.10 and Proposition 2.19 we may con-
clude for (a1, a2) ∈ Z2
≥0 that the combinatorial element defined by (2.12) satisfies (2.10) and thus that
it is actually the greedy element x[a1, a2]. For (a1, a2) ∈ Z2
<0 both terms in (2.10) are always zero and
D[a1]+,[a2]+ = D0,0, thus the combinatorially defined element and the greedy element x[a1, a2] are both equal
to the initial cluster monomial x−a1
.
1 x−a2
2
Now suppose a1 > 0 and a2 < 0. Notice that the first term in (2.10) is always zero for a2 ≤ 0. Using the
second term in (2.10) we see that c(p, 0) = 0 for all p > 0 and thus c(p, q) = 0 whenever p > 0. Comparing
with (2.8) we may conclude that x[a1, a2] = x−a2
2 x[a1, 0], where we note that x[a1, 0] has already identified
with the combinatorial element defined by (2.12). Using Proposition 2.17 we may conclude that x[a1, 0] = xa1
3
and thus x[a1, a2] = x−a2
3 = σ2(x[−a1, a2]) coincides with the combinatorially defined element.
2 xa1
The case a1 < 0 and a2 > 0 follows by a similar argument.
(cid:3)
We are now ready to complete the proof of our main theorem.
Proof of Theorem 1.1. The existence of greedy elements is the content of Theorem 2.15, uniqueness and
indecomposability follow from the definition (2.10). This gives parts (a) and (b).
In the proof of Theorem 2.15 above we have seen that x[a1, a2] = z1[a1, a2] is a standard monomial for
(a1, a2) ∈ Z2 \ Z2
pointed elements in A(P1, P2), thus part (c) follows from Proposition 2.9.
>0, i.e. the set (cid:8)x[a1, a2](cid:9)a1,a2∈Z of greedy elements is a complete bounded collection of
To see parts (d) and (e) we note that x[a1, a2] = x−a1
is an initial cluster monomial for every
(a1, a2) ∈ Z2
<0. Since any other cluster monomial can be obtained from an initial cluster monomial by a
reflection σp for some p ∈ Z (or better, by alternately applying σ1 and σ2), we see that both part (d) and
(e) follow from Proposition 2.17.
(cid:3)
1 x−a2
2
To complete this section we describe explicitly which greedy elements correspond to cluster monomials.
To do so we introduce two-parameter Chebyshev polynomials uk,j (k, j ∈ Z) defined recursively by:
u−1,j = 0,
u0,j = 1,
uk+1,j+1 = dj uk,j − uk−1,j−1 where dj =(d1,
d2,
if j is odd;
if j is even.
Remark 2.20. The two-parameter Chebyshev polynomials can be seen as the components of positive roots
in the rank two root system associated to the Cartan matrix (cid:18) 2
fact we leave the verification to the reader.
−d2
−d1
2 (cid:19). Since we will not need this
Using Proposition 2.17 and the definition of the Chebyshev polynomials we easily see that the cluster
variables are given by
(2.19)
xk =(x[uk−3,1, uk−4,2]
x[u−k−1,1, u−k,2]
for k ≥ 2;
for k ≤ 1.
Applying a sequence of reflections σ1 and σ2 we can transform
(x1, x2) 7→((xk, xk+1)
(xk+1, xk)
for odd k;
for even k.
Applying the same sequence of reflections to the initial cluster monomial x[a1, a2] = x−a1
Z2
≤0 and applying (2.19) we see that the kth cluster monomial zk[a1, a2] is given by
1 x−a2
2
for (a1, a2) ∈
(2.20)
zk[a1, a2] = x−a1
k x−a2
k+1 =
x[−a1uk−3,1 − a2uk−2,1, −a1uk−4,2 − a2uk−3,2]
x[a1, a2]
x[−a1u−k−1,1 − a2u−k−2,1, −a1u−k,2 − a2u−k−1,2]
for k ≥ 2;
for k = 1;
for k ≤ 0.
Combining (2.19) and (2.20) we get the following factorization properties for greedy elements corresponding
to cluster monomials:
Rank 2 Generalized Greedy Bases
13
x[a1uk−3,1 + a2uk−2,1, a1uk−4,2 + a2uk−3,2] = x[uk−3,1, uk−4,2]a1 x[uk−2,1, uk−3,2]a2
x[a1u−k−1,1 + a2u−k−2,1, a1u−k,2 + a2u−k−1,2] = x[u−k−1,1, u−k,2]a1 x[u−k−2,1, u−k−1,2]a2
(2.21)
for (a1, a2) ∈ Z2
(2.22)
for (a1, a2) ∈ Z2
≥0 and k ≥ 2;
≥0 and k ≤ 0.
Remark 2.21. Consider any Dyck path D of the form
Da1uk−3,1+a2uk−2,1,a1uk−4,2+a2uk−3,2
or Da1u−k−1,1+a2u−k−2,1,a1u−k,2+a2u−k−1,2
giving rise to a cluster monomial. Using (2.12), the factorization equations above imply that the "wrap
around" of subpaths in D assumed in Section 3 can be dropped when considering compatible pairs. In other
words, there is a principle of non-interaction between the various subpaths of D of the form
or
Duk−3,1,uk−4,2
and Duk−2,1,uk−3,2
Du−k−1,1,u−k,2
and Du−k−2,1,u−k−1,2 ,
i.e. when considering compatible pairs we may view compatible pairs on each such subpath independent of
all other such subpaths of D.
3. Structure of Maximal Dyck Paths
For non-negative integers a1, a2 ∈ Z≥0, let R = Ra1,a2 denote the rectangle in R2 with corner vertices
(0, 0) and (a1, a2). A Dyck path in R is a lattice path in Z2 ⊂ R2 beginning at (0, 0), taking North and East
steps, and ending at (a1, a2), where the path never crosses above the main diagonal of R. We will consider
the maximal Dyck path D = Da1,a2 which lies closest to (and possibly touches) the main diagonal of R, i.e.
any lattice point above D is also above the main diagonal.
Label the horizontal and vertical edges of D as D1 = {h1, . . . , ha1} and D2 = {v1, . . . , va2} respectively,
where edges are labeled by their maximum distance from the vertical and horizontal axes respectively. We
will call the distance from a horizontal edge to the horizontal axis its height and the distance from a vertical
edge to the vertical axis its depth. It will be convenient to fold the rectangle R into a torus and identify the
vertices (0, 0) and (a1, a2), in this way D becomes a closed loop. Thus we will sometimes allow arbitrary
integer subscripts hj and vj with the understanding that hj = hj+a1 and vj = vj+a2 for j ∈ Z.
For edges e, e′ ∈ D we will let ee′ denote the subpath of D beginning at e and ending at e′. We consider
ee to be the path which only contains the edge e. Note that when e lies to the North-East of e′ the path ee′
will contain the vertex (0, 0) ≡ (a1, a2) and "wrap around" D. We will write (ee′)1 for the set of horizontal
edges in the subpath ee′ and (ee′)2 for its set of vertical edges. It will also be convenient to denote by ee′,
ee′, and ee′ the paths obtained from ee′ by removing the edge e, removing the edge e′, or removing both
edges e and e′ respectively.
Example 3.1. For (a1, a2) = (5, 2) the maximal Dyck path D has horizontal edges D1 = {h1, h2, h3, h4, h5}
and vertical edges D2 = {v1, v2} which can be visualized as
v2
h5
h4
v1
.
h1
h2
h3
Then (h3v2)1 = {h3, h4, h5} and (h3v2)2 = {v1, v2}, while (v2h3)1 = {h1, h2, h3} and (v2h3)2 = {v2}. The
path h3h3 has length 1, while v1h3 = D has length 7.
We will often need the following easy but incredibly useful Lemma presented in [LLZ] which precisely
describes the Dyck path D.
14
D. RUPEL
Lemma 3.2.
(a) The height of the horizontal edge hj is ⌊(j − 1)a2/a1⌋, and so the vertical distance (hihj)2 between
hi and hj for 1 ≤ i < j ≤ a1 is equal to
(hihj)2 = ⌊(j − 1)a2/a1⌋ − ⌊(i − 1)a2/a1⌋.
(b) The depth of the vertical edge vj is ⌈ja1/a2⌉, and so the horizontal distance (vivj)1 between vi and
vj for 1 ≤ i < j ≤ a2 is equal to
(vivj)1 = ⌈ja1/a2⌉ − ⌈ia1/a2⌉.
In particular, this implies the following result giving a bound on the slope of certain subpaths of D in
relation to the slope of the main diagonal of R.
Corollary 3.3. For any h ∈ D1 and vj ∈ D2 with h to the left/below vj we have a1(cid:0)(hvj )2−1(cid:1) < a2(hvj)1.
Proof. Suppose h has height j′ − 1 < j. Then (hvj )1 ≥ (vj′ vj )1 + 1 and (hvj)2 − 1 = j − j′. Then by
Lemma 3.2 we have
a2(hvj)1 ≥ a2(cid:0)(vj′ vj)1 + 1(cid:1) = a2(cid:0)⌈ja1/a2⌉ − ⌈j′a1/a2⌉ + 1(cid:1)
> a2(ja1/a2 − j′a1/a2) = a1(j − j′) = a1(cid:0)(hvj )2 − 1(cid:1).
(cid:3)
4. Combinatorics of Graded Compatible Pairs: Shadows
Here we introduce our main combinatorial object of study, namely graded compatible pairs. We will begin
with general results that do not require an upper bound on the gradings and restrict to the bounded case
when it becomes necessary. Most of the results in this section are generalizations of, or are inspired by,
similar results from [LLZ, Section 3]. Let D = Da1,a2 be any maximal Dyck path with horizontal edges D1
and vertical edges D2.
4.1. Unbounded Gradings.
Definition 4.1. Consider functions S1 : D1 → Z≥0 and S2 : D2 → Z≥0 which we call horizontal (resp.
vertical) gradings. We call the pair (S1, S2) (graded) compatible if for every h ∈ D1 and v ∈ D2 there exists
an edge e ∈ D so that at least one of the following conditions on the gradings is satisfied:
(HGC)
(VGC)
he is a proper subpath of hv
and
ev is a proper subpath of hv
and
(he)2 = Xh′∈(he)1
(ev)1 = Xv′∈(ev)2
S1(h′);
S2(v′).
Write C = Ca1,a2 for the set of all such compatible pairs (S1, S2).
Remark 4.2. For h ∈ D1 with S1(h) = 0 the horizontal grading condition (HGC) of Definition 4.1 is
trivially satisfied for all v ∈ D2 by taking e = h. Similarly for v ∈ D2 with S2(v) = 0 the vertical grading
condition (VGC) is trivially satisfied.
Example 4.3. We continue Example 3.1. Here we will consider bounded gradings S1 : D1 → {0, 1, 2} and
S2 : D2 → {0, 1, 2, 3} on the Dyck path D. All compatible pairs are given below, we draw all horizontal
gradings compatible with a given vertical grading where for example " 012" written over an edge h means
S1(h) can be either 0, 1, or 2 and the pair (S1, S2) will be compatible:
012 012 0
012 012 0
012 012 0
012 012 0
012 012 012 0
(1 + x2 + x2
2)5
012 012 0 1
x1(1 + x2 + x2
2)4
012 0
0 2
x2
1(1 + x2 + x2
2)3
0
0
0 3
x3
1(1 + x2 + x2
2)2
Rank 2 Generalized Greedy Bases
15
012 0 1
012 0 1
012 0 1
012 0 1
012 012 012 0
x1(1 + x2 + x2
2)4
012 012 0 1
x2
1(1 + x2 + x2
2)3
012 0
0 2
x3
1(1 + x2 + x2
2)2
0
0 3
0
1(1 + x2 + x2
x4
2)
0
0 2
0
0 2
0
0 2
0
0 2
012 012 012 0
x2
1(1 + x2 + x2
2)3
012 012 0 1
x3
1(1 + x2 + x2
2)2
012 0
0 2
1(1 + x2 + x2
x4
2)
0
0 3
0
0 3
0
0 3
012 012 01 0
x3
1(1 + x2 + x2
2)2(1 + x2)
0 1
012 01
x4
1(1 + x2 + x2
2)(1 + x2)
01
0 2
0
x5
1(1 + x2)
0
0
0
0
0 3
x5
1
0 3
x6
1
0
0 3
.
We will illustrate the compatibility of S1 = (2, 1, 0, 0, 0) and S2 = (1, 3), where we have written
S1 =(cid:0)S1(h1), S1(h2), S1(h3), S1(h4), S1(h5)(cid:1)
and
S2 =(cid:0)S2(v1), S2(v2)(cid:1).
As in Remark 4.2 the condition (HGC) is automatically satisfied for the horizontal edges h3, h4, h5 and each
vertical edge. The condition (HGC) cannot be satisfied for h1 and any vertical edge. The condition (VGC)
is satisfied for h1 and v1 by taking e = h3, while the condition (VGC) is satisfied for h1 and v2 by taking
e = h2. The condition (HGC) cannot be satisfied for h2 and v1, but the condition (VGC) is satisfied by
taking e = h3. The condition (HGC) is satisfied for h2 and v2 by taking e = v1, while the condition (VGC)
cannot be satisfied.
For a horizontal grading S1 : D1 → Z≥0, write C(S1) for the set of all vertical gradings S2 : D2 → Z≥0
with (S1, S2) ∈ C. Define C(S2) similarly. We define the magnitude of a horizontal grading S1 or a vertical
S2(v). The following result greatly simplifies
grading S2 respectively by S1 = Ph∈D1
the check for membership in C(S1) and C(S2).
S1(h) and S2 = Pv∈D2
Lemma 4.4.
(a) For every S1 : D1 → Z≥0 there exist subsets rsh(S1) ⊂ sh(S1) ⊂ D2 such that:
(i) for S2 : D2 → Z≥0, S2 ∈ C(S1) if and only if S2(v) = 0 for all v ∈ sh(S1) \ rsh(S1) and one of
the conditions (HGC) or (VGC) is satisfied for each h ∈ supp(S1) and v ∈ rsh(S1);
(ii) sh(S1) = min(a2, S1).
(b) For every S2 : D2 → Z≥0 there exist subsets rsh(S2) ⊂ sh(S2) ⊂ D1 such that
(i) for S1 : D1 → Z≥0, S1 ∈ C(S2) if and only if S1(h) = 0 for all h ∈ sh(S2) \ rsh(S2) and one of
the conditions (HGC) or (VGC) is satisfied for each h ∈ rsh(S2) and v ∈ supp(S2);
(ii) sh(S2) = min(a1, S2).
We will refer to the sets sh(Si) and rsh(Si) respectively as the shadow and remote shadow of Si (i = 1, 2).
Before proving Lemma 4.4 we require some preparation. The shadow of each Si will be defined in terms of
certain "local shadows".
Definition 4.5.
(a) For each h ∈ D1, define sh(h; S1), the local shadow of S1 at h, as the set of vertical edges (he)2 in
S1(h′). Write D(h; S1) = he for this minimal
path. If there is no such subpath then we set sh(h; S1) = D2 and D(h; S1) = D. We define the shadow
the shortest subpath he of D such that (he)2 = Ph′∈(he)1
of S1 as the union of local shadows: sh(S1) = Sh∈D1
the shortest subpath ev of D such that (ev)1 = Pv′∈(ev)2
of S2 as the union of local shadows: sh(S2) = Sv∈D2
sh(h; S1).
sh(v; S2).
(b) For each v ∈ D2, define sh(v; S2), the local shadow of S2 at v, as the set of horizontal edges (ev)1 in
S2(v′). Write D(v; S2) for this minimal path
ev. If there is no such subpath then we set sh(v; S2) = D1 and D(v; S2) = D. We define the shadow
16
D. RUPEL
Remark 4.6. When S1(h) = 0, we have sh(h; S1) = ∅ and D(h; S1) = hh is the subpath of D consisting of
only the edge h. A similar claim holds if S2(v) = 0.
It will be useful to consider certain shadow statistics for subpaths of D which are closely related to the
compatibility conditions for a pair (S1, S2) ∈ C.
Definition 4.7. For a subpath ee′ ⊂ D define the horizontal shadow statistic
for each horizontal grading S1 and the vertical shadow statistic
fS1(ee′) := −(ee′)2 + Xh∈(ee′)1
S1(h)
fS2(ee′) := −(ee′)1 + Xv∈(ee′)2
S2(v)
for each vertical grading S2.
It immediately follows from the definitions that the functions fS1 and fS2 satisfy the following additivity
properties with respect to concatenation of paths:
fS1(ee′′′) = fS1(ee′) + fS1(e′′e′′′),
fS2(ee′′′) = fS2(ee′) + fS2(e′′e′′′),
whenever ee′′′ = ee′ ∐ e′′e′′′. The next lemma relates the minimal shadow paths D(h; S1) and D(v; S2) to
the shadow statistics.
Lemma 4.8.
(a) Suppose h ∈ supp(S1) and write D(h; S1) = he.
(i) fS1(D(h; S1)) = 0.
(ii) For any proper subpath he′ ⊂ D(h; S1) we have fS1(he′) > 0.
(iii) For any proper subpath e′e ⊂ D(h; S1) we have fS1(e′e) < 0.
In particular, one may conclude that the last edge e of D(h; S1) is vertical.
(b) Suppose v ∈ supp(S2) and write D(v; S2) = ev.
(i) fS2(D(v; S2)) = 0.
(ii) For any proper subpath e′v ⊂ D(v; S2) we have fS2(e′v) > 0.
(iii) For any proper subpath ee′ ⊂ D(v; S2) we have fS2(ee′) < 0.
In particular, one may conclude that the first edge e of D(v; S2) is horizontal.
Proof. We will prove (a), the proof of (b) is similar. Recalling the definition of D(h; S1) establishes (i).
Consider the value of fS1(he′) as e′ traverses the path D(h; S1) from left to right. When e′ = h, we have
fS1(hh) = S1(h) > 0 since h ∈ supp(S1). As e′ moves to the right, fS1(he′) will either increase, stay constant,
or decrease by 1. It follows that fS1(he′) will remain positive until it first reaches zero, at which point we
must have e′ = e by minimality. This proves (ii), (iii) is an immediate consequence of (i) and (ii) using the
additivity property of fS1.
(cid:3)
The next easy result is an immediate consequence of Lemma 4.8 but will be useful in the proofs of
Proposition 4.20 and Proposition 4.22 below.
Corollary 4.9. Suppose h ∈ supp(S1) and let v ∈ D(h; S1) be any vertical edge. Then (HGC) is not
satisfied for h and v, however for any S2 ∈ C(S1) the condition (VGC) is satisfied for h and v. In particular,
D(v; S2) is a proper subpath of hv for each S2 ∈ C(S1).
Proof. By Lemma 4.8 we have fS1(he) > 0 for every edge e ∈ hv and thus (HGC) cannot be satisfied. Since
S2 ∈ C(S1) the condition (VGC) must be satisfied.
(cid:3)
We can now understand the relationships between the various local shadows associated to a horizontal
grading S1 or a vertical grading S2.
Lemma 4.10.
(a) Let S1 be a horizontal grading. For any horizontal edges h, h′ ∈ D1, the local shadows sh(h; S1) and
sh(h′; S1) are either disjoint or one is contained in the other.
Rank 2 Generalized Greedy Bases
17
(b) Let S2 be a vertical grading. For any vertical edges v, v′ ∈ D2, the local shadows sh(v; S2) and sh(v′; S2)
are either disjoint or one is contained in the other.
Proof. We will prove (1), the proof of (2) is similar. The proof will be in terms of the local shadow paths
D(h; S1) and D(h′; S1).
For h = h′ the claim is trivial. If either D(h; S1) = D or D(h′; S1) = D the claim is again trivially true,
thus we assume that both D(h; S1) and D(h′; S1) are proper subpaths of D.
Suppose D(h; S1) and D(h′; S1) intersect but not in a proper containment. Then either h ∈ D(h′; S1)
or h′ ∈ D(h; S1), without loss of generality assume h′ ∈ D(h; S1) = he. Then e ∈ D(h′; S1) since there is
not a proper containment of subpaths. Denote by e′ the edge of D immediately preceding h′. It follows
from Lemma 4.8 that fS1(he′) > 0 and fS1(h′e) > 0. But notice that the path he is just the concatenation
of he′ and h′e, in particular adding the two inequalities above gives fS1(he) = fS1(he′) + fS1(h′e) > 0,
contradicting the definition of D(h; S1).
(cid:3)
Definition 4.11.
(a) The remote shadow of S1 is the set rsh(S1) obtained from sh(S1) by removing for each d ∈ [1, a1] the
(up to) S1(hd) vertical edges of depth d immediately following hd. By considering the definition of
compatibility one may alternatively described rsh(S1) ⊂ sh(S1) as the subset consisting of all vertical
edges v for which there exists a vertical grading S2 ∈ C(S1) with S2(v) 6= 0.
(b) The remote shadow of S2 is the set rsh(S2) obtained from sh(S2) by removing for each ℓ ∈ [1, a2] the
(up to) S2(vℓ) horizontal edges of height ℓ−1 immediately preceding vℓ. By considering the definition of
compatibility one may alternatively describe rsh(S2) ⊂ sh(S2) as the subset consisting of all horizontal
edges h for which there exists a horizontal grading S1 ∈ C(S2) with S1(h) 6= 0.
We are now ready to prove Lemma 4.4. We will explain part (a), part (b) follows by a similar argument.
Proof of Lemma 4.4.a. Fix a horizontal grading S1. We have constructed the sets rsh(S1) and sh(S1). Thus
it remains to check that they satisfy the desired conditions (i) and (ii). Suppose S2 ∈ C(S1). From the
definition of rsh(S1) we must have S2(v) = 0 for all v ∈ sh(S1) \ rsh(S1). Since S1 and S2 are compatible we
see that either (HGC) or (VGC) is satisfied for every pair h ∈ D1 and v ∈ D2, in particular for h ∈ supp(S1)
and v ∈ rsh(S1). This proves the forward implication of (i).
Now suppose S2 : D2 → Z≥0 satisfies S2(v) = 0 for all v ∈ sh(S1) \ rsh(S1) and one of the conditions
(HGC) or (VGC) is satisfied for each h ∈ supp(S1) and v ∈ rsh(S1). As in Remark 4.2 we know that, when
S1(h) = 0, (HGC) is satisfied for h and each v ∈ D2. Thus to check compatibility we may restrict attention
to h ∈ supp(S1).
Consider the path D(h; S1) = he. Using the definition of sh(S1) as a union of local shadows we see
that any v ∈ D2 \ sh(S1) lies outside the path he, in other words the edge e is contained in the path hv.
Thus (HGC) is satisfied for h and v using exactly this edge e. Thus we have further reduced the check for
compatibility to h ∈ supp(S1) and v ∈ sh(S1).
Furthermore since S2(v) = 0 for v ∈ sh(S1) \ rsh(S1), it again follows from Remark 4.2 that (VGC) is
trivially satisfied for all h ∈ supp(S1) and v ∈ sh(S1) \ rsh(S1). Thus we only need one of the conditions
(HGC) or (VGC) for h ∈ supp(S1) and v ∈ rsh(S1), but this is guaranteed by assumption. Thus S2 ∈ C(S1),
establishing the reverse implication of (i).
We now move on to proving (ii). If there exists an edge h ∈ D1 with sh(h; S1) = D2 then fS1(he) ≥ 0
for every edge e. In particular, when e is the edge immediately preceding h in D we have sh(S1) = a2 =
S1(h′) = S1. This establishes (ii) in this case.
(he)2 ≤ Ph′∈(he)1
Now suppose sh(h; S1) 6= D2 for all h ∈ D1. It follows from Lemma 4.10 that for any fixed horizontal
grading S1 we can find horizontal edges η1, . . . , ηp ∈ supp(S1) (labeled in the natural order along the Dyck
path D) such that
• sh(ηi; S1) ∩ sh(ηj ; S1) = ∅ for i 6= j;
• sh(S1) = sh(η1; S1) ∐ · · · ∐ sh(ηp; S1) is a partition of sh(S1).
18
D. RUPEL
Define vertical edges ν1, . . . , νp by D(ηj ; S1) = ηjνj. Then each horizontal edge h ∈ supp(S1) is contained
in one of the maximal local shadows sh(ηj; S1) and thus we see that
sh(S1) = sh(η1; S1) + · · · + sh(ηp; S1) =
(ηjνj)2 =
p
Xj=1
p
Xj=1 Xh′∈(ηj νj )1
S1(h′) = S1,
Since sh(S1) ⊂ D2, we have S1 = sh(S1) ≤ a2. This completes the proof of (ii).
(cid:3)
It will be useful to partition the remote shadows according to which local shadow contains a given edge.
Definition 4.12.
(a) For 0 < j ≤ a1 and 0 < d ≤ a1, denote by rsh(S1)j;d the set of v ∈ rsh(S1) of depth d such
that v ∈ sh(hj; S1) and hj is the first edge before v with this property (i.e. the path hjv is shortest
(b) For 0 < k ≤ a2 and 0 ≤ ℓ < a2, denote by rsh(S2)k;ℓ the set of h ∈ rsh(S2) of height ℓ such that
h ∈ sh(vk; S2) and vk is the first edge after h with this property (i.e. the path hvk is shortest possible).
possible). Define the local remote shadow of the edge hj as rsh(hj; S1) = `d∈[1,a1]
Define the local remote shadow of the edge vk as rsh(vk; S2) = `ℓ∈[0,a2−1]
rsh(S1)j;d.
rsh(S2)k;ℓ.
Remark 4.13. For a horizontal grading S1 there are only finitely many j with rsh(hj; S1) 6= ∅. Thus there
exists a sequence 1 ≤ j1 < · · · < jn ≤ a1 such that rsh(hjt ; S1) 6= ∅ for each t ∈ [1, n] and there is a partition
rsh(S1) = rsh(hj1 ; S1) ∐ · · · ∐ rsh(hjn ; S1).
The next lemma gives precise conditions describing when the remote shadows rsh(S1)j;d and rsh(S2)j;ℓ
are non-empty. First note that rsh(S1)d;d = ∅ for every d and rsh(S2)ℓ+1;ℓ = ∅ for every ℓ.
Lemma 4.14.
(a) Let 0 < d, j ≤ a1 with j 6= d. Then rsh(S1)j;d 6= ∅ if and only if fS1(hhd+1) < 0 < fS1(hjh) for every
horizontal edge h ∈ (hjhd+1)1.
(b) Let 0 ≤ ℓ < a2 and 0 < j ≤ a2 with j 6= ℓ + 1. Then rsh(S2)j;ℓ 6= ∅ if and only if fS2(vℓv) < 0 <
fS2(vvj) for every vertical edge v ∈ (vℓvj)2.
Proof. We will establish (a), the proof of (b) is similar.
If rsh(S1)j;d 6= ∅ then hjhd is a proper subpath of D(hj; S1). So Lemma 4.8 implies 0 < fS1(hje) for every
edge e ∈ hjhd, taking e to be horizontal gives half of the forward implication.
Let v ∈ rsh(S1)j;d and suppose for sake of contradiction that there exists a horizontal edge h ∈ (hjhd+1)1
If hℓ /∈ supp(S1) then
such that fS1(hhd+1) ≥ 0. Let ℓ ≤ d be the largest so that fS1(hℓhd+1) ≥ 0.
fS1(hℓ+1hd+1) ≥ fS1(hℓhd+1) ≥ 0 contradicting the maximality of hℓ. Thus we must have S1(hℓ) 6= 0, in
particular D(hℓ; S1) is a non-trivial subpath of D. If hℓhd+1 is a subpath of D(hℓ; S1), then v ∈ D(hℓ; S1)
contradicting the minimality of the path hjv. Thus D(hℓ; S1) must be a proper subpath of hℓhd+1.
Write e′ for the first edge after D(hℓ; S1), since fS1(cid:0)D(hℓ; S1)(cid:1) = 0 additivity gives fS1(e′hd+1) ≥ 0. If
D(hℓ; S1) does not contain hd, we may move to the right from e′ until reaching a horizontal edge hℓ′ which
will satisfy fS1(hℓ′hd+1) ≥ fS1(e′hd+1) ≥ 0, this again contradicts the maximality of hℓ. Thus D(hℓ; S1)
must contain hd. But then every edge in the path e′hd+1 is vertical which is absurd since fS1(e′hd+1) ≥ 0.
This contradiction shows that we must have fS1(hhd+1) < 0 for every horizontal edge h ∈ (hjhd+1)1.
Now suppose rsh(S1)j;d = ∅. This can happen in one of two ways:
• the local shadow sh(hj; S1) is "too short";
• for each v ∈ sh(hj; S1) of depth d there exists j′ = j′(v) so that the edge hj′ is closer to v than hj
with v ∈ sh(hj′ ; S1).
The local shadow sh(hj; S1) is "too short" if sh(hj; S1) does not contain any vertical edges of depth d.
Then we have fS1(hje) = 0 where D(hj; S1) = hje with e a vertical edge to the left of hd. Taking hj′ ∈ hjhd
to be the first horizontal edge to the right of e will give fS1(hjhj′ ) ≤ 0, establishing the reverse implication
in this case.
Thus we may assume for each v ∈ sh(hj; S1) of depth d the existence of a number j′ > j such that
v ∈ sh(hj′ ; S1). Take v to be the last edge of depth d in sh(hj; S1). By Lemma 4.8.a.ii we have fS1(hjhj′ ) > 0
Rank 2 Generalized Greedy Bases
19
and fS1(hj′ v) ≥ 0, so by additivity we have fS1(hjv) > 0. This inequality implies that a vertical edge of depth
d which comes after v is also contained in sh(hj; S1), but such cannot exist by the choice of v. This implies
hj′ hd+1 = hj′ v and so fS1(hj′ hd+1) = fS1(hj′ v) ≥ 0, completing the proof of the reverse implication.
(cid:3)
Remark 4.15. The inequalities above imply that when rsh(S1)j;d 6= ∅ the set rsh(S1)j′;d′ will be empty
whenever j < j′ < d < d′.
Now we are able to give an explicit formula for the sizes of the remote shadows.
Lemma 4.16.
(a) Let 0 < d, j ≤ a1 with d 6= j and suppose fS1(hhd+1) < 0 < fS1(hjh) for every horizontal edge
(b) Let 0 ≤ ℓ < a2 and 0 < j ≤ a2 with ℓ 6= j and suppose fS2(vℓv) < 0 < fS2(vvj) for every vertical edge
h ∈ (hjhd+1)1. Then rsh(S1)j;d =
min
h∈(hj hd+1)1
v ∈ (vℓvj)2. Then rsh(S2)j;ℓ = min
v∈(vℓvj )2
min(cid:0) − fS1(hhd+1), fS1(hjh)(cid:1).
min(cid:0) − fS2(vℓv), fS2(vvj)(cid:1).
Proof. As always, we will prove (a) and leave the adaptation of the proof to (b) for the reader.
By Lemma 4.14 the remote shadow rsh(S1)j;d is non-empty. Consider any element v ∈ rsh(S1)j;d. Since
v has depth d we have
(hjhd)2 ≤ (hjv)2 ≤ (hjhd+1)2.
For each h ∈ (hjhd+1)1 we have v /∈ sh(h; S1) but v ∈ sh(hj; S1) so that fS1(hv) < 0 and fS1(hjv) ≥ 0.
Expanding the definition of fS1 gives
(hjh)2 + Xh′∈(hhd+1)1
S1(h′) < (hjh)2 + (hv)2 = (hjv)2 ≤ Xh′∈(hj hd+1)1
S1(h′)
for every h ∈ (hjhd+1)1. Combining the two displayed sequences of inequalities above we get
for each v ∈ rsh(S1)j;d. Thus the number of possible v ∈ rsh(S1)j;d is equal to
h∈(hj hd+1)1
max
rsh(S1)j;d = min
= min
(hjh)2 + Xh′∈(hhd+1)1
S1(h′)
(hjhd+1)2 , Xh′∈(hj hd+1)1
(hjhd+1)2 , Xh′∈(hj hd+1)1
(hjhd+1)2 , Xh′∈(hj hd+1)1
< (hjv)2 ≤ min
h∈(hj hd+1)1
S1(h′)
− max
h∈(hj hd+1)1
S1(h′)
+
min
(hhd+1)2 − Xh′∈(hhd+1)1
S1(h′) , −(hjh)2 + Xh′∈(hj h)1
min(cid:0)−fS1(hhd+1), fS1(hjh)(cid:1) .
(hjh)2 + Xh′∈(hhd+1)1
−(hjh)2 − Xh′∈(hhd+1)1
S1(h′)
min
=
=
min
h∈(hj hd+1)1
min
h∈(hj hd+1)1
S1(h′)
S1(h′)
S1(h′)
(cid:3)
a2m}.
4.2. Bounded Gradings. Fix a vertical grading S2 and suppose r ≥ max{S2(v) : v ∈ D2} ∪ {l a1
ra2−a1} and vertical edges D′
Write D′ = Dra2−a1,a2 with horizontal edges D′
We identify the vertical edges of D and the vertical edges of D′ via a map ϕr : D′
ϕr(v′
is well-defined (non-negative) since S2(vj) is always bounded above by r.
j ) = va2+1−j. Define the vertical grading ϕ∗
2 = {v′
a2}.
2 → D2 given by
j ) = r − S2(va2+1−j), it
2 → Z≥0 by the rule ϕ∗
1, . . . , h′
1, . . . , v′
rS2 : D′
1 = {h′
rS2(v′
Lemma 4.17. For any v′
i, v′
j ∈ D′
2 we have fϕ∗
r S2(v′
iv′
j ) = −fS2(va2−jva2−i).
20
D. RUPEL
Proof. This follows from a direct calculation using the definitions and Lemma 3.2:
fϕ∗
r S2(v′
iv′
iv′
iv′
j )2
ϕ∗
iv′
j)1
ϕ∗
rS2(v′) − (v′
j ) = Xv′∈(v′
= Xv′∈(v′
rS2(v′) −(cid:0) ⌈j(ra2 − a1)/a2⌉ − ⌈i(ra2 − a1)/a2⌉(cid:1)
Xv∈(va2+1−j va2+1−i)2(cid:0)r − S2(v)(cid:1) − ⌈jr − ja1/a2⌉ + ⌈ir − ia1/a2⌉
Xv∈(va2+1−j va2 +1−i)2
S2(v) − ⌈−ja1/a2⌉ + ⌈−ia1/a2⌉
= −
j )2
=
= − Xv∈(va2 −j va2 −i)2
= − Xv∈(va2 −j va2 −i)2
= −fS2(va2−jva2−i).
S2(v) + ⌈(a2 − i)a1/a2⌉ − ⌈(a2 − j)a1/a2⌉
S2(v) + (va2−jva2−i)1
Corollary 4.18. Suppose 0 ≤ ℓ < a2 and 0 < j ≤ a2 with ℓ 6= j. Then we have
rsh(S2)j;ℓ = rsh(ϕ∗
rS2)a2−ℓ;a2−j.
Proof. According to Lemma 4.16 and Lemma 4.17 we have
rsh(S2)j;ℓ = min
v∈(vℓvj )2
min (−fS2(vℓv), fS2 (vvj))
=
v′∈(v′
= rsh(ϕ∗
min
a2 −j v′
rS2)a2−ℓ;a2−j.
min(cid:0)fϕ∗
a2−ℓ)2
r S2(v′v′
a2−ℓ), −fϕ∗
r S2(v′
(cid:3)
(cid:3)
a2−jv′)(cid:1)
Thus for each ℓ and j we may define a bijection θj;ℓ : rsh(S2)j;ℓ → rsh(ϕ∗
rS2)a2−ℓ;a2−j which preserves
the natural left-to-right order. Following Lemma 4.4 we will define Crs(S2) ⊂ C(S2) to be the subset of
horizontal gradings S1 compatible with S2 whose support is contained in the remote shadow of S2, i.e.
supp(S1) ⊂ rsh(S2). Now we can define a map Ω : Crs(S2) → Crs(ϕ∗
rS2) as follows:
(4.1)
Ω(S1)(h′) =(0
S1(h)
1 \ rsh(ϕ∗
if h′ ∈ D′
if h′ = θj;ℓ(h) for h ∈ rsh(S2)j;ℓ.
rS2);
Note that Ω admits an obvious inverse map. The following result is as much of an analogue of Lemma 4.17
as one can hope for, however it is the essential ingredient to show that Ω is well-defined, i.e. that the pair
(Ω(S1), ϕ∗
rS2) is compatible.
Lemma 4.19. Suppose h′ = θj;ℓ(h) for h ∈ rsh(S2)j;ℓ ∩ supp(S1). Then fΩ(S1)(h′v′
a2−ℓ) = fS1(hvj).
Proof. Since h ∈ rsh(S2)j;ℓ it follows from Lemma 4.10 that each horizontal edge in hvj is contained in some
rsh(S2)j′;ℓ′ where ℓ ≤ ℓ′ < j′ ≤ j. Using this we may partition the edges in supp(S1) ∩ (hvj )1 to get:
S1(h′′) − (hvj)2
fS1(hvj ) = Xh′′∈(hvj )1
S1(h′′) + Xh′′∈(vℓ+1vj )1
= Xh′′∈(hvℓ+1)1
S1(h′′) + Xj′ :
X
h′′∈(hvℓ+1)1∩rsh(S2)j;ℓ
=
S1(h′′) − (vℓ+1vj )2
ℓ<j′<j Xh′′∈rsh(S2)j′ ;ℓ
S1(h′′) + Xℓ′,j′ :
ℓ<ℓ′<j′≤j Xh′′∈rsh(S2)j′ ;ℓ′
S1(h′′) − (vℓ+1vj)2.
Applying the definition of Ω and the same logic as above this becomes
Rank 2 Generalized Greedy Bases
21
h′′∈(h′v′
a2+1−j )1∩rsh(ϕ∗
r S2)a2−ℓ;a2−j
X
Ω(S1)(h′′) + Xj′ :
ℓ<j′<j
X
h′′∈rsh(ϕ∗
r S2)a2 −ℓ;a2−j′
Ω(S1)(h′′)
X
h′′∈rsh(ϕ∗
r S2)a2 −ℓ′ ;a2−j′
Ω(S1)(h′′) − (v′
a2+1−jv′
a2−ℓ)2
ℓ<ℓ′<j′≤j
+ Xℓ′,j′ :
= Xh′′∈(h′v′
= Xh′′∈(h′v′
a2−ℓ)1
= fΩ(S1)(h′v′
a2−ℓ).
Ω(S1)(h′′) +
Xh′′∈(v′
a2 +1−j v′
a2 −ℓ)1
Ω(S1)(h′′) − (v′
a2+1−jv′
a2−ℓ)2
a2+1−j )1
Ω(S1)(h′′) − (h′v′
a2−ℓ)2
Proposition 4.20. Let S1 : D1 → Z≥0 be a horizontal grading. Then S1 ∈ Crs(S2) if and only if Ω(S1) ∈
Crs(ϕ∗
rS2).
Proof. We will prove the forward implication. The reverse implication can be obtained by a similar argument
with Ω−1.
(cid:3)
To check compatibility it suffices to consider h′ ∈ supp(Ω(S1)) ⊂ rsh(ϕ∗
h ∈ rsh(S2)j;ℓ ∩ supp(S1). The containment h′ ∈ rsh(ϕ∗
with ℓ < ℓ′ ≤ j that D(v′
for h′ and v′
satisfied for h′ and each other vertical edge v′
rS2) is a proper subpath of h′v′
a2−ℓ′ . We claim that D(h′; Ω(S1)) is a subpath of h′v′
a2−ℓ′ ; ϕ∗
a2−ℓ′ with ℓ′ ≤ ℓ or ℓ′ > j.
rS2)a2−ℓ;a2−j implies for any vertical edge v′
rS2). Suppose h′ = θj;ℓ(h) for
a2−ℓ′
a2−ℓ′ , i.e. the condition (VGC) is satisfied
a2−ℓ which implies the condition (HGC) is
Indeed, since h ∈ D(vj ; S2) each horizontal edge h′′ ∈ (hvj )1 is also contained in D(vj; S2). Then
Corollary 4.9 implies we may partition hvj into shadow paths D(h′′; S1), horizontal edges outside the support
of S1, and the remaining vertical edges. Thus we have fΩ(S1)(h′v′
a2−ℓ) = fS1(hvj ) ≤ 0. In particular, arguing
as in the proof of Lemma 4.8 we see that there exists v′ ∈ (h′v′
a2−ℓ)2 so that fΩ(S1)(h′v′) = 0, i.e. D(h′; Ω(S1))
is a subpath of h′v′
(cid:3)
a2−ℓ as desired.
4.3. Magnitudes of Bounded Gradings. Fix nonnegative integers d1, d2 ≥ 0. In this section we make
explicit the possibilities for the pair of magnitudes (S1, S2) associated to a compatible pair (S1, S2) ∈ Ca1,a2.
We assume throughout that S1(h) ≤ d1 for each h ∈ D1 and S2(v) ≤ d2 for each v ∈ D2.
Lemma 4.21. For any compatible pair (S1, S2) ∈ Ca1,a2 we must have S1 < a2 or S2 < a1.
Proof. Indeed, we will show that S1 ≥ a2 and S2 ≥ a1 is impossible. Thus we assume that S1 ≥ a2 and
will deduce S2 < a1, the other case can be proven by a similar argument.
Notice that under this assumption Lemma 4.4 says sh(S1) = a2, in other words sh(S1) = D1.
It
follows that each vertical edge v ∈ D2 is contained in some maximal local shadow. Let sh(h; S1) be such
a maximal local shadow. By Corollary 4.9, for each vertical edge v ∈ sh(h; S1) the maximal shadow path
D(v; S2) is a proper subpath of hv. It follows that h /∈ sh(S2) and thus the inequality S2 < a1 follows from
Lemma 4.4.
(cid:3)
Proposition 4.22. Let (S1, S2) ∈ Ca1,a2 be a bounded compatible pair.
(a) If 0 ≤ d2a2 ≤ a1, then (S1, S2) is contained in the (possibly degenerate) trapezoid with corner
vertices (0, 0), (d1a1, 0), (0, d2a2), and (d1a1 − d1d2a2, d2a2).
(b) If 0 ≤ d1a1 ≤ a2, then (S1, S2) is contained in the (possibly degenerate) trapezoid with corner
vertices (0, 0), (d1a1, 0), (0, d2a2), and (d1a1, d2a2 − d1d2a1).
(c) If 0 < a1 < d2a2 and 0 < a2 < d1a1, then (S1, S2) is contained in the non-convex quadrilateral with
corner vertices (0, 0), (d1a1, 0), (0, d2a2), and (a2, a1) with the convention that the boundary segments
(cid:2)(0, 0), (d1a1, 0)(cid:3) and (cid:2)(0, 0), (0, d2a2)(cid:3) are included, while the boundary segments (cid:0)(d1a1, 0), (a2, a1)(cid:3)
and (cid:0)(0, d2a2), (a2, a1)(cid:3) are excluded.
22
S2
D. RUPEL
S2
S2
(0,d2a2)
(0,d2a2)
(0,d2a2) (d1a1−d1d2a2,d2a2)
(d1a1,d2a2−d1d2a1)
(a1,a2)
(0,0)
S1
(0,0)
S1
(0,0)
(d1a1,0)
(d1a1,0)
(d1a1,0)
S1
Case (a)
Case (b)
Case (c)
Proof. There will be many cases to consider, we begin by handling the easier degenerate cases.
• Suppose a1 = a2 = 0. Then the maximal Dyck path D is a single vertex, i.e. D1 = D2 = ∅, so that
(S1, S2) = (0, 0). This verifies the result in this most degenerate of cases.
• Suppose a1 > a2 = 0. Then the maximal Dyck path D consists of a1 consecutive horizontal edges and no
vertical edges. It follows that (S1, S2) is contained in the segment (cid:2)(0, 0), (d1a1, 0)(cid:3). This verifies (a) in
this degenerate case.
• Suppose a2 > a1 = 0. Then the maximal Dyck path D consists of a2 consecutive vertical edges and no
horizontal edges. It follows that (S1, S2) is contained in the segment(cid:2)(0, 0), (0, d2a2)(cid:3). This verifies (b) in
this degenerate case.
• Suppose a1 ≥ d2a2 > 0. We first note that S1 ≤ d1a1 and S2 ≤ d2a2. Thus it only remains to justify the
boundary segment(cid:2)(d1a1, 0), (d1a1 − d1d2a2, d2a2)(cid:3). In the present case we have rsh(S2) = ∅, thus according
to Lemma 4.4 we have sh(S2) = S2 and S1(h) = 0 for each h ∈ sh(S2). It follows that S1 ≤ d1a1 − d1S2
and thus (S1, S2) lies on or to the left of the segment. This complete the proof of (a).
• Suppose a2 ≥ d1a1 > 0. By a similar argument as above we have S2 ≤ d2a2 − d2S1, from which (b)
follows.
• Suppose 0 < a1 < d2a2 and 0 < a2 < d1a1. First note that S1 ≤ d1a1 and S2 ≤ d2a2. Thus we only
need to justify the boundary segments (cid:2)(d1a1, 0), (a2, a1)(cid:3) and (cid:2)(0, d2a2), (a2, a1)(cid:3).
Following Lemma 4.21 we will assume 0 < S2 < a1 and justify the boundary segment(cid:2)(d1a1, 0), (a2, a1)(cid:3),
the other boundary segment follows by a similar argument under the assumption 0 < S1 < a2. Note that
the condition that (S1, S2) lies strictly to the left of this segment is equivalent to the inequality
(4.2)
S1 < −
d1a1 − a2
a1
S2 + d1a1.
This translates into the inequality a1S1 < d1a1(cid:0)a1 −S2(cid:1)+a2S2, where we note that the quantity a1 −S2
is exactly the number of edges outside the shadow of S2. Thus it suffices to show
(4.3)
a1S1 < a2S2 for any S1 ∈ Crs(S2).
Since sh(S2) ⊂ D1 is a proper subset, Lemma 4.10 implies that sh(S2) is a disjoint union of maximal local
shadows each of which is a proper subset of D1. Let sh(v; S2) denote such a maximal local shadow and write
D(v; S2) = ev for the corresponding minimal shadow path. Then, by Corollary 4.9, for any h ∈ (ev)1 the
local shadow path D(h; S1) is a proper subset of ev. It follows that
(4.4)
S1(ev)1(cid:12)(cid:12)(cid:12)
a1(cid:12)(cid:12)(cid:12)
≤ a1(cid:16)(ev)2 − 1(cid:17) < a2(ev)1 = a2(cid:12)(cid:12)(cid:12)
S2(ev)2(cid:12)(cid:12)(cid:12)
,
Rank 2 Generalized Greedy Bases
23
where the second inequality follows from Corollary 3.3 and the last equality is the condition (VGC) for the
edges e and v.
Since S1(h) = 0 for any horizontal edge h outside the shadow of S2 and S2(v) = 0 for any vertical edge
outside of a minimal shadow path, (4.4) implies (4.3). This completes the proof of the final case and thus
concludes the proof of Proposition 4.22.
(cid:3)
5. Appendix: Multinomial Coefficients
In this section we collect certain lesser-known analogues for multinomial coefficients of well-known iden-
tities involving binomial coefficients.
The binomial coefficients describe the coefficients when expanding powers of binomials, generalizing this
we define multinomial coefficients (cid:0)
k1,...,kr(cid:1) by
n
(5.1)
(x1 + · · · + xr)n = X(k1,...,kr )⊢n(cid:18)
n
k1, . . . , kr(cid:19)xk1
1 · · · xkr
r ,
where (k1, . . . , kr) ⊢ n denotes a partition of the positive integer n into r positive parts, i.e. k1, . . . , kr ∈ Z≥0
with k1 + · · · + kr = n. By convention, for n ≥ 0 we will take (cid:0)
will discuss n < 0 below.
k1,...,kr(cid:1) = 0 whenever ki < 0 for some i. We
To begin, the multinomial coefficients satisfy an analogue of the Pascal identity for binomial coefficients.
n
Lemma 5.1. Suppose n ≥ 1. For any partition (k1, . . . , kr) ⊢ n we have the "Pascal identity"
(cid:18)
n
k1, . . . , kr(cid:19) =(cid:18)
n − 1
k1 − 1, . . . , kr(cid:19) + · · · +(cid:18)
n − 1
k1, . . . , kr − 1(cid:19).
Proof. We apply the definition of (cid:0)
k1,...,kr(cid:1) and expand in two different ways:
n
(x1 + · · · + xr)n = (x1 + · · · + xr)(x1 + · · · + xr)n−1
1 · · · xkr
r
k1, . . . , kr(cid:19)xk1
n − 1
= (x1 + · · · + xr) X(k1,...,kr )⊢n−1(cid:18) n − 1
= X(k1,...,kr )⊢n(cid:20)(cid:18)
k1 − 1, . . . , kr(cid:19) + · · · +(cid:18)
in (x1 + · · · + xr)n is thus described by (cid:0)
n − 1
k1, . . . , kr − 1(cid:19)(cid:21) xk1
1 · · · xkr
r .
The coefficient of xk1
1 · · · xkr
r
n
k1,...,kr(cid:1), so these must be equal.
(cid:0)
n−1
k1−1,...,kr(cid:1) + · · · +(cid:0)
n−1
k1,...,kr−1(cid:1) and also
(cid:3)
Using this recursive description of the multinomial coefficients one may derive an expression in terms of
factorials, again analogous to the well-known formula for binomial coefficients.
Lemma 5.2. Suppose n ≥ 0. For any partition (k1, . . . , kr) ⊢ n we have
n
(cid:18)
k1, . . . , kr(cid:19) =
n!
k1! · · · kr!
.
Proof. We will work by induction, when n = 0 there is only the trivial partition (0, . . . , 0) ⊢r 0, where ⊢r
denotes a partition into r parts. On both sides of the desired equality we have 1, establishing the base of
the induction.
24
D. RUPEL
Suppose n > 0 and consider a partition (k1, . . . , kr) ⊢ n. Using Lemma 5.1 and the induction hypothesis
we have
n
(cid:18)
k1, . . . , kr(cid:19) =(cid:18)
n − 1
k1 − 1, . . . , kr(cid:19) + · · · +(cid:18)
n − 1
k1, . . . , kr − 1(cid:19)
= k1
(n − 1)!
k1! · · · kr!
+ · · · + kr
(n − 1)!
k1! · · · kr!
= (k1 + · · · + kr)
(n − 1)!
k1! · · · kr!
=
n!
k1! · · · kr!
,
where we remark for the reader that there is no conflict in the second equality with our convention for
nonnegative n that (cid:0)
(n−1)!
k1!···kr! are zero.
and k1
k1,...,kr(cid:1) = 0 whenever ki < 0 for some i, for example if k1 = 0 then both (cid:0)
n
n−1
k1−1,...,kr(cid:1)
(cid:3)
Corollary 5.3. For n, k ≥ 0 and (k1, . . . , kr) ⊢ k we have the following identity relating binomial coefficients
and multinomial coefficients
(cid:18)n
k(cid:19)(cid:18)
k
k1, . . . , kr(cid:19) =(cid:18)
n
n − k, k1, . . . , kr(cid:19).
Proof. This is an immediate consequence of Lemma 5.2 and the well-known identity (cid:0)n
k(cid:1) =
Recall that we can make sense of (cid:0)n
(cid:0)−n
k (cid:1) = (−1)k(cid:0)n+k−1
k0, k1, . . . , kr(cid:19) =(cid:18) n
(cid:1). Thus we can make sense of(cid:0)
n − k0(cid:19)(cid:18) n − k0
k1, . . . , kr(cid:19) = (−1)n−k0(cid:18)−k0 − 1
k(cid:1) for k ≥ 0 and n ∈ Z, i.e.
n − k0 (cid:19)(cid:18) n − k0
precisely, when n, k0 < 0 we have
k1, . . . , kr(cid:19),
k0,k1,...,kr(cid:1) for n, k0 ∈ Z by applying Corollary 5.3. More
for n ≥ 1 we have the equality
n!
k!(n−k)! .
(cid:3)
(cid:18)
n
n
k
where −k0 − 1 ≥ 0. By our conventions, the binomial coefficient(cid:0)−k0−1
the multinomial coefficient (cid:0)
k0,k1,...,kr(cid:1) is zero whenever n < k0.
Our primary goal in this section is to understand how to use multinomial coefficients to expand as power
series the negative powers of a polynomial in a single variable. The following lemma is slightly more general
but will serve our purposes here.
n−k0 (cid:1) is zero for n − k0 < 0, in particular
n
Lemma 5.4. For any n > 0 and c ∈ k× we have
(5.2)
(c + x1 + · · · + xr)−n =Xk≥0 X(k1,...,kr )⊢k
(−1)k(cid:18) n + k − 1
n − 1, k1, . . . , kr(cid:19)c−n−kxk1
1 · · · xkr
r .
Proof. We apply the standard binomial formula with negative exponents and then the definition of the
multinomial coefficients to get
(c + x1 + · · · + xr)−n = c−n(cid:0)1 + (x1 + · · · + xr)/c(cid:1)−n
(−1)k(cid:18)n + k − 1
k
(cid:19)c−n−k(x1 + · · · + xr)k
=Xk≥0
=Xk≥0 X(k1,...,kr)⊢k
=Xk≥0 X(k1,...,kr)⊢k
k
k
(−1)k(cid:18)n + k − 1
(cid:19)(cid:18)
k1, . . . , kr(cid:19)c−n−kxk1
(−1)k(cid:18) n + k − 1
n − 1, k1, . . . , kr(cid:19)c−n−kxk1
1 · · · xkr
r ,
1 · · · xkr
r
where the last equality follows from Corollary 5.3.
(cid:3)
Rank 2 Generalized Greedy Bases
25
Remark 5.5. Notice that the result of Lemma 5.4 makes sense for any n ∈ Z. Indeed, applying the definition
of the standard (positive) multinomial coefficients to (c + x1 + · · · + xr)n for n ≥ 0 we get
n
n
=
n
1 · · · xkr
(c + x1 + · · · + xr)n = X(k0,k1,...,kr )⊢n(cid:18)
k0, k1, . . . , kr(cid:19)ck0 xk1
Xk=0 X(k1,...,kr )⊢k(cid:18)
n − k, k1, . . . , kr(cid:19)cn−kxk1
=Xk≥0 X(k1,...,kr )⊢k(cid:18)n
k(cid:19)(cid:18)
k1, . . . , kr(cid:19)cn−kxk1
(−1)k(cid:18)−n + k − 1
(cid:19)(cid:18)
k1, . . . , kr(cid:19)cn−kxk1
=Xk≥0 X(k1,...,kr )⊢k
(−1)k(cid:18) −n + k − 1
=Xk≥0 X(k1,...,kr )⊢k
−n − 1, k1, . . . , kr(cid:19)cn−kxk1
1 · · · xkr
1 · · · xkr
r
r
k
k
k
r
1 · · · xkr
r
1 · · · xkr
r ,
which agrees with (5.2) as claimed.
To complete our discussion of multinomial coefficients consider a polynomial P (z) ∈ k[z] of degree d and
write P (z) = ρ0 + ρ1z + · · · + ρdzd.
Corollary 5.6. For n ∈ Z we have
P (z)n =Xk≥0 X(k1,...,kd)⊢k
(−1)kρn−k
0
1 · · · ρkd
ρk1
d (cid:18) −n + k − 1
−n − 1, k1, . . . , kd(cid:19)zk1+2k2+···+dkd .
References
[BFZ] A. Berenstein, S. Fomin, and A. Zelevinsky, Cluster algebras III: Upper bounds and double Bruhat cells. Duke Math. J.
126 (2005), no. 1, pp. 1 -- 52.
[CZ] P. Caldero and A. Zelevinsky, Laurent expansions in cluster algebras via quiver representations. Mosc. Math. J. 6 (2006),
no. 3, pp. 411 -- 429, 587.
[CLS] G. Cerulli Irelli, D. Labardini-Fragoso, and J. Schroer, Caldero-Chapoton algebras. Preprint: arXiv:1208.3310v2 (2012).
[CS] L. Chekhov and M. Shapiro, Teichmuller spaces of Riemann surfaces with orbifold points of arbitrary order and cluster
variables. Preprint: arXiv:1111.3963 (2011).
[LLZ] K. Lee, L. Li, and A. Zelevinsky, Greedy elements in rank 2 cluster algebras. Preprint: arXiv:1208.2391 (2012), to
appear in Selecta Math.
[LS1] K. Lee and R. Schiffler, Proof of a positivity conjecture of M. Kontsevich on non-commutative cluster variables. Compos.
Math. 148 (2012), no. 6, pp. 1821 -- 1832.
[LS2] K. Lee and R. Schiffler, Positivity for cluster algebras. Preprint: arXiv:1306:2415 (2013).
[R1] D. Rupel, On a quantum analogue of the Caldero-Chapoton formula. Int. Math. Res. Not. (2011), no. 14, pp. 3207 -- 3236.
[R2] D. Rupel, Proof of the Kontsevich non-commutative cluster positivity conjecture. C. R. Math. Acad. Sci. Paris 350
(2012), no. 21-22, pp. 929 -- 932.
[R3] D. Rupel, Rank 2 non-commutative Laurent phenomenon and positivity. In preparation.
[S]
[U]
R. Schiffler, A cluster expansion formula (An case). Electron. J. Combin. 15 (2008), no. 1, Research paper 64, 9 pp.
A. Usnich, Non-commutative Laurent phenomenon for two variables. Preprint arXiv:1006.1211 (2010).
Department of Mathematics, Northeastern University, Boston, MA 02115
E-mail address: [email protected]
|
1506.00510 | 1 | 1506 | 2015-06-01T14:34:16 | On the growth of graded polynomial identities of sl_n | [
"math.RA"
] | Let K be a field of characteristic 0 and L be a G-graded Lie PI-algebra, where G is a finite group. We define the graded Gelfand-Kirillov dimension of L. Then we measure the growth of the Z_n-graded polynomial identities of the Lie algebra of n x n traceless matrices sl_n(K) giving an exact value of its Z_n-graded Gelfand-Kirillov dimension. | math.RA | math | ON THE GROWTH OF GRADED POLYNOMIAL IDENTITIES
OF sln
LUCIO CENTRONE AND MANUELA DA SILVA SOUZA
Abstract. Let K be a field of characteristic 0 and L be a G-graded Lie PI-
algebra, where G is a finite group. We define the graded Gelfand-Kirillov
n-graded polynomial
dimension of L. Then we measure the growth of the Z
identities of the Lie algebra of n × n traceless matrices sl
n(K) giving an exact
value of its Z
n-graded Gelfand-Kirillov dimension.
]
.
A
R
h
t
a
m
[
1
v
0
1
5
0
0
.
6
0
5
1
:
v
i
X
r
a
1. introduction
All fields we refer to are to be considered of characteristic 0. Let A be an
algebra with non-trivial polynomial identity (or simply PI-algebra) and denote by
T (A) the T -ideal of its polynomial identities. In general the description of a T -
ideal is a hard problem. In order to overcome this difficulty one introduces some
functions measuring the growth of T (A) in some sense. For every integer k ≥ 1 one
defines the Gelfand-Kirillov (GK) dimension of A in k variables as the Gelfand-
Kirillov dimension of the relatively free algebra of rank k with respect to the ideal
of polynomial identities of A. The Gelfand-Kirillov dimension of arbitrary finitely
generated algebra is a measure of the rate of growth in terms of any finite generating
set. For any associative PI-algebra the GK dimension of A in k variables is always
an integer (see [4] and [5]). For a more detailed background about Gelfand-Kirillov
dimension see the book of Krause and Lenagan [18].
After the powerful theory developed by Kemer in the 80's in order to solve the
Specht problem, i.e., the existence of a finite generating set for any T -ideal over a field
of characteristic zero, other kinds of identities have become object of much interest
in the theory of polynomial identities. For example if A is a G-graded algebra, where
G is a group, one may be interested in the study of G-graded polynomial identities
of A. In [1] Aljadeff and Kanel-Belov proved the analog of the Specht problem in
the graded case, for G-graded PI-algebra when G is a finite group. We have to cite
the work by Sviridova [25] who solved the Specht problem for associative graded
algebras graded by a finite abelian group.
When the group G is finite, Centrone in [9] defined the G-graded GK dimension
in k graded variables for a G-graded PI-algebra. In particular if A is a G-graded PI
algebra, one may consider the relatively-free G-graded algebra of A in k variables
F G
k (A) and define the G-graded Gelfand-Kirillov dimension of A as the GK dimen-
sion of F G
k (A). In [10] Centrone computed the graded Gelfand-Kirillov dimension
2010 Mathematics Subject Classification. 17B01,17B70.
Key words and phrases. Graded lie algebras, graded identities, growth of algebras.
The first author was supported by grant from FAPESP (Proc. 2013/06752-4) of Brazil.
The second author was partially supported by grant from FAPESP (Proc. 2013/04590-7) of
Brazil.
1
2
CENTRONE AND SOUZA
of the verbally prime algebras Mn(K), Mn(E) and Ma,b(E), where K is a field
and E is the infinite dimensional Grassmann algebra, endowed with a "Vasilovsky
type"-grading (see [26]). We recall that the verbally prime algebras are the building
blocks of the theory of Kemer.
The graded polynomial identities of Lie algebras have seldom been studied. Most
of the known results about the topic are related to the algebra of 2 × 2 traceless
matrices sl2. Razmyslov found a finite basis of the ordinary identities satisfied by
sl2 and proved that the variety of Lie algebras generated by sl2 is Spechtian i.e., the
identities of any subvariety have a finite basis ([23]). Up to graded isomorphism, sl2
can only be graded by {0} , Z2, Z2 × Z2 and Z. The structure of the relatively-free
algebra for each non-trivial grading over a field of characteristic 0 was described
by Repin in [24].
In [17] Koshlukov described the graded polynomial identities
of sl2 for the above gradings when the base field is infinite and of characteristic
6= 2. Recently Giambruno and Souza proved in [15] that the variety of graded Lie
algebras generated by sl2 is also Spechtian.
In this paper we define the graded Gelfand-Kirillov dimension on a finite set
of graded variables of any G-graded Lie PI-algebra according to the associative
case. We compute explicitly the graded Gelfan-Kirillov dimension of sl2 over a
field of characteristic 0 for each of its non-trivial gradings. Next we consider the
Lie algebra sln of n × n traceless matrices endowed with the grading of Vasilovsky,
then we measure the growth of the graded polynomial identities of the Lie algebra
of n×n traceless matrices sln giving the exact value of their graded Gelfand-Kirillov
dimension. We recall that the ordinary GK dimension in k variables of the relatively
free algebra of sl2 has been already computed with two different methods in [14]
and [19].
2. Preliminaries
Throughout the paper K will be a fixed field of characteristic zero, L a Lie
algebra over K and sln = sln(K) the Lie algebra of n × n traceless matrices over K.
Moreover we refer to a left-normed product of elements l1, . . . , lk of a Lie algebra
as the product [[· · · [[l1, l2], l3] · · · ], lk].
Recall that the growth function gV (n)(A) of a finitely generated algebra A gen-
erated by the set V = {v1, . . . , vr} is the dimension of the space spanned by words
of length at most n. We say that gV (n)(A) is the growth function with respect to
the vector space generated by V over K. The Gelfand-Kirillov dimension of A is
the superior limit
GKdim(A) = lim sup
n→+∞
ln gV (n)(A)
ln n
when it exists. The Gelfand-Kirillov dimension does not depend on the choice of
the generators of algebra, then we shall use g(n)(A) instead of gV (n)(A).
From now on G will be an abelian group and L be a G-graded Lie algebra over
subspaces such that LgLh ⊆ Lg+h, for all g, h ∈ G.
K. Recall that L is a G-graded Lie algebra if L = Lg∈G Lg is a direct sum of
on the set X =Sg∈G Xg, where for any g ∈ G the sets Xg = {xg
The free G-graded Lie algebra L(X) is the G-graded Lie algebra freely generated
i ; i ≥ 1} of variables
of homogeneous degree g are infinite and disjoint. A polynomial f of L(X) is a
G-graded polynomial identity of L if f vanishes under all graded substitutions i.e.,
ON THE GROWTH OF GRADED POLYNOMIAL IDENTITIES OF sln
3
for any g ∈ G, we evaluate the variables xg
into elements of the homogeneous
i
component Lg. We denote by TG(L) the ideal of L(X) of G-graded polynomial
identities of L. It is easily checked that TG(L) is a TG-ideal i.e., an ideal invariant
under all G-graded endomorphisms of L(X). We say that L is a graded PI-algebra
if TG(L) 6= 0.
Here we shall always consider gradings on L such that the support Supp(L) =
{g ∈ G : Lg 6= 0} is finite. Suppose for instance that Supp(L) = {g1, . . . , gs}. We
denote by
LG
k (L) =
L (xg1
1 , . . . , xg1
L (xg1
1 , . . . , xg1
k , . . . , xgs
1 , . . . , xgs
k )
k , . . . , xgs
1 , . . . , xgs
k ) ∩ TG(L)
the relatively-free G-graded algebra of L in k variables. We define the G-graded
Gelfand-Kirillov dimension of L in k variables similarly to the associative case (see
[9, 10]),
GKdimG
k (L) := GKdim(LG
k (L)).
See the survey of Drensky [12] and Centrone [8] for more details on the Gelfand-
Kirillov dimension of PI-algebras.
Suppose that L is a G-graded PI-algebra. It is well known that the relatively-free
k , ¯xg2
1 , . . . , ¯xgs
k ,
k (L), is a Z∞-graded Lie alge-
k (L) in the
G-graded algebra LG
where ¯xg
i
bra when considering the usual multigrading. The Hilbert series of LG
variables t1, t2, . . . , tr, r = sk is the formal power series
k , . . . , ¯xgs
k (L) in the variables ¯xg1
is the homomorphic image of xg
i
1 , . . . , ¯xg1
in LG
1 , . . . , ¯xg2
H(L, t1, t2, . . . , ts) = Xm=(m1,m2,...,ms)
dim LG
k (L)(m1,m2,...,ms)tm1
1 tm2
2
· · · tms
s
,
where LG
k (L)(m1,m2,...,ms) is the homogeneous component of degree (m1, m2, . . . , ms).
The growth function of LG
k (L) with respect to the vector space generated by
V = {¯xg1
1 , . . . , ¯xg1
k , . . . , ¯xgs
1 , . . . , ¯xgs
k }
g(n)(LG
k (sl2)) = Xm≤n
am,
is
where
(1)
am =
Xm=m1+m2+···+ms
dimK LG
k (L)(m1,m2,...,ms).
Let T be a tableau of shape σ filled in with natural numbers {1, . . . , k} and let
di be the multiplicity of i in T . A tableau is said to be semistandard if the entries
weakly increase along each row and strictly increase down each column. We say
that
Sσ(t1, . . . , tk) = XTσ semistandard
td1
1 td2
2 · · · tdk
k
is the Schur Function of σ in the variables t1, . . . , tk.
The G-graded cocharacters of L are strictly related with the Hilbert series
H(L, t1, t2, . . . , tr). We have the following (see [7],[13]).
Proposition 2.1. Let L be a G-graded Lie algebra and m1,. . ., ms ≥ 0. Suppose
that
χm1,...,ms(L) = Xhσi⊢m
mhσiχσ1 ⊗ · · · ⊗ χσs
4
CENTRONE AND SOUZA
is the (m1, . . . , ms)-cocharacter of L where hσi = (σ1, . . . , σs) is a multipartition of
n i.e., σ1 ⊢ m1, . . . , σs ⊢ ms and m = m1 + · · · + ms. Then
(2)
H(L, t1, t2, . . . , tr) =
Xm=(m1,m2,...,mr)
mσ1,...,σsSσ1 (T1) · · · Sσs(Ts)
where Sσ1(T1), . . ., Sσs (Ts) are Schur functions with shape σ1, . . . , σs in the vari-
ables T1 = {t1, t2, . . . , tk}, . . ., Ts = {tsk−k+1, tsk−k+2, . . . , tr}.
3. Graded Gelfand-Kirilov dimension for sl2
In this section we calculate the G-graded Gelfand-Kirillov dimension of sl2, for
any non-trivial group G.
In order to simplify the notation we write
0
h =(cid:18)1
0 −1(cid:19) ,
e =(cid:18)0 1
0 0(cid:19) ,
f =(cid:18)0 0
1 0(cid:19) ,
for the standard basis of L = sl2. Up to equivalence, sl2 has three non-trivial
G-gradings (see [2]):
(1) G = Z2: L0 = Kh, L1 = Ke ⊕ Kf ;
(2) G = Z2 × Z2: L(0,0) = 0, L(1,0) = Kh, L(0,1) = K(e + f ), L(1,1) = K(e − f );
(3) G = Z: L−1 = Ke, L0 = Kh, L1 = Kf, Li = 0, i /∈ {−1, 0, 1}.
We shall write slG
2 for the algebra sl2 endowed with the correspondent G-grading
(when necessary). We consider each grading separately.
We recall that ⌊w⌋ = max{u ∈ Z : u ≤ w} is the integer part of w when w is a
real number.
3.1. The Z2-grading. We assume that L = sl2 is Z2-graded i.e., L = L0 ⊕ L1
where L0 = Kh and L1 = Ke ⊕ Kf . In order to compute the Z2-graded Gelfand-
Kirillov dimension of sl2 we use the Repin's description of the Z2-graded cocharacter
sequence of sl2.
Proposition 3.1 ([24]). Let
χp,n−p(slZ2
2 ) = Xσ⊢p, τ ⊢n−p
mσ,τ χσ ⊗ χτ
be the (p, n − p)-th cocharacter of sl2. Then, for every σ ⊢ p and τ ⊢ n − p,
mσ,τ ≤ 1. Moreover, mσ,τ = 1 if and only if σ = (p), τ = (q + r, q) and the
following conditions hold:
(1) p 6= n;
(2) r 6= n;
(3) r ≡ 1 or p + q ≡ 1(mod 2).
Then we have the following result.
Proposition 3.2. The growth function g(n) = g(n)(LZ2
polynomial in n of degree 3k − 1.
k (sl2)) of LZ2
k (sl2) is a
Proof. By Proposition 3.1,
χp,m−p(slZ2
2 ) =X p ⊗
m − p − q
q
ON THE GROWTH OF GRADED POLYNOMIAL IDENTITIES OF sln
5
where p 6= m, r = m − p − 2q 6= m and r ≡ 1 or p + q ≡ 1(mod 2). If m is even we
have that
m−2
2
⌊ m−(2p1 +1)
2
⌋
Xq1=0
χp,m−p(slZ2
2 ) =
⌊ m−2p2 −2
⌋
Xp1=0
Xt2=0
4
+
m−2
2
Xp2=0
2p2 ⊗
m − 2p2 − (2t2 + 1)
2t2 + 1
.
2p1 + 1 ⊗
m − (2p1 + 1) − q1
q1
By the definition of Hilbert series given in (1) and (2) we get:
m−2
2
⌊ m−(2p1+1)
2
⌋
2
m−2
am =
⌊ m−2p2 −2
Xp1=0
Xp2=0
Xq1=0
Xt2=0
where sT (cid:0) p (cid:1), sU(cid:18) m − p − q
+
q
⌋
4
shape p ,
m − p − q
q
q1
(cid:19)
sT (cid:0) 2p1 + 1 (cid:1) sU(cid:18) m − (2p1 + 1) − q1
sT (cid:0) 2p2 (cid:1) sU(cid:18) m − 2p2 − (2t2 + 1)
(cid:19) ,
(cid:19) is the number of semistandard tableaux with
2t2 + 1
in the variables T and U respectively. For m odd we observe
that am can be calculated as above. Since (see [16, 27])
sT (cid:0) p (cid:1) = (cid:18)p + k − 1
k − 1 (cid:19)
m − p − 2q + 1
sU(cid:18) m − p − q
q
(cid:19) =
k − 1
(cid:18)m − p − q + k − 1
k − 2
(cid:19)(cid:18)q + k − 2
k − 2 (cid:19)
we can assume that for all m, am is asymptotically equivalent to
m−2
2
⌊ m−(2p1 +1)
2
⌋
⌊ m−(2p1 +1)
⌋
Xp1=0
Xq1=0
2
Xq1=0
q1
sT (cid:0) 2p1 + 1 (cid:1) sU(cid:18) m − (2p1 + 1) − q1
(cid:19) =
(cid:19)(cid:18)q1 + k − 2
k − 2 (cid:19).
(cid:18)m − 2p1 − 2q1 − k − 2
k − 2
m − 2p1 − 2q1
k − 1
Using standard combinatorial arguments, it is easy to see that am is a polynomial
in m of degree 3k − 2. Therefore g(n) is a polynomial in n of degree 3k − 1 since
(cid:3)
m−2
2
Xp1=0(cid:18)2p1 + k
k − 1 (cid:19)
g(n) = Xm≤n
am.
As an immediate consequence of the previous proposition we obtain the Z2-
graded Gelfand-Kirillov dimension of sl2.
Corollary 3.3. GKdimZ2
k (sl2) = 3k − 1.
3.2. The Z2 × Z2-grading. We consider the Z2 × Z2-grading of L = sl2 where
L(0,0) = 0, L(1,0) = Kh, L(0,1) = K(e + f ) and L(1,1) = K(e − f ).
The next result by Repin [24] shows the Z2 × Z2-cocharacters decomposition of
sl2.
6
CENTRONE AND SOUZA
Proposition 3.4. Let
χn(slZ2×Z2
2
) = Xσ⊢p, τ ⊢q
π⊢r
mσ,τ,πχσ ⊗ χτ ⊗ χπ
be the (p, q, r)-th cocharacter of sl2. Then, for every σ ⊢ p, τ ⊢ q, π ⊢ r, mσ,τ,π ≤ 1.
Moreover, mσ,τ,π = 1 if and only if σ = (p), τ = (q), π = (r) and the following
conditions hold:
(1) p 6= n, q 6= n, r 6= n;
(2) p + q ≡ 1 or q + r ≡ 1(mod 2).
We have the following result.
Proposition 3.5. The growth function g(n) = g(n)(LZ2×Z2
a polynomial in n of degree 3k + 1.
k
(sl2)) of LZ2×Z2
k
(sl2) is
Proof. By Proposition 3.4,
χp,q,r(slZ2×Z2
2
) =X p ⊗ q ⊗ r
where m = p + q + r, p 6= m, q 6= m, r 6= m and p + q ≡ 1 or q + r ≡ 1(mod 2).
If m is even then
χp,q,r(slZ2×Z2
2
) =
⌊ m−1
2 ⌋
⌊ m−2s1−1
2
⌋
m−2s1−2t1 −2
2
m−2s2−2
m−2s2 −2t2 −2
Xs1=0
2
Xt1=0
Xu2=0
Xu3=0
2
+
+
m−2
2
m−2
Xs2=0
Xs3=0
2
2
Xt2=0
Xt3=0
2
⌊ m−2s3 −1
⌋
m−2s3−2t3 −2
2s1 ⊗ 2t1 + 1 ⊗ 2u1 + 1
Xu1=0
2s2 + 1 ⊗ 2t2 + 1 ⊗ 2u2
2s3 + 1 ⊗ 2t3 ⊗ 2u3 + 1 .
Using arguments similar to those of Proposition 3.2 we have that
⌊ m−1
2 ⌋
⌊ m−2s1−1
2
⌋
m−2s1 −2t1 −2
2
am ≈
=
Xs1=0
Xs1=0
Xt1=0
Xt1=0
2
Xu1=0
Xu1=0
2
⌊ m−1
2 ⌋
⌊ m−2s1−1
⌋
m−2s1 −2t1 −2
sT (cid:0) 2s1 (cid:1) sU(cid:0) 2t1 + 1 (cid:1) sV (cid:0) 2u1 + 1 (cid:1)
(cid:18)2s1 + k − 1
k − 1 (cid:19)(cid:18)2u1 + k
k − 1 (cid:19)
k − 1 (cid:19)(cid:18)2t1 + k
for all m. It is easy to see that am is a polynomial in m of degree 3k. Therefore
g(n) is a polynomial in n of degree 3k + 1.
(cid:3)
Corollary 3.6. GKdimZ2×Z2
k
(sl2) = 3k + 1.
3.3. The Z-grading. We consider now the Z-grading of L = sl2, where L−1 = Ke,
L0 = Kh, L1 = Kf and Li = 0 for all i /∈ {−1, 0, 1}.
We have the following result by Repin [24].
Proposition 3.7. Let
χn(slZ
2 ) = Xσ⊢p, τ ⊢q
π⊢r
mσ,τ,πχσ ⊗ χτ ⊗ χπ
ON THE GROWTH OF GRADED POLYNOMIAL IDENTITIES OF sln
7
be the (p, q, r)-th cocharacter of sl2. Then, for every σ ⊢ p, τ ⊢ q, π ⊢ r, mσ,τ,π ≤ 1.
Moreover, mσ,τ,π = 1 if and only if σ = (p), τ = (q), π = (r) and the following
conditions hold:
(1) p 6= n, q 6= n, r 6= n;
(2) p − r ≤ 1.
As in the previous two cases, we get the asympthotics for the growth function
of LZ
k(sl2).
Proposition 3.8. The growth function g(n) = g(n)(LZ
nomial in n of degree 3k − 1.
k(sl2)) of LZ
k(sl2) is a poly-
Proof. By Proposition 3.7,
where m = p + q + r, p 6= m, q 6= m, r 6= m and p − r ≤ 1.
χp,q,r(slZ
2 ) =X p ⊗ q ⊗ r
For m even we have that
⌊ m−1
2 ⌋
Xs1=0
χp,q,r(slZ
2 ) =
m−2
2
m−2s1
2
⊗ 2s1 ⊗ m−2s1
2
⌊ m−2s2−1
2
⌋ + 1 ⊗ 2s2 + 1 ⊗ ⌊ m−2s2−1
2
⌋
⌊ m−2s2−1
2
⌋ ⊗ 2s2 + 1 ⊗ ⌊ m−2s2−1
2
⌋ + 1 .
+
+
m−2
Xs2=0
Xs2=0
2
Using arguments similar to those of Proposition 3.2 we obtain
am ≈
=
m−1
2
2
m−1
Xs1=0
sT (cid:0) m−2s1
Xs1=0(cid:18) m−2s1
k − 1
2
2 + k − 1
2
(cid:1) sU(cid:0) 2s1 (cid:1) sV (cid:0) m−2s1
(cid:18)2s1 + k − 1
(cid:19)
k − 1 (cid:19).
2
(cid:1)
and consequently g(n) is a polynomial in n of degree 3k − 1.
(cid:3)
Corollary 3.9. GKdimZ
k(sl2) = 3k − 1.
4. The Zn-graded GK dimension of sln
In this section we compute the exact value of the Zn-graded GK dimension of
sln.
Let us consider the relatively-free algebra of sln graded by Zn with the grading
of Vasilovsky (see [26] and [11]). Let (sln)i = spanK{epq : q − p = i}, for all
i ∈ {0, . . . , n− 1}, where the epq's are the usual matrix units and · is the reduction
modulo n.
We shall construct LZn
consider X = {x(r)
k (sln) as a generic matrix algebra. For this purpose we
ij i, j = 1, . . . , n−1, r = 1, . . . , k} and for every r = 1, . . . , k, let us
8
CENTRONE AND SOUZA
consider the generic n×n matrices with entries from the algebra of the commutative
polynomials K[X]:
n−1
x(r)
ii eii − n−1
Xi=1
x(r)
pq epq.
Ar
Ar
0 =
Xi=1
i = Xq−p=i
x(r)
ii ! enn,
We denote by Lk(A) and Kk(A) respectively the Lie algebra and the associative
let us observe that Lk(A) is contained in
unitary algebra generated by the Ar
Kk(A). We have the following results (see [23]).
i 's.
Theorem 4.1. For every n, k ∈ N, k ≥ 2, we have LZn
k (sln) ∼= Lk(A).
In [3] the author proved the existence of a graded field of quotients Q for asso-
ciative unitary G-prime algebras provided G to be either an abelian or an ordered
group. The next result is easy to be proved.
Proposition 4.2. For each k, n ∈ N, n ≥ 2 the algebra Kk(A) is a Zn-prime
algebra.
We consider Q to be the Zn-graded algebra of central quotients of Kk(A), always
existing due to Proposition 4.2. We denote by Z the (graded)-center of Q which
is a (graded) field. We observe that Z contains K. Following word by word the
computations by Procesi (see [22], Part II, Section 1), we have the next result.
Proposition 4.3. Let k, n ∈ N, where k ≥ 2, then
tr.degKZ = k(n2 − 1) − n + 1.
Because Lk(A) is contained in Kk(A), due to Proposition 4.3 and Theorem 5.7
of [10], we have the following.
Proposition 4.4. Let k, n ∈ N, where k ≥ 2, then
GKdimsln
k ≤ k(n2 − 1) − n + 1.
We observe that the growth function g(r) of Lk(A) depends on the number of
linearly independent polynomials of degree r appearing in one of its entries, to say
(a, b), hence from now on we are going to investigate such a number in such an
entry. Let us set
X ∗ = {x(r)
ij i, j = 1, . . . , n − 1, i 6= j, r = 1, . . . , k}.
We observe that for each polynomial m of any degree appearing in the entry (a, b)
of Kk(A) having no variables appearing in one of the A0
i (so in the variables from
X ∗), we can construct a non-zero monomial M of the same degree of Lk(A) having
m as one of its summands in its (a, b)-entry. More precisely, if m is generated by
the product Ai1 · · · Air we consider M = [Ai1 , . . . , Air ].
Let σ ∈ Sr, then we consider the natural left action of GLr on Ai1 · · · Air such
that σ(Ai1 · · · Air ) = Aiσ−1 (1) · · · Aiσ−1 (r)
where Φ = 2r−1 and each permutation of Φ is a product of cycles in which at most
one cycle has length greater than 2. We have the next result.
. We observe that M =Pϕ∈Φ σ(Ai1 · · · Air ),
Proposition 4.5. In the previous notation and hipothesis, the left-normed Lie
monomial M = [Ai1 , . . . , Air ] contains m in its (a, b)-entry.
ON THE GROWTH OF GRADED POLYNOMIAL IDENTITIES OF sln
9
Proof. Straightforward computations show that the result is true for r = 2. Let
r > 2, then suppose the proposition true for r − 1. We observe that the (a, b) entry
of M is pab′ xb′b−xab′′ pb′′b, where pab′, pb′′b are entries of [Ai1 , . . . , Air−1 ]. Due to the
induction hipotheses, pab′ = mab′ + p′, where mab′ is the monomial appearing in the
(a, b′)-entry of Ai1 ·· · · Air−1 and p′ is a polynomial. If the assertion is not true, there
exists a monomial xb′′b′′′ · · · xa′b of pb′′b such that mab′ xb′b − xab′′ xb′′b′′′ · · · xa′b =
0. We note that at least the first and last variable of mab′ are different than
xb′′b′′′ and xa′b.
It turns out that there exist a submonomial l := xij · · · xa′b of
xab′′ xb′′ b′′′ · · · xa′b such that xij · · · xa′bxab′′ xb′′b′′′ · · · xa′′j = mab′ xb′b that means
a = b = i. It also means that there exists a permutation, which is a product of
two cycles, ϕ ∈ Φ such that xij · · · xa′bxab′′ xb′′ b′′′ · · · xa′′j and mab′xb′b are the (a, a)
entry of ϕ(Ai1 , . . . , Air−1 ) and Ai1 , . . . , Air−1 respectively, i.e., the latter are linearly
dependent. Because each permutation of Φ is a product of cycles in which at most
one cycle has length greater than 2, we have ϕ(Ai1 , . . . , Air−1 ) and Ai1 , . . . , Air−1
are linearly dependent if and only if mab′ = l, that is (see [26]) if and only if Air
has Zn-degree 0.
(cid:3)
From now on we shall consider sln, n ≥ 3. We observe that left-normed mono-
mials with different multidegree are linearly independent. We consider the ordered
set S = {m1, . . . , ms} of linearly independent monomials of degree r appearing in
the entry (a, b) of a product of generators of Kk(A) of Zn-degree different than 0.
Due to the previous observation, we may consider the mi's as generated by the
non-zero monomials M1, . . . , Ms in Kk(A) with the same multidegree, i.e., if m1 is
generated by M1 := Ai1 · · · Air , then for each t ∈ {2, . . . , s} there exists σt ∈ Sr
such that mt is generated by σt(M1) := Mt = Ai
. We take now
the set S′ of non-zero left-normed monomials Lt of Lk(A) as above corresponding
to each t. We observe that each summand of Lt is a monomial of Kk(A) obtained
by Mt applying a permutation ϕ ∈ Φ.
· · · Ai
−1
t
(r)
σ
−1
t
(1)
σ
Due to the fact that linearly independent monomials of Kk(A) grow polyno-
mially whereas the monomial identities of a single monomial grow factorially, for
sufficiently large r and for each Mi, Mj, it is possible to find permutations σ, τ such
that σ(Mi) = Mi, τ (Mj) = Mj and ϕ(σ(Mi)) 6= ϕ(τ (Mj )), ϕ ∈ Φ. Hence we have
the following result.
Proposition 4.6. For sufficiently large r, S′ is a linearly independent set.
In an analogous way, we consider now polynomials m of any degree appearing in
the entry (a, b) of Kk(A) in the variables from X. We observe that GKdim(Kk(A)) =
GKdim(M), where M is the finitely generated K[X]-module contained in Kk(A)
formed by all products Ai1 · · · Air such that Ai1 6= Ai2 and for each initial sub-
monomial Ai1 · · · Ail , l ≤ r such that Ai1 · · · Ail has Zn-degree 0 we have that Ail+1
has not Zn-degree 0. In light of this, we have that Proposition 4.6 gives us the
next.
Proposition 4.7. Let k ∈ N, then GKdimZn
k (sln) ≥ GKdim(M).
Combining Propositions 4.4 and 4.7 we have the following result.
Theorem 4.8. Let k ∈ N, then GKdimZn
k (sln) = k(n2 − 1) − n + 1.
10
CENTRONE AND SOUZA
5. Conclusions
We want to draw out a parallelism with the arguments used by Procesi and the
first of the authors in order to compute the Zn-graded Gelfand-Kirillov dimension
of Mn(K) (see [22] and [10]).
We introduce the notion of G-prime (Lie) algebra.
Definition 5.1. A G-graded ideal P of L is called G-prime if aH ⊆ P for G-
homogeneous element a ∈ L and a G-graded ideal H of L, then either a ∈ P or
H ⊆ P . If (0) is a G-prime we say that L is a G-prime algebra.
When the grading group is not specified, we shall refer to graded prime algebras.
We have that the relatively-free algebras of sl2 endowed with the gradings intro-
duced in the previous section are graded prime.
Proposition 5.2. The relatively-free G-graded algebra of sl2 in k variables LG
is G-prime.
k (sl2)
Proof. We argue only in the case G = Z2 because the other cases may be treated
analogously. Suppose that aH = 0 for a Z2-homogeneous element a of degree i and
a Z2-graded ideal H 6= 0 of LZ2
k (sl2). Since H = H0 ⊕ H1 is a non-trivial Z2-graded
ideal then H0 6= 0 and H1 6= 0 where H0 and H1 are the homogeneous component
of H of the degree 0 and 1 respectively. By simple calculations aHj = 0 with j 6= i
implies that a = 0.
(cid:3)
We have already pointed out that the associative algebras Kk(A) generated by
the generic homogeneous elements of the Zn-graded relatively-free algebra of sln
are Zn-prime (Proposition 4.2). Of course, due to the theory of Balaba [3], we
have the existence of a (graded) field of quotients whose transcendence degree over
the ground field equals the GK dimension of Kk(A). The fact that the relatively-
free algebras Lk(A) of sln is prime in a Lie sense (Proposition 5.2) gives us a
hope that what happened in the associative case is maybe true in the Lie case,
i.e. there exists a graded central localization of Lk(A) such that the "generalized
transcendence degree" of its "center" over the ground field gives a measure of the
graded GK dimension of Lk(A). We recall that a definition of central localization
for Lie algebras may be found in [20].
For the sake of completeness we want to show a consequence of some results of
Procesi and Razmyslov (see [21] and [23]) in the ordinary case. In particular, let
C(0)
kn be the trace algebra of Kk(A), the associative unitary algebra generated by
k ungraded generic traceless n × n matrices over the field K, and T (0)
kn the mixed
trace algebra of Kk. Then we have the next result.
Theorem 5.3. For k, n ≥ 2 we have
GKdim(T (0)
kn ) = GKdim(C(0)
kn ) = (k2 − 1)(n − 1).
Due to the fact that Kk(A) is a prime algebra, we have the ungraded analog of
Proposition 4.3.
Proposition 5.4. Let k, n ∈ N, where k ≥ 2, then
GKdim(Kk(A)) = (k2 − 1)(n − 1).
ON THE GROWTH OF GRADED POLYNOMIAL IDENTITIES OF sln
11
In light of Theorem 4.8 and Proposition 5.4, it is reasonable to consider a graded
generalization of Theorem 5.3.
References
[1] E. Aljadeff, A. Kanel-Belov, Representability and Specht problem for G-graded algebras, Adv.
Math. 225(5) (2010), 2391-2428.
[2] Y. Bahturin and M. Kochetov, Classification of group gradings on simple Lie algebras of types
A, B, C and D, J. Algebra 324(11) (2010), 2971-2989.
[3] I. N. Balaba, Graded prime PI-algebras, J. Math. Sciences 128(6) (2005), 3345-3349.
[4] A. Y. Belov, Rationality of Hilbert series of relatively free algebras (Russian), Uspekhi Mat.
Nauk 52(2) (1997), 153-154.
[5] A. Berele, Homogeneous polynomial identities, Israel J. Math. 42 (1982) 258-272.
[6] A. Berele, Generic verbally prime PI-algebras and their GK-dimensions, Comm. Algebra 21(5)
(1993), 1487-1504.
[7] A. Berele, Cocharacters of Z/2Z-graded algebras, Israel J. Math. 61 (1988), 225-234.
[8] L. Centrone, On some recent results about the graded Gelfand-Kirillov dimension of graded
PI-algebras, Serdica Math. J. 38(1-3) (2012), 43-68.
[9] L. Centrone, A note on graded Gelfand-Kirillov dimension of graded algebras, J. Algebra Appl.
10(5) (2011), 865-889.
[10] L. Centrone, The graded Gelfand-Kirillov dimension of verbally prime algebras, Linear Mul-
tilinear Algebra 59(12) (2011), 1433-1450.
[11] O. M. Di Vincenzo, On the graded identities of M1,1(E), Israel J. Math. 80 (1992), 323-335.
[12] V. Drensky, Gelfand-Kirillov dimension of PI-algebras. Methods in ring theory (Levico Terme,
1997), 9713, Lecture Notes in Pure and Appl. Math., 198, Dekker, New York, 1998.
[13] V. Drensky, A. Giambruno, Cocharacters, codimensions and Hilbert series of the polynomial
identities for 2 × 2 matrices with involution, Can. J. Math 46 (1994), 718-733.
[14] V. Drensky, P. Koshlukov, G. G. Machado, GK-dimension of the lie algebra of generic 2 × 2
matrices, http://arxiv.org/pdf/1503.02091.pdf.
[15] A. Giambruno, M. S. Souza, Graded polynomial identities and Specht property of the Lie
algebra sl2, J. Algebra 389 (2013), 6-22.
[16] A. Kerber, Representations of permutation groups, Lect. Notes Math. 240 (1971), Springer-
Verlag, Berlin Heidelberger, New York.
[17] P. Koshlukov, Graded polynomial identities for the Lie algebra sl2(K), Internat. J. Algebra
Comput. 18(5) (2008), 825-836.
[18] G. R. Krause, T. H Lenagan, Growth of algebras and Gelfand-Kirillov dimension. Revised
edition. Graduate Studies in Mathematics, 22. American Mathematical Society, Providence,
RI, 2000.
[19] G. G. Machado, P. Koshlukov, GK dimension of the relatively free algebra for sl2, Monatsh.
Math. 175(4) (2014), 543-553.
[20] M. Siles Molina, Algebras of quotients of Lie algebras, J. Pure Appl. Algebra 188(1-3) (2004),
175-188.
[21] C. Procesi, Computing with 2 × 2 matrices, J. Algebra 87 (1984), 342-359.
[22] C. Procesi, Non-commutative affine rings. Atti Accad. Naz. Lincei Mem. Cl. Sci. Fis. Mat.
Natur. Sez. I (8) 8 (1967), 237-255.
[23] Y. Razmyslov, Finite basing of the identities of a matrix algebra of second order over a field
of characteristic zero (Russian), Algebra i Logika 12 No. 1 (1973) 83-113; Translation: Algebra
and Logic 12 (1973), 47-73.
[24] D. V. Repin, Graded identities of a simple three-dimensional Lie algebra (Russian), Vestn.
Samar. Gos. Univ. Estestvennonauchn. Ser. 2004, Special Issue 2, 5-16.
[25] I. Sviridova, Identities of pi-algebras graded by a finite abelian group, Comm. Algebra 39(9)
(2011), 3462-3490.
[26] S. Yu. Vasilovsky, Z
n-graded polynomial identities of the full matrix algebra of order n, Proc.
Amer. Math. Soc. 127(12) (1999), 3517-3524.
[27] H. Weyl, The Classical Groups, Their Invariants and Representations, Princeton University
Press, Princeton, N. J., 1946.
12
CENTRONE AND SOUZA
Departamento de Matem´atica, IMECC-UNICAMP, Rua S´ergio Buarque de Holanda
651, 13083-859 Campinas-SP, Brazil
E-mail address: [email protected]
Departamento de Matem´atica, IME-USP, Rua do Matao, 1010, Cidade Universit´aria,
05508-090 Sao Paulo-SP, Brazil
E-mail address: [email protected]
|
1309.7390 | 1 | 1309 | 2013-09-27T23:49:07 | A family of varieties of pseudosemilattices | [
"math.RA",
"math.GR"
] | In [3], a basis of identities {u_n = v_n | n\geq 2} for the variety SPS of all strict pseudosemilattices was determined. Each one of these identities u_n = v_n has a peculiar 2-content D_n. In this paper we study the varieties of pseudosemilattices defined by sets of identities, all with 2-content the same D_n. We present here the family of all these varieties and show that each variety from this family is defined by a single identity also with 2-content D_n. This paper ends with the study of the inclusion relation between the varieties of this family. | math.RA | math |
A FAMILY OF VARIETIES OF PSEUDOSEMILATTICES
LU´IS OLIVEIRA
Abstract. In [3], a basis of identities {un ≈ vn n ≥ 2} for the
variety SPS of all strict pseudosemilattices was determined. Each one
of these identities un ≈ vn has a peculiar 2-content Dn. In this paper
we study the varieties of pseudosemilattices defined by sets of identities,
all with 2-content the same Dn. We present here the family of all these
varieties and show that each variety from this family is defined by a
single identity also with 2-content Dn. This paper ends with the study
of the inclusion relation between the varieties of this family.
1. Introduction
A regular semigroup is a semigroup S for which every x ∈ S has an x′ ∈ S
such that xx′x = x. Thus xx′ and x′x are idempotents in S and we shall
denote the set of all idempotents of S by E(S). Consider the following two
binary relations on E(S):
e ≤R f ⇔ e = f e
and
e ≤L f ⇔ e = ef ;
and set ≤ = ≤R ∩ ≤L . Then ≤R and ≤L are quasi-orders on E(S), while ≤
is a partial order on E(S). We shall denote by (f ]R the set of idempotents
e such that e ≤R f . Similarly, we define (f ]L and (f ]≤ . We can introduce
also the following two equivalence relations on E(S):
e R f ⇔ (e]R = (f ]R
and
e L f ⇔ (e]L = (f ]L ,
or equivalently, R = ≤R ∩ ≥R and L = ≤L ∩ ≥L for ≥R and ≥L the
expected reverse relations corresponding to ≤R and ≤L , respectively. Thus
R ∩ L is just the identity relation.
Any finite sequence f1, f2, · · · , fm of alternately R- or L -equivalent idem-
potents of S contains a subsequence f1 = e1, e2, · · · , en = fm with the same
property but where no two consecutive elements are equal. An E-chain is
then a finite sequence of alternately R- or L -equivalent idempotents with
no two consecutive elements equal. Thus two idempotents are (R ∨ L )-
equivalent, and we shall say they are connected, if there exists an E-chain
starting at one of them and ending at the other one. The connected compo-
nents of E(S) are the (R ∨ L )-equivalence classes.
The equivalence relations R and L are clearly the restriction to E(S)
of the homonymous Green's relations on the semigroup S. However, if we
consider the Green's relation D (= R ∨ L ) on S, the relation R ∨ L
2010 Mathematics Subject Classification. 08B15, 08B05, 08B20, 20M17.
1
2
LU´IS OLIVEIRA
on E(S) defined above may not be the restriction of D to E(S). In fact,
the set of idempotents of a D-class of S is the (disjoint) union of several
connected components of E(S). Nevertheless, we can guarantee that the
relation R ∨ L on E(S) is the restriction of D if S is generated by its
idempotents.
A regular semigroup S is locally inverse if for every ordered pair (e, f ) of
idempotents of S, (e]R ∩ (f ]L = (g]≤ for some g ∈ E(S). The idempotent
g is clearly unique since ≤ is a partial order. Thus, we can define a new
binary algebra (E(S), ∧) for every locally inverse semigroup S by setting
e ∧ f = g. The algebra (E(S), ∧) so obtained is called the pseudosemilattice
of idempotents of S, and the quasi-orders ≤R and ≤L on E(S) can be
recovered from the binary operation ∧ by setting:
e ≤R f ⇔ f ∧ e = e
and
e ≤L f ⇔ e ∧ f = e .
The pseudosemilattices of idempotents of locally inverse semigroups are
idempotent binary algebras, although they are often not semigroups them-
selves. Nevertheless, the class of all these binary algebras constitutes a
variety (Nambooripad [8]) given by the identities:
(P S1) x ∧ x ≈ x ;
(P S2) (x ∧ y) ∧ (x ∧ z) ≈ (x ∧ y) ∧ z ;
(P S3) ((x ∧ y) ∧ (x ∧ z)) ∧ (x ∧ w) ≈ (x ∧ y) ∧ ((x ∧ z) ∧ (x ∧ w)) ;
together with the left-right duals (PS2') and (PS3') of (PS2) and (PS3),
respectively. We shall denote this variety by PS.
The structure of pseudosemilattices is related to the notion of semilat-
tice. Every pseudosemilattices is the union of its maximal subsemilattices,
and further, it is a homomorphic image of another pseudosemilattice whose
maximal subsemilattices are disjoint [7]. A pseudosemilattice with disjoint
maximal subsemilattices can be described [4, 10] as the union of disjoint
maximal semilattices Eiλ for (i, λ) ∈ I × Λ such that:
(i) if xiλ ∈ Eiλ and xjµ ∈ Ejµ, then xiλ ∧ xjµ ∈ Eiµ;
(ii) ∪i∈I Eiλ is a left normal band, that is, an idempotent semigroup
satisfying xyz ≈ xzy;
(iii) ∪λ∈ΛEiλ is a right normal band, that is, an idempotent semigroup
satisfying xyz ≈ yxz.
We can say even more, every pseudosemilattice divides an elementary pseu-
dosemilattice [7], that is, a pseudosemilattice with disjoint maximal sub-
semilattices all isomorphic (the semilattices Eiλ above are all isomorphic).
An e-variety of regular semigroups [5, 6] is a class of these algebras closed
for taking homomorphic images, direct products and regular subsemigroups.
The class LI of all locally inverse semigroups is an example of an e-variety.
Nambooripad's result [8] was generalized by Auinger [2] who proved that
the mapping
ϕ : Le(LI) −→ L(PS), V 7−→ {(E(S), ∧) S ∈ V}
3
is a well-defined complete homomorphism from the lattice Le(LI) of e-varie-
ties of locally inverse semigroups onto the lattice L(PS) of varieties of pseu-
dosemilattices. Thus, any information about L(PS) is useful to understand
the structure of Le(LI) itself.
A strict pseudosemilattice is the pseudosemilattice of idempotents of some
[combinatorial] strict regular semigroup, that is, of some subdirect product
of completely simple and/or 0-simple semigroups. The class SPS of all
strict pseudosemilattices is a variety, and in fact it is the smallest variety
of pseudosemilattices containing algebras that are not semigroups. On the
other hand, the largest variety of pseudosemilattices whose algebras are all
semigroups is the variety NB of all normal bands. It is a well known fact
that NB ⊆ SPS. The set of identities satisfied by all strict pseudosemi-
lattices was characterized by Auinger [2]: an identity u ≈ v is satisfied by
all strict pseudosemilattices if and only if the words u and v have the same
leftmost letter, the same rightmost letter, and the same 2-content (see sec-
In [3] a basis {un ≈ vn : n ≥ 2}
tion 2 for the definition of 2-content).
of identities for SPS was determined. The 2-content Dn of the word un
has a very peculiar nature.
In this paper we shall study the varieties of
pseudosemilattices defined by a single identity whose words have some Dn
(n ≥ 2) as their 2-content. This will allow us to define a family of varieties
of pseudosemilattices which will gives us some incite into the structure of
the lattice L(PS).
The terminology introduced in [3] and the results obtained in that paper
are crucial for the unwind of the present one. Thus we shall devote the next
section to recall the concepts and results from [3]. We shall call an identity
non-trivial if there is a pseudosemilattice which does not satisfies it.
For each n ≥ 2, a family
{un,k,i ≈ vn,k,i, u∗
n,k,j ≈ v∗
n,k,j k ≥ 1, 1 ≤ i, j ≤ 2n and j odd}
of non-trivial identities will be introduced in section 3 which generalizes the
identity un ≈ vn (in fact un ≈ vn corresponds to un,1,1 ≈ vn,1,1). We shall
see that this family contains all the identities needed to describe varieties of
pseudosemilattices defined by identities with 2-content Dn. The list of all
such varieties will be presented in section 4 together with the description
of the inclusion relation between them. Further, it is shown in that same
section that every variety from that list is defined by a single identity with 2-
content Dn. Finally, in the last section, we shall study the inclusion relation
between varieties of pseudosemilattices defined by identities with 2-content
Dn and varieties of pseudosemilattices defined by identities with 2-content
Dm for n 6= m.
2. Recalling concepts and results from [3]
The present section is entirely devoted to recall concepts and results ob-
tained and used in [3]. Thus most details are left to the reader to consult
[3].
4
LU´IS OLIVEIRA
Let X be a non-empty set whose elements shall be called letters. We
shall denote by T(X) the set of all finite downward (connected) trees with
(i) a unique top vertex, the root; (ii) each non-root vertex has a unique
predecessor (a vertex placed above it in the tree but connected to it by an
edge); (iii) each non-leaf vertex a (vertex of degree at least 2) has exactly
two successors (vertices placed below a in the tree but connected to a by
an edge), one to the left of a and one to the right of a; and (iv) the leaves
(vertices of degree at most 1) are labeled by letters of X. Thus the leaves
appear at the bottom of the trees of T(X). A non-root vertex of a tree of
T(X) is called a left/right vertex if that vertex is placed to the left/right of
its predecessor. Hence each non-root vertex is either a left vertex or a right
vertex. The root is considered neither a left vertex, nor a right vertex.
We shall denote by (F2(X), ∧) the absolutely free binary algebra on X.
Thus, the elements of F2(X) are well-formed words on the alphabet X ∪
{(, ), ∧}. We can associate inductively a tree from T(X) to each word of
F2(X) by setting Γ(x) = •
x
for each x ∈ X and then letting
Γ(u ∧ v) :=
•
Γ(u) Γ(v)
for u, v ∈ F2(X). The mapping Γ : F2(X) → T(X) so obtained is in fact
a bijection (see [9]), and if we introduce the binary operation ∧ on T(X)
by setting Γ(u) ∧ Γ(v) = Γ(u ∧ v) for any u, v ∈ F2(X), we obtain a model
(T(X), ∧) for the absolutely free binary algebra on X.
There is always some ambiguity when referring to subwords of a word
u since u can have several distinct copies of the same word as a subword.
For example, the letter x occurs twice as a subword of x ∧ x. We shall use
Γ(u) to avoid this ambiguity. Let Sub(u) denote the set of all subwords of u
including repetitions, that is, if v is a subword of u then Sub(u) as a distinct
copy of v for each occurrence of v as a subword of u; and for each vertex a
of Γ(u), let Γ(u, a) denote the downward subtree of Γ(u) having a as the top
vertex. The graph Γ(u) captures all the subword structure of u in the sense
that there is a natural bijection ηu : V (Γ(u)) → Sub(u), where V (Γ(u)) is
the set of vertices of Γ(u), such that if ηu(a) = v then Γ(v) is (isomorphic
to) Γ(u, a). Thus ηu(a) = Γ(u) for a the root of Γ(u) and the leaves of
Γ(u) are in bijection with the one-letter subwords of u. Further, if b and c
are respectively the left and right successors of a in Γ(u), then ηu(a) is the
subword ηu(b) ∧ ηu(c) of u. We can use now the vertices of Γ(u) to pinpoint
the concrete subword we are referring to and avoid in this way any possible
ambiguity that may occur.
In the following we define some combinatorial invariants of the words
u ∈ F2(X). Let l(u) and r(u) be respectively the leftmost and rightmost
letter in u, and let c(u) be the content of u, that is, the set of letters
that occur in u. We define also the 2-content c2(u) inductively by setting
5
c2(x) = {(x, x)} for each x ∈ X, and letting
c2(u ∧ v) = c2(u) ∪ {(l(u), r(v))} ∪ c2(v)
for u, v ∈ F2(X). These combinatorial invariants could be defined using
instead Γ(u) in the obvious way. We shall use however Γ(u) to introduce two
other combinatorial invariants: the left content cl(u) and the right content
cr(u) of a word u ∈ F2(X) are respectively the labels of the left leaves and
the labels of the right leaves of Γ(u). In particular cl(x) = cr(x) = ∅ for
every x ∈ X since the root vertex is considered neither a left vertex nor a
right vertex.
There is another type of trees introduced in [3] with all vertices labeled
by letters of X that we shall recall now. So, let B′(X) be the set of all
finite non-trivial (connected) trees γ whose vertices are labeled by letters of
X and with an ordered pair (lγ, rγ ) of distinguished vertices connected by
an edge. The vertices lγ and rγ shall be called respectively the left root and
the right root of γ. We can now partition the vertices of γ into two disjoint
sets accordingly to their distance to lγ (or to rγ): the set Lγ of all vertices
with even distance to lγ (or with odd distance to rγ) and the set Rγ of all
vertices with odd distance to lγ (or with even distance to rγ). Thus lγ ∈ Lγ
and rγ ∈ Rγ, and any edge of γ connects a vertex of Lγ with a vertex of Rγ.
Thus, we shall consider γ always as a bipartite graph (in fact a bipartite
tree), and if a ∈ Lγ and b ∈ Rγ are connected by an edge, then we shall
represent that edge by the ordered pair (a, b). Further, a vertex from Lγ
shall be called a left vertex of γ while a vertex from Rγ shall be called a
right vertex of γ.
Let B(X) = B′(X) ∪ {•
x
: x ∈ X} and define lγ = rγ = γ for γ = •
x
. For
each γ ∈ B′(X) let Lγ coincide with γ but now only with the left root as
a distinguished vertex. However, this unique distinguished vertex continues
to be seen as a left vertex in Lγ, that is, one still sees the left/right vertices
of γ as left/right vertices of Lγ. For each γ = •
let Lγ be just the graph
x
γ but now seen as a bipartite graph with only one left vertex and no right
vertices; the only vertex of Lγ is now distinguished as a left root. We define
γR dually and introduce the binary operation ⊓ on B(X) by setting
α ⊓ β = Lα ∪ {(lα, rβ)} ∪ βR
for α, β ∈ B(X). In other words, we construct α ⊓ β by taking the disjoint
union of α and β, adding the edge (lα, rβ), and setting lα and rβ respectively
as the left root and the right root of α ⊓ β.
The binary algebra (B(X), ⊓) is generated as such by the set {•
x
: x ∈ X}.
Thus, there is a unique surjective homomorphism ∆ : F2(X) → B(X) such
that ∆(x) = •
= Γ(x) . There is a standard procedure to obtain ∆(u) from
x
Γ(u): first set the leaves of Γ(u) (together with their labels) as the vertices
of ∆(u); then let l∆(u) and r∆(u) be the leftmost and the rightmost leaves of
Γ(u) respectively; and finally add an edge to ∆(u) for each non-leaf vertex
6
LU´IS OLIVEIRA
a of Γ(u), which connects the leftmost and rightmost leaves of Γ(u, a). In
particular, L∆(u) is the set of left leaves of Γ(u) while R∆(u) is the set of
right leaves of Γ(u) for each u ∈ F2(X) \ {X}.
It is convenient to see the procedure just described as a partial bijection
χu : Γ(u) → ∆(u) whose domain is the set of vertices of Γ(u) and whose
image is the all graph ∆(u). Thus χu induces a label preserving bijection
from the set of leaves of Γ(u) onto the set of vertices of ∆(u) such that if
a is a non-leaf vertex of Γ(u) and b and c are respectively the leftmost and
rightmost leaves of Γ(u, a), then χu(a) is the edge (χu(b), χu(c)) of ∆(u).
This procedure was described in [3] using a different but equivalent method
involving the notion of contraction of subtrees. We recommend the reader to
consult [3] for more details including the illustration of a concrete example.
In particular, it was pointed out in that paper that if all vertices of γ ∈ B(X)
have degree at most two, then there exists a unique u ∈ F2(X) such that
γ = ∆(u) (in general ∆ is not injective). This last observation will be useful
for this paper.
We shall denote by ca the label of a labeled vertex a. The combinatorial
invariants l(u), r(u), c(u), c2(u), cl(u) and cr(u) of a word u ∈ F2(X) can
be described using ∆(u):
l(u) is the label of l∆(u) while r(u) is the label
of r∆(u); c(u) is the set of labels of all vertices of ∆(u) while cl(u) and
cr(u) are respectively the set of labels of all vertices of L∆(u) and R∆(u);
and c2(u) is constituted by the pairs (ca, cb) for all edges (a, b) of ∆(u)
together with the pairs (ca, ca) for all vertices a of ∆(u). We can define now
the combinatorial invariants l(γ), r(γ), c(γ), c2(γ), cl(γ) and cr(γ) for each
γ ∈ B(u) as expected: l(γ) = clγ , r(γ) = crγ , and so on.
Let a be a vertex of Γ(u) and let s = ηu(a), a subword of u. Denote by
u(s → t) the word obtained by replacing in u the subword s = ηu(a) with
some word t ∈ F2(X). Thus Γ(s) = Γ(u, a) and Γ(u(s → t)) is obtained
from Γ(u) by substituting Γ(t) for the downward subtree Γ(u, a).
If a is
the root of Γ(u), then s = u and u(s → t) = t, whence ∆(s) = ∆(u)
and ∆(u(s → t)) = ∆(t). Now, assume that a is a left vertex of Γ(u)
and let b be the leftmost leaf of Γ(u, a). Then L∆(s) is just the connected
subtree χu(Γ(u, a)) of ∆(u) with the vertex χu(b) as the distinguished vertex.
Furthermore, χu(b) is the only vertex of χu(Γ(u, a)) connected by an edge
to vertices outside χu(Γ(u, a)). Thus if c denotes the left root of L∆(t), then
∆(u(s → t)) is obtained from ∆(u) by substituting L∆(t) for the subtree
χu(Γ(u, a)) (and each edge (χu(b), d) ∈ Γ(u) with d 6∈ χu(Γ(u, a)) is replaced
by the edge (c, d)). A similar situation occurs if a is a right vertex of Γ(u).
If ψ is an endomorphism of F2(X), then uψ is the word obtained by
replacing in u each one-letter subword x with xψ, or equivalently, Γ(uψ) is
the graph obtained from Γ(u) by replacing each leaf a with Γ(caψ). Thus
∆(uψ) is the graph resulting from replacing in ∆(u) each left vertex a by
the left-rooted tree L(a, ψ) = L∆(caψ) and each right vertex b by the right-
rooted tree R(b, ψ) = ∆(cbψ)R (all these graphs are assumed to be pairwise
disjoint); and then setting lL(l∆(u),ψ) as the left root and rR(r∆(u),ψ) as the right
root. Alternatively, ∆(uψ) can be obtained as follows:
form the disjoint
union
7
[a∈L∆(u)
L(a, ψ) ∪ [b∈R∆(u)
R(b, ψ)
of all graphs L(a, ψ) and R(b, ψ) and add the edge (lL(a,ψ), rR(b,ψ)) for each
edge (a, b) of ∆(u). For (a, b) = (l∆(u), r∆(u)) this yields the connection
between the distinguished vertices lL(l∆(u),ψ) and rR(r∆(u),ψ) of ∆(uψ). The
following corollary has been already stated in [3] and is an immediate con-
sequence of the previous description of ∆(uψ).
Corollary 2.1. For all u, v ∈ F2(X) and each endomorphism ψ : F2(X) →
F2(X), if ∆(u) = ∆(v) then ∆(uψ) = ∆(vψ). In particular, ker∆ is a fully
invariant congruence on F2(X).
The skeleton sk(u, ψ) of ∆(uψ) is the subtree spanned by the set of vertices
{lL(a,ψ) a ∈ L∆(u)} ∪ {rR(b,ψ) b ∈ R∆(u)}
L(l∆(u),ψ) and r
(or spanned by all edges (lL(a,ψ), rR(b,ψ)) for (a, b) an edge in ∆(u)). Fur-
ther, we set l
R(r∆(u),ψ) as the left and right roots of sk(u, ψ)
respectively. Then sk(u, ψ) has the same graph structure as ∆(u) (including
the same distinguished vertices) although with different vertex labels. To be
more precise, the label of each a ∈ L∆(u) is changed from ca to l(caψ) while
the label of each b ∈ R∆(u) is changed from cb to r(cbψ). In case the left
content of u is disjoint from its right content, the skeleton sk(u, ψ) itself can
be viewed as a graph of the form ∆(uψ′) for any endomorphism ψ′ satisfying
xψ′ = l(xψ) if x ∈ cl(u) and xψ′ = r(xψ) if x ∈ cr(u).
We need to make some conventions about the graphical representation
of the bipartite graphs from B′(X). For each γ ∈ B′(X) we shall arrange
their vertices either in two columns (the left/right column representing the
left/right vertices) or in two horizontal rows (the bottom/top row represent-
ing the left/right vertices); and we shall distinguish the left and right roots
by especially representing the unique edge connecting these two vertices by
. When referring to one-vertex only distinguished graphs,
a double line
we shall use an "encircled bullet" • to indicate the distinguished vertex.
Next we shall introduce two rules for changing the graphs of B′(X), but
we need to define first the notion of thorn. A thorn in a graph α ∈ B′(X)
is a pair {e, a} consisting of a degree one vertex a together with the edge e
having a as one of its endpoints such that both a and the other endpoint
of e have the same label. The thorn {e, a} is called essential if a is one of
the distinguished vertices of α; otherwise it is called a non-essential thorn.
These notions of essential and non-essential thorns shall be considered also
for one-vertex only distinguished graphs and for non-distinguished vertex
graphs with the obvious adaptations. The two reduction rules are now
introduced as follows:
8
LU´IS OLIVEIRA
(i) remove a non-essential thorn {e, a} from α. This rule may be visu-
alized graphically as
•
x
•
x
7→
•
x
and
•
x
7→
•
x
•
x
(ii) suppose that two edges e and f have a vertex in common and that
the two other (distinct) vertices a and b have the same label; then
identify the two edges e and f and the vertices a and b (and retain
their label). If one of the merged vertices happens to be a distin-
guished one then so is the resulting vertex. Graphically, this rule
may be visualized as
•
x
y
•
•
y
7→ •
x
• and
y
y
•
•
y
7→
•
x
•
y
•
x
Rule (i) is referred to as the deletion of a thorn while rule (ii) is called an
edge-folding.
A graph α ∈ B′(X) is called reduced if none of the two rules above
If α ∈ B′(X) is not reduced, then we can always
can be applied to it.
obtain a reduced graph by applying the rules (i) and (ii) until no more
reductions are possible. Of course, we can apply these reductions in many
different orders. Nevertheless, we always get the same reduced graph from
a given α independently of the order of reductions we choose to apply. We
shall denote by α the reduced form of α, that is, the unique reduced graph
obtained from α by applying the rules (i) and (ii). We can however obtain
α by first carry out all possible edge-foldings and then carry out all possible
thorn deletions (note that a thorn deletion does not produce any possible
new edge-folding). Furthermore, the edge-folding reduced graph obtained
form α is always the same independently of the order in which we apply
graph obtained from α, and thus we can see the reduction from α into α as
left and right roots. In the second step of this process, we can see α both
the edge-folding reductions. We shall denote by eα the edge-folding reduced
the two step process α → eα → α. The first step of this process induces a
natural graph homomorphism from α onto eα preserving the labels and the
as a subgraph of eα (preserving the labels and the left and right roots) and
as the graph obtained fromeα by 'contracting' to a vertex each non-essential
thorn {a, e} together with the other endpoint of e, say b, and keeping the
label of b (or a).
Let A(X) be the set of all reduced graphs from B′(X) and define a binary
operation ∧ on A(X) by setting
α ∧ β = α ⊓ β .
Set also α =
•
x
• for α = •
x
x
and let Θ(u) = ∆(u). The mapping
F2(X) → A(X), u → Θ(u) is a surjective homomorphism, and in fact it
is the canonical homomorphism which extends the mapping x →
•
x
•
x
9
for each x ∈ X.
pseudosemilattice on X if we identify each x ∈ X with
It was proved in [3] that A(X) is a model for the free
• . For future
x
•
x
reference we restate this result in the following proposition together with
some more information about A(X) obtained in [3]:
Proposition 2.2. The binary algebra (A(X), ∧) is a model for the free
pseudosemilattice on X if we identify each x ∈ X with
• . Further, the
x
•
x
maximal subsemilattices of A(X) are the sets
Sx,y(X) = {α ∈ A(X) l(α) = x and r(α) = y}
for x, y ∈ X, while the maximal right normal subbands and the maximal left
normal subbands are respectively
Rx(X) = {α ∈ A(X) l(α) = x}
Lx(X) = {α ∈ A(X) r(α) = x}
and
for x ∈ X.
Given an endomorphism ϕ of F2(X), we shall denote by ϕ the unique
endomorphism of A(X) that makes the following diagram commute:
F2(X)
Θ
A(X)
ϕ
ϕ
F2(X)
Θ
A(X)
•
•
y
•
•
x
)ϕ = y
Thus (Θ(u))ϕ = Θ(uϕ) = ∆(uϕ) for any u ∈ F2(X), and in particular
if xϕ = y ∈ X. Hence, if Xϕ ⊆ X, then we can
( x
obtain αϕ for α ∈ A(X) by first setting β to be the graph α but with each
label x changed to xϕ, and then reducing β; that is, αϕ = β. Conversely,
given an endomorphism ψ of A(X), we can construct an endomorphism ϕ
of F2(X) such that ψ = ϕ, namely by setting for each x ∈ X, xϕ ∈ F2(X)
)ψ. We should alert the reader that we shall
such that ∆(xϕ) = ( x
jump very often between endomorphisms ϕ of F2(X) and their corresponding
endomorphisms ϕ of A(X) without further notice.
x
•
•
x
•
x
if α =
distinguished roots and removing the existing thorns (that is, the thorns that
Let α ∈ A(X) and set bα = •
let bα be the (non-rooted) bipartite tree obtained from α by un-marking the
were essential in α). However, in this last case, the vertices of bα continue to
be divided in left and right vertices as they were in α. Define also
• for some x ∈ X; otherwise,
x
lα =(Lα \ {(lα, rα), rα}
Lα
if {(lα, rα), rα} is a thorn
otherwise
10
LU´IS OLIVEIRA
and αr dually. If α =
•
x
• then lα is the singleton graph •
x
x
considered as
a bipartite graph with one (distinguished) left vertex and no right vertex;
the dual is assumed for αr. The next result compiles the main results of
subsection 3.3 of [3].
Proposition 2.3. Let α, β ∈ A(X); then
(1) β ≤R α if an only if lα is a left-rooted subtree of lβ;
(2) β ≤L α if and only if αr is a right-rooted subtree of βr;
(3) β ≤ α if and only if α is a bi-rooted subtree of β;
(4) α R β if and only if lα = lβ;
(5) α L β if and only if αr = βr;
(6) α (R ∨ L ) β if and only if bα = bβ.
In particular, if α covers β (that is, β ≤ α and if β ≤ γ ≤ α for some γ then
β = γ or α = γ) and cl(β) ∩ cr(β) = ∅, then β has exactly one more vertex
(and one more edge) than α.
In this paper we shall talk about identities in the context of varieties
of pseudosemilattices. For example, when we say that two identities are
equivalent, we mean that the pseudosemilattices that satisfy one of them
are the same that satisfy the other one. For each variety V of pseu-
dosemilattices there exists a fully invariant congruence ρV(X) on A(X)
such that A(X)/ρV(X) is the relatively free algebra on X for the variety
V. Then, a set of identities I is a basis of identities for V if and only if
{(Θ(u), Θ(v)) u ≈ v ∈ I} generates ρV(X) as a fully invariant congruence
for X a countably infinite set. We shall write only ρV instead of ρV(X)
when we are considering the set X to be countably infinite. We shall say
that a binary relation σ on A(X) is a consequence of another binary rela-
tion τ if σ is contained in the fully invariant congruence generated by τ (or
equivalently, if the variety defined by the identities induced by σ contains
the variety defined by the identities induced by τ ). Two binary relations
on A(X) are said to be equivalent if they generate the same fully invariant
congruences (or equivalently, if the corresponding identities define the same
variety of pseudosemilattices).
By [1] an identity u ≈ v is satisfied by all strict pseudosemilattices if and
only if
(l(u), c2(u), r(u)) = (l(v), c2(v), r(v)) .
Thus (α, β) ∈ ρSPS if and only if c2(α) = c2(β) and α and β belong to the
same maximal subsemilattice of A(X). A pair (α, β) of elements of A(X) is
called elementary if
(1) (l(α), c2(α), r(α)) = (l(β), c2(β), r(β)),
(2) cl(β) ∩ cr(β) = ∅ (or equivalently cl(α) ∩ cr(α) = ∅),
(3) α covers β,
(4) and the unique degree 1 vertex in β \ α is adjacent to a distinguished
vertex of β.
11
Thus elementary pairs induce non-trivial identities satisfied by all strict
pseudosemilattices. In fact, for each elementary pair (α, β) we can always
find an identity u ≈ v such that (∆(u), ∆(v)) = (α, β) and v is obtained
from u by replacing either the first letter of u or the last letter of u, say
x, respectively with x ∧ y or y ∧ x for some y ∈ X. We compile in the
following result some conclusions obtained in [3] although not all of them
are explicitly stated their.
Proposition 2.4. If (α, β) is a non-trivial pair of ρSPS, then (α, β) is
equivalent to a finite set I of elementary pairs whose graphs all belong to
the same maximal subsemilattice of A(X) and have the same 2-content.
Further, if α has disjoint left and right contents, then (l(α1), c2(α1), r(α1)) =
(l(α), c2(α), r(α)) for each (α1, β1) ∈ I; otherwise, we can always say that
c(α1) ≤ 2c(α).
It was shown also in [3] that we do not need to be too rigorous about the
last condition in the definition of elementary pair.
Proposition 2.5. Let (α, β) be an elementary pair and let (a, b) be an edge
of α. Let α1 and β1 be respectively the graphs α and β but now with the
vertices a and b as the distinguished vertices. Then (α, β) and (α1, β1) are
equivalent.
For each integer n ≥ 2 let x1, . . . , x2n be distinct letters from X. We
designate by αn and βn the following graphs from A(X):
x2
•
x4
•
x2n
•
. . .
•
x3
•
x2n−1
•
x1
αn =
•
x1
x2n
•
βn =
x2
•
x4
•
x2n
•
. . .
and
•
x1
•
x3
•
x2n−1
•
x1
The pairs (αn, βn) for n ≥ 2 are obviously elementary pairs, and we shall
designate by Dn the 2-content of αn (or of βn). Thus, Dn is constituted by
{(x2i−1, x2i), (x2i+1, x2i) 1 ≤ i < n} ∪ {(x1, x2n), (x2n−1, x2n)}
together with {(xi, xi) 1 ≤ i ≤ 2n}. Since all vertices of both αn and
βn have degree at most 2, there are unique words un and vn such that
∆(un) = αn and ∆(vn) = βn. The main result of section 4 of [3] states that
the set of all identities un ≈ vn, n ≥ 2, is a basis of identities for the variety
of all strict pseudosemilattices.
Proposition 2.6. The set {un ≈ vn n ≥ 2} is a basis of identities for the
variety of all strict pseudosemilattices, or equivalently, the set {(αn, βn)
n ≥ 2} generates ρSPS as a fully invariant congruence for X a countably
infinite set.
We shall end this section by associating another graph to each γ ∈ B′(X).
Let γ be the graph underlying γ, that is, γ is just the graph γ but now with
no labels on the vertices and with no distinguished vertices. However, we
continue to see γ as a bipartite graph with 'left' and 'right' vertices as in γ,
12
LU´IS OLIVEIRA
that is, a left vertex of γ continues to be considered a left vertex in γ and
the same occurs for the right vertices.
3. Pairs of ρSPS with 2-content Dn
Fix n ≥ 2 for the next two sections. In this section we begin the study of
the pairs (α, β) ∈ ρSPS such that c2(α) = Dn, or in other words, we shall
begin to investigate the identities u ≈ v satisfied by all strict pseudosemi-
lattices and whose 2-content is Dn (that is, c2(u) = Dn). We shall prepare
here the ground for the next section where we determine all varieties of pseu-
dosemilattices defined by identities satisfied by all strict pseudosemilattices
and whose 2-content is Dn.
We first reinforce that we are fixing n ≥ 2. Note that for n = 1, the nat-
ural definition for D1 would be D1 = {(x1, x2), (x1, x1), (x2, x2)} and the
x2
only three graphs of A(X) with 2-content D1 are x1
x1
, x1
x2
•
•
•
•
•
•
•
x2•
x1
and x2
; no non-trivial pair formed by these graphs exists in ρSPS.
If α ∈ A(X) has 2-content Dn and has no essential thorn, then the under-
lying graph α is a 'zig-zag segment', that is, has one of the following four
configurations:
•
•
•
•
•
•
•
•
. . .
•
•
•
if it begins with a right vertex, and
•
•
•
. . .
or
or
•
•
•
•
. . .
. . .
•
•
•
•
•
•
•
•
•
•
•
•
if it begins with a left vertex (the first graph of each row are in fact isomor-
phic). Moreover, α becomes completely determined once we identify in α
the distinguished vertices and their labels. In other words, once we know
which vertices are the distinguished vertices and we know their labels, the
labels of all other vertices of α become fixed and easily determined. Fur-
thermore, since all vertices of α have degree at most 2, there exists a unique
word u ∈ F2(X) such that ∆(u) = α.
Lemma 3.1. Let α ∈ A(X) be such that c2(α) = Dn. If α has no essential
thorn, then there exists a unique word u ∈ F2(X) such that ∆(u) = α. If a
and b are two distinct and non-connected vertices of α with the same label,
then the geodesic path from a to b has either 2nk + 2 vertices if a or b belong
to some essential thorn of α, or 2nk + 1 otherwise, for some k ≥ 1.
Proof. The first part has been observed already above and follows from
the fact that all vertices of α have degree at most two. For the second
part, assume first that a and b do not belong to some essential thorn. Let
a = a0, a1, · · · , am = b be the sequence of vertices in the geodesic path from
a to b (we are assuming that two consecutive vertices are connected by an
13
edge). Let ca0 = xi. Then either ca1 = xi+1 (or ca1 = x1 if i = 2n), or
ca1 = xi−1 (or ca1 = x2n if i = 1). We shall assume that ca1 = xi+1 and
prove only this case since the other one is similar. Now, the labels of all
other vertices aj are fixed recursively by the rules: (i) if caj = x2n then
caj+1 = x1, and (ii) if caj = xl with l 6= 2n then caj+1 = xl+1. We can now
easily observe that caj = ca0 if and only if j = 2nk for some k ≥ 0.
In
particular m = 2nk for some k ≥ 1 and there are 2nk + 1 vertices in the
geodesic path from a = a0 to b = am.
Finally, assume that a belongs to some essential thorn. Then a is a
distinguished vertex of α with degree 1 and if c is the other distinguished
vertex of α then c and a have the same label. Thus c and b are two vertices
of α with the same label and not belonging to some essential thorn, whence
there are 2nk + 1 vertices in the geodesic path from c to b for some k ≥ 1.
It is clear now that there are 2nk + 2 vertices in the geodesic path from a
to b.
(cid:3)
Let (α, β) be an elementary pair with c2(α) = Dn. Then both α and β
have no essential thorns and their underlying graphs have each one of the
four configurations depicted above.
In fact, since (α, β) is an elementary
pair, we can assume that the distinguished vertices of α are the two first
vertices (going from the left to the right); and then β is obtained from α
by adding a vertex a on the left side and connecting it by an edge to the
leftmost vertex of α (if the leftmost vertex of α is a left vertex, then a is a
right vertex; otherwise a is a left vertex).
Let l(α) = xi (i odd) and consider the permutation
· · ·
· · · x2n+1−i (cid:19) .
x2n
σ =(cid:18)
x1
x2
x2n+2−i x2n+3−i
· · · xi−1 xi xi+1
· · ·
x2
x2n
x1
Extend σ to an automorphism ϕ of F2(X) by setting xϕ = x for any x 6∈
{x1, · · · , x2n}. Then, for each γ ∈ A(X), γϕ is just the graph γ but with
each label xi changed to σ(xi). Hence (αϕ, βϕ) is another elementary pair
equivalent to (α, β) and with c2(αϕ) = Dn. Further l(αϕ) = x1 and so
r(αϕ) = x2 or r(αϕ) = x2n. If r(αϕ) = x2n then consider the automorphism
xψ =( x2n+2−i
x
if x = xi and 1 < i ≤ 2n
otherwise
of F2(X) and observe that ((α)ϕ ◦ ψ, (β)ϕ ◦ ψ) is once more an elementary
pair equivalent to (α, β), with c2((α)ϕ◦ψ) = Dn, but now with distinguished
vertices labeled by x1 and x2.
Summing up the conclusions of the previous paragraph, we can assume
that the labels of the distinguished vertices of the graphs of an elementary
pair (α, β) with c2(α) = Dn are always x1 and x2, and so (α, β) becomes
completely determined once we know the number of vertices of α and if the
vertex in β \ α is a left vertex or a right vertex. Assume that α has m
vertices and write m = 2nk + i for k ≥ 1 and 1 ≤ i ≤ 2n (there is obviously
14
LU´IS OLIVEIRA
only one choice for k and i). Assume further that the vertex in β \ α is a
right vertex and let
x2
•
x4
•
xi
•
x2
•
x4
•
xi−1
•
λi =
. . .
or
λi =
. . .
•
x1
•
x3
•
xi−1
•
x1
•
x3
•
xi−2
•
xi
accordingly to i being even or odd. Then α is the graph
•
•
•
•
•
•
•
αn,k,i =
. . .
. . .
•
λ2n
•
•
•
λ2n
•
•
•
λ2n
λi
while β is the graph
x2n•
βn,k,i =
•
•
•
•
•
•
•
. . .
. . .
•
λ2n
•
•
•
λ2n
•
•
•
λ2n
λi
where the segment λ2n occurs k times in both αn,k,i and βn,k,i (if i = 2n
then the segment λ2n occurs in fact k + 1 times because the last segment λi
becomes another copy of λ2n).
Let us assume now that the vertex in β \ α is a left vertex. To highlight
the dual nature of this case, we need to introduce some dual concepts. Let
D∗
n = {(xj, xi) (xi, xj) ∈ Dn} and
x1
•
x3
•
λ∗
i =
xi−1
•
. . .
•
x2
•
x4
•
xi
or
λ∗
i =
x1
•
x3
•
xi−2
•
xi
•
. . .
•
x2
•
x4
•
xi−1
accordingly to i being even or odd. Consider the following permutation
τ =(cid:18) x1 x2
x3
x4
x2 x1 x2n x2n−1
· · · x2n
· · ·
x3 (cid:19) ,
and extend it to an automorphism ψ∗ of F2(X) in the obvious way. Then
(αψ∗, βψ∗) is another elementary pair equivalent to (α, β), with distin-
guished vertices labeled by x1 and x2, but now with c2(αψ∗) = D∗
n. Hence
αψ∗ is the graph
•
•
•
•
•
•
•
α∗
n,k,i =
. . .
•
λ∗
2n
•
•
•
λ∗
2n
•
•
λ∗
2n
. . .
•
λ∗
i
while βψ∗ is the graph
•
β∗
n,k,i =
•
•
•
•
•
•
. . .
•
x2n
•
λ∗
2n
•
•
•
λ∗
2n
•
•
λ∗
2n
. . .
•
λ∗
i
15
where the segment λ∗
2n occurs k times in both α∗
n,k,i and β∗
n,k,i.
We gather the previous conclusions in the following proposition.
Proposition 3.2. Let (α, β) be an elementary pair with c2(α) = Dn and
such that α has 2nk + i vertices. Then (α, β) is equivalent to (αn,k,i, βn,k,i)
if the vertex of β \ α is a right vertex or to (α∗
n,k,i) otherwise.
n,k,i, β∗
Our next result states that no two pairs (αn,k,i, βn,k,i) and (αn,l,j, βn,l,j)
are incomparable, that is, one of them is always a consequence of the other.
Proposition 3.3. Let k, l ≥ 1 and i, j ∈ {1, · · · , 2n}. If 2nk + i ≥ 2nl + j,
then (αn,k,i, βn,k,i) is a consequence of (αn,l,j, βn,l,j) and (α∗
n,k,i) is a
consequence of (α∗
n,k,i, β∗
n,l,j, β∗
n,l,j).
Proof. We shall prove only the (αn,k,i, βn,k,i) case since the (α∗
n,k,i) case
follows by symmetry. Further, the (αn,k,i, βn,k,i) case becomes proved once
we show both that (i) (αn,l,j+1, βn,l,j+1) is a consequence of (αn,l,j, βn,l,j) for
j < 2n and that (ii) (αn,l+1,1, βn,l+1,1) is a consequence of (αn,l,2n, βn,l,2n).
Let ϕ be the endomorphism of F2(X) fixing all x 6∈ {x1, · · · , x2n} and such
that
n,k,i, β∗
Then αn,l,jϕ is obtained from αn,l,j by first adding, for each vertex a labeled
with xj, a new vertex b connected to a by an edge and with label xj+1 (or
x1 if j = 2n); and then reducing this last graph. In the reducing process
all new vertices b are eliminated by edge-folding except for the last one
which is labeled with xj+1 (or x1 if j = 2n). Thus αn,l,jϕ = αn,l,j+1 for
j 6= 2n and αn,l,2nϕ = αn,l+1,1. Similarly βn,l,jϕ = βn,l,j+1 for j 6= 2n
and βn,l,2nϕ = βn,l+1,1. We have just proved statements (i) and (ii) as
desired.
(cid:3)
In the next two results we show that the two pairs (αn,k,i, βn,k,i) and
n,k,i, β∗
(α∗
n,k,i) are equivalent if i even. We shall see later on the next section
that the similar result for i odd does not hold true (we shall prove they are
incomparable for i odd).
Lemma 3.4. (α∗
i < 2n, and (α∗
n,k,i+1, β∗
n,k,i+1) is a consequence of (αn,k,i, βn,k,i) for each
n,k+1,1, β∗
n,k+1,1) is a consequence of (αn,k,2n, βn,k,2n).
Proof. Note that the second part of this result is the i = 2n version of the
first part, and its proof follows the same arguments as of the case i < 2n
with the expected adaptations. Therefore, we shall present here only the
proof of the case i < 2n.
Let a and b be respectively the left and right roots of α∗
c be the other vertex of α∗
β be respectively the graphs α∗
n,k,i+1 and let
n,k,i+1 connected to a by an edge. Let α′ and
n,k,i+1 but with the right root
n,k,i+1 and β∗
xjϕ =
xj ∧ xj+1
if j odd
x1 ∧ x2n
if j = 2n
xj+1 ∧ xj
otherwise.
16
LU´IS OLIVEIRA
changed from b to c. Then (α′, β) and (α∗
n,k,i+1) are equivalent by
Proposition 2.5. Let now α ∈ A(X) be the graph obtained from α′ by
deleting the vertex b and the edge (a, b). Then (α′, β) is a consequence of
(α, β) since β ≤ α′ ≤ α by Proposition 2.3.(3).
n,k,i+1, β∗
Consider the endomorphism ϕ of F2(X) fixing all x 6∈ {x1, · · · , x2n} and
such that x2nϕ = x2n ∧ x1 and xjϕ = xj+1 for 1 ≤ j < 2n. Then for each
γ ∈ A(X), γϕ = γ′ where γ′ ∈ B′(X) is the graph obtained from γ by
replacing each label xj with xj+1 for 1 ≤ j < 2n and each vertex •
with
x2n
•
•
x1
. Observe now that αn,k,iϕ = α and βn,k,iϕ = β, whence (α, β)
n,k,i+1) is
(cid:3)
x2n
is a consequence of (αn,k,i, βn,k,i). We have shown that (α∗
a consequence of (αn,k,i, βn,k,i) for each i < 2n.
Proposition 3.5. If i is even, then (αn,k,i, βn,k,i) and (α∗
equivalent.
n,k,i+1, β∗
n,k,i) are
n,k,i, β∗
Proof. We shall assume that i 6= 2n and prove only this case. The proof of
the case i = 2n follows the same arguments but with minor changes due to
the fact that (x1, x2n) belongs to Dn and not (x2n+1, x2n).
Let a be the only non-distinguished vertex of degree 1 of αn,k,i. Since
i is even, a is a right vertex labeled with xi. Let b be the left vertex of
αn,k,i connected to a by an edge, and let α and β be respectively the graphs
αn,k,i and βn,k,i but now with the vertices a and b as the distinguished
vertices. Then (αn,k,i, βn,k,i) and (α, β) are equivalent by Proposition 2.5.
Observe that we can obtain now (α∗
n,k,i+1) from (α, β) by relabeling
the vertices. To be more precise, consider the permutation
n,k,i, α∗
and extend it to an automorphism ϕ of F2(X); then α∗
α∗
n,k,i+1 = βϕ. Thus (α∗
by Lemma 3.4 so is (α∗
we conclude that (α∗
these two pairs are equivalent by symmetry of the arguments used.
n,k,i = αϕ and
n,k,i+1) is a consequence of (αn,k,i, βn,k,i), and
n,k,i+1). Finally, since β∗
n,k,i,
n,k,i) is a consequence of (αn,k,i, βn,k,i), and so
(cid:3)
n,k,i, α∗
n,k,i, β∗
n,k,i+1 ≤ β∗
n,k,i ≤ α∗
n,k,i, β∗
The arguments presented in the previous proof do not work properly for
the case i odd mainly because αϕ and βϕ become respectively the graphs
αn,k,i and αn,k,i+1 (and not α∗
n,k,i+1). For the case i odd, those
arguments allow us to conclude however that the pairs (αn,k,i, βn,k,i) and
(αn,k,i, αn,k,i+1) are equivalent. Curiously, we can use Proposition 3.5 itself
to prove that the previous conclusion also holds true for i even.
n,k,i and α∗
Lemma 3.6. The pairs (αn,k,i, βn,k,i) and (αn,k,i, αn,k,i+1) are equivalent for
i 6= 2n. Further (αn,k,2n, βn,k,2n) and (αn,k,2n, αn,k+1,1) are also equivalent.
Proof. As mentioned above we just need to prove this result for the case i
even. In fact, we shall assume also that i 6= 2n (the case i = 2n is similar
and only needs minor adaptations by the same reason mentioned in the
σ =(cid:18) x1
x2
xi xi−1
· · · xi−1 xi xi+1
· · ·
x2n
x2
x1
· · ·
x2n
· · · xi+1 (cid:19)
17
proof of Proposition 3.5). Let a and b be the vertices of αn,k,i considered in
the proof of Proposition 3.5. Let α and β′ be respectively the graphs αn,k,i
and αn,k,i+1 but with the vertices a and b as their distinguished vertices.
Thus (αn,k,i, αn,k,i+1) and (α, β′) are equivalent pairs by Proposition 2.5.
Consider again the permutation σ and the automorphism ϕ used in the
proof of Proposition 3.5; thus αϕ = α∗
n,k,i. A close analysis to the image
of β′ under ϕ allows us to conclude that β′ϕ = β∗
n,k,i. Hence (α, β′) and
(α∗
n,k,i) and (αn,k,i, βn,k,i) are
equivalent by Proposition 3.5 since i is even, whence (αn,k,i, αn,k,i+1) and
(αn,k,i, βn,k,i) are equivalent pairs too.
(cid:3)
n,k,i) are equivalent pairs. But (α∗
n,k,i, β∗
n,k,i, β∗
The graphs βn,k,i and αn,k,i+1 (or αn,k+1,1 if i = 2n) are the only graphs
covered by αn,k,i for the natural partial order with 2-content Dn. We shall
use this fact to prove that if (αn,k,i, βn,k,i) is a consequence of some I ⊆
A(X)× A(X), then (αn,k,i, βn,k,i) is a consequence of a single pair (α, β) ∈ I.
To prove this claim we shall mix the notion of pair of elements from A(X)
with the notion of identity. Let un,k,i and vn,k,i be (the unique) words of
F2(X) such that
∆(un,k,i) = αn,k,i
and ∆(vn,k,i) = βn,k,i ,
and let u∗
n,k,i and v∗
∆(u∗
n,k,i be (the unique) words of F2(X) such that
n,k,i) = α∗
n,k,i .
n,k,i
and ∆(v∗
n,k,i ≈ v∗
n,k,i) = β∗
n,k,i correspond respectively
Thus, the identities un,k,i ≈ vn,k,i and u∗
to the elementary pairs (αn,k,i, βn,k,i) and (α∗
n,k,i, β∗
n,k,i).
Lemma 3.7. If (αn,k,i, βn,k,i) is a consequence of I ⊆ A(X) × A(X), then
there exists (α, β) ∈ I such that (αn,k,i, βn,k,i) is a consequence of (α, β).
Proof. We shall prove the following equivalent statement: if un,k,i ≈ vn,k,i is
a consequence of a set J of identities, then there exists u ≈ v ∈ J such that
un,k,i ≈ vn,k,i is a consequence of u ≈ v. We may assume that all identities
in J are satisfied by all strict pseudosemilattices since an identity not satis-
fied by all strict pseudosemilattices defines a variety of normal bands, and
therefore implies un,k,i ≈ vn,k,i. Thus (l(u), c2(u), r(u)) = (l(v), c2(v), r(v))
for any u ≈ v ∈ J.
Let u ≈ v ∈ J and u′ ∈ F2(X) such that αn,k,i ≤ Θ(u′) and uϕ is a
subword of u′ for an endomorphism ϕ of F2(X). Let v′ be the word obtained
from u′ by replacing the subword uϕ with vϕ. This lemma becomes proved
once we show that if αn,k,i (cid:2) Θ(v′) then un,k,i ≈ vn,k,i is a consequence of u ≈
v. Indeed, if un,k,i ≈ vn,k,i is a consequence of J, then there exist a sequence
un,k,i = w0, w1, · · · , wl = vn,k,i of words from F2(X), identities rj ≈ sj ∈ J
(or sj ≈ rj ∈ J) and endomorphisms ϕj of F2(X) for j = 1, · · · , l such that
rjϕj is a subword of wj−1 and wj is obtained from wj−1 by replacing the
subword rjϕj with sjϕj. Now, since αn,k,i ≤ Θ(w0) and αn,k,i (cid:2) Θ(wl),
there exists some j such that αn,k,i ≤ Θ(wj−1) and αn,k,i (cid:2) Θ(wj). Thus,
18
LU´IS OLIVEIRA
after showing our claim above, we can conclude that un,k,i ≈ vn,k,i is a
consequence of rj ≈ sj ∈ J.
Let u, v, u′, v′ and ϕ be as above. First note that αn,k,i, Θ(u′) and
Θ(v′) belong to the same maximal subsemilattice of A(X) and that c2(u′) =
c2(v′) ⊆ Dn. Further, un,k,i ∧ u′ ≈ un,k,i ∧ v′ is also a consequence of u ≈ v.
We must have now
and
Θ(un,k,i ∧ u′) = αn,k,i ∧ Θ(u′) = αn,k,i
Θ(un,k,i ∧ v′) = αn,k,i ∧ Θ(v′) < αn,k,i
respectively because αn,k,i ≤ Θ(u′) and αn,k,i (cid:2) Θ(v′), whence (αn,k,i, α) is
a consequence of (Θ(u), Θ(v)) for α = Θ(un,k,i ∧ v′). Since βn,k,i and αn,k,i+1
(or αn,k+1,1 if i = 2n) are the only graphs of A(X) covered by αn,k,i and
with the same 2-content as αn,k,i, we conclude that
α ≤ βn,k,i ≤ αn,k,i or α ≤ αn,k,i+1 ≤ αn,k,i .
Thus (αn,k,i, βn,k,i) or (αn,k,i, αn,k,i+1) is a consequence of (Θ(u), Θ(v)). Fi-
nally, by Lemma 3.6, (αn,k,i, βn,k,i) is a consequence of (Θ(u), Θ(v)), that is,
un,k,i ≈ vn,k,i is a consequence of u ≈ v ∈ J.
(cid:3)
For n ≥ 2, let
In = {(αn,k,i, βn,k,i) , (α∗
n,k,j, β∗
n,k,j) k ≥ 1, 1 ≤ i, j ≤ 2n and j odd} .
By Propositions 3.2 and 3.5, any elementary pair with 2-content Dn is equiv-
alent to an elementary pair from In. By Lemma 3.7 and its dual, if a pair
from In is a consequence of a subset I of ρSPS, then it is a consequence of a
single pair from I. We shall end this section by proving that any subset of
ρSPS composed by non-trivial pairs all with 2-content Dn is equivalent to a
pair from In or to a set composed by two pairs from In.
Proposition 3.8. Let I be a subset of ρSPS composed by non-trivial pairs,
all with 2-content Dn. Then I is equivalent to a pair from In or to a subset
{(αn,k,i, βn,k,i), (α∗
n,k,i)} of In for some i odd.
n,k,i, β∗
Proof. This result becomes proved once we show that a non-trivial pair
(α, γ) from ρSPS with γ ≤ α and with 2-content Dn is equivalent to a
(finite) subset of In. Indeed, since any pair (α, β) ∈ ρSPS is equivalent to
the set {(α, α ∧ β), (β, α ∧ β)} and α ∧ β ≤ α and α ∧ β ≤ β, we can then
conclude that I is equivalent to a subset of In. This proposition then follows
immediately from Propositions 3.3 and 3.5 and Lemma 3.4.
So, let (α, γ) be a non-trivial pair of ρSPS with γ ≤ α and with 2-content
Dn, and let us prove that (α, γ) is equivalent to a subset of In. Thus α is
a bi-rooted subtree of γ. Assume first that lα belongs to an essential thorn
of α. Then lα = rα = lγ = rγ = xi for some i even. Let a be another vertex
connected to rα in α but distinct from lα. Set
α′ = Θ(ca) ∧ α and γ′ = Θ(ca) ∧ γ .
19
Then α′ is the graph obtained from α by deleting the vertex lα and the
edge (lα, rα), setting a as the left root and keeping rα as the right root (γ′
is obtained from γ similarly). But since
α = Θ(xi) ∧ α′ and γ = Θ(xi) ∧ γ′ ,
the pairs (α, γ) and (α′, γ′) are equivalent, and α′ has no essential thorn.
Further, γ′ ≤ α′ and the 2-content of α′ continues to be Dn. Thus, by
symmetry, we can assume that α (and γ) has no essential thorn. Now, since
Dn is the 2-content of α and α has no essential thorn, then the left and right
contents of α are disjoint. Thereby, by Proposition 2.4, (α, γ) is equivalent
to a finite subset of elementary pairs, all with 2-content Dn; and finally by
Proposition 3.2, (α, γ) is equivalent to a (finite) subset of In.
(cid:3)
We can formulate Proposition 3.8 in terms of varieties of pseudosemilat-
tices.
Corollary 3.9. Let V be a variety of pseudosemilattices defined by a set
of identities, all with 2-content Dn. Then V is defined by a single identity
un,k,i ≈ vn,k,i, or by a single identity u∗
n,k,i with i odd, or by {un,k,i ≈
vn,k,i, u∗
n,k,i ≈ v∗
n,k,i} with i odd.
n,k,i ≈ v∗
We shall see in the next section that no two pairs from In are equivalent,
n,k,i} with i odd is equivalent to a
n,k,i ≈ v∗
and that no set {un,k,i ≈ vn,k,i, u∗
pair from In.
4. Varieties defined by a single identity with 2-content Dn
Let k ≥ 1 and 1 ≤ i, j ≤ 2n with j odd, and set Vn,k,i and V∗
n,k,j
as the varieties of pseudosemilattices defined respectively by the identities
un,k,i ≈ vn,k,i and u∗
n,k,j ≈ v∗
n,k,j.
Proposition 4.1. The varieties Vn,k,i and V∗
∩-irreducible in the lattice L(PS).
n,k,j with j odd are complete
Proof. Let {Uj j ∈ J} be a family of varieties of pseudosemilattices such
that ∩j∈J Uj ⊆ Vn,k,i, and let Bj be a basis of identities for each Uj. Then
un,k,i ≈ vn,k,i is a consequence of ∪i∈J Bj. By Lemma 3.7, un,k,i ≈ vn,k,i is a
consequence of a single identity from ∪i∈J Bj; whence some Uj is contained
in Vn,k,i. The proof for V∗
(cid:3)
n,k,j is similar.
Corollary 4.2. Each variety Vn,k,i and each variety V∗
a unique cover in the lattice L(PS).
n,k,j with j odd has
Proof. We shall prove only the Vn,k,i case since the other one is similar. Let
V be the variety of pseudosemilattices obtained by intersecting all varieties
of pseudosemilattices containing Vn,k,i properly. Then Vn,k,i is properly
contained in V by Proposition 4.1 and all varieties containing Vn,k,i prop-
erly, contain also V. We have proved this corollary.
(cid:3)
20
LU´IS OLIVEIRA
Consider the set
{Vn,k,i, V∗
n,k,j, Vn,k,j ∩ V∗
n,k,j k ≥ 1, 1 ≤ i, j, ≤ 2n, j odd}
(4.1)
of varieties of pseudosemilattices. By Corollary 3.9, this set is composed
by all varieties of pseudosemilattices defined by sets of identities, all with
2-content Dn. This section has two main results. The first one will state
that no two varieties from (4.1) are the same. The second one will state that
(4.1) is also the set of all varieties of pseudosemilattices defined by a single
identity with 2-content Dn. Thus, for the latter result, it will be enough to
show that {(αn,k,j, βn,k,j), (α∗
n,k,j)} with j odd is equivalent to a single
identity with 2-content Dn. Of course, this last identity will not belong to
In because of the first result.
n,k,j, β∗
n,k,j, β∗
The key ingredient to prove that no two varieties from (4.1) are the same is
to show that the pairs (αn,k,j, βn,k,j) and (α∗
n,k,j) are incomparable for
j odd. But, to do so, we need to go back to [3] and recall Lemma 5.1 of that
paper which states that the words un+1,k,2n+2 with k ≥ 1 are isoterms for
the identity un,1,1 ≈ vn,1,1 relative to PS. In other words, if un+1,k,2n+2 ≈ v
is a consequence of un,1,1 ≈ vn,1,1 (or equivalently if (αn+1,k,2n+2, Θ(v)) is
a consequence of (αn,1,1, βn,1,1)), then un+1,k,2n+2 ≈ v is a trivial identity
(or equivalently αn+1,k,2n+2 = Θ(v)). Looking carefully to the proof of that
lemma one realizes that the proof works for any word u such that c2(u) = Dm
for m > n. In particular, we have the following result:
Lemma 4.3. For each m > n ≥ 2, k ≥ 1 and 1 ≤ i ≤ 2m, the word um,k,i
is an isoterm for the identity un,1,1 ≈ vn,1,1.
n,k,i, β∗
Although the previous lemma will be used only in the next section,
it is similar to the result that we need to prove that (αn,k,i, βn,k,i) and
(α∗
n,k,i is an isoterm for un,k,i ≈ vn,k,i
if i odd. The proof of this last result follows the same strategy used in Lemma
5.1 of [3] although it is more complex. We begin with the following auxiliary
lemma.
n,k,i) are incomparable if i odd: u∗
Lemma 4.4. Let ϕ be an endomorphism of F2(X) such that c2(un,k,iϕ) ⊆
D∗
n. Then un,k,iϕ ≈ vn,k,iϕ is a trivial identity or sk(un,k,i, ϕ) ∈ A(X) with
c2(sk(un,k,i, ϕ)) = D∗
n.
Proof. We show first that we can consider k = 1 and i = 1 and prove only
this case. Observe that sk(un,k,i, ϕ) = ∆(un,k,iψ) for ψ an endomorphism
of F2(X) such that xiψ = l(xiϕ) if i odd and xiψ = r(xiϕ) if i even. Thus
sk(un,k,i, ϕ) is just the graph αn,k,i but with each label xi changed to xiψ.
Since the labels in the graphs sk(un,k,i, ϕ) are 'periodic', we immediately
conclude that sk(un,k,i, ϕ) is reduced if and only if sk(un,1,1, ϕ) is reduced too,
and in this case both c2(sk(un,k,i, ϕ)) and c2(sk(un,1,1, ϕ)) are D∗
n. On the
other hand, by Proposition 3.3, if un,1,1ϕ ≈ vn,1,1ϕ is a trivial identity, then
so is un,k,iϕ ≈ vn,k,iϕ since un,k,i ≈ vn,k,i is a consequence of un,1,1 ≈ vn,1,1.
Summing up, we can assume that k = 1 and i = 1 and prove this result only
for this case.
Let α be the subtree sk(un,1,1, ϕ) = ∆(un,1,1ψ) of ∆(un,1,1ϕ) and assume
that α is not reduced. We need to prove that un,1,1ϕ ≈ vn,1,1ϕ is a trivial
identity. Let a = lα and b be the two vertices of degree 1 of α (and of αn,1,1);
then ca = cb = x1ψ in α. We shall prove first that a and b merge into a
21
of edge-foldings to α. Let xi = x1ψ. We shall consider two cases: i odd and
i even.
single vertex in eα, that is, we can identify a with b by applying a sequence
We begin assuming i even. If eα = α then α must have a non-essential
thorn since it is not reduced; but the only candidate to a non-essential thorn
is {(b, c), b} where c is the only vertex connected to b by an edge. Let d be the
other vertex connected by an edge to c in α. Since i is even and c2(α) ⊆ D∗
n,
cd = cc = cb = xi and we can merge d with b by an edge-folding, whence
eα 6= α which contradicts our assumption. Thereby, we can assume that
eα 6= α. Let β be the geodesic path from a to b in eα. Then β has at most
2n − 1 vertices. If β has more than one vertex, then fix a and the other
vertex connected to a by an edge in β as the distinguished vertices of β
(a is obviously the left root). Then β belongs to B′(X) and is reduced for
edge-foldings. Since i is even we can conclude as above that β has no non-
essential thorn, whence β ∈ A(X). Finally, by the dual of Lemma 3.1, any
graph γ ∈ A(X) with c2(γ) = D∗
n and with two distinct vertices with the
same label must have at least 2n+1 vertices in the geodesic path connecting
those two vertices.
In other words, β must have at least 2n + 1 vertices
which is not the case. We can now conclude that β has only one vertex and
therefore a and b are identified by a sequence of edge-foldings applied to α.
Assume now that x1ψ = xi with i odd, and let c and d be the only two
vertices connected respectively to a and b in α. Then cd = cc = cb =
ca = xi since c2(α) ⊆ D∗
n. An argumentation similar to the one applied in
the previous paragraph allows us to conclude that c and d can be merged
together by applying a sequence of edge-foldings to α. Thus we can identify
a with b by applying one more edge-folding.
Summing up the two previous paragraphs, we proved that we can merge
together the two vertices a and b by applying a sequence of edge-foldings to
α. Since α is a subgraph of ∆(vn,1,1ϕ), we can apply that same sequence of
edge-foldings to ∆(vn,1,1ϕ) and merge together the vertices a and b also in
∆(vn,1,1ϕ). Observe now that we have two copies of x2nϕ attached to the
vertex that results from merging together a and b. These two copies of x2nϕ
can be reduced to a single copy of x2nϕ by another sequence of edge-foldings.
Finally, we just have to observe that this last graph (the one resulting from
reducing the two copies of x2nϕ to a single copy) is the graph obtained from
∆(un,1,1ϕ) by applying the sequence of edge-foldings mentioned above that
merge together a and b. Therefore ∆(vn,1,1ϕ) = ∆(un,1,1ϕ) and un,1,1ϕ ≈
vn,1,1ϕ is a trivial identity.
(cid:3)
22
LU´IS OLIVEIRA
Lemma 4.5. Let k ≥ 1 and 1 ≤ i ≤ 2n. If i odd, then u∗
for the identity un,k,i ≈ vn,k,i (relative to PS).
n,k,i is an isoterm
Proof. Let u ∈ F2(X) such that u∗
n,k,i ≈ u is a trivial identity and let ϕ be
an endomorphism of F2(X) such that un,k,iϕ or vn,k,iϕ is a subword of u.
In particular, un,k,iϕ is always a subword of u. Let v be the word obtained
from u by replacing the subword un,k,iϕ or vn,k,iϕ with respectively vn,k,iϕ
or un,k,iϕ. This result becomes proved once we show that u ≈ v is a trivial
identity.
Since un,k,iϕ is a subword of u and u∗
n,k,i ≈ u is a trivial identity, we
must have c2(un,k,iϕ) ⊆ D∗
n. Let α = sk(un,k,i, ϕ). Now, by Lemma 4.4,
un,k,iϕ ≈ vn,k,iϕ is a trivial identity or α ∈ A(X) with c2(α) = D∗
n. If we
show that the former case must occur, then u ≈ v is also a trivial identity
and we are done. So, assume otherwise that α ∈ A(X) with c2(α) = D∗
n.
Since α is a subgraph of ∆(un,k,iϕ) which in turn is a subgraph of ∆(u),
we conclude that α is a subgraph of ∆(u). Thus α = eα is a subgraph of
]∆(u). But note that α has no essential thorn either, and so α is in fact a
subgraph of ∆(u) = α∗
n,k,i. Finally, we get a contradiction by counting the
number of left vertices of α and α∗
n,k,i has one less left vertex than
α since i is odd. Hence α cannot be a subgraph of α∗
n,k,i. Consequently,
un,k,iϕ ≈ vn,k,iϕ is a trivial identity, and we have proved this lemma.
(cid:3)
n,k,i: α∗
We have now all the tools we need to prove that the pairs (αn,k,i, βn,k,i)
and (α∗
n,k,i, β∗
n,k,i) are incomparable for i odd.
Proposition 4.6. The pairs (αn,k,i, βn,k,i) and (α∗
rable if i odd.
n,k,i, β∗
n,k,i) are incompa-
Proof. Assume i odd. By the last lemma the identity u∗
n,k,i is not a
consequence of un,k,i ≈ vn,k,i, that is, (α∗
n,k,i) is not a consequence of
(αn,k,i, βn,k,i). By symmetry (using the dual version of the previous lemma),
neither (αn,k,i, βn,k,i) is a consequence of (α∗
n,k,i), whence these two
pairs are incomparable.
(cid:3)
n,k,i ≈ v∗
n,k,i, β∗
n,k,i, β∗
We can now prove that no two varieties from the list (4.1) are the same.
Proposition 4.7. The varieties listed in (4.1) are all pairwise distinct va-
rieties. Further, (4.1) lists all varieties defined by sets of identities, all with
2-content Dn.
Proof. The second part is just a reformulation of Corollary 3.9. Thus we
only need to prove the first part. Let k ≥ 1 and i ∈ {1, · · · , 2n} odd, and
set
U2nk+i = {Vn,k,i, V∗
n,k,i, Vn,k,i ∩ V∗
n,k,i, Vn,k,i+1} .
Then the list (4.1) is the union of all sets U2nk+i. The varieties from each
U2nk+i are pairwise distinct by Propositions 3.3, 3.5 and 4.6, and the inclu-
sion relation between them is given by the following scheme:
23
Vn,k,i+1
Vn,k,i
V∗
n,k,i
Vn,k,i ∩ V∗
n,k,i
Furthermore, by these same results and Proposition 4.1, if 2nk + i < 2nl + j
with l ≥ 1 and j ∈ {1, · · · , 2n} odd, then any variety from U2nk+i is properly
contained in any variety from U2nl+j; and we have shown this result.
(cid:3)
In the previous proof we have shown more than what it is stated in the
result. We have proved that the list (4.1), under the inclusion relation,
constitutes an infinite ascending 'chain of diamonds' like the one depicted
in the proof above. The following corollary is now obvious.
Corollary 4.8. Let k, l ≥ 1 and i, j ∈ {1, · · · , 2n}.
(i) (αn,k,i, βn,k,i) is a consequence of (αn,l,j, βn,l,j) if and only if 2nk+i ≥
2nl + j.
(ii) If i and j are odd, then (α∗
n,k,i, β∗
if and only if 2nk + i ≥ 2nl + j.
n,k,i) is a consequence of (α∗
n,l,j, β∗
n,l,j)
(iii) If i odd, then (α∗
n,k,i, β∗
n,k,i) is a consequence of (αn,l,j, βn,l,j) if and
only if 2nk + i > 2nl + j.
(iv) If j odd, then (αn,k,i, βn,k,i) is a consequence of (α∗
n,l,j, β∗
n,l,j) if and
only if 2nk + i > 2nl + j.
We end this section by showing that the list (4.1) is also the list of all
varieties of pseudosemilattices defined by a single identity with 2-content
Dn. In fact, we just need to prove that un,k,i ≈ vn,k,i−1 defines the variety
Vn,k,i ∩ V∗
n,k,i for i > 1 odd and that un,k,1 ≈ vn,k−1,2n defines the variety
Vn,k,1 ∩ V∗
n,k,1.
Proposition 4.9. The varieties from (4.1) are precisely the varieties of
pseudosemilattices defined by a single identity with 2-content Dn.
Proof. We just need to prove the two claims above. As for earlier re-
sults, the proof of the case i = 1 is similar to the proof of the case i 6=
1 but needs minor obvious adaptations. Therefore, we shall prove only
the general case i 6= 1. So, assume that i is odd and greater than 1.
Since αn,k,i ∧ βn,k,i−1 = βn,k,i, the pair (αn,k,i, βn,k,i−1) is equivalent to
the set {(αn,k,i, βn,k,i), (βn,k,i−1, βn,k,i)}. Let a be the only vertex from
βn,k,i \ βn,k,i−1; then a is a left vertex. Let b and c be the vertices of βn,k,i
such that (a, b) and (c, b) are edges of βn,k,i (b and c are clearly unique);
and let α′ and β′ be respectively the graphs βn,k,i−1 and βn,k,i but with
b and c as the distinguished vertices. Thus (βn,k,i−1, βn,k,i) and (α′, β′) are
equivalent by Proposition 2.5. Finally, we can obtain α∗
n,k,i respec-
tively from α′ and β′ by relabeling the vertices using a permutation; whence
n,k,i and β∗
24
LU´IS OLIVEIRA
(α′, β′) is equivalent to (α∗
and {un,k,i ≈ vn,k,i, u∗
n,k,i ≈ v∗
defines the variety Vn,k,i ∩ V∗
n,k,i, β∗
n,k,i). We have proved that un,k,i ≈ vn,k,i−1
n,k,i} are equivalent, that is, un,k,i ≈ vn,k,i−1
n,k,i .
(cid:3)
5. Comparing Vn,k,i with Vm,l,j
In the previous section we found and listed all the varieties of pseudosemi-
lattices defined by sets of identities, all with 2-content Dn, and we studied
their inclusion relation. Further, we showed that this list is also the list of
all varieties of pseudosemilattices defined by a single identity with 2-content
Dn.
In the present section we shall study the inclusion relation between
varieties of pseudosemilattices defined by identities all with 2-content Dn
and varieties of pseudosemilattices defined by identities all with 2-content
Dm for n 6= m. We begin by comparing Vn,k,i with Vm,l,j for n, m ≥ 2,
k, l ≥ 1, 1 ≤ i ≤ 2n and 1 ≤ j ≤ 2m, and we claim that Vn,k,i ⊆ Vm,l,j if
and only if m ≤ n and either l > k, or k = l and j ≥ i + 2m − 2n. The first
result of this section is precisely the 'if' part of our claim although stated in
terms of elementary pairs from ρSPS.
Proposition 5.1. Let 2 ≤ m ≤ n, l, k ≥ 1, i ∈ {1, · · · , 2n} and j ∈
{1, · · · , 2m}. If either l > k, or l = k and j ≥ i+2m−2n, then (αm,l,j, βm,l,j)
is a consequence of (αn,k,i, βn,k,i) .
Proof. Let ϕ be an endomorphism of F2(X) such that
xpϕ =( x1
xp+2m−2n
for 1 ≤ p ≤ 2n − 2m
for 2n − 2m < p ≤ 2n ,
and let j1 = max{1, i + 2m − 2n}. Then αn,k,iϕ = αm,k,j1 and βn,k,iϕ =
βm,k,j1, and so (αm,k,j1, βm,k,j1) is a consequence of (αn,k,i, βn,k,i) . Now, if
l > k then 2ml + j ≥ 2mk + j1; and if l = k and j ≥ i + 2m − 2n, then j ≥ j1
and 2ml + j ≥ 2mk + j1. By Proposition 3.3 we conclude that (αm,l,j, βm,l,j)
is a consequence of (αm,k,j1, βm,k,j1), and we have shown that (αm,l,j, βm,l,j)
is a consequence of (αn,k,i, βn,k,i) as desired.
(cid:3)
We begin working now towards the proof of the 'only if' part of our
claim. Our first observation is that m must be less than or equal to n.
Indeed, if m > n then um,l,j is an isoterm for the identity un,k,i ≈ vn,k,i by
Lemma 4.3 and Proposition 3.3; thus um,l,j ≈ vm,l,j cannot be a consequence
of un,k,i ≈ vn,k,i. Before we can give a formal proof of the 'only if' part
of our claim, we need to do a deep analysis onto the structure of αn,k,iϕ
for ϕ an endomorphism of F2(X) such that c2(un,1,1ϕ) = c2(un,k,iϕ) ⊆
Dm. This analysis will culminate with the proof of Lemma 5.4. So, fix an
endomorphism ϕ of F2(X) such that c2(un,k,iϕ) ⊆ Dm for m ≤ n.
Let a1, a2, · · · , a2n+1 designate sequentially the vertices of αn,1,1 with a1
its left root. Hence, a2 is its right root and the edges of αn,1,1 are the
pairs (ap−1, ap) and (ap+1, ap) for p even in between 2 and 2n. Set β =
sk(un,1,1, ϕ). Thus β has the same underlying graph as αn,1,1 but with each
25
label xp changed to l(xpϕ) if p odd and to r(xpϕ) if p even. Consider the
natural graph homomorphism πβ : β → eβ from β onto eβ.
Lemma 5.2. If a1πβ = a2n+1πβ then un,k,iϕ ≈ vn,k,iϕ is a trivial identity
for any k ≥ 1 and any i ∈ {1, · · · , 2n}.
Proof. First note that the general case follows from the case case k = 1
and i = 1 since un,k,i ≈ vn,k,i is a consequence of un,1,1 ≈ vn,1,1. We shall
conclude that un,1,1ϕ ≈ vn,1,1ϕ is a trivial identity by an argumentation
already used in previous results. We can start by identifying the vertices a1
and a2n+1 in ∆(vn,1,1ϕ) by the same sequence of edge-foldings used in β to
identify these same vertices. Then two copies of x2nϕ become attached to
the vertex a1 and they can be reduced to a single copy again by a sequence
of edge-foldings. We can observe now that this latter graph is just the graph
obtained from ∆(un,1,1ϕ) by identifying the vertices a1 and a2n+1 using again
the same sequence of edge-foldings used in β. Thus Θ(un,1,1ϕ) = Θ(vn,1,1ϕ)
or equivalently un,1,1ϕ ≈ vn,1,1ϕ is a trivial identity.
(cid:3)
Now, assume that a1πβ 6= a2n+1πβ. Thus a1πβ and a2n+1πβ are two
l(x1ϕ), the label of a1 in β. If h is odd then let a1πβ = b1, b2, · · · , br+1 =
distinct left vertices of eβ with the same label. Since eβ is reduced for edge-
folding and c2(eβ) ⊆ Dm, we must have c2(β) = c2(eβ) = Dm. Let xh =
a2n+1πβ be the geodesic path from a1πβ to a2n+1πβ in eβ; otherwise consider
instead the geodesic path a2πβ = b1, b2, · · · , br+1 = a2nπβ from a2πβ to
a2nπβ; designate this geodesic path by β′. Thus b1 6= br+1 and cb1 = cbr+1
independently of h being odd or even (note that ca2 = ca1 = ca2n+1 = ca2n if
h even). Further, β′ is obviously edge-folding reduced and has no (essential
and non-essential) thorn. Hence, β′ is also a geodesic path in β ∈ A(X). By
Lemma 3.1, r = 2ms for some s ≥ 1. If cb2 = cbr , then cb3 = cbr−1 because
c2(β) = Dm, cb3 6= cb1 and cbr−1 6= cbr+1; continuing this process we would
conclude that cbms = cbms+2, which contradict the fact that β′ is edge-folding
reduced. Hence cb2 6= cbr . We have proved the following lemma.
Lemma 5.3. With the notation introduced above, if a1πβ 6= a2n+1πβ then
r = 2ms for some s ≥ 1 and cb2 6= cbr .
Consider now the graph αn,k,i and let a1, a2, · · · , a2nk+i designate sequen-
tially the vertices of αn,k,i with a1 its left root. We can view αn,1,1 as the sub-
graph of αn,k,i spanned over the vertices a1, a2, · · · , a2n+1 (or spanned over
the vertices a2n(t−1)+1, a2n(t−1)+2, · · · , a2nt+1 for 1 ≤ t ≤ k with a2n(t−1)+1
and a2n(t−1)+2 as the distinguished vertices). Set α = sk(un,k,i, ϕ). Thus
α has the same underlying graph as αn,k,i but with each label xp changed
to l(xpϕ) if p odd and to r(xpϕ) if p even.
In particular α is also a 'pe-
riodic' graph in the sense that the vertices ap and ap+2n have the same
label in α. Thus, if βt denotes the subgraph of α spanned over the vertices
{aq 2n(t − 1) + 1 ≤ q ≤ 2nt + 1} for each t ∈ {1, · · · , k}, then all these
subgraphs βt are isomorphic to β (considering a2n(t−1)+1 and a2n(t−1)+2 the
26
LU´IS OLIVEIRA
distinguished vertices of βt). Let also βk+1 be the subgraph of α spanned
over the vertices {a2nk+1, · · · , a2nk+i} (that is, βk+1 is the last incomplete
copy of β inside α). The (underlying structure of the) graph α can be
depicted as follows (for i even):
a2
•
a2n
•
a2n+2
•
a4n
•
a2n(k−1)+2
•
a2nk
•
a2nk+2
•
a2nk+i
•
α =
. . .
. . .
. . .
. . .
. . .
•
a1
•
a2n−1
•
a2n+1
•
a4n−1
•
a4n+1
β
β
•
•
a2n(k−1)+1
a2nk−1
β
•
a2nk+1
•
a2nk+i−1
βk+1
Let us look to the subgraph γ = sk(un,k,1, ϕ) of α under the assumption
that a1πβ 6= a2n+1πβ. Note that γ is a sequence of k graphs β1, β2, · · · , βk,
each one a copy of β, such that the last vertex of βt−1 is the first vertex
of βt for 1 < t ≤ k. Let γ′ be the graph obtained from γ by reducing
inside γ each subgraph βt to eβt by edge-folding. Thus each subgraph eβt of
γ′ contains a geodesic path b1,t, b2,t · · · , br+1,t isomorphic to b1, b2, · · · , br+1.
Further, if h is odd, we can assume that br+1,t−1 = b1,t; and if h is even, we
can assume that we can merge together br+1,t−1 and b1,t by an edge-folding,
for 1 < t ≤ k. Now, set γ+ = γ′ if h odd and set γ+ to be the graph obtained
from γ′ by merging together by edge-folding each br+1,t−1 and b1,t if h even.
Thus
b1,1, · · · , br,1, b1,2, · · · , br,2, · · · , b1,k, · · · , br,k, br+1,k
is a path in γ+ with no thorns. Since cb2 6= cbr , this path is edge-folding
reduced too. Hence γ contains this path. Finally, since r = 2ms for some
s ≥ 1, the previous path has at least 2mk + 1 vertices.
Lemma 5.4. Let m ≤ n, k ≥ 1 and 1 ≤ i ≤ 2n, and consider an endomor-
phism ϕ of F2(X) such that c2(un,k,iϕ) ⊆ Dm. If un,k,iϕ ≈ vn,k,iϕ is not
a trivial identity, then αn,k,iϕ has a geodesic path with no (essential) thorn
and with at least 2mk + j vertices where
(i) j = max{1, i + 2m − 2n} if the geodesic path starts with a left vertex,
or
(ii) j = max{1, i + 1 + 2m − 2n} if the geodesic path starts with a right
vertex.
Proof. We shall assume that un,k,iϕ ≈ vn,k,iϕ is not a trivial identity and so
we need to prove that αn,k,iϕ has a geodesic path with no essential thorn
and with at least 2mk + j vertices. Since α = sk(un,k,i, ϕ) is a subgraph of
∆(un,k,iϕ) (with the same distinguished vertices), α is also a subgraph of
αn,k,iϕ = ∆(un,k,iϕ). Hence, we just need to show that α has a geodesic
path with no thorns and with at least 2mk + j vertices.
Let γ = sk(un,k,1, ϕ) and denote by η the geodesic path
b1,1, · · · , br,1, b1,2, · · · , br,2, · · · , b1,k, · · · , br,k, br+1,k
inside γ constructed above. Let now γ1 = sk(un,k+1,1, ϕ). By the same
argumentation made prior to this result, we can extend η to a geodesic path
η1 inside γ1:
b1,1, · · · , br,1, b1,2, · · · , br,2, · · · , b1,k, · · · , br,k, b1,k+1, · · · , br,k+1, br+1,k+1
(note that br+1,k = b1,k+1). Further, the path b1,k+1, · · · , br,k+1, br+1,k+1 is
another copy of b1, b2, · · · , br+1, and η1 has r(k + 1) + 1 = 2ms(k + 1) + 1
vertices and no thorn.
27
a maximal p ∈ {1, · · · , r + 1} such that
b1,1, · · · , br,1, b1,2, · · · , br,2, · · · , b1,k, · · · , br,k, b1,k+1, · · · , bp,k+1
Now, γ is a subgraph of α which in turn is a subgraph of γ1. Thuseγ is a
subgraph of eα which in turn is a subgraph of eγ1. In particular, there exists
is a geodesic path ofeα. Let η2 be this geodesic path and let πγ1 : γ1 → eγ1 be
the natural graph homomorphism from γ1 onto eγ1. Since γ1 \α has 2n+1−i
vertices, eγ1 \ (απγ1 ) has at most 2n + 1 − i vertices. Hence
p ≥ r + 1 − (2n + 1 − i) = r + i − 2n ≥ 2m + i − 2n.
However, if b1,1 is a right vertex (that is, b1 = a2πβ), then a2n(k+1)+1πγ1 6∈ η1
and so
p ≥ r + 1 − (2n − i) ≥ 2m + i + 1 − 2n.
Let j = max{1, 2m + i − 2n} if b1,1 is a left vertex and let j = max{1, 2m +
i + 1 − 2n} if b1,1 is a right vertex. Then η2 has at least 2mk + j vertices.
Since η2 is edge-folding reduced and has no thorns, we conclude that η2 is
also a geodesic path in α with at least 2mk + j vertices.
(cid:3)
We can finish now the proof of our claim.
Proposition 5.5. Let n, m ≥ 2, i ∈ {1, · · · , 2n}, j ∈ {1, · · · , 2m} and
k, l ≥ 1. Then (αm,l,j, βm,l,j) is a consequence of (αn,k,i, βn,k,i) if and only if
(i) n ≥ m and l > k, or
(ii) n ≥ m, l = k and j ≥ i + 2m − 2n.
Proof. By Proposition 5.1 we only need to prove the 'only if' part. Assume
that (αm,l,j, βm,l,j) is a consequence of (αn,k,i, βn,k,i). We have observed
already that we must have m ≤ n. So, we just need to prove that either
l > k, or l = k and j ≥ i + 2m − 2n. We shall prove this by assuming
the opposite and getting a contradiction. Hence, assume that l < k or that
l = k and j < i + 2m − 2n.
Let um,l,j ≈ u be a trivial identity and let ϕ be an endomorphism of F2(X)
such that un,k,iϕ or vn,k,iϕ is a subword of u. In particular, un,k,iϕ is always
a subword of u and c2(un,k,iϕ) ⊆ Dm. By the previous lemma, un,k,iϕ ≈
vn,k,iϕ is a trivial identity or αn,k,iϕ = Θ(un,k,iϕ) has a geodesic path with
no thorns and with at least 2mk + j1 vertices for j1 = max{1, i + 2m − 2n}.
But if the latter case occurs, then αm,l,j = Θ(u) would contain that same
geodesic path, whence 2ml + j ≥ 2mk + j1 and either l > k, or l = k and j ≥
j1 ≥ i+2m−2n. By the assumption we made, we must have the former case,
that is, un,k,iϕ ≈ vn,k,iϕ is a trivial identity. It is evident now that, under
our assumption, the word um,l,j is an isoterm for the identity un,k,i ≈ vn,k,i,
28
LU´IS OLIVEIRA
and so (αm,l,j, βm,l,j) cannot be a consequence of (αn,k,i, βn,k,i). Therefore,
for (αm,l,j, βm,l,j) to be a consequence of (αn,k,i, βn,k,i), we must have, beside
m ≤ n, either l > k, or l = k and j ≥ i + 2m − 2n.
(cid:3)
To deal with the case where one of the pairs is the dual pair (α∗
m,l,j, β∗
m,l,j)
for j odd, we begin with the following lemma.
Lemma 5.6. Let n > m ≥ 2, k ≥ 1 and i ∈ {1, · · · , 2n} odd such that j =
i + 2m − 2n ≥ 1. Then (α∗
m,k,j) is not a consequence of (αn,k,i, βn,k,i).
m,k,j, β∗
Proof. Let ψ be the automorphism of F2(X) induced by the permutation
τ =(cid:18) x1 x2
x3
x4
x2 x1 x2m x2m−1
· · · x2m
· · ·
x3 (cid:19) ,
m,k,jψ and β = β∗
and let α = α∗
m,k,jψ. Thus α and β are obtained from
α∗
m,k,j and β∗
m,k,j, respectively, by replacing each label xt with xtψ. Then
(α, β) is equivalent to (α∗
m,k,j), and c2(α) = Dm. Let u, v ∈ F2(X)
be such that ∆(u) = α and ∆(v) = β. Let also u ≈ u′ be a trivial identity
and ϕ be an endomorphism of F2(X) such that un,k,iϕ is a subword of u′.
This result becomes proved once we show that un,k,iϕ ≈ vn,k,iϕ is a trivial
identity. Indeed, this fact implies that u is an isoterm for un,k,i ≈ vn,k,i, and
so (α∗
m,k,j) is not a consequence of (αn,k,i, βn,k,i).
m,k,j, β∗
m,k,j, β∗
So, assume that un,k,iϕ ≈ vn,k,iϕ is not a trivial identity. Then, by Lemma
5.4, αn,k,iϕ has a geodesic path η with no thorn and with at least 2mk + j1
vertices for j1 = j if η starts with a left vertex and for j1 = j + 1 if η starts
with a right vertex. Since un,k,iϕ is a subword of u′ and η has no thorn,
than η is a subgraph of α. But note that the longest geodesic path in α has
2mk + j vertices and it starts with right vertices. Hence un,k,iϕ ≈ vn,k,iϕ
must be a trivial identity.
(cid:3)
Corollary 5.7. Let n, m ≥ 2, i ∈ {1, · · · , 2n}, j ∈ {1, · · · , 2m} odd and
k, l ≥ 1.
n,k,i, β∗
n,k,i) if and only if
(1) (α∗
m,l,j) is a consequence of (α∗
m,l,j, β∗
(i) n ≥ m and l > k; or
(ii) n ≥ m, l = k and j ≥ i + 2m − 2n.
m,l,j, β∗
(i) n ≥ m and l > k; or
(ii) n ≥ m, l = k and j > i + 2m − 2n.
(2) (α∗
m,l,j) is a consequence of (αn,k,i, βn,k,i) if and only if either
(3) (αn,k,i, βn,k,i) is a consequence of (α∗
m,l,j, β∗
m,l,j) if and only if either
(i) m ≥ n and k > l; or
(ii) m ≥ n, k = l and i > j + 2n − 2m.
Proof. (1) is just the dual of Proposition 5.5, while (3) is the dual of (2).
Hence, we shall prove only (2). If i is even, then (αn,k,i, βn,k,i) is equivalent
to (α∗
n,k,i) and (2) follows from (1) in this case (j > i + 2m − 2n
if j ≥ i + 2m − 2n because j is odd and i is even). Thus, assume i is
odd, and note that (α∗
n,k,i+1) is a consequence of (αn,k,i, βn,k,i) by
n,k,i+1, β∗
n,k,i, β∗
29
Corollary 4.8. Thereby, if either n ≥ m and l > k, or n ≥ m, l = k and
j > i + 2m − 2n, then (α∗
m,l,j) is a consequence of (αn,k,i, βn,k,i) since
it is a consequence of (α∗
n,k,i+1) by (1); we have shown the 'if' part
of (2) for i odd. But if m ≤ n, l = k and j = i + 2m − 2n, we know from
Lemma 5.6 that (α∗
m,l,j) is not a consequence of (αn,k,i, βn,k,i). The
'only if' part for i odd follows now from Corollary 4.8.
(cid:3)
m,l,j, β∗
n,k,i+1, β∗
m,l,j, β∗
Acknowledgments: This work was partially supported by the European
Regional Development Fund through the programme COMPETE and by
the Portuguese Government through the FCT Funda¸cao para a Ciencia e
a Tecnologia under the project PEst-C/MAT/UI0144/2011.
References
[1] K. Auinger, The word problem for the bifree combinatorial strict regular semigroup,
Math. Proc. Cambridge Philos. Soc. 113 (1993), 519 -- 533.
[2] K. Auinger, On the lattice of existence varieties of locally inverse semigroups, Canad.
Math. Bull. 37 (1994), 13 -- 20.
[3] K. Auinger and L. Oliveira, On the variety of strict pseudosemilattices, Studia Sci.
Math. Hungarica 50 (2013), 207 -- 241.
[4] K. Byleen, J. Meakin and F. Pastijn, Building bisimple idempotent-generated semi-
groups, J. Algebra 65 (1980), 60 -- 83.
[5] T. Hall, Identities for existence varieties of regular semigroups, Bull. Austral. Math.
Soc. 40 (1989), 59 -- 77.
[6] J. Kadourek and M. B. Szendrei, A new approach in the theory of orthodox semi-
groups, Semigroup Forum 40 (1990), 257 -- 296.
[7] J. Meakin and F. Pastijn, The structure of pseudo-semilattices, Algebra Universalis
13 (1981), 355 -- 372.
[8] K. S. S. Nambooripad, Pseudo-semilattices and biordered sets I, Simon Stevin 55
(1981), 103 -- 110.
[9] L. Oliveira, A solution to the word problem for free pseudosemilattices, Semigroup
Forum 68 (2004), 246 -- 267.
[10] F. Pastijn, Rectangular bands of inverse semigroups, Simon Stevin 56 (1982), 3 -- 95.
Departamento de Matem´atica Pura, Faculdade de Ciencias da Universidade
do Porto, R. Campo Alegre, 687, 4169-007 Porto, Portugal
E-mail address: [email protected]
|
1707.03610 | 1 | 1707 | 2017-07-12T09:15:08 | Infinite dimensional Jordan algebras and symmetric cones | [
"math.RA"
] | A celebrated result of Koecher and Vinberg asserts the one-one correspondence between the finite dimensional formally real Jordan algebras and Euclidean symmetric cones. We extend this result to the infinite dimensional setting. | math.RA | math |
Infinite dimensional Jordan algebras and symmetric cones
Cho-Ho Chu
Abstract. A celebrated result of Koecher and Vinberg asserts the one-one correspon-
dence between the finite dimensional formally real Jordan algebras and Euclidean sym-
metric cones. We extend this result to the infinite dimensional setting.
MSC. 17C65; 22E65; 46B40; 46H70
Keywords. Jordan algebra; Symmetric cone; Order-unit space; Banach Lie group;
Banach Lie algebra
1. Introduction
Finite dimensional formally real Jordan algebras were first introduced by Jordan, von
Neumann and Wigner for quantum mechanics formalism in [9], where these algebras were
completely classified. Since then, many far reaching connections to Lie algebras, geometry
and analysis have been found. One such connection to geometry is the seminal result of
Koecher [11] and Vinberg [18], which establishes the one-one correspondence between the
formally real Jordan algebras and a class of Reimmanian symmetric spaces, namely, the
symmetric cones. The latter plays a useful role in the study of automorphic functions
on bounded homogeneous domains in complex spaces and harmonic analysis (see, for
example, [8, 15, 18] and references therein).
In recent decades, infinite dimensional Jordan algebras and Jordan triple systems have
gradually become a significant part of the theory of bounded symmetric domains. While
much of the theory of finite dimensional bounded symmetric domains, which are Rie-
mannian symmetric spaces, can be extended to infinite dimension via Jordan theory, an
infinite dimensional generalisation of symmetric cones and the result of Koecher and Vin-
berg has not yet been accomplished. Our objective in this paper is to carry out this task
by introducing infinite dimensional symmetric cones and show in Theorem 3.1 that they
correspond exactly to a class of infinite dimensional real Jordan algebras with identity,
called unital JH-algebras.
In finite dimensions, the unital JH-algebras are exactly the
formally real Jordan algebras and our result is identical to that of Koecher and Vinberg.
Although our approach is a natural extension of the finite dimensional one, there are
some infinite dimensional pitfalls in Lie theory and other obstructions that different argu-
ments are required to circumvent, for instance, a closed subgroup of an infinite dimensional
Lie group need not be a Lie group in the relative topology [10] and an infinite dimensional
orthogonal group need not be compact and lacks an invariant measure. Unlike the finite
dimensional case, the order-unit structures play a prominent role in infinite dimension and
we make use of the weak topology as well as Kakutani's fixed-point theorem to achieve the
final result. This approach also provides some new perspectives for the finite dimensional
1
2
case. Our focus, however, is to present a proof of Theorem 3.1 as simply as possible, but
not discuss all its ramifications.
For completeness, we review briefly some relevant basics of Jordan algebras and refer to
[5, 17] for more details. In what follows, a Jordan algebra A is a real vector space, which
can be infinite dimensional, equipped with a bilinear product (a, b) ∈ A × A 7→ ab ∈ A
that is commutative and satisfies the Jordan identity
a(ba2) = (ab)a2
(a, b ∈ A).
A Jordan algebra is called unital if it contains an identity. There are two fundamental
linear operators on a Jordan algebra A, namely, the left multiplication La : x ∈ A 7→
ax ∈ A and the quadratic map Qa : x ∈ A 7→ {a, x, a} ∈ A, where the Jordan triple
product {·, ·, ·} is defined by
{a, b, c} = (ab)c + a(bc) − b(ac)
(a, b, c ∈ A).
We have the identities
Qa = 2L2
a − La2, Q2
a = Qa2
(a ∈ A).
The operator Lab + [La, Lb] : A → A, where [·, ·] denotes the Lie brackets, is often denoted
by a b and is called a box operator.
An element a in a Jordan algebra with identity e is called invertible if there exists an
element a−1 ∈ A (which is necessarily unique) such that aa−1 = e and (a2)a−1 = a. This
is equivalent to the invertibility of the quadratic operator Qa, in which case a−1 = Q−1
a (a).
If the left multiplication La is invertible, then a is invertible with inverse a−1 = L−1
a (e).
n = 0 implies a1 = · · · = an = 0
for any a1, . . . , an ∈ A. A finite dimensional formally real Jordan algebra A is necessarily
unital (cf. [5, Proposition 1.1.13]) and contains an abundance of idempotents, which are
elements p satisfying p2 = p (cf. [5, Theorem 1.1.14]). It is a real Hilbert space in the
trace norm kak2 = trace (a a) for a ∈ A.
A Jordan algebra A is called formally real if a2
1 + · · · + a2
Following [14], we call a real Jordan algebra H a JH-algebra if it is also a Hilbert space
in which the inner product, always denoted by h·, ·i, is associative, that is,
hab, ci = hb, aci
(a, b, c ∈ H).
A finite dimensional JH-algebra is called Euclidean in [8]. In fact, the finite dimensional
formally real Jordan algebras are exactly the Euclidean Jordan algebras with identity [5,
Lemma 2.3.7]. JH-algebras are examples of non-associative H*-algebras which have been
studied by many authors and references are detailed in [4, p. 222].
Throughout, all vector spaces are over the real scalar field unless stated otherwise.
2. Symmetric cones
Let V be a real vector space. By a cone C in V , we mean a nonempty subset of V
satisfying (i) C + C ⊂ C and (ii) αC ⊂ C for all α > 0. We note that a cone is necessarily
convex. A cone C is called proper if C ∩ −C = {0}. The partial ordering on V induced
by a proper cone C will be denoted by ≤C, or by ≤ if C is understood, so that x ≤ y
whenever y − x ∈ C. Conversely, if V is equipped with a partial ordering ≤, we let
V+ = {v ∈ V : 0 ≤ v} denote the corresponding proper cone.
Given a real topological vector space V and a set E ⊂ V , we will denote its closure and
interior by E and int E respectively. If C is a cone in V , then its closure C is also a cone.
3
If C is an open cone, then we have int C = C. For completeness, we include a proof of
this fact in the next lemma.
Lemma 2.1. Let C be an open convex set in a real topological vector space V . Then
int C = C.
Proof. There is nothing to prove if C is empty. Pick any q ∈ C. Let p ∈ int C. Then
p is an internal point of C, that is, every line through p meets C in a set containing an
interval around p (cf. [7, p.410, 413]).
In particular, for the line joining p and q, there
exists δ ∈ (0, 1) such that p ± δ(q − p) ∈ C. Since C is open and q is an interior point of
C, we have λ(p − δ(q − p)) + (1 − λ)q ∈ C for 0 < λ < 1 (cf. [7, p.413]). Hence
p =
1
1 + δ
(p − δ(q − p)) +
δ
1 + δ
q ∈ C.
(cid:3)
Given an open cone C such that C ∩ −C = {0}, we must have 0 /∈ C since every point
in C is an internal point of C.
Let V be a real vector space equipped with a proper cone V+ and induced partial
ordering ≤. An element e ∈ V is called an order-unit if
V = [λ>0
{x ∈ V : −λe ≤ x ≤ λe}.
We call V an order-unit space if it admits an order-unit. The real field R with the usual
partial ordering is an order-unit space with order-unit 1. We note that an order-unit space
V , with order-unit e, is determined by its cone V+ completely in that V = V+−V+. Indeed,
each x ∈ V with −λe ≤ x ≤ λe can be written as x = x1 − x2, where 2x1 = λe + x ∈ V+
and 2x2 = λe − x ∈ V+.
An order-unit e ∈ V is called Archimedean if for each v ∈ V , we have v ≤ 0 whenever
λv ≤ e for all λ ≥ 0. We call V an Archimedean order-unit space if it is equipped with
an Archimedean order-unit. An Archimedean order-unit e ∈ V induces a norm k · ke on
V , called an order-unit norm, and is defined by
kxke = inf{λ > 0 : −λe ≤ x ≤ λe}
(x ∈ V )
which satisfies −kxkee ≤ x ≤ kxkee. We denote by (V, e) an Archimedean order-unit
space V equipped with an Archimedean order unit e and the order-unit norm k · ke, where
the subscript e will be omitted if it is understood. We call (V, e) a complete Archimedean
order-unit space if the order-unit norm k · ke is complete.
We note that the cone V+ in (V, e) is closed in the order-unit norm and the Archimedean
order-unit e belongs to the interior int V+ of V+ [1, Theorem 2.2.5]. In fact, each interior
point u ∈ int V+ is an order-unit (cf. Lemma 2.5) and the corresponding order-unit norm
k · ku is equivalent to k · ke.
A linear map T : V → W between two order-unit spaces (V, e) and (W, u) is called
positive if T (V+) ⊂ W+. It is called a positive linear functional on V if (W, u) = (R, 1).
+, which consists of continuous
The dual V ∗ of (V, e) is partially ordered by the dual cone V ∗
positive linear functionals on V and is precisely the polar set
(2.1)
− V 0
+ := −{f ∈ V ∗ : f (x) ≤ 1, ∀x ∈ V+} = {f ∈ V ∗ : f (x) ≥ 0, ∀x ∈ V+}
(cf. [1, p.30]).
4
Lemma 2.2. Let T : V → W be a positive linear map between two Archimedean order-
unit spaces (V, e) and (W, u). Then T is continuous and kT k = kT (e)ku.
Proof. We need to show sup{kT xku : −e ≤ x ≤ e} < ∞, where {x ∈ V : −e ≤ x ≤ e} is
the closed unit ball of (V, e). Indeed, given −e ≤ x ≤ e, we have
−kT (e)kuu ≤ −T (e) ≤ T (x) ≤ T (e) ≤ kT (e)kuu
by positivity. This implies kT (x)ku ≤ kT (e)ku and hence T is continuous with kT k ≤
kT (e)ku. Since kT k = sup{kT xku : −e ≤ x ≤ e} ≥ kT (e)ku, we have kT k = kT (e)ku. (cid:3)
Proposition 2.3. Let T : (V, e) → (V, e) be a linear isomorphism such that T (V+) = V+.
Then T is an isometry if, and only if, T (e) = e.
Proof. Let T be an isometry. By Lemma 2.2, we have kT (e)k = kT k = 1 = kT −1k =
kT −1(e)k which implies −e ≤ T (e) ≤ e and −e ≤ T −1(e) ≤ e. Therefore T (e) = e by
positivity.
Conversely, Lemma 2.2 implies
kT k = kT (e)k = kek = 1 = kT −1(e)k = kT −1k.
Hence T is an isometry.
(cid:3)
We now introduce the concept of an infinite dimensional symmetric cone (cf. [5, p. 105])
which is a natural generalisation of the finite dimensional one.
Definition 2.4. Let V be a real Hilbert space with inner product h·, ·i. An open cone Ω
in V is called symmetric if it satisfies the following two conditions:
(i) (self-duality) Ω = {v ∈ V : hv, xi > 0, ∀x ∈ Ω\{0}};
(ii) (homogeneity) given x, y ∈ Ω, there is a continuous linear isomorphism g : V → V
such that g(x) = y.
Since a cone Ω in a Hilbert space V is convex, its weak and norm closures in V coincide
and is denoted by Ω. If Ω is open, we have Ω = int Ω from Lemma 2.1. By self-duality,
the closure Ω of a symmetric cone Ω is proper and in view of (2.1), also 'self-dual' in the
sense of Connes [6], namely,
Indeed, given v ∈ V with hv, xi ≥ 0 for all x ∈ Ω, we have, by picking some e ∈ Ω,
Ω = {v ∈ V : hv, xi ≥ 0, ∀x ∈ Ω}.
hv +
1
n
e, xi = hv, xi +
1
n
he, xi > 0
(n = 1, 2, . . .)
for x ∈ Ω\{0}. Hence v + 1
ne ∈ Ω for n = 1, 2, . . . and v ∈ Ω.
For finite dimensional Euclidean spaces, the preceding definition is the same as the usual
one for a symmetric cone [8]. In finite dimensions, Koecher and Vinberg's celebrated result
states that the interior of the cone {x2 : x ∈ A} in a formally real Jordan algebra A is
a symmetric cone and conversely, every symmetric cone is of this form. We will extend
this result to the infinite dimensional setting in the next section. We first discuss how a
symmetric cone relates to the underlying Hilbert space structure.
Lemma 2.5. Let V be a real vector space, equipped with a norm k · k and partially ordered
by the closure Ω of an open cone Ω. Then each point e ∈ Ω is an order-unit. If moreover,
e is Archimedean, then the order-unit norm k · ke satisfies k · ke ≤ ck · k for some c > 0.
Proof. Let e ∈ Ω. Since Ω is open, there exists r > 0 such that the open ball e − B(0, r)
is contained in Ω, where B(0, r) = {x ∈ V : kxk < r}. Let v ∈ V \{0}. Then we have
±(r/2kvk)v ∈ B(0, r) which implies e ∓ (r/2kvk)v ∈ e − B(0, r) ⊂ Ω, that is
5
−
2kvk
r
e ≤ v ≤
2kvk
r
e.
This proves that e is an order-unit. If e is Archimedean, it also implies that the order-unit
norm k · ke satisfies kvke ≤ (2/r)kvk for all v ∈ V .
(cid:3)
From now on, the inner product norm of a Hilbert space V will always be denoted by
k · k = h·, ·i1/2. The following result reveals that a real Hilbert space equipped with a
symmetric cone is a complete Archimedean order-unit space.
Lemma 2.6. Let V be a real Hilbert space partially ordered by the closure Ω of a symmetric
cone Ω and let e ∈ Ω. Then e is an Archimedean order-unit and the order-unit norm k · ke
is equivalent to the inner product norm k · k of V .
Proof. By lemma 2.5, e is an order-unit. To see that e is Archimedean, let λv ≤ e for
all λ ≥ 0. By self-duality of Ω, we have he − λv, xi ≥ 0 for all x ∈ Ω, which gives
he, xi ≥ λhv, xi for all λ ≥ 0. If x ∈ Ω\{0}, then he, xi > 0. It follows that hv, xi ≤ 0 for
all x ∈ Ω and hence −v ∈ Ω by self-duality.
For the second assertion, Lemma 2.5 already implies k · ke ≤ ck · k for some c > 0.
To complete the proof, we show that kvk2 ≤ he, eikvk2
e for all v ∈ V . Indeed, we have
kvkee±v ∈ Ω and by self-duality, hkvkee+v, kvkee−vi ≥ 0. Expanding the inner product
gives kvk2
(cid:3)
ehe, ei ≥ kvk2.
Let V be a real Hilbert space partially ordered by the closure of a symmetric cone Ω and
let e ∈ Ω. Then the previous lemma implies that a linear map T : V → V is continuous
with respect to the Hilbert space norm of V if, and only if, it is continuous with respect
to the order-unit norm k · ke. In the sequel, we will denote by kT k and kT ke the operator
norm of T : V → V with respect to the Hilbert space norm and order-unit norm of V ,
respectively.
We refer to [3, 17] for definitions and properties of infinite dimensional Banach Lie
groups and Lie algebras, which are analytic manifolds. Let L(V ) be the Banach algebra
of bounded linear operators on V , equipped with the Hilbert space operator norm k · k
and the usual involution ∗. Then L(V ) is a real Banach Lie algebra in the commutator
product [S, T ] = ST − T S for S, T ∈ L(V ). We denote by GL(V ) the open subgroup
of invertible elements in L(V ), which is a real Banach Lie group with Lie algebra L(V ).
Given S, T ∈ GL(V ), we have kS −1 − T −1k ≤ kS −1kkS − T kkT −1k. Hence the orthogonal
group
O(V ) = {T ∈ GL(V ) : kT k = kT −1k = 1}
of V , consisting of linear isometries of V and equipped with the norm topology, is a closed
subgroup of GL(V ) and a real Banach Lie group.
In contrast to the finite dimensional case, a closed subgroup H of an infinite dimen-
sional real Banach Lie group G need not be a Lie group in the relative topology [10].
Nevertheless, it can still be topologised (with a finer topology T ) to form a Banach Lie
group, by [17, 7.8]. In fact, if g is the Lie algebra of G, then the Lie algebra of H is given
6
by
h = {X ∈ g : exp tX ∈ H, ∀t ∈ R}
and the inclusion map (H, T ) ֒→ G is analytic.
Henceforth, let Ω be a symmetric cone in a real Hilbert space V . The open cone Ω is a
real Hilbert manifold modelled on V , where the tangent space TωΩ at ω ∈ Ω is identified
with V . The positive linear maps in GL(V ) with positive inverse, with respect to the
cone Ω, form a subgroup of GL(V ), which will be denoted by
G(Ω) = {g ∈ GL(V ) : g(Ω) = Ω}.
An element g ∈ GL(V ) belongs to G(Ω) if and only if g(Ω) = Ω. Hence G(Ω) is a closed
subgroup of GL(V ) and can be topologised in a finer topology to a real Banach Lie group
with Lie algebra
g(Ω) = {X ∈ L(V ) : exp tX ∈ G(Ω), ∀t ∈ R}.
For each ω ∈ Ω, the map g ∈ GL(V ) 7→ g(ω) ∈ V is analytic and its derivative at the
identity of GL(V ) is the linear map X ∈ L(V ) 7→ X(ω) ∈ V . Since the inclusion map
G(Ω) ֒→ GL(V ) is analytic, the orbital map g ∈ G(Ω) 7→ g(ω) ∈ Ω is analytic.
By a Lie subgroup of G(Ω), we mean a subgroup and submanifold of G(Ω). For instance,
the connected component G0 of the identity in G(Ω) is a Lie subgroup of G(Ω).
[3,
Chap. III,§1.3]. The subgroup K = G(Ω) ∩ O(V ) of G(Ω) is closed in the norm and any
finer topologies of G(Ω), and also a real Banach Lie group.
Given g ∈ G(Ω), we note that its adjoint g ∗ in L(V ) also belongs to G(Ω). Indeed, for
v ∈ Ω and x ∈ Ω\{0}, we have hg ∗(v), xi = hv, g(x)i > 0 and hence g ∗(v) ∈ Ω.
Lemma 2.7. Let V be a real Hilbert space partially ordered by the closure of a symmetric
cone Ω and let e ∈ Ω with αk · k ≤ k · ke ≤ βk · k for some β > α > 0. Then we have
β e ≤ g(e) ≤ β
α
Proof. Let g ∈ G(Ω) ∩ O(V ). Then we have
αe for all g ∈ G(Ω) ∩ O(V ).
α
β
kgk ≤ kgke ≤
β
α
kgk
where kgk = 1. By Lemma 2.2, kg(e)ke = kgke since g(Ω) ⊂ Ω, and the same for g−1. It
follows that
α
β
e ≤ g(e) ≤
β
α
e.
(cid:3)
Theorem 2.8. Let V be a real Hilbert space partially ordered by the closure of a symmetric
cone Ω. Then there exists ω ∈ Ω such that g(ω) = ω for all g ∈ G(Ω) ∩ O(V ).
Proof. Fix an Archimedean order-unit e ∈ Ω and let αk · k ≤ k · ke ≤ βk · k for some
β > α > 0. By Lemma 2.7, the orbit
is contained in the order interval
S = {g(e) : g ∈ G(Ω) ∩ O(V )}
[[(α/β)e, (β/α)e]] := {x ∈ V : (α/β)e ≤ x ≤ (β/α)e}
which is convex, bounded in the order-unit norm k · ke and hence bounded in the Hilbert
space norm of V . It is clearly closed in the order-unit norm, and hence in the Hilbert
7
space norm of V . It follows that the order interval [[(α/β)e, (β/α)e]] is weakly compact
in V . Let Q = co S be the closed convex hull of the orbit S. Then Q ⊂ [[(α/β)e, (β/α)e]]
and is weakly compact.
Since G(Ω) ∩ O(V ) is a group and each g ∈ O(V ) is continuous on V , we see that
g(Q) ⊂ Q for all g ∈ G(Ω) ∩ O(V ). Further, G(Ω) ∩ O(V ) is equicontinuous on Q in the
weak topology (as defined in [7, V.10.7]), that is, given any weak neighbourhood N of 0,
there exists a weak neighbourhood U of 0 such that for x, y ∈ Q with x − y ∈ U, we have
g(x − y) ∈ N for all g ∈ G(Ω) ∩ O(V ).
To prove equicontinuity, we first observe that for any weak neighbourhood of 0 of the
form N0 = {v ∈ V : hv, zi < ε} for some z ∈ V \{0}, one can find z1, z2 ∈ Ω\{0} such
that N0 contains the following weak neighbourhood of 0:
Indeed, write z = z1 − z2 with z1, z2 ∈ Ω\{0}, then we have
{v ∈ V : hv, zji < ε/2, j = 1, 2}.
hv, zi = hv, z1 − z2i ≤ hv, z1i + hv, z2i < ε.
Hence, given any weak neighbourhood N of 0, there are positive elements z1, . . . , zk in
Ω\{0} such that
N ⊃ {v ∈ V : hv, zji < ε, j = 1, . . . k}
for some ε > 0. We can find c > 1 such that zj ≤ ce for j = 1, . . . , k. Now, for each
g ∈ G(Ω) ∩ O(V ) and j = 1, . . . , k, we have
which gives
(2.2)
0 ≤ g(zj) ≤ cg(e) ≤
cβ
α
e
0 ≤ hx, g(zj)i ≤ hx, (cβ/α)ei
for all x ∈ Ω. Pick γ ∈ (0, 1) such that γ <
εα2
c(β 2 − α2)he, ei
. Then
U = {v ∈ V : hγ(v + (β 2 − α2)/(αβ) e), (cβ/α) ei < ε}.
is a weak neighbourhood of 0 in V . Since (α/β)e ≤ g(e) ≤ (β/α)e for all g ∈ G(Ω)∩O(V ),
by Lemma 2.7, we have
(α/β)he, ei ≤ hg(e), ei ≤ (β/α)he, ei
and hence
0 <
γ(β 2 − α2)
β 2
he, ei ≤
γ(β 2 − α2)
αβ
hg(e), ei ≤
cγ(β 2 − α2)
α2
he, ei < ε
for all g ∈ G(Ω) ∩ O(V ).
Let x, y ∈ Q ⊂ [[(α/β)e, (β/α)e]]. We have
−(cid:18) β
α
−
α
β(cid:19) e ≤ x − y ≤ (cid:18) β
α
−
α
β(cid:19) e =
β 2 − α2
αβ
e
8
and in particular, 0 ≤ (x − y) + (β 2 − α2)/(αβ) e ≤ (x − y)/γ + (β 2 − α2)/(αβ) e where
γ ∈ (0, 1). Hence x − y ∈ γU implies
0 ≤ hg(γ((x − y)/γ + (β 2 − α2)/(αβ) e), zji
= hγ{(x − y)/γ + (β 2 − α2)/(αβ) e}, g ∗(zj)i
≤ hγ{(x − y)/γ + (β 2 − α2)/(αβ) e}, (cβ/α)ei < ε
by (2.2), which gives
γ(β 2 − α2)
−
αβ
where (2.2) implies
hg(e), zji ≤ hg(x − y), zji ≤ ε −
γ(β 2 − α2)
αβ
hg(e), zji < ε
γ(β 2 − α2)
αβ
hg(e), zji ≤
γ(β 2 − α2)
αβ
he,
cβ
α
ei < ε
and hence
hg(x − y), zji < ε
for all g ∈ G(Ω) ∩ O(V ) and j = 1, . . . , k. This proves equicontinuity of G(Ω) ∩ O(V ).
Hence, by Kakutani's fixed-point theorem [7, V.10.8], the group G(Ω) ∩ O(V ) has a
(cid:3)
common fixed point, say, ω ∈ Q ⊂ [[(α/β)e, (β/α)e]] ⊂ Ω. The proof is complete.
We note that a bounded linear operator T on a complex Hilbert space H is hermitian
if, and only if, k exp itT k = 1 for all t ∈ R (cf. [2, p. 46]). If k exp itT k ≤ M for some
M > 0 and all t ∈ R, then exp itT has spectral radius
ρ(exp itT ) = lim
n→∞
k(exp itT )nk1/n = lim
n→∞
k exp intT k1/n ≤ lim
n→∞
M 1/n = 1
and if iT is hermitian as well, then we have k exp itT k = ρ(exp itT ) ≤ 1 for all t ∈ R,
which implies that T is hermitian and hence T = 0. Given X ∈ L(V ), by considering its
complexification XC : VC → VC on the complex Hilbert space VC and noting that exp tX
is an isometry on V if, and only if, exp tXC is a unitary operator on VC, we see that
k exp tXk = 1 for all t ∈ R if, and only if, X is skew-symmetric, that is, X ∗ = −X. On
the other hand, if X ∗ = X and if k exp tXk ≤ M for some M > 0 and for all t ∈ R, then
we must have X = 0.
The above observation implies that the Lie algebra of the Lie group K = G(Ω) ∩ O(V )
is given by
k = {X ∈ g(Ω) : exp tX ∈ G(Ω) ∩ O(V ), ∀t ∈ R}
= {X ∈ g(Ω) : X ∗ = −X} ⊂ L(V ).
For X ∈ g(Ω), we have exp tX ∈ G(Ω) and exp tX ∗ = (exp tX)∗ ∈ G(Ω) for all t ∈ R.
Hence X ∗ ∈ g(Ω). Let
p = {X ∈ g(Ω) : X ∗ = X}.
Then we have the direct sum decomposition g(Ω) = k ⊕ p with Lie brackets
[k, p] ⊂ p,
[p, p] ⊂ k.
This implies that the inclusion map ι : K ֒→ G(Ω) is an immersion, as its differential at
the identity, dι : k → g(Ω), has an image with a direct sum complement p in g(Ω) (cf. [5,
Definition 2.1.16]).
By Theorem 2.8, there is an order-unit ω ∈ Ω which is a fixed-point of the group
K = G(Ω) ∩ O(V ). Given X ∈ k, we have exp tX ∈ K and
ω = exp tX(ω) = ω + tX(ω) +
t2
2!
X(ω) + · · ·
for all t ∈ R
9
which implies X(ω) =
exp tX(ω) = 0. In fact, the converse also holds.
d
dt(cid:12)(cid:12)(cid:12)(cid:12)t=0
Lemma 2.9. Let ω ∈ Ω be a fixed point of K = G(Ω) ∩ O(V ) with Lie algebra k. Then
we have
k = {X ∈ g(Ω) : X(ω) = 0}.
Proof. Let X ∈ g(Ω) and X(ω) = 0. We have the decomposition X = Xk + Xp ∈ k ⊕ p.
Since Xk(ω) = 0 as noted above, we have Xp(ω) = 0 and hence exp tXp(ω) = ω for
all t ∈ R. This implies k exp tXpkω = kωkω = 1 and there exists M > 0 such that
k exp tXpk ≤ M for all t ∈ R since the order-unit norm k · kω is equivalent to the Hilbert
space norm k · k. On the other hand, Xp ∈ p implies X ∗
p = Xp. Therefore we must have
Xp = 0 and X = Xk ∈ k.
(cid:3)
Remark 2.10. The above lemma implies that the Lie algebra
kω = {X ∈ g(Ω) : exp tX(ω) = ω, ∀t ∈ R}
of the isotropy subgroup
Kω = {g ∈ G(Ω) : g(ω) = ω} ⊃ K
coincides with the Lie algebra k of K, where g(Ω) = k ⊕ p = kω ⊕ p. Hence, analogous to
the case of K, the inclusion map Kω ֒→ G(Ω) is an immersion.
The differential of the orbital map ρω : g ∈ G(Ω) 7→ g(ω) ∈ Ω at the identity of G(Ω)
is the evaluation map X ∈ g(Ω) 7→ X(ω) ∈ V . By homogeneity of Ω, we can identify
the Lie algebra g(Ω) = kω ⊕ p with the Lie algebra aut Ω of analytic vector fields on
Ω that generate one-parameter subgroups of G(Ω) (cf. [19, p. 110]). This implies that
the evaluation map X ∈ g(Ω) 7→ X(ω) ∈ TωΩ = V is surjective.
It follows that the
orbital map ρω is a submersion since kω is the kernel of the evaluation map, which is
complemented in g(Ω). In particular, ρω is an open map and the identity component G0
also acts transitively on Ω since Ω is connected and a disjoint union of open G0-orbits.
3. JH-algebras
In this final section, we extend to infinite dimension the celebrated result of Koecher
[11] and Vinberg [18] on the one-one correspondence between finite dimensional symmetric
cones and formally real Jordan algebras. This correspondence is furnished by the assertion
that given a finite dimensional formally real Jordan algebra A, which is a Hilbert space
in the trace norm, the interior of the cone {x2 : x ∈ A} is a symmetric cone and, every
symmetric cone in an Euclidean space is of this form. We show in Theorem 3.1 below
that symmetric cones of all dimensions are in one-one correspondence with the unital
JH-algebras.
Let H be a JH-algebra with identity 1 and let
H+ = {x2 : x ∈ H}.
10
It has been shown in [5, Lemma 2.3.17] that H+ is a closed cone and its interior
(3.1)
int H+ = {y ∈ H+ : hy, ai > 0, ∀a ∈ H+\{0}}
is a symmetric cone consisting of invertible elements in H+. For each x2 ∈ H+, the left
multiplication Lx2 : H → H is a positive self-adjoint operator on the Hilbert space H [5,
p. 108]. Further, int H+ is a Hilbert manifold and a Riemannian symmetric space, which
can be infinite dimensional, with Riemannian metric
gω(u, v) = hQω−1(u), vi
(ω ∈ int H+, u, v ∈ H)
where the quadratic map Qx2 = 2L2
In what follows, we retain the notation in the previous section.
x2 −Lx4 = Q2
x is a positive operator on H, for x2 ∈ H+.
Theorem 3.1. Let Ω be an open cone in a real Hilbert space. Then Ω is a symmetric
cone if, and only if, it is of the form
(3.2)
Ω = int {a2 : a ∈ H}
for a unique unital JH-algebra H.
Proof. Given a unital JH-algebra H, the cone
Ω = int {a2 : a ∈ H}
is a symmetric cone by [5, Lemma 2.3.17], as noted above.
Conversely, let Ω be a symmetric cone in a real Hilbert space V . We show that it is of
the form in (3.2) for a unique JH-algebra H. In fact, we show that H is the Hilbert space
V itself, equipped with a suitable Jordan product. Our arguments are a natural extension,
albeit with some infinite dimensional adaptation, of Satake's proof in [16, Theorem I.8.5]
for the finite dimensional case (see also [8]).
We first note that the JH-algebra H in (3.2) must be unique since (H, Ω) is an order-unit
space and hence H = Ω − Ω.
As before, let K = G(Ω) ∩ O(V ). By Theorem 2.8, K is contained in the isotropy
subgroup Kω of G(Ω) at some ω ∈ Ω. The Lie algebra g(Ω) of G(Ω) has a direct sum
decomposition
g(Ω) = kω ⊕ p
where kω is the Lie algebra of Kω.
As noted at the end of the previous section, homogeneity of Ω implies that the evaluation
map
X ∈ g(Ω) 7→ X(ω) ∈ V
is surjective. It follows that the map Φ : X ∈ p 7→ X(ω) ∈ V is a linear isomorphism
since it has kernel
{X ∈ g : X(ω) = 0} = kω
by Lemma 2.9 and Remark 2.10, where kω ∩p = {0}. Denote the inverse of Φ by L : V → p
so that
On V , we defined a product
L(x)ω = x
(x ∈ V ).
xy := L(x)y = L(x)L(y)ω
(x, y ∈ V ).
11
We show that V is a Jordan algebra with this product and moreover, it is a JH-algebra
with identity ω such that
Ω = int {x2 : x ∈ V }
which would complete the proof.
To see that V is a Jordan algebra in the above product, let x, y ∈ V . Then we have
xy = L(x)y = L(x)L(y)ω and xy − yx = [L(x), L(y)]ω = 0
where L(x), L(y) ∈ p implies [L(x), L(y)] ∈ kω. To prove the Jordan identity, we observe
x2(yx) = x2(L(y)L(x)ω) = L(x2)L(L(y)L(x)ω)ω = L(x2)L(y)L(x)ω
and
x(x2y) = L(x)L(L(x2)L(y)ω)ω = L(L(x2)L(y)ω)L(x)ω = L(x2)L(y)L(x)ω
which equals x2(yx). This proves that V is a Jordan algebra.
Evidently ω is the identity in V as xω = L(x)ω = x for all x ∈ V . To see that the
inner product h·, ·i is associative, we first note that L(x)∗ = L(x) since L(x) ∈ p. Given
x, y, z ∈ V , we have
hxy, zi = hL(x)L(y)ω, zi = hL(y)ω, L(x)zi = hy, xzi
which shows that V is a JH-algebra.
Finally, we show that Ω is identical with the symmetric cone
C := int {x2 : x ∈ V }.
Let x2 ∈ C. We have noted previously that the left multiplication La2 is a positive self-
adjoint operator on the Hilbert space V , for each a ∈ V . Since x2 is an interior point of
the cone, it has been shown in [5, p. 109] that x2 = a2 + βω for some a ∈ V and β > 0.
This implies that Lx2 = La2 + βLω, where Lω is the identity operator on V . Hence Lx2 is
a positive invertible operator in L(V ) and by spectral theory, there exists T ∈ L(V ) such
that Lx2 = exp T . It follows that
x2 = Lx2(ω) = exp T (ω) = exp X(ω)
for some X ∈ p. Since exp X ∈ G(Ω), we have exp X(ω) ∈ Ω, that is, x2 ∈ Ω. We have
shown that C ⊂ Ω and they must be equal by self-duality.
(cid:3)
In finite dimensions, the above theorem is exactly the aforementioned result of Koecher
and Vinberg since the finite dimensional unital JH-algebras coincide with the formally
real Jordan algebras. The following corollary of the theorem is immediate.
Corollary 3.2. A symmetric cone in a Hilbert space V carries the structure of a Rie-
mannian symmetric space.
We conclude with two examples of infinite dimensional unital JH-algebras.
Example 3.3. Given any real Hilbert space H with inner product h·, ·i, the Hilbert space
direct sum H ⊕ R, called a spin factor, is a JH-algebra with identity 0 ⊕ 1 in the following
Jordan product
(a ⊕ α)(b ⊕ β) := (βa + αb) ⊕ (ha, bi + αβ).
In particular, for a, b ∈ H, we have (a ⊕ 0)(b ⊕ 0) = ha, bi(0 ⊕ 1), which is a scalar multiple
of the identity.
12
Given a spin factor V , we denote by V0 the orthogonal complement of the identity in
V . In the above example, V0 is just the Hilbert space H. For each v ∈ V0, we observe
that v2 is a scalar multiple of the identity.
Example 3.4. Let Z = V ⊕ W be the Hilbert space direct sum of two spin factors V and
W , equipped with coordinatewise Jordan product. Then it can be verified readily that
Z is a JH-algebra with identity (eV , eW ), where eV is the identity of V and eW that of
W . Further, Z is not a spin factor. Indeed, the orthogonal complement Z0 of the identity
contains V0 ⊕ W0 and for z = (v, w) ∈ V0 ⊕ W0, we have z2 = (v2, w2) which need not be
a scalar multiple of (eV , eW ).
References
[1] L. Asimow and A.J. Ellis, Convexity theory and its applications in functional analysis, Academic
Press, London, 1980.
[2] F.F. Bonsall and J. Duncan, Numerical ranges of operators on normed spaces and of elements of
normed algebras, (Cambridge: Cambridge Univ. Press, 1971).
[3] N. Bourbaki, Lie groups and Lie algebras, Springer-Verlag, Berlin, 1989.
[4] M. Cabrera Garcia and A. Rodr´ıguez Palacios, Non-associative normed algebras, Vol. 1, Encyclop.
Math. Appl. 154, Cambridge Univ. Press, Cambridge, 2014.
[5] C-H. Chu, Jordan structures in geometry and analysis, Cambridge Tracts in Math. 190, Cambridge
Univ. Press, Cambridge, 2012.
[6] A. Connes, Caract´erisation des espaces vectoriels ordonn´es sous jacents aux alg`ebres de von Neu-
mann, Ann. Inst. Fourier 24 (1974) 121-155.
[7] N. Dunford and J.T. Schwartz, Linear operators, Wiley Classics Library Edition (1988) New York.
[8] J. Faraut and A. Koranyi, Analysis on symmetric cones, Oxford Univ. Press, Oxford, 1994.
[9] P. Jordan, J. von Neumann and E. Wigner, On an algebraic generalisation of the quantum mechanical
formalism, Ann. of Math. 36 (1934) 29 -- 64.
[10] L. A. Harris and W. Kaup, Linear algebraic groups in infinite dimensions, Illinois J. Math. 21 (1977)
666-674.
[11] W. Kaup, Algebraic characterization of symmetric complex Banach manifolds, Math. Ann. 228
(1977) 39-64.
[12] W. Kaup and H. Upmeier, Jordan algebras and symmetric Siegel domains in Banach spaces, Math.
Z. 157 (1977) 179-200.
[13] M. Koecher, Jordan algebras and their applications, University of Minnesota, Minneapolis, 1962
(lecture Notes in Math. 1710 Springer-Verlag, Heidelberg, 1999).
[14] T. Nomura, Grassmann manifold of a JH-algebra, Ann. Global Analysis and Geometry, 12 (1994)
237-260.
[15] I.I. Pjatetskij-Shapiro, Automorphic functions and the geometry of classical domains, Gordon-
Breach, New York, 1969.
[16] I. Satake, Algebraic strutures of symmetric domains, Princeton Univ. Press, Princeton, 1980.
[17] H. Upmeier, Symmetric Banach manifolds and Jordan C*-algebras (North Holland Math. Studies
104) North Holland, Amsterdam 1985.
[18] E. B.Vinberg, The theory of convex homogeneous cones, Trudy Moskov. Mat. Obsc. 12 (1963) 303-
358.
[19] D.R. Wilkin, Infinite-dimensional homogeneous manifolds, Proc. Royal Irish Acad. 94A (1994) 105-
118.
School of Mathematical Sciences, Queen Mary, University of London, London E1 4NS,
UK
E-mail address:
[email protected]
|
1209.3987 | 5 | 1209 | 2013-09-17T13:09:53 | Universal geometric cluster algebras | [
"math.RA",
"math.CO",
"math.RT"
] | We consider, for each exchange matrix B, a category of geometric cluster algebras over B and coefficient specializations between the cluster algebras. The category also depends on an underlying ring R, usually the integers, rationals, or reals. We broaden the definition of geometric cluster algebras slightly over the usual definition and adjust the definition of coefficient specializations accordingly. If the broader category admits a universal object, the universal object is called the cluster algebra over B with universal geometric coefficients, or the universal geometric cluster algebra over B. Constructing universal coefficients is equivalent to finding an R-basis for B (a "mutation-linear" analog of the usual linear-algebraic notion of a basis). Polyhedral geometry plays a key role, through the mutation fan F_B, which we suspect to be an important object beyond its role in constructing universal geometric coefficients. We make the connection between F_B and g-vectors. We construct universal geometric coefficients in rank 2 and in finite type and discuss the construction in affine type. | math.RA | math |
UNIVERSAL GEOMETRIC CLUSTER ALGEBRAS
NATHAN READING
Abstract. We consider, for each exchange matrix B, a category of geomet-
ric cluster algebras over B and coefficient specializations between the cluster
algebras. The category also depends on an underlying ring R, usually Z, Q,
or R. We broaden the definition of geometric cluster algebras slightly over the
usual definition and adjust the definition of coefficient specializations accord-
ingly. If the broader category admits a universal object, the universal object is
called the cluster algebra over B with universal geometric coefficients, or the
universal geometric cluster algebra over B. Constructing universal geomet-
ric coefficients is equivalent to finding an R-basis for B (a "mutation-linear"
analog of the usual linear-algebraic notion of a basis). Polyhedral geometry
plays a key role, through the mutation fan FB, which we suspect to be an im-
portant object beyond its role in constructing universal geometric coefficients.
We make the connection between FB and g-vectors. We construct universal
geometric coefficients in rank 2 and in finite type and discuss the construction
in affine type.
Contents
Introduction
1.
2. Cluster algebras of geometric type
3. Universal geometric coefficients
4. Bases for B
5. B-cones and the mutation fan
6. Positive bases and cone bases
7. Properties of the mutation fan
8. g-Vectors and the mutation fan
9. The rank-2 case
10. Rays in the Tits cone
Acknowledgments
References
1
3
6
9
14
20
25
27
33
40
47
48
1. Introduction
In this paper, we consider the problem of constructing universal geometric cluster
algebras: cluster algebras that are universal (in the sense of coefficient specializa-
tion) among cluster algebras of geometric type with a fixed exchange matrix B. In
2010 Mathematics Subject Classification. Primary 13F60, 52B99; Secondary 05E15, 20F55.
This material is based upon work partially supported by the National Security Agency under
Grant Number H98230-09-1-0056, by the Simons Foundation under Grant Number 209288 and
by the National Science Foundation under Grant Number DMS-1101568.
1
2
NATHAN READING
order to accommodate universal geometric cluster algebras beyond the case of finite
type, we broaden the definition of geometric type by allowing extended exchange
matrices to have infinitely many coefficient rows. We have also chosen to broaden
the definition in another direction, by allowing the coefficient rows to have entries
in some underlying ring (usually Z, Q, or R). We then narrow the definition of
coefficient specialization to rule out pathological coefficient specializations.
There are at least two good reasons to allow the underlying ring R to be some-
thing other than Z. First, when R is a field, a universal geometric cluster algebra
over R exists for any B (Corollary 4.7). Second, the polyhedral geometry of muta-
tion fans, defined below, strongly suggests the use of underlying rings larger than Q.
(See in particular Section 9.)
On the other hand, there are good reasons to focus on the case where R = Z.
In this case, the broadening of the definition of geometric type (allowing infinitely
many coefficient rows) is mild: Essentially, we pass from polynomial coefficients to
formal power series coefficients. The only modification of the definition of coefficient
specializations is to require that they respect the formal power series topology. The
case R = Z is the most natural in the context of the usual definition of geometric
type. We have no proof, for general B, that universal geometric cluster algebras
exist over Z, but we also have no counterexamples.
In this paper and [21, 22]
we show that universal geometric cluster algebras over Z exist for many exchange
matrices.
Given an underlying ring R and an exchange matrix B, the construction of a
universal geometric cluster algebra for B over R is equivalent to finding an R-basis
for B (Theorem 4.4). The notion of an R-basis for B is a "mutation-linear" analog
of the usual linear-algebraic notion of a basis. The construction of a basis is in
turn closely related to the mutation fan FB, a (usually infinite) fan of convex cones
that are essentially the domains of linearity of the action of matrix mutation on
coefficients. In many cases, a basis is obtained by taking one nonzero vector in each
ray of FB, or more precisely in each ray of the R-part of FB. (See Definition 6.9.)
Indeed, we show (Corollary 6.12) that a basis with the nicest possible properties
only arises in this way.
The fan FB appears to be a fundamental object. For example, conditioned
on a well-known conjecture of Fomin and Zelevinsky (sign-coherence of principal
coefficients), we show (Theorem 8.7) that there is a subfan of FB containing all cones
spanned by g-vectors of clusters of cluster variables for the transposed exchange
matrix BT . In particular, for B of finite type, the fan FB coincides with the g-
vector fan for BT , and as a result, universal coefficients for B are obtained by
making an extended exchange matrix whose coefficient rows are exactly the g-
vectors of cluster variables for BT (Theorem 10.12). This is a new interpretation of
the universal coefficients in finite type, first constructed in [12, Theorem 12.4]. We
conjecture that a similar statement holds for B of affine type as well, except that
one must adjoin one additional coefficient row beyond those given by the g-vectors
for BT (Conjecture 10.15). We intend to prove this conjecture in general in a future
paper using results of [26, 27]. We will also describe the additional coefficient row
in terms of the action of the Coxeter element. Here, we work out the rank-2 case
and one rank-3 case of the conjecture. The cases of the conjecture where B arises
from a marked surface are proved in [21]. (See [21, Remark 7.15]). Further, but
more speculatively, we suspect that FB should play a role in the problem of finding
UNIVERSAL GEOMETRIC CLUSTER ALGEBRAS
3
nice additive bases for cluster algebras associated to B. In finite type, the cluster
monomials are an additive basis. Each cluster variable indexes a ray in the g-
vector fan (i.e. in FB), and cluster monomials are obtained as combinations of
cluster variables whose rays are in the same cone of FB. Beyond finite type, the
rays of FB may in some cases play a similar role to provide basic building blocks
for constructing additive bases.
When B arises from a marked surface in the sense of [7, 8], the rational part
of the mutation fan FB can in most cases be constructed by means of a slight
modification of the notion of laminations. This leads to the construction, in [21],
of universal geometric coefficients for a family of surfaces including but not lim-
ited to the surfaces of finite or affine type. (The family happens to coincide with
the surfaces of polynomial growth identified in [7, Proposition 11.2].) In [22], we
explicitly construct universal geometric coefficients for an additional surface: the
once-punctured torus. A comparison of [6] and [22] suggests a connection between
universal geometric coefficients and the cluster X -varieties of Fock and Goncharov
[5, 6], but the precise nature of this connection has not been worked out.
The definitions and results given here are inspired by the juxtaposition of several
results on cluster algebras of finite type from [12, 19, 23, 30]. Details on this
connection are given in Remark 10.13.
Throughout this paper, [n] stands for {1, 2, . . . , n}. The notation [a]+ stands for
max(a, 0), and sgn(a) is 0 if a = 0 or a/a if a (cid:54)= 0. Given a vector a = (a1, . . . , an)
in Rn, the notation min(a, 0) stands for (min(a1, 0), . . . , min(an, 0)). Similarly,
sgn(a) denotes the vector (sgn(a1), . . . , sgn(an)).
Notation such as (ui : i ∈ I) stands for a list of objects indexed by a set I of
arbitrary cardinality, generalizing the notation (u1, . . . , un) for an n-tuple. We use
the Axiom of Choice throughout the paper without comment.
2. Cluster algebras of geometric type
In this section we define cluster algebras of geometric type. We define geometric
type more broadly than the definition given in [12, Section 2]. The exposition here
is deliberately patterned after [12, Section 2] to allow easy comparison.
Definition 2.1 (The underlying ring). All of the definitions in this section depend
on a choice of an underlying ring R, which is required be either the integers Z
or some field containing the rationals Q as a subfield and contained as a subfield of
the reals R. At present, we see little need for the full range of possibilities for R.
Rather, allowing R to vary lets us deal with a few interesting cases simultaneously.
Namely, the case R = Z allows us to see the usual definitions as a special case of
the definitions presented here, while examples in Section 9 suggest taking R to be
the field of algebraic real numbers, or some finite-degree extension of Q. The case
R = R also seems quite natural given the connection that arises with the discrete
(real) geometry of the mutation fan. (See Definition 5.12.)
Definition 2.2 (Tropical semifield over R). Let I be an indexing set. Let (ui : i ∈
I) be a collection of formal symbols called tropical variables. Define TropR(ui :
i ∈ I) to be the abelian group whose elements are formal products of the form
(cid:81)
i∈I uai
i with each ai ∈ R and multiplication given by
uai+bi
.
i
ubi
i =
uai
i
(cid:89)
i∈I
(cid:89)
i∈I
·(cid:89)
i∈I
indexing set I, of copies of R. The multiplicative identity (cid:81)
Thus, as a group, TropR(ui : i ∈ I) is isomorphic to the direct product, over the
i is abbreviated
i∈I u0
NATHAN READING
4
as 1.
We define an auxiliary addition ⊕ in TropR(ui : i ∈ I) by
(cid:89)
i∈I
i ⊕(cid:89)
uai
i∈I
(cid:89)
i∈I
ubi
i =
umin(ai,bi)
i
.
We endow R with the discrete topology and endow TropR(ui
: i ∈ I) ,⊕ , · ) is a semifield : an abelian multiplicative
The triple (TropR(ui
group equipped with a commutative, associative addition ⊕, with multiplication
distributing over ⊕. Specifically, this triple is called a tropical semifield over R.
: i ∈ I) with
the product topology as a product of copies of the discrete set R. The product
topology is sometimes called the Tychonoff topology or the topology of point-
wise convergence. Details on the product topology are given later in Section 3.
The reason we impose this topology is seen in Proposition 3.7. When I is finite,
the product topology on TropR(ui : i ∈ I) is the discrete topology, and when I
is countable, the product topology on TropR(ui : i ∈ I) is the usual topology of
formal power series (written multiplicatively).
As a product of copies of R, the semifield TropR(ui : i ∈ I) is a module over R.
: i ∈ I) is written multiplicatively, the
(cid:81)
Since the group operation in TropR(ui
scalar multiplication corresponds to exponentiation. That is, c ∈ R acts by sending
to(cid:81)
i∈I ucai
i
.
i∈I uai
i
One recovers [12, Definition 2.2] by requiring I to be finite and taking the un-
derlying ring R to be Z. One may then ignore the topological considerations.
The role of the tropical semifield P in this paper is to provide a coefficient ring
ZP for cluster algebras. The notation ZP denotes the group ring, over Z, of the
multiplicative group P, ignoring the addition ⊕. The larger group ring QP also
makes a brief appearance in Definition 2.3.
Definition 2.3 (Labeled geometric seed of rank n). Let P = TropR(ui : i ∈ I) be a
tropical semifield. Let K be a field isomorphic to the field of rational functions in
n independent variables with coefficients in QP. A (labeled) geometric seed of
rank n is a pair (x, B) described as follows:
is referred to as a matrix over R.
• B is a function from ([n] ∪ I) × [n] to R. For convenience, the function B
• The first rows of B indexed by [n] have integer entries and form a skew-
symmetrizable matrix B. (That is, there exist positive integers d1, . . . , dn
such that dibij = −djbji for all i, j ∈ [n]. If the integers di can be taken to
all be 1, then B is skew-symmetric.)
• x = (x1, . . . , xn) is an n-tuple of algebraically independent elements of K
which generate K (i.e. K = QP(x1, . . . , xn)).
The n-tuple x is the cluster in the seed and the entries of x are called the cluster
variables. The square matrix B is the exchange matrix or the principal part
of B and the full matrix B is the extended exchange matrix . The rows of
B indexed by I are called coefficient rows; the coefficient row indexed by i is
written bi = (bi1, . . . , bin). The semifield P is called the coefficient semifield
and is determined up to isomorphism by the number of coefficient rows of B (i.e.
UNIVERSAL GEOMETRIC CLUSTER ALGEBRAS
5
by the cardinality of I). The elements yj = (cid:81)
i∈I ubij
i
of P are the coefficients
associated to the seed.
Definition 2.4 (Mutation of geometric seeds). Fix I, P = TropR(ui : i ∈ I), and
K as in Definitions 2.2 and 2.3. For each k ∈ [n] we define an involution µk on the
set of labeled geometric seeds of rank n. Let (x, B) be a labeled geometric seed.
We define a new seed µk(x, B) = (x(cid:48), B(cid:48)) as follows. The new cluster x(cid:48) is
1, . . . , x(cid:48)
j = xj whenever j (cid:54)= k and
n) with x(cid:48)
(x(cid:48)
(cid:81)n
+(cid:81)n
i=1 x[−bik]+
i
,
i=1 x[bik]+
i
x(cid:48)
k =
yk
(2.1)
where, as above, yk is the coefficient(cid:81)
(cid:81)
i . Thus x(cid:48)
k ∈ K.
i∈I u0
words, x(cid:48)
The new extended exchange matrix B(cid:48) has entries
xk(yk ⊕ 1)
i∈I ubik
(cid:26)−bij
(2.2)
b(cid:48)
ij =
bij + sgn(bkj) [bikbkj]+ otherwise.
if i = k or j = k;
and 1 is the multiplicative identity
k is a rational function in x with coefficients in ZP, or in other
i
constants di as the top part of B. The map µk is an involution.
The top part of µk( B) is skew-symmetrizable with the same skew-symmetrizing
The notation µk also denotes the mutation µk( B) = B(cid:48) of extended exchange
matrices, which does not depend on the cluster x. Given a finite sequence k =
kq, . . . , k1 of indices in [n], the notation µk stands for µkq ◦ µkq−1 ◦ ··· ◦ µk1. We
have indexed the sequence k so that the first entry in the sequence is on the right.
Definition 2.5 (Mutation equivalence of matrices). Two exchange matrices (or ex-
tended exchange matrices) are called mutation equivalent if there is a sequence k
such that one matrix is obtained from the other by applying µk. The set of all ma-
trices mutation equivalent to B is the mutation class of B.
Definition 2.6 (Regular n-ary tree). Let Tn denote the n-regular tree with edges
labeled by the integers 1 through n such that each vertex is incident to exactly one
-- -- t(cid:48) indicates that vertices t and t(cid:48) are
k
edge with each label. The notation t
connected by an edge labeled k.
Definition 2.7 (Cluster pattern and Y -pattern of geometric type). Fix a vertex
t0 in Tn. Given a seed (x, B), the assignment t0 (cid:55)→ (x, B) extends to a unique
map t (cid:55)→ (xt, Bt) from the vertices of Tn by requiring that (xt(cid:48), Bt(cid:48)) = µk(xt, Bt)
-- -- t(cid:48). This map is called a cluster pattern (of geometric type).
whenever t
The map t (cid:55)→ Bt is called a Y -pattern (of geometric type). The cluster variables
ij). For each j ∈ [n],
and matrix entries of (xt, Bt) are written (x1;t, . . . , xn;t) and (bt
k
the coefficient(cid:81)
bt
ij
i∈I u
i
is represented by the symbol yj;t.
Definition 2.8 (Cluster algebra of geometric type). The cluster algebra A asso-
ciated to a cluster pattern t (cid:55)→ (xt, Bt) is the ZP-subalgebra of K generated by all
cluster variables occurring in the cluster pattern. That is, setting
X = {xi;t : t ∈ Tn, i ∈ [n]},
6
NATHAN READING
we define A = ZP[X ]. Since the seed (x, B) = (xt0, Bt0) uniquely determines the
cluster pattern, we write A = AR(x, B). Up to isomorphism, the cluster algebra is
determined entirely by B, so we can safely write AR( B) for AR(x, B).
Remark 2.9. Definitions 2.3, 2.4, 2.7, and 2.8 are comparable to [12, Definition 2.3],
[12, Definition 2.4], [12, Definition 2.9], and [12, Definition 2.11] restricted to the
special case described in [12, Definition 2.12]. Essentially, the only difference is the
broader definition of the notion of a tropical semifield, which forces us to allow B
to have infinitely many rows, and to allow non-integer entries in the coefficient rows
of B. Definition 2.6 is identical to [12, Definition 2.8].
3. Universal geometric coefficients
In this section, we define and discuss a notion of coefficient specialization of clus-
ter algebras of geometric type and define cluster algebras with universal geometric
coefficients. These definitions are comparable, but not identical, to the correspond-
ing definitions in [12, Section 12], which apply to arbitrary cluster algebras.
Definition 3.1 (Coefficient specialization). Let (x, B) and (x(cid:48), B(cid:48)) be seeds of
rank n, and let P and P(cid:48) be the corresponding tropical semifields over the same un-
derlying ring R. A ring homomorphism ϕ : AR(x, B) → AR(x(cid:48), B(cid:48)) is a coefficient
specialization if
(i) the exchange matrices B and B(cid:48) coincide;
(ii) ϕ(xj) = x(cid:48)
(iii) the restriction of ϕ to P is a continuous R-linear map to P(cid:48) with ϕ(yj;t) = y(cid:48)
j for all j ∈ [n];
j;t
and ϕ(yj;t ⊕ 1) = y(cid:48)
j;t ⊕ 1 for all j ∈ [n] and t ∈ Tn.
Continuity in (iii) refers to the product topology described in Definition 2.2.
Definition 3.2 (Cluster algebra with universal geometric coefficients). A cluster
algebra A = AR( B) of geometric type with underlying ring R is universal over R
if for every cluster algebra A(cid:48) = AR( B(cid:48)) of geometric type with underlying ring R
sharing the same initial exchange matrix B, there exists a unique coefficient spe-
cialization from A to A(cid:48). In this case, we also say B is universal over R and call
the coefficient rows of B universal geometric coefficients for B over R.
The local conditions of Definition 3.1 imply some global conditions, as recorded
in the following proposition, whose proof follows immediately from (2.1) and (2.2).
t coincide for any t.
j;t for all j ∈ [n] and t ∈ Tn.
Proposition 3.3. Continuing the notation of Definition 3.1, if ϕ is a coefficient
specialization, then
(i(cid:48)) the exchange matrices Bt and B(cid:48)
(ii(cid:48)) ϕ(xj;t) = x(cid:48)
Remark 3.4. A converse to Proposition 3.3 is true for certain B. Specifically,
suppose the exchange matrix B has no column composed entirely of zeros. If the
restriction of ϕ to P is a continuous R-linear map and conditions (i(cid:48)) and (ii(cid:48)) of
Proposition 3.3 both hold, then ϕ is a coefficient specialization. This fact is argued
in the proof of [12, Proposition 12.2], with no restrictions on B. (Naturally, the
argument there makes no reference to continuity and R-linearity, which are not
part of [12, Definition 12.1].) The argument given there is valid in the context of
Definition 3.1, over any R, as long as B has no zero column.
UNIVERSAL GEOMETRIC CLUSTER ALGEBRAS
7
j;t from p−
The reason [12, Proposition 12.2] can fail when B has a zero column is that it
may be impossible to distinguish p+
j;t in [12, (12.2)]. Rather than defining
j;t and p−
p+
j;t here, we consider a simple case in the notation of this paper. Take
R = Z and let B have a 1 × 1 exchange matrix [0] and a single coefficient row 1,
so that in particular P = TropZ(u1). The cluster pattern associated with (x1, B)
has two seeds: One with cluster variable x1 and coefficient u1, and the other with
cluster variable x−1
1 . On the other hand, let B(cid:48) also have
exchange matrix [0] but let its single coefficient row be −1. Let P(cid:48) = TropZ(u(cid:48)
1).
The cluster pattern associated with (x(cid:48)
1 and
1)−1 and a seed with cluster variable (x(cid:48)
coefficient (u(cid:48)
1) and coefficient
u(cid:48)
1. The map ϕ sending x1 to x(cid:48)
1 extends to a ring homomorphism
satisfying condition (ii(cid:48)). However, condition (iii) fails.
1, B(cid:48)) has a seed with cluster variable x(cid:48)
1 (u1 +1) and coefficient u−1
1 and u1 to u(cid:48)
1)−1(1 + u(cid:48)
The characterization in [12, Theorem 12.4] of universal coefficients in finite type
is valid except when B has a zero column, or in other words when A(B) has a
irreducible component of type A1. It can be fixed by separating A1 as an exceptional
case or by taking [12, Proposition 12.2] as the definition of coefficient specialization.
Remark 3.5. Here we discuss the requirement in Definition 3.1 that the restriction
of ϕ be R-linear. This is strictly stronger than the requirement that the restriction
be a group homomorphism, which is all that is required in [12, Definition 12.1].
The additional requirement in Definition 3.1 is that the restriction of ϕ commutes
with scaling (i.e. exponentiation) by elements of R. This requirement is forced by
the fact that Definitions 2.1 and 2.2 allow the underlying ring R to be larger than
Q. When R is Z or Q, the requirement of commutation with scaling is implied by
the requirement of a group homomorphism.
In general, however, consider again the example of a cluster algebra A of rank 1,
given by the 2 × 1 matrix with rows 0 and 1 and the initial cluster x with a single
entry x1. The coefficient semifield is TropR(u1). The cluster variables are x1 and
x−1
1 (u1 +1). Suppose the underlying ring R is a field. Then R is a vector space over
Q, and homomorphisms of additive groups from R to itself correspond to Q-linear
If R strictly contains Q, then there are infinitely many
maps of vector spaces.
such maps fixing 1. Thus, if we didn't require commutation with scaling, there
would be infinitely many coefficient specializations from A to itself fixing 1, x1 and
x−1
1 (u1 + 1). The requirement of linearity ensures that the identity map is the only
coefficient specialization fixing 1, x1 and x−1
Remark 3.6. Here we discuss the requirement in Definition 3.1 that the restriction
of ϕ be continuous. This requirement is forced, once we allow the set of tropical
variables to be infinite. Take I to be infinite, let P = TropR(ui : i ∈ I) and consider
a cluster algebra A of rank 1 given by the exchange matrix [0] and each coefficient
row equal to 1. Consider the set of R-linear maps ϕ : P → P with ϕ(ui) = ui for all
i ∈ I. Linearity plus the requirement that ui (cid:55)→ ui only determine ϕ on elements
i with only finitely many nonzero ai).
This leaves infinitely many R-linear maps from P to itself that send every tropical
variable ui to itself. Dropping the requirement of continuity in Definition 3.1, each
of these maps would be a coordinate specialization.
of P with finite support (elements (cid:81)
1 (u1 + 1).
i∈I uai
The following proposition shows that the requirements of R-linearity and conti-
nuity resolve the issues illustrated in Remarks 3.5 and 3.6.
8
NATHAN READING
Proposition 3.7. Let TropR(ui : i ∈ I) and TropR(vk : k ∈ K) be tropical semi-
fields over R and fix a family (pik : i ∈ I, k ∈ K) of elements of R. Then the
following are equivalent.
(i) There exists a continuous R-linear map ϕ : TropR(ui : i ∈ I) → TropR(vk :
(ii) For all k ∈ K, there are only finitely many indices i ∈ I such that pik is
for all i ∈ I.
k∈K vpik
k
k ∈ K) with ϕ(ui) =(cid:81)
(cid:16)(cid:89)
nonzero.
When these conditions hold, the unique map ϕ is
(cid:17)
(cid:89)
k∈K
(cid:80)
v
k
uai
i
=
ϕ
i∈I
i∈I pikai
.
We conclude this section by giving the details of the definition of the product
topology and proving Proposition 3.7. We assume familiarity with the most basic
ideas of point-set topology. Details and proofs regarding the product topology are
found, for example, in [14, Chapter 3]. Let (Xi : i ∈ I) be a family of topological
i∈I Xi,
and write the elements of X as (ai : i ∈ I) with each ai ∈ Xi. Let Pj : X → Xj
be the jth projection map, sending (ai : i ∈ I) to aj. The product topology on X
is the coarsest topology on X such that each Pj is continuous. Open sets in X are
i∈I Oi where each Oi is an open set in Xi and
the identity Oi = Xi holds for all but finitely many i ∈ I.
spaces, where I is an arbitrary indexing set. Let X be the direct product(cid:81)
arbitrary unions of sets of the form(cid:81)
(meaning that every subset of R is open). We write XI for (cid:81)
We are interested in the case where each Xi is R with the discrete topology
i∈I Xi in this case.
For each i ∈ I, let ei ∈ XI be the element whose entry is 1 in the position indexed
by i, with entries 0 in every other position. Later in the paper, we use the bold
symbol ei for a standard unit basis vector in Rn. Here, we intend the non-bold
symbol ei to suggest a similar idea, even though a basis for XI may have many
more elements than I. The open sets in XI are arbitrary unions of sets of the form
i∈I Oi where each Oi is a subset of R and the identity Oi = R holds for all but
finitely many i ∈ I. The following is a rephrasing of Proposition 3.7.
Proposition 3.8. Let XI and XK be as above for indexing sets I and K. Fix a
family (pik : i ∈ I, k ∈ K) of elements of R. Then the following are equivalent.
(i) There exists a continuous R-linear map ϕ : XI → XK with ϕ(ei) = (pik : k ∈
(ii) For all k ∈ K, there are only finitely many indices i ∈ I such that pik is
K) for all i ∈ I.
(cid:81)
nonzero.
When these conditions hold, the unique map ϕ is
ϕ(ai : i ∈ I) =
pikai : k ∈ K
(cid:18)(cid:88)
i∈I
(cid:19)
.
To simplify the proof of Proposition 3.8, we appeal to the following lemma.
Lemma 3.9. A map ϕ : XI → XK is continuous and R-linear if and only if the
map Pk ◦ ϕ : XI → R is continuous and R-linear for each k ∈ K.
The assertions of the lemma for continuity and linearity are independent of each
other. A direct proof for continuity is found in [14, Chapter 3]. The assertion for
linearity is immediate because the linearity of each Pk ◦ ϕ is the linearity of ϕ in
UNIVERSAL GEOMETRIC CLUSTER ALGEBRAS
9
each entry, which is equivalent to the linearity of ϕ. Lemma 3.9 reduces the proof
of Proposition 3.8 to the following proposition:
i∈I piai.
Proposition 3.10. Let XI be as above and endow R with the discrete topology.
Fix a family (pi : i ∈ I) of elements of R. Then the following are equivalent.
(i) There exists a continuous R-linear map ϕ : XI → R with ϕ(ei) = pi for all
(ii) There are only finitely many indices i ∈ I such that pi is nonzero.
i ∈ I.
When these conditions hold, the unique map ϕ is ϕ(ai : i ∈ I) =(cid:80)
each Om is a Cartesian product(cid:81)
U =(cid:81)
Proof. Suppose ϕ : XI → R is a continuous R-linear map with ϕ(ei) = pi for all
i ∈ I. The continuity of ϕ implies in particular that ϕ−1({0}) is some open set O in
XI . The set O is the union, indexed by some set M , of sets (Om : m ∈ M ), where
i∈I Omi with Omi ⊆ R for all i ∈ I and with the
identity Omi = R holding for all but finitely many i ∈ I. The set M is nonempty
because R-linearity implies that φ(0 · ei) = 0 for any i ∈ I. We choose M and the
sets Om to be maximally inclusive, in the following sense: If O contains some set
i∈I Ui with Ui ⊆ R for all i ∈ I and with the identity Ui = R holding for all
but finitely many i ∈ I, then there exists m ∈ M such that U = Om.
For each m ∈ M , let Sm be the finite set of indices i ∈ I such that Omi (cid:54)= R
and choose µ ∈ M to minimize the set Sµ under containment. A priori, there may
not be a unique minimum set, but some minimal set exists because M is nonempty
and the sets Sm are finite. We now show that pi = 0 if and only if i (cid:54)∈ Sµ.
If pi = 0 for some i ∈ Sµ, then for any element z of Oµ, the element z + cei is
j∈I Oµj by replacing Oµi with R
is contained in O, and so equals Oµ(cid:48) for some µ(cid:48) ∈ M . This contradicts our choice of
µ to minimize Sµ, and we conclude that pi (cid:54)= 0 for all i ∈ Sµ. Now suppose i (cid:54)∈ Sµ.
Let x ∈ Oµ and let y be obtained from x by subtracting 1 in the entry indexed by i.
Since Oµi = R, we have y ∈ Oµ as well. Thus pi = ϕ(ei) = ϕ(x − y) = 0 − 0 = 0.
Given (i), we have established (ii), and now we show that the map ϕ is given by
i∈I piai, which is a finite sum by (ii). Let a = (ai : i ∈ I) ∈ XI .
Let b = (bi : i ∈ I) ∈ XI have bi = 0 whenever i ∈ I \ Sµ and bi = ai whenever
i ∈ Sµ. Then a and b differ by an element of Oµ, so ϕ(a − b) = 0 by linearity, and
biei, so ϕ(a) = ϕ(b) =
i∈Sµ
Finally, suppose (ii) holds, let S be the finite set {i ∈ I : pi (cid:54)= 0}, and define
i∈I piai. It is immediate that this map is R-linear.
The map is also continuous: Given any subset U of R, let O = ϕ−1(U ). Then, for
i∈I Oai where Oai = {ai} if
i ∈ S or Oai = R if i (cid:54)∈ S. The map ϕ is constant on Oa, so {a} ⊆ Oa ⊆ O for each
(cid:3)
ϕ(ai : i ∈ I) =(cid:80)
thus ϕ(a) = ϕ(b). But b is a finite linear combination(cid:80)
(cid:80)
pibi =(cid:80)
ϕ : XI → R by ϕ(ai : i ∈ I) =(cid:80)
every a = (ai : i ∈ I) ∈ O, let Oa be the product (cid:81)
a ∈ O. Thus O =(cid:83)
in O. In particular, the set obtained from Oµ =(cid:81)
a∈A Oa, and this is an open set in XI .
piai. Since pi = 0 for i (cid:54)∈ Sµ, we have ϕ(a) =(cid:80)
i∈Sµ
i∈Sµ
i∈I piai.
4. Bases for B
In this section, we define the notion of an R-basis for an exchange matrix B and
show that an extended exchange matrix is universal for B over R if and only if its
coefficient rows constitute an R-basis for B. This amounts to a simple rephrasing
of the definition combined with a reduction to single components.
10
NATHAN READING
Definition 4.1 (Mutation maps). The exchange matrix B defines a family of
k : Rn → Rn : k ∈ [n]) as follows. (The use of the real
piecewise linear maps (ηB
numbers R here rather than the more general R is deliberate.) Given a vector
a = (a1, . . . , an) ∈ Rn, let B be an extended exchange matrix having exchange
matrix B and having a single coefficient row a = (a1, . . . , an). Let ηB
k (a) be the
coefficient row of µk( B). By (2.2), ηB
n), where, for each j ∈ [n]:
−ak
aj + akbkj
aj − akbkj
aj
1, . . . , a(cid:48)
k (a) = (a(cid:48)
if j = k;
if j (cid:54)= k, ak ≥ 0 and bkj ≥ 0;
if j (cid:54)= k, ak ≤ 0 and bkj ≤ 0;
otherwise.
(4.1)
a(cid:48)
j =
k is a continuous, piecewise linear, invertible map with inverse ηµk(B)
The map ηB
so in particular it is a homeomorphism from Rn to itself.
For any finite sequence k = kq, kq−1, . . . , k1 of indices in [n], let B1 = B and
define Bi+1 = µki (Bi) for i ∈ [q]. Equivalently, Bi+1 = µki,...,k1(B) for i ∈ [q]. As
before, we index the sequence k so that the first entry in the sequence is on the
right. Define
k
,
= ηBq
kq
◦ ηBq−1
kq−1
◦ ··· ◦ ηB1
k = ηB
ηB
kq,kq−1...,k1
. When k is the empty sequence, ηB
(4.2)
This is again a piecewise linear homeomorphism from Rn to itself, with inverse
ηBq+1
k are
k1,...,kq
collectively referred to as the mutation maps associated to B.
Definition 4.2 (B-coherent linear relations). Let B be an n× n exchange matrix.
Let S be a finite set, let (vi : i ∈ S) be vectors in Rn and let (ci : i ∈ S) be elements
i∈S civi is a B-coherent linear relation with
k is the identity map. The maps ηB
k1
of R. Then the formal expression(cid:80)
coefficients in R if the equalities(cid:88)
(cid:88)
(4.3)
i∈S
(4.4)
ciηB
k (vi) = 0, and
cimin(ηB
k (vi), 0) = 0
i∈S
hold for every finite sequence k = kq, . . . , k1 of indices in [n]. The requirement that
(4.3) holds when k is the empty sequence ensures that a B-coherent linear relation
is in particular a linear relation in the usual sense. A B-coherent linear relation
(cid:80)
i∈S civi is trivial if ci = 0 for all i ∈ S.
The definition of an R-basis for B is parallel to the definition of a linear-algebraic
basis, with B-coherent linear relations replacing ordinary linear relations.
Definition 4.3 (R-Basis for B). Let I be some indexing set and let (bi : i ∈ I)
be a collection of vectors in Rn. Then (bi : i ∈ I) is an R-basis for B if and only
if the following two conditions hold.
(i) If a ∈ Rn, then there exists a finite subset S ⊆ I and elements (ci : i ∈ S) of
R such that a −(cid:80)
(ii) If S is a finite subset of I and(cid:80)
coefficients in R, then ci = 0 for all i ∈ S.
i∈S cibi is a B-coherent linear relation with
If condition (i) holds, then (bi : i ∈ I) is called an R-spanning set for B, and if
condition (ii) holds, then (bi : i ∈ I) is called an R-independent set for B.
i∈S cibi is a B-coherent linear relation.
UNIVERSAL GEOMETRIC CLUSTER ALGEBRAS
11
(i)
(ii)
i∈I
pibt
i = at for every t ∈ Tn.
i, 0) = min(at, 0) for every t ∈ Tn.
Theorem 4.4. Let B be an extended exchange matrix with exchange matrix B and
whose coefficient rows have entries in R. Then B is universal over R if and only
if the coefficient rows of B are an R-basis for B.
Proof. Let (bi : i ∈ I) be the coefficient rows of B. We show that the following
conditions are equivalent.
(a) B is universal over R.
(b) For every extended exchange matrix B(cid:48) with exactly one coefficient row, sharing
the exchange matrix B with B and having entries in R, there exists a unique
coefficient specialization from AR( B) to AR( B(cid:48)).
(c) For every extended exchange matrix B(cid:48) with exactly one coefficient row a =
(a1, . . . , an) ∈ Rn, sharing the exchange matrix B with B, there exists a unique
choice (pi : i ∈ I) of elements of R, finitely many nonzero, such that both of
the following conditions hold:
(cid:88)
(cid:88)
pimin(bt
i∈I
i is the coefficient row of Bt indexed by i ∈ I in the Y -pattern t (cid:55)→ Bt
Here bt
with Bt0 = B and at is the coefficient row of B(cid:48)
t with
B(cid:48)
If (a) holds, then (b) holds by definition. Condition (b) is the assertion that there
exists a unique map satisfying Definition 3.1(iii). By Proposition 3.7, choosing a
continuous R-linear map is equivalent to choosing elements (pi : i ∈ I) of R with
finitely many nonzero. The remainder of Definition 3.1(iii) is rephrased in (c) as
conditions (i) and (ii). We see that Condition (c) is a rephrasing of condition (b).
Now suppose (c) holds and let B(cid:48)(cid:48) be an extended exchange matrix with coefficient
rows indexed by an arbitrary set K. For each k ∈ K, condition (c) implies that
there is a unique choice of elements pik of R satisfying, in the kth component, the
conditions of Proposition 3.7 and of Definition 3.1. The elements pik, taken together
for all k ∈ K, satisfy the conditions of Propositions 3.7 and 3.1, and thus define the
unique coordinate specialization from AR( B) to AR( B(cid:48)(cid:48)). We have verified that (c)
implies (a), so that the three conditions are equivalent.
But (c) is equivalent to the assertion that, for each a ∈ Rn, there exists a unique
finite subset S ⊆ I and unique nonzero elements (ci
: i ∈ S) of R such that
i∈S cibi is a B-coherent linear relation. This is equivalent to the assertion
that (bi : i ∈ I) is an R-basis for B.
(cid:3)
t in the Y -pattern t (cid:55)→ B(cid:48)
a −(cid:80)
= B(cid:48).
t0
Remark 4.5. Given a universal extended exchange matrix B over R and an extended
exchange matrix B(cid:48) sharing the exchange matrix B with B and having entries in R,
the proof of Theorem 4.4 provides an explicit description of the unique coefficient
specialization from AR(x, B) to AR(x(cid:48), B(cid:48)). It is the map sending each xj to x(cid:48)
and acting on coefficient semifields as follows: Let (bi : i ∈ I) be the coefficient
rows of B and let (ak : k ∈ K) be the coefficient rows of B(cid:48). For each k, there
is a unique choice (pik : i ∈ I) of elements of R, finitely many nonzero, such
i∈S pikbi is a B-coherent linear relation. Then the restriction of ϕ to
that ak −(cid:80)
j
coefficient semifields is the map described in Proposition 3.7.
12
NATHAN READING
Proposition 4.6. Suppose the underlying ring R is a field. For any exchange
matrix B, there exists an R-basis for B. Given an R-spanning set U for B, there
exists an R-basis for B contained in U .
Proof. Let U be an R-spanning set for B. Given any chain U1 ⊆ U2 ⊆ ··· of R-
independent sets for B contained in U , the union of the chain is an R-independent
set for B. Thus Zorn's Lemma says that among the R-independent subsets of U ,
there exists a maximal set M . We show that if R is a field, then M is an R-spanning
set for B. If not, then there exists a vector a ∈ Rn such that no B-coherent linear
i∈S cibi over R exists with each bi in M . In particular, a (cid:54)∈ M , so
M ∪ {a} strictly contains M . If M ∪ {a} is not an R-independent set for B, then
since M is an R-independent set for B, there exists a B-coherent linear relation
ci
c bi
is a B-coherent linear relation over R, and this contradiction shows that M ∪ {a}
is an R-independent set for B. That contradicts the maximality of M , and we
conclude that M is an R-spanning set for B. We have proved the second statement
of the proposition. The first statement follows because Rn is an R-spanning set
(cid:3)
for B.
i∈S cibi over R with c (cid:54)= 0 and each bi in M . Since R is a field, a−(cid:80)
relation a −(cid:80)
ca−(cid:80)
i∈S
Theorem 4.4 and Proposition 4.6 combine to prove the following corollary.
Corollary 4.7. Suppose the underlying ring R is a field. For any exchange matrix
B, there exists an extended exchange matrix B with exchange matrix B that is
universal over R. The cluster algebra AR( B) has universal geometric coefficients
over R.
Remark 4.8. The proof of Proposition 4.6 echoes the standard argument showing
that a vector space has a (Hamel) basis. As in the linear algebraic case, this proof
provides no general way of constructing a basis, and thus Corollary 4.7 provides no
general way of constructing a universal extended exchange matrix.
Remark 4.9. The definitions in Section 2 are closest to the original definitions
in [12] when R = Z, and arguably it is most important to find Z-bases for exchange
matrices (and thus universal geometric coefficients over Z). Unfortunately, the
second assertion of Proposition 4.6 can fail when R = Z, as shown in Example 4.10
below. We have no proof of the assertion that every B admits a Z-basis, but also
no counterexample.
In Section 9, we construct R-bases for any B of rank 2 and any R. In Section 10
we construct R-bases for any B of finite type and any R. We also conjecture a form
for R-bases for B of affine type and any R. In [21] and [22], we use laminations to
construct an R-basis (with R = Z or Q) for certain exchange matrices arising from
marked surfaces.
Example 4.10. Suppose B = [0]. If R is a field, then {x, y} is an R-basis for B
if and only if x and y are elements of R with strictly opposite signs. The set {±1}
is the unique Z-basis for B. In particular, the extended exchange matrix
is
universal over any R. The set {−1, 2, 3} is a Z-spanning set for B, but contains
no Z-basis for B. In particular, the second assertion of Proposition 4.7 may fail
without the hypothesis that R is a field.
(cid:104) 0
(cid:105)
1−1
The remainder of this section is devoted to further details on B-coherent linear
relations and bases. First, in the most important cases, condition (4.4) can be
ignored when verifying that a linear relation is B-coherent.
13
UNIVERSAL GEOMETRIC CLUSTER ALGEBRAS
dition (4.3) holds for all sequences k.
Proposition 4.11. Suppose B has no column consisting entirely of zeros. Then
i∈S civi is a B-coherent linear relation if and only if con-
the formal expression(cid:80)
Proof. Suppose condition (4.3) holds for all k for the formal expression(cid:80)
i∈S civi.
We need to show that condition (4.4) holds as well. Given any sequence k of
indices in [n] and any additional integer k ∈ [n], we show that (4.4) holds in the kth
k (vi) for each i ∈ S, let S>0 be the subset of S consisting
component. Write wi = ηB
of indices i such that the kth coordinate of wi is positive, and let S≤0 = S \ S>0.
Consider condition (4.3) for the sequences k and kk. These conditions are
(wi).
i∈S>0
ciwi = (−a1, . . . ,−an). Since
all of the vectors wi for i ∈ S>0 have positive kth coordinate, (4.1) says that
ciwi = −(cid:80)
ciwi = (a1, . . . , an), so that (cid:80)
(wi) = −(cid:80)
ciηµk(B)
k
i∈S≤0
ciηµk(B)
i∈S≤0
i∈S≤0
i∈S>0
i∈S>0
k
n) given by:
k
i∈S>0
ciηµk(B)
1, . . . , a(cid:48)
(wi) is (a(cid:48)
(cid:80)
Let (cid:80)
(cid:80)
ciwi and(cid:80)
−ak
ak
The requirement that (cid:80)
aj + akbkj
aj
−aj + akbkj
−aj
given by:
a(cid:48)(cid:48)
j =
a(cid:48)
j =
ciηµk(B)
where bkj is the kj-entry of µk(B). Similarly,(cid:80)
if j = k;
if j (cid:54)= k and bkj ≥ 0;
otherwise,
i∈S≤0
ciηµk(B)
k
(wi) is (a(cid:48)(cid:48)
1 , . . . , a(cid:48)(cid:48)
n),
if j = k;
if j (cid:54)= k and bkj ≤ 0;
otherwise.
(wi) = −(cid:80)
ciηµk(B)
kth coordinate of(cid:80)
Proposition 4.12. Let (cid:80)
i∈S≤0
k
1, . . . , a(cid:48)
1 , . . . , a(cid:48)(cid:48)
n) = −(a(cid:48)(cid:48)
(wi) means that
i∈S>0
(a(cid:48)
n). Therefore akbkj = 0 for all j. The property of having
a column consisting entirely of zeros is preserved under mutation, so µk(B) has no
column of zeros. Since µk(B) is skew-symmetrizable, it also has no row consisting
entirely of zeros. We conclude that ak = 0. In particular, we have showed that the
(cid:3)
ciwi is zero. This is the kth component of (4.4).
i∈S≤0
k
We record a simple but useful observation about B-coherent linear relations.
i∈S civi be a B-coherent linear relation. Suppose, for
some i ∈ S, for some j ∈ [n], and for some sequence k of indices in [n], that the jth
entry of ηB
k (vi) is strictly positive (resp. strictly negative) and that the jth entry
k (vi(cid:48)) with i(cid:48) ∈ S \ {i} is nonpositive (resp. nonnegative). Then
of every vector ηB
ci = 0.
k (vi) is strictly positive and the jth entries of
Proof. Suppose the jth entry of ηB
the other vectors ηB
k (vi(cid:48)) are nonpositive. Consider the jth coordinate of (4.3)
and (4.4), for the chosen sequence k. The difference between the two left-hand
sides is the jth coordinate of ciηB
k (vi) has a
positive entry in its jth position, so ci = 0.
k (vi), which is therefore zero. But ηB
If the jth entry of ηB
vectors are nonnegative, then the jth entry of ηB
jth entries of the other vectors ηB
ci = 0.
k (vi) is strictly negative and the jth entries of the other
jk(vi) is strictly positive and the
jk(vi(cid:48)) are nonpositive, and we conclude that
(cid:3)
14
NATHAN READING
To further simplify the task of finding universal geometric coefficients, we con-
clude this section with a brief discussion of reducibility of exchange matrices.
Definition 4.13 (Reducible (extended) exchange matrices). Call an exchange ma-
trix B reducible if there is some permutation π of [n] such that simultaneously
permuting the rows and columns of B results in a block-diagonal matrix. Otherwise
call B irreducible.
The following proposition is immediate, and means that we need only construct
bases for irreducible exchange matrices.
Proposition 4.14. Suppose B is a p×p exchange matrix and B(cid:48) is a q×q exchange
matrix. If (bi : i ∈ I) is an R-basis for B1 and (b(cid:48)
j : j ∈ J) is an R-basis for B2,
then (bi × 0q : i ∈ I) ∪ (0p × b(cid:48)
j : j ∈ J) is an R-basis for(cid:2) B1 0
Here bi × 0q represents the vector bi ∈ Rp included into Rn by adding q zeros
(cid:3).
0 B2
at the end, and 0p × b(cid:48)
j is interpreted similarly.
Proposition 4.14 also allows us to stay in the case where B has no column of
zeros, so that the definition of B-coherent linear relations simplifies as explained
in Proposition 4.11. This is because an exchange matrix with a column of zeros is
reducible and has the 1 × 1 exchange matrix [0] as a reducible component. This
component is easily dealt with as explained in Example 4.10.
5. B-cones and the mutation fan
In this section, we use the mutation maps associated to an exchange matrix B
to define a collection of closed convex real cones called B-cones. These define a fan
called the mutation fan for B.
Definition 5.1 (Cones). A convex cone is a subset of Rn that is closed under
positive scaling and under addition. A convex cone, by this definition, is also convex
in the usual sense. A polyhedral cone is a cone defined by finitely many weak
linear inequalities, or equivalently it is the nonnegative R-linear span of finitely
many vectors. A rational cone is a cone defined by finitely many weak linear
inequalities with integer coefficients, or equivalently it is the nonnegative R-linear
span of finitely many integer vectors. A simplicial cone is the nonnegative span
of a set of linearly independent vectors.
Definition 5.2 (B-classes and B-cones). Let B be an n × n exchange matrix.
Define an equivalence relation ≡B on Rn by setting a1 ≡B a2 if and only if
sgn(ηB
k (a2)) for every finite sequence k of indices in [n]. Recall
that sgn(a) denotes the vector of signs of the entries of a. Thus a1 ≡B a2 means
k (a2) ∈ H for every open coordinate halfspace H
that ηB
of Rn and every sequence k. The equivalence classes of ≡B are called B-classes.
The closures of B-classes are called B-cones. The latter term is justified in Propo-
sition 5.4, below.
k (a1)) = sgn(ηB
k (a1) ∈ H if and only if ηB
Proposition 5.3. Every mutation map ηB
k is linear on every B-cone.
Proof. Let C be a B-class and let k = kq, . . . , k1. We show by induction on q
that the map ηB
k is linear on C. The base case q = 0 is trivial. If q > 0 then let
k(cid:48) = kq−1, . . . , k1. By induction, the map ηB
k(cid:48) is linear on C. Since C is a B-class,
k(cid:48) takes C to a set C(cid:48) on which the function sgn is constant. In particular, C(cid:48)
ηB
UNIVERSAL GEOMETRIC CLUSTER ALGEBRAS
15
is contained in one of the domains of linearity of ηµk(cid:48) (B)
linear on C. Since ηB
k(cid:48) is
k is also a continuous map, it is linear on the closure of C. (cid:3)
Proposition 5.4. Each B-class is a convex cone containing no line and each B-
cone is a closed, convex cone containing no line.
k = ηµk(cid:48) (B)
, so ηB
kq
◦ ηB
kq
k (a1) + ηB
Proof. First, notice that each B-class is closed under positive scaling because each
map ηB
k commutes with positive scaling, and because the function sgn is unaffected
by positive scaling. Furthermore, if a1 ≡B a2 then sgn(ηB
k (a2)), for
any finite sequence k of indices in [n]. Thus by Proposition 5.3, sgn(ηB
k (a1 +a2)) =
k (a1)), so a1 ≡B a1 + a2. Since B-classes are
sgn(ηB
closed under positive scaling and addition, they are convex cones. The requirement
k (· )) is constant within B-classes implies in particular (taking k to be
that sgn(ηB
the empty sequence) that each B-class is contained in a coordinate orthant.
In
particular, no B-class contains a line. We have verified the assertion for B-classes,
(cid:3)
and the assertion for B-cones follows.
k (a2)) = sgn(ηB
k (a1)) = sgn(ηB
Proposition 5.5. Let k be a finite sequence of indices in [n]. Then a set C is a
B-class if and only if ηB
k (C) is a µk(B)-class. A set C is a B-cone if and only if
ηB
k (C) is a µk(B)-cone.
Proof. Let C be a B-class and let k(cid:48) = k(cid:48)
1 be another finite sequence of
indices in [n]. Then since C is a B-class, the function sgn is constant on ηB
k(cid:48)k(C),
where k(cid:48)k is the concatenation of k(cid:48) and k. Letting k(cid:48) vary over all possible
k (C) is contained in some µk(B)-class C(cid:48). Symmetrically,
sequences, we see that ηB
if k(cid:48)(cid:48) is the reverse sequence of k, ηµk(B)
(C(cid:48)) is contained in some B-class C(cid:48)(cid:48). But
k )−1, so C ⊆ C(cid:48)(cid:48). By definition, distinct B-classes are disjoint, so
ηµk(B)
k(cid:48)(cid:48)
C(cid:48)(cid:48) = C, and thus C(cid:48) = ηB
k is a
(cid:3)
homeomorphism.
k (C). The assertion for B-cones follows because ηB
q(cid:48), . . . , k(cid:48)
= (ηB
k(cid:48)(cid:48)
The following is [12, Definition 6.12].
Definition 5.6 (Sign-coherent vectors). A collection X of vectors in Rn is sign-
coherent if for any k ∈ [n], the kth coordinates of the vectors in X are either all
nonnegative or all nonpositive.
As pointed out in the proof of Proposition 5.4, each B-cone is contained in some
coordinate orthant. In other words:
Proposition 5.7. Every B-cone is a sign-coherent set.
Our understanding of B-cones allows us to mention the simplest kind of B-
coherent linear relation.
Definition 5.8 (B-local linear relations). Let B be an n × n exchange matrix.
Let S be a finite set, let (vi : i ∈ S) be vectors in Rn and let (ci : i ∈ S) be real
i∈S civi is a B-local linear relation if
i∈S civi = 0 holds and if {vi : i ∈ S} is contained in some B-cone.
numbers. Then the formal expression (cid:80)
the equality(cid:80)
Proof. Let(cid:80)
i∈S civi be a B-local linear relation. Then by definition, (4.3) holds
for k empty. Now Proposition 5.3 implies that (4.3) holds for all k. Propositions 5.5
and 5.7 imply that, for any k, each coordinate of (4.4) is either the tautology 0 = 0
(cid:3)
or agrees with some coordinate of (4.3).
Proposition 5.9. A B-local linear relation is B-coherent.
16
NATHAN READING
In order to study the collection of all B-cones, we first recall some basic defini-
tions from convex geometry.
Definition 5.10 (Face). A subset F of a convex set C is a face if F is convex and
if any line segment L ⊆ C whose interior intersects F has L ⊆ F . In particular, the
empty set is a face of C and C is a face of itself. Also, if H is any hyperplane such
that C is contained in one of the two closed halfspaces defined by H, then H ∩ C
is a face of C. The intersection of an arbitrary set of faces of C is another face of
C. A face of a closed convex set is closed.
Definition 5.11 (Fan). A fan is a collection F of closed convex cones such that if
C ∈ F and F is a face of C, then F ∈ F, and such that the intersection of any two
cones in F is a face of each of the two. In some contexts, a fan is required to have
finitely many cones, but here we allow infinitely many cones. A fan is complete if
the union of its cones is the entire ambient space. A simplicial fan is a fan all of
whose cones are simplicial. A subfan of a fan F is a subset of F that is itself a
fan. If F1 and F2 are complete fans such that every cone in F2 is a union of cones
in F1, then we say that F1 refines F2 or equivalently that F2 coarsens F1.
Definition 5.12 (The mutation fan for B). Let FB be the collection consisting
of all B-cones, together with all faces of B-cones. This collection is called the
mutation fan for B. The name is justified by the following theorem.
Theorem 5.13. The collection FB is a complete fan.
To prove Theorem 5.13, we introduce some additional background on convex
sets and prove several preliminary results.
Definition 5.14 (Relative interior ). The affine hull of a convex set C is the union
of all lines defined by two distinct points of C. The relative interior of C, written
relint(C), is its interior as a subset of its affine hull, and C is relatively open if it
equals its relative interior. The relative interior of a convex set is nonempty. (See
e.g. [29, Theorem 2.3.1].)
We need several basic facts about convex sets. Proofs of the first four are found,
for example, in [29, Theorem 2.3.4], [29, Theorem 2.3.8], and [29, Corollary 2.4.11].
Lemma 5.15. Let C be a convex set in Rn. Let x be in the closure of C, let y be
in the relative interior of C and let ε ∈ [0, 1). Then εx + (1 − ε)y is in the relative
interior of C.
Lemma 5.16. The relative interior of a convex set C in Rn equals the relative
interior of the closure of C.
Lemma 5.17. A closed convex set in Rn is the closure of its relative interior.
Lemma 5.18. If C and D are nonempty convex sets in Rn with disjoint relative
interiors, then there exists a hyperplane H, defining halfspaces H+ and H− such
that C ⊆ H+, D ⊆ H− and C ∪ D (cid:54)⊆ H.
Lemma 5.19. Let F , G, M , and N be convex sets in Rn such that F is a face
of M and G is a face of N . Suppose M ∩ N is a face of M and of N . Then F ∩ G
is a face of F and of G.
UNIVERSAL GEOMETRIC CLUSTER ALGEBRAS
17
Proof. If F ∩ G is empty, then we are done. Otherwise, let L be any line segment
contained in F whose relative interior intersects F ∩ G. Then L is in particular a
line segment in M whose relative interior intersects M ∩ N . By hypothesis, M ∩ N
is a face of M , so L is contained in M ∩ N . Since G is a face of N , since L is
contained in N , and since the relative interior of L intersects G, we see that L is
contained in G. Thus L is contained in F ∩ G. We have shown that F ∩ G is a face
of F , and the symmetric argument shows that F ∩ G is a face of G.
(cid:3)
Lemma 5.20. If C is a sign-coherent convex set then sgn is constant on relint(C).
Proof. Let x, y ∈ relint(C) and suppose sgn(xi) (cid:54)= sgn(yi) for some index i ∈ [n].
Since x and y are in relint(C), the relative interior of C contains an open interval
about x in the line containing x and y. If sgn(xi) = 0, then sgn(yi) (cid:54)= 0, and thus
this interval about x contains points whose ith coordinate have all possible signs.
Arguing similarly if sgn(yi) = 0, we see that in any case, relint(C) contains points
whose ith coordinates have opposite signs. This contradicts the sign-coherence
(cid:3)
of C.
We now prove some preliminary results about B-cones.
Proposition 5.21. Every B-cone C is the closure of a unique B-class, and this
B-class contains relint(C).
Proof. The B-cone C is the closure of some B-class C(cid:48). In light of Proposition 5.4,
Lemma 5.16 says that relint(C(cid:48)) = relint(C). In particular, relint(C) ⊆ C(cid:48). If C
is the closure of some other B-class C(cid:48)(cid:48), then also C(cid:48)(cid:48) contains relint(C). Since
(cid:3)
relint(C) is nonempty and since distinct B-classes are disjoint, C(cid:48) = C(cid:48)(cid:48).
To further describe B-cones, we will use a partial order on sign vectors that
appears in the description of the ("big") face lattice of an oriented matroid. (See
[1, Section 4.1].)
Definition 5.22 (A partial order on sign vectors). A sign vector is an n-tuple
whose entries are in {−1, 0, 1}, or in other words, a vector that arises by applying
the operator sgn to a vector in Rn. For sign vectors x and y, say x (cid:22) y if x
agrees with y, except possibly that some entries 1 or −1 in y become 0 in x. The
point of this definition is to capture a notion of limits of sign vectors: If v is the
limit of a sequence or continuum of vectors all having the same sign vector y, then
sgn(v) (cid:22) y.
Proposition 5.23. Let C(cid:48) be a B-class whose closure is the B-cone C. Let y be
any point in C(cid:48). Then C is the set of points x such that sgn(ηB
k (y))
for all sequences k of indices in [n].
Proof. Let y(cid:48) be any point in relint(C). Proposition 5.21 says that C(cid:48) contains
relint(C), so the vectors sgn(ηB
Suppose x ∈ C and let k be a sequence of indices. By Propositions 5.3 and 5.5,
k (x) is in ηB
k (C),
k (C(cid:48)). By Lemma 5.15, the entire line segment from ηB
k (y(cid:48))
k (C), and
k (C(cid:48)) by Proposition 5.21. Therefore sgn(· ) is constant
k (y(cid:48))) =
the point ηB
which is the closure of ηB
to ηB
thus in the µk(B)-class ηB
on the line segment, except possibly at ηB
sgn(ηB
k (y(cid:48)) is in the relative interior of ηB
k (x), is in the relative interior of ηB
k (x)) (cid:22) sgn(ηB
k (x), so sgn(ηB
k (x), except possibly the point ηB
k (y)).
k (x)) (cid:22) sgn(ηB
k (y(cid:48))) and sgn(ηB
k (y)) agree for any sequence k.
k (C) and the point ηB
18
NATHAN READING
k (x)) (cid:22) sgn(ηB
k (x) and ηB
On the other hand, suppose sgn(ηB
k (L) is the line segment defined by ηB
k (y(cid:48)), and this constant sign vector equals sgn(ηB
k (y)) for all sequences k of
indices in [n]. Then the function sgn is constant on the relative interior of line seg-
ment defined by ηB
k (y)).
Let L be the line segment defined by x and y(cid:48). We show by induction that ηB
k is
k (y(cid:48)). The
linear on L, so that ηB
base case where k is the empty sequence is easy because ηB
k is the identity map. If k
is not the empty sequence, then write k = jk(cid:48) for some index j and some sequence
k(cid:48). By induction on the length of the sequence k, the map ηB
k(cid:48) is linear on L, so
k(cid:48)(y(cid:48)). Since sgn is constant on
ηB
k(cid:48)(L) is the line segment defined by ηB
the relative interior of ηB
k(cid:48)(L), the map ηB
k (L) is the
k(cid:48)(y(cid:48)).
line segment defined by ηB
We have shown that sgn(ηB
k (y)) on relint(L)
for all k. Thus relint(L) is in C(cid:48). All of L, including x, is therefore in the closure
(cid:3)
of C(cid:48), which is the B-cone C.
k(cid:48)(x) and ηB
k is also linear on L and thus ηB
k (· )) is constant and equals sgn(ηB
k(cid:48)(x) and ηB
k (x) and ηB
Proposition 5.24. Every B-cone is a union of B-classes.
Proof. For any B-class C, Proposition 5.23 implies that a B-class D is either com-
(cid:3)
pletely contained in C or disjoint from C.
The following proposition generalizes Proposition 5.21.
Proposition 5.25. Let C be a B-cone and let F be a nonempty convex set con-
tained in C. Then the relative interior of F is contained in a B-class D with
D ⊆ C.
Proof. Let k be some sequence of indices in [n]. By Proposition 5.5, ηB
a µk(B)-cone. By Proposition 5.3, ηB
Thus ηB
interior of F to the relative interior of ηB
k (C) is
k (C) is the image of C under a linear map.
k maps the relative
k (F ) is a nonempty convex set contained in ηB
Proposition 5.7 says that ηB
k (F ) is sign-coherent, so Lemma 5.20 says that sgn
is constant on the relative interior of ηB
k is constant
on the relative interior of F . Since this was true for any k, the relative interior
of F is contained in some B-class D. Since D is a B-class which intersects C,
(cid:3)
Proposition 5.24 says that D is contained in C.
k (F ).
k (F ). Equivalently, sgn ◦ ηB
k (C), and ηB
Then C =(cid:84)
Proposition 5.26. An arbitrary intersection of B-cones is a B-cone.
Proof. Let I be an arbitrary indexing set and, for each i ∈ I, let Ci be a B-cone.
i∈I Ci is non-empty because it contains the origin. Furthermore, it is
a closed convex cone because it is an intersection of closed convex cones. Let i ∈ I.
Then Proposition 5.25 says that the relative interior of C is contained in a B-class
D with D ⊆ Ci. Since distinct B-classes are disjoint, applying Proposition 5.25
to each i, we obtain the same D, and we conclude that D is contained in C. By
Lemma 5.17, C is the closure of relint(C). Since relint(C) ⊆ D ⊆ C and C is
(cid:3)
closed, we conclude that C is the closure of D. Thus C is a B-cone.
Lemma 5.27. If C and D are B-cones with C ⊆ D, then C is a face of D.
Proof. Let F be the intersection of all faces of D which contain C. Since D is itself
a face of D, this intersection is well-defined. Then F is a face of D, and we claim
that the relative interior of C intersects the relative interior of F .
UNIVERSAL GEOMETRIC CLUSTER ALGEBRAS
19
Suppose to the contrary that the relative interior of C does not intersect the
relative interior of F . Then Lemma 5.18 says that there exists a hyperplane H,
defining halfspaces H+ and H− such that C ⊆ H+, F ⊆ H− and C ∪ F (cid:54)⊆ H. Since
F contains C, we have C ⊆ H−, so C ⊆ H+ ∩ H− = H. Therefore F (cid:54)⊆ H. Now
H ∩ F is a proper face of F containing C, contradicting our choice of F as the
intersection of all faces of D which contain C. This contradiction proves the claim.
By Proposition 5.25, there is some B-class F (cid:48) containing relint(F ). By Propo-
sition 5.21, C is the closure of a B-class C(cid:48) containing relint(C). By the claim,
relint(C) ∩ relint(F ) is nonempty, so F (cid:48) ∩ C(cid:48) is nonempty. Since distinct B-classes
are disjoint, we conclude that F (cid:48) = C(cid:48). Thus relint(F ) ⊆ C(cid:48). Proposition 5.17 says
that F is the closure of relint(F ), so applying closures to the relation relint(F ) ⊆ C(cid:48),
we see that F ⊆ C. By construction C ⊆ F , so C = F , and thus C is a face of D. (cid:3)
Remark 5.28. One can phrase a converse to Lemma 5.27: If D is a B-cone and C is
a face of D, then C is a B-cone. This statement is false; a counterexample appears
in the proof of Proposition 9.9.
We are now prepared to prove the main theorem of this section.
Proof of Theorem 5.13. By Proposition 5.4, each element of FB is a convex cone.
By the definition of FB, if C ∈ FB then all faces of C are also in FB. Thus, to show
that FB is a fan, it remains to show that the intersection of any two cones in FB is
a face of each. Since every cone in FB is a face of some B-cone, Lemma 5.19 says
that it is enough to consider the case where both cones are B-cones. If C and D are
distinct B-cones, then Proposition 5.26 says that C ∩ D is a B-cone. Lemma 5.27
says that C ∩ D is a face of C and that C ∩ D is a face of D.
Since the union of all B-classes is all of Rn, the union of the B-cones is also all
of Rn. Thus FB is complete.
(cid:3)
We conclude the section with two more useful facts about B-cones.
Proposition 5.29. For any a ∈ Rn, there is a unique smallest B-cone contain-
ing a. This B-cone is the closure of the B-class of a.
Proof. By Proposition 5.26, the intersection of all B-cones containing a is a B-
cone C. Proposition 5.24 says that any B-cone containing a contains the entire
(cid:3)
B-class of a, so C is the closure of the B-class of a.
Proposition 5.30. A set C ⊆ Rn is contained in some B-cone if and only if the
set ηB
k (C) is sign-coherent for every sequence k of indices in [n].
Proof. The "only if" direction follows immediately from Proposition 5.23. Suppose
the set ηB
k (C) is sign-coherent for every sequence k of indices in [n]. Then the set
of finite nonnegative linear combinations of vectors in ηB
k (C) is sign-coherent for
every k. Let D be the set of finite nonnegative linear combinations of vectors in C.
Arguing by induction as in the second half of the proof of Proposition 5.23, we see
k is linear on D, so that ηB
that every map ηB
k (D) is the set of finite nonnegative
linear combinations of vectors in ηB
k (D) is sign-
k (· )) is constant on relint(D) by Lemma 5.20. Thus relint(D)
coherent, so sgn(ηB
is contained in some B-class and D is contained in the closure of that B-class. (cid:3)
k (C). Therefore, for all k, the set ηB
Proposition 5.30 suggests a method for computing approximations to the muta-
tion fan FB. The proposition implies that vectors x and y are not contained in a
20
NATHAN READING
common B-cone if and only if, for some k and j, they are strictly separated by the
k )−1, of the jth coordinate hyperplane. Thus we approximate FB
image, under (ηB
by computing these inverse images for all sequences k of length up to some m. The
inverse images define a decomposition of Rn that may be coarser than FB but that
approaches FB as m → ∞.
Example 5.31. The approximation to FB, for B =
and m = 9, is shown
in Figure 1. The picture is interpreted as follows: Intersecting each inverse image
(cid:104) 0 −3
2
0
0 −3
(cid:105)
0
2
0
e2
e3
e1
Figure 1. The mutation fan FB for B =
(cid:104) 0 −3
2
0
0 −3
(cid:105)
0
2
0
of a coordinate plane with a unit sphere about the origin, we obtain a collection of
arcs of great circles. These are depicted in the plane by stereographic projection.
The projections of the unit vectors e1, e2, and e3 are labeled. We suspect that the
differences between this approximation and FB are unnoticeable at this resolution.
6. Positive bases and cone bases
In this section, we discuss two special properties that an R-basis for B may have.
One of these properties is a notion of positivity. It is not clear what consequences
positivity has for cluster algebras, but it is a very natural notion for R-bases and for
coefficient specializations. Many of the bases that have been constructed, here and
UNIVERSAL GEOMETRIC CLUSTER ALGEBRAS
21
in [21, 22], are positive. The second special property, a condition on the interaction
of the basis with the mutation fan, follows from the positivity property.
subset S ⊆ I and positive elements (ci : i ∈ S) of R such that a −(cid:80)
Definition 6.1 (Positive basis and positive universal extended exchange matrix ).
An R-basis (bi : i ∈ I) for B is positive if, for every a ∈ Rn, there exists a finite
i∈S cibi
is a B-coherent linear relation. The corresponding universal extended exchange
matrix for B is also called positive in this case. Looking back at Remark 4.5, we
see that a positive universal extended exchange matrix B has a special property
with respect to coefficient specializations: Suppose B(cid:48) is another extended exchange
matrix sharing the same exchange matrix with B. When we describe the unique
coefficient specialization ϕ from AR( B) to AR( B(cid:48)) as in Proposition 3.7, all of the
pik are nonnegative.
k∈Sj
j∈S
k∈Sj
(cid:80)
an index i ∈ I, there is a B-coherent linear relation ai −(cid:80)
(cid:80)
coherent relation ai −(cid:80)
Proposition 6.2. For any fixed R, there is at most one positive R-basis for B, up
to scaling each basis element by a positive unit in R.
Proof. Suppose (ai : i ∈ I) and (bj : j ∈ J) are positive R-bases for B. Given
j∈S cjbj with positive
coefficients cj ∈ R. For each j ∈ S, there is a B-coherent linear relation bj −
djkak with positive coefficients djk ∈ R. Combining these, we obtain a B-
cjdjkak. Since (ai : i ∈ I) is an R-basis for B,
this B-coherent relation is trivial. Thus the summation over j ∈ S and k ∈ Sj has
only one term ai. In particular, the set S has exactly one element j, the set Sj has
only one element i, and cjdji = 1. We see that ai = cjbj for some j ∈ J, where cj
is a positive unit with inverse dji. Thus every basis element ai is obtained from a
(cid:3)
basis element bj by scaling by a positive unit.
Definition 6.3 (Cone basis for B). Let I be some indexing set and let (bi : i ∈ I)
be a collection of vectors in Rn. Then (bi : i ∈ I) is a cone R-basis for B if and
only if the following two conditions hold.
(i) If C is a B-cone, then the R-linear span of {bi : i ∈ I} ∩ C contains Rn ∩ C.
(ii) (bi : i ∈ I) is an R-independent set for B.
Proposition 6.4. If (bi : i ∈ I) is a cone R-basis for B, then (bi : i ∈ I) is an
R-basis for B. An R-basis is a cone R-basis if and only if its restriction to each
B-cone C is a basis (in the usual sense) for the R-linear span of the vectors in
Rn ∩ C.
Proof. Condition (ii) of Definition 4.3 is identical to condition (ii) of Definition 6.3.
Suppose (bi : i ∈ I) is a cone R-basis for B. Let a ∈ Rn and let C be the B-
cone that is the closure of the B-class of a. By condition (i) of Definition 6.3,
i∈S cibi of vectors in {bi : i ∈ I} ∩ C. The linear
i∈S cibi is B-local and thus B-coherent by Proposition 5.9. We have
established condition (i) of Definition 4.3, thus proving the first assertion of the
proposition.
Still supposing (bi : i ∈ I) to be a cone R-basis for B, let C now be any B-cone.
If there is some non-trivial linear relation among the vectors in {bi : i ∈ I} ∩ C,
then that relation is B-local and thus B-coherent. This contradicts condition (ii)
of Definition 6.3, so {bi : i ∈ I} ∩ C is linearly independent. Thus {bi : i ∈ I} ∩ C
is a basis for its R-linear span, which equals the R-linear span of C by condition
(i) in Definition 6.3.
a is an R-linear combination (cid:80)
relation a−(cid:80)
22
NATHAN READING
On the other hand, suppose (bi : i ∈ I) is an R-basis for B whose restriction
to each B-cone is a basis (in the usual sense) for the span of that B-cone. Then
(cid:3)
condition (i) of Definition 6.3 holds.
Remark 6.5. Suppose B is a universal extended exchange matrix over R whose
coefficient rows constitute a cone R-basis. Then coefficient specializations from B
can be found more directly than in the case where we don't necessarily have a cone
basis. (See Remark 4.5.) Each ak is in some B-cone C. Knowing that we have a
cone basis, we can choose the elements (pik : i ∈ I) by taking pik = 0 for bi (cid:54)∈ C and
i∈S pikbi is a linear relation
choosing the remaining pik (uniquely) so that ak −(cid:80)
in the usual sense.
For every exchange matrix B for which the author has constructed an R-basis,
the R-basis is in fact a cone R-basis. The following question thus arises.
Question 6.6. If (bi : i ∈ I) is an R-basis for B is it necessarily a cone R-basis?
We now relate the notion of a cone R-basis to the notion of a positive R-basis.
Proposition 6.7. Given a collection (bi : i ∈ I) of vectors in Rn, the following
conditions are equivalent.
(i) (bi : i ∈ I) is a positive R-basis for B.
(ii) (bi : i ∈ I) is a positive cone R-basis for B.
(iii) (bi : i ∈ I) is an R-independent set for B with the following property: If C
is a B-cone, then the nonnegative R-linear span of {bi : i ∈ I} ∩ C contains
Rn ∩ C.
Since (bi
relation −a +(cid:80)
Proof. Condition (ii) implies condition (i) trivially. Conversely, suppose (i) holds,
let C be a B-cone and let a ∈ Rn ∩ C. To show that (ii) holds, we need to show
that the R-linear span of {bi : i ∈ I} ∩ C contains a.
: i ∈ I) is a positive R-basis for B, there is a B-coherent linear
i∈S cibi with the ci positive. Proposition 5.29 says that there is
a unique smallest B-cone D containing a, and that D is the closure of the B-
class of a. We claim that {bi : i ∈ S} is contained in D. Otherwise, for some
k (bi)) (cid:54)(cid:22)
i ∈ S, Proposition 5.23 says that there is a sequence k such that sgn(ηB
k (a)). That is, there is some index j ∈ [n] such that the jth coordinate
sgn(ηB
of sgn(ηB
k (bi)) is nonzero and different from the jth coordinate of sgn(ηB
k (a)).
Possibly replacing k by jk, we can assume that the jth coordinate of sgn(ηB
k (bi))
is negative and the jth coordinate of sgn(ηB
k (a)) is nonnegative. Thus we write S
as a disjoint union S1 ∪ S2, with S1 nonempty, such that ηB
k (bi) has negative jth
k (bi) has nonnegative jth coordinate for i ∈ S2,
coordinate for i ∈ S1, such that ηB
i∈S cibi is B-coherent,
k (bi) is zero.
i∈S1
But since the ci are all positive, we have reached a contradiction, thus proving
i∈S cibi in
particular writes a as a positive linear combination of elements of D. Since D ⊆ C,
we have established (ii). Since the relation is positive, we have also shown that (i)
implies (iii).
If (iii) holds, then for any a in Rn, there exists a finite subset S ⊆ I and positive
i∈S cibi is a B-local linear relation.
Proposition 5.9 implies that (bi : i ∈ I) is a positive R-basis for B, so (i) holds. (cid:3)
and such that a has nonnegative jth coordinate. Since −a+(cid:80)
we appeal to (4.4) to conclude that the jth coordinate of (cid:80)
the claim. The claim shows that the B-coherent linear relation −a +(cid:80)
elements (ci : i ∈ S) of R such that a −(cid:80)
ciηB
UNIVERSAL GEOMETRIC CLUSTER ALGEBRAS
23
Remark 6.8. A positive basis is necessarily a cone basis, by Proposition 6.7, so ex-
plicit coefficient specializations from the corresponding universal extended exchange
matrix are found as described in Remark 6.5. In Section 9, there are examples of
exchange matrices B having a cone R-basis but not a positive R-basis for R = Z
or R = Q.
The existence of a positive R-basis has implications for the structure of the
mutation fan. We describe these implications by constructing another fan closely
related to FB.
Definition 6.9 (R-part of a fan). Suppose F is a fan and R is an underlying ring.
Suppose F(cid:48) is a fan satisfying the following conditions:
(i) Each cone in F(cid:48) is the nonnegative R-linear span of finitely many vectors
(ii) Each cone in F(cid:48) is contained in a cone of F.
(iii) For each cone C of F, there is a unique largest cone (under containment)
in Rn. (For example, if R = Z or Q, then these are rational cones.)
among cones of F(cid:48) contained in C. This largest cone contains Rn ∩ C.
Then F(cid:48) is called the R-part of F. The R-part of F might not exist. For example,
the fans discussed later in Proposition 9.9 have no Q-part. However, if the R-part
of F exists, then it is unique: For each cone C of F, condition (iii) implies that
the largest cone of F(cid:48) contained in C is the nonnegative R-linear span of Rn ∩ C.
Condition (iii) implies that every cone in F(cid:48) is a face of one of these largest cones.
Proposition 6.10. Suppose F is a fan, R is an underlying ring, and F(cid:48) is the
R-part of F. Every cone of F spanned by vectors in Rn is a cone in F(cid:48). The
full-dimensional cones in F are exactly the full-dimensional cones in F(cid:48).
Proof. Suppose C is a cone of F spanned by vectors in Rn. Condition (iii) of
Definition 6.9 says that there is a cone D of F(cid:48) contained in C and containing
Rn ∩ C. But then D contains the vectors spanning C, so C = D.
Suppose C is a full-dimensional cone of F. Since Qn is dense in Rn, if C is strictly
larger than the largest cone D of F(cid:48) contained in C, then there are rational vectors
(and thus integer vectors and thus vectors in Rn) in C \ D. This is a contradiction
to condition (iii) of Definition 6.9. Thus C equals the cone D of F(cid:48).
Conversely, suppose D is a full-dimensional cone of F(cid:48). Then condition (ii) of
Definition 6.9 says that D is contained in some full-dimensional cone C of F. Since
F(cid:48) is a fan and D is full dimensional, D is the largest cone of F(cid:48) contained in C,
and as argued above, D = C. Thus D is a cone of F.
(cid:3)
Now suppose a positive R-basis (bi : i ∈ I) exists for B. Let F R
B be the collection
of all cones spanned by sets {bi : i ∈ I} ∩ C, where C ranges over all B-cones,
together with all faces of such cones. In light of Proposition 6.2, the collection F R
does not depend on the choice of positive R-basis.
Proposition 6.11. If a positive R-basis (bi : i ∈ I) exists for B, then F R
simplicial fan and is the R-part of FB.
Proof. If some cone C in F R
B is not simplicial, then C is generated by a set U of
vectors in Rn that is linearly dependent over R. We conclude that U is also linearly
dependent over R. (Suppose U is a set of vectors in Rn that is linearly independent
over R. Write a matrix whose rows are U . Use row operations over R to put the
matrix into echelon form. Since there are no nontrivial R-linear relations on the
B is a
B
24
NATHAN READING
Now suppose C1 and C2 are maximal cones in F R
rows, the echelon form has full rank, so U is linearly independent over R.) The
relation expressing this linear dependence is B-local by the definition of F R
B , so by
Proposition 5.9 it is a B-coherent linear relation over R among the basis vectors.
B , spanned respectively by
subsets U1 and U2 of the positive basis. Since C1 and C2 are spanned by vectors in
Rn, if R is a field, then C1 and C2 are each intersections of halfspaces with normal
vectors in Rn. Thus C1 ∩ C2 is an intersection of halfspaces with normal vectors
in Rn, and thus is spanned by vectors in Rn. In particular, there exists a vector
x ∈ Rn contained in the relative interior of C1 ∩ C2. The vector x can be expressed
both as a nonnegative R-linear combination of U1 and as a nonnegative R-linear
combination of U2. Both of these expressions are B-local and thus B-coherent by
Proposition 5.9, so their difference is a B-coherent R-linear relation. But since
U1 and U2 are part of an R-basis (and in particular an R-independent set) for B,
the two expressions must coincide, so that each writes x as a nonnegative R-linear
combination of U1∩ U2. We conclude that C1∩ C2 is contained in the cone spanned
by U1 ∩ U2, and the opposite containment is immediate. Thus, since C1 and C2 are
simplicial, C1 ∩ C2 is a face of both. If instead R = Z, then the usual arguments
by clearing denominators show that (bi : i ∈ I) is also a positive Q-basis for B, so
that FQ
B is a simplicial fan.
B = FZ
B and thus FZ
We now verify the conditions of Definition 6.9. Conditions (i) and (ii) hold by
construction. The first part of condition (iii) holds by construction and the second
part follows from the implication (i) =⇒ (iii) in Proposition 6.7.
(cid:3)
The following corollary is immediate by Propositions 6.7 and 6.11.
Corollary 6.12. If a positive Z-basis exists for B, then the unique positive Z-basis
for B consists of the smallest nonzero integer vector in each ray of the Z-part of
FB. If R is a field and a positive R-basis exists for B, then a collection of vectors
is a positive R-basis for B if and only if it consists of exactly one nonzero vector
in each ray of the R-part of FB.
Propositions 6.10 and 6.11 imply that FR
B = FB. Thus we have the following
corollary, which we emphasize is specific to the case R = R.
Corollary 6.13. If a positive R-basis for B exists, then FB is simplicial. The basis
consists of exactly one vector in each ray of FB.
For any exchange matrix B, Proposition 5.9 implies that a collection consisting
of exactly one nonzero vector in each ray of FB is an R-spanning set for B and
thus contains an R-basis for B by Proposition 4.6.
In particular, the following
proposition holds.
Proposition 6.14. A collection consisting of exactly one nonzero vector in each
ray of FB is a positive R-basis for B if and only if it is an R-independent set for B.
In this case, FB is simplicial by Corollary 6.13. On the other hand, if FB is
simplicial, then a collection consisting of exactly one nonzero vector in each ray
of FB could conceivably fail to be an R-independent set for B, in which case no
positive R-basis exists for B. However, we know of no exchange matrix B for which
this happens.
Example 6.15. In Sections 9 and 10 and in [21, 22], we encounter examples of
mutation fans that are simplicial. There also appear to exist exchange matrices B
UNIVERSAL GEOMETRIC CLUSTER ALGEBRAS
such that FB is not simplicial, and one such example appears to be B =
.
This "appearance" is based on computing approximations to FB as explained in
the paragraph after Proposition 5.30. See Figure 1. It appears that as m → ∞,
the quadrilateral region at the middle of the picture near the top will decrease
in size but neither disappear nor lose its quadrilateral shape. (Compare with the
(cid:3), depicted in Figure 4.) If FB is indeed not simplicial, then
example B = (cid:2) 0 2−3 0
2
0
0 −3
Corollary 6.13 says that no positive R-basis exists for B.
0
2
0
(cid:104) 0 −3
25
(cid:105)
7. Properties of the mutation fan
In this section, we prove some properties of mutation fans that are useful in
constructing FB is some cases. We begin by pointing out several symmetries.
It is apparent from Definition 4.1 that
k (−a)
k (a) = −η−B
ηB
for any sequence k.
(7.1)
Since also the antipodal map a (cid:55)→ −a commutes with the map sgn, we see that
a1 ≡B a2 if and only if (−a1) ≡−B (−a2). Thus we have the following proposition,
in which −FB denotes the collection of cones −C = {−a : a ∈ C} such that C is a
cone in FB.
Proposition 7.1. F−B = −FB.
Given any permutation π of [n], let π(B) be the exchange matrix whose ij-entry
is bπ(i)π(j), where the entries of B are bij. Then µπ(k)(πB) = π(µk(B)). Let π
also denote the linear map sending eπ(i) to ei. Then ηπB
k . Thus
we have the following proposition, in which πFB denotes the collection of cones
πC = {πa : a ∈ C} such that C is a cone in FB.
Proposition 7.2. FπB = πFB.
π(k) ◦ π = π ◦ ηB
Proposition 5.5 implies the following proposition, in which ηB
k C =(cid:8)ηB
k (a) : a ∈ C(cid:9) such that C is a cone in FB.
k FB denotes the
collection of cones ηB
Proposition 7.3. Fµk(B) = ηB
k FB.
A less obvious symmetry is a relationship called rescaling.
σi
times the ij-entry of B.
Every matrix is a rescaling of itself, taking Σ = cI for some positive c.
Definition 7.4 (Rescaling of exchange matrices). Let B and B(cid:48) be exchange matri-
ces. Then B(cid:48) is a rescaling of B if there exists a diagonal matrix Σ with positive
entries such that B(cid:48) = Σ−1BΣ.
In this case, if Σ = diag(σ1, . . . , σn), then the
ij-entry of B(cid:48) is σj
If
B(cid:48) = Σ−1BΣ for some Σ not of the form cI, then B(cid:48) is a nontrivial rescaling
of B. A given exchange matrix may or may not admit any nontrivial rescalings.
Proposition 7.5. If B and B(cid:48) are exchange matrices, then B(cid:48) is a rescaling of B
ji for all i, j ∈ [n].
if and only if sgn(bij) = sgn(b(cid:48)
ijb(cid:48)
Proof. If Σ is a diagonal matrix such that B(cid:48) = Σ−1BΣ, then b(cid:48)
ji = σj
bji =
σi
ji for all i, j ∈ [n].
ij) and bijbji = b(cid:48)
bijbji. Conversely, suppose sgn(bij) = sgn(b(cid:48)
ijb(cid:48)
Let d1, . . . dn be the skew-symmetrizing constants for B, so that dibij = −djbji. Let
d−1
D be the diagonal matrix diag(
n ), and define sB = D−1BD. The
ij) and bijbji = b(cid:48)
d−1, . . . ,
(cid:112)
ijb(cid:48)
√
σi
σj
bij
ij-entry of sB is
(cid:113) di
dj
bij. Since the di are skew-symmetrizing constants, we calculate
26
bij =
dj
didj
NATHAN READING
(cid:113) di
(cid:113) 1
ij-entry of sB is sgn(bij)(cid:112)−bijbji. Define matrices D(cid:48) and ĎB(cid:48) similarly in terms of
of sB is sgn(b(cid:48)
(cid:113) dj
(cid:113) 1
ji. Thus ĎB(cid:48) = sB, so B(cid:48) = D(cid:48)D−1BD(D(cid:48))−1.
the skew-symmetrizing constants for B(cid:48). Arguing similarly, we see that the ij-entry
(cid:3)
(cid:113)−b(cid:48)
(cid:113) di
(cid:113) dj
bji = bijbji, the
djbji. Since
djbji =
dibij =
ijb(cid:48)
bij
dj
ij)
didj
di
di
Proposition 7.5 relies on the assumption that B and B(cid:48) are exchange matrices in
the sense of Definition 2.3. That is, they are skew-symmetrizable integer matrices.
Proposition 7.6. If B and B(cid:48) are exchange matrices such that B(cid:48) is a rescaling
of B, then the diagonal matrix Σ with B(cid:48) = Σ−1BΣ can be taken to have integer
entries.
Proof. Fixing B and B(cid:48), the matrix Σ = diag(σ1, . . . , σn) satisfies B(cid:48) = Σ−1BΣ if
and only if σib(cid:48)
ij = bijσj for all i and j. If these equations can be solved for the σi,
(cid:3)
then there is a rational solution and therefore an integer solution.
i
i
i α∨
i α∨
Σ−1AΣ is the matrix expressing K in the basis(cid:8)σ−1
simple roots (cid:8)σ−1
: i ∈ [n](cid:9) (on the left) and
: i ∈ [n](cid:9) and simple co-roots {σiαi : i ∈ [n]}. In this situa-
Remark 7.7. The definition of rescaling is motivated by root system considera-
tions. The Cartan companion A of B (see Definition 10.2) defines a root system,
and in particular simple roots and simple co-roots, as well as a symmetric bilin-
ear form K. The matrix A expresses K in terms of the simple co-root basis (on
the left) and the simple root basis (on the right). If Σ = diag(σ1, . . . , σn), then
{σiαi : i ∈ [n]} (on the right).
In other words, Σ−1AΣ is a Cartan matrix with
tion, every root in the root system for the Cartan matrix Σ−1AΣ is a scaling of a
corresponding root for A, and vice versa.
Proposition 7.8. Suppose B(cid:48) is a rescaling of B, specifically with B(cid:48) = Σ−1BΣ.
(1) µk(B(cid:48)) is a rescaling of µk(B) for any sequence k of indices in [n]. Specif-
(2) If a ∈ Rn is a row vector, then ηB(cid:48)
k (a)Σ for any sequence k of
(3) The mutation fan FB(cid:48) is the collection of all cones CΣ = {aΣ : a ∈ C},
ically, µk(B(cid:48)) = Σ−1µk(B)Σ.
k (aΣ) = ηB
indices in [n].
where C ranges over cones in FB.
(4) The expression (cid:80)
(cid:80)
i∈S ci(viΣ) is a B(cid:48)-coherent linear relation.
entries, then (biΣ : i ∈ I) is a Z-independent set for B(cid:48).
i∈S civi is a B-coherent linear relation if and only if
(5) If (bi : i ∈ I) is a Z-independent set for B and Σ is taken to have integer
: i ∈ I) is an R-
independent set, R-spanning set, R-basis, cone R-basis and/or positive R-
basis for B if and only if (biΣ : i ∈ I) is the same for B(cid:48).
(6) Suppose the underlying ring R is a field. Then (bi
Proof. For (1) and (2), it is enough to replace the sequence k with a single index
k ∈ [n]. For (1), we use (2.2) to verify that the ij-entry of µk(B(cid:48)) is σj
times the
ij-entry of µk(B). For (2), we use (4.1) to verify that the jth entry of ηB(cid:48)
k (aΣ) is
k (a). To prove (3), we show that the map a (cid:55)→ aΣ takes
σj times the jth entry of ηB
B-classes to B(cid:48) classes. Indeed, by (2), we have sgn(ηB(cid:48)
k (a1Σ)) = sgn(ηB(cid:48)
k (a2Σ)) if
and only if sgn(ηB
k (a2)Σ), but since multiplication by Σ preserves
sgn, this is if and only if sgn(ηB
k (a1)Σ) = sgn(ηB
k (a1)) = sgn(ηB
k (a2)).
σi
i∈S ciηB(cid:48)
i∈S ciηB
i∈S cimin(ηB
27
UNIVERSAL GEOMETRIC CLUSTER ALGEBRAS
For any k, (2) says that the sum (cid:80)
k (viΣ), 0) equals(cid:0)(cid:80)
and that the sum(cid:80)
third matrix in (9.1), with exchange matrix B = (cid:2) 0 1−2 0
(cid:3)
conclude that (4) holds. Now (5) and (6) follow.
Remark 7.9. If R is not a field (that is, if R = Z), then the conclusion of Propo-
sition 7.8(6) can fail. For an example, we look ahead to Section 9. Let B be the
k (vi)(cid:1) Σ
k (vi), 0)(cid:1) Σ. We
(cid:3). Then B(cid:48) = (cid:2) 0 2−1 0
(cid:3) is
k (viΣ) equals (cid:0)(cid:80)
i∈S cimin(ηB(cid:48)
a rescaling of B for Σ any matrix of the form [ a 0
0 2a ] for positive integer a. The
coefficient rows bi of B are a positive Z-basis for B, but there is no choice of a such
that the vectors biΣ are a Z-basis for B(cid:48). There is, however, a positive Z-basis for
B(cid:48) whose elements are positive rational multiples of the vectors biΣ.
The final proposition of this section concerns limits of B-cones.
Definition 7.10 (Limits of rays and of cones). Given a sequence (ρm : m =
1, 2, . . .) of rays in Rn, we say that the sequence converges if the sequence (vm :
m = 1, 2, . . .) consisting of a unit vector vm in each ray ρm converges in the usual
topology on Rn. The limit of the sequence is the ray spanned by the limit of the
vectors vm. A sequence (Cm : m = 1, 2, . . .) of closed cones in Rn converges
if, for some fixed p, each cone Cm is the nonnegative linear span of some rays
{ρm;i : i = 1, 2, . . . , p} and the sequence (ρm;i : m = 1, 2, . . .) converges for each
i ∈ [p]. The limit of the sequence is the nonnegative linear span of the limit rays.
Proposition 7.11. Given a sequence of B-cones that converges in the sense of
Definition 7.10, the limit of the sequence is contained in a B-cone.
Proof. Suppose (Cm : m = 1, 2, . . .) is a sequence of cones, each the nonnegative
linear span of rays {ρm;i : i = 1, 2, . . . , p} such that (ρm;i : m = 1, 2, . . .) converges
for each i ∈ [p]. Write C for the limiting cone. For each m ≥ 1 and i ∈ [p],
let vm;i be the unit vector in ρm;i, and write vi for the limit of the sequence
(vm;i : m = 1, 2, . . .) for each i.
If ηB
k (vi) and ηB
Proposition 5.30 because vm;i an vm;j are both in Cm, so(cid:8)ηB
k (vj) have strictly opposite signs in some coordinate, for some
sequence k, then by the continuity of ηB
k (vm;i)
and ηB
k (vm;j) have strictly opposite signs in that coordinate. This contradicts
is sign-coherent for each k. Proposition 5.30 says that the set {vi : i = 1, 2, . . . , p}
is contained in some B-cone. Now Proposition 5.4 implies that the set C of nonneg-
ative linear combinations of {vi : i = 1, 2, . . . , p} is contained in that B-cone.
(cid:3)
k (vi) : i = 1, 2, . . . , p(cid:9)
k , for large enough m, the vectors ηB
8. g-Vectors and the mutation fan
In this section, we show that the mutation fan FB contains an embedded copy
of a fan defined by g-vectors for BT . The result is conditional on the following con-
jecture, which shown in Proposition 8.9 to be equivalent to a conjecture from [12].
Conjecture 8.1. The nonnegative orthant (R≥0)n is a B-cone.
At the time of writing, the conjecture is known for some exchange matrices B
and not for others (Remark 8.14), so we need to be precise about what we require.
Definition 8.2 (Standard Hypotheses on B). The Standard Hypotheses on B
are that Conjecture 8.1 holds for every exchange matrix in the mutation class of B
and for every exchange matrix in the mutation class of −B. We use the symbol O
28
NATHAN READING
for the nonnegative orthant (R≥0)n.
In light of Proposition 7.1, we restate the
Standard Hypotheses as the requirement that both O and the nonpositive orthant
−O are B(cid:48)-cones for every exchange matrix B(cid:48) in the mutation class of B.
The Standard Hypotheses allow us to relate the mutation fan for B to the g-
vectors associated to cluster variables in a cluster algebra with principal coefficients.
Definition 8.3 (Principal coefficients). Given an exchange matrix B, consider the
2n × n extended exchange matrix B whose top n rows are B and whose bottom
n rows are the n × n identity matrix. A cluster pattern or Y -pattern with B in
its initial seed is said to have principal coefficients at the initial seed. We write
xB;t0
for the cluster indexed by t in the cluster pattern with exchange matrix B
t
and principal coefficients at the vertex t0 of Tn. The notation xB;t0
i;t denotes the ith
cluster variable in the cluster xB;t0
.
t
The g-vectors are most naturally defined as a Zn-grading on the cluster algebra.
(See [12, Section 6].) For convenience, however, we take as the definition a recursion
(in fact two recursions) on g-vectors established in [12, Proposition 6.6].
Definition 8.4 (g-Vectors). We define a g-vector gB;t0
for each cluster variable
xB;t0
. The g-vector associated to the cluster variable xB;t0
in the initial seed is the
standard unit basis vector e(cid:96) ∈ Rn. The remaining g-vectors are defined by the
i;t
(cid:96),t0
following recurrence relation, in which we have suppressed the superscripts B; t0.
i;t
(cid:40)
−gk,t +(cid:80)n
g(cid:96),t
ik]+gi;t −(cid:80)n
i=1[bt
i=1[bt
n+i,k]+col(i, B)
if (cid:96) (cid:54)= k,
if (cid:96) = k.
(8.1)
g(cid:96),t(cid:48) =
k
-- -- t(cid:48) in Tn. The entries bt refer to the Y -pattern of geometric
for each edge t
type with exchange matrix B at t0 and principal coefficients. The notation col(j, B)
refers to the jth column of the initial exchange matrix B.
The g-vectors are also defined by the recurrence relation
−gk,t +(cid:80)n
g(cid:96),t
i=1[−bt
ik]+gi;t −(cid:80)n
i=1[−bt
n+i,k]+col(i, B)
if (cid:96) (cid:54)= k,
if (cid:96) = k.
(8.2)
g(cid:96),t(cid:48) =
(cid:40)
The equivalence of (8.1) and (8.2) is not obvious in this context, but the non-
recursive definition given in [12, Section 6] validates the equivalence and the well-
definition of the recursive definitions. See the discussion in the proof of [12, Propo-
sition 6.6].
t
: i ∈ [n]
for the g-vector cone associated to xB;t0
, the associated g-vector
gB;t0
. We write
i;t
Definition 8.5 (g-Vector cone). For each cluster xB;t0
t
cone is the nonnegative linear span of the vectors
ConeB;t0
Definition 8.6 (Transitive adjacency and the subfan F◦
B). Two full-dimensional
cones are adjacent if they have disjoint interiors but share a face of codimension 1.
Two full-dimensional cones C and D in a fan F are transitively adjacent in F
if there is a sequence C = C0, C1, . . . , Ck = D (possibly with k = 0) of full-
dimensional cones in F such that Ci−1 and Ci are adjacent for each i ∈ [k]. If the
nonnegative orthant O is a B-cone, then the full-dimensional cones in FB that are
transitively adjacent to O in FB are the maximal cones of a subfan F◦
B of FB.
.
t
(cid:110)
(cid:111)
UNIVERSAL GEOMETRIC CLUSTER ALGEBRAS
29
The following is the main result of this section.
Theorem 8.7. Assume the Standard Hypotheses on B. Then the subfan F◦
FB is the set of g-vector cones
, together with their faces.
: t ∈ Tn
ConeBT ;t0
B of
t
(cid:110)
(cid:111)
Before proving the theorem, we discuss the Standard Hypotheses further. The
following assertion appears in the proof of [12, Proposition 5.6] as condition (ii'),
which is shown there to be equivalent to [12, Conjecture 5.4].
Conjecture 8.8. For each extended exchange matrix in a Y -pattern with initial
exchange matrix B and principal coefficients, rows n + 1 through 2n are a set of
sign-coherent vectors.
Many cases of Conjecture 8.8 are known. See Remark 8.14.
Proposition 8.9. Conjecture 8.8 holds for a given B if and only if Conjecture 8.1
holds for B.
Proof. Conjecture 8.8 is equivalent to the following assertion: For each sequence k,
the set {ηk(ei) : i ∈ [n]} is sign-coherent. Proposition 5.30 says that the latter
assertion is equivalent to the assertion that {ei : i ∈ [n]} is contained in some B-
cone. Any set O(cid:48) strictly larger than O contains a vector with a strictly negative
entry and also a vector with all entries positive. Thus Proposition 5.30 (with k
the empty sequence) implies that O(cid:48) is not a B-cone. Since a B-cone containing
{ei : i ∈ [n]} is no larger than O, the set {ei : i ∈ [n]} is contained in some B-cone
(cid:3)
if and only if O is a B-cone.
A crucial consequence of the Standard Hypotheses is another conjecture, [12,
Conjecture 7.12]. We quote the following weak version.
-- -- t1 be an edge in Tn and let B0 and B1 be exchange
Conjecture 8.10. Let t0
matrices such that B1 = µk(B0). Then, for any t ∈ Tn and i ∈ [n], the g-vectors
i;t = (g(cid:48)
gB0;t0
i;t = (g1, . . . , gn) and gB1;t1
n) are related by
1, . . . , g(cid:48)
k
(8.3)
g(cid:48)
j =
gj + [bjk]+gk − bjk min(gk, 0)
if j = k;
if j (cid:54)= k,
where the quantities bjk are the entries of the matrix B0.
More to the point is a restatement of Conjecture 8.10 in the language of mutation
maps. First, we rewrite (8.3) as
(cid:40)−gk
(8.4)
g(cid:48)
j =
−gk
gj + bjkgk
gj − bjkgk
gj
if j = k;
if j (cid:54)= k, gk ≥ 0 and bjk ≥ 0;
if j (cid:54)= k, gk ≤ 0 and bjk ≤ 0;
otherwise.
Then, comparing with (4.1), we see that the following conjecture is equivalent to
Conjecture 8.10. (See [12, Remark 7.15] for a related restatement of [12, Conjec-
ture 7.12].)
k
-- -- t1 be an edge in Tn and let B0 and B1 be exchange
Conjecture 8.11. Let t0
matrices such that B1 = µk(B0). Then, for any t ∈ Tn and i ∈ [n], the g-vectors
gB0;t0
0 is the transpose
i;t
of B0.
are related by gB1;t1
), where BT
and gB1;t1
= ηBT
i;t
(gB0;t0
i;t
0
k
i;t
30
NATHAN READING
Nakanishi and Zelevinsky proved [17, Proposition 4.2(v)] that if Conjecture 8.8 is
true for all B, then [12, Conjecture 7.12] (the strong form of Conjecture 8.10) is true
for all B. Their argument also proves the following statement, with weaker hypothe-
ses and weaker conclusions, and with Conjecture 8.10 replaced by the equivalent
Conjecture 8.11.
Theorem 8.12. If the Standard Hypotheses hold for B, then Conjecture 8.11 holds
for all exchange matrices B0 and B1 mutation equivalent to B.
The recursive definition of g-vectors implies that if t and t(cid:48) are connected by an
are adjacent. Thus Theorem 8.7 is an
edge in Tn, then ConeBT ;t0
immediate corollary to the following proposition.
and ConeBT ;t0
t(cid:48)
t
Proposition 8.13. Assume the Standard Hypotheses on B. If t0, t1, . . . , tq is a
path in Tn, with the edge from ti−1 to ti labeled ki, then ConeBT ;t0
is the full-
(O), where O is the nonnegative orthant (R≥0)n.
dimensional B-cone η
tq
µkq ,...,k1 (B)
k1,...,kq
Cone
µk1(BT );t1
tq
. Recalling that η
µk1 (B)
k1
µk1(BT );t1
tq
Proof. We argue by induction on q.
In the course of the induction, we freely
replace B by elements of its mutation class. If q = 0 then the proposition follows
by the base of the inductive definition of g-vectors and by Conjecture 8.1 for B. If
q > 0, then take B0 = BT and B1 = µk1 (BT ). Matrix mutation commutes with
matrix transpose, so B1 = (µk1 (B))T . By Theorem 8.12, Conjecture 8.11 holds
for B0 and B1, so ηB
to the extreme rays of
k1
takes the extreme rays of ConeBT ;t0
tq
is the inverse of ηB
k1
, we see that η
µk1 (B)
k1
takes
the extreme rays of Cone
(because the path from t1 to tq has length q − 1), the g-vector cone Cone
equals η
(O), and this cone is a µk1 (B)-cone. Since Cone
to the extreme rays of ConeBT ;t0
µkq ,...,k2 (µk1 (B))
k2,...,kq
tq
. By induction
µk1(BT );t1
tq
µk1(BT );t1
tq
is
a µk1(B)-cone, the map η
µk1 (B)
k1
the entire cone Cone
extreme rays to extreme rays). Thus
µk1(BT );t1
tq
is linear on it by Proposition 5.3, so this map takes
to the cone ConeBT ;t0
(rather than only mapping
tq
ConeBT ;t0
tq
= η
µk1 (B)
k1
µkq ,...,k2 (µk1 (B))
η
k2,...,kq
(O).
Referring to (4.2), we see that
µkq ,...,k2 (µk1 (B))
η
k2,...,kq
(O) = η
µk2k1 (B))
k2
◦ η
µk3k2 k1 (B))
k3
◦ ··· ◦ η
µkq kq−1···k1 (B))
kq
(O),
so that
ConeBT ;t0
tq
= η
µk1 (B)
k1
◦ ··· ◦ η
µk2k1 (B))
k2
◦ η
(O) is a µk1(B)-cone, Proposition 5.5 says that ConeBT ;t0
µkq kq−1···k1 (B))
kq
µkq ,...,k1 (B)
k1,...,kq
(O) = η
(O).
tq
µkq ,...,k2 (µk1 (B))
k2,...,kq
Since η
a µk1 (µk1 (B))-cone, or in other words, a B-cone.
is
(cid:3)
Remark 8.14. Conjecture 8.8 (and thus Conjecture 8.1) is known in many cases, but
currently not in full generality. In particular, it was proved in [4] for skew-symmetric
exchange matrices.
(In particular, since skew-symmetry is
preserved under matrix mutation, Conjecture 8.11 is true whenever B0 and B1
are skew-symmetric.) Conjecture 8.8 is not specifically mentioned in [4], but [4,
(See also [16, 18].)
UNIVERSAL GEOMETRIC CLUSTER ALGEBRAS
31
Theorem 1.7] establishes [12, Conjecture 5.4], which is shown in the proof of [12,
Proposition 5.6] to be equivalent to Conjecture 8.8. In [2], the construction of [4] is
extended to some non-skew-symmetric exchange matrices. In particular, [2, Propo-
sition 11.1] establishes Conjecture 8.8 for a class of exchange matrices including
all matrices of the form DS, where D is an integer diagonal matrix and S is an
integer skew-symmetric matrix. (This is a strictly smaller class than the class of
all skew-symmetrizable matrices, which are the integer matrices of the form D−1S
where D is an integer diagonal matrix and S is a skew-symmetric matrix.) In a
personal communication to the author [3], Demonet indicates that his results can be
extended to prove Conjecture 8.8 for all exchange matrices that are mutation equiv-
alent to an acyclic exchange matrix. See also [2, Remark 7.2] and the beginning of
[2, Section 11].
As a corollary to Theorem 8.7, we see that any positive R-basis for B must
involve the g-vectors for BT .
Corollary 8.15. Assuming the Standard Hypotheses on B, if a positive R-basis
exists for B, then the positive basis includes a positive scaling of the g-vector as-
sociated to each cluster variable for BT . If R = Z, then the positive basis includes
the g-vector associated to each cluster variable for BT .
To prove Corollary 8.15 (in particular the second statement), we consider the
following conjecture, which is [12, Conjecture 7.10(2)].
Conjecture 8.16. For each t ∈ Tn, the vectors (gB;t0
for Zn.
i;t
: i ∈ [n]) are a Z-basis
In [17, Proposition 4.2(iv)], it is shown that if Conjecture 8.8 is true for all B,
then Conjecture 8.16 is true for all B. Again, the argument from [17] also proves a
statement with weaker hypothesis and weaker conclusion:
Theorem 8.17. If Conjecture 8.8 is true for some B, then Conjecture 8.16 is also
true for B.
Proof of Corollary 8.15. Corollary 6.12 says that the positive basis consists of one
nonzero vector in each ray in the R-part of FB. Thus by Proposition 6.10, the
positive basis contains one nonzero vector in each ray of F◦
B. Theorem 8.7 implies
that the basis contains a positive scalar multiple of the g-vector of each cluster
variable for BT . If R = Z, then in particular each vector in the positive basis is
the shortest integer vector in the ray it spans. By Theorem 8.17, the same is true
(cid:3)
of each g-vector for BT .
We now present some results about rescalings of exchange matrices as they relate
to the Standard Hypotheses and to g-vectors. The first result is immediate from
Proposition 7.8(1) and (3).
Proposition 8.18. Suppose B(cid:48) is a rescaling of B. The Standard Hypotheses hold
for B(cid:48) if and only if they hold for B.
Proposition 8.19. The Standard Hypotheses for B, for −B, for BT , and for −BT
are all equivalent.
Proof. The Standard Hypotheses for B and −B are syntactically equivalent. Propo-
sition 7.5 implies that −BT is a rescaling of B. Proposition 8.18 completes the
(cid:3)
argument.
32
NATHAN READING
Proposition 8.20. Suppose B(cid:48) is a rescaling of B, specifically with B(cid:48) = Σ−1BΣ.
If the Standard Hypotheses hold for B, then gB(cid:48);t0
is the smallest nonzero integer
vector in the ray spanned by gB;t0
i;t
i,t Σ.
Proof. Using Propositions 8.18 and 8.19 to obtain the Standard Hypotheses for
all the relevant matrices, we appeal to Propositions 8.13 and 7.8(3) to conclude
that gB(cid:48);t0
is the
(cid:3)
smallest nonzero integer vector in that ray.
i,t Σ span the same ray. Theorem 8.17 implies that gB(cid:48);t0
and gB;t0
i;t
i;t
We next discuss two conjectures suggested by Corollary 8.15. The first is a
separation property for cluster variables.
Conjecture 8.21. Suppose xB;t0
i;t(cid:48)
algebra with principal coefficients. Then the following are equivalent.
j;t(cid:48)(cid:48) are cluster variables in the cluster
and xB;t0
i;t(cid:48) and xB;t0
(i) xB;t0
(ii) There exists t ∈ Tn and k ∈ [n] such that gBt;t
j;t(cid:48)(cid:48) are not contained in any common cluster.
i;t(cid:48) and gBt;t
j,t(cid:48)(cid:48) have strictly opposite
signs in their kth entry.
Proposition 8.22. Assuming the Standard Hypotheses on B, condition (ii) of
Conjecture 8.21 implies condition (i).
Proof. By Proposition 5.30 and Theorem 8.12, condition (ii) implies that gB;t0
i;t(cid:48)
and gB;t0
j;t(cid:48)(cid:48) are not contained in any common BT -cone. By Proposition 8.19, the
Standard Hypotheses hold for BT , so all of the g-vector cones for B are BT -cones
by Theorem 8.7. Thus gBt;t
j,t(cid:48)(cid:48) are not contained in any common g-vector
(cid:3)
cone, and condition (i) follows.
i;t(cid:48) and gBt;t
We cannot reverse the argument for Proposition 8.22 because conceivably the
two g-vectors are contained in some B-cone that is not in F◦
B.
The second conjecture suggested by Theorem 8.7 is an independence property
of g-vectors.
Conjecture 8.23. Suppose t1, . . . , tm are vertices of Tn, suppose ii, . . . , im are
= 0 for all
indices in [n], and suppose c1, . . . , cm are integers. If (cid:80)m
i=1 cmgBt;t
im;tm
vertices t of Tn, then cj = 0 for all j = 1, . . . , m.
Proposition 8.24. Assuming the Standard Hypotheses on B, if there exists an
underlying ring R such that a positive R-basis exists for BT , then Conjecture 8.23
holds for B.
Proof. Given the Standard Hypotheses on B, Proposition 8.19 and Theorem 8.12
allow us to restate Conjecture 8.23: There is no nontrivial BT -coherent linear
relation over Z among the g-vectors for cluster variables associated to B.
Suppose there exists an underlying ring R such that a positive R-basis exists
for BT .
In light of Proposition 8.19, Corollary 8.15 says that the positive basis
contains a positive scalar multiple of the g-vector of each cluster variable for B.
In particular, a BT -coherent linear relation over Z among these g-vectors can be
rewritten as a BT -coherent linear relation among basis elements. Thus any BT -
(cid:3)
coherent linear relation over Z among these g-vectors is trivial.
UNIVERSAL GEOMETRIC CLUSTER ALGEBRAS
33
9. The rank-2 case
In this section, we construct R-bases for exchange matrices of rank 2. We prove
the following theorem and give explicit descriptions of the additional vectors men-
tioned. The details are given in Propositions 9.4, 9.8 and 9.9.
Theorem 9.1. Let B be an exchange matrix of rank 2. Then an R-basis for B can
be constructed by taking the g-vectors associated to BT and possibly adding one or
two additional vectors.
The exchange matrices of rank 2 are the matrices B of the form [ 0 a
b 0 ] where
a and b are integers with sgn(b) = − sgn(a).
(This is skew-symmetrizable with
d1 = b and d2 = a.) We label the vertices of the infinite 2-regular tree T2 as
. . . , t−1, t0, t1, . . . with ti adjacent to ti−1 by an edge labeled k ∈ {1, 2} with k ≡ i
mod 2. The Standard Hypotheses (Definition 8.2) are known to hold for exchange
matrices of rank 2. (For example, one may apply [2, Proposition 11.1] as explained
in Remark 8.14.) Since B is of rank 2, the fan FB is a fan in R2, and since no
B-cone contains a line, FB is therefore simplicial.
The following proposition accomplishes most of the work towards proving The-
orem 9.1. We again write R(B) for the set of rays of the mutation fan FB. We
write R◦(B) for the subset of R(B) consisting of rays of F◦
B. (For the definition of
F◦
B, see the paragraph before Theorem 8.7.)
Proposition 9.2. Suppose B is of rank 2 and, for each ray ρ ∈ R(B), choose a
nonzero vector vρ contained in ρ. Then any B-coherent linear relation supported
on the set {vρ : ρ ∈ R(B)} is in fact supported on {vρ : ρ ∈ R(B) \ R◦(B)}.
Proof. Suppose ρ is in R◦(B), so that ρ is an extreme ray of a g-vector cone for BT
by Theorem 8.7. In fact, ρ is an extreme ray of exactly two g-vector cones for BT ,
for some q ∈ Z. We argue the case
and these cones are ConeBT ;t0
q ≥ 0; the other case is the same except for notation. Proposition 8.13 says that
ConeBT ;t0
Inverting
the maps η in these two expressions, we see that the nonnegative orthant O equals
ηB
kq−1,...,k1
(O) and that ConeBT ;t0
and ConeBT ;t0
tq−1
µkq−1,...,k1 (B)
k1,...,kq−1
µkq ,...,k1 (B)
k1,...,kq
ConeBT ;t0
tq−1
and also equals
tq−1 = η
(cid:16)
(cid:17)
(O).
= η
tq
tq
ηB
kq,...,k1
ConeBT ;t0
(cid:17)
By Proposition 7.3, the cones ηB
intersect in a ray of Fµkq−1,...,k1 (B), namely the ray ηB
µkq−1,...,k1 (B)
kq
ConeBT ;t0
tq−1
ηB
kq−1,...,k1
kq−1,...,k1
(cid:16)
= η
tq
ConeBT ;t0
tq
and ηB
kq−1,...,k1
kq−1,...,k1
(ρ) = ηB
kq,...,k1
(ρ).
µkq−1,...,k1 (B)
kq
kq,...,k1
kq−1,...,k1
(ρ) and ηB
But the map η
fixes one extreme ray of O and sends the other outside
(ρ) both equal R≥0ej, where j
of O. We conclude that ηB
is the unique element of {1, 2} \ {kq}. In rank 2, every single mutation operation
µk is simply negation of the matrix. Thus, either µkq−1,...,k1 (B) or µkq,...,k1 (B)
has a nonnegative ij-entry. If µkq−1,...,k1 (B) has a nonnegative ij-entry, then write
k = kq−1, . . . , k1 and t = tq−1 and t(cid:48) = tq. Otherwise, write k = kq, . . . , k1 and
t = tq and t(cid:48) = tq−1. Write i = kq. The edge connecting t to t(cid:48) is labeled t
-- -- t(cid:48).
= ej span the cone O = Coneµk(BT );t
.
The matrix µk(BT ) has a nonnegative ji-entry, so the recursion (8.2) says that
The g-vectors gµk(BT );t
= ei and gµk(BT );t
j;t
i;t
t
i
(cid:16)
(cid:17)
(cid:16)
(cid:17)
(cid:16)
(cid:17)
.
ConeBT ;t0
tq
34
NATHAN READING
t
= −ei and gµk(BT );t
j;t(cid:48)
= ej. Thus the cones Coneµk(BT );t
Now suppose(cid:80)
and Coneµk(BT );t
gµk(BT );t
i;t(cid:48)
cover the closed halfspace consisting of vectors with non-negative jth entry. These
are cones in the fan Fµk(B) by Theorem 8.7, so their common ray ηB
k (ρ) is the
unique ray of Fµk(B) that intersects the open halfspace consisting of vectors with
positive jth entry.
i∈S civi is a B-coherent linear relation with S a finite subset of
R(B). By Proposition 7.3, for each ρ ∈ S, the vector ηk(vρ) spans a ray of Fµk(B).
By the argument above, if ρ ∈ S ∩ R◦(B), then the jth entry of ηB
k (vρ) is strictly
positive and every vector vρ(cid:48) with ρ(cid:48) (cid:54)= ρ has non-positive jth entry. We conclude
(cid:3)
from Proposition 4.12 that cρ = 0.
t(cid:48)
To complete the proof of Theorem 9.1, we explicitly construct the mutation fan
for B and to prove that the remaining vectors {vρ : ρ ∈ R(B)} beyond the rays
in g-vector cones must also appear with coefficient zero in any B-coherent linear
relation. We consider several cases separately.
Definition 9.3 (Cluster algebra/exchange matrix of finite type). A cluster algebra
is of finite type if and only if its associate cluster pattern contains only finitely
many distinct seeds. Otherwise it is of infinite type.
The classification [9, 11] of cluster algebras of finite type says in particular that
b 0 ] is of finite type if and only if ab ≥ −3. The possibilities for (a, b) are
B = [ 0 a
(0, 0), (±1,∓1), (±1,∓2), (±2,∓1), (±1,∓3), and (±3,∓1).
Proposition 9.4. If B is a rank-2 exchange matrix of finite type, then for any R,
the g-vectors for BT constitute a positive R-basis for B.
Proof. In the case where B is of finite type, it is known that the g-vector cones are
the maximal cones of a complete fan. For example, in Section 10, we explain how
this fact (Theorem 10.6), for arbitrary rank and finite type, follows from results of
[23, 26, 30]. (See Remark 10.13.) This can also be verified directly in all rank-2
In particular, FB consists of the g-vector cones associated to
finite-type cases.
BT , and their faces. The collection of all g-vectors is a positive R-basis for B by
(cid:3)
Theorem 8.17 and Proposition 9.2.
One can calculate these bases explicitly in all cases. Some of the resulting uni-
versal extended exchange matrices are shown below in (9.1).
0
0
0
0
0
1
1
0
−1
0
0 −1
0
1
−1
0
1
0
0
1
−1
0
0 −1
1 −1
0
1
−2
0
1
0
0
1
−1
0
0 −1
1 −1
2 −1
0
1
−3
0
1
0
0
1
−1
0
0 −1
1 −1
3 −2
2 −1
3 −1
(9.1)
In light of Propositions 7.1 and 7.2, the remaining cases are obtained from the
cases shown by applying one or both of the following operations: (a) negating all
entries and/or (b) swapping the columns and then swapping the first two rows.
The mutation fans corresponding to the four cases in (9.1) are shown in Figure 2.
The standard unit basis vector e1 points to the right in the pictures and e2 points
upwards. Black lines indicate the rays.
UNIVERSAL GEOMETRIC CLUSTER ALGEBRAS
35
B =(cid:2) 0 1−1 0
(cid:3)
B =(cid:2) 0 1−2 0
(cid:3)
B =(cid:2) 0 1−3 0
(cid:3)
B = [ 0 0
0 0 ]
Figure 2. Mutation fans FB for the finite examples in (9.1)
exchange matrix B in (9.1) whose underlying exchange matrix is B =(cid:2) 0 1−2 0
Example 9.5. We now consider a detailed example based on the universal extended
write the indexing set I as {a, b, c, d, e, f}, so that the matrix is indexed as
(cid:3). We
(9.2)
1
2
a
b
c
d
e
f
1
0
2
1
−2
0
1
0
0
1
−1
0
0 −1
1 −1
2 −1
We continue to label the vertices of T2 as . . . , t−1, t0, t1, . . . with ti adjacent to ti−1
by an edge labeled k ∈ {1, 2} such that k ≡ i mod 2. In this case, the labeled seed
at ti depends only on i mod 6. The extended exchange matrices, coefficients, and
cluster variables in this cluster algebra are shown in Tables 1 and 2.
Now consider another extended exchange matrix B(cid:48) with the same underlying
(cid:3). This time, the indexing set I(cid:48) is {α, β, γ}, so that B(cid:48)
exchange matrix B =(cid:2) 0 1−2 0
is indexed as
(9.3)
1
2
α
β
γ
(cid:35)
(cid:34) 0
1
−2
2
1
0
3 −2
2
1
−1
1
The extended exchange matrices, coefficients, and cluster variables in this cluster
algebra are shown in Tables 3 and 4.
The coefficient rows of B, indexed by a, b, c, d, e, f , are vectors contained in the
rays of FB. We name these vectors va, vb, etc. Let vα, vβ, and vγ similarly
name the coefficient rows of B(cid:48). We find the unique coefficient specialization from
AR(x, B) to AR(x(cid:48), B(cid:48)) as described in Remarks 6.5 and 6.8. First, vα is in the
B-cone spanned by ve and vf , with vα = ve + vf . Similarly, vβ is in the B-cone
spanned by va and vb, with vβ = va + 2vb, and vγ is in the B-cone spanned by
vb and vc, with vγ = vb + vc. Accordingly, we obtain a coefficient specialization
mapping
x1 (cid:55)→ x(cid:48)
x2 (cid:55)→ x(cid:48)
1
2
ua (cid:55)→ uβ
ub (cid:55)→ u2
βuγ
uc (cid:55)→ uγ
ud (cid:55)→ 1
ue (cid:55)→ uα
uf (cid:55)→ uα.
We have dealt with the rank-2 exchange matrices of finite type. It remains to
consider the cases with ab ≤ −4. For m = 0, 1, 2, . . ., define
Pm = (−1)(cid:98)m/2(cid:99)(cid:88)
(cid:18)m − i
(cid:19)
i≥0
i
(ab)(cid:98)m/2(cid:99)−i.
(9.4)
(9.5)
36
NATHAN READING
0
t0
1
−2
0
1
0
0
1
−1
0
0 −1
1 −1
2 −1
uaueu2
f
uc
ub
udueuf
t
Bt
y1,t
y2,t
t1
0 −1
2
0
−1
1
0
1
1
0
0 −1
−1
0
−2
1
t2
−2
0
1
0
1 −1
2 −1
1
0
0
1
−1
0
0 −1
t3
0 −1
2
0
−1
0
−2
1
−1
1
0
1
1
0
0 −1
t4
−2
−1
0
1
0
0
0 −1
1 −1
2 −1
0
1
0
1
t5
0 −1
2
0
1
0
0 −1
−1
0
−2
1
−1
1
0
1
uc
u1ueu2
f
uaubuf
ud
b uc
uau2
ue
ud
uaubuf
ue
uau2
b uc
ubucud
uf
due
ucu2
ua
uf
ubucud
ua
ucu2
due
udueuf
ub
Table 1. Extended exchange matrices and coefficients for Example 9.5 (universal)
t
t0
t1
t2
t3
t4
t5
t
B(cid:48)
t
y1,t
x1;t
x1
x2
2uc+uaueu2
f
x1
x2
2uc+uaueu2
f
x1
x2
1uau2
b +2x1uaubudueuf +x2
2ucu2
due+uau2
du2
eu2
f
x1x2
2
x2
1uau2
b +2x1uaubudueuf +x2
2ucu2
due+uau2
du2
eu2
f
x1x2
2
x1
x2;t
x2
x2
x1uaubuf +x2
x1uaubuf +x2
f
2ucud+uaudueu2
x1x2
2ucud+uaudueu2
x1x2
f
x1ub+udueuf
x2
x1ub+udueuf
x2
Table 2. Cluster variables for Example 9.5 (universal)
(cid:35) (cid:34) 0 −1
t1
(cid:35) (cid:34) 0
(cid:35) (cid:34) 0 −1
t3
(cid:35) (cid:34) 0
(cid:35) (cid:34) 0 −1
t5
(cid:35)
(cid:34) 0
t0
−2
1
0
3 −2
2
1
−1
1
2
−3
−1
1
0
1
3
1
u3
αuβ
uγ
uγ
αuβ
u3
t2
1
−2
0
−1 −1
5 −3
3 −1
β u3
u5
uα
γ
t4
1
−2
0
1
1
−1 −2
1 −2
0
2
−1
2
1 −2
−1 −1
uαuγ
uβ
uβ
uαuγ
2
0
1 −1
−5
2
−3
2
u5
uα
β u3
γ
u2
β u2
uα
γ
u2
β uγ
u2
α
y2,t
Table 3. Extended exchange matrices and coefficients for Example 9.5
βuγ
u2
u2
uαu3
1
uαu3
β uγ
uα
β u2
γ
u2
α
β uγ
t
t0
t1
t2
t3
t4
t5
1
2
1
x(cid:48)
αuβ
αuβ
(x(cid:48)
(x(cid:48)
2)2uγ +u3
2)2uγ +u3
x1;t
x(cid:48)
x2;t
x(cid:48)
x(cid:48)
2
β uγ +(x(cid:48)
2)2uγ +u3
x(cid:48)
1x(cid:48)
β uγ +(x(cid:48)
2)2uγ +u3
x(cid:48)
1x(cid:48)
x(cid:48)
1u2
β uγ +u2
x(cid:48)
β uγ +u2
x(cid:48)
Table 4. Cluster variables for Example 9.5
x(cid:48)
β uγ +(x(cid:48)
αu3
1(x(cid:48)
x(cid:48)
β uγ +(x(cid:48)
αu3
1(x(cid:48)
x(cid:48)
2)2
x(cid:48)
2)2uαuγ +u4
2)2uαuγ +u4
x(cid:48)
1uαu3
x(cid:48)
1uαu3
γ +2x(cid:48)
γ +2x(cid:48)
x(cid:48)
1u2
1u2
1u2
αuβ
αuβ
2)2
1
α
α
2
2
1
2
2
αuβ
αuβ
(x(cid:48)
1)2u5
β u2
(x(cid:48)
1)2u5
β u2
UNIVERSAL GEOMETRIC CLUSTER ALGEBRAS
37
This is a polynomial in −ab. The first several polynomials Pm are shown in Table 5.
m
0
1
2
3
4
5
Pm
1
1
−ab − 1
−ab − 2
a2b2 + 3ab + 1
a2b2 + 4ab + 3
Table 5. The polynomials Pm
Then the g-vectors associated to BT are(cid:2) ±1
Proposition 9.6. Suppose a and b are integers with ab ≤ −4 and write B = [ 0 a
b 0 ].
(cid:3), and
(cid:3),(cid:2) 0±1
0
−aPm+1
(cid:104) sgn(a)Pm
(cid:104) −bPm
(cid:104) −bPm+1
(cid:104) sgn(a)Pm+1
sgn(b)Pm+1
sgn(b)Pm
−aPm
(cid:105)
(cid:105)
(cid:105)
(cid:105)
(9.6)
(9.7)
(9.8)
(9.9)
The rays of F◦
for m even and m ≥ 0,
for m odd and m ≥ 1,
for m even and m ≥ 0, and
for m odd and m ≥ 1.
B are spanned by the g-vectors given above.
To prove Proposition 9.6, one matches (9.5) with a well-known formula for Um(x),
the Chebyshev polynomials of the second kind, to see that
(cid:40)
√−ab
(9.10)
Pm =
Um(
1√−ab
)
√−ab
2
Um(
2
if m is even,
if m is odd.
)
Using known formulas for the roots of Um(x), one shows that Pm is positive for
all m ≥ 0. One then computes g-vectors using Conjecture 8.11, which holds in
this case by Theorem 8.12. The second assertion follows by Theorem 8.7. We omit
further details. (The calculation of g-vectors draws on notes shared with the author
by Speyer [28] in connection with joint work on [26].)
The rays described in (9.6) and (9.7) interlace and approach a limit. Using (9.10)
and a well-known formula for Um(x), we see that the limiting ray is spanned by
(9.11)
v∞(a, b) =
(9.12)
v−∞(a, b) =
(cid:104)
(cid:104) −b(
−a(
(cid:105)
(cid:105)
.
√−ab
√−ab−4)
2 sgn(a)
√−ab+
√−ab+
2 sgn(b)
√−ab−4)
√−ab
.
Similarly, the rays described in (9.8) and (9.9) interlace and approach the limit
Let C∞(a, b) be the closed cone whose extreme rays are v∞(a, b) and v−∞(a, b).
Then C∞(a, b)\{0} is the set of all points in R2 not contained in any 2-dimensional
cone transitively adjacent to O in FB. For k ∈ {1, 2}, the mutation map ηB
k takes
C∞(a, b) to C∞(−a,−b). Similarly, η−B takes C∞(−a,−b) to C∞(a, b). (Recall
that µk(B) = −B for k = 1 or k = 2.) Both of these cones are contained in an
38
NATHAN READING
open coordinate orthant, and we conclude that C∞(a, b) \ {0} is contained in a B-
class. That B-class cannot be any larger than C∞(a, b) \ {0} without overlapping
a B-cone that is transitively adjacent to O. Thus C∞(a, b) \ {0} is a B-class and
C∞(a, b) is a B-cone. We have proven the following proposition.
Proposition 9.7. Suppose a and b are integers with ab ≤ −4 and write B = [ 0 a
b 0 ].
Then the mutation fan FB consists of F◦
B and the cone C∞(a, b).
For each ray ρ in R◦(B), choose vρ to be the corresponding vector in Proposi-
tion 9.6. Each other ray ρ in R(B) is spanned by v∞(a, b) or v−∞(a, b) or both,
and we thus choose vρ to be v∞(a, b) or v−∞(a, b). Now, Proposition 9.2 tells
us that in any B-coherent linear relation supported on {vρ : ρ ∈ R(B)}, all of the
vectors appearing in Proposition 9.6 appear with coefficient zero. The one or two
remaining vectors form a linearly independent set, so they also appear with coeffi-
cient zero. We conclude that there exists no nontrivial B-coherent linear relation
supported on a finite subset of {vρ : ρ ∈ R(B)}. Theorem 8.17 implies that each
pair of vectors vρ and vρ(cid:48) spanning a g-vector cone for BT are a Z-basis for Z2.
When v∞(a, b) and v−∞(a, b) are related by a positive scaling, the cone C∞
degenerates to a limiting ray spanned by v∞(a, b) =
. This happens if and
only if ab = −4, so that (a, b) is (±1,∓4), (±2,∓2), or (±4,∓1). These are exactly
the rank-2 cases where B is of affine type, as we define later in Definition 10.14. In
these cases, the mutation fan FB consists of the g-vector cones for BT , together
with the limiting ray ρ∞. We have proved the following.
Proposition 9.8. Suppose a and b are integers with ab = −4 and write B = [ 0 a
b 0 ].
Then for any R, the g-vectors for BT , together with the shortest integer vector that
is a positive rational scaling of
, constitute a positive R-basis for B.
(cid:104) 2 sgn(a)−a
(cid:105)
(cid:104) 2 sgn(a)−a
(cid:105)
Some of the rank-2 universal extended exchange matrices of affine type are given
below in (9.13).
(9.13)
0
−4
−1
1
0
0
1 −1
3 −2
·
·
·
·
·
·
0 −1
4 −3
8 −5
·
·
·
·
·
·
0
1
4 −1
8 −3
·
·
·
·
·
·
0
1
3 −1
5 −2
·
·
·
·
·
·
2 −1
0
−2
−1
2
0
0
0 −1
1 −2
2 −3
3 −4
4 −5
·
·
·
·
·
·
0
1
1
0
2 −1
3 −2
4 −3
5 −4
·
·
·
·
·
·
1 −1
Each matrix contains several infinite sequences of coefficient rows, separated by
horizontal lines for the sake of clarity. The calculations are easy when −ab = 4,
because it is known that Um(1) = m + 1, so that Pm = m + 1 for m even and
Pm = 1
2 (m + 1) for m odd. As before, Propositions 7.1 and 7.2 ensure that the
remaining cases are obtained from the cases shown by negating all entries and/or
swapping the columns and then swapping the first two rows. The mutation fans
for the two cases in (9.13) are shown in Figure 3.
UNIVERSAL GEOMETRIC CLUSTER ALGEBRAS
39
B =(cid:2) 0 1−4 0
(cid:3)
B =(cid:2) 0 2−2 0
(cid:3)
Figure 3. Mutation fans FB for the affine examples in (9.13)
When ab ≤ −5, the set C∞(a, b) is a 2-dimensional cone whose extreme rays
are algebraic, but not rational. (Consider (9.11) and (9.12) and note that the only
pair of perfect squares differing by 4 is {0, 4}.) Two of these cases are shown in
Figure 4, with C∞(a, b) shaded gray. The additional points drawn in the figure are
explained in the proof of Proposition 9.9.
B =(cid:2) 0 1−5 0
(cid:3)
B =(cid:2) 0 2−3 0
(cid:3)
Figure 4. Mutation fans FB in two infinite non-affine rank-2 cases
Proposition 9.9. Suppose a and b are integers with ab ≤ −5 and write B = [ 0 a
√−ab − 4], then the g-vectors for BT ,
b 0 ].
(1) If R is a field containing Q[
√−ab,
together with v∞(a, b) and v−∞(a, b), constitute a positive R-basis for B.
(2) No positive Z-basis or Q-basis exists for B.
(3) If R is a field, then the g-vectors for BT , together with any two linearly
independent vectors in C∞(a, b) ∩ R2, constitute a cone R-basis for B.
BT , constitute a cone Z-basis for B.
(4) There exist integer vectors in C∞(a, b) that, together with the g-vectors for
Proof. Assertion (1) is immediate from the discussion above. Assertion (2) also
follows in light of Corollary 8.15. To establish Assertions (3) and (4), we point
out that the extreme rays of C∞(a, b) are not themselves B-cones. (If an extreme
ray of C∞(a, b) were a B-cone, then the nonzero points on that ray would be a
B-class, by Proposition 5.21. But C∞(a, b) \ {0} is already a B-class.) To satisfy
Definition 6.3, we therefore don't need vectors in the extreme rays of C∞(a, b). We
40
NATHAN READING
only need two vectors in C∞(a, b) that span R2. If R is a field, these can be any
linearly independent vectors in C∞(a, b).
If R = Z, then we need to find an Z-basis for Z2 in C∞(a, b). If a(cid:48) ≥ a ≥ 0 and
0 ≥ b ≥ b(cid:48), then C∞(a, b) ⊆ C∞(a(cid:48), b(cid:48)). Thus, by symmetry, it is enough to check
the minimal cases (a, b) = (1,−5) and (a, b) = (2,−3).
In both cases, C∞(a, b)
contains a Z-basis for Zn, shown by the white dots in Figure 4. (The black and
(cid:3)
white dots in the pictures are the points Z2.)
10. Rays in the Tits cone
In this section, we define a subset R±Tits(B) of the rays R(B) of FB. These
are the rays spanned by g-vectors for BT that are contained in the Tits cone or in
its antipodal cone, in a sense that we make precise below. We prove the following
proposition.
Proposition 10.1. Suppose B is acyclic and satisfies the Standard Hypotheses.
Suppose also that every submatrix B(cid:104)j(cid:105) of B is of finite Cartan type. Choose a
nonzero vector vρ in each ray ρ ∈ R(B). Then any B-coherent linear relation
supported on {vρ : ρ ∈ R(B)} is in fact supported on(cid:8)vρ : ρ ∈ R(B) \ R±Tits(B)(cid:9).
Proposition 10.1 will allow us to construct positive bases for exchange matrices of
finite type, generalizing Proposition 9.4. To illustrate the usefulness of the propo-
sition beyond finite type, we also construct positive bases for a rank-3 exchange
matrix of affine type. A similar construction can be carried out for any rank-3
exchange matrix of affine type. Based on these constructions and on insight from
[27], we make a general conjecture about bases for exchange matrices of affine type.
We now proceed to fully explain and then prove Proposition 10.1. For Standard
Hypotheses, see Definition 8.2. An exchange matrix B is acyclic if, possibly after
reindexing by a permutation of [n], it has the following property: If bij > 0 then
i < j. Define B(cid:104)j(cid:105) to be the matrix obtained from B by deleting row j and column j.
Definition 10.2 (Cartan companion of B). Recall from Definition 2.3 that B is an
n × n skew-symmetrizable integer matrix, and specifically that δ(i)bij = −δ(j)bji
for all i, j ∈ [n]. The Cartan companion of B is the n × n matrix A with
diagonal entries 2 and off-diagonal entries aij = −bij. Then δ(i)aij = δ(j)aji for
all i, j ∈ [n], and accordingly A is symmetrizable. In fact, A is a Cartan matrix
in the usual sense. (See [13], or for a treatment specific to the present purposes,
see [24, Section 2.2].)
Definition 10.3 (Tits cone and R±Tits(B)). Let V be a real vector space of di-
mension n with a basis Π = {αi : i ∈ [n]} and write V ∗ for its dual vector space.
The basis vectors αi are called the simple roots. The simple co-roots associated
to A are α∨
i = δ(i)−1αi. These are the simple roots associated to the transpose
AT . Let (cid:104)·,·(cid:105) : V ∗ × V → R denote the canonical pairing. We identify V ∗ with
Rn by identifying dual basis to Π (called the fundamental co-weights) with the
standard unit basis. Specifically, we identify the dual basis vector dual to αi with ei.
Associated to the Cartan matrix A is a Coxeter group W , defined as a group
generated by reflections on V . For each i ∈ [n], we define a reflection si by spec-
ifying its action on the simple roots: si(αj) = αj − aijαi. The group W is the
group generated by all of these reflections, and it is a Coxeter group. We define a
symmetric bilinear form K on V by setting K(α∨
i , αj) = aij. The element si is a
UNIVERSAL GEOMETRIC CLUSTER ALGEBRAS
41
specifically, the action of si fixes ej for j (cid:54)= i, and sends ei to −ei +(cid:80)
Define D =(cid:84)
reflection with respect to the form K, and thus W acts by isometries on V with
respect to the form K. The action of W on V induces a dual action on V ∗ in the
usual way. The action of si on V ∗ is a reflection fixing the hyperplane α⊥
i . More
j(cid:54)=i aijej,
i∈[n] {x ∈ V ∗ : (cid:104)x, αi(cid:105) ≥ 0} ⊂ V ∗. Under the identification of V ∗
with Rn, the cone D is the nonnegative orthant O. The cones wD are distinct,
for distinct w ∈ W . The union of these cones is called the Tits cone Tits(A).
As the name suggests, the Tits cone is a cone. By definition, it is preserved under
the action of W on V ∗. We write −Tits(A) for the image of Tits(A) under the
antipodal map.
Continuing to identify V ∗ with Rn as above, we define R±Tits(B) to be the set
of rays in R(B) that are contained in Tits(A) ∪ (−Tits(A)). (Since Tits(A) is a
cone and contains 0, any ray is either completely contained in Tits(A)∪ (−Tits(A))
or intersects Tits(A) ∪ (−Tits(A)) only at 0.)
Definition 10.4 (Cartan type of B). A Cartan matrix A is of finite type if the
associated Coxeter group W is finite. When A is of finite type, then Tits(A) is
all of V ∗. Cartan matrices of finite type are classified (e.g. as "type Dn," etc.).
We define Cartan matrices of affine type in Definition 10.14. If A is the Cartan
companion of B, then we use the phrase Cartan type of B to refer to the type
of A.
The following result [11, Theorem 1.4] links Definition 10.4 to Definition 9.3.
Theorem 10.5. An exchange matrix B is of finite type if and only if it is mutation
equivalent to an exchange matrix of finite Cartan type.
Having explained Proposition 10.1, we now prepare to prove it. We begin with
a known result on the g-vector cones in finite type.
Theorem 10.6. If B is of finite type, then the g-vector cones associated to B are
the maximal cones of a complete simplicial fan.
One way to obtain Theorem 10.6 from the literature is the following: If B is
of finite Cartan type, then the g-vector cones coincide with the maximal cones
of the Cambrian fan of [23], which is complete by definition and simplicial by [23,
Theorem 1.1]. This was conjectured (and proved in a special case) in [23, Section 10]
and proved (for all B of finite Cartan type) in [30]. To obtain the theorem for B of
finite type but not of finite Cartan type, one applies Theorems 8.12 and 10.5 and
appeals to the case of finite Cartan type.
Next, we need a known result on g-vectors.
Proposition 10.7. Suppose the entries in column n of B are nonpositive and
suppose t is a vertex of Tn connected to t0 by a path with no edges labeled n. Then
the g-vector cone ConeB;t0
is the nonnegative linear span of en and Cone
B(cid:104)n(cid:105);t0
t
t
.
Here, we realize Tn−1 by deleting all edges labeled n from Tn and taking the
⊂ Rn−1 is embedded into Rn by
B(cid:104)n(cid:105);t0
connected component of t0. The cone Cone
t
appending an nth entry 0 to all vectors in the cone.
The fact that ConeB;t0
has en as an extreme ray is obvious from Definition 8.4.
The rest of Proposition 10.7 is not obvious from Definition 8.4. We sketch how the
rest of the proposition can be obtained from [26].
t
42
NATHAN READING
t
By [26, Theorem 3.27] we associate a framework to any exchange matrix B;
this is an assignment of n vectors to each vertex of Tn satisfying certain conditions.
By [26, Theorem 3.24(3)], the vectors assigned to t ∈ Tn are inward-facing normals
to the facets of the cone ConeB;t0
. A framework in particular satisfies the Transition
condition and the Sign condition. The Sign condition says that each inward-facing
normal β has a sign sgn(β) ∈ {±1} such that every entry of β weakly agrees in sign
with sgn(β). If t and t(cid:48) are adjacent vertices of Tn then the cones ConeB;t0
and
ConeB;t0
share a facet. Say β is normal to that facet, facing inward with respect
to ConeB;t0
. The
Transition condition is assertion that γ + [sgn(β)ω(β∨, γ)]+ β is an inward facing
. Here β∨ is a certain nonzero scaling of β and ω is bilinear
normal to ConeB;t0
form whose matrix is essentially B.
, and suppose γ is some other inward-facing normal to ConeB;t0
t(cid:48)
t(cid:48)
t
t
t
Now ConeB;t0
t0
t0
has en as an inward-facing normal. Taking β to be some other
inward-facing normal of ConeB;t0
, the assumption that the entries in column n of B
are nonpositive translates to the assertion that sgn(β)ω(β∨, en) is nonpositive, so
the Transition condition implies that the g-vector cones adjacent to ConeB;t0
also
have en as an inward-facing normal. Repeating the argument, we see that en is an
inward-facing normal to any cone ConeB;t0
with t connected to t0 by a path with
no labels n. Restricting the Transition condition for cones ConeB;t0
to the facets
B(cid:104)n(cid:105);t0
defined by en, we obtain exactly the Transition condition for cones Cone
,
t
and we conclude by induction that the facet of ConeB;t0
defined by en is exactly
Cone
t0
.
t
t
t
B(cid:104)n(cid:105);t0
t
Theorem 10.6 and Proposition 10.7 allow us to prove the following Lemma which
is the key step in the proof of Proposition 10.1.
Lemma 10.8. Let B be an exchange matrix satisfying the Standard Hypotheses.
Suppose the entries in column n of B are nonnegative and suppose B(cid:104)n(cid:105) is of finite
type. Then the unique ray of FB intersecting {x ∈ Rn : xn > 0} is R≥0en.
Proof. Theorem 10.6 (applied to BT ) says that Rn−1 is covered by cones Cone
BT(cid:104)n(cid:105);t0
for t ∈ Tn−1. Thus Proposition 10.7 implies that the closed halfspace of Rn con-
t
sisting of vectors with nonnegative nth entry is covered by cones ConeBT ;t0
with t
connected to t0 by paths with no labels n. By Theorem 8.7, each of these cones is
in FB. Each of these cones contains the ray R≥0en, which is therefore the unique
ray of FB intersecting {x ∈ Rn : xn > 0}.
(cid:3)
t
Definition 10.9 (Length of a ray in Tits(A)). The length of an element w ∈ W
is the number of letters in a shortest expression for w as a product of generators
si. We write (cid:96)(w) for this length. Given an element w and a generator si, it is
well-known that (cid:96)(siw) < (cid:96)(w) if and only if wD is contained in the halfspace
{x ∈ V ∗ : (cid:104)x, αi(cid:105) ≤ 0}. This is identified with the set of points in Rn with non-
positive ith coordinate. We define the length (cid:96)(ρ) of a ray ρ contained in Tits(A)
to be the minimum length of w such that ρ ⊂ wD. If siρ is the image of ρ under
the reflection si, then we have (cid:96)(siρ) < (cid:96)(ρ) if and only if the ith coordinate of
nonzero vectors in ρ is negative.
We now prove Proposition 10.1.
UNIVERSAL GEOMETRIC CLUSTER ALGEBRAS
43
Consider a B-coherent linear relation(cid:80)
Proof of Proposition 10.1. Since B is acyclic, we may as well assume that B is
indexed so that if bij > 0 then i < j. The skew-symmetrizability of B means that if
bji < 0 then i < j. In particular, the entries in column n of B are all nonnegative.
ρ∈S cρvρ with S a finite subset of R(B).
Let ρ be a ray in S contained in Tits(A). (Rays in −Tits(A) are dealt with later.)
We argue by induction on (cid:96)(ρ), letting B vary, that cρ = 0.
First, suppose the nth coordinate of vρ is positive. Then Lemma 10.8 implies
that vρ is the unique vector in {vρ(cid:48) : ρ(cid:48) ∈ S} whose nth entry is positive. Now
Proposition 4.12 implies that cρ = 0.
Next suppose the nth coordinate of vρ is negative. Write vρ = (cid:80)
n (vρ) = vρ − 2vnen −(cid:80)
ρ∈R(B) cρηB
i∈[n] viei.
The entries bnj are all nonpositive, so ηB
j∈[n−1] vnbnjej.
Since the bnj are nonpositive, they equal the entries anj of the Cartan companion
(cid:80)
of B, and we observe that ηB
n (vρ) = sn(vρ). Proposition 7.3 implies that the
rays R(µn(B)) of Fµn(B) are obtained by applying ηB
n to each ray in R(B). Thus
n (vρ) is a µn(B)-coherent linear relation. Since all entries in column
n are nonnegative, all entries in row n are nonpositive. It is thus apparent that
the operation µn does nothing to B other than reverse signs in row n and column
n. In particular, µn(B) is acyclic and the Cartan type of submatrices of µn(B)
is the same as the Cartan type of submatrices of B. We see that µn(B) satisfies
the hypotheses of the proposition. Since the ith coordinate of vρ is negative, we
have (cid:96)(ηB
n (ρ)) = (cid:96)(siρ) < (cid:96)(ρ), so by induction on (cid:96)(ρ), we conclude that cρ = 0 as
desired.
Finally, suppose the nth coordinate of vρ is zero. As before,(cid:80)
n (vρ)
is a µn(B)-coherent linear relation, and as before, µn(B) is obtained from B by
reversing signs in row n and column n. We observe that ηB
n (ρ) = ρ. The matrix
µn(B) has nonnegative entries in column n − 1. Now we consider the sign of the
(n − 1)st coordinate of vρ. If it is positive or negative, then we conclude as above
that cρ = 0. If it is zero, the we replace µn(B) with µ(n−1)n(B) and consider the
sign of the (n − 2)nd coordinate of vρ. Continuing in this manner, we eventually
find a positive or negative coefficient, since vρ is nonzero, and when we do, we
conclude that cρ = 0.
ρ∈R(B) cρηB
If ρ is in −Tits(A), then (7.1) implies that (cid:80)
The base case (cid:96)(ρ) = 0 is handled as part of the above argument: If (cid:96)(ρ) = 0
then all coordinates of vρ are nonnegative and we eventually complete the argument
without induction.
This completes the proof that cρ = 0 for a ray ρ in R(B) contained in Tits(A).
ρ∈R(B) cρ(−vρ) is a (−B)-coherent
linear relation. Proposition 7.1 implies that the vectors −vρ span the rays of F−B.
(cid:3)
The argument above shows that cρ = 0.
Remark 10.10. Proposition 10.1 can also be proved using the Cambrian frameworks
of [19, Section 5]. Readers familiar with Cambrian lattices and sortable elements
will recognize that the proof given here is patterned after the usual induction on
length and rank common to proofs involving sortable elements.
Remark 10.11. It is natural to ask whether Proposition 10.1 can be proved with
weaker hypotheses. Indeed, it was stated without assuming the Standard Hypothe-
ses and without the hypotheses on submatrices in an earlier version of this paper,
but an error was later found in the proof.
44
NATHAN READING
We now use Proposition 10.1 to prove the following theorem.
Theorem 10.12. Let B be a skew-symmetrizable exchange matrix of finite type
satisfying the Standard Hypotheses and let R be any underlying ring. Then the
g-vectors associated to BT constitute a positive R-basis for B.
We expect that every skew-symmetrizable exchange matrix of finite type satisfies
the Standard Hypotheses. (Indeed, as indicated in Remark 8.14, since every such
matrix is mutation equivalent to an acyclic matrix, the result may soon appear in
print.)
Proof. Theorem 8.7 says that the g-vector cones associated to BT are the maximal
cones in a subfan of FB. Theorem 10.6 says that this subfan is complete, and
therefore it must coincide with the entire fan FB. When B is of finite Cartan type
with Cartan companion A, the Tits cone Tits(A) is all of Rn. Thus Proposition 10.1
says that there is no non-trivial B-coherent linear relation supported on the g-
vectors. By Theorem 8.17 and Proposition 6.7, we conclude that the g-vectors are
a positive R-basis for B, for any R. The case where B is of finite type but not of
finite Cartan type now follows by Theorems 8.12 and 10.5 and by the observation
(cid:3)
that a mutation map ηB
k takes an R-basis for B to an R basis for µk(B).
The g-vectors for B of finite Cartan type can be found explicitly in various ways,
including using sortable elements and Cambrian lattices as described in [26], or by
the methods of [30].
Remark 10.13. Theorems 4.4 and 10.12 say that the g-vectors for BT are the coeffi-
cient rows of a positive universal extended exchange matrix. Another construction
of universal coefficients, not conditioned on any conjectures, was already given in
[12, Theorem 12.4]. Furthermore, the construction from [12] yields cluster algebras
with completely universal coefficients, rather than only universal geometric coef-
ficients. That is, an arbitrary cluster algebra (not necessarily of geometric type)
with initial exchange matrix B admits a unique coefficient specialization from the
universal cluster algebra. (The relevant definition of coefficient specialization is [12,
Definition 12.1].) We conclude the section by explaining the connection between
Theorem 10.12 and [12, Theorem 12.4]. This connection provided the original mo-
tivation for this research.
A more detailed description of [12, Theorem 12.4], in the language of this paper,
is the following: Let B be a bipartite exchange matrix of finite Cartan type with
Cartan companion A. (An exchange matrix B is bipartite if there is a function
ε : [n] → {±1} such that bij > 0 implies that ε(i) = 1 and that ε(j) = −1.) The
co-roots associated to A are the vectors in the W -orbits of the simple co-roots.
Since A is of finite type, there are finitely many co-roots. A co-root is positive if it
is in the nonnegative linear span of the simple co-roots, and almost positive if it is
positive or if it is the negative of a simple co-root. We construct an integer extended
exchange matrix extending B whose coefficient rows are indexed by almost positive
co-roots. The coefficient row indexed by a co-root β∨ has entries ε(i)[β∨ : α∨
i ] for
i = 1, . . . , n, where [β∨ : α∨
in the expansion of
β∨ in the basis of simple co-roots. The assertion of [12, Theorem 12.4] is that this
extended exchange matrix defines a universal cluster algebra for B.
Let L be the linear map taking a positive root αi to −ε(i)ei. (Recall that we
have identified ei with an element in the basis for V ∗ dual to the basis of simple
i ] stands for the coefficient of α∨
i
UNIVERSAL GEOMETRIC CLUSTER ALGEBRAS
45
roots in V .) One can define almost positive roots by analogy to the definitions in
the previous paragraph, so that the almost positive co-roots for A are the almost
positive roots for AT . As conjectured in [19, Conjecture 1.4] and proved in [23, The-
orem 9.1], the map L takes almost positive roots into rays in the Cambrian fan for
B and induces a bijection between almost positive roots and rays of the Cambrian
fan. (The difference in signs between [19, Conjecture 1.4] and [23, Theorem 9.1] is
the result of a difference in sign conventions. See the end of [25, Section 1].)
It now becomes possible to relate [12, Theorem 12.4] to Theorem 10.12. To
apply [12, Theorem 12.4] to BT , we pass from co-roots to roots (thus exchanging A
with AT ) and we introduce a global sign change to the function ε. Thus we write
a universal extended exchange matrix for BT by taking, for each almost positive
root of A, a coefficient row with entries −ε(i)[β : αi] for i = 1, . . . , n, where [β : αi]
stands for the coefficient of αi in the expansion of β in the basis of simple roots. In
other words, the coefficient row indexed by an almost positive root β is L(β). This
means we have chosen a nonnegative vector in each ray of the Cambrian fan for BT .
But as mentioned above, the Cambrian fan for BT is the fan whose maximal cones
are the g-vector cones for BT , and thus we have essentially recovered Theorem 10.12
for the case of bipartite B.
We conclude by sketching the construction of an R-basis for the rank-3 exchange
. This exchange matrix is of affine Cartan type in the sense
−1
(cid:104) 0
(cid:105)
matrix B =
of the following definition.
2
0
0 −2
0
1
0
simple roots coordinates of a vector β =(cid:80)
Definition 10.14 (Affine type). A Cartan matrix is of affine type if the associ-
ated symmetric bilinear form is positive semidefinite and every proper principal
submatrix is of finite type. For our purposes, the key property of a Cartan matrix
A of affine type is that the closure of Tits(A) is a half-space. More specifically, A
has a 0-eigenvector t = (t1, . . . , tn) with nonnegative entries. We think of t as the
i∈[n] tiαi in V . The closure of Tits(A)
is then {x ∈ V ∗ : (cid:104)x, β(cid:105) ≥ 0}, where, under our identification of Rn with V ∗, we
interpret (cid:104)x, β(cid:105) as the usual pairing of x with t. The Tits cone is the union of the
open halfspace {x ∈ V ∗ : (cid:104)x, β(cid:105) > 0} with the singleton {0}. The Cartan matrices
of affine type are classified, for example, in [15].
As in Definition 10.4, an exchange matrix B is of affine Cartan type if its Cartan
companion is of affine type. By analogy with Theorem 10.5, we say that an exchange
matrix B is of affine type if it is mutation equivalent to an acyclic exchange matrix
of affine Cartan type. (The classification of Cartan matrices of finite type implies
that an exchange matrix of finite Cartan type is acyclic. A non-acyclic exchange
matrix of affine Cartan type is of finite type.)
Examples including Proposition 9.8 and the example worked out below, together
with insights from [27], suggest the following conjecture. More partial results to-
wards the conjecture are described in the introduction.
Conjecture 10.15. Suppose B is an exchange matrix of affine type. There is a
unique integer vector v∞ such that the g-vectors associated to BT , together with
v∞, constitute a positive R-basis for B for every R.
For the rest of the section, we take B to be the exchange matrix
This is of affine Cartan type with Cartan companion A =
. The matrix
(cid:104) 0
−1
2
0
0 −2
0
1
0
(cid:105)
.
(cid:104) 2 −2
−1
0
2 −1
2
0 −2
(cid:105)
46
NATHAN READING
B has a special property not shared by all exchange matrices of affine Cartan type.
It is of the form DS, where D is an integer diagonal matrix and S is an integer
skew-symmetric matrix. Specifically, B =
. The mutation class
of B is
1
0
0 −1
0
1
0
(cid:110)±B, ±(cid:104) 0 −2
1
0
0 −2
(cid:105)
0
1
0
(cid:105)
(cid:104) 2 0 0
(cid:105)(cid:104) 0
, ±(cid:104) 0 −2
0 1 0
0 0 2
−1
1
−2
2
0 −1
0
2
(cid:105)(cid:111)
,
(10.1)
and each of these exchange matrices shares the same special property for the
same D. By results of [2], as explained in Remark 8.14, we conclude that the
Standard Hypotheses hold for B.
The vector t = [ 1 1 1 ] is a 0-eigenvector of A. The closure of the Tits cone
Tits(A) is the set of vectors in Rn whose scalar product with t is nonnegative. In
light of Theorem 8.12, one can use Conjecture 8.11 and induction to calculate the
g-vectors for cluster variables associated to BT . One finds that there are exactly
two g-vectors
(10.2)
v+ = [ 0
1 −1 ]
and v− = [ 1 −1
0 ]
in the boundary of Tits(A). The remaining g-vectors consist of six infinite families,
with the g-vector rays in each family limiting to the same ray in the boundary of
Tits(A). The vector
(10.3)
v∞ = [ 1
0 −1 ]
is the smallest nonzero integer vector in this limiting ray. The infinite families of
g-vectors are {vi + nv∞ : n ∈ Z≥0} with (vi : i = 1 . . . 6) being the vectors
(10.4)
[ 0 −1 0 ]
[ 1 −2 0 ]
[ −1 0 0 ]
[ 0 2 −1 ]
[ 0 0 1 ]
[ 0 1 0 ]
The g-vector cones for BT are the maximal cones of a simplicial fan occupying
almost all of R3. This fan is depicted in Figure 5. The picture is interpreted as
follows: Intersecting each nonzero cone with a unit sphere about the origin, we
obtain a collection of points, arcs and spherical triangles. These are depicted in
the plane by stereographic projection, with the ray spanned by t projecting to the
origin. The rays spanned by v+ and v− are indicated by blue (or dark gray) dots,
the rays spanned by vi + nv∞ are indicated by red (or medium gray) dots, and the
limiting ray spanned by v∞ is indicated by a green (or light gray) dot. The dotted
circle indicates the boundary of Tits(A).
Theorem 8.7 says that the g-vector cones determine a subfan of FB. The points
not contained in this subfan form an open cone consisting of positive linear com-
binations of v+ and v−. This open cone is covered by the nonnegative span C+
of v+ and v∞ and the nonnegative span C− of v− and v∞. Since C+ is a limit
of B-cones, it is contained in a B-cone by Proposition 7.11, and similarly C− is
contained in a B-cone. These two cones cover the set of points not contained in
g-vector cones, so either C+ and C− are each B-cones or C+∪C− is a single B-cone.
The latter possibility is ruled out by Proposition 5.30, taking k to be the empty
sequence. Thus the maximal cones of FB are the g-vector cones and C+ and C−.
Having determined the fan FB, we now prove Conjecture 10.15 for this B. The
rays of FB are spanned by the g-vectors for BT and the vector v∞. We choose
the vectors (vρ : ρ ∈ R(B)) to be these g-vectors and v∞. Proposition 10.1
says that any B-coherent linear relation supported on {vρ : ρ ∈ R(B)} is in fact
supported on {v+, v∞, v−}. Taking k to be the empty sequence and j = 2 in
Proposition 4.12 (twice), we see that the relation is supported on {v∞}, and thus
UNIVERSAL GEOMETRIC CLUSTER ALGEBRAS
47
v4
v5
v3
v1
v+
v2
v−
v6
Figure 5. The mutation fan FB for B =
(cid:104) 0
−1
2
0
0 −2
(cid:105)
0
1
0
is trivial. Appealing to Theorem 8.17 and Proposition 6.7, we have established the
following proposition.
Proposition 10.16. For B =
and any underlying ring R, the set
(cid:104) 0
{v+, v∞, v−} ∪ 6(cid:91)
2
0
0 −2
−1
(cid:105)
0
1
0
(10.5)
{vi + nv∞ : n ∈ Zn≥0}
is a positive R-basis for B, where v+, v∞, v− and the vi are as in (10.2), (10.3),
and (10.4).
i=1
Acknowledgments
Thanks to Ehud Hrushovski for enlightening the author on the subject of en-
domorphisms of the additive group R (in connection with Remark 3.5). Thanks
to David Speyer for pointing out the role of the polynomials Pm in describing the
g-vectors associated to rank-2 exchange matrices of infinite type. (See Section 9.)
Thanks to an anonymous referee of [21] for pointing out that Proposition 4.6 needs
the hypothesis that R is a field. Thanks to Kiyoshi Igusa and Dylan Rupel for
48
NATHAN READING
helping to detect an error in an earlier version. Thanks to an anonymous referee of
this paper for many helpful suggestions which improved the exposition.
References
[1] A. Bjorner, M. Las Vergnas, B. Sturmfels, N. White and G. Ziegler, Oriented matroids.
Encyclopedia of Mathematics and its Applications, 46, Cambridge Univ. Press, 1999.
[2] L. Demonet, Mutations of group species with potentials and their representations. Applica-
tions to cluster algebras. Preprint, 2010 (arXiv:1003.5078)
[3] L. Demonet, Personal communication, 2012.
[4] H. Derksen, J. Weyman, and A. Zelevinsky, Quivers with potentials and their representations
II: applications to cluster algebras. J. Amer. Math. Soc. 23 (2010), no. 3, 749 -- 790.
[5] V. V. Fock and A. B. Goncharov, Cluster ensembles, quantization and the dilogarithm. Ann.
Sci.´Ec. Norm. Sup´er. (4) 42 (2009), no. 6, 865 -- 930.
[6] V. V. Fock, and A. B. Goncharov, Cluster X-varieties at
infinity. Preprint, 2011
(arXiv:1104.0407).
[7] S. Fomin, M. Shapiro, and D. Thurston, Cluster algebras and triangulated surfaces. I. Cluster
complexes. Acta Math. 201 (2008), no. 1, 83 -- 146.
[8] S. Fomin and D. Thurston, Cluster algebras and triangulated surfaces. Part II: Lambda
lengths. Preprint, 2008.
[9] S. Fomin and A. Zelevinsky, Y -systems and generalized associahedra. Ann. of Math. (2) 158
(2003), no. 3, 977 -- 1018.
[10] S. Fomin and A. Zelevinsky, Cluster algebras. I. Foundations. J. Amer. Math. Soc. 15 (2002),
no. 2, 497 -- 529.
[11] S. Fomin and A. Zelevinsky, Cluster Algebras II: Finite Type Classification. Inventiones
Mathematicae 154 (2003), 63 -- 121.
[12] S. Fomin and A. Zelevinsky, Cluster Algebras IV: Coefficients. Compositio Mathematica 143
(2007), 112 -- 164.
[13] V. Kac, Infinite-dimensional Lie algebras. Third edition. Cambridge University Press, Cam-
bridge, 1990.
[14] J. L. Kelley, General topology. Graduate Texts in Mathematics 27. Springer-Verlag, New
York-Berlin, 1975.
[15] I. G. Macdonald, Affine root systems and Dedekind's η-function. Invent. Math. 15 (1972),
91 -- 143.
[16] K. Nagao, Donaldson-Thomas theory and cluster algebras. Preprint, 2010. (arXiv:1002.4884)
[17] T. Nakanishi and A. Zelevinsky, On tropical dualities in cluster algebras. Proceedings of
Representation Theory of Algebraic groups and Quantum groups 10, Contemp. Math. 565
(2012), 217 -- 226.
[18] P.-G. Plamondon, Cluster algebras via cluster categories with infinite-dimensional morphism
spaces. Compositio Mathematica 147 (2011), 1921 -- 1954.
[19] N. Reading, Cambrian Lattices. Adv. Math. 205 (2006), no. 2, 313 -- 353.
[20] N. Reading, Clusters, Coxeter-sortable elements and noncrossing partitions. Trans. Amer.
Math. Soc. 359 (2007), no. 12, 5931 -- 5958.
[21] N. Reading, Universal geometric cluster algebras
from surfaces. Preprint,
2012.
(arXiv:1209.4095)
[22] N. Reading, Universal geometric coefficients for the once-punctured torus. Preprint, 2012.
(arXiv:1212.1351)
[23] N. Reading and D. E. Speyer, Cambrian Fans. J. Eur. Math. Soc. (JEMS) 11 (2009), no. 2,
407 -- 447.
[24] N. Reading and D. Speyer, Sortable elements in infinite Coxeter groups. Trans. Amer. Math.
Soc. 363 (2011) no. 2, 699 -- 761.
[25] N. Reading and D. Speyer, Sortable elements for quivers with cycles. Electron. J. Combin.
17(1) (2010), Research Paper 90, 19 pp.
[26] N. Reading and D. E. Speyer, Combinatorial frameworks for cluster algebras. Preprint, 2011.
(arXiv:1111.2652)
[27] N. Reading and D. E. Speyer, Cambrian frameworks for cluster algebras of affine Cartan
type. In preparation, 2012.
[28] D. E. Speyer, Personal communication, 2011.
UNIVERSAL GEOMETRIC CLUSTER ALGEBRAS
49
[29] R. Webster, Convexity. Oxford Science Publications. The Clarendon Press, Oxford University
Press, New York, 1994.
[30] S. Yang and A. Zelevinsky Cluster algebras of finite type via Coxeter elements and principal
minors, Transformation Groups 13 (2008), no. 3 -- 4, 855 -- 895.
|
1007.4975 | 1 | 1007 | 2010-07-28T13:50:52 | Derived $H$-module endomorphism rings | [
"math.RA",
"math.KT"
] | Let $H$ be a Hopf algebra, $A/B$ be an $H$-Galois extension. Let $D(A)$ and $D(B)$ be the derived categories of right $A$-modules and of right $B$-modules respectively. An object $M^\cdot\in D(A)$ may be regarded as an object in $D(B)$ via the restriction functor. We discuss the relations of the derived endomorphism rings $E_A(M^\cdot)=\op_{i\in\mathbb{Z}}\Hom_{D(A)}(M^\cdot,M^\cdot[i])$ and $E_B(M^\cdot)=\op_{i\in\mathbb{Z}}\Hom_{D(B)}(M^\cdot,M^\cdot[i])$. If $H$ is a finite dimensional semisimple Hopf algebra, then $E_A(M^\cdot)$ is a graded subalgebra of $E_B(M^\cdot)$. In particular, if $M$ is a usual $A$-module, a necessary and sufficient condition for $E_B(M)$ to be an $H^*$-Galois graded extension of $E_A(M)$ is obtained. As an application of the results, we show that the Koszul property is preserved under Hopf Galois graded extensions. | math.RA | math |
DERIVED H-MODULE ENDOMORPHISM RINGS
JI-WEI HE, FRED VAN OYSTAEYEN AND YINHUO ZHANG
Abstract. Let H be a Hopf algebra, A/B be an H-Galois extension.
Let D(A) and D(B) be the derived categories of right A-modules and of
right B-modules respectively. An object M ·
∈ D(A) may be regarded as
an object in D(B) via the restriction functor. We discuss the relations
of the derived endomorphism rings EA(M ·) = ⊕i∈Z HomD(A)(M ·, M ·[i])
and EB(M ·) = ⊕i∈Z HomD(B)(M ·, M ·[i]).
If H is a finite dimensional
semisimple Hopf algebra, then EA(M ·) is a graded subalgebra of EB(M ·).
In particular, if M is a usual A-module, a necessary and sufficient condition
for EB(M ) to be an H ∗-Galois graded extension of EA(M ) is obtained. As
an application of the results, we show that the Koszul property is preserved
under Hopf Galois graded extensions.
Keywords: Hopf Galois extension, derived endomorphism ring
MSC(2000): 16E45, 16E40, 16W50
1. Introduction
Let A/B be a Hopf Galois extension over a finite dimensional Hopf algebra
H. The motivation of this note is to try to understand how much homological
properties, such as Koszul property and Gorenstein property, may be preserved
under Hopf Galois extensions. To this aim, we need to discuss the derived
endomorphism rings of certain modules. The endomorphism ring extension of
an A-module has been studied by the second and third author in [13]. Let
M be an A-module. It was proved that there is an isomorphism of algebras
EndA(M ⊗B A) ∼= EndB(M )#H. Also necessary and sufficient conditions for
an endomorphism ring extension to be a Hopf Galois extension were obtained
in [13]. We investigate whether these properties hold when the endomorphism
rings are replaced by the derived endomorphism rings. To do this, we deal with
the relevant derived functors and work with complexes or differential graded
modules instead of usual modules over an algebra A.
Let H be a Hopf algebra with a bijective antipode, and let A be a right H-
comodule algebra with coinvariant subalgebra B = AcoH. We say that A/B is
H-Galois if A is a Hopf Galois extension of B over H. Let D(A) be the derived
category of complexes of right A-modules, and D(B) be the derived category
1
2
JI-WEI HE, FRED VAN OYSTAEYEN AND YINHUO ZHANG
of complexes of right B-modules. The restriction functor D(A) −→ D(B)
is an exact functor. For M ·, N · ∈ D(A), M · and N · may be regarded as
objects in D(B) via the restriction functor. As usual, write Ext∗
A(M ·, N ·) for
⊕i∈Z HomD(A)(M ·, N ·[i]). Endowed with the Yoneda product Ext∗
A(M ·, M ·)
becomes a graded algebra. Since A/B is H-Galois, there is a natural H-action
on Ext∗
B(M ·, M ·) so that it becomes a graded H-module algebra. In general,
the graded algebra Ext∗
A(M ·, M ·) can not be embedded into the graded algebra
Ext∗
B(M ·, M ·), which differs from the usual endomorphism rings of A-modules.
However, if H is semisimple, then Ext∗
A(M ·, M ·) is the H-invariant subalgebra
of Ext∗
B(M ·, M ·). The main result of this paper is the following one (Theorem
2.10).
Theorem. Let H be a finite dimensional semisimple Hopf algebra, and let
A/B be an H-Galois extension. For a right A-module M , Ext∗
B(M, M ) is a
graded H ∗-Galois extension of Ext∗
A(M, M ) if and only if M ⊗B A ∈ add(M ),
where add(M ) is the category consisting of all the direct summands of finite
direct sums of M .
If M is a right (A, H)-Hopf module, then Ext∗
H ∗-Galois extension of Ext∗
(Corollary 2.3), which can be regarded as an extension of [11, Theorem 3.2].
A(M, M ).
B(M, M ) is not merely an
In fact, we have the following result
Theorem. Let H be a finite dimensional semisimple and cosemisimple Hopf
algebra. If M is a right (A, H)-Hopf module, then
Ext∗
B(M, M ) ∼= Ext∗
A(M, M )#H ∗.
From this theorem, one can easily show that the Artin-Schelter Gorenstein
property is preserved under Hopf Galois extensions when the Hopf algebra is
semisimple and cosemisimple. With a bit more effort we can also show that the
Koszul property is preserved under Hopf Galois graded extensions (Theorem
2.11).
Theorem. Let H be a finite dimensional semisimple and cosemisimple Hopf
algebra, let A = ⊕n≥0An be a graded right H-module algebra such that Ai is
finite dimensional for all i ≥ 0, and let B = AcoH . Assume that A/B is an H-
Galois graded extension. If B is an N -Koszul algebra, then A is an N -Koszul
algebra.
In particular, for a pointed cocommutative Hopf algebra A, the graded al-
gebra gr(A) associated to the coradical filtration is a Koszul algebra if A has
finitely many group-like elements and the dimension of P (A), the vector space
of primitive elements, is finite.
DERIVED H -MODULE ENDOMORPHISM RINGS
3
The main tools used in this note are differential graded algebras and differen-
tial graded modules. Let us recall some notations and properties of differential
graded (dg, for short) algebras and dg modules which will be used later. By
a dg algebra we mean a graded algebra R = ⊕n∈ZRn with a differential d of
degree 1 which is also a derivation of R. Similarly we have dg modules. We
sometimes view a usual algebra A as a dg algebra concentrated in degree 0 with
a trivial differential. Then a complex of A-modules can be regarded as a dg
A-module. Let R be a dg algebra, and let XR and YR be right dg R-modules.
Write HomR(X, Y ) for ⊕i∈Z Homi
R(X, Y ) is the set of all
graded R-module morphisms of degree i. HomR(X, Y ) is a complex of vector
space. The differential on HomR(X, Y ) is given by δ(f ) = dY ◦f −(−1)f f ◦dX ,
where f is the degree of f . Also HomR(X, X) is a dg algebra and HomR(X, Y )
is a right dg HomR(X, X)-module. Let H be a Hopf algebra. We say that R
is a dg left H-module algebra if R is a left H-module algebra and the H-action
respects the the grading of R and is compatible with the differential of R. In
this case, the smash product R#H is a dg algebra and the cohomology alge-
bra H ·(R#H) ∼= H ·(R)#H. Notation and terminology used in this note is the
same as in [1, 4, 8]. For example, we use RHomR(−, −) to denote the right
derived functor of HomR(−, −), and − ⊗L
R − to denote the left derived functor
of the tensor functor − ⊗R − over a dg algebra R.
R(X, Y ), where Homi
We work over a fixed field k. All the algebras and modules are defined over
k. Unadorned ⊗ means ⊗k, and Hom means Homk.
2. Derived endomorphism rings
Throughout H will be always a Hopf algebra with a bijective antipode. Let
A/B be an H-Galois extension with the canonical map β : A ⊗B A −→ A ⊗ H.
for β−1(1 ⊗ h). For right A-modules M
i ⊗ Y h
i
and N , there is a natural left H-module action on HomB(M, N ):
For any h ∈ H, we write P X h
For h ∈ H, m ∈ M and f ∈ HomB(M, N ),
(h · f )(m) =X f (mX Sh
i
)Y Sh
i
.
Similarly, if M · and N · are complexes of right A-modules, then HomB(M ·, N ·)
is a complex of left H-modules. Hence for any M · ∈ D−(A) and N · ∈ D(A),
RHomB(M ·, N ·) is a complex of left H-modules.
If M is a right A-module, then there is an isomorphism of right B-modules
and right H-comodules:
M ⊗B A −→ M ⊗ H,
m ⊗ a 7→X ma0 ⊗ a1,
4
JI-WEI HE, FRED VAN OYSTAEYEN AND YINHUO ZHANG
with the inverse given by:
M ⊗ H −→ M ⊗B A,
These isomorphisms may be generalized to complexes. If M · is a complex of
A-modules, then we have an isomorphism of complexes: M · ⊗B A −→ M · ⊗ H.
m ⊗ h 7→X mX h
i ⊗ Y h
i .
Theorem 2.1. Let A/B be an H-Galois extension, and M · an object in
D−(A). If H is finite dimensional, then there is an isomorphism of graded
algebras
Ext∗
A(M · ⊗L
B A, M · ⊗L
B A) −→ Ext∗
B(M ·, M ·)#H.
Proof. Let P · be a bounded above complex of projective A-modules which is
B A ∼= P · ⊗B A. Of course
quasi-isomorphic to the complex M ·. Then M · ⊗L
P · ⊗B A is homotopically projective (see [8, Ch. 8]). Hence
RHomA(M · ⊗L
B A, M · ⊗L
B A) ∼= RHomA(P · ⊗B A, P · ⊗B A)
= HomA(P · ⊗B A, P · ⊗B A)
∼= HomB(P ·, P · ⊗B A)
∼= HomB(P ·, P · ⊗ H)
∼= HomB(P ·, P ·) ⊗ H
= RHomB(M ·, M ·) ⊗ H.
Therefore as graded spaces, we have Ext∗
It remains to show that this isomorphism is an isomorphism of graded algebras.
BA, M ·⊗L
A(M ·⊗L
B(M ·, M ·)#H.
BA) ∼= Ext∗
From the above isomorphisms, we get an isomorphism of complexes
(1)
ϕ : HomB(P ·, P ·) ⊗ H −→ HomA(P · ⊗B A, P · ⊗B A),
such that for f ∈ HomB(P ·, P ·), h ∈ H, p ∈ P and a ∈ A,
ϕ(f ⊗ h)(p ⊗B a) =X f (p)X h
i ⊗B Y h
i a.
We have to show that ϕ is an isomorphism of dg algebras from HomB(P ·, P ·)#H
to HomA(P · ⊗B A, P · ⊗B A). Since ϕ is compatible with the differentials, it
suffices to show that it is an algebra morphism. The proof is essentially the
same as that of [13, Theorem 2.3], so we omit it. Now by taking the cohomol-
ogy algebra of the dg algebras HomB(P ·, P ·)#H and HomA(P · ⊗B A, P · ⊗B A)
we get the desired result.
(cid:3)
Let M · be a complex of A-modules. We have HomA(M ·, M ·) ∼= HomB(M ·, M ·)H
as dg algebras, where HomB(M ·, M ·)H is the H-invariant dg subalgebra of
HomB(M ·, M ·).
If H is a finite dimensional semisimple Hopf algebra, then
DERIVED H -MODULE ENDOMORPHISM RINGS
5
it is not difficult to see that Ext∗
bras. In other words, the graded algebra Ext∗
Ext∗
A(M ·, M ·) ∼= Ext∗
A(M ·, M ·). We ask when this H ∗-extension is an H ∗-Galois extension.
Let M · be a complex of right (A, H)-Hopf modules. Then HomA(M ·, M ·)
is a dg left H ∗-module algebra with the left H ∗-action given by, for α ∈ H ∗
and f ∈ HomA(M ·, M ·),
B(M ·, M ·)H as graded alge-
B(M ·, M ·) is an H ∗-extension of
which induces a left H ∗-module algebra structure on Ext∗
A(M ·, M ·).
(α · f )(m) =X α(1)f (S(α(2))m),
Theorem 2.2. Let H be a finite dimensional semisimple and cosemisimple
Hopf algebra. Let A/B be an H-Galois extension. If M · is a bounded above
complex of right (A, H)-Hopf modules, then there is an isomorphism of graded
algebras
Ext∗
B(M ·, M ·) ∼= Ext∗
A(M ·, M ·)#H ∗.
Proof. The complex M · of right (A, H)-Hopf modules may be regarded as a
complex of right A#H ∗-modules in a natural way. By [2, Theorem 2.2] the
functor HomB(A, −) : Mod-B −→ Mod-A#H ∗ is an equivalence of abelian
categories with inverse functor − ⊗A#H ∗ A. Hence we have
Ext∗
∼= Ext∗
∼= Ext∗
A#H ∗(M · ⊗ H ∗, M · ⊗ H ∗)
B ((M · ⊗ H ∗) ⊗A#H ∗ A, (M · ⊗ H ∗) ⊗A#H ∗ A)
B(M ·, M ·).
The last isomorphism holds because (M ·⊗H ∗)⊗A#H ∗A ∼= (M ·⊗A(A#H ∗))⊗A#H ∗
A ∼= M · as a complex of right B-modules. On the other hand, since as com-
plexes of right A#H ∗-modules
M · ⊗ H ∗ ∼= M · ⊗L
A (A#H ∗),
we have
Ext∗
∼= Ext∗
∼= Ext∗
A#H ∗(M · ⊗ H ∗, M · ⊗ H ∗)
A#H ∗(M · ⊗L
A(M ·, M ·)#H ∗.
A (A#H ∗), M · ⊗L
A (A#H ∗))
The last isomorphism follows from Theorem 2.1 since A#H ∗ is an H ∗-Galois
extension of A. Now we obtain the desired isomorphism
Ext∗
B(M ·, M ·) ∼= Ext∗
A(M ·, M ·)#H ∗. (cid:3)
From the above theorem, we obtain the following corollary which may be
regarded as a natural generalization of [11, Theorem 3.2].
6
JI-WEI HE, FRED VAN OYSTAEYEN AND YINHUO ZHANG
Corollary 2.3. Let H be a finite dimensional semisimple and cosemisimple
Hopf algebra, and let M be a right (A, H)-Hopf module. Then
Ext∗
B(M, M ) ∼= Ext∗
A(M, M )#H ∗.
For an A-module M , the algebra Ext∗
B(M ·, M ·) may not be a smash product
A(M ·, M ·) with H ∗. However, if the Hopf algebra H is unimodular, then
B(M ·, M ·) is an H ∗-
A(M ·, M ·). Firstly, we make explicit the exact
of Ext∗
we will see that in certain cases the graded algebra Ext∗
Galois graded extension of Ext∗
meaning of a Galois graded extension.
Let E be a Z-graded algebra, H a Hopf algebra. We say that E is a graded
right H-comodule algebra if E is a right H-comodule algebra and the H-
coaction on E respects the grading of E, that is; for x ∈ En, ρ(x) ∈ En ⊗ H.
Let D = EcoH. Then D is a graded subalgebra of E. E/D is called an
H-Galois graded extension if the canonical map [10]
E ⊗D E −→ E ⊗ H,
is bijective.
We have the following observation.
x ⊗ x′ 7→X xx′
(0) ⊗ x′
(1)
Lemma 2.4. Let E/D be an H-Galois graded extension. If E = ⊕n≥0En is
positively graded, then E0/D0 is an H-Galois extension.
For graded right E-modules V and W , let HomE(V, W ) = ⊕i∈ZHomi
E(V, W ),
where Homi
E(V, W ) is the set of all the graded E-module morphisms of degree
i, and End(VE) denotes the sum of all the graded endomorphisms. Then
HomE(V, W ) is a graded vector space and End(VE) is a graded algebra. We
use Ext∗
E(−, −) to denote the derived functor of HomE(−, −). Note that
Extn
E(V, W ) is a graded vector space for all n ∈ Z, where the grading is induced
from the gradings of V and W . Hence Ext∗
E(V, W ) is a bigraded vector space.
To avoid possible confusions, following [6], we say an element x in Extn
E(V, W )
has ext-degree n when we ignore the grading induced from the gradings of V
and W .
For H-Galois graded extensions, we have the following result which cor-
responds to [2, Theorem 1.2]. The proof is exactly the same as that of [2,
Theorem 1.2] except that we should keep in mind that everything needs to
preserve the gradings.
Lemma 2.5. Let H be a finite dimensional Hopf algebra. The following are
equivalent:
DERIVED H -MODULE ENDOMORPHISM RINGS
7
(i) E/D is an H-Galois graded extension.
(ii) (a) The natural map E#H ∗ −→ End(ED) is an isomorphism of graded
algebras.
(b) ED is a finitely generated projective graded D-module.
(iii) As a left graded E#H ∗-module, E is a generator of the category of
graded E#H ∗-modules.
Remark 2.6. If the H-Galois extension A/B in Theorem 2.1 (resp. 2.2) is
H-Galois graded extension and M · is a bounded above complex of graded
A-modules (resp. of graded right (A, H)-Hopf modules), then the result still
holds when the functor Ext is replaced by Ext. Moreover, one can check that
the isomorphisms in the theorems are bigraded morphisms. In particular, the
isomorphism in Corollary 2.3 has the following form, which we will use later,
Ext∗
B(M, M ) ∼= Ext∗
A(M, M )#H ∗,
where the bigrading of the right hand is induced from the bigrading of Ext∗
A(M, M ).
For later use, we need to generalize some results in [3, 13].
Lemma 2.7. Let H be a finite dimensional Hopf algebra, and let A/B be an
H-Galois extension. If M · ∈ D−(A) and N · is a complex of right B-modules,
then there is an isomorphism in D(k)
RHomB(M ·, N ·) ∼= RHomA(M ·, N · ⊗L
B A).
Proof. Replace M · and N · by complexes P · and Q· of projective A-modules
respectively. Since A/B is an H-Galois extension, AB is a projective mod-
ule. Hence P · and Q· are complexes of projective B-modules. We have
RHomB(M ·, N ·) = HomB(P ·, Q·), and RHomA(M ·, N ·⊗L
A).
Following [3], we define a map
BA) = HomA(P ·, Q·⊗B
(2)
ξ : HomB(P ·, Q·) −→ HomA(P ·, Q· ⊗B A)
i ) ⊗B Y t
by ξ(f )(p) = P f (pX t
i where t is a nonzero right integral in H. We
claim that ξ is an isomorphism of complex of vector spaces. By essentially
the same calculations as in the proof of [3, Theorem 5], one sees that ξ is
bijective. What left to show is that ξ is compatible with the differentials. Now
8
JI-WEI HE, FRED VAN OYSTAEYEN AND YINHUO ZHANG
for f ∈ HomB(P ·, Q·) and p ∈ P ·, we have
ξ(d(f ))(p) = ξ(dQ· ◦ f − (−1)f f ◦ dP ·)(p)
i )) ⊗B Y t
= P dQ· (f (pX t
= dQ·⊗B A(P f (pX t
= dQ·⊗B A ◦ ξ(f )(p) − (−1)f ξ(f ) ◦ dP · (p)
= d(ξ(f ))(p).
i ) ⊗B Y t
i −P(−1)f f (dP ·(pX t
i ) − (−1)f P f (dP ·(p)X t
i )) ⊗B Y t
i
i ) ⊗B Y t
i
Hence we get the desired result.
(cid:3)
A(M · ⊗L
B A, M · ⊗L
B A) ∼= Ext∗
B(M ·, M ·) and D = Ext∗
Let H be a finite dimensional semisimple Hopf algebra. By Theorem 2.1,
we have K = Ext∗
B(M ·, M ·)#H. Write E =
Ext∗
A(M ·, M ·). Then E is a graded left K-module
and a graded right D-module. Since HomB(P ·, P ·) is a dg H-module alge-
bra, it is a left dg HomB(P ·, P ·)#H-module. Let U = HomB(P ·, P ·)H =
HomA(P ·, P ·) be the fixed dg subalgebra of HomB(P ·, P ·). Then HomB(P ·, P ·)
is a dg HomB(P ·, P ·)#H-U -bimodule. By taking the cohomology, we obtain
that E = H · HomB(P ·, P ·) is a graded left K- and right D-bimodule since
K ∼= E#H and D ∼= H ·U .
Proposition 2.8. With notation as above, there is a graded bimodule isomor-
phism
KED ∼= K Ext∗
B(M ·, M · ⊗L
B A)D.
B(M ·, M · ⊗L
Proof. By Lemma 2.7, E ∼= Ext∗
B A) as graded spaces. It suffices
to prove that the isomorphism is a bimodule isomorphism. Going back to the
morphism defined by (2) in the proof of Lemma 2.7, one sees that ξ is a dg
right HomA(P ·, P ·)-module morphism. We show that ξ : HomB(P ·, P ·) −→
HomA(P ·, P · ⊗B A) is a left dg HomB(P ·, P ·)#H module morphism. Note
that the left dg HomB(P ·, P ·)#H module action is given by the pullback of
dg algebra morphism ϕ defined by (1) in the proof of Theorem 2.1. Now for
g, f ∈ HomB(P ·, P ·), h ∈ H and p ∈ P ·, we have
ξ((g#h) · f )(p) = ξ(g ◦ (h · f ))(p)
Since A/B is an H-Galois extension and H is semisimple, it is not hard to check
in A ⊗B A ⊗B A.
j ⊗B Y Sh
j ⊗B Y t
j ⊗B Y h
j Y t
i
i X Sh
(see [13]) P X t
Then we obtain
i ) ⊗B Y t
i
j )Y Sh
j
) ⊗B Y t
i .
i X Sh
= P g ◦ (h · f )(pX t
= P g(f (pX t
i =P X t
i ⊗B X h
ξ((g#h) · f )(p) = P g(f (pX t
= ϕ(g#h) ◦ ξ(f )(p)
= (g#h) · ξ(f )(p).
i )X h
j )) ⊗B Y h
j Y t
i
DERIVED H -MODULE ENDOMORPHISM RINGS
9
Therefore ξ is a dg bimodule isomorphism. By taking the cohomologies of
HomB(P ·, P ·) and HomA(P ·, P · ⊗B A) respectively we arrive at the desired
graded bimodule isomorphism.
(cid:3)
Let T and D be triangulated categories. Let F : T −→ D and G : D −→ T
be a pair of adjoint exact functors with F left adjoint to G. Recall that the
Auslander class is the subcategory of T :
A(T ) = {X ∈ T the adjunction map X −→ GF X is isomorphic},
and the Bass class is the subcategory of D:
B(D) = {Y ∈ Dthe adjunction map F GY −→ Y is isomorphic}.
It is well known that both the Auslander class and Bass class are thick trian-
gulated subcategories of the respective triangulated categories, and the adjoint
functors F and G induce a pair of inverse equivalences between triangulated
categories A(T ) and B(D).
The following lemma is straightforward.
Lemma 2.9. For M · ∈ D−(A), if M · ⊗L
sum of M ·, then Ext∗
Ext∗
A(M ·, M ·)-module.
A(M ·, M · ⊗L
B A is a direct summand of finite direct
B A) is a finitely generated graded projective
Let P · ∈ D−(A) be a complex of projective A-modules. We regard A as
a dg algebra concentrated in degree 0. Then P · is a dg right A-module. Let
R = HomA(P ·, P ·) be the dg algebra of endomorphisms. Then P · is a left
dg R-module, and it is also a dg R-A-bimodule. Therefore we have an exact
functor
RHomA(P ·, −) : D(A) −→ Ddg(R),
where Ddg(R) is the derived category of right dg R-modules. This functor is
naturally left adjoint to the functor
− ⊗L
R P · : Ddg(R) −→ D(A).
Clearly the dg A-module P · lies in the Auslander class. Since the Auslander
class is a thick triangulated subcategory, all the direct summands of finite
direct sums of copies of P · belong to the Auslander class.
Let M be a right A-module. Denote by add(M ) the category of A-modules
isomorphic to direct summands of finite sums of M . Recall that E = Ext∗
D = Ext∗
A(M, M ) and K = E#H.
B(M, M ),
10
JI-WEI HE, FRED VAN OYSTAEYEN AND YINHUO ZHANG
Theorem 2.10. Let H be a finite dimensional semisimple Hopf algebra, and
A/B be an H-Galois extension. For a right A-module M , the following are
equivalent.
(i) M ⊗B A ∈ add(M ).
(ii) EndB(M )/ EndA(M ) is an H ∗-Galois extension.
(iii) E/D is an H ∗-Galois graded extension.
Moreover, if M ⊗B A ∈ add(M ), then D and E#H are graded Morita
equivalent.
Proof. The equivalence between (i) and (ii) was proved in [13, Theorem 2.4].
It remains to show that (i) and (iii) are equivalent.
Assume that E/D is an H ∗-Galois graded extension. Since E = Ext∗
B(M, M )
B(M, M ) is an H ∗-Galois extension of
B(M, M ), that is; EndB(M )/EndA(M ) is an H ∗-Galois extension. Hence
is positively graded, by Lemma 2.4, Ext0
Ext0
M ⊗B A ∈ add(M ).
Conversely, assume M ⊗B A ∈ add(M ). By Proposition 2.8, KED ∼=
B A)D as graded K-D-bimodules. It follows from Lemma 2.9
B A), and hence E, is a finitely generated graded right
B(M, M ⊗L
B(M, M ⊗L
K Ext∗
that Ext∗
D-module.
Now we prove that End(ED) is isomorphic to the graded algebra K = E#H.
Let P · ∈ D−(A) be a projective resolution of MA. Viewing A as a dg algebra
concentrated in degree 0, we see that P · is a dg R-A-bimodule, where R =
HomA(P ·, P ·). As mentioned before, (RHomA(P ·, −), − ⊗L
R P ·) is a pair of
adjoint functors; and M ∼= P · lies in the Aulander class of the adjoint pair.
Therefore add(M ) is contained in the Auslander class. We have the following
graded algebra isomorphism:
Li∈Z HomDdg(R)(RHomA(P ·, M ⊗B A), RHomA(P ·, M ⊗B A)[i])
∼= Li∈Z HomD(A)(M ⊗B A, M ⊗B A[i]).
Since P · is a projective resolution, RHomA(P ·, M ⊗B A) ∼= HomA(P ·, P · ⊗B A).
Since M ⊗BA ∈ add(M ), P ·⊗BA is homotopically equivalent to Q·. Thus there
exist an integer n and some complex ¯Q· such that Q· ⊕ ¯Q· is homotopically
It follows that HomA(P ·, P · ⊗B A) ∼= HomA(P ·, Q·).
equivalent to (P ·)⊕n.
Moreover, HomA(P ·, Q·) is a homotopically projective dg R-module. Now we
DERIVED H -MODULE ENDOMORPHISM RINGS
11
have the following graded algebra isomorphisms:
Li∈Z HomDdg(R)(RHomA(P ·, M ⊗B A), RHomA(P ·, M ⊗B A)[i])
∼= H · RHomR(RHomA(P ·, M ⊗B A), RHomA(P ·, M ⊗B A))
∼= H · HomR(HomA(P ·, Q·), HomA(P ·, Q·))
(a)
∼= HomH ·R(H · HomA(P ·, Q·), H · HomA(P ·, Q·))
= HomD(Ext∗
∼= End(ED).
A(M, M ⊗B A), Ext∗
A(M, M ⊗B A))
The isomorphism (a) holds because HomA(P ·, Q·) is a direct summand (in the
homotopy category) of a free dg R-module. More precisely, in the homotopy
category of right dg R-modules, we have HomA(P ·, Q·) ⊕ HomA(P ·, ¯Q·) ∼=
HomA(P ·, Q· ⊕ ¯Q·) ∼= HomA(P ·, P ·⊕n) ∼= R⊕n. Let X = HomA(P ·, Q·) and
Y = HomA(P ·, ¯Q·). We have the following commutative diagram:
H · HomR(X ⊕ Y, X)
/ HomH ·R(H ·(X ⊕ Y ), H ·X)
∼=
∼=
H · HomR(R⊕n, X)
∼= /
/ HomH ·R(H ·(R⊕n), H ·X).
Since all the morphisms in the diagram are natural, we obtain a natural iso-
morphism H · HomR(X ⊕ Y, X) ∼= HomH ·R(H ·(X ⊕ Y ), H ·X), which implies
the isomorphism (a): H · HomR(X, X) ∼= HomH ·R(H ·X, H ·X).
On the other hand, we have
HomD(A)(M ⊗B A, M ⊗B A[i]) ∼= Ext∗
A(M ⊗B A, M ⊗B A) ∼= E#H.
Mi∈Z
Thus End(ED) ∼= E#H as graded algebras. Now applying Lemma 2.5, we
obtain that E/D is an H ∗-Galois extension.
Assume now M ⊗B A ∈ add(M ). Then E/D is an H ∗-Galois extension. By
Lemma 2.5(iii), E#HE is a graded generator. Since H is semisimple, E is a
graded projective E#H-module. Therefore, E is finitely generated projective
generator of the category of graded E#H-modules. Thus E#H is graded
Morita equivalent to D (see [5]).
(cid:3)
We next want to show that Galois extensions over a semisimple and cosemisim-
ple Hopf algebra preserve Koszul property. Let us recall the definition of Koszul
algebras. Let A = ⊕n≥0An be a graded algebra such that Ai is finite dimen-
sional for all i ≥ 0, A0 is semisimple and AiAj = Ai+j for all i, j ≥ 0. Let
N ≥ 2 be an integer. A graded A-module M is called an N -Koszul module
(see, [6] for instance) if M has a graded projective resolution:
· · · −→ P −n −→ · · · −→ P −1 −→ P 0 −→ M −→ 0
/
12
JI-WEI HE, FRED VAN OYSTAEYEN AND YINHUO ZHANG
such that the graded projective module P −n (n > 0) is generated in degree
2 N if n is even, or n−1
n
2 N + 1 if n is odd. If the trivial graded right A-module
A0 (via the projection) is N -Koszul, then A is called an N -Koszul algebra.
Theorem 2.11. Let H be a finite dimensional semisimple and cosemisimple
Hopf algebra, let A = ⊕n≥0An be a graded right H-module algebra such that
Ai is finite dimensional for all i ≥ 0, and let B = AcoH . Assume that A/B
is an H-Galois graded extension. If B is an N -Koszul algebra, then A is an
N -Koszul algebra.
Proof. By the assumption, B0 is a finite dimensional semisimple algebra. Since
A/B is an H-Galois graded extension, A0/B0 is an H-Galois extension by
Lemma 2.4. Now that H is semisimple implies that A0#H is semisimple
since it is Morita equivalent with B0. Since H is cosemisimple, (A0#H)#H ∗
is semisimple by Maschke's theorem. It follows from the Morita equivalence
between A0 and (A0#H)#H ∗ that A0 is a semisimple algebra. As a right B0-
module, A0 = B0 ⊕ S for some finite dimensional B0-module S. By Remark
2.6, we have an isomorphism of bigraded algebras
(3)
Ext∗
B(A0, A0) ∼= Ext∗
A(A0, A0)#H.
B(S, B0) and Ext1
B(B0, S), (Ext2
Since B is N -Koszul, the graded space Ext1
B(B0, B0) is concentrated in de-
gree 1, and Ext2
B(B0, B0) is concentrated in degree N (see [6]). Therefore
Ext1
B(B0, S), (Ext1
B(S, S), respectively) is concentrated in
degree 1, and Ext2
B(S, S) respectively) is con-
centrated in degree N as S is a direct summand of of a finite sum of B0. Hence
Ext1
B(S, S) is
concentrated in degree 1. Since we already know that A0 is semisimple, A
must be generated in degrees 0 and 1. Similarly, we see that Ext2
B(A0, A0) is
concentrated in degree N . Thus A must be homogeneous in the sense of [6].
B(A0, A0) ∼= Ext1
B(S, B0) and Ext2
B(B0, S) ⊕ Ext1
B(B0, B0) ⊕ Ext1
B(B0, S) ⊕ Ext1
B(B0, B0), Dn = Extn
By the isomorphism (3), the graded algebra Ext∗
A(A0, A0) is generated in
ext-degrees 0, 1 and 2 if and only if Ext∗
B(A0, A0) is. Next we show that
Ext∗
B(A0, A0) is generated in ext-degrees 0, 1 and 2. For convenience, we
let En = Extn
B(S, B0) and V n =
Extn
B(B0, S). Write E and D for the graded algebras ⊕n≥0En and ⊕n≥0Dn
respectively. Similarly let U and V be the respective graded E-D-bimodule
⊕n≥0U n and D-E-bimodule ⊕n≥0V n. Since S as a graded left E-module is
a direct summand of of a finite sum of B0, U is generated in ext-degree 0.
Similarly, V is generated in ext-degree 0 as a graded right E-module. The
B(S, S), U n = Extn
DERIVED H -MODULE ENDOMORPHISM RINGS
13
graded algebra Ext∗
B(A0, A0) is isomorphic to the following matrix algebra
E U
V D !
where the products U V and V U are the Yoneda products of the extensions.
Now if B is N -Koszul, then E is generated by E0, E1 and E2. Since V , as a
graded right E-module, is generated in degree 0, we have V = V 0E. Hence each
element of V can be written as a sum of some multiplications of elements in E0,
E1, E2 and V 0. Similarly, each element of U can be written as a sum of some
multiplications of elements in E0, E1, E2 and U 0. Since B0 is semisimple, we
may assume that there is a semisimple B0-module T such that S ⊕ T ∼= ⊕n
i=1Qi
as right B0-modules, where Qi = B0 for all i = 1, . . . , n. Notice that the
actions of an element in ⊕i≥1Bi on both sides of the foregoing isomorphism
are trivial. Thus the isomorphism is in fact a right B-module isomorphism. Let
ι : S −→ ⊕n
i=1Qi −→ S be the projection
B(S, S) −→ Extn
map. Then π ◦ι = id. For any n ≥ 0, let ιn
i=1Qi)
i=1Qi) −→ Extn
ext : Extn
be the map induced by ι, and πn
B(S, S) be the
map induced by π. Then πn
B(S, S),
ext(x) ∈ Extn
i=1 Extn
we have y := ιn
i=1Qi) = ⊕n
B(S, Qi). Hence y can be
written as (y1, y2, . . . , yn) for some yi ∈ Extn
B(S, B0) = U n (i = 1, . . . , n).
While π ∈ HomB0(⊕n
S, π is corresponding to a sequence
of elements (s1, . . . , sn) of S. Now x = πn
i=1 siyi,
where the multiplication siyi is the Yoneda product of the extensions, that is
to say, x may be written as a sum of multiplications of some elements in V 0
and some elements in U n. Summary, we have proved that if E is generated in
ext-degrees 0, 1 and 2, then the graded matrix algebra
i=1Qi be the inclusion map, and π : ⊕n
ext : Extn
B(S, ⊕n
ext = id. For any element x ∈ Dn = Extn
ext(y) = Pn
ext◦ιn
B(S, ⊕n
B(S, ⊕n
i=1Qi, S) = ⊕
n copies
ext ◦ ιn
ext(x) = πn
E U
V D !
is also generated in degrees 0, 1 and 2, or equivalently, the graded algebra
Ext∗
B(A0, A0) is generated in ext-degrees 0, 1 and 2. Hence the graded algebra
Ext∗
A(A0, A0) is generated in ext-degree 0, 1 and 2. Now applying [6, Theorem
4.1], we obtain that A is N -Koszul.
(cid:3)
At this moment, we don't know whether the converse of the theorem above
is true. However, for some special cases, the converse holds.
Let A = ⊕n≥0An be a positively graded Hopf algebra such that A0 is a
semisimple and cosemisimple Hopf algebra, Ai is finite dimensional for all i ≥ 0
and AiAj = Ai+j. Let H = A0. Then the natural projection p : A −→ H is a
Hopf algebra map; and A becomes a graded right H-comodule algebra through
14
JI-WEI HE, FRED VAN OYSTAEYEN AND YINHUO ZHANG
the projection p. Let B = AcoH. Then B is a positively graded algebra with
B0 ∼= k. Now we assume that A/B is a graded H-Galois extension. View A0
as a right A-module via the projection p. We have the following corollary.
Corollary 2.12. With the notation as above, A is a Koszul algebra if and only
if the coinvariant subalgebra B is.
B(A0, A0) is generated in ext-degrees 0, 1 and 2 if and only if Ext∗
Proof. From the proof of the theorem above we see that the graded algebra
Ext∗
A(A0, A0)
is generated in ext-degrees 0, 1 and 2. Since B0 = k, A0 as a graded B-module
is a finite sum of copies of k. Hence Ext∗
B(A0, A0) is isomorphic to a matrix
algebra of the graded algebra Ext∗
B(A0, A0) is generated
in ext-degrees 0, 1 and 2 if and only if Ext∗
B(k, k) is. Also from the proof of
the theorem above, we see that B is a homogeneous algebra. Now the result
follows from [6, Theorem 4.1].
(cid:3)
B(k, k). Hence Ext∗
Let A be a cocommutative Hopf algebra over an algebraic closed field of
characteristic 0. It is well-known that A ∼= U (P (A))#kG as Hopf algebras,
where P (A) is the Lie algebra of primitive elements of A and U (P (A)) is
the universal enveloping algebra of P (A), and G is the group of group-like
elements of A. Let gr(A) be the graded Hopf algebra associated to the coradical
filtration of A. Then gr(A) ∼= gr(U (P (A)))#kG.
In particular, gr(A) is a
Galois (graded) extension over kG. If dim(P (A)) < ∞, then gr(U (P (A))) is
isomorphic to a polynomial algebra. Hence we have the following corollary,
which results in that a cocommutative Hopf algebra is a PBW-deformations
of a Koszul algebra. By the Koszul property, we may lift certain homological
properties of gr(A) to A, e.g. Calabi-Yau property, see [7].
Corollary 2.13. Let A be a cocommutative Hopf algebra over an algebraic
closed field of characteristic 0 such that G is a finite group and dim(P (A)) <
∞. Then gr(A) is a Koszul algebra.
We end this note with an example of a cocommutative Hopf algebra which
satisfies Corollary 2.12.
Example 2.14. Let Q be the following quiver.
0
s
⑥⑥
⑦⑦
1
s
x0
y0
x1
y1
As an algebra, let A = kQ/I where the ideal I is generated by the relations
{x0y1 −y0x1, x1y0 −y1x0}. Denote by e0 and e1 the idempotents corresponding
DERIVED H -MODULE ENDOMORPHISM RINGS
15
to the vertices. We define a coproduct and a counit on A so that it becomes
a Hopf algebra. It is enough to define the comultiplication and the counit on
the generators.
∆(e0) = e0 ⊗ e0 + e1 ⊗ e1,
∆(e1) = e1 ⊗ e0 + e0 ⊗ e1,
∆(x0) = e0 ⊗ x0 + e1 ⊗ x1 + x0 ⊗ e0 + x1 ⊗ e1,
∆(y0) = e0 ⊗ y0 + e1 ⊗ y1 + y0 ⊗ e0 + y1 ⊗ e1,
∆(x1) = e1 ⊗ x0 + x1 ⊗ e0 + e0 ⊗ x1 + x0 ⊗ e1,
∆(y1) = e1 ⊗ y0 + y1 ⊗ e0 + e0 ⊗ y1 + y0 ⊗ e1,
and
ε(e0) = 1,
ε(x0) = ε(x1) = ε(y0) = ε(y1) = 0.
ε(e1) = 0,
One can check that A is a cocommutative bialgebra. In fact, A is also a Hopf
algebra, and the antipode is given by
S(e0) = e0,
S(x0) = −x1,
S(x1) = −x0,
S(e1) = e1,
S(y0) = −y1,
S(y1) = −y0.
It is clear that H = A0 ∼= kZ2 is a semisimple and cosemisimple Hopf algebra.
One can also check that B = AcoH ∼= k[x, y] is a Koszul algebra. Thus A is a
Koszul algebra.
Acknowledgement. The first named author is supported by an FWO-grant
and NSFC (No. 10801099).
References
[1] L.L. Avramov, H.-B. Foxby and S. Halperin, Differential Graded Homological Algebra,
preprint.
[2] M. Cohen, D. Fischman and S. Montgomery, Hopf Galois extensions, smash products,
and Morita equivalence, J. Algebra 133 (1990), 351 -- 372.
[3] Y. Doi, Hopf extensions of algebras and Maschke type theorems, Israel J. Math. 72
(1990), 99 -- 108.
[4] Y. F´elix, S. Halperin and J.-C. Thomas, Rational Homotopy Theory, Grad. Texts Math.
205, Springer-Verlag, New York, 2001.
[5] R. Gordon and E.L. Green, Graded Artin algebras, J. Algebra 76 (1982), 111 -- 137.
[6] E.L. Green, E.N. Marcos, R. Mart´ınez-Villa and P. Zhang, D-Koszul algebras, J. Pure
Appl. Algebra 193 (2004), 141 -- 162.
[7] J.-W. He, F. Van Oystaeyen and Y. Zhang, Cocommutative Calabi-Yau Hopf algebras
and deformations, arXiv:0906.1911, to appear in J. Algebra.
[8] S. Konig and A. Zimmermann, Derived Equivalences for Group Rings, Lect. Notes Math.
1685, Spring-Verlag, New York, 1998.
[9] S. Priddy, Koszul resolutions, Trans. Amer. Math. Soc. 152 (1970), 39 -- 60.
16
JI-WEI HE, FRED VAN OYSTAEYEN AND YINHUO ZHANG
[10] P. Schauenburg, Hopf-Galois extensions of graded algebras, Proceedings of the 2nd Gauss
Symposium. Conference A: Mathematics and Theoretical Physics (Munich, 1993), 581 --
590, Sympos. Gaussiana, de Gruyter, Berlin, 1995.
[11] H.J. Schneider, Hopf Galois extensions, crossed products, and clifford theory, Advances
in Hopf algebras (Chicago, IL, 1992), 267 -- 297. Lecture Notes in Pure and Appl. Math.
158, Dekker, New York, 1994.
[12] D. Stefan, Hochschild cohomology on Hopf Galois extensions, J. Pure Appl. Algebra 103
(1995), 221 -- 233.
[13] F. Van Oystaeyen and Y. Zhang, H-module endomorphism rings, J. Pure Appl. Algebra
102 (1995), 207 -- 219.
J.-W. He
Department of Mathematics, Shaoxing College of Arts and Sciences, Shaoxing
Zhejiang 312000, China
Department of Mathematics and Computer Science, University of Antwerp,
Middelheimlaan 1, B-2020 Antwerp, Belgium
E-mail address: [email protected]
F. Van Oystaeyen
Department of Mathematics and Computer Science, University of Antwerp,
Middelheimlaan 1, B-2020 Antwerp, Belgium
E-mail address: [email protected]
Y. Zhang
Department WNI, University of Hasselt, 3590 Diepenbeek, Belgium
E-mail address: [email protected]
|
1601.08126 | 1 | 1601 | 2016-01-29T14:27:05 | Symmetry vs Symmetry | [
"math.RA"
] | This preprint deals with the symmetry of parametrized families of systems and the changes therein as the parameter changes. There are (at least ?) two kinds of symmetry: generic and specific which behave in almost totally opposite ways as the parameter changes: generic symmetry has links with entropy while specific symmetry has to do with symmetry breaking as it is usually understood in physics. | math.RA | math | Notes and reflections on the gain and loss of symmetries as a system evolves.
Symmetry vs symmetry
Symmetry vs entropy.
Michiel Hazewinkel
<[email protected]>
Burg. ‘s Jacob laan 18
NL-1401BR BUSSUM
The Netherlands
Abstract. This preprint deals with the symmetry of parametrized families of
systems and the changes therein as the parameter changes. There are (at least ?) two kinds of
symmetry: generic and specific which behave in almost totally opposite ways as the
parameter changes: generic symmetry has links with entropy while specific symmetry has to
do with symmetry breaking as it is usually understood in physics.
Classification (MSC 2010): 16W20
1. Introduction. These notes, reflections and (mainly) examples are concerned with the
gain and loss of symmetry as a parameter varies. More precisely the setting is that of a family
of objects
typically a smooth manifold. And one is interested in how the symmetry of
varies. For instance
could be the family of three dimensional algebras
, where P, the parameter space, is
(continuously) parametrized by
changes as t
(1.1)
where the symmetry of an algebra
automorphisms of
is by definition the group
of algebra
. This example will be described in some detail below.
It seems to me that to discuss gain and loss of symmetry precisely one must first specify
a GPS (Group of Possible Symmetries). In the example just mentioned a natural GPS is
. The use of the indefinite article ‘a’ in the previous two sentences is deliberate. As
the example of four lines in the plane, also discussed below in some detail, shows, it is
entirely possible to have more than one, in this case two, GPS’s. Moreover the gain and loss
of symmetry behaviors in the two cases of this example are precisely opposite. This could go
some way to clarify the different viewpoints of Ilya Prigogine and Joe Rosen concerning gain
and loss of symmetry (symmetry breaking) as described in the book review [4 Lin].
I term the two different kinds of symmetry generic symmetry and specific symmetry (or
design symmetry). See [3 Hazewinkel], section 3 for some early discussion of the matter.
Generic symmetry is almost always there and characteristically is subject to sudden
decreases. Nontrivial specific symmetry is almost never there and tends to have sudden
increases. In both cases the word ‘sudden’ means that these phenomena take place on a subset
Att!PAtAtAt=C[X]/(X(X!t)(X!1)),!!!t"CAtAut(At)AtGL(C3) The following well-known theorem of Louis Michel is very much a theorem about
of parameter space of codimension
specific (design) symmetry.The setting is that of a (Lie) group acting smoothly on a smooth
manifold
.
,
(1.2)
The symmetry of an
is, by definition, the isotropy subgroup
there
The theorem, see [7 Michel], now says that if G is compact, then for every
is a neighborhood U of m such that
for all
is, up to conjugacy, larger than, or equal to
.
Note that the GPS in this setting is obviously the group G.
This is very much a theorem about specific (design) symmetry The theorem should
generalize considerably. But it is not universally true as is shown by example 2.5 of [3
Hazewinkel].
On the other hand the Curie-Rosen symmetry principle as well as positive correlation
between entropy and symmetry (number of symmetries) as discussed in [4 Lin; 5 Lin; 9
Rosen; 10 Rosen], see also [1 Ben-Naim], belong to the domain of generic symmetry. In this
connection there is the famous remark made by Pierre Curie in 1894: “ C’est la dissymmétrie
qui crée le phénomène”.
2. Configurations of four lines in the plane.
Most of this note is concerned with algebras A over the complex numbers with
, the automorphism group of A (as an algebra), interpreted as the symmetry of A.
First, however, I will try to illustrate the idea of generic symmetry vs specific (design)
symmetry by means of the geometric example of four lines in the plane. These four line
configurations (4LC’s) form an eight dimensional ‘manifold’ M. Generically for such a 4LC
every two lines intersect in precisely one point and no three lines meet in one point as in the
example depicted in figure (2.1) below. Let
denote the subset of all such 4LC’s.
(2.1)
!1G!M"#$#Mgm=!(g,m)m!MGm={g!G:gm=m}m!MGmG!m!m"UAut(A)V!M . I.e.
has codimension at least 1. Moreover for every
is open and
Then
neighborhood U of m such that
The idea is now to look at a 4LC as an intersection system. The natural GPS to take is
then the group of permutations of the four lines and the symmetry subgroup of a given 4LC
such that line i and line j intersect or coincide if
m consists of those permutations
and only if such is the case for lines
and
. For instance if for that given m lines
1 and 2 are parallel and distinct and and lines 3 and 4 are a second such pair, different from
the first pair then the (isotropy) symmetry subgroup of m is the Klein four group
there is an open
is closed.
.
The system of subgroups
there is a neighborhood U such that for every
.
Note that this is almost exactly the opposite property as that which obtains in the case of the
Michel theorem mentioned in the introduction above. This illustrates the way generic
symmetry behaves.
has the following property. For every
one has the inclusion
Next let G be the group of Euclidean motions of the plane (generated by rotations
translations and reflections. This group acts of course naturally on the manifold M of all
4LC’s. So apart from the fact that this G is not compact the setting is the one of the Louis
Michel theorem. Still the conclusions hold in this case. This of course pertains to design
symmetry.
To further illustrate the opposite behaviour of design symmetry and generic symmetry
consider the example of the family of 4LC’s parametrized by a parameter t depicted in the
figure on the next page. To fix ideas the parameter can be taken to run from 1/2 to 1. For the
description in words of this example the plane is coordinatized. These coordinates are not
themselves part of the example.
as t rises to one rotates clockwise around (2,4) to reach a vertical position for
indicated by the dashed vertical line in the picture below.
Line 1 at parameter value t smaller than 1 runs south-east through the point (2,4) and
Line 2 is the X-axis for all t.
Line 3 is the Y-axis for all t.
Line 4 runs north-east through (0,4) and rotates clockwise as t rises to reach a
Then for
the generic symmetry of the 4LC depicted is maximal and equal to
horizontal position at
.
while the design symmetry is minimal, viz only the identity. But at
symmetry decreases to the group off eight elements generated by switching the two lines of
one of the two pairs of parallel lines and switching the two pairs of parallel lines, and the
design symmetry suddenly increases to the Klein four group (the symmetry group of a
rectangle).
the generic
M\Vm!VU!VVM\V!"S4!(i)!(j){(12),(34),(12)(34),(1)}!S4Sm,m!Mm!M!m"USm!S"mt=1t=1t!1S4t=1 L4(t)
L4(1)
L2
L3
L1(1)
L1(t)
3. Monogenic finite dimensional algebras over the complexes.
All but one of the further examples in this note concern families of monogenic (i.e. one
generator) algebras over the complexes( or, basically any algebraically closed field) and their
automorphism groups (as algebras) and how these groups can change as a parameter varies.
This section contains some preliminary material on these algebras.
A finite dimensional monogenic algebra over the complexes is of course necessarily of
the form
where
is a monic polynomial of degree n with n roots
.
3.3. The multiplicity free case. In this subsection it is supposed that all the
are distinct, i.e. that
is multiplicity free. Define for all
(3.1)
(3.2)
(3.4)
The first result is that the
are orthogonal idempotents. That they are orthogonal, i.e. that
for
is immediate as such a product is divisible by
. As to the
idempotency, observe that for
Also observe that for
It follows that the
polynomial
points
is of degree
. So it must be equal to 1.
are all different and that they sum to one. As to this last point the
and takes the value 1 at the n distinct
The result is of course that in the multiplicity free case
This is pretty elementary algebraic geometry and can be concluded without the explicit
(3.5)
A=C[X]/(f(X))f(X)=(i=1n!X"zi) zi,!i!I={1,!!,!n}zi,!i!If(X)i!Iei=(X!zj)(zi!zj)j"I,j#i$eieiej=0i!j(X!zi)i"I#j!iX!zjzi!zjei=X!zi+zi!zjzi!zjei=X!zizi!zjei+zi!zjzi!zjei=0+ei=eij!iei(zj)=0,!!!ei(zi)=1ei e1+e2+!!!+enn!1 z1,!!!,zn A=C[X]/(f(X))!Cn expressions for the idempotents.
Obviously the symmetry (automorphism) group of
is
. This is very much
a matter of generic symmetry (at least within the class of monogenic finite dimensional
algebras over the complex numbers but it seems to me also for larger classes). Indeed if
is monic and has n different roots than so has any monic
sufficiently
near
. Further if
has roots with multiplicity larger than 1 there are monic
arbitrarily close to it that have n distinct roots. These are precisely the kind of
properties for generic symmetry as hinted at in the introduction.
However, in order to study automorphism groups as a parameter varies as in the
example
(3.6)
which will be studied in the next section, it seems that explicit formulas are needed.
What seems to be needed is the transition matrix from the basis .
to the basis given by the n idempotents
this end observe that
of
of (3.4) above. To
and hence
so that the transition matrix is given by the Vandermonde matrix
(3.7)
and its automorphisms for
not
3.8. Remarks on
necessarily multiplicity-free.
sort of complete the picture.
The following remarks are not needed for the examples to be discussed below; they just
For each
consider the algebra
A!CnSnf(X)!C[X]g(X)f(X)f(X)g(X)At=C[X]/(X(X!t)(X!1)) 1,X,X2,!!!,Xn!1A=C[X]/(f(X)) e1,!!,!enXei=X(X!zj)(zi!zj)j"I,j#i$=(X!zi+zi)(X!zj)(zi!zj)j"I,j#i$=0+zi(X!zj)(zi!zj)j"I,j#i$=ziei X=X.1=Xe1+!!!+Xen=z1e1+!!!+znen 1z1!z1n!11z2!z2n!1"""1zn!znn!1"#$$$$$%&'''''A=C[X]/(f(X))f(X) n!N={1,2,!!} (3.9)
is non-reduced).
The acronym FPA stands for ‘fat point algebra’, a term from algebraic geometry. Scheme
theoretically Spec(FPA(n)) is a point with the ‘sheaf of rings’ FPA(n) over it (which for
with the trivial one element group as automorphism group;
, given by
has as automorphism group the group of non-zero complex numbers
; the automorphism group
under multiplication,
of
with the additive groups
6 Mathoverflow] for more details about fat points and their automorphism groups.
the following proposition.
The fat point algebras are the building blocks of the monogenic algebras in the sense of
of complex numbers under addition. See [2 Cooper et al.;
times repeated extension of
is an
for
3.10. Proposition. Let
complex numbers. Then
be a monic polynomial of degree n over the
(3.11)
Here the
are the multiplicities of the roots of
that actually occur and for each i
is the number of distinct roots that occur with multiplicity
.
It is not difficult to describe the automorphisms of an algebra of the form (3.11). Indeed,
be such an automorphism. As such it must take indecomposable idempotents into
let
indecomposable idempotents. So let
be an indecomposable idempotent and let
be the corresponding factor in the direct product decomposition (3.11) of
. Let
and
the corresponding factor in (3.11).
Then
(3.12)
in (3.12) is injective; it is also obviously
is a permutation of the copies
and
As the restriction of an injective morphism, the
surjective. Hence an isomorphism and
of
automorphisms of the algebra (3.11).
that occur in (3.11). Thus one obtains the following description of the
Let
and
a permutation that takes the subsets
automorphism of the algebra (3.11) is now given by such a permutation together with for
each
an isomorphism of the corresponding
with
,
of
into themselves. An
FPA(n)=C[X]/(Xn)n!2FPA(1)=CFPA(2)=C[X]/(X2)Gm(C) X!bX,!!b!C\{0}FPA(n)=C[X]/(Xn)n!3n!2Gm(C)Ga(C)f(X)!C[X] C[X]/(f(X))!FPA(m1)r1!!"!!!FPA(ms)rs,!!!r1m1+!"!+!rsms=nmif(X)rimi!eiFPA(m)jA=C[X]/(f(X))!(ei)=e"iFPA(!m)!jFPA(m)j=eiAei!"#"e$iAe$i=FPA($m)$j!m=!m j!!jFPA(m) r=r1+!!!+rs!"Sr {r1+!!+ru+1,!!!r1+!!!ru+1} u=0,!!,!s!1 {1,!!,!r} i!{1,!!,!r}FPA(m)j=eiAei .
4 Examples of families of algebras and their automorphism groups.
In this section k is generally an algebraically closed field which, if the reader
finds that convenient, can be taken to be the complex numbers.
4.1.The algebra automorphism group
An algebra endomorphism of the algebra
two polynomial giving the image of X mod
.
is given by specifying a degree
The requirement that (4.1.1) is such that
goes to zero mod
then implies that
(4.1.1)
(and that suffices for this condition). The resulting endomorphism
an automorphism iff
. Let G be this group of automorphisms:
Let
Writing
for
the composition law in G is
is
(4.1.2)
(4.1.3)
(4.1.4)
(4.1.5)
.
showing that G is non-commutative. The sets H and N are subgroups and
The group N is in fact a normal subgroup (but H is not normal). Thus G is in fact a
non-trivial extension of H by N. The corresponding action of H on N is given by the
formula
That is, the multiplicative group
multiplication.
acts on the additive group
by
FPA(m)!j=e"(i)Ae"(i)G=Aut(k[X]/(X3))k[X]/(X3)(X3) X!c+aX+bX2X3(X3)c=0 X!aX+bX2a!0 G={X!aX+bX2:a!k\{0},!b!k} H={!a:X!aX:a"k\{0}} N={!b:X!X+bX2:b"k}!a,b X!aX+bX2 !a,b!!"a,"b=(X"a"aX+(a"b+"a2b)X2)G=HN=NH!a"b!a#1="ab H!k\{0} N!k In view of the examples below concerning automorphisms of families of algebras there
is special interest in automorphisms of order 2 and 3. It follows immediately from (4.1.5)
that
and
These formulas generalize:
as is proved by a straightforward induction.
Let
denote the subset of G of automorphisms of order precisely two.
(4.1.6)
(4.1.7)
(4.1.8)
Automorphisms of order 2 when
Thus the automorphisms of order two are the non-identity elements
. In this case a must be 1 or
and hence
of the normal subgroup N.
(4.1.9)
Automorphisms of order 2 when
. If
and so
. If
. Thus there are no automorphisms of order two with
. In this case a must be 1 or
and b can be arbitrary. Thus in this case
Automorphisms of order 3 when
of unity. In this case
and
and k contains no primitive third root
. So b must be zero. Thus in this case
(4.1.11)
(4.1.10)
Automorphisms of order 3 when
of unity. Denote a primitive root of unity by
it must be the case that
so that
and k contains a primitive third root
. Then for an automorphism to be of order 3
in
. When
!a,b2(X)=a2X+(a+a2)bX2!a,b3(X)=a3X+(a2+a3+a4)bX2 !a,bn(X)=anX+(an"1+an+!!!+a2n"2)bX2G2char(k)=2!1=1(a+a2)=0G2={!a,b:a=1,b"0}char(k)!2!1a=1a+a2=2!0b=0a=1a=!1a+a2=0G2={!a,b:a="1,b#k}char(k)=2a=1a2+a3+a4=1!0G3=!char(k)=2!3a3=1a=1,!!3,!or !32a=1a2+a3+a4=1 k and, hence b must be zero. So that gives no automorphisms of order two. on the other
hand if
case
and b can be any element in k. Thus in this
then
(4.1.12)
the union of two (disjoint) congruence classes of N in G.
Automorphisms of order 3 when
of unity. In this case
and so
and
. and k contains no primitive third root
and thus b can be anything
(4.1.13)
Automorphisms of order 3 when
of unity. Denote a primitive third root of unity by
three cases
giving
and k contains a primitive third root
and in all
. In this case
Automorphisms of order 3 when
root of unity. In this case
and
(4.1.14)
. and k contains no primitive third
and so b must be zero to give
(4.1.15)
Automorphisms of order 3 when
and k contains a primitive third root
of unity. Denote such a primitive root of unity by
first case
automorphism of order three of this form. Thus there remains
cases
and so b can be anything, giving (again)
so that b must be zero and thus there is, in this case, no
. In both these
. In this case
. In the
It is worth noting that it never happens that there are 2 automorphisms of order 2 whose
product is of order 3. But in the symmetric group
product of two different elements of order 2 is of order 3. So there is no way to inject
into the group of unity preserving algebra automorphisms of the algebra
.
of permutations on three letters every
(4.1.16)
a=!3!or!!32a2+a3+a4=0G3={!a,b:a"{#3,#32},!b"k}char(k)=3a=1a2+a3+a4=3=0!0G3={!1,b:b"k\{0}}char(k)=3!3a=1,!!3,!or !32a2+a3+a4=0G3={!a,b:a"{#3,#32},!b"k}${!1,b:b%0}char(k)!2,3a=1a2+a3+a4=3!0G3=!char(k)!2,3!3a=1,!!3,!or !32a2+a3+a4=3!0a=!3!or!!32a2+a3+a4=0G3={!a,b:a"{#3,#32},!b"k}S3S3k[X]/(X3) 4.2. The algebra
and its automorphism group.
Here k is a field, which can be taken to be the field of complex numbers (for the
convenience of the reader). The symbol J is used to denote the principal ideal
The algebra
has two orthogonal idempotents, viz.
and
and correspondingly splits into a direct sum of two sub-algebras. As a matter of fact
.
One isomorphism is given by
,!!with inverse
(4.2.1)
(4.2.2)
The unity preserving algebra automorphisms of
are given by
where a is a nonzero element of k. This corresponds to
Thus the (unity preserving) algebra automorphism group of the algebra
the multiplicative group
subgroup isomorphic to
is
. This group is commutative and hence contains no
.
(4.2.3)
4.3. The family of algebras
and their
automorphisms.
Here k is a field, which can be taken to be the field of complex numbers (for the
convenience of the reader). The symbol t is seen as a parameter (with values in k) and
are two unequal elements of k. The question to be investigated is how the
automorphism groups of these algebras change as t varies.
Let
For
this polynomial has two different roots. So then the algebra
is isomorphic to
and its sole non-identity automorphism consists of interchanging the two factors k.
Thus there is a generic symmetry group
.
In slightly more detail there are two orthogonal idempotents, viz.
(4.3.1)
k[X]/(X3!X2)(X3!X2)k[X]/(X3!X2)X21!X2 k[X]/(X3!X2)!!!k[Y]/(Y2)"k !:X!(Y,1) !"1:(Y,0)!X"X2,!!(0,1)!X2k[Y]/(Y2)!k (Y,0)!(aY,0),!!(0,1)!(0,1) X!aX+(1!a)X2k[X]/(X3!X2)k\{0}=Gm(k)S3At=k[X]/(X!tx1)(X!tx2))x1 and x2pt(X)=(X!tx1)(X!tx2)t!0Atk!kS2 and the nontrivial automorphism is given by interchanging these two. At the level of X this
works out as
(4.3.2)
This still makes sense when t becomes zero and gives a nontrivial automorphism as long as
and there is neither gain or loss of symmetry in this case.
there is loss of generic symmetry (but no gain in specific
But if
symmetry).
4.4. The family of algebras
and their
automorphisms.
Here k is a field, which can be taken to be the field of complex numbers (for the
convenience of the reader). The symbol t is seen as a parameter (with values in k). As
before the question to be investigated is how the automorphism groups of these algebras
change as t varies.
(4.4.1)
Let
For
to
this polynomial has three different roots. So then the algebra
, or, in other words, there are three pairwise orthogonal idempotents
and for each i the Pierce decomposition component
because of commutativity) is equal to k. Thus the algebra automorphisms for
(which is equal to
is isomorphic
the t are given by the permutations of these three idempotents and so there is a generic
symmetry group
, he group of permutations on three letters.
The three orthogonal idempotents can be written down explicitly (as in section 3
above). For instance the idempotent corresponding to the root t is equal to
but this is not of immediate use for calculations. What is of great use is the observation that
the transition from the basis
to the basis
formed by the three idempotents is the Vandermonde matrix M of the three
of the algebra (vector space)
roots 0,t,1 with inverse
as follows
e1=X!tx2tx1!tx2,!!!e2=X!tx1tx2!tx1 X!!X+tx1+tx2char(k)!2char(k)=2At=k[X]/(X(X!t)(X!1))pt(X)=X(X!t)(X!1)t!0,1Atk!k!kei,!i=1,2,3eiAteieiAt=AteiS3et=(X!0(X!1)(t!0)(t!1){1,X,X2}At{e0,et,e1}M!1 ,
Thus the coordinates of, respectively
,
are respectively
(4.4.2)
with respect to the basis of idempotents
,
, the column matrices of M.
When t becomes zero the first two roots coincide (become a double root) but remain
different from the third one. Thus, intuitively, one expects that the transposition of the first
two roots (really the corresponding idempotents) might survive as t becomes zero; but that
disappear. Here are
the other four non-identity elements of the generic symmetry group
some explicit calculations. Switching the first two entries of
transforms this vector to
. Thus the coordinates of the image of X under this automorphism are given by
so that
(4.4.3)
For
this becomes
which is indeed an automorphism of
so that the generic symmetry ‘switching the first two roots’ survives
when t becomes zero. Note, however, that (4.4.3) makes no sense for
showing that
this symmetry does not survive when the second and third root become equal (and remain
different from the first).
of order three of the three idempotents.
Now consider the cyclic permutation
This takes
to
. Now
and so this automorphism is given by
M=1001tt2111!"###$%&&&,!!!!M'1=1(1't)tt't200'(1't2)1't21't'1t!"###$%&&&1,!X,!X2{e0,et,e1}(1,1,1)(0,t,1)(0,t2,1)S3(0,t,1)(t,0,1)M!1t01"#$$%&''=tt2!t!11!t2!t1!t"#$$$$$$%&'''''' X!t+t2!t!11!tX+2!t1!tX2t=0 X!!X+2X2A0=k[X]/(X3!X2)t=1(132)(0,!t,!1)(t,!1,!0)Mt!1t10"#$$%&''=1t!t2t2!t3!t(1!t2)+1t(1!t)"#$$$%&''' (4.4.4)
which is undefined for both
and
.
This would appear to be sufficient evidence to show that the generic symmetry
the family of algebra under consideration collapses to a subgroup of order two when t
becomes zero or one.
of
4.5. The family of algebras
and their
Here k is a field, which can be taken to be the field of complex numbers (for the
automorphisms.
convenience of the reader). The symbol t is seen as a parameter (with values in k). Again
the question to be investigated is how the automorphism groups of these algebras change as t
varies.
Let
(4.5.1)
For
to
precisely permuting the three idempotents. Thus there is \a generic symmetry group
this polynomial has three different roots. So then the algebra
and the automorphisms consists of permuti ng the three factors k; more
.
When t becomes zero all three roots become zero. So intuitively it could be the case
is isomorphic
that generic symmetry is preserved.
This time the transition matrix from the basis
idempotents and the inverse of this matrix are
to the basis formed by the
(4.5.2)
(4.5.3)
The coordinates of X in the basis of idempotents
Interchanging the first two basis elements changes this to
are
. Now
.
X!t+1!t+t3t(1!t)X+!1+t!t2t(1!t)X2t=0t=1S3At=k[X]/(X(X!t)(X!t2))pt(X)=X(X!t)(X!t2))t!0,1Atk!k!kS3{1,X,X2}Mt=1001tt21t2t4!"###$%&&&,!!!det(Mt)=(t2't)(t2'0)(t'0)=t4(t'1)Mt!1=1t4(t!1)t5!t400!(t4!t2)t4!t2t2!t!t2t"#$$$%&'''{e0,et,et2}(0,t,t2)(t,0,t2) So this automorphism is given by
(4.5.4)
(4.5.5)
This is not defined for
Now consider the symmetry given by the cyclic permutation that takes
and so this generic symmetry disappears at t equal to zero.
to
. Then
so that his automorphism is given by
which is not defined for
so that also this generic symmetry disappears.
(4.5.6)
Strictly speaking more calculations should be done. But it seems clear that for this
. In return the
family of algebras all of the generic symmetry
specific symmetry increases from trivial to that of the two dimensional group G described in
section 4.1 above.
This still makes sense when t becomes one and gives a nontrivial automorphism as long as
disappears at
and there is neither gain or loss of symmetry in this case.
there is loss of generic symmetry (but no gain in specific
But if
symmetry).
4.6. The family of algebras
automorphisms.
and their
Mt!1t0t2"#$$$%&'''=1t4(t!1)t6!t5!t(t4!t2)!t4(t2!t)t+t3"#$$$$%&'''' X!t+1!t!t2t(t!1)X+!1+2tt2(t!1)X2t=0(0,t,t2)(t,t2,0)Mt!1tt20"#$$$%&'''=t1t!1!1+t!t2t2(t!1)"#$$$$$$%&'''''' X!t+1(t!1)X+!1+t!t2t2(t!1)X2t=0S3t=0char(k)!2char(k)=2At=k[X]/((X!tx1)(X!tx2)(X!tx3)) Here k is a field, which can be taken to be the field of complex numbers (for the
convenience of the reader). The symbol t is seen as a parameter (with values in k) and
are three pairwise unequal elements of k. The question to be investigated is
how the automorphism groups of these algebras change as t varies.
Let
For
this polynomial has three different roots. So then the algebra
and its automorphisms consist of permuting the three factors k. Thus there is a
is isomorphic to
generic symmetry group
. The three idempotents corresponding to the three factors k can
be easily written down (as, more generally, in the case of any polynomial with distinct roots
in k; see section 3 above). The three idempotents are
(4.6.1)
amounts to, and what happens
but it is not easy to see what, say, interchanging
to the automorphism as t becomes zero. Given the symmetry (sic!) of the situation one
survives as t becomes 0 or that none
expects that either all of the generic symmetry
survives (except the identity). The symmetry (= automorphism) group of
was described in section 4.1 above and as was remarked at the end of that section, this group
does not admit
as a subgroup. thus one expects all generic symmetry to disappear when
t becomes zero. As the calculations below show, thus is, surprisingly perhaps, not always the
case. True, for most triples
, in fact all but those in a subspace
of codimension 1, the generic symmetry
completely disappears when t becomes zero;
so, generically (sic!) this is the case. But there are certain triples
for which some of the generic symmetry survives when t
becomes zero. Here follow some calculations.
The Vandermonde matrix giving the transition from the basis
basis
is
of
to the
(4.6.2)
In particular
x1,!x2 and x3pt(X)=(X!tx1)(X!tx2)(X!tx3)t!0Atk!k!kS3e1=(X!tx2)(X!tx3)(tx1!tx2)(tx1!tx3),!!!e2=(X!tx1)(X!tx3)(tx2!tx1)(tx2!tx3),!!e3=(X!tx1)(X!tx2)(tx3!tx1)(tx3!tx2)e1 and e2S3A0=k[X]/(X3)S3(x1,x2,x3),!!x1!x2!x3!x1S3(x1,x2,x3),!!x1!x2!x3!x1{1,X,X2}At{e1,e2,e3}Mt=1tx1t2x121tx2t2x221tx3t2x32!"####$%&&&& So to see what interchanging
calculate
means in terms of the basis
we must
(4.6.3)
The determinant and inverse of
are, respectively
Obviously (4.6.4) is equal to something of the form
where
is a rational function in the indicated variables with denominator
numerator of
is zero. This numerator is equal to
. Thus (4.6.7) is well-defined at
if and only if the
So it is zero if and only if
(4.6.4)
(4.6.5)
(4.6.6)
(4.6.7)
(4.6.8)
X=(tx1)e1+(tx2)e2+(tx3)e3 or, in vector notation X=tx1tx2tx3!"###$%&&&e1 and e2{1,X,X2}Mt!1tx2tx1tx3"#$$$%&'''Mtdet(Mt)=t3(x2!x1)(x3!x2)(x3!x1)Mt!1=det(Mt)!1t3(x2x32!x3x22)!t3(x1x32!x3x12)t3(x1x22!x2x12)!t2(x32!x22)t2(x32!x12)!t2(x22!x12)t(x3!x2)!t(x3!x1)t(x2!x1)"#$$$$%&''''Mt!1tx2tx1tx3"#$$$%&'''=trat1(x1,x2,x3)rat2(x1,x2,x3)t!1rat3(x1,x2,x3)"#$$$%&'''rati(x3!x2)(x3!x1)(x2!x1)t=0rat3(x3!x2)x2!(x3!x1)x1+(x2!x1)x3=(x2!x1)(!x1!x2+2x3)2x3=x1+x2 It is now a straightforward calculation to show that if (4.6.8) holds
that at
the automorphism becomes
so
(4.6.9)
So in the special case (4.6.8) some of the generic symmetry survives.
have to calculate
Now let’s consider the case when the first and third roots are interchanged. Then we
(4.6.10)
Where of course the denominators of the three rational functions are again equal to
. Then the numerator of
is
So the third entry of (4.6.10) is zero iff
(4.6.11)
(4.6.12)
and then, as is bound to happen,
.
It is now no surprise to find that the symmetry ‘interchanging the second and third root
survives as
iff
It is a striking fact that the conditions (4.6.8), (4.6.12), (4.6.13) that make a symmetry
survive as t becomes zero are precisely symmetric
from the generic symmetry group
under that symmetry. This is something I do not really understand (yet).
Now consider the cyclic order three symmetry
(4.6.13)
To see when this one survives setting
calculate
(4.6.14)
rat2(x1,x2,x3)=!1t=0 X!!XMt!1tx3tx2tx1"#$$$%&'''=tra(t1(x1,x2,x3)ra(t2(x1,x2,x3)t!1ra(t3(x1,x2,x3)"#$$$%&'''(x3!x2)(x3!x1)(x2!x1)ra!t3(x3!x2)x3!(x3!x1)x2+(x2!x1)x1=(x3!x1)(x3+x1!2x2)x1+x3=2x2ra!t2="1 X!!Xx2+x3=2x1S3 tx1tx2tx3!"###$%&&&!tx2tx3tx1!"###$%&&&t=0 For this to have meaning it is necessary and sufficient that the denominator of
(4.6.15)
(4.6.16)
This condition can be fulfilled. Indeed,
third root of unity is a solution1.
It follows that
and that at
survives as the order three rotation
where
is a primitive
the third order cyclic symmetry (4.6.14)
(4.6.17)
Note again that the survival condition (4.6.16) respects the symmetry that is to survive.
4.7. The family of algebras
and their automorphisms.
Here k is a field, which can be taken to be the field of complex numbers (for the
convenience of the reader). The symbol t is seen as a parameter (with values in k). The
question to be investigated is how the automorphism groups of these algebras change as t
varies. So far all the examples have dealt with generic symmetry groups that are finite and
with finite loss of symmetry. That need not always be the case as the present example will
show.
For
the defining polynomial
has two equal roots and one additional root different from this double one. So for these t the
algebra
4.2. It is equal to the multiplicative group
. Thus the generic symmetry of the family
. The automorphism group of
was studied in section
is isomorphic to
is this one dimensional group.
(4.7.1)
For
the algebra becomes
whose automorphism group was
described in section 4.1. This automorphism group is a semi-direct product of the the
multiplicative group
and the additive group
and contains the multiplicative
1.!Indeed up to isomorphism it is the only solution: translation sees to it that
taken to be zero; and then rescaling t can be used to make
equal to one (as
can be
; and then, necessarily
Mt!1tx2tx3tx1"#$$$%&'''=tra((t1(x1,x2,x3)ra((t2(x1,x2,x3)t!1ra((t3(x1,x2,x3)"#$$$%&'''ra!!t3="(x12+x22+x32)+x1x2+x1x3+x2x3=0x1=0,!x2=1,!x3=!"3!3ra!!t2="3t=0 X!!3XAt=k[X]/(X3!tX2)t!0pt(X)=k[X]/(X3!tX2)AtA1A1Gm(k)Att=0A0=k[X]/(X3)Gm(k)Ga(k)x1x2x2!x1=0x2!x1=0 group as a subgroup. Thus it is a priori possible that all the generic symmetry is preserved
when t becomes zero. This turns out not to be the case.
The isomorphism
is given by
defined by
the automorphism of
is given by
corresponding to a is
. The automorphism of
. Tracing things back one sees that
Let’s check this. Note that
are modulo the principal ideal generated by
. Thus
where all the congruences
(4.7.2)
and
So (4.7.2) does indeed define a endomorphism of algebras which is an automorphism when
.
When t becomes zero (4.7.2) is undefined and so the complete generic symmetry
group of this family disappears. In return there is a gain in specific symmetry given by the
two-dimensional group
.
4.8. A family of triangular matrix algebras and their automorphisms.
Let k be a field, which can be taken to be the field of complex numbers (for the
convenience of the reader). The symbol t is seen as a parameter (with values in k).
Consider the three dimensional algebra
with basis
and defining
relations
and, as suggested by the notation,
, the unit element. Mapping
(4.8.1)
At!!"!A1 X!tXA1a!Gm(k) X!aX+(1!a)X2At X!aX+t!1(1!a)X2X4=X(X3)!X(tX2)=tX3!t2X2pt(X)(aX+t!1(1!a)X2)2=a2X2+(1!a)2t!2X4+2a(1!a)t!1X3"a2X2+(1!a)2X2+2a(1!a)X2=X2X2(aX+(1!a)X2)=aX3+t!1(1!a)X4"taX2+(1!a)tX2=tX2a!0G=Aut(k[X]/(X3))T1e1=1,!e2,e3e22=1,!!e32=0,!!e2e3=e3,!!e3e2=!e3e1=1 e1!1001!"#$%&,!e2!100'1!"#$%&,!!e3!0100!"#$%& shows that this is just the matrix algebra of upper triangular
let
be the three dimensional algebra with basis
matrices. More generally,
and defining relations
The
are isomorphic to
as long as
. The isomorphism is given by
(4.8.2)
(4.8.3)
is not isomorphic to
But
isomorphic to
. For one thing because it is commutative. It is in fact
The next step is to calculate the automorphism group of
given by
(4.8.4)
. An automorphism
is
Preserving the relation
then immediately gives
it follows that
and then
and then because
because
automorphism. So
is of the form
(4.8.5)
is an
Now use
so that
of elements of k of
which the second one is non-zero. The composition of two such pairs is easily calculated to
be given by
. Thus an automorphism of
to find respectively
is given by a pair
and
This fits with the matrix multiplication
(4.8.6)
2!2Tt{1,!!e2,!!e3}(!e2)2=t2,!!(!e3)2=0,!!!e2!e3=t!e3,!!!e3!e2="t!e3TtT1t!0!e2=te2,!!!e3=e3T0T1k[X,Y]/(X2,Y2,XY)T1! !:!e2!ae2+be3+ce3!"ae2+"be3+"c#$%e32=0!c=0e2e3=!e3e2,!!e32=0,!!e22=1!a=0!b"0!! !:!e2!ae2+be3+ce3!"be3#$%!!with "b&0e2e3=e3 and e3e2=!e3a+c=1c!a=!1c=0,!!a=1T1(b,!b)(b2,!b2)(b1,!b1)=(b2+b1!b2,!b1!b2)1b10!b1"#$$%&''1b20!b2"#$$%&''=1b2+b1!b20!b1!b2"#$$%&'' Note also that the
the quotient by this normal subgroup is the multiplicative group
automorphism group is a slightly differently written version of the group G discussed in
section 4.1 above.
form a normal subgroup N isomorphic to
. In fact this
and that
Via the isomorphism
one finds the corresponding automorphism of
to be
,
Thus there is a generic symmetry group G for the family
for all
. The (specific) symmetry group of
obviously is
(4.8.7)
in that
.
When t becomes zero (4.8.7) still makes sense in the form given by the matrix
(4.8.8)
So here there is a situation that a two dimensional generic symmetry group degenerates to a
one dimensional quotient group that is a subgroup of the four dimensional specific symmetry
group at
.
5. Conclusion. I have no doubt that in general the situation will be that a generic
symmetry group of a family changes to a sub-quotient that is a subgroup of the specific
symmetry group at critical values of the parameter. Examples that a subgroup occurs and that
a quotient group occurs have been given above. Currently I have no example when a true
sub-quotient arises.
The simple calculations above can certainly be extended to the case of 4-dimensional
algebras as have been classified in [8 Rakhimov et al.].
The remarks above I regard as a mere beginning; much more remains to be done.
References.
Ben-Naim, Arieh, A farewell to entropy, World Scientific, 2014
Cooper, Susan and Brian Harbourne, Regina lectures on fat points, 2013
Hazewinkel, Michiel, Symmetrry, bifurcations and pattern formation, in M.
Hazewinkel, R. Jurkovich and J. H. C. Paelinck, eds., Bifurcation analysis: principles,
applications and synthesis, Reidel Publ. cy, 1985, 201 - 232
Lin, Shu-Kun, Book review of Joe Rosen, Symmetry rules: how science and nature are
founded on symmetry, Springer 2008, Entropy 10 (2008), 55 - 57
Lin, Shu-Kun, ed., Symmetry and entropy. Preface to a special issue of the journal
Entropy, 2009
1.
2.
3.
4.
5.
(b,1),!!b!kGa(k)Gm(k) T1!Tt,!!!e2=te2,!e3=e3Tt !e2=te2!t(e2+be3)=!e2+tbe3 !e3=e3!!be3=!b!e3{Tt,!t!k}Aut(Tt)=Gt!0T0GL2(k)100!b"#$%&'t=0 6.
7.
8.
9.
10.
Mathoverflow, Automorphisms of fat points, 2014
Michel, Louis, Applications of group theory to quantum physics: algebraic aspects, in
V. Bargmann, ed., Group representations in mathematics and physics, Springer, 1970,
36 - 143
Rakhimov, I S, I M Rikshiboev and W Basri, Complete lists of low dimensional
complex associative algebras, 2009. arXiv:0910.0932 v2 [math.RA]
Rosen, Joe, The symmetry principle, Entropy 7 (2005), 308 - 313
Rosen, Joe, Symmetry rules: how science and nature are founded on symmetry,
Springer, 2008
|
1702.06734 | 1 | 1702 | 2017-02-22T10:13:59 | Weakly $r$-clean rings and weakly $\star$-clean rings | [
"math.RA"
] | Motivated by the concept of weakly clean rings, we introduce the concept of weakly $r$-clean rings. We define an element $x$ of a ring $R$ as weakly $r$-clean if it can be expressed as $x=r+e$ or $x=r-e$ where $e$ is an idempotent and $r$ is a regular element of $R$. If all the elements of $R$ are weakly $r$-clean then $R$ is called a weakly $r$-clean ring. We discuss some of its properties in this article. Also we generalise this concept of weakly $r$-clean ring to weakly $g(x)$-$r$-clean ring, where $g(x) \in C(R)[x]$ and $C(R)$ is the centre of the ring $R$. Finally we introduce the concept of weakly $\star$-clean and $\star$-$r$-clean ring and discuss some of their properties. | math.RA | math |
Weakly r-clean rings and weakly ⋆-clean rings
Ajay Sharma
Department of Mathematical Sciences, Tezpur University,
Napaam, Tezpur-784028, Assam, India.
Email: [email protected]
Dhiren Kumar Basnet
Department of Mathematical Sciences, Tezpur University,
Napaam, Tezpur-784028, Assam, India.
Email: [email protected]
Abstract: Motivated by the concept of weakly clean rings, we introduce
the concept of weakly r-clean rings. We define an element x of a ring R as
weakly r-clean if it can be expressed as x = r + e or x = r − e where e
is an idempotent and r is a regular element of R. If all the elements of R
are weakly r-clean then R is called a weakly r-clean ring. We discuss some
of its properties in this article. Also we generalise this concept of weakly
r-clean ring to weakly g(x)-r-clean ring, where g(x) ∈ C(R)[x] and C(R) is
the centre of the ring R. Finally we introduce the concept of weakly ⋆-clean
and ⋆-r-clean ring and discuss some of their properties.
Key words: r-clean ring, Weakly r-clean ring, ⋆-clean ring, Weakly ⋆-clean ring,
⋆-r-clean ring.
2010 Mathematics Subject Classification: 16N40, 16U99.
1
INTRODUCTION
Here rings R are associative with unity unless otherwise indicated. The Jacob-
son radical, set of units, set of idempotents, set of regular elements, set of nilpotent
elements and centre of a ring R are denoted by J(R), U (R), Idem(R), Reg(R),
N il(R) and C(R) respectively. Nicholson[8] called an element x of R, a clean
element, if x = e + u for some e ∈ Idem(R), u ∈ U (R) and called the ring R as
clean ring if all its elements are clean. Weakening the condition of clean element,
M.S. Ahn and D.D. Anderson[1] defined an element x as weakly clean if x can be
expressed as x = u + e or x = u − e, where u ∈ U (R), e ∈ Idem(R). Generalising
1
the idea of clean element, N. Ashrafi and E. Nasibi [2, 3] introduced the concept
of r-clean element as follows: an element x of a ring R is said to be r-clean if it
can be written as a sum of an idempotent and a regular element. Further if all
the elements of R are r-clean then the ring R is called r-clean ring. Motivated by
these ideas we define an element of a ring R as weakly r-clean if x can be expressed
as x = r + e or x = r − e, where r ∈ Reg(R), e ∈ Idem(R) and call the ring R
to be weakly r-clean if all its elements are so. We show that in case of an abelian
ring the concept of weakly r-clean rings coincides with that of weakly clean rings
and hence with that of weakly exchange rings. We also show that the center of
weakly r-clean rings are not so. However if R has no non-trivial idempotents then
the center of a weakly r-clean ring is weakly r-clean. Further we discuss some
interesting properties of weakly g(x)-r- clean ring, where g(x) ∈ C(R)[x]. Finally
we define the concept of weakly ⋆-clean ring and ⋆-r-clean ring. A ring R is a
⋆-ring (or ring with involution) if there exists an operation ⋆ : R → R such that
for all x, y ∈ R, (x + y)⋆ = x⋆ + y⋆, (xy)⋆ = y⋆x⋆ and (x⋆)⋆ = x. An element
p of a ⋆-ring is a projection if p2 = p = p⋆. Obviously, 0 and 1 are projections
of any ⋆-ring. Henceforth P (R) will denote the set of all projections in a ⋆-ring.
We define an element x in a ⋆-ring R is weakly ⋆-clean element if x = u + p or
x = u − p, where u ∈ U (R) and p ∈ P (R). If all the elements of R are weakly
⋆-clean then the ⋆-ring R is called weakly ⋆-clean. We show that for an abelian
⋆-clean ring R, if every idempotent of the form e = ry or e = yr is projection,
where r ∈ Reg(R), then R is ⋆-r-clean if and only if R is ⋆-clean.
2 Weakly r-clean ring
Definition 2.1. An element x in a ring R is called weakly r-clean if it can be
written as x = r + e or x = r − e, for some r ∈ Reg(R) and e ∈ Idem(R). If all the
elements of R are weakly r-clean, then the ring R is called a weakly r-clean ring.
For example all clean rings, weakly clean rings and r-clean rings are weakly
r-clean ring. The following theorem is obvious.
Theorem 2.2. Homomorphic image of weakly r-clean ring is weakly r-clean.
However the converse is not true as Z6 ∼= Z/h6i is a weakly r-clean ring, but
Z is not weakly r-clean ring.
Theorem 2.3. Let {Rα} be a collection of rings. Then the direct product R =
Rα is weakly r-clean if and only if each Rα is weakly r-clean ring and at most
Qα
one Rα is not r-clean.
2
Proof. Let R be weakly r-clean ring. Then being homomorphic image of R each
Rα is weakly r-clean. Suppose Rα1 and Rα2 are not r-clean, where α1 6= α2.
Since Rα1 is not r-clean so not all elements x ∈ Rα1 of the form x = r − e where
r ∈ Reg(Rα1 ) and e ∈ Idem(Rα1 ). As Rα1 is weakly r-clean so there exists
xα1 ∈ Rα1 with xα1 = rα1 + eα1, where rα1 ∈ Reg(Rα1 ) and eα1 ∈ Idem(Rα1 )
but xα1 6= r − e for any r ∈ Reg(Rα1 ) and e ∈ Idem(Rα1 ). Similarly there exists
xα2 ∈ Rα2 with xα2 = rα2 − eα2, where rα2 ∈ Reg(Rα2 ) and eα2 ∈ Idem(Rα2 ), but
xα2 6= r + e for any r ∈ Reg(Rα2 ) and e ∈ Idem(Rα2 ). Define x = (xα) ∈ R by
xα = xα
= 0
if α ∈ {α1, α2}
if α /∈ {α1, α2}
Then clearly x 6= r ± e for any r ∈ Reg(R) and e ∈ Idem(R). Hence at most one
Rα is not r-clean.
(⇐) If each Rα is r-clean then R = Q Rα is r-clean and hence weakly r-clean.
Assume Rα0 is weakly r-clean but not r-clean and that all other Rα's are r-clean.
If x = (xα) ∈ R then in Rα0 we can write xα0 = rα0 + eα0 or xα0 = rα0 − eα0,
where rα0 ∈ Reg(Rα0 ) and eα0 ∈ Idem(Rα0 ). If xα0 = rα0 + eα0 then for α 6= α0
let, xα = rα + eα and if xα0 = rα0 − eα0 then for α 6= α0 let, xα = rα − eα then
r = (rα) ∈ Reg(R) and e = (eα) ∈ Idem(R) such that x = r + e or x = r − e and
consequently R is weakly r-clean ring.
Lemma 2.4. Let R be a ring with no zero divisor. Then R is weakly clean if and
only if R is weakly r-clean.
Proof. Let r(6= 0) ∈ Reg(R), then r = ryr for some y ∈ R, i.e., r(1 − yr) = 0,
which implies that r is a unit. Now the result follows immediately.
Lemma 2.5. Let R be a ring and e2 = e ∈ R, a = u−f ∈ eRe, where u ∈ U (eRe),
f ∈ Idem(eRe). Then there exist elements v ∈ U (R) and e ∈ Idem(R) such that
a = v − e.
Proof. Since u ∈ U (eRe), so there exists w ∈ U (eRe) such that uw = e = wu.
Assume v = u + (1 − e), then v(w + (1 − e)) = 1 = (w + (1 − e))v, because
v(1 − e) = 0 = (1 − e)w. Clearly a − v = a − u − (1 − e) = −(f + (1 − e)). It is
easy to see that e = f + (1 − e) is idempotent and hence the result follows.
For the sake of completeness of the article we prove a result on weakly clean
element.
Proposition 2.6. Let R be an abelian ring and a ∈ R is weakly clean element
then ae is weakly clean for any idempotent e ∈ R.
3
Proof. Let a ∈ R be a weakly clean element. So a = u + e′ or a = u − e′, where
u ∈ U (R) and e′ ∈ Idem(R)
Case 1: If a = u + e′, then ae is clean for any idempotent e by Lemma 2.1 [2].
Case 2: If a = u − e′, then ae = ue − e′e. But ue = ue2 = eue ∈ U (eRe) and
e′e ∈ Idem(eRe), as R is abelian. So by Lemma 2.5, ae = v −e, for some v ∈ U (R)
and e ∈ Idem(R). Hence ae is weakly clean.
Lemma 2.7. Let R be an abelian ring such that a and −a both are clean. Then
the followings hold:
(i) a + e is clean for any idempotent e of R.
(ii) a − e is weakly clean for any idempotent e of R.
Proof. (i) It follows from Lemma 2.1 [2] .
(ii) Consider a = u + e′ and −a = u′ + e′′ where, u, u′ ∈ U (R) and e, e′ ∈ Idem(R).
Now
a − e = (ae + a(1 − e)) − e
= (a − 1)e + a(1 − e)
= −(1 − a)e + a(1 − e)
= −(1 − e′ − u)e + (−u′ − e′′)(1 − e)
= −{(1 − e′)e + e′′(1 − e)} + {ue − u′(1 − e)}
Since {(1 − e′)e + e′′(1 − e)} ∈ Idem(R) and {ue − u′(1 − e)} ∈ U (R), so a − e is
weakly clean.
Here we define a weakly r-clean (respectively weakly clean) element x ∈ R
is of type 1 form if x = r + e, where r ∈ Reg(R) (respectively r ∈ U (R)) and
e ∈ Idem(R) and of type 2 form if x = r − e, where r ∈ Reg(R) (respectively
r ∈ U (R)) and e ∈ Idem(R).
Theorem 2.8. Let R be an abelian ring. Then R is weakly clean if and only if R
is weakly r-clean.
Proof. (⇒) Obviously weakly clean rings are weakly r-clean.
(⇐) Let R be weakly r-clean ring and x ∈ R. So x = r ± e′, where r = ryr for
some y ∈ R and e′ ∈ Idem(R).
First let x = r + e′. Assume yr = e, then clearly e is an idempotent and therefore
(ye + (1 − e))(re + (1 − e)) = 1. Let u = re + (1 − e), then u ∈ U (R) and ue = r,
so ue + f ∈ U (R), where (1 − e) = f . Thus −r = −(ue + f ) + f is clean. Hence
by lemma 2.7, r + e′ is clean.
If x = r − e′, then clearly r and −r both are clean and so by lemma 2.7, r − e′ is
weakly clean. Hence R is weakly clean.
4
Consider the ring Z(6) of all rational numbers a
b of lowest form, where b is
relatively prime to 6. Since 21
11 both are not units and 0,
1 are the only idempotents of Z(6) so Z(6) is not weakly clean and hence not weakly
r-clean. The following is an example of weakly r-clean but not r-clean ring.
11 + 1 = 32
11 and 21
11 − 1 = 10
Example 2.9. Z(3) ∩ Z(5) is weakly r-clean but not r-clean.
Since Z(3) ∩ Z(5) is weakly clean but not clean [1] and if R is an abelian ring
then R is clean if and only if R is r-clean by Theorem 2.2 [2]. So by Theorem 2.8,
Z(3) ∩ Z(5) is weakly r-clean but not r-clean.
Definition 2.10. A ring R is called weakly exchange ring if for any x ∈ R, there
exists an idempotent e ∈ xR such that 1 − e ∈ (1 − x)R or 1 − e ∈ (1 + x)R.
The relation between weakly r-clean ring and weakly exchange ring is given
below.
Theorem 2.11. If R is abelian ring then R is weakly r-clean ring if and only if
R is weakly exchange ring.
Proof. (⇒) It is equivalent to show R is weakly clean ring if and only if R is
weakly exchange ring.
Let x ∈ R, if x = u + e where u ∈ U (R) and e ∈ Id(R) then clearly it satisfies the
exchange property [8].
Suppose, x = u − e where u ∈ U (R) and e ∈ Idem(R). Let f = u−1(1 − e)u then
f 2 = f . Now
u(x + f ) = u(u − e + u−1(1 − e)u)
= (u − e)2 + (u − e)
= x2 + x
∴ x + f ∈ R(x2 + x). So x satisfies weakly exchange property by Lemma 2.1
[5]. Hence R is weakly exchange ring. (⇐) Suppose x ∈ R then there exists an
idempotent e ∈ Rx such that 1 − e ∈ R(1 − x) or 1 − e ∈ R(1 + x).
Case 1: Suppose 1 − e ∈ R(1 + x) and e = ax, 1 − e = b(1 + x), for some
a, b ∈ R. Assume that ea = a and (1 − e)b = b so that axa = ea = a and
b(1 + x)b = b. Here ax, xa, b(1 + x), (1 + x)b all are central idempotent and
ax = (ax)(ax) = (ax)(xa) = x(ax)a = xa similarly (1 + x)b = b(1 + x). Now
(a + b)(x + (1 − e)) = ax + bx + a(1 − e) + b(1 − e) = 1 so x + (1 − e) is unit. Hence
x is weakly clean element.
Case 2: If 1−e ∈ R(1−x) then clearly x is weakly clean by Proposition 1.8 [8].
5
A polynomial ring over a commutative ring R can never be weakly r-clean as
x can not be represented in weakly clean decomposition. N. Ashrafi and E. Nasibi
[2], showed that R is r-clean if and only if R[[x; α]] is r-clean, where R[[x; α]] is
the ring of skew formal power series over R i.e. all formal power series in x with
coefficients from R and α is an ring endomorphism. Multiplication is defined by
xr = α(r)x, for all r ∈ R. In particular R[[x]] = R[[x; IR]] is the ring of formal
power series over R.
Proposition 2.12. Let R be an abelian ring and α be an endomorphism of R.
Then the following are equivalent.
(i) R is weakly r-clean ring.
(ii) The formal power series ring R[[x]] over R is weakly r-clean.
(iii) The skew power series ring R[[x; α]] over R is weakly r-clean.
Proof. Follows from Theorem 2.8, and similar results of weakly clean rings.
Consider a ring R[D, C] constructed in [10], for any subring C of a ring D. The
following result gives the relation between R[D, C] with D and C, about weakly
r-clean ness.
Proposition 2.13. Let C be a subring of a ring D then R[D, C] is weakly r-clean
ring if and only if D is r-clean and C is weakly r-clean ring.
Proof. (⇒) As R[D, C] = D ⊗ D so by Theorem 2.3 D is r-clean. Clearly C is
weakly r-clean ring by Theorem 2.2.
(⇐) Let D be r-clean and C is weakly r-clean ring. Let x = (a1, a2, · · ·, an, c, c, c, · ·
·) ∈ R[D, C]. Since C is weakly r-clean ring, so c = r ± e, where r ∈ Reg(R) and
e ∈ Idem(R). If c = r + e, write ai = ri + ei then x = r + e, where r = (r1, r2, · ·
·, rn, r, r, · · ·) ∈ Reg(R[D, C]) and e = (e1, e2, · · ·, en, e, e, · · ·) ∈ Idem(R[D, C]). If
c = r − e, write ai = ri − ei then x = r − e, where r = (r1, r2, · · ·, rn, r, r, · · ·) ∈
Reg(R[D, C]) and e = (e1, e2, · · ·, en, e, e, · · ·) ∈ Idem(R[D, C]).
Clearly every abelian semi regular ring is weakly clean and hence weakly r-
clean. However the converse is not true as shown by the following example.
Example 2.14. Let Q be a field of rational numbers and L the ring of all rational
numbers with odd denominators. Then by example 2.7 [2] and Theorem 2.8,
R[Q, L] is commutative exchange ring and hence R[Q, L] is weakly r-clean ring
but not semi regular.
If R is weakly r-clean ring then R/I is weakly r-clean being homomorphic
image of weakly r-clean ring. The theorem is a partial converse of this statement.
6
Theorem 2.15. Let I be a regular ideal of a ring R, where idempotents can be
lifted modulo I. Then R is weakly r-clean if and only if R/I is weakly r-clean.
Proof. Following theorem 2.8 [2], it is enough to show that for any a ∈ R if
a + I has type 2 weakly r-clean decomposition in R/I then a is weakly r-clean
in R. Let a + I ∈ R/I has type 2 weakly r-clean decompositions in R/I then
there exists idempotent e + I ∈ R/I such that (a + I) + (e + I) ∈ Reg(R/I) i.e.
(a + e) + I ∈ Reg(R/I). Therefore ((a + e) + I)(x + I)((a + e) + I) = (a + e) + I,
for some x ∈ R, so (a + e)x(a + e) − (a + e) ∈ I but I is regular ideal, Hence (a + e)
is regular by Lemma 1, [4] . Since idempotents can be lifted modulo I, we may
assume that e is an idempotent of R.
Proposition 2.16. Let M be a A − B bi-module. If T = (cid:18) A 0
M B (cid:19), a formal
triangular matrix ring is weakly r-clean then A and B are weakly r-clean ring.
Proof. Let a ∈ A and b ∈ B and m ∈ M , consider t = (cid:18) a
m b (cid:19) ∈ T .
0
Case I: If t is r-clean then clearly a and b both are r-clean in their respective rings
A and B by Theorem 16, [3] .
Case II: If t = −(cid:18) f1
and (cid:18) r1
r3 (cid:19) ∈ Reg(T ).
r2
0
0
f2 f3 (cid:19) + (cid:18) r1
r2
0
r3 (cid:19) where (cid:18) f1
0
f2 f3 (cid:19)2
= (cid:18) f1
f2 f3 (cid:19)
0
It is simple calculation to show that f1 and f3 are
idempotents in A and B respectively and r1 and r3 are regular elements in A and
B respectively. Finally a = −f1 + r1 and b = −f3 + r3. Hence A and B both are
weakly r-clean rings.
In [10], T. Kosan, S. Sahinkaya and Y. Zhou showed that the center of weakly
clean rings need not be weakly clean.
In that example End(RE(R)) is weakly
clean whereas the center is not weakly clean as R is not weakly clean. But since
R is abelian in that example so by Theorem 2.8, the center of End(RE(R)) is
not weakly r-clean. Hence the center of weakly r-clean ring need not be weakly
r-clean.
Theorem 2.17. Let R be a weakly r-clean ring with no nontrivial idempotents.
Then the center of R is also a weakly r-clean ring.
Proof. Let R be weakly r-clean ring and Z(R) be the center of R. Let x ∈ Z(R)
then there exists regular element r ∈ R such that either x = r or x = r − 1 or
x = r + 1. If x = r then clearly x is weakly r-clean in Z(R). If x = r ± 1 then
r ∈ Z(R) as x ± 1 ∈ Z(R). Hence Z(R) is weakly r-clean ring.
7
A ring R is called graded ring(or more precisely, Z-graded) if there exists a
Rn (as abelian groups) and
family of subgroups {Rn}n∈Z of R such that R = ⊕
n
RnRm ⊆ Rn+m, for all n, m ∈ Z. The relation of graded ring and weakly r-clean
ring is given below.
Proposition 2.18. Let R = R0 ⊕ R1 ⊕ R2 ⊕ · · · be a graded ring then the following
hold:
(i) If R is weakly r-clean ring then R0 is weakly r-clean ring.
(ii) If R0 is weakly r-clean ring and each Rn is a torsion-less R0 module, i.e.
r /∈ Z(R0) ⇒ r /∈ Z(Rn), then R is weakly r-clean ring.
Proof.
(i) Let r0 ∈ R0, so r0 = r + e or r0 = r − e, where r ∈ Reg(R) and
e ∈ Idem(R). Since Idem(R) = Idem(R0), so e ∈ R0 and r ∈ R0∩Reg(R) ⊆
Reg(R0). Hence R0 is weakly r-clean ring.
(ii) Let x = x0 + x1 + x2 + ···+ xn ∈ R, where xi ∈ Ri. Write x0 = r0 ± e0, where
r0 ∈ Reg(R0) and e0 ∈ Idem(R0) = Idem(R). Put x′ = r0+x1+x2+···+xn,
so x = x′ ± e0 where e0 ∈ Idem(R). If x′ ∈ Z(R) then there exists a non
zero homogeneous t ∈ Rn with tx′ = 0. But then r0t = 0, a contradiction.
3 Weakly g(x)-r-clean rings
Definition 3.1. Let g(x) be a fixed polynomial in C(R)[x]. An element x ∈ R is
called weakly g(x)-r-clean if x = r ± e, where r ∈ Reg(R) and g(e) = 0. We say
that R is weakly g(x)-r-clean if every element is weakly g(x)-r-clean.
If g(x) = x2 −x, then weakly g(x)-r-clean ring is similar to weakly r-clean ring.
An element s of R is called root of the polynomial g(x) ∈ C(R)[x] if g(s) = 0. For
g(x) = x2 − x, the ring Z(3) ∩ Z(5) is weakly g(x)-r-clean ring but not g(x)-r-clean
ring.
Let R and S be rings and θ : C(R) → C(S) be a ring homomorphism with
θ(1) = 1. Then θ induces a map θ′ from C(R)[x] to C(S)[x] such that for g(x) =
Pn
i=0 aixi ∈ C(R)[x], θ′(g(x)) := Pn
i=0 θ(ai)xi ∈ C(S)[x].
Proposition 3.2. Let θ : R → S be a ring epimorphism. If R is weakly g(x)-r-
clean ring then S is weakly θ′(g(x))-r-clean ring.
Proof. Let g(x) = a0 +a1x+···+anxn ∈ C(R)[x], then θ′(g(x)) = θ(a0)+θ(a1)x+
· · · + θ(an)xn ∈ C(S)[x]. Since θ is ring epimorphism so for any s ∈ S, there exists
8
x ∈ R such that θ(x) = s. Let x = r ± s0 where r ∈ Reg(R) and g(s0) = 0, as
R is weakly g(x)-r-clean ring. Now s = θ(x) = θ(r ± s0) = θ(r) ± θ(s0). Clearly
θ(r) ∈ Reg(S) and
θ′(g(θ(s0))) = θ(a0) + θ(a1)θ(s0) + · · · + θ(an)(θ(s0))n
= θ(a0 + a1s0 + · · · + ansn
0 )
= θ(0)
= 0
Hence S is θ′(g(x))-r-clean ring.
Next we extend the Theorem 2.3 to g(x)-r-clean ring.
Theorem 3.3. Let g(x) ∈ Z[x] and {Ri}i∈I be a family of rings. Then Qi∈I
g(x)-r-clean ring if and only if Ri's are weakly g(x)-r-clean ring and at most one
Rα is not g(x)-r-clean ring.
Ri is
Proof. (⇒) Similar to Theorem 2.3.
(⇐) If each Rα is g(x)-r-clean ring, then R = Q Rα is g(x)-r-clean. Assume Rα0 is
weakly g(x)-r-clean but not g(x)-r-clean and that all other Rα's are g(x)-r-clean.
Let x = xα ∈ R so in Rα0 we can write xα0 = rα0 + sα0 or xα0 = rα0 − sα0,
where rα0 ∈ Reg(Rα0 ) and g(sα0) = 0 in Rα0 .
If xα0 = rα0 + sα0 then for
α 6= α0 assume, xα = rα + sα and if xα0 = rα0 − sα0 then for α 6= α0 assume,
xα = rα − sα, where rα ∈ Rα and g(sα) = 0 in Rα. Then r = (rα) ∈ Reg(R) and
g(s = {sα}) = a0{1Ri} + a1{si} + · · · + an{sn
i } = {a0 + a1si + · · · + ansn
i }=0.
Theorem 3.4. Let R be a ring and a, b ∈ R then R is weakly (ax2n − bx)-r-clean
ring if and only if R is weakly (ax2n + bx)-r-clean ring, for n ∈ N.
Proof. Suppose R is weakly (ax2n − bx)-r-clean ring. Since (as2n − bs) = 0 ⇒
(a(−s)2n + b(−s)) = 0. So for a ∈ R, −a = r ± s, where (as2n − bs) = 0 and r ∈ R.
Hence a = (−r) ± (−s) and (a(−s)2n + b(−s)) = 0. Similarly the converse is also
true.
Proposition 3.5. Let 2 ≤ n ∈ N. If for every a ∈ R, a = r ± t, where r ∈ Reg(R)
and tn−1 = 1 then R is weakly (xn − x)-r-clean ring.
Proof. The proof follows from Theorem 3.4.
Proposition 3.6. Let g(x) = (xn−x) ∈ C(R)[x] and g(e) = 0. Let a ∈ en−1Ren−1
be strongly g(x)-clean in en−1Ren−1 then a is strongly g(x)-clean in R.
9
Proof. Let a = u + f , where u ∈ U (en−1Ren−1) and g(f ) = 0 in en−1Ren−1. Then
uw = en−1 = wu, for some w ∈ U (en−1Ren−1). Consider v = u − (1 − en−1), then
{u − (1 − en−1)}{w − (1 − en−1)} = 1, so v = u − (1 − en−1) is a unit in R. Now
a − v = f + (1 − en−1) and {f + (1 − en−1)}n = f n + (1 − en−1)n = f + (1 − en−1),
so a is g(x)-clean ring in R.
Proposition 3.7. Let R be an abelian ring and g(x) ∈ C(R)[x] then if a is g(x)-
clean element in R then aen−1 is also g(x)-clean element in R for any root e of
g(x).
Proof. Let a = u + f , where u ∈ U (R) and g(f ) = 0 in R. Suppose e be any
root of g(x) in R. Then aen−1 = uen−1 + f en−1 but uen−1 ∈ U (en−1Ren−1) and
(f en−1)n = f nen−1 = f en−1. So aen−1 is g(x)-clean in R.
4 Weakly ⋆-clean rings and ⋆-r-clean ring
A ring R is a ⋆-ring (or ring with involution) if there exists an operation ⋆ : R → R
such that for all x, y ∈ R, (x + y)⋆ = x⋆ + y⋆, (xy)⋆ = y⋆x⋆ and (x⋆)⋆ = x. An
element p of a ⋆-ring is a projection if p2 = p = p⋆. Obviously, 0 and 1 are
projections of any ⋆-ring. Henceforth P (R) will denote the set of all projections
in a ⋆-ring.
Here we define the concept of weakly ⋆-clean ring and discuss some properties
of weakly ⋆-clean ring.
Definition 4.1. An element x of a ⋆-ring R is said to be weakly ⋆-clean if x = u+p
or x = u − p, where u ∈ U (R) and p ∈ P (R). A ⋆-ring R is said to be weakly
⋆-clean ring if every elements are weakly ⋆-clean.
Example 4.2.
(i) Units, elements in J(R) and nilpotents of a ⋆-ring R are
weakly ⋆-clean.
(ii) Idempotents of a ⋆-regular rings are weakly ⋆-clean.
Lemma 4.3. Let R be a boolean ⋆-ring. Then R is weakly ⋆-clean if and only if
⋆ = 1R is the identity map of R.
Proof. It is clear that boolean rings are clean. Suppose that R is weakly ⋆-clean.
Given any a ∈ R, we have −a = u±p = ±p+1 = ±p−1, for some p ∈ P (R). So we
have a = 1 ± p ∈ P (R). Hence a⋆ = a, which implies ⋆ = 1R. Conversely if ⋆ = 1R
then every idempotent of R is a projection. Thus, R is a weakly ⋆-clean.
Example 4.4. R = Z2 ⊕ Z2 with involution ⋆ defined by (a, b)⋆ = (b, a) is weakly
clean but not weakly ⋆-clean.
10
Lemma 4.5. Let R be a ⋆-ring. If 2 ∈ U (R), then for any u2 = 1, u⋆ = u ∈ R if
and only if every idempotent of R is a projection.
Proof. Proof is given in Lemma 2.3 [11].
Corollary 4.6. Let R be a ⋆-ring with 2 ∈ U (R). The following are equivalent:
(i) R is weakly clean and every unit of R is self-adjoint.(i.e., u⋆ = u for every
unit u ).
(ii) R is weakly ⋆-clean and ⋆ = 1R.
Proof. (1) ⇒ (2). Let a ∈ R. Then a = u ± p, for some p ∈ P (R) and u ∈ U (R).
Note that (1 − 2p) ∈ U (R), so (1 − 2p)⋆ = (1 − 2p) ⇒ p⋆ = p by Lemma 4.5. Also
a⋆ = (u ± p)⋆ = u⋆ ± p⋆ = u ± p = a.
(2) ⇒ (1). is trivial.
For a ⋆-ring R, an element x ∈ R is called self adjoint square root of 1 if x2 = 1
and x⋆ = x.
Theorem 4.7. Let R be a ⋆-ring, the following are equivalent:
(1) R is weakly ⋆-clean and 2 ∈ U (R).
(2) Every element of R is a sum of unit and a self adjoint square root of 1 or an
element of the form 2p + 1, where p2 = p = p⋆.
2 = u ± p, where u ∈ U (R) and
2 = u − p, then a = 2u + (1 − 2p), where 2u ∈ U (R) and 1 − 2p
2 = u + p, then a = 2u + (1 + 2p), where
Proof. (1) ⇒ (2). Consider a ∈ R, then a−1
p ∈ P (R). If a−1
is a self adjoint square root of 1. If a−1
p2 = p = p⋆.
(2) ⇒ (1) First we show that 2 ∈ (R). By assumption, 1 = u+x or 1 = u+(2p+1),
where u ∈ U (R), x is self adjoint square root of 1 and p ∈ P (R). If 1 = u + x,
then clearly 2 ∈ U (R) by Theorem 2.5 [11]. If 1 = u + (2p + 1), then u = −2p, so
2 ∈ U (R). Next for showing R is weakly ⋆-clean ring, let a ∈ R, so 2a + 1 = u + x
or 2a + 1 = u + (2p + 1), where u ∈ U (R), x is self adjoint square root of 1 and
p ∈ P (R).
Case I: If 2a + 1 = u + x, then a = u
2 ∈ (R), so a is weakly ⋆-clean element.
Case II: If 2a + 1 = u + (2p + 1), then a = u
2 )2 = 1−x
2 − 1−x
2 . Since ( 1−x
2 = ( 1−x
2 )⋆ and
2 + p, a ⋆-clean element.
Lemma 4.8. Let R be weakly ⋆-clean ring. If I is a ⋆ invariant ideal of R, then
R/I is weakly ⋆-clean. In particular, R/J(R) is weakly ⋆-clean ring.
Proof. The result follows from Lemma 2.7 [11].
11
Let R be a ⋆-ring. Then ⋆ induces an involution in the power series ring R[[x]],
defined by (P∞
i=0 aixi)⋆ = P∞
i=0 a⋆
i xi.
Proposition 4.9. Let R be a ⋆-ring. Then R[[x]] is weakly ⋆-clean if and only if
R is weakly ⋆-clean.
Proof. Let R[[x]] be weakly ⋆-clean. Since R ∼= R[[x]]/ < x > and < x > is
⋆-invariant ideal of R[[x]]. So R is weakly ⋆-clean ring. Conversely, suppose that
i=0 aixi ∈ R[[x]]. Write a0 = u ± p
i=1 aixi), where p ∈
i=1 aixi ∈ U (R[[x]]). Hence g(x) is weakly ⋆-clean in
R is weakly ⋆-clean ring. Let g(x) = P∞
with p ∈ P (R) and u ∈ U (R). Then g(x) = ±p + (u + P∞
P (R) ⊆ P (R[[x]]) and u +P∞
R[[x]].
Theorem 4.10. Homomorphic image of weakly ⋆-clean ring is weakly ⋆-clean.
Similarly we extend the Theorem 2.3 to ⋆-clean ring given below.
Theorem 4.11. Let {Rα} be a collection of ⋆-rings. Then the direct product
Rα is weakly ⋆-clean if and only if each Rα is weakly ⋆-clean ring and at
R = Qα
most one Rα is not ⋆-clean.
Proof. Similar to the proof of Theorem 2.3.
Definition 4.12. An element x in a ⋆-ring R is said to be ⋆-r-clean if x = r + p
where r ∈ Reg(R) and p ∈ P (R). A ⋆-ring R is said to be ⋆-r-clean ring if every
element of R is ⋆-r-clean.
Proposition 4.13. Let R be a ⋆-ring and e ∈ P (R). If a ∈ eRe is strongly ⋆-clean
in eRe then a is strongly ⋆-clean in R.
Proof. Let a = f + v, where f ∈ P (eRe) and v ∈ U (eRe), so there exists w ∈
U (eRe) such that vw = e = wv. Clearly (v − (1 − e))(w − (1 − e)) = 1 implies
v − (1 − e) ∈ U (R) and a − (v − (1 − e)) = f + (1 − e) ∈ P (R). Hence a is strongly
⋆ clean in R.
Proposition 4.14. Let R be an abelian ⋆-ring. If a is ⋆-clean element in R then
ae is ⋆-clean for any e ∈ P (R).
Proof. Let a = u + p1, where u ∈ U (R) and p1 ∈ P (R). Now ae = ue + p1e.
Clearly ue ∈ U (eRe) and p1e ∈ P (eRe) imply ae is strongly ⋆-clean in eRe, so ae
is strongly ⋆-clean in R.
Proposition 4.15. Let R be an abelian ⋆-ring and a be a ⋆-clean element in R
and e ∈ P (R). If −a is ⋆-clean then a + e is also ⋆-clean.
12
Proof. Clearly a and a + 1 are ⋆-clean, as a is ⋆-clean implies 1 − a is so. Let
a = u + f and 1 + a = v + g, where f, g ∈ P (R) and u, v ∈ U (R). a + e =
(1 + a)e + a(1 − e) = (ve + u(1 − e)) + (ge + f (1 − e)). But (ve + u(1 − e)) ∈ U (R)
and (ge + f (1 − e)) ∈ P (R). So a + e is ⋆-clean.
Lemma 4.16. Let R be an abelian ⋆-ring where every idempotent of the form ry
or yr is projection, for any regular element r, then r is ⋆-clean.
Proof. Let r ∈ Reg(R) then r = ryr, for some y ∈ R. So f = yr ∈ P (R). Let
e = f + (1 − f )rf = (y + (1 − f )ry)r = ar, where a = y + (1 − f )ry. Here
e2 = e ∈ P (R) and (1 − e) = (1 − f )(1 − r) = b(1 − r) ∈ R(1 − r), where
b = (1 − f ). Assume ea = a so that ara = a. Since idempotents are central so
ra = r(ar)a = ra(ra) = (ra)ra = a(ra)r = ar. Similarly (1 − r)b = b(1 − r),
where it is assumed that (1 − e)b = b. By Proposition 1.8 [8], a − b is the inverse
of r − (1 − e) and so r is ⋆-clean.
Theorem 4.17. Let R be an abelian ⋆-clean, where any idempotent of the form
e = ry or yr is projection, for any r ∈ Reg(R). Then R is ⋆-r-clean if and only if
R is ⋆-clean.
Proof. (⇐) Obviously R is ⋆-clean ⇒ R is ⋆-r-clean.
(⇒) Let R be a ⋆-r-clean ring and x ∈ R, then x = r + e′, where e′ ∈ P (R)
and r ∈ Reg(R). Therefore r = ryr, for some y ∈ R. Taking e = ry we see
that (re + (1 − e))(ye + (1 − e)) = 1 and e ∈ P (R). Hence u = re + (1 − e) is
a unit and r = eu. Set f = 1 − e then −(eu + f ) is a unit and f ∈ P (R). So,
−r = −(ue + f ) + f is ⋆-clean. Also by Lemma 4.16, r is ⋆-clean. Hence by
Proposition 4.15, x = r + e′ is ⋆-clean.
References
[1] Ahn M-S. and Anddreson D.D., Weak clean rings and almost clean rings,
Rocky mount. Jorn. math. 36, (2006) no 3, 783-798.
[2] Ashrafi N. and Nasibi E., Rings in which elements are the sum of idempotent
and a regular element, Buttet. Iranian. Math. Soc 39, 2013 No. 3, 579 − 588.
[3] Ashrafi N. and Nasibi E., r-clean Rings, Math. Report 15(65), 2013, No. 2.
[4] Brown B. and McCoy N., The Maximal Regular ideal of a ring, Proc. Amer.
Math. Soc. 1, (1950) 165-171.
[5] Danchev P. V., On Weakly Exchange Rings, Univ. of Plovdiv, 13, 2014.
13
[6] Diesl A. J., Classes of Strongly Clean Rings, Ph D. Thesis, University of
California, Berkeley, (2006).
[7] Goodearl K.R., Von Neuman Regular Rings, Robert E. Krieger Publishing
Co. Inc. Malabar, 1991.
[8] Nicholson W.K., Lifting idempotents and exchange rings(1). Trans. Amer.
Math. Soc, (1977).
[9] Nicholson W. K. and Zhou Y., Rings in which elements are uniquely the sum
of an idempotent and a unit, Glasg. Math. J., 46, (2004), No. 2, 227 − 236.
[10] Tamer Kosan, Serap Sahinkaya and Zhou Y. On weakly clean rings, Comm.
in Algebra, 2015.
[11] Wang Z., Cui J., A Note on ⋆-Clean Rings. Dept. of Math. Southern. Univ.,
(2015).
14
|
1907.06958 | 2 | 1907 | 2019-11-08T21:41:32 | Actions of cocommutative Hopf algebras | [
"math.RA"
] | Let $H$ be a cocommutative Hopf algebra acting on an algebra $A$. Assuming the base field to be algebraically closed and the $H$-action on $A$ to be integral, that is, it is given by a coaction of some Hopf subalgebra of the finite dual $H^\circ$ that is an integral domain, we stratify the prime spectrum $\mbox{Spec}\, A$ in terms of the prime spectra of certain commutative algebras. For arbitrary $H$-actions in characteristic $0$, we show that the largest $H$-stable ideal of $A$ that is contained in a given semiprime ideal of $A$ is semiprime as well. | math.RA | math |
ACTIONS OF COCOMMUTATIVE HOPF ALGEBRAS
MARTIN LORENZ, BACH NGUYEN, AND RAMY YAMMINE
ABSTRACT. Let H be a cocommutative Hopf algebra acting on an algebra A. Assuming the
base field to be algebraically closed and the H-action on A to be integral, that is, it is given by
a coaction of some Hopf subalgebra of the finite dual H ◦ that is an integral domain, we stratify
the prime spectrum Spec A in terms of the prime spectra of certain commutative algebras. For
arbitrary H-actions in characteristic 0, we show that the largest H-stable ideal of A that is
contained in a given semiprime ideal of A is semiprime as well.
INTRODUCTION
0.1. Let H be a Hopf algebra over a field k and let A be an arbitrary associative k-algebra. An
action of H on A is given by a k-linear map H⊗A → A, h⊗a 7→ h.a, that makes A into a left
H-module and satisfies the "measuring" conditions h.(ab) = (h1.a)(h2.b) and h.1 = hε, hi1
for h ∈ H and a, b ∈ A. Here, ⊗ = ⊗k , ∆h = h1 ⊗ h2 denotes the comultiplication of H,
and ε is the counit. We will write H A to indicate such an action. Algebras equipped with
an H-action are called left H-module algebras. With algebra maps that are also H-module
maps as morphisms, left H-module algebras form a category, H Alg.
For example, an action of a group algebra kG on A amounts to the datum of a group
homomorphism G → Aut A, the automorphism group of the algebra A. For the enveloping
algebra U g of a Lie k-algebra g, an action U g A is given by a Lie homomorphism g →
Der A, the Lie algebra of all derivations of A. In both these prototypical cases, the acting
Hopf algebra is cocommutative. This article investigates the effect of a given action H A of
an arbitrary cocommutative Hopf algebra H on the prime and semiprime ideals of A, partially
generalizing prior work of the first author on rational actions of algebraic groups [11], [12],
[13]. The interesting article [18] by Skryabin covers related territory, but the actual overlap
with our work is insubstantial.
0.2. For now, let H be arbitrary and let A ∈ H Alg. An ideal I of A that is also an H-
submodule of A is called an H-ideal. The action H A then passes down to an H-action on
the quotient algebra A/I. The sum of all H-ideals that are contained in an arbitrary ideal I,
clearly the unique largest H-ideal of A that is contained in I, will be referred to as the H-core
of I and denoted by I:H. Explicitly,
(1)
If A 6= 0 and the product of any two nonzero H-ideals of A is again nonzero, then A is said
to be H-prime. An H-ideal I of A is called H-prime if A/I is H-prime. It is easy to see
I:H = {a ∈ A H.a ⊆ I}.
2010 Mathematics Subject Classification. 16T05, 16T20.
Key words and phrases. Hopf algebra, action, quantum invariant theory, prime spectrum, stratification, prime
ideal, semiprime ideal, integral action, rational action, algebraic group, Lie algebra, derivation.
1
2
MARTIN LORENZ, BACH NGUYEN, AND RAMY YAMMINE
that H-cores of prime ideals are H-prime. Denoting the collection of all H-primes of A by
H-Spec A, we thus obtain a map Spec A → H-Spec A , P 7→ P :H. The fibers
SpecI A def= {P ∈ Spec A P :H = I}
are called the H-strata of Spec A. The stratification Spec A = FI∈H-Spec A SpecI A was
pioneered by Goodearl and Letzter [7] in the case of group actions or, equivalently, actions
of group algebras. It has proven to be a useful tool for investigating Spec A, especially for
rational actions of a connected affine algebraic group G over an algebraically closed field k.
In this case, one has a description of each stratum SpecI A in terms of the prime spectrum of
a suitable commutative algebra [12, Theorem 9]. Our principal goal is to generalize this result
to the context of cocommutative Hopf algebras. This is carried out in Section 3 of this article.
The first two sections serve to deploy some generalities on actions of Hopf algebras that are
needed for the proof, with Section 2 focusing on the cocommutative case.
0.3. To state our main result, we make the following observations; see Sections 1-3 for details.
Let H be cocommutative and k algebraically closed. Assume that the action H A is locally
finite, that is, dimk H.a < ∞ for all a ∈ A. Then A becomes a right comodule algebra over
the (commutative) finite dual H ◦ of H:
A
a
A ⊗ H ◦
a0 ⊗ a1
with h.a = a0ha1, hi
(h ∈ H, a ∈ A).
(2)
The action H A will be called integral if the image of the map (2) is contained in A ⊗ O
for some Hopf subalgebra O ⊆ H ◦ that is an integral domain. This condition serves as
a replacement for connectedness in the case of algebraic group actions. Assuming it to be
satisfied, it follows that each I ∈ H-Spec A is in fact a prime ideal of A. Consequently, the
extended center C(A/I) = Z Q(A/I) is a k-field, where Q(A/I) denotes the symmetric ring
of quotients of A/I. The action H A/I extends uniquely to an H-action on Q(A/I) and
this action stabilizes the center C(A/I). Furthermore, O ∈ H Alg via the "hit" action ⇀ that
is given by hh⇀f, ki = hf, khi for f ∈ O and h, k ∈ H. The actions ⇀ and H C(A/I)
combine to an H-action on the tensor product; so
CI := C(A/I) ⊗ O ∈ H Alg .
(3)
The algebra CI is a commutative integral domain. We let SpecHCI denote the subset of
Spec CI consisting of all prime H-ideals of CI .
Theorem 1. Let H be a cocommutative Hopf algebra over an algebraically closed field k
and let A be a k-algebra that is equipped with an integral action H A. Then, for any
I ∈ H-Spec A, there is a bijection
c : SpecI A ∼ SpecHCI
having the following properties, for any P, P ′ ∈ SpecI A:
(i) c(P ) ⊆ c(P ′) if and only if P ⊆ P ′, and
(ii) Fract(CI /c(P )) ∼= C((A/P ) ⊗ O).
ACTIONS OF COCOMMUTATIVE HOPF ALGEBRAS
3
0.4. As a first example, let G be an affine algebraic k-group and let O = O(G) be the algebra
of polynomial functions on G. Then O ⊆ H ◦ with H = kG. A rational G-action on A, by
definition, is a locally finite action H A such that the image of (2) is contained in A ⊗ O.
If G is connected, then O is an integral domain and so the action is integral. In this setting,
Theorem 1 is covered by [12, Theorem 9].
Next, let g be a Lie k-algebra acting by derivations on A and assume that every a ∈ A is
contained in some finite-dimensional g-stable subspace of A. With H = U g, we then have a
locally finite action H A and hence a map (2). If char k = 0, then the convolution algebra
H ∗ is a power series algebra over k and hence H ∗ is a commutative domain; see §4.3 below.
Since H ◦ is a subalgebra of H ∗, we may take O = H ◦ and so we have an integral action.
Theorem 1 appears to be new in this case.
0.5. In Section 4, we show that if char k = 0 and H is cocommutative, then the core operator
· :H preserves semiprimess. Recall that an ideal I of A is called semiprime if I is an inter-
section of prime ideals or, equivalently, A/I has no nonzero nilpotent ideals. In Section 4,
we turn to the question as to whether I:H is then semiprime as well. This question may be
reformulated in various alternative ways (Lemma 15). In general, the answer is negative: even
for a cocommutative Hopf algebra H, semiprimeness may be lost upon passage to the H-core.
For instance, consider the group algebra A = kG with its standard G-grading. For G finite,
this grading amounts to an action of the Hopf dual H = (kG)∗ on A. The only H-ideals
of A are 0 and A; so the H-core of every proper ideal of A is 0. If G is abelian, then H is
cocommutative. If the operator · :H is to preserve semiprimeness for any such G, then we
must require that char k = 0 by Maschke's Theorem. It turns out that this is also sufficient in
general:
Theorem 2. Let A ∈ H Alg and assume that H is cocommutative and char k = 0. Then I:H
is semiprime for every semiprime ideal I of A.
0.6. In future work, we hope to pursue the general theme of this article for Hopf algebras that
are not necessarily cocommutative. In particular, we plan to address "rationality" of prime
ideals and explore the Dixmier-Moeglin equivalence in the context of Hopf algebra actions,
generalizing the work on group actions in [11], [12].
Notations and conventions. We work over a base field k and continue to write ⊗ = ⊗k.
Throughout, H is a Hopf k-algebra, cocommutative when so specified. The counit and the
antipode of H will be denoted by ε and S, respectively. Furthermore, we fix the following
notations for the remainder of this article:
A will be a left H-module algebra, with H-action h ⊗ a 7→ h.a (h ∈ H, a ∈ A);
AH = {a ∈ A h.a = hε, hia for all h ∈ H} is the subalgebra of H-invariants;
B = Homk(H, A) will denote the convolution algebra.
1. GENERALITIES ON ACTIONS
1.1. The convolution algebra B. We begin by introducing certain subalgebras and automor-
phisms of B as well as some H-operations on B that will be used throughout this article.
First, the counit of H and the action H A give rise to the following maps ι, δ : A → B,
which are easily seen to be k-algebra embeddings:
ιa = (h 7→ hε, hia)
δa = (h 7→ h.a)
(h ∈ H, a ∈ A).
4
MARTIN LORENZ, BACH NGUYEN, AND RAMY YAMMINE
We will generally identify A with ιA and we will also identify the linear dual H ∗ with the
image of the algebra map u∗ : H ∗ ֒→ B that is given by the unit u : k → A. In this way, we
view A, H ∗, and A ⊗ H ∗ as k-subalgebras of B, with A ⊗ H ∗ consisting of all k-linear maps
H → A that have finite rank. Explicitly, a⊗ f = (h 7→ ahf, hi), ιa = a⊗ ε and u∗f = 1⊗ f
for a ∈ A, f ∈ H ∗.
The algebra B becomes a left H-module algebra via the ⇀-action, which is defined by
(h⇀b)(k) = b(kh)
(4)
We will write (B, ⇀) when viewing B ∈ H Alg with (4). The map δ : A → (B, ⇀) is a
morphism in H Alg and A ⊗ H ∗ is an H-module subalgebra of (B, ⇀).
Our main focus later will be on the following alternative left H-module structure on B,
which takes into account the given action H A:
(h, k ∈ H, b ∈ B).
(5)
If H is cocommutative, then B ∈ H Alg with (5) as well; see Proposition 6 below. In general,
we may pass between (4) and (5) by means of the k-linear automorphisms Φ, Ψ : B ∼ B
that are defined by
(h · b)(k) = h1.b(kh2).
Φb = (h 7→ h1.b(h2))
and Ψb = (h 7→ S(h1).b(h2)).
These maps are inverse to each other and they satisfy the following "intertwining" formulas,
for any a ∈ A, b ∈ B and h ∈ H:
Φ((ιa)b) = δa Φb, Ψ((δa)b) = ιa Ψb
(6)
Φ(h · b) = h⇀(Φb),
h · (Ψb) = Ψ(h⇀b).
(7)
The Ψ-identities follow from those for Φ = Ψ−1. To verify (6), one checks that both sides
send a given h ∈ H to h1.(cid:0)a b(h2)(cid:1). For (7), observe that (Φb)(h) = (h · b)(1) and so
Φ(h · b)(k) = (kh · b)(1) = (Φb)(kh) = (h⇀(Φb))(k) for h, k ∈ H and b ∈ B. Finally,
note that Φ and Ψ fix the unit element 1B = u∗(ε) and they are right linear for the subalgebra
Homk(H, AH ) ⊆ B. In particular, Φ and Ψ restrict to the identity map on H ∗ and are right
H ∗-linear.
1.2. Invariants and the locally finite part of (B, ⇀). The locally finite part of any H-
module algebra A is defined by
Afin : = {a ∈ A dimk H.a < ∞}
= {a ∈ A I.a = 0 for some cofinite ideal I of H}.
Here, "cofinite" is short for "having finite codimension." The locally finite part Afin is always
an H-module subalgebra of A containing the algebra of invariants, AH .
The following lemma determines the invariants and the locally finite part of (B, ⇀).
Lemma 3.
(a) (B, ⇀)H = ιA ∼= A;
(b) (B, ⇀)fin = {b ∈ B Ker b contains some cofinite ideal of H} ∼= A ⊗ H ◦.
Proof. (a) Note that b ∈ (B, ⇀)H if and only if b(kh) = hε, hib(k) for all h, k ∈ H. Equiv-
alently, b(h) = hε, hib(1) for all h ∈ H, which in turn states that b = ι(b(1)). The assertion
about H-invariants follows.
ACTIONS OF COCOMMUTATIVE HOPF ALGEBRAS
5
(b) For any ideal I of H and any b ∈ B, the equality I ⇀b = 0 is equivalent to I ⊆
Ker b; so b ∈ (B, ⇀)fin if and only if Ker b contains some cofinite ideal of H. In particular,
ιA ⊆ (B, ⇀)fin and the embedding u∗ : H ∗ ֒→ B sends the finite dual H ◦ to (B, ⇀)fin. The
isomorphism between the subalgebra of B consisting of all finite-rank maps and A ⊗ H ∗
(§1.1) restricts to an isomorphism of subalgebras, (B, ⇀)fin ∼= A ⊗ H ◦.
(cid:3)
The isomorphisms in Lemma 3 will be treated as identifications as in §1.1. In particular,
we will write A ⊗ H ◦ = (B, ⇀)fin and identify A and H ◦ with the subalgebras ιA = A ⊗ ε
and u∗H ◦ = 1 ⊗ H ◦, respectively. As was mentioned, (B, ⇀)fin is an H-module subalgebra
of (B, ⇀); explicitly, with b = a ⊗ f (a ∈ A, f ∈ H ◦) formula (4) becomes
h⇀(a ⊗ f ) = a ⊗ (h⇀f ) = a ⊗ f1hf2, hi.
(8)
Furthermore, (B, ⇀)fin = A ⊗ H ◦ is also stable under the H-operation (5), which becomes
the standard Hopf operation on tensor products:
h · (a ⊗ f ) = h1.a ⊗ (h2⇀f ) = h1.a ⊗ f1hf2, h2i.
More generally, for any Hopf subalgebra O ⊆ H ◦, we will consider the subalgebra
(9)
AO := A ⊗ O ⊆ (B, ⇀)fin .
Each such AO is stable under both ⇀ and · by (8) and (9).
Lemma 4. Let O ⊆ H ◦ be a Hopf subalgebra. The k-subspaces of AO that are stable under
right multiplication by O and under the ⇀-action (8) are exactly those of the form W ⊗ O,
where W is an arbitrary k-subspace of A.
Proof. Certainly, each W ⊗ O is stable under right multiplication by O and under (8). For
the converse, equip AO with the "trivial" right O-Hopf module structure: the right O-module
and right O-comodule structures are given by IdA ⊗mO and d = IdA ⊗∆O, where mO
and ∆O are the multiplication and comultiplication of O, respectively [14, Examples 10.3,
10.4]. Then AO ֒→ AO ⊗ O ֒→ Homk(H, AO) via d, with (db)(h) = h⇀b for b ∈ AO.
Now let V ⊆ AO be a k-subspace that is stable under right multiplication by O and under
(8). Then dV ⊆ Homk(H, V ) ∩ (AO ⊗ O) = V ⊗ O and so V is a O-Hopf submodule
of AO. By the Structure Theorem for Hopf modules [14, §10.1.2], V is generated as right
O-module by the subspace of O-coinvariants, V co O = {v ∈ V δv = v ⊗ 1}. Since
V co O ⊆ (AO)co O = A ⊗ Oco O = A ⊗ k, it follows that V = W ⊗ O with W = V co O. (cid:3)
1.3. Coefficient Hopf algebras of locally finite actions.
1.3.1. Assume that the action H A is locally finite, that is, A = Afin. Then, since δ : A →
(B, ⇀) is a morphism in H Alg (§1.1), the image δA is contained in (B, ⇀)fin = A ⊗ H ◦ and
A becomes a right H ◦-comodule algebra via δ [14, Proposition 10.26]. Any Hopf subalgebra
O ⊆ H ◦ such that δA is contained in the subalgebra AO = A⊗O ⊆ (B, ⇀)fin will be called a
coefficient Hopf algebra for H A. We then have the following version of (2), which makes
A a right O-comodule algebra:
AO
with h.a = a0ha1, hi
(h ∈ H, a ∈ A).
(10)
δ : A
a
a0 ⊗ a1
6
MARTIN LORENZ, BACH NGUYEN, AND RAMY YAMMINE
Any coefficient Hopf algebra for H A will also serve as a coefficient Hopf algebra for the
H-action on quotients of A modulo H-ideals and on H-module subalgebras of A, and the
intersection of all coefficient Hopf subalgebras for H A is the unique smallest one.
Lemma 5. Let H A be locally finite and let O ⊆ H ◦ be a coefficient Hopf algebra. Then:
(a) All subspaces of AO of the form W ⊗ O, where W ⊆ A is an H-stable k-subspace,
(b) If I is an ideal of A, then I:H = Ψ(I ⊗ O) ∩ A.
are stable under the maps Φ, Ψ (§1.1).
Proof. (a) If W ⊆ A is H-stable, then δW ⊆ W ⊗ O. Thus, for any a ∈ W and f ∈ O,
right O-linearity of Φ and (6) now give Φ(a ⊗ f ) = δ(a)f = a0 ⊗ a1f ∈ W ⊗ O. Similarly,
Ψ(a ⊗ f ) = a0 ⊗ S∗(a1)f ∈ W ⊗ O , proving stability of W ⊗ O under Φ and Ψ.
(b) By (1), I:H = δ−1(Homk(H, I)) = δ−1(I ⊗ O). Putting I ′ = Ψ(I ⊗ O) and using
the identity Φ ◦ ι = δ from (6), we obtain I:H = δ−1(I ⊗ O) = (ι)−1(I ′) = I ′ ∩ A.
1.3.2. The action H A is certainly locally finite if the H-module algebra A under consid-
eration is generated as k-algebra by a finite-dimensional H-stable subspace V ⊆ A. Assume
this to be the case and let ρ : H → Endk(V ) denote the algebra map given by the operation of
i ) denote the dual basis of V ∗, we obtain
H on V . Fixing a k-basis (vi)n
the basis (vi ⊗ v∗
j ) of Endk(V ) ∼= V ⊗ V ∗ and linear forms ρi,j such that
1 of V and letting (v∗
(cid:3)
vi ⊗ v∗
jhρi,j, hi
(h ∈ H).
(11)
ρh = X
i,j
δv = X
i,j
The isomorphism Endk(V ) ∼= Mn(k) given by our choice of basis for V allows us to write
ρ = (ρi,j) : H → Mn(k): the scalar hρi,j, hi ∈ k is the (i, j)-entry of the matrix ρh. Thus,
ρi,j ∈ H ◦, because ρi,j vanishes on the cofinite ideal Ker ρ, and ∆ρi,j = Pk ρi,k ⊗ ρk,j. On
the generating subspace V ⊆ A, the algebra map (10) takes the form
vihv∗
j , vi ⊗ ρi,j
(v ∈ V )
(12)
and we may take O to be the Hopf subalgebra of H ◦ that is generated by the matrix coefficient
functions ρi,j. If H is involutory, then O is the k-subalgebra of H ◦ that is generated by all
ρi,j and S∗ρi,j. We will call O the coefficient Hopf algebra of the representation V ∈ Rep H;
a more general situation is discussed in [14, Exercise 9.2.3].
2.1. The · -action.
2. THE COCOMMUTATIVE CASE
2.1.1. We recall some general ring-theoretic material that will be tacitly used in the next
proposition and throughout the remainder of this article. A ring homomorphism f : R → S
is called centralizing if the ring S is generated by the image f R and its centralizer in S. If f
makes S a free R-module having a basis that centralizes f R, then f is called free centralizing.
Any centralizing homomorphism f restricts to a map of centers, ZR → ZS; the assignment
I 7→ (f I)S sends (two-sided) ideals of R to ideals of S; and P 7→ f −1P gives a well-defined
map Spec S → Spec R [11, §1.5].
ACTIONS OF COCOMMUTATIVE HOPF ALGEBRAS
7
2.1.2. The proposition below shows that, for H cocommutative, we may view B and various
subalgebras of B as H-module algebras with action (5), which we will refer to as the · -action,
rather than the ⇀-action (4). When using the latter H-action, we will continue write (B, ⇀);
otherwise, the · -action is assumed.
Proposition 6. Let H be cocommutative and let O ⊆ H ◦ be a coefficient Hopf algebra for
the action H Afin . Then:
(a) Φ, Ψ are algebra automorphisms of B stabilizing the subalgebra Afin,O := Afin ⊗ O.
(b) B ∈ H Alg with the · -action and BH = ΨA.
(c) AO is an H-module subalgebra of B, with (AO)fin = Afin,O and (AO)H = Ψ(Afin).
The inclusion (AO)H ⊆ (AO)fin is free centralizing.
(d) The following maps are bijections that are inverse to each other:
{ideals of Afin}
∼
I
Φ(J ′′) ∩ Afin
{H-ideals of Afin,O}
I ′ := Ψ(I ⊗ O)
J ′′ := JAfin,O
∼
{ideals of (AO)H}
I ′ ∩ (AO)H
J
Proof. (a) For b, b′ ∈ B and h ∈ H, one computes using cocommutativity,
Φ(bb′)(h) = h1.(b(h2)b′(h3)) = (h1.b(h3))(h2.b′(h4))
= (h1.b(h2))(h3.b′(h4)) = (Φ(b)Φ(b′))(h).
Thus Φ is multiplicative and hence it is an algebra automorphisms of B; likewise for Ψ =
Φ−1. Stability of the subalgebra Afin,O under Φ and Ψ follows from Lemma 5(a) for Afin.
(b) Since Ψ is an algebra automorphism of B, the intertwining formula (7) together with
the fact that (B, ⇀) ∈ H Alg implies that B = (B, · ) ∈ H Alg as well. Also by (7), Ψ
yields an algebra isomorphism (B, ⇀)H ∼= BH and so Lemma 3(a) gives the isomorphism
Ψ◦ι : A ∼ (B, ⇀)H ∼ BH. Identifying A with ιA = A⊗ε as usual, we obtain the claimed
equality BH = ΨA.
(c) The subalgebra AO ⊆ B is stable under the · -action by (9) and so it is an H-module
subalgebra of B. The equality (AO)fin = Afin,O follows from [10, Corollary 6], because the
H-action ⇀ on O is locally finite. Since S is an involution of H, the formula (Ψa)(h) =
S(h1).ahε, h2i = S(h).a (a ∈ A, h ∈ H) shows that Ψa vanishes on some cofinite ideal
of H if and only if a ∈ Afin. In that case, Ψa ∈ Afin,O ⊆ AO by (a). Thus, in view of
part (b) and Lemma 3(b), Ψ(Afin) = BH ∩ (B, ⇀)fin = (AO)H as asserted. Finally, since
(AO)fin = Afin,O = Ψ(Afin,O) is generated by the commuting subalgebras Ψ(Afin) = (AO)H
and ΨO, with ΨO providing an (AO)H -basis of (AO)fin, the inclusion (AO)H ⊆ (AO)fin is
(d) Let J be an ideal of (AO)H . Since the inclusion (AO)H ⊆ (AO)fin = Afin,O is central-
izing by (c), JAfin,O is an ideal of Afin,O, clearly an H-ideal. Further, JAfin,O ∩ (AO)H = J
by freeness of Afin,O over (AO)H . Now let L be any H-ideal of Afin,O. Then ΦL is an ideal
of Afin,O that is stable for the ⇀-action by (7). Thus, Lemma 4 implies that ΦL = I ⊗ O for
some ideal I of Afin. Therefore, L = Ψ(I ⊗ O) = ΨI ΨO = (ΨI)Afin,O with ΨI being an
ideal of (AO)H = Ψ(Afin). This proves that extension and contraction give inverse bijections
between the sets of ideals of (AO)H and H-ideals of Afin,O. The bijection between the sets of
ideals of (AO)H and of Afin is a consequence of the equality Ψ(Afin) = (AO)H .
(cid:3)
free centralizing.
8
MARTIN LORENZ, BACH NGUYEN, AND RAMY YAMMINE
2.2. Integral actions.
2.2.1. A locally finite action H A will be called integral if it has a coefficient Hopf algebra
that is a commutative integral domain. Even though H need not a priori be cocommutative,
any integral H-action factors through a cocommutative Hopf quotient of H.
Proposition 7. Let H A be integral and k algebraically closed. Then P :H is prime for
every P ∈ Spec A. Furthermore, H-Spec A is the set of all prime H-ideals of A.
Proof. Let O ⊆ H ◦ be a coefficient Hopf algebra of the action H A that is an integral
domain, and assume that H is cocommutative, as we may. Then Ψ is an algebra automorphism
of B that restricts to an automorphism of the subalgebra AO (Proposition 6). For any P ∈
Spec A, the ideal P ⊗ O of AO is prime [14, Lemma 11.19], and hence
P ′ := Ψ(P ⊗ O) ∈ Spec AO.
(13)
By Lemma 5(b), P :H = P ′∩A, which is a prime ideal of A, because A ֒→ AO is centralizing.
The final assertion also follows, because the map Spec A → H-Spec A , P 7→ P :H, is
surjective for any locally finite action H A [14, Exercise 10.4.4].
2.2.2. Returning to the situation considered in §1.3.2, assume that A is affine, generated by
an H-stable subspace V ⊆ A with n = dimk V < ∞. We use our earlier notation, but we
now also assume that H is cocommutative and so involutory. The k-subalgebra O ⊆ H ◦ that
is generated by the functions ρi,j and S∗ρi,j is thus a coefficient Hopf algebra for H A.
(cid:3)
Example 8 (Group algebras). Let H = kG be a group algebra. Then H ∗ is the algebra kG
of all functions G → k with pointwise addition and multiplication. The subalgebra H ◦ is
commonly referred to as the algebra of representative functions on G and denoted by Rk(G)
(e.g., [8, Chapter 1]). For any g ∈ G, the determinant det ρg is nonzero and hS∗ρi,j, gi =
hρi,j, g−1i = 1
det ρghcj,i, gi, where hcj,i, gi denotes the (j, i)-cofactor of the matrix ρg ∈
GLn(k), a polynomial in the entries of ρg. Thus, cj,i ∈ R := k[ρi,j i, j = 1, . . . , n] and
S∗ρi,j = cj,i
det ρ ∈ R[(det ρ)−1], the subalgebra of H ◦ that is generated by the functions ρi,j
and (det ρ)−1 = S∗ det ρ. We obtain
O = R[(det ρ)−1] = k[(det ρ)−1, ρi,j i, j = 1, . . . , n].
Now let k be algebraically closed. The group GLn(k) is affine algebraic, with associated
Hopf algebra O(GLn) as in [14, Example 9.19]. The Hopf algebra O is the image of O(GLn)
under restriction to ρG ≤ GLn(k). Finally, O is a domain if and only if ρG is an irreducible
subset of GLn(k) in the Zariski topology or, equivalently, the closure ρG is a connected affine
algebraic group.
Example 9 (Enveloping algebras). Let g be a Lie k-algebra and H = U g its enveloping
algebra. If char k = 0, then H ∗ is a (commutative) domain; see [4, Chap. II, §1, no 5] or
Section 4.3 below. Therefore, the subalgebra H ◦ and all coefficient Hopf algebras O are
integral domains as well in characteristic 0. The situation is different for char k = p > 0.
Indeed, in this case, hf p, gi = 0 for any f ∈ H ∗. If f ∈ H ◦ is grouplike or, equivalently,
an algebra map, then so is f p and hence Ker f p is the ideal of H that is generated by g.
Therefore, f p = ε and so (f − ε)p = 0. If g 6= [g, g] then we may choose f 6= ε, giving a
nonzero nilpotent element of H ◦.
ACTIONS OF COCOMMUTATIVE HOPF ALGEBRAS
9
2.3. The symmetric ring of quotients and the extended center.
2.3.1. We briefly recall some background material on the symmetric ring of quotients QR of
an arbitrary ring R and its center, CR := Z(QR), the so-called extended center of R. See [14,
Appendix E] for details. The ring R is a subring of QR and CR coincides with the centralizer
of R in QR. In particular, ZR ⊆ CR. If ZR = CR, then R is called centrally closed. In
general, we may consider the following subring of QR, possibly strictly larger than R:
eR := R(CR) ⊆ QR.
If R is semiprime, then eR is a centrally closed ring [2, Theorem 3.2], called the central closure
of R. If R is a k-algebra, then so is eR, because ZR ⊆ CR = Z eR.
2.3.2. The next lemma concerns the extension of a given action H A to an action on QA.
Part (a), which summarizes known facts, shows that this is always possible, in a unique way,
in the situation that we are interested in. Indeed, any cocommutative Hopf algebra over an
algebraically closed field is pointed [19, Lemma 8.0.1]. However, local finiteness of an action
may be lost in the process. For instance, consider the action of the group k× on the polynomial
algebra A = k[x] that is determined by λ.x = λx for λ ∈ k× and the extended action on the
field QA = Fract A = k(x). If k is infinite, then k(x)fin = k[x±1]. Part (b) of the lemma
focuses on the action H (QA)fin rather than H QA.
Lemma 10. Let H be pointed cocommutative. Then:
(a) The H-action on A extends uniquely to an action H QA and this action stabilizes
(b) If the action H A is locally finite with coefficient Hopf algebra O, then the extended
the subalgebras eA = A(CA) and CA.
action H (QA)fin also has coefficient Hopf algebra O.
(b) Put R = (QA)fin. So A ⊆ R and the action H
Proof. (a) Since H is pointed, [16, Cor. 3.5] tells us that the H-action on A extends uniquely
to an action on QA. This action stabilizes the center, CA, because H is cocommutative [5,
Prop. 4]. Therefore, eA is stable as well.
QA restricts to a locally finite
action H R extending the H-action on A. We must show that the map δ : R → R ⊗ H ◦
(§1.3) has image in RO = R ⊗ O, given that δA ⊆ AO. Fix r ∈ R and consider the
subspace V = H.r ⊆ R, which is finite dimensional and H-stable. Therefore, the ideal
I = {a ∈ A aAV ⊆ A} of A has zero (left and) right annihilator in R [14, Proposition E.1]
and I is an H-ideal as is readily verified using the following standard identity for H-module
algebras:
(h.x)y = h1.(x(S(h2).y))
(14)
By Lemma 5(a), Φ(I ⊗ O) = I ⊗ O. Furthermore, by Proposition 6(a), Φ gives an algebra
automorphism of R ⊗ H ◦ stabilizing AO and satisfying Φ ◦ ι = δ by (6). Therefore,
(I ⊗ ε)δV ⊆ Φ((I ⊗ O)(V ⊗ ε)) = Φ(IV ⊗ O) ⊆ Φ(A ⊗ O) = AO .
(h ∈ H, x, y ∈ R).
Thus, (I ⊗ ε)δr ⊆ RO . Since δr ∈ R ⊗ H ◦ and I has zero right annihilator in R, it follows
that δr ∈ RO, as desired.
(cid:3)
10
MARTIN LORENZ, BACH NGUYEN, AND RAMY YAMMINE
3. PRIME STRATA
3.1. Prime correspondences. The central closure eR of a prime ring R is also prime and its
center, the extended center CR, is a field. Thus, for an arbitrary ring R, we may associate to
any P ∈ Spec R the field C(R/P ), called the heart of P .
Proposition 11. Let R be a centrally closed prime k-algebra and put K = CR. Let S be any
K-algebra and put U = R ⊗K S. Then we have the following bijections, which are inverse
to each other:
{P ∈ Spec U P ∩ R = 0}
∼
P
U p = R ⊗K p
Spec S
P ∩ S
p
This correspondence preserves hearts: C(U/P ) ∼= C(S/P ∩ S). Moreover, if U ∈ H Alg with
S being an H-module subalgebra, then H-ideals are matched with H-ideals.
Proof. Apart from the last assertion, involving an H-action, the proposition is identical with
[12, Proposition 5]. The bijections in the proposition, contraction and extension of ideals,
both evidently send H-ideals to H-ideals in the given situation.
(cid:3)
Proposition 12. Assume that H is pointed cocommutative and let R, T ∈ H Alg, with R being
prime. Put V := R ⊗ T ⊆ eV := eR ⊗ T , where eR denotes the central closure of R. Then
there is a bijection
{P ∈ Spec eV P ∩ eR = 0}
∼
{Q ∈ Spec V Q ∩ R = 0}
P
P ∩ V
This bijection is an order isomorphism for ⊆ and it preserves hearts. Moreover, viewing
eR ∈ H Alg with the extended H-action (Lemma 10) and V, eV ∈ H Alg with the standard
H-action on tensor products, the bijection gives a bijection on H-stable primes.
Proof. Again, this proposition is covered by [12, Proposition 6] except for the statement about
H-stability. To justify this assertion, note that tensor products of H-module algebras are again
H-module algebras, with the standard H-action via ∆, because H is cocommutative. Thus,
eV ∈ H Alg and all of R, eR, T and V are H-module subalgebras of eV . If P is an H-ideal
of eV , then its contraction, P ∩ V , is clearly an H-ideal of V . Conversely, let Q ∈ Spec V
with Q ∩ R = 0 be H-stable. In order to show that the preimage of Q under the bijection
in the proposition is an H-ideal of eV , we recall the construction of the preimage from [12].
The canonical epimorphism π : V ։ W := V /Q is a map in H Alg, and the restriction
ρ := π(cid:12)(cid:12)R : R ֒→ W is an embedding of prime algebras and a centralizing map in H Alg. By
[12, Lemma 4], ρ extends uniquely to an embedding of central closures, eρ : eR ֒→ fW , which
Claim. eρ is a map in H Alg for the extended action H fW (Lemma 10).
To prove the claim, let er ∈ eR and h ∈ H. Fix a nonzero ideal I of R such that (h.er)I
and all er(S(h2).I) are contained in R; this is possible by continuity of the action H R [16,
is also centralizing.
ACTIONS OF COCOMMUTATIVE HOPF ALGEBRAS
11
Proposition 2.2]. Using H-equivariance of ρ one computes, with x ∈ I,
eρ(h.er)ρx = ρ((h.er)x) = ρ(h1.(er(S(h2).x))) = h1.ρ(er(S(h2).x))
= h1.(eρer ρ(S(h2).x)) = h1.(eρer (S(h2).ρx)) = (h.eρer)ρx,
where the last equality uses the identity (14). Thus, (eρ(h.er) − h.eρer)ρI = 0. Since (ρI)fW is
a nonzero ideal of fW , it follows that eρ(h.er) = h.eρer, proving the claim.
The image πT ⊆ W ⊆ fW centralizes ρR and hence also eρeR = ρR eρ(Z eR), because
eρ(Z eR) ⊆ ZfW . Therefore, eρ and π(cid:12)(cid:12)T give a homomorphism eπ : eV = eR ⊗ T → fW . The
preimage of Q in Proposition 12 is Kereπ; see [12, proof of Proposition 6]. Finally, since
eρ and π(cid:12)(cid:12)T are maps in H Alg, it follows that eπ is likewise. Hence Kereπ is an H-ideal, as
desired.
(cid:3)
3.2. Proof of Theorem 1. Let H be a cocommutative Hopf algebra over an algebraically
closed field k and let H A be an integral action. Fix a coefficient Hopf algebra O ⊆ H ◦
that is an integral domain and write AO = A ⊗ O as before. Given I ∈ H-Spec A, our goal
is to describe the stratum SpecI A. Replacing A by A/I, we may assume that A is prime
(Proposition 7) and focus on the set Spec0 A = {P ∈ Spec A P :H = 0}.
3.2.1. Put X := {Q ∈ Spec AO Q ∩ A = 0}. We first establish a bijection between
Spec0 A and the subset X H ⊆ X consisting of all Q ∈ X that are stable for the · -action (9).
Recall that Ψ gives an automorphism of the algebra AO and, for any P ∈ Spec A, we have
P ′ := Ψ(P ⊗ O) ∈ Spec AO; see Proposition 6 and (13). If P ∈ Spec0 A, then P ′ ∈ X by
Lemma 5(b). So we have a map Spec0 A → X, P 7→ P ′, which is evidently injective. By
Proposition 6(d), the image consists of H-ideals; so P ′ ∈ X H . Conversely, let Q ∈ X H be
given and put P = Φ(Q) ∩ A; this is a prime ideal of A, because Φ(Q) ∈ Spec AO and the
embedding A ֒→ AO is centralizing. Also, Lemma 5(b) gives P :H = P ′ ∩ A ⊆ Q ∩ A = 0;
so P ∈ Spec0 A. Finally, Q = P ′ by Proposition 6(d). Thus, we have the desired bijection:
Spec0 A
∼
P
X H = {Q ∈ Spec AO Q ∩ A = 0 and Q is stable under (9)}
P ′ = Ψ(P ⊗ O)
This bijection has the following properties, with (i) being evident and (ii) resulting from the
algebra isomorphism AO/P ′ ∼= AO/(P ⊗ O) ∼= (A/P ) ⊗ O that comes from Ψ:
(i) P1 ⊆ P2 if and only if P ′
1 ⊆ P ′
(ii) C(AO/P ′) ∼= C((A/P ) ⊗ O).
2, and
3.2.2. Now put K = CA and C = K ⊗O; the former is a k-field and the latter a commutative
integral domain, identical to the algebra C0 in (3). Let eA = AK denote the central closure of
A and put eAO = eA⊗O ∼= eA⊗K C. Then we have the following isomorphisms of posets (for
⊆), with d1 coming from Proposition 12 and d2 from Proposition 11:
d : X = {Q ∈ Spec AO Q ∩ A = 0}
∼d1
{P ∈ Spec eAO P ∩ eA = 0}
∼d2
Spec C.
12
MARTIN LORENZ, BACH NGUYEN, AND RAMY YAMMINE
The composite of the bijection ′ : Spec0 A ∼ X H in §3.2.1, the inclusion X H ֒→ X, and
the above bijection d = d2 ◦ d1 yields the map
X H
Spec C.
X
∼d
c : Spec0 A
∼′
Since ′ and both di are order isomorphisms and preserve hearts, the same holds for c:
(i) c(P1) ⊆ c(P2) if and only if P1 ⊆ P2, and
(ii) Fract(C/c(P )) = C(C/c(P )) ∼= C((A/P ) ⊗ O).
Finally, Im c consists exactly of the H-ideals in Spec C. This amounts to checking that Q ∈
X is an H-ideal if and only if d(Q) ∈ Spec C is an H-ideal. But, by Lemma 10(a) and
Proposition 6(c), eAO ∈ H Alg for the · -action, with AO and C = Z( eAO) being H-module
subalgebras. By Propositions 11 and 12, we know that a given P ∈ Spec eAO with P ∩ eA = 0
is an H-ideal if and only if d2(P ) = P ∩ C ∈ Spec C is an H-ideal and if and only if
d−1
1 (P ) = P ∩ AO ∈ X is an H-ideal. This completes the proof of Theorem 1.
3.3. Tensor algebras. As an application of Theorem 1, we offer the following result on the
tensor algebra TV of a finite-dimensional representation V ∈ Rep H. The H-action on V
extends uniquely to an action H TV [14, 10.4.2].
(cid:3)
Corollary 13. Let H be a cocommutative Hopf algebra over an algebraically closed field k
and let V be a representation of H with 2 ≤ dimk V < ∞ and such that the coefficient Hopf
algebra of V (§1.3.2) is a domain. Then every nonzero prime ideal of the tensor algebra TV
contains a nonzero H-stable prime ideal.
A = TV is integral and we may apply Theorem 1 with I = 0.
Proof. The action H
By a result of Kharchenko [9], QA = A and so CA = ZA = k. Therefore, the algebra
C0 in Theorem 1 coincides with O, the coefficient Hopf algebra of the representation V .
Since C0 = O is H-simple by Lemma 4, we have SpecHC0 = {0} and Theorem 1 gives
Spec0 A = {0}. Consequently, if P is a nonzero prime ideal of A, then P /∈ Spec0 A and so
P :H 6= 0. Finally, P :H is an H-stable prime ideal by Proposition 7.
3.4. Torus actions. For rational torus actions, the set SpecHCI in Theorem 1 has a simpler
description as Spec ZI for an explicit commutative domain ZI . This was shown in [13], but
the presentation contains some inaccuracies which we will now repair.
(cid:3)
To start with, let H be pointed cocommutative and consider an arbitrary locally finite action
H A with coefficient Hopf subalgebra O ⊆ H ◦. Associated to any I ∈ H-Spec A, we have
the commutative algebra CI = C(A/I)⊗O ∈ H Alg as in (3); the H-action on CI is (9) using
the action H C(A/I) from Lemma 10(a). This action need not be locally finite (§2.3.2).
Following [13], we consider the locally finite part and the invariants:
ZI := C(A/I)fin ⊇ FI := C(A/I)H .
The invariant algebra FI is in fact a field [15, Lemma 1.4]. By Lemma 10(b), the action
H ZI has coefficient Hopf algebra O and Proposition 6(c) gives the following isomorphism
which generalizes [13, Equation (2)]:
ZI ∼= Ψ(ZI ) = (CI )H.
(15)
ACTIONS OF COCOMMUTATIVE HOPF ALGEBRAS
13
Turning to torus actions, let k be algebraically closed and let G be an algebraic k-torus
acting rationally on the k-algebra A. As was remarked in §0.4, the algebra of polynomial
functions O = O(G) serves as a coefficient Hopf algebra for this action. Moreover, O is a
Laurent polynomial algebra over k, the algebra CI is a Laurent polynomial algebra over the
field C(A/I), and the isomorphism ZI ∼= (CI )G in (15) realizes ZI as a Laurent polynomial
algebra over the field of G-invariants FI = C(A/I)G; see the Stratification Theorem in [13]
or [13, Equation (4)]. The set SpecH CI in Theorem 1 now is the set SpecGCI consisting of
all G-stable prime ideals of CI and Spec (CI )G ∼= Spec ZI by (15).
Lemma 14 (notation as above). Contraction and extension of ideals give a G-equivariant
order isomorphism SpecGCI ∼ Spec (CI )G.
Proof. By [13, Equation (5)], all G-stable ideals of S := CI are generated by their intersection
with R := SG. Therefore, the contraction map SpecGS → Spec R, P 7→ P ∩ R, is injective.
If a is an ideal of R, then its extension ea := aS is a G-stable ideal of S. Moreover, ea∩ R = a,
because S is free over R by [13, 2.3]. Thus, ea is the only G-stable ideal of S that contracts
to a. Now let p ∈ Spec R be given and choose an ideal P of S maximal subject to the
condition P ∩ R = p. Then P is prime and hence so is its G-core by Proposition 7. Thus,
P:G ∈ SpecGS and P:G = ep by the foregoing. This proves surjectivity of the contraction
map and that the inverse is given by extension. The statements about G-equivariance and
preservation of inclusions in the lemma are clear.
(cid:3)
4. SEMIPRIMENESS
4.1. Reformulations. We start this section by giving several reformulations, in terms of the
semiprime radical operator √ · , of the conclusion of Theorem 2, which is equivalent to (ii)
in the lemma below. The semiprime radical of a subset X ⊆ A, by definition, is the unique
smallest semiprime ideal of A containing X:
√X = \
P .
P ∈Spec A
P ⊇X
We continue to assume that A ∈ H Alg; the Hopf algebra H can be arbitrary for now.
Lemma 15. The following are equivalent:
(i) If J is an H-ideal of A, then so is √J ;
(ii) for all ideals I of A, the H-core √I:H is semiprime;
(iii) H.√I ⊆ √H.I for any ideal I of A.
Proof. (i) ⇒ (ii). We may assume that I is semiprime. Then √I:H ⊆ √I = I, since √ ·
preserves inclusions. In fact, √I:H ⊆ I:H, because √I:H is an H-ideal by (i). The reverse
inclusion being trivial, it follows that I:H = √I:H is semiprime.
(ii) ⇒ (iii). Let J denote the ideal of A that is generated by the subset H.I ⊆ A. Then
√I ⊆ √J = √H.I and J is easily seen to be an H-ideal. (If the antipode S is bijective, then
J = H.I [14, Exercise 10.4.3].) Thus, J = J:H ⊆ √J:H and the latter ideal is semiprime
by (ii). It follows that √J = √J:H ⊆ √J:H. Again, the reverse inclusion is clear; so
√J = √J:H. Therefore, H.√I ⊆ H.√J = √J = √H.I.
(iii) ⇒ (i). Specialize (iii) to the case where I = J is an H-ideal.
(cid:3)
14
MARTIN LORENZ, BACH NGUYEN, AND RAMY YAMMINE
4.2. Extending the base field. For the proof of Theorem 2, we may work over an alge-
braically closed base field. This follows by taking K to be an algebraic closure of k in the
argument below.
Let K/k be any field extension and put H ′ = H ⊗ K and A′ = A ⊗ K. Then A′ ∈ H ′Alg
and H ′ is cocommutative if H is so. Assuming Theorem 2 to hold for A′, our goal is to show
that it also holds for A. So let I be a semiprime ideal of A. Viewing A as being contained
in A′ in the usual way, IA′ is an ideal of A′ satisfying IA′ ∩ A = I. By Zorn's Lemma, we
may choose an ideal I ′ of A′ that is maximal subject to the condition I ′ ∩ A = I. Then I ′ is
semiprime. For, if J is any ideal of A′ such that J % I ′, then J ∩ A % I by maximality of I ′,
and so (J ∩A)2 * I by semiprimeness of I. Since (J ∩A)2 ⊆ J 2∩A, it follows that J 2 * I ′,
proving that I ′ is semiprime. Therefore, by our assumption, the core I ′:H ′ is semiprime. Since
the extension A ֒→ A′ is centralizing, it follows that (I ′:H ′) ∩ A is a semiprime ideal of A.
Finally, (I ′:H ′) ∩ A = {a ∈ A H ′.a ⊆ I ′} = {a ∈ A H.a ⊆ I ′ ∩ A = I} = I:H by (1),
giving the desired conclusion that I:H is semiprime.
4.3. Enveloping algebras.
4.3.1. For any ring R, let RJXλKλ∈Λ denote the ring of formal power series in the commuting
variables Xλ (λ ∈ Λ) over R; see [3, Chap. III, §2, no 11].
Lemma 16. Let R be a ring, let Λ be any set, and let S be a subring of RJXλKλ∈Λ such that
S maps onto R under the homomorphism RJXλKλ∈Λ → R, Xλ 7→ 0. If R is prime (resp.,
semiprime, a domain) then so is S.
Proof. We write monomials in the variables Xλ as X n = Qλ X n(λ)
(n ∈ M ), where M =
Z(Λ)
+ denotes the additive monoid of all functions n : Λ → Z+ such that n(λ) = 0 for almost
all λ ∈ Λ. Fix a total order < on M having the following properties (e.g., [1, Example
2.5]): every nonempty subset of M has a smallest element; the zero function 0 is the smallest
element of M ; and n < m implies n + r < m + r for all n, m, r ∈ M .
For any 0 6= s = Pn∈M snX n ∈ RJXλKλ∈Λ, we may consider its lowest coefficient,
smin := sm with m = min{n ∈ M sn 6= 0}. If R is prime and 0 6= s, t ∈ S are given,
then 0 6= sminrtmin for some r ∈ R. By assumption, there exists an element u ∈ S having
the form u = r + Pn6=0 unX n. It follows that sut 6= 0, with (sut)min = sminrtmin. This
proves that S is prime. For the assertions where R is semiprime or a domain, take s = t or
r = 1, respectively.
(cid:3)
λ
4.3.2. Now let H = U g be the enveloping algebra of an arbitrary Lie k-algebra g and assume
that char k = 0. For the proof of the next proposition, we recall the structure of the convo-
lution algebra Homk(H, R) for an arbitrary k-algebra R. Let (eλ)λ∈Λ be a k-basis of g and
fix a total order of the index set Λ. Put M = Z(Λ)
+ as in the proof of Lemma 16 and, for each
n ∈ M , put en = Q<
∈ H, where the superscript < indicates that the factors
occur in the order of increasing λ. The elements en form a k-basis of H by the Poincar´e-
Birkhoff-Witt Theorem, and the comultiplication of H is given by ∆en = Pr+s=n er ⊗ es;
see [14, Example 9.5]. Writing X n = Qλ X n(λ)
as in the proof of Lemma 16, we obtain an
λ
λ
1
n(λ)! en(λ)
λ
ACTIONS OF COCOMMUTATIVE HOPF ALGEBRAS
15
isomorphism of k-algebras,
ϕ : Homk(H, R) ∼ RJXλKλ∈Λ ,
f 7→ X
n∈M
f (en)X n.
Under this isomorphism, the algebra map u∗ : Homk(H, R) → R , f 7→ f (1) , coming from
the unit map u = uH : k → H translates into the map RJXλKλ∈Λ ։ R, Xλ 7→ 0, as
considered in Lemma 16.
The proposition below is not new; it can be found in Dixmier's book [6, 3.3.2 and 3.8.8],
albeit with a very different proof. Recall that an ideal of a ring is said to be completely prime
if the quotient is a domain.
Proposition 17. Let H = U g be the enveloping algebra of a Lie k-algebra g, let A ∈ H Alg,
and let I be an ideal of A. Assume that char k = 0. If I is prime, semiprime or completely
prime, then I:H is likewise.
Proof. By (1), the core I:H is identical to the kernel of the map δI : A → Homk(H, A/I)
that is given by δI (a) = (h 7→ h.a + I). We need to show that the properties of being
prime, semiprime, or a domain all transfer from A/I to the subring δI A ⊆ Hom(H, A/I)
or, equivalently, to the subring S ⊆ (A/I)JXλKλ∈Λ that corresponds to δI A under the above
isomorphism ϕ. Consider the map u∗ : Homk(H, A/I) → A/I , f 7→ f (1), and note that
(u∗◦δI )(a) = a+I for a ∈ A. Therefore, S maps onto A/I under the map (A/I)JXλKλ∈Λ →
A/I, Xλ 7→ 0. Now all assertions follow from Lemma 16.
4.3.3. For an arbitrary cocommutative Hopf algebra H, we cannot expect a result as strong as
Proposition 17: group algebras provide easy counterexamples to the primeness and complete
primeness assertions. Indeed, let H = kG be the group k-algebra of the group G and let
A ∈ H Alg. Then I:H = Tg∈G g.I for any ideal I of A. If I is semiprime, then so are all
g.I, because each g ∈ G acts on A by algebra automorphisms, and hence Tg∈G g.I will be
(cid:3)
semiprime also. However, primeness and complete primeness, while inherited by each g.I,
are generally lost upon taking the intersection.
4.4. Proof of Theorem 2. Let H be cocommutative Hopf and assume that char k = 0 and
that k is algebraically closed, as we may by §4.2. Then H has the structure of a smash product,
H ∼= U #V , where U is the enveloping algebra of the Lie algebra of primitive elements of H
and V is the group algebra of the group of grouplike elements of H; see [19, §13.1] or [17,
§15.3]. Thus, both U and V are Hopf subalgebras of H and H = U V , the k-space spanned
by all products uv with u ∈ U and v ∈ V . Viewing A ∈ U Alg and A ∈ V Alg by restriction,
repeated application of (1) gives the following equality for any ideal I of A:
I:H = {a ∈ A U V.a ⊆ I} = {a ∈ A V.a ⊆ I:U} = (I:U ):V.
If I is semiprime, then so is I:U (Proposition 17). Our remarks on group algebras in the first
paragraph of this proof further give semiprimeness of (I:U ):V . Thus, I:H is semiprime and
Theorem 2 is proved.
(cid:3)
Acknowledgment. We thank the referee for pointing out reference [6] in connection with
Proposition 17 above.
16
MARTIN LORENZ, BACH NGUYEN, AND RAMY YAMMINE
REFERENCES
[1] Matthias Aschenbrenner and Christopher J. Hillar, Finite generation of symmetric ideals, Trans. Amer. Math.
Soc. 359 (2007), no. 11, 5171 -- 5192. MR 2327026 (2008g:13030)
[2] W. E. Baxter and W. S. Martindale, III, Jordan homomorphisms of semiprime rings, J. Algebra 56 (1979),
no. 2, 457 -- 471. MR 528587 (80f:16008)
[3] Nicolas Bourbaki, Alg`ebre. Chapitres 1 `a 3, Hermann, Paris, 1970. MR 43 #2
[4]
, Groupes et alg`ebres de Lie. Chapitre II: Alg`ebres de Lie libres. Chapitre III: Groupes de Lie,
´El´ements de math´ematique. Fasc. XXXVII., Hermann, Paris, 1972, Actualit´es Scientifiques et Industrielles,
No. 1349. MR 58 #28083a
[5] Miriam Cohen, Smash products, inner actions and quotient rings, Pacific J. Math. 125 (1986), no. 1, 45 -- 66.
MR 860749 (88a:16019)
[6] Jacques Dixmier, Enveloping algebras, Graduate Studies in Mathematics, vol. 11, American Mathematical
Society, Providence, RI, 1996, Revised reprint of the 1977 translation. MR 1393197 (97c:17010)
[7] Kenneth R. Goodearl and Edward S. Letzter, The Dixmier-Moeglin equivalence in quantum coordinate
rings and quantized Weyl algebras, Trans. Amer. Math. Soc. 352 (2000), no. 3, 1381 -- 1403. MR 1615971
(2000j:16040)
[8] Gerhard P. Hochschild, Basic theory of algebraic groups and Lie algebras, Graduate Texts in Mathematics,
vol. 75, Springer-Verlag, New York-Berlin, 1981. MR 620024 (82i:20002)
[9] Vladislav K. Kharchenko, Algebras of invariants of free algebras, Algebra i Logika 17 (1978), no. 4, 478 --
487, 491. MR 538309
[10] Stefan Kolb, Martin Lorenz, Bach Nguyen, and Ramy Yammine, On the adjoint representation of a Hopf
algebra, arXiv:1905.03020v2, 2019.
[11] Martin Lorenz, Group actions and rational ideals, Algebra Number Theory 2 (2008), no. 4, 467 -- 499.
[12]
[13]
[14]
MR 2411408 (2009i:16058)
, Algebraic group actions on noncommutative spectra, Transform. Groups 14 (2009), no. 3, 649 -- 675.
MR 2534802 (2010m:14060)
, On the stratification of noncommutative prime spectra, Proc. Amer. Math. Soc. 142 (2014), 3013 --
3017.
, A tour of representation theory, Graduate Studies in Mathematics, vol. 193, American Mathematical
Society, Providence, RI, 2018.
[15] Jerzy Matczuk, Centrally closed Hopf module algebras, Comm. Algebra 19 (1991), no. 7, 1909 -- 1918.
MR 1121113 (92i:16031)
[16] Susan Montgomery, Bi-invertible actions of Hopf algebras, Israel J. Math. 83 (1993), no. 1-2, 45 -- 71.
MR 1239716 (94g:16047)
[17] David E. Radford, Hopf algebras, Series on Knots and Everything, vol. 49, World Scientific Publishing Co.
Pte. Ltd., Hackensack, NJ, 2012. MR 2894855
[18] Serge Skryabin, Hopf algebra orbits on the prime spectrum of a module algebra, Algebr. Represent. Theory
13 (2010), no. 1, 1 -- 31. MR 2585121 (2011e:16058)
[19] Moss E. Sweedler, Hopf algebras, Mathematics Lecture Note Series, W. A. Benjamin, Inc., New York, 1969.
MR 0252485 (40 #5705)
DEPARTMENT OF MATHEMATICS, TEMPLE UNIVERSITY, PHILADELPHIA, PA 19122
|
1406.2396 | 3 | 1406 | 2016-10-14T17:35:02 | On complex H-type Lie algebras | [
"math.RA",
"math.DG"
] | H-type Lie algebras were introduced by Kaplan as a class of real Lie algebras generalizing the familiar Heisenberg Lie algebra $\mathfrak{h}^3$. The H-type property depends on a choice of inner product on the Lie algebra $\mathfrak{g}$. Among the H-type Lie algebras are the complex Heisenberg Lie algebras $\mathfrak{h}^{2n+1}_{\mathbb{C}}$, for which the standard Euclidean inner product not only satisfies the H-type condition, but is also compatible with the complex structure, in that it is Hermitian. We show that, up to isometric isomorphism, these are the only complex Lie algebras with an inner product satisfying both conditions. In other words, the family $\mathfrak{h}^{2n+1}_{\mathbb{C}}$ comprises all of the complex H-type Lie algebras. | math.RA | math |
On complex H-type Lie algebras
Nathaniel Eldredge ∗
October 17, 2016
Abstract
H-type Lie algebras were introduced by Kaplan as a class of real
Lie algebras generalizing the familiar Heisenberg Lie algebra h3. The
H-type property depends on a choice of inner product on the Lie al-
gebra g. Among the H-type Lie algebras are the complex Heisenberg
Lie algebras h2n+1
, for which the standard Euclidean inner product
not only satisfies the H-type condition, but is also compatible with the
complex structure, in that it is Hermitian. We show that, up to iso-
metric isomorphism, these are the only complex Lie algebras with an
inner product satisfying both conditions. In other words, the family
h2n+1
comprises all of the complex H-type Lie algebras.
C
C
MSC 2010: 17B30 22E30 22E25 32M05 32Q99
1
Introduction
Since their introduction by Kaplan [8], H-type Lie algebras, and their cor-
responding nilpotent Lie groups, have attracted interest as a natural gener-
alization of the classical real Heisenberg Lie algebra h3 of dimension 3 and
the corresponding real Heisenberg group H3. The Heisenberg group is a
motivating example in many areas of mathematics, and in many cases, facts
about the Heisenberg group carry over into the H-type setting. For instance,
H-type groups carry a natural structure as sub-Riemannian manifolds, and
the analysis of their sub-Laplacians has attracted considerable interest. As
a sampling, we mention [1, 3, 5, 6, 7, 9].
The H-type condition for a (real) Lie algebra g is dependent on a choice
of inner product h·, ·i (i.e. a positive definite, symmetric, bilinear form) on
∗School of Mathematical Sciences, University of Northern Colorado, 501 20th St., Cam-
pus Box 122, Greeley, CO, 80639, USA. Email: [email protected]. This work was
supported by a grant from the Simons Foundation (#355659, Nathaniel Eldredge).
1
g, so it is really a property of the pair (g, h·, ·i). For example, in h3, the
natural Euclidean inner product will do.
C
In this note, we focus on the complex Heisenberg (or Heisenberg–Weyl)
Lie algebras h2n+1
, which, when considered as real Lie algebras and equipped
with their natural Euclidean inner products, are likewise H-type. But these
Lie algebras also carry a complex structure, and the Euclidean inner product
is Hermitian with respect to this structure, which is a natural compatibility
condition. As such, analysis on the complex Heisenberg groups H2n+1
can
take advantage of all the tools of complex geometry, together with the many
results for H-type groups mentioned above. However, the purpose of this
note is to show that this harmonious relationship between these structures
is essentially unique to these specific Lie algebras (and their respective Lie
groups).
C
As an application, we refer to [4], in which we studied a property known
as strong hypercontractivity for the hypoelliptic heat kernel on a stratified
complex Lie group. An essential hypothesis for this result was that the heat
kernel should satisfy a logarithmic Sobolev inequality. For most Lie groups,
it remains an open problem to determine whether this inequality holds, but
it follows from the results of [3, 6] that the inequality holds in every H-type
Lie group. Thus, the strong hypercontractivity theorem proved in [4] holds
in particular for any complex Lie group which, when considered as a real Lie
group, is also H-type. The result of the present note implies that these Lie
groups are precisely the family H2n+1
. As this is a relatively limited class of
examples, we see this as further motivation to try to extend the logarithmic
Sobolev inequality beyond the H-type case.
C
2 Definitions and examples
We begin by recalling the definition of an H-type Lie algebra, as formulated
in [2, Definition 18.1.1]. (Kaplan's original definition [8] is equivalent, but
slightly less convenient for our purposes.) Let g be a real finite-dimensional
Lie algebra equipped with an inner product h·, ·i : g × g → R. Let z be the
center of g, and let v = z⊥. For z ∈ z and u ∈ v, define Jzu as the unique
element of v satisfying
hJzu, vi = hz, [u, v]i
for all v ∈ v.
(1)
It is clear that each Jz : v → v is a linear map, and moreover z 7→ Jz is
linear in z.
2
Definition 1. We say that (g, h·, ·i) is H-type if the following two condi-
tions hold:
1. [v, v] ⊂ z
2. For each z ∈ z with kzk = 1, Jz : v → v is an isometry with respect to
h·, ·i.
We observe that an H-type Lie algebra is necessarily nilpotent of step
2. A simply-connected Lie group is said to be H-type if its Lie algebra is
H-type in the above sense.
Now suppose that g is a complex Lie algebra, whose complex structure
we denote by i. If we wish to equip g with a real inner product, it is natural
to demand some compatibility with the complex structure. Specifically, we
would like the inner product to be Hermitian, i.e., for x, y ∈ g we have
hix, iyi = hx, yi. We may then define J in terms of this inner product by (1).
We observe for later use that, as a consequence of the Hermitian property
of the inner product, we have for α, β ∈ C and u, z ∈ g,
Jαz(βu) = α ¯βJzu.
(2)
That is, Jzu is complex linear in z and conjugate linear in u.
The question of interest in this note is when both of the above properties
hold, motivating the following definition.
Definition 2. A complex H-type Lie algebra is a pair (g, h·, ·i), where
g is a complex Lie algebra and h·, ·i is an inner product on g, such that the
following two conditions hold:
• The inner product h·, ·i is Hermitian with respect to the complex struc-
ture of g.
• Forgetting the complex structure on g, the pair (g, h·, ·i) is H-type in
the sense of Definition 1.
We can likewise define a complex H-type Lie group as a connected
and simply connected complex Lie group G equipped with a Hermitian left-
invariant Riemannian metric g which, when viewed as an inner product on
the Lie algebra of G, satisfies the above conditions.
Example 3. The complex Heisenberg Lie algebra of complex dimen-
sion 2n + 1 is the complex Lie algebra h2n+1
generated (over C) by the
basis of the 2n + 1 vectors {x1, y1, . . . , xn, yn, z}, with the bracket defined
C
3
C
by [xk, yk] = z, and for j 6= k, [xj , yk] = [xj, z] = [yj, z] = 0. We may equip
h2n+1
with the real inner product h·, ·i that makes all of xk, ixk, yk, iyk, z, iz
orthonormal; it is clear that this inner product is Hermitian. The center z
of h2n+1
is spanned (over C) by z, so we clearly have [v, v] = z. Defining J
as above, it is easy to compute
C
Jzxk = yk Jzyk = −xk Jzixk = −iyk Jziyk = ixk
so that Jz is an isometry. Moreover, every element w ∈ z is of the form
w = αz for some α ∈ C, and kwk = α, so using (2) we see that Jw is an
isometry whenever kwk = 1. Thus (h2n+1
, h·, ·i) is a complex H-type Lie
algebra.
C
Of course, the complex Heisenberg Lie algebras are a very special family
within the far larger class of all complex Lie algebras. Likewise, the class of
H-type Lie algebras, although fairly restrictive, is still much broader than
this specific family. For instance, there exist H-type Lie algebras having
centers of any given real dimension [8], while the complex Heisenberg Lie
algebras all have centers of real dimension 2.
Nevertheless, we shall now prove that the complex Heisenberg Lie alge-
bras are, up to isometric isomorphism, the only complex H-type Lie algebras.
3 Main result
Theorem. Let (g, h·, ·i) be a complex H-type Lie algebra as defined above.
Then for some n, (g, h·, ·i) is isometrically isomorphic to h2n+1
with its
standard Hermitian inner product.
C
In particular, complex H-type Lie algebras are completely classified by
their dimension. We also immediately obtain the analogous classification of
complex H-type Lie groups.
Proof. Suppose (g, h·, ·i) is complex H-type, and let v, z and J be defined as
above.
We recall the well-known Clifford algebra identity for H-type Lie alge-
bras:
JzJw + JwJz = −2 hz, wi I,
z, w ∈ z.
(3)
To prove this, first consider the case when w = z and kzk = 1. Then for
any u, v ∈ v, we have
(cid:10)J 2
z u, v(cid:11) = hz, [Jzu, v]i = − hz, [v, Jzu]i = − hJzv, Jzui = − hv, ui
4
since Jz is an isometry. So J 2
and polarization.
z = −I. The general case follows by scaling
We begin by showing that z must have complex dimension 1. If not, then
we can find z, w ∈ z with kzk = kwk = 1 and hz, wi = hiz, wi = 0. Then by
(3) and (2) we have
0 = −2 hz, wi I = JzJw + JwJz
0 = −2 hiz, wi I = JizJw + JwJiz = iJzJw + JwiJz = i(JzJw − JwJz).
Thus JwJz = JzJw = 0, contradicting the requirement that Jz, Jw be isome-
tries.
Therefore, z is the complex span of a single unit vector z. We recursively
construct an orthonormal basis for v over R, of the form {xk, ixk, yk, iyk :
k = 1, . . . , n}. Suppose {xk, ixk, yk, iyk : k = 1, . . . , m − 1} have been
constructed and do not span v. Let xm be any unit vector orthogonal to all
of xk, ixk, yk, iyk for k = 1, . . . , m. Then set ym = Jzxm. We have kymk = 1,
and a few straightforward computations verify that {xk, ixk, yk, iyk : k =
1, . . . , m} are now orthogonal. When the process terminates, we have the
desired orthonormal basis.
To compute brackets, for j 6= k we have
hz, [xk, yk]i = hJzxk, yki = hyk, yki = 1
hz, [xk, xj]i = hJzxk, xji = hyk, xji = 0
hz, [yk, yj]i = hJzyk, yji = hJzyk, Jzxji = hyk, xji = 0
hz, [xk, yj]i = hJzxk, yji = hyk, yji = 0.
Similar computations show that if z is replaced by iz, all of the above ex-
pressions vanish. Each bracket is in z and hence a complex scalar multiple
of z, so we have
[xk, yk] = z,
[xk, xj] = [yk, yj] = [xk, yj] = 0.
The corresponding brackets for ixk, iyk, etc, follow from the complex bi-
linearity of the bracket. These are precisely the same relations as for the
complex Heisenberg Lie algebra h2n+1
, and the basis is orthonormal, just
as for the standard inner product on h2n+1
. Therefore, the unique complex
linear map g → h2n+1
sending x1, y1, . . . , xn, yn, z ∈ g to the standard basis
for h2n+1
(described in Example 3) is an isometric isomorphism of complex
Lie algebras.
C
C
C
C
Acknowledgments. The author would like to thank Maria Gordina,
Leonard Gross, Martin Moskowitz, and Laurent Saloff-Coste for helpful sug-
gestions.
5
References
[1] Zolt´an M. Balogh and Jeremy T. Tyson.
Polar
in Carnot
nates
ISSN 0025-5874.
http://dx.doi.org/10.1007/s00209-002-0441-7.
groups.
doi:
Math. Z.,
10.1007/s00209-002-0441-7.
241(4):697–730,
coordi-
2002.
URL
[2] A. Bonfiglioli, E. Lanconelli, and F. Uguzzoni. Stratified Lie groups
and potential theory for their sub-Laplacians. Springer Monographs in
Mathematics. Springer, Berlin, 2007.
ISBN 978-3-540-71896-3; 3-540-
71896-6.
[3] Nathaniel Eldredge.
kernel on H-type groups.
2010.
http://dx.doi.org/10.1016/j.jfa.2009.08.012. arXiv:0904.1781.
Gradient estimates for the subelliptic heat
258(2):504–533,
URL
10.1016/j.jfa.2009.08.012.
J. Funct. Anal.,
ISSN 0022-1236.
doi:
[4] Nathaniel Eldredge, Leonard Gross, and Laurent Saloff-Coste. Strong
hypercontractivity and logarithmic Sobolev inequalities on strati-
fied complex Lie groups. Preprint. arXiv:1510.05151, 2015. URL
http://arxiv.org/abs/1510.05151.
[5] Jianxun He and Mingkai Yin.
half-space on the H-type group.
10, 2016.
http://dx.doi.org/10.1186/s13660-016-1070-8.
Lp Hardy type inequality in the
J. Inequal. Appl., pages 2016:129,
ISSN 1029-242X. doi: 10.1186/s13660-016-1070-8. URL
[6] Jun-Qi Hu and Hong-Quan Li. Gradient estimates for the heat
33(4):355–386,
URL
semigroup on H-type groups.
2010.
doi:
http://dx.doi.org/10.1007/s11118-010-9173-1.
10.1007/s11118-010-9173-1.
ISSN 0926-2601.
Potential Anal.,
[7] J. Inglis, V. Kontis, and B. Zegarli´nski. From U -bounds to isoperime-
try with applications to H-type groups. J. Funct. Anal., 260(1):76–
116, 2011.
doi: 10.1016/j.jfa.2010.08.003. URL
http://dx.doi.org/10.1016/j.jfa.2010.08.003.
ISSN 0022-1236.
[8] Aroldo Kaplan. Fundamental solutions for a class of hypoelliptic PDE
generated by composition of quadratic forms. Trans. Amer. Math. Soc.,
258(1):147–153, 1980. ISSN 0002-9947.
[9] Hans Martin Reimann. Rigidity of H-type groups. Math. Z., 237
ISSN 0025-5874. doi: 10.1007/PL00004887. URL
(4):697–725, 2001.
http://dx.doi.org/10.1007/PL00004887.
6
|
1312.2782 | 1 | 1312 | 2013-12-10T12:36:54 | On sets of eigenvalues of matrices with prescribed row sums and prescribed graph | [
"math.RA"
] | Motivated by a work of Boros, Brualdi, Crama and Hoffman, we consider the sets of (i) possible Perron roots of nonnegative matrices with prescribed row sums and associated graph, and (ii) possible eigenvalues of complex matrices with prescribed associated graph and row sums of the moduli of their entries. To characterize the set of Perron roots or possible eigenvalues of matrices in these classes we introduce, following an idea of Al'pin, Elsner and van den Driessche, the concept of row uniform matrix, which is a nonnegative matrix where all nonzero entries in every row are equal. Furthermore, we completely characterize the sets of possible Perron roots of the class of nonnegative matrices and the set of possible eigenvalues of the class of complex matrices under study. Extending known results to the reducible case, we derive new sharp bounds on the set of eigenvalues or Perron roots of matrices when the only information available is the graph of the matrix and the row sums of the moduli of its entries. In the last section of the paper a new constructive proof of the Camion-Hoffman theorem is given. | math.RA | math |
On sets of eigenvalues of matrices with prescribed row
sums and prescribed graph
Gernot Michael Engela,∗, Hans Schneiderb , Sergeı Sergeevc,1
aTransversal Networks Corp., 2753 Mashal l Parkway, Madison, WI 53713,USA
bDeaprtment of Mathematics, University of Wisconsin-Madison, Madison, WI 53706, USA
cUniversity of Birmingham, School of Mathematics, Edgbaston B15 2TT, UK
Abstract
Motivated by a work of Boros, Brualdi, Crama and Hoffman, we consider the
sets of (i) possible Perron roots of nonnegative matrices with prescribed row
sums and associated graph, and (ii) possible eigenvalues of complex matrices
with prescribed associated graph and row sums of the moduli of their entries.
To characterize the set of Perron roots or possible eigenvalues of matrices in
these classes we introduce, following an idea of Al’pin, Elsner and van den
Driessche, the concept of row uniform matrix, which is a nonnegative matrix
where all nonzero entries in every row are equal. Furthermore, we completely
characterize the sets of possible Perron roots of the class of nonnegative matrices
and the set of possible eigenvalues of the class of complex matrices under study.
Extending known results to the reducible case, we derive new sharp bounds
on the set of eigenvalues or Perron roots of matrices when the only information
available is the graph of the matrix and the row sums of the moduli of its entries.
In the last section of the paper a new constructive proof of the Camion-Hoffman
theorem is given.
Keywords: Gersgorin, eigenvalues, Perron root, row sums, row uniform
matrices, graphs, diagonal similarity, sum scaling, Camion-Hoffman,
AMS Classification: 15A18, 15A29, 15A80
1. Introduction
1.1. Background and motivation
The use of the row sums of a matrix to determine nonsingularity or to bound
its spectrum has its origins in the 19th century [18, Section 2] and has led to a
vast literature associated with the name of Gersgorin and his circles [21]. One of
∗Corresponding author. Email: [email protected]
Email addresses: [email protected] (Gernot Michael Engel),
[email protected] (Hans Schneider), [email protected] (Sergeı Sergeev)
1 Supported by EPSRC grant EP/J00829X/1
Preprint submitted to Elsevier
December 11, 2013
the first observations, due to Frobenius, was that the Perron root ρ(A) (i.e., the
biggest nonnegative eigenvalue, or the spectral radius) of a nonnegative matrix
A ∈ Rn×n
is bounded by
+
n
min
i=1
n
max
i=1
(1)
ri (A)
ri (A) ≤ ρ(A) ≤
where ri denotes the ith row sum of the elements of A. If A is irreducible then
the inequalities in (1) are strict except when minn
i=1 ri (A) = maxn
i=1 ri (A).
In a recent development, Al’pin [2], Elsner and van den Driessche [11] sharp-
ened the classical bounds of Frobenius by considering a matrix B which has the
same zero-nonzero pattern as A, and whose entries are equal to the row sums of
A in the corresponding rows. We formalize this idea in the following definition.
Definition 1.1. For A ∈ Rn×n
we define the auxiliary matrix B = Aux(A)
+
defined by
(bij = Pk aik ,
if aij 6= 0,
bij = 0,
if aij = 0.
For a general complex matrix A ∈ C n×n , its auxiliary matrix is defined as
Aux(A).
Next, recall the concepts of minimal and maximal cycle (geometric) means.
For an arbitrary matrix A ∈ Rn×n
these quantities are defined as follows
+
(2)
ν (A) =
µ(A) =
(ai1 i2 · ai2 i3 · . . . · aiℓ i1 )1/ℓ ,
min
(i1 ,...,iℓ )∈C (A)
(ai1 i2 · ai2 i3 · . . . · aiℓ i1 )1/ℓ ,
max
(i1 ,...,iℓ )∈C (A)
where C (A) denotes the set of cycles of the associated graph. Recall that the
directed weighted graph, associated with an arbitrary complex matrix A ∈ C nn ,
is defined by the set of nodes N = {1, . . . , n} and set of edges E such that
(i, j ) ∈ E if and only if aij 6= 0, in which case edge (i, j ) is assigned the weight
aij .
According to Al’pin [2], Elsner and van den Driessche [11], we have
(3)
(4)
B = Aux(A),
ν (B ) ≤ ρ(A) ≤ µ(B ),
for any nonnegative matrix A.
If A and hence B are irreducible then either
ν (B ) = ρ(A) = µ(B ) or (if ν (B ) < µ(B )) the inequalities in (4) are strict.
Exploiting similar ideas, Boros, Brualdi, Crama and Hoffman [4] investigated
a class of complex matrices A ∈ C n×n with prescribed off-diagonal row sums of
the moduli of their entries, prescribed associated graph, and prescribed moduli
of all diagonal entries. In the case when G (A) is strongly connected with at
least two cycles (scwaltcy), they investigated the existence of a positive vector
x satisfying
aii xi ≥ Xj 6=i
aij xj ,
i = 1, . . . , n
(5)
2
for all matrices from the class simultaneously, and described the cases when all
inequalities in (5) are strict [4, Theorem 1.1], at least one of the inequalities
is strict [4, Theorem 1.2], or all inequalities hold with an equality [4, Theorem
1.3]. These results imply generalizations of Gersgorin’s theorem due to Brualdi
[5]. Following the statement of [4, Theorem 1.4] the authors provide a detailed
outline for the proof that Brualdi’s conditions are sharp.
In this paper we mainly deal with the two classes of matrices described in the
abstract. These classes are similar to those in [4], but we drop the requirement
that G (B ) is scwaltcy. In particular we also handle the reducible (not strongly
connected) case. However we do not prescribe the moduli of diagonal entries,
and include these moduli in the row sums instead. This allows us, in particular,
to combine the problem statement of Boros, Brualdi, Crama and Hoffman [4]
with that of Al’pin [2], Elsner and van den Driessche [11] and to generalize
all above mentioned results removing the restriction that B is irreducible. The
main results of this paper characterize the Perron roots or the sets of eigenvalues
of the classes of matrices under consideration.
At the end of the paper we present a new constructive proof of the Camion-
Hoffman theorem [8] (see also [10]). This theorem characterizes regularity of
a class of complex matrices with prescribed moduli of their entries. The scal-
ing result of Section 2.3 is crucial for our new proof (which also makes use of
one of the previously mentioned characterization results). Since we are dealing
with complex rather than with nonnegative matrices here, the triangle inequlity
(implicit in Lemma 4.10) also plays a role.
Other proofs of the Camion-Hoffman theorem have been given by Levinger
and Varga [16], and Engel [12].
1.2. Contents of the paper
The rest of this paper is organized as follows. Section 1.3 is a reminder of
the Frobenius normal form of nonnegative matrices.
Section 2 is devoted to a form of diagonal similarity scaling called visu-
alization scaling [19] or Fiedler-Pt´ak scaling [14] (see also [1]). Interest in this
scaling has been motivated by its use in max algebra, see for example [6] and [7].
Lemmas 2.4 and 2.5 can be used to generalize the simultaneous scaling results
of Boros, Brualdi, Crama and Hoffman [4, Theorems 1.1-1.3] to include the re-
ducible case. This also yields a derivation of the bounds of Al’pin, Elsner and
van den Driessche (Theorem 2.6). Theorem 2.8 establishes the existence of an
advanced visualization scaling, which is applied in the proof of the Camion-
Hoffman theorem.
In Section 3 we consider the class of nonnegative matrices with prescribed
graph and prescribed row sums. Theorem 3.7 characterizes the set of possible
Perron roots of such matrices also when B is reducible. This is one of the main
results of this paper. The proof is based on analyzing the sunflower subgraphs
of G (B ), a technique well-known in max algebra [15]. As an immediate corollary
it follows from Theorem 3.7 that for irreducible B with ν (B ) < µ(B ) and any
r, ν (B ) < r < µ(B ) there exists A with Aux(A) = B such that ρ(A) = r.
3
In Section 4.1 we consider the class of complex matrices with prescribed
graph and prescribed row sums of the moduli of their entries. We seek a char-
acterization of the set of nonzero eigenvalues of such matrices, starting with
the irreducible case in Theorem 4.4 . In this case we show in particular that
when B has more than one cycle, the set of possible nonzero eigenvalues of A
satisfying Aux(A) = B consists either of all s satisfying 0 < s < µ(B ) when
ν (B ) < µ(B ), or 0 < s ≤ µ(B ) if ν (B ) = µ(B ). Then, based on the irreducible
case the full characterization in the reducible case is given in Theorem 4.9. In
addition to this, the occurance of a 0 eigenvalue is treated in Theorem 4.2.
In Section 4.2 a new proof of the Camion-Hoffman theorem [8] is given, based
on the advanced visualization scaling of Section 2.3 and the characterization
result of Theorem 4.9.
0
0
1.3. Frobenius normal form
Let A be a square nonnegative matrix. If A is irreducible (i.e., the associated
digraph is strongly connected) then according to the Perron-Frobenius theorem
A has a unique (up to a multiple) positive eigenvector corresponding to the
Perron root ρ(A) (which is also the greatest modulus of all eigenvalues of A). If
A is reducible then by means of simultaneous permutations of rows and columns
or, equivalently, an application of P −1AP similarity where P is a permutation
matrix, A can be brought to the following form:
0
A1
0
0
∗ A2
. . .
0
∗
∗
∗ Am
∗
∗
where the square blocks A1 , . . . , Am correspond to the maximal strongly con-
nected components of the associated graph. These diagonal blocks A1 , . . . , Am
will be further referred to as classes of A. Note that each class Ai is either a
nonzero irreducible matrix, in which case it is called nontrivial, or a zero di-
agonal entry (and then it is called trivial).
If some component G (Ai ) of the
associated graph G (A) does not have access to any other component, which
means that there is no edge connecting one of its nodes to a node in another
component, then this component or the corresponding class Ai are called final.
Otherwise, this component or the corresponding class are called transient.
The entries denoted by 0 are actually off-diagonal blocks of zeros of ap-
propriate dimension, and ∗ denote submatrices of approriate dimensions whose
zero-nonzero pattern is unimportant.
.
2. Visualization scaling
2.1. Visualization of auxiliary matrices
In this section we assume that A is a nonnegative matrix such that G (A)
contains at least one cycle. Let us introduce some terminology related to max
algebra and visualization.
4
Definition 2.1. For a nonnegative matrix A, the critical graph C (A) = (Nc (A), Ec (A))
is defined as the subgraph of G (A) consisting of al l nodes Nc (A) and edges Ec (A)
on the cycles whose geometric mean equals µ(A). These nodes and edges are
also cal led critical. A node is cal led strictly critical if al l edges emanating from
it are critical.
Similarly, by anticritical graph we mean the subgraph of G (A) consisting
of al l nodes and edges on the cycles whose geometric mean equals ν (A) (also
speaking of anticritical nodes and edges). A node is cal led strictly anticritical if
al l edges emanating from it are anticritical.
Definition 2.2. A positive vector x is cal led a visualizing, resp. strictly vi-
sualizing, vector of A if aij xj ≤ µ(A)xi for al l (i, j ) ∈ E (A), resp.
if also
aij xj = µ(A) if and only if (i, j ) is critical.
Existence of such vector was proved by Engel and Schneider [13, Theorem 7.2]
in the irreducible case, and was extended to reducible matrices in [19].
Definition 2.3. A positive vector x is cal led an antivisualizing, resp. a strictly
antivisualizing, vector of A if aij xj ≥ ν (A)xi for al l (i, j ) ∈ E (A), resp. if also
aij xj = ν (A) if and only if (i, j ) is anticritical.
An existence of such scaling follows from the existence of visualization scaling,
applied to a matrix resulting from A after elementwise inversion of the entries.
The following lemmas are based on the results on simultaneous scaling found
in [4, Theorems 1.1-1.3]. We make arguments of [4] more precise by basing them
on the existence of strictly visualizing vectors [19].
Lemma 2.4 (cf. [4]). Let A be a nonnegative matrix and let B = Aux(A)
with µ(B ) 6= 0. Let x be a strictly visualizing vector of B . Then we have
Ax ≤ µ(B )x and, more precisely, (Ax)i = µ(B )xi if i is a strictly critical node
of B and (Ax)i < µ(B )xi otherwise.
Proof: Assume that µ(B ) = 1. Then
max
j
max
j
bij xj
xi ≤ 1 for all i,
bij xj
xi
= 1 for all critical i.
(6)
If i is strictly critical, we have
= 1,
bij xj
∀j : (i, j ) ∈ E (A)
xi
which implies that xj = xk for all j and k such that both (i, j ) ∈ E (A) and
(i, k) ∈ E (A). Hence we can take any k with (i, k) ∈ E (A), and obtain
(Pj aij )xk
Pj aij xj
bik xk
xi
xi
xi
5
(7)
(8)
=
=
= 1
If i is not strictly critical then let us denote
xk = max
j {xj : (i, j ) ∈ E (A)}.
If i is not critical then
Pj aij xj
(Pj aij )xk
xi
xi
If i is critical (but not strictly) then
≤
=
bik xk
xi
< 1
∃l, h
bil xl
xi
= 1,
bihxh
xi
< 1,
(9)
(10)
(11)
which implies xl = xk > xh for these l and h. In particular, note that we have
(i, k) ∈ Ec (B ). Hence
Pj aij xj
xi
(Pj aij )xk
xi
= 1.
(12)
<
=
bik xk
xi
(cid:3)
Lemma 2.5 (cf. [4]). Let A be a nonnegative matrix and let B = Aux(A)
with ν (B ) 6= 0. Let x be a strictly antivisualizing vector of B . Then we have
Ax ≥ ν (B )x and, more precisely, (Ax)i = ν (B )xi if i is a strictly anticritical
node of B and (Ax)i > ν (B )xi otherwise.
2.2. Bounds of Alpin, Elsner, van den Driessche
We call a nonnegative matrix A truly substochastic, if Pj aij ≤ 1 for all i
and Pj aij < 1 for some i. In a similar way, A is called truly superstochastic if
Pj aij ≥ 1 for all i and Pj aij > 1 for some i.
The following known result can be now obtained from Lemmas 2.4 and 2.5.
Theorem 2.6 ([2], [11][Theorem A]). Let A be an irreducible nonnegative
matrix and let B = Aux(A).
(i) If µ(B ) = ν (B ), then A is diagonal ly similar to a stochastic matrix mul-
tiplied by µ(B ). In this case, ρ(A) = µ(B ) = ν (B ).
(ii) If ν (B ) < µ(B ), then A is diagonal ly similar to a truly substochastic
matrix multiplied by µ(B ). In this case, ν (B ) < ρ(A) < µ(B ).
Proof:
(i): As B is irreducible and µ(B ) = ν (B ), all nodes of G (B ) are strictly
critical. Taking any visualization2 x of B we have Ax = µ(B )x, which implies
2 not necessarily strict
6
ρ(A) = µ(B ) = ν (B ). We also have that X −1AX , with X = diag(x), is a
stochastic matrix multiplied by ρ(A) = µ(B ) = ν (B )
(ii): As µ(B ) > ν (B ), not all nodes of G (B ) are strictly critical. Taking any
strictly visualizing vector x of B we have Ax ≤ µ(B )x where (Ax)i < µ(B )xi for
some i. We also have that X −1AX with X = diag(x), is a truly substochastic
matrix multiplied by µ(B ), as claimed. As X −1AX is also irreducible, it follows
that ρ(A) = ρ(X −1AX ) < µ(B ). The inequality ρ(A) < µ(B ) can be also
obtained (following an argument found, for instance in [11]) by multiplying the
system Ax ≤ µ(B )x, where at least one of the inequalities is strict, from the left
by a row vector z such that zA = ρ(A)z (which does not have 0 components if
A is irreducible).
Not all nodes of G (B ) are strictly anticritical, either. Taking any strictly an-
tivisualizing vector y of B = Aux(A) we have Ay ≥ ν (B )y where (Ay )i > ν (B )yi
for some i. We also have that Y −1AY with Y = diag(y ), is a truly superstochas-
tic matrix multiplied by µ(B ), as claimed. As Y −1AY is also irreducible, it
follows that ρ(A) = ρ(Y −1AY ) > ν (B ). The inequality ρ(A) > ν (B ) can be
also obtained by multiplying the system Ay ≥ µ(B )y , where at least one of
the inequalities is strict, from the left by a row vector z such that zA = ρ(A)z
(which does not have 0 components if A is irreducible). (cid:3)
2.3. Sum visualization
Definition 2.7. For A ∈ Rn×n
and a > 0, a vector x ∈ Rn
+ is cal led an a-sum
+
visualizing vector of A, if the entries of C = X −1AX with X = diag(x) satisfy
cij ≤ a for al l i, j and Pj cij ≥ a for al l i. In this case C is cal led an a-sum
visualization of A.
Recall that we have µ(A) ≤ ρ(A) for any nonnegative matrix. Indeed, since
for any positive x and any cycle (i1 , . . . iℓ ) we have that
≤
xi1 (cid:19)1/ℓ
(cid:18)ai1 i2
xiℓ
Yk∈{i1 ,...,iℓ } Xj
it follows by taking x satisfying Ax = ρ(A)x, that µ(A) ≤ ρ(A).
Theorem 2.8. Let A ∈ Rn×n
be irreducible, and define α(A) as the set of
+
positive numbers a for which an a-sum visualization of A exists. Then α(A) =
[µ(A), ρ(A)].
xi3
xi2 · . . . · aiℓ i1
xk
xj
xi2
xi1 · ai2 i3
1/ℓ
,
akj
Proof: 1. α(A) ⊆ [µ(A), ρ(A)]:
Let a ∈ α(A) and let C = X −1AX (for some diagonal X ) be such that cij ≤ a
for all i, j and Pj cij ≥ a for all i. Then µ(C ) ≤ a and ρ(C ) ≥ a, and as
µ(A) = µ(C ) and ρ(A) = ρ(C ) we obtain that a ∈ [µ(A), ρ(A)].
2. [µ(A), ρ(A)] ⊆ α(A):
7
Let µ(A) ≤ a ≤ ρ(A). We can assume without loss of generality (dividing A by
a if necessary) that a = 1 and µ(A) ≤ 1 ≤ ρ(A).
As µ(A) ≤ 1, there exists a non singular diagonal matrix X such that all
entries gij of G := X −1AX satisfy 0 ≤ gij ≤ 1. Since G is diagonally similar to
A, ρ(A) is also the spectral radius of G and hence the exists a vector z whose
zj
entries zi satisfy 1 = maxi zi and Pj gij
zi ≥ 1 for all i .
We will now construct an entrywise nonincreasing sequence of vectors {y (s)}s≥0
bounded from below by z . Such a sequence obviously converges, and as we will
yj
yj
yi ≤ 1 for all i, j , and Pj gij
argue, the limit denoted by y satisfies gij
yi ≥ 1
for all i (and, obviously, y ≥ z ).
Let us define a continuous mapping f : (R+\{0})n → (R+\{0})n , by its
components
fi (x) = min(xi , Xj
(13)
i = 1, . . . , n.
gij xj ),
Now let y (0) = (1, 1 . . . 1) and consider a sequence {y (s)}s≥0 defined by
y (s+1) := f (y (s) ) (that is, the orbit of y (0) under f ).
Observe that y (s+1) ≤ y (s) , as f (x) ≤ x for all x ∈ (R+ \{0})n.
It follows by induction that y (s) ≥ z for all s. The case s = 0 is the basis
of induction (since zi ≤ 1 for all i). We have to show that y (s+1) ≥ z knowing
that y (s) ≥ z . It amounts to verify that y (s+1)
≥ zk for the indices k where
k
y (s+1)
< y (s)
k . For such indices we have
k
gkj y (s)
j ≥ Xj
= Xj
As the sequence {y (s)}s≥0 is nonincreasing and bounded from below, it has
a limit which we denote by y . As f is continuous, this limit satisfies f (y ) = y ,
yj
which by the definition of f implies that Pj gij
yi ≥ 1 for all i.
y (s)
We now show by induction that gij
j
i ≤ 1, for all i 6= j and s. Denote by
y (s)
= Pj gij y (s)
. Thus y (s+1)
j < y (s)
Is the set of indices i where Pj gij y (s)
and
j
i
i
= y (s)
for i ∈ Is , while y (s+1)
< y (s)
y (s+1)
for i /∈ Is .
i
i
i
i
Observe that s = 0 is the basis of induction, so we assume that the claim
holds for s and we have to prove it for s + 1. For i, j /∈ Is the inequality
y (s+1)
j
gij
≤ 1 holds trivially. If i ∈ Is then
y (s+1)
i
gkj z (s)
j ≥ zk .
y (s+1)
k
gij
y (s+1)
j
y (s+1)
i
≤ gij
y (s)
j
y (s+1)
i
= gij y (s)
j (Xk
(where the last inequality follows since gij y (s)
is just one of the nonnegative
j
terms of the sum in the denominator).
Finally if i /∈ Is and j ∈ Is : then we have gij
gik y (s)
k )−1 ≤ 1
y (s)
j
i ≤ 1.
y (s)
y (s)
j
y (s+1)
i
< gij
= gik
y (s+1)
j
y (s+1)
i
8
Thus the inequalities gij
y (s)
j
i ≤ 1 hold for all i 6= j and s, and this implies
y (s)
yj
yi ≤ 1 hold as well. The case
that for the limit point y , all the inequalities gij
yi
i = j is trivial since the inequality gii
= gii ≤ 1 holds for all i.
yi
Let D be the diagonal matrix with dii = yixi for all i. For the entries cij of
C = D−1AD we have cij ≤ 1 for all i, j and Pj cij ≥ 1 for all i so the theorem
is proved. (cid:3)
Denote by A[−1] = (a[−1]
) the Hadamard inverse of A ∈ Rn×n
+ :
ij
ij = ( 1
if aij > 0,
,
a[−1]
aij
0,
if aij = 0.
Observe that µ(A[−1] ) = (ν (A))−1 (however, there is no such inversion for the
Perron root), and let us formulate the following corollary of Theorem 2.8.
be irreducible. The fol lowing are equivalent:
Corollary 2.9. Let A ∈ Rn×n
+
a ∈ [ 1
(i) 1
ν (A) , ρ(A[−1] )];
(ii) ∃x > 0 such that for C = X −1AX with X = diag (x) we have that cij ≥ a
a
for al l i, j . and Pj
cij ≥ 1 for al l i.
Proof: The corollary follows by elementwise inversion of the nonzero entries
and applying Theorem 2.8. (cid:3)
3. Nonnegative reducible matrices
Here we characterize Perron roots of nonnegative matrices with prescribed
row sums and prescribed graph. Section 3.1 is devoted to sunflower graphs,
which will be used in the proof of the main result. Section 3.2 contains the
main result and example.
3.1. Sunflowers
We introduce the following definition, inspired by description of the Howard
algorithm in [9] and [15, Chapter 6].
Definition 3.1. Let G be a weighted graph. A subgraph G of G is cal led a
sunflower subgraph of G if the fol lowing conditions hold:
(i) If a node in G has an outgoing edge then it has a unique outgoing edge in
G ;
(ii) Every edge in G has the same weight as the corresponding edge in G .
9
It is easy to see ([15]) that such a digraph can be decomposed into several
isolated components, each of them either acyclic or consisting of a unique cycle
and some walks leading to it. A sunflower subgraph G of G is called a simple
γ -sunflower subgraph of G , if γ is the unique cycle of G . The set of all sunflower
subgraphs of the weighted digraph G (B ), with full node set 1, . . . , n, will be
denoted by S (B ).
Denoting by µ(G ) the maximal cycle mean of a subgraph G ⊆ G (B ), we
introduce the following parameters:
M (B ) := max
G∈S (B )
µ(G ), m(B ) := min
G∈S (B )
µ(G ).
(14)
Lemma 3.2. Let G be a strongly connected graph. Then, for any cycle γ of G
there exists a simple γ -sunflower subgraph of G .
Proof: Let {1, . . . , n} be the nodes of G . Suppose that {1, . . . , k} are the nodes
in γ , and k + 1, ..., n are the rest of the nodes.
Observe first that we can construct a simple γ -sunflower on nodes 1, . . . , k :
this is just the cycle γ itself.
The proof is by contradiction. Assume that a simple γ -sunflower G can be
constructed for a subgraph induced by the set of nodes M , which contains the
nodes 1, . . . , k and is a proper subset of {1, . . . , n}, and that M is a maximal
such set. However, since G is connected, there is a walk W from {1, . . . , n}\M
to M , and we can pick the last edge of that walk and its last node before
it enters M . Adding that node and that edge to G we increase it while it
remains a simple γ -sunflower (of a subgraph induced by a larger node set). The
contradiction shows that we can construct a simple γ -sunflower of G . (cid:3)
Let us also recall the following.
Lemma 3.3. Let A be a nonnegative square matrix such that the digraph asso-
ciated with A is a sunflower graph. Then ρ(A) = µ(A).
Proof: Clearly, the cycles of G (A) are exactly the nontrivial classes of the
Frobenius Normal Form. Hence it suffices to observe that ρ(A) = µ(A) if G (A)
is a Hamiltonian cycle γ . Indeed, we can set xi = 1 for any i ∈ γ and then
calculating all the rest of coordinates from the equalities aij xj = µ(A)xi for
aij 6= 0. This computation does not lead to contradiction, since µ(A) is the
cycle mean of γ . (cid:3)
The following proposition expresses m(B ) and M (B ) in terms associated
with the Frobenius normal form.
Proposition 3.4. Let B be a nonnegative matrix. Then
M (B ) = µ(B ), m(B ) = max
Ni is final
ν (Bi ).
(15)
10
Proof:
ν (Bi ), since any sunflower
M (B ): It is obvious from (14) that M (B ) ≤ µ(B ). The reverse inequality
M (B ) ≥ µ(B ) follows since we can take a cycle α of G (B ) whose cycle mean
equals to µ(B ) and construct a sunflower subgraph of G (B ) that contains α as
one of its cycles.
m(B ): It is obvious from (14) that m(B ) ≥ max
Ni is final
subgraph of B contains a cycle in every nontrivial final class. So we show that
m(B ) ≤ max
ν (Bi ). For this, in each submatrix Bi corresponding to a final
Ni is final
class we take a cycle αi whose mean value is ν (Bi ) and using Lemma 3.2 build
a simple αi sunflower of the strongly connected component associated with
Bi . Unite all these sunflowers. If Bi is not final then it has access to another
class from some node ki .
In this case build a spanning tree on the nodes of
Bi , directed to ki , and for ki choose an edge going to another class. Finally,
for each trivial node of Bi we choose an arbitrary outgoing edge if it exists.
Adjoin these spanning trees and outgoing edges to the above union of simple
sunflowers. This leads to a sunflower subgraph G of G (B ), for which we have
µ(G ) = max
ν (Bi ), hence m(B ) ≤ max
ν (Bi ) and the required equality
Ni is final
Ni is final
follows. (cid:3)
Remark 3.5. Observe that m(B ) = 0 if and only if al l final classes of B are
trivial.
A sunflower subgraph which has cycles only in the final classes of G (B )
will be called thin. In the proof of Proposition 3.4 we actually established the
following result.
Lemma 3.6. Let G be a graph where each node has an outgoing edge and let
Gi for i = 1, . . . , q be the nontrivial final components of G .
For each col lection of cycles αi ∈ Gi for i = 1, . . . , q , there is a (thin) sunflower
subgraph of G whose cycles are α1 , . . . , αq .
If al l final components of G are trivial then there exists an acyclic sunflower
subgraph of G (i.e., a directed forest).
3.2. Range of the Perron root
For a row uniform nonnegative matrix B , denote
η(B ) := {ρ(A) : A ∈ Rn×n
+ , Aux(A) = B}.
We are going to extend Theorem 2.6 to include the reducible case and de-
scribe η(B ) for a general row uniform nonnegative matrix B .
(16)
Theorem 3.7. Let B be a nonnegative row uniform matrix.
(i) η(B ) ⊆ [m(B ), M (B )].
11
(ii) M (B ) ∈ η(B ) if and only if there is at least one final class Bi with µ(Bi ) =
ν (Bi ) = M (B ).
(iii) If m(B ) > 0 then m(B ) ∈ η(B ) if and only if µ(Bi ) = ν (Bi ) = m(B )
for al l final (nontrivial) classes Ni attaining the maximum in (15).
If
m(B ) = 0 then m(B ) ∈ η(B ) of and only if G (B ) is acyclic, in which case
η(B ) = {0}.
(iv) If M (B ) = m(B ) then η(B ) = {m(B )}.
(v) If M (B ) > m(B ) then (m(B ), M (B )) ⊆ η(B ).
Proof: Throughout the proof, let A be such that Aux(A) = B . Let Ai and
Bi for i = 1, . . . , m be the classes of the Frobenius normal form of A and B
respectively, and let Ni be the corresponding node sets (or classes).
(i): We have to show that ρ(A) ∈ [m(B ), M (B )]. Note that for any class Ai of
A we have Aux(Ai ) ≤ Bi , and Theorem 2.6 implies that ρ(Ai ) ≤ µ(Bi ), but we
do not have ρ(Ai ) ≥ ν (Bi ) in general. However, Aux(Ai ) = Bi holds for a final
class, and hence ν (Bi ) ≤ ρ(Ai ) ≤ µ(Bi ) for any final class.
With above considerations, the inequality ρ(A) ≤ M (B ) follows since M (B ) =
µ(B ) and ρ(Ai ) ≤ µ(Bi ) for all classes. To show that ρ(A) ≥ m(B ) we first de-
fine matrix A formed from A by zeroing out all the entries except for the entries
in final classes, and we similarly define B = Aux( A). Then we have ρ(A) ≥ ρ( A)
by monotonicity of the spectral radius. Since m(B ) = max
ν (Bi ) by (15)
Ni is final
and ρ(Ai ) = ρ( Ai ) ≥ ν ( Bi ) = ν (Bi ) for each final class we obtain that ρ(A) ≥
ρ( A) ≥ m(B ), hence the claim.
(ii): When G (B ) is acyclic the proof of (ii) is trivial. If G (B ) is not acyclic then
M (B ) = µ(B ) > 0 and if ρ(Ai ) = µ(B ) then Ai must be nontrivial. We first
argue that ρ(Ai ) = M (B ) is impossible if Ai has access to other classes. Indeed,
if there is such access then we only have Aux(Ai ) ≤ Bi with strict inequalities
i such that Aux(A′
in some rows. This implies that we can find A′
i ) = Bi and
A′
i ≥ Ai , with strict inequalities in the same rows. But then we have ρ(Ai ) <
ρ(A′
i ) ≤ µ(Bi ) so ρ(Ai ) = µ(B ) is impossible. Thus ρ(Ai ) = µ(B ) can be
attained only in a final class, which happens if and only if ρ(Ai ) = µ(Bi ), and
by Theorem 2.6, if and only if µ(Bi ) = ν (Bi ) = µ(B ) for one such class.
(iii): In the case when m(B ) = 0 but B has at least one nontrivial class Bi , we
have ρ(Ai ) > 0 and hence ρ(A) > 0 for any A such that Aux(A) = B . Therefore
in this case m(B ) = 0 ∈ η(B ) if and only if all classes of B are trivial (i.e., G (B )
is acyclic).
If m(B ) > 0, we first show that the given condition is necessary: if ν (Bi ) <
µ(Bi ) for at least one of these final classes then we have ν (Bi ) < ρ(Ai ) < µ(Bi )
by Theorem 2.6, hence m(B ) < ρ(Ai ) ≤ ρ(A), so m(B ) = ρ(A) does not
hold. Following the proof of Proposition 3.4, we can construct a thin sunflower
subgraph of G (B ) with the cycles attaining ν (Bi ) in all final classes. Denote
the matrix associated with this subgraph by C and the submatrices extracted
12
from the node sets Ni by Ci . For each submatrix Ci with Ni not final, we
have ρ(Ci ) = 0. By the continuity of Perron root we can find a small enough
ǫ such that ρ((1 − ǫ)Ci + ǫAi ) is smaller than m(B ) for all classes that are
not final and for all classes that are final but have ν (Bi ) < m(B ). This is
while Aux((1 − ǫ)C + ǫA) = B and, by Theorem 2.6, ρ((1 − ǫ)Ci + ǫAi ) =
m(B ) for all classes where the maximum in (15) is attained. This implies that
ρ((1 − ǫ)C + ǫA) = m(B ), hence the claim.
(iv): By part (i), ρ(A) can be only equal to m(B ) = M (B ). However, the set of
A such that Aux(A) = B is nonempty for any row uniform B , hence the claim.
(v): Let us first observe that by definition of m(B ) and M (B ) (14), there
exist matrices A and A whose associated graphs are the sunflower subgraphs of
G (B ) attaining the maximum and the minimum value of µ(G ) over all possible
sunflower subgraphs of G (B ). By Lemma 3.3 we have that ρ(A) = m(B ) and
ρ(A) = M (B ).
Now we argue that there exists A0 with Aux(A0 ) = B and ρ(A0 ) arbitrarily
close to m(B ) = ρ(A). Indeed, let D be any matrix with Aux(D) = B , and
consider the family of matrices Cǫ = (1 − ǫ)A + ǫD for ǫ > 0. Then (for any
ǫ > 0) we have Aux(Cǫ ) = B and since ρ(Cǫ ) is a continuous function of ǫ it
follows that limǫ→0 ρ(Cǫ ) = ρ(A). Similarly, there exists A1 with Aux(A1 ) = B
and ρ(A1 ) arbitrarily close to M (B ) = ρ(A).
Thus for each ǫ we have some A0 and A1 with Aux(A0 ) = Aux(A1 ) = B
and ρ(A0 ) < m(B ) + ǫ and ρ(A1 ) > M (B ) − ǫ. For λ, where 0 < λ < 1, let
Aλ := λA1 + (1 − λ)A0 interpolate between A0 and A1 . Since Aux(Aλ ) = B for
each λ and ρ(Aλ ) is continuous in λ, the claim follows. (cid:3)
As an immediate corollary we obtain the following result in the irreducible
case.
Corollary 3.8. Let B be an irreducible nonnegative row uniform matrix. Then
(i) If ν (B ) < µ(B ) then η(B ) = (ν (B ), µ(B )) .
(ii) If ν (B ) = µ(B ) then {ν (B )} = η(B ) = {µ(B )}.
Example. Given an irreducible row uniform matrix B ∈ Rn×n
and a con-
+
stant ρ ∈ (ν (B ); µ(B )) = (m(B ); M (B )), we describe a method for constructing
a matrix A such that Aux(A) = B and ρ(A) = ρ. Take two simple γ - sun-
flowers: one where γ has cycle mean equal to µ(B ), and the other where γ
has cycle mean equal to ν (B ). Denote by A1 the matrix associated with the
first sunflower, and by A2 the matrix associated with the second sunflower. We
have ρ(A1 ) = µ(B ) and ρ(A2 ) = ν (B ). For the convex combinations of these
matrices, we have that ρ(Aλ ), where Aλ := (1 − λ)A1 + λA2 and 0 ≤ λ ≤ 1, will
assume all values between ν (B ) and µ(B ). This follows from the continuity of
spectral radius as a function of λ(as in the more general construction above).
The value of λ for which ρ(Aλ ) = ρ, can be found from the system A(λ)x = ρx,
which has n + 1 variables (n components of x and the parameter λ). However,
13
since x can be multiplied by any scalar, one of the coordinates of x can be cho-
sen equal to 1. Then, for at least one of such choices, the existence of solution
is guaranteed.
For example, consider
(17)
(18)
B =
0 0
0 0
0 0
0 0
0 0
0 8
2 0
2 0
0 3
0 3
8
0
0
0
0
0 8
2 2
0 0
3 3
0 0
0 8
0 0
2 0
0 3
0 3
0 0
0 0
0 0
0 0
0 0
0 8
2 0
2 0
0 3
0 3
We see that the cycle (1, 2) is critical, with the cycle mean µ(B ) = 4, and
the cycle (2, 5) is anticritical with the cycle mean ν (B ) = √6. For the matrices
A1 and A2 assiciated with the corresponding sunflower graphs, we can take
A1 =
, A2 =
0
0
0
2
0
0
0
0
0
0
Equation Aλx = ρx, where we put x1 = 1, can be written as
1
0
0
x2
0
2 − y
x3
0
2
x4
0
0
x5
0
0
where y ∈ [0, 2] (so that y = 2λ). Observe that Aλ is irreducible, so the existence
of a solution with x1 = 1 (as well as a solution with any other component set
to 1) is guaranteed.
System (19) can be solved explicitly. Indeed, from the first equation of this
system we have 8x2 = ρ so x2 = ρ/8, from the third equation we have 2 = ρx3
so x3 = 2/ρ, from the fourth and the fifth equation we have 3x2 = 3ρ/8 =
ρx4 = ρx5 so x4 = 3/8 = x5 . Using the second equation of the system, we
obtain 2 − y + (3/8)y = ρx2 = (ρ2 )/8. Thus y = 16−ρ2
.
5
8 0
0 0
0 0
3 0
3 0
= ρ
,
(19)
0
y
0
0
0
1
x2
x3
x4
x5
4. Complex matrices
In Section 4.1 we characterize the set of eigenvalues of complex matrices
with prescribed graph and prescribed row sums of the moduli of their entries.
In Section 4.2, a new proof of the Camion-Hoffman theorem is presented.
4.1. Complex matrices with prescribed row sums of moduli
Definition 4.1. For B a row uniform nonnegative matrix, let σ(B ) denote the
set
σ(B ) = (cid:8)λ : ∃A ∈ C n×n , Aux(A) = B , det(A − λI ) = 0(cid:9).
14
Here A denotes the matrix whose entries are the moduli of (complex) entries
of A.
We first consider the conditions when 0 ∈ σ(B ). In what follows, the imag-
inary number “i” is denoted by ℑ. By a generalized diagonal product of B we
mean a product of the form Qn
i=1 biσ(i) where σ is an arbitrary permutation of
{1, . . . , n}.
Theorem 4.2. Let B be a row uniform nonnegative matrix. Then the fol lowing
are equivalent:
(i) 0 ∈ σ(B ).
(ii) The number of generalized nonzero diagonal products of B is not 1.
Proof: Suppose the number of generalized nonzero diagonal products of B is
one. Let A be such that Aux(A) = B . The determinant of A equals the signed
sum of the nonzero generalized diagonal products of A. Since all but one of
the generalized diagonal products of A are zero, we have det(A) 6= 0. Thus
0 /∈ σ(B ).
Suppose that B has no generalized nonzero diagonal products and let A be
such that Aux(A) = B . Since all the generalized diagonal products of A are
zero, we have det(A) = 0 and 0 ∈ σ(B ).
Suppose that B has two or more non zero generalized diagonal products.
Let us permute the columns of B in order to put one of the generalized diagonal
products on the (main) diagonal.
In other words, consider BP where P is
a permutation matrix and all diagonal entries of BP are nonzero. We have
Aux(A) = B if and only if Aux(AP ) = BP , and det(A) = det(AP ), therefore
0 ∈ σ(B ) if and only if 0 ∈ σ(BP ). As BP has at least one nonzero diagonal
product different from the main diagonal, the Frobenius normal form of BP has
a nontrivial diagonal block of dimension greater than 1.
Denote the index set of that block by M , and let us take any row uniform
nonnegative matrix D = (dkl ) such that Aux(D) = BP . For each k ∈ M , denote
by nk the number of outgoing edges of the kth node in M in the associated
digraph of BP that go to the nodes in M . As the block is irreducible and has
all diagonal entries nonzero, we have nk > 1. Let tk be a bijection between the
outgoing edges of k and {1, 2, . . . nk }, and define matrix C = (ckl ) by
ckl = (dkl exp(ℑ tk (l)2π
if k , l ∈ M and dkl 6= 0
),
nk
dkl ,
otherwise.
Then Aux(C ) = BP . In addition CMM v = 0 where CMM is the principal
submatrix of C extracted from rows and columns with indices in M , and v is the
vector with all components equal to 1. This implies that det(C ) = det(CMM ) =
0 so 0 ∈ σ(BP ) and 0 ∈ σ(B ).
(cid:3)
(20)
We now describe σ(B )\{0} starting from the irreducible case.
15
Definition 4.3. An irreducible matrix B is cal led unicyclic if G (B ) consists of
a single Hamiltonian cycle, and multicyclic otherwise.
Theorem 4.4. Let B be a row uniform nonnegative irreducible matrix.
(i) If B is unicyclic then σ(B ) = {s : s = µ(B )};
(ii) If B is multicyclic and ν (B ) < µ(B ) then
σ(B ) \ {0} = {s : 0 < s < µ(B )};
(iii) If B is multicyclic and ν (B ) = µ(B ) then
σ(B ) \ {0} = {s : 0 < s ≤ µ(B )}.
Proof: (i): In this case, all complex matrices A satisfying Aux(A) = B are
formed by multiplying the entries of B (that is, the entries of its only cycle) by
some complex numbers of modulus 1. The claim follows.
(ii), (iii): We first show that σ(B ) is contained in the above mentioned
intervals. For that we first recall a known result of Frobenius (see e.g. [3], p.31,
Theorem 2.14) that for any square complex matrix A, we have
(21)
ρ(A) := max{λ > 0 : det(A − λI ) = 0} ≤ ρ(A).
As Aux(A) = B , Theorem 3.7 implies that ρ(A) ≤ µ(B ) if µ(B ) ∈ η(B ) and
ρ(A) < µ(B ) if µ(B ) /∈ η(B ). Combining these inequalities with (21), we have
the desired inclusion.
We are left to show that each number in the intervals can be realized as an
eigenvalue of a complex matrix A with Aux(A) = B . Select λ ∈ (0, µ(B )) if
µ(B ) /∈ η(B ) or λ ∈ (0, µ(B )] if µ(B ) ∈ η(B ).
If λ ∈ η(B ) where η(B ) = {µ(B )} if ν (B ) = µ(B ) or η(B ) an interval whose
interior is (ν (B ), µ(B )), then there is an irreducible nonnegative matrix E such
that Aux(E ) = B with λ = ρ(A).
In the remaining case λ ≤ ν (B ) we will construct a row uniform matrix H
so that ν (H ) ≤ λ ≤ µ(H ). Since B has at least two cycles and it is irreducible,
there exists a row with index belonging to one of those cycles and with at least
two nonzero elements one of which must be on that cycle. Let t be the index
of such row. Consider a cycle α going through that row, with cycle mean c and
length ℓ.
If we have c ≤ λ, it follows that ν (B ) ≤ λ ≤ µ(B ) and we select
H = B . If c > λ then we multiply all entries of row t by z such that c · z 1/ℓ = λ.
Let H be the resulting matrix, so we have 0 < ν (H ) ≤ λ ≤ µ(H ).
If ν (H ) < µ(H ) ≤ ν (B ) < µ(B ) and λ = µ(H ) then µ(H ) is the new mean
value of the cycle α, which previously had c > λ. In this case, the corresponding
factor z < 1 can be slightly increased so that ν (H ) < λ < µ(H ) is satisfied. If
ν (H ) < µ(H ) and λ = ν (H ), then multiplying the row t by a value 1 − ǫ for
small enough ǫ we can also ensure that ν (H ) < λ < µ(H ).
Thus we can assume that ν (H ) = λ = µ(H ) or ν (H ) < λ < µ(H ), where H
is obtained from B by multiplying the row t with at least two nonzero entries
by a nonnegative scalar z ≤ 1. Then by Theorem 3.7, there is a nonnegative
16
matrix E with an eigenvector v such that E v = λv and Aux(E ) = H , where
row t has at least two nonzero entries that we denote by etk and etl . Since E
is irreducible, all components of v are positive. We now modify row t of E to
form a matrix C such that Aux(C ) = B and C v = λv . Let x be such that
tk + (x/vk )2 + qe2
ets + qe2
Xs6=k,l
tl + (x/vl )2 = btk .
It can be observed that this equation can be explicitly resolved with respect to
x.
crs =
if r = t, s = k ;
etk − ℑ(x/vk ),
if r = t, s = l;
etl + ℑ(x/vl ),
otherwise.
ers ,
Then Aux(C ) = B and C v = λv , so λ is an eigenvalue of C . The claim
follows.
(cid:3)
(22)
We call a class of complex matrices regular if all matrices in the class are
nonsingular.
Corollary 4.5. Let B be an irreducible row uniform nonnegative multicyclic
matrix with al l diagonal elements equal to 0. Let Γ(B ) consist of al l complex
matrices I − A with Aux(A) = B .
(i) If µ(B ) < 1 then Γ(B ) contains only regular matrices.
(ii) If µ(B ) = 1 then Γ(B ) contains only regular matrices
if and only if ν (B ) < 1.
(iii If µ(B ) > 1 then Γ(B ) contains a singular matrix.
Proof: Γ(B ) contains a singular matrix if and only if 1 ∈ σ(B ). By Theo-
rem 4.4 this happens if and only if either µ(B ) > 1 or µ(B ) = 1 = ν (B ). This
establishes all the claims. (cid:3)
Remark 4.6. As noted in the abstract and introduction of [4], the theorems in
that paper imply Brualdi’s [5] conditions for the non-singularlty of matrices and
show that they are sharp. There is no essential difference or simplification in
assuming that the main diagonal of the matrices considered there is the identity,
and in that case the spectral content of [4] Theorems 1.1 – 1.4 is recaptured by
Corollary 4.5 via standard Gersgorin theory, e.g.[20]. More precisely, Corollary
4.5(i) corresponds to Theorem 1.1 of [4], 4.5(ii) corresponds to Theorems 1.2
and 1.3, and 4.5(iii) corresponds to Theorem 1.4.
For A ∈ Rn×n
+ , index set K and row uniform matrix B we write Aux(A) (cid:12)K
B when the following conditions hold.
(a) For Aux(A) = B = (bij ) we have bij = 0 ⇔ bij = 0 for all i, j .
17
(b) For all i ∈ K we have bij < bij for all j where bij > 0.
(c) For all i /∈ K and all j we have bij = bij .
We will also need the following variation of Definition 4.1.
Definition 4.7. For B a row uniform matrix, let σK (B ) denote the set σK (B ) =
(cid:8)λ : ∃A ∈ C n×n , Aux(A) (cid:12)K B , det(A − λI ) = 0(cid:9).
The following corollary of Theorem 4.4 is immediate.
Corollary 4.8. Let B be a row uniform nonnegative irreducible matrix. Then
for any non-empty index set K ,
σK (B )\{0} = {s : 0 < s < µ(B )}.
Proof: Let us analyze the following three cases.
Case 1: µ(B ) > ν (B ). There exists an A such that µ(Aux(A) is arbitrarily
close to µ(B ) and ν (Aux(A) < µ(Aux(A), and for each A with Aux(A) (cid:12)K
B we have ν (Aux(A) < µ(B ).
Case 2: µ(B ) = ν (B ) and each cycle contains an index from K . In this case
µ(Aux(A), where Aux(A) (cid:12)K B , assumes all values in (0, µ(B )).
Case 3: µ(B ) = ν (B ) and there is a cycle avoiding the nodes with indices
In this case µ(Aux(A)) = µ(B ) for all A with Aux(A) (cid:12)K B , but
in K .
ν (Aux(A)) < µ(Aux(A)) for all such matrices.
In all three cases we obtain the claim by applying Theorem 4.4 to all Aux(A)
satisfying Aux(A) (cid:12)K B . (cid:3)
We are now ready to deal with the general reducible case.
Theorem 4.9. Let B be a row uniform nonnegative matrix, and let
M (B ) := max{µ(Bi ) where
Bi is a transient class or a final multicyclic class of B}.
(23)
Then
(i) If M (B ) is attained at some final multicyclic class Bs with ν (Bs ) = µ(Bs )
then
σ(B )\{0} = {s : 0 < s ≤ M (B )}∪
∪i {s : s = µ(Bi ), Bi is a final unicyclic class and µ(Bi ) > M (B )}.
(24)
(ii) Otherwise,
σ(B )\{0} = {s : 0 < s < M (B )}∪
∪i {s : s = µ(Bi ), Bi is a final unicyclic class and µ(Bi ) ≥ M (B )}.
(25)
18
Proof: It is known that λ is an eigenvalue of a matrix A ∈ C n×n if and only
if det(A − λI ) = 0, which implies that the spectrum of A ∈ C n×n (i.e., the
set of eigenvalues of A) is the union of spectra of its nontrivial classes in the
Frobenius normal form. Furthermore, if a principal submatrix As corresponds to
a transient class then it can be any matrix satisfying Aux(As ) (cid:12)Ks Bs , where
Ks is the (non-empty) set of indices of all nodes in this transient class that
have a connection to another class. Observe that the entries in different rows of
matrices with the same Aux(A) vary independently and hence the same is true
about the sets of rows belonging to different classes. Therefore σ(B )\{0} can
be found as union of σ(Bi )\{0} over all final classes Bi and σKs (Bs )\{0} over
all transient classes Bs , for some non-empty index sets Ks . Using Theorem 4.4
and Corollary 4.8 and taking the above mentioned union, it can be verified that
σ(B )\{0} is as claimed. (cid:3)
Example. To illustrate the last theorem, let us consider the following row
uniform matrices:
B =
, C =
0 0
0 4
4 0
0 0
0 0
0 0
0 0
0 0
3 3
3 3
0 0
4 0
0 0
0 3
0 3
5 0
4 0
0 4
3 0
0 3
0
5
0
0
0
0
3
0
3
0
That is, C is formed from B by cutting all connections between the classes.
The moduli of the eigenvalues in σ(B ) assume all the values in (0, 4) ∪ {5}.
Note that M (B ) = max{3, 4}, but 4 /∈ σ(B ) because the class extracted from
rows and columns 2 and 3 is transient (b12 > 0). Therefore condition (ii) of the
Theorem 4.9 is used in computing σ(B ).
The moduli of eigenvalues in σ(C ) assume all values in (0, 3] ∪ {4} ∪ {5}.
Here M (C ) = 3, which is the maximum cycle mean of the only final class which
is multicyclic. As the means of all cycles in that class are equal to each other,
the value of M (C ) belongs to σ(C ).
4.2. Camion-Hoffman theorem
We now will apply Theorem 2.8 and Theorem 4.9 to provide a new proof for
a theorem of Camion and Hoffman [8].
Let us first recall the following known facts and a definition:
Lemma 4.10 ([8]). Let a1 , . . . an be nonnegative numbers such that each num-
ber does not exceed the sum of other numbers. Then there exist complex numbers
c1 , . . . , cn such that ci = ai for i = 1, . . . , n and c1 + . . . + cn = 0.
Corollary 4.11 ([8]). Let the entries of A = (aij ) ∈ Rn×n
satisfy aii = 1,
+
Pj 6=i aij ≥ 1 for al l i and aij ≤ 1 for al l i, j . Then there exists a complex
matrix C with C = A and det(C ) = 0.
19
Proof: Since the condition of Lemma 4.10 are satisfied for ai1 , . . . , ain for all
i, there exists a complex matrix C with C = A such that Pj cij = 0 for all i.
This implies det(C ) = 0. (cid:3)
Definition 4.12. A matrix A = (aij ) ∈ Rn×n
is cal led strictly diagonal ly dom-
+
inant if aii > Pj 6=i aij for al l i.
We will investigate the following matrix class:
Definition 4.13. For A ∈ Rn×n
define Ω(A) = {E : eij = aij 1 ≤ i, j ≤ n}.
+
Theorem 4.14 (Camion-Hoffman). For A ∈ Rn×n
the fol lowing are equiv-
+
alent:
(i) Ω(A) does not contain a singular matrix;
(ii) There exists a permutation matrix P and a diagonal matrix D such that
P AD is strictly diagonal ly dominant;
(iii) There exists a permutation matrix P and nonsingular diagonal matri-
ces D1 , D2 such that al l diagonal entries of D1P AD2 are equal to 1 and
µ(Aux(D1P AD2 − I )) < 1.
Proof: (i)⇒ (ii): Assume that Ω(A) is regular. Let P be a permutation matrix
such that the diagonal product of P A is greater then or equal to any generalized
diagonal product of A. Since A is nonsingular the diagonal elements of E = P A
are nonzero. Let D be the diagonal matrix with entries equal to the inverse of
the corresponding diagonal elements of P A. Since all diagonal entries of P AD
are equal to 1, for any cycle α we can find a generalized diagonal product of
P AD equal to the product of the entries of α. Since any generalized diagonal
product of P AD is less than or equal to 1, it follows that µ(P AD) = 1.
We will now establish that ρ(P AD − I ) < 1. The proof is by contradiction.
Assume that ρ(P AD − I ) ≥ 1. Then P AD − I has a class B such that ρ(B ) > 1
and for all i we have bii = 0. Since µ(P AD − I ) ≤ 1 we also have µ(B ) ≤ 1.
Applying Theorem 2.8 to B , we obtain a diagonal nonnegative matrix Y such
that matrix E := Y −1 (B + I )Y has entries satisfying 0 ≤ eij ≤ 1 and eii = 1
for all i, j, and Pk 6=i eik ≥ 1 for all i. By Corollary 4.11 there is a matrix
H = (hij ) with complex entries satisfying H = E and det(H ) = 0. Replacing
the class B + I in F by Y H Y −1 we obtain a matrix G with det(G) = 0 and
G ∈ Ω(P AD). As P is a permutation matrix and D diagonal, there is a
bijective correspondence between Ω(P AD) and Ω(A) in which the singularity
and nonsingularity are preserved. This contradicts that Ω(A) does not contain
a singular matrix and hence ρ(P AD − I ) < 1.
Since ρ(P AD−I ) < 1, there exists a diagonal matrix Z such that Z −1 (P AD−
I )Z has all row sums strictly less than 1, see [3, Chapter 6] or [17] for a detailed
argument. (Such a diagonal matrix Z can be constructed using Perron eigen-
vectors of nontrivial classes.) As all row sums in the matrix Z −1 (P AD − I )Z =
20
Z −1P ADZ − I are strictly less than 1, it follows that the matrix P ADZ is
strictly diagonally dominant, with P a permutation matrix and DZ a diagonal
matrix, as required.
(ii)⇒ (iii) If P AD is strictly diagonally dominant then there is a digonal
matrix D1 such that the diagonal entries of D1P AD are equal to 1 and the row
sums of D1P AD − I are strictly less than 1. As each entry in Aux(D1P AD − I )
is strictly less than 1, we also have µ(Aux(D1P AD − I )) < 1 as claimed.
(iii)⇒ (i): The proof is by contradiction. Assume that (iii) holds but (i)
does not hold. That is, assume that there exists a permutation matrix P and
nonsingular diagonal matrices D1 , D2 such that µ(Aux(D1P AD2 − I )) < 1, and
that (in contradiction with (i)) there exists C ∈ Ω(A) with det(C ) = 0. Then
µ(Aux(D1P C D2 − I )) = µ(Aux(D1P AD2 − I )) < 1, and by Theorem 4.9 we
have 1 /∈ σ(Aux(D1P AD2 − I )). However, we have det(D1P CD2 ) = 0, and
we can multiply the rows of D1P CD2 by some complex numbers with moduli
1 to obtain a matrix with zero determinant and with all diagonal entries equal
to −1. Adding the identity matrix to this matrix we obtain a matrix in the
class Ω(D1P AD2 − I ), for which 1 is an eigenvalue. The set of eigenvalues
of matrices in Ω(D1P AD2 − I ) is a subset of σ(Aux(D1P AD2 − I )), so 1 ∈
σ(Aux(D1P AD2 − I )), a contradiction. (cid:3)
Let us also reformulate the Camion-Hoffman theorem in terms of M -matrices
and comparison matrices. Recall that a real matrix B is a nonsingular M -matrix
if B = ρI −C where C is a nonnegative matrix and the Perron root of C is strictly
less than ρ (see [3] for many other equivalent definitions). For a nonnegative
matrix A ∈ Rn×n
+ , its comparison matrix E = comp(A) has entries eii = aii for
i = 1, . . . , n and eij = −aij for i 6= j .
Theorem 4.15. For a nonnegative matrix A, the fol lowing are equivalent:
(i) Ω(A) does not contain a singular matrix,
(ii) For P a permutation matrix corresponding to the greatest generalized di-
agonal product of A, the matrix comp(P A) is a nonsingular M -matrix.
References
References
[1] S.N. Afriat. The system of inequalities ars > Xr − Xs . Proc. of Cambridge
Phylosophical Society 59:125-133, 1963.
[2] Yu.A. Al’pin, Bounds for the Perron root of a nonnegative matrix involving
properties of its graph Math. Notes 58:1121-1123, 1995.
[3] A. Berman, R. Plemmons, Nonnegative Matrices in the Mathematical Sci-
ences. Society for Industrial and Applied Mathematics, Philadelphia, 1994.
21
[4] E. Boros, R.A Brualdi, Y. Crama, A.J. Hoffman. Gersgorin variations
III:On a theme of Brualdi and Varga Linear Algebra Appl., 428:14–19,
2007.
[5] R.A. Brualdi. Matrices, eigenvalues, and directed graphs. Linear and Mul-
tilinear Algebra, 11:143-165, 1983.
[6] P. Butkovic. Max-linear Systems: Theory and Algorithms. Springer, Lon-
don, 2010.
[7] P. Butkovic and H. Schneider. Applications of max algebra to diagonal
scaling of matrices. Electronic J. Linear Algebra 13:262-273, 2005.
[8] P. Camion and A.J. Hoffman. On the nonsingularity of complex matrices.
Pacific J. Math., 17(2): 211-214, 1966.
[9] J. Cochet-Terrasson, G. Cohen, S. Gaubert, M. Gettrick and J.P. Quadrat.
Numerical computation of spectral elements in max-plus algebra. In Proc.
of the IFAC Conference on System Structure and Control, Nantes, July
1998.
[10] D. Coppersmith and A.J. Hoffman. On the singularity of matrices. Linear
Algebra Appl. 411:277-280, 2005.
[11] L. Elsner, P. van den Driessche, Bounds for the Perron root using max
eigenvalues, Linear Algebra Appl., 428:2000-2005, 2007.
[12] G.M. Engel. Regular equimodular sets of matrices for generalized matrix
functions. Linear Algebra Appl. 7:243-274 (1973).
[13] G.M. Engel and H. Schneider. Diagonal similarity and equivalence for ma-
trices over groups with 0. Czechoslovak Math. J., 25(3):389-403 (1975).
[14] M. Fiedler and V. Pt´ak. Diagonally dominant matrices. Czechoslovak Math.
J., 92:420-433, 1967.
[15] B. Heidergott, G.-J. Olsder, and J. van der Woude. Max-plus at Work.
Princeton Univ. Press, 2005.
[16] B.W. Levinger, R.S. Varga. On a problem of O. Taussky. Pacific J. Math.
19:473-487, 1966.
[17] H. Schneider. An inequality for latent roots of a matrix applied to determi-
nants with dominant main diagonal. J. London Math. Soc. 28:8-20, 1953.
[18] H. Schneider. Olga Taussky-Todd’s influence on matrix theory and matrix
theorists. Linear and Multilinear Algebra5:197-224, 1977.
[19] S. Sergeev, H. Schneider and P. Butkovic, On visualization scaling,
subeigenvectors and Kleene stars in max algebra. Linear Algebra Appl.,
431:2395-2406, 2009.
22
[20] O. Taussky, A recurring theorem on determinants, American Mathematical
Monthly 56:672-676,1949.
[21] R. S. Varga. Gersgorin and his circles Springer, 2004.
23
|
1301.6123 | 1 | 1301 | 2013-01-25T18:52:23 | Classifying Several Classes of Leibniz Algebras | [
"math.RA"
] | We extend results related to maximal subalgebras and ideals from Lie to Leibniz algebras. In particular, we classify minimal non-elementary Leibniz algebras and Leibniz algebras with a unique maximal ideal. In both cases, there are types of these algebras with no Lie algebra analogue. We also give a classification of E-Leibniz algebras which is very similiar to its Lie algebra counterpart. | math.RA | math |
CLASSIFYING SEVERAL CLASSES OF LEIBNIZ ALGEBRAS
CHELSIE BATTEN RAY, ALLISON HEDGES AND ERNEST STITZINGER
ABSTRACT
We extend results related to maximal subalgebras and ideals from Lie to Leibniz algebras.
In particular, we classify minimal non-elementary Leibniz algebras and Leibniz algebras
with a unique maximal ideal. In both cases, there are types of these algebras with no Lie
algebra analogue. We also give a classification of E-Leibniz algebras which is very similiar
to its Lie algebra counterpart. Note that a classification of elementary Leibniz algebras
has been shown in [3].
I. PRELIMINARIES
Loday introduced Leibniz algebras as a noncommutative generalization of Lie algebras.
Lie algebra results have been extended to these new algebras. Properties of the Frattini
subalgebra and ideal are studied in [3] for Leibniz algebras.
In particular, elementary
Leibniz algebras, those algebras with the property that the Frattini ideal is 0 for every
subalgebra, were classified over algebraically closed fields, extending Lie algebra results of
Towers [12], [13] and Towers and Varea [14]. Algebras closely related to elementary Lie
algebras are E-Lie algebras and minimal non-elementary Lie algebras, which have have
been investigated in [13]. We extend these results to Leibniz algebras.
In the minimal
non-elementary case, the classification contains many new algebras. Similar in definition
to the Frattini subalgebra is the Jacobson radical, which is the intersection of all maximal
ideals. This concept appears in [5] and [9] for Lie algebras. In particular, Lie algebras with
a unique maximal ideal are classified in [5]. We investigate these ideas in Leibniz algebras,
again getting new cases in the classification result.
Let A be an algebra. The intersection of all maximal subalgebras, F(A), is called the
Frattini subalgebra and the largest ideal of A contained in F(A), Φ(A), is called the Frattini
ideal. For Lie algebras these concepts have been widely studied, see [13]. If A is a solvable
Lie algebra or is over a field of characteristic 0, then F(A) is an ideal, but not generally.
For Leibniz algebras F(A) is an ideal at characteristic 0, but not necessarily when A is
solvable [3]. The intersection of all maximal ideals, J(A), is called the Jacobson radical.
The nilradical, Nil(A), is the maximal nilpotent ideal of A; it exists by [6]. The sum of all
minimal abelian ideals is denoted by Asoc(A). An algebra is elementary if Φ(B)=0 for all
1
2
CHELSIE BATTEN RAY, ALLISON HEDGES AND ERNEST STITZINGER
subalgebras B of A and is minimal non-elementary if Φ(B)=0 for all proper subalgebras of
A but Φ(A) 6= 0. A is called an E-algebra if Φ(B)⊆ Φ(A) for all subalgebras of B of A.
Following Barnes [1], we call an algebra L (left) Leibniz if left multiplication by each
x in L is a derivation. Many authors consider right Leibniz algebras instead. Thus an
algebra L is Leibniz if x(yz)=(xy)z+y(xz) is an identity on L. The product need not be
antisymmetric.
II. E-LEIBNIZ ALGEBRAS
L is an E-algebra if Φ(B)⊆ Φ(L) for all subalgebras B of L. The following result is an
extension of Proposition 2 of [11] to Leibniz algebra. The proof is the same as the original,
hence we omit it.
Theorem 1. Let L be a Leibniz algebra. Then L is an E-Leibniz algebra if and only if
L/Φ(L) is elementary.
Proposition 2. If L is solvable over a field of characteristic 0, then L is an E-Leibniz
algebra.
Proof. Since L is solvable, L2 is nilpotent. By Theorem 3.5 of [3] the result holds.
Lemma 3. If L=BLC and B and C are elementary, then L is elementary.
Proof. Let S be a subalgebra of BLC. We show that Φ(S)=0. For x∈B+S, x=b+c, where
b∈B and c∈C and c=x-b ∈C∩(B+S) The projection mapping from B+S onto (B+S)∩ C,
where x goes to c, has kernel B. Hence S/(B∩S)∼=(B+S)/B∼= (B+S)∩C. Since C is elemen-
tary, Φ((B+S)∩C)=0. Hence Φ(S)⊆B∩S. Similarly Φ(S)⊆C∩S. Hence Φ(S)⊆B∩C=0.
Theorem 4. Let L be a Leibniz algebra over K, an algebraically closed field of charac-
teristic 0. Then L is an E- Leibniz algebra if and only if:
(1) L is solvable, or
(2) L∼= sl2(K)L ... Lsl2(K), or
(3) L=R+S, where R is a solvable ideal, S ∼= sl2(K)L ... Lsl2(K), and RS+SR is con-
tained in Φ(L).
Proof. If L is solvable, then L is an E-Leibniz algebra by Proposition 2. If L is as in
(2), then L is a Lie algebra and L is elementary by Theorem 3.2 of [13]. Let L be as in
(3). S is elementary as in (2). Φ(L) is a nilpotent ideal in L, hence Φ(L) ⊆ R. Thus,
L/(Φ(L))∼=R/Φ(L)+S. Since RS+SR⊆ Φ(L), the previous sum is a direct sum. Thus,
by Theorem 4.8 of [12], 0=Φ(L/Φ(L))=Φ(S)L Φ(R/Φ(L)). This shows that R/Φ(L) is
CLASSIFYING SEVERAL CLASSES OF LEIBNIZ ALGEBRAS
3
elementary using Proposition 2. Then L/Φ(L) is elementary by Lemma 3 and L is an
E-algebra by Theorem 1.
Conversely, suppose that L is an E-algebra. Then L/Φ(L) is elementary from Theorem 1.
L/Φ(L) is the direct sum of its radical, R, and a semisimple ideal S ∼= sl2(K)L ... Lsl2(K)
by Theorem 4.3 of [3], either of which may be 0. If S=0, then L is solvable and (1) holds.
If R=0, the L is the direct sum of copies of sl2(K) as in (2). If neither R nor S is 0, then
RS+SR ⊆ Φ(L) as a consequence of Theorem 4.3 of [3].
The following result addresses the special case in which L is a perfect Leibniz algebra.
Corollary 5. Let L be a perfect Leibniz algebra (L2=L) over K, an algebraically closed
field of characteristic 0. Then L is an E-Leibniz algebra if and only if L ∼= sl2(K)L ... Lsl2(K).
Proof. Let L=R+S be the Levi decomposition for L as in [2], where R is the radical and S
is a semisimple subalgebra. Then L2=S2+RS+SR+R2. R=RS+SR+R2 since L is perfect.
Since L is an E-algebra, RS+SR ⊆ Φ(L). Hence R2+Φ(L)=R and R2+Φ(L)+S=L. Hence
R2+S=L since no proper subalgebra can supplement Φ(L). Thus R2=R, which implies
that R=0 since R is solvable. Hence L=S and L is Lie. Then, by Corollary 4.5 of [13], L∼=
sl2(K)L ... Lsl2(K).
III. MINIMAL NON-ELEMENTARY LEIBNIZ ALGEBRAS
An algebra, L, is called minimal non-elementary if L is not elementary but all proper
subalgebras of L are elementary. In [5], conditions for a Lie algebra to be minimal non-
elementary are found when L2 is nilpotent. The next result is the Leibniz algebra version.
Note that cases 3 and 4 have no Lie algebra counterpart and case 1 has a case not found
in Lie algebras. When L2 is nilpotent, Φ(L)=0 if and only if L2 ⊆ Asoc(L) and L2 is
complemented in L by Proposition 3.1 of [3]. We often use this fact in the following proof.
Theorem 6. Let L be a finite dimensional Leibniz algebra over an algebraically closed
field. Suppose that L2 is nilpotent. L is minimal non-elementary if and only if:
(1) L is three dimensional non-nilpotent with basis x,y,z and non-zero multiplication as:
(a) xz=cz, xy=cy+z, zx=0 and yx=0, where c is a non-zero scalar, or
(b) xz=cz, xy=cy+z, zx=-cz, yx=-cy-z where c is a non-zero scalar, or
(2) L is Heisenberg, or
(3) L is generated by a, where a2 6= 0, L is nilpotent and dim L ≥ 2, or
(4) L is the four dimensional non-nilpotent algebra generated by a,b,x and y with mul-
tiplication:
4
CHELSIE BATTEN RAY, ALLISON HEDGES AND ERNEST STITZINGER
ax=α x, ay=αy, bx=βx, by=βy, xa=-αx, ya=0, xb=y-βx and yb=0 where α, β are non
zero scalars.
Proof. Let L be minimal non-elementary. L is solvable by the conditions. Suppose
that L is not nilpotent. Then Φ(L)6=L2 and there exists a maximal subalgebra, M, of L
such that L=L2+M. Let B be an algebra of minimum dimension such that L=-L2+B. By
Lemma 7.1 of [12], L2∩B ⊆ Φ(B)=0. L2 is nilpotent and elementary, hence it is abelian.
Clearly B is also abelian.
Suppose that dim B >1. For any a∈B , let H(a)=L2+(a). Then Φ(H(a))=0 since all
proper subalgebras are elementary. Then L2 ⊆ Nil(H(a))=Asoc(H(a)). L2 is abelian since
it is elementary and nilpotent, and it is completely reducible under the action of a. On each
minimal ideal, either Ra=-La or Ra=0 [1]. Hence the minimal ideals are one dimensional
eigenspaces for La and Ra acting on L2 and La and Ra are simultaneously diagonalizable
on L2. This holds for all a in B, and since the left multiplications, La, commute, they are
simultaneously diagonalizable.
Consider B acting on W=L2 by left multiplication and decompose L2 as the direct sum of
weight modules, {x ∈ W: ax=αax for all a in B} Each of these weight modules is invariant
under Rb, b ∈B. If there is more than one weight module, then, for each weight α, let
Bα=Wα+B. Since Φ(Bα)=0, Wα ⊆ Asoc(Bα)=Nil(Bα) and Wα is completely reducible as
a B-bimodule. Hence W ⊆ Asoc(L). Since B complements W, Asoc(L) is complemented
and Φ(L)=0, a contradiction.
Hence there is one weight module and each left multipliction is a scalar. Pick a∈ B
and suppose that ax=αx on W. Let W0= {x ∈ W: xa=0} and W1={x∈W: xa= -αx}.
Since W is the direct sum of one dimensional a-invariant submodules, W=W0+W1. If,
for each a ∈ B, W=W0 or W=W1, then each right multiplication is a scalar on L2, and
L2 ⊆Asoc(L). Again Asoc(L) is complemented in L and Φ(L)=0, a contradiction. Thus
assume there is an a ∈ B such that neither W0 nor W1 is 0. W0 is a submodule. If W1 is a
submodule, then by induction both components are completely reducible under B, and W
⊆ Asoc(L). Thus Asoc(L) is complemented and Φ(L) = 0, a contradiction. Hence assume
there is an x in W1 and b in B such that xb is not in W1. Let N=(x,xb), bx=βx, and set
y=βx+xb. The following multiplications hold: ax=αx, ay=αy, bx=βx, by=βy, xa=-αx,
ya=(βx+xb)a=βxa-b(xa)=-αβx+αβx=0, xb=y-βx, and yb=(βx+xb)b=βxb-b(xb) =βxb-
βxb=0. Let C = (a,b). Hence N is a C-bimodule and (y) is a submodule which is not
complemented in N, for suppose that σx+τ y is C-invariant where σ 6= 0. Then (σx+τ y)a=-
σαx which yields that τ = 0. Then σx b= σy-σβ x and σ=0 since xb is not in W1. Thus
no complement exists. Hence Asoc(N+C) is not complemented in N+C and Φ(N+C) 6=0.
Thus B=C, N=W=L2, dim B=2 and dim L2=2 and the multipliction for B acting on L2
is the one given in this paragraph.
CLASSIFYING SEVERAL CLASSES OF LEIBNIZ ALGEBRAS
5
Let B=(a,b) and L2=(x,y). From the forgoing ax=αx, ay=αy bx=βx, by=βy, xa=-αx,
ya=0, yb=0, and xb=y-βx. N=(y) is a one dimensional submodule of L, and L2 is not
completely reducible under the action of L. This is the algebra in (4).
Suppose that dim B =1. Hence L=L2+B and L2 is abelian. We now show that there
exists a chain of ideals 0 ⊆ L1...Ln−1 ⊆=L2 ⊆ Ln=L where dim Li=i.
If P ⊆ Q⊆L2
are ideals of L with Q/P irreducible under the action of L, then dim Q/P=1 since the
action of x on Q/P determines the action of L on Q/P, and either Rx=-Lx or Rx=0.
Then any eigenvector of Lx in Q/P must span Q/P and dim Q/P=1. Hence there exists
a flag from 0 to L as claimed. Now M=Ln−2+B is a maximal subalgebra of L, hence
Φ(M) = 0 by assumption. Hence Ln−2 ⊆ Asoc(M) = Nil(M). Since L2 is abelian and B
is one dimensional, Asoc(M) ⊆ Asoc(L). Thus Ln−2 ⊆ Asoc(L). If Ln−2 6= Asoc(L), then
Asoc(L)=Ln−1=Nil(L), and L splits over Nil(L)=Asoc(L); hence, Φ(L)=0 from Theorem
3.1 of [3], and L is elementary. Otherwise, Asoc(L)=Ln−2 and Asoc(L) has co-dimension
two in L. If Asoc(L) is not contained in Φ(L), then L splits over Asoc(L) by Theorem 7.1
of [12]. Again Φ(L)=0. and L is elementary. Thus assume that Asoc(L)=Φ(L). Since all
minimal ideals are one dimensional, they are eigenspaces for Lx where B=<x>. Note that
Rx=0 or Rx=-Lx on each of these minimal ideals. Since L2/Asoc(L) is one-dimensional,
there exists a scalar α with (Lx − αI)2=0 and Lx-αI 6= 0. Hence there exist y,z ∈ L2 with
the following possible multiplications.
Case 1: Let xz=cz, xy=cy +z, zx=0, yx=-dcy+ez, where d=0 or 1 and c is a non-
zero scalar since L is not nilpotent. From x(yx)=(xy)x+y(xx) we obtain that -cd +ce=ce.
Hence d=0. From y(xx)=(yx)x+x(yx), we obtain that -cde-cd+ce=0. Hence e=0, and
yx=0.
Case 2: Let xz=cz, xy=cy+z, zx=-cz, yx=-dcy+ez where d=0 or 1 and c 6= 0. Using
x(yx)=(xy)x +y(xx), we obtain that -dc=-c. Hence d=1. From y(xx)=(yx)x+x(yx), we
obtain that cde=-cd. Hence e=-1, and yx=-cy-z.
Hence H=< x, y, z > has Φ(H)=< z > and L=H, which is the algebra in (1).
Let L be nilpotent with all proper subalgebras elementary.
If there exists an a ∈ L
with a2 6= 0, then the subalgebra B with basis a, a2,...,an and aan=0 has Φ(B)=B2. Since
a2 ∈ Φ(B), B=L, which is the algebra in (3). If no such a exists, then L is Lie and hence
Heisenberg by Theorem 4.7 of [8], yielding case (2).
Conversely, in cases (1) and (2) and (4), clearly all proper subalgebras are elementary.
Suppose that L is as in case (3). Then L=<a,a2,...,an > with aan=0. Since L is nilpotent,
L2=Φ(L). Now b= c1a+c2a2+...+cnan is in Φ(L) if and only if c1=0. If the subalgebra b
contains an element that is not in Φ(L), then B+Φ(L)=L since L/Φ(L) is one dimensional.
6
CHELSIE BATTEN RAY, ALLISON HEDGES AND ERNEST STITZINGER
Hence B=L. Therefore all proper subalgebras of L are contained in L2, which is abelian,
and hence are elementary. Thus L satisfies the conditions of the theorem.
We turn to the Leibniz algebra version of a Lie algebra result of Towers [13].
Theorem 7. Let L be a Leibniz algebra over K, an algebraically closed field of charac-
teristic 0. L is minimal non-elementary if and only if:
(1) L is three dimensional non-nilpotent with basis x,y,z and non-zero multiplication as:
(a) xz=cz, xy=cy+z, zx=0 and yx=0, where c is a non-zero scalar, or
(b) xz=cz, xy=cy+z, zx=-cz, yx=-cy-z where c is a non-zero scalar, or
(2) L is Heisenberg, or
(3) L is generated by a, where a2 6= 0, L is nilpotent and dim L ≥ 2, or
(4) L is the four dimensional non-nilpotent algebra generated by a,b,x and y with mul-
tiplcation. ax=α x, ay=αy, bx=βx, by=βy, xa=-αx, ya=0, xb=y-βx and yb=0 where α
and β are non zero scalars.
Proof: Suppose that L is minimal non-elementary. Then L is an E-algebra.
Suppose that L is not solvable. Hence there exists a k such that L(k)=L(k+1) If k=1,
then L is a perfect Lie algebra and L is elementary by Corollary 5, a contradiction.
If
k≥ 2, then Lk is perfect and L(k) ∼= sl2(K)L ... Lsl2(K). If R is the radical of L, then,
then L=RLL(k). Since both summands are elementary, L is elementary by Lemma 3 , a
contradiction. Hence L must be solvable. The result now follows from Theorem 6.
IV. THE JACOBSON RADICAL
The Jacobson radical J(L), is the intersection of maximal ideals of L. This concept was
considered in [5] and [9] when L is a Lie algebra. If L is nilpotent, then J(L)=Φ(L), since
then all maximal subalgebras are ideals [1]. Clearly J(L)⊆L2, since if x is not in L2, then
we can find a complementary subspace, M, of x in L that contains L2 and, since L2 ⊆ M,
M is a maximal ideal of L and x is not in M.
If L is a linear Lie algebra, let R=Rad(L), and let Rad(L∗) be the radical of the associative
envelope, L∗ of L. Then, by corollary 2 p. 45 of [7], L∩ Rad(L∗) = all nilpotent elements
of R and [R,L] ⊆Rad(L∗).
Theorem 8. Let L be a Leibniz algebra and R=Rad(L) be the radical of L. Then
LR+RL⊆N=Nil(L).
CLASSIFYING SEVERAL CLASSES OF LEIBNIZ ALGEBRAS
7
Proof: Let L (L)={Lx: x∈N}. The map π:L→ L (L) is a homomorphism and L (L)
is a Lie algebra under commutation.. Also π :R→R(L (L)), the radical of L (L). By the
result in the first paragraph, [L (L),R(L (L))]⊆R(L (L)∗). Hence there exists an n such
that [Lsn,Ltn]...[Ls2 ,Lt2][Ls1,Lt1]=0, where si ∈ L , ti R or si ∈ R and ti ∈ L. Hence
Lsntn...Ls1t1 = 0. Hence sntn(...(s1t1(x))...) = 0 for all x∈L. Hence sntn(...(s1t1(s0t0))...) =
0 Therefore(LR+RL)n+1=0. Since LR+RL is an ideal in L, LR+RL ⊆ N.
Proposition 9. If L is of characteristic 0, then L2∩R=LR+RL ⊆N.
Proof: L=R+S as in the Levi decomposition for Leibniz algebras ([2]). Then L2 =S2+LR+RL.
and L2∩R=S2∩R+(LR+RL)∩R=LR+RL since S∩R=0 and LR+RL⊆R.
Lemma 10. If L is solvable, then J(L)=L2.
Proof: If M is a maximal ideal of L, then L/M is abelian and, in fact, one-dimensional
since L is solvable. Hence L2 ⊆ M for all M and L2 ⊆ J(L). Since J(L) ⊆ L2 always holds,
L2=J(L).
Proposition 11 . Let L be a Leibniz algebra over a field of characteristic 0. Let R=R(L).
Then J(L)=LR+RL.
Proof. Let S be a Levi factor of L. Then S is Lie and S=S1 L ... LSt where each Si is
simple. Let Mi be the sum of R and all of the Sj except Si. Then Mi is a maximal ideal
of L. Then J(L) ⊆ T Mi=R and J(L) ⊆ R∩ L2.
If M is a maximal ideal of L, then L/M is abelian or simple. In the first case, L2 ⊆ M
and in the second case, R ⊆M. The intersection of all of the first type of M contains L2,
and all of the second type contains R. Therefore, R∩ L2 ⊆ J(L) and the result follows.
Corollary 12. J(L) is nilpotent.
Proof; J(L)=LR+RL ⊆ Nil(L) by Propositions 9 and 11.
Corollary 13. Φ(L) ⊆ J(L) when L has characteristic 0.
Proof: L=R+S as in the Levi decomposition and S is Lie. Hence Φ(S)=0. Thus Φ(L)
⊆ R. Since Φ(L) ⊆L2, it follows that Φ(L)⊆R∩L2=LR+RL=J(L) using Proposition 9 and
11.
Proposition 14. Let B be a nilpotent ideal in a Leibniz algebra, L. Then J(B) ⊆ J(L).
8
CHELSIE BATTEN RAY, ALLISON HEDGES AND ERNEST STITZINGER
Proof: Since B is nilpotent, J(B)=B2, which is an ideal in L. Suppose that x ∈ J(B),
x /∈ J(L). Let M be a maximal ideal of L such that x is not in M. Then L=J(B)+M and
B=J(B)+(M∩B). M∩B is a proper ideal of B that supplements J(B), a contradiction.
V. LEIBNIZ ALGEBRAS WITH A UNIQUE MAXIMAL IDEAL.
Lie algebras with a unique maximal ideal were classified in [5] when the field of scalars
has characteristic 0. We extend this result to Leibniz algebras.
Lemma 15. Suppose that L=N+<x> where <x> is a one dimensional vector space,
x2 6= 0, and N is an abelian ideal. Then xL is an ideal in L.
Proof. Since xL ⊆ N, N(xL)=(xL)N=0. (xL)x=(x(x+N))x=(x2+xN)x=(xN)x=x(Nx)-
N(x2) ⊆ xL since x2 ∈ N and N is abelian. Furthermore x(xL) ⊆ xL. Hence the result
holds.
Lemma 16. Suppose that L=N+<x> where <x> is a one dimensional vector space,
x2 6= 0 and N is an abelian ideal. If xL+Lx=N, then xL=N.
Proof. If xL 6= N, then xL is contained in an ideal M of L , M ⊆ N, and N/M is a
minimal ideal of L/M. Note that x2 ∈ M. We may take M=0. Then xN=0, which yields
Nx=0 by [1]. Then Lx=(N+x)x =0. Therefore xL+Lx=0, a contradiction
We now show
Theorem 17. Let L be a Leibniz algebra over a field of characteristic 0. L has a unique
maximal ideal if and only if one of the following holds.
1) L is nilpotent and cyclic, or
2) L=Nil(L)+S where S is simple and N/N2=S(N/N2)+(N/N2)S, or
3) L=Nil(L)+< x > where < x > is a one dimensional vector space, x2 6= 0 and
N/(N2)=x(L/N2), or
4) L=Nil(L)+< x > where x2=0 and N/(N2)=x(N/N2)+(N/N2)x.
Proof. Let L have only one maximal ideal which is then J(L)=J. Then L/J is simple
or one-dimensional. Since J ⊆ Nil(L)=N, if L/J is simple, then J=N and L=J+S=N+S.
To show the second part of (2), assume that N2=0 and let R=Rad(L). Then N=R and
J=LR+RL=SN+NS.
CLASSIFYING SEVERAL CLASSES OF LEIBNIZ ALGEBRAS
9
Suppose that dim(L/J)=1. Since J ⊆ N ⊆ L, either J=N or N=L. If N=L, then L is
nilpotent and J=Φ(L)=L2, so L is cyclic generated by some a and L is nilpotent and (1)
holds.
Suppose that N=J. Then L=N+< x > where < x > is a one dimensional subspace. Sup-
pose that x2 6= 0 . By Theorem 3.1 of [4], L/N2 is not nilpotent, hence Nil(L/N2) =N/N2
=J/N2=J(L/N2)=x(L/N2)+(L/N2)x. Suppose that N2=0 and T is the algebra generated
by x. Then Nil(L)=J(L)=xL+Lx=T2+xN+Nx. Then J(L/T2)=x(N/T2)+(N/T2)x=x(N/T2).
by Lemma 16. Hence Nil(L)= J(L)=T2+xN=xL and (3) holds. Assume that x2=0 and
N2=0. As in the last case, Nil(L)=J(L)=LR+RL=xN+Nx. Hence (4) holds.
We now show that each of the algebras in 1-4 have a unique maximal ideal. For the
algebras in (3), we have the following lemma.
Lemma 18. Suppose that L is Leibniz, and L=Nil(L)+< x > where x2 6= 0. Suppose
that xL=N and N2=0 where N=Nil(L). Then Nil(L)=J(L). The algebras in (3) have a
unique maximal ideal.
Proof. Let J=J(L). Since xL=N, N=L2. Let T be the subalgebra generated by x. Since
dim(Ker LxL)=1 and 0 6= Ker LxT ⊆ Ker LxL, it follows that Ker LxT = Ker LxL. Since
dim(Ker LxT )=1, the Fitting null component of LxT is the union T0 ⊆ T1 ⊆ ...Tk where
dim(Tj)=j and xTj ⊆ Tj−1. The same holds for the Fitting null component of Lx on L
where the invariant subspaces end with Lm. Clealy Ti=Li for i ≤ k. We claim that m=k.
There is a Ti that is not contained in N since the Fitting one component of Lx on T is
contained in N. Then αx+n ∈Ti, α 6= 0 n ∈ N. If i 6= k, then there exists βx+p such that
x(βx+p)=αx+n, which is impossible since x(βx+p) ∈ N=L2. Thus Tk−1 ⊆ N while Tk is
not contained in N. Since Tk=Lk is not contained in N, the chain of Li's must stop at Lk
since only the final term in the string is not in N. Hence the Fitting null component for Lx
acting on L is the same as Lx acting on T and the Fitting null component of Lx on N and
T2 are the same, both equal to Tk−1 ⊆ T2.
Let J=J(L). Since xL=N , N=L2. Let N=N0+N1 be the Fitting decomposition of Lx on
N. Let K be a maximal ideal of L and suppose that K is not contained in L2. There exists
y ∈ K such that y=αx+n0+n1 where α 6= 0, n0 ∈ N0 and n1 ∈ N1. For any t∈ N1, there
exists s ∈ N1 such that xs=t. Then ys=(αx +n0+n1)s=αt since N2 = 0. Thus αt ∈ K and
N1 ⊆ K. Also n0 ∈ N0=Tk−1 ⊆ T2 by the last paragraph. Hence n0 = α2x2 + ... + αtxt
and yx=(αx+α2x2 + ... + αtxt +n1)x =αx2 +n1x ∈ K. Since n1x ∈K, αx2 ∈ K. Therefore
x2 ∈ K and T2 ⊆ K. Since N0 ⊆ T2 and N1 ⊆ K, N ⊆ K. Since J(L)=N has codimension 1,
J(L)=N=K for all maximal ideals K of L. Hence algebras as in (3) have a unique maximal
ideal.
10
CHELSIE BATTEN RAY, ALLISON HEDGES AND ERNEST STITZINGER
We now consider the algebras in (1), (2), and (4). Suppose that L is as in (2). We may
assume that N2=0 since N/N2=Nil(L/N2). Thus J=LR+RL=SN+NS=Nil(L) and J(L) is
the only maximal ideal of L. If L is as in (4), then assume that N2=0 since N/N2=Nil(L/N2).
Then J(L)=LR+RL=xN+Nx. Since dim(L/N)=1, N is the only maximal ideal of L.
If L is as in (1), then L2=J(L) since L is nilpotent, and J(L) is the unique maximal ideal.
REFERENCES
1. D.W.Barnes, Some theorems on Leibniz algebras, Comm. In Alg. 39. , 2011, 2463-
2472
2. D. W. Barnes, On Levi's theorem for Leibniz algebras, Bull. Aust. Math. Soc. 86,
2012, 184-185
3. C. Batten Ray, L. Bosko-Dunbar, A. Hedges, J.T.Hird, K. Stagg, E. Stitzinger, A
Frattini theory for Leibniz algebras, Comm. in Alg. To appear, arXiv 1108.2451
4. C. Batten Ray, A. Combs, N. Gin, A. Hedges, J.T.Hird, L. Zack, Nilpotent Lie and
Leibniz algebras, Comm. In Alg. To appear, arXiv 1207.3739
5. P. Benito Clavijo, Lie algebras in which the lattice of ideals form a chain, Comm. In
Alg. 20, 1992, 93-108,
6. L. Bosko, A. Hedges, J.T. Hird, N. Schwartz, K. Stagg, Jacobson's refinement of
Engel's theorem for Leibniz algebras, Involve 4(3), 2011, 293-296
7. N. Jacobson, Lie algebras, Dover, 1979
8. J. Loday, Une version non commutative des algebres de Lie: les algebres de Leibniz,
Enseign Math. 39, 1993, 269-293
9. E. I. Marshall, The Frattini Subalgebra of a Lie Algebra, J. Lond. Math. Soc. 42,
1967 416-422
10. K. Stagg and E. Stitzinger, Minimal non-elementary Lie algebras, Proc. Amer.
Math. Soc. 137(7), 2010 , 2435-2437
11. E. Stitzinger, Frattini subalgebras of a class of solvable Lie algebras, Pac. J. Math.
34(1), 1970, 177-182
12. D. A. Towers, A Frattini theory for algebras, Proc. Lond. Math. Soc. 27(3), 1973,
440-462
13. D. A. Towers, Elementary Lie Algebras, J. Lond. Math. Soc. 7(2), 1973, 295-302
14. D. A. Towers, V. Varea, Elementary Lie Algebras and Lie A-Algebras, J. Alg. 312,
2007, 891-901
|
1210.4675 | 1 | 1210 | 2012-10-17T09:08:25 | Isomorphisms between strongly triangular matrix rings | [
"math.RA"
] | We describe isomorphisms between strongly triangular matrix rings that were defined earlier in Berkenmeier et al. (2000) as ones having a complete set of triangulating idempotents, and we show that the so-called triangulating idempotents behave analogously to idempotents in semiperfect rings. This study yields also a way to compute theoretically the automorphism groups of such rings in terms of corresponding automorphism groups of certain subrings and bimodules involved in their structure. | math.RA | math |
Isomorphisms between strongly triangular matrix rings
P.N. ´Anh and L. van Wyk
Abstract. We describe isomorphisms between strongly triangular matrix rings
that were defined earlier in [3] as ones having a complete set of triangulating
idempotents, and we show that the so-called triangulating idempotents be-
have analogously to idempotents in semiperfect rings. This study yields also a
way to compute theoretically the automorphism groups of such rings in terms
of corresponding automorphism groups of certain subrings and bimodules in-
volved in their structure, which completes the project started in [1].
1. Introduction
Triangular matrix rings appear naturally in the theory of certain algebras, like
nilpotent and solvable Lie algebras, Kac-Moody, Virasoro and Heisenberg algebras
(see, for example, [6]), as well as in algebras of certain directed trees. In the latter
case the triangular matrix rings may be seen to provide the abstract description of
such quiver algebras without mentioning the associated directed tree and without
appropriate numbering of the vertices.
Triangular matrix rings have become an important object of intense research,
for example, it is a key tool in the description of semiprimary hereditary rings
(see, for example, [4]), and certain triangular matrix rings are natural examples of
representation-finite hereditary algebras (see, for example, [2] and [5]).
On the other hand, Birkenmeier et al in [3] developed the general theory of gen-
eralized triangular matrix rings and used it to describe several particular classes of
rings. Combining their terminology with ones (introduced later) in [1] we say that
a ring A admits an m-strongly (upper) triangular matrix decomposition with respect
2010 Mathematics Subject Classification. 16S50, 15A33, 16D20.
Key words and phrases. triangular matrix ring, semicentral idempotent, semicentral reduced
ring, bimodule isomorphism.
Corresponding author: L. van Wyk
The research was carried out in accordance with the Hungary / South Africa Agreement
on Cooperation in Science and Technology. In particular, the first author was supported by the
Hungarian National Foundation for Scientific Research under Grants no. K-101515 and K-81928,
and the second author was supported by the National Research Foundation of South Africa under
grant no. UID 72375. Any opinion, findings and conclusions or recommendations expressed in
this material are those of the authors and therefore the National Research Foundation does not
accept any liability in regard thereto.
The authors are very grateful to the anonymous referee of a previous version of this work.
He/She pointed out some inaccuracies and called our attention to results in [3]. His/Her critical
remarks helped to improve this work.
The authors also thank L. M´arki and J. Szigeti for fruitful consultations.
1
2
P.N. ´ANH AND L. VAN WYK
to an ordered sequence {e1, . . . , em} if the ei's are pairwise orthogonal idempotents
in A such that 1 = e1 + · · · + em, ejAei = 0 for all j > i and eiAei is semicentral
reduced for every i, or equivalently, {e1, . . . , em} is a complete set of left triangulat-
ing idempotents by terminology of [3] . Here, according to [3], an idempotent e in
a ring A is called semicentral if (1 − e)Ae = 0, and A is called semicentral reduced
if 0 and 1 are the only semicentral idempotents in A, i.e., A is semicentral reduced
if and only if A is strongly indecomposabble in the sense of [1]. Therefore, an
idempotent e ∈ A is semicentral reduced if it is semicentral and the subring eAe is
a strongly indecomposable ring. If we set Ri := eiAei and Lij := eiAej for i < j,
then A can be written as a generalized upper triangular matrix ring
R1 L12 L13
0 R2 L23
· · ·
· · ·
. . .
. . .
...
0
0
L1m
L2m
...
· · ·
0 Rm−1 Lm−1,m
· · ·
· · ·
0
Rm
with the obvious matrix addition and multiplication. It was pointed out in [3] that
by reversing the order of the sequence {e1, . . . , em} one obtains a new sequence
providing the lower triangular matrix representation for A. Therefore it is not a
restriction to study rings with a complete set of left triangulating idempotents.
The aim of this paper is to describe isomorphisms between strongly triangular
matrix rings, thereby finishing the project initiated in [1]. As a by-product we
show that triangulating idempotents behave similarly to idempotents in semiper-
fect rings. Namely, if one fixes a complete set {e1, . . . , em} of triangulating idem-
potents, then a left ideal generated by any semicentral idempotent is isomorphic
to one generated by an appropriate partial sum of some idempotents from the set
{e1, . . . , em}.
For more information and detailed treatment of triangular matrix rings and
their applications in other areas of mathematics we refer to [3], and for some inter-
esting related questions on matrix rings we refer to [7].
2. Strongly triangular matrix rings
A strongly (upper) triangular matrix decomposition of a ring A depends essen-
tially on the ordered sequence {e1, . . . , em} of pairwise orthogonal idempotents
with sum 1. However, in particular cases, another ordering of the set {e1, . . . , em}
may also give a strongly triangular matrix decomposition of A.
Furthermore, if there is no room for misunderstanding, then for short we some-
times say that a ring A is a strongly triangular matrix ring, without stating exactly
the ordering on the set {e1, . . . , em}. Therefore one has to see clearly that all Ri
are semicentral reduced, but all ei, except e1, need not be even semicentral idem-
potents of A, i.e., ei for i ≧ 1 is certainly reduced semicentral only in the subring
Ai = (ei + · · · + em)A(ei + · · · + em) of A = A1 but not necessarily in Aj with j < i.
For example, if A is a strongly triangular matrix ring with respect to the ordered
ISOMORPHISMS BETWEEN STRONGLY TRIANGULAR MATRIX RINGS
3
sequence {e1, e2, e3}, then the generalized matrix decompositions of A with respect
to the ordered sequence {e2, e1, e3} and {e2, e3, e1} are
R2
0 L23
L12 R1 L13
0
0 R3
and
R2 L23
0
R3
0
0
L12 L13 R1
,
respectively, which are definitely not triangular matrix decompositions of A.
Next, let B be an n-strongly triangulated matrix ring with respect to an ordered
sequence {f1, . . . , fn}, i.e., the fi's are pairwise orthogonal idempotents in B with
sum 1, fjBfi = 0 for all j > i, Si := fiBfi is semicentral reduced for every i, and fi
is a semicentral reduced idempotent of the ring Bi = (fi + · · · + fn)B(fi + · · · + fn)
for i = 1, . . . , n. For each i 6= j, let Mij = fiBfj. Therefore Mij = 0 for all j < i.
Moreover, for each i let Mi be the i-th truncated row of B, i.e.,
Mi = ⊕k>iMik = ⊕k6=iMik.
If σ is any permutation on {1, . . . , n}, then σ induces a new (generalized) matrix ring
decomposition on B with respect to the ordered sequence {f σ
n :=
fσ(n)}. According to this notation, if we write gi = f σ
i , then one can identify the
above convention as follows. Let Ti = giBgi, Ci = (gi + · · · + gn)B(gi + · · · +
gn), Nij = giBgj for all i 6= j, Ni = ⊕k6=iNik. It is important to emphasize that B
is not necessarily an n-strongly triangular matrix ring with respect to the ordered
sequence {f σ
1 := fσ(1), . . . , f σ
1 , . . . , f σ
Ti = Sσ
i
n }. From the above one gets
:= Sσ(i), Ci = Bσ
i
:= Bσ(i), Ni = M σ
i
:= Mσ(i).
Now we are in a position to state the main result precisely.
Theorem. Let A and B be m- and n-strongly triangular matrix rings with respect
to ordered sequences {e1, . . . , em} and {f1, . . . , fn}, respectively. Then A and B are
isomorphic via an isomorphism ϕ : A → B iff m = n and there is a permutation σ
of {1, . . . , m} such that B is also an m-strongly triangular matrix ring with respect
to the ordered sequence {f σ
i =
Sσ(i), i = 1, . . . , m = n, and for i = 1, . . . , m − 1 there are elements mi ∈ M σ
i
and ring isomorphisms ϕi+1 : Ai+1 → Bσ
i+1 and Ri − Ai+1-bimodule isomorphisms
χi : eiAi(ei+1 + · · · + em(=n)) = Li → M σ
i with respect to ρi, ϕi+1, such that for
i = 1, . . . , m − 1 and
m}, there are ring isomorphisms ρi : Ri → Sσ
1 , . . . , f σ
ai =
ri
ℓi
0 ai+1
∈ Ai =
Ri
Li
0 Ai+1
,
ϕi(ai) =
ρi(ri) ρi(ri)mi + χi(ℓi) − miϕi+1(ai+1)
0
ϕi+1(ai+1)
.
Moreover, all isomorphisms between isomorphic rings A and B can be described in
this manner. (Keep in mind that ϕ1 = ϕ, ϕm = ρm; Am = Rm.)
4
P.N. ´ANH AND L. VAN WYK
Remark 1. The equality m = n as well as some invariants associated to a com-
plete set of triangulating idempotents up to a permutation σ were already obtained
as Theorem 2.10 in [3], where an isomorphism is (surprisingly enough) an inner
automorphism. However, these results are by-products of our description of gen-
eral isomorphisms between such rings and our treatment is both elementary and
direct. For further details for structural discussion we refer to Theorems 2.10, 3.3
and Corollary 3.4 in [3].
Proof of Theorem. We prove this theorem by induction on m, a number of
pairwise orthogonal idempotents ei in the ordered sequence {e1, . . . , em} giving a
strongly triangular matrix ring decomposition on A. The case m = 1 is obvious
by the definition, because B must be also semicentral reduced, i.e., m = n = 1.
Assume now that m ≥ 2 and the theorem holds for m − 1.
The first induction step is the following obvious but interesting result (by direct
computation, see also [1]). Because of its importance we state it separately as a
self-contained assertion.
Proposition. Let e ∈ A and f ∈ B be semicentral idempotents. Put R =
eAe, S = f Bf, ¯A = (1 − e)A(1 − e), ¯B = (1 − f )B(1 − f ), L = eA(1 − e), M =
f B(1 − f ), i.e., A = (cid:20) R L
¯B (cid:21), and let ϕ : A → B be a
ring isomorphism. Then ϕ(e) ∈ f + M if and only if there are ring isomorphisms
ρ : R → S and ¯ϕ : ¯A → ¯B and an R − ¯A-bimodule isomorphism χ : L → M (M is
an R − ¯A-bimodule via ρ and ¯ϕ) and an element m in M such that
¯A (cid:21) , B = (cid:20) S M
0
0
ϕ
r
ℓ
0 a
=
ρ(r) ρ(r)m + χ(ℓ) − m ¯ϕ(a)
0
¯ϕ(a)
.
(1)
In particular, χ is just the restriction of ϕ to L. Moreover, all isomorphisms ϕ
from A to B satisfying ϕ(e) ∈ f + M can be obtained from a quadruple (ρ, ¯ϕ, χ, m)
in this manner.
For the verification of the Proposition one observes ϕ(e) ∈ f +M if and only if
ϕ(e) = f +m = fm for some m ∈ M . Therefore ϕ(1−e) = 1−f −m = (1−f )−m =
gm. Put g = 1 − f . Then by direct calculations (see also Lemma 2.2 in [1]) one has
M = f Bg = fmBgm and canonical isomorphisms S ∼= fmBfm : v ∈ S 7→ v + vm ∈
fmBfm and ¯B ∼= gmBgm : w ∈ B1 7→ w − mw ∈ gmBgm. Consequently, ϕ induces
the isomorphisms ρ : R −→ S : r ∈ R 7→ ρ(r) = v, ¯ϕ : ¯A −→ ¯B : a ∈ ¯A 7→ ¯ϕ(a) = w
if ϕ(r) = v + vm ∈ fmBfm = ϕ(e)ϕ(A)ϕ(e) = fmBfm, ϕ(a) = w − mw ∈ ϕ(1 −
e)ϕ(A)ϕ(1 − e) = gmBgm. Therefore for an arbitrary element of A, i.e., for an
arbitrary triple r ∈ R, ℓ ∈ L, a ∈ A, one obtains (1) immediately.
Finally, it is clear that every quadruple (ρ, ¯ϕ, χ, m) as described in the state-
ment of the proposition leads to one of the desired isomorphisms, completing the
justification of the proposition.
Now we continue with the proof of the Theorem. Consider the
Main Step. Let A and B be m-and n-strongly upper triangular matrix rings
with respect to {e1, . . . , em} ⊆ A and {f1, . . . , fn} ⊆ B, respectively, and let ϕ :
A → B be a ring isomorphism. Let Ri = eiAei, Lij = eiAej for i < j, and
Si = fiBfi, Mij = fiBfj for i < j, i.e.,
ISOMORPHISMS BETWEEN STRONGLY TRIANGULAR MATRIX RINGS
5
A =
R1 L12 L13
· · · L1m
0 R2 L23
· · · L2m
...
...
. . .
. . .
. . .
. . .
...
...
0
· · ·
· · ·
0
Rm
, B =
S1 M12 M13
· · · M1n
0
...
...
S2 M23
· · · M2n
. . .
. . .
. . .
. . .
...
...
0
· · ·
· · ·
0
Sn
.
Then either ϕ(e1) ∈ f1 + M1 or there is a j ≥ 2 such that ϕ(e1) ∈ fj + Mj and
M1j = 0, . . . , Mj−1,j = 0.
Since f = ϕ(e1) ∈ B is a semicentral reduced idempotent, the statement of the
Main Step can be reformulated in an equivalent, but little sharper, form, namely:
If f is a semicentral reduced idempotent in the n-strongly triangulated matrix
ring B, then either f ∈ f1 + M1 or there is a j ≥ 2 such that f ∈ fj + Mj and
M1j = 0, . . . , Mj−1,j = 0.
Proof of the Main step. Again we use induction for the verification. Let F1 =
1 − f1, B2 = F1BF1, M = f1BF1, i.e., B = (cid:20) S1 M
for n = 1. Assume n ≥ 2. Writing f = (cid:20) α µ
from f = f 2 =
B2 (cid:21). The statement is obvious
B2 (cid:21) it follows
that α2 = α, β2 = β and αµ + µβ = µ, and so
(cid:21) and b = (cid:20) 0 µβ
(cid:21) we get s2 = s, b2 = b and
αµβ = 0. Writing s = (cid:20) α αµ
β (cid:21) ∈ B = (cid:20) S1 M
α2 αµ + µβ
0 β
β2
0
0
0
0
0
0
sb = 0 = bs. Hence, f = s + b implies that f s = s = sf and f b = b = f b, i.e.,
s, b ∈ f Bf , which is semicentral reduced. Moreover,
(cid:20) 0 µβ
0 β
(cid:21) (cid:20) S1 M
B2 (cid:21) (cid:20) α αµ
0
0
0
(cid:21) = (cid:20) 0 µβB2
0 βB2
(cid:21) (cid:20) α αµ
α 0
(cid:21) = 0,
i.e., s ∈ f Bf is a semicentral. Consequently,
0
0
or
f = (cid:20) 0 µβ
f = s = (cid:20) α αµ
(cid:21)
Assume the first case: f = (cid:20) α µ
0 (cid:21). Then (1 − f )Bf = 0 implies that
0 (cid:21) = 0,
0 (cid:21) = (cid:20) (1 − α)S1α ∗
1 (cid:21) (cid:20) S1 M
B2 (cid:21) (cid:20) α µ
(cid:20) 1 − α −µ
(cid:21) .
0 β
0
0
0
0
0
i.e., (1 − α)S1α = 0, hence α is a semicentral idempotent in S1. Since α 6= 0 and
S1 is semicentral reduced we obtain α = 1, i.e., f ∈ f1B.
6
P.N. ´ANH AND L. VAN WYK
Next, consider the case f = (cid:20) 0 µ
0 β (cid:21). Again (1 − f )Bf = 0 implies that
(cid:20) 1 −µ
0 1 − β (cid:21)(cid:20) S1 M
0 B2 (cid:21)(cid:20) 0 µ
0 β (cid:21) = (cid:20) 0
0
∗
(1 − β)B2β (cid:21) = 0,
showing that (1 − β)B2β = 0, i.e., β is semicentral. Since βB2β = βBβ = f Bf , we
have that β is also reduced. Since B2 is (n−1)-strongly triangular, the induction hy-
pothesis shows that there is a j ≥ 2 such that, in B2 =
S2 M23
0
S3
...
. . .
· · ·
0
· · · M2n
· · · M3n
...
. . .
0
Sn
,
∗
0
...
0
,
f is of the form
j − 1
0 · · ·
. . .
0
...
0
0
...
0
1 ∗
0
(cid:13)
· · ·
· · ·
. . .
i.e., M2j = 0, . . . , Mj−1,j = 0. Therefore, in B, f is of the form
0
· · ·
. . .
· · ·
. . .
0 x1
...
...
0
0
...
0
1
j
where x1 ∈ M1j. Now 0 = (1 − f )Bf =
1
0
. . .
· · ·
0 −x1
. . .
1
0
...
0
0
∗
1
· · ·
0
. . .
(cid:13)
· · ·
· · ·
. . .
. . .
∗
0
...
0
1
B
∗
0
· · ·
· · ·
. . .
(cid:13)
· · ·
· · ·
. . .
0
· · ·
. . .
· · ·
. . .
,
∗
0
...
...
0
0 x1
...
...
0
0
...
0
1
∗ · · ·
0 · · ·
. . .
(cid:13)
· · ·
· · ·
. . .
∗
0
...
...
0
implies both x1 = 0 and M1j = 0, completing the proof of the Main Step. (cid:3)
The following observation is the last piece in the proof of the Theorem.
ISOMORPHISMS BETWEEN STRONGLY TRIANGULAR MATRIX RINGS
7
Lemma. If
B =
S1 M12 M13
· · · M1n
0
...
...
S2 M23
· · · M2n
. . .
. . .
. . .
. . .
...
...
0
· · ·
· · ·
0
Sn
is n-strongly triangular with respect to the ordered sequence {f1, · · · , fn} of pair-
wise orthogonal idempotents such that M1j = 0, . . . , Mj−1,j = 0 for some index
j > 1, then B is also n-strongly triangular with respect to the ordered sequence
{fj, f1, . . . , fj−1, fj+1, . . . , fn}.
Proof. Obvious by definition. (cid:3)
If we define now σ(1) = j, then the above Lemma together with the Proposition
1 = Sσ(1), ϕ2 : A2 ∼=
shows that ϕ induces the ring isomorphisms ρ1 : R1 ∼= Sj = Sσ
¯B = (1 − fj)B(1 − fj) and the bimodule isomorphism χ1 : L1 = e1A(1 − e1) ∼=
M σ
1 such that for an
1 = Mσ(1) = fjB(1 − fj) together with an element m1 ∈ M σ
arbitrary a = a1 = (cid:20) r1
0 a2 (cid:21) ∈ A = A1 = (cid:20) R1 L1
ℓ1
ρ1(r1) ρ1(r1)m1 + χ1(ℓ1) − m1ϕ2(a2)
0 A2 (cid:21) , ϕ = ϕ1 satisfies
,
ϕ2(a2)
ϕ(a) = ϕ1(a1) =
0
and every such ϕ can be described in this manner. Since A1 is an (m − 1)-strongly
triangular matrix ring and ¯B is an (n − 1)-strongly triangular matrix ring, the
theorem follows now immediately from the induction hypothesis which makes the
proof of the Theorem complete. (cid:3)
We emphasize three important remarks.
Remark 2. If Mi,i+1 6= 0 for i = 1, . . . , n − 1, then {f1, . . . , fn} is the unique order
(up to isomorphism) which induces the n-strongly triangular matrix decomposition
on B.
Remark 3. If i < j and Mij = 0, . . . , Mj−1,j = 0, then {f1, . . . , fi−1, fj, fi, fi+1, . . . ,
fj−1, fj+1, . . . , fn} also induces an n-strongly triangular matrix decomposition on B.
In this case, the ring (fi + · · · + fj−1 + fj)B(fi + · · · + fj−1 + fj) is the direct sum of
the two rings (fi +· · ·+fj−1)B(fi +· · ·+fj−1) and Sj = fjBfj. In particular, in the
case i = 1, j = n the ring B is the direct sum of (f1 + · · · + fn−1)B(f1 + · · · + fn−1)
and Sn if the truncated last column is 0.
Remark 4. Specializing the theorem for the case A = B one obtains the description
of the automorphism group of the strongly triangular matrix rings in terms of the
corresponding automorphism groups of reduced rings Ri and of the corresponding
bimodules Li similar to one given in [1].
8
P.N. ´ANH AND L. VAN WYK
The proof of the main step shows also that for an arbitrary semicentral idem-
potent e in a strongly triangular matrix ring A each semicentral reduced idem-
potent g ∈ A is either g = (cid:20) α µ
0
0 (cid:21) or g = (cid:20) 0 ν
0 β (cid:21) where α ∈ eAe and
β ∈ (1 − e)A(1 − e) are semicentral reduced idempotents in A, the associated sub-
rings gAg, αAα, βAβ are isomorphic, and µ, ν are appropriate elements in eA(1−e).
Observing that C = (1 − α)A(1 − α) in the first case or C = (1 − β)A(1 − β) in the
second case is an (m − 1)-strongly triangular matrix ring, by considering ¯e = e − α
in the first case or in view of e = (1 − β)e(1 − β) in the second case, respectively, the
Main Step and the Theorem together with an obvious induction imply immediately
the following
Corollary. Any semicentral idempotent e in a m-strongly triangular matrix ring A
with a complete set of triangulating idempotents can be written as a sum of l
pairwise orthogonal idempotents {e1, . . . , el} where l ≤ m is uniquely determined
by e and this set of idempotents can be extended to the first l idempotents in a
complete set of triangulating idempotents of A.
References
[1] P.N. Anh and L. van Wyk, Automorphism groups of generalized triangular matrix rings,
Linear Algebra Appl. 434 (2011), 1018 -- 1025.
[2] I.N. Bernstein, I.M. Gelfand and V.A. Ponomarev, Coxeter functors, and Gabriel's theorem
(Russian), Uspehi Mat. Nauk 28 (1973), 19 -- 33.
[3] G.F. Birkenmeier, H.E. Heatherly, J.Y. Kim, J.K. Park, Triangular matrix representations,
J. Algebra 230 (2000), 558 -- 595.
[4] C. Faith, "Algebra II: Ring Theory", Springer, 1976.
[5] P. Gabriel, Unzerlegbare Darstellungen. I. (German) Manuscripta Math. 6 (1972), 71 -- 103.
[6] R.V. Moody and A. Pianzola, "Lie algebras with triangular decompositions", Wiley, New
York, 1995.
[7] J. Szigeti, Linear algebra in lattices and nilpotent endomorphisms of semisimple modules, J.
Algebra 319 (2008), 296 - 308.
R´enyi Institute of Mathematics, Hungarian Academy of Sciences, 1364 Budapest,
Pf. 127, Hungary
E-mail address: [email protected]
Department of Mathematical Sciences, Stellenbosch University, P/Bag X1,
Matieland 7602, Stellenbosch, South Africa
E-mail address: [email protected]
|
1911.10626 | 2 | 1911 | 2019-12-13T19:12:33 | The center of the total ring of fractions | [
"math.RA"
] | Let $A$ be a right Ore domain, $Z(A)$ be the center of $A$ and $Q_r(A)$ be the right total ring of fractions of $A$. If $K$ is a field and $A$ is a $K$-algebra, in this short paper we prove that if $A$ is finitely generated and ${\rm GKdim}(A)<{\rm GKdim}(Z(A))+1$, then $Z(Q_r(A))\cong Q(Z(A))$. Many examples that illustrate the theorem are included, most of them within the skew $PBW$ extensions. | math.RA | math |
The center of the total ring of fractions
Oswaldo Lezama
[email protected]
Helbert Venegas
[email protected]
Seminario de ´Algebra Constructiva - SAC2
Departamento de Matem´aticas
Universidad Nacional de Colombia, Sede Bogot´a
Abstract
Let A be a right Ore domain, Z(A) be the center of A and Qr(A) be the right total ring of fractions
of A. If K is a field and A is a K-algebra, in this short paper we prove that if A is finitely generated
and GKdim(A) < GKdim(Z(A)) + 1, then Z(Qr(A)) ∼= Q(Z(A)). Many examples that illustrate the
theorem are included, most of them within the skew P BW extensions.
Key words and phrases. Ore domains, total ring of fractions, center of a ring, Gelfand-Kirillov
dimension, skew P BW extensions.
2010 Mathematics Subject Classification. Primary: 16S85, 16U70. Secondary: 16P90, 16S36.
1 Introduction
Given an Ore domain A, it is interesting to know when Z(Q(A)) ∼= Q(Z(A)), where Z(A) is the center
of A and Q(A) is the total ring of fractions of A. This question became important after the formulation
of the Gelfand-Kirillov conjecture in [5]: Let G be an algebraic Lie algebra of finite dimension over a field
K, with char(K) = 0. Then, there exist integers n, k ≥ 1 such that
Q(U(G)) ∼= Q(An(K[s1, . . . , sk])),
(1.1)
where U(G) is the enveloping algebra of G and An(K[s1, . . . , sk]) is the general Weyl algebra over K.
In the investigation of this famous conjecture the isomorphism between the center of the total ring of
fractions and the total ring of fractions of the center occupies a special key role. There are remarkable
examples of algebras for which the conjecture holds and they satisfy the isomorphism. For example, if G
is a finite dimensional nilpotent Lie algebra over a field K, with char(K) = 0, then the conjecture holds
and Z(Q(U(G))) ∼= Q(Z(U(G))) ([5], Lemma 8). More recently, the quantum version of the Gelfand-
Kirillov conjecture has occupied the attention of many researchers. One example of this is the following
(see [2], Theorem 2.15): Let U +
q (slm) be the quantum enveloping algebra of the Lie algebra of strictly
q (slm)) ∼=
superior triangular matrices of size m × m, m ≥ 3, over a field K. If m = 2n + 1, then Q(U +
Q(Kq[x1, . . . , x2n2 ]), where Kq[x1, . . . , x2n2 ] is the quantum ring of polynomials, K := Q(Z(U +
q (slm))) and
1
q := [qij ] ∈ M2n2 (K), with qii = 1 = qijqji, and qij is a power of q for every 1 ≤ i, j ≤ 2n2. If m = 2n, then
q (slm)) ∼= Q(Kq[x1, . . . , x2n(n−1)]), where q := [qij ] ∈ M2n(n−1)(K), with qii = 1 = qij qji, and qij is
Q(U +
a power of q for every 1 ≤ i, j ≤ 2n(n − 1). Moreover, in both cases Z(Q(U +
q (slm))).
Let A be a right Ore domain. In this paper we study the isomorphism Z(Qr(A)) ∼= Q(Z(A)), where
Z(A) is the center of A and Qr(A) is the right total ring of fractions of A. The main tool that we will use
is the Gelfand-Kirillov dimension, so we will assume that A is a K-algebra, where K is an arbitrary field.
The principal result is Theorem 2.3 proved in Section 2. The result can also be interpreted as a way of
computing the center of Qr(A). In Section 3 we include many examples that illustrate the theorem, most
of them within the skew P BW extensions.
q (slm))) ∼= Q(Z(U +
We start with the following known facts about skew P BW extensions that will be used in the examples.
Definition 1.1 ([4]). Let R and A be rings. We say that A is a skew P BW extension of R (also called
a σ − P BW extension of R) if the following conditions hold:
(i) R ⊆ A.
(ii) There exist finitely many elements x1, . . . , xn ∈ A such A is a left R-free module with basis
Mon(A) := {xα = xα1
1 · · · xαn
n α = (α1, . . . , αn) ∈ Nn}, with N := {0, 1, 2, . . . }.
The set Mon(A) is called the set of standard monomials of A.
(iii) For every 1 ≤ i ≤ n and r ∈ R − {0} there exists ci,r ∈ R − {0} such that
xir − ci,rxi ∈ R.
(iv) For every 1 ≤ i, j ≤ n there exists ci,j ∈ R − {0} such that
Under these conditions we will write A := σ(R)hx1, . . . , xni.
xjxi − ci,jxixj ∈ R + Rx1 + · · · + Rxn.
(1.2)
(1.3)
Associated to a skew P BW extension A = σ(R)hx1, . . . , xni there are n injective endomorphisms
σ1, . . . , σn of R and σi-derivations, as the following proposition shows.
Proposition 1.2 ([4], Proposition 3). Let A be a skew P BW extension of R. Then, for every 1 ≤ i ≤ n,
there exist an injective ring endomorphism σi : R → R and a σi-derivation δi : R → R such that
for each r ∈ R.
xir = σi(r)xi + δi(r),
A particular case of skew P BW extension is when all σi are bijective and the constants cij are
invertible.
Definition 1.3 ([4]). Let A be a skew P BW extension. A is bijective if σi is bijective for every 1 ≤ i ≤ n
and ci,j is invertible for any 1 ≤ i < j ≤ n.
Proposition 1.4 ([1], Theorem 3.6). Let A = σ(R)hx1, . . . , xni be a bijective skew P BW extension of a
right Ore domain R. Then A is also a right Ore domain.
Proposition 1.5 ([11], Theorem 14). Let R be a K-algebra with a finite dimensional generating subspace
V and let A = σ(R)hx1, . . . , xni be a bijective skew P BW extension of R. If σi, δi are K-linear and
σi(V ) ⊆ V , for 1 ≤ i ≤ n, then
GKdim(A) = GKdim(R) + n.
2
Proposition 1.6. Let R be a commutative domain, σ an automorphism of R and
Rσ := {r ∈ Rσ(r) = r}.
If σ has infinite order, then Z(R[x; σ]) = Rσ. If σ has finite order v, then Z(R[x; σ]) = Rσ[xv].
Proof. The proof when R is a field can be found in [12], Proposition 1.6.25. For completeness we include
the proof in the general case. Firstly observe that Rσ is a subring of R[x; σ], whence Rσ[xv] is the subring
of R[x; σ] generated by Rσ and xv, this implies that the elements of Rσ[xv] are polynomial in xv with
coefficients in Rσ.
Let p(x) := p0 + p1x + · · · + pnxn ∈ Z(R[x; σ]), then for every r ∈ R, rp(x) = p(x)r, so for every
0 ≤ i ≤ n, we get rpi = piσi(r) = σi(r)pi. If the order of σ is infinite, then σi 6= iR for every i ≥ 1,
hence pi = 0 for i ≥ 1. Thus, in this case p(x) = p0; moreover, p(x) commutes with x, so p0x = xp0,
whence p0 = σ(p0), i.e., p0 ∈ Rσ. Therefore, if σ has infinite order, then Z(R[x; σ]) ⊆ Rσ, but clearly,
Rσ ⊆ Z(R[x; σ]). Suppose that σ has finite order, say, v. If σi 6= iR, i.e., if v ∤ i, then pi = 0, hence
p(x) = p0 + p1xv + p2(xv)2 + · · · + pt(xv)t. Since p(x) commutes with x, then σ(pi) = pi for every
0 ≤ i ≤ t. This proves that Z(R[x; σ]) ⊆ Rσ[xv]. But, Rσ[xv] ⊆ Z(R[x; σ]) since every element of the
form r(xv)t, with r ∈ Rσ and t ≥ 0, commutes with every element s ∈ R and with x.
2 Main theorem
We start with the following easy proposition. We include the proof for completeness.
Proposition 2.1. Let A be a right Ore domain.
(i) If p
q ∈ Z(Qr(A)) then
(a) pq = qp.
(b) For every s ∈ A − {0}, psq = qsp.
(c) p ∈ Z(A) if and only if q ∈ Z(A).
(ii) Let p ∈ A. Then, p
1 ∈ Z(Qr(A)) if and only if p ∈ Z(A). Thus, Z(A) ֒→ Z(Qr(A)).
(iii) If K is a field and A is a K-algebra such that Z(Qr(A)) = K, then Q(Z(A)) = K = Z(Qr(A)).
Proof. (i) (a) We have p
q
q
1 = q
1
p
q , so p
1 = qp
(b) For s = 0 is clear. Let s ∈ A − {0}, then p
(c) If p
q = 0, then p
q = 0
q
q , whence p
q = ps
1 , i.e., pq
1 = qp
1
qs , so by (a), psqs = qsps, whence psq = qsp.
1 , thus pq = qp.
1 = qp
q
q
(ii) If p
1 and the claimed trivially holds. We can assume that p 6= 0; by (b), for every
s ∈ A − {0}, psq = qsp, hence if p ∈ Z(A), then sqp = qsp, whence sq = qs, i.e., q ∈ Z(A). On the other
hand, since q
p ∈ Z(Qr(A)), then if q ∈ Z(A) we get p ∈ Z(A).
1 ∈ Z(Qr(A)), then by (i), ps = sp for every s 6= 0, whence p ∈ Z(A). Conversely, let
r = acp
rp ,
1 ∈ Z(Qr(A))), then for every a
s ∈ Qr(A) we have p
s and a
s = pa
1 = ac
p ∈ Z(A)−{0} (for p = 0, p
where sc = pr = rp, with c, r 6= 0. From this we get acp
a
rp = apc
sc = ap
s = pa
s , i.e., p
1 ∈ Z(Qr(A)).
1
p
s
(iii) From (ii), K ⊆ Z(A) ⊆ Z(Qr(A)) = K, so Z(A) = K, and hence Q(Z(A)) = K = Z(Qr(A)).
The next example illustrates the part (iii) of Proposition 2.1.
Example 2.2. We consider the quantum plane A := Kq[x, y], where q is not a root of unity. We will
1 xγj
show that Z(Qr(A)) = K. Let p
2 ,
with ri, uj ∈ K − {0}. From px1s = sx1p and since q is not a root of unity, we get βi + βiθj = γj + γjαi
for every 1 ≤ i ≤ t and 1 ≤ j ≤ l. Similarly, from px2s = sx2p we obtain θjβi + θj = αiγj + αi for all
i, j, whence βi + αi = γj + θj, so fixing i and then fixing j we conclude that p and s are homogeneous of
the same degree (this condition is not enough since x1
x2
s ∈ Z(Qr(A)) − {0}, where p := Pt
2 and s := Pl
/∈ Z(Qr(A))). Now,
j=1 ujxθj
i=1 rixαi
1 xβi
3
p
s
x1
1 = x1
1
p
s , i.e., Pt
Pl
i=1 rix
j=1 uj x
αi
1 x
θj −1
1
βi
2
γj
2
x
= Pt
Pl
i=1 rix
j=1 uj x
αi+1
1
θj
1 x
βi
x
2
γj
2
,
hence there exist c := xm
1 pm(x2) + · · · + p0(x2), d := xk
2 )c = (Pt
xγj
2 )c = (Pl
1 xβi
i=1 rixαi
j=1 ujxθj −1
(Pt
(Pl
1 qk(x2) + · · · + q0(x2) ∈ A − {0} such that
i=1 rixαi+1
j=1 ujxθj
xβi
2 )d,
1 xγj
2 )d.
1
1
Since p and q are homogeneous, we can assume α1 > · · · > αt and θ1 > · · · > θl, whence β1 < · · · < βt
and γ1 < · · · < γl. Then,
1 xβ1
(r1xα1
r1xα1+1
2 )(xm
xβ1
2 xk
1 pm(x2)) = r1qmβ1xα1+m
1qk(x2) = r1qkβ1 xα1+1+k
xβ1
2 pm(x2),
xβ1
2 qk(x2),
1
1
1
whence α1 + m = α1 + 1 + k, i.e., m = k + 1. Moreover, let pm be the leader coefficient of pm(x2) and qk
be the leader coefficient of qk(x2), then qβ1pm = qk. Similarly, we can prove that qγ1 pm = qk, but since
q is not a root of unity, β1 = γ1. From α1 + β1 = θ1 + γ1 we get that α1 = θ1 (considering instead the
identity p
s
p
s we obtain the same result). Thus, we have
1 = x2
x2
1
Notice that
α1 = θ1 and β1 = γ1.
p
s = r1u−1
1 + p−r1u−1
1 s
s
, with r1u−1
1 ∈ K ⊆ Z(Qr(A)),
∈ Z(Qr(A)). But observe that p−r1u−1
1 s
hence p−r1u−1
1 s
= 0, contrary, we could repeat the previous
procedure and find that there exists, either i ≥ 2 such that αi = θ1 = α1, βi = γ1 = β1, or j ≥ 2 such
that θj = θ1, γj = γ1, a contradiction. Thus, p
1 ∈ K, and hence, Z(Qr(A)) = K.
s = r1u−1
s
s
The previous example shows that the proof of the isomorphism Z(Qr(A)) ∼= Q(Z(A)) by direct
computation of the center of the total ring of fractions is tedious. An alternative more practical method
is given by the following theorem.
Theorem 2.3. Let K be a field and A be a right Ore domain. If A is a finitely generated K-algebra such
that GKdim(A) < GKdim(Z(A)) + 1, then
Z(Qr(A)) = { p
q p, q ∈ Z(A), q 6= 0} ∼= Q(Z(A)).
Proof. We divide the proof in three steps.
Step 1. As in the proof of Theorem 4.12 in [6], we will show that
Qr(A) ∼= A(Z(A)0)−1 = { p
q p ∈ A, q ∈ Z(A)0}, with Z(A)0 := Z(A) − {0}.
First observe that Z(A)0 is a right Ore set of A, so A(Z(A)0)−1 exists. From the canonical injec-
tion Q(Z(A)) ֒→ A(Z(A)0)−1, p
q , we get that A(Z(A)0)−1 is a vector space over Q(Z(A)),
q
moreover, A(Z(A)0)−1 = AQ(Z(A)). We will show that the dimension of this vector space is fi-
nite. Let V be a frame that generates A. Since {V n}n≥0 is a filtration of A, then A = Sn≥0 V n
and AQ(Z(A)) = Sn≥0 V nQ(Z(A)). Arise two possibilities: Either there exists n ≥ 0 such that
V nQ(Z(A)) = V n+1Q(Z(A)), or else
7→ p
Q(Z(A)) ( V Q(Z(A)) ( V 2Q(Z(A)) ( · · ·
In the first case AQ(Z(A)) = V nQ(Z(A)) and we get the claimed. In the second case,
dimQ(Z(A)) Q(Z(A)) (cid:12) dimQ(Z(A)) V Q(Z(A)) (cid:12) dimQ(Z(A)) V 2Q(Z(A)) (cid:12) · · ·
and we will show that this produces a contradiction. In fact, for every n ≥ 0,
4
dimQ(Z(A)) V nQ(Z(A)) ≥ n + 1;
let u1, . . . , ud(n) be a Q(Z(A))-basis of V nQ(Z(A)), thus, d(n) ≥ n + 1; we can assume that ui ∈ V n for
every 1 ≤ i ≤ d(n); let W be an arbitrary K-subspace of Z(A) of finite dimension, then
(V + W )2n ⊇ V nW n ⊇ u1W n ⊕ · · · ⊕ ud(n)W n
(the sum is direct since the elements ui are linearly independent over Z(A)); from this we get
dimK(V + W )2n ≥ d(n) dimk(W n) ≥ (n + 1) dimK(W n),
but since V + W is a frame of A, then GKdim(A) ≥ 1 + GKdim(Z(A)), false.
Now we can prove the claimed isomorphism. For this consider the canonical injective homomorphism
g : A → A(Z(A)0)−1, a 7→ a
1 is invertible in A(Z(A)0)−1. In fact, the map
p
h : A(Z(A)0)−1 → A(Z(A)0)−1, p
q , is an injective Q(Z(A))-homomorphism since A is a domain,
but as was observed above, A(Z(A)0)−1 is finite-dimensional over Q(Z(A)), therefore h is surjective,
whence, there exists p
1 : In fact, since p 6= 0,
there exists p′
q ∈ A(Z(A)0)−1 such that a
1
1 . If a ∈ A − {0}, then a
1 . Observe that p
p
1 , then
q′ in A(Z(A)0)−1 such that p
p′
q′ = 1
q 7→ a
1
1 = 1
q = 1
q = 1
a
1
p
q
q
1 , and since p
1 = p′
1 , whence a
q′
q
a
1
1
q
p
1
p′
q′ = 1
1
p′
q′ , i.e., a
1
1
q = p′
q′ , so a
1
1
q
q
1 = p′
q′
q
1 , so p
q
a
1 = p
q
p′
q′
q
1 = 1
q
p
1
p′
q′
q
1 = 1
q
q
1 = 1
1 .
p
In order to conclude the proof of the isomorphism A(Z(A)0)−1 ∼= Qr(A), observe that any element
q ∈ A(Z(A)0)−1 can be written as p
q = g(p)g(q)−1.
Step 2. Let C := { p
q p, q ∈ Z(A), q 6= 0}. If p
q ∈ Z(Qr(A)), then by the first step we can assume that
q ∈ Z(A)0, and from the part (i)-(c) of Proposition 2.1, we get p ∈ Z(A). Therefore, Z(Qr(A)) ⊆ C.
Conversely, let p
q ∈ C, then p, q ∈ Z(A), with q 6= 0, whence, by the part (ii) of Proposition 2.1,
1 , q
p
1 ∈ Z(Qr(A)), so 1
Step 3. According to the part (ii) of Proposition 2.1, we have the canonical injective homomorphism
1 , that sends invertible elements of Z(A) in invertible elements of Z(Qr(A)),
q = ι(p)ι(q)−1. This proves the
1
q ∈ Z(Qr(A)). Thus, C ⊆ Z(Qr(A)).
q ∈ Z(Qr(A)) can be written p
Z(A) ι−→ Z(Qr(A)), p 7→ p
moreover, by the step 2, every element p
isomorphism Z(Qr(A)) ∼= Q(Z(A)).
q ∈ Z(Qr(A)), and hence, p
q = p
1
3 Examples
Next we present many examples of K-algebras that satisfy the hypotheses of Theorem 2.3. Most of them
within the skew P BW extensions.
Example 3.1. (i) Any domain A such that dimK A < ∞. For example, the real algebra H of quaternions
since dimR(H) = 4.
(ii) Any right Ore domain A finitely generated as Z(A)-module.
Example 3.2. Applying Propositions 1.4 and 1.5, we will check next that the following skew P BW
extensions are K-algebras that satisfy the hypotheses of Theorem 2.3. The precise definition of any of
these algebras can be found in [8].
(i) Consider a skew polynomial ring A := R[x; σ], with R a commutative domain R that is a K-
algebra generated by a subspace V of finite dimension such that σ(V ) ⊆ V , σ is K-linear of finite order
m, Rσ = K and GKdim(R) = 0. Then, GKdim(A) = 1, and from Proposition 1.6, Z(A) = K[xm],
and hence, GKdim(Z(A)) = 1. Thus, Z(Qr(A)) ∼= Q(Z(A)) = K(xm). A particular case of this general
example is A := C[x; σ] as R-algebra, with σ(z) := z, z ∈ C (here C and R are the fields of complex and
real numbers, respectively). In this case the order of σ is two and GKdim(C) = 0 since dimR(C) = 2.
(ii) Let char(K) = p > 0 and A := Sh := K[t][xh; σh] be the algebra of shift operators. Then,
h(t) = t, i.e., the order of σh is p,
GKdim(A) = 2. Moreover, for every k ≥ 0, σk
h(t) = t − kh, then σp
5
therefore, Z(A) = K[t]σh [xp
h]. Since K[tp] ⊆ K[t]σh ⊆ K[t] and K[t] is finitely generated over K[tp], then
GKdim(K[tp]) = GKdim(K[t]) = 1, whence GKdim(K[t]σh ) = 1. Therefore, GKdim(Z(A)) = 2. Thus,
Z(Qr(A)) ∼= Q(Z(A)) = Q(K[t]σh )(xp
h).
(iii) Let char(K) = p > 0 and A := K[t][x; d
Then, GKdim(A) = 3 and can be proved that Z(A) = K[xp, xp
Z(Qr(A)) ∼= Q(Z(A)) = K(xp, xp
− tp).
h, tp2
dt ][xh; σh] be the algebra of shift differential operators.
− tp]. Since GKdim(Z(A)) = 3, then
h, tp2
(iv) Let char(K) = p > 0 and A := An(K) be the Weyl algebra. Since GKdim(A) = 2n and
n] (see [7], ejemplo 1.3.), then GKdim(Z(A)) = 2n. Therefore, Z(Qr(A)) ∼=
1, . . . , tp
Z(A) = K[tp
Q(Z(A)) = K(tp
n, xp
1, . . . , tp
1, . . . , xp
n, xp
1, . . . , xp
n).
(v) Let char(K) = p > 0 and A := J := K{x, y}/hyx − xy − x2i be the Jordan algebra. Since
GKdim(J ) = 2 and Z(A) = K[xp, yp] (see Theorem 2.2 in [13]), then GKdim(Z(A)) = 2, whence
Z(Qr(A)) ∼= Q(Z(A)) = K(xp, yp).
(vi) Consider the quantum plane A := Kq[x, y], with q 6= 1 a root of unity of degree m ≥ 2. Then
GKdim(A) = 2 and Z(A) = K[xm, ym] (see [13]). Therefore, Z(Qr(A)) ∼= Q(Z(A)) = K(xm, ym).
(vii) The previous example can be extended to the quantum polynomials A := Kq[x1, . . . , xn], where
n ≥ 2 and q ∈ K − {0, 1}, defined by
If q is a root of unity of degree m ≥ 2, then can be proved that if n even, then Z(A) = K[xm
Therefore, GKdim(Z(A)) = n = GKdim(A) and hence
1 , . . . , xm
n ].
xjxi = qxixj, with 1 ≤ i < j ≤ n.
Z(Qr(A)) ∼= Q(Z(A)) = K(xm
1 , . . . , xm
n ).
(viii) Let Aq be the quantum Weyl algebra generated by x, y with rule of multiplication yx = qxy + a,
where q, a ∈ K − {0}. If q is a primitive root of unity of degree m ≥ 2, then Z(Aq) = K[xm, ym] (see
[3]). Since GKdim(Aq) = 2, then Z(Qr(Aq)) ∼= Q(Z(A)) = K(xm, ym).
(ix) In [9] has been computed the center of the following algebras. In every example we assume that
the parameters q's are root of unity of degree l ≥ 2, or li ≥ 2, appropriately:
(a) Algebra of q-differential operators, then Z(A) = K[xl, yl] and GKdim(A) = 2, so Z(Qr(A)) ∼=
Q(Z(A)) = K(xl, yl).
(b) Additive analogue of the Weyl algebra, Z(A) = K[xl1
1 , . . . , yln
n ).
so Z(Qr(A)) ∼= Q(Z(A)) = K(xl1
1 , . . . , xln
n , yl1
1 , . . . , xln
n , yl1
1 , . . . , yln
n ] and GKdim(A) = 2n,
(c) Algebra of linear partial q-dilation operators, in this case we have GKdim(A) = 2n and Z(A) =
K[tl
1, . . . , tl
n, H l
1, . . . , H l
n]. Therefore, Z(Qr(A)) ∼= Q(Z(A)) = K(tl
1, . . . , tl
n, H l
1, . . . , H l
n).
(d) Algebra of linear partial q-differential operators, in this case we have GKdim(A) = 2n and Z(A) =
K[tl
1, . . . , tl
n, Dl
1, . . . , Dl
n]. Hence, Z(Qr(A)) ∼= Q(Z(A)) = K(tl
1, . . . , tl
n, Dl
1, . . . , Dl
n).
(x) Let sl(n, K) be the Lie algebra of 2 × 2 matrices with null trace with K-basis e, f, h. If char(K) =
2, then Z(U(sl(2, K))) = K[e2, f 2, h] (see [7], p. 147). Moreover, GKdim(U(sl(2, K))) = 3. Thus,
Z(Qr(A)) ∼= Q(Z(A)) = K(e2, f 2, h).
Remark 3.3. As occurs for the Gelfand-Kirillov conjecture (see [5]), if the hypotheses of Theorem 2.3 fail,
then the isomorphism Z(Qr(A)) ∼= Q(Z(A)) could hold or fail. Thus, the hypotheses are not necessary
conditions. For example,
(a) H is not finitely generated as Q-algebra, however Qr(H) = H and Z(Qr(H)) ∼= R ∼= Q(Z(H)).
(b) Let K be a field with char(K) = 0, and let G be a three-dimensional completely solvable Lie
algebra with basis x, y, z such that [y, x] = y, [z, x] = λz and [y, z] = 0, λ ∈ K − {0} (see Example 14.4.2
in [10]). If λ ∈ K − Q, then Z(U (G)) = K and GKdim(U(G)) = 3, thus, in this case GKdim(U(G)) ≮
GKdim(Z(U(G))) + 1, and 14.4.7 in [10] shows that Z(Qr(U(G))) ≇ Q(Z(U(G))).
6
(c) Let A := U +
q (slm) be the quantum enveloping algebra of the Lie algebra of strictly superior
triangular matrices of size m × m over a field K, where q ∈ K − {0} is not a root of unity.
In [2]
was proved that Z(A) is the classical commutative polynomial algebra over K in n variables, with
m = 2n or m = 2n + 1, whence, GKdim(Z(A)) = n. On the other hand, according to [2], p. 236,
A is an iterated skew polynomial ring of K of m(m−1)
. Thus,
GKdim(A) ≮ GKdim(Z(A)) + 1, however, Z(Qr(A)) ∼= Q(Z(A)).
variables, hence GKdim(A) = m(m−1)
2
2
References
[1] Acosta, J.P., Chaparro, C., Lezama, O., Ojeda, I., and Venegas, C., Ore and Goldie
theorems for skew P BW extensions, Asian-European J. Math., 06, 2013, 1350061 [20 pages].
[2] Alev, J. and Dumas, F., Sur le corps des fractions de certaines alg`ebres quantiques, J. of Algebra,
170, 1994, 229-265.
[3] Chan, K., Young, A., and Zhang, J.J., The discriminant formulas and applications, Algebra
Number Theory, 10, 2016, 557-596.
[4] Lezama, O. and Gallego, C., Grobner bases for ideals of sigma-PBW extensions, Communica-
tions in Algebra, 39 (1), 2011, 50-75.
[5] Gelfand, I. and Kirillov, A., Sur le corps li´es aux alg`ebres enveloppantes des alg`ebres de Lie,
Math. IHES, 31, 1966, 509-523.
[6] Krause, G.R., and Lenagan, T.H., Growth of algebras and Gelfand-Kirillov dimension, (Re-
vised edition), Graduate Studies in Mathematics, 22, AMS, Providence, RI (2000).
[7] Levandovskyy, V., Non-commutatve Computer Algebra for Polynomial Algebras: Grobner Bases,
Applications and Implementation, Doctoral Thesis, Universitat Kaiserslautern, Germany, 2005.
[8] Lezama, O. and Reyes, A., Some homological properties of skew P BW extensions, Comm.
Algebra 42, 2014, 1200-1230.
[9] Lezama, O. and Venegas, H., Center of skew PBW extensions, arXiv: 1804.05425 [math.RA],
2018.
[10] McConnell, J. and Robson, J., Noncommutative Noetherian Rings, Graduate Studies in Math-
ematics, AMS, 2001.
[11] Reyes, A., Gelfand-Kirillov dimension of skew PBW extensions, Rev. Col. Mat., 47 (1), 2013,
95-111.
[12] Rowen, L.H., Ring Theory, Vol.1, Academic Press, 1988.
[13] Shirikov, E. N., Two-generated graded algebras, Algebra Discrete Math., 3, 2005, 64-80.
7
|
1204.3159 | 1 | 1204 | 2012-04-14T10:18:53 | Multialternating graded polynomials and growth of polynomial identities | [
"math.RA"
] | Let G be a finite group and A a finite dimensional G-graded algebra over a field of characteristic zero. When A is simple as a G-graded algebra, by mean of Regev central polynomials we construct multialternating graded polynomials of arbitrarily large degree non vanishing on A. As a consequence we compute the exponential rate of growth of the sequence of graded codimensions of an arbitrary G-graded algebra satisfying an ordinary polynomial identity. In particular we show it is an integer.
The result was proviously known in case G is abelian. | math.RA | math |
MULTIALTERNATING GRADED POLYNOMIALS AND
GROWTH OF POLYNOMIAL IDENTITIES
ELI ALJADEFF AND ANTONIO GIAMBRUNO
Abstract. Let G be a finite group and A a finite dimensional G-graded alge-
bra over a field of characteristic zero. When A is simple as a G-graded algebra,
by mean of Regev central polynomials we construct multialternating graded
polynomials of arbitrarily large degree non vanishing on A. As a consequence
we compute the exponential rate of growth of the sequence of graded codi-
mensions of an arbitrary G-graded algebra satisfying an ordinary polynomial
identity. If cG
n (A), n = 1, 2, . . ., is the sequence of graded codimensions of A,
we prove that expG(A) = limn→∞
n (A), the G-exponent of A, exists and
is an integer. This result was proved in [1] and [9], in case G is abelian.
npcG
Introduction
Let F be a field of characteristic zero and A an F -algebra graded by a finite
group G. We shall assume throughout that A is a PI-algebra, i.e., it satisfies an
ordinary polynomial identity. The graded polynomial identities satisfied by A and
their growth have been extensively studied in the last years in an effort to develop
a general theory that generalizes the theory of ordinary polynomial identities.
For instance in analogy with a basic result in Kemer's theory (see [15]), it was
recently proved in [2] (and independently in [20] for G abelian) that if A is a finitely
generated algebra, then A satisfies the same graded identities as a finite dimensional
graded algebra. As a consequence one can reduce the study of an arbitrary G-graded
PI-algebra to that of the Grassmann envelope of a finite dimensional Z2 × G-graded
algebra.
Also, in analogy with the results in [11] and [12] concerning the existence of the
exponent of a PI-algebra, in [1] and [9] it was shown that when G is an abelian
group expG(A), the graded exponent of A, exists and is an integer.
Here we focus on the growth of the graded codimensions of A. Recall that if P G
n
is the space of multilinear graded polynomials of degree n and IdG(A) is the ideal
of graded identities of A, then cG
n (A), the nth G-codimension of A, measures the
dimension of P G
n mod. IdG(A).
It was known that the sequence cG
n (A), n = 1, 2, . . . , is exponentially bounded
([10]), but only recently its exponential rate of growth was captured in case G is
an abelian group. In fact in [1] for finitely generated algebras and in [9] in general,
it was shown that if A is any G-graded PI-algebra and G is a finite abelian group,
then expG(A) = limn→∞
n (A) exists and is an integer called the G-exponent of
npcG
2010 Mathematics Subject Classification. Primary 16R50, 16P90, 16R10, 16W50.
Key words and phrases. graded algebra, polynomial identity, growth, codimensions.
The first author was supported by the ISRAEL SCIENCE FOUNDATION (grant No. 1283/08)
and by the E. SCHAVER RESEARCH FUND. The second author was partially supported by
MIUR of Italy.
1
2
ELI ALJADEFF AND ANTONIO GIAMBRUNO
A. Moreover the G-exponent can be explicitly computed and equals the dimension
of a suitable finite dimensional semisimple graded algebra related to A.
In this note we shall extend the above result by proving that the G-exponent
exists and is an integer in case of arbitrary (non necessarily abelian) groups G. We
notice that the proof given in [9] works also when G is a non abelian group modulo
two basic ingredients that we manage to prove here. The first result says that
the multiplicities in the nth graded cocharacter of A are polynomially bounded,
the second is of independent interest and consists on the construction of suitable
multialternating graded polynomials non vanishing in a finite dimensional G-graded
algebra which is simple as a graded algebra.
Throughout the paper F will be a field of characteristic zero and A an associa-
tive F -algebra satisfying a non-trivial polynomial identity. We assume that A is
graded by a finite group G = {g1 = 1, g2, . . . , gs} and we let A = ⊕g∈GAg be its
decomposition into a sum of homogeneous components.
1. Multialternating polynomials on G-simple algebras
F . The set X decomposes as X = Ss
We start by introducing some standard notation.
We let F hX, Gi be the free associative G-graded algebra of countable rank over
i=1 Xgi , where the sets Xgi = {x1,gi , x2,gi , . . .}
are disjoint, and the elements of Xgi have homogeneous degree gi. The algebra
F hX, Gi is endowed with a natural G-grading F hX, Gi = ⊕g∈GFg, where Fg is
of homogeneous degree
the subspace spanned by the monomials xi1 ,gj1
g = gj1 · · · gjt .
· · · xit ,gjt
Recall that a graded polynomial f ∈ F hX, Gi is a graded (polynomial) identity
of A if f vanishes under all graded substitutions xi,g = ag ∈ Ag. Let also IdG(A) =
{f ∈ F hX, Gi f ≡ 0 on A} be the ideal of graded identities of A.
We say that an algebra A is G-graded simple if A is a G-graded algebra which
is simple as a graded algebra.
Let A be a finite dimensional G-graded simple algebra over an algebraically
closed field of characteristic zero. The purpose of this section it to produce non
identity G-graded polynomials with arbitrary many alternating sets of variables
which correspond to the homogeneous components of A and with a bounded number
of extra variables.
A key ingredient in the construction of these polynomials is a presentation of
any G-graded simple algebra as a tensor product of two types of G-graded simple
algebras, namely a twisted group algebra (with fine grading) and a matrix algebra
with an elementary grading. Here is the precise statement. It is due to Bahturin,
Sehgal and Zaicev.
Theorem 1.1. [4] Let A be a finite dimensional G-graded simple algebra. Then
there exists a subgroup H of G, a 2-cocycle α : H×H → F ∗ where the action of H on
F is trivial, an integer r and an r-tuple (g1, g2, . . . , gr) ∈ Gr such that A is G-graded
isomorphic to Λ = F αH ⊗ Mr(F ) where Λg = spanF {πh ⊗ ei,j g = g−1
i hgj}. Here
πh ∈ F αH is a representative of h ∈ H and ei,j ∈ Mr(F ) is the (i, j) elementary
matrix.
In particular the idempotents 1 ⊗ ei,j as well as the identity element of A are
homogeneous of degree e ∈ G.
MULTIALTERNATING GRADED POLYNOMIALS AND GROWTH
3
Let t1, . . . , ts > 0 be integers and, for i = 1, . . . s, define
X j
gi = {xj
1,gi
, . . . , xj
mi,gi } ⊆ Xgi , 1 ≤ j ≤ ti,
ti distinct sets consisting of mi ≥ 0 variables each of homogeneous degree gi. Let
i=1 Xgi be another set of homogeneous variables disjoint from the pre-
also Y ⊆ Ss
vious sets.
Also, let f = f (X 1
g1 , . . . , X t1
g1 , . . . , X 1
gs , . . . , X ts
graded polynomial in the variables from the sets X j
ti.
gs , Y ) ∈ F hX, Gi be a multilinear
gi and Y, 1 ≤ i ≤ s and 1 ≤ j ≤
This section is devoted to the proof of the following
Theorem 1.2. Let F be an algebraically closed field of characteristic zero and A
a finite dimensional G-graded simple algebra over F . For any t ≥ 1, there exist
integers
and a G-graded polynomial
2t ≤ t1, . . . , ts ≤ 2tG
ft(X (t1,...,ts)
G
; Y ) = ft(X 1
g1
, . . . , X t1
g1
, X 1
g2
, . . . , X t2
g2
, . . . , X 1
gs
, . . . , X ts
gs; Y )
such that
(1) ft(X (t1,...,ts)
G
in A of the form 1 ⊗ ei,j.
; Y ) is not an identity of A; in particular it has an evaluation
(2) the cardinality of Y depends on the order of G and the dimension of A and
not on the parameter t. In particular, the cardinality of Y is bounded.
(3) For every j and g ∈ G, X j
(4) ft(X (t1,...,ts)
G
g = dim Ag.
; Y ) is alternating on each one of the sets X j
g .
In view of the theorem above we claim that it is sufficient to construct G-graded
polynomials, which are non identities of A, and correspond to the cyclic subgroups
of G.
In order to make the statement precise, let g be any element of G and let S = hgi
be the subgroup it generates. We denote by d the order of S.
Proposition 1.3. It is sufficient to construct, for any integer t ≥ 1, a G-graded
polynomial (non identity of A)
ft,g(XS; YS) = ft,g(X 1
e , . . . , X 2t
e , X 1
g , . . . , X 2t
g , . . . , X 1
gd−1 , . . . , X 2t
gd−1; YS)
where
(1) YS ≤ r − 1 (r is the cardinality of the tuple which provides the elementary
grading on A).
(2) for every i = 1, . . . , 2t, and every 0 ≤ j ≤ d − 1, we have that X i
gj =
dim Agj .
(3) ft,g(XS; YS) is alternating on the set X i
gj for every i = 1, . . . , 2t, and every
0 ≤ j ≤ d − 1.
(4) ft,g(XS; YS) admits a non-zero G-graded evaluation on A of the form πgl ⊗
e1,r.
Remark 1.4. Clearly, by adding an extra variable if necessary, we may assume
that the value of the polynomial above is 1 ⊗ e1,1.
4
ELI ALJADEFF AND ANTONIO GIAMBRUNO
Proof. Indeed, having constructed the polynomials ft,g = ft,g(XS; YS) above, the
required polynomials are given by
ft(X (t1,...,ts)
G
; Y ) = Yg∈G
ft,g(XS; YS),
where Y = Pg∈G YS.
Consider the subalgebra AS = ⊕d−1
i=0 Agi ⊆ A. By [17, Theorem 1.6] (see also
[16, Theorem 18.13]), it is semisimple and so it decomposes into the direct sum of
S-graded simple algebras
(cid:3)
AS ∼= B1 ⊕ B2 ⊕ · · · ⊕ Bl.
It follows from Bahturin, Sehgal and Zaicev' result that for every i = 1, . . . , l, there
exists a subgroup Ci ≤ S and a pi-tuple (wi,1, . . . , wi,pi ) of elements in S (which
determines the elementary grading on Mpi(F )) such that
Bi ∼= F Ci ⊗ Mpi(F )
as S-graded algebras.
Notice that since Ci is cyclic, H 2(Ci, F ∗) = 0, 1 ≤ i ≤ l.
The structure of AS is given here up to an S-graded isomorphism. But we need
more. In fact, in order to "bridge" the S-simple components Bi by elements of A,
we need to realize the algebra AS as a subalgebra of A in terms of its presentation
(i.e. with the terminology of Theorem 1.1). Here is the precise statement.
First we make a definition:
for a homogeneous subspace D of A we define
weight(D) = {gi ∈ G D ∩ Agi 6= 0}.
Proposition 1.5. With the above notation:
(1) For every i = 1, ..., l, Bi ∼= F Ci ⊗ Mpi(F ) where Ci = H gji ∩ S.
(2) In terms of the presentation of A, after a possible permutation of the r-
tuple (g1, · · · , gr) (the tuple which provides the elementary grading in the
presentation of A), we have
Bi = span{πh ⊗ eu,v ∈ AS p1 + p2 + · · · + pi−1 + 1 ≤ u, v ≤ p1 + p2 + · · · + pi}.
In particular p1 + p2 + · · · + pl = r.
(3) For every i = 1, ..., l,
Bi = Bi,1 ⊕ · · · ⊕ Bi,ci
(direct sum of vector spaces that we shall call "pages") where
(a) Bi,k = span{πhu,v ⊗ eu,v ∈ AS p1 + p2 + · · · + pi−1 + 1 ≤ u, v ≤
p1 + p2 + · · · + pi} (for any pair (u, v) as above we choose a suitable
h = hu,v ∈ H).
(b) Any homogeneous component of Bi is "concentrated" in a unique page.
More precisely, weight(Bi,k) ∩ weight(Bi,l) = ∅, if k 6= l.
Before proving the proposition let us show how to construct the polynomial
ft,g(XS; YS).
For each "page" Bi,k we construct a Regev polynomial f k
(see [8]), with 2p2
i
variables whose homogeneous degrees are as the homogeneous degrees of Bi,k and
let
2p2
i
MULTIALTERNATING GRADED POLYNOMIALS AND GROWTH
5
fi = f 1
2p2
i
f 2
2p2
i
· · · f ci
2p2
i
.
Now, the evaluation of f k
on a basis of Bi,k gives an element of the form
πbk ⊗ 1pi×pi where πbk is a trivial unit of F Ci. This is a slight abuse of notation
since in fact the identity matrix 1pi×pi is located in the block diagonal between
rows p1 + p2 + · · · + pi−1 + 1 and p1 + p2 + · · · + pi.
2p2
i
Remark 1.6. This is the place where we use the fact that Ci is a cyclic group.
Indeed, the evaluation of a Regev polynomial on the space Bi,k has the same effect
as the evaluation on pi × pi matrix algebra where all monomials are multiplied by
the same trivial unit which is obtained as the product of commuting trivial units
of F Ci.
′
′
From the construction of fi we see that it has an evaluation of the form πb
⊗
1pi×pi where πb
is a trivial unit of F Ci. Note that since the homogeneous degrees
of the spaces Bi,k are disjoint for different k, the polynomial fi is alternating (in
particular) on sets of gν-variables, any gν ∈ S, of cardinality which is equal to the
dimension of the gν-homogeneous subspace of Bi.
i
i
In order to get arbitrary many alternating sets we let f t
i be the product of t-
copies (with disjoint sets of variables) of fi. Clearly, the evaluation of f t
i on a basis
of Bi gives an element of the form πai ⊗ 1pi×pi where πai is a trivial unit in F Ci.
i for any Bi. We can now "bridge"
Thus we have constructed a polynomial f t
these polynomials and get a polynomial
φS = x0f t
1x1f t
2 · · · f t
l xl+1.
We observe that with suitable evaluations on the x's the polynomial φS has an
evaluation which is equal to 1 ⊗ e1,1. But we are not done yet. We still need to
alternate variables with the same homogeneous degrees in the different polynomials
f t
i in a suitable way. More precisely, for every s = 1, . . . , t, we alternate all variables
2 , . . . , f s
with the same homogeneous degrees which appear in the polynomials f s
l .
Clearly, because of the bridging variables xj, the resulting polynomial ft,g(XS; YS)
admits the value 1 ⊗ e1,1 and has the required form.
1 , f s
Let us prove now the proposition above. To set up the notation again we recall
that A has a presentation given by F αH ⊗ Mr(F ) and the elementary grading is
given by the r-th tuple (g1, g2, . . . , gr). We let S be the cyclic group generated by
g ∈ G and we denote by d its order. Let us introduce an equivalence relation on
the index set {1, . . . , r} by setting i ∼ j if and only if g−1
i Hgj ∩ S 6= ∅. It is easy
to see that this is indeed an equivalence relation and we let
Ω1, . . . , Ωl
be the equivalence classes. We may clearly assume (after reordering the tuple
(g1, g2, ..., gr)) that equivalent indices are adjacent to each other. In other words
we have integers p1, . . . , pl such that
Ω1 = {1, . . . , p1}, Ω2 = {p1 + 1, . . . , p1 + p2},
. . . , Ωl = {p1 + · · · + pl−1 + 1, . . . , p1 + · · · + pl = r}.
We shall replace (as we may) elements from the r-tuple which represent the same
right H-coset by the same representative.
6
ELI ALJADEFF AND ANTONIO GIAMBRUNO
Consider indices i, j ∈ Ωk. We know (by the definition of the equivalence rela-
tion) that there exists an h ∈ H such that g−1
i hgj ∈ S. We claim that the number
of elements h ∈ H with that property depends on the index k but not on i and j.
In other words we claim the following.
Lemma 1.7. For every i, j, k0 ∈ Ωk, g−1
the sets g−1
i Hgj ∩ S are g−1
k0
Hgk0 ∩ S-cosets in S.
i Hgj ∩ S = g−1
k0
Hgk0 ∩ S. Furthermore,
i Hgk0 ∩ S and hence g−1
i Hgj ∩ S. For different elements q ∈ g−1
Proof. Take z ∈ g−1
j Hgk0 ∩ S, we obtain
different elements zq ∈ g−1
j Hgk0 ∩ S. On
the other hand, taking inverses we see that g−1
Hgj ∩ S. Being
i, j and k0 arbitrary the first part of the lemma follows. For the second part, note
that g−1
Hgk0 ∩ S are subgroups of the cyclic group S. By the first
part of the proof, they have the same order and hence they coincide. Following the
first part of the proof we see that g−1
i Hgj ∩ S is a (right, and hence 2-sided (by
commutativity)) g−1
i Hgi ∩ S-coset and the lemma is proved.
j Hgk0 ∩ S = g−1
k0
i Hgi ∩ S and g−1
k0
i Hgk0 ∩ S ≥ g−1
Now, observe that for i = 1, . . . , l, the algebra
Ui = span{πh ⊗ eu,v ∈ AS u, v ∈ Ωi}
(cid:3)
is a direct summand of AS and so, the proof of the proposition will be completed
if we show that Ui, i = 1, . . . , l, is S-simple. To see this we exhibit an explicit
presentation of Ui as a S-simple algebra.
Fix 1 ≤ k ≤ l and k0 ∈ Ωk. Let wi−n(k−1) ∈ S be a g−1
k0
Hgk0 ∩ S-coset rep-
Hgi ∩ S, where p1 + · · · + pk−1 + 1 ≤ i ≤ p1 + · · · + pk, and
resentative of g−1
k0
n(k−1) = p1 + · · · + pk−1. Then the map
φk : πh ⊗ ei,j 7−→ wi−n(k−1) g−1
i hgjw−1
j−n(k−1)
⊗ ei−n(k−1) ,j−n(k−1)
determines an isomorphism of the S-graded algebra Uk with F (g−1
Hgk0 ∩ S) ⊗
k0
Mpk (F ). In the latter, the elementary grading is given by the pk-tuple (w1, . . . , wpk ).
Details are omitted. Finally we note that Proposition 1.5 (b) follows from Lemma
1.7 (by ordering the coset's elements) and the proof of Proposition1.5 is completed.
Remark 1.8. Note that since the group g−1
k0
vanishes and hence we may use group elements rather then representatives.
Hgk0 ∩ S is cyclic, its cohomology
2. Graded exponent
Throughout F will be an algebraically closed field of characteristic zero and A
a G-graded PI-algebra over F with G = {g1 = 1, g2, . . . , gs} a finite group.
In this section we shall prove that the G-exponent of A exists and is an integer.
This result was proved in case G is an abelian group in [1] for finitely generated
algebras and in [9] in general. Here we shall follow closely that approach.
We start by recalling the general setting. The ideal of G-graded polynomial
identities of A is denoted IdG(A). For every n ≥ 1,
P G
n = spanF {xσ(1),giσ(1)
· · · xσ(n),giσ(n)
σ ∈ Sn, gi1 , . . . , gin ∈ G}
MULTIALTERNATING GRADED POLYNOMIALS AND GROWTH
7
is the space of multilinear G-graded polynomials in the homogeneous variables
x1,gi1
, . . . , xn,gin , gij ∈ G. We construct the quotient space
P G
n (A) =
P G
n
n ∩ IdG(A)
P G
and the non-negative integer
cG
n (A) = dimF P G
n (A), n ≥ 1,
is the nth G-graded codimension of A. Our aim is to prove that expG(A) =
limn→∞
n (A) exists and is an integer. Moreover we shall relate such an in-
teger to the dimension of a finite dimensional semisimple algebra related to A.
npcG
For every n ≥ 1, write n = n1 + · · · + ns a sum of non-negative integers and let
Pn1,...,ns ⊆ P G
n be the space of multilinear graded polynomials in which the first
n1 variables have homogeneous degree g1, the next n2 variables have homogeneous
degree g2 and so on. Then P G
n is the direct sum of subspaces isomorphic to Pn1,...,ns,
for every choice of the integers n1, . . . , ns. Moreover such decomposition is inherited
by Pn1,...,ns ∩ IdG(A) and we define
Pn1,...,ns(A) =
Pn1,...,ns
Pn1,...,ns ∩ IdG(A)
.
If we let
cn1,...,ns(A) = dim Pn1,...,ns(A)
then, by checking dimensions we have
(1)
cG
n (A) = Xn1+···+ns=n
n
(cid:18)
n1, . . . , ns(cid:19)cn1,...,ns(A),
where (cid:18)
n
n1, . . . , ns(cid:19) =
n!
n1! · · · ns!
denotes the multinomial coefficient. In order to
compute an upper and a lower bound for cG
and then apply (1).
n (A), it is enough to do so for cn1,...,ns(A)
The space Pn1,...,ns(A) is naturally endowed with a structure of Sn1 × · · · × Sns-
module in the following way. The group Sn1 × · · · × Sns acts on the left on Pn1,...,ns
by permuting the variables of the same homogeneous degree; hence Sn1 permutes
the variables of homogeneous degree g1, Sn2 those of homogeneous degree g2, etc..
Since IdG(A) is invariant under this action, Pn1,...,ns(A) inherits a structure of
Sn1 × · · · × Sns-module and we denote by χn1,...,ns(A) its character.
If λ is a partition of n, we write λ ⊢ n. It is well-known that there is a one-to-
one correspondence between partitions of n and irreducible Sn-characters. Hence if
λ ⊢ n, we denote by χλ the corresponding irreducible Sn-character. Now, if λ(1) ⊢
n1, . . . , λ(s) ⊢ ns, are partitions, then we write hλi = (λ(1), . . . , λ(s)) ⊢ (n1, . . . , ns)
and we say that hλi is a multipartition of n = n1 + · · · + ns.
Since char F = 0, by complete reducibility χn1,...,ns(A) can be written as a sum
of irreducible characters and let
(2)
χn1,...,ns(A) = Xhλi⊢n
mhλiχλ(1) ⊗ · · · ⊗ χλ(s),
where hλi = (λ(1), . . . , λ(s)) ⊢ (n1, . . . , ns) is a multipartition of n = n1 + · · · + ns
and mhλi ≥ 0 is the multiplicity of χλ(1) ⊗ · · · ⊗ χλ(s) in χn1,...,ns(A).
8
ELI ALJADEFF AND ANTONIO GIAMBRUNO
Our first objective is to prove that the multiplicities in (2) are polynomially
bounded.
Let E = he1, e2, . . . eiej = −ejeii be the infinite dimensional Grassmann al-
gebra over F and let E = E0 ⊕ E1 be its standard Z2-grading. Now, if B =
⊕(g,i)∈G×Z2B(g,i) is a G × Z2-graded algebra, then B has an induced Z2-grading,
B = B0 ⊕ B1, where B0 = ⊕g∈GB(g,0) and B1 = ⊕g∈GB(g,1), and an induced G-
grading B = ⊕g∈GBg where, for all g ∈ G, Bg = B(g,0)⊕B(g,1). Then one defines the
Grassmann envelope of B as E(B) = (B0 ⊗ E0) ⊕ (B1 ⊗ E1). Clearly E(B) has a G-
grading given by E(B) = ⊕g∈GE(B)g, where E(B)g = (B(g,0) ⊗E0)⊕(B(g,1) ⊗E1).
As in the ordinary case, the Grassmann envelope is an important object. In fact
by a result of Aljadeff and Belov ([2]), any variety of G-graded PI-algebras can be
generated by the Grassmann envelope of a suitable finite dimensional G×Z2-graded
algebra.
Let V = varG(A) denote the variety of G-graded algebras generated by A and
define V ∗ = {B = G × Z2-graded algebra such that E(B) ∈ V}. Then V ∗ is a
variety (see [13, Theorem 3.7.5]).
Now, according to [9, Corollary 1], there exist integers k ≥ l ≥ 0 such that in
(2) λ(1), . . . , λ(s) ∈ H(k, l), where H(k, l) = {λ ⊢ n λk+1 ≤ l} is the infinite k × l
hook.
Let L be the relatively free G × Z2-graded algebra of V ∗ on the (k + l)s graded
generators
(3)
(gi,0), vp
uj
(gi,1),
1 ≤ i ≤ s, 1 ≤ j ≤ k, 1 ≤ p ≤ l.
Then V = varG(E(L)) (see for instance [13, Theorem 4.8.2]).
Since L is a finitely generated G × Z2-graded PI-algebra, by [2, Theorem 1.1]
there exists a finite dimensional G × Z2-graded algebra W such that varG×Z2(L) =
varG×Z2(W ). Moreover L is a homomorphic image of a relatively free graded algebra
T of such variety on ks even generators and ls odd generators.
The algebra T can be constructed by "generic" elements as follows: fix a basis
{a1 . . . , am} of W of homogeneous elements. Let ξ(t)
, 1 ≤ i ≤ m, 1 ≤ t ≤ (k + l)s
be commutative variables and define ξ(t)
, 1 ≤ t ≤ k, where the
sum runs over all ij such that aij is of homogeneous degree (gi, 0). Similarly define
ξ(r)
, 1 ≤ r ≤ l, where the aij are of homogeneous degree (gi, 1).
Then T is generated by the "generic" elements
(gi,0) = P aij ⊗ ξ(t)
(gi,1) = P aij ⊗ ξ(r)
ij
ij
i
(4)
(gi,0), ξ(r)
ξ(t)
(gi,1), 1 ≤ i ≤ s, 1 ≤ t ≤ k, 1 ≤ r ≤ l.
Denote by Ln the subspace of L spanned by all products w1 · · · wi, 1 ≤ i ≤ n,
where w1, . . . , wi run over the generators given in (3). Define Tn accordingly on
the relatively free generators given in (4). Since L is a homomorphic image of T ,
dim Ln ≤ dim Tn and we compute an upper bound of this last dimension.
Notice that every monomial of degree at most n in the generic elements in (4),
]n is the subspace of polynomials of degree at
belongs to W ⊗ F [ξ(t)
most n in the commutative variables ξ(t)
]n, where F [ξ(t)
i
i
, 1 ≤ i ≤ m, 1 ≤ t ≤ (k + l)s.
i
Since dim F [ξ(t)
]n = (cid:0)(k+l)sm+n
i
dim Ln ≤ dim Tn ≤ m(n + (k + l)s)(k+l)sm ≤ Cnt,
(cid:1) ≤ (n + (k + l)s)(k+l)sm, we get that
n
(5)
for some constants C, t.
MULTIALTERNATING GRADED POLYNOMIALS AND GROWTH
9
We are now ready to prove the following.
Lemma 2.1. There exist constants C, t such that for all n ≥ 1, mhλi ≤ Cnt in (2).
Proof. Suppose that there exists n and hλi ⊢ n such that mhλi > Cnt ≥ dim Ln,
where C and t are defined in (5). Hence there exist mhλi = r irreducible Sn1 ×
· · · × Sns-modules M1, . . . , Mr ∈ Pn1,...,ns with character χλ(1) ⊗ · · · ⊗ χλ(s) and
(M1 ⊕ · · · ⊕ Mr) ∩ IdG(V) = 0. Now, as in [9, Lemma 4], we may take Mi =
F (Sn1 × · · · × Sns)hi where, by eventually adding some empty sets of variables,
we may assume that each hi is a polynomial in the homogeneous sets of variables
gs, i ≤ j ≤ k, 1 ≤ p ≤ l, and hi is symmetric in each Y j
g1 , . . . , Z p
g1 . . . , Y j
Y j
gt and
gt . Since (M1 ⊕ · · · ⊕ Mr) ∩ IdG(V) = 0, for every choice of
alternating in each Z p
gs , Z p
i=1 αihi 6∈ IdG(V).
α1, . . . , αr ∈ F not all zero, we have that h = Pr
gt as even variables and those of Z p
each set Y j
the polynomials hi, 1 ≤ i ≤ r. Then hi is symmetric on each Y j
Let be the map defined in [9, Section 5]. Then, if we regard the variables of
gt as odd variables, we can construct
gt and on each Z p
gt.
ri,gi }.
Define S to be the relatively free G-graded algebra of the variety V on the graded
generators
For every i and j, let Y j
mi,gi } and Z j
gi = {yj
gi = {zj
, . . . , yj
, . . . , zj
1,gi
1,gi
p,gi , ¯zj
¯yj
q,gi ,
1 ≤ i ≤ s, 1 ≤ p ≤ mi, 1 ≤ q ≤ ri, j = 1, 2, . . . .
Then the algebra Q = (S ⊗ E0) ⊕ (S ⊗ E1) has a natural G × Z2-grading and we
can take its Grassmann envelope
E(Q) = (S ⊗ E0 ⊗ E0) ⊕ (S ⊗ E1 ⊗ E1) ⊆ S ⊗ (E0 ⊗ E0 ⊗ E1 ⊗ E1).
Since E0 ⊗ E0 ⊗ E1 ⊗ E1 is commutative, E(Q) and S satisfy the same G-graded
identities. Hence E(Q) ∈ V and, so, Q ∈ V ∗.
Now in each polynomial hi, 1 ≤ i ≤ r, we identify the variables in each set
gi and in each set Z j
Y j
i be the corresponding polynomial. Since
deg hi = n, under the evaluation
gi and we let h◦
ϕ(Y j
gi ) = uj
we have that ϕ(h◦
gi ) = vp
(gi,0), ϕ(Z p
i ) ∈ Ln, for all 1 ≤ i ≤ r.
(gi,1),
1 ≤ i ≤ s, 1 ≤ j ≤ k, 1 ≤ p ≤ l,
Since by hypothesis r > dim Ln, there exist scalars α1, . . . , αr not all zero, such
i ) = 0 in L. Recalling that L is a relatively free algebra of V ∗, we
i ∈ IdG×Z2(V ∗). Now, Q ∈ V ∗ and, so, h◦ ∈ IdG×Z2(Q).
i=1 αih◦
If we consider the evaluation in Q = (S ⊗ E0) ⊕ (S ⊗ E1) defined by
i=1 αih◦
that ϕ(Pr
obtain that h◦ = Pr
gi ) = ¯yj
gi ) = ¯zj ′
ϕ(Y j
1,gi ⊗ αj
1,gi ⊗ βj ′
1 + · · · + ¯yj
1 + · · · + ¯zj ′
mi,gi ⊗ αj
mi ,
ri,gi ⊗ βj ′
ri ,
1 ≤ i ≤ s, 1 ≤ j ≤ k,
t , βj ′
ϕ(Z j ′
t are disjoint monomials of E of the correct homogeneous degree (αj
where αj
homogeneous degree 0 and βj ′
By computing explicitly we obtain 0 = ϕ(h◦) = ϕ′(
is an evaluation such that
t of
t of homogeneous degree 1), we get that ϕ(h◦) = 0.
h) ⊗ γ where 0 6= γ ∈ E and ϕ′
1 ≤ i ≤ s, 1 ≤ j′ ≤ l,
ϕ′(yj
p,gi) = ¯yj
p,gi
, ϕ′(zj ′
q,gi ) = ¯zj ′
q,gi
,
with 1 ≤ p ≤ mi, 1 ≤ q ≤ ri, 1 ≤ j ≤ k, 1 ≤ j′ ≤ l, 1 ≤ i ≤ s. Since
obtain that ϕ′(h) = ϕ′(
h) = 0.
h = h, we
10
ELI ALJADEFF AND ANTONIO GIAMBRUNO
Now, recall that the elements ¯yj
graded algebra of V of countable rank. Hence ϕ′(h) = 0 says that h = Pr
IdG(V), and this contradiction completes the proof.
, j ≥ 1, generate the relatively free G-
i=1 αihi ∈
(cid:3)
, ¯zj
q,gi
p,gi
Next we shall prove the existence of the G-exponent of A. Let B be a finite
dimensional G × Z2-graded algebra such that varG(A) = varG(E(B)). By the
Wedderburn-Malcev theorem [7] and the result in [21], we can write B = B1 ⊕ · · · ⊕
Bk + J, where B1, . . . , Bk are G × Z2-graded simple algebras and J is the Jacobson
radical of B. Recall that according to [9, Section 3], a semisimple subalgebra
D = D1 ⊕ · · · ⊕ Dh, where D1, . . . , Dh ∈ {B1, . . . , Bk} are distinct, is admissible if
D1J D2J · · · J Dh 6= 0. Then one defines
(6)
p = p(B) = max (dim D)
where D runs over all admissible subalgebras of B.
We shall prove that p coincides with the G-exponent of A. In fact we have.
Theorem 2.2. Let B be a finite dimensional G × Z2-graded algebra over an al-
gebraically closed field of characteristic zero. Then there exist constants C1 >
0, C2, k1, k2 such that
where p = p(B) is the integer defined in (6).
C1nk1 pn ≤ cG
n (E(B)) ≤ C2nk2 pn,
Proof. This theorem is proved in [9] for G abelian but the proof carries over to the
non abelian case by making the following changes.
In the computation of the upper bound cG
n (E(B)) ≤ C2nk2 pn one should use
Lemma 2.1 above instead of [9, Remark 1].
For the computation of the lower bound C1nk1 pn ≤ cG
n (E(B)) one should use
Theorem 1.2 instead of [1, Lemma 18], concerning the construction of multialter-
nating polynomials of arbitrary large degree for finite dimensional G-graded simple
algebras. We should point out that while the polynomial constructed in [1, Lemma
18] depends on a parameter t, the one constructed in Theorem 1.2 depends on s
parameters t1, . . . , ts (each corresponding to a homogeneous component of the al-
gebra) and these parameters are squeezed between 2t and 2Gt. Then one notices
that in [9, Section 6] the proofs are carried over for each homogeneous component
separately. This completes the proof of the theorem.
(cid:3)
Since graded codimensions do not change by extension of the base field, we get
the following.
Theorem 2.3. Let G be a finite group and A a G-graded PI-algebra over any
field F of characteristic zero. Then expG(A) = limn→∞
n (A) exists and is an
integer.
npcG
References
[1] E. Aljadeff, A. Giambruno and D. La Mattina, Graded polynomial identities and exponential
growth, J. Reine Angew. Math. 650 (2011), 83-100.
[2] E. Aljadeff and A. Kanel-Belov, Representability and Specht problem for G-graded algebras,
Adv. Math. 225 (2010), 2391-2428.
[3] Y. A. Bahturin and V. Drensky, Graded polynomial identities of matrices, Linear Algebra
Appl. 357 (2002), 15 -- 34.
[4] Y. A. Bahturin, S. K. Sehgal and M. V. Zaicev, Finite-dimensional simple graded algebras,
Sb. Math. 199 (2008), no. 7, 965-983.
MULTIALTERNATING GRADED POLYNOMIALS AND GROWTH
11
[5] A. Berele and A. Regev, Hook Young diagrams with applications to combinatorics and to
representations of Lie superalgebras, Adv. Math. 64 (1987), 118-175.
[6] A. Berele and A. Regev, Exponential growth for codimensions of some p.i. algebras, J. Algebra
241 (2001), 118-145.
[7] C. W. Curtis and I. Reiner, Representation theory of finite groups and associative algebras,
Wiley Classics Library, John Wiley & Sons, Inc., New York, 1988.
[8] E. Formanek, A conjecture of Regev about the Capelli polynomial, J. Algebra 109 (1987),
93-114.
[9] A. Giambruno and D. La Mattina Graded polynomial identities and codimensions: computing
the exponential growth, Adv. Math. 225 (2010), 859-881.
[10] A. Giambruno and A. Regev, Wreath products and P.I. algebras, J. Pure Applied Algebra,
35 (1985), 133-149.
[11] A. Giambruno and M. Zaicev, On codimension growth of finitely generated associative alge-
bras, Adv. Math. 140 (1998), 145-155.
[12] A. Giambruno and M. Zaicev, Exponential codimension growth of P.I. algebras: an exact
estimate, Adv. Math. 142 (1999), 221-243.
[13] A. Giambruno and M. Zaicev, Polynomial Identities and Asymptotic Methods, AMS, Math-
ematical Surveys and Monographs Vol. 122, Providence, R.I., 2005.
[14] G. James and A. Kerber, The Representation Theory of the Symmetric Group, Encyclopedia
of Mathematics and its Applications, Vol. 16, Addison-Wesley, London, 1981.
[15] A. Kemer, Ideals of identities of associative algebras, AMS Translations of Mathematical
Monograph, Vol. 87, Providence, R.I., 1988.
[16] D. S. Passman, Infinite crossed products, Academic Press, San Diego, CA, 1989.
[17] D. Quinn, Group-graded rings and duality, Trans. Amer. Math. Soc. 292 (1985), no. 1,
155167.
[18] A. Regev, Existence of identities in A ⊗ B, Israel J. Math. 11 (1972), 131-152.
[19] A. Regev, Asymptotic values for degrees associated with stripes of Young diagrams, Adv. in
Math. 41 (1981), 115-136.
[20] I. Sviridova, Identities of PI-algebras graded by a finite abelian group, Comm. Algebra, 39
(2011), 3462-3490.
[21] E. Taft, Orthogonal conjugacies in associative and Lie algebras, Trans. Amer. Math. Soc.
113 (1964), 18 -- 29.
Department of Mathematics, Technion-Israel Institute of Technology, Haifa 32000,
Israel
E-mail address: [email protected]
Dipartimento di Matematica e Informatica, Universit`a di Palermo, Via Archirafi 34,
90123 Palermo, Italy
E-mail address: [email protected]
|
1705.01293 | 1 | 1705 | 2017-05-03T08:22:57 | Order 3 elements in G2 and idempotents in symmetric composition algebras | [
"math.RA",
"math.GR"
] | Order three elements in the exceptional groups of type G2 are classified up to conjugation over arbitrary fields. Their centralizers are computed, and the associated classification of idempotents in symmetric composition algebras is obtained. Idempotents have played a key role in the study and classification of these algebras.
Over an algebraically closed field, there are two conjugacy classes of order three elements in G2 in characteristic not 3 and four of them in characteristic 3. The centralizers in characteristic 3 fail to be smooth for one of these classes. | math.RA | math |
ORDER 3 ELEMENTS IN G2 AND IDEMPOTENTS IN
SYMMETRIC COMPOSITION ALGEBRAS
ALBERTO ELDUQUE⋆
Abstract. Order three elements in the exceptional groups of type G2 are
classified up to conjugation over arbitrary fields. Their centralizers are com-
puted, and the associated classification of idempotents in symmetric compo-
sition algebras is obtained. Idempotents have played a key role in the study
and classification of these algebras.
Over an algebraically closed field, there are two conjugacy classes of order
three elements in G2 in characteristic 6= 3 and four of them in characteristic 3.
The centralizers in characteristic 3 fail to be smooth for one of these classes.
1. Introduction
Symmetric composition algebras constitute an important class of algebras strongly
related to the triality phenomenon (see [KMRT98, Chapter VIII]). They were classi-
fied in [EM91, EM93] over fields of characteristic 6= 3 and in [Eld97] in characteristic
3.
Idempotents in these algebras have proved to be a key tool in this classification,
and these idempotents, in dimension 8, are related to order 3 automorphisms of
Cayley algebras, that is, to order 3 rational elements in simple linear algebraic
groups of type G2: the groups of automorphisms of Cayley algebras.
The goal of this paper is to classify, up to conjugation, the order 3 elements in the
groups of automorphisms of Cayley algebras over arbitrary fields, and to deduce
from here a complete description of the idempotents in symmetric composition
algebras.
Over an algebraically closed field of characteristic 6= 3, it is easy to check that, up
to conjugation, there are two different order 3 elements, which correspond to the
two nonisomorphic eight-dimensional symmetric composition algebras: the para-
Cayley algebra and the Okubo algebra. Over arbitrary fields of characteristic 6= 3,
still there are two types of order 3 elements in the group of automorphisms of a
Cayley algebra (Theorem 5.1).
The situation is more interesting in characteristic 3, where the existence of a
unique 'quaternionic idempotent' in the split Okubo algebra was of crucial impor-
tance in the determination of the affine group scheme of automorphisms of this
algebra in [CEKT13], and explains why the group of automorphisms is a subgroup
of the exceptional algebraic group of type G2. In characteristic 3, once a torus is
fixed on the linear algebraic group of type G2 over an algebraically closed field, for
any root α relative to this torus and any nonzero scalar t, the element "exp(txα)"
has order 3. It turns out that there are four different types of order 3 elements in the
automorphism group of a split Cayley algebra over an arbitrary field of characteritic
3 (Theorem 6.3).
2010 Mathematics Subject Classification. Primary 17A75; Secondary 20G15, 14L15, 17B25.
Key words and phrases. Symmetric composition algebra; Okubo algebra; Automorphism
group; Centralizer; Idempotent.
⋆ Supported by the Spanish Ministerio de Econom´ıa y Competitividad -- Fondo Europeo de
Desarrollo Regional (FEDER) MTM2013-45588-C3-2-P .
1
2
A. ELDUQUE
In Section 2, the basic definitions and results on both unital composition algebras
and on symmetric composition algebras will be recalled. The construction of the
split Cayley algebra as the algebra of Zorn matrices will be reviewed. In Section
3, a specific Chevalley basis of the Lie algebra of type G2 will be given, and some
noteworthy subgroups of automorphisms will be considered.
Section 4 will highlight some specific order 3 automorphisms of Cayley algebras
related to para-Cayley algebras. Idempotents in para-Cayley algebras and in sym-
metric composition algebras of dimension 6= 8 will be described too in this section.
Section 5 is devoted to describe the conjugacy classes of order 3 elements in
the group of automorphisms of a Cayley algebra over a field of characteristic 6=
3, together with their centralizers, and this is used to describe the idempotents
in eight-dimensional symmetric composition algebras over these fields. The split
Okubo algebra is characterized as the only Okubo algebra with isotropic norm
and idempotents (Theorem 5.9). The number of conjugacy classes of idempotents
depends on the ground field.
The remaining sections will deal with the much more difficult case of fields of
characteristic 3. Section 6 will classify the conjugacy classes of order 3 automor-
phisms of Cayley algebras. There are four different types of such automorphisms.
Section 7 will be devoted to compute their centralizers. Some of them are not
smooth. Finally, Section 8 deals with idempotents of Okubo algebras over fields of
characteristic 3. In the split case, there is a unique quaternionic idempotent, some
conjugacy classes of quadratic idempotents, the number of which depends on the
ground field, and a unique conjugacy class of singular idempotents (Theorem 8.5).
Over non perfect fields of characteristic 3, there are non split Okubo algebras with
idempotents, and all these idempotents are quadratic.
Throughout the paper, the functorial approach to algebraic groups (i.e.; algebraic
affine group schemes) will be followed (see, e.g., [KMRT98, Chapter VI]). Thus, an
algebraic group is a representable functor from the category of unital, commutative,
associative F-algebras AlgF to the category of groups, so that the (Hopf) algebra
that represents it is finitely generated (as an algebra).
Any linear algebraic group G defined over a field F in the classical sense (as in,
e.g., [Spr09]) defines a smooth algebraic group which will be denoted too by G.
Also, for a finite dimensional algebra (A, ·) over F, Aut(A, ·) will denote its
automorphism group. This is the group of rational points of the algebraic group
Aut(A, ·): Aut(A, ·) = Aut(A, ·)(F). The Lie algebra of Aut(A, ·) is the Lie algebra
of derivations Der(A, ·).
Note that, in general, Aut(A, ·) may fail to be smooth. This is what happens
for the Okubo algebras over fields of characteristic 3 ([CEKT13, §10,11]).
2. Composition algebras
Definition 2.1. A composition algebra over a field F is a triple (C, ∗, n) such that
• (C, ∗) is a nonassociative (i.e., not necessarily associative) algebra.
• n : C → F is a nonsingular multiplicative (i.e. n(x ∗ y) = n(x)n(y) for any
x, y ∈ C) quadratic form.
The unital composition algebras are termed Hurwitz algebras.
The quadratic form being nonsingular means that either {x ∈ C : n(x, C) = 0}
is trivial, or it is a nonisotropic one-dimensional subspace. Here we will denote the
polar form: n(x, y) := n(x + y) − n(x) − n(y), by the same symbol n.
The reader is referred to [KMRT98, Chapter VIII] or to [ZSSS82, Chapter 2] for
the basic facts about composition algebras.
ORDER 3 ELEMENTS IN G2 AND IDEMPOTENTS
3
If the multiplication and the norm are clear from the context, we will simply
refer to the algebra C.
Hurwitz algebras form a well-known class of composition algebras. Any element
of a Hurwitz algebra (C, ·, n) satisfies the 'Cayley-Hamilton equation'
x·2 − n(x, 1)x + n(x)1 = 0.
Besides, (C, ·, n) is endowed with an involution: the standard conjugation, defined
by
¯x := n(x, 1)1 − x,
which satisfies
¯¯x = x,
for any x, y ∈ C.
x · y = ¯y · ¯x,
x + ¯x = n(x, 1)1,
x · ¯x = ¯x · x = n(x)1,
The dimension of any Hurwitz algebra is finite and restricted to 1, 2, 4 or 8,
and any Hurwitz algebra is either the ground field, a quadratic ´etale algebra, a
quaternion algebra or a Cayley (or octonion) algebra (see [KMRT98, (33.17)]).
Example 2.2. The Zorn matrix algebra is the compostion algebra (C, ·, n) with
v β(cid:19) : α, β ∈ F, u, v ∈ F3(cid:27) ,
C =(cid:26)(cid:18)α u
v′ β′(cid:19) =(cid:18) αα′ + (u v′)
α′v + βv′ + u × u′
v β(cid:19) ·(cid:18)α′ u′
(cid:18)α u
αu′ + β′u − v × v′
ββ′ + (v u′) (cid:19) ,
with
and
n(cid:18)(cid:18)α u
v β(cid:19)(cid:19) = αβ − (u v),
where, for u = (µ1, µ2, µ3) and v = (ν1, ν2, ν3) ∈ F3, (u v) denotes the usual
i=1 µiνi, and u × v the usual vector product: u × v =
(µ2ν3 − µ3ν2, µ3ν1 − µ1ν3, µ1ν2 − µ2ν1).
Let {a1, a2, a3} be the standard basis of F 3. The elements
scalar product: (u v) = P3
0(cid:19) , f1 =(cid:18)0 0
e1 =(cid:18)1
0
0
0 1(cid:19) , ui =(cid:18)0 −ai
0 (cid:19) , vi =(cid:18) 0
0
ai
0
0(cid:19) , i = 1, 2, 3,
form a hyperbolic basis, called the canonical basis of the Zorn algebra (C, ·, n). The
multiplication table in this basis is given in Table 1.
e1
e1
0
0
0
0
v1
v2
v3
e1
e2
u1
u2
u3
v1
v2
v3
e2
0
e2
u1
u1
0
u2
u2
0
u3
u3
0
v1
0
v1
0
u1
u2 −v3
u3
v2 −v1
v3 −v2 −e1
0
0
v1
0
v2
0
v2
0
−e1
v3
0
v3
0
0
0
−e1
0
0
0 −e2
0
0
0
−e2
0
0
0
0
0
−e2
u3 −u2
0
u1
0
−u3
u2 −u1
Table 1. Multiplication table in a canonical basis of the Zorn
matrix algebra.
4
A. ELDUQUE
Remark 2.3. Given the Zorn matrix algebra and its Peirce decomposition
C = Fe1 ⊕ Fe2 ⊕ U ⊕ V,
(2.1)
where U = span {u1, u2, u3} = e1 · C · e2, and V = span {v1, v2, v3} = e2 · C · e1, the
trilinear map U × U × U → F, (x, y, z) 7→ n(x, y · z), is alternating and nonzero.
Then, given any basis {u1, u2, u3} of U with n(u1, u2 · u3) = 1, its dual basis relative
to n in V is {v1 = u2 · u3, v2 = u3 · u1, v3 = u1 · u2}, and {e1, e2, u1, u2, u3, v1, v2, v3}
is another canonical basis, that is, it has the same multiplication table.
The subalgebra Fe1 ⊕ Fe2 is isomorphic to the algebra F × F, the subalgebra
Fe1 ⊕ Fe2 ⊕ Fu1 ⊕ Fv1 is isomorphic to the algebra M2(F) of two by two matrices.
The algebras F × F, M2(F) and Zorn matrix algebra exhaust, up to isomorphism,
the Hurwitz algebras with isotropic norm. Together with the ground field, these
are called the split Hurwitz algebras.
We will make use several times of the following helpful result. Throughout the
paper, an idempotent of an algebra (A, ∗) is a nonzero element e ∈ A such that
e ∗ e = e.
Lemma 2.4. Let (C, ·, n) be the Zorn matrix algebra over a field F. Then,
• Any idempotent of (C, ·, n), different from the unity 1, is conjugate to e1.
(That is, there is an automorphism ϕ ∈ Aut(C, ·, n) such that ϕ(e) = e1.)
• Any nonzero element x ∈ C with x·2 = 0 is conjugate to u1.
Proof. The elements e and 1 − e are orthogonal idempotents, so there is an isomor-
phism ϕ : Fe1 ⊕ Fe2 → F1 ⊕ Fe such that ϕ(e1) = e. By means of the Cayley-
Dickson doubling process (as, for instance, in [SV00, Corollary 1.7.3] or [EK13,
proof of Corollary 4.7]), ϕ can be extended to an automorphism of (C, ·, n).
Now, if x·2 = 0, let y ∈ C with n(x, y) = −1 and n(1, y) = 0. Changing y by
y − 1
n(y) x we may assume n(y) = 0 too. Then ¯x = −x, ¯y = −y and 1 = −n(x, y)1 =
x · y + y · x. It follows that f = x · y and 1 − f = y · x are orthogonal idempotents.
The assignment e1 7→ f , e2 7→ 1 − f , u1 7→ x, v1 7→ y, gives an isomorphism
Fe1 ⊕ Fe2 ⊕ Fu1 ⊕ Fv1(cid:0)∼= M2(F)(cid:1) onto the subalgebra generated by x and y that
extends to an automorphism of (C, ·, n).
(cid:3)
Definition 2.5. A composition algebra (S, ∗, n) is said to be symmetric if the polar
form of the norm is associative:
for any x, y, z ∈ S. This is equivalent to the condition
n(x ∗ y, z) = n(x, y ∗ z)
(x ∗ y) ∗ x = n(x) = x ∗ (y ∗ x)
for any x, y ∈ C.
(2.2)
(2.3)
If e is an idempotent of a symmetric composition algebra (S, ∗, n), then the map
τ : x 7→ e ∗ (e ∗ x) = n(e, x)e − x ∗ e
(2.4)
is an automorphism of (S, ∗, n) such that τ 3 = id (see [EP96, Theorem 2.5]). More-
over, the new multiplication on S given by
x · y = (e ∗ x) ∗ (y ∗ e)
(2.5)
makes (S, ·, n) a Hurwitz algebra with unity e, and the original multiplication is
recovered as
x ∗ y = τ (¯x) · τ 2(¯y),
(2.6)
for any x, y ∈ S. Moreover, τ is also an automorphism of (S, ·, n).
ORDER 3 ELEMENTS IN G2 AND IDEMPOTENTS
5
Proposition 2.6. Let e be an idempotent of a symmetric composition algebra
(S, ∗, n), let τ be the automorphism defined in (2.4), and let (S, ·, n) be the Hur-
witz algebra with multiplication given in (2.5). Then,
(i) the subalgebra of fixed points by τ :
Fix(τ ) := {x ∈ S : τ (x) = x}
coincides with the centralizer of e:
Cent(S,∗,n)(e) := {x ∈ S : e ∗ x = x ∗ e}.
(ii) The centralizer in the group scheme of automorphisms of (S, ·, n) of τ co-
incides with the stabilizer of e in the group scheme of automorphisms of
(S, ∗, n):
CentAut(S,·,n)(τ ) = StabAut(S,∗,n)(e).
Proof. For x ∈ S, τ (x) = x if and only if e ∗ (e ∗ x) = x, if and only if (e ∗ (e ∗ x))∗ e =
x ∗ e, if and only if (because of (2.3)) e ∗ x = x ∗ e. This proves the first part.
If ϕ ∈ Aut(S, ∗, n) fixes e: ϕ(e) = e, then clearly ϕ is an automorphism of
Aut(S, ∗, n) that commutes with τ . Conversely, if ϕ ∈ Aut(S, ·, n) and ϕτ = τ ϕ,
then by (2.6), ϕ ∈ Aut(S, ∗, n), and ϕ(e) = e because e is the unity of (S, ·, n).
Besides, all of this is functorial, so it is valid at the level of group schemes.
(cid:3)
Definition 2.7. Given a Hurwitz algebra (C, ·, n) over a field F and an automor-
phism τ ∈ Aut(C, ·, n) with τ 3 = id, the new algebra defined on C by means of (2.6),
and with the same norm, is called a Petersson algebra, and denoted by Cτ .
This multiplication appeared for the first time in [Pet69]. Any Petersson algebra
is a symmetric composition algebra. If τ is the identity automorphism: τ = id, the
Petersson algebra Cid is called the para-Hurwitz algebra associated to the Hurwitz
algebra (C, ·, n). We will talk about para-quadratic algebras in dimension 2, para-
quaternion algebras in dimension 4, and para-Cayley (or para-octonion) algebras
in dimension 8.
Usually the multiplication in a para-Hurwitz algebra will be denoted by •: x•y =
¯x · ¯y. The unity 1 of C becomes an idempotent of Cid that satisfies 1 • x = x • 1 =
Idempotents with this property are called para-units. Any
para-unit lies in the commutative center K(C, •) := {x ∈ C : x • y = y • x ∀y ∈ C},
which is the whole C if dimF C = 1 or 2, and equals F1 otherwise. Hence there is
a unique para-unit if the dimension is 4 or 8. In particular, this implies that the
group scheme of automorphisms Aut(C, ·, n) and Aut(C, •, n) coincide and that
any form of a para-Hurwitz algebra (i.e., any algebra that becomes isomorphic to
a para-Hurwitz algebra after an extension of scalars) is itself para-Hurwitz (if the
dimension is 4 or 8).
¯x(cid:16)= n(x, 1)1 − x(cid:17).
There is a natural order 3 automorphism of the Zorn matrix algebra:
τst(cid:18)(cid:18)
α
(µ1, µ2, µ3)
(ν1, ν2, ν3)
β
(cid:19)(cid:19) =(cid:18)
α
(µ3, µ1, µ2)
(ν3, ν1, ν2)
β
(cid:19) .
(2.7)
Definition 2.8 ([EP96]). Let (C, ·, n) be the Zorn matrix algebra, and let τst be
its order 3 automorphism in (2.7). The Petersson algebra Cτst is called the split
Okubo algebra.
The forms of the split Okubo algebra are called Okubo algebras.
This is not the original definition of these algebras given by Okubo in [Oku78].
Theorem 2.9 ([OO81a, OO81b, EP96]). Any symmetric composition algebra is
either a form of a para-Hurwitz algebra (hence a para-Hurwitz algebra if the dimen-
sion is 6= 2), or an Okubo algebra.
6
A. ELDUQUE
The proof in [EP96] works as follows. Let (S, ∗, n) be a symmetric composition
algebra. Then either it contains an idempotent or it contains an idempotent after a
cubic field extension of degree 3 ([KMRT98, ((34.10)]). Assuming the existence of
an idempotent, the arguments above show that the symmetric composition algebra
is a Petersson algebra. Then in [EP96] it is shown how to find, assuming the field is
algebraically closed, an idempotent such that either the automorphism τ in (2.4) is
the identity (i.e., e is a para-unit) or the dimension is 8 and τ is, up to conjugation,
the automorphism τst in (2.7).
3. Zorn matrix algebra and automorphisms
Let us consider the Zorn matrix algebra (C, ·, n), as defined in Example 2.2. The
stabilizer in Aut(C, ·, n) of the orthogonal idempotents e1 and e2:
{ϕ ∈ Aut(C, ·, n) : ϕ(ei) = e1, i = 1, 2}
(3.1)
is isomorphic to the special linear group SL(U) ≃ SL3 (see [Jac58] for the rational
points, but the arguments are valid in general), where U is the Peirce component
in (2.1), with the action of f ∈ SL(U) given by
f ·(cid:18)α u
v β(cid:19) =(cid:18)
α
(f t)−1(v)
f (u)
β (cid:19) ,
where f t ∈ SL(V) is the adjoint of f relative to the scalar product (. .). The torus
T consisting of the diagonal matrices in SL(U), relative to the basis {u1, u2, u3} in
Table 1, is a split maximal torus of Aut(C, ·, n).
In the same vein, {d ∈ Der(C, ·, n) : d(ei) = 0, i = 1, 2} is a Lie subalgebra of
Der(C, ·, n) isomorphic to sl(U) ≃ sl3(F), and its diagonal subalgebra is a Cartan
subalgebra of Der(C, ·, n).
The weights of the action of T on C are {0, ±εi : 1 ≤ i ≤ 3} (additive notation),
where for f ∈ T, f = diag(α1, α2, α3) (with α1α2α3 = 1), εi(f ) = αi. Hence
ε1 + ε2 + ε3 = 0.
The root system in Der(C, ·, n) is
Φ = {±εi, εi − εj : 1 ≤ i 6= j ≤ 3}
(the root system of type G2):
ε2 − ε1
ε2
ε1
ε3
Then ∆ = {ε1, ε2 − ε1} is a set of simple roots. The associated set of positive roots
is Φ+ = {ε1, ε2, −ε3, ε2 − ε1, ε2 − ε3, ε1 − ε3}.
ORDER 3 ELEMENTS IN G2 AND IDEMPOTENTS
7
With F = C (the field of complex numbers), a concrete Chevalley basis is com-
puted in [EK13, 4.4]:
xεi−εj = [Lui , Rvj ] = Eij ∈ sl3(F) ⊆ Der(C, ·, n), 1 ≤ i 6= j ≤ 3,
xεi = de1,ui ,
h1 = 2E11 − E22 − E33,
x−εi = −de2,vi ,
h2 = E22 − E11,
where Eij denotes the 3 × 3 matrix with 1 in the (i, j)-position and 0's elsewhere,
Lx (respectively Rx) denotes the left (resp. right) multiplication by x, and dx,y =
[Lx, Ly]+[Lx, Ry]+[Rx, Ry] = ad[x,y]· +3[Lx, Ry] (with adx(y) = [x, y]· = x·y −y ·x
for any x, y ∈ C).
The Z-span of the canonical basis is an admissible lattice, that is, it is invariant
under the action of xn
for each root α. Actually, for each long root α = εi − εj,
α
n!
x2
α = 0, and for each short root α = εi (resp. −εi), x3
α takes vi to 2ui
(resp. ui to 2vi), and kills all the other elements in the canonical basis. Hence, for
an indeterminate T , exp(T xα) is an automorphism of (CZ[T ], ·, n), where CZ is the
Z-span of the canonical basis, and CZ[T ] = CZ ⊗Z Z[T ].
α = 0, and x2
As usual, specializing T to elements in an arbitrary algebra R ∈ AlgF, it makes
sense to consider the induced automorphisms, denoted by exp(txα), t ∈ R, thus
obtaining group homomorphisms Ga → Aut(C, ·, n) for each root α, where Ga
denotes the additive group (G(R) is the additive group of R for any R ∈ AlgF).
Recall [Spr09, Lemma 7.3.3] that for each connected, reductive, algebraic group G
with a maximal split torus T, and for each root α relative to this torus, there is
an essentially unique homomorphism of algebraic groups uα : Ga → G, satisfying
some natural restrictions. Here uα is the map t 7→ exp(txα) above. If Uα denotes
the image of uα, then G is generated by T and the Uα's.
(see [Spr09, 8.1.1]).
Actually, over an algebraically closed field, the group of rational points of the Uα's
generate the group of rational points of G as an abstract group.
If the characteristic of F is 3, all these automorphisms exp(txα) (for t 6= 0) have
order 3 and will play an important role in Theorem 6.3.
Let B be the standard Borel subgroup associated to our set ∆ of simple roots.
Then B has dimension 8 and it is generated by T and the Uα's, with α ∈ Φ+, and
its unipotent radical U is generated by the Uα's (α ∈ Φ+). Also, for any simple root
γ ∈ ∆, we will use the corresponding parabolic subgroup P{γ}. Recall (see [Spr09,
§8.4]) that P{γ} is the semidirect product of the Levi subgroup L{γ}, generated
by T and U±γ, and its unipotent radical Ru(P{γ}), generated by the Uα's, with
α ∈ Φ+ \ {γ}. The derived subgroup [P{γ}, P{γ}] = [L{γ}, L{γ}] ⋉ Ru(P{γ}) is
eight-dimensional, with [L{γ}, L{γ}] being isomorphic to SL2 (see Remark 5.6).
4. Para-Cayley algebras as Petersson algebras
The order 3 automorphisms of a Cayley algebra whose associated Petersson
algebra is para-Cayley are of a very specific nature. These automorphisms were
known to Okubo1.
Lemma 4.1. Let w be an element of a Cayley algebra (C, ·, n) such that w·3 = 1.
Then the map x 7→ w · x · w·2 is an automorphism of (C, ·, n). It is the identity if
w ∈ F1, and its order is 3 otherwise.
1Private communication (June 2013)
8
A. ELDUQUE
Proof. For any x, y ∈ C,
(cid:0)w · x · w·2(cid:1) ·(cid:0)w · y · w·2(cid:1)
= w ·(cid:16)(x · w·2) · (w · y · w)(cid:17) · w
= w ·(cid:16)(cid:0)(cid:0)(x · w·2) · w(cid:1)·y(cid:1)·w(cid:17)·w
= w ·(cid:0)(x · y) · w(cid:1) · w = w · (x · y) · w·2,
(Middle Moufang Identity)
(Right Moufang Identity)
where the Moufang identities ([Sch95]p.28) have been used, together with the fact
that any two elements generate an associative subalgebra. This also shows that the
order of this automorphism is 1 if w ∈ F1 and 3 otherwise.
(cid:3)
Theorem 4.2. Let τ be an order 3 automorphism of a Cayley algebra (C, ·, n).
Then the Petersson algebra Cτ is para-Cayley if and only if there is an element
w ∈ C \ F1 with w·2 + w + 1 = 0 such that τ (x) = w · x · w·2 for any x ∈ C.
In this case, the element w is the para-unit of Cτ .
Proof. Let Cτ = (C, ∗, n), where x ∗ y = τ (¯x) · τ 2(¯y) for any x, y ∈ C. Assume first
that Cτ is para-Cayley, and let e be its para-unit. Since τ is also an automorphism
of Cτ , τ (e) = e. Also n(e) = 1 and e 6∈ F1 (otherwise τ would be the identity).
For any x ∈ C,
x ∗ e =(n(e, x)e − x,
τ (¯x) · τ 2(¯e) = τ (¯x) · ¯e.
because e is the para-unit,
Therefore
n(e, x)e·2 − x · e = (x ∗ e) · e = (τ (¯x) · ¯e) · e = τ (¯x),
and we get, for any x ∈ C,
τ (x) = n(e, ¯x)e·2 − ¯x · e.
In particular e = n(e, ¯e)e·2 − 1 and, since e·2 − n(e, 1)e + n(e)1 = 0 and 1 and e are
linearly independent, we conclude that n(e, 1) = −1 = n(e, ¯e), so e·2 + e + 1 = 0
and
τ (x) = n(e, ¯x)e·2 − ¯x · e = (e · x + ¯x · ¯e) · e·2 − ¯x · e = e · x · e·2,
for any x ∈ C.
Conversely, for w ∈ C \ F1 with w·2 + w + 1 = 0 such that τ (x) = w · x · w·2, and
with x ∗ y = τ (¯x) · τ 2(¯y), one gets
w ∗ x = τ ( ¯w) · τ 2(¯x) = ¯w · w·2 · ¯x · w
= w · ¯x · w =(cid:0)n(w, x)1 − x · ¯w(cid:1) · w = n(w, x)w − x,
and, in the same vein, x ∗ w = n(x, w)w − x = w ∗ x. Therefore, w is a para-unit
of Cτ , so that Cτ is para-Cayley.
(cid:3)
Remark 4.3. Let (C, ·, n) be a Cayley algebra and let w ∈ C \ F1 be an element
such that w·2 + w + 1 = 0. Denote by τw the order 3 automorphism x 7→ w · x · w·2.
• If char F 6= 3, then the subalgebra generated by w is alghwi = F1 + Fw,
which is isomorphic to the quadratic ´etale algebra F[X]/(X 2 + X + 1). For
any element x ∈ C orthogonal to alghwi,
τw(x) = w · x · w·2 = w · x · ¯w = x · ¯w·2 = x · w.
But n(1 − w) = 3 6= 0, so the right multiplication by 1 − w is a bijection,
and hence τw(x) 6= x. Therefore, the subalgebra Fix(τw) of the elements
fixed by τw coincides with alghwi = F1 + Fw.
If w is another element in C \ F1 with w·2 + w + 1 = 0, there there is
an isomorphism ϕ : alghwi → algh wi with ϕ(w) = w. By means of the
ORDER 3 ELEMENTS IN G2 AND IDEMPOTENTS
9
Cayley-Dickson doubling process, this isomorphism ϕ can be extended to
an automorphism of (C, ·, n), also denoted by ϕ. Then τ w = ϕ ◦ τw ◦ ϕ−1,
so that all these order 3 automorphisms are conjugate.
• If char F = 3, then 0 = w·2 + w + 1 = (w − 1)·2, so the nonzero element
u = w−1 satisfies u·2 = 0, and hence (C, ·, n) is the split Cayley algebra and,
by Lemma 2.4, there is a canonical basis with u = u1, so that w = 1 + u1,
w·2 = 1 − u1.
In particular, again all these order 3 automorphisms are
conjugate. Computing in this canonical basis one gets:
1 τw−id
τw−id
−−−−→ (1 + u1) · v1 · (1 − u1) − v1 = −e1 + e2 + u1
−−−−→ 0,
v1
τw−id
−−−−→ (1 + u1) · e1 · (1 − u1) − e1 = −u1
τw−id
−−−−→ 0,
u2
τw−id
−−−−→ (1 + u1) · u2 · (1 − u1) − u2 = −v3
τw−id
−−−−→ 0,
τw−id
−−−−→ (1 + u1) · u3 · (1 − u1) − u3 = v2
u3
Thus, the Segre symbol of the nilpotent linear map τw − id is (3, 22, 1).
τw−id
−−−−→ 0,
Corollary 4.4. Let (C, ·, n) be a Cayley algebra and (C, •, n) the associated para-
Cayley algebra (x • y = ¯x · ¯y for any x, y ∈ C). Then
{idempotents of (C, •, n)} = {1} ∪ {w ∈ C \ F1 : w·2 + w + 1 = 0},
and all the idempotents, with the exception of the para-unit 1, are conjugate under
Aut(C, ·, n) = Aut(C, •, n). In particular,
• If either char F 6= 3 and (C, •, n) contains a subalgebra isomorphic to the
quadratic ´etale algebra K = F[X]/(X 2 + X + 1), or char F = 3 and (C, ·, n)
is split, then (C, •, n) contains the para-unit and a unique conjugacy class
of other idempotents.
• Otherwise, the only idempotent of (C, •, n) is its para-unit.
Proof. If w ∈ C \ F1 satisfies w • w = w, then n(w) = 1 and w·2 = ¯w = n(1, w)1 − w.
As w·2 − n(1, w)w + n(w)1 = 0, it follows that n(1, w) = −1 and w·2 + w + 1 = 0.
Conversely, if w·2 + w + 1 = 0 for w ∈ C \ F1, then n(1, w) = −1 so w·2 = ¯w, and
w • w = w. Now the arguments in Remark 4.3 apply.
(cid:3)
We finish this section by looking at the (easier) situation in dimension 2 and 4.
(See also [EP96, Theorem 3.2].)
Proposition 4.5. Let τ be an order 3 automorphism of a quaternion algebra
(Q, ·, n). Then there exists an element w ∈ Q \ F1 with w·2 + w + 1 = 0 such
that τ (x) = w · x · w·2 for any x ∈ Q. All these automorphisms are conjugate.
1
Proof. By the Noether-Skolem Theorem, there is an invertible element a ∈ Q \ F1
such that τ (x) = a · x · a−1 and a·3 = λ1, λ ∈ F×. Then a·4 ∈ F×a, so with w =
n(a·2) a·4, τ (x) = w · x · w−1 and n(w) = 1. Besides, w·3 = 1
λ4 (λ1)·4 = 1.
Then w·2 = ¯w ·w·3 = ¯w and, as in the proof of Corollary 4.4, we get w·2 + w + 1 = 0.
The conjugation of these automorphisms follows as in Remark 4.3.
(cid:3)
n(a)6 a·12 = 1
Remark 4.6. Under the conditions of Proposition 4.5, and as in Theorem 4.2,
the Petersson algebra Qτ is a para-quaternion algebra with para-unit w. Since the
norm is the same n, Qτ is isomorphic to (Q, •, n).
Corollary 4.7. Let (Q, ·, n) be a quaternion algebra and (Q, •, n) the associated
para-quaternion algebra. Then
{idempotents of (Q, •, n)} = {1} ∪ {w ∈ Q \ F1 : w·2 + w + 1 = 0},
10
A. ELDUQUE
and all the idempotents, with the exception of the para-unit 1, are conjugate under
Aut(Q, ·, n) = Aut(Q, •, n). In particular,
• If either char F 6= 3 and (Q, •, n) contains a subalgebra isomorphic to the
quadratic ´etale algebra K = F[X]/(X 2 + X + 1), or char F = 3 and (Q, ·, n)
is split, then (Q, •, n) contains the para-unit and a unique conjugacy class
of other idempotents.
• Otherwise, the only idempotent of (Q, •, n) is its para-unit.
Finally, the next remark settles the situation for two-dimensional symmetric
composition algebras.
Remark 4.8. The automorphism group Aut(K, ·, n) for a two-dimensional Hurwitz
algebra (i.e., an ´etale quadratic algebra) is the cyclic group of order 2. In particular,
it does not contain order 3 automorphisms. Hence, if a two-dimensional symmetric
composition algebra (S, ∗, n) contains an idempotent, this idempotent is a para-unit,
so that (S, ∗, n) is a para-quadratic algebra. Moreover,
• If char F = 3, the para-unit is the only idempotent of a para-quadratic
algebra.
• If char F 6= 3, the only para-quadratic algebra containing other idempotents
is, up to isomorphism, the para-quadratic algebra (K, •, n) associated to the
quadratic ´etale algebra (K, ·, n), with K = F1 ⊕ Fw and w·2 + w + 1 = 0.
Then
{idempotents of (Q, •, n)} = {1, w, w·2},
and Aut(K, •, n) is the symmetric group S3 of degree 3 (see [KMRT98,
(34.5)]), which acts permuting the three idempotents. Actually, the affine
group scheme of automorphisms Aut(K, •, n) is the corresponding constant
group scheme, also denoted by S3.
5. Order 3 elements in G2 and idempotents, char F 6= 3
Over an algebraic closed field of characteristic 6= 3, any order 3 element of the
algebraic group of type G2 is semisimple and contained in a torus. By conjugacy
of the maximal tori, any such element τ is conjugate to an element in the torus
consisting of the diagonal matrices in the subgroup SL3(F) in (3.1). Therefore,
there is a primitive cubic root ω of 1 such that τ is conjugate to diag(ω, ω, ω) or
diag(1, ω, ω2) (changing ω by ω2 gives a conjugate element). This shows that there
are exactly two conjugacy classes of such elements. The same can be deduced from
the description of finite order automorphisms in [Kac90, Ch. VIII] in characteristic
0 and its extension to the modular case by Serre [Ser06] (see also [PL15, Theorem
6.3]).
Over arbitrary fields of characteristic 6= 3, the situation is not much worse.
Theorem 5.1. Let (C, ·, n) be a Cayley algebra over a field F of characteristic not
3, and let τ be an order 3 automorphism of (C, ·, n). Then one of the following
conditions holds:
(1) There is an element w ∈ C \ F1 with w·2 + w + 1 = 0 such that
τ (x) = w · x · w2
for any x ∈ C. In this case, the Petersson algebra Cτ is para-Cayley with
para unit w, and the subalgebra Fix(τ ) of the elements fixed by τ is F1 + Fw
(a quadratic ´etale subalgebra). Any two such automorphisms are conjugate
in Aut(C, ·, n).
ORDER 3 ELEMENTS IN G2 AND IDEMPOTENTS
11
(2) The subalgebra of fixed elements by τ is a quaternion subalgebra of C con-
taining an element w ∈ C \ F1 such that w·2 + w + 1 = 0. In this case the
Petersson algebra Cτ is an Okubo algebra. Any two such automorphisms
are conjugate in Aut(C, ·, n) if and only if the corresponding quaternion sub-
algebras are isomorphic.
In particular, if F contains the cubic roots of 1, then C is the split Cayley
algebra, Cτ is the split Okubo algebra, and any two such automorphisms are
conjugate.
Proof. By [EP96, Proposition 3.4 and Theorem 3.5], either Fix(τ ) is a quadratic
´etale subalgebra and Cτ is para-Cayley, or Fix(τ ) is a quaternion subalgebra.
In the first case, Theorem 4.2 and Remark 4.3 give us the desired result.
In
the second case, Q = Fix(τ ) is a quaternion subalgebra and C = Q ⊕ Q · u for a
nonisotropic element u orthogonal to Q. Then τ (u) = w · u for a norm 1 element
w ∈ Q such that w·3 = 1 6= w. Then w·2 = ¯w · w·3 = ¯w = n(1, w)1 − w and, as in
the proof of Proposition 4.5, we have w·2 + w + 1 = 0.
If τ ′ is another such automorphism and the quaternion subalgebra Q′ = Fix(τ ′)
is isomorphic to Q, then the orthogonal complements Q⊥ and (Q′)⊥ are isometric,
so there is an element u′ ∈ (Q′)⊥ with n(u′) = n(u). As before, τ ′(u′) = w′ · u′ with
w′ ∈ Q′ such that (w′)·2 + w′ + 1 = 0. The isomorphism ϕ : F1 ⊕ Fw → F1 ⊕ Fw′
that takes w to w′ can be extended to an isomorphism ϕ : Q → Q′, and then to
an automorphism ϕ of (C, ·, n), also denoted by ϕ, with ϕ(u) = u′. It follows that
τ ′ = ϕ ◦ τ ◦ ϕ−1, as required.
If F contains the cubic roots of 1, then the subalgebra F1 ⊕ Fw is isomorphic to
the split quadratic algebra F × F, so (C, ·, n) is the split Cayley algebra, and the
uniqueness up to conjugacy of τ shows that there is a canonical basis such that
τ (ui) = ui+1 (indices modulo 3), so that Cτ is the split Okubo algebra in this case.
If F does not contain the cubic roots of 1, this shows that Cτ is an Okubo algebra,
as it is so over a field extension.
(cid:3)
Example 5.2. Let (C, ·, n) be the split Cayley algebra over the field of real numbers
R. Then C contains subalgebras isomorphic to both the real quaternion division
algebra H and the split real quaternion algebra M2(R). Both of them contain copies
of C and hence contain an element w 6∈ R1 with w·2 + w + 1 = 0. Therefore, there
are two conjugacy classes of order 3 automorphisms of (C, ·, n) such that Cτ is an
Okubo algebra.
The centralizers in the affine group scheme of automorphisms Aut(C, ·, n) of the
automorphisms τ in Theorem 5.1 will be computed now. Because of Proposition 2.6,
these centralizers coincide with the stabilizers of the idempotent 1 of the symmetric
composition algebra (Petersson algebra) Cτ .
Some preliminaries are needed first.
Let K be a quadratic ´etale subalgebra of a Cayley algebra (C, ·, n) over a field F.
Then (see [Jac58, EM95]), the orthogonal subspace W = K⊥ is a free left K-module
of rank 3 endowed with
• a hermitian nondegenerate form σ : W × W → K (i.e., σ is F-bilinear,
σ(a · x, y) = a · σ(x, y) and σ(y, x) = σ(x, y), for any a ∈ K and x, y ∈ W,
and σ induces a K-linear isomorphism W → HomK(W, K), x 7→ σ(., x)),
and
• an anticommutative product W × W → W, (x, y) 7→ x × y, with (a · x) × y =
¯a · (x × y) = x × (a · y) for any a ∈ K and x, y ∈ W,
such that, for any x, y ∈ W,
x · y = −σ(x, y) + x × y.
(5.1)
12
A. ELDUQUE
The K-trilinear form Φ : W × W × W → K, given by
Φ(x, y, z) = σ(x, y × z),
is alternating, and satisfies
for any x1, x2, x3 ∈ W.
n(cid:0)Φ(x1, x2, x3)(cid:1) = det(cid:16)σ(xi, xj)(cid:17)
(5.2)
(5.3)
Conversely, if (K, ·) is a quadratic ´etale algebra with norm n and standard con-
jugation a 7→ ¯a, and W is a free left K-module of rank 3 endowed with:
• a nondegenerate hermitian form σ : W × W → K,
• a K-trilinear alternating form Φ : W × W × W → K satisfying (5.3),
then the vector space direct sum C = K ⊕ W is a Cayley algebra, where the multi-
plication · and the norm n are defined as follows:
• K is a subalgebra of C and for any a ∈ K and w ∈ W, a · w is given by
the product of the scalar a on the element w in the K-module W, while
w · a := ¯a · w.
• x · y = −σ(x, y) + x × y for any x, y ∈ W, where x × y is defined by the
equation σ(z, x × y) = Φ(z, x, y) for any z ∈ W.
• K and W are orthogonal for n, the restriction of n to K is the generic norm
of K, and n(x) = σ(x, x) for any x ∈ W.
Assume now that we are in the situation of Theorem 5.1.(1), so char F 6= 3, and
τ ∈ Aut(C, ·, n) is given by τ (x) = w · x · w·2, with w ∈ C \ F1, w·2 + w + 1 = 0.
Then K = F1 + Fw is a quadratic ´etale subalgebra of (C, ·, n). Hence W = K⊥ is
endowed with the hermitian form σ and product × in (5.1).
Lemma 5.3. Under the conditions above, StabAut(C,·,n)(w) is naturally isomor-
phic to the special unitary group SU(W, σ).
Proof. Any K-linear map of W of determinant 1 preserves the alternating K-
trilinear form Φ in (5.2) and hence, if it preserves σ, it preserves also × (see [Jac58]
if char F 6= 2, but the arguments are valid too if char F = 2), so it extends to an
automorphism of (C, ·, n) which restricts to the identity map on K. Conversely, any
automorphism that fixes w restricts to a K-linear map that preserves σ and ×, and
hence Φ, so it lies in the special unitary group. Besides, all this is functorial, so the
result is valid at the level of schemes.
(cid:3)
Theorem 5.4. Let (C, ·, n) be a Cayley algebra over a field F of characteristic 6= 3,
and let w ∈ C \ F1 with w·2 + w + 1 = 0. Let τ be the order 3 automorphism given
by τ (x) = w · x · w·2 for any x ∈ C. Then,
CentAut(C,·,n)(τ ) = StabAut(C,·,n)(w)(cid:0)≃ SU(W, σ)(cid:1).
Proof. Any ϕ in CentAut(C,·,n)(τ ) stabilizes K = Fix(τ ) so (locally) we have either
ϕ(w) = w or ϕ(w) = w·2, because the group scheme Aut(K, ·) is the constant
group scheme C2 (cyclic group of order 2). But ϕ(w) · ϕ(x) · ϕ(w)·2 = w · ϕ(x) · w·2
for any x ∈ C ⊗F R (if ϕ is a point over R), so w−1 · ϕ(w) lies in the commutative
center of (C, ·), which is F1. Hence ϕ(w) = w.
(cid:3)
For automorphisms in Theorem 5.1.(2), consider first a Cayley algebra (C, ·, n)
over an arbitrary field, and let Q be a quaternion subalgebra. Then C = Q ⊕ Q · u for
some nonisotropic element u orthogonal to Q. Consider the group scheme SL1(Q),
whose set of points over an algebra R ∈ AlgF is SL1(Q)(R) = {q ∈ Q ⊗F R : n(q) =
1}. The scheme µ2 of square roots of unity embeds naturally in SL1(Q) as the
scalar elements.
ORDER 3 ELEMENTS IN G2 AND IDEMPOTENTS
13
Lemma 5.5. Under the conditions above, the stabilizer StabAut(C,·,n)(Q) is iso-
morphic to SL1(Q) × SL1(Q)/µ2 (where µ2 embeds diagonally on the product of the
two copies of SL1(Q)).
Proof. For any q ∈ Q of norm 1, the maps ψq, ϕq : C → C given by
ψq(x) = q · x · q−1 = q · x · ¯q,
ϕq(x) = x,
ψq(x · u) = (x · q−1) · u,
ϕq(x · u) = (q · x) · u,
for x ∈ Q, are automorphisms of (C, ·, n). The same happens for C ⊗F R and
q ∈ Q ⊗F R of norm 1, for any algebra R ∈ AlgF.
Moreover, ψq and ϕp commute for any q, p ∈ SL1(Q). Therefore, we have the
natural homomorphism
Φ : SL1(Q) × SL1(Q) −→ StabAut(C,·,n)(Q)
(q, p)
7→ ψq ◦ ϕp.
(5.4)
Its kernel consists of the pairs (q, p) such that ψq ◦ψp = id, and this forces ψqQ = id,
so q ∈ µ2. Then u = ψq ◦ϕp(u) = (p·q−1)·u, so p = q. Thus the kernel is µ2 embed-
ded diagonally. This shows, in particular, that the dimension of StabAut(C,·,n)(Q)
is at least 6.
Finally, Φ is a quotient map as Φ¯F is surjective (this follows easily using, for
instance, [SV00, §2.1]) and StabAut(C,·,n)(Q) is smooth, because its Lie algebra
s = {δ ∈ Der(C, ·, n) : δ(Q) ⊆ Q} has dimension 6. To check this, consider the
homomorphism
s −→ Der(Q, ·) ≃ ad(Q) ≃ Q/F1
δ 7→ δQ,
whose kernel consists of those derivations δ ∈ Der(C, ·, n) such that δ(Q) = 0. But
any such derivation is determined by δ(u), which belongs to {x ∈ Q·u : n(x, u) = 0},
whose dimension is 3.
(cid:3)
Remark 5.6. If {e1, e2, u1, u2, u3, v1, v2, v3} is a canonical basis of the split Cayley
algebra (C, ·, n) over an arbitrary field F, let Q be the quaternion subalgebra spanned
by e1, e2, u1, and v1. The group SL2 ≃ SL1(Q) embeds in Aut(C, ·, n) both by
means of q 7→ ψq and by q 7→ ϕq. With the notation in Section 3, the image is the
derived subgroup of the Levi subgroup L{ε1} in the first case, and of L{ε2−ε3} in
the second case.
Theorem 5.7. Let (C, ·, n) be a Cayley algebra over a field F of characteristic
6= 3, and let τ be an order 3 automorphism of (C, ·, n) that fixes elementwise a
quaternion subalgebra Q. Let u ∈ Q⊥ be a nonisotropic element and let w ∈ Q \ F1
with w2 + w + 1 = 0 such that τ (u) = w · u. Denote by K the quadratic ´etale
subalgebra K = F1 + Fw. Then the centralizer CentAut(C,·,n)(τ ) is isomorphic to
SL1(Q) × SL1(K)/µ2.
Proof. As Fix(τ ) = Q, it follows that the centralizer CentAut(C,·,n)(τ ) is contained
in StabAut(C,·,n)(Q). The homomorphism Φ in (5.4) restricts to a homomorphism
SL1(Q) × SL1(K) → CentAut(C,·,n)(τ ).
(Observe that ψq ◦ τ = τ ◦ ψq for any
q ∈ SL1(Q), but ϕp ◦ τ = τ ◦ ϕp if and only if p · w = w · p, if and only if p ∈ K
(and the same over any R). Now the same arguments as in the proof of Lemma 5.5
apply.
(cid:3)
Due to the relationship between idempotents in symmetric composition algebras
and automorphisms of Cayley algebras of order 1 or 3, we immediately get the
next consequence of Theorem 5.1. Here K will denote the quadratic ´etale algebra
F[X]/(X 2 + X + 1), and K the corresponding para-quadratic algebra.
14
A. ELDUQUE
Corollary 5.8. An Okubo algebra (O, ∗, n) over a field F of characteristic 6= 3
contains an idempotent if and only if it contains a subalgebra isomorphic to K.
In this case, the centralizer of any idempotent e is a para-quaternion subalgebra
containing a subalgebra isomorphic to K, and two idempotents are conjugate if and
only if the corresponding quaternion algebras are isomorphic.
In particular, if F contains the cubic roots of 1, then (O, ∗, n) contains a unique
conjugacy class of idempotents.
Proof. Let e be an idempotent of the Okubo algebra (O, ∗, n), and let τe ∈ Aut(O, ∗, n)
be the order 3 automorphism given by τe(x) = e ∗ (e ∗ x) for any x ∈ O as in (2.4).
Consider the Cayley algebra (O, ·, n), with unity e, where x · y = (e ∗ x) ∗ (y ∗ e), as
in (2.5). Then τe is an automorphism of (O, ·, n) too. By Theorem 5.1, Fix(τe) =
Cent(O,∗,n)(e) is a quaternion subalgebra of (O, ·, n) (and a para-quaternion subal-
gebra of (O, ∗, n)) containing a subalgebra isomorphic to K, so that (O, ∗, n) con-
tains a subalgebra isomorphic to K. Moreover, if f is another idempotent, τf the
corresponding order 3 automorphism, and (O, ⋄, n) the associated Cayley algebra
(x⋄ y = (f ∗ x)∗ (y ∗ f )), then both Cayley algebras (O, ·, n) and (O, ⋄, n) are isomor-
phic, as their norms are isometric. If the para-quaternion algebras Cent(O,∗,n)(e)
and Cent(O,∗,n)(f ) are isomorphic, then by Theorem 5.1 there is an isomorphism
ϕ : (O, ·, n) → (O, ⋄, n) such that ϕ ◦ τe = τf ◦ ϕ. Note that ϕ(e) = f because e
(respectively f ) is the unity of (O, ·, n) (resp., of (O, ⋄, n)). Then, for any x, y ∈ O,
ϕ(x ∗ y) = ϕ(cid:0)τe(¯x) · τ 2
e (¯y)(cid:1) = τf (ϕ(x)) ⋄ τ 2
f (ϕ(y)) = ϕ(x) ∗ ϕ(y),
and ϕ ∈ Aut(O, ∗, n) too, with ϕ(e) = f .
(cid:3)
Up to isomorphism there is a unique Cayley algebra with isotropic norm: the
Zorn matrix algebra (or split Cayley algebra). However, the split Okubo algebra
(Definition 2.8) is not characterized by its norm being isotropic. If the characteristic
of F is not 3 and F contains the cubic roots of 1, or if the characteristic of F is 3,
then the norm of any Okubo algebra is isotropic (see [EM93, Proposition 7.3] and
[Eld97]).
Theorem 5.9. Let (O, ∗, n) be an Okubo algebra over a field F of characteristic
6= 3. Then (O, ∗, n) is split (i.e., isomorphic to the split Okubo algebra in Definition
2.8) if and only if n is isotropic and (O, ∗, n) contains an idempotent.
Proof. The split Okubo algebra is the Petersson algebra Cτst = (C, ∗, n), where
(C, ·, n) is Zorn matrix algebra, τst is defined in (2.7), and
x ∗ y = τst(¯x) · τ 2
st(¯y)
for any x, y ∈ C. The norm n is isotropic and the unity 1 of (C, ·, n) is an idempotent
of Cτst .
Conversely, let (O, ∗, n) be an Okubo algebra with isotropic norm n and let
0 6= e = e ∗ e be an idempotent. Let τe ∈ Aut(O, ∗, n) as in (2.4), and consider the
Cayley algebra (O, ·, n) with unity e, where x · y = (e ∗ x) ∗ (y ∗ e) as in (2.5). Then
x ∗ y = τe(¯x) · τ 2
e (¯y), where ¯x = n(e, x)e − x is the 'conjugation' relative to e. By
Theorem 5.1, Q = Fix(τe) is a quaternion subalgebra of (O, ·, n) and O = Q ⊕ Q · u
for a nonisotropic element u ∈ O orthogonal to Q. Besides, τe(u) = w · u for a norm
1 element w ∈ Q such that w·3 = 1 6= w. If the restriction of n to Q is isotropic,
then (Q, ·, n) is isomorphic to M2(F), so the restriction nQ is universal. Hence,
by multiplying by a suitable q ∈ Q, we may asume n(u) = 1 (this will change
the element w). Then the multiplication in (O, ∗, n) is completely determined (see
[CEKT13, Example 9.4]) so, up to isomorphism, (O, ∗, n) is the split Okubo algebra.
If the restriction nQ is anisotropic, then since n itself is isotropic, there are
nonzero elements q1, q2 ∈ Q such that n(q1 +q2 ·u) = 0, so that n(cid:0)q−1
1 ·(q2 ·u)(cid:1) = −1.
ORDER 3 ELEMENTS IN G2 AND IDEMPOTENTS
15
Replacing u by q−1
element w). Note then that the element f = w·2 is an idempotent:
· (q2 · u) we may assume n(u) = −1 (again, this will change the
1
w·2 ∗ w·2 = τe(cid:16)w·2(cid:17) · τ 2
e (cid:16)w·2(cid:17) = w·2 · w·2 = w · w = w·2,
and
f ∗ (1 + u) = w · (1 − w·2 · u) = w − u,
(1 + u) ∗ f = (1 − w · u) · w = w − w · u · w = w − w·2 · (w − u) = w − u.
Hence if τf denotes the corresponding order 3 automorphism: τf (x) = f ∗ (f ∗ x),
and we consider the Cayley algebra (O, ⋄, n) with x ⋄ y = (f ∗ x) ∗ (y ∗ f ), the fixed
subalgebra Q = {x ∈ O : f ∗ x = x ∗ f } is a quaternion subalgebra of (O, ⋄, n) with
isotropic norm, because 1 + u ∈ Q and n(1 + u) = 1 − 1 = 0. By the arguments
above, (O, ∗, n) is the split Okubo algebra.
(cid:3)
Example 5.10. According to Example 5.2, there are two different conjugacy
classes of order 3 automorphisms τ of the split Cayley algebra (C, ·, n) over R such
that Cτ is an Okubo algebra. By Theorem 5.9, Cτ is always the split Okubo algebra
and hence, by Corollary 5.8, the real split Okubo algebra contains two conjugacy
classes of idempotents.
6. Order 3 elements in G2, char F = 3
Over a field F of characteristic 3, the situation for order 3 elements in the auto-
morphism group of a Cayley algebra is quite different. To begin with, only the split
Cayley algebra admits order 3 automorphisms (Theorem 6.3 below). In this case
the group of automorphisms is the Chevalley group of type G2, and given any split
maximal torus and any root, all the elements exp(δ), where δ is a nonzero element
in a root space (these elements make sense, as shown in Section 3), have order 3 as
its cube is exp(3δ) = exp(0) = id.
Remark 6.1. Over fields of characteristic 3, the Lie algebra Der(C, ·, n) of deriva-
tions of a Cayley algebra is not simple, as it contains the simple seven-dimensional
ideal adC (ad1 = 0), which is a twisted form of psl3. Moreover, the quotient
Der(C, ·, n)/adC is isomorphic to adC. (See, for example, [AEMN02] or [AE16].)
The next definition extends [CEKT13, Definition 9.2] (see also [Eld15, Definition
22]).
Definition 6.2. Let f be an idempotent of the Okubo algebra (O, ∗, n) (char F =
3). Then f is said to be:
• quaternionic, if its centralizer contains a para-quaternion algebra,
• quadratic, if its centralizer contains a para-quadratic algebra and no para-
quaternion subalgebra,
• singular, otherwise.
It is proved in [CEKT13, Proposition 9.9 and Theorem 9.13] that only the split
Okubo algebra contains a quaternionic idempotent, and there is only one such
idempotent.
The next result extends [Eld15, Lemma 21]. The Peirce component U in Remark
2.3 generates the split Cayley algebra. Hence any automorphism is determined by
its action on U.
Theorem 6.3. Let (C, ·, n) be a Cayley algebra over a field F of characteristic 3,
and let τ be an order 3 automorphism of (C, ·, n). Then (C, ·, n) is the split Cayley
algebra and one of the following conditions holds:
16
A. ELDUQUE
(1) (τ − id)2 = 0. In this case, the Petersson algebra Cτ is the split Okubo alge-
bra, 1 is its (unique) quaternionic idempotent, and there exists a canonical
basis of C such that
τ (ui) = ui, i = 1, 2,
τ (u3) = u3 + u2.
In other words, τ = exp(δ), where δ = [Lu2, Rv3 ], which is a derivation in
the root space Der(C, ·, n)ε2−ε3. (ε2 − ε3 is a long root.)
In particular, up to conjugacy, there is a unique such automorphism τ .
The Segre symbol of the nilpotent endomorphism τ − id is (22, 14).
(2) (τ − id)2 6= 0 and there is a quadratic ´etale subalgebra K of C fixed ele-
mentwise by τ . In this case, the Petersson algebra Cτ is an Okubo algebra,
1 is a quadratic idempotent of Cτ , and the Segre symbol of the nilpotent
endomorphism τ − id is (32, 12).
If F is algebraically closed, then there is a canonical basis of C such that
τ (ui) = ui+1
(indices modulo 3)
so τ fixes e1 and e2, and it belongs to the normalizer of the associated
maximal torus. (Note that τ is the automorphism τst in (2.7).)
(3) There is a canonical basis such that
τ (ui) = ui, i = 1, 2,
τ (u3) = u3 + v3 − (e1 − e2).
In this case, τ (x) = w · x · w·2, with w = 1 − v3. Hence (Theorem 4.2)
the Petersson algebra Cτ is a para-Cayley algebra with para-unit w. Also,
τ = exp(δ), where δ = −adv3 : x 7→ −[v3, x] = x · v3 − v3 · x is a derivation
in the root space Der(C, ·, n)−ε3 . (−ε3 is a short root.)
In particular, up to conjugacy, there is a unique such automorphism τ .
The Segre symbol of the nilpotent endomorphism τ − id is (3, 22, 1).
(4) There is a canonical basis such that
τ (ui) = ui, i = 1, 2,
τ (u3) = u3 + u2 + v3 − (e1 − e2).
In this case, the Petersson algebra Cτ is the split Okubo algebra, and 1 is
a singular idempotent of Cτ . Such an automorphism τ is unique, up to
conjugacy, and it is the composition of the (commuting) automorphisms in
items (1) and (3). The Segre symbol of the nilpotent endomorphism τ − id
is again (3, 22, 1).
Proof. Fix(τ )⊥ is a subspace invariant under τ and, since τ − id is nilpotent ((τ −
id)3 = 0), Fix(τ ) ∩ Fix(τ )⊥ is a nonzero isotropic subspace. Hence (C, ·, n) is the
split Cayley algebra.
As in [EP96, Lemma 21] either:
(i) (τ −id)2 = 0, in which case there is a quaternion subalgebra Q of C contained
in Fix(τ ), or
(ii) (τ − id)2 6= 0 and there is a quadratic ´etale subalgebra K in Fix(τ ), or
(iii) (τ −id)2 6= 0, Fix(τ ) = F1⊕(cid:0)Fix(τ )∩C0(cid:1), where C0 = (F1)⊥, and Fix(τ )∩C0
is a maximal isotropic subspace (hence of dimension 3) of C0.
In case (i), for any nonisotropic u ∈ Q⊥, C = Q ⊕ Q · u, and τ (u) = w · u, for
a norm 1 element w ∈ Q such that w·3 = 1. Then w·2 = ¯w = w−1 and, as in the
proof of 4.2, we get w·2 + w + 1 = 0. That is, (w − 1)·2 = 0 since the characteristic
is 3. We conclude that Q is split: Q ≃ M2(F). But then the norm is universal in Q,
and we may take u ∈ Q⊥ with n(u) = 1. Also, there is an isomorphism Q ≃ M2(F)
that takes w to ( 1 1
0 1 ). Then there is a canonical basis with Q = span {e1, e2, u1, v1},
ORDER 3 ELEMENTS IN G2 AND IDEMPOTENTS
17
w = 1 + u1, and u = u2 + v2. Therefore, we get
τ (u1) = u1,
τ (v1) = v1,
τ (u2) = τ (e1 · u) = e1 ·(cid:0)(1 + u1) · (u2 + v2)(cid:1)
= e1 · (u2 + v2 − v3) = u2,
τ (v2) = τ (e2 · u) = e2 · (u2 + v2 − v3) = v2 − v3,
τ (u3) = τ (v1 · v2) = v1 · (v2 − v3) = u3 + u2,
τ (v3) = τ (u1 · u2) = v3,
Hence Fix(τ ) = Q + {x ∈ Q : w · x = x} · u = span {e1, e2, u1, v1, u2, v3} and we
obtain the first possibility in the Theorem.
Case (ii) above corresponds to the second possibility in the Theorem, which will
be developed in Theorem 6.7.
We are left with case (iii) above, so Fix(τ ) = F1 ⊕ W, where W is a three-
dimensional isotropic subspace of C0. Let W′ be another three-dimensional isotropic
subspace of C0 paired with W (that is, the bilinear map W × W′ → F, (x, y) 7→
n(x, y), is nondegenerate). Then C0 = W ⊕ W′ ⊕ Fa, with Fa = (W ⊕ W′)⊥ ∩ C0
(nonisotropic). By Witt's Cancellation Theorem, the restriction of n to F1 ⊕ Fa
is hyperbolic, so −n(a) ∈ (F×)2, and we may scale a so that n(a) = −1, that is,
a·2 = 1.
For any x ∈ C0 and y ∈ W, since τ (y) = y we have
n(cid:0)(τ − id)(x), y(cid:1) = n(cid:0)τ (x), y(cid:1) − n(x, y) = n(cid:0)τ (x), τ (y)(cid:1) − n(x, y) = 0,
so (τ − id)(C0) ⊆ W⊥ ∩ C0 = W ⊕ Fa. Then 0 6= (τ − id)2(C0) ⊆ F(τ − id)(a), and
it follows that the Segre symbol of (τ − id)C0 is (3, 22), so that the Segre symbol
of τ − id is (3, 22, 1). Moreover, (τ − id)(C0) = W ⊕ Fa.
Now, τ (a) = a + w for some 0 6= w ∈ W. Also,
n(a · W, W) = n(a, W·2) ⊆ n(a, W) = 0, n(a · W, a) = n(a·2, W) = n(1, W) = 0,
so we obtain
Moreover, for any x ∈ W,
a · W = W · a ⊆ W.
w · x = τ (a) · x − a · x = τ (a · x) − a · x = 0
because a · x is in W ⊆ Fix(τ ). Hence
w · W = 0 = W · w.
2 (1 − a) and e2 = 1
The elements e1 = 1
2 (1 + a) are orthogonal idempotents, and let
C = Fe1 ⊕ Fe2 ⊕ U ⊕ V be the corresponding Peirce decomposition: U = e1 · C · e2,
V = e2 · C · e1. Then W is invariant under multiplication by e1 and e2, so that
W = (W ∩ U) ⊕ (W ∩ V). Besides, neither F1 ⊕ U nor F1 ⊕ V are subalgebras, so
W ∩ U 6= W 6= W ∩ V. Changing a by −a if necessary (i.e., interchanging e1 and e2),
we may (and will) assume dimF(W ∩ U) = 2, dimF(W ∩ V) = 1. Since w · W = 0, it
follows that Fw = W ∩ V and hence a · w = (e1 − e2) · w = −w.
The subspace W′ may now be chosen so that W′ = (W′ ∩ U) ⊕ (W′ ∩ V), with
dimF(W′ ∩ U) = 1 and dimF(W′ ∩ V) = 2. Take the element w′ ∈ W′ ∩ U such that
n(w′, w) = 1. Then (τ − id)(w′) ∈ W ⊕ Fa, so
τ (w′) = w′ + αa + u + βw
18
A. ELDUQUE
for some α, β ∈ F and u ∈ W ∩ U. From τ (a) = a + w we deduce:
τ (e1) = τ(cid:18) 1
τ (e2) = τ(cid:18) 1
2
2
(1 + a)(cid:19) =
(1 − a)(cid:19) =
1
2
1
2
(1 + a + w) = e1 − w,
(1 − a − w) = e2 + w.
Then we get
e1 − w = τ (e1) = −τ (w′ · w) = −(w′ + αa + u + w) · w = −w′ · w − αa · w = e1 + αw,
where we have used that W · w = 0. Hence α = −1. Also,
so β = 1 and, therefore,
0 = n(w′) = n(cid:0)τ (w′)(cid:1) = n(a) + βn(w′, w) = −1 + β,
τ (w′) = w′ − a + u + w.
We are left with two cases:
• If u = 0, we get
(1 − w) · w′ · (1 + w) = (w′ + e2) · (1 + w) = w′ − e1 + e2 + w = τ (w′),
and for any x ∈ W ∩ U, (1 − w) · x · (1 + w) = x (recall that W · w =
0 = w · W). Since U generates C, it follows that τ is the automorphism
x 7→ (1 − w) · x · (1 − w)·2. (Note that the element c = 1 − w satisfies
c·2 + c + 1 = 0.)
Hence Cτ is para-Cayley (Theorem 4.2). Let u3 = w′ and take u1, u2 ∈
W∩ U with n(u1 ·u2, u3) = 1. In this situation, with vi = ui+1 ·ui+2 (indices
modulo 3), {e1, e2, u1, u2, u3, v1, v2, v3} is a canonical basis of (C, ·, n). Then
w = v3, because n(ui, w) = 0 for i = 1, 2 as W is isotropic, and n(u3, w) = 1;
and
τ (ui) = ui, i = 1, 2,
τ (u3) = u3 − (e1 − e2) + v3,
thus obtaining the third possibility of the Theorem.
• If u 6= 0, let u3 = w′, u2 = u and take u1 ∈ W ∩ U with n(u1 · u2, u3) = 1.
Then w = v3 again, and we obtain:
τ (ui) = ui, i = 1, 2,
τ (u3) = u3 − (e1 − e2) + u2 + v3.
It remains to prove that the Petersson algebra Cτ = (C, ∗, n) is the split
Okubo algebra. The element e = 1 − v3 satisfies:
-- e ∗ e = τ (¯e) · τ 2(¯e) = (1 + v3)·2 = 1 − v3 = e, so e is an idempotent of
Cτ .
-- For any x ∈ W, x ∗ e = τ (¯x) · τ 2(¯e) = −x · (1 + v3) = −x, as W · w =
W · v3 = 0, so e ∗ x = −e ∗ (x ∗ e) = −x too.
-- a∗e = τ (e1 − e2)·τ 2(¯e) = −(e1 −e2 +v3)·(1−v3) = −a−v3 +v3 = −a,
so that e ∗ a = −e ∗ (a ∗ e) = −a too.
-- τ (v1) = τ (u2 · u3) = u2 · τ (u3) = v1 + u2, so v1 ∗ e = τ (¯v1) · τ 2(¯e) =
−(v1 + u2) · (1 − v3) = −v1 − u2 + u2 = −v1, so that e ∗ v1 = −v1 too
as before.
Hence span {e1, e2, u1, u2, v1, v3} is contained in the centralizer in Cτ of e.
But τ (v2) = τ (u3 · u1) = τ (u3) · u1 = v2 − u1 − v3 and v2 ∗ e = τ (¯v2) · τ 2(¯e) =
−(v2 − u1 − v3) · (1 − v3) = −v2 + u1 + u1 + v3 = −v2 − u1 + v3, and hence e
is not a para-unit of Cτ . Consider the Cayley algebra (C, ⋄, n) with x ⋄ y =
(e ∗ x)∗ (y ∗ e), whose unity is e. Then τ ′ : x 7→ e ∗ (e ∗ x) is an automorphism
of (C, ⋄, n) with (τ ′)3 = id, and τ ′
6= id because e is not a para-unit of
Cτ = (C, ∗, n). As Fix(τ ′), which is the centralizer in Cτ of e, has dimension
at least 6, τ ′ is necessarily an order 3 element of Aut(C, ⋄, n) as in the first
ORDER 3 ELEMENTS IN G2 AND IDEMPOTENTS
19
item of the Theorem, and hence e is the quaternionic idempotent of the
split Okubo algebra Cτ .
(cid:3)
Remark 6.4. Let F be an algebraically closed field of characteristic 3. In [Ste68,
§5] it is shown that the automorphisms in items (1) and (3) of Theorem 6.3 generate
the Chevalley group of type G2, which is the whole group Aut(C, ·, n). In particular,
Aut(C, ·, n) is generated (as an abstract group) by its order 3 elements.
Corollary 6.5. If an Okubo algebra (O, ∗, n) contains a singular idempotent, then
it is split.
Example 6.6. Let K be a quadratic ´etale algebra over a field F of characteristic
3, a an element of K of generic norm 1, W a free left K-module. Let {w1, w2, w3}
be a K-basis of W and let σ : W × W → K be the nondegenerate hermitian form
such that
σ(wi, wi) = 0,
σ(wi, wj ) = −1
for 1 ≤ i 6= j ≤ 3.
Finally, let Φ : W × W × W → K be the alternating K-trilinear form such that
Φ(w1, w2, w3) = a.
char F = 3), we get
j=1 aij · wj, for some elements aij ∈ K. Then,
For any x1, x2, x3 ∈ W, xi = P3
since Φ(w1, w2, w3) = a satisfies n(a) = 1, and det(cid:0)σ(wi, wj )(cid:1) = 1 (because
n(cid:0)Φ(x1, x2, x3)(cid:1) = n(cid:0)det(aij )Φ(w1, w2, w3)(cid:1)
= det(aij )det(aij)n(cid:0)Φ(w1, w2, w3)(cid:1) = n(cid:0)det(aij)(cid:1),
det(cid:0)σ(xi, xj )(cid:1) = det(cid:16)(aij)(cid:0)σ(wi, wj )(cid:1)(¯aji)(cid:17) = det(aij) det(¯aji) = n(cid:0)det(aij )(cid:1).
Therefore, (5.3) is satisfied and hence C = K ⊕ W is a Cayley algebra as in the
paragraph previous to Lemma 5.3. Actually, it is the split Cayley algebra, as
n(wi) = σ(wi, wi) = 0.
The K-linear map on W permuting cyclically w1 7→ w2 7→ w3 7→ w1 preserves σ
and Φ, and hence we obtain an order 3 automorphism τ of C determined by
τ K = id,
τ (wi) = wi+1 (indices modulo 3).
This automorphism will be denoted by τK,a.
(cid:3)
Case (2) of Theorem 6.3 can be settled now.
Theorem 6.7. Let τ be an order 3 automorphism of a Cayley algebra (C, ·, n)
over a field F of characteristic 3 such that (τ − id)2 6= 0 and τ fixes elementwise a
quadratic ´etale subalgebra. Then the Petersson algebra Cτ is an Okubo algebra and
τ is conjugate to an automorphism τK,a as in Example 6.6.
Moreover, two such automorphisms τK,a and τK′,a′ are conjugate if and only if
there is an isomorphism ϕ : K → K′ such that ϕ(a) ∈ (K′)·3 · a′ (i.e., there is an
element b′ ∈ K′, necessarily of norm 1, such that ϕ(a) = (b′)·3 · a′).
Proof. Let K be a quadratic ´etale subalgebra whose elements are fixed by τ . Let
W = K⊥ and consider σ, × and Φ as above. We will follow several steps.
(i) There is an element u ∈ W such that {u, τ (u), τ 2(u)} is a K-basis of W:
This is trivial if K is a field. Otherwise, K = Fe1 ⊕ Fe2, with e·2
i = ei, i = 1, 2,
e1 · e2 = 0 = e2 · e1. The Peirce subspaces U = e1 · C · e2 and V = e2 · C · e1 are
invariant under τ . Since U and V are isotropic subspaces coupled by the norm,
and τ is an orthogonal transformation, the minimal polynomial of τ U and τ V is
T 3 − 1. If x ∈ U and y ∈ V are elements such that {x, τ (x), τ 2(x)} is a basis of U
and {y, τ (y), τ 2(y)} is a basis of V, then we may take u = x + y. We also check
with this argument that the Segre symbol of τ is (32, 12).
20
A. ELDUQUE
Moreover, if F is algebraically closed, we may take a basis {u1, u2, u3} of U with
n(u1, u2 · u3) = 1 and τ (ui) = ui+1 (indices modulo 3). This shows that Cτ is the
split Okubo algebra over F (Definition 2.8). If F is not algebraically closed, extend
scalars to the algebraic closure to show that Cτ is an Okubo algebra over F.
(ii) With δ = τ − id, for any u ∈ W as in (i),
0 6= n(cid:0)Φ(u, τ (u), τ 2(u)(cid:1) = −n(cid:0)δ(u)(cid:1)3
:
Note first that, by linearity on each component, Φ(cid:0)u, τ (u), τ 2(u)(cid:1) = Φ(cid:0)u, δ(u), δ2(u)(cid:1).
Now, for x, y ∈ W, as σ is invariant under τ ,
Therefore, as δ3 = 0, for any x, y ∈ W,
σ(cid:0)δ(x), y(cid:1) = σ(cid:0)τ (x), y(cid:1) − σ(x, y)
= σ(cid:0)x, τ 2(y)(cid:1) − σ(x, y) = σ(cid:0)x, (δ2 − δ)(y)(cid:1).
σ(cid:0)δ(x), δ(y)(cid:1) = −σ(cid:0)x, δ2(y)(cid:1).
σ(cid:0)δ(x), δ2(y)(cid:1) = 0,
Thus, if α = n(cid:0)δ(u)(cid:1) = σ(cid:0)δ(u), δ(u)(cid:1), we get σ(cid:0)u, δ2(u)(cid:1) = −α = σ(cid:0)δ2(u), u(cid:1), and
0 6= n(cid:16)Φ(cid:0)u, τ (u), τ 2(u)(cid:1)(cid:17) = n(cid:16)Φ(cid:0)u, δ(u), δ2(u)(cid:1)(cid:17)
= det(cid:16)σ(cid:0)δi(u), δj(u)(cid:1)(cid:17)0≤i,j≤2
∗ −α
∗
∗
α
0
−α 0
= det
0
= −α3,
as required.
(iii)
If u and v are two elements as in (i), then
Indeed, given u as in (i), any other such v is of the form v = a · u + δ(w), for
Φ(cid:0)v, τ (v), τ 2(v)(cid:1) ∈ K·3 · Φ(cid:0)u, τ (u), τ 2(u)(cid:1) :
some a ∈ K and w ∈ W. Then, since Φ is alternating,
(6.1)
Φ(cid:0)v, τ (v), τ 2(v)(cid:1) = Φ(cid:0)v, δ(v), δ2(v)(cid:1)
(iv) There exists an element u as in (i) satisfying
= Φ(a · u, δ(a · u), δ2(a · u)(cid:1) = a·3 · Φ(cid:0)u, δ(u), δ2(u)(cid:1).
σ(cid:0)τ i(u), τ j(u)(cid:1) = −1,
for 1 ≤ i 6= j ≤ 3:
This has been essentially done in [EP96, p. 107]. Here a slightly different argu-
σ(cid:0)τ i(u), τ i(u)(cid:1) = 0,
ment will be used.
Let u be an element as in (i) and let a = Φ(cid:0)u, τ (u), τ 2(u)(cid:1) and α = n(cid:0)δ(u)(cid:1). By
(ii), n(a) = −α3. Take u = (αa−1) · u, then
n(cid:0)δ(u)(cid:1) = n(αa−1)n(cid:0)δ(u)(cid:1) =
α2
n(a)
α = −1.
Therefore we may assume that n(cid:0)δ(u)(cid:1) = −1 and hence, by (6.1), σ(cid:0)u, δ2(u)(cid:1) = 1.
For a, b ∈ K, consider the element u′ = u + a · δ(u) + b · δ2(u) ∈ W. Then
δ(u′) = δ(u) + a · δ2(u) so, using (6.1), we have
Hence σ(cid:0)u′, δ2(u′)(cid:1) = 1. Now,
n(cid:0)δ(u′)(cid:1) = σ(cid:0)δ(u′), δ(u′)(cid:1) = σ(cid:0)δ(u), δ(u)(cid:1) = n(cid:0)δ(u)(cid:1) = −1.
σ(cid:0)u′, δ(u′)(cid:1) = σ(cid:0)u + a · δ(u) + b · δ2(u), δ(u) + a · δ2(u)(cid:1)
= σ(cid:0)u, δ(u)(cid:1) + σ(cid:0)u, δ2(u)(cid:1) · ¯a + a · σ(cid:0)δ(u), δ(u)(cid:1)
= σ(cid:0)u, δ(u)(cid:1) + (a − ¯a) · σ(cid:0)δ(u), δ(u)(cid:1)
= σ(cid:0)u, δ(u)(cid:1) + (a − ¯a)n(cid:0)δ(u)(cid:1) = σ(cid:0)u, δ(u)(cid:1) − (a − ¯a),
(by (6.1))
ORDER 3 ELEMENTS IN G2 AND IDEMPOTENTS
21
and we may take a so that σ(cid:0)u′, δ(u′)(cid:1) ∈ F. But then we have
σ(cid:0)u′, δ(u′)(cid:1) = σ(cid:0)δ(u′), u′(cid:1) = σ(cid:0)u′, (δ2 − δ)(u′(cid:1) = 1 − σ(cid:0)u′, δ(u′)(cid:1),
so σ(cid:0)u′, δ(u′)(cid:1) = −1. Finally,
σ(u′, u′) = σ(cid:0)u + a · δ(u), u + a · δ(u)(cid:1) + σ(cid:0)u, b · δ2(u)(cid:1) + σ(cid:0)b · δ2(u), u(cid:1)
and we can take b so that σ(u′, u′) = 0. Therefore, the coordinate matrix of σ in
= n(cid:0)u + a · δ(u)(cid:1) + (b + ¯b),
the K-basis {u′, δ(u′), δ2(u′)} is(cid:16) 0 −1 1
0 0(cid:17), so that the coordinate matrix in the K-
basis {u′, τ (u′), τ 2(u′)} is (cid:16) 0 −1 −1
−1 −1 0 (cid:17), as required. (For instance, σ(cid:0)u′, τ 2(u′)(cid:1) =
σ(cid:0)u′, (δ2 − δ + id)(u′)(cid:1) = 1 − (−1) + 0 = −1.)
of Example 6.6, with a = Φ(cid:0)u, τ (u), τ 2(u)(cid:1), so that τ is, up to conjugation, the
Now, if u is an element as in (iv), let wi = τ i(u) and we are in the situation
automorphism τK,a.
−1 −1 0
1
−1 0 −1
If K and K′ are quadratic ´etale subalgebras of (C, ·, n), a ∈ K, a′ ∈ K′ are norm 1
elements and there is an isomorphism ϕ : K → K′ and an element b′ ∈ K′ such that
ϕ(a) = (b′)·3 · a′, then n(b′) = 1. The automorphism τ = τK,a is obtained by taking
a K-basis {w1, w2, w3} of K⊥ with σ(wi, wi) = 0, σ(wi, wj) = −1 for 1 ≤ i 6= j ≤ 3,
such that Φ(w1, w2, w3) = a and τ (wi) = wi+1 (indices modulo 3). Similarly,
denoting by σ′, Φ′ the corresponding maps on (K′)⊥, there is a K′-basis {w′
3}
of (K′)⊥ with σ′(w′
j) = −1 for 1 ≤ i 6= j ≤ 3, such that
Φ′(w′
3}
is another K′-basis of (K′)⊥ and, since n(b′) = 1, it also satisfies σ′(b′·w′
i) = 0,
3) = (b′)·3 · a′ =
σ′(b′ · w′
ϕ(a). Therefore, the isomorphism ϕ : K → K′ extends to an automorphism of
(C, ·, n) by ϕ(wi) = b′ · w′
j ) = −1, for 1 ≤ i 6= j ≤ 3, while Φ′(b′ · w′
i+1 (indices modulo 3). Then {b′·w′
i, 1 ≤ i ≤ 3, and ϕ ◦ τ = τ ′ ◦ ϕ.
3) = a′ and τ (w′
i) = 0, σ′(w′
2, b′ · w′
1, b′ · w′
i) = w′
1, w′
2, w′
1, b′·w′
2, b′·w′
1, w′
2, w′
i, b′ · w′
i, b′·w′
i, w′
i, w′
Conversely, assume that the order 3 automorphisms τ = τK,a and τ ′ = τK′,a′ are
conjugate. Then there is an element u ∈ W = K⊥ such that {u, τ (u), τ 2(u)} is a K-
basis of W in which the coordinate matrix of σ is(cid:16) 0 −1 −1
a and similarly for τ ′. Let φ ∈ Aut(C, ·, n) be such that φ ◦ τ ′ = τ ◦ φ. Then
K = φ(K′) is a quadratic ´etale subalgebra fixed elementwise by τ : K ⊆ Fix(τ ) =
K ⊕ δ2(W). In ( K)⊥, the element u = φ(u′) satisfies that {u, τ (u), τ 2(u)} is a K-
−1 −1 0 (cid:17) and Φ(cid:0)u, τ (u), τ 2(u)(cid:1) =
−1 0 −1
basis of W = ( K)⊥, and Φ(cid:0)u, τ (u), τ 2(u)(cid:1) = a = φ(a′). Then we must prove that
there is an isomorphism ϕ : K → K such that ϕ(a) ∈ ( K)·3 · a, as then φ−1 ◦ ϕ gives
an isomorphism K → K′ that takes a to (K′)·3 · a′.
Let 0 6= k ∈ K orthogonal to 1, so K = F1 ⊕ Fk. For any x, y ∈ W,
n(x, y) = −n(x · y, 1) = n(cid:0)σ(x, y), 1(cid:1),
n(k · x, y) = n(x · y, k) = −n(cid:0)σ(x, y) = −n(cid:0)σ(x, y), k(cid:1).
Then,
σ(x, y) = −n(x, y)1 +
1
n(k)
n(k · x, y)k.
The map g : C → F given by
g(x) = n(cid:0)x, τ (¯x) · τ 2(¯y)(cid:1)
(6.2)
(6.3)
22
A. ELDUQUE
is semilinear: g(x + y) = g(x) + g(y), g(αx) = α3g(x), for any x, y ∈ C and α ∈ F
(see [EP96, §5]). Then, for any x ∈ W,
Φ(cid:0)x, τ (x), τ 2(x)(cid:1) = σ(cid:0)x, τ (x) × τ 2(x)(cid:1)
But (k · y) × (k · z) = ¯k2(y × z) = −n(k)(y × z), so we get:
= −n(cid:0)x, τ (x) × τ 2(x)(cid:1)1 +
= −n(cid:0)x, τ (x) · τ 2(x)(cid:1)1 +
Φ(cid:0)x, τ (x), τ 2(x)(cid:1) = −g(x)1 −
1
n(k)
1
n(k)
n(cid:0)k · x, τ (x) × τ 2(x)(cid:1)k
n(cid:0)k · x, τ (x) · τ 2(x)(cid:1)k.
1
n(k)2 g(k · x)k.
(6.4)
Now, K is a quadratic ´etale subalgebra of Fix(τ ) = K ⊕ δ2(W) = K ⊕ K · δ2(u).
Since n(cid:0)δ2(W), Fix(τ )(cid:1) = 0, K = F1 ⊕ Fk, with k = k + l · δ2(u) for some l ∈ K.
n(k) k. Then n(v, 1) = 0 and
Let v = u + l+¯l
n(k, v) = −(l + ¯l) + n(cid:0)l · δ2(u), u(cid:1).
But σ(cid:0)l · δ2(u), u(cid:1) = l · σ(cid:0)δ2(u), u(cid:1) = l (recall that the coordinate matrix of σ in
the K-basis {u, δ(u), δ2(u)} is (cid:16) 0 −1 1
and hence n(k, v) = 0. Thus v ∈ W = ( K)⊥. By step (iii) above, Φ(cid:0)v, τ (v), τ 2(v)(cid:1) ∈
n(cid:0)l · δ2(u), u(cid:1) = σ(cid:0)l · δ2(u), u(cid:1) + σ(cid:0)u, l · δ2(u)(cid:1) = l + ¯l,
( K)·3 · a. As for (6.4), we have
0 0(cid:17)), so
−1 −1 0
1
Φ(cid:0)v, τ (v), τ 2(v)(cid:1) = −g(v)1 −
n(k)2
1
n(k)2 g(k · v)k.
On the other hand, g(v) = g(u), because g(k) = n(k, k·2) = 0, so
= −g(v)1 −
g(k · v)k
1
k · v =(cid:0)k + l · δ2(v)(cid:1) ·(cid:16)u +
1
n(k)
(l + ¯l) · k(cid:17)
= k · u − (l + ¯l) +(cid:0)l · δ2(u)(cid:1) · u +(cid:0)l · δ2(u)(cid:1)(cid:16) 1
n(k)
(l + ¯l) · k(cid:17).
For any x ∈ C, g(x) = g(cid:0)τ (x)(cid:1), so g(cid:0)δ(W)(cid:1) = 0. In particular, the last term above
is in δ2(W) ⊆ δ(W), so it is sent to 0 by g. Also, g(l + ¯l) = 2g(l) = −g(l), and
(6.5)
(6.6)
(cid:0)l · δ2(u)(cid:1) · u = −σ(cid:0)l · δ2(u), u(cid:1) +(cid:0)l · δ2(u)(cid:1) × u
= −l · σ(cid:0)δ2(u), u(cid:1) + ¯l ·(cid:0)δ2(u) × u(cid:1)
= −l + ¯l ·(cid:0)δ2(u) × u(cid:1).
As σ(cid:0)δ2(u), δ2(u) × u(cid:1) = Φ(cid:0)δ2(u), δ2(u), u(cid:1) = 0, because Φ is alternating,
so ¯l ·(cid:0)δ2(u) × u(cid:1) ∈ δ(W), g(cid:16)¯l ·(cid:0)δ2(u) × u(cid:1)(cid:17) = 0, and g(cid:16)(cid:0)l · δ2(u)(cid:1) · u(cid:17) = −g(l).
δ2(u) × u ∈ {x ∈ W : σ(cid:0)δ2(u), x(cid:1) = 0} = δ(W),
Therefore, using (6.6), g(k · v) = g(k · u) − 2g(l) − g(l) = g(k · u), so that, by
(6.5), we get
Φ(cid:0)v, τ (v), τ 2(v)(cid:1) = −g(u) −
1
n(k)
g(k · u)k,
and the isomorphism ϕ : K → K taking k to k satisfies that ϕ(a) = Φ(cid:0)v, τ (v), τ 2(v)(cid:1)
belongs to ( K)·3 · a, as required.
(cid:3)
ORDER 3 ELEMENTS IN G2 AND IDEMPOTENTS
23
Example 6.8. Let (C, ·, n) be the split Cayley algebra over a field F of characteristic
3, and let K be a subalgebra isomorphic to F × F, so K = Fe1 ⊕ Fe2, for orthogonal
idempotents e1 and e2. For 0 6= α, β ∈ F, the elements a = αe1 + α−1e2 and
b = βe1 + β−1e2 have norm 1 and the automorphisms τK,a and τK,b are conjugate
if and only if there is an automorphism ϕ of K such ϕ(a) ∈ K·3 · b, if and only if
there is a nonzero scalar 0 6= µ ∈ F such that either α = µ3β or α = µ3β−1, if
and only if F3α ∪ F3α−1 = F3β ∪ F3β−1. If F is perfect, we conclude that all these
automorphisms are conjugate.
Corollary 6.9. Let (C, ·, n) be the split Cayley algebra over a perfect field F of
characteristic 3. Let K and K′ be quadratic ´etale subalgebras, and let a ∈ K and
a′ ∈ K′ be norm 1 elements. Then the automorphisms τK,a and τK′,a′ are conjugate
if and only if K is isomorphic to K′. In particular, τK,a is conjugate to τK,1 and,
if F is algebraically closed, all these automorphisms are conjugate.
Proof. If ϕ : K → K′ is an isomorphism, then either both K and K′ are split,
and we are in the situation of Example 6.8, or K and K′ are perfect fields, so
ϕ(a) ∈ K′ = (K′)·3 · a′.
(cid:3)
7. Centralizers of order 3 automorphisms, char F = 3
In this section the centralizers of the different order 3 automorphisms in Theorem
6.3 will be computed.
For the automorphism in Theorem 6.3.(1), this has been done in [CEKT13, §10]
as part of the computation of the algebraic group of automorphisms of the split
Okubo algebra, which is not smooth. Here a different approach will be used, which
sheds light on the corresponding results in [CEKT13].
Actually, items (1) and (3) in Theorem 6.3 can be dealt with together. The
notation of Section 3 will be used throughout.
Theorem 7.1. Let (C, ·, n) be the split Cayley algebra over a field F of characteristic
3.
(1) For the order 3 automorphism τ of (C, ·, n) in item (1) of Theorem 6.3,
CentAut(C,·,n)(τ ) is the derived subgroup of the parabolic subgroup P{ε1}.
(2) For the order 3 automorphism τ of (C, ·, n) in item (3) of Theorem 6.3,
CentAut(C,·,n)(τ ) is the derived subgroup of the parabolic subgroup P{ε2−ε1}.
Proof. For simplicity write G = Aut(C, ·, n), L = Der(C, ·, n).
The automorphism in Theorem 6.3.(1) is τ = exp(δ), where δ = [Lu2 , Rv3] is in
the root space Lε2−ε3 . Note that δ2 = 0. More precisely,
δ(e1) = δ(e2) = 0,
δ(u1) = 0 = δ(u2), δ(u3) = u2 · (u3 · v3) − (u2 · u3) · v3 = u2,
δ(v1) = 0 = δ(v3), δ(v2) = −v3.
Hence τ = exp(δ) = id + δ. For ϕ ∈ G, ϕτ = τ ϕ if and only if Adϕ(δ) = δ. Denote
by H the centralizer of τ in G, and by H the stabilizer in G of Lε2−ε3 , so that H
is a subgroup of H. Moreover, the derived subgroup of H is contained in H, as its
action on the one-dimensional space Lε2−ε3 is trivial.
For any δ′ ∈ Lα with α ∈ Φ+ ∪ {−ε1}, α + (ε2 − ε3) is not a root, so [δ′, δ] = 0.
Hence the Uα's, with α ∈ Φ+ ∪{−ε1} are contained in H, and T and the Uα's, with
α ∈ Φ+ ∪ {−ε1}, are contained in H. It follows that H is the parabolic subgroup
P{ε1}, and H = TH. But T ∩ H = ker(ε2 − ε3) is a maximal torus of the derived
subgroup of P{ε1}, and hence H coincides with this derived subgroup.
24
A. ELDUQUE
As for the automorphism in Theorem 6.3.(3), τ = exp(δ), where δ = −adv3 . For
any R-point ϕ ∈ G(R), Adϕ(δ) = δ occurs if and only if adv3 = adϕ(v3) in L ⊗F R,
if and only if ϕ(v3) − v3 ∈ R1. But ϕ preserves the subspace orthogonal to 1, so we
get ϕ(v3) = v3. Denote again by H the centralizer of τ in G, so H is the stabilizer
of v3, and by H the stabilizer in G of Fv3, so that H is a subgroup of H. Moreover,
the derived subgroup of H is contained in H.
As before, for any δ′ ∈ Lα with α ∈ Φ+ ∪ {ε1 − ε2}, α + ε3 is not a root, so
[δ′, δ] = 0. Hence the Uα's, with α ∈ Φ+ ∪ {ε1 − ε2} are contained in H, and T
and the Uα's, with α ∈ Φ+ ∪ {ε1 − ε2}, are contained in H. It follows that H is the
parabolic subgroup P{ε2−ε1}. As above, H coincides with its derived subgroup. (cid:3)
Remark 7.2. Let (C, ·, n) be the split Cayley algebra over a field F of charac-
teristic 3, and let τ be its order 3 automorphism in item (1) of Theorem 6.3.
Then e = 1 is the unique quaternionic idempotent of the split Okubo algebra
Cτ = (C, ∗, n). By Proposition 2.6, StabAut(C,∗,n)(e) = CentAut(C,·,n)(τ ), and this
stabilizer is precisely Aut(C, ∗, n)red (the largest smooth subgroup of Aut(C, ∗, n))
by [CEKT13, §10]. Theorem 7.1 gives a more clear picture of this last group. It
turns out (see [CEKT13, §11]) that the group of automorphisms Aut(C, ∗, n) fac-
2
tors as StabAut(C,∗,n)(e)µ
3, where neither StabAut(C,∗,n)(e) nor the non smooth
subgroup µ
2
3 are normal.
The subgroup µ
2
3 corresponds to a grading by Z2
3 of the split Okubo algebra
(see [Eld09] or [EK13, §4.6]) and it is easy to check that it is not contained in the
normalizer of any maximal torus. Actually, the maximal tori of Aut(C, ∗, n) have
dimension one, because they are contained in Aut(C, ∗, n)red, which is the derived
subgroup of the parabolic subgroup P{ε1}, and if µ3 is in the normalizer of a rank
2
one torus, then it centralizes it. Hence, if µ
3 were contained in the normalizer of a
2
3 is self-centralized, because the corresponding
torus, it would centralize it. But µ
grading is fine with one-dimensional homogeneous components.
The fact [Pla68, Theorem 3.15] that quasitori, and more generally supersolvable
subgroups consisting of semisimple elements, are contained in normalizers of max-
imal tori over algebraically closed fields of characteristic 0 has been an important
tool in the classification of gradings on some exceptional Lie algebras (see, for in-
stance, [DM09]). The above shows that this is no longer true over fields of prime
characteristic.
The order 3 automorphism τ in Theorem 6.3.(4) is the composition of the au-
tomorphisms in items (1) and (3) of the same Theorem. Therefore, the unipotent
radical U of the standard Borel subgroup B, which is contained in the derived
subgroups of both P{ε1} and P{ε2−ε1}, centralizes τ . Actually, it is the whole
CentAut(C,·,n)(τ ).
Theorem 7.3. Let (C, ·, n) be the split Cayley algebra over a field F of characteristic
3. Then the centralizer of the order 3 automorphism of (C, ·, n) in item (4) of
Theorem 6.3 is the unipotent radical U of the standard Borel subgroup B.
Proof. Any ϕ in CentAut(C,·,n)(τ ) stabilizes the subalgebra of elements fixed by τ :
Fix(τ ) = span {1, u1, u2, v3}, so it stabilizes too its nilpotent radical: rad(cid:0)Fix(τ )(cid:1) =
span {u1, u2, v3}, and its square Fv3. But the stabilizer of Fv3 has been shown to be
the parabolic subgroup P{ε2−ε1} in the proof of Theorem 7.1. On the other hand,
with the notations of Section 3, both the torus T and the subgroups U±(ε2−ε1)
are contained in the subgroup SL(U), and hence so is the Levi subgroup L{ε2−ε1}.
Hence CentAut(C,·,n)(τ ) = U(cid:0)CentAut(C,·,n)(τ ) ∩ L{ε2−ε1}(cid:1). For any algebra R ∈
ORDER 3 ELEMENTS IN G2 AND IDEMPOTENTS
25
AlgF, any R-point ϕ of L{ε2−ε1} has a matrix in our basis {u1, u2, u3} of the form
a c
b
0
0
d 0
0
r
with (ad−bc)r = 1. Hence ϕ(v3) = ϕ(u1·u2) = (au1+bu2)·(cu1+du2) = (ad−bc)v3,
so
ϕτ (u3) = ϕ(cid:0)u3 + u2 + v3 − (e1 − e2)(cid:1) = ru3 + cu1 + du2 + (ad − bc)v3 + (e1 − e2),
τ ϕ(u3) = eτ (u3) = ru3 + ru2 + rv3 + r(e1 − e2).
Hence, if ϕ centralizes τ , we get r = 1, c = 0, and d = 1, so that ϕ is an element in
Uε2−ε1 ⊆ U. Therefore, CentAut(C,·,n)(τ ) = U.
(cid:3)
Finally, we must deal with the order 3 automorphisms in Theorem 6.3.(2), which
correspond to the automorphisms in Example 6.6. We will begin with the case that
appears explicitly in Theorem 6.3. Thus, consider the order 3 automorphism τst of
the split Cayley algebra (C, ·, n) given in (2.7) that permutes cyclically the elements
of our basis of the Peirce component U. That is,
τ (ej) = ej,
τ (ui) = ui+1,
τ (vi) = vi+1,
for j = 1, 2, and for i = 1, 2, 3 (indices taken modulo 3).
In this case, the subalgebra of elements fixed by τ is Fix(τ ) = span {e1, e2, u, v},
with u = u1 + u2 + u3 and v = v1 + v2 + v3. This is an associative algebra with
nilpotent radical rad(cid:0)Fix(τ )(cid:1) = Fu + Fv. Actually (Fu + Fv)·2 = 0. Any element in
the centralizer H = CentAut(C,·,n)(τ ) stabilizes Fix(τ ) and its radical, and hence
there is a natural homomorphism:
Φ : H −→ Aut(cid:16)Fix(τ )/ rad(cid:0)Fix(τ )(cid:1)(cid:17) ≃ Aut(F × F) ≃ C2,
where C2 denotes the constant group scheme corresponding to the cyclic group of
order 2. The homomorphism Φ is surjective with a section given by the embedding
of C2 into H corresponding to the order 2 automorphism σ ∈ Aut(C, ·, n) given by
σ(e1) = e2, σ(ui) = vi, i = 1, 2, 3. Thus, H is the semidirect product
(7.1)
H = ker Φ ⋊ C2.
(7.2)
As u · v = 3e1 = 0 = v · u, the derivations adu and adv commute. Moreover, we
v = 0. Their matrices in the canonical basis have integral entries,
have ad3
and if T and S are indeterminates,
u = ad3
exp(T adu) exp(Sadv) = exp(adT u+Sv) ∈ Aut(C ⊗F F[T, S]),
and hence, over any algebra R ∈ AlgF, it makes sense to consider the group K(R) =
{exp(adtu+sv) : t, s ∈ R}, which is a subgroup of Aut(C ⊗F R, ·, n). In this way
we obtain a subgroup K of Aut(C, ·, n), which is contained in H, and which is
isomorphic to G2
a. Actually, K is contained in ker Φ and its intersection with the
subgroup SL(U) is trivial (recall that SL(U) is identified with the subgroup of
Aut(C, ·, n) that fixes e1 and e2). Moreover, for any R ∈ AlgF, any R-point ϕ ∈
ker(Φ) satisfies ϕ(e1) = e1 +tu+sv for some t, s ∈ R, so ϕ(e1) = exp(ad−tu+sv)(e1).
Thus ϕ ◦ exp(adtu−sv) fixes e1 and e2 = 1 − e1, and hence it belongs to H ∩ SL(U).
Therefore, we have
ker Φ = K ⋊(cid:0)H ∩ SL(U)(cid:1).
(7.3)
26
A. ELDUQUE
There are two natural subgroups contained in H ∩ SL(U). One of them is
isomorphic to µ3, as µ3 embeds in H ∩ SL(U) as follows:
µ3 ֒→ H,
id
on Fe1 + Fe2,
rid
on U,
r2id on V.
r 7→
The other natural subgroup of H ∩ SL(U) is N := ker Ψ, where Ψ is the restriction
homomorphism
It is clear that the subgroups K and N commute. Given any R ∈ AlgF and any
ϕ ∈ N(R), ϕ(u1) = au1 + bu2 + cu3 for a, b, c ∈ R. As ϕ commutes with τ , the
matrix, relative to the basis {u1, u2, u3} of the restriction of ϕ to U is
Ψ : H −→ Aut(cid:0)Fix(τ )(cid:1).
(7.4)
(7.5)
(7.6)
a c
b
b a c
c
b a
0
1
c
c − b
1
c − b
0
0
1
with a + b + c = 1 because ϕ(u) = u. Hence, in the basis {u1, u1 − u2, u1 + u2 + u3},
the matrix is
and this shows that N is a two-dimensional unipotent group. Actually, the assign-
ment
(r, s) 7→
1
r
1
2 (r + r2) + s
0
1
r
shows that N is isomorphic to G2
a.
0
0
1
Proposition 7.4. Let (C, ·, n) be the split Cayley algebra over a field F of charac-
teristic 3. Then the centralizer of the order 3 automorphism τst in (2.7) is, with
the notations above,
H =(cid:16)(cid:0)K ⋊ µ3(cid:1) × N(cid:17) ⋊ C2.
Proof. Because of (7.2) and (7.3), it is enough to show that H ∩ SL(U) = µ3 × N.
But for any R ∈ AlgF and any R-point ϕ ∈ H ∩ SL(U), if ϕ(u1) = au1 + bu2 + cu3
(a, b, c ∈ R), then the matrix of the restriction of ϕ to U is the matrix in (7.5),
with determinant 1, that is, with (a + b + c)3 = 1, so a + b + c ∈ µ3. The result
follows.
(cid:3)
Note that the centralizer in Proposition 7.4 is four-dimensional and not smooth,
because of the presence of the subgroup µ3. The same situation appears in our next
result, which deals with the general case in Theorem 6.3.(2). We must consider the
order 3 automorphisms τK,a in Example 6.6:
Theorem 7.5. Let (C, ·, n) be the split Cayley algebra over a field F of characteristic
3. Let τ be the order 3 automorphism in Example 6.6 and let H = CentAut(C,·,n)(τ )
be its centralizer. Then, with the notations in this Example, there is a short exact
sequence
where
1 →(cid:0)K ⋊ µ3[K](cid:1) × N → H → C2 → 1,
ORDER 3 ELEMENTS IN G2 AND IDEMPOTENTS
27
• µ3[K] is the twisted form of µ3 such that for any R ∈ AlgF, µ3[K](R) =
{r ∈ K ⊗F R : r3 = 1, n(r) = 1} (see, for instance, [KMRT98, p. 418]).
Here n denotes the R-extension of the generic norm of K.
µ3[K] embeds in H by sending any r ∈ µ3[K](R) to the automorphism of
C ⊗F R that is the identity on K ⊗F R and r times the identity on W ⊗F R.
R} for any R ∈ AlgF, so that K is isomorphic to G2
a.
• N is the two-dimensional unipotent subgroup obtained as the kernel of the
• K is the abelian subgroup such that K(R) = {exp(adx) : x ∈ rad(cid:0)Fix(τ )(cid:1)⊗F
restriction map H → Aut(cid:0)Fix(τ )(cid:1). It is isomorphic too to G2
Φ : H −→ Aut(cid:16)Fix(τ )/ rad(cid:0)Fix(τ )(cid:1)(cid:17) ≃ Aut(K) ≃ C2,
Proof. Here Fix(τ ) = K ⊕ K · (w1 + w2 + w3) and its radical is K · (w1 + w2 + w3).
As for Proposition 7.4, we have a homomorphism as in (7.1)
and ker Φ = K⋊(cid:0)H∩SU(W, σ)(cid:1), with K as above and where SU(W, σ) is identified
naturally with the subgroup of automorphisms of (C, ·, n) which restrict to the
identity on K. Moreover, Φ is surjective, as so it is over the algebraic closure F,
where we are in the situation of (7.1). (Note that C2 is smooth.)
a.
For R ∈ AlgF, if ϕ is an R-point of H ∩ SU(W, σ), then ϕ(w1) = aw1 + bw2 + cw3
for a, b, c ∈ K⊗FR. As ϕ commutes with τ , the matrix of ϕ in the basis {w1, w2, w3}
is the matrix in (7.5), whose determinant is (a + b + c)3, which must be 1. Write
r = a + b + c.
Now, with N being defined as the kernel of the restriction homomorphism Ψ :
so a¯b + b¯c + c¯a = 1
n(a + b + c) = n(a) + n(b) + n(c) = 1. Therefore, r ∈ µ3[K].
only if −(a¯b + a¯c + b¯a + b¯c + c¯a + a¯b) = 0, if and only if n(a, b) + n(a, c) + n(b, c) = 0,
where x 7→ ¯x denotes the involution on K ⊗F R obtained by the extension of
As ϕ ∈ SU(W, σ)(R), σ(cid:0)ϕ(w1), ϕ(w1)(cid:1) = σ(w1, w1) = 0, and this happens if and
the nontrivial involution on K. Also σ(cid:0)ϕ(w1), ϕ(w2)(cid:1) = σ(w1, w2) = −1, which is
equivalent to −(cid:0)n(a)+n(b)+n(c)+a¯b+b¯c+c¯a(cid:1) = −1. This implies a¯b+b¯c+c¯a ∈ R,
2(cid:0)n(a, b) + n(a, c) + n(b, c)(cid:1) = 0. It follows then that n(r) =
H → Aut(cid:0)Fix(τ )(cid:1), as in (7.4), the same arguments as for the proof of Proposition
ker Φ = (cid:0)K ⋊ µ3[K](cid:1) × N. More precisely, any R-point of N (cid:0)⊆ SU(W, σ)(cid:1) has a
coordinate matrix as in (7.5), relative to the basis {w1, w2, w3}, with a, b, c ∈ K⊗F R
satisfying a+b+c = 1, or as in (7.6), relative to the basis {w1, w1−w2, w1+w2+w3}.
But, as above, a¯b + b¯c + c¯a = 0, so (1 − b − c)¯b + b¯c + c(1 − ¯b − ¯c) = 0. This gives
c + ¯b = (b − c)(¯b − ¯c) = n(b − c) ∈ R and, as b + ¯b ∈ R, we get r := c − b ∈ R. Hence
r2 = n(b − c) = c + ¯b, so c = r2 − ¯b = r + b and thus c = −(r + r2) − (b − ¯b). Let
K = F1 + Fk, with n(1, k) = 0, then ¯b − b = sk for some s ∈ R, and the assignment
7.4 give that N is a two-dimensional unipotent subgroup of H ∩ SU(W, σ) and that
(r, s) 7→
a ≃ N.
gives an isomorphism G2
1
r
−(r + r2) + sk
0
1
r
0
0
1
(cid:3)
We will see in Remark 8.6 that the short exact sequence in Theorem 7.5 does not
necessarily split, and even that the projection of H onto C2 may fail to be surjective
for rational points.
8. Idempotents in Okubo algebras, char F = 3
The next definition is based on Theorem 6.7.
Definition 8.1. Let F be a field of characteristic 3.
28
A. ELDUQUE
• Let K (respectively, K′) be a quadratic ´etale algebra over F, and let a ∈ K
(resp. a′ ∈ K′) be a norm 1 element. Then the pair (K, a) is said to be
equivalent to (K′, a′) if and only if there is an isomorphism ϕ : K → K′
such that ϕ(a) ∈ (K′)·3 · a′.
The equivalence class of the pair (K, a) will be denoted by [K, a].
• Let e be a quadratic idempotent of and Okubo algebra (O, ∗, n) and let,
as in (2.4), τe : x 7→ e ∗ (e ∗ x) be the associated order 3 automorphism of
(O, ∗, n) and of the split Cayley algebra (O, ·, n), where x·y = (e∗x)∗(y ∗e).
Then τe is conjugate in Aut(O, ·, n) to an automorphism τK,a as in Example
6.6. Define the class of e as cl(e) := [K, a].
Proposition 8.2. Let (O, ∗, n) be an Okubo algebra over a field F of characteristic
3, and let e, f be two quadratic idempotents. Then e and f are conjugate (by an
element of Aut(O, ∗, n)) if and only if cl(e) = cl(f ).
Proof. Consider the order 3 automorphisms τe and τf and the Cayley algebras
(O, ·, n) and (O, ⋄, n) with multiplications given by x · y = (e ∗ x) ∗ (y ∗ e) and
x ⋄ y = (f ∗ x) ∗ (y ∗ f ) respectively. Since both (O, ·, n) and (O, ⋄, n) are the split
Cayley algebra up to isomorphism, there is an isomorphism ϕ : (O, ·, n) → (O, ⋄, n).
Note that ϕ(e) = f , because e is the unity of (O, ·, n) and f the one in (O, ⋄, n).
Now τe is conjugate to some τK,a in Aut(O, ·, n) and τf to τK′,a′ in Aut(O, ⋄, n).
Then ϕ ◦ τe ◦ ϕ−1 is conjugate to τϕ(K),ϕ(a) and trivially [ϕ(K), ϕ(a)] = [K, a].
Then [K, a] = [K′, a′] if and only if ϕ ◦ τe ◦ ϕ−1 is conjugate to τf in Aut(O, ⋄, n),
if and only if there is an automorphism ψ ∈ Aut(O, ⋄, n) such that ψ ◦ ϕ ◦ τe ◦ (ψ ◦
ϕ)−1 = τf , if and only if there is an isomorphism φ : (O, ·, n) → (O, ⋄, n) such that
φ ◦ τe = τf ◦ φ. In this case, for any x, y ∈ O,
φ(x ∗ y) = φ(cid:16)τe(cid:0)n(e, x)e − x(cid:1) · τ 2
e(cid:0)n(e, y)e − y(cid:1)(cid:17)
= τf(cid:16)n(f, φ(x))f − φ(x)(cid:17) ⋄ τ 2
= φ(x) ∗ φ(y),
f(cid:16)n(f, φ(y)) − φ(y)(cid:17)
so φ ∈ Aut(O, ∗, n), and φ(e) = f .
Conversely, if φ ∈ Aut(O, ∗, n) and φ(e) = f , then φ ◦ τe = τf ◦ φ, and φ is also
(cid:3)
an isomorphism (O, ·, n) → (O, ⋄, n).
By [Eld97], if an Okubo algebra (O, ∗, n) over a field F of characteristic 3 contains
an idempotent, then either it is the split one (and this happens if and only if
g(O) = F3, where g : O → F is the semilinear map given by g(x) = n(x, x ∗ x) as
in (6.3)), or g(O) is a cubic (purely inseparable) field extension of F3, i.e., there
is α ∈ F \ F3 such that g(O) = F3(α). In the latter case, two such algebras are
isomorphic if and only if g(O1) = g(O2).
Proposition 8.3. Let (C, ·, n) be the split Cayley algebra over a field F of charac-
teristic 3, let K be a quadratic ´etale subalgebra, and a ∈ K an element of norm 1
as in Example 6.6. Consider the Petersson algebra CτK,a. Then
(the subfield of F generated by F3 and a + ¯a).
g(cid:0)CτK,a(cid:1) = F3(a + ¯a)
Proof. As in Example 6.6, C = K ⊕ K · w1 ⊕ K · w2 ⊕ K · w3, with W = K⊥ =
K · w1 ⊕ K · w2 ⊕ K · w3, σ(wi, wi) = 0, σ(wi, wj ) − 1, for 1 ≤ i 6= j ≤ 3, and
Φ(w1, w2, w3) = a. The automorphism τ = τK,a restricts to the identity on K and
satisfies τ (wi) = wi+1 (indices modulo 3). Take k ∈ K with n(1, k) = 0 6= n(k), so
that K = F1 ⊕ Fk.
ORDER 3 ELEMENTS IN G2 AND IDEMPOTENTS
29
Since g(x) = n(x, x ∗ x) is semilinear and τ -invariant, g(cid:0)CτK,a(cid:1) = F3 + g(cid:0)K · w1(cid:1),
and it is enough to compute g(w1) and g(k · w1). From
σ(wi, w2 × w3) = Φ(wi, w2, w3) =(0
if i = 2, 3,
a if i = 1,
we obtain w2 × w3 = ¯a · (w2 + w3 − w1), so
w1 ∗ w1 = τ (w1) · τ 2(w1) = w2 · w3 = −σ(w2, w3) + w2 × w3 = 1 + ¯a · (w2 + w3 − w1).
Also, n(K, W) = 0 and n(x, y) = σ(x, y) + σ(y, x) = σ(x, y) + σ(x, y) for any
x, y ∈ W. Hence,
g(w1) = n(cid:0)w1, 1 + ¯a · (w2 + w3 − w1)(cid:1)
= n(w1, ¯a · w2) + n(w1, ¯a · w3)
= σ(w1, ¯a · w2) + σ(w1, ¯a · w2) + σ(w1, ¯a · w3) + σ(w1, ¯a · w3)
= −a − ¯a − a − ¯a = a + ¯a,
g(k · w1) = n(cid:0)k · w1), (k · w1) ∗ (k · w1)(cid:1) = n(cid:0)k · w1, τ (k · w1) · τ 2(k · w1)(cid:1)
= n(cid:0)k · w1, (k · w2) × (k · w3)(cid:1) = −n(k)n(cid:0)k · w1, w2 × w3(cid:1)
= −n(k)n(cid:0)k · w1, ¯a · (w2 + w3 − w1)(cid:1)
= −n(k)(k · a + k · a).
But a = α + βk, α, β ∈ F, with α2 + β2n(k) = 1, so a + ¯a = −α, and
−n(k)(k · a + k · a) = βn(k)2 =(0 if β = 0, so that α = ±1 ∈ F3,
β 3 (1 − α2)2 ∈ F3(α),
= 1
Hence g(k · w1) ∈ F3(a + ¯a), as required.
1
β 3(cid:0)β2n(k)(cid:1)2
otherwise.
(cid:3)
Corollary 8.4. The Petersson algebra CτK,a is the split Okubo algebra if and only
if a ∈ K·3.
Proof. We have g(cid:0)CτK,a(cid:1) = F3 if and only if a + ¯a ∈ F3. But with a = α + βk as
in the proof of Proposition 8.3, a ∈ K·3 if and only if α = −(a + ¯a) ∈ F3. Indeed,
if a = (µ + νk)·3 = µ3 − n(k)ν3k, then −(a + ¯a) = α = µ3. And conversely, if
α = µ3, then β2n(k) = 1 − α2 = 1 − µ6 = (1 − µ2)3 ∈ F3, and either β = 0 or
βk = − 1
(cid:3)
β 2n(k) (βk)·3 ∈ K·3. Hence both α and βk are in K·3, so a ∈ K·3.
Therefore, the situation for idempotents in Okubo algebras over fields of char-
acteristic 3 is now almost settled. Only a couple of things remain to be checked:
Theorem 8.5. Let (O, ∗, n) be an Okubo algebra over a field F of characteristic 3.
(1) If dimF3 g(O) = 8, then (O, ∗, n) contains no idempotents.
(2) If dimF3 g(O) = 3, (O, ∗, n) contains a unique quadratic idempotent for each
class [K, a] with a 6∈ K·3 and g(O) = F3(a + ¯a).
(3) If (O, ∗, n) is the split Okubo algebra, then it contains:
(a) a unique quaternionic idempotent,
(b) a conjugacy class of quadratic idempotents for each isomorphism class
of quadratic ´etale algebras,
(c) a unique conjugacy class of singular idempotents.
Proof. The uniqueness of the conjugacy class of singular idempotents in the split
Okubo algebra needs to be proved, but if e and f are two such idempotents, using
the notation in the proof of Proposition 8.2, τe is the unique, up to conjugacy,
order 3 automorphism of (O, ·, n) in Theorem 6.3.(4), and the same happens for
30
A. ELDUQUE
τf in Aut(O, ⋄, n). Then there is an isomorphism ϕ : (O, ·, n) → (O, ⋄, n) with
ϕ ◦ τe = τf ◦ ϕ and, as in the proof of 8.2, ϕ is an automorphism of (O, ∗, n) and
ϕ(e) = f .
Also, if dimF3 g(O) = 3, only quadratic idempotents exist (see Corollary 6.5),
and there exists a conjugacy class of quadratic idempotentes for each class [K, a]
with a 6∈ K·3 and g(O) = F3(a + ¯a). It remains to show that these conjugacy classes
are singletons. In other words, that the idempotents are fixed by automorphisms
in this case. Note first that here F is not perfect, and hence it is infinite. Let
e be a quadratic idempotent, let (O, ·, n) be the Cayley algebra where x · y =
(e ∗ x) ∗ (y ∗ e) as in (2.5), and let G = Aut(O, ·, n). The automorphism group
Aut(O, ∗, n) contains the group of rational points of CentG(τ ) = StabAut(O,∗,n)(e).
But the arguments in the proof of [Eld99, Theorem 13] show that the matrix group
Aut(O, ∗, n) is a four-dimensional unipotent affine algebraic group over F in the sense
of [Wat79, Chapter 4], and it is isomorphic, as an algebraic set, to the affine space
of dimension 4 (i.e.; its coordinate ring is isomorphic to the ring of polynomials in
four variables over F). However, with the notation in Theorem 7.5, CentG(τ )(F)
contains the closed subgroup K(F) × N(F), isomorphic to G4
a(F). By dimension
count, Aut(O, ∗, n) = K(F) × N(F), and hence e is fixed by any automorphism. (cid:3)
Remark 8.6. In the case of dimF3 g(O) = 3 in Theorem 8.5, the arguments in
the proof show that Aut(O, ∗, n) coincides with the group of rational points of
(cid:0)K ⋊ µ3[K](cid:1) × N in Theorem 7.5, but then this gives
Aut(O, ∗, n) =(cid:16)(cid:0)K ⋊ µ3[K](cid:1) × N(cid:17)(F) = H(F),
and this shows that the projection H → C2 is not surjective on rational points. In
particular, the short exact sequence in Theorem 7.5 does not split.
Theorem 8.5 shows, in particular, that over an algebraically closed field of char-
acteristic 3, the Okubo algebra contains a unique quaternionic idempotent, a unique
conjugacy class of quadratic idempotents, and a unique conjugacy class of singular
idempotents. This was proved in [Eld15].
The singular idempotents of the split Okubo algebra are strongly connected with
the quaternionic idempotent.
Proposition 8.7. Let (O, ∗, n) be the split Okubo algebra over a field F of charac-
teristic 3, and let e be its quaternionic idempotent. Then:
(i) The set of singular idempotents is given by:
{singular idempotents of (O, ∗, n)}
= {e + x : 0 6= x ∈ Cent(O,∗,n)(e) ∩ Cent(O,∗,n)(e)⊥}.
(ii) The set of idempotents is:
{idempotents of (O, ∗, n)} = {e + x : x ∈ Cent(O,∗,n)(e), x ∗ x = 0}.
Proof. Let τ be the order 3 automorphism in (2.4) attached to the quaternionic
idempotent e, and let (O, ·, n) be the associated Cayley algebra, with unity 1 = e,
and with multiplication given by x · y = (e ∗ x) ∗ (y ∗ e), so that x ∗ y = τ (¯x) · τ 2(¯y),
for any x, y ∈ O.
By Theorem 6.3.(1), there is a canonical basis {e1, e2, u1, u2, u3, v1, v2, v3} of
(O, ·, n) such that
τ (ui) = ui, i = 1, 2,
τ (u3) = u3 + u2.
The proof of Theorem 6.3 also shows that Fix(τ ) = span {e1, e2, u1, u2, v1, v3}, so
that R := Cent(O,∗,n)(e) ∩ Cent(O,∗,n)(e)⊥ = Fu2 + Fv3. We have R · R = 0, so that
ORDER 3 ELEMENTS IN G2 AND IDEMPOTENTS
31
R ∗ R = 0 too and hence, for any x ∈ R:
(1 + x) ∗ (1 + x) = τ (1 + x) · τ 2(1 + x) = (1 − x) · (1 − x) = 1 − 2x = 1 + x,
and 1+x = e+x is an idempotent. If x 6= 0, this idempotent is not the quaternionic
idempotent, and hence to check that this is a singular idempotent, it is enough to
find a three-dimensional isotropic subspace in its centralizer. But x = αu2 + βv3,
with α, β ∈ F. If α 6= 0 = β, then span {u2, v3, v1} is contained in the centralizer
of e + x, if α = 0 6= β, then span {u2, v3, u1} works, and if α 6= 0 6= β, then
span(cid:8)u2, v3, (e1 − e2) + α−1βu1 + αβ−1v1(cid:9) does the job.
Conversely, if now f is a singular idempotent, and τ ′ is the associated order
3 automorphism: τ ′(x) = f ∗ (f ∗ x) for any x ∈ O, the proof of Theorem 6.3
shows that there is a canonical basis {e1, e2, u1, u2, u3, v1, v2, v3} of (O, ⋄, n), where
x ⋄ y = (f ∗ x) ∗ (y ∗ f ) for any x, y, such that
τ ′(ui) = ui, i = 1, 2,
τ ′(u3) = u3 − (e1 − e2) + u2 + v3,
and it also shows that e = 1−v3 = f −v3 is the quaternionic idempotent of (O, ∗, n),
and that Cent(O,∗,n)(e) equals span {e1, e2, u1, u2, v1, v3}, so that f = e + v3, and
v3 ∈ Cent(O,∗,n)(e) ∩ Cent(O,∗,n)(e)⊥. This proves the first part.
Now, for any x ∈ Cent(O,∗,n)(e), Proposition 2.6 shows that τ (x) = e∗(e∗x) = x,
and e ∗ x = x∗ e = n(e, x)e − x. If x∗ x = 0, then 0 = x∗ x = ¯x·2, so that x·2 = 0 and
n(x) = 0 = n(x, e). Hence e ∗ x = x ∗ e = −x and (e + x) ∗ (e + x) = e + e ∗ x + x ∗ e =
e − 2x = e + x.
Conversely, singular idempotents are of this form by the first part of the Propo-
sition, so it is enough to show that quadratic idempotents are of this form too,
and to do that we may extend scalars and assume that F is algebraically closed.
Let us prove first that any idempotent is contained in Cent(O,∗,n)(e). Since all
quadratic idempotents are conjugate (by our assumption on F and Theorem 8.5),
and since Cent(O,∗,n)(e) is invariant under automorphisms (because so is the unique
quaternionic idempotent e), it is enough to show that Cent(O,∗,n)(e) contains a qua-
dratic idempotent. With the notations in Theorem 6.3.(1), the element 1 + u1 is
an idempotent and it is nonsingular as u1 6∈ Cent(O,∗,n)(e)⊥.
Now, if 0 6= f ∈ Cent(O,∗,n)(e) is an idempotent, f ∈ Cent(O,∗,n)(e) so f =
a + b with a ∈ Q = span {e1, e2, u1, v1} and b ∈ Cent(O,∗,n)(e) ∩ Cent(O,∗,n)(e)⊥ =
span {u2, v3}. Then a ∗ a = a, so by the proof of Corollary 4.4, (a − 1)·2 = 0, and
f = 1 + c = e + c, with c = (a − 1) + b. Moreover, n(c) = 0 = n(1, c), so c·2 = 0
and c ∗ c = ¯c·2 = 0, as required.
(cid:3)
Finally, if a non split Okubo algebra (O, ∗, n) over a field F of characteristic 3
contains idempotents, then dimF3 g(O) = 3 and all the idempotents are quadratic.
In this case any idempotent determines all the idempotents:
Proposition 8.8. Let (O, ∗, n) be an Okubo algebra over a field F of characteristic
3 with dimF3 g(O) = 3, and let e be an idempotent. Then the set of idempotents is
given by:
{idempotents of (O, ∗, n)} =(cid:8)e + x : x ∈ rad(cid:0)Cent(O,∗,n)(e)(cid:1)(cid:9) .
Proof. Let (O, ·, n) be the Cayley algebra where x · y = (e ∗ x) ∗ (y ∗ e) as in (2.5),
where 1 = e. The proof of Theorem 8.5 shows that any idempotent is fixed by
Aut(O, ∗, n) = K(F)×N(F). By Theorems 6.7 and 7.5, O = K⊕K·w1⊕K·w2⊕K·w3
as in Example 6.6, and the automorphism τ : x 7→ e ∗ (e ∗ x) fixes elementwise
the quadratic ´etale algebra K and permutes cyclically the wi's. Then, with w =
32
A. ELDUQUE
w1 + w2 + w3 we have Cent(O,∗,n)(e) = K ⊕ K · w, and
Fix(cid:0)Aut(O, ∗, n)(cid:1) = Fix(cid:0)K(F)(cid:1) ∩ Fix(cid:0)N(F)(cid:1)
= Fix(cid:0)K(F)(cid:1) ∩(cid:0)K ⊕ K · w(cid:1)
= F1 ⊕ K · w = F1 ⊕ rad(cid:0)Cent(O,∗,n)(e)(cid:1).
Any idempotent in F1 ⊕ rad(cid:0)Cent(O,∗,n)(e)(cid:1) projects into an idempotent in F1, and
hence it is of the form e + x, with x ∈ rad(cid:0)Cent(O,∗,n)(e)(cid:1). Conversely, any element
of the form e+x with x ∈ rad(cid:0)Cent(O,∗,n)(e)(cid:1) satisfies (e+x)∗2 = e∗2 +e∗x+x∗e =
e − 2x = e + x.
(cid:3)
References
[AEMN02] P. Alberca-Bjerregaard, A. Elduque, C. Mart´ın-Gonz´alez, and F.J. Navarro-M´arquez,
[AE16]
On the Cartan-Jacobson theorem, J. Algebra 250 (2002), no. 2, 397 -- 407.
A. Castillo-Ram´ırez and A. Elduque, Some special features of Cayley algebras, and
G2, in low characteristics, J. Pure Appl. Algebra 220 (2016), no. 3, 1188 -- 1205.
[CEKT13] V. Chernousov, A. Elduque, M.-A. Knus, and J.-P. Tignol, Algebraic groups of type
[DM09]
[Eld97]
[Eld99]
[Eld09]
[Eld15]
[EK13]
[EM91]
[EM93]
[EM95]
[EP96]
[Jac58]
[Kac90]
D4, triality, and composition algebras, Doc. Math. 18 (2013), 413 -- 468.
C. Draper and C. Mart´ın-Gonz´alez, Gradings on the Albert algebra and on f4, Rev.
Mat. Iberoam. 25 (2009), no. 3, 841 -- 908.
A. Elduque, Symmetric composition algebras J. Algebra 196 (1997), no. 1, 282 -- 300.
A. Elduque, Okubo algebras in characteristic 3 and their automorphisms Comm. Al-
gebra 27 (1999), no. 6, 3009 -- 3030.
A. Elduque, Gradings on symmetric composition algebras, J. Algebra 322 (2009),
no. 10, 3542 -- 3579.
A. Elduque, Okubo algebras: automorphisms, derivations and idempotents, Lie alge-
bras and related topics, 61 -- 73, Contemp. Math., 652, Amer. Math. Soc., Providence,
RI, 2015.
A. Elduque and M. Kochetov, Gradings on simple Lie algebras, Mathematical Surveys
and Monographs 189, American Mathematical Society, Providence, RI, 2013.
A. Elduque and H.-C. Myung, Flexible composition algebras and Okubo algebras
Comm. Algebra 19 (1991), no. 4, 1197 -- 1227.
A. Elduque and H.-C. Myung, On flexible composition algebras, Comm. Algebra 21
(1993), no. 7, 2481 -- 2505.
A. Elduque and H.-C. Myung, Colour algebras and Cayley-Dickson algebras, Proc.
Roy. Soc. Edinburgh Sect. A 125 (1995), no. 6, 1287 -- 1303.
A. Elduque, J.-M. P´erez-Izquierdo, Composition algebras with associative bilinear
form, Comm. Algebra 24 (1996), no. 3, 1091 -- 1116.
N. Jacobson, Composition algebras and their automorphisms, Rend. Circ. Mat.
Palermo (2) 7 (1958), 55 -- 80.
V.G. Kac, Infinite-dimensional Lie algebras, third edition, Cambridge University Press,
Cambridge, 1990.
[Oku78]
[OO81a]
[OO81b]
[KMRT98] M.-A. Knus, A. Merkurjev, M. Rost, and J.-P. Tignol, The book of involutions, Amer-
ican Mathematical Society Colloquium Publications, vol. 44, American Mathematical
Society, Providence, RI, 1998, With a preface in French by J. Tits.
S. Okubo, Pseudo-quaternion and pseudo-octonion algebras, Hadronic J. 1 (1978),
no. 4, 1250 -- 1278.
S. Okubo and J.M. Osborn Algebras with nondegenerate associative symmetric bilinear
forms permitting composition, Comm. Algebra 9 (1981), no. 12, 1233 -- 1261.
S. Okubo and J.M. Osborn Algebras with nondegenerate associative symmetric bilinear
forms permitting composition. II, Comm. Algebra 9 (1981), no. 20, 2015 -- 2073.
S. Pepin Lehalleur, Subgroups of maximal rank of reductive groups, in "Autour des
sch´emas en groupes. Vol. III. A celebration of SGA3. B. Edixhoven, P. gille, G. Prasad,
and P. Polo, editors. Panoramas et Synth`eses 47, pages 147 -- 172. Socit Mathmatique
de France, Paris, 2015.
H.P. Petersson, Eine Identitat funften Grades, der gewisse Isotope von Kompositions-
Algebren genugen, Math. Z. 109 (1969) 217 -- 238.
V.P. Platonov, The theory of algebraic linear groups and periodic groups, American
Society Translations (2) 69 (1968), 61 -- 110.
[PL15]
[Pet69]
[Pla68]
ORDER 3 ELEMENTS IN G2 AND IDEMPOTENTS
33
[Sch95]
[Ser06]
[Spr09]
[SV00]
[Ste68]
R.D. Schafer, An introduction to nonassociative algebras, corrected reprint of the 1966
original. Dover Publications, Inc., New York, 1995.
J.P. Serre, Coordonn´ees de Kac, Oberwolfach Reports 3 (2006), 1787 -- 1790.
T.A. Springer, Linear algebraic groups, reprint of the 1998 second edition. Modern
Birkhuser Classics. Birkhuser Boston, Inc., Boston, MA, 2009.
T.A. Springer, F.D. Veldkamp, Octonions, Jordan algebras and exceptional groups,
Springer Monographs in Mathematics. Springer-Verlag, Berlin, 2000.
R. Steinberg, Lectures on Chevalley groups. Notes prepared by John Faulkner and
Robert Wilson. Yale University, New Haven, Conn., 1968.
[Wat79] W.C. Waterhouse, Introduction to affine group schemes, Graduate Texts in Mathe-
matics 66. Springer-Verlag, New York -- Berlin, 1979.
[ZSSS82] K.A. Zhevlakov, A.M. Slin'ko, I.P. Shestakov, and A.I. Shirshov, Rings that are Nearly
Associative, Pure Appl. Math., vol. 104, Academic Press Inc. [Harcourt Brace Jo-
vanovich Publishers], New York, 1982.
Departamento de Matem´aticas e Instituto Universitario de Matem´aticas y Aplica-
ciones, Universidad de Zaragoza, 50009 Zaragoza, Spain
E-mail address: [email protected]
|
1909.12039 | 1 | 1909 | 2019-09-26T11:51:27 | Weakly $I$-clean rings | [
"math.RA"
] | In this article, we introduce the concept of weakly $I$-clean ring, for any ideal $I$ of a ring $R$. We show that, for an ideal $I$ of a ring $R$, $R$ is uniquely weakly $I$-clean if and only if $R/I$ is semi boolean and idempotents can be lifted uniquely weakly modulo $I$ if and only if for each $a\in R$, there exists a central idempotent $e\in R$ such that either $a-e\in I$ or $a+e\in I$ and $I$ is idempotent free. As a corollary, we characterize weakly $J$-clean ring. Also we study various properties of weakly $I$-clean ring. | math.RA | math |
Weakly I-clean rings
Ajay Sharma and Dhiren Kumar Basnet
Department of Mathematical Sciences, Tezpur University,
Napaam, Tezpur-784028, Assam, India.
Email: [email protected]
[email protected]
In this article, we introduce the concept of weakly I-clean ring,
Abstract:
for any ideal I of a ring R. We show that, for an ideal I of a ring R, R is
uniquely weakly I-clean if and only if R/I is semi boolean and idempotents
can be lifted uniquely weakly modulo I if and only if for each a ∈ R, there
exists a central idempotent e ∈ R such that either a − e ∈ I or a + e ∈ I and
I is idempotent free. As a corollary, we characterize weakly J-clean ring.
Also we study various properties of weakly I-clean ring.
Key words: Clean ring, I-clean ring, weakly I-clean ring, weakly J-clean ring.
2010 Mathematics Subject Classification: 16N40, 16U99.
1
INTRODUCTION
In this article, all rings are associative ring with unity unless otherwise indi-
cated. Here Jacobson radical of a ring R is denoted by J(R). Set of all units, set of
all nilpotent elements and set of all idempotent elements are respectively denoted
by U (R), N il(R) and Idem(R). A ring R is called abelian if every idempotent
commutes with every element of R. An ideal I of a ring R is said to be idempotent
free if e2 = e ∈ I, then e = 0. For an ideal I of a ring, we say that idempotents can
be lifted (respectively, lifted uniquely, lifted centrally) modulo I, if for any x ∈ R
with x − x2 ∈ I, there exists e ∈ Idem(R) (respectively, unique e ∈ Idem(R),
central e ∈ Idem(R)) such that x − e ∈ I. A ring R is said to be exchange ring if
for any x ∈ R there exists e ∈ Idem(R) such that e ∈ aR and 1 − e ∈ (1 − a)R. In
1977, W. K. Nicholson [4] introduced the concept of clean rings as a subclass of
exchange rings. He defined a ring R to be clean ring if every element of R can be
written as a sum of a unit and an idempotent. Again as a subclass of clean rings
A. J. Diesl [2] introduced the notion of nil clean ring in the year 2013. A ring R is
1
said to be a nil clean ring if for any x ∈ R, x = n + e, for some e ∈ Idem(R) and
n ∈ N il(R). In 2010, H. Chen [1] introduced the notion of J-clean ring as a ring
R, where every element of R can be written as a sum of an idempotent and an
element from the Jacobson radical. V. A. Hiremath and H. Sharad [3] generalised
the concepts of nil clean ring and J-clean ring to I-clean ring in the year 2013. For
an ideal I of a ring R, R is said to be I-clean ring if for any x ∈ R, there exists
e ∈ Idem(R) such that x − e ∈ I and if e is unique in the expression, then the ring
R is said to be uniquely I-clean ring.
Here we introduce the notion of weakly I-clean ring, for any ideal I of a ring
R. For an ideal I of a ring R, we say R is weakly I-clean ring if for any x ∈ R
there exists e ∈ Idem(R) such that x − e ∈ I or x + e ∈ I and if e is unique, then R
is said to be uniquely weakly I-clean ring. We study various properties of weakly
I-clean ring and uniquely weakly I-clean ring.
2 Weakly I-clean ring
Definition 2.1. For an ideal I of R, we say that R is weakly I-clean ring, if for
each x ∈ R there exists e ∈ Idem(R) such that x − e ∈ I or x + e ∈ I. Also R
is said to be uniquely weakly I-clean ring if for any x ∈ R, there exists a unique
e ∈ Idem(R) such that x − e ∈ I or x + e ∈ I.
It is easily seen that every I-clean ring is weakly I-clean. The converse is not
true because Z is weakly 3Z-clean but not 3Z-clean. Note that if R is weakly
I1-clean and not weakly I2-clean then R is not necessarily weakly I1 ∩ I2-clean.
For a commutative ring R, R is weakly nil clean ring if I = N il(R).
Theorem 2.2. Let {Rα} be a collection of rings and Iα 6 Rα. Then R = Q Rα
is weakly I = Q Iα-clean if and only if each Rα is weakly Iα-clean and at most one
Rα is not Iα-clean.
Proof. If R is weakly I clean, then each Rα is weakly Iα-clean. Suppose that
Rα1 and Rα2 (α1 6= α2) are not Iα1-clean and Iα2-clean respectively. There exists
xα1 ∈ Rα1 such that xα1 6= w1 − e1, where w1 ∈ Iα1 and e1 ∈ Idem(Rα1 ).
Similarly there exists xα2 ∈ Rα2 such that xα2 6= w2 + e2, where w2 ∈ Iα2 and
e2 ∈ Idem(Rα2 ). Define x = (xα) ∈ R by xα = xα for α ∈ {α1, α2}, otherwise
xα = 0. Then x 6= w + e or x 6= w − e, for any w ∈ I and e ∈ Idem(R).
Conversely, let each Rα be Iα-clean, then clearly R is I-clean. Assume that Rα0 is
weakly Iα0-clean but not Iα0-clean and other Rα's are Iα-clean. Let x = (xα) ∈ R,
so in Rα0, xα0 = wα0 + eα0 or xα0 = wα0 − eα0 , where wα0 ∈ Iα0 and eα0 ∈
Idem(Rα0 ).
If xα0 = wα0 + eα0, then assume xα = wα + eα (α 6= α0) and if
2
xα0 = wα0 − eα0, then we assume xα = wα + eα (α 6= α0), where wα ∈ Iα and
eα ∈ Idem(Rα). Set w = (wα) and e = (eα), then either x = w+e or x = w−e.
Lemma 2.3. If R is weakly I-clean ring, then J(R) ⊆ I.
Proof. Let x ∈ J(R), then there exists e ∈ Idem(R) such that either x − e ∈ I
or x + e ∈ I. If x − e ∈ I, then x − e = w, so (x − w)2 = x − w, which implies
x(1 − x) ∈ I. But 1 − x is unit, hence x ∈ I.
The converse of Lemma 2.3 is not true as for the ideal I = 5Z in Z, J(Z) =
{0} ⊆ 5Z, but Z is not weakly 5Z-clean. Note that, if R is weakly I-clean, then R
is weakly B-clean for any ideal B of R with I ⊆ B.
Definition 2.4. Let I be an ideal of R. We say that idempotents can be lifted
weakly modulo I if for x2 − x ∈ I, there exists e ∈ Idem(R) such that either
x − e ∈ I or x + e ∈ I. Also we say that idempotents can be lifted uniquely weakly
modulo I if for x2 − x ∈ I, there exists a unique e ∈ Idem(R) such that either
x − e ∈ I or x + e ∈ I.
Theorem 2.5. If I is an ideal of R such that R/I is boolean and idempotents lift
weakly modulo I, then R is weakly I-clean.
Proof. Let x ∈ R. We have x2 − x ∈ I, so there exists e ∈ Idem(R) such that,
either x − e ∈ I or x + e ∈ I. Hence R is weakly I-clean.
Clearly Z is weakly 3Z-clean but Z3 = Z/3Z is not boolean, so the converse of
Theorem 2.5 is not true.
Lemma 2.6. Let R be a ring and e = f + n, where e, f ∈ Idem(R), n ∈ N il(R)
and f ∈ C(R), then n = 0.
Proof. See Lemma 2.5[3].
For any ring R, let Tn(R) be the ring of all upper triangular matrices over
R with usual addition and multiplication. Using Lemma 2.6 we can say that if
idempotents of a ring R are central, then the idempotents of the ring S = {[aij ] ∈
Tn(R) aii = ajj for all i, j} is the set Idem(S) = {eIn e ∈ Idem(R)}.
Theorem 2.7. If R is weakly I-clean ring and idempotents in R are central, then
S = {[aij] ∈ Tn(R) aii = ajj for all i, j} is weakly I ′-clean, where I ′ = {[aij ] ∈
S aii ∈ I}.
3
Proof. Clearly I ′ is an ideal of S. Since idempotents are central, so Idem(S) =
{eIn e ∈ Idem(R)}. Let A = [aij] ∈ S and aii = a. Since R is weakly I-clean, so
there exist e ∈ Idem(R) and w ∈ I such that a = w + e or a = w − e. If a = w + e,
then set A = B + eIn and B = (bij), where bij = w if i = j, otherwise bij = aij.
Hence B ∈ I ′ and eIn ∈ Idem(S). Similarly if a = w − e, then we can prove that
A = B − eIn, where B ∈ I ′ and eIn ∈ Idem(S).
An element x of a ring R is said to be quasi-regular if 1 − x ∈ U (R).
Proposition 2.8. If R is a uniquely weakly I-clean ring for an ideal I of R, then
the following hold:
(i) I is idempotent free.
(ii) I contains all the quasi-regular elements.
(iii) R is abelian.
Proof.
(i) Let e2 = e ∈ I. Since 1 = 1 + 0 = (1 − e) + e, hence e = 0.
(ii) Let a ∈ R be quasi-regular. Since R is uniquely weakly I-clean ring, so
1 − a = (1 − e) + y or 1 − a = −(1 − e) + y, for some e ∈ Idem(R) and y ∈ I.
Now (1 − a)e = ye ∈ I, but 1 − a ∈ U (R) which implies e ∈ I. So by (i),
e = 0 and hence a ∈ I.
(iii) Let e ∈ Idem(R). By (ii) N il(R) ⊆ I. Let x = e + er − ere, clearly
x ∈ Idem(R). Now x = (e + er − ere) + 0 = e + (er − ere), Hence by
uniqueness er − ere = 0 i.e. er = ere. Similarly we can show that re = ere.
Lemma 2.9. Let I be an ideal of a ring R with N (R) ⊆ I. Then the following
are equivalent:
(i) Idempotents can be lifted uniquely weakly modulo I.
(ii) Idempotents can be lifted weakly modulo I, R is abelian and I is idempotent
free.
(iii) Idempotents can be lifted centrally weakly modulo I and I is idempotent free.
Proof. (i) ⇒ (ii) proof is same as Lemma 2.8(1) [3].
(ii) ⇒ (iii) is trivial.
(iii) ⇒ (i)
Let x be an idempotent in R/I. By assumption, there exists a central idempotent
e ∈ R such that either x − e ∈ I or x + e ∈ I. Suppose that f ∈ Idem(R) for
4
which either x − f ∈ I or x + f ∈ I, then
Case I:
If x − e ∈ I and x − f ∈ I or x + e ∈ I and x + f ∈ I, then e − f ∈ I. Similar to
the proof of Lemma 2.8 [3] we can show that e = f .
Case II:
If x − e ∈ I and x + f ∈ I or x + e ∈ I and x − f ∈ I, then e + f ∈ I. As I is an
ideal so (e + f )(1 − e) ∈ I, implies f (1 − e) ∈ I. But e is central idempotent, so
f (1 − e) ∈ Idem(R), i.e. f = f e, as I is idempotent free. Similarly we can show
that e = ef . Hence e = f .
From both the cases we conclude that idempotents can be lifted uniquely weakly
modulo I.
Definition 2.10. A ring R is called semi boolean ring if either x2 = x or x2 = −x,
for all x ∈ R.
Clearly boolean rings are semi boolean rings but converse is not true as Z3 is
semi boolean ring but not boolean.
Theorem 2.11. For an ideal I of a ring R the following are equivalent:
(i) R is uniquely weakly I-clean ring.
(ii) R/I is semi boolean and idempotents can be lifted uniquely weakly modulo I.
Proof. (i)⇒ (ii)
Let x ∈ R/I. Then there exists unique e ∈ Idem(R) such that either x = i + e or
x = i − e, where i ∈ I. If x = i + e, then x2 = x, also if x = i − e, then x2 = −x,
hence R/I is semi boolean. Next part is obvious.
(ii)⇒ (i)
Let x ∈ R, by (ii) either x2 = x or x2 = −x. If x2 = x, then by assumption there
exists unique e ∈ Idem(R) such that either x − e ∈ I or x + e ∈ I. Also if x2 = −x,
then (−x)2 = −x, again by assumption there exists unique e ∈ Idem(R) such that
either −x + e ∈ I or −x − e ∈ I.
A ring R is said to be weakly J-clean ring if for any x ∈ R, x = j + e or
x = j − e, where j ∈ J(R) and e ∈ Idem(R). Also R is said to be uniquely
weakly J-clean ring if for any x ∈ R, there exists a unique e ∈ Idem(R) such that
x − e ∈ J(R) or x + e ∈ J(R).
Corollary 2.12. For a ring R, the following are equivalent.
(i) R is uniquely weakly J-clean.
5
(ii) R/J(R) is semi boolean and idempotents can be lifted uniquely weakly modulo
J(R).
A ring R is said to be weakly nil clean ring if for any x ∈ R, x = e + n or
x = −e + n, where n ∈ N il(R) and e ∈ Idem(R). Also R is said to be uniquely
weakly nil clean ring if for any x ∈ R, there exists a unique e ∈ Idem(R) such that
x − e ∈ N il(R) or x + e ∈ N il(R).
Corollary 2.13. For a commutative ring R, the following are equivalent.
(i) R is uniquely weakly nil clean.
(ii) R/N il(R) is semi boolean and idempotents can be lifted uniquely weakly mod-
ulo N il(R).
Corollary 2.14. Let I be an ideal of a ring R. Then the following are equivalent.
(i) R is uniquely weakly I-clean.
(ii) R/I is semi boolean and idempotents can be lifted uniquely weakly modulo I.
(iii) R/I is semi boolean and idempotents can be lifted weakly modulo I, R is
abelian and I is idempotent free.
(iv) For each a ∈ R there exists a central idempotent e ∈ R such that either
a − e ∈ I or a + e ∈ I and I is idempotent free.
Proof. If R/I is semi boolean, then for n ∈ N (R), nk = 0, for some k ∈ N. Either
nk = n or nk = −n but nk = I, implies n ∈ I. Hence by Lemma 2.9 the result
follows.
Proposition 2.15. For an ideal I of a ring R the following are equivalent.
(i) I is prime and R is uniquely weakly I-clean.
(ii) R is uniquely weakly I-clean and 0, 1 are the only idempotents in R.
(iii) R/I ∼= Z2 or Z3 and I is idempotent free.
(iv) I is maximal and R is uniquely weakly I-clean.
Proof. (i)⇒ (ii)
Let e ∈ Idem(R). By Proposition 2.8, R is abelian, so eR(1 − e) = 0 ⊆ I.
Therefore e ∈ I or 1 − e ∈ I, but again by Proposition 2.8, I is idempotent free
and hence e = 0 or e = 1.
(ii)⇒ (iii)
6
From Theorem 2.11, R/I is semi boolean. Let x ∈ R/I, then either x is boolean
or −x is boolean. If x is boolean, then x ∈ I or x − 1 ∈ I and if −x is boolean,
then −x ∈ I or −x − 1 ∈ I. From both the cases, either x = 0 or x = 1 or x = −1.
Hence R/I ∼= Z2 or Z3.
(iii)⇒ (iv)
Assume R/I ∼= Z2 or Z3, then clearly I is a maximal ideal and R/I is semi boolean.
(iv)⇒ (i) It is obvious.
Corollary 2.16. The following are equivalent for a non trivial ring R.
(i) R is local and uniquely weakly J-clean.
(ii) R is uniquely weakly J-clean and 0, 1 are the only idempotents in R.
(iii) R/J(R) ∼= Z2 or Z3
For a ring R, let V be an R − R bimodule, which is a ring not necessarily with
1. Let I(R; V ), the ideal extension of R by V , is defined to be the additive abelian
group I(R; V ) = R ⊕ V , where the multiplication is defined by (r, v)(s, w) =
(rs, rw + vs + vw), for all v, w ∈ V and r, s ∈ R.
Lemma 2.17. Let R be a ring, V be an R-R-bimodule which is also a ring
(not necessarily with unity) and let S = I(R; V ) be the ideal extension.
If V
is idempotent free and for each e ∈ Idem(R), ev = ve for all v ∈ V , then
Idem(S) = {(e, 0) e ∈ Idem(R)}.
Proof. See Lemma 2.22 [3].
Lemma 2.18. Let R be a ring, V be an R-R-bimodule which is also a ring (not
necessarily with 1) and let S = I(R; V ) be the ideal extension. Then the following
are equivalent.
(i) For every v ∈ V , there exists w ∈ V such that v + w + vw = 0.
(ii) (1, v) ∈ U (S) for all v ∈ V .
(iii) U (S) = {(u, v) ∈ S u ∈ U (R)}.
(iv) J(S) = J(R) × V .
Further if any of four equivalent conditions holds, then J(R) = {r ∈ R (r, 0) ∈
J(S)}.
Proof. See Lemma 2.23 [3].
7
Proposition 2.19. Let R be a ring and let V be an R-R-bimodule which is also an
idempotent-free ring not necessarily with 1. Let S = I(R; V ) be the ideal extension
of R by V . Then the following are equivalent.
(i) S = I(R; V ) is uniquely weakly I ′-clean for some ideal I ′ of S.
(ii)
(a) R is uniquely weakly I-clean for some ideal I of R.
(b) If e ∈ Idem(R), then ev = ve for all v ∈ V .
Proof. (i)⇒ (ii)
(b) Let e ∈ Idem(R), then clearly (e, 0) ∈ Idem(S). Since S is weakly I ′-clean,
so by Proposition 2.8, S is abelian. For v ∈ V , (e, 0)(v, 0) = (v, 0)(e, 0) and hence
ev = ve, as required.
(a) Consider I = {x ∈ R (x, 0) ∈ I ′}, then I be an ideal of R. Since S is abelian,
so by Lemma 2.17, Idem(S) = {(e, 0) e ∈ Idem(R)}.
Claim: R is uniquely weakly I-clean ring.
Let x ∈ R. As S is uniquely weakly I ′-clean, so by Corollary 2.14, there exists a
central idempotent (e, 0) ∈ S such that (x, 0) = (e, 0) + (y, 0) or (x, 0) = −(e, 0) +
(y, 0), for some (y, 0) ∈ I ′. Therefore x − e = y ∈ I or x + e = y ∈ I, so R is weakly
I-clean ring. Let e2 = e ∈ I, which implies (e, 0) ∈ I ′. Since I ′ is idempotent free,
so e = 0 and hence I is idempotent free. By Corollary 2.14, R is uniquely weakly
I-clean ring.
(ii) ⇒ (i)
Consider I ′ = I × V , then I ′ is an ideal of S. Let (r, v) ∈ S. Since R is uniquely
weakly I-clean ring, so there exists a central idempotent e ∈ R such that r − e ∈ I
or r + e ∈ I. Now (r, v) = (e, 0) + (r − e, v) or (r, v) = −(e, 0) + (r + e, v) implies
(r, v) − (e, 0) ∈ I ′ or (r, v) + (e, 0) ∈ I ′ and hence S is weakly I ′-clean ring. By (b)
and 2.17, we have Idem(S) = {(e, 0) ∈ Idem(R)}. For (e, 0)2 = (e, 0) ∈ I ′, e ∈ I,
but by Corollary 2.14, I is idempotent free, so (e, 0) = (0, 0). Hence S is uniquely
weakly I ′-clean ring.
3 Acknowledgement
The first Author was supported by Government of India under DST(Department
of Science and Technology), DST-INSPIRE registration no IF160671.
References
[1] Chen, H. On strongly J-clean rings, Comm. Algebra, 38(10) : 3790 − 3804,
2010.
8
[2] Diesl, A. J. Nil clean rings, J. Algebra 383: 197 − 211, 2013.
[3] Hiremath, V. A. and Sharad, H. Using ideals to provide a unified approach
to uniquely clean rings. J. Aust. Math. Soc., 96(2): 258 − 274, 2013.
[4] Nicholson, W. K., Lifting idempotents and exchange rings, Trans. Amer.
Math. Soc., 229: 269 − 278, 1977.
9
|
1806.08580 | 1 | 1806 | 2018-06-22T09:58:13 | Gradings on the real form $\mathfrak{e}_{6,-14}$ | [
"math.RA"
] | Six fine gradings on the real form $\mathfrak{e}_{6,-14}$ are described, precisely those ones coming from fine gradings on the complexified algebra. The universal grading groups are $\mathbb Z_2^3\times\mathbb Z_3^2$, $\mathbb Z_2^6$, $\mathbb Z\times\mathbb Z_2^4$, $\mathbb Z_2^7$, $\mathbb Z\times\mathbb Z_2^5$ and $\mathbb Z^2\times\mathbb Z_2^3$. | math.RA | math |
GRADINGS ON THE REAL FORM e6,−14
CRISTINA DRAPER∗ AND VALERIO GUIDO
Abstract. Six fine gradings on the real form e6,−14 are described, precisely
those ones coming from fine gradings on the complexified algebra. The uni-
versal grading groups are Z3
3, Z6
2, Z7
2 × Z2
2, Z × Z4
2, Z × Z5
2 and Z2 × Z3
2.
1. Introduction
This work continues a series of papers devoted to describing the fine gradings
on the exceptional real Lie algebras. The interest of gradings has been present
in the theory of Lie algebras from the very beginning. Let us recall that, at the
end of the nineteenth century, W. Killing used the root decomposition (which
is, in fact, a grading over the free-torsion abelian group Zl for l the rank of the
algebra) in order to classify complex finite-dimensional semisimple Lie algebras.
More recently, the physicist J. Patera and his collaborators proposed to start a
systematic study of the gradings on Lie algebras in [PZ89]. This seminal work
was followed by some others for real classical Lie algebras, as [HPPe00] or [Sv08].
Since then, many authors shifted the attention to the complex case. A remarkable
amount of these works was compiled in [EK]. This monograph was published in
2013 and it contains almost completely the classification of the gradings on the
complex simple finite-dimensional Lie algebras. Precisely, the classification of
the gradings on type-E algebras is only conjectured in the book, and had to wait
until [DrV16], [DrE17] (finite groups) and [Y16] to be completed. A review of the
fine gradings on Lie algebras of type-E is [DrEl14], which tries to give a version
as unified as possible of the gradings on the three complex Lie algebras of type
E. Having finished the complex case, it is the moment to return to the real case,
which is relevant for many applications. The problem of getting a classification
of the gradings on real simple Lie algebras is being tackled by several authors
almost simultaneously. On one hand, some recent papers devoted to the classical
simple real Lie algebras are [EK18], [BKR18a] and [BKR18b]. On the other hand,
as mentioned, this is the third of our papers about exceptional real Lie algebras:
first, in [CalDrM10], we classified fine gradings on the real algebras of types G2
and F4, and second, in [DrG16b], we described the fine gradings on the algebra
e6,−26 coming from fine gradings on the complex algebra e6. Let us recall that
there are three non-split and non-compact real algebras of type E6, characterized
2010 Mathematics Subject Classification. Primary 17B70; Secondary 17B25, 17B60.
Key words and phrases. Gradings, exceptional Lie algebras, real forms, signature −14.
∗ Partially supported by MCYT grant MTM2016-76327-C3-1-P and by the Junta de An-
daluc´ıa PAI project FQM-336.
1
2
C. DRAPER AND V. GUIDO
by the signatures of their Killing forms, namely, -26, -14 and 2. Here we focus
on e6,−14, an algebra with frequent appearances in Physics (see below). The
difficulty in defining explicitly e6,−14 renders extremely useful the knowledge of
its structure, gradings, and different models. Such gradings and models could
be potentially related with different physical phenomena. The gradings on e6,−14
have nothing to do with the gradings on e6,−26 obtained in [DrG16b]. Indeed,
our main result, Theorem 1, states that there are 6 fine gradings (and essentially
only 6) on e6,−14, whose universal grading groups are Z3
2, Z7
2,
Z × Z5
2. In contrast, the fine gradings occurring on e6,−26 are 4,
with universal grading groups Z6
2. Only two pairs
of those gradings have the same origin, i.e., isomorphic complexifications (the
two Z2 × Z3
2-gradings have quite different properties). Regarding the tools used
here for finding the gradings on e6,−14, they are indeed related to the tools in
[DrG16b], but new approaches have been necessary too. For each grading, we
have tried to do a self-contained treatment, providing a related suitable model
of the algebra e6,−14.
2 × Z2
2 and Z4 × Z4
2 and Z2 × Z3
2, Z × Z4
3, Z6
2, Z × Z4
2, Z2 × Z3
The relation between graded Lie algebras and Physics is well-known, see, for
instance, the classical paper [CoNSt] of 1975. Some recent reference is [At15], de-
voted to the applications of graded Lie algebras for studying dynamic symmetries
of atomic nucliei. A survey of algebraic methods that are widely used in nuclear
and molecular physics is [I97], which states a spectrum generating algebra as the
basic mathematical tool of all such methods, and presents some models where
such spectrum generating algebra is just a graded Lie algebra. More applica-
tions of the gradings on Lie algebras to mathematical physics and particularly
to particle physics are referenced in our previous work [DrG16b]. To illustrate
them with some examples, recall first the Pauli matrices, which provide a fine
Z2
In quantum mechanics, these matrices occur in the Pauli
equation which takes into account the interaction of the spin of a particle with
an external electromagnetic field. On the other hand, the Gell-Mann matrices
provide a fine Z3
2-grading on the compact Lie algebra su3 (see Section 6). The
Gell-Mann matrices are used in the study of the strong interaction in particle
physics, and they serve to study the internal (color) rotations of the gluon fields
associated with the colored quarks of quantum chromodynamics. Also the gen-
eralized Pauli matrices of order 3 provide a Z2
3 , but note that it
has been necessary to complexify here, since the only groups being the grading
groups of a compact Lie algebra are the direct product of copies of Z2.
3-grading on suC
2-grading on su2.
Moreover, the role of the Lie algebras of type E6 in Physics is also quite relevant
[W95], giving rise to a large number of references, some of them appearing in
[DrG16b]. We can add, for instance, [KL05], on the discovery limits of different
E6 models at some colliders (as Tevatron); [GV05], on order-two twisted D-branes
of E6; [HM06], which proposes a model of dark energy and dark matter based
on E6 very different from the usual E6-unification encountered in the literature;
[MPW91], which proposes U28 → SU27×U1 → E6 → G2 → SO3 for the extension
of the interacting boson model; [DuF07], on N = 8 black holes in five dimensions
FINE GRADINGS ON e6,−14
3
where the common symmetry is E6; or [GLM07], which analyses heterotic string
compactifications on specific classes of manifolds with SU(3)-structure, proving
that E6 is still the resulting gauge group in four dimensions.
Finally, we would also like to mention the growing number of specific appear-
ances of the real form e6,−14 in the literature. This algebra appears as U-duality
algebra when considering the theories coupled to gravity for N = 5 supersym-
metries and D = 3 spacetime dimensions ([HPS08, §7.3]). Also, according to
[GPR18, Proposition 7.1], the Toledo invariant of e6,−14 corresponding to the
Hitchin-Kostant-Rallis section for the moduli space of its Higgs bundles is zero.
Relative to incidence geometry, a recent paper concerned with the group E6,−14
is [KGK10]. It describes a projective embedding (called a Veronese embedding)
of the building of the group E6,−14 (a generalized quadrangle) over the reals,
trying to follow the spirit of Freudenthal's description of the Cayley plane. The
paper [Do14] reviews the progress of the classification and construction of in-
variant differential operators for non-compact semisimple Lie groups, including
many references about the importance of these invariant differential operators in
the description of physical symmetries. As the study of such classification should
proceed group by group, the author chooses some groups. This choice, which in-
cludes E6,−14, is supported by these groups being non-compact groups that have
a discrete series of representations (the rank of the group G coincides with the
rank of the maximal compact subgroup K), and, besides, G/K belongs to the
list of Hermitian symmetric spaces (each one admits a complex structure which is
invariant by the group of isometries). The two exceptional cases in such list (the
list of possible groups G with G/K irreducible non-compact Hermitian symmet-
ric space) are E6,−14 and E7,−25, corresponding to the only Lie exceptional real
algebras with highest weight representations. The same author devotes the work
[Do09] to study invariant differential operators focused on E6,−14. Amazingly,
the symmetric space E7,−25/E6,−14 appears in [BFGM06] corresponding to the
orbit of a large black hole with central charge equal to zero. It is just one of the
three species of the solutions to the black hole attractor equations giving rise to
different mass spectra of the scalar fluctuations. These authors have devoted a
considerable amount of papers to study exceptional Lie algebras in relation to
physical theories. In their own words ([MT16]): "While describing the results of
our recent work on exceptional Lie and Jordan algebras, so tightly intertwined in
their connection with elementary particles, we will try to stimulate a critical dis-
cussion on the nature of spacetime and indicate how these algebraic structures
can inspire a new way of going beyond the current knowledge of fundamental
physics."
Outline. In Section 2, we summarize the basics about gradings on a Lie algebra,
while recalling the classification of the fine gradings on e6 obtained in [DrV16].
There are 14 fine gradings on the complex Lie algebra e6, and some invariants are
provided, including the universal groups. Also, the necessary background about
real forms is recalled, and some notations and results about signatures are stated.
In Section 3, two gradings on e6,−14 over the groups Z6
2 are constructed
2 and Z×Z4
4
C. DRAPER AND V. GUIDO
2 × Z2
based on a model of e6,−14 as a sum of the set of traceless elements of certain
Albert algebra with the Lie algebra of derivations of such Albert algebra (a non-
split and non-compact real form of f4). Section 4 provides a Z3
3-grading on
e6,−14 based on the Tits' construction. This requires to determine the signatures
of the real forms obtained when applying Tits' construction to different Jordan
and composition algebras. In Section 5, we exhibit a model of e6,−14 based on a
contact Z-grading, in which two gradings over the groups Z × Z5
2 and Z2 × Z3
2
are considered. The last fine grading is studied in Section 6. Similarly to the
2-grading on su2 and to the Z3
Z2
2-grading on su3, related to the Pauli matrices
and to the Gell-Mann matrices respectively, we find now a Z7
2-grading on e6,−14.
This provides a basis of e6,−14 with interesting properties (Corollary 1). Most of
the real simple (finite-dimensional) Lie algebras of rank l admit a fine grading
over the group Zl+1
, with very few exceptions, as for instance e6,−26 which is not
Z7
2-graded [DrG16b, Proposition 11]. The last section is devoted to prove that
there do not exist fine gradings on e6,−14 whose complexification is any of the
remaining 8 fine gradings on the complex algebra e6. This is a case-by-case proof
for some of the gradings on e6, based on the knowledge of their features and of
compatible subalgebras. Some representation theory of real Lie algebras is used
too. We finish with some open problems: apart from the mathematical questions
(not difficult to tackle although very technical), we would feel deeply interested
in finding the physical meaning of the obtained gradings, specially those ones
involving groups with 3-torsion.
2
2. Preliminaries
We review now the basic concepts about gradings and real forms. Most of
them appear in [DrG16b], but we sketch them here for self-containedness.
2.1. About gradings. The main reference about the topic of gradings on Lie
algebras is the monograph [EK].
Let A be a finite-dimensional algebra over a field F ∈ {R, C}, and G an abelian
group. A G-grading Γ on A is a vector space decomposition
(1)
Γ : A =Mg∈G
Ag
such that AgAh ⊂ Ag+h for all g, h ∈ G. The subspace Ag is called homogeneous
component of degree g and its elements homogeneous elements of degree g. The
support of the grading is the set Supp Γ := {g ∈ G : Ag 6= 0}. The type of Γ
is the sequence of integers (h1, . . . , hr), where hi is the number of homogeneous
components of dimension i, with i = 1, . . . , r and hr 6= 0. Obviously, dimA =
Pr
i=1 ihi.
If Γ : A = ⊕g∈GAg and Γ′ : A = ⊕h∈HA′
h are gradings over two abelian groups
G and H, Γ is said to be a refinement of Γ′ (or Γ′ a coarsening of Γ) if for
any g ∈ G, there is h ∈ H such that Ag ⊂ A′
h. A refinement is proper if
some inclusion Ag ⊂ A′
h is proper. A grading is said to be fine if it admits
no proper refinement. Also, Γ and Γ′ are said to be equivalent if there is a
FINE GRADINGS ON e6,−14
5
bijection α : Supp Γ → Supp Γ′ and ϕ ∈ Aut(A) such that ϕ(As) = A′
α(s) for
all s ∈ Supp Γ. Note that in this case the grading groups G and H are not
necessarily isomorphic groups. For any group grading Γ on A, there exists a
distinguished group among the ones which are grading groups of gradings on A
equivalent to Γ such that they are generated by the support of the grading (more
details in [EK]). It is usually called the universal (grading) group of Γ.
If the ground field is the complex field, there is a close relationship between
automorphisms and gradings. First, it is obvious that every diagonalizable au-
tomorphism f ∈ Aut(A) produces a grading on A over the subgroup of C×
generated by the spectrum Spec(f ), and also that any collection of commut-
ing diagonalizable automorphisms produces a grading where the homogeneous
components are the simultaneous eigenspaces. Conversely, if Γ is a G-grading
as in (1) and we have any character χ ∈ X(G) = Hom(G, C×), then the map
ϕχ : A → A is an automorphism of A, being ϕχ(x) := χ(g)x for any g ∈ G and
x ∈ Ag. Moreover ϕχ belongs to the diagonal group of Γ, defined as follows:
Diag(Γ) := {ϕ ∈ Aut(A) : ∀g ∈ G, ∃αg ∈ C× such that ϕAg = αg id}.
Then X(Diag(Γ)) is precisely the universal group of Γ. Furthermore, Γ is a
fine grading if and only if Diag(Γ) is a maximal abelian diagonalizable subgroup
(usually called a MAD-group) of Aut(A) ([PZ89, Theorem 2]). For instance, if
S is a (complex) semisimple Lie algebra, and Γ : S = Pα∈Φ∪{0} Sα is the root
decomposition relative to a Cartan subalgebra, which is obviously a fine grading,
then Diag(Γ) = T is a maximal torus of the automorphism group, namely, if
{α1, . . . , αl} is a set of simple roots of Φ,
(2)
where the automorphism ts1,...,sl ∈ Aut(S) acts scalarly on the root space Sα
with eigenvalue sk1
i=1 kiαi. Thus the universal group is
X(Diag(Γ)) ∼= Zl.
T = {ts1,...,sl : si ∈ C×} ∼= (C×)l
l , for α = Pl
1 . . . skl
The fine gradings on the complex algebra S = e6 were classified up to equiv-
alence in [DrV16], that is, the MAD-groups of Aut(e6) were classified up to
conjugation. According to this classification, there are 14 non-equivalent fine
gradings on S, whose universal grading groups and types are listed in Table 1.
Note that these two data together give the equivalence class of the grading. We
Sg, which will
have also added a column with the interval dimC Se ± dimCP2g=e
be a tool to apply Proposition 2 lately.
g6=e
2.2. About real forms. A real Lie algebra L is called a real form of a complex
Lie algebra S if LC = L ⊗R C = L ⊕ iL = S. Recall that L is semisimple if its
Killing form κL is non-degenerate. As, for any x, y ∈ L, κL(x, y) = κLC(x, y),
consequently the semisimplicity of L equivales to that one of LC. We will abuse
of notation calling the signature of L (and denoting sign(L)) to the signature of
κL, that is, the difference between the number of positive and negative entries
in the diagonal of any matrix of the symmetric bilinear form κL relative to
6
C. DRAPER AND V. GUIDO
Grading Universal group
Type
Γ1
Γ2
Γ3
Γ4
Γ5
Γ6
Γ7
Γ8
Γ9
Γ10
Γ11
Γ12
Γ13
Γ14
Z4
3
3
Z2 × Z2
3 × Z3
Z2
Z2 × Z3
2
2
Z6
Z4 × Z2
Z6
2
2
Z × Z4
Z3
3 × Z2
Z2 × Z3
Z4 × Z4
Z × Z5
2
2
2
Z7
2
Z3
4
(72, 0, 2)
(60, 9)
(64, 7)
(48, 1, 0, 7)
(72, 0, 0, 0, 0, 1)
(72, 1, 0, 1)
(48, 1, 0, 7)
(57, 0, 7)
(26, 26)
(60, 7, 0, 1)
(48, 13, 0, 1)
(73, 0, 0, 0, 1)
(72, 0, 0, 0, 0, 1)
(48, 15)
Table 1. Fine gradings on e6
Interval
0 ± 0
2 ± 0
0 ± 14
2 ± 28
6 ± 0
4 ± 2
0 ± 78
1 ± 29
0 ± 0
2 ± 16
0 ± 46
1 ± 35
0 ± 78
0 ± 14
an orthogonal basis. For most of the complex simple Lie algebras S, including
S = e6, two real forms of S are isomorphic if and only if their signatures coincide,
due to the correspondence (3) below. There always exists a compact real form,
with negative definite Killing form, and a split real form, with signature equal to
the rank of S, that is, the dimension of the Cartan subalgebra. In case S = e6,
there are five real forms up to isomorphism, characterized by their signatures,
namely: −78, −26, −14, 2 and 6. We will use the notation e6,s to refer to any
real form of e6 of signature s, without specifying a concrete representative of the
isomorphism class. So, e6,−78 is compact and e6,6 is split.
There is a strong relationship between real forms of a complex Lie algebra S
and involutive automorphisms of S. If L is a real form of S = L ⊕ iL, then we
can consider the map
σ : S → S,
σ(x + iy) = x − iy,
(x, y ∈ L)
which is a conjugation (conjugate-linear order two map) such that Sσ := {x ∈ S :
σ(x) = x} equals L. Moreover, two real forms Sσ1 and Sσ2 are isomorphic (real)
Lie algebras if and only if there is ϕ ∈ Aut(S) such that ϕσ1ϕ−1 = σ2. If CS and
AS denotes the set of conjugations of S and the set of involutive automorphisms
of S, respectively, then
(3)
CS/ ∼= −→ AS/ ∼=,
[σ] 7→ [θσ],
FINE GRADINGS ON e6,−14
7
is well defined and bijective, where if σ is any conjugation of S, we denote by
θσ ∈ Aut(S) any involutive automorphism commuting with σ such that the
conjugation θσσ is compact. (τ is called a compact conjugation if the real form
Sτ is compact.) Furthermore,
sign(Sσ) = dim S − 2 dim fix(θσ),
(4)
where, for θ ∈ Aut(S), we denote the fixed subalgebra by fix(θ) = {x ∈ S :
θ(x) = x}. In particular, Eq. (3) implies that there are 4 order two automor-
phisms of e6 up to conjugation, characterized by the isomorphy class of their
fixed subalgebra. This correspondence for S = e6 is detailed in Table 2.
sign(Sσ)
dim fix(θσ)
fix(θσ)
2
38
−78
78
E6
−26
52
F4
Table 2. Automorphisms versus signatures
A5 ⊕ A1
D5 ⊕ C
−14
46
6
36
C4
The order two automorphisms of the complex simple Lie algebras were classi-
fied by ´E. Cartan in [Ca27]. This gives the corresponding real forms by (3). In
case S is a classical (complex) simple Lie algebra, the possibilities for a real form
L are:
An If S = sln+1(C),
signature −(n2 + 2n) + 4pq;
◦ sup,q = {x ∈ sln+1(C) : Ip,qx + ¯xtIp,q = 0}, with p + q = n + 1, of
◦ sln+1(R), of signature n;
◦ slm(H) = {x ∈ glm(H) : Re(tr(x))} = 0, with odd n = 2m − 1 > 1,
of signature −2m − 1 = −n − 2.
Bn, Dn If S ∈ {so2n+1(C), so2n(C)},
◦ sop,q(R) = {x ∈ glp+q(R) : Ip,qx + xtIp,q = 0}, with either p + q =
2n + 1 (type Bn) or p + q = 2n (type Dn), of signature (p+q)−(p−q)2
;
◦ so∗
2n(R) = {x ∈ gln(H) : xth + h¯x = 0}, where h = diag(i, . . . , i), of
signature −n. This case only happens when n > 4 (Dn).
2
Cn If S = sp2n(C),
◦ sp2n(R) = {x ∈ gl2n(R) : x + ¯xt = 0}, of signature n;
◦ spp,q(H) = {x ∈ gln(H) : Ip,qx + ¯xtIp,q = 0}, with p + q = n, of
signature −2(p − q)2 − (p + q).
Here Ip,q = diag(1, . . . , 1,−1, . . . ,−1) (the first p entries in the diagonal are 1
and the q remaining entries are −1) and In ≡ In,0 is so the identity matrix. Also
we denote by son(R) ≡ son,0(R), and similar conventions are used for sp and su,
and for C and H. Each real form will also be denoted according to its signa-
ture and the type of the complexification, for instance, slm(H) ≡ a2m−1,−2m−1.
(In particular, a2m−1,−2m−1 denotes also any algebra in the isomorphy class of
slm(H).) Throughout the text, Eij will denote a matrix, of size depending on
8
C. DRAPER AND V. GUIDO
the context, where the only non-zero entry will be the (i, j)th, equal to 1. So
diag(s1, . . . , sn) means Pn
i=1 siEii.
Note that the signatures of the subalgebras of a Lie algebra are related, in a
certain way, to the signatures of the whole Lie algebra, although the Killing form
of the subalgebra does not coincide with the restriction of the Killing form of the
algebra. To this aim, we state the next result for complex Lie algebras.
Lemma 1. Let S0 be a simple subalgebra of a complex Lie algebra S. Then there
is a positive rational number r ∈ Q>0 such that κSS0 = r κS0.
Proof. If g is a simple Lie algebra and f1, f2 : g× g → C are non-zero g-invariant
maps, then fi : g → g∗, x 7→ fi(x,−) are isomorphisms of g-modules and hence
−1 f1 ∈ Homg(g, g). Now φ ∈ C idg (Schur's lemma) because any eigenspace
φ = f2
must be a non-zero g-submodule of the adjoint module g, which is irreducible,
so the eigenspace is the whole g. If we apply this fact to g = S0, f1 = κSS0 and
f2 = κS0, we get that there exists r ∈ C such that κSS0 = r κS0.
Let us denote by Φ a root system of S0 relative to a Cartan subalgebra h and
{αi}l
i=1 a set of simple roots of Φ. Thus h is spanned by {hαi}l
i=1, which satisfy
λi(hαj ) = δij for λi the fundamental dominant weights. As h = hα1 ∈ S0, in par-
ticular κSS0(h, h) = r κS0(h, h). The last Killing form is easy to compute since
ad h acts scalarly on each root space (S0)α with scalar α(h) = hα, α1i ∈ Z, so that
κS0(h, h) = Pα∈Φ α2(h) is a non-negative integer. In fact, it is positive, since
α1(h) = 2. Furthermore, for any representation ρ : S0 → gl(V ), the endomor-
phisms ρ(h) are simultaneously diagonalizable, and the simultaneous eigenspaces
are the weight spaces [H]. If we denote by Λ(S) ⊂ h∗ the weights of the S0-module
S, then Λ(S) ⊂ Pl
i=1 Zλi, so that κS(h, h) = Pµ∈Λ(S) µ2(h) ∈ Z≥0 and hence
r ∈ Q≥0. Besides, as Φ ⊂ Λ(S), then κS(h, h) = κS0(h, h) +Pµ∈Λ(S)\Φ µ2(h) is
strictly positive, and so r is.
(cid:3)
This gives immediately the required facts for signatures of simple subalgebras.
Lemma 2. Let L0 be a simple subalgebra of a real Lie algebra L. Then there
is a positive rational number r ∈ Q>0 such that κLL0 = r κL0. In particular,
sign(κLL0) = sign(L0).
Proof. The first result is clear from Lemma 1, since κLL0 = κLCL0 and also
κL0 = κLC
0L0. For the claim relative to the signatures, observe that the key point
is that the scalar r is positive.
(cid:3)
2.3. About gradings on real forms. If L is a real form of a complex Lie
algebra S, there is a close relationship between gradings on L and on S.
If Γ : L = ⊕g∈GLg is a grading on a real Lie algebra L, let us denote by ΓC
the grading on LC given by ΓC : LC = ⊕g∈G(Lg)C = ⊕g∈G(Lg ⊕ iLg). It will be
called the complexified grading. It is important to observe that, if ΓC is a fine
grading on LC, then Γ is a fine grading on L.
FINE GRADINGS ON e6,−14
9
Let ΓS : S = ⊕g∈GSg be a grading on a complex Lie algebra S and let L be a
real form of S = LC. We will say that L inherits the grading ΓS if there exists
a grading Γ on L such that ΓC = ΓS, that is, L = ⊕g∈G(L ∩ Sg).
By abuse of notation, we will say that e6,−14 inherits the grading Γi in Table 1
(i = 1, . . . , 14) if some real form of e6 of signature −14 has a grading whose
complexified grading is equivalent to Γi.
A useful criterion to check if a determined real form of a complex Lie algebra
inherits a grading, assumed we know another real form inheriting it, can be
stated as a direct consequence of [CalDrM10, Proposition 3]:
Proposition 1. Let σ be a conjugation of S such that Sσ inherits certain fine
grading Γ : S = ⊕g∈GSg. For any other conjugation µ of S, the real form Sµ
inherits Γ if and only if µσ−1 ∈ Diag(Γ) ≤ Aut(S).
Gradings and Killing forms are closely related. For instance, if L = ⊕g∈GLg is
a grading over an abelian group G, then two homogeneous components Lg and
Lh are orthogonal for the Killing form unless g + h is the neutral element e ∈ G.
In particular, the homogeneous components corresponding to degrees of order
not 2 are totally isotropic subspaces.
Lemma 3. For L = L¯0 ⊕ L¯1 a real Z2-graded algebra and 0 6= t ∈ R, denote by
Lt := (L, [ ,
]t) the Lie algebra with the same underlying vector space but new
product given by
[x0 + v0, x1 + v1]t = [x0, x1] + [x0, v1] + [v0, x1] + t[v0, v1],
(5)
if xi ∈ L¯0 and vi ∈ L¯1. Then, for ε(t) = t/t ∈ {±1}:
a) sign(Lt) = sign(κLL¯0) + ε(t) sign(κLL¯1),
b) sign(L) + sign(L−1) = 2 sign(κLL¯0),
c) sign(L) + sign(L−1) = 2 sign(L¯0) in case L¯0 is simple.
Proof. For item a), note that κLt(x0, x1) = κL(x0, x1); κLt(x0, v1) = 0 = κL(x0, v1)
and κLt(v0, v1) = t κL(v0, v1). Item b) is obtained when adding the two equations
obtained in item a) for t = 1 and t = −1. Item c) is consequence of item b) and
Lemma 2: sign(κLL¯0) = sign(L¯0).
This result relates the signature of a Z2-graded Lie algebra L = L¯0 ⊕ L¯1 with
that one of L¯0⊕iL¯1 = L−1, which is a different real form of LC. This is important
because any grading on L which is a refinement of such Z2-grading is immediately
a grading of the other real form L−1 (and conversely).
(cid:3)
Finally, we put our attention on the next result, extracted from [DrG16b,
Proposition 9], which bounds the admissible signatures of a real form inheriting
a determined grading.
Proposition 2. Let S be a complex simple Lie algebra and L a real form of S.
Suppose that L inherits a fine grading on S given by S =Pg∈G Sg. Then
dim Sg.
sign(L) − dim Se ≤ Xe6=g∈G
2g=e
10
C. DRAPER AND V. GUIDO
e5
e7
e2
e3
e4
e1
e6
Figure 1. Fano plane
This is the reason why we have added, in the last column in Table 1, the
Sg for each grading Γi.
possible interval for the signature dimC Se ± dimCP2g=e
2.4. About related structures. Throughout this paper we will use several
non-associative algebras.
g6=e
First, we denote by O the octonion algebra, which is the real division algebra
endowed with a norm n such that {1, ei : 1 ≤ i ≤ 7} is an orthogonal basis and
the product is given by recalling Fano plane in Figure 1: each triplet {ei, ej, ek}
in one of the seven lines (the circle is also a line) spans a subalgebra isomorphic
to R3 with the usual cross product. (Thus Rh1, ei, ej, eki is isomorphic to the
quaternion algebra H when ei, ej and ek are the three different elements in one
line. That is, eiejek = −1 in the sense of the arrows.) Recall that a := r1−P siei
if a = r1 +P siei, so that the norm and the trace are given by n(a) = aa and
tO(a) = a + a. The subspace of zero trace octonions (or imaginary octonions) is
denoted by O0.
Second, an F-algebra is called a Jordan algebra if it is commutative and it
satisfies the Jordan identity (x2 · y)· x = x2 · (y · x). The examples of real Jordan
algebras more relevant for our purposes are
Jc = H3(O, γ1), J = H3(O, γ2), M = H3(C, γ3), Ms = H3(R ⊕ R, γ1);
where if C is either O, or C, or R⊕R (with the exchange involution (a, b) = (b, a)),
H3(C, γ) := {x ∈ Mat3×3(C) : γ ¯xtγ−1 = x},
the so called symmetrized product is given by
(6)
x · y :=
1
2
(xy + yx),
and γ is either γ1 = I3 or γ2 = diag(1,−1, 1) or γ3 = E11 + E23 + E32. (Note
that M = H3(C, γ3) is of course isomorphic to the Jordan algebra H3(C, γ2),
but our preference is due to the fact the gradings in Eq. (11) will be more easily
formulated in terms of M.)
If A is an associative algebra, A+ denotes the Jordan algebra obtained when
considering the symmetrized product (6) on the vector space A. Observe that
Ms is isomorphic to Mat3×3(R)+.
FINE GRADINGS ON e6,−14
11
3. From the Albert algebra to gradings over Z6
2 and Z × Z4
2
3.1. The model. Nathan Jacobson described all the real forms of e6 in his book
about exceptional Lie algebras [Ja, Eq. (147)]. The five of them can be obtained
as L = (Der(J) ⊕ J0)± for J a real form of the Albert algebra J C
c , which is the
only exceptional complex Jordan algebra, of dimension 27. The product on L is
given by
• The natural action of Der(J) on J0;
• If x, y ∈ J0, [x, y] := ±[Rx, Ry] ∈ Der(J), where Rx : J → J denotes the
operator multiplication Rx(y) = x · y.
Thus L becomes a Z2-graded Lie algebra in both cases ±, with even part L¯0 =
Der(J) and odd part L¯1 = J0. A first model of e6,−14 is
L = (Der(J ) ⊕ J0)−, for J = H3(O, diag{1,−1, 1}).
Indeed, the even and the odd part of any Z2-graded Lie algebra are orthogonal
for the Killing form κ of L, so sign(L) = sign κL¯0 + sign κL¯1. On one hand,
Der(J ) ∼= f4,−20, and, by Lemma 2, the identity sign(κL¯0) = sign(L¯0) = −20
holds. On the other hand, the traceform of the Jordan algebra J ((x, y) 7→
tr(x· y)) has signature equal to −6 [Ja, p. 114], which coincides with − sign κL¯1,
due to choice of sign when multiplying two odd elements (see also Lemma 3a).
3.2. The gradings. All the G-gradings on J induce naturally G-gradings on
Der(J ) and G × Z2-gradings on L. The Jordan algebra J is Z5
2-graded and
Z × Z3
2-graded [CalDrM10, Theorem 7 and Corollary 1(3)]. Furthermore, the
complexifications of the Z6
2-grading induced on L are
In particular, the real form e6,−14 admits a fine
just Γ7 and Γ8, respectively.
Z6
2-grading as well as a fine Z × Z4
The fine gradings on J were recalled in [DrG16b] since (DerJ ⊕J0)+ has just
signature −26, and hence both e6,−26 and e6,−14 share the gradings coming from
gradings on the Jordan algebra J . For completeness we enclose a description
here.
2-grading and of the Z × Z4
2-grading.
The octonion algebra O is Z3
2-graded:
(7)
O(¯1,¯0,¯0) = Re1, O(¯0,¯1,¯0) = Re2, O(¯0,¯0,¯1) = Re7.
This induces a Z3
2-grading on the Jordan algebra J by means of
(8)
where an arbitrary element in J is denoted by
J(¯0,¯0,¯0) =Pi REii ⊕Pi Rιi(1),
if e 6= g ∈ Z3
Jg =Pi ιi(Og)
2,
=:Xi
siEii +Xi
a3
s2
s1
¯a2
−¯a3
a1
a2 −¯a1 s3
ιi(ai),
12
C. DRAPER AND V. GUIDO
if si ∈ R and ai ∈ O. The grading (8) is compatible with the Z2
defined by
2-grading on J
J(¯0,¯0) =Pi REii,
J(¯1,¯0) = ι2(O),
J(¯0,¯1) = ι1(O),
J(¯1,¯1) = ι3(O),
getting then the mentioned Z5
2-grading on J .
Also, the Z × Z3
2-grading on J is obtained by refining the Z3
2-grading (8) with
the next Z-grading on J :
J−2 = R(E22 − E33 − ι1(1)),
J−1 = {ι2(a) − ι3(a) : a ∈ O},
J0 = h{E11, E22 + E33, ι1(a) : a ∈ O0}i,
J1 = {ι2(a) + ι3(a) : a ∈ O},
J2 = R(E22 − E33 + ι1(1)),
which is just the eigenspace decomposition relative to the derivation 4[Rι1(1), RE22] ∈
Der(J ).
4. A Z3
2 × Z2
3-grading based on the Tits construction
4.1. The model. In 1966 [T66], Tits provided a beautiful unified construction
of all the exceptional simple Lie algebras. We review here only some particular
case of Tits' construction to get a (not usual) model of e6,−14.
Consider, for the Jordan algebra M = H3(C, γ3), the vector space
(9)
T (O,M) := Der(O) ⊕ (O0 ⊗ M0) ⊕ Der(M),
which is made into a Lie algebra over R by defining the multiplication (bilinear
and anticommutative) which agrees with the ordinary commutator on the Lie
algebras Der(O) and Der(M) and it satisfies
(10)
• [ Der(O), Der(M)] = 0,
• [d, a ⊗ x] = d(a) ⊗ x,
• [D, a ⊗ x] = a ⊗ D(x),
• [a ⊗ x, b ⊗ y] = 1
3 tr(x · y)da,b + [a, b] ⊗ (x ∗ y) + 2tO(ab)[Rx, Ry],
for all d ∈ Der(O), D ∈ Der(M), a, b ∈ O0 and x, y ∈ M0. The used notations
are, for a, b, c ∈ O and x, y ∈ M,
◦ [a, b] = ab − ba ∈ O0,
◦ da,b := [la, lb] + [la, rb] + [ra, rb] ∈ Der(O), for la(b) = ab and ra(b) = ba the
left and right multiplication operators in O respectively, so that da,b(c) =
[[a, b], c] + 3(ac)b − 3a(cb);
◦ x ∗ y := x · y − 1
3 tr(x · y)I3 ∈ M0,
and again Rx is the multiplication operator in the Jordan algebra and [Rx, Ry]
the commutator.
Proposition 3. T (O,M) ∼= e6,−14
FINE GRADINGS ON e6,−14
13
Killing form ([Ja, p. 116]).
Proof. It is well-known that T (OC,MC) is a complex Lie algebra of type e6, so
that we only need to compute the signature of the Killing form κ of the real
form T (O,M). As far as we know, it can not be found in the literature. For the
computation, we follow the lines of the proof of [DrG16b, Proposition 2], although
the ideas are mainly based in Jacobson's book [Ja]. Observe the following facts:
a) Der(O), O0 ⊗ M0 and Der(M) are three orthogonal subspaces for the
b) If d, d′ ∈ Der(O), then κ(d, d′) = 12 tr(dd′) = 3κg2(d, d′), denoting by κg2
the Killing form of the algebra Der(O) = g2,−14. This implies that the
signature of κDer(O) is the same as the one of κg2, that is, -14.
c) If D, D′ ∈ Der(M), then κ(D, D′) = 8 tr(DD′) = 8κa2(D, D′), denoting
by κa2 the Killing form of the algebra Der(M) ∼= su1,1 = a2,0. This
implies that the signature of κDer(M) is the same as the one of κa2, that
is, 0.
d) For each a, b ∈ O0 and x, y ∈ M0, we compute κ(a ⊗ x, b ⊗ y) =
−60 n(a, b) tr(x · y). As n is positive definite, the signature of κO0⊗M0
will coincide with minus 7 times the signature of the traceform of M0
(the bilinear form (x, y) 7→ tr(x · y)), which is equal to 0.
Consequently, the signature of T (O,M) turns out to be −14.
(cid:3)
It can be observed that the facts about signatures in items b) and c) are also
consequences of Lemma 2, and that it was no necessary to compute the signature
of the traceform in d) since -14 is the only signature of a real form of e6 which is
multiple of 7.
4.2. The grading. A Z3
grading on O given by Eq. (7) and the following fine Z2
3-grading is achieved when considering the Z3
2-
3-grading on M:
since M is generated (as an algebra) by these two matrices. There is a multi-
plicative basis of M (the product of two elements in such a basis is multiple of
a third one) formed by homogeneous elements of the Z2
3-grading, in fact, all of
them are invertible matrices. In particular M0 breaks into eight one-dimensional
3-grading on T (O,M) ∼= e6,−14
homogeneous components. The obtained Z3
is fine, since its complexified grading is just Γ3.
2 × Z2
Remark 1. Tits' construction can be applied by replacing in T (O,M) the Jor-
dan algebra M with H3(C, γi), for C ∈ {C, R ⊕ R}, or the octonion algebra O
with the split octonion algebra Os, getting in this way all the five real forms
of e6. The same arguments as in Proposition 3 can be used to conclude that
T (Os,M) ∼= e6,2, which implies that e6,2 possesses also a fine Z3
3-grading
whose complexified grading is Γ3.
2 × Z2
(11)
2 × Z2
deg
0 1 0
0 0 1
1 0 0
= (¯1, ¯0);
1 0
0
0 ω 0
0 0 ω2
deg
= (¯0, ¯1),
14
C. DRAPER AND V. GUIDO
2 and Z2 × Z3
2
5. Flag model and related gradings over Z × Z5
Recall that Γ12 and Γ10, the fine Z × Z5
2-gradings on S = e6, are
both refinements of the Z-grading on S produced by the one-dimensional torus
{t1,s,1,1,1,1 : s ∈ C×} contained in the maximal torus T considered in (2).
In
other words, if S = h ⊕ (⊕α∈ΦSα) is the root decomposition relative to a Cartan
subalgebra h, and {αi}6
i=1 is a set of simple roots of the root system Φ (ordered
as in Figure 2), the Z-grading is S = ⊕2
(12)
2 and Z2 × Z3
n=−2Sn, for
Sn = {⊕α∈ΦSα : α =Xi
kiαi, k2 = n}.
By counting roots, we see that S0 = h ⊕ {⊕α∈ΦSα : α = Pi kiαi, k2 = 0} has
dimension 36, dim S±1 = 20 and dim S±2 = 1. Also, the Dynkin diagram of
the semisimple part of the homogeneous component S0 is obtained by removing
the node α2 of the Dynkin diagram, so S0 has type A5 ⊕ C (C denoting here a
one-dimensional centre). A linear model useful for our purposes is developed in
[ADrGu14, §3.4]. Take V a 6-dimensional vector space over C, and then
(13)
can be endowed with a Lie algebra structure such that SV ∼= e6. The products
can be described with multilinear algebra: in terms of contractions, of the wedge
products, and also making use of the dualization of the product · : Sn×S−n → C,
which gives the bracket [Sn, S−n] ⊂ S0. Precisely, if u ∈ Sn, f ∈ S−n, then [u, f ]
denotes the only element in gl(V ) characterized by
SV := ∧6V ∗ ⊕ ∧3V ∗ ⊕ gl(V ) ⊕ ∧3V ⊕ ∧6V
tr(g ◦ [u, f ]) = g(u) · f ∈ C
(14)
for all g ∈ gl(V ).
The first question, then, is if e6,−14 inherits this Z-grading. This can be an-
swered affirmatively as a consequence of the following result due to Cheng [Ch87,
Theorem 3].
Proposition 4. A simple real Lie algebra L admits a grading L = L−2 ⊕ L−1 ⊕
L0⊕L1⊕L2 such that dim L2 = 1 if and only if there is a long root corresponding
to a white node such that its restricted multiplicity is equal to 1.
This can be applied to L = e6,−14 as detailed in [DrG16a, 17.6]. Precisely, the
Satake diagram is the following
α2
α1 α3 α4 α5 α6
Figure 2. Satake diagram of e6,−14
The restricted root system is of type BC2, namely,
±{β1, β2, β1 + β2, 2β1 + β2, 2β1, 2β1 + 2β2} ⊂ a∗,
for a a maximal abelian subalgebra of the eigenspace of a Cartan involution
relative to the eigenvalue −1. There is a (long) root, the maximal one α =
FINE GRADINGS ON e6,−14
15
α1 + 2α2 + 2α3 + 3α4 + 2α5 + α6, such that the restricted root αa = 2β1 6= 0 has
restricted multiplicity equal to 1 (in other words, {α ∈ Φ : αa = αa} = { α}).
So, the diagonalization of any non-zero element in the corresponding root space
produces the desired Z-grading.
Remark 2. Note that the only Z-grading S = ⊕2
n=−2Sn of the complex Lie
algebra e6 such that dim S2 = 1 is that one in (12), up to isomorphism. This
type of gradings usually receives the name of contact gradings. We can argue
as follows. By [OV, Chapter 3], any Z-grading on S comes from a choice of
non-negative integers (l1, . . . , l6) in such a way that (for n 6= 0)
kili}.
Sn = {⊕α∈ΦSα : α =Xi
kiαi, n =Xi
For instance, the grading (12) corresponds to the choice (0, 1, 0, 0, 0, 0). For a
grading with just 5 pieces, as the maximal root is α = α1 + 2α2 + 2α3 + 3α4 +
2α5 + α6, then 2 = l1 + 2l2 + 2l3 + 3l4 + 2l5 + l6, and the only new possibilities
are (1, 0, 0, 0, 0, 1), (0, 0, 1, 0, 0, 0) and (0, 0, 0, 0, 1, 0). But in the three cases, the
dimension of the corner S2 is strictly greater than 1, since the root space related
to the root α1 + α2 + 2α3 + 3α4 + 2α5 + α6 would be contained in S2 too.
We need a more specific model of the Z-grading on e6,−14, in order to obtain
refinements whose complexified gradings are, respectively, Γ12 and Γ10.
5.1. The model. Consider L0 = su5,1 ⊕ RI6. We can describe the semisimple
part1 as
L′
0 = [L0, L0] = {x ∈ sl(V ) : b(xu, v) + b(u, xv) = 0 ∀u, v ∈ V },
where b is the hermitian form (linear in the first variable and conjugate-linear in
the second one) given, relative to a fixed C-basis BV = {e1, . . . , e6} of V , by
b : V × V −→ C,
b(Σsiei, Σtiei) = s1¯t1 + · · · + s5¯t5 − s6¯t6.
First, note that the hermitian form b induces the conjugate-linear (φ(sv) = sv)
L′
0-module isomomorphism (φ(x · v) = x · φ(v))
V −→ V ∗
v
φ :
7−→ b(−, v),
ψ :
∧3V
v1 ∧ v2 ∧ v3
−→ (∧3V )∗
7−→ ψv1,v2,v3,
1It is convenient to describe the semisimple part separately, because the center of the real
Lie algebra {x ∈ gl(V ) : b(xu, v) + b(u, xv) = 0 ∀u, v ∈ V } is RiI6.
for s ∈ C, x ∈ L′
L′
0-module isomorphism
φ :
0, v ∈ V . This can be extended to another conjugate-linear
∧3V
v1 ∧ v2 ∧ v3
−→ ∧3V ∗
7−→ φ(v1) ∧ φ(v2) ∧ φ(v3).
Second, our choice of basis BV allows us to identify S0 = gl(V ) with gl6(C) and
∧6V with C (by e1 ∧ · · · ∧ e6 7→ 1). Then we have the S0-module isomorphism
16
where
ψv1,v2,v3 :
C. DRAPER AND V. GUIDO
∧3V
v4 ∧ v5 ∧ v6
−→ C ≡ ∧6V
7−→ v1 ∧ v2 ∧ v3 ∧ v4 ∧ v5 ∧ v6.
Third, recall that ∧3V ∗ and (∧3V )∗ can be naturally identified by means of the
S0-module isomorphism
where
ρf1,f2,f3 :
ρ :
∧3V ∗
f1 ∧ f2 ∧ f3
−→ (∧3V )∗
7−→ ρf1,f2,f3,
∧3V
v1 ∧ v2 ∧ v3
−→ C
7−→ det(fi(vj)).
We are taking vi ∈ V , fi ∈ V ∗.
We consider the composition of the maps to obtain an S0-module isomorphism
ρ−1ψ : ∧3V −→ ∧3V ∗,
which is in particular an L0-module isomorphism (L0
inverse of this map and the above (conjugate-linear) L′
we obtain a conjugate-linear L′
0-module isomorphism
ψ−1ρ φ : ∧3V −→ ∧3V.
C = S0). By composing the
0-module isomorphism φ,
It has order two, so ∧3V is the sum of the eigenspaces corresponding to the
eigenvalues ±1. We will construct a real form L = ⊕Ln of SV , with L1 one of
the eigenspaces (and, after identifying with the dual, L−1 the other one). Note
0-module ∧3V is 2 dimC(∧3V ) = 40 and its
that the (real) dimension of our L′
complexification is the (non-irreducible) S′
Lemma 4. Under the previous assumptions, the eigenspaces Ker(ψ−1ρ φ ∓ id)
are two L′
0 = [S0, S0]-module ∧3V ⊕ ∧3V .
0-submodules of ∧3V , each one of dimension 20.
Proof. We use the notation ei1...is := ei1 ∧ · · · ∧ eis, so that {eijk, ieijk : 1 ≤ i <
j < k ≤ 6} is a real basis of ∧3V . Besides we denote si = 1 if i = 1, . . . 5 and
s6 = −1. A basis of the real space Ker(ψ−1ρ φ − id) is provided by
l m n(cid:17) ∈ S6
σ =(cid:16)1
i < j < k, l < m < n, i < l ) .
(15)
k
3
2
j
i
4
5
6
i(eijk + sgn(σ) sisjsk elmn) (cid:12)(cid:12)(cid:12)(cid:12)(cid:12)
( eijk − sgn(σ) sisjsk elmn,
Similarly {eijk + sgn(σ) sisjsk elmn, i(eijk − sgn(σ) sisjsk elmn)}, with the permu-
tation σ as above, gives a basis of the eigenspace related to −1.
(cid:3)
(This lemma follows being true if we take the hermitian form b of signature
(3, 3), but it is false for signature (4, 2).)
We have already the tools to provide a convenient model of the real form.
Proposition 5. Take the real vector space
L = L−2 ⊕ L−1 ⊕ L0 ⊕ L1 ⊕ L2 ⊂ SV ,
FINE GRADINGS ON e6,−14
17
for L0 = su5,1 ⊕ RI6 and
(16)
L1 := {v ∈ ∧3V : ρ φ(v) = ψ(v)},
L2 := [L1, L1],
L−1 := φ(cid:0){v ∈ ∧3V : ρ φ(v) = −ψ(v)}(cid:1),
L−2 := [L−1, L−1].
Then L is a real form of SV ∼= e6 of signature −14.
Proof. Note that Li are not only L′
0-modules but L0-modules, since I6 acts as the
i=1 ⊂ V ∗ is
i}6
identity on V , and hence, as n id∧nV and −n id∧nV ∗ for n = 3, 6. If {e∗
the dual basis of {ei}6
i1...is := e∗
i1∧· · ·∧e∗
is,
it is easily checked that
i (ej) = δij), and we denote by e∗
i=1 (i.e., e∗
L−1 = h{e∗
ijk − sgn(σ) sisjsk e∗
lmn)}i,
lmn, i(e∗
ijk + sgn(σ) sisjsk e∗
with σ ∈ S6 as above. Also, L2 = Rie123456 and L−2 = Rie∗
123456. In particular,
i = Si and LC = S = e6. To conclude that L is indeed a real form of S,
LC
we need to check that [L, L] ⊂ L. Note that [L2, L−2] = RI6, and [L1, L−2] ⊂
[[L1, L−1], L−1] by the Jacobi identity, so that the only non-trivial fact to be
checked is that [L1, L−1] ⊂ L. This computation requires of the products in
[ADrGu14, §3.4] recalled in Eq. (14). According to them, the map F = ad(e∗
ijk −
sgn(σ) sisjsk e∗
lmn) ∈ ad(L−1), for an arbitrary permutation σ ∈ S6, satisfies
F (eijk − sgn(σ) sisjsk elmn) = 0,
F (i(eijk + sgn(σ) sisjsk elmn)) ∈ Ri(Eii + Ejj + Ekk − Ell − Emm − Enn) ⊂ L′
0,
F (eijl − sgn(σ′) sisjsl ekmn) ∈ hEkl − slsk Elk, i(Ekl + slsk Elk)i ⊂ L,
F (i(eijl + sgn(σ′) sisjsl ekmn)) ∈ hEkl − slsk Elk, i(Ekl + slsk Elk)i ⊂ L,
for σ′ = (34) ◦ σ, and hence ad(L−1)(L1) ⊂ L, as required.
In order to compute the signature of the obtained real form L, note that
κ(I6, I6) = tr(ad2 I6) = 2(32 dim L1 + 62 dim L2) > 0. Besides the nilpotent
pieces Ln with n 6= 0 do not contribute to the computation of the signature, so
0 (the center of L0
that the signature of L coincides with sign κL0 = 1 + sign κL′
and the derived algebra L′
0 is a positive multiple of
the Killing form κL′
0 = sign(su5,1) = −15 and the
signature of L turns out to be −14.
C ∼= ∧3V is well-
Remark 3. The existence of an L0-module L1 such that L1
known. It is usually said that the complex su5,1-module ∧3V determines a real
representation. This does not happen for the module V , i.e., there does not exist
an L0-module whose complexification is isomorphic to the S0-module V . The
hypotheses to be satisfied by a complex module to determine a real representation
can be consulted in [O, §8]. A more specific construction of L1, quite similar to
the one in Eq. (16), has been sketched in [CS, p. 425-426], although without
using such existence to construct real forms of e6.
0 by Lemma 2, so that sign κL′
0 are orthogonal). But κL′
(cid:3)
18
C. DRAPER AND V. GUIDO
5.2. The gradings. A suitable description of the Z2 × Z3
2-grading Γ12, adapted
to the model of e6 in Eq. (13), is developed in [ADrGu14, §4.5]. Take θ : S → S
defined by:
σ(1) ∧ e∗
θ(eσ(1) ∧ eσ(2) ∧ eσ(3)) := sg(σ) ieσ(4) ∧ eσ(5) ∧ eσ(6),
σ(5) ∧ e∗
θ(e∗
σ(2) ∧ e∗
σ(6),
θ(sI6 + x) := sI6 − xt,
θS2⊕S−2 := − id,
σ(3)) := −sg(σ) ie∗
σ(4) ∧ e∗
for any s ∈ C, x ∈ sl(V ), σ ∈ S6. It is checked in [ADrGu14, §3.4] that θ is
an (outer) order 2 automorphism of S fixing a subalgebra of type c4. For each
A ∈ GL(V ), let ϕA : S → S be the linear map defined by
ϕA(u1 ∧ . . . ∧ ur) := Au1 ∧ . . . ∧ Aur,
ϕA(f1 ∧ . . . ∧ fr) := (A · f1) ∧ . . . ∧ (A · fr),
ϕA(sI6 + x) := sI6 + AxA−1,
for any ui ∈ V , fi ∈ V ∗, s ∈ C and x ∈ sl(V ), where A · fi ∈ V ∗ is given by
(A · fi)(v) = fi(A−1v). Again ϕA is an (inner) automorphism of S. Take the
invertible matrices
A1 = diag(−1,−1, 1, 1, 1, 1), A2 = diag(−1, 1,−1, 1, 1, 1),
A3 = diag(−1, 1, 1,−1, 1, 1), A4 = diag(−1, 1, 1, 1,−1, 1),
and the order 2 automorphisms Fi = ϕAi ∈ Aut(S).
Then Γ12 is produced when refining the Z-grading (13) on S by consider-
ing the simultaneous eigenspaces relative to all the automorphisms in Q =
{F1, . . . , F4, θ} ⊂ Aut(S). In order to prove that Γ12 is inherited by e6,−14, it
is sufficient to prove that θ(L) ⊂ L and Fi(L) ⊂ L for all i = 1, . . . , 4, so that the
restriction to L of the elements in Q are automorphisms of L. These are straight-
forward computations. For instance, F1 acts on eijk − sgn(σ) sisjsk elmn ∈ L1
with eigenvalue −1 if the set {i, j, k} ∩ {1, 2} has cardinal equal to 1 and with
eigenvalue 1 otherwise; and so on.
Now we deal with the Z2 × Z3
2-grading Γ10. In this case, it is convenient for us
to use neither the description of the grading in [ADrGu14, §4.6], nor in [DrV16,
§5.3], but we need to find an equivalent description of Γ10 compatible with some
real form of signature −14. Take the element
Note that the endomorphism ad E : L → L is diagonalizable with eigenvalues
±2,±1, 0, producing a 5-grading on L. Moreover, {x ∈ L : [E, x] = 2x} =
R(i(E55 − E66) + E56 + E65) has dimension 1, so that it is a contact grading.
By Remark 2, the Z-grading on S produced by the eigenspace decomposition
of ad E : S → S is equivalent (in fact, isomorphic) to that one in Eq. (13).
Moreover, both Z-gradings on L are compatible: as E ∈ L0, then ad E(Ln) ⊂ Ln,
0
0
0
0
0
0
E =
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0 −i
0
0
0
0
i
0
= i(E56 − E65) ∈ su5,1.
FINE GRADINGS ON e6,−14
19
so that we have a Z2-grading on L given by L(m,n) = {x ∈ Ln : [E, x] = mx}.
Now, each of the automorphisms q ∈ {θ, F1, F2} satisfies q(E) = E and q(I6) =
I6, so that q(L(m,n)) ⊂ L(m,n). This means that the group generated by {θ, F1, F2}
induces a Z3
2-grading on L compatible with the above grading over the group
2 × Z2-grading on L whose complexified grading is just Γ10. In
Z2, getting a Z3
particular, it is fine.
6. A Z7
2-grading
It was proved in [DrG16b, Proposition 11] the existence of this fine Z7
2-grading
on e6,−14. We will briefly describe the main ideas. Our interest is also to study
the basis provided by the grading.
Let Be6 = {hi, eα, fα : i = 1, . . . , 6, α ∈ Φ+} be a Chevalley basis of S = e6,
so that the structure constants are integers, Φ is the root system relative to
: i = 1, . . . , 6i, and eα ∈ Lα, fα ∈ L−α and
a Cartan subalgebra h = hhi
hi ≡ hαi = [eαi, fαi], for {αi : i = 1, . . . , 6} a set of simple roots of Φ. The
Z7
2-grading Γ13 on S is given by the simultaneous diagonalization relative to the
following MAD-group of the automorphism group:
{t, ωt : t ∈ T, t2 = 1},
where the maximal torus T is given by Eq. (2) and ω ∈ Aut(S) is the involutive
automorphism determined by
ω(eαi) = −fαi, ω(fαi) = −eαi, ω(hi) = −hi,
for all i = 1, . . . , 6. Let σ0 : S → S be a conjugation such that σ0Be6 = id.
The algebra Sσ0 (R-spanned by Be6) is a split real form of S which inherits the
Z7
2-grading, since σ0 commutes with ω and with every order two automorphism
in the torus. By Proposition 1, the algebra Sσ0ωt inherits the Z7
2-grading too
for any t = ts1,...,s6 with s2
i = 1. The signature of the real form Sσ0ωt obviously
depends on t. To be precise, sign(Sσ0ωt) = 78 − 2 dim fix(t) by Eq. (4), since
σ0ω is a compact conjugation commuting with t. But dim fix(t) − 6 is equal to
the cardinal of {(k1, . . . , k6) : P6
6 = 1}. For instance, if
we choose t = t−1,1,1,1,1,1, then dim fix(t) = 46 because there are 20/16 positive
roots with k1 even/odd (respectively), that implies that Sσ0ωt is a real form of
signature -14 with a fine Z7
i=1 kiαi ∈ Φ, sk1
1 . . . sk6
2-grading.
Let us take a closer look at the properties of the obtained grading, and, more
concretely, at the basis provided by it.
Corollary 1. There is a basis B = {ui : i = 1, . . . , 78} of e6,−14 satisfying the
following properties:
a) B is an orthogonal basis for the Killing form.
b) Every element in B is semisimple.
c) If [ui, uj] =Pk f ijkuk, then the structure constants f ijk are rational num-
bers and completely antisymmetric in the three indices.
20
C. DRAPER AND V. GUIDO
As mentioned in Introduction, this generalizes what happens with the Pauli
matrices as well as with the Gell-Mann matrices. Note that Gell-Mann matri-
ces provide a Z3
2-grading on su3 with zero neutral component, one homogeneous
component spanned by two commuting Gell-Mann matrices and each of the re-
maining 6 homogeneous components spanned by one of the remaining matrices.
Proof. Recall that κh0 is negative definite for h0 = P6
j=1 Rihj (a real form of
h), which is the only homogeneous component of the Z7
2-grading on e6,−14 of
dimension strictly greater than 1. Every element in h0 is semisimple with purely
imaginary spectrum. Take B0 = {ih′
j : j = 1, . . . , 6} the following orthogonal
basis of h0:
h′
1 = h1, h′
h′
5 = 2h1 + 3h2 + 4h3 + 6h4 + 5h5, h′
3 = h1 + 2h3, h′
2 = h2, h′
4 = 2h1 + 3h2 + 4h3 + 6h4,
6 = 2h1 + 3h2 + 4h3 + 6h4 + 5h5 + 4h6.
The required basis is provided by this one joint with homogeneous elements of
the grading, namely,
B = B0 ∪ {eα − fα, i(eα + fα) : α ∈ Φ+
0 } ∪ {eα + fα, i(eα − fα) : α ∈ Φ+
1 },
where Φ+
:= {α = P kiαi ∈ Φ+ : k1 = i} if i = 0, 1. Clearly B is orthogonal,
i
recalling that κ(eα, eβ) = 0 = κ(fα, fβ) for all α, β ∈ Φ+ = Φ+
1 , and that
κ(eα, fβ) 6= 0 only for β = α. Moreover, the first 46 elements in B have negative
norm, and the last 32 elements in B have positive norm, thus recovering the fact
of being the signature of κ equal to −14.
Item b) is a direct consequence of [DrV16, Lemma 1]: If a fine grading on a
complex simple Lie algebra satisfies that the universal grading group is finite,
then every homogeneous element is semisimple.
0 ∪ Φ+
In order to study the structure constants, recall that [eα, eβ] = Nα,βeα+β for
Nα,β = 1 + pα,β, where pα,β is the greatest integer such that β − pα,βα is a
root [H, §25.4]. Here the used notation is e−α = fα for α ∈ Φ+. In the case
of e6, the strings have length at most 2:
if α, β, α + β ∈ Φ, then α + 2β /∈ Φ
and α − β /∈ Φ. Hence [eα, eβ] = ±eα+β if α + β is a root, and 0 otherwise.
Now, most of the structure constants f ijk are integers (α(h′
j) ∈ Z), except for
[eα + fα, i(eα − fα)] = −2iPj kjhj ∈ Z[ 1
1 (and similarly
60 ]B0 if α =Pj kjαj ∈ Φ+
Finally, the equality Nα,β = Nβ,γ = Nγ,α for roots such that α + β + γ = 0
(all of them have necessarily the same length) proves the antisymmetry of the
structure constants.
(cid:3)
for Φ+
0 ).2
Remark 4. A basis with the same properties as in Corollary 1 can be found in
e6,−78, in e6,6 and in e6,2 too.
2If we change B by an orthonormal basis, then f ijk /∈ Q, but f ijk ∈ Q[√2,√3,√5] ⊂ R.
FINE GRADINGS ON e6,−14
21
7. No more fine gradings
As a consequence of the above sections, the gradings Γ3, Γ7, Γ8, Γ10, Γ12 and
Γ13 are inherited by e6,−14. The purpose now is to prove that these are the only
gradings on e6,−14 whose complexified grading is fine.
The gradings Γ1, Γ2, Γ5, Γ6 and Γ9 cannot be inherited by e6,−14 by Proposi-
tion 2, taking into account the data dimC Se ± dimCP2g=e
We will prove that e6,−14 inherits neither Γ4, nor Γ11, nor Γ14, by using ad-hoc
Sg in Table 1.
g6=e
arguments for each case.
7.1. No fine Z3
4-grading. Proposition 2 implies that e6,−26 and e6,−78 do not
inherit Γ14, but it does not say anything about e6,−14. We will prove that e6,−14
neither inherits Γ14 by reductio ad absurdum. We will use the knowledge of some
features about Γ14 extracted from [DrV16, §5.4], joint with some representation
theory of real Lie algebras.
Recall first that the complex exceptional Lie algebra e6 = S has a fine grading
Sg such that its coarsening S = S¯0 ⊕ S¯1 ⊕ S¯2 ⊕ S¯3 defined
entitled Γ14 : S = ⊕g∈Z3
by S¯i = ⊕h∈Z2
S¯0 = a3 ⊕ sl(V ), S¯1 = V (2λ1) ⊗ V, S¯2 = V (2λ2) ⊗ C, S¯3 = V (2λ3) ⊗ V,
where V is now a two-dimensional (complex) vector space, and λi's denote again
the fundamental weights but this time of the simple Lie algebra a3 = sl4(C)
(i.e., the maximal weights of its basic representations, defined by λi(hαj ) = δij if
i, j = 1, 2, 3). Thus, dim S¯0 = 15 + 3 = 18 while dim S¯i = 20 for ¯i = ¯1, ¯2, ¯3. The
Z2
4-grading on sl4(C) ⊂ S¯0 is given by the generalized Pauli matrices according
to the following degree assignment:
S(¯i,h) satisfies
4
4
(17)
= (¯i, ¯j),
0 0 0 1
1 0 0 0
0 1 0 0
0 0 1 0
deg
i
1 0
0
0 i
0
0 0 −1
0 0
j
0
0
0
0 −i
4
Suppose that there exists a grading on L = ⊕g∈Z3
for all i, j ∈ {0, 1, 2, 3}.
Lg, a real form of S, such
that the complexified grading is Γ14. That is, Lg ⊗R C = Sg. In particular, L has
L(¯i,h), and L¯i⊗R C = S¯i.
a Z4-grading L = L¯0⊕L¯1⊕L¯2⊕L¯3 defined by L¯i = ⊕h∈Z2
This means that L¯0 is a real form of sl4(C) ⊕ sl2(C). Hence L¯0 = L1
¯0, for
L1
¯0 and L2
¯0 real forms of sl4(C) and sl2(C) respectively. (In general, if σ is the
conjugation of a complex algebra S = LC with Sσ = L, and S is sum of two
ideals S = S1 ⊕ S2, then L = L1 ⊕ L2 for Li = (Si)σ.)
(see Section 2.2), the only ones which inherit the Z2
are su3,1 and su2,2, according to [Sv08]. Thus,
Among the 5 real forms of sl4(C), that is, su4, su3,1, su2,2, sl4(R) and sl2(H)
4-grading given by Eq. (17),
¯0 ⊕ L2
4
L1
¯0 ∈ {su3,1, su2,2}, L2
¯0 ∈ {su2, sl2(R)}.
22
C. DRAPER AND V. GUIDO
Note that L¯1 is an L1
as (L1
¯0) ⊗R C ∼= sl4(C)-module.
¯0-module whose complexification S¯1 is isomorphic to 2V (2λ1)
We now recall some basic facts on representations of real Lie algebras, for in-
stance from [CS, 2.3.14]. If W is a complex vector space, we denote by W the
new complex vector space with the same underground set and scalar multiplica-
tion given by C × W → W , (a + ib, w) 7→ (a − ib)w. For g a real Lie algebra,
W is called a complex representation of g if there is a homomorphism of real
Lie algebras g → glC(W ). In this case, W is naturally a complex representation
of g called the conjugate representation. As real representations, W and W are
always isomorphic (the identity map is an isomorphism), but this is not the case
as complex representations in general.
If U is a (real) g-module, then W = U C is a complex representation and in
this case the map R : W → W , R(u1 + iu2) = u1 − iu2 (ui ∈ U), provides an
isomorphism of complex representations of g. This can be applied to our setting,
since S¯1 is an L1
¯0-complex representation which is the complexification of L¯1,
so that S¯1 ∼= S¯1 is a self-conjugate L1
¯0-module. On one hand we know that S¯1
is isomorphic to 2V (2λ1) as sl4(C)-module, and hence as L1
¯0-module. On the
other hand, L1
¯0 ∈ {su3,1, su2,2}. But for both algebras it is well known ([CS,
p. 230] or [O, Table 5]) that the conjugate representation V (2λ1) ∼= V (2λ3),
which gives a contradiction since 2V (2λ1) cannot be a self-conjugate complex
L1
¯0-representation. Hence we have proved that
Proposition 6. None of the real forms of e6 inherit Γ14.
7.2. No inner fine Z2 × Z3
lows. The algebra S = e6 is modeled by the Tits' construction
2-grading. The grading Γ4 can be described as fol-
S = T (OC, M) = Der(OC) ⊕ (OC
0 ⊗ M0) ⊕ Der(M),
If we consider the Z3
C = Mat3×3(C)+, where again the products are
for the Jordan algebra M = Ms
2-grading on OC (complexified of the
given by Eq. (10).
grading in (7)) and the Z2-grading on M given by the assignment deg(E12) =
(1, 0) and deg(E23) = (0, 1), the obtained Z2 × Z3
2-grading on S is precisely Γ4 in
Table 1. Observe that the Z2-grading induced on the Lie subalgebra Der(M) ∼=
sl3(C) is precisely the root decomposition of a2.
2
Suppose that there exists a grading on e6,−14 = L = ⊕g∈Z2×Z3
L(0,0,h). Then we study L. Its complexification is S = ⊕h∈Z3
Lg such that the
complexified grading is Γ4. By arguments as in Proposition 2, the signature of L
equals the signature of the restriction of the Killing form of L to its subalgebra
L := ⊕h∈Z3
S(0,0,h),
the subalgebra fixed by the two-dimensional torus of the automorphism group
of S producing the Z2-grading, that is, S = Der(OC) ⊕ (OC
0 ⊗ H) ⊕ H, for
H = hE11 − E22, E22 − E33i seen as a subset of M0, but also of sl3(C) = Der(M).
This second piece H ⊂ Der(M) is a two-dimensional centre of S, while the
complementary subspace [ S, S] = Der(OC) ⊕ (OC
0 ⊗ H) is a simple subalgebra of
S isomorphic to d4 = so8(C).
2
2
FINE GRADINGS ON e6,−14
23
Thus [ L, L] is a simple subalgebra of L and a real form of d4. By Section 2.2,
the only possibilities for [ L, L] are so8, so7,1, so6,2, so5,3, so4,4, whose signatures
are, respectively, −28,−14,−4, 2, 4.
In the proof of Proposition 2, it is not only proved that
sign κL = sign κ L = sign κLe + sign κP2g=e,g6=e Lg,
but also that κLe is positive definite, so that sign κLe = dim Le. In our case
Le = Le = H ⊕ [ L, L]e, so that sign κLe = 2 + sign κ[ L, L]e and hence
sign κL = 2 + sign κ[ L, L].
But sign κ[ L, L] = sign([ L, L]) by Lemma 2, which gives a contradiction as −14 /∈
2 +{−28,−14,−4, 2, 4}. (By the way, the number 2 neither belongs to such set.)
Hence, we conclude that
Proposition 7. Neither e6,−14 nor e6,2 inherit Γ4.
7.3. No Z4×Z4
2-grading. There are two classes of order 2 outer automorphisms
of S = e6: those ones fixing a subalgebra of type c4 = sp8(C) and those ones
fixing a subalgebra of type f4 (see Table 2). For the grading Γ11, all the order 2
outer automorphisms belonging to the MAD-group of automorphisms producing
the grading are of the first type [DrV16, §5.3]. Thus, we need to know which
of the real forms of c4 are even parts of a Z2-grading on e6,−14. Take in mind
that there are 4 real forms of c4 = sp8(C), namely, sp4(H), sp3,1(H), sp2,2(H) and
sp8(R), of signatures −36, −12, −4 and 4, respectively (Section 2.2).
Lemma 5. If L = L¯0 ⊕ L¯1 is a Z2-grading on e6,−14 such that L¯0 is a real form
of c4, then L¯0 is a Lie algebra isomorphic to c4,−4 = sp2,2(H).
Proof. By Lemma 3, sign(L−1) = 2 sign(L¯0) + 14. As L−1 is also a real form of
e6, its signature must belong to {6, 2,−14,−26,−78}. But, taking into account
that sign(L¯0) ∈ {4,−4,−12,−36}, the only true possibility is sign(L¯0) = −4.
(The argument is a trivial computation: 2 · 4 + 14 = 22 is not admissible and so
on.)
(cid:3)
In fact, it is not difficult to prove that there exists a Z2-grading L = L¯0 ⊕ L¯1
on L = e6,−14 with L¯0 ∼= sp2,2(H) (so that L−1 split), but it is not necessary for
our purposes.
Lemma 6. Take the fine grading Γ11 : S = ⊕g∈Z4
2×Z4Sg and consider a coarsening
S = S¯0⊕ S¯1 with S¯0 ∼= c4. Let us denote by Γ′
2× Z4-
fine grading on c4 obtained by restriction of Γ11 to the even part S¯0 (perhaps after
reordering the indices). The only real forms of c4 which inherit Γ′
11 are those ones
with signatures 4 and −12.
11 : S¯0 = ⊕h∈Z3
2×Z4S(¯0,h) the Z3
The existence of real forms c4,4 and c4,−12 inheriting Γ′
11 was proved in [DrG16b,
Proposition 8]. But precisely we are interested in that they are the only cases.
The advantage of working with c4 instead of e6 is that all the computations can
be done with matrices.
24
C. DRAPER AND V. GUIDO
Proof. Recall from [DrG16b, V.C] that the grading Γ′
11 on the complex Lie alge-
I4
−I4
0 (cid:17), is
bra S¯0 = spC(8, C) = {x ∈ Mat8×8(C) : xC + Cxt = 0}, for C = (cid:16) 0
given by the simultaneous diagonalization of S¯0 relative to the automorphisms
Ad Ai ∈ Aut(spC(8, C)) ∼= PSpC(8, C), x 7→ AixAi
−1, for the invertible matrices
I2
0
0 σ1
0
0
0
0
A2 = i diag{I4,−I4},
A3 = diag{σ1, σ1, σ1, σ1},
A4 = diag{1,−1,−i, i, 1,−1, i,−i},
A1 = i
0
0
0
0
0
I2
0 σ1
,
1
1
0 (cid:17).
where we denote σ1 =(cid:16) 0
Recall, also from [DrG16b, V.C], that there is a basis B0 of S¯0 formed by
homogeneous elements (simultaneous eigenvectors), all of them matrices with
entries in the set {1, 0,−1}. In particular B0 ⊂ spR(8, C) = {x ∈ Mat8×8(R) :
xC + Cxt = 0}, which is a split real form of c4 that, obviously, inherits Γ′
11.
(Take into account that a real algebra L inherits a grading Γ on LC if and only
if there is a basis of L formed by homogeneous elements of Γ.)
If σ0 denotes the conjugation fixing spR(8, C), then, by Proposition 1, the set
of real forms inheriting Γ′
11 is exactly
(18)
{Sσ0 Ad A : A ∈ hA1, A2, A3, A4i, (σ0 Ad A)2 = id}.
Then the task is to study the signatures of all the real forms in the set in Eq. (18).
According to Eq. (4), sign(Sσ) = 36 − 2 dim fix(θσ), for θσ ∈ Aut(S¯0) of order 2,
commuting with σ, and such that σθσ is compact. For instance, τ = σ0 Ad(C)
is a compact conjugation ([DrG16b, Remark 2]) which commutes with all the
conjugations in (18). Now we compute (tediously, but straightforwardly)
dim fix(Ad(CA)) =(cid:26) 24 if A = A1A2As
16 otherwise.
3Ar
4, for s = 0, 1, r = 0, 1, 2, 3,
Hence, the signatures of the real forms in (18) are:
36 − 2 dim fix(Ad(CA)) ∈ {−12, 4}.
(cid:3)
The immediate consequence is
Corollary 2. The real form e6,−14 does not inherit Γ11.
Proof. By Lemma 5, if e6,−14 inherited Γ11, then sp2,2(H) would inherit Γ′
would contradict Lemma 6.
11. This
(cid:3)
FINE GRADINGS ON e6,−14
25
7.4. Conclusions about gradings on e6,−14. By summarizing the previous
sections, we have proved our main result:
Theorem 1. There are exactly 6 fine gradings on e6 producing fine gradings on
e6,−14, namely, the Z3
2-grading Γ8,
the Z7
2-grading Γ10.
2-grading Γ7, the Z× Z4
2-grading Γ12 and the (outer) Z2 × Z3
2-grading Γ13, the Z × Z5
2 × Z2
3-grading Γ3, the Z6
As happened in the study of e6,−26, the following situations have still to be
studied if we want to have a complete knowledge of the fine gradings on e6,−14
up to equivalence:
a) There could exist a fine grading on e6,−14 whose complexification would
not be a fine grading on e6.
b) There could be two fine gradings on e6,−14 not isomorphic but with iso-
morphic complexifications.
These questions have mainly mathematical interest. For physical purposes, an
interesting topic could be to study the different bases provided by the gradings
and their properties: are they involved in some physical phenomenon? Taking
into account that every fine grading on a simple Lie algebra over a finite group
provides a basis of semisimple elements, and that we have described a fine grading
over Z3
3, what type of properties or processes can be described or modeled
by an abelian group with 3-torsion?
2 × Z2
The authors thank Professors A. Elduque for his read of the manuscript and
M. Atencia for his English language revision.
Acknoledgements
References
[ADrGu14] D. Aranda, C. Draper and V. Guido. Weyl groups of the fine gradings on e6.
J.Algebra 417(1) (2014), 353–390.
[At15] A. Atanasov. Graded Lie Algebras, Supersymmetry, and Applications. Paper consulted
in the web page http://abatanasov.com/
[BKR18a] Y. Bahturin, M. Kochetov, and A. Rodrigo-Escudero. Classification of involutions
on graded-division simple real algebras, Linear Algebra Appl. 546 (2018), 1–36.
[BKR18b] Y. Bahturin, M. Kochetov, and A. Rodrigo-Escudero. Gradings on classical central
simple real Lie algebras, J. Algebra 506 (2018), 1–42.
[BFGM06] S. Bellucci, S. Ferrara, M. Gunaydin and A. Marrani. Charge orbits of symmetric
special geometries and attractors. Internat. J. Modern Phys. A 21 (2006), no. 25, 5043–
5097.
[CalDrM10] A.J. Calder´on, C. Draper and C. Mart´ın. Gradings on the real forms of g2 and f4.
J. Math. Phys. 51(5) (2010), 053516, 21 pp.
[CS] A. Cap and J. Slov´ak. Parabolic Geometries I, Background and General Theory. Mathe-
matical Surveys and Monographs, Vol 154. Amer. Math. Soc. 2009.
[Ca27] ´E. Cartan. La g´eom´etrie des groupes simples. Ann. di Mat. 4 (1927), 209–256.
[Ch87] J-H. Cheng. Graded Lie algebras of the Second kind. Trans. Amer. Math. Soc. 302
(1987), no. 2, 467–488.
[CoNSt] L. Corwin, Y. Ne'eman and S. Sternberg. Graded Lie algebras in mathematics and
physics (Bose-Fermi symmetry). Rev. Mod. Phys. 47 (1975), no. 3, 573–603.
26
C. DRAPER AND V. GUIDO
[Cv] P. Cvitanovi´c. Group theory. Birdtracks, Lie's, and exceptional groups. Princeton Uni-
versity Press, Princeton, NJ, 2008. xiv+273 pp.
[Do09] V.K. Dobrev. Invariant Differential Operators for Non-Compact Lie Groups:
the
E6(−14) case. Proceedings, Eds. B. Dragovich, Z. Rakic, (Institute of Physics, Belgrade,
SFIN Ser. A: Conferences; A1 (2009)) pp. 95–124, arXiv:0812.2655 [math-ph]
[Do14] V.K. Dobrev. Classification of Invariant Differential Operators for Non-Compact Lie
Algebras via Parabolic Relations. J. Phys.: Conf. Ser. 512 012020
[DrE17] C. Draper and A. Elduque. Maximal finite abelian subgroups of E8. Proc. Roy. Soc.
Edinburgh Sect. A 147 (2017), no. 5, 993–1008.
[DrEl14] C. Draper and A. Elduque. Fine gradings on the simple Lie algebras of type E. Note
Mat. 34 (2014), no. 1, 53–86.
[DrG16a] C. Draper and V. Guido. On the real forms of the exceptional Lie algebra e6 and their
Satake diagrams. Non-associative and non-commutative algebra and operator theory, 211–
226, Springer Proc. Math. Stat., 160, Springer, Cham, 2016.
[DrG16b] C. Draper and V. Guido. Gradings on the real form e6,−26. J. Math. Phys. 57(10)
(2016), 18 pp.
[DrV16] C. Draper and A. Viruel. Fine gradings on e6, Publ. Mat. 60(1) (2016), 113–170.
[DuF07] M.J. Duff and S. Ferrara. E6 and the bipartite entanglement of three qutrits. Phys.
Rev. D 76 (2007), no. 12, 124023, 7 pp.
[EK] A. Elduque and M. Kotchetov. Gradings on simple Lie algebras. Mathematical Surveys
and Monographs 189, American Mathematical Society, Providence, RI; Atlantic Associa-
tion for Research in the Mathematical Sciences (AARMS), Halifax, NS, 2013.
[EK18] A. Elduque and M. Kotchetov. Gradings on the simple real Lie algebras of types G2
and D4 Preprint arXiv:1803.10949.
[GV05] T. Gannon and M. Vasudevan. JHEP 0507:035 (2005), hep-th/0504006.
[GPR18] O. Garc´ıa-Prada, A. Pe´on-Nieto and S. Ramanan. Higgs bundles for real groups and
the Hitchin-Kostant-Rallis section. Trans. Amer. Math. Soc. 370 (2018), no. 4, 2907–2953.
[Ge] H. Georgi. Lie algebras in particle physics. From isospin to unified theories. Frontiers in
Physics, 54. Benjamin/Cummings Publishing Co., Inc., Advanced Book Program, Read-
ing, Mass., 1982. xxii+255 pp.
[GLM07] S. Gurrieri, A. Lukas and A. Micu. JHEP 0712:081 (2007), arXiv:0709.1932.
[HPS08] M. Henneaux, D. Persson and P. Spindel.
Hidden Symmetries
http://www.livingreviews.org/lrr-2008-1
of Gravity,
Living Rev. Relativity 11 (2008),
Spacelike Singularities and
1–232.
[HPPe00] M. Havl´ıcek, J. Patera and E. Pelantov´a. On Lie gradings III. Gradings of the real
forms of classical Lie algebras (dedicated to the memory of H. Zassenhaus). Linear Algebra
Appl. 314 (2000), 1–47.
[H] J.E. Humphreys. Introduction to Lie algebras and representation theory, Graduate Texts
in Mathematics 9, Springer-Verlag, New-York, 1978.
[HM06] P.Q. Hung and P. Mosconi. hep-ph/0611001.
[I97] F. Iachello. Algebraic methods in physics. Symmetries in science, X (Bregenz, 1997),
135–143, Plenum, New York, 1998.
[Ja] N. Jacobson. Exceptional Lie algebras. Lecture notes in pure and applied mathematics.
Marcel Dekker, Inc., New York, 1971.
[KL05] J. Kang and P. Langacker. Phys. Rev. D71 (2005) 035014, hep-ph/0412190.
[KGK10] T. Kurth, R. Gramlich and L. Kramer. The real quadrangle of type E6. Adv. Geom.
10 (2010), no. 3, 505–526.
[MT16] A. Marrani and P. Truini. Exceptional Lie algebras at the very foundations of space
and time. p-Adic Numbers Ultrametric Anal. Appl. 8 (2016), no. 1, 68–86.
[MPW91] I. Morrison, P.W. Pieruschka and B.G. Wybourne. The interacting boson model
with the exceptional groups G2 and E6. J. Math. Phys. 32 (1991), no. 2, 356–372.
FINE GRADINGS ON e6,−14
27
[O] A.L. Onishchnik. Lectures on Real Semisimple Lie Algebras and Their Representations.
European Mathematical Society, 2004. Series ESI Lectures in Mathematics and Physics.
[OV] A.L. Onishchnik, `E.B. Vinberg (Editors). Lie Groups and Lie Algebras III, Encyclopaedia
of Mathematical Sciences, Vol. 41. Springer-Verlag, Berlin, 1991.
[PZ89] J. Patera and H. Zassenhaus. On Lie gradings. I. Linear Algebra Appl. 112 (1989),
87–159.
[Sv08] M. Svobodov´a. Fine Gradings of Low-Rank Complex Lie Algebras and of Their Real
Forms. SIGMA 4 (2008), 039, 13 pages.
[T66] J. Tits. Alg`ebres alternatives, alg`ebres de Jordan et alg`ebres de Lie exceptionelles. I.
Construction. Nederl. Akab. Wetensch. Proc. Ser. A 69 = Indag. Math. 28 (1966), 223–
237.
[W95] B.G. Wybourne. Exceptional Lie groups in physics. Lithuanian Journal of Physics.
Volume 35, 1995, no. 2, 123–132.
[Y16] J. Yu. Maximal abelian subgroups of compact simple Lie groups of type E. Geom.
Dedicata 185 (2016), 205–269.
Cristina Draper Fontanals: Departamento de Matem´atica Aplicada, Escuela
de Ingenier´ıas Industriales, Ampliaci´on Campus de Teatinos, S/N, 29071 M´alaga,
Spain, [email protected]
Valerio Guido: [email protected]
|
1806.06844 | 1 | 1806 | 2018-06-18T17:45:59 | Classification of quadratic and cubic PBW algebras on three generators | [
"math.RA",
"math-ph",
"math.AT",
"math.GR",
"math-ph",
"math.QA"
] | We give a complete classification of quadratic algebras A, with Hilbert series $H_A=(1-t)^{-3}$, which is the Hilbert series of commutative polynomials on 3 variables. Koszul algebras as well as algebras with quadratic Gr\"obner basis among them are identified. We also give a complete classification of cubic algebras A with Hilbert series $H_A=(1+t)^{-1}(1-t)^{-3}$. These two classes of algebras contain all Artin-Schelter regular algebras of global dimension 3. As far as the latter are concerned, our results extend well-known results of Artin and Schelter by providing a classification up to an algebra isomorphism. | math.RA | math |
Classification of quadratic and cubic PBW algebras on three
generators
Natalia Iyudu and Stanislav Shkarin
Abstract
We give a complete classification of quadratic algebras A, with Hilbert series HA = (1 − t)−3,
which is the Hilbert series of commutative polynomials on 3 variables. Koszul algebras as well
as algebras with quadratic Grobner basis among them are identified. We also give a complete
classification of cubic algebras A with Hilbert series HA = (1 + t)−1(1 − t)−3. These two classes of
algebras contain all Artin–Schelter regular algebras of global dimension 3. As far as the latter are
concerned, our results extend well-known results of Artin and Schelter by providing a classification
up to an algebra isomorphism.
MSC: 17A45, 16A22
Keywords: Quadratic algebras, Cubic algebras, Koszul algebras, Hilbert series, Sklyanin algebras, PBW-
algebras, potential algebras
1
Introduction
Throughout this paper K is an algebraically closed field of characteristic different from 2 or 3 (some
arguments break down in the absence of any of these assumptions). If B is a Z+-graded vector space,
Bm stands for the mth component of B. We always assume that each Bm is finite dimensional, which
allows to consider the obvious generating function of the sequence of dimensions of graded components
called the Hilbert series of B:
HB(t) =
∞
∑
m=0
dim Bm tm ∈ Z[[t]].
assumed to be degree graded. If R is a subspace of the n2-dimensional space V 2, then the quotient
If V is an n-dimensional vector space over K, then F = F(V) is the tensor algebra of V . For any
choice of a basis x1, . . . , xn in V , F is naturally identified with the free algebra K⟨x1, . . . , xn⟩, always
A of F(V) by the ideal generated by R is called a quadratic algebra and denoted A(V, R). If R is a
subspace of the n3-dimensional space V 3, then the quotient A of F(V) by the ideal generated by R is
called a cubic algebra and denoted B(V, R). In both cases, the ideal generated by R is known as the
ideal of relations of A. We say that
A is a P BWS-algebra if HA(t) = HK[x1,...,xn](t) =(1 − t)−n.
These algebras are called PBW (Poincare-Birkhoff-Witt) by a number of authors, for example, Odesskii
[17]. In the book by Polishchuk and Positselski [18], however, the term PBW algebra is reserved for a
quadratic algebra with a quadratic Grobner basis in the ideal of relations. We call them here PBWB-
algebras. More precisely, a quadratic algebra A = A(V, R) is a P BWB-algebra if there are linear bases
x1, . . . , xn and g1, . . . , gm in V and R respectively such that with respect to some compatible with
multiplication well-ordering on the monomials in xj, g1, . . . , gm form a Grobner basis of the ideal of
relations of A. References to Poincare-Birkhoff-Witt properties are relevant in both cases, in both cases
we deal with generalisations of the PBW theorem on the Hilbert series of the universal enveloping of
a Lie algebra: the series concept refers to the conclusion, while the basis concept refers to the method
1
of the proof of this theorem. However, one has to keep in mind that none of PBWS or PBWB yield
the other: a PBWB algebra may very well have exponential growth or be finite dimensional, while a
PBWS algebra may fail to even have a finite Grobner basis in the ideal of relations.
Another concept playing an important role in this paper is Koszulity. A quadratic algebra A =
A → K → 0, where the second last arrow is the augmentation map and the matrices of the maps
Mm → Mm−1 with respect to some free bases consist of homogeneous elements of degree 1. If we pick
A(V, R) is called Koszul if K as a graded right A-module has a free resolution ⋅ ⋅ ⋅ → Mm → ⋅ ⋅ ⋅ → M1 →
a basis x1, . . . , xn in V , we get a bilinear form on the free algebra K⟨x1, . . . , xn⟩ defined by [u, v] = δu,v
for every monomials u and v. The quadratic algebra A! = A(V, R⊥), where R⊥ = {u ∈ V 2 ∶ [r, u] =
0 for each r ∈ R}, is called the dual algebra of A. Note that up to an isomorphism A! does not depend
on the choice of a basis in V . We shall use the following well-known properties of Koszul algebras:
every PBWB-algebra is Koszul; A is Koszul ⇐⇒ A! is Koszul;
if A is Koszul, then HA(−t)HA!(t) = 1.
(1.1)
Artin and Schelter [1] characterize the regular algebras of global dimension 3. These naturally
split into two classes: some quadratic algebras A satisfying HA = (1 − t)−3 and some cubic algebras
A with HA = (1 + t)−1(1 − t)−3. As far as quadratic algebras are concerned, Artin and Schelter
characterize a subclass of PBWS quadratic algebras on three generators (additional properties imposed
are Gorenstein and global dimension 3). The purpose of this article is to complete their characterization
to incorporate all quadratic PBWS algebras on three generators, identifying Koszul algebras on the
way. We also characterize all cubic algebras with the Hilbert series (1 + t)−1(1 − t)−3. We push it to
the limit providing a canonical form up to isomorphism. For the sake of brevity we denote
Ω ={A ∶ A is a quadratic algebra satisfying HA =(1 − t)−3}.
We split the class Ω into three disjoint parts: Ω = Ω0 ∪ Ω+
∪ Ω−, where
Ω0 ={A ∈ Ω ∶ A is PBWB},
Ω+ ={A ∈ Ω ∶ HA! =(1 + t)3, but A is not PBWB},
Ω− ={A ∈ Ω ∶ HA! ≠(1 + t)3}.
Note also that Ω ⊂ Ω′, where
Ω′ ={A ∶ A is a quadratic algebra satisfying dim A1 = 3, dim A2 = 6 and dim A3 = 10}.
Observe that A ∈ Ω− can not be Koszul and therefore can not be PBWB since the equality HA(−t)HA!(t) =
1 fails.
As for cubic algebras we denote
Λ ={A ∶ A is a cubic algebra satisfying HA =(1 + t)−1(1 − t)−3}.
Note that Λ ⊂ Λ′, where
Λ′ ={A ∶ A is a cubic algebra satisfying dim A1 = 2, dim A2 = 4, dim A3 = 6 and dim A4 = 9}.
Let us mention, that as a consequence of this classification we were able to answer an old ques-
tion of Ufnarovski, namely to provide an example of automaton algebra (one from the family N1 in
Theorem 1.11), which does not have a finite Grobner basis. The proof of this result one can find in
[14].
Before stating the main result, we would like to say a few words about the key idea as well as to
introduce some further notations, which will be used throughout the paper.
2
1.1 Quasipotentials
Notation 1.1. If V is a finite dimensional vector space over K, k ⩾ 2 and Q ∈ V k+1 = V ⊗(k+1), then
there are the smallest (in the inclusion sense) subspaces
Ej = Ej(Q) of V and Fj = Fj(Q) of V k such that Q ∈ E1 ⊗ F1 and Q ∈ F2 ⊗ E2.
Clearly,
while
We also denote
n1(Q) = dim E1 = dim F1 is the rank of Q as an element of V ⊗ V k,
n2(Q) = dim E2 = dim F2 is the rank of Q as an element of V k
⊗ V .
RQ = F1 + F2, which is a subspace of V k.
Lemma 1.2. Let n, k be integers such that n, k ⩾ 2, V be an n-dimensional vector space over K and R
Proof. Obviously, dim Ak+1 = nk+1 − dim Ik+1 and Ik+1 = RV + V R. Since dim RV = dim V R = n2, we
be an n-dimensional subspace of the nk-dimensional space V k. Assume also that A = F(V)~I, where
I is the ideal generated by R. Then dim Ak+1 = nk+1 − 2n2 + 1 if and only if dim(RV ∩ V R) = 1.
have dim Ik+1 = 2n2 − dim(RV ∩ V R). The result immediately follows.
be an n-dimensional subspace of the nk-dimensional space V k and I be the ideal in F(V) generated by
R. The algebra A = F(V)~I is called a quasipotential algebra if dim Ak+1 = nk+1−2n2+1. By Lemma 1.2,
Q of F(V). That is, RV ∩ V R = span{Q}. We call Q a quasipotential for A. We call Q ∈ V k+1 a
Definition 1.3. Let n, k be integers such that n, k ⩾ 2, V be an n-dimensional vector space over K, R
RV ∩ V R is one-dimensional and therefore is spanned by a single degree k + 1 homogeneous element
quasipotential if it is a quasipotential for some algebra.
Remark 1.4. Note that to be a quasipotential algebra is an isomorphism invariant. Moreover, the
quasipotential Q of an algebra A is unique up to a scalar multiple and every linear substitution
providing an isomorphism between two quasipotential algebras must transform the quasipotential of
the first into the quasipotential of the second up to a non-zero scalar multiple.
Remark 1.5. By Lemma 1.2, all algebras in Ω′ are quasipotential, each with a degree 3 quasipotential,
while all algebras in Λ′ are quasipotential, each with a degree 4 quasipotential.
Remark 1.6. Let Q be a quasipotential for a quasipotential algebra A and V , R be as in Definition 1.3.
We easily have that RQ ⊆ R, where RQ = F1(Q) + F2(Q) is introduced in Notation 1.1. In particular,
if RQ happens to be n-dimensional, we must have RQ = R. That is, if Q ∈ V k+1 is a quasipotential
and dim RQ = n, then there is exactly one algebra for which Q is the quasipotential.
The bulk of the paper is devoted to providing a canonical form of quasipotentials in the cases
(n, k) = (3, 2) and (n, k) = (2, 3) under the natural action of GLn(K) by linear substitutions. In the
case (n, k) =(3, 2), this task can in a way be treated as an extension of the canonical form results for
ternary cubics (these go way back to Weierstrass). It turns out that only in the case n1(Q) = n2(Q) = 1,
there are multiple algebras with desired Hilbert series corresponding to the same quasipotential. This
case stands out a lot.
As usual, an invariant is some characteristic of an algebra from a given class, which remains the
same when we replace an algebra by an isomorphic one. By Remark 1.4,
the ranks n1(Q) and n2(Q) are invariants for quasipotential algebras.
(1.2)
3
Definition 1.7. Let n, k be integers such that n, k ⩾ 2, V be an n-dimensional vector space over K
and Q ∈ V k+1 be a quasipotential. We call Q a twisted potential if n1(Q) = n2(Q) = n. If a twisted
potential Q is cyclicly invariant (that is, invariant under the linear map C ∶ V k+1
x0x1 . . . xk ↦ x1 . . . xkx0), then Q is called a potential.
→ V k+1 defined by
Remark 1.8. Assume that dim V = n and Q ∈ V k+1 is a twisted potential. Then for every linear basis
X ={x1, . . . , xn} in V , we have
Q =
xj fj =
gj xj,
n
∑
j =1
n
∑
j =1
where both {f1, . . . , fn} and {g1, . . . , gn} are linear bases in the n-dimensional space R = RQ. Thus
we have a matrix MQ(X) ∈ GLn(K) of coefficients of gj with respect to the basis f1, . . . , fn. Now if
Y = {y1, . . . , yn} is another linear basis in V and C ∈ GLn(K) is the matrix of coefficients of xj with
respect to the basis y1, . . . , yn, then a routine computation shows that MQ(Y) = BMQ(X)B −1, where
B = C T is the transpose of C. This observation yields that the Jordan normal form of MQ(X) is an
invariant if and only if MQ(X) is the identity matrix for some (=for any) basis X in V . That is, Q
is a potential if and only if n1(Q) = n2(Q) = n and MQ(X) is the identity matrix for some (=for any)
invariant for the class of twisted potential algebras. Note also that a twisted potential Q is cyclicly
basis X in V .
Remark 1.9. The concepts of potential and twisted potential algebras go beyond degree graded
algebras. However, in the case of degree graded algebras with certain non-degeneracy assumed, our
definition is equivalent to the original definition of potential algebras of Kontsevich [15] (see [5, 9] for
alternative equivalent definitions). What we call twisted potential algebras here and in [11] was first
introduced under the name of algebras defined by multilinear forms by Dubois-Violette [7, 8]. Since
we intend to never wander outside the class Ω′
∪ Λ′, we stick with the above definitions (ignore the
non-graded case as well as the degenerate potentials).
Lemma 1.10. Let n, k be integers such that n, k ⩾ 2, V be an n-dimensional vector space over K and
Not all pairs of numbers between 1 and n occur as (n1(Q), n2(Q)) for a quasipotential Q.
Q ∈ V k+1 be a quasipotential. Then n1(Q) = n ⇐⇒ n2(Q) = n.
Proof. Assume the contrary: min{n1(Q), n2(Q)} < max{n1(Q), n2(Q)} = n. Reversing the order of
letters in each of the monomials featuring in Q yields another quasipotential Q′ with n1(Q′) = n2(Q)
and n2(Q′) = n1(Q). This allows us, without loss of generality, assume that n1(Q) = n and n2(Q) =
m < n. Let x1, . . . , xm be a linear basis in W = E2(Q). Pick any xm+1, . . . , xn such that x1, . . . , xn
form a basis in V . Then
Q =
n
Q
j =1
xj fj =
m
Q
j =1
gj xj,
(1.3)
where f1, . . . , fn form a basis in the n-dimensional space F1(Q) ⊂ V k, while g1, . . . , gm form a basis
in the m-dimensional space F2(Q) ⊂ V k. Since dim(F1(Q) + F2(Q)) ⩽ n (see Remark 1.6), we have
F2(Q) ⊂ F1(Q) and therefore,
each gj is a linear combination of f1, . . . , fn.
(1.4)
Now by (1.3), fj ∈ V k−1W for each j. By (1.4), gj ∈ V k−1W for each j. Plugging this back into
(1.3), we get fj ∈ V k−2W 2 for each j. According to (1.4), gj ∈ V k−2W 2 for each j.
Iterating this
procedure, we eventually see that fj ∈ W k and gj ∈ W k for all j. The latter plugged into the equality
Q = g1x1 + . . . + gmxm of (1.3), yields Q ∈ W k+1, which is incompatible with the first equality in (1.3)
since fj are linearly independent. This contradiction completes the proof.
4
The proof of the main results goes along the following lines. Assuming A ∈ Ω′, for the quasipotential
Q for A, we have that one of following mutually exclusive options:
• Q is a cube (of a degree 1 element);
• n1(Q) = n2(Q) = 1 and Q is not a cube;
• n1(Q) = 1 and n2(Q) = 2;
• n1(Q) = 2 and n2(Q) = 1;
• n1(Q) = n2(Q) = 2;
• n1(Q) = n2(Q) = 3.
All possibilities are covered since, n1(Q) = 3 ⇐⇒ n2(Q) = 3 for A ∈ Ω′ by Lemma 1.10. Since nj(Q)
stratify further by isomorphism classes of the algebra A0 = A~I, where I is the ideal generated by z
are invariants, we do not have to worry of algebras corresponding to different items of the above list
In the case when Q is a cube: Q = zzz for a non-zero z ∈ V , we
being isomorphic to each other.
(A0 is uniquely determined by A ∈ Ω′ up to an isomorphism). In each particular case, we use the
Grobner basis technique to figure out which of the algebras have the same series as the commutative
polynomials. We also identify Koszul and PBWB algebras among them.
Similarly, if A ∈ Λ′, for the quasipotential Q for A, we have that one of following mutually exclusive
options:
• Q is a fourth power (of a degree 1 element);
• n1(Q) = n2(Q) = 1 and Q is not a fourth power;
• n1(Q) = n2(Q) = 2.
Again, all possibilities are covered since, n1(Q) = 2 ⇐⇒ n2(Q) = 2 for A ∈ Λ′ by Lemma 1.10.
Before formulating main results, we introduce some more notation. Everywhere afterwards, θ and
i are fixed elements of K satisfying
θ3 = 1 ≠ θ
and i2 = −1.
(1.5)
Note that θ does exist since K is algebraically closed and has characteristic different from 3, i does
exist since K is algebraically closed and i ∉{1, −1} since char K ≠ 2.
1.2 Main results
The results are presented in tables. The first column provides a label for further references. Gen-
erators of algebras from Ω are denoted x, y, z, while generators of algebras from Λ are denoted x, y.
We use the letters a, b, c, d for the parameters from K (we never need more than 4 parameters). The
exceptions column says which values of the parameters are excluded. The isomorphism column pro-
vides generators of a group action on the space of parameters such that corresponding algebras are
isomorphic precisely when the parameters are in the same orbit. All isomorphisms are meant in the
graded algebras sense. For shortness, we occasionally use the following notation. If u1, . . . , un ∈ V ,
⟲ stands for the sum of all n cyclic permutations of the word u1 . . . un. The PBW column
then u1 . . . un
indicates whether the algebra is PBWB or not: the Y entry stands for the algebra being PBWB, while
the N entry for the opposite. Algebras featuring with different labels are non-isomorphic.
5
Theorem 1.11. I. An algebra A belongs to Ω and its quasipotential Q = QA is the cube of a degree
one element if and only if A is isomorphic to an algebra from the following table. All such algebras
are NON-Koszul and therefore non-P BWB.
Exceptions
Isomorphisms
xy+yx+zx+azy; y2−xz+(1−a)yz−zx−azy; z2
Quasipotential QA Defining Relations of A
z3
z3
z3
z3
xy+yx; y2−xz−yz−zx−zy; z2
xy+yx; y2−xz−zx; z2
xy − yx; y2 − xz − zx; z2
axy−a2yx−a2(a2−1)zx−(a−1)(a3 +1)zy;
y2−xz+(a−2)yz−a2zx−(a2−a+1)zy; z2
xy−ayx; y2−xz−a2zx; z2
x2−xy−yz; yx−azx−bzy; z2
x2−xy; yx−azx+zy; z2
yx−bxz−azx−azy; y2−zx−zy; z2
yx−axz−zx; y2−zx; z2
yx−xz; y2−zx; z2
yx; y2−zx; z2
xy−bzx−axz−ayz; y2−xz−yz; z2
xy−azx−xz; y2−xz; z2
xy−zx; y2−xz; z2
xy; y2−xz; z2
xy−yz + zx; yx−xz+zy; z2
xy−yz − zx; yx−xz−zy; z2
xy−azx−zy; yx−xz; z2
x2+yz+azy; y2+xz+
a zx; z2
1
R1
R2
R3
R4
R5
z3
R8
R7
R6
z3
z3
z3
z3
R9
R10 z3
R11 z3
R12 z3
R13 z3
R14 z3
R15 z3
R16 z3
R17 z3
R18 z3
R19 z3
R20 z3
none
none
none
none
a ≠ 0, a2 ≠ 1
a ≠ 0, a2 ≠ 1
a ≠ 0, a ≠ −1
none
a(1+. . .+bk) ≠ 1 for all k ∈ Z+
none
none
none
none
none
none
none
none
none
a ≠ 0
(1+a+. . .+ak) ≠ 0 for all k ∈ Z+
trivial
trivial
trivial
trivial
trivial
trivial
trivial
trivial
trivial
trivial
trivial
trivial
trivial
trivial
trivial
trivial
trivial
trivial
trivial
1
a
a ↦
trivial
trivial
trivial
trivial
trivial
trivial
(a, b, c, d) ↦( 1
a , 1
b , d
b)
b , c
trivial
trivial
trivial
trivial
trivial
trivial
trivial
trivial
trivial
trivial
trivial
trivial
R21 z3
R22 z3
R23 z3
R24 z3
R25 z3
R26 z3
R27 z3
R28 z3
R29 z3
R30 z3
R31 z3
R32 z3
R33 z3
R34 z3
R35 z3
R36 z3
R37 z3
R38 z3
R39 z3
xy−yx−y2−zx−czy; xz−azx−bzy; z2
xy−yx−y2−zy; xz−azx−bzy; z2
xy−yx−y2; xz−azx−bzy; z2
xy−yx−y2−zx − bzy; zx−azy; z2
xy−yx−y2−zy; zx−azy; z2
xy−yx−y2; zx−azy; z2
xy−ayx−zx−zy; xz+byz+czx+dzy; z2
xy−ayx−zx−zy; xz+bzx+czy; z2
xy−ayx−zx; xz+yz+bzx+czy; z2
xy−ayx−zx; xz+bzx+zy; z2
xy−ayx−zx; yz+zx+bzy; z2
xy−ayx; xz+yz+bzx+czy; z2
xy−ayx; yz+bzx+zy; z2
xy−yx−zx−azy; xz−zy; z2
xy−yx−zy; xz−zy; z2
xy−yx; xz−zy; z2
xy−yz−czx; xz+ayz+bzx+zy; z2
xy−zx; xz+ayz+bzx+zy; z2
xy; xz+ayz+bzx+zy; z2
na + b ≠ 0 for all n ∈ Z+
na + b ≠ 0 for all n ∈ Z+
na + b ≠ 0 for all n ∈ Z+
a ≠ 0
a ≠ 0
a ≠ 0
a ≠ 0, a ≠ 1, b ≠ 0
d ≠ anbc for all n ∈ Z+
a ≠ 0, a ≠ 1, c ≠ 0
a ≠ 0, a ≠ 1
c ≠ anb for all n ∈ Z+
a ≠ 0, a ≠ 1
a ≠ 0, a ≠ 1
a ≠ 0, a ≠ 1
c ≠ anb for all n ∈ Z+
a ≠ 0, a ≠ 1
none
none
none
ab ≠ 1
ab ≠ 1
ab ≠ 1
6
II. An algebra A belongs to Ω and its quasipotential Q = QA satisfies n1(Q) = 1, n2(Q) = 2 if and
only if A is isomorphic to an algebra from the following table. All such algebras are Koszul.
Quasipotential QA
Defining Relations of A
xyz − axzy
xy; xz; yz − azy
xyz − axzy + xzx
xy; xz; yz − azy + zx
xyz − axzy + xzx + xyx
xy; xz; yz − azy + zx + yx
xyz − xzy − xz2 + xyx
xyz − xzy − xz2 + xzx
xyz − xzy − xz2
xy; xz; yz − zy − z2 + yx
xy; xz; yz − zy − z2 + zx
xy; xz; yz − zy − z2
xzy + xyx
xy; xz; zy + yx
M1
M2
M3
M4
M5
M6
M7
Exceptions
a ≠ 0
a ≠ 0
a ≠ 0, a ≠ 1
none
none
none
none
Isomorphisms
PBW
a ↦
1
a
trivial
1
a
a ↦
trivial
trivial
trivial
trivial
Y
Y
Y
Y
Y
Y
Y
M8
xzy + xyx − xzx
xy; xz; zy + yx − zx
none
trivial
Y
III. An algebra A belongs to Ω and its quasipotential Q = QA satisfies n1(Q) = 2, n2(Q) = 1 if and
only if A is isomorphic to an algebra from the following table. All such algebras are Koszul.
Quasipotential QA
Defining Relations of A
zyx − ayzx
yx; zx; zy − ayz
zyx − ayzx + xzx
yx; zx; zy − ayz + xz
zyx − ayzx + xzx + xyx
yx; zx; zy − ayz + xz + xy
zyx − yzx − z2x + xyx
zyx − yzx − z2x + xzx
zyx − yzx − z2x
yx; zx; zy − yz − z2 + xy
yx; zx; zy − yz − z2 + xz
yx; zx; zy − yz − z2
yzx + xyx
yx; zx; yz + xy
L1
L2
L3
L4
L5
L6
L7
Exceptions
a ≠ 0
a ≠ 0
a ≠ 0
none
none
none
none
Isomorphisms
PBW
a ↦
1
a
trivial
1
a
a ↦
trivial
trivial
trivial
trivial
Y
Y
Y
Y
Y
Y
Y
L8
yzx + xyx − xzx
yx; zx; yz + xy − xz
none
trivial
Y
IV. An algebra A belongs to Ω and its quasipotential Q = QA satisfies n1(Q) = n2(Q) = 2 if and only
if A is isomorphic to an algebra from the following table. All such algebras are Koszul. The algebra
in (S18) is an odd one out. It is well-defined and belongs to Ω′ whenever char K ≠ 2. However it is in
Ω only when K is of characteristic zero. This explains the weird entry in the exceptions column.
Quasipotential QA
xzy + xyx + yxy
bxyz + xzy + xz2 + x2z + ayxz
x2z + axyz + xzy + byxz
xz2 + axyz + yxz + bxzy
axyz + xzy + byxz
xyz + ayxz + yzy + yz2
S1
S2
S3
S4
S5
S6
S7
S8
xyz + ayxz + yzy
xyz − yxz + ay2z + yzy + yz2
xyz − yxz + ay2z + yzy
S9
S10 xyz + axzy + yxy + x2y
Defining Relations of A
xy; zy + yx; xz + yx
bxy + ayx + x2
byz + zy + z2;
xz;
xz; ayz + zy; axy + byx + x2
xz; ayz + bzy + z2; axy + yx
xz; ayz + zy; axy + byx
yz; axz + zy + z2; xy + ayx
yz; axz + zy; xy + ayx
yz; xz − zy − z2; xy − yx + ay2
yz; xz − zy; xy − yx + ay2
xy; yz + azy; axz + yx + x2
S11 xyz + axzy + yxy
S12 xyz − xzy + axy2 + yxy + x2y
S13 xyz − xzy + axy2 + yxy
S14 x2z + axyz + xzy + yxz + y3
S15 xz2 + axyz + yxz + xzy + y3
S16 axyz + xzy + yxz + y3
xy; yz + azy; axz + yx
xy; yz − zy + ay2; xz − yx − x2
xy; yz − zy + ay2; xz − yx
ayz + zy − y2; xz + y2; axy + yx + x2
ayz + zy + z2; xz + y2; axy + yx − y2
ayz + zy; xz + y2; axy + yx
S17
S18
S19
S20
yxz−ayxy−azxz+ay2z−yz2+az3
+a2zxy − a2y3 − a2z2y + a3zy2
+(a−1−a3)zyz +(a3−a2+1)yzy
4yxy + 4y3 + 4zxy + 4zy2
−2z2y − 2yz2 + 2zyz − z3
axy − xz + a2y2 −(a3 + 1)zy + z2;
yx − azx + ay2 −(a3 + 1)zy + az2; yz − azy
2 z2
2 z2; yz−zy+
2 z2; xy+y2−
yx+y2+zx+zy−
1
1
1
axy2 − axyz − xzy + x2y
−azxy −
b2
−(a+1)2
4
z3
axy2 − axyz − xzy + x2y
−azxy +
(a+1)2
4
z3
x2−xz−azx−
b2
−(a+1)2
4
z2; xy+
b2
−(a+1)2
4a
z2;
y2−yz−
1
a zy−
b2
−(a+1)2
4a2
z2
x2−xz−azx+
(a+1)2
4
z2; xy−
(a+1)2
4a z2;
y2−yz−
1
a zy+
(a+1)2
4a2 z2
7
Exceptions
Isomorphisms
PBW
none
ab ≠ 0
ab ≠ 0
ab ≠ 0
ab ≠ 0
a ≠ 0
a ≠ 0
a ≠ 0
a ≠ 0
a ≠ 0
a ≠ 0
a ≠ 0
a ≠ 0
a ≠ 0
a ≠ 0
a ≠ 0
a ≠ 0, ak ≠ 1 for all k ∈ N
char K = 0
a≠0, a≠ − 1, b2≠(a+1)2,
1+a+b ≠ (1−a+b)n
(1−a−b)n
1+a−b
for all n∈Z+
trivial
trivial
trivial
trivial
trivial
trivial
trivial
trivial
trivial
trivial
trivial
trivial
trivial
trivial
trivial
trivial
a ↦
1
a
trivial
(a, b) ↦(a, −b)
a≠0, a≠ − 1,
na ≠ n + 2 for all n ∈ N
trivial
Y
Y
Y
Y
Y
Y
Y
Y
Y
Y
Y
Y
Y
Y
Y
Y
N
N
N
N
V. There is no algebra A ∈ Ω such that its quasipotential Q = QA satisfies max{n1(Q), n2(Q)} =
3 and min{n1(Q), n2(Q)} < 3. There is no A ∈ Λ such that its quasipotential Q = QA satisfies
max{n1(Q), n2(Q)} = 2 and min{n1(Q), n2(Q)} < 2.
VI. An algebra A belongs to Ω and is potential if and only if A is isomorphic to an algebra from the
following table. All such algebras are Koszul.
The potential QA
x3 + y3 + z3 + axyz⟲ + bxzy⟲
xyz⟲ + axzy⟲
(y + z)3 + xyz⟲ + axzy⟲
z3 + xyz⟲ + axzy⟲
y3 + xz2⟲
xz2⟲
+ y2z
⟲
+ xyz⟲ − xzy⟲
+ xyz⟲ − xzy⟲
y3 + z3 + xyz⟲ − xzy⟲
yz2⟲
+ xyz⟲ − xzy⟲
P1
P2
P3
P4
P5
P6
P7
P8
Defining Relations of A
x2 + ayz + bzy;
y2 + azx + bxz;
z2 + axy + byx
yz + azy; zx + axz; xy + ayx
yz + azy; axz + zx +(y + z)2;
xy + ayx +(y + z)2
yz + azy; axz + zx; xy + ayx + z2
yz − zy + z2; −xz + zx + y2; xy − yx + xz + zx
yz − zy + z2; −xz + zx + yz + zy;
xy − yx + xz + zx + y2
yz − zy; −xz + zx + y2; xy − yx + z2
yz − zy; −xz + zx + z2; xy − yx + yz + zy
a ≠ 0, a ≠ −1
a ↦ a−1
a ≠ 0
none
none
none
none
a ↦ a−1
trivial
trivial
trivial
trivial
Exceptions
(a, b) ≠(0, 0)
(a3, b3) ≠(1, 1)
(a + b)3 + 1 ≠ 0
a ≠ 0
Isomorphisms
PBW
(a, b) ↦(θa, θb)
(a, b) ↦ θa+θ2b+1
a+b+1 , θ2a+θb+1
a+b+1 N
a ↦ a−1
Y
Y
Y
Y
Y
Y
Y
VII. An algebra A belongs to Ω and is twisted potential and non-potential if and only if A is isomorphic
to an algebra from the following table. All such algebras are Koszul.
+
Twisted potential QA
bxyz + ayzx + czxy
−abyxz − bcxzy − aczyx
axyz + byzx + azxy
−abyxz − a2xzy − abzyx − az3
xzy⟲−xyz⟲−
1+a
2 yzy
1−a
+a(xz2+z2x+z2y)
2 (y2z+zy2−2zxz−zyz)
3 z2x
2
3 yzy
xzy⟲−xyz⟲+
3 y2z+
3 zyz+ a
3 xz2+
3 zy2−
27 z3
zyx+byxz+b2xzy−bzxy−b2xyz
2
3 zxz +
1
1
3 z2y−
−yzx+(ab−1)zxz+az2x+ab2xz2
−xyz − zxy +(a − 1)yzy
yxz − xzy + zyx + yzx
+
−
1
1
1
1
+ay2z + azy2 + z3
xzy⟲ − xyz⟲ − yzy
+ay2z⟲ + by3 + z3
xzy⟲−xyz⟲−yzy+yz2⟲
+ay3
a2xyz+yzx+azxy−a2xzy−zyx
−ayxz + a2xz2 + zyz + azxz
xyz − yzx + zxy − yxz + xzy
−zyx + y2z − yzy + zy2 + z3
x2z + axzx + a2zx2
+y2z − ayzy + a2zy2
z2y + izyz − yz2 + y2x
−yxy + xy2 + x3
z2y − izyz − yz2 + y2x
−yxy + xy2 + x3
xyx + yxy + zyx + yzy + zyz
+θxzy + θzxz + θ2xzx + θ2yxz
xyx + yxy + zyx + yzy + zyz
+θ2xzy + θ2zxz + θxzx + θyxz
T1
T2
T3
T4
T5
T6
T7
T8
T9
T10
T11
T12
T13
T14
T15
⟲
T16 y2z
T17 xy2⟲
T18 y3+yz2⟲
+ z3 + x2z − xzx + zx2
+ y3 + xz2 − zxz + z2x
+az3+x2z−xzx+zx2
Defining Relations of A
xy − ayx; zx − bxz; yz − czy
xy − byx − z2; zx − axz; yz − azy
Exceptions
abc ≠ 0
(a−b, a−c)≠(0, 0)
ab ≠ 0
a ≠ b
(1+a)(1−2a)
4
zy+
a2(1−a)
2
z2
yz − zy − az2;
xz − zx − azy +
xy−yx+(1−2a)zx+ a−1
2
a(1−a)
z2;
2 y2+
yz − zy −
xz − zx −
xy − yx −
1
3 z2;
1
3 zy −
1
3 y2 +
1
9 z2;
1
3 zx +
bxy+(1−ab)xz−yx−azx;
yz − zy − az2
−xy + yx + ay2 + z2;
xz + zx +(a−1)zy + ayz;
yz + zy
−xy + yx + ay2 + z2; yz − zy;
xz+by2+ayz−zx+(a−1)zy
−xy + yx + yz + zy; yz − zy;
xz + ay2 − zx − zy + z2
2
9 zy +
1−a
27 z2
bxz − zx;
axy − yx + 2zx; axz − zx; yz − zy + z2
xy − yx + y2 + z2; xz + zx + 2zy; yz + zy
xz + azx; yz − azy; x2 + y2
x2 + y2; xy − yx + z2; zy + iyz
x2 + y2; xy − yx + z2; zy − iyz
a ≠ 1
3
none
b ≠ 0
none
none
a ≠ 0
a ≠ 0
none
a ≠ 0
none
none
yx + θzy + θ2zx; xy + zy + θ2xz; yx + yz + θxz
none
yx + θ2zy + θzx; xy + zy + θxz; yx + yz + θ2xz
none
x2 + y2 + z2; xz − zx; yz + zy
xz − zx; xy + yx + y2; y2 + z2
xz − zx; yz + zy + x2 + az2; y2 + z2
none
none
a2 + 4 ≠ 0
8
Isomorphisms
(a, b, c) ↦(b, c, a)
(a, b, c) ↦(a−1, c−1, b−1)
(a, b) ↦(a−1, b−1)
trivial
trivial
trivial
trivial
(a, b) ↦(a, −b)
trivial
trivial
trivial
a ↦ −a
trivial
trivial
trivial
trivial
trivial
trivial
a ↦ −a
PBW
Y
Y
Y
Y
Y
Y
Y
Y
Y
Y
Y
N
N
N
N
Y
Y
N
VIII. An algebra A belongs to Ω and its quasipotential Q = QA satisfies n1(Q) = n2(Q) = 1 with
Q NOT being a cube of a degree one element if and only if A is isomorphic to an algebra from the
following table. All such algebras are Koszul.
Exceptions
Isomorphisms
PBW
none
none
none
none
none
none
none
none
none
none
none
none
none
none
none
Defining Relations of A
xy; yz; x2 + xz + azx + yx + by2 + czy
xy; yz; x2 + xz + azx + by2 + zy
xy; yz; x2 + xz + azx + y2
xy; yz; x2 + xz + azx
xy; yz; xz + azx + yx + by2 + czy + z2
xy; yz; xz + azx + by2 + zy + z2
xy; yz; xz + azx + y2 + z2
xy; yz; xz + azx + z2
xy; yz; xz + azx + yx + by2 + zy
xy; yz; xz + azx + yx + y2
xy; yz; xz + azx + yx
xy; yz; xz + azx + y2 + zy
xy; yz; xz + azx + zy
xy; yz; xz + azx + y2
xy; yz; xz + azx
xy; yz;
x2 − xz − azx − yx − cy2 − dzy −
xy; yz;
x2 − xz − azx − cy2 − zy −
b2
−(a+1)2
4
z2
xy; yz;
x2 − xz − azx − y2 −
b2
−(a+1)2
4
xy; yz;
x2 − xz − azx −
b2
−(a+1)2
4
z2
b2
−(a+1)2
z2
4
z2
Quasipotential QA
N1
N2
N3
N4
N5
N6
N7
N8
N9
xyz
xyz
xyz
xyz
xyz
xyz
xyz
xyz
xyz
N10 xyz
N11 xyz
N12 xyz
N13 xyz
N14 xyz
N15 xyz
N16 xyz
N17 xyz
N18 xyz
N19 xyz
N20 xyz
N21 xyz
N22 xyz
N23 xyz
trivial
trivial
trivial
trivial
trivial
trivial
trivial
trivial
trivial
trivial
trivial
trivial
trivial
trivial
trivial
(a, b) ↦(a, −b)
(a, b) ↦(a, −b)
(a, b) ↦(a, −b)
(a, b) ↦(a, −b)
Y
Y
Y
Y
Y
Y
Y
Y
Y
Y
Y
Y
Y
Y
Y
N
N
N
N
N
N
N
N
(1−a−b)n
a≠ − 1, b2≠(a+1)2,
1+a+b ≠ (1−a+b)n
a≠ − 1, b2≠(a+1)2,
1+a+b ≠ (1−a+b)n
(1−a−b)n
a≠ − 1, b2≠(a+1)2,
1+a+b ≠ (1−a+b)n
a≠ − 1, b2≠(a+1)2,
1+a+b ≠ (1−a+b)n
(1−a−b)n
(1−a−b)n
1+a−b
1+a−b
1+a−b
for all n∈Z+
for all n∈Z+
for all n∈Z+
1+a−b
for all n∈Z+
xy; yz;
x2 − xz − azx − yx − cy2 − dzy +
(a+1)2
4
z2
xy; yz;
x2 − xz − azx − cy2 − zy +
(a+1)2
4
z2
xy; yz;
x2 − xz − azx − y2 +
(a+1)2
4
z2
xy; yz;
x2 − xz − azx +
(a+1)2
4
z2
a≠−1, na≠n+2 for all n ∈ N trivial
a≠−1, na≠n+2 for all n ∈ N trivial
a≠−1, na≠n+2 for all n ∈ N trivial
a≠−1, na≠n+2 for all n ∈ N trivial
IX. An algebra A belongs to Λ and is potential if and only if A is isomorphic to an algebra from the
following table.
Potential QA
F1
x4 + ax2y2⟲
+ bxyxy⟲ + y4
F2
F3
F4
x2y2⟲
+ a
2 xyxy⟲
x4 + x2y2⟲
+ a
2 xyxy⟲
x3y⟲ + x2y2⟲
− xyxy⟲
Defining relations of AQ
x3 + axy2 + ay2x + 2byxy;
ax2y + ayx2 + 2bxyx + y3
xy2 + y2x + ayxy;
x2y + yx2 + axyx
x3 + xy2 + y2x + ayxy;
x2y + yx2 + axyx
x2y⟲ + xy2⟲
− 3yxy;
x3 + x2y + yx2 − 2xyx
Exceptions
4(a + b)2 ≠ 1
(a, b) ≠(0, 0)
(a, b) ≠ ±(1, 1~2)
Isomorphisms
(a, b) ↦(−a, −b)
(a, b) ↦ 1−2b
1+2a+2b ,
1−2a+2b
2(1+2a+2b)
none
none
none
trivial
trivial
trivial
9
X. An algebra A belongs to Λ and is twisted potential and non-potential if and only if A is isomorphic
to an algebra from the following table.
Twisted potential QA
x2y2 + a2y2x2 + axy2x
+ayx2y + bxyxy + abyxyx
x2y2 + y2x2 − xy2x − yx2y +(a − 1)x2yx
+(1 − a)xyx2 + ayx3 − ax3y + a
2 x4
x2y2⟲
+ax3y +(a − 1)xyx2 +(a + 1)x2yx
x2y2⟲
− xyxy⟲ − xyx2 + x2yx + ax4
− xyxy⟲ + ayx3
x2y2 + a2y2x2 + axy2x − ayx2y
x3y + yx3 + θxyx2 + θ2x2yx + y4
x3y + yx3 + θ2xyx2 + θx2yx + y4
x4 − iyx3 − y2x2 + iy3x
+y4 + xy3 + x2y2 + x3y
x4 + iyx3 − y2x2 − iy3x
+y4 + xy3 + x2y2 + x3y
x2y2 − yx2y + y2x2 − xy2x
+y3x − xy3 + yxy2 − y2xy
G1
G2
G3
G4
G5
G6
G7
G8
G9
G10
Defining relations of AQ
a2yx2 + ax2y + abxyx;
xy2 + ay2x + byxy
yx2 − x2y + ax3
xy2−y2x+(a−1)xyx+(1−a)yx2−ax2y+ a
2 x3;
xy2+y2x−2yxy+ax2y+(a−1)yx2+(a+1)xyx;
ax3 + x2y + yx2 − 2xyx
x2y + yx2 − 2xyx;
xy2 + y2x − 2yxy − yx2 + xyx + ax3
a2yx2 − ax2y; xy2 + ay2x
x2y + θyx2 + θ2xyx; x3 + y3
x2y + θ2yx2 + θxyx; x3 + y3
Exceptions
a ≠ 0
a ≠ 1
none
none
none
a ≠ 0
none
none
x3 + x2y + xy2 + y3; x3 − iyx2 − y2x + iy3
none
x3 + x2y + xy2 + y3; x3 + iyx2 − y2x − iy3
none
x2y − yx2 − y2x − xy2 + yxy; xy2 − y2x − y3 none
Isomorphisms
trivial
trivial
trivial
trivial
trivial
trivial
trivial
trivial
trivial
trivial
XI. An algebra A belongs to Λ and its quasipotential Q = QA is the fourth power of a degree 1 element
if and only if A is isomorphic to an algebra from the following table.
Quasipotential QA
y4
y4
y4
y4
y4
y4
y4
y4
y4
y4
Z1
Z2
Z3
Z4
Z5
Z6
Z7
Z8
Z9
Z10
Defining relations of AQ
y3; x3 − xy2 − ayxy − y2x
y3; x2y + xyx + yx2 − yxy
y3; x2y − xyx + yx2 − yxy
y3; x2y − yxy − ay2x
y3; yx2 − yxy − axy2
y3; x2y − ayx2 − yxy − by2x
y3; x2y − ayx2 − y2x
y3; yx2 − xy2
y3; x2y − axyx + a2yx2
y3; x2y − xyx + yx2 − xy2 − ayxy + y2x
Exceptions
a ≠ 0
none
none
a ≠ 0
a ≠ 0
a ≠ 0, a + a2 + . . . + ak + b ≠ 0
for all k ∈ N
none
none
a ≠ 0
none
Isomorphisms
trivial
trivial
trivial
trivial
trivial
trivial
trivial
trivial
trivial
trivial
XII. An algebra A belongs to Λ and its quasipotential Q = QA satisfies n1(Q) = n2(Q) = 1 with Q
NOT being a fourth power of a degree one element if and only if A is isomorphic to an algebra from
the following table.
Quasipotential QA
(x − by)(xy − ayx)(x − y)
(x − y)(xy − yx − ayy)x
x(xy − yx)y
(x − ay)xy(x − y)
(x − y)(xy − ayx)y
x(xy − ayx)y
x(xy − yx − yy)y
y(xy − yx − yy)x
Y1
Y2
Y3
Y4
Y5
Y6
Y7
Y8
Defining relations of AQ
x2y − axyx + byxy − aby2x;
xyx + xy2 − ayx2 − ayxy
x2y − xyx − axy2 − yxy + y2x + ay3;
xyx + xy2 − yx2
x2y − xyx; xy2 − yxy
x2y − ayxy; xyx − xy2
x2y − axyx + yxy − ay2x; xy2 − ayxy
x2y − axyx; xy2 − ayxy
x2y − xyx − xy2; xy2 − yxy − y3
yxy − y2x − y3; xyx − yx2 − y3
Exceptions
a ≠ 0, a ≠ 1, akb ≠ 1 for all k ∈ Z+
Isomorphisms
(a, b) ↦( 1
b)
a , 1
a ≠ 0, na + 1 ≠ 0 for all n ∈ N
none
a ≠ 1
a ≠ 1
a ≠ 1
none
none
trivial
trivial
trivial
trivial
trivial
trivial
trivial
10
Note, that graded 3-Calabi-Yau algebras are known to be potential [?].
It is also true that the
potential complex for a 3-Calabi-Yau algebra must be exact. In case of quadratic algebras on three
generators and cubic algebras on two generators, this yields that the Hilbert series of the Hilbert series
of the algebra must be (1 − t)−3 or (1 + t)(1 − t)−3 respectively. It is easy to check that algebras P 1 − P 8
and F 1 − F 4 of Theorem 1.11 are 3-Calabi-Yau. Since they are the only potential in the above tables,
they form of a complete list of quadratic 3-Calabi-Yau algebras on three generators P 1 − P 8 and cubic
3-Calabi-Yau algebras on two generators F 1 − F 4. Thus this part of classification coincide with lists
given by [19] and [20] respectively.
Let us note that the geometric meaning of our classification is that we classify orbits of the natural
action of GL3(K) on the Grassmann variety Gr(3, 9). Fix a basis x, y, z in a 3-dimensional vector space
V over K. Quadratic algebras A(V, R) with R being 3-dimensional (=given by 3 linearly independent
quadratic relations) can be interpreted as points in the 18-dimensional Grassmanian manifold G of
3-dimensional subspaces of the 9-dimensional space V 2. This turns
Ω′′ ={A ∶ A is a quadratic algebra satisfying dim A1 = 3, dim A2 = 6 and dim A3 ⩾ 10}
into an algebraic subvariety of G. Note that Ω′′ is in a sense almost the same as Ω′: Ω′ is Zarisski open
in Ω′′. The natural action of GL3(K) cuts Ω′′ into orbits with algebras from Ω′′ being isomorphic
precisely when they are in the same orbit. What we do is the following: we determine which orbits
correspond to algebras from Ω and pick a single element (a canonical form) in each orbit corresponding
to an algebra from Ω. Note that Ω′′
it is the
union of countably many subvarieties of Ω′′. Similar interpretation is available for the part of the
above theorem dealing with cubic algebras.
∖ Ω, although not Zarisski closed, is nearly like that:
Lemma 1.12. Let A = A(V, R) ∈ Ω′, Q be the corresponding quasipotential and u ∈ V ∖{0}. Then Q
is a scalar multiple of u3 if and only if u2 ∈ R.
Proof. If Q is a scalar multiple of u3, then u2 ∈ RQ ⊆ R. Now assume that u2 ∈ R. Then u3 ∈ RV ∩ V R.
Since RV ∩ V R is one-dimensional and Q ∈ V R ∩ RV , Q is a scalar multiple of u3.
Here we also give a related statement about the dual algebras.
that uv, vu ∈ R. Then A! is infinite dimensional.
Lemma 1.13. Let A = A(V, R) be a quadratic algebra and u, v be non-zero elements of V . Assume
Proof. Choose a basis x1, . . . , xn in V such that u = x1 and v = axj with a ∈ K∗ and j ∈ {1, 2}. Since
generators xk. It follows that (x1xj)n for n ∈ N are linearly independent in A! and therefore A! is
x1xj , xj x1 ∈ R, x1xj and xjx1 do not feature at all in any of the defining relations of A! written in
infinite dimensional.
The above two lemmas explain why algebras with the labels containing the letter R in Theorem 1.11
can not be Koszul. They all fall into Ω−. Thus Theorem 1.11 yields the following funny corollary.
Corollary 1.14. If a quadratic algebra A = A(V, R) over a field whose characteristic is different from
2 or 3 satisfies HA =(1 − t)−3, then
A is Koszul ⇐⇒ HA! =(1 + t)3
⇐⇒ u2 ∉ R for every non-zero u ∈ V.
Remark 1.15. The quadratic algebras among the Artin–Schelter regular algebras of global dimension
3 are precisely the twisted potential algebras (including the potential ones) in Theorem 1.11. The
classiffication of Artin–Schelter does not provide a canonical form up to an isomorphism. As it is
observed in [1], different sets of parameters in their description may lead to isomorphic algebras and
when this actually happens was left a mystery.
11
Remark 1.16. The groups featuring in the isomorphism column of the tables in Theorem 1.11 are
all finite, most being trivial. The largest order 24 occurs for Sklyanin algebras (P1).
Throughout the paper we perform linear substitutions. When describing a substitution, we keep
the same letters for both old and new variables. We introduce a substitution by showing by which
linear combination of (new) variables must the (old) variables be replaced. For example, if we write
x → x + y + z, y → z − y and z → 7z, this means that all occurrences of x (in the relations, potential
etc.) are replaced by x + y + z, all occurrences of y are replaced by z − y, while z is swapped for 7z.
A scaling is a linear substitution with a diagonal matrix. That is it swaps each variable with it own
scalar multiple. For example, the substitution x → 2x, y → −3y and z → iz is a scaling.
2 Auxiliary results
In this section, we prepare some tools needed for proving the main result. The following lemma is
very useful in dealing with algebras from Ω0.
Lemma 2.1. Let A = A(V, R) be a quadratic algebra. Then the following statements are equivalent∶
(1) A ∈ Ω0;
(2) A ∈ Ω′ and A is PBWB;
(3) dim V = dim R = 3, dim A3 ⩾ 10 and there is a basis x, y, z in V and a well-ordering on x, y, z
monomials compatible with multiplication, with respect to which the set of leading monomi-
als of elements of a basis in R is one of {xy, xz, yz}, {xy, xz, zy}, {xy, zx, zy}, {yx, yz, xz},
{yx, yz, zx} or {yx, zy, zx}.
(4) dim V = dim R = 3, dim A3 ⩾ 10 and there is a basis x, y, z in V and a well-ordering on x, y, z
monomials compatible with multiplication, with respect to which the set of leading monomials of
elements of a basis in R is {xy, xz, yz}.
Proof. The implication (1)Ô⇒(2) is obvious. Next, assume A is P BWB and A ∈ Ω′. Then dim V =
dim R = 3 and dim A3 = 10. Let x, y, z be a PBW-basis for A, while f, g, h be corresponding PBW-
generators. Since f , g and h form a Grobner basis of the ideal of relations of A, it is easy to see that
dim A3 is 9 plus the number of overlaps of the leading monomials f , g and h of f , g and h. Since
dim A3 = 10, the monomials f , g and h must produce exactly one overlap. Now it is a straightforward
routine check that if at least one of three degree 2 monomials in 3 variables is a square, these monomials
overlap at least twice. The same happens, if the three monomials contain uv and vu for some distinct
u, v ∈ {x, y, z}. Finally, the triples {xy, yz, zx} and {yx, xz, zy} produce 3 overlaps apiece. The only
option left, is for (f , g, h) to be one of the triples listed in (3). This completes the proof of implication
of the 3-element set {x, y, z} acts transitively on the 6-element set of triples from (3). Finally, assume
(2)Ô⇒(3). The implication (3)Ô⇒(4) follows from the observation that the group S3 of permutations
that (4) is satisfied. Then the leading monomials of defining relations have exactly one overlap. If
this overlap produces a non-trivial degree 3 element of the Grobner basis of the ideal of relations of
A, then dim A3 = 9, which contradicts the assumptions. Hence, the overlap resolves. That is, a linear
basis in R is actually a Grobner basis of the ideal of relations of A. Then A is PBWB. Furthermore,
the leading monomials of the defining relations are the same as for K[x, y, z] for the left-to-right or
right-to-left degree lexicographical ordering with x > y > z. Hence A and K[x, y, z] have the same
Hilbert series: HA =(1 − t)−3. That is, A ∈ Ω0. This completes the proof of implication (4)Ô⇒(1).
2.1 One and two-dimensional subspaces of V 2 with dim V = 2
The following observations are very well-known. We sketch the proofs for the sake of convenience.
12
Lemma 2.2. Let K be an arbitrary algebraically closed field (characteristics 2 and 3 are allowed here),
V be a 2-dimensional vector space over K and S be a 1-dimensional subspace of V 2 = V ⊗ V . Then S
satisfies exactly one of the following conditions∶
there is a basis x, y in V such that
(I.1) S = span{yy};
(I.2) S = span{yx};
(I.3) S = span{xy − αyx} with α ∈ K∗;
(I.4) S = span{xy − yx − yy}.
Furthermore, if S = span{xy − αyx} = span{x′y′
− βy′x′} with αβ ≠ 0 for two different bases x, y and
Proof. If V is spanned by a rank one element, then S = span{uv}, where u, v are non-zero elements
of V uniquely determined by S. If u and v are linearly independent, we set y = u and x = v to see
that (I.2) is satisfied. If u and v are linearly dependent, we set y = u and pick an arbitrary x ∈ V such
that y and x are linearly independent. In this case (I.1) is satisfied. Obviously, (I.1) and (I.2) can not
happen simultaneously. Since S in (I.3) and (I.4) are spanned by rank 2 elements, neither of them
can happen together with either (I.1) or (I.2).
x′, y′ in V , then either α = β or αβ = 1.
Now let u, v be an arbitrary basis in V and S be spanned by a rank 2 element f = auu+buv +cvu+dvv
with a, b, c, d ∈ K. A linear substitution, which keeps u intact and replaces v by v + su with an
appropriate s ∈ K turns a into 0 (one must use the fact that f has rank 2 and that K is algebraically
closed: s is a solution of a quadratic equation). Thus we can assume that a = 0. Since f has rank 2, it
follows that bc ≠ 0. If b + c ≠ 0, we set x = u + dv
b ≠ 1.
Note also that the only linear substitutions which send xy − αyx to xy − βyx (up to a scalar multiple)
with αβ ∈ K∗, α ≠ 1 are the scalings of the variables and the scalings composed with swapping x and
y. In the first case α = β. In the second case αβ = 1. Finally, if b + c = 0, then we have two options. If,
additionally, d = 0, S is spanned by xy − yx with x = u and y = v, which falls into (I.3) with α = 1. Note
that any linear substitution keeps the shape of xy − yx up to a scalar multiple. If d ≠ 0, we set x = u
and y = dv
b to see that S is spanned by xy − yx − yy yielding (I.4). The remarks on linear substitutions,
we have thrown on the way complete the proof.
b+c and y = bv to see that (I.3) is satisfied with α = c
Lemma 2.3. Let K be an arbitrary algebraically closed field (characteristics 2 and 3 are allowed here),
V be a 2-dimensional vector space over K and S be a 2-dimensional subspace of V 2 = V ⊗ V . Then
Then S satisfies exactly one of the following conditions∶
there is a basis x, y in V such that
(II.1) S = span{xx, yy};
(II.2) S = span{xx − yx, yy};
(II.3) S = span{xy, yy};
(II.4) S = span{yx, yy};
(II.5) S = span{xy − αyx, yy} with α ∈ K∗;
(II.6) S = span{xy, yx};
(II.7) S = span{xx − xy, yx};
(II.8) S = span{xx − axy − yy, yx} with a ∈ K, a2 + 1 ≠ 0.
Furthermore, α in (II.5) is uniquely determined by S. Finally, if S = span{xx − axy − yy, yx} =
− y′y′, y′x′} with (a2 + 1)(b2 + 1) ≠ 0 for two different bases x, y and x′, y′ in V , then
span{x′x′
− bx′y′
either a = b or a + b = 0.
13
Proof. Case 1: S contains yy for some non-zero y ∈ V .
Picking a basis y, w in V , we see that S is spanned by {yy, aww + bwy + cyw} for some non-zero
(a, b, c) ∈ K3. First, assume that a ≠ 0. Replacing w by αx + βy with appropriately chosen α ∈ K∗ and
β ∈ K, we can turn S into the span of either {xx, yy} or {xx − yx, yy}. If a = 0, we just set x = w and
see that S is the span of either {xy, yy}, or {yx, yy}, or {xy − αyx, yy} with α ∈ K∗. Thus we fall into
with u ∈ V ∖{0} (in a manner of speaking, S is square-free). Thus each of (II.1–II.5) is incompatible
one of (II.1–II.5). The spaces S in (II.6–II.9) are easily seen to contain no element of the shape uu
with each of (II.6–II.9). Next, (II.1) is singled out by S having two linearly independent squares. As
for S from (II.2–II.5), it contains just one square: yy. Thus any linear substitution transforming one
such S into another must send y to its own scalar multiple. Without loss of generality it sends y to
itself, while sending x to sx + ty with s ∈ K∗, t ∈ K. An easy cases by case consideration shows that
each substitution like this preserves each S from (II.5), only the substitutions with t = 0 are eligible
for S from (II.3) and (II.4), preserving it, while only the identity substitution is eligible for S from
(II.2). Thus different conditions from the list (II.1–II.5) can not happen simultaneously and α in (II.5)
is uniquely determined by S.
Case 2: S contains no square of a non-zero element of V .
It is easy to see that for every finite dimensional space V and for every 2-dimensional subspace S0 of
V ⊗V , there is a non-zero element in S0 of rank strictly less than the dimension of V (this follows easily
from the fact that the spectrum of a matrix over an algebraically closed field is always non-empty).
Hence there is an element of rank 1 in S. That is, there are non-zero u, v ∈ V such that uv ∈ S. By
the assumption of Case 2, u and v are linearly independent and therefore form a basis of V . Since S
S contains uu, contradicting the assumption of Case 2. If αβ ≠ 0, we set y = u and x = −
α = β = 0, then S contains vv, contradicting the assumption of Case 2. If α = 0, β ≠ 0, we set y = −βu
is 2-dimensional, it is spanned by uv and αuu + βvu + γvv with non-zero (α, β, γ) ∈ K3. First, consider
the case γ = 0. If α = 0, then we set x = u, y = v and observe that S is spanned by {xy, yx}. If β = 0,
that S = span{xx − xy, yx}. It remains to consider the case γ ≠ 0. Without loss of generality, γ = 1. If
and x = v to see that S = span{xx − xy, yx}. Finally, assume α ≠ 0. Since K is algebraically closed,
there is s ∈ K such that α = −s2. Setting x = v and y = su, we see that S = span{xx − axy − yy, yx} for
s ∈ K. Finally, if a2 + 1 = 0, then xx − axy − ayx − yy = (x − ay)(x − ay) ∈ S and the assumption
a = β
of Case 2 is violated. Thus a2 + 1 ≠ 0. Hence at least one of (II.6–II.8) is satisfied. Incompatibility of
different conditions from (II.6–II.8) and the fact that a in (II.8) is uniquely determined by S up to
the sign is another easy case by case consideration.
β
α v, we see
2.2 A remark on Koszulity
Let A = A(V, R) be a quadratic algebra. Fix a basis x1, . . . , xn in V . Recall that there is a specific
complex of free right A-modules, called the Koszul complex, whose exactness is equivalent to the
Koszulity of A:
. . .
dk+1
k)∗
Ð→(A!
⊗ A
dk
Ð→(A!
k−1)∗
⊗ A
dk−1
Ð→ . . .
d1
Ð→(A!
0)∗
⊗ A = A Ð→ K → 0,
(2.1)
where the tensor products are over K, the second last arrow is the augmentation map, each tensor
ϕj ⊗ xju, where ϕj ∈ (A!
product carries the natural structure of a free right A-module and dk are given by dk(ϕ ⊗ u) =
k−1)∗, ϕj(v) = ϕ(xj v). Although A! and the Koszul complex seem to
n
∑
j =1
depend on the choice of a basis in V , it is not really the case up to the natural equivalence [18]. If we
know a Grobner basis in the ideal of relations of A!, we know the multiplication table (=structural
constants) of A! with respect to the basis of normal words, which allows us to explicitly write the
matrices of the maps dk.
Note that the Koszul complex is finite (=bounded) precisely when A! is finite dimensional. Some
authors have remarked (see [11] for detailed proof) that if A! is finite dimensional and the Koszul
14
complex (2.1) of A is exact at all entries with one possible exception, then it is exact (equivalently, A
is Koszul) if and only if the duality formula HA(t)HA!(−t) = 1 is satisfied. It is also well-known that
for every quadratic algebra A, the Koszul complex (2.1) is automatically exact at its three rightmost
entries. These observations lead to the following lemma, which is our main tool in proving Koszulity.
3 ≠ {0} and A!
Lemma 2.4. Let A = A(V, R) be a quadratic algebra such that A!
Koszul if and only if HA(t)HA!(−t) = 1 and the map d3 in the Koszul complex
Ð→(A!
0)∗
0 Ð→(A!
3)∗
Ð→(A!
2)∗
Ð→(A!
1)∗
⊗ A
⊗ A
⊗ A
d1
d3
d2
⊗ A = A Ð→ K → 0
4 = {0}. Then A is
is injective.
Proof. If A is Koszul, then HA(t)HA!(−t) = 1 and the above complex is exact. In particular, d3 is
injective. Now assume that HA(t)HA!(−t) = 1 and d3 is injective. This injectivity is the same as the
exact at its three rightmost entries. Thus the exactness can potentially break at one entry (A!
2)∗
only. As we have mentioned above, the equality HA(t)HA!(−t) = 1 now yields that A is Koszul.
exactness of the above complex at its leftmost entry. As we have already mentioned the complex is
⊗ A
2.3 Remarks on generic algebras
Definition 2.5. If W is an irreducible affine algebraic variety of positive dimension over an uncount-
able algebraically closed field K, then we say that generic points of W have a property P if P holds for
all x ∈ W outside the union of countably many proper subvarieties of W . Since the field is uncountable,
such a union can never cover the whole of W . Obviously, if each of P1, . . . , Pn holds for generic x ∈ W ,
then P1 ∧ . . . ∧ Pn holds for generic x ∈ W .
We sketch the proof of the following elementary and known fact about the varieties of quadratic
algebras for the sake of convenience.
Definition 2.6. Let n, m, d ∈ N, K be any field, V be an n-dimensional vector space over K and
W ⊆ Km be an affine algebraic variety. For 1 ⩽ j ⩽ d, let qj ∶ Km → V kj be a polynomial map, where
kj ⩾ 2. For b ∈ W , we set Ab = F(V)~Ib, where Ib is the ideal generated by q1(b), . . . , qd(b). The family
{Ab ∶ b ∈ W} will be called a variety of graded algebras. If each kj equals 2, every Ab is quadratic and
we say that {Ab ∶ b ∈ W} is a variety of quadratic algebras.
Lemma 2.7. Let {Ab ∶ b ∈ W} be a variety of graded algebras. For k ∈ Z+, let hk = min{dim Ab
W}. Then the non-empty set {b ∈ W ∶ dim Ab
Proof. We can assume that k ⩾ 2 (for k ∈ {0, 1}, the set in question is the entire W ). Pick c ∈ W
k is the linear span of uqj(b)v,
where 1 ⩽ j ⩽ d, u, v are monomials and deg(uv) + kj = k, dim I b
K-matrix M(b) of the coefficients of all such uqj(b)v. Let M1(b), . . . , MN(b) be all (nk −hk) ×(nk −hk)
submatrices of M(b). For each j, let δj(b) be the determinant of the matrix Mj(b). Clearly, each δj
is a (commutative) polynomial in the variables b =(b1, . . . , bm). Obviously,
k = hk} is Zarissky open in W .
k is exactly the rank of the rectangular
k = nk − hk. Note that since I b
k = hk. Then dim I c
such that dim Ac
∶ b ∈
k
G ={b ∈ W ∶ dim Ab
k < nk
is Zarissky closed. Then U = W ∖ G = {b ∈ W ∶ dim Ab
U ={b ∈ W ∶ dim Ab
k = hk} and the result follows.
k > hk} ={b ∈ Km
∶ dim I b
− hk} ={b ∈ Km
k ⩽ hk} is Zarisski open. By definition of hk,
∶ δ1(b) = . . . = δN(b) = 0}
The following result of Drinfeld [6] features also as Theorem 2.1 in Chapter 6 in [18]. To explain it
properly, we need to remind the characterization of Koszulity in terms of the distributivity of lattices
subspaces of V n generated by the spaces V kRV n−2−k for 0 ⩽ k ⩽ n − 2 (as usual, the lattice operations
of vector spaces. Let A = A(V, R) be a quadratic algebra. For n ⩾ 3, let Ln(V, R) be the finite lattice of
are sum and intersection). Then A is Koszul if and only if Ln(V, R) is distributive for each n ⩾ 3 (see
[18, Chapter 3]). The mentioned result of Drinfeld is as follows.
15
Lemma 2.8. Let {Ab ∶ b ∈ W} be a variety of quadratic algebras and U is a non-empty Zarissky open
3 do not depend on b for b ∈ U . Then for each k ⩾ 3, the set
subset of W such that dim Ab
2 and dim Ab
{b ∈ U ∶ Lj(V, Rb) for 3 ⩽ j ⩽ k are distributive}
is Zarissky open in W .
The proof of the above lemma is a classical blend of the same argument as in the proof of Lemma 2.7
with an appropriate inductive procedure. We need the following corollary of Lemmas 2.7 and 2.8.
Lemma 2.9. Let {Ab ∶ b ∈ W} be a variety of quadratic algebras. Assume also that K is un-
hk = min{dim Ab
countable and algebraically closed and W is irreducible and has positive dimension. As above, let
hktk. Furthermore, exactly
∶ b ∈ W} for k ∈ Z+. Then for generic b ∈ W , HAb(t) =
∑
k=0
∞
k
one of the following statements holds true∶
(1) Ab is non-Koszul for every b ∈ W satisfying dim Ab
(2) Ab is Koszul for generic b ∈ W.
3 = h3 and dim Ab
2 = h2;
Proof. By Lemma 2.7, {b ∈ W ∶ dim Ab
HAb =
∞
hntn for generic b ∈ W . Next, by Lemma 2.7,
k ≠ hk} is a proper subvariety of W for each k ∈ N. By definition,
∑
n=0
U ={b ∈ W ∶ dim Ab
3 = h3 and dim Ab
2 = h2}
is a non-empty Zarissky open subset of W .
Assume now that (1) fails. Then there is c ∈ U for which Ac is Koszul. By Lemma 2.8, Wk = {b ∈ U ∶
Lj(V, Rb) for 3 ⩽ j ⩽ k are distributive} is Zarissky open in W . Since Ac is Koszul, c ∈ Wk for every
k ⩾ 3. Since for b from the intersection of Wk with k ⩾ 3, Ab is Koszul and each Wk is Zarissky open
and non-empty, (2) is satisfied. Obviously, (1) and (2) are incompatible.
If Ab is non-Koszul for every b ∈ U , (1) is satisfied.
We adjust the above results to the situation we are interested in.
Ac is Koszul. Then for every n ∈ Z+ and every b ∈ W , dim Ab
and algebraically closed, W is irreducible and has positive dimension, there is a non-empty Zarisski
Lemma 2.10. Let {Ab ∶ b ∈ W} be a variety of quadratic algebras. Assume also that K is uncountable
open subset U of W such that for each b ∈ U , H(Ab)! =(1 + t)3 and there is at least one c ∈ U for which
b ∈ W , HAb =(1 − t)−3 and Ab is Koszul.
Proof. As above, let hk = min{dim Ab
dim A2
Zarisski open, Lemma 2.7 yields h1 = 3, h2 = 6 and h3 = 10. By Lemma 2.9, HAb =
Koszul for generic b ∈ W . On the other hand, the duality formula in (1.1), implies that HAb =(1 − t)−3
whenever b ∈ U and Ab is Koszul. Thus HAb = (1 − t)−3 for generic b ∈ W . Hence (1 − t)−3 =
and therefore hn = (n+1)(n+2)
1 = 3,
3 = 10 for each b ∈ U . Since W is irreducible and U ⊆ W is non-empty and
hntn and Ab is
∶ b ∈ W}. Since H(Ab)! = (1 + t)3, we have that dim Ab
for all n ∈ Z+. By definition of hn, dim Ab
∑
n=0
for all b ∈ W .
. Moreover, for generic
b = 6 and dim Ab
n ⩾ (n+1)(n+2)
n ⩾ (n+1)(n+2)
hntn
∞
∞
∑
n=0
k
2
2
2
2.4 Degenerate algebras
Here we provide a sufficient condition for a quadratic algebra to fall outside Ω. More specifically, we
show that if an algebra is given by generators x, y, z and three quadratic relations, of which at least
two are devoid of x, then A ∉ Ω.
16
Lemma 2.11. Let K be an arbitrary field (characteristics 2 and 3 are allowed here) and A = A(V, R) ∈
Ω. Then dim R ∩ M 2 ⩽ 1 for every 2-dimensional subspace M of V .
Proof. It is easy to see that if this lemma holds for a ground field K then it holds when K is replaced
by any subfield. Thus, without loss of generality, we can assume that K is algebraically closed. Since
dim A1 = 3 and dim A2 = 6, we have dim V = dim R = 3.
−t3
Assume the contrary. Then there is a 2-dimensional subspace M of V such that dim R ∩ M 2 ⩾ 2.
First, we get rid of the trivial case R ⊆ M 2. If x, y, z is a basis in V such that y, z span M , then,
provided R ⊆ M 2, R⊥ is spanned by xx, xy, xz, yx, zx, h, where h ∈ M 2 is non-zero. By Lemma 2.3,
making a y, z linear substitution, we can turn h into one of the following forms yy, yz − zy − zz or
yz − azy with a ∈ K. Whatever the case xx, xy, xz, yx, zx and h form a Grobner basis of the ideal
of relations of A! with respect to the left-to-right degree lexicographical ordering assuming x > y > z.
Thus A! is Koszul and we can compute its Hilbert series: HA! = 1 + 3t + 3t2 + 4t3 + 5t4 + . . . = 1+t−2t2
+t3
(1−t)2
if h = yz − zy − zz or h = yz − azy and HA! = 1 + 3t + 3t2 + 5t3 + 8t4 + . . . = 1+2t−t2
if h = yy. In both
1−t−t2
cases the duality formula in (1.1) yields dim A3 > 10, which contradicts the assumptions. Thus for the
rest of the proof we can assume that S = R ∩ M 2 is two-dimensional.
By Lemma 2.3, there is a basis y, z in M such that S is spanned by one of the following sets∶ {yy, zz},
{yy − zy, zz}, {yz, zz}, {zy, zz}, {yz − αzy, zz} with α ∈ K∗, {yz, zy}, {yy − yz, zy}, {yy − zy, yz} or
{yy − αyz − zz, zy} with α ∈ K, α2 + 1 ≠ 0. Pick x ∈ V ∖ M . Then x, y, z is a basis in V . For the rest of
this proof we use the following ordering on monomials in x, y, z. First, a monomial of bigger degree is
considered to be bigger. For two monomials of equal degree, the one with bigger x-degree (x occurs
more times) is bigger. Finally we use the left-to-right lexicographical ordering with x > y > z to break
the ties. One easily sees that this order is compatible with multiplication and therefore allows to use
the Grobner basis technique. Since R is 3-dimensional, R is spanned by S together with one element
f ∈ R ∖ S.
Case 1: S is the span of one of {yy, zz}, {yz, zz}, {zy, zz}, {yz, zy}, or {yz − αzy, zz} with α ∈ K∗.
In this case, regardless what shape f has, at least two overlaps of leading monomials of the defining
relations resolve without producing a degree 3 element of the Grobner basis of the ideal of relations
of A. This yields dim A3 > 10, contradicting the assumption A ∈ Ω.
These two options reduce to each other by passing to the opposite multiplication. Thus we can
Case 2: S = span{yy − yz, zy} or S = span{yy − zy, yz}.
assume that S = span{yy − yz, zy}. Then R is spanned by yy − yz, zy and f = a1xx + a2xy + a3xz +
a4yx + a5zx + a6yz + a7zz with a =(a1, . . . , a7) ∈ K7. The overlaps yyy and zyy produce just one degree
3 element of the Grobner basis of the ideal of relations of A being yzz.
If a1 ≠ 0, the leading monomial of f is xx. Any substitution of the form x → x +αy +βz, y → y, z → z
with α, β ∈ K leaves the relations yy − yz and zy intact and preserves the shape of f . Furthermore, α
and β can be chosen to turn a2 and a3 into 0. After such a substitution the defining relations of A
turn into xx − byx − czx − dyz − pzz, yy − yz and zy with b, c, d, p ∈ K. Provided b ≠ 0, the only overlap
(other than yyy and zyy) xxx produces the second degree 3 element g of the Grobner basis of the
ideal of relations of A:
g = bxyx + cxzx + dxyz + pxzz −(d + b2
+ bc)yzx −(p + c2)zzx − cpzzz
with the leading monomial being xyx. Degree 4 overlaps produce just one degree 4 element of the
Grobner basis (comes from the overlap xxyx):
h =(b2
+ bc + d)xyzx +(p + c2)xzzx + cpxzzz −(c3
+ 2cp)zzzx −(c2p + p2)zzzz.
In the case b ≠ 0 and h ≠ 0, we obtain dim A4 = 16. Thus dim A4 = 16 for Zarisski generic(b, c, d, p) ∈ K4.
By Lemma 2.7, dim A4 ⩾ 16 for all (b, c, d, p) ∈ K4. Thus dim A4 ⩾ 16 whenever a1 ≠ 0. Hence
dim A4 ⩾ 16 for Zarisski generic a ∈ K7. By Lemma 2.7, dim A4 ⩾ 16 for all a ∈ K7. This contradicts
the assumption A ∈ Ω and concludes Case 2.
17
Case 3: S = span{yy − zy, zz}.
a = (a1, . . . , a7) ∈ K. The overlaps yyy and zzz produce just one degree 3 element of the Grobner
As above, R is spanned by yy − zy, zz and f = a1xx + a2xy + a3xz + a4yx + a5zx + a6yz + a7zy with
basis of the ideal of relations of A being yzy. We proceed as in the second case. If a1 ≠ 0, the leading
monomial of f is xx. Making a substitution x → x+αy +βz, y → y, z → z with α, β ∈ K with appropriate
α, β ∈ K, we turn the defining relations of A turn into xx − byx − czx − dyz − pzy, yy − zy and zz with
b, c, d, p ∈ K. Provided b ≠ 0, we can scale x to turn b into −1. Assuming d−c ≠ 0 and pc+c−c2 +cd−d ≠ 0,
we compute the Grobner basis of the ideal of relations of A, which turns out finite and consisting of
xx+yx−czx−dyz −pzy, yy −zy, zz, yzy, −xyx+cxzx+dxyz +pxzy +(c−d)yzx+(c−p−1)zyx+d(1−c)zyz,
(d − c)xyzx +(p + 1 − c)xzyx + d(c − 1)xzyz +((d − c)(1 − c) − cp)zyzx and xzyzx. Now we easily get
dim A7 = 37. Thus dim A7 ⩾ 37 for Zarisski generic (b, c, d, p) ∈ K4. By Lemma 2.7, dim A7 ⩾ 37 for
all (b, c, d, p) ∈ K4. Thus dim A7 ⩾ 37 whenever a1 ≠ 0. Hence dim A7 ⩾ 37 for Zarisski generic a ∈ K7.
Case 4: S = span{yy − αyz − zz, zy} with α2 + 1 ≠ 0.
a = (a1, . . . , a7) ∈ K. Since R is not contained in M 2, we have (a1, a2, a3, a4, a5) ≠ (0, 0, 0, 0, 0). The
By Lemma 2.7, dim A7 ⩾ 37 for all a ∈ K7. This contradicts the assumption A ∈ Ω. Indeed, one has
dim A7 = 36 for A ∈ Ω. This concludes Case 3.
Then R is spanned by yy − ayz − zz, zy and f = a1xx + a2xy + a3xz + a4yx + a5zx + a6yz + a7zz with
overlaps yyy and zyy produce two linearly independent degree 3 element of the Grobner basis of the
ideal of relations of A being yzz and zzz.
First, consider the case a1 ≠ 0. Using the same substitution as in Case 2, we can turn the defining
relations of A into xx − byx − czx − dyz − pzz, yy − αyz − zz and zy with b, c, d, p ∈ K. The only overlap
(other than yyy and zyy) xxx produces
g = bxyx + cxzx + dxyz + pxzz −(d + αb2
+ bc)yzx − bczyx −(p + b2
+ c2)zzx.
If g ≠ 0, we have dim A3 = 9, which is impossible since we know that dim A3 = 10. Thus g = 0, or
equivalently, b = c = d = p = 0. In the latter case, xx, yy − ayz − zz, zy, yzz and zzz form a Grobner
basis of the ideal of relations of A, yielding dim A4 = 18 > 15, which contradicts the assumptions. Thus
a1 = 0.
Now we consider the case a1 = 0 and a2 ≠ 0. In this case the defining relations of A have the shape
xy − bxz − cyx − dzx − pyz − qzz, yy − αyz − zz and zy with b, c, d, p, q ∈ K. The only overlap (other than
yyy and zyy) xyy produces
g = −(αb + 1)xzz + c(b − α)yxz + c(d + αc)yzx + d(b − α)zxz +(d2
+ c2)zzx.
If g ≠ 0, we have dim A3 = 9, which is impossible. Thus g = 0. Using the fact that α2 + 1 ≠ 0, we see
that this only happens if c = d = 0 and αb + 1 = 0. Thus α ≠ 0 and the defining relations of A have the
α xz − pyz − qzz, yy − αyz − zz and zy with p, q ∈ K. Now the defining relations together
shape xy +
with zzz and yzz form a Grobner basis of the ideal of relations of A. This yields dim A4 = 19, which
contradicts the assumptions. Thus a2 = 0.
1
Next, consider the case a1 = a2 = 0 and a3 ≠ 0. In this case the defining relations of A have the
shape xz − byx − czx − pyz − qzz, yy − αyz − zz and zy with b, c, p, q ∈ K. The only overlap (other than
yyy and zyy) xzy produces g = byxy + czxy. If g ≠ 0, we have dim A3 = 9, which is impossible. Thus
b = c = 0. Then the defining relations of A have the form xz − pyz + qzz, yy − αyz + zz and zy with
p, q ∈ K. Now the defining relations together with zzz and yzz form a Grobner basis of the ideal of
relations of A. This yields dim A4 = 19, which contradicts the assumptions. Thus a3 = 0.
In the case a1 = a2 = a3 = 0 and a4 ≠ 0, the defining relations of A have the shape yx − bzx − cyz − dzz,
yy − αyz − zz and zy = 0 with b, c, d ∈ K. The two overlaps (other than yyy and zyy) yyx and zyx
produce g = (b − α)yzx − zzx and h = bzzx. If g and h are linearly independent, we have dim A3 = 9,
which is impossible. Thus g and h are linearly dependent. This only happens if either b = α or b = 0.
If b = α, then yx − αzx − cyz − dzz, yy − αyz − zz, zy, zzz, yzz and zzx form a Grobner basis of
18
the ideal of relations of A. If b = 0, then yx − cyz − dzz, yy − αyz − zz, zy, zzz, yzz and αyzx + zzx
form a Grobner basis of the ideal of relations of A. In both cases dim A4 = 19, which contradicts the
assumptions. Thus a4 = 0.
Since (a1, a2, a3, a4, a5) ≠ (0, 0, 0, 0, 0), the only option left is a1 = a2 = a3 = a4 = 0 and a5 ≠ 0. Then
the defining relations of A have the shape zx − byz − czz, yy − αyz − zz and zy with b, c ∈ K. In this
case zx − byz − czz, yy − αyz − zz, zy, zzz and yzz form a Grobner basis of the ideal of relations of A,
which yields dim A3 = 9. This contradicts the assumptions and concludes the last case.
Remark 2.12. What we have actually proved in the above lemma, is that if A = A(V, R) is a quadratic
algebra, dim V = dim R = 3 and dim R ∩ M 2 ⩾ 2 for some 2-dimensional subspace M of V , then the
first 8 terms of HA can not be the same as for an algebra from Ω: HA ≠ 1 + 3t + 6t2 + 10t3 + 15t4 +
21t5 + 28t6 + 36t7 + . . . However, they can coincide up to the t6 term inclusively.
3 Absence of PBWB condition
In order to prove Theorem 1.11, we need to show that certain algebras are not PBWB. In this section,
we deal with this.
Lemma 3.1. Let A be the quadratic algebra given by the generators x, y, z and three quadratic relations
r1, r2 and r3 from the following list
(A1) r1 = xy, r2 = yz, r3 = a0xx+a1xz+a2yx+a3yy+a4zx+a5zy+a6zz for aj ∈ K such that a0a6 ≠ 0;
(A2) r1 = xx−xz−azx−bzz, r2 = xy+ b
a2 zz, where a, b ∈ K∗, (a, b) ≠(1, −1);
(A3) r1 = xy+byz+zz, r2 = zx+(b−1)yz−bzy−zz, r3 = yy+yz+bzy+zz, where b ∈ K, b ≠ 0, b ≠ 1.
a zz, r3 = yy−yz−
1
a zy− b
Then either dim A3 ≠ 10 or A is non-P BWB.
Proof. Assume the contrary. That is, dim A3 = 10 and A is PBWB. Since A is PBWB and A ∈
Ω′, Lemma 2.1 yields that there exist a well-ordering ⩽ on the x, y, z monomials compatible with
multiplication and satisfying x > y > z (this we can acquire by permuting the variables) and a non-
degenerate linear substitution x ↦ ux + α1y + β1z, y ↦ vx + α2y + β2z, z ↦ wx + α3y + β3z such that
the leading monomials m1, m2, m3 of the new space of defining relations satisfy
{m1, m2, m3}∈{xy, xz, yz},{xy, xz, zy},{xy, zx, zy},{yx, yz, xz},{yx, yz, zx},{yx, zy, zx}
Since xx is the biggest degree 2 monomial,
Since < satisfies x > y > z and is compatible with multiplication,
xx is absent in rj after the substitution.
(3.1)
(3.2)
four biggest degree 2 monomials are either xx, xy, yx, xz or xx, xy, yx, zx
(3.3)
(not necessarily in this order). Since each of the classes of algebras with relations from (A1), (A2) or
(A3) is closed (up to isomorphism) with respect to passing to the opposite multiplication and the two
options in (3.3) reduce to one another via passing to the opposite multiplication, for the rest of the
proof we can assume that
the set of four biggest degree 2 monomials is {xx, xy, yx, xz}.
(3.4)
Case 1: rj are given by (A1).
Since a0a6 ≠ 0, by scaling we can without loss of generality assume that a0 = a6 = 1 from the
start (that is, r3 = xx + a1xz + a2yx + a3yy + a4zx + a5zy + zz to begin with). Then (3.2) reads
0 = uv = vw = u2 +(a1 + a4)uw + a3v2 + w2.
19
Case 1a:
v = 0. Since our substitution is non-degenerate (v, w) ≠ (0, 0). Since 0 = u2 +(a1 +
a4)uw + w2, we get uw ≠ 0. By scaling x (this does not effect the leading monomials), we can assume
that u = 1. Then w is a solution of the quadratic equation w2 +(a1 + a4)w + 1 = 0. The following table
shows the coefficients in rj in front of certain monomials (after substitution). The coefficients whose
shape we do not care about are replaced by ∗.
xx xy
yx
xz
zx
yy
yz
r1
r2
r3
0
0 α2
0
0
β2
0 α2w 0
∗
∗
∗
0
α1α2 α1β2
β2w α2α3 α2β3
∗
∗
∗
If both xy and yx columns of the above matrix vanish, then (3.1) is violated. Thus at least one of
these columns is non-zero. First, assume that α2 ≠ 0. Then the yx column of the above matrix is not
in the linear span of the xy and xz columns. Using (3.4), the linear independence of the xy and yx
columns and the the obvious inequality xy > xz, we have that both xy and yx are among the leading
monomials of the defining relations. Since this contradicts (3.1), we must have α2 = 0. Since v = 0 and
our substitution is non-degenerate, we have β2 ≠ 0. Now by (3.1) and (3.4), the two biggest leading
monomials m1 and m2 of the defining relations are either xy and xz or yx and xz. Now using α2 = 0
and β2 ≠ 0, we see that both yy and yz columns are in the linear span of m1 and m2 columns, while
the zx column is not in the said span. Since zx > zy > zz, the third leading monomial must be zx.
Again, we have arrived to a contradiction with (3.1). This concludes Case 1a.
Case 1b:
v ≠ 0. As above, without loss of generality, v = 1. The equations 0 = uv = vw =
u2 +(a1 + a4)uw + a3v2 + w2 now yield u = w = a3 = 0. Taking this into account, we write the following
table of the coefficients in rj in front of certain monomials (after substitution):
xx
xy
yx
xz
zx
yy
yz
r1
r2
r3
0
α3
0
0
0 α1a2 α3a5 β1a2 β3a5
α1
0
0
β3
β1
0
α1α2 α1β2
α2α3 α2β3
q1
q2
where q1 = α2
q2 = α1β1 + α1β3a1 + α2β1a2 + α3β1a4 + α3β2a5 + α3β3.
1 + α1α3(a1 + a2 + a4) + α2α3a5 + α2
3,
If α1α3 ≠ 0 or α1a2 ≠ 0 or α1 = 0 and α3a5 ≠ 0, then the yx column of the above matrix is not in the
linear span of the xy and xz columns. Using (3.4) and the inequality xy > xz, we see that both xy
and yx are among the leading monomials of the defining relations. Since this contradicts (3.1), we
must have α1a2 = α1α3 = 0 and α3a5 = 0 if α1 = 0.
First, assume that α1 = 0. Since our substitution is non-degenerate, this yields β1α3 ≠ 0. Since
α3a5 = 0, we have a5 = 0. Then the yx column is zero. If a2 ≠ 0, the xy and xz columns are linearly
independent, which makes xy and xz the first two leading monomials of the defining relations. One
easily sees that yy and yz columns are in the linear span of xy and xz columns, while the zx column
is not in the said span. Since zx > zy > zz, zx is the last leading monomial. Thus both xz and zx
are among the leading monomials, which contradicts (3.1). This contradiction implies a2 = 0. In this
case the yx column is zero, the xy column is non-zero and the xz column is a scalar multiple of the
xy one. Thus the biggest leading monomial is xy. Moreover, using β1α3 ≠ 0, one easily checks that
xy, zx and yy columns are linearly independent. Since yy > zy > zz and yy > yz, this implies that yy
is among the leading monomials of the defining relations, which contradicts (3.1).
The above contradiction yields α1 ≠ 0. Since α1a2 = α1α3 = 0, we have a2 = α3 = 0. Since our
substitution is non-degenerate, we must have β3 ≠ 0. Now the xy column is zero, while the yx and
xz columns are linearly independent. By (3.4), this makes yx and xz leading monomials. By (3.1),
the third leading monomial must be either yz or zy. Since yy > yz and yy > zy, this means that the
20
yy column must be in the linear span of yx and xz columns. However, it is easily seen to be not the
case. This contradiction completes the proof in Case 1.
Case 2: rj are given by (A2).
In this case (3.2) reads
0 = u2
−(a + 1)uw − bw2 =(av)2
−(a + 1)(av)w − bw2 = u(av) + bw2.
Since (u, v, w) ≠ (0, 0, 0), it easily follows that w ≠ 0. Normalizing, we can assume w = 1. Then the
equation in the above display is satisfied if and only if t2 −(a + 1)t − b = (t − u)(t − av) in K[t]. This
yields u + av = a + 1 and auv + b = 0.
b ≠ −a.
Case 2a:
1−v , b = uv 1−u
1−v zx − uv 1−u
In this case, one easily sees that v ≠ 1, u ≠ 1 and uv ≠ 1 and the system
u + av = a + 1, auv + b = 0, gives a = u−1
1−v . Since b ≠ 0, we also have uv ≠ 0. Now we rewrite the
1−u
defining relations as r1 = xx − xz +
1−v zz, r2 = xy − uvzz, r3 = yy − yz +
1−u zz. We
split the substitution x ↦ ux + α1y + β1z, y ↦ vx + α2y + β2z, z ↦ wx + α3y + β3z into two consecutive
subs: first x ↦ ux, y ↦ vx + y, z ↦ x + z and, second, x ↦ x + α1y + β1z, y ↦ α2y + β2z, z ↦ α3y + β3z
(αj , βj are not the same). After the first substitution, the space R of defining relations is spanned by
(new) rj with r1 =(1 − uv)xy − v(2 − u − v)zx − v(1 − v)zz, r2 =(1 − uv)xz −(1 − u)(1 − v)zx + v(1 − u)zz,
r3 = −(1 − u)(1 − v)yx +(1 − u)yy −(1 − u)yz + v(2 − u − v)zx +(1 − v)zy + v(1 − u)zz. After performing the
second sub, we arrive to table of the coefficients in rj in front of certain monomials (the coefficients
we do not care about are replaced by ∗):
1−u zy − uv 1−v
1−v
xx
xy
0 α2(1 − uv)
0 α3(1 − uv)
0
0
r1
r2
r3
yx
∗
∗
xz
β2(1 − uv)
β3(1 − uv)
0
zx
∗
∗
q
−α2(1 − u)(1 − v) + v(2 − u − v)α3
with q = α1(−α2(1−u)(1−v)+v(2−u−v)α3)+(1−u)α2
2+(u−v)α2α3+v(1−u)α2
3.
If α2 ≠ v(2−u−v)
leading monomials of defining relations are xy, yx and xz, contradicting (3.1). Hence α2 = v(2−u−v)
Since the second sub must be non-degenerate, α3 ≠ 0. Using the last equality, we get
(1−u)(1−v) α3, then the 3 × 3 matrix of xy, yx and xz coefficients is non-degenerate. Hence the
(1−u)(1−v) α3.
q = α2
3 v2(2 − u − v)2
(1 − v)2
+ (u − v)(2 − u − v)
1 − v
+ v(1 − u)2 = α2
3
v(1 − uv)2
(1 − u)(1 − v)2 ≠ 0.
Since q ≠ 0, we easily see that either yy is among leading monomials or both xz and zx are. Either
way, we get a contradiction with (3.1).
b = −a.
Case 2b:
In this case u = v = 1 and our sub takes the shape x ↦ x + α1y + β1z,
y ↦ x + α2y + β2z, z ↦ x + α3y + β3z. After performing this sub, we arrive to table of the coefficients
in rj in front of certain monomials:
xx
0
0
0
r1
r2
r3
xy
(1−a)(α1−α3)
(1−
a)(α2 −α3)
α2−α3
1
yx
0
α1−α3
0
xz
(1−a)(β1 −β3)
a)(β2 −β3)
(1−
β2−β3
1
zx
0
β1−β3
0
yy
(α1−α3)(α1−aα3)
a α3)
(α2−α3)(α2 −
α1α2−α2
3
1
If α1 ≠ α3, then the 3 × 3 matrix of xy, yx and xz coefficients is non-degenerate. Hence the leading
monomials of defining relations are xy, yx and xz, contradicting (3.1). Thus α1 = α3. In this case
both 3 × 3 matrices of xy, xz and zx coefficients and of xy, xz and yy coefficients are non-degenerate
(to prove all these invertibilities, we use the fact that our sub is non-degenerate). This yields that
either yy or both xz and zx are among the leading monomials, which is incompatible with (3.1). This
contradiction completes Case 2.
21
Case 3: rj are given by (A3).
In this case (3.2) reads 0 = uv + bvw + ww2 = uw − vw − w2 = v2 +(b + 1)vw + w2.
Case 3a: w = 0. Since (u, v, w) ≠ (0, 0, 0), it easily follows that v = 0 and u ≠ 0. Normalizing, we
can assume u = 1. After performing the sub, we arrive to table of the coefficients in rj in front of
certain monomials (the coefficients we do not care about are replaced by ∗):
xx xy yx xz
zx yy
0 α2
0
0
0
0 α3
0
0
β2
0
0
0
β3
0
∗
∗
q
r1
r2
r3
with q = α2
2 + α2
3 +(b + 1)α2α3.
b = −1 − v −
1
1
monomial, which contradicts (3.1) and concludes Case 3a.
If α2α3 ≠ 0, one easily sees that both xy and yx must be among the leading monomials, contradicting
Case 3b: w ≠ 0. Normalizing, we can assume w = 1. Now the equations 0 = uv + bvw + ww2 =
(3.1). Thus α2α3 = 0. Since (α2, α3) ≠ (0, 0), it follows that q ≠ 0. This forces yy to be a leading
uw −vw −w2 = v2 +(b+1)vw +w2 are satisfied precisely when v2 +(b+1)v +1 = 0 and u = v +1. In this case
v)zy − zz
and r3 = yy + yz −(1 + v +
v)zy + zz. Now our sub takes form x ↦(1 + v)x + α1y + β1z, y ↦ vx + α2y + β2z,
v and the relations read: r1 = xy −(1 + v +
v)yz + zz, r2 = zx −(2 + v +
z ↦ x + α3y + β3z. After performing this sub, one easily sees that the vectors of xy and xz coefficients
are always linearly independent, while of the vectors of yx and yy coefficients, at least one is not in
the span of the vectors of xy and xz coefficients (more precisely, the opposite happens only if v = −1,
which corresponds to the excluded case b = −1). The formulas for the coefficients in this case are
rather unwieldy, so we skip them leaving the verification to the reader. The above relations between
vectors of coefficients imply that among the leading coefficients of the defining relations we find either
both xy and yx or both xz and zx or yy. Either way, (3.1) is violated. This contradiction completes
the proof.
1
v)yz +(1 + v +
1
1
4 Parts V–VII, IX and X of Theorem 1.11 and general comments
It is elementary to verify that
each Q in Theorem 1.11 is indeed the quasipotential for the corresponding algebra A.
(4.1)
Moreover, each Q has the properties claimed in Theorem 1.11: has the declared n1(Q) and n2(Q), is
Since n1(Q), n2(Q) and the Jordan normal form of MQ (in case n1(Q) = n2(Q) = n) are invariants,
or os not cyclicly invariant etc. We skip this elementary and routine linear algebra exercise. It is very
easy in each case, but since we have so many of them, we do not spell out this verification.
we immediately see that
algebras from Theorem 1.11 with different letters in their labels can not be isomorphic.
(4.2)
By (4.2), parts of Theorem 1.11 are indeed independent.
Parts VI, VII, IX and X of Theorem 1.11 follow from the description by the authors of all potential
and twisted potential quadratic algebras on three generators and cubic algebras on two generators
[11, Theorems 1.6 and 1.7]. Part V of Theorem 1.11 is a direct corollary of Lemma 1.10. Apart from
mentioning this, we prove the following lemma. Although it slightly overlaps with Parts VI and VII
of Theorem 1.11, we give it in full.
Lemma 4.1. Every algebra A = A(V, R) from (P2–P8), (T1–T11), (T16–T17), (S1–S16), (M1–M8),
(L1–L8) and (N1–N15) (in short, every algebra in Theorem 1.11 specified as P BWB) belongs to Ω0.
That is, A is indeed P BWB and HA =(1 − t)−3.
22
Proof. By (4.1) and Lemma 2.1, it suffices to show that there is a basis x, y, z in V and a well-ordering
on x, y, z monomials compatible with multiplication, with respect to which the set of leading monomials
If we keep the original basis x, y, z and use the left-to-right degree-lexicographical
(T1–T10), (S15–S16), (M1–M2), (M5–M6) and (N5–N15). If we keep the original basis x, y, z and use
the right-to-left degree-lexicographical ordering with z > y > x, then the triple of leading monomials
of elements of a basis in R is one of {xy, xz, yz}, {xy, xz, zy}, {xy, zx, zy}, {yx, yz, xz}, {yx, yz, zx}
or {yx, zy, zx}.
ordering with x > y > z, then the triple of leading monomials for R is {xy, xz, yz} for A from (P2–P8),
for R is {xy, xz, yz} for A from (M3), (S4), (N1–N4) and (S14). With respect to the same order, the
triple of leading monomials for R is {yx, zx, yz} for A from (L1–L3), (L7–L8) and is {xy, xz, zy} for
ordering with y > z > x, then the triple of leading monomials for R is {yx, zx, yz} for A from (L4–L6)
and is {yz, zx, xy} for A from (M4). If we keep the original basis x, y, z and use the left-to-right degree-
lexicographical ordering with y > x > z, then the triple of leading monomials for R is {xz, yz, yx} for
ordering with z > x > y, then the triple of leading monomials for R is {zy, xz, xy} for A from (S10–S13).
A from (M7–M8). If we keep the original basis x, y, z and use the left-to-right degree-lexicographical
A from (S2–S9). If we keep the original basis x, y, z and use the left-to-right degree-lexicographical
These considerations take care of all algebras in question except for (T11), (T16) and (T17), for which
the original basis does not work regardless which order on monomials we consider. Thus a change of
basis is needed.
For A from (T11), we perform the substitution x → x, y → y + ix, z → z, which turns the defining
relations into xz + azx, yz − azy − 2aizx and xy + yx − iyy. For A from (T16), we perform the same
substitution x → x, y → y + ix, z → z, which turns the defining relations into xz − zx, yz + zy − 2izx and
xy + yx − iyy − izz. For A from (T17), we first swap y and z turning the defining relations into yy + zz,
xy − yx and xz + zx + zz. Next, we follow up with the substitution x → x, y → y and z → z + iy, turning
the defining relations into xy − yx, yz + zy − izz and xz + zx + 2iyx − yy. In the new x, y, z basis the
leading monomials for R for each of these three algebras are {xy, xz, yz} with respect to left-to-right
degree-lexicographical ordering with x > y > z. As we have mentioned at the start, an application of
Lemma 2.1 completes the proof.
5 Proof of Part VIII of Theorem 1.11
We start by studying the family (W) of quadratic algebras Aa given by the generators x, y, z and
relations xy, yz and a1xx + xz + a2yx + a3yy + a4zx + a5zy + a6zz for a ∈ K6. Note that this family
includes all of (N1–N24).
α2.
α , a6
Lemma 5.1. Let a, b ∈ K6 and Aa, Ab be the corresponding algebras from (W). Then Aa and Ab are
isomorphic if and only if there exist α, β ∈ K∗ such that b =α2a1, αβa2, β2a3, a4, βa5
Proof. Let Aa = A(V, R) be an algebra from (W). It is easy to see that y is up to a scalar multiple the
only non-zero element v of V for which there exist non-zero u, w ∈ V satisfying uv, vw ∈ R. Next, up
to a scalar multiple, x is the only non-zero element u of V for which uy ∈ R. Finally, up to a scalar
multiple, z is the only non-zero element u of V for which yu ∈ R. These observations imply that every
linear substitution providing an isomorphism of Aa and Ab must send each of x, y, z to its own scalar
multiple (=is a scaling). Now, applying a scaling to relations of Aa, we immediately see that Aa and
Ab are isomorphic if and only if there exist α, β ∈ K∗ such that b =α2a1, αβa2, β2a3, a4, βa5
and xx − xz − αyx − βyy − azx − γzy − bzz for some (a, b, α, β, γ) ∈ K5. Then A ∈ Ω0 if b = 0 and the
Lemma 5.2. Let A be the quadratic algebra given by the generators x, y, z and the relations xy, yz
α2.
α , a6
following statements hold true∶
(5.2.1) dim An ⩾ (n+1)(n+2)
(5.2.2) if (a, b) ≠(1, −1), then HA! =(1 + t)3;
for all n ∈ Z+;
2
23
(5.2.3) if b ≠ 0 and (a, b) ≠(1, −1), then the (right module) Koszul complex of A is given by
where d0 is the augmentation, d1(u, v, w) = xu + yv + zw, d3(u) =(zu, 0, 0)
and d2(u, v, w) =(yu +(z − x)w, zv +(αx + βy)w,(ax + γy + bz)w).
d0
Ð→ K → 0
Ð→ A3 d1
Ð→ A3 d2
0 Ð→ A
Ð→ A
d3
(5.2.4) if (a, b) =(1, −1), then dim A5 ⩾ 24 and therefore A ∉ Ω;
(5.2.5) if b ≠ 0, (a, b) ≠(1, −1) and there exists a non-zero homogeneous of degree k element u of A such
that zu = 0 in A, then dim Aj > (j +1)(j +2)
for some j ⩽ k + 2.
(5.1)
2
Proof. If b = 0, then with respect to the right-to-left degree lexicographical ordering satisfying z > y > x,
the leading monomials of the defining relations of A are xy, xz and yz. By Lemma 2.1, A ∈ Ω0.
shape xx +
Throughout the rest of the proof, we use the left-to-right degree lexicographical ordering on x, y, z
monomials assuming x > y > z. Since proving the above statements in the case when K is replaced by
any field extension will yield their validity for original K, we can without loss of generality, assume
that K is uncountable.
As we have already observed, A ∈ Ω0 if b = 0.
K3. By Lemma 2.7, dim A5 ⩾ 24 for all α, β, γ justifying (5.2.4).
x by x + z. The defining relations of A take the shape xx − αyx − βyy − γzy, xy + zy, yz. A direct
of relations of A is finite and the leading monomials of its members are xx, xy, yz, yyx, xzy, zyyy,
Assuming (a, b) = (1, −1), we make the substitution leaving y and z as they were and replacing
computation shows that for (α, β, γ) from a Zarisski open subset of K3, the Grobner basis of the ideal
yyyy, zzzyx, zzyxz and xzzyx. This yields dim A5 = 24 for (α, β, γ) from a Zarisski open subset of
In particular, A is Koszul, HA = (1 − t)−3 and
HA! = (1 = t)3 provided b = 0. Now assume b ≠ 0. In this case, the defining relations of A! take the
that provided (a, b) ≠ (1, −1), the defining relations together with yzz and zzz form a Grobner basis
and xyz yielding HA! =(1 + t)3 and justifying (5.2.2). Furthermore, the normal words in A! furnish us
constants. Now a routine computation yields (5.2.3). By the already verified (5.2.2), HA! = (1 + t)3
provided (a, b) ≠ (1, −1). We also know that A is Koszul if b = 0. By Lemma 2.10 (the corresponding
Now we shall verify (5.2.5). Assume that b ≠ 0, (a, b) ≠ (1, −1) and there exists a non-zero homo-
of the ideal of relations of A!. The complete list of normal words for A! now is 1, x, y, z, xy, yz, zz
with a graded linear basis in A!, while the above Grobner basis provides the corresponding structural
geneous of degree k element u of A such that zu = 0 in A. Without loss of generality, we can assume
that k is the minimal positive integer for which such an u exists. Towards a contradiction assume
that (5.2.5) fails. That is, dim Aj ⩽ aj for all j ⩽ k + 2, where aj = (j +1)(j +2)
. According to the already
verified (5.2.1), dim Aj = aj for all j ⩽ k + 2. Now we consider the following graded 'slice' of the Koszul
complex (5.1) of A:
variety W is the affine space K5), (5.2.1) is satisfied.
γ
b zz. A direct computation shows
β
b zz and zy −
b zz, yx − α
b zz, zx − a
1
b zz, xz −
b zz, yy −
1
2
δ2
Ð→ A3
δ3
Ð→ A3
k
δ1
Ð→ Ak+2 → 0,
k+1, A3
k+1
0 Ð→ Ak−1
(5.2)
where δ1, δ2 and δ3 are restrictions of d1, d2 and d3 to A3
k and Ak−1 respectively. Since (5.1)
(as a Koszul complex of a quadratic algebra) is exact at its three rightmost terms, (5.2) is exact at
k+1 and at Ak+2. Since dim Aj = aj for all j ⩽ k + 2, this yields that the dimension of the kernel of
A3
δ2 is 3ak − 3ak+1 + ak+2 = ak−1 (the last equality follows from the definition of aj). The minimality of
k and the shape of d3 yields that δ3 is injective. Hence (5.2) is exact at Ak−1 and the image of δ3 has
the dimension ak−1. Since the latter coincides with the dimension of the kernel of δ2 the said image
and kernel coincide. On the other hand, one easily sees that δ2(0, u, 0) = d2(0, u, 0) = 0 (since zu = 0),
while (0, u, 0) is clearly not in the image of δ3. This contradiction concludes the proof of (5.2.5).
We also need two sequences of polynomials. For n ∈ Z+, let pn, qn ∈ K[x, y] be defined by
(5.3)
pn
qn = 1
y
1 −x n 1
1 .
24
qn(a, b) = (1−a+d)n+1
qn(a, b) = (n+1)(1−a)n
2n
,
−(1−a−d)n+1
2nd
, pn(a, b) = qn(a, b) +
pn(a, b) = (n+2−na)(1−a)n
2n−1
(a+b)((1−a+d)n −(1−a−d)n )
2n−1d
if d ≠ 0;
if d = 0.
(5.4)
Equivalently, polynomials pn and qn can be defined recurrently by p0 = q0 = 1, pn+1 = pn + yqn,
qn+1 = pn − xqn. On few occasions, we will have to deal with the condition pn(a, b) ≠ 0 for all n ∈ N
imposed on (a, b) ∈ K2. The following remark simplifies this condition.
Remark 5.3. By expressing the matrix
M = 1
1 −a
b
with a, b ∈ K as a conjugate of a matrix in Jordan normal form (the eigenvalues of M are α = 1−a+d
and β = 1−a−d
2 with d2 =(a + 1)2 + 4b), one easily gets explicit formulas for pn(a, b) and qn(a, b):
2
Furthermore, if pn(a, b) = 0 and d ≠ 0, then (1 − a + d)(1 + a − d) ≠ 0 and one easily concludes that
1−a+dn+1
pn(a, b) = 0 ⇐⇒ 1−a−d
pn(a, b) = 0 ⇐⇒ na = n + 2
= 1+a+d
1+a−d
if d ≠ 0;
if d = 0.
(5.5)
Lemma 5.4. Let A be the quadratic algebra given by the generators x, y, z and the relations xy, yz
and xx − xz − αyx − βyy − azx − γzy − bzz for some a, b, α, β, γ ∈ K5 such that b ≠ 0 and (a, b) ≠(1, −1).
Let also n ∈ N be such that pk(a, b) ≠ 0 for 0 ⩽ k < n and qk(a, b) ≠ 0 for 0 ⩽ k ⩽ n. Then the equalities
xzkx − skxzk+1
− tkyk+1x − hkyk+2
− zuk = 0 and xzky − rkyk+2
− zvk = 0
(5.6)
γ
sk
hold in A for 0 ⩽ k ⩽ n, where sk = pk(a,b)
, hk+1 = βsk(rk −t−k)−γhk
by u0 = ax + γy + bz, v0 = 0, t0 = α, h0 = β, r0 = 0, tk+1 = α(rk −tk)−hk
sk −auk(z − x −
uk+1 = 1
imposed upon a, b not only sk ≠ 0, but also sk ≠ a for 0 ⩽ k < n). Moreover, the equalities
(sn − a)xzn+1x −(sn + b)xzn+2 +(αtn + hn − αrn)yn+2x + β(tn − rn)yn+3 − γxzn+1y
qk(a,b) , while tk, hk, rk ∈ K and uk, vk ∈ Ak+1 are defined inductively
hk
,
sk
uy for 0 ⩽ k < n (note that by the conditions
y) + vk(αx + βy) and vk+1 = −
−z(un(z − x) + vn(αx + βy)) = 0 and snxzn+1y + hnyn+3 + zuny = 0
hold in A. Furthermore, the left-hand sides of (5.6) for 0 ⩽ k ⩽ n together with yz are all the elements
of degree up to n + 2 of the reduced Grobner basis of the ideal of relations of A and dim Aj = (j +1)(j +2)
for 0 ⩽ j ⩽ n + 2.
, rk+1 = −
sk(sk −a)
(5.7)
sk −a
1
sk
2
Proof. We shall prove (5.6) for 0 ⩽ k ⩽ n inductively. The defining relations xy = 0 and xx − xz −
αyx − βyy − azx − γzy − bzz = 0 justify (5.6) for k = 0, providing the basis of induction. Assume that
0 ⩽ k ⩽ n − 1 and (5.6) holds for k. We shall verify (5.6) for k replaced by k + 1, which will complete
the inductive proof. We do this by resolving certain overlaps as in the Buchberger algorithm using
the induction hypothesis and the defining relations of A in the process. The monomials to which we
apply reduction are indicated by underlining:
xzkxy → 0 = skxzk+1y + tkyk+1xy + hkyk+3
+ zuky = skxzk+1y + hkyk+3
+ zuky,
which yields
Since sk ≠ 0, we can divide by sk and use the definitions of rj and vj to see that
skxzk+1y + hkyk+3
+ zuky = 0 in A.
(5.8)
xzk+1y − rk+1yk+3
− zvk+1 = 0 in A,
25
which is the second equality in (5.6). Now we proceed in the same manner with another overlap:
xzkxx → 0 = skxzk+1x + tkyk+1xx + hkyk+2x + zukx − xzkxz
−axzk+1x − bxzk+2 − αxzkyx − βxzkyy − γxzk+1y
= skxzk+1x + tkyk+1xz + atkykyzx + btkykyzz + αtkyk+2x + βtkyk+3 + γtkykyzy
+hkyk+2x + zukx − skxzk+2 − tkyk+1xz − hkyk+2yz − zukz − axzk+1x
−bxzk+2 − αrkyk+2x − αzvkx − βrkyk+3 − βzvky − γxzk+1y.
After obvious cancelations and rearrangements, we get
(sk − a)xzk+1x −(sk + b)xzk+2 +(αtk + hk − αrk)yk+2x + β(tk − rk)yk+3
−γxzk+1y − z(uk(z − x) + vk(αx + βy)) = 0 in A.
(5.9)
Using the already verified equality xzk+1y − rk+1yk+3 − zvk+1, dividing by sk − a and using the obvious
equality sk+1 = sk +b
sk −a together with the definitions of tj, hj and uj, we obtain the first equality in (5.6).
This completes the inductive proof of (5.6) for 0 ⩽ k ⩽ n. Now (5.7) is the combination of (5.8) and
(5.9) for k = n, which hold since (5.6) holds for k = n.
Now the only degree ⩽ n + 2 monomials, which do not contain yz as well as xzkx and xzky for
0 ⩽ k ⩽ n as submonomials are the words of the shape zj ym and zj ymxzp of degree ⩽ n + 2 with
j, m, p ∈ Z+. The number of such monomials of degree k is exactly (k+1)(k+2)
. Since by Lemma 5.2,
dim Ak ⩾ (k+1)(k+2)
elements of degree up to n + 2 of the reduced Grobner basis of the ideal of relations of A and we must
have dim Ak = (k+1)(k+2)
, yz together with the left-hand sides of (5.6) for 0 ⩽ k ⩽ n must comprise all the
for 0 ⩽ k ⩽ n + 2.
2
2
2
Lemma 5.5. Let A be the quadratic algebra given by the generators x, y, z and the relations xy, yz
and xx − xz − αyx − βyy − azx − γzy − bzz for some a, b, α, β, γ ∈ K5 such that b ≠ 0. Assume also that
from (N16–N19) belong to Ω+ are Koszul and non-P BWB.
pk(a, b) ≠ 0 for all k ∈ N. Then A ∈ Ω+, A is Koszul and A is non-P BWB. In particular, algebras
Proof. Note that p1(a, b) ≠ 0 yields b ≠ −1 and therefore (a, b) ≠(1, −1).
Case 1: qk(a, b) ≠ 0 for all k ∈ N.
By Lemma 5.4, HA =(1 − t)−3 and the leading monomials of the reduced Grobner basis of the ideal
of relations of A are yz, xzkx and xzky for k ∈ Z+. Since none of these monomials starts with the
smallest variable z, we have zu ≠ 0 for all non-zero u ∈ A. The exact shape of the Koszul complex of
A provided by Lemma 5.2 now allows us to say that this complex is exact at its left-most term (d3 is
injective). By Lemma 2.4, A is Koszul.
Case 2: qk(a, b) = 0 for some k ∈ N.
Let n ∈ Z+ be the minimal non-negative integer for which qn+1(a, b) = 0. By Lemma 5.4, (5.6)
is satisfied for 0 ⩽ k ⩽ n and (5.7) holds. Since qn+1(a, b) = 0, we have sn = a for sj defined in By
Lemma 5.4. Next, observe that a + b ≠ 0 and a ≠ 0. Indeed, if a + b = 0, then qk(a, b) = pk(a, b) ≠ 0,
while if a = 0, then qk(a, b) = 1 for all k ∈ Z+ thus contradicting the assumption of Case 2. Taking this
into account, we can rewrite (5.7) as follows:
xzn+2 −
xzn+1y + hn
αtn +hn−αrn
a+b
yn+2x −
β(tn −rn)
a+b
yn+3 +
γ
a+b xzn+1y − zu = 0;
a yn+3 + zv = 0 for some u, v ∈ An+2.
(5.10)
Now we observe that the number of monomials of degree m, which do not contain any of yz, xzky
for 0 ⩽ k ⩽ n + 1, xzkx for 0 ⩽ k ⩽ n and xzn+2 as submonomials is exactly (m+1)(m+2)
for every m ∈ Z+.
Since by Lemma 5.2, dim Am ⩾ (m+1)(m+2)
, yz together with the left-hand sides of (5.6) for 0 ⩽ k ⩽ n
and the left-hand sides of (5.10) must form the reduced Grobner basis of the ideal of relations of A
and we must have HA = (1 − t)−3. Since none of the leading monomials of this basis starts with the
2
2
26
smallest variable z, we have zu ≠ 0 for all non-zero u ∈ A. Exactly as in Case 1, we can now use
Lemma 5.2 and Lemma 2.4 to conclude that A is Koszul. Note that in the second case the Grobner
basis turns out to be finite. Finally, A is non-PBWB according to Lemma 3.1. The comment about
algebras in (N16–N19) is a direct corollary of the above.
Lemma 5.6. Let A be the quadratic algebra given by the generators x, y, z and the relations xy, yz
and xx − xz − αyx − βyy − azx − γzy − bzz for some a, b, α, β, γ ∈ K5 such that pk(a, b) = 0 for some
k ∈ N. Then A ∉ Ω.
Proof. As on few occasions above, we can without loss of generality assume that K is uncountable. By
defined by one non-trivial polynomial equation, W1 must be one-dimensional. Then W = W1 × K3 is a
4-dimensional irreducible affine variety. The plan of the proof is the following. Instead of dealing with
Lemma 5.2, A ∉ Ω if (a, b) =(1, −1). Thus for the rest of the proof we can assume that (a, b) ≠(1, −1).
Let n be the smallest positive integer for which pn(a, b) = 0. Consider the variety W0 = {(s, t) ∈ K2 ∶
pn(s, t) = 0}. We strongly suspect that W0 is irreducible, however, we do not see an easy way to
demonstrate it. Thus we take an irreducible component W1 of W0, which contains (a, b). As W0 is
the given specific algebra, we shall demonstrate that for generic (a, b, α, β, γ) ∈ W , HA ≠(1 − t)−3. On
the other hand by Lemma 2.9, for such generic A, HA = Hmin, where Hmin is coefficient-wise minimum
of HA for A with parameters from W . Furthermore, by Lemma 5.2, dim Aj ⩾ (j +1)(j +2)
for all j ∈ Z+.
Combining these, we see that if HA ≠ (1 − t)−3 for generic (a, b, α, β, γ) ∈ W , then HA ≠ (1 − t)−3 for
all (a, b, α, β, γ) ∈ W , which will include the original specific algebra. Thus the proof will be complete
if we demonstrate that HA ≠(1 − t)−3 for generic (a, b, α, β, γ) ∈ W .
Using the explicit description of pn and qn given in Remark 5.3, one easily sees that qk(a, b) ≠ 0
for all k ∈ Z+ for all (a, b) ∈ W0 with countably many possible exceptions. Thus qk(a, b) ≠ 0 for all
k ∈ Z+ for generic (a, b, α, β, γ) ∈ W . Let sk, tk, hk, rk, uk and vk be as in Lemma 5.4. From their
definition, one easily sees that hn ≠ 0 for generic (a, b, α, β, γ) ∈ W . Now if α = β = γ = 0, then from
the recurrent definition of uk, vk it follows that vk = 0 for 0 ⩽ k ⩽ n, while uk = zk(akx + bkz) for
to check that bn = 0 only for finitely many (a, b) ∈ W0. From the inequality dim Aj ⩾ (j +1)(j +2)
it
follows that provided hn ≠ 0, the left-hand sides of (5.6) and (5.7) form the degree up to n + 3 part of
a Grobner basis of the ideals of relations of A. Thus, if α = β = γ = 0, unyy = bnzn+1yy ≠ 0 provided
bn ≠ 0 with the latter being true with finitely many exceptions. Now one easily sees that unyy ≠ 0 for
the defining relations, we get hnyn+4 = 0 in A. Now multiplying hnyn+3 + zuny = 0 by y on the right
generic (a, b, α, β, γ) ∈ W . By (5.7), hnyn+3 + zuny = 0 in A. Multiplying by y on the left and using
and using hnyn+4 = 0, we get zunyy = 0. Thus for generic (a, b, α, β, γ) ∈ W , the non-zero degree n + 3
element unyy satisfies zunyy = 0 in A. By Lemma 5.2, there is j ⩽ n + 5 for which dim Aj > (j +1)(j +2)
Thus for generic (a, b, α, β, γ) ∈ W , HA ≠(1 − t)−3, which completes the proof.
0 ⩽ k ⩽ n, where a0 = a, b0 = b, ak+1 = −
for 0 ⩽ k ⩽ n. It is straightforward
aak +bk
sk −a and bk+1 = bk −bak
sk −a
2
2
.
2
Lemma 5.7. All algebras in (N1–N23) belong to Ω, all these algebras are Koszul, algebras from (N1–
N15) are P BWB, while algebras from (N1–N15) are non-P BWB. Algebras in (N1–N23) with different
labels are non-isomorphic and the isomorphism conditions of Theorem 1.11 within algebras with a
given label from (N1–N23) are satisfied.
Proof. The isomorphism statement follows easily from Lemma 5.1. Algebras from (N1–N15) belong
to Ω0 by Lemma 4.1. Algebras in (N16–N23) belong to Ω+ and are Koszul according to Lemma 5.5
and Remark 5.3. These algebras are non-PBWB by Lemma 3.1.
Lemma 5.8. Let A = A(V, R) ∈ Ω and the corresponding quasipotential Q is not a cube and satisfies
n1(Q) = n2(Q) = 1. Then there is a basis x, y, z in V such that Q = xyz and R is spanned by xy, yz,
a0xx + a1xz + a2yx + a3yy + a4zx + a5zy + a6zz with some a =(a0, . . . , a6) ∈ K7 satisfying a1 ≠ 0.
27
possible linear dependencies between x, u and v, one easily sees that y, z ∈ V can be chosen in such a
Proof. Since n1(Q) = n2(Q) = 1, Q = xuv, where x, u, v are non-zero elements of V . Considering
way that x, y, z is a basis in V and Q has one of the following forms: xxx, xxz, xzz, xzx, x(x − z)z,
or Q = x(x − z)z, RQ intersects M 2 by at 2-dimensional space, where M is spanned by x and z. By
xyz. If Q = xxx, Q = xxz or Q = xzz, RQ ⊆ R contains a square and therefore Q must be a cube
according to Lemma 1.12. Since this is not the case, Q can not be xxx or xxz or xzz. If Q = xzx
Lemma 2.11, A ∉ Ω, which is a contradiction. This leaves only one option:
Q = xyz.
In this case RQ is spanned by xy and yz. Thus R must be spanned by xy, yz and a0xx + a1xz +
a2yx + a3yy + a4zx + a5zy + a6zz for some non-zero (a0, . . . , a6) ∈ K7. It remains to show that a1 ≠ 0.
Assume the contrary. That is, a1 = 0. If a0a2a3 ≠ 0, we can scale to get a0 = a2 = −a3 = 1. The defining
relations of A now take the shape xy = yz = 0 and xx = pyx − yy + zx + azy + bzz with a, b, p ∈ K. We
use the usual left-to-right degree lexicographical ordering on monomials assuming x > y > z. A direct
computation shows that for (a, b, p) from a Zarisski open subset of K3, the leading monomials of the
members of the Grobner basis of the ideal of relations of A of degrees up to 5 are xx, xy, yz, yyy,
xzx, xzyy, zzyy, xzyx, zzzzy and xzzyx. More specifically, for this to be the case, one needs a, b,
b + a2, a + p and p2b − pa − b − a2 to be non-zero. In this case we get dim A5 = 22. By Lemma 2.7,
dim A5 ⩾ 22 for all a, b, p ∈ K. This means that dim A5 ⩾ 22 whenever a0a2a3 ≠ 0 (and a1 = 0 still).
Again, Lemma 2.7 yields that dim A5 ⩾ 22 if a1 = 0 regardless what other aj are. Since this inequality
is incompatible with A ∈ Ω, we arrive to a contradiction, which proves that a1 ≠ 0.
We are ready to prove Part VIII of Theorem 1.11. According to Lemma 5.7, the only thing which
we have to verify is that every A = A(V, R) ∈ Ω such that the corresponding quasipotential Q is not
a cube and satisfies n1(Q) = n2(Q) = 1 is isomorphic to one of the algebras in (N1–N23). Using
Lemma 5.8 and a suitable scaling to turn a1 into 1, we can assume that A is given by generators
x, y, z and relations xy, yz and a1xx + xz + a2yx + a3yy + a4zx + a5zy + a6zz for some a ∈ K6. Now
using Lemma 5.1, Lemma 5.6, Remark 5.3 and considering possible distributions of zeros among
the parameters aj, it is easy to see that A is isomorphic to one the algebras in (N1–N23). Indeed,
Lemma 5.1 describes possible isomorphisms within our family of algebras, Lemma 5.6 pinpoints the
exceptional parameters and Remark 5.3 translates the description of these exceptions to the form used
in Theorem 1.11 (that is, without polynomials pk).
6 Proof of parts II and III of Theorem 1.11
Lemma 6.1. The algebras in (M1–M8) and (L1–L8) belong to Ω0 and therefore are P BWB and
Koszul. Algebras in (M1–M8) and (L1–L8) with different labels are non-isomorphic and the isomor-
phism conditions of Theorem 1.11 within algebras with a given label from (M1–M8) and (L1–L8)are
satisfied.
Proof. Algebras in (M1–M8) and (L1–L8) belong to Ω0 according to Lemma 4.1. It remains to deal
with isomorphisms. Recall that we already know that algebras from Theorem 1.11 with different letters
in the labels are non-isomorphic. Thus the families (M1–M8) and (L1–L8) can be treated separately.
First, observe that for every algebra in (M1–M8), the rank one quadratic relations are non-zero
members of the linear span of xy and xz. Thus any linear substitution providing an isomorphism
between two algebras from (M1–M8) must preserve the linear span of x as well as the linear span
of y and z. Now for every A from (M1–M8), the quotient B = A~I by the ideal generated by x
is given by generators y, z and one quadratic relation: yz − azy with a ∈ K∗ for A from (M1–M3),
yz − zy − zz for A from (M4–M5) and zy for A from (M7–M8). Since our substitution sends x into
its own scalar multiple, it must provide an isomorphism of the corresponding algebras B as well. By
28
Lemma 2.2, we now have that algebras from different groups (M1–M3), (M4–M6) and (M7–M8) can
not be isomorphic. Furthermore, the only automorphisms of each of the two algebras K⟨y, z⟩~Id(zy)
and K⟨y, z⟩~Id(yz − zy − zz) are given by scalings. Just by looking at the sets of monomials involved
in defining relations, it now becomes obvious that algebras in (M4–M8) are pairwise non-isomorphic.
These considerations take (M4–M8) out of the picture, leaving us to deal with (M1–M3). Now for
each presentation (M1–M3), it is easy to see that a scaling either throws it outside the class (M1–M3)
or provides an automorphism. Other than scalings, the only other y, z substitutions that transform
yz −azy with a ∈ K∗ into yz −bzy with b ∈ K∗ are certain scalings composed with the swapping of y and
z. Such a substitution (combined with a scaling of x) throws every presentation from (M2) outside
the class (M1–M3). As for each of (M1) and (M3), such a substitution provides an isomorphism of A
and an algebra from the same class with the parameter a replaced by 1
a . This completes the proof of
the isomorphism statement for (M1–M8). Since algebras in (L1–L8) are isomorphic to algebras from
(M1–M8) with the opposite multiplication, the isomorphism statement for (L1–L8) follows as well.
quadratic relations from (M1–M8) of Theorem 1.11.
Lemma 6.2. Let A = A(V, R) ∈ Ω be such that the corresponding quasipotential Q = QA satisfies
n1(Q) = 1 and n2(Q) = 2. Then A is isomorphic to to a K-algebra given by generators x, y, z and three
Proof. Since n1(Q) = 1 and n2(Q) = 2, there is x ∈ V such that Q = xf , where f ∈ V 2 has rank 2. Let
1. By Lemma 2.2, a basis y, z in M can be chosen in such a way that g ∈{zz, yz, yz − αzy, yz − zy − zz},
Case 1a: g = zz or g = yz. In this case Q = pxxy + qxxz + xwz with p, q ∈ K, where w ∈{y, z}. Then
Case 1: x ∉ M . Then Q = xxu + xg, where u ∈ M and g ∈ M 2. Clearly, g ≠ 0: otherwise f has rank
M be the (unique) 2-dimensional subspace of V such that f ∈ V M .
RQ is spanned by pxy + qxz + wz, pxx and qxx + xw. If p ≠ 0, then xx ∈ RQ ⊆ R. By Lemma 1.12, Q
is a scalar multiple of xxx, which is obviously not the case. Hence p = 0. Then f = qxz + wz has rank
1. This contradiction completes Case 1a.
where α ∈ K∗. Clearly, x, y, z is a basis in V .
px−z
1
α xyz + xzy − qxzx +
α , Q acquires the shape Q = −
Case 1b: g = yz −αzy with α ∈ K∗. In this case Q = pxxy +qxxz +xyz −αxzy with p, q ∈ K. After the
p
α xyx.
linear substitution x → x, y → y − qx, z →
Depending on which of p, q is or is not zero, a scaling turns Q into one of the following forms:
Q = xyz − αxzy, Q = xyz − αxzy + xyx, Q = xyz − αxzy + xzx or Q = xyz − αxzy + xyx + xzx. The
families of quasipotentials Q = xyz − αxzy + xyx for α ∈ K∗ and Q = xyz − αxzy + xzx for α ∈ K∗ turn
into each other by swapping y and z together with a scaling. Thus we end up with three families of
quasipotentials: Q = xyz − αxzy, Q = xyz − αxzy + xzx and Q = xyz − αxzy + xyx + xzx with α ∈ K∗.
If α = 1, the quasipotential Q = xyz − αxzy + xyx + xzx transforms into Q = xyz − xzy + xzx by means
of the substitution x → x, y → y − z, z → z. Thus we can exclude α = 1 from the last family. For the
three families, we have obtained, RQ = R is spanned by {xy, xz, yz − αzy}, {xy, xz, yz − αzy + zx} and
{xy, xz, yz − αzy + zx + yx} respectively. Thus A is isomorphic to an algebra from (M1–M3).
g = yz − zy − zz. In this case Q = pxxy + qxxz + xyz − xzy − xzz with p, q ∈ K. After
the substitution x → x, y → y + z − qx, z → z + px, Q acquires the shape Q = xyz − xzy − xzz + qxzx +
pxyx − pxzz − p2xxx. If p ≠ 0, a substitution x → x, y → y + sz, z → z with an appropriate s ∈ K kills
q. After that, a scaling turns p into 1. This yields Q = xyz − xzy − xzz + xyx − xzz − xxx. After the
sub x → x, y → x + y + z, z → z we arrive to Q = xyz − xzy − xzz + xyx, the quasipotential of (M4). If
p = 0 and q ≠ 0, a scaling turns q into 1: Q = xyz − xzy − xzz + xzx, which is the quasipotential from
(M5). Finally, if p = q = 0, we have Q = xyz − xzy − xzz, the quasipotential from (M6). This concludes
Case 1.
Case 1c:
Case 2: x ∈ M .
Now we can pick y, z ∈ V such that x, y, z is a basis in V , while x, y form a basis in M . Then
Q = xzu + xg, where u ∈ M and g ∈ M 2.
Case 2a: u = 0. Then g = f must have rank 2. Hence RQ ⊂ R ∩ M 2 is at least 2-dimensional. By
Lemma 2.11, A ∉ Ω, which is a contradiction.
29
Case 2b: x and u are linearly independent. Without loss of generality, we can then assume that
y = u. Then Q = xzy + axxx + bxxy + cxyx + dxyy with a, b, c, d ∈ K. The substitution x → x, y → y,
z → z + sx + ty with appropriate s, t ∈ K kills b and d. This makes Q = xzy + axxx + cxyx with a, c ∈ K.
If c = 0, then a ≠ 0 (otherwise n2(Q) = 1) and therefore xx ∈ RQ ⊆ R. This, however, can not happen
according to Lemma 1.12. Scaling, we can make c = 1 and a ∈{0, 1}. If a = 0, Q = xzy + xyx and R is
spanned by{zy +yx, xz, xy}. If a = 1, Q = xzy +xyx+xxx and R is spanned by{zy +yx+xx, xz, xy +xx}.
The first algebra is (M7), while the second is isomorphic to (M8): just use the sub x → x, y → y − x,
z → z.
Case 2c: u and x are linearly dependent and u ≠ 0. Without loss of generality, we can assume
u = x. Then Q = xzx + axxx + bxxy + cxyx + dxyy with a, b, c, d ∈ K. The substitution x → x, y → y,
z → z + sx + ty with appropriate s, t ∈ K kills a and c. This makes Q = xzx + bxxy + dxyy with
b, d ∈ K. If d = 0, we have xx ∈ R, which can not happen according to Lemma 1.12. Thus d ≠ 0. A
normalization turns d into 1: Q = xzx + bxxy + xyy. The sub x → x, y → y − bx, z → z turns Q into
Q = xzx + xyy − bxyx. Then R is spanned by xy, xz and yy + zx − byx. If b ≠ 0, we use the usual
left-to-right degree-lexicographical ordering assuming x > y > z to see that xy, xz, yy − byx + dzx, yyy
and yyz form a Grobner basis of the ideal of relations of A. This yields dim A4 = 17 provided b ≠ 0.
By Lemma 2.7, dim A ⩾ 17 for all b, contradicting A ∈ Ω. This concludes Case 2.
Part II of Theorem 1.11 follows straight away from Lemmas 6.1 and 6.2. Part III of Theorem 1.11
is equivalent to Part II: just pass to the opposite multiplication.
7 Proof of Part IV of Theorem 1.11
First we prove the following lemma, which deals with a family of algebras containing (S19) and (S20)
Lemma 7.1. Let A = Aa,b be the quadratic algebra given by the generators x, y, z and the relations
xx − xz − azx − bzz, xy + b
are
non-isomorphic provided (a, b) ≠(a′, b′). If b = 0, then A ∈ Ω0. If b ≠ 0 and pn(a, b) = 0 for some n ∈ N
(again, pn are polynomials defined in (5.3)), then A ∉ Ω. If b ≠ 0 and pn(a, b) ≠ 0 for all n ∈ N, then
a2 zz with a ∈ K∗ and b ∈ K. Then Aa,b and Aa′,b′
a zz and yy − yz −
a zy − b
1
A ∈ Ω+, A is Koszul and A is non-P BWB.
In order to prove above lemma we need some preparation.
Lemma 7.2. Let A be the quadratic algebra given by the generators x, y, z and the relations xx − xz −
azx − bzz, xy + b
a2 zz with a ∈ K∗ and b ∈ K. Then the following statements
hold∶
a zz and yy − yz −
a zy − b
1
for all n ∈ Z+;
(1) dim An ⩾ (n+1)(n+2)
(2) if (a, b) ≠(1, −1), then HA! =(1 + t)3;
(3) if b ≠ 0 and (a, b) ≠(1, −1), then the (right module) Koszul complex of A is given by
2
0 Ð→ A
d3
Ð→ A3 d2
Ð→ A3 d1
Ð→ A
d0
Ð→ K → 0
where d0 is the augmentation, d1(u, v, w) = xu + yv + zw, d3(u) =(yu, a(y − z)u, 0)
and d2(u, v, w) =((x − z)u + yv,(y − z)w, −(ax + bz)u + b
(4) if (a, b) =(1, −1), then dim A3 = 12 and therefore A ∉ Ω;
a zv −( 1
a2 z)w).
a y + b
(5) if b = 0, then A ∈ Ω0.
(7.1)
Proof. Since proving the above statements in the case when K is replaced by any field extension will
yield their validity for original K, we can without loss of generality, assume that K is uncountable.
30
If b = 0, then with respect to the right-to-left degree lexicographical ordering assuming z > y > x, the
leading monomials of the defining relations are yz, xz and xy. By Lemma 2.1, A ∈ Ω0. This verifies
(5).
If b ≠ 0 and (a, b) ≠ (1, −1), then the defining relations of A! are xx +
b zz,
1
a zx, yx, yy + azy and yz − azy. A direct computation shows that with respect to the left-to-right
xz −
degree lexicographical ordering assuming x > y > z, the reduced Grobner basis of the ideal of relations
of A! consists of the defining relations together with zzx, zzy and zzzz. The complete list of normal
with a graded linear basis in A!, while the above Grobner basis provides the corresponding structural
words consists of 1, x, y, z, zx, zy, zz and zzz yielding HA! =(1 + t)3. If b = 0, we already know that
A ∈ Ω0 giving HA! = (1 + t)3. This takes care of (2). Furthermore, the normal words in A! furnish us
constants. Now a routine computation yields (3). By the already verified (2), HA! =(1 + t)3 provided
(a, b) ≠ (1, −1). We also know that A is Koszul if b = 0. By Lemma 2.10 (the corresponding variety
a zx, xy + zx + zy − a
1
W is the affine space K2), (1) is satisfied.
Finally, (4) is easily verified by a direct computation.
Lemma 7.3. Let A be the quadratic algebra given by the generators x, y, z and the relations xx − xz −
azx − bzz, xy + b
a2 zz with a, b ∈ K∗ such that (a, b) ≠(1, −1). Let also n ∈ N be
such that pk(a, b) ≠ 0 for 0 ⩽ k < n and qk(a, b) ≠ 0 for 0 ⩽ k ⩽ n. Then the equalities
a zz and yy − yz −
a zy − b
1
xzkx − αkxzk+1 − ukzk+1x − skzk+2 = 0;
xzky − βkxzk+1 − vkzk+1y − tkzk+2 = 0;
yzky − γkyzk+1 − wkzk+1y − rkzk+2 = 0
(7.2)
hold in A for 0 ⩽ k ⩽ n, where αk, βk, γk, uk, vk, wk, sk, tk, rk ∈ K, αk = pk(a,b)
the rest of the numbers are defined inductively by u0 = a, w0 = 1
uk+1 = −
qk(1~a,b~a2) , while
a , β0 = v0 = 0,
and tk+1 = buk
for 0 ⩽ k < n (note that by the conditions imposed upon a, b, we never divide by zero in these fractions).
aαk
a2γk +b
a2γk −a , rk+1 = a2rk −bwk
αk −a , sk+1 = sk −buk
a2γk −a , βk+1 = − b
qk(a,b) , γk = pk(1~a,b~a2)
αk −a , wk+1 = −
a , s0 = b, r0 = b
a2 , t0 = − b
, vk+1 = −
sk +auk
sk
αk
aαk
Moreover, the equalities
αnxzn+1y + b
(αn − a)xzn+1x −(αn + b)xzn+2 +(sn + aun)zn+2x +(bun − sn)zn+3 = 0;
a2 − rn)zn+3 = 0
(γn −
a)yzn+1y −(γn + b
a2)yzn+2 +(rn +
a )zn+2y +( bwn
a xzn+2 + snxzn+2y − b
a unzn+3 = 0;
wn
1
(7.3)
hold in A. Furthermore, the left-hand sides of (7.2) for 0 ⩽ k ⩽ n together with yz are all the elements
of degree up to n + 2 of the reduced Grobner basis of the ideal of relations of A and dim Aj = (j +1)(j +2)
for 0 ⩽ j ⩽ n + 2.
2
Proof. We shall prove (7.2) for 0 ⩽ k ⩽ n inductively. The defining relations justify (7.2) for k = 0,
providing the basis of induction. Assume that 0 ⩽ k ⩽ n − 1 and (7.2) holds for k. We shall verify
(7.2) for k replaced by k + 1, which will complete the inductive proof. We do this by resolving certain
overlaps as in the Buchberger algorithm using the induction hypothesis and the defining relations of
A in the process. The monomials to which we apply reduction are indicated by underlining:
xzkxx → 0 = αkxzkx + ukzk+1xx + skzk+2x − xzkxz − axzk+1x − bxzk+2
=(αk − a)xzkx + skzk+2x − bxzk+2 + ukzk+1xz + aukzk+2x + bukzk+3
−αkxzk+2 − ukzk+1xz − skzk+3,
which yields
(αk − a)xzk+1x −(αk + b)xzk+2
+(sk + auk)zk+2x +(buk − sk)zk+3 = 0 in A.
(7.4)
From the definition of αk and the recurrent formulas for polynomials pk and qk it follows that αk+1 =
αk +b
αk −a . By the assumptions (recall that k < n) it follows that αk ≠ a. Thus dividing (7.4) by αk − a and
31
using the definition of uj and sj, we see that the first equality in (7.2) is satisfied. We proceed in the
same manner:
yzkyy → 0 = γkyzky + wkzk+1yy + rkzk+2y − yzkyz −
wk
a zk+2y
a2 yzk+2 + wkzk+1yz +
a)yzky + rkzk+2y − b
=(γk −
bwk
a2 zk+3 − γkxzk+2 − wkzk+1xz − rkzk+3,
+
1
1
a yzk+1y − b
a2 yzk+2
which yields
(γk −
1
a)yzk+1y −(γk + b
a2)yzk+2 +(rk +
wk
a )zk+2y +( bwk
a2 − rk)zk+3 = 0 in A.
(7.5)
From the definition of γk and the recurrent formulas for polynomials pk and qk it follows that γk+1 =
a2γk +b
a2γk −a . By the assumptions (recall that k < n) it follows that γk ≠ 1
a . To see this, one has to observe
that from the explicit formulae for pj and qj given in Remark 5.3 it follows that qj(a, b) = 0 ⇐⇒
qj(1~a, b~a2) = 0. Thus dividing (7.5) by γk −
1
a and using the definition of wj and rj, we see that the
third equality in (7.2) is satisfied. Finally, we deal with the third overlap:
xzkxy → 0 = αkxzk+1y + ukzk+1xy + skzk+2y + b
= αkxzk+1y −
buk
a zk+3 + skzk+2y + b
a xzk+2,
a xzk+2
which yields
αkxzk+1y + b
a xzk+2 + skzk+2y −
buk
a zk+3 = 0 in A.
(7.6)
By assumptions αk ≠ 0. Dividing (7.6) by αk and using the definition of βj, vj and tj, we see that the
second equality in (7.2) is satisfied. This completes the inductive proof of (7.2) for 0 ⩽ k ⩽ n. Now
(7.3) is the combination of (7.4), (7.5) and (7.6) for k = n, which hold since (7.2) holds for k = n.
Now the only degree ⩽ n + 2 monomials, which do not contain xzkx, yzky and xzky for 0 ⩽ k ⩽ n
as submonomials are the words of the shape zj, zj xzm, zjyzm and zj yzmxzp of degree ⩽ n + 2 with
j, m, p ∈ Z+. The number of such monomials of degree k is exactly (k+1)(k+2)
. Since by Lemma 5.2,
dim Ak ⩾ (k+1)(k+2)
degree up to n + 2 of the reduced Grobner basis of the ideal of relations of A and we must have
dim Ak = (k+1)(k+2)
, the left-hand sides of (7.2) for 0 ⩽ k ⩽ n must comprise all the elements of
for 0 ⩽ k ⩽ n + 2.
2
2
2
1
a zy − b
a zz and yy − yz −
Lemma 7.4. Let A be the quadratic algebra given by the generators x, y, z and the relations xx − xz −
azx − bzz, xy + b
a2 with a, b ∈ K∗. If pk(a, b) ≠ 0 for all k ∈ N, then A ∈ Ω+ and
A is Koszul. On the other hand, if pk(a, b) = 0 for some k ∈ N, then A ∉ Ω.
Proof. First, assume that pk(a, b) ≠ 0 for all k ∈ N. Note that p1(a, b) ≠ 0 yields b ≠ −1 and therefore
(a, b) ≠(1, −1).
Case 1: qk(a, b) ≠ 0 for all k ∈ N.
By Lemma 7.3, HA =(1 − t)−3 and the leading monomials of the reduced Grobner basis of the ideal
of relations of A are xzky, xzkx and yzky for k ∈ Z+. Since none of these monomials starts with the
smallest variable z, we have zu ≠ 0 for all non-zero u ∈ A. The shape of the Koszul complex of A
provided by Lemma 7.2 now allows us to say that this complex is exact at its left-most term (d3 is
injective). By Lemma 2.4, A is Koszul.
Case 2: qk(a, b) = 0 for some k ∈ N.
Let n ∈ Z+ be the minimal non-negative integer for which qn+1(a, b) = 0. By Lemma 7.3, (7.2) is
satisfied for 0 ⩽ k ⩽ n and (7.3) holds. Since qn+1(a, b) = 0, we have αn = a for αj defined in Lemma 7.3.
As we have already mentioned, from the explicit formula for pn and qn provided by Remark 5.3 it
follows that qn+1(1~a, b~a2) = 0, from which one sees that γn = 1
a . Next, observe that a + b ≠ 0. Indeed,
32
if a + b = 0, then qk(a, b) = pk(a, b) ≠ 0, contradicting the assumption of Case 2. Taking this into
account, we can rewrite (7.3) as follows:
xzn+2 −
xzn+1y + b
bun −sn
a+b zn+3 = 0;
sn+aun
a+b zn+2x −
sn
a xzn+2y −
a2 xzn+2 +
a2rn+awn
bwn−a2rn
yzn+2 −
zn+2y +
a+b
a+b
bun
a2 zn+3 = 0;
zn+3 = 0
(7.7)
Now we observe that the number of monomials of degree m, which do not contain any of yzky, xzkx
for 0 ⩽ k ⩽ n, xzky for 0 ⩽ k ⩽ n + 1, xzn+2 and yzn+2 as submonomials is exactly (m+1)(m+2)
for every
m ∈ Z+. Since by Lemma 5.2, dim Am ⩾ (m+1)(m+2)
, yz together with the left-hand sides of (7.2) for
0 ⩽ k ⩽ n and the left-hand sides of (7.7) must form the reduced Grobner basis of the ideal of relations
of A and we must have HA = (1 − t)−3. Since none of the leading monomials of this basis starts with
the smallest variable z, we have zu ≠ 0 for all non-zero u ∈ A. Exactly as in Case 1, we can now use
Lemma 5.2 and Lemma 2.4 to conclude that A is Koszul. Note that in the second case the Grobner
basis turns out to be finite.
2
2
Since A is non-PBWB according to Lemma 3.1, A ∈ Ω+ in both cases. Thus A ∈ Ω+ and A is Koszul
provided pk(a, b) ≠ 0 for all k ∈ N.
Assume now that pk(a, b) = 0 for some k ∈ N. To complete the proof, we have to demonstrate
that A ∉ Ω. Note that p1(1, −1) = 0 and that by Lemma 7.2, A ∉ Ω if (a, b) = (1, −1). Thus we can
assume (a, b) ≠(1, −1). Let n be the smallest non-negative integer satisfying pn+1(a, b) = 0. Using the
explicit formulas for pj and qj given in Remark 5.3, we see that qj(a, b) ≠ 0 and qj(1~a, b~a2) ≠ 0 for
0 ⩽ j ⩽ n + 1 ((1, −1) is the only exception for this rule and was excluded because of that). Consider
the variety W0 = {(s, t) ∈ K2 ∶ pn+1(s, t) = 0}. Clearly, (a, b) ∈ W0. By Lemma 7.3, (7.2) is satisfied
for 0 ⩽ k ⩽ n and (7.3) holds. Furthermore, αn = −b. For (a, b) ∈ W0 with finitely many exceptions,
αn ≠ a and γn ≠ 1
a . By Lemma 7.3, the left hand sides of (7.2) with 0 ⩽ k ⩽ n provide all elements of
degree ⩽ n + 2 of the reduced Grobner basis of the ideal of relations of A. The corresponding leading
monomials are xzkx, xzky and yzky with 0 ⩽ k ⩽ n. By Lemma 7.2, dim An+3 ⩾ (n+4)(n+5)
. It follows
that the left-hand sides of (7.3) give the degree n + 3 elements of the said Grobner basis. With finitely
many exceptions (in W0) the new leading monomials are xzn+1x, yzn+1y and xzn+2. Proceeding the
same way as in the proof of Lemma 7.3, we can get few more steps of the Grobner basis. With finitely
many exceptions (in W0) the leading monomials turn out to be: zn+2yx and yzn+2y in degree n + 4
and zn+2yzx, yzn+3y and yzn+3x in degree n + 5. This still is in agreement with the PBWS-series:
dim Aj = (j +1)(j +2)
for j ⩽ n + 5. At degree n + 6, we have to deal with all 13 overlaps, to find out that
(still with finitely many exceptions) the leading monomials of degree n + 6 of the reduced Grobner
basis are zn+2yzzx, yzn+4y and yzn+4x. This yields dim An+6 = (n+7)(n+8)
+ 1. Since this holds for all
(a, b) ∈ W0 with finitely many exceptions, Lemma 2.7 (applied to all irreducible components of W0)
for all (a, b) ∈ W0, which includes the original pair (a, b). Hence A ∉ Ω, as
yields dim An+6 > (n+7)(n+8)
required.
2
2
2
2
Proof of Lemma 7.1. The fact that A ∈ Ω0 if b = 0 follows from Lemma 7.2. That A ∉ Ω if b ≠ 0 and
from Lemma 7.4. Now if A = Aa,b and A′ = Aa′,b′
are isomorphic, then the isomorphism (a linear
substitution) must turn the corresponding quasipotentials Q and Q′ one to the other. It is easy to
pn(a, b) = 0 for some n ∈ N and that A ∈ Ω+ and is Koszul if b ≠ 0 and pn(a, b) ≠ 0 for all n ∈ N follows
check that Q and Q′ satisfy E1(Q) = E1(Q′) = span{x, z} and E2(Q) = E2(Q′) = span{y, z}. Thus
our substitution must have both span{x, z} and span{y, z} as invariant subspaces. Hence it is given
by x → αx + sz, y → βy + tz, z → γz with α, β, γ ∈ K∗ and s, t ∈ K. Without loss of generality, γ = 1.
Applying this kind of a sub to the defining relations, we see that we have the relation of the form
xy + czz with c ∈ K present only if s = t = 0. Thus our sub has the form x → αx, y → βy, z → z. Now
it is easy to see that the shape of defining relations is not preserved unless α = β = 1. This leaves the
identity map only and therefore we must have (a, b) =(a′, b′).
33
Next, we move to the following family of algebras, which as we shall see, contains (S17) and (S18).
Lemma 7.5. Let A = Ab be the quadratic algebra given by the generators x, y, z and the relations
xy + byz + zz, zx +(b − 1)yz − bzy − zz and yy + yz + bzy + zz with b ∈ K. Then Ab is non-isomorphic
to Ab′
provided b ≠ b′.
Proof. The value b = 1 is the only one for which dim A3 ≠ 10 (easily checked using Grobner basis). Thus
we can assume that b ≠ 1 and b′ ≠ 1. For b other than 1, Ab ∈ Ω′ and the corresponding quasipotential
Qb satisfies E1(Qb) = E2(Qb) = M = span{y, z}. Thus every linear sub providing an isomorphism
between Ab and Ab′
must keep M invariant. Analyzing the space Rb of quadratic relations of Ab
(pay attention to how x occurs), one sees that our shape of relations is not preserved unless the
sub sends each of y and z to their scalar multiples. Without loss of generality (just normalize the
relations), it sends y to y and z to az with a ∈ K∗. Moreover, Rb ∩ M 2 is the one-dimensional space
spanned by yy + yz + bzy + zz. Thus our sub must transform yy + yz + bzy + zz into a scalar multiple
of yy + yz + b′zy + zz. The latter only happens when a = 1 and b = b′. The result follows.
hold true∶
Lemma 7.6. Let A = Ab be the quadratic algebra given by the generators x, y, z and the relations
xy + byz + zz, zx +(b − 1)yz − bzy − zz and yy + yz + bzy + zz with b ∈ K. Then the following statements
(1) dim An ⩾ (n+1)(n+2)
(2) if b ∉{0, 1}, then HA! =(1 + t)3 and the (right module) Koszul complex of A is given by
for all n ∈ Z+;
2
d3
Ð→ A3 d2
Ð→ A3 d1
Ð→ A
d0
Ð→ K → 0
0 Ð→ A
where d0 is the augmentation, d1(u, v, w) = xu + yv + zw, d3(u) =(0, yu,(y + z)u)
and d2(u, v, w) =(yu, bzu +(b − 1)zv +(y + z)w, zu +(x − by − z)v +(by + z)w).
(7.8)
(3) if b = 1, then A ∉ Ω.
Proof. Since proving the above statements in the case when K is replaced by any field extension will
yield their validity for original K, we can without loss of generality, assume that K is uncountable.
In this proof we always use the left-to-right degree lexicographical ordering assuming x > y > z. The
validity of (3) is a matter of routine verification: a direct Grobner basis calculation yields dim A3 = 11.
Next, it is a matter of an easy calculation to see that A! is given by generators x, y, z and the
A direct computation shows that the reduced Grobner basis of the ideal of relations of A! consists
of the defining relations together with zzx, zzy and zzzz. The complete list of normal words is: 1,
relations xx, bxy + zy − bzz, xz, yx, byy − bzx − zy and byz − b2zx +(b − 1)zy − b2zz provided b ∉{0, 1}.
x, y, z, zx, zy, zz and zzz yielding HA! = (1 + t)3. Furthermore, the normal words in A! furnish us
with a graded linear basis in A!, while the above Grobner basis provides the corresponding structural
constants. Now a routine computation yields (7.8), which completes the proof of (2).
It remains to prove (1). First, observe that it is enough to prove (1) in the case b ≠ 0: Lemma 2.7
fills the gap. If b ≠ 0, we perform the sub x → bx, y → y, z → z turning the defining relations of Ab
that Ab = A1~a is isomorphic to the algebra B = Ba, given by the generators x, y, z and the relations
into bxy + byz + zz, bzx +(b − 1)yz − bzy − zz and yy + yz + bzy + zz. Now replacing b by 1~a, we see
xy + yz + azz, zx +(1 − a)yz − zy − azz and ayy + ayz + zy + azz with a ∈ K. We already know that
for a ∉{0, 1}, HB! =(1 + t)3. Next, for a = 0, the leading monomials of the defining relations of B are
Lemma 2.1, B0 ∈ Ω0 and therefore B0 is Koszul and HB! = (1 + t)3 for a = 0. By Lemma 2.10 (the
xy, zx and zy with respect to the left-to-right degree lexicographical ordering assuming x > z > y. By
corresponding variety W is K), (1) is satisfied.
Lemma 7.7. Let A = Ab be the quadratic algebra given by the generators x, y, z and the relations
If b ≠ −3, then
xy + byz + zz, zx + (b − 1)yz − bzy − zz and yy + yz + bzy + zz with b ∈ K, b ≠ 1.
34
Ab is isomorphic to the quadratic algebra B = Bβ given by the generators x, y, z and the relations
βxy − xz + β2yy −(β3 + 1)zy + zz, yx − βzx + βyy −(β3 + 1)zy + βzz and yz − βzy with β ∈ K∗, β2 ≠ 1,
β . If b = −3, then A is isomorphic to the quadratic algebra B
where b and β are related by b = −1 − β −
1
given by the generators x, y, z and the relations xy + yy −
2 zz.
In both cases the linear substitution facilitating the isomorphism can be chosen preserving the linear
span of y and z.
1
2 zz, yx + yy + zx + zy −
1
2 zz and yz − zy +
1
1
Proof. First, assume that b ≠ −3. Since K is algebraically closed and b ≠ 1, we can solve a quadratic
equation to find β ∈ K∗ such that b = −1 − β −
the algebra A = Ab. The sub x → x + by + z, y → y, z → z turns the defining relations of A into
β . Since b ∉ {1, −3}, we have β2 ≠ 1. We start with
xy − (b2 − 1)zy − (b − 1)zz, xy +(b − 1)yz and yy + yz + bzy + zz. We follow up with scaling of x:
x → (b − 1)x, y → y, z → z: the relations become xy −(b + 1)zy − zz, zx + yz and yy + yz + bzy + zz.
Finally, we perform the y, z sub y → βy − z, z → y − βz, x → x (non-degenerate since β2 ≠ 1). This
turns the relations into those of Bβ. Thus the composition of these three linear subs provides an
isomorphism between Ab and Bβ. Since each preserves the span of y and z, so does the resulting
isomorphism.
Now assume b = −3. In this case the defining relations of A transformed by the sub x → x, y → y,
z → y + z span the same space as the defining relations of B. Again, the span of y and z is preserved
by the isomorphism.
Lemma 7.8. Let B = Bβ be the quadratic algebra given by the generators x, y, z and the relations
βxy − xz + β2yy −(β3 + 1)zy + zz, yx − βzx + βyy −(β3 + 1)zy + βzz and yz − βzy with β ∈ K∗, β2 ≠ 1.
If βk ≠ 1 for all k ∈ N, then B ∈ Ω+ and B is Koszul. On the other hand, if k is the smallest positive
integer for which βk+2 = 1, then dim Bk+5 = (k+6)(k+7)
+ 1 and therefore B ∉ Ω.
2
Proof. Throughout the proof we use the left-to-right degree lexicographical ordering on x, y, z mono-
mials assuming x > y > z. First, we use induction by k to prove that, the equality
xzky − β −k−1xzk+1 + βk+1zkyy −
holds in B.
β3
+1
β zk+1y + β −k−1zk+2 = 0 (k ∈ Z+)
(7.9)
Indeed, the first defining relation yields the validity of (7.9) for k = 0. Now if we assume that k ∈ N
and (7.9) holds with k replaced by k − 1, then we resolve the overlap xzk−1yz =(xzk−1y)z = xzk−1(yz)
using (7.9) for k − 1 and the defining relations. After obvious cancelations (7.9) for k follows. Next,
we show that
(1−βk+1)xzkx+(βk−1−1)(β2 −β −k)xzk+1+(βk+3−1)(1−βk)zk+1x+βk+2(βk+1−1)zkyy
+(1−β2k+2)(β3−β+1)zk+1y+(1−βk+1)(β2 +β −1−β −k −β−βk+2)zk+2 = 0 (k ∈ N)
(7.10)
holds in B.
These equalities are obtained by resolving the overlap xzk−1yx = (xzk−1y)x = xzk−1(yx) using the
already verified (7.9) and the defining relations. After obvious cancelations (7.10) follows.
Now assume that
n ∈ Z+ and βj ≠ 1 for 1 ⩽ j ⩽ n + 1.
(7.11)
Note that the only monomials that do not have any of the monomials of the shape xzj y for j ∈ Z+,
xzj x for j ∈ N, yx or yz as submonomials are exactly the monomials of the form zkym or zkxpzm with
k, m, ∈ Z+, p ∈ N. Observe that the number of such monomials of degree s is precisely (s+1)(s+2)
By Lemmas 7.6 and 7.7, dim Bk ⩾ (k+1)(k+2)
left-hand sides of (7.9) and (7.10) for k ⩽ n together with yx − βzx + βyy −(β3 + 1)zy + βzz and yz − βzy
form the degree up to n +2 part of a Grobner basis of the ideal of relations of B and dim Bk = (k+1)(k+2)
for k ⩽ n + 2.
for all k. It follows that if (7.11) is satisfied, then the
.
2
2
2
35
Now assume that βj ≠ 1 for all j ∈ N. Then (7.11) is satisfied for all n ∈ Z+. Hence the left-hand
sides of (7.9) and (7.10) for all k together with yx − βzx + βyy −(β3 + 1)zy + βzz and yz − βzy form a
Grobner basis of the ideal of relations of B and dim Bk = (k+1)(k+2)
for k ⩽ n + 2. Then B ∈ Ω. Since
none of the leading monomials of the above Grobner basis starts with the smallest variable z, zu ≠ 0
for every non-zero u ∈ B. Hence zu = yu = 0 fails for every non-zero u ∈ B. By Lemma 7.7, B is
isomorphic to the quadratic algebra A given by the generators x, y, z and the relations xy + byz + zz,
1
β . Since the isomorphism
is given by a linear sub preserving the span of y and z, we have that zu = yu = 0 fails for every
non-zero u ∈ B. Hence the map d3 from (7.8) is injective. By Lemma 2.4, A is Koszul and so is B.
By Lemma 3.1, A is not PBWB and therefore A and B are in Ω+.
zx +(b − 1)yz − bzy − zz and yy + yz + bzy + zz with b ∈ K satisfying b = −1 − β −
2
It remains to deal with the case when β is a root of 1. Let m be the smallest positive integer satisfying
βm+1 = 1. Since β2 ≠ 1, m ⩾ 2. Then (7.11) is satisfied for n = m − 1. Then the left-hand sides of (7.9)
and (7.10) for k ⩽ m − 1 together with yx − βzx + βyy −(β3 + 1)zy + βzz and yz − βzy form the degree
up to m + 1 part of a Grobner basis of the ideal of relations of B and dim Bk = (k+1)(k+2)
for k ⩽ m + 1.
Plugging βm+1 = 1 into (7.10) and (7.9) with k = m and using the inequality dim Bm+2 ⩾ (m+3)(m+4)
,
we easily see that the left-hand sides of (7.9) and (7.10) for k = m form the degree up to m + 2 part
of a Grobner basis of the ideal of relations of B and dim Bm+2 = (m+3)(m+4)
. The said left-hand sides
β3
(up to a scalar multiple) are xzmy − zm+1x + zmyy −
β zm+1y + zm+2 and xzm+1 − zm+1x. Resolving
the overlap xzmyx = xzm(yx) =(xzmy)x, we get the following equality in B:
+1
2
2
2
zm+1xx − β2yy +(β − β −1)xz +(β4
+ β3
− β2
+ 1)zy +(β −1
− 1 − β)zz = 0.
2
2
Now it is easy to see that the number of degree m + 3 monomials, which do not contain any of
yx, yz, xzj y for 0 ⩽ j ⩽ m, xzj x for 1 ⩽ j ⩽ m − 1, xzm+1 or zm+1xx as a submonomial is exactly
(m+4)(m+5)
+ 1.
, while the number of degree m + 4 monomials with the same property is (m+5)(m+6)
2
, the degree m + 3 part of the Grobner basis consists of just one element:
Since dim Bm+3 ⩾ (m+4)(m+5)
the left-hand side in the above display and dim Bm+3 = (m+4)(m+5)
. Finally, dealing with all (this time)
overlaps of the leading monomials of the Grobner basis (so far) of degree m + 4 (there are m + 3 of
them: yzm+1xx, xzm+1xx, zm+1xy, xzxzm+1, xzj xzpx and xzj xzpy with j, p ∈ N, j + p = m + 1, p ⩾ 2),
we find that all of them resolve without producing a degree m + 4 member of the Grobner basis. This
part is easy but tedious: we leave the details to an interested reader.
2
As a result, we have dim Bm+4 = (m+5)(m+6)
2
+ 1 and therefore B ∉ Ω. This completes the proof.
Lemma 7.9. Let B be the quadratic algebra given by the generators x, y, z and the relations xy + yy −
1
2 zz. If the characteristic of K is 0, then B ∈ Ω+ and B is
2 zz, yx + yy + zx + zy −
Koszul. If K has prime characteristic, then B ∉ Ω.
1
2 zz and yz − zy +
1
Proof. Throughout the proof we use the left-to-right degree lexicographical ordering on x, y, z mono-
mials assuming x > y > z. First, we use induction by k to prove that, the equality
xzky − k
2 xzk+1 + zkyy − kzk+1y +
(k−1)(k+2)
4
zk+2 = 0 (k ∈ Z+)
(7.12)
holds in B.
Indeed, the first defining relation yields the validity of (7.9) for k = 0. Now if we assume that k ∈ N
and (7.12) holds with k replaced by k − 1, then we resolve the overlap xzk−1yz =(xzk−1y)z = xzk−1(yz)
using (7.12) for k − 1 and the defining relations. After obvious cancelations (7.12) for k follows.
Next, we show that
(k + 1)xzkx + k(k+1)
−(k + 1)zkyy +
k(k+1)
2
2
− 1xzk+1 − (k+1)(k+2)
(k−2)(k+1)(k+3)
2
zk+1y −
− 1zk+1x
zk+2 = 0 (k ∈ N)
4
36
(7.13)
holds in B.
These equalities are obtained by resolving the overlap xzk−1yx = (xzk−1y)x = xzk−1(yx) using the
already verified (7.12) and the defining relations. After obvious cancelations (7.13) follows.
Now assume that
(7.14)
Note that the only monomials that do not have any of the monomials of the shape xzj y for j ∈ Z+,
xzj x for j ∈ N, yx or yz as submonomials are exactly the monomials of the form zkym or zkxpzm with
k, m, ∈ Z+, p ∈ N. Note also that the number of such monomials of degree s is precisely (s+1)(s+2)
n ∈ Z+ and char K ∉{2, . . . , n + 1}.
.
By Lemmas 7.6 and 7.7, dim Bk ⩾ (k+1)(k+2)
for all k. It follows that if (7.14) is satisfied, then the
1
left-hand sides of (7.12) and (7.13) for k ⩽ n together with yx + yy + zx + zy −
2 zz form
the degree up to n + 2 part of a Grobner basis of the ideal of relations of B and dim Bk = (k+1)(k+2)
for k ⩽ n + 2.
1
2 zz and yz − zy +
2
2
2
Now assume that K has characteristic 0. Then (7.14) is satisfied for all n ∈ Z+. Hence the left-hand
1
2 zz form a
sides of (7.12) and (7.13) for all k together with yx + yy + zx + zy −
Grobner basis of the ideal of relations of B and dim Bk = (k+1)(k+2)
for k ⩽ n + 2. Then B ∈ Ω. Since
none of the leading monomials of the above Grobner basis starts with the smallest variable z, zu ≠ 0
for every non-zero u ∈ B. Hence zu = yu = 0 fails for every non-zero u ∈ B. By Lemma 7.7, B is
isomorphic to the quadratic algebra A given by the generators x, y, z and the relations xy − 3yz + zz,
zx − 4yz + 3zy − zz and yy + yz − 3zy + zz (b = −3). Since the isomorphism is given by a linear sub
preserving the span of y and z, we have that zu = yu = 0 fails for every non-zero u ∈ B. Hence the map
d3 from (7.8) is injective. By Lemma 2.4, A is Koszul and so is B. By Lemma 3.1, A is not PBWB
and therefore A and B are in Ω+.
1
2 zz and yz − zy +
2
It remains to deal with the case when char K = p is a prime number. Since we have excluded
characteristics 2 and 3, p ⩾ 5. Then (7.14) is satisfied for n = p − 1 ⩾ 4. Then the left-hand sides of
(7.12) and (7.13) for k ⩽ p − 2 together with form the degree up to p part of a Grobner basis of the
ideal of relations of B and dim Bk = (k+1)(k+2)
for k ⩽ p. Using the equality char K = p, (7.13) and
(7.12) with k = p − 1 and using the inequality dim Bp+1 ⩾ (p+2)(p+3)
, we easily see that the left-hand
sides of (7.12) and (7.13) for k = p − 1 form the degree p + 1 part of a Grobner basis of the ideal
of relations of B and dim Bp+1 = (p+2)(p+3)
. The said left-hand sides have the same linear span as
xzp−1y +
(xzp−1y)x, we get the following equality in B:
2 zp+1 and xzp − zpx. Resolving the overlap xzp−1yx = xzp−1(yx) =
2 zpx + zp−1yy + zpy −
2
1
1
2
2
zp(xx − xz − yy − zy +
3
2 zz) = 0.
Now it is easy to see that the number of degree p + 2 monomials, which do not contain any of yx, yz,
xzj y for 0 ⩽ j ⩽ p−1, xzj x for 1 ⩽ j ⩽ p−2, xzp or zpxx as a submonomial is exactly (p+3)(p+4)
, while the
number of degree p + 3 monomials with the same property is (p+4)(p+5)
+ 1. Since dim Bp+2 ⩾ (p+3)(p+4)
,
the degree p + 2 part of the Grobner basis consists of just one element: the left-hand side in the
above display and dim Bp+2 = (p+3)(p+4)
. Finally, as in the previous lemma all overlaps of the leading
monomials of the Grobner basis (so far) of degree p + 3 (they are listed in the proof of the previous
lemma: one just has to assume m + 1 = p) resolve. As a result, dim Bp+3 = (p+4)(p+5)
+ 1 and therefore
B ∉ Ω. This completes the proof.
2
2
2
2
2
Lemma 7.10. The algebras in (S1–S20) belong to Ω and are Koszul. The algebras in (S1–S16) are
P BWB, while the algebras in (S17–S20) are non-P BWB. Algebras in (S1–S20) with different labels
are non-isomorphic and the isomorphism conditions of Theorem 1.11 within algebras with a given label
from (S1–S20) are satisfied.
Proof. Algebras from (S1–S16) belong to Ω0 by Lemma 4.1. Algebras in (S17) and (S18) are Koszul
and belong to Ω+ by Lemmas 7.9, 7.8 and 7.7, while algebras in (S19) and (S20) are Koszul and
37
belong to Ω+ by Lemma 7.1 and Remark 5.3. Algebras in (S17–S20) are non-PBWB according to
Lemma 3.1. It remains to deal with isomorphisms. As algebras in (S17–S20) are the only non-PBWB
ones of the batch, they can not be isomorphic to any of the algebras from (S1–S16). Thus the families
of algebras in (S1–S16) and in (S17–S20) can be treated separately. Note that quasipotentials Q
for algebras in (S17–S18) satisfy E1(Q) = E2(Q), while E1(Q) ≠ E2(Q) for algebras in (S19–S20).
Thus algebras in (S17–S18) can not be isomorphic to any of the algebras in (S19–S20). Within the
family (S17–S18), the required isomorphism statement now follows from Lemmas 7.5 and 7.7, while
the same for (S19–S20) is ensured by the isomorphism part of Lemma 7.1. These considerations take
(S17–S20) out of the picture, leaving us with (S1–S16). The algebra (S1) is the only one of the lot
with quasipotential Q satisfying E1(Q) = E2(Q). This singles it out, leaving us with (S2–S16). The
quasipotential Q for each algebra in (S2–S16) satisfies E1(Q) = span{x, y} and E2(Q) = span{y, z}.
a member of span{x, y} and z to a member of span{y, z}.
Thus every isomorphism (linear substitution) between two algebras in (S2–S16) must leave each of
these two spaces invariant. That is, such an isomorphism must send y to its own scalar multiple, x to
Keeping this in mind, observe that algebras in (S14–S16) are the only ones in (S2–S16) with no
rank one elements in the space of quadratic relations. Note as well that the defining relations of each
algebra in (S14–S16) contain xz + yy, one rank 2 relation depending on y and z only and one rank 2
relation depending on x and y only. When a substitution of the above described form is applied to
one of the algebras in (S14–S16), the only way for the resulting algebra to still possess the quadratic
relation of the form sxz + tyy with s, t ∈ K∗, is for the substitution to be a scaling. Checking scalings,
one easily sees that no scaling provide an isomorphism of algebras from (S14–S16) with different labels
and the only scalings providing isomorphisms between two algebras from (S14–S16) with the same
y
label are x → sx, y → sy and z → sz with s ∈ K∗ for algebras from (S14–S15) and x → sx, y →
st , z → tz
with s, t ∈ K∗ for algebras from (S16). It is clear that in each case the parameter a is preserved, which
proves that algebras in (S14–S16) are pairwise non-isomorphic. These considerations take (S14–S16)
out of the picture, leaving us to deal with (S2–S13).
Note that (S2–S5) are singled out from (S2–S13) by having a rank one element f in the space of
not be isomorphic to algebras from (S6–S13). Note that the above element f is actually xz for all
algebras in (S2–S5) and that such an element is unique up to a scalar multiple. Thus a substitution
quadratic relations, which belongs to neither E1(Q)2 nor E2(Q)2. Thus algebras from (S2–S5) can
providing an isomorphism between two algebras in (S2–S5) must not only preserve span{x, y} and
span{y, z} but also transform xz to its own scalar multiple. Again, only scalings do that. It is obvious
that a scaling can not provide an isomorphism between algebras from (S2–S5) with different labels.
As for scalings transforming an algebra from (S2–S5) to another one with the same label, they are
x → sx, y → sy and z → sz with s ∈ K∗ for algebras from (S2), x → sx, y → sy and z → tz with s, t ∈ K∗
for algebras from (S3), x → sx, y → ty and z → tz with s, t ∈ K∗ for algebras from (S4) and arbitrary
scalings for algebras from (S5). However, it is clear that in each case the parameter a is preserved,
which proves that algebras in (S2–S5) are pairwise non-isomorphic. These considerations take (S2–S5)
out of the picture, leaving us to deal with (S6–S13).
Now (S6–S9) are singled out from (S6–S13) by having a rank one element f in the space of quadratic
from (S10–S13). Note that the above element f is actually yz for all algebras in (S6–S9) and that such
an element is unique up to a scalar multiple. Thus a substitution providing an isomorphism between
relations, which belongs to E2(Q)2. Thus algebras from (S6–S9) can not be isomorphic to algebras
two algebras in (S6–S9) must not only preserve span{x, y} and span{y, z} but also transform yz to
its own scalar multiple. The only substitutions, which do this are x → αx + ty, y → βy and z → γz
with α, β, γ ∈ K∗ and t ∈ K. Such substitutions can not give birth or eliminate the presence of zz in
the defining relations. Thus algebras from (S6) and (S8) can not be isomorphic to any of the algebras
from (S7) and (S9). Since xy + byx and xy − yx + ayy with a, b ∈ K∗ can not be obtained from one
another by an x, y linear substitution (one can use Lemma 2.2), algebras from (S6–S7) can not be
isomorphic to any of the algebras from (S8–S9). It follows that algebras from (S6–S9) with different
38
labels are non-isomorphic. Next, xy + ayx + pyy and xy + byx + qyy with different a, b ∈ K∗ can not
be transformed to one another by a substitution of the form x → αx + ty, y → βy with α, β ∈ K∗ and
t ∈ K. It follows that algebras in (S6) and (S7) are pairwise non-isomorphic. Finally any substitution
x → αx + ty, y → βy and z → γz with α, β, γ ∈ K∗ and t ∈ K is an automorphism of each of the algebras
in (S8–S9). As a result, algebras in (S6–S9) are pairwise non-isomorphic. Since algebras in (S10–S13)
are isomorphic to algebras from (S6–S9) with the opposite multiplication, algebras in (S10–S13) are
pairwise non-isomorphic as well. The proof is now complete.
of V . Then they may either coincide or intersect by a one-dimensional space.
4-dimensional. In this case though Q is not a quasipotential and therefore is of no interest to us. The
key part of our job in the remaining part of this section is identifying Q for which RQ is 3-dimensional.
If V is a 3-dimensional vector space and Q ∈ V 3 satisfies n1(Q) = n2(Q) = 2, then, generically, RQ is
We split our search for such Q in two. Note that both E1(Q) and E2(Q) are 2-dimensional subspaces
Lemma 7.11. Let A = A(V, R) ∈ Ω be such that the corresponding quasipotential Q = QA satisfies
n1(Q) = n2(Q) = 2 and E1(Q) = E2(Q). Then A is isomorphic to an algebra from (S1, S17, S18) of
Proof. Start by choosing a basis x, y, z in V such that x and y span M = E1(Q) = E2(Q). Then
Theorem 1.11.
Q = Q0 + Q1, where Q1 ∈ M 3 and Q0 ∈ M LM , where L is the one-dimensional space spanned by z. If
Q0 = 0, then Lemma 2.11 ensures that A ∉ Ω. This contradiction yields Q0 ≠ 0. By Lemma 2.3, there
is an x, y sub bringing Q0 to one of the following forms: Q0 = xzy − yzx − yzy, Q0 = xzy − αyzx with
α ∈ K∗, Q0 = yzy or Q0 = xzy.
Case 1: Q0 = xzy − yzx − yzy.
By a sub x → x, y → y, z → sx + ty with appropriately chosen s, t ∈ K we kill the xxy and xyy
coefficients of Q. Then Q acquires form
Q = xzy − yzx − yzy + a1xxx + a2xyx + a3yxx + a4yxy + a5yyx + a6yyy
with aj ∈ K. Then RQ = F1(Q) + F2(Q) is spanned by f1 = zy + a1xx + a2yx, f2 = −zx − zy + a3xx +
a4xy + a5yx + a6yy, f3 = −yz + a1xx + a2xy + a3yx + a5yy and f4 = xz − yz + a4yx + a6yy. The 4 × 4
matrix S of the zx, xz, zy and yz coefficients of f2, f4, f1 and f3 (in this order) is
S =
⎛⎜⎜⎜⎝
−1 0 −1
0
0
0
0
0 −1
1
0
0 −1
1
0
0
⎞⎟⎟⎟⎠
.
Since S is obviously invertible, dim RQ = 4 > 3. Hence Q fails to be a quasipotential. Thus Case 1
produces no algebras from Ω.
Case 2: Q0 = xzy − αyzx with α ∈ K∗.
By a sub x → x, y → y, z → sx + ty with appropriately chosen s, t ∈ K we kill the xxy and xyy
coefficients of Q. Then Q acquires form
Q = xzy − αyzx + a1xxx + a2xyx + a3yxx + a4yxy + a5yyx + a6yyy
with aj ∈ K. Then RQ = F1(Q) + F2(Q) is spanned by f1 = zy + a1xx + a2yx, f2 = −αzx + a3xx + a4xy +
a5yx + a6yy, f3 = −αyz + a1xx + a2xy + a3yx + a5yy and f4 = xz + a4yx + a6yy. The 4 × 4 matrix S of
the zx, xz, zy and yz coefficients of f2, f4, f1 and f3 (in this order) is diagonal with the numbers −α,
1, 1, −α on the main diagonal. Since α ≠ 0, this matrix is invertible. Hence dim RQ = 4 > 3 and Q fails
to be a quasipotential. Thus Case 2 produces no algebras from Ω.
Case 3: Q0 = yzy.
39
By a sub x → x, y → y, z → sx + ty with appropriately chosen s, t ∈ K we kill the yxy and yyy
coefficients of Q. Then Q acquires form
Q = yzy + a1xxx + a2xxy + a3xyx + a4xyy + a5yxx + a6yyx
that f2 = αf1 ≠ 0.
with aj ∈ K. Then RQ = F1(Q) + F2(Q) is spanned by f1 = a1xx + a2xy + a3yx + a4yy, f2 = a1xx + a3xy +
a5yx + a6yy, f3 = zy + a5xx + a6yx and f4 = yz + a2xx + a4xy. Since Q is a quasipotential, dim RQ ⩽ 3
and therefore fj are linearly dependent. The monomials yz and zy feature with non-zero coefficients
only in f4 and f3 respectively. Hence linear dependence of fj yields linear dependence of f1 and f2.
If a1 ≠ 0, we must have α = 1. In this case f2 = αf1 reads a2 = a3 = a5 and a4 = a6. By scaling x,
Next, neither f1 nor f2 is zero (otherwise either n1(Q) < 2 or n2(Q) < 2). Hence there is α ∈ K∗ such
we can turn a1 into 1. Then Q = yzy + xxx + a(xxy + xyx + yxx) + b(xyy + yyx) with a, b ∈ K. By
shape of Q) kill a. Then Q = yzy + xxx + b(xyy + yyx) with b ∈ K. If b = 0, then xx ∈ R, which in
a sub x → x − ay, y → y, z → z + sx + ty with appropriately chosen s, t ∈ K, we can (preserving the
view of Lemma 1.12 leads to a contradiction. Thus b ≠ 0. A further scaling turns b into 1. Then
Q = yzy + xxx + xyy + yyx. In this case R is spanned by xx + yy, zy + yx and yz + xy. A direct Grobner
basis computation yields dim A5 = 22, which is incompatible with A ∈ Ω.
It remains to consider the case a1 = 0. In this case, the equality f2 = αf1 implies that Q has the
shape Q = yzy + a(xxy + αxyx + α2yxx) + b(xyy + αyyx) with a, b ∈ K (recall that α ∈ K∗). If a = b = 0,
n1(Q) = 1. If a = 0 and b ≠ 0, yy ∈ R. Since both contradict the assumptions, a ≠ 0. A normalization
If b = 0, then Q = yzy + xxy + αxyx + α2yxx with α ∈ K∗. Now R is spanned by
turns a into 1.
xy + αyx, zy + α2xx and yz + xx. A Grobner basis computation gives dim A5 = 22 for generic α (with
finitely many exceptions). By Lemma 2.7, dim A5 ⩾ 22 for all α. Since this is incompatible with
A ∈ Ω, case b = 0 does not occur. For every s ∈ K the sub x → x + sy, y → y, z → z + px + qy with
appropriately chosen p, q ∈ K preserves the shape of Q changing the parameter b according to the rule
b ↦ b +(1 + α)s. Thus we can reduce the general situation to the already considered case b = 0 unless
α = −1. Thus it remains to consider the case b ≠ 0 and α = −1. Further scaling turns b into 1, leaving
us with Q = yzy + xxy − xyx + yxx + xyy − yyx. Then R is spanned by xy − yx + yy, zy + xx − yx and
yz + xx + xy. A Grobner basis computation yields dim A5 = 22, which is incompatible with A ∈ Ω.
Thus Case 3 yields no algebras from Ω.
Case 4: Q = xzy.
By a sub x → x, y → y, z → sx + ty with appropriately chosen s, t ∈ K we kill the xxy and xyy
coefficients of Q. Then Q acquires form
Q = xzy + a1xxx + a2xyx + a3yxx + a4yxy + a5yyx + a6yyy
with aj ∈ K. Then RQ = F1(Q) + F2(Q) is spanned by f1 = a3xx + a4xy + a5yx + a6yy, f2 = a1xx + a2xy +
a3yx + a5yy, f3 = zy + a1xx + a2yx and f4 = xz + a4yx + a6yy. Since Q is a quasipotential, dim RQ ⩽ 3
and therefore fj are linearly dependent. The monomials xz and zy feature with non-zero coefficients
only in f4 and f3 respectively. Hence linear dependence of fj yields linear dependence of f1 and f2.
such that f2 = αf1 ≠ 0. Solving this system of linear equations, we see that Q must have the form
1
α y, z → z
Next, neither f1 nor f2 is zero (otherwise either n1(Q) < 2 or n2(Q) < 2). Hence there is α ∈ K∗
Q = xzy + a(xxx + αyxx + α2yyx + α3yyy) + b(xyx + αyxy) with a, b ∈ K. The sub x → x, y →
turns α into 1, while preserving the overall shape of Q: Q = xzy +a(xxx +yxx +yyx +yyy)+b(xyx +yxy)
with a, b ∈ K. If a = 0, then b ≠ 0 (otherwise n1(Q) = 1). By scaling z, we turn b into 1 arriving at
Q = xzy + xyx + yxy. Then R is spanned by zy + yx, xy and xz + yx. That is, we have arrived
It remains to consider the case a ≠ 0. By scaling z, we can turn a into 1.
at the algebra (S1).
Thus Q = xzy + xxx + yxx + yyx + yyy + bxyx + byxy with b ∈ K. After swapping x and z, we get
Q = zxy + zzz + yzz + yyz + yyy + bzyz + byzy. Then R is spanned by xy + zz + byz zz + yz + yy + bzy
zx + yy + byz. These relations span the same space as that of the algebra Ab of Lemma 7.5. By
Lemmas 7.7, 7.8 and 7.9, A is isomorphic to an algebra from (S17–S18).
40
x, y, z and three quadratic relations from (S2–S16) or (S19–S20) of Theorem 1.11.
Lemma 7.12. Let A = A(V, R) ∈ Ω be such that the corresponding quasipotential Q = QA satisfies
n1(Q) = n2(Q) = 2 and E1(Q) ≠ E2(Q). Then A is isomorphic to to a K-algebra given by generators
Proof. Let y ∈ V be such that y spans the one-dimensional space E1(Q) ∩ E2(Q). Pick x and z in V
such that x, y is a basis in E1(Q), while y, z is a basis in E2(Q). Clearly, x, y, z form a basis of V .
Then Q has the form
Q = a1xxy + a2xxz + a3xyy + a4xyz + a5xzy + a6xzz
+a7yxy + a8yxz + a9yyy + a10yyz + a11yzy + a12yzz,
where aj ∈ K. Clearly RQ is spanned by
f1 = a1xy + a2xz + a3yy + a4yz + a5zy + a6zz,
f2 = a7xy + a8xz + a9yy + a10yz + a11zy + a12zz,
f3 = a1xx + a3xy + a5xz + a7yx + a9yy + a11yz,
f4 = a2xx + a4xy + a6xz + a8yx + a10yy + a12yz.
Since Q is a quasipotential dim RQ ⩽ 3. Hence fj must be linearly dependent. Note also that f1 and
f2 must be linearly independent (otherwise n2(Q) < 2) and f3 and f4 must be linearly independent
(otherwise n1(Q) < 2).
The 4 × 4 matrices S1 of xz, yz, zy and zz coefficients and S2 of xx, yx, xy and xz coefficients of
f1, f2, f3 and f4 (all in given order) are
S1 =
⎛⎜⎜⎜⎝
a5
a4
a2
a6
a8 a10 a11 a12
0
a5 a11
0
a6 a12
0
0
⎞⎟⎟⎟⎠
and S2 =
⎛⎜⎜⎜⎝
0
0
0
a1 a2
0
a7 a8
a1 a7 a8 a5
a2 a8 a4 a6
⎞⎟⎟⎟⎠
.
If either of the two matrices
S3 = a1 a7
a2 a8 or S4 = a5
a11 a12
a6
is non-degenerate, then at least one of S1 or S2 is invertible leading to linear independence of fj. Thus
both S3 and S4 must be non-invertible.
Case 1: a2a6 ≠ 0.
A scaling turns a2 and a6 into 1. Hence, we can assume that a2 = a6 = 1. A substitution x → x + sy,
y → y, z → z + ty with appropriate s, t ∈ K kills a1 and a12. Thus we can assume a1 = a12 = 0. Now
since S3 and S4 are non-invertible, we have a7 = a11 = 0. Thus Q = xxz + a3xyy + a4xyz + a5xzy +
xzz + a8yxz + a9yyy + a10yyz. In this case f1 = xz + a3yy + a4yz + a5zy + zz, f2 = a8xz + a9yy + a10yz,
f3 = a3xy + a5xz + a9yy and f4 = xx + a4xy + xz + a8yx + a10yy. The monomial zz features with non-zero
coefficient only in f1, while xx features with non-zero coefficient only in f4. Hence linear dependence of
fj yields linear dependence of f2 and f3. If a3 ≠ 0, linear dependence of f2 and f3 occurs only if f2 = 0.
In this case, f1 and f2 are linearly dependent, which is a contradiction. Thus a3 = 0. If a10 ≠ 0, linear
dependence of f2 and f3 occurs only if f3 = 0. In this case, f3 and f4 are linearly dependent, which is
a contradiction. Hence a10 = 0. Plugging a3 = a10 = 0 back into fj, we get f1 = xz + a4yz + a5zy + zz,
f2 = a8xz +a9yy, f3 = a5xz +a9yy and f4 = xx +a4xy +xz +a8yx. Since f2 and f3 are linearly dependent
and neither is 0, there is α ∈ K∗ such that f3 = αf2 ≠ 0.
If a9 ≠ 0, we have α = 1 and a5 = a8. A scaling turns a9 into 1. Thus Q = xxz + xzz + yyy + sxzy +
syxz + txyz with s, t ∈ K. Clearly, R is spanned by xy + yy + syz + tzy, zz + sxy and xx + xy + szx + txz.
Then s ≠ 0 (otherwise zz ∈ R and Lemma 1.12 provides a contradiction). If t = 0, then computing the
reduced Grobner basis, we see that for generic algebras in our one-parametric family dim A6 = 29. By
41
Lemma 2.7, dim A6 ⩾ 29 for all s whenever t = 0, which is incompatible with the membership in Ω.
Hence t ≠ 0. Now swapping of y and z together with an appropriate scaling turns these relations into
those from Lemma 7.1. Now by Lemma 7.1 and Remark 5.3, our algebra is isomorphic to an algebra
from (S19–S20). It remains to deal with the case a9 = 0. Since f2 and f3 are non-zero a5a8 ≠ 0. A
scaling turns a5 into 1. Now Q = xxz + bxyz + xzy + xzz + ayxz with b ∈ K and a ∈ K∗. Then R is
spanned by byz + zy + zz, xz and xx + bxy + ayx. If b = 0, applying the Grobner basis technique, we
see that dim A3 > 10, contradicting the assumptions. Thus b ≠ 0 and we have an algebra from (S2).
This concludes Case 1.
Case 2: a2 ≠ 0 and a6 = 0.
Scaling, we can make a2 = 1. A sub x → x, y → y, z → z + sy with an appropriate s ∈ K kills a7.
Since S3 is degenerate, a7 = 0 and a2 = 1, we have a1 = 0. Since a6 = 0 and S4 is degenerate, we have
a5a12 = 0.
Case 2a: additionally, a5 ≠ 0 and a12 = 0.
By a sub x → x + ty, y → y, z → z, we can kill a11. Further scaling turns a5 into 1. Thus
Q = xxz +a3xyy +a4xyz +xzy +a8yxz +a9yyy +a10yyz, f1 = xz +a3yy +a4yz +zy, f2 = a8xz +a9yy +a10yz,
f3 = a3xy + xz + a9yy and f4 = xx + a4xy + a8yx + a10yy. Since zy features with non-zero coefficient
only in f1, while xx features with non-zero coefficient only in f4, linear dependence of fj yields linear
dependence of f2 and f3. Since neither is 0, there is α ∈ K∗ such that f3 = αf2. For this to happen, we
must have a3 = a10 = 0. Thus Q = xxz +a4xyz +xzy +a8yxz +a9yyy, f1 = xz +a4yz +zy, f2 = a8xz +a9yy,
f3 = xz +a9yy and f4 = xx +a4xy +a8yx. If a9 ≠ 0, we must have α = 1 and a8 = 1. Further scaling allows
us to turn a9 into 1. In this case Q = xxz + axyz + xzy + yxz + yyy, f1 = xz + ayz + zy, f2 = f3 = xz + yy,
and f4 = xx + axy + yx, where a = a4 ∈ K. If a = 0, a Grobner basis computation yields dim A4 = 16,
contradicting the assumptions. Thus a ≠ 0. Now A is an algebra from (S14).
It remains to deal with the case a9 = 0. Then Q = xxz + axyz + xzy + byxz, f1 = xz + ayz + zy,
f2 = bxz, f3 = xz and f4 = xx + axy + byx, where a = a4 and b = a8 are in K. Since f2 ≠ 0, we must
have b ≠ 0. Now R is spanned by xz, ayz + zy and xx + axy + byz. Again, if a = 0, then dim A4 = 16,
contradicting A ∈ Ω (use Grobner basis). Thus a ≠ 0 and A becomes an algebra from (S3).
Case 2b: additionally (to assumptions of Case 2), a5 = 0 and a12 ≠ 0.
By scaling, we can turn a12 into 1. By a sub x → x + sy, y → y, z → z with an appropriate
s ∈ K, we can kill a4. Then Q = xxz + a3xyy + a8yxz + a9yyy + a10yyz + a11yzy + yzz, f1 = xz + a3yy,
f2 = a8xz + a9yy + a10yz + a11zy + zz, f3 = a3xy + a9yy + a11yz and f4 = xx + a8yx + a10yy + yz. Since
xx features with non-zero coefficient only in f4, while zz features with non-zero coefficient only in f2,
linear dependence of fj yields linear dependence of f1 and f3. Since xz features in f1 (with non-zero
coefficient) but not in f3, the latter fails. Thus Case 2b carries no algebras from Ω.
Case 2c: additionally (to assumptions of Case 2), a5 = a12 = 0.
By a sub x → x + sy, y → y, z → z with an appropriate s ∈ K, we can kill a10. Since a2 = 1
and a1 = a5 = a6 = a7 = a10 = a12 = 0, we get Q = xxz + a3xyy + a4xyz + a8yxz + a9yyy + a11yzy,
f1 = xz + a3yy + a4yz, f2 = a8xz + a9yy + a11zy, f3 = a3xy + a9yy + a11yz and f4 = xx + a4xy + a8yx. If
a11 ≠ 0, then fj are linearly independent. Indeed, xx features only in f4, zy features only in f2, out of
f1 and f3 only f1 sports xz and f3 must be non-zero. Since fj are linearly dependent, we have a11 = 0.
Plugging this back in Q and fj, we get Q = xxz + a3xyy + a4xyz + a8yxz + a9yyy, f1 = xz + a3yy + a4yz,
f2 = a8xz + a9yy, f3 = a3xy + a9yy and f4 = xx + a4xy + a8yx. Now a3a8 ≠ 0. Indeed, otherwise, f2 = 0
or f3 = 0 or yy ∈ R. By means of scaling, we can turn a3 and a8 into 1. If a4 ≠ 0, then fj are linearly
independent. Indeed, xx features only in f4, yz features only in f1, while f2 and f3 are obviously
linearly independent. Hence a4 = 0. Plugging a4 = 0 and a3 = a8 = 1 this back into Q and fj, we get
Q = xxz + xyy + yxz + ayyy, f1 = xz + yy, f2 = xz + ayy, f3 = xy + yy and f4 = xx + yx. The only case
when fj are linearly dependent is a = 1. Thus Q = xxz + xyy + yxz + yyy and R is spanned by xz + yy,
xy + yy and xx + yx. A direct computation shows that dim A3 = 12, contradicting the assumption
A ∈ Ω.
Case 3: a2 = 0 and a6 ≠ 0.
42
This case is obtained from Case 2 by passing to opposite multiplication. That is, in this case A
must be isomorphic to and algebra from (S4) and (S15). Indeed, up to an isomorphism, algebras in
(S4) and (S15) are the algebras from (S3) and (S14) with the opposite multiplication.
Case 4: a2 = a6 = 0.
Since S3 and S4 are degenerate, a1a8 = a5a12 = 0.
Case 4a: additionally, a1 = a5 = 0.
Then Q = a3xyy + a4xyz + a7yxy + a8yxz + a9yyy + a10yyz + a11yzy + a12yzz, f1 = a3yy + a4yz, f2 =
a7xy+a8xz+a9yy+a10yz+a11zy+a12zz, f3 = a3xy+a7yx+a9yy+a11yz and f4 = a4xy+a8yx+a10yy+a12yz.
Now a4 ≠ 0 (otherwise either f1 = 0 or yy ∈ R). We can normalize to make a4 = 1 and use the sub
x → x, y → y, z → z + sy with an appropriate s ∈ K to kill a3.
First, we show that a7 = 0. Assume the contrary: a7 ≠ 0. Then we can normalize to make a7 = 1 and
use the sub x → x +ty, y → y and z → z to kill a9. Now Q = xyz +yxy +a8yxz +a10yyz +a11yzy +a12yzz,
f1 = yz, f2 = xy + a8xz + a10yz + a11zy + a12zz, f3 = yx + a11yz and f4 = xy + a8yx + a10yy + a12yz.
If a12 ≠ 0, fj are easily seen to be linearly independent. Hence a12 = 0. Using this equality, we see
that if a8 ≠ 0, then fj are still linearly independent. Hence a8 = 0. Now if a10 ≠ 0, fj persist in being
linearly independent. Then a10 = 0. Plugging all this back, we get Q = xyz + yxy + a11yzy, f1 = yz,
f2 = xy + a11zy, f3 = yx + a11yz and f4 = xy. Since xy, yz ∈ R, xyz ∈ RV ∩ V R. The latter space is
supposed to be one-dimensional spanned by Q. This contradiction proves that a7 = 0.
First, we consider the case a10 = 0. Plugging a7 = a10 = 0 back into formulas for Q and fj, we get
Q = xyz + a8yxz + a9yyy + a11yzy + a12yzz, f1 = yz, f2 = a8xz + a9yy + a11zy + a12zz, f3 = a9yy + a11yz
and f4 = xy + a8yx + a12yz. Now a9 = 0. Indeed, otherwise either f3 = 0 or yy ∈ R. Since a9 = 0,
we have a11 ≠ 0 (otherwise f3 = 0). We arrive at Q = xyz + a8yxz + yzy + a12yzz, f1 = f3 = yz,
f2 = a8xz + zy + a12zz and f4 = xy + a8yx + a12yz. If a8 = 0, Lemma 2.11 says A ∉ Ω. Hence a8 = a ∈ K∗.
If a12 ≠ 0, it can be turned into 1 by scaling. Thus we have two options Q = xyz + ayxz + yzy + yzz or
Q = xyz + ayxz + yzy with a ∈ K∗. In the first case R is spanned by yz, axz + zy + zz and xy + ayx + yz,
while in the second case R is spanned by yz, axz + zy and xy + ayx landing us into (S6) and (S7).
Now assume a10 ≠ 0. If a8 ≠ −1, then we can use the sub x → x +ty, y → y, z → z with an appropriate
t ∈ K to kill a10, bringing us back to the case a10 = 0, already dealt with. Thus we can assume that
a8 = −1. Next, a11 ≠ 0 (otherwise f3 = 0 or yy ∈ R). By scaling, we can turn a11 into 1. Next, a9 = 0
(otherwise yy ∈ R). If a12 ≠ 0, it can be turned into 1 by scaling.
Thus we have two options Q = xyz − yxz + ayyz + yzy + yzz and Q = xyz − yxz + ayyz + yzy, where
a = a10 ∈ K∗. In the first case R is spanned by yz, −xz + ayz + zy + zz and xy − yx + ayy + yz, while in
the second case R is spanned by yz, −xz + ayz + zy and xy − yx + ayy and we arrive to (S8) and (S9).
Case 4b: additionally (to a2 = a6 = 0), a8 = a12 = 0.
This case is obtained from Case 4a by passing to opposite multiplication. Thus the list of algebras
Indeed, the classes (S10–S13) can (up to an
to one of which A must be isomorphic is (S10–S13).
isomorphism) be obtained from (S6–S9) in this order by passing to the opposite multiplication.
Case 4c: additionally (to a2 = a6 = 0), a1 = a12 = 0 and a5a8 ≠ 0.
Since a5 ≠ 0, we can scale to make a5 = 1. By the sub x → x + sy, y → y, z → z, we can kill
a11. Since a8 ≠ 0 the sub x → x, y → y, z → z + ty with an appropriate t ∈ K kills a7. Plugging
a2 = a6 = a1 = a12 = a11 = a7 = 0 and a5 = 1 into the formula for Q and fj, we get Q = a3xyy + a4xyz +
xzy + a8yxz + a9yyy + a10yyz, f1 = a3yy + a4yz + zy, f2 = a8xz + a9yy + a10yz, f3 = a3xy + xz + a9yy and
f4 = a4xy + a8yx + a10yy. Since zy features only in f1 and yx features only in f4, linear dependence
of fj yields linear dependence of f2 and f3. Since neither is zero, f3 = αf2 for some α ∈ K∗. Then
a3 = a10 = 0. Plugging this back into the formula for Q and fj, we get Q = a4xyz + xzy + a8yxz + a9yyy,
f1 = a4yz + zy, f2 = a8xz + a9yy, f3 = xz + a9yy and f4 = a4xy + a8yx. First, consider the case a9 ≠ 0.
In this case a scaling makes a9 = 1. Then α = 1 and therefore a8 = 1. Thus Q = axyz + xzy + yxz + yyy,
f1 = ayz + zy, f2 = f3 = xz + yy and f4 = axy + yx with a = a4 ∈ K. If a = 0, one easily sees that
dim A3 > 10. Thus a ≠ 0 and we fall under the jurisdiction of (S16). It remains to deal with the case
a9 = 0. Then Q = axyz +xzy +byxz, f1 = ayz +zy, f2 = bxz, f3 = xz and f4 = axy +byx, where a = a4 ∈ K
43
and b = a8 ∈ K∗. If a = 0, one easily sees that dim A3 > 10. Thus a ≠ 0 and we arrive to an algebra
from (S5).
Case 4d: additionally (to a2 = a6 = 0), a5 = a8 = 0 and a1a12 ≠ 0.
This case is obtained from Case 4c by passing to opposite multiplication. Since both classes (S5)
and (S16) are closed under passing to the opposite multiplication, we again have A isomorphic from
an algebra of (S5) or (S16).
Between cases 4a–4d all options for aj satisfying a1a8 = a5a12 = 0 are exhausted.
Part IV of Theorem 1.11 now follows from Lemmas 7.11, 7.12 and 7.10.
8 Proof of Part I of Theorem 1.11
Part I of Theorem 1.11 is a rather odd one out and is technically more difficult than each of the
other parts. We start with some general comments on algebras in Ω having a square in the space of
quadratic relations.
Lemma 8.1. Let L be a 1-dimensional subspace of the 3-dimensional vector space V over K and let R
be a 3-dimensional subspace of the 5-dimensional space LV + V L such that L2 ⊂ R. Then the quadratic
Proof. Let z ∈ V be such that z spans L. Then zz ∈ R. By passing to the opposite multiplication, if
algebra A = A(V, R) satisfies dim A3 ⩾ 12. In particular, A ∉ Ω′.
necessary, we can assume that R ~⊆ LV . A standard linear algebra argument yields that x, y ∈ V can
{xz − azx, yz − bzy, zz} with a, b ∈ K, {xz − azx − zy, yz − azy, zz} with a ∈ K, {xz − azx, zy, zz} with
a ∈ K, {xz − zy, zx, zz} or {xz, zx, zz}. With respect to the chosen basis, the triples serve as defining
be chosen in such a way that x, y, z form a basis in V and R is spanned by one of the following triples
relations for A. Using the usual left-to-right degree-lexicographical ordering with x > y > z, we can
compute the members of the reduced Grobner basis in the ideal of relations of A of degree up to 3
(actually, the defining relations form the Grobner basis already in all cases except for the second last,
for which two degree 3 members of the Grobner basis occur: zyz and zyx). This gives dim A3 = 12 in
all cases except for {xz − azx, zy, zz}, in which we have dim A3 = 13.
Lemma 8.2. Assume that A = A(V, R) ∈ Ω′ and the corresponding quasipotential is Q = z3. Then the
quadratic algebra B = A~I, where I is the ideal generated by z, is uniquely (up to an isomorphism)
determined by A. For B = A(V0, R0), we have dim V0 = 2 and 1 ⩽ dim R0 ⩽ 2 and exactly one of the
following holds true
• R0 = span{yy};
• R0 = span{xy};
• R0 = span{xy − αyx} with α ∈ K∗ being uniquely determined up to replacing it by 1
• R0 = span{xy − yx − yy};
• R0 = span{xx, yy};
• R0 = span{xx − yx, yy};
• R0 = span{xy, yy};
• R0 = span{yx, yy};
• R0 = span{xy − αyx, yy} with α ∈ K∗ being uniquely determined;
• R0 = span{xy, yx};
• R0 = span{xx − xy, yx};
• R0 = span{xx − αxy − yy, yx} with α ∈ K, α2 + 1 ≠ 0 with α being uniquely determined
α ;
for some x, y ∈ V such that x, y, z is a basis in V ({x, y} is now naturally interpreted as a basis in V0).
44
Proof. As the quasipotential for an algebra from Ω′ is unique up to a scalar multiple, z is uniquely
determined up to a non-zero scalar multiple as well. Thus B is uniquely determined by A. Obviously,
for B = A(V0, R0), we have dim V0 = 2 and dim R0 ⩽ 2. However, we can not have dim R0 = 0
since otherwise A satisfies the assumptions of Lemma 8.1, which yields dim A3 ⩾ 12 contradicting the
inclusion A ∈ Ω′. The last statement is a direct application of Lemmas 2.2 and 2.3.
in Lemma 8.2) is spanned by
Note that for A = A(V, R) from (R1–R39), the space R0 (of the algebra B = A(V0, R0), as defined
• {xy + yx, yy} if A is from (R1–R3);
• {xy − yx, yy} if A is from (R4);
• {xy − ayx, yy} with a ∉{0, 1, −1} if A is from (R5–R6);
• {xx − xy, yx} if A is from (R7–R8);
• {yx, yy} if A is from (R9–R12);
• {xy, yy} if A is from (R13–R16);
• {xy, yx} if A is from (R17–R19);
• {xx, yy} if A is from (R20);
• {xy − yx − yy} if A is from (R21–R26);
• {xy − ayx} with a ∉{0, 1} if A is from (R27–R33);
• {xy − yx} if A is from (R34–R36);
• {xy} if A is from (R37–R39).
By Lemma 8.2, algebras from different groups out of (R1–R3), (R4), (R5–R6), (R7–R8), (R9–
R12), (R13–R16), (R17–R19), (R20), (R21–R26), (R27–R33), (R34–R36) and (R37–R39) are non-
isomorphic. This splits the proof into 12 independent parts.
In this section we shall often use the following ordering on monomials in three variables x, y, z. To
introduce it smoother, for an x, y, z monomial u, we denote by u′ the (same degree) monomial obtained
from u by replacing all occurrences of y by x. We also use the symbol ≺ to denote the left-to-right
degree lexicographical ordering assuming x > y > z. The ordering < we shall use is defined as follows.
We say that v < u if and only if
u has higher degree than v,
or u and v have the same degree, but z-degree of v is higher,
or u and v have the same degree and the same z-degree and v′ ≺ u′,
or u and v have the same degree and the same z-degree and v′ = u′ and v ≺ u.
(8.1)
For instance, for monomials of degree up to 3 the just defined order looks like
1 < z < y < x < zz < zy < zx < yz < xz < yy < yx < xy < xx < zzz < zzy < zzx < zyz
< zxz < yzz < xzz < zyy < zyx < zxy < zxx < yzy < yzx < xzy < xzx < yyz
< yxz < xyz < xxz < yyy < yyx < yxy < yxx < xyy < xyx < xxy < xxx.
It is easy to see that the ordering (8.1) is compatible with multiplication and therefore can be used in
computing Grobner bases. Curiously, for the purpose of dealing with algebras in Ω, whose quasipoten-
tial is a cube, this order in most cases proves much more convenient than the degree lexicographical
one. On a number of occasions it even yields a finite Grobner basis while the degree lexicographical
ordering provides an infinite one.
45
8.1 Case R0 = span {xy + ayx, yy}
Lemma 8.3. Let A be a quadratic algebra given by generators x, y, z and relations
xy − αyx − azx − bzy, yy − xz − qyz − czx − dzy and zz,
where α, a, b, c, d, q ∈ K and α ≠ 0. Then A ∈ Ω′. Moreover, A ∈ Ω if and only if
c = α2, a = α(d − q), b = d(d−q)
α
and (α + 1)((2 − α)d −(α2 − α + 1)q) = 0.
(8.2)
(8.3)
Furthermore, two algebras from this family with parameters α, a, b, c, d, q and α′, a′, b′, c′, d′, q′ are iso-
morphic if and only if either α = α′ ≠ 1 and (a′, b′, c′, d′, q′) = (at, bt2, c, dt, qt) for some t ∈ K∗ or
α = α′ = 1 and (a′, b′, c′, d′, q′) =(at, t2(b + as), c, t(d + s), t(q + s)) for some t ∈ K∗ and s ∈ K.
Proof. First, we deal with the isomorphism question. Assume that two algebras from our family
with parameters α, a, b, c, d, q and α′, a′, b′, c′, d′, q′ are isomorphic. Then there is a linear substitution
facilitating the isomorphism. Since z2 is the only square in the space of quadratic relations for both
algebras, z is mapped to its own scalar multiple by this substitution: z → wz with w ∈ K∗. Since
by z for both algebras, our substitution must map y to vy + sz with v ∈ K∗, s ∈ K. By Lemma 8.2
(the uniqueness), α is an isomorphism invariant. Thus we must have α = α′. First, consider the case
y2 is the only square in the space of quadratic relations for A~I with I being the ideal generated
α = α′ ≠ 1. Then it is easy to see that the shape of the space of quadratic relations of the quotient A~I
will not be preserved unless x → ux + tz with u ∈ K∗, t ∈ K. Now applying the substitution x → ux + tz,
y → vy + sz, z → wz, we see that the shape of the space of quadratic relations of the algebra A will not
be preserved unless s = t = 0. We are left with scalings only. Now it is a matter of direct verification
to see that a scaling provides a required isomorphism if and only if (a′, b′, c′, d′, q′) = (at, bt2, c, dt, qt)
for some t ∈ K∗. It remains to consider the case α = α′ = 1. Similar considerations show that in this
case the shape of the space quadratic relations of A is preserved (stays in our family) if and only if
the substitution has the form x → ux + rz, y → vy, z → wz with u, v, w, r ∈ K, uvw ≠ 0 and uw = v2.
Applying these substitutions, we directly see that one of them provides a required isomorphism if and
only if (a′, b′, c′, d′, q′) =(at, t2(b + as), c, t(d + s), t(q + s)) for some t ∈ K∗ and s ∈ K.
Now we deal with the Hilbert series of A. Throughout the proof we use the order (8.1) on x, y, z
monomials. Resolving the overlaps yyy and xyy, we see that the degree 3 part of the Grobner basis
of the ideal of relations of A consists of two members
g1 = yxz−xzy+cyzx+(d−q)yzy−αczyx+(qc−d)zxz,
g2 = xxz+(c−α2)xzx+(αq+d)xzy−α(qc+αq+a)yzx−α(b+qd−q2)yzy
−α2czxx−α(a + αd−αqc)zyx+(qa+αqd−b−αq2c)zxz.
Since g1 and g2 are inearly independent, it follows that dim A3 = 10 regardless what the values of
the parameters are. Thus A ∈ Ω′. There are 4 degree 4 overlaps yxzz, xxzz, yyxz and xyxz. The
members of the ideal of relations obtained from the last two overlaps always belong to the linear span
of the ones obtained from the first two overlaps. We denote the latter r1 and r2 respectively. The
explicit formulae for rj are as follows:
r1 =xzyz−cyzxz+(q−d)yzyz+αczxzy−αc2zyzx+αc(q−d)zyzy;
+qd)yzyz+α2c(α2
r2 =(α2
−c)xzxz+(α2q+αa−cd)yzxz+(αb−d2
−c)zxzx+α(cd−α2qc−αcd+a+αd)zxzy
+α(α2qc2
−c2d+α3qc+α2ac−ac−αcd)zyzx+α(α2bc+αqcd−α2q2c+qa+αqd+qcd−ad−αd−cd2)zyzy
Note that there are exactly 16 degree 4 monomials which do not contain the leading terms xy, yy,
zz, yxz and xxz of the members of the Grobner basis of degree up to 3. For A to be in Ω, we must
have dim A4 = 15. This happens precisely when the dimension of the space L spanned by r1 and r2
is exactly 16 − 15 = 1. On the other hand, the monomial xzyz features in r1 with non-zero coefficient
46
and does not feature in r2. Hence the only way for the dimension of L to be 1 is to have r2 = 0. Thus
we have a system of algebraic equations on the parameters coming from the coefficients of r2 being
zero. The equation coming from the xzxz-coefficient is c = α2. Plugging this into the yzxz one, we
. Plugging all this into the zxzy
get a = α(d − q). Plugging both into yzyz one, we get b = d(d−q)
coefficient, we get (α + 1)((2 − α)d −(α2 − α + 1)q) = 0. Now the rest of the equations are automatically
satisfied provided these four are. Thus we have that dim A4 = 15 if and only if r2 = 0 if and only if
(8.3) is satisfied.
α
Now if r2 = 0, one easily checks that the Grobner basis of the ideal of relations of A actually ends
with r1: it consists of the defining relations, g1, g2 and r1 (great advantage of the ordering we picked!).
The leading monomials of the elements of the basis are xy, yy, zz, yxz, xxz and xzyz. Knowing
these, we easily confirm that HA = (1 − t)−3 and A is indeed in Ω. Thus A ∈ Ω if and only if (8.3) is
satisfied.
The main result of this section is the following lemma.
Lemma 8.4. Let A = A(V, R) ∈ Ω be a quadratic algebra such that dim V = dim R = 3 and with respect
to some basis x, y, z in V , zz ∈ R and the quadratic algebra B = A~I with I being the ideal generated
by z, is given by the relations xy − αyx and yy with α ∈ K∗. Then A is isomorphic to to a K-algebra
given by generators x, y, z and three quadratic relations from (R1–R6) of Theorem 1.11. Furthermore,
the algebras in (R1–R6) belong to Ω− and are pairwise non-isomorphic.
Proof. By the assumptions, R is spanned by zz, xy −αyx +f and yy +g with f, g ∈ span{xz, zx, yz, zy}.
Using a substitution x → x + sz, y → y + tz, z → z with appropriate s, t ∈ K, we can kill the xz and
yz coefficients of f . We shall see later that the xz coefficient in g must be non-zero. In the meantime
though, we view it as an extra assumption.
Case 1: the xz coefficient in g is non-zero.
By means of scaling, we can turn this coefficient into −1. Then A is given by the generators x, y, z
and the relations from (8.2) for some α, a, b, c, d, q ∈ K and α ≠ 0. Since A ∈ Ω, Lemma 8.3 implies that
(8.3) is satisfied.
q = 2 − α, d = α2 − α + 1 and plugging this into the rest of the equations in (8.3), we get c = α2,
. We have arrived to an algebra from (R5) with a = α. If q = d = 0,
Case 1a: α ≠ 1 and α ≠ −1. The last equation in (8.3) reads (2 − α)d =(α2 − α + 1)q. According to
the isomorphism part of Lemma 8.3, we can (by a scaling) multiply (q, d) by any non-zero constant
without breaking up the overall shape of relations. Then if (q, d) ≠(0, 0), since (2−α, α2 −α +1) ≠(0, 0)
and (2 − α)d = (α2 − α + 1)q, by doing this we can turn (q, d) into (2 − α, α2 − α + 1). After this sub
a = α(α2 − 1) and b = (α−1)(α3
a = q − d and b = d(q − d). As above, a scaling allows to multiply (q, d) by any non-zero constant.
yy − xz(1 − d)yz − zx − dzy and zz and we have an algebra from (R1). It remains to consider the case
Thus if q ≠ d, we can turn d − q into 1. Then the defining relations take form xy + yx + zx + dzy,
Case 1b: α = −1. Then the last equation in (8.3) is satisfied automatically. The rest yield c = 1,
the equations in (8.3) read c = α2 and a = b = 0. In this case we have an algebra from (R6).
q = d. Then a = b = 0 and the defining relations take the form xy + yx, yy − xz − dyz − zx − dzy and
zz. If d ≠ 0, a scaling transforms them into xy + yx, yy − xz − yz − zx − zy and zz and we arrive to the
algebra (R2). If d = 0, we have the algebra (R1).
+1)
α
Case 1c: α = 1. In this case the equations (8.3) read a = b = 0, c = 1 and d = q. The defining
relations take the form xy − yx, yy − xz − dyz − zx − dzy and zz. The isomorphism part of Lemma 8.3
in the case α = 1 implies that all these algebras (when d varies) are isomorphic to each other. Hence
they are isomorphic to the algebra with d = 0, which is (R4).
Case 2: the xz coefficient in g is zero, while the zx coefficient is non-zero.
We shall show that this case does not occur. A sub of the form x → x + sz, y → y + tz, z → rz with
appropriately chosen s, t ∈ R and r ∈ K∗ will kill the zx and zy coefficients in f (the monomials xz
and yz will creep back in) and turn the non-zero zx coefficient of g into −1. The defining relations of
47
A take form xy − αyx − axz − byz, yy − qyz − zx − dzy and zz with α, a, b, d, q ∈ K, α ≠ 0. Since A ∈ Ω,
so is A with the opposite multiplication. The latter is isomorphic to the quadratic algebra C given by
generators x, y, z and relations yx − αxy − azx − bzy, yy − qzy − xz − dyz and zz. Up to scalar multiples,
the relations of C are of the form (8.2). Hence the parameters must satisfy (8.3), the first equation in
which yields 0 = 1
α2 . Since this is obviously faulty, Case 2 actually does not occur.
Case 3: both xz and zx do not feature in g.
In this case Lemma 2.11 yields A ∉ Ω, contradicting the assumptions. Hence this case does not
occur as well. It remains to notice that algebras in (R1–R6) are all in the family given by relations
(8.2). The isomorphism part of Lemma 8.3 easily implies that algebras in (R1–R6) are pairwise
non-isomorphic.
8.2 Case R0 = span {xx − xy, yx}
Lemma 8.5. Let A be the quadratic algebra given by the generators x, y, z and the relations xx−xy −yz,
yx − azx − bzy and zz with a, b ∈ K. If either b = 0 and a ≠ −1 or b ≠ 0 and a(1 + b + . . . + bk) ≠ 1 for all
k ∈ Z+, then A ∈ Ω−. Otherwise, A ∉ Ω.
Proof. The fact that A ∈ Ω− whenever A ∈ Ω follows from Lemma 1.13. It remains to prove that A ∈ Ω
First, we get rid of the easy case b = 0. In this case we use the left-to-right degree lexicographical
if and only if either b = 0 and a ≠ −1 or b ≠ 0 and a(1 + b + . . . + bk) ≠ 1 for all k ∈ Z+.
ordering assuming z > y > x. The ideal of relations of A always has finite Grobner basis. If a ∉{0, −1},
the said basis consists of the defining relations together with zyx, xxz + xxy − xxx, yyx + axzx − axxx
and xxyx − a
a+1 xxxx. If a = 0, the basis consists of the defining relations together with xxz + xxy − xxx.
In both cases it follows that HA =(1 − t)−3 and therefore A ∈ Ω. On the other hand, if a = −1, then the
Grobner basis consists of the defining relations together with zyx, xxz + xxy − xxx, yyx − xzx + xxx
and xxxx, yielding dim A5 = 23. Hence A ∉ Ω in this case. For the rest of the proof we shall assume
that b ≠ 0.
We start by a substitution, which turns the defining relations into a more convenient form. After
the permutation z → y → x → z, the defining relations take the form yy, xz − ayz − byx and xy + zx − zz.
Now we perform the sub x → x + ay, y → y and z → z, after which the relations take the form yy,
xz − byx and xy + zx + azy − zz. Finally, after scaling x → bx, y → y and z → bz, the relations acquire
the shape yy, xz − yx and xy + bzx + azy − bzz, which shows that A is isomorphic to the algebra B
given by the generators x, y, z and the relations
yy, xz − yx and xy + bzx + azy − bzz.
(8.4)
We shall compute the Grobner basis of the ideal of relations of B with respect to the left-to-right
degree lexicographical ordering assuming x > y > z. First, a direct computation shows that the only
degree 3 and 4 members of the reduced Grobner basis of the ideal of relations of B are
zzx −
1−a
b zzy − zzz and zzyx −
1−a
b zzyz − zzzz.
Similarly, the degree 5 part of the Grobner basis consists of
(a − 1)zzyzz + bz5 and b2zzyzx +(ab + a − 1)zzyzy − b2zzyzz + bz5.
Consider the recurrent sequence α1 = a−1
b and αk+1 = αk +a
b
following statement (k ∈ N):
for k ∈ N. We shall prove inductively the
(Gk) If αj ≠ 0 for 1 ⩽ j ⩽ k, then the complete list of the degree up to 2k + 3 elements of the
reduced Grobner basis of the ideal of relations of B consists of the defining relations yy, xz − yx,
xy + bzx + azy − bzz, the degree 3 element zzx −
1−a
b zzy − zzz, zz(yz)j −1yx + αj zz(yz)j + z4uj,
b z4ujy + γjz2k+3 for 1 ⩽ j ⩽ k, where um =
1
zz(yz)j z + γj z2j +3 and zz(yz)j x + αj +1zz(yz)j y +
48
m−1
∑
j =0
βm,j z2m−2j(yz)j ∈ B2m−2 (βm,j ∈ K) and γm ∈ K are defined recurrently: u1 = −1 (that is,
b um−1yz + γmz2m−2
1−a and for m > 1, γm =
γj βm−1,j, um = 1
β1,0 = −1), γ0 = 1, γ1 = −
= b
1
α1
m−1
∑
j =0
(yielding a recurrent formula for βm,j as well).
If α1 ≠ 0 (that is, a ≠ 1), then the degree up to 5 elements of the Grobner basis, collected in the
above two displays, easily justify (G1). Now assume that k ∈ N and (Gk) holds. We shall verify (Gk+1)
b z4uky.
The rest of degree 2k + 4 overlaps resolve. Using the recurrent definitions of um and αm, we see that
provided αj ≠ 0 for 1 ⩽ j ⩽ k + 1. The overlap zz(yz)kxz produces zz(yz)kyx +
+ γkz2k+3 + 1
αk +a
b
Next,
the degree 2k + 4 part of the reduced Grobner basis of B
consists of one element zz(yz)kyx + αk+1zz(yz)j + z4uk+1.
the overlap zz(zy)kyxz produces αk+1zz(yz)k+1z + z4uk+1z.
αk+1, the expression in (8.6) acquires the shape zz(yz)k+1z + γk+1z2k+5. Finally,
the overlap zz(zy)kyxz produces zz(yz)k+1x+αk+1zz(yz)k+1y−zz(yz)k+1z+
By assumption, αk+1 ≠ 0. Using the recurrent definitions of um and αm, we see that after dividing by
1
b z4uk+1y.
(8.7)
(8.5)
(8.6)
1
bm+1
b z4uk+1y + γk+1z2k+3. This concludes the proof of (Gk+1).
other overlaps of degree 2k + 5 resolve, we see that the degree 2k + 5 part of the reduced Grobner
Since when αk+1 ≠ 0, we already have zz(yz)k+1z + γk+1z2k+5 in the ideal of relations and since the
basis of the ideal of relations of B consists of two members: zz(yz)k+1z + γk+1z2k+5 and zz(yz)k+1x +
αk+1zz(yz)k+1y +
Now assume that a(1 + b + . . . + bk) ≠ 1 for all k ∈ Z+. It easily follows that αk ≠ 0 for every k ∈ N.
Indeed, one easily checks that αm+1 = a(1+...+bm)−1
. Then the assumption of each (Gk) is satisfied
and (Gk) produce the complete reduced Grobner basis of the ideal of relations of B. The set of
its leading monomials consists of xy, xz, yy, zz(yz)k−1x, zz(yz)kyx and zz(yz)kz for k ∈ N. As a
result, the corresponding normal words are zδ(yz)myεxk, yδ(zy)mzpyxq and yδ(zy)mzp(yz)kyε, where
ε, δ ∈ {0, 1}, k, m ∈ Z+, q ∈ N and p ⩾ 2. Now it is easy to see that the number of normal words of
degree n is exactly (n+1)(n+2)
It remains to deal with the case when a(1 + b + . . . + bm) = 1 for some m ∈ N. Then there is the
(8.6) is z4(yz)kz). Now all degree 2k + 6 overlaps resolve except for zz(yz)k+1xz, which yields a basis
member with zz(yz)k+1 as the leading monomial. Knowing the leading monomials of the Grobner
basis members of degrees up to 2k +6, we get that dim A2k+6 = dim B2k+6 = (2k+7)(2k+8)
A ∉ Ω.
smallest k ∈ Z+ for which αk+1 = 0. Then (Gk) is satisfied. Furthermore, (8.5–8.7) still give the degrees
2k + 4 and 2k + 5 parts of the Grobner basis of the ideal of relations of B (the leading monomial in
and therefore B ∈ Ω. Hence, A ∈ Ω.
+1 and therefore
2
2
Lemma 8.6. Let A = A(V, R) ∈ Ω be a quadratic algebra such that dim V = dim R = 3 and with
respect to some basis x, y, z in V , zz ∈ R and the quadratic algebra B = A~I with I being the ideal
generated by z, is given by the relations xx − xy and yx. Then A is isomorphic to to a K-algebra given
by generators x, y, z and three quadratic relations from (R7–R8) of Theorem 1.11. Furthermore, the
algebras in (R7–R8) belong to Ω− and are pairwise non-isomorphic.
Proof. By the assumptions, R is spanned by zz, xx − xy + f and yx + g with f, g ∈ span{xz, zx, yz, zy}.
Using a substitution x → x + sz, y → y + tz, z → z with appropriate s, t ∈ K, we can kill the xz and zx
coefficients of f . Then A is given by the generators x, y, z and the relations
xx − xy − ayz − bzy, yx − pxz − qzx − cyz − dzy and zz,
(8.8)
49
where a, b, c, d, p, q ∈ K.
Then Aopp, being A with the opposite multiplication, is isomorphic to the algebra C given by the
generators x, y, z and the relations xx − yx − azy − byz, xy − pzx − qxz − czy − dyz and zz. After the
substitution x → x +(p + c)z, y → x − y +(q + d)z, z → z, the defining relations of C take the shape
xx − xy + dyz − pzy, yx − bxz +(p + q + c + d)zx +(p + b + c)yz + azy. That is,
Aopp is isomorphic to an algebra given by (8.8) with the
parameters (−d, p, −p − b − c, −a, b, −p − q − c − d) in place of (a, b, c, d, p, q)
(8.9)
Throughout the proof we again use the order (8.1) on x, y, z monomials. Resolving the overlaps
xxx and yxx, we see that the degree 3 part of the Grobner basis of the ideal of relations of A consists
of two members
g1 = ayyz−pxzx+pxzy−cyzx+(b+c)yzy+dzyy−pdzxz−(aq+cd)zyz,
g2 = xyy+(a−c−p)xyz−qxzx+(b−d)xzy−ayzx+ayzy+bzyy−pbzxz−(bc+pb)zyz.
(8.10)
Case 1: a ≠ 0.
By means of scaling, we can turn a into 1. The leading monomials of g1 and g2 are now yyz and
xyy. One easily sees that dim A3 = 10 regardless what the values of other parameters are. There are 5
degree 4 overlaps yyzz, xyyz, xyyx, xxyy and yxyy. The members of the ideal of relations obtained
from the last two overlaps always belong to the linear span of the ones obtained from the first three.
We denote these r1, r2 and r3 respectively. The explicit formulae for rj (we assume a = 1) are as
follows:
r1 =pxzxz−pxzyz+cyzxz−(b+c)yzyz−pdzxzx+pdzxzy−cdzyzx+d(b+c)zyzy;
r2 =−(p+c)xyzx+(p+b+c)xyzy+dxzyy+(q−pd)xzxz+(d−b−q−cd)xzyz+yzxz
−yzyz−pbzxzx+pbzxzy−(bc+pb)zyzx+b(b+c+p)zyzy;
r3 =(p+c−q−1)xyzx−dxyzy+qxzxy+yzxy+(pd−pb−pq−cq)xzxz+(q−pd)xzyz
−(p+c)yzxz+yzyz+pbzxzx+(bc+pb−qb)zyzx−bdzyzy.
Note that there are exactly 17 degree 4 monomials which do not contain the leading terms xx, yx,
zz, yyz and xyy of the members of the Grobner basis of degree up to 3. Since A ∈ Ω, we must have
dim A4 = 15. This happens precisely when the dimension of the space L spanned by r1, r2 and r3 is
exactly 17 − 15 = 2. On the other hand, yzxy features in r3 with coefficient 1 and does not feature in
r1 or r2. Thus dim L = 2 implies that the space L1 spanned by r1 and r2 is one-dimensional. The 2 × 2
matrix S1 of yzxz and yzyz coefficients in r1 and r2 is
S1 = c −b − c
−1
1
.
Since S1 must have rank at most 1, we have b = 0. Plugging this back into r1 and r2, we see that L1
is one-dimensional precisely means that the rank of S2 is 1, S2 being the matrix of coefficients of r1
and r2:
S2 =
0
0
p
−p
c pd cd
p + c d q − pd d − q − cd 1
0
0 .
Now it is elementary to see that rank of S2 equals 1 precisely when either p = c = 0 or d = p +c = q +1 = 0.
In the case d = p +c = q +1 = 0, the defining relations of A take the shape xx −xy −yz, yx −pxz +zx +pyz
and zz. Using Grobner basis, one easily sees that for generic p, dim A = 23 in this case. By Lemma 2.7,
dim A ⩾ 23 for all p, which violates the assumption A ∈ Ω. Thus the case d = p + c = q + 1 = 0 does
not occur. This leaves us with the option p = c = 0. Then the defining relations of A take the shape
xx − xy − yz, yx − qzx − dzy and zz. By Lemma 8.5, A falls into (R7) any algebra in (R7) belongs to
Ω−. This concludes Case 1.
50
Case 2: a = 0 and d ≠ 0.
By means of scaling, we can turn d into −1. By (8.9), Aopp is isomorphic to an algebra given by
(8.8) with the parameters (1, p, −p −b −c, 0, b, −p −q −c +1) in place of (a, b, c, d, p, q). The latter is under
the jurisdiction of Case 1. From Case 1 we know that the inclusion Aopp ∈ Ω yields p = b = −p − b − c = 0.
That is p = b = c = 0. Plugging a = b = c = p = 0 and d = −1 back into (8.8), we see that the defining
relations of A take the shape xx − xy, yx − qzx + zy and zz. If additionally q = 0, a direct Grobner
basis computation yields dim A5 = 23, which is incompatible with A ∈ Ω. Thus q ≠ 0 and we fall into
(R8). Plugging a = b = c = p = 0 and d = −1 into (8.10), we get g1 = −zyy and g2 = xyy − qxzx + xzy.
Continuing the computation of the Grobner basis, we see that it actually turns out finite comprising
the defining relations, g1, g2, qzyzx − zyzy and qxyzx − xyzy − qxzxy. The leading monomials of the
members of the Grobner basis are xx, yx, zz, zyy, xyy, zyzx and xyzx. This allows to compute the
Hilbert series of A: HA =(1 − t)−3. Thus algebras in (R8) are in Ω. By Lemma 1.13, they are in Ω−.
Case 3: a = d = 0 and p ≠ 0.
By means of scaling, we can turn p into 1. Plugging a = d = 0 and p = 1 into (8.8) and (8.10), we
see that the defining relations of A are xx − xy − bzy, yx − xz − qzx − cyz and zz, while the degree
3 members of the Grobner basis of the ideal of relations of A are xzx − xzy + cyzx −(b + c)yzy and
xyy −(c + 1)xyz +(b − q)xzy + qcyzx − q(b + c)yzy + bzyy − bzxz − b(c + 1)zyz. Resolving the degree 4
overlaps of the leading monomials, we find that the degree 4 part of the Grobner basis is spanned by
4 elements r1, r2, r3 and r4 (they come from the overlaps yxzx, xxzx, xzxx and xyyx respectively),
where
r1=−cyyyz+(b+c)yyzy+qczyzx−q(b+c)zyzy;
r2=−(c+1)xyzx+(b+c+1)xyzy−bzyzx+bzyzy;
r3=−xzyy−(b+c)yzyy+(c+1)xzyz+byzxz+(c+1)(b+c)yzyz;
r4=(c+1−q)xyzx+qxzyy−qcyzxy+q(b+c)yzyy−(b+q(c+1))xzyz+(qc(c+1)−b2)yzxz
−q(c+1)(b+c)yzyz+bzxzy+b(1−q)zyzx+b(b+c)zyzy.
Note that there are exactly 18 degree 4 monomials which do not contain the leading terms xx, yx,
zz, xyy and xzx of the members of the Grobner basis of degree up to 3. Since we have assumed that
A ∈ Ω, we must have dim A4 = 15. This happens precisely when the dimension of the space L spanned
by r1, r2, r3, r4 is exactly 18 − 15 = 3. If b ≠ 0, one easily sees that rj are linearly independent (just look
at the 4 × 6 matrix of zxzy, xyzx, xyzy, yyzx, yyzy and xzyy coefficients of rj). Thus b = 0. Next,
it is easy to see that if c is neither 0 nor −1, rj are still linearly independent. Thus c = 0 or c = −1.
If c = −1, the defining relations of A are xx − xy, yx − xz − qzx + yz and zz. Using Grobner basis
technique, one easily sees that in this case for generic q, dim A5 = 22. By Lemma 2.7, dim A5 ⩾ 22 for
all q, contradicting A ∈ Ω. If c = 0, the defining relations of A are xx − xy, yx − xz − qzx and zz. Using
Grobner basis technique, one easily sees that in this case for generic q, dim A5 = 23. By Lemma 2.7,
dim A5 ⩾ 23 for all q, contradicting A ∈ Ω. Thus this case does not occur.
Case 4: a = d = p = 0 and c ≠ 0.
By means of scaling, we can turn c into 1. Plugging a = d = p = 0 and c = 1 into (8.8) and (8.10), we
see that the defining relations of A are xx −xy −bzy, yx −qzx −yz and zz, while the degree 3 members of
the Grobner basis of the ideal of relations of A are yzx−(b+1)yzy and xyy−xyz−qxzx+bxzy+bzyy−bzyz.
Resolving the degree 4 overlaps of the leading monomials, we find that the degree 4 part of the Grobner
basis is spanned by 2 elements r1 and r2 (they come from the overlaps yzxx and xyyx respectively),
where
r1=(b+1)(yzyy−yzyz) and r2 =(1−q)(b+1)(xyzy+bzyzy)+q(xzxy−xzxz).
Note that there are exactly 16 degree 4 monomials which do not contain the leading terms xx, yx,
zz, xyy and yzx of the members of the Grobner basis of degree up to 3. Since we have assumed that
A ∈ Ω, we must have dim A4 = 15. This happens precisely when the dimension of the space L spanned
by r1 and r2 is exactly 16 − 15 = 1. This can only happen when b = −1. Thus the defining relations of
A are xx − xy + zy, yx − qzx − yz and zz. Using Grobner basis technique, one easily sees that in this
51
case for generic q, dim A5 = 23. By Lemma 2.7, dim A5 ⩾ 23 for all q, contradicting A ∈ Ω. Thus this
case does not occur.
Case 5: a = d = p = c = 0.
The defining relations of A are xx−xy −bzy, yx−qzx and zz. If b = 0, one easily sees that dim A3 > 10,
which is incompatible with A ∈ Ω. This b ≠ 0. By scaling, we can make b = 1. The defining relations of
A become xx − xy − zy, yx − qzx and zz. Using Grobner basis technique, one easily sees that in this
case for generic q, dim A5 = 22. By Lemma 2.7, dim A5 ⩾ 22 for all q, contradicting A ∈ Ω. Thus this
final case does not occur.
It remains to show that algebras in (R7–R8) are pairwise non-isomorphic. For B = A~I for algebras
A from (R7–R8), x is up to a scalar multiple the only degree 1 for which there exist degree 1 elements
v and w satisfying vx = xw = 0 in B. Next, y is up to a scalar multiple the only degree 1 element
satisfying yx = 0 in B. Since zz is the only square in R, a linear substitution providing an isomorphism
between to algebras in (R7–R8) must be of the form x → ux + tz, y → vy + sz, z → wz with u, v, w ∈ K∗
and s, t ∈ K. Applying these to the relations in (R7–R8), we see that the only way such a sub transforms
a space of quadratic relations corresponding to an algebra from (R7–R8) to another such space is to
be of the form x → ux, y → uy, z → uz with u. Since this substitution provides an automorphism of
each algebra in (R7–R8), we see that algebras in (R7–R8) are pairwise non-isomorphic.
8.3 Cases R0 = span {yy, yx} and R0 = span {yy, xy}
Lemma 8.7. Let A be a quadratic algebra given by generators x, y, z and relations
yx − pxz − azx − bzy, yy − qxz − czx − dzy and zz,
where a, b, c, d, p, q ∈ K. Then A ∈ Ω if and only if
c ≠ 0, q = 0 and bc = ad.
(8.11)
(8.12)
Furthermore, two algebras from this family with parameters a, b, c, d, p, q and a′, b′, c′, d′, p′, q′ are iso-
morphic if and only if (a′, b′, c′, d′, p′, q′) =ta, tb
s , tsc, td, tp, tsq for some t, s ∈ K∗.
Proof. Throughout the proof we again use the order (8.1) on x, y, z monomials. Resolving the overlaps
yyy and yyx, we see that the degree 3 part of the Grobner basis of the ideal of relations of A consists
of two members
g1 = −czxy−qxzy+cyzx+dyzy+q(a−d)zxz+qbzyz;
g2 = −czxx−qxzx+ayzx+byzy+p(a−d)zxz+pbzyz.
(8.13)
Case 1: c = 0.
By Lemma 2.11, A ∉ Ω if q = 0. Assume that q ≠ 0. By means of scaling, we can turn q into 1:
q = 1. Plugging c = 0 and q = 1 into (8.11) and (8.13), we see that the defining relations of A are
yx − pxz − azx − bzy, yy − xz − dzy and zz, while the degree 3 elements of the Grobner basis of the ideal
of relations of A are −xzy + dyzy +(a−d)zxz + bzyz and −xzx + ayzx + byzy + p(a−d)zxz + pbzyz. The
degree 4 part of the Grobner basis of the ideal of relations is spanned by 2 elements r1 and r2 arising
from the overlaps xzyy and xzyx (other overlaps resolve). The explicit formulae for rj (we assume
q = 1) are as follows:
r1=(d−a)yzxz−byzyz+(b+ad−d2)zyzy and r2 =(b+a2
−ad)zyzx+((b−pd)(a−d)−pb)zyzy.
Note that there are exactly 16 degree 4 monomials which do not contain the leading terms yx, yy,
zz, xzy and xzx of the members of the Grobner basis of degree up to 3. Since A ∈ Ω, we must have
dim A4 = 15. This happens precisely when the dimension of the space L spanned by r1 and r2 is
exactly 16 − 15 = 1. That is, the rank of the matrix S of coefficients of r1 and r2 is 1:
rk S = 1, where S = d − a −b
0
0
0
b + a(a − d)
b + d(a − d)
(b − pd)(a − d) − pb .
52
It is a routine exercise to see that S has rank 1 only if p = a and b = −a(a − d). Now the defining
relations of A are yx − axz − azx + a(a − d)zy, yy − xz − dzy and zz. A direct Grobner basis computation
shows that for generic a, d, dim A5 = 22. By Lemma 2.7, dim A5 ⩾ 22 for all a, d. Hence in Case 1,
A ∉ Ω.
Case 1: c ≠ 0.
By means of scaling, we can turn c into 1: c = 1. One easily sees that dim A3 = 10 regardless what
the values of other parameters are. The leading monomials of g1 and g2 are zxy and zxx respectively.
The degree 4 part of the Grobner basis of the ideal of relations is spanned by 2 elements r1 and r2
arising from the overlaps zzxx and zzxy (other overlaps resolve). The explicit formulae for rj (we
assume c = 1) are as follows:
r1=−qzxzx+azyzx+bzyzy and r2 =−qzxzy+zyzx+dzyzy.
Note that there are exactly 16 degree 4 monomials which do not contain the leading terms yx, yy,
zz, zxy and zxx of the members of the Grobner basis of degree up to 3. If A ∈ Ω, we must have
dim A4 = 15. This happens precisely when the dimension of the space L spanned by r1 and r2 is
exactly 16 − 15 = 1. The latter happens precisely when r1 is a scalar multiple of r2. This, in turn,
happens if and only if q = 0 and b = ad, which corresponds to q = 0 and bc = ad before scaling. Thus
dim A4 = 15 if and only if (8.12) is satisfied. In this case, computing the next step of the Grobner
basis, we see that no new terms appear: the defining relations, g1, g2 and r2 form the Grobner basis
of the ideal of relations of A. The leading monomials of the members of the basis are yx, yy, zz, zxy,
zxx and zyzx. This allows to compute the Hilbert series of A and confirm that HA =(1 − t)−3. Thus,
A ∈ Ω if and only if (8.12) is satisfied.
Now standard argument shows that only scalings transform the space of quadratic relations R for
any given algebra from (8.11) to another space like that. Applying scalings to relations in (8.11), we
easily confirm the required isomorphism statement.
Lemma 8.8. Let A be a quadratic algebra given by generators x, y, z and relations (8.11). Then A ∈ Ω
if and only if A is isomorphic to an algebra given by generators x, y, z and relations
yx − pxz − azx − abzy, yy − zx − bzy and zz,
(8.14)
where α, a, b, p ∈ K. Furthermore, two algebras from this family with parameters a, b, p and a′, b′, p′ are
isomorphic if and only if (a′, b′, p′) =(ta, tb, tp) for some t ∈ K∗.
Proof. By Lemma 8.7, A ∈ Ω if and only if c ≠ 0, q = 0 and bc = ad. A scaling allows to turn c into
1, which brings A into the subfamily (8.14). By the same criterion for A to be in Ω, every algebra
given by the relations (8.14) is in Ω. As for isomorphisms, the result immediately follows from the
isomorphism part of Lemma 8.7.
Lemma 8.9. Let A = A(V, R) ∈ Ω be a quadratic algebra such that dim V = dim R = 3 and with respect
to some basis x, y, z in V , zz ∈ R and the quadratic algebra B = A~I with I being the ideal generated
by z, is given by the relations yy and yx. Then A is isomorphic to to a K-algebra given by generators
x, y, z and three quadratic relations from (R9–R12) of Theorem 1.11. Furthermore, the algebras in
(R9–R12) belong to Ω− and are pairwise non-isomorphic.
Proof. The assumptions imply that R is spanned by zz, yx+f and yy+g with f, g ∈ span{xz, zx, yz, zy}.
Using a linear substitution x → x + sz, y → y + tz, z → z with appropriate s, t ∈ K, we can kill the
yz coefficients of f and g. Then A is given by the generators x, y, z and the relations (8.11). By
Lemma 8.8, A is isomorphic to an algebra given by relations (8.14). If b ≠ 0, then a scaling turns b
into 1 and we have an algebra from (R9). If b = 0 and a ≠ 0, a scaling turns a into 1, while keeping
the equality b = 0 and we have an algebra from (R10). If a = b = 0 and p ≠ 0, a scaling turns p into 1,
while keeping the equality a = b = 0 and we have the algebra (R11). Finally, if a = b = p = 0, we already
53
have the algebra (R12). Finally, Lemma 8.8 immediately implies that algebras in (R9–R12) belong to
Ω and are pairwise non-isomorphic. They belong to Ω− by Lemma 1.13.
Lemma 8.10. Let A = A(V, R) ∈ Ω be a quadratic algebra such that dim V = dim R = 3 and with
respect to some basis x, y, z in V , zz ∈ R and the quadratic algebra B = A~I with I being the ideal
generated by z, is given by the relations yy and xy. Then A is isomorphic to to a K-algebra given by
generators x, y, z and three quadratic relations from (R13–R16) of Theorem 1.11. Furthermore, the
algebras in (R13–R16) belong to Ω− and are pairwise non-isomorphic.
Proof. This lemma is obviously equivalent to Lemma 8.9. One just has to pass to the opposite
multiplication.
9 Cases R0 = span {xy, yx}
Lemma 9.1. Let A = Aa be the quadratic algebra given by the generators x, y, z and the relations
xy − azx − zy, yx − xz and zz with a ∈ K. If for some k ∈ N, 1 + a + . . . + ak = 0, then A ∉ Ω. On the
other hand, if 1 + a + . . . + ak ≠ 0 for each k ∈ N, then A ∈ Ω−. Furthermore, the algebras Aa for a ∈ K
are pairwise non-isomorphic.
Proof. The isomorphism statement is easy and routine. We have done similar verifications many times
now. The fact that A ∈ Ω− whenever A ∈ Ω follows from Lemma 1.13. It remains to prove that A ∈ Ω
if and only if 1 + a + . . . + ak ≠ 0 for each k ∈ N.
We start by a substitution, which turns the defining relations into a more convenient form. After
swapping x and z, the defining relations take the form xx, yz − zx and xy + axz − zy. Now we perform
an extra sub x → x, y → y − az and z → z, which shows that A is isomorphic to the algebra B given
by the generators x, y, z and the relations
xy − zy + azz, yz − zx − azz and xx.
(9.1)
We shall compute the Grobner basis of the ideal of relations of B with respect to the left-to-right
degree lexicographical ordering assuming x > y > z. For the sake of convenience, we denote:
r−1 = 0, r1 = 1 and rk = 1 + a + . . . + ak for k ∈ N.
We shall prove inductively the following statement:
(Gk) If rj ≠ 0 for 1 ⩽ j ⩽ k, then the complete list of the degree up to 2k + 3 elements of the reduced
xz2j +1 −
Grobner basis of the ideal of relations of B consists of yz − zx − azz, xz2j x, xz2j y −
arj−1
rj
1
rj
z2j +1y + a
rj
z2j +2, xz2j +1y − axz2j +2 and xz2j +1x +
aj (a−1)
rj
xz2j +2 −
1
rj
z2j +2x for 0 ⩽ j ⩽ k.
We start with the basis of induction. The overlap xxx resolves (produces no Grobner basis elements
of degree 3). The overlap xyz produces xzx+(a−1)xz2j +2−z2j +2x, while the only remaining overlap xxy
the ideal of relations of B consists of the defining relations together with xzx+(a−1)xz2j +2 −z2j +2x and
yields xzy −axzz. Thus the complete list of the degree up to 3 elements of the reduced Grobner basis of
xzy − axzz. This is exactly the statement (G0). Now assume that (Gk) holds. We shall verify (Gk+1).
According to (Gk), we already know the members of the basis of degrees up to 2k + 3. The only degree
2k + 4 overlaps are xxz2k+1x, xxz2k+1y, xz2k+1xx, xz2k+1xy and xz2k+1yz. Each of xxz2k+1x, xz2k+1xx
and xz2k+1yz produce the same element xz2k+2x of the reduced Grobner basis, while xxz2k+1y resolves.
Finally, xz2k+1xy produces
rk+1xz2k+2y − arkxz2k+3
− z2k+3y + az2k+4.
(9.2)
54
1
ark
rk+1
xz2k+3 −
Provided rk+1 ≠ 0, we get xz2k+2y −
z2k+4. Now at degree 2k +4 we have just
two members of the reduced Grobner basis: xz2k+2x and xz2k+2y −
z2k+4.
The only degree 2k + 5 overlaps are xxz2k+2x, xxz2k+2y, xz2k+2xx, xz2k+2xy and xz2k+2yz. Now
xxz2k+2x and xz2k+2xx resolve, both xxz2k+2y and xz2k+2xy produce xz2k+3y −axz2k+4, while xz2k+2yz
z2k+4x. This proves (Gk+1) and concludes the inductive proof
produces xz2k+3x +
of (Gj ).
z2k+3y + a
rk+1
xz2k+4 − 1
rk+1
z2k+3y + a
rk+1
xz2k+3 −
ak+1(a−1)
ark
rk+1
rk+1
rk+1
rk+1
1
2
Now assume that 1 + a + . . . + ak ≠ 0 for each k ∈ N (that is, rk ≠ 0 for k ∈ N). Then the assumption of
each (Gk) is satisfied and (Gk) produce the complete reduced Grobner basis of the ideal of relations
of B. The set of its leading monomials consists of yz, xzkx and xzky with k ∈ Z+. As a result, the
corresponding normal words are zmyn and zmynxzj for k, m, j ∈ Z+. One easily sees that the number
of these words of degree k is (k+1)(k+2)
and therefore HB = HA =(1 − t)−3. Hence A ∈ Ω.
It remains to deal with the case when rj = 0 for some j ∈ N. Then we can pick the minimal k ∈ Z+
such that rk+1 = 0. Then the assumptions of (Gk) are satisfied. By (Gk) the complete list of the degree
up to 2k +3 elements of the reduced Grobner basis of the ideal of relations of B consists of yz −zx −azz,
z2j +2x
xz2j x, xz2j y −
for 0 ⩽ j ⩽ k. We already know that the degree 2k + 4 part of the basis consists of xz2k+2x and (9.2).
Using the equation rk+1 = 0, we can (up to scaling) rewrite (9.2) as xz2k+3 −z2k+3x +az2k+4. Continuing
the process, we obtain the degrees 2k + 5, 2k + 6 and 2k + 7 parts of the Grobner basis, which comprise
z2k+3(yy − azy − azx) and zxz2k+2yy − a2zxz2k+4 − az2k+4yx + a2z2k+5x (one element in each of degrees
2k + 5 and 2k + 6 and nothing of degree 2k + 7). Now it is easy to verify that dim A2k+7 = (2k+8)(2k+9)
and therefore A ∉ Ω.
z2j +2, xz2j +1y − axz2j +2 and xz2j +1x +
z2j +1y + a
rj
xz2j +1 −
xz2j +2 −
aj (a−1)
arj−1
+ 1
1
rj
1
rj
rj
rj
2
Lemma 9.2. Let A = A(V, R) ∈ Ω be a quadratic algebra such that dim V = dim R = 3 and with respect
to some basis x, y, z in V , zz ∈ R and the quadratic algebra B = A~I with I being the ideal generated
by z, is given by the relations yx and xy. Then A is isomorphic to to a K-algebra given by generators
x, y, z and three quadratic relations from (R17–R19) of Theorem 1.11. Furthermore, the algebras in
(R17–R19) belong to Ω− and are pairwise non-isomorphic.
Proof. Algebras in (R19) are pairwise non-isomorphic by Lemma 9.1. The two algebras from (R17)
and (R18) are easily seen to be non-isomorphic to each other and to any of the algebras from (R19).
Thus we can forget about isomorphisms and concentrate on A.
The assumptions imply that R is spanned by zz, xy + f and yx + g with f, g ∈ span{xz, zx, yz, zy}.
Using a substitution x → x +sz, y → y +tz, z → z with appropriate s, t ∈ K, we can kill the xz coefficient
of f and the yz coefficient of g. Then A is given by the generators x, y, z and the relations
xy − pyz − azx − bzy, yx − qxz − czx − dzy and zz,
(9.3)
where a, b, c, d, p, q ∈ K.
Note also that Aopp is isomorphic to the quadratic algebra C given by the generators x, y, z and the
relations yx − pzy − axz − byz, xy − qzx − cxz − dyz and zz. After the sub x → x + bz, y → y + cz and
z → z, the relations of C take the shape xy − dyz − qzx + bzy, yx − axz + czx − pzy and zz. Hence
Aopp is isomorphic to an algebra given by (9.3) with the
parameters (q, −b, −c, p, d, a) in place of (a, b, c, d, p, q)
(9.4)
Throughout the proof we again use the order (8.1) on x, y, z monomials. Resolving the overlaps
yxy and xyx, we see that the degree 3 part of the Grobner basis of the ideal of relations of A consists
of two members
g1 = pyyz−qxzy+ayzx+byzy−dzyy−pczyz and g2 = qxxz+cxzx+dxzy−pyzx−azxx−qbzxz.
(9.5)
Case 1: pq ≠ 0.
55
Scaling x and y, we can turn p and q into 1: p = q = 1. One easily sees that dim A3 = 10 regardless
what the values of other parameters are. The leading monomials of g1 and g2 are yyz and xxz
respectively. The degree 4 part of the Grobner basis of the ideal of relations is spanned by 2 elements
r1 and r2 arising from the overlaps xxzz and yyzz (other overlaps resolve). The explicit formulae for
rj (we assume p = q = 1) are as follows:
r1=−cxzxz−dxzyz+yzxz−aczxzx−adzxzy+azyzx; r2=xzyz−ayzxz−byzyz+dzxzy−adzyzx−bdzyzy.
Note that there are exactly 16 degree 4 monomials which do not contain the leading terms yx, xy,
zz, xxz and yyz of the members of the Grobner basis of degree up to 3. Since A ∈ Ω, we must have
dim A4 = 15. This happens precisely when the dimension of the space L spanned by r1 and r2 is exactly
16 − 15 = 1. It is easy to observe that r1 and r2 are linearly independent if b ≠ 0 or c ≠ 0. This yields
and we end up with two algebras from (R17) and (R18). After checking that in the case p = q = 1,
b = c = 0. Now L has dimension 1 precisely when the vectors (−d, 1, −ad, a) and (1, −a, d, −ad) are
proportional. The latter occurs exactly when a2 = d2 = ad = 1. That is, (a, d) =(1, 1) or (a, d) =(−1, 1)
b = c = 0 and (a, d) ∈ {(1, 1),(−1, −1)}, the Grobner basis of the ideal of relations of A is finite and
and confirm that HA = (1 − t)−3. Thus, indeed, A ∈ Ω for A from (R17) and (R18). By Lemma 1.13,
consists of defining relations, g1, g2 and r2, one gets the complete set of leading monomials of the
members of the basis: xy, yx, zz, xxz, yyz and xzyz. This allows to compute the Hilbert series of A
A ∈ Ω− for A from (R17) and (R18).
Case 2: ad ≠ 0.
According to (9.4), Aopp falls under the jurisdiction of Case 1. Since both algebras in (R17) and
(R18) are isomorphic to their own opposites, in this case we again get only algebras from (R17) and
(R18).
Case 3: ad = pq = 0.
There are options here. We have either a = p = 0, or q = d = 0, or p = d = 0, or q = a = 0.
Case 3a: a = p = 0.
In this case the defining relations of A are xy − bzy, yx − qxz − czx − dzy and zz, while g1 and g2
take the form g1 = −qxzy + byzy − dzyy and g2 = qxxz + cxzx + dxzy − qbzxz. The leading monomials
of g1 and g2 now are xzy and xxz respectively. The degree 4 part of the Grobner basis of the ideal of
relations is spanned by 3 elements r1, r2 and r3 arising from the overlaps xxzz, xzyx and xxzy (other
overlaps resolve). The explicit formulae for rj (we assume q = 1) are as follows:
r1=−cxzxz−bdyzyz+d2zyyz; r2 =−xzxz+byzxz−dczyzx−d2zyzy; r3=byyzy−dyzyy−(d+bc)zyzy.
Note that there are exactly 17 degree 4 monomials which do not contain the leading terms yx, xy,
zz, xzy and xxz of the members of the Grobner basis of degree up to 3. Since A ∈ Ω, we must have
dim A4 = 15. This happens precisely when the dimension of the space L spanned by r1, r2 and r3 is
exactly 17 − 15 = 2. One easily sees that r1, r2 and r3 are linearly independent provided d ≠ 0. Thus
d = 0. Again, it is easily observed that r1, r2 and r3 are linearly independent if bc ≠ 0. Hence bc = 0.
Next, if b = d = 0, then L is one-dimensional. Hence b ≠ 0. The only option left is c = d = 0 and b ≠ 0.
y
Now the defining relations of A are xy − bzy, yx − xz and zz with b ∈ K∗. After the sub x → x, y →
b
z
and z →
b , the space of defining relations is spanned by xy − zy, yx − xz and zz and we have an algebra
from (R19) corresponding to the parameter a equal 0.
It remains to deal with the case q = 0. That is, a = p = q = 0. Then the defining relations of A are
xy − bzy, yx − czx − dzy and zz, while g1 and g2 take the form g1 = byzy − dzyy and g2 = cxzx + dxzy. If
bc ≠ 0, we can turn b and c into 1 by scaling the variables. The leading monomials of g1 and g2 are now
xzx and yzy. The degree 4 part of the Grobner basis now is spanned by dxzyy and d(zyzx + dzyzy).
If d = 0, this yields dim A4 = 16, while if d ≠ 0, we have dim A4 = 14. Since both are incompatible with
A ∈ Ω, we must have bc = 0. If b = 0 and c ≠ 0, we turn c into 1 by scaling. Then the defining relations
of A are xy, yx − zx − dzy and zz, while g1 and g2 take the form g1 = −dzyy and g2 = xzx + dxzy. If
56
d = 0, we have dim A3 = 11. Hence d ≠ 0. The degree 4 part of the Grobner basis now is spanned by
xzyy and zyzx + dzyzy yielding dim A4 = 14. Since A ∈ Ω, this can not occur. If b ≠ 0 and c = 0, we
turn b into 1 by scaling. The defining relations of A are xy − zy, yx − dzy and zz, while g1 and g2
take the form g1 = yzy − dzyy and g2 = dxzy. Again dim A3 = 11 if d = 0. Hence d ≠ 0. Computing
the degree 4 part of the Grobner basis, we get dim A4 = 16, contradicting A ∈ Ω. Finally, if b = c = 0,
then the defining relations of A are xy, yx − dzy and zz, while g1 and g2 take the form g1 = −dzyy and
g2 = dxzy. If d = 0, dim A3 = 12, while if d ≠ 0, then dim A4 = 16. Since both are incompatible with
A ∈ Ω, we conclude that the case q = 0 does not occur.
Case 3b: q = d = 0.
This case is obtained from Case 3a by the sub swapping x and y. Thus no new algebras occur here.
Case 3c: p = d = 0.
If q = 0 or a = 0, we fall into the already considered Cases 3a or 3b. Thus we can assume that
qa ≠ 0. By scaling, we can turn q into 1 (can not normalize a at the same time). In this case the
defining relations of A are xy − azx − bzy, yx − xz − czx and zz with a ≠ 0, while g1 and g2 take the
form g1 = −xzy + ayzx + byzy and g2 = xxz + cxzx − azxx − bzxz. The leading monomials of g1 and g2
now are xzy and xxz respectively. The degree 4 part of the Grobner basis of the ideal of relations is
spanned by 3 elements r1, r2 and r3 arising from the overlaps xxzz, yxxz and xzyx (other overlaps
resolve). The explicit formulae for rj are as follows:
r1 =c(xzxz+azxzx); r2=ayzxx−xzxz+byzxz; r3 =ayyzx+byyzy−aczyzx−bczyzy.
Note that there are exactly 17 degree 4 monomials which do not contain the leading terms yx, yy,
zz, xzy and xxz of the members of the Grobner basis of degree up to 3. Since A ∈ Ω, we must have
dim A4 = 15. This happens precisely when the dimension of the space L spanned by r1, r2 and r3 is
exactly 17 − 15 = 2. That is, the rank of the matrix S of the coefficients of rj must be 2:
rk S = 2, where S =⎛⎜⎝
0 0 0
0 0 a −1 b
a b 0
0
0
0 ac
0
0
0 −ac −bc
0
0
c
0
⎞⎟⎠ .
Recall that a ≠ 0. If c ≠ 0, then rk S = 3. This leaves only the case c = 0 in which the rank of S
is indeed 2. Then the defining relations of A are xy − azx − bzy, yx − xz and zz. If b = 0, a direct
Grobner basis computation shows that dim A5 = 22 for generic a. By Lemma 2.7, dim A5 ⩾ 22 for all
a, contradicting A ∈ Ω. Hence b ≠ 0. An extra scaling turns b into 1. Then the defining relations of A
are xy − azx − zy, yx − xz and zz with a ∈ K∗. By Lemma 9.1, we fall into (R19) and A ∈ Ω−.
Case 3d: q = a = 0.
This case is obtained from Case 3c by swapping x and y. Thus no new algebras occur here. The
proof is now complete.
9.1 Case R0 = span {xx, yy}
Lemma 9.3. Let A = A(V, R) ∈ Ω be a quadratic algebra such that dim V = dim R = 3 and with
respect to some basis x, y, z in V , zz ∈ R and the quadratic algebra B = A~I with I being the ideal
generated by z, is given by the relations xx and yy. Then A is isomorphic to to a K-algebra given
by generators x, y, z and three quadratic relations from (R20) of Theorem 1.11. Furthermore, the
isomorphism conditions in (R20) are satisfied and all algebras in (R20) belong to Ω−.
Proof. The assumptions imply that R is spanned by zz, xx+f and yy+g with f, g ∈ span{xz, zx, yz, zy}.
Using a substitution x → x +sz, y → y +tz, z → z with appropriate s, t ∈ K, we can kill the xz coefficient
of f and the yz coefficient of g. Then A is given by the generators x, y, z and the relations
xx − pyz − azx − bzy, yy − qxz − czx − dzy and zz,
(9.6)
57
where a, b, c, d, p, q ∈ K.
Note also that Aopp is isomorphic to the quadratic algebra C given by the generators x, y, z and the
relations xx − pzy − axz − byz, yy − qzx − cxz − dyz and zz. After the sub x → x + az, y → y + dz and
z → z, the relations of C take the shape xx − byz + azx − pzy, yy − cxz − qzx + dzy and zz. Hence
Aopp is isomorphic to an algebra given by (9.6) with the
parameters (−a, p, q, −d, b, c) in place of (a, b, c, d, p, q)
(9.7)
Another useful observation is that swapping x and y preserves the shape of the relations and changes
the parameters by (a, b, c, d, p, q) →(d, c, b, a, q, p)
Throughout the proof we again use the order (8.1) on x, y, z monomials. Resolving the overlaps
xxx and yyy, we see that the degree 3 part of the Grobner basis of the ideal of relations of A consists
of two members:
g1 = pxyz+axzx+bxzy−pyzx−bzyx−pazyz; g2 = qyxz−qxzy+cyzx+dyzy−czxy−qdzxz.
(9.8)
Case 1: pq ≠ 0.
By a scaling, we can turn p and q into −1: p = q = −1. One easily sees that dim A3 = 10 regardless
what the values of other parameters are. The leading monomials of g1 and g2 are xyz and yxz
respectively. The degree 4 part of the Grobner basis of the ideal of relations is spanned by 2 elements
r1 and r2 arising from the overlaps xyzz and yxzz (other overlaps resolve). The explicit formulae for
rj (we assume p = q = 1) are as follows:
r1 =axzxz+bxzyz+yzxz−bzxzy−bczyzx−bdzyzy; r2 =xzyz+cyzxz+dyzyz−aczxzx−bczxzy−czyzx.
Note that there are exactly 16 degree 4 monomials which do not contain the leading terms xx, yy,
zz, xyz and yxz of the members of the Grobner basis of degree up to 3. Since A ∈ Ω, we must have
dim A4 = 15. This happens precisely when the dimension of the space L spanned by r1 and r2 is
exactly 16 − 15 = 1. It is easy to observe that r1 and r2 are linearly independent if a ≠ 0 or d ≠ 0.
This yields a = d = 0. Now L has dimension 1 precisely when the vectors (b, 1, b, bc) and (1, c, bc, c)
are proportional. The latter occurs exactly when bc = 1. That is, b = −α and c = − 1
α for some α ∈ K∗.
α zx and zz with α ∈ K∗ landing
Then the defining relations of A take the form xx + yz + αzy, yy + xz +
us into (R20). Swapping of x and y provides an isomorphism between two such algebra parameters
α and 1
α . Observing that the shape (9.6) of relations is preserved only by scaling of the variables
and scaling combined with swapping of x and y. This allows to easily verify that there are no other
isomorphic algebras in (R20) than indicated in Theorem 1.11. Next, as soon as we have gotten into
(R20) the Grobner basis of the ideal of relations of A became finite consisting of defining relations,
g1, g2 and r2. The complete set of leading monomials of the members of the basis is: xx, yy, zz, xyz,
1
yxz and xzyz. This allows to compute the Hilbert series of A and confirm that HA =(1 − t)−3. Thus
A ∈ Ω. By Lemma 1.13, A ∈ Ω−.
Case 2: bc ≠ 0.
According to (9.7), Aopp falls under the jurisdiction of Case 1. Since the class (R20) is closed under
passing to the opposite multiplication, in this case we again get only algebras from (R20).
Case 3: b = p = 0 or c = q = 0.
In this case Lemma 2.11 yields A ∉ Ω. This contradiction shows that the case does not occur. What
is left to consider are the cases p = c = 0 and q = b = 0.
Case 4: q = b = 0.
If p = 0 or c = 0, we fall into the already considered Cases 3. Thus we can assume that pc ≠ 0. By
scaling, we can turn p and c into 1. In this case the defining relations of A are xx−yz −azx, yy −zx−dzy
and zz, while g1 and g2 take the form g1 = xyz + axzx − yzx − azyz and g2 = yzx + dyzy − zxy. The
leading monomials of g1 and g2 now are xyz and yzx respectively. The degree 4 part of the Grobner
58
basis of the ideal of relations is spanned by 3 elements r1, r2 and r3 arising from the overlaps xyzz,
yyzx and yzxx (other overlaps resolve). The explicit formulae for rj are as follows:
r1=axzxz+dyzyz+azxzx+dzyzy; r2=zxzx−d2zyzy; r3=−zxyx+dyzyx+yzyz.
Note that there are exactly 17 degree 4 monomials which do not contain the leading terms xx, yy,
zz, xyz and yzx of the members of the Grobner basis of degree up to 3. Since A ∈ Ω, we must have
dim A4 = 15. This happens precisely when the dimension of the space L spanned by r1, r2 and r3 is
exactly 17 − 15 = 2. That is, the rank of the matrix S of the coefficients of rj must be 2:
rk S = 2, where S =⎛⎜⎝
−1 d 0 1 0
0
0
0
0 0 0 1 −d2
0 a d a
d
⎞⎟⎠ .
If d ≠ 0, then rk S = 3. Hence d = 0. Now if a ≠ 0, then rk S = 3. Hence a = 0. Then the defining
relations of A are xx − yz, yy − zx and zz. A direct Grobner basis computation shows that dim A6 = 31,
which is incompatible with A ∈ Ω. Hence the case does not occur.
Case 5: p = c = 0.
This case is obtained from Case 4 by the sub swapping x and y.
9.2 Case R0 = span {xy − yx − yy}
Lemma 9.4. Let A = Aa,b,p,q be the quadratic algebra given by the generators x, y, z and the relations
xy − yx − yy − azx − bzy, xz − pzx − qzy and zz with a, b, p, q ∈ K, while B = Ba,b,q be the quadratic
algebra given by the generators x, y, z and the relations xy − yx − yy − azx − bzy, zx − qyz and zz with
a, b, q ∈ K. Then A ∈ Ω if and only if A ∈ Ω− if and only if q + np ≠ 0 for every n ∈ Z+. Similarly,
B ∈ Ω if and only if B ∈ Ω− if and only if q ≠ 0. Furthermore, Aa,b,p,q and Aa′,b′,p′,q′
are isomorphic
are isomorphic if and only if
if and only if (a′, b′, p′, q′) = (ta, tb, p, q) with t ∈ K∗, Ba,b,q and Ba′,b′,q′
(a′, b′, q′) =(ta, tb, q) with t ∈ K∗ and none of algebras A is isomorphic to an algebra B.
Proof. Any isomorphism between two of algebras A, two of algebras B or an algebra A and an algebra
B must send z to its scalar multiple (indeed zz is the only square in the space of defining relations
up to a scalar multiple). The same isomorphism, after z is factored out, must preserve xy − yx − yy
up to a scalar multiple. Thus the substitution facilitating such an isomorphism must be of the form
x → ux + αy + βz, y → uy + γz, z → vz with u, v ∈ K∗ and α, β, γ ∈ K. Now the isomorphism statements
of the lemma are easily verified.
It remains to deal with the Hilbert series statement. First, we deal with algebras A. We use the
left-to-right degree lexicographical ordering assuming x > y > z. If k ∈ Z+ and q + jp ≠ 0 for 0 ⩽ j < k,
then we can easily compute the members of the reduced Grobner basis of the ideal of relations of A
of degrees up to k + 3. Indeed, the only degree 3 overlap other than zzz (this one resolves producing
nothing) is xzz. It resolves producing the member qzyz of the ideal of relations. If k > 0, then q ≠ 0
and zyz comprises the degree 3 part of the said basis. The only degree 4 overlaps are zzyz, zyzz and
xzyz. The first two resolve to 0, while the last produces (q + p)zyyz. If k > 1, then q + p ≠ 0 and zyyz
comprises the degree 4 part of the Grobner basis. Proceeding in the same manner, we find that the
the members of the reduced Grobner basis of the ideal of relations of A of degrees up to k + 2 are the
defining relations together with zyz, zyyz, . . . , zykz. If q + kp ≠ 0, the only degree k + 3 member is
zyk+1z, while if q + kp = 0, then there are no degree k + 3 members.
Now observe that the only monomials, which do not contain any of xy, xz and zyjz for j ∈ Z+ as
submonomials are ymxn and yjzymxn with j, m, n ∈ Z+. The number of such words of degree n is
exactly (n+1)(n+2)
hand, if k is the smallest non-negative integer satisfying q + kp = 0, then dim Ak+3 = (k+4)(k+5)
. Thus if q + kp ≠ 0 for all k ∈ Z+, then HA = (1 − t)−3 and A ∈ Ω. On the other
+ 1 (the
2
2
59
preceding dimensions match that of an algebra from Ω) and therefore A ∉ Ω. Since zz is a relation of
A, Lemma 1.13 yields that A ∈ Ω− provided A ∈ Ω.
Now we deal with algebras B. One way is to repeat the above process using the right-to-left degree
If q = 0, then B ∉ Ω according
lexicographical ordering instead. However, there is a quicker way.
to Lemma 2.11. Thus we are left with showing that B ∈ Ω provided q ≠ 0 (non-Koszulity and
membership on Ω− are dealt with in exactly the same way as with algebras A). Clearly, B with the
opposite multiplication Bopp is isomorphic to the quadratic algebra C given by the generators x, y, z
and the relations yx − xy − yy − axz − byz, xz − qzy and zz with a, b ∈ K, q ∈ K∗ (each monomial in
the defining relations of B is reversed). After the sub x → x, y → −y and z → z, the relations of C
take the form xy − yx − yy − axz + byz, xz + qzy and zz. We follow up with the sub x → x +(b − a)z,
y → y + az, z → z, which turns the defining relations of C into xy − yx − yy − azx +(b − 2a)zy, xz + qzy
and zz. These are the defining relations of one of the algebras A, which we have already dealt with.
By the criterion of the PBWS condition for algebras A, verified in the first part of the proof, C ∈ Ω.
Since B and C share the Hilbert series, B ∈ Ω.
Lemma 9.5. Let A = A(V, R) ∈ Ω be a quadratic algebra such that dim V = dim R = 3 and with respect
to some basis x, y, z in V , zz ∈ R and the quadratic algebra B = A~I with I being the ideal generated
by z, is given by one relation xy − yx − yy. Then A is isomorphic to to a K-algebra given by generators
x, y, z and three quadratic relations from (R21–R26) of Theorem 1.11. Furthermore, the algebras in
(R17–R19) belong to Ω− and are pairwise non-isomorphic.
Proof. The fact that algebras in (R17–R19) belong to Ω− and are pairwise non-isomorphic follows
directly from Lemma 9.4. Thus we can concentrate on the algebras A. The assumptions imply that
R is spanned by zz, xy − yx − yy + f and g with f, g ∈ span{xz, zx, yz, zy}, g ≠ 0.
Case 1: xz features in g with non-zero coefficient.
A sub x → x + uy, y → y, z → z with an appropriately chosen u ∈ K kills the yz coefficient in g.
After doing this, we perform a sub x → x + vz, y → y + wz with appropriately chosen v, w ∈ K to kill
the xz and yz coefficients in f . Now the defining relations of A take the form xy − yx − yy − azx − bzy,
xz − pzx − qzy and zz, where a, b, p, q ∈ K. If a ≠ 0, Lemma 9.4 implies that A is isomorphic to an
algebra given by relations of the same shape only with a = 1. By Lemma 9.4, A is now isomorphic to
an algebra from (R21). If a = 0 and b ≠ 0, same argument shows that A is isomorphic to an algebra
from (R22). Similarly, if a = b = 0, A is isomorphic to an algebra from (R23).
Case 2: xz does not feature in g.
If zx does not feature in g as well, Lemma 2.11 yields that A ∉ Ω, providing a contradiction. Thus
zx does feature in g with non-zero coefficient. A sub x → x + uy, y → y, z → z with an appropriately
chosen u ∈ K kills the zy coefficient in g. Now if yz does not feature in g, Lemma 2.11 kicks in again
furnishing a contradiction. Hence yz features in g with non-zero coefficient. After an appropriate
scaling, the defining relations of A take the form xy − yx − yy − azx − bzy, zx − pyz and zz, where
p, a, b ∈ K, p ≠ 0. If a ≠ 0, Lemma 9.4 ensures that A is isomorphic to an algebra from (R24). Same
lemma implies that A is isomorphic to an algebra from (R25) if a = 0 and b ≠ 0 and to an algebra from
(R29) if a = b = 0. The proof is now complete.
9.3 Case R0 = span {xy − αyx} with α ≠ 0, α ≠ 1
Lemma 9.6. Let A be the quadratic algebra given by the generators x, y, z and the relations xy − αyx −
azx − bzy, sxz + pyz + qzx + rzy and zz with a, b, s, p, q, r ∈ K and α ∈ K∗. Then A ∈ Ω if and only if
A ∈ Ω− if and only if sr − αnpq ≠ 0 for all n ∈ Z+.
Proof. The proof is exactly the same as the proof in Lemma 9.4 of the condition for membership of A
(from the said lemma) in Ω. The only difference is that the coefficient popping up in front of zyk+1z
when the overlap xzykz is considered is now sr − αkpq (instead of q + kp of Lemma 9.4).
60
Aη and Aη′
Lemma 9.7. Let A = Aη be the quadratic algebra given by the generators x, y, z and the relations
xy − αyx − azx − bzy, sxz + pyz + qzx + rzy and zz with η = (α, a, b, s, p, q, r) ∈ K7, α ∉ {0, 1}. Then
generated by the involution (α, a, b, s, p, q, r) ↦ 1
α , p, s, r, q and the maps (α, a, b, s, p, q, r) ↦
α , − b
(α, t1a, t2b, t1t3s, t2t3p, t1t3q, t2t3r) with t1, t2, t3 ∈ K∗.
are isomorphic if and only if η and η′ belong to the same orbit of the group action
α , − a
α , − a
Proof. Any isomorphism between two algebras A must send z to its scalar multiple (zz is the only
square in the space of defining relations up to a scalar multiple). The same isomorphism, after z
is factored our, must send xy − αyx to a scalar multiple of an expression of the same shape (with
different α perhaps). Thus the substitution facilitating such an isomorphism must be of the form
x → ux + αz, y → vy + βz, z → wz with u, v, w ∈ K∗ and α, β ∈ K or the same composed with
swapping of x and y. The swapping yields the transformation of parameters (α, a, b, s, p, q, r) ↦
α , p, s, r, q. As for the first collection of subs, exactly as in Lemma 9.11, one must have
1
α , − b
α = β = 0. This leaves us with scaling only, resulting in transformations of parameters of the form
(α, a, b, s, p, q, r) ↦(α, t1a, t2b, t1t3s, t2t3p, t1t3q, t2t3r) with t1, t2, t3 ∈ K∗. The result follows.
Lemma 9.8. Let A = A(V, R) ∈ Ω be a quadratic algebra such that dim V = dim R = 3 and with
respect to some basis x, y, z in V , zz ∈ R and the quadratic algebra B = A~I with I being the ideal
generated by z, is given by one relation xy − αyx with α ∈ K∗, α ≠ 1. Then A is isomorphic to to a
K-algebra given by generators x, y, z and three quadratic relations from (R27–R33) of Theorem 1.11.
Furthermore, algebras with different labels are non-isomorphic, the isomorphism conditions in (R27–
R33) are satisfied and all algebras in (R27–R33) belong to Ω−.
Proof. The fact that algebras with different labels from (R27–R33) are non-isomorphic, the isomor-
phism conditions in (R27–R33) are satisfied and all algebras in (R27-R33) belong to Ω− follows from
Lemmas 9.6 and 9.7. Now we can focus on A. By the assumptions, R is spanned by zz, xy − αyx + f
and g with f, g ∈ span{xz, zx, yz, zy}, g ≠ 0.
A sub x → x+vz, y → y+wz, z → z with appropriately chosen v, w ∈ K to kill the xz and yz coefficients
in f . Now the defining relations of A take the form xy − αyx − azx − bzy, sxz + pyz + qzx + rzy and zz,
where α, a, b, s, p, q, r ∈ K, α ≠ 0, α ≠ 1. Using Lemmas 9.6 and 9.7 and considering options for possible
distribution of zeros among the parameters, one easily sees that A must be isomorphic to one of the
algebras from (R27–R33).
9.4 Case R0 = span {xy − yx}
Lemma 9.9. Let A = Aa,b be the quadratic algebra given by the generators x, y, z and the relations
xy − yx − azx − bzy, xz − zy and zz with a, b ∈ K. Then Aa,b and Aa′,b′
are isomorphic if and only if
(a′, b′) =(ta, tb) for some t ∈ K∗.
Proof. Any isomorphism between two algebras A must send z to its scalar multiple (zz is the only
square in the space of defining relations up to a scalar multiple). It should also keep xz − zy in the
space of relations. These two properties are satisfied only for subs of the form x → ux +αz, y → uy +βz,
z → vy with u, v ∈ K∗ and α, β ∈ K. After applying this sub to Aa,b, the space of defining relations
becomes spanned by xy − yx − tazx − tbzy, xz − zy and zz with t = v
u . The result follows.
Lemma 9.10. Let A = A(V, R) ∈ Ω be a quadratic algebra such that dim V = dim R = 3 and with respect
to some basis x, y, z in V , zz ∈ R and the quadratic algebra B = A~I with I being the ideal generated
by z, is given by one relation xy − yx. Then A is isomorphic to to a K-algebra given by generators
x, y, z and three quadratic relations from (R34–R36) of Theorem 1.11. Furthermore, algebras from
(R34–R36) belong to Ω− and are pairwise non-isomorphic.
Proof. Lemmas 9.6 and 9.9 imply that algebras from (R34–R36) belong to Ω− and are pairwise non-
isomorphic. Now we can focus on A. By the assumptions, R is spanned by zz, xy − yx + f and g with
61
f, g ∈ span{xz, zx, yz, zy}, g ≠ 0. A sub x → x + vz, y → y + wz with appropriately chosen v, w ∈ K to
kill the xz and yz coefficients in f . Now the defining relations of A take the form xy − yx − azx − bzy,
sxz + pyz + qzx + rzy and zz, where a, b, s, p, q, r ∈ K. By Lemma 9.6, we must have sr − pq ≠ 0. It
easily follows that there is a substitution of the form x → u1x + u2y + u3z, y → v1x + v2y + v3z and
z → wz, which preserves the overall shape of relations and turns the second one into xz − zy. Hence
the defining relations of A become xy −yx −azx −bzy, xz −zy and zz with a, b ∈ K. Now by Lemmas 9.6
and 9.9, A is isomorphic to an algebra from (R34) if a ≠ 0, A is isomorphic to an algebra from (R35)
if a = 0 and b ≠ 0 and A is isomorphic to the algebra (R36) if a = b = 0.
9.5 Case R0 = span {xy}
Lemma 9.11. Let A = Aa,b,s,p,q,r be the quadratic algebra given by the generators x, y, z and the
relations xy − azx − byz, sxz + pyz + qzx + rzy and zz with a, b, s, p, q, r ∈ K. Then Aa,b,s,p,q,r and
Aa′,b′,s′,p′,q′,r′
t1, t2, t3 ∈ K∗. That is, 6-tuples of parameters give rise to isomorphic algebras if they are in the same
are isomorphic if and only if (a′, b′, s′, p′, q′, r′) = (t1a, t2b, t1t3s, t2t3p, t1t3q, t2t3r) with
orbit of the action (a, b, s, p, q, r) ↦(t1a, t2b, t1t3s, t2t3p, t1t3q, t2t3r) of K∗
× K∗
× K∗.
Proof. Any isomorphism between two algebras A must send z to its scalar multiple (zz is the only
square in the space of defining relations up to a scalar multiple). The same isomorphism, after z is
factored our, must preserve xy up to a scalar multiple. Thus the substitution facilitating such an
isomorphism must be of the form x → ux + αz, y → vy + βz, z → wz with u, v, w ∈ K∗ and α, β ∈ K.
Taking into account the specifics of the first relation (absence of xz and zy), one sees that our sub
preserves the general shape of the space of relations only if α = β = 0. This leaves us with scaling only
and the verification of the desired description of isomorphic algebras A becomes trivial.
Lemma 9.12. Let A be the quadratic algebra given by the generators x, y, z and the relations xy −
azx − byz, sxz + pyz + qzx + rzy and zz with a, b, s, p, q, r ∈ K. Then A ∈ Ω if and only if A ∈ Ω− if and
only if sr ≠ 0 and sr − pq ≠ 0.
Proof. First, we eliminate the case s = r = 0. In this case Lemma 2.11 implies that A ∉ Ω if pq = 0.
Thus we can assume that s = r = 0 and pq ≠ 0. After an appropriate scaling of the variables, the space
R of defining relations of A is spanned either by xy, yz − zx and zz or by xy − zx, yz − zx and zz.
In both cases an easy Grobner basis calculation yields dim A4 = 17, which is incompatible with the
membership in Ω. Thus A ∉ Ω if s = r = 0.
The two cases s ≠ 0 and r ≠ 0 are reduced to one another by passing to the opposite multiplication
followed up by swapping of x and y: this procedure results in an algebra with relations of the same
the rest of the proof we can assume that s ≠ 0. We use the left-to-right degree lexicographical ordering
assuming x > y > z. The only degree 3 overlaps of the leading monomials are zzz and xzz. The first
shape, parameters being transformed according to the rule (a, b, s, p, q, r) ↦ (b, a, r, q, p, s). Thus for
resolves, while the second results in (sr − pq)zyz. If sr − pq = 0, we have dim A3 = 11 and therefore
A ∉ Ω. From now on, we assume sr − pq ≠ 0. In this case the degree 3 part of the reduced Grobner
basis of the ideal of relations of A consists of zyz. The only degree 4 overlaps are zzyz, zyzz and
xzyz. The first two resolve, while the third produces rzyyz. If r = 0 ⇐⇒ sr = 0, dim A4 = 16 and
therefore A ∉ Ω. It remains to show that A ∈ Ω provided sr ≠ 0 and sr − pq ≠ 0. In this case the degree
4 part of the Grobner basis consists of zyyz. Now we proceed inductively to show that for each k ∈ N,
the degree k + 2 part of the reduced Grobner basis of the ideal of relations of A consists of zykz. We
already have the basis of induction. Assume that k ⩾ 3 and the statement holds for all smaller k. Then
the only degree k + 2 overlaps are zzyk−1z, zyk−1zz and xzyk−1z. The first two resolve, while the last
is easily seen to produce zykz (we have to use r ≠ 0 here). Thus the reduced Grobner basis of the ideal
of relations of A consists of the defining relations together with zykz for k ∈ N. The corresponding
normal words are ymxn and yjzymxn with j, m, n ∈ Z+.
It easily follows that HA = (1 − t)−3: the
62
number of normal words of degree n is (n+1)(n+2)
completes the proof.
2
. By Lemma 1.13, A ∈ Ω−, whenever A ∈ Ω, which
Lemma 9.13. Let A = A(V, R) ∈ Ω be a quadratic algebra such that dim V = dim R = 3 and with respect
to some basis x, y, z in V , zz ∈ R and the quadratic algebra B = A~I with I being the ideal generated
by z, is given by one relation xy. Then A is isomorphic to a K-algebra given by generators x, y, z
and three quadratic relations from (R37–39) of Theorem 1.11. Furthermore, algebras from (R34–R36)
belong to Ω− and are pairwise non-isomorphic.
Proof. The fact that algebras with different labels from (R27–R33) are non-isomorphic, the isomor-
phism conditions in (R27–R33) are satisfied and all algebras in (R27–R33) belong to Ω− follows from
Lemmas 9.12 and 9.11. Now we can focus on A. The assumptions imply that R is spanned by zz,
xy + f and g with f, g ∈ span{xz, zx, yz, zy}, g ≠ 0. A sub x → x + vz, y → y + wz with appropriately
chosen v, w ∈ K kills the xz and zy coefficients in f . Now the defining relations of A take the form
xy − ayz − bzx, sxz + pyz + qzx + rzy and zz, where a, b, s, p, q, r ∈ K. By Lemma 9.12, sr ≠ 0 and
sr − pq ≠ 0.
By scaling we can turn s into 1.
Thus we fall into (R15). The inclusion A ∈ Ω−, non-Koszulity of A along with the isomorphism
statements follow from Lemmas 9.12 and 9.11. Usin the same lemmas and considering all options for
possible distribution of zeros among the parameters, one easily sees that A must be isomorphic to one
of the algebras from (R37–R39).
Note that now we have run out of algebras in the first part of Theorem 1.11, but we still have
unexplored options for R0. This is due to the fact that the latter provide no algebras from Ω.
9.6 Case R0 = span {yy}
generated by z, is given by one relation yy. Then A ∉ Ω.
Lemma 9.14. Let A = A(V, R) ∈ Ω be a quadratic algebra such that dim V = dim R = 3 and with
respect to some basis x, y, z in V , zz ∈ R and the quadratic algebra B = A~I with I being the ideal
Proof. The assumptions imply that R is spanned by zz, yy + f and g with f, g ∈ span{xz, zx, yz, zy},
g ≠ 0. If both xz and zx do not feature in g, the result follows from Lemma 2.11. Passing to the
opposite multiplication does not change membership in Ω and leads to an algebra of the same shape
with the xz and zx coefficients in g (among other things) swapped. Thus for the rest of the proof we
can without loss of generality assume that xz features in g with non-zero coefficient. A sub x → x + sy,
y → y, z → z with an appropriate s ∈ K kills the yz term in g. Subtracting g with an appropriate
coefficient from the first relation, we can assume that xz does not feature in f . Now if either zy does
not feature in g or zx does not feature in f , the result follows from Lemma 2.11. Thus we can assume
that both zy in g and zx in f have non-zero coefficients. By scaling, we can turn these coefficients
into −1. Thus the defining relations of A acquire the form yy − zx − azy, xz − bzx − zy and zz with
a, b ∈ K. We use the ordering (8.1) on the monomials. The members of the reduced Grobner basis of
the ideal of relations of A of degrees three and four are easily seen to be zyz, yzx + ayzy − zxy, zxyz
and zxzx + azxzy, which yields dim A4 = 16. Hence A ∉ Ω.
9.7 Case R0 = span {xx − yx, yy}
Lemma 9.15. Let A = A(V, R) ∈ Ω be a quadratic algebra such that dim V = dim R = 3 and with
respect to some basis x, y, z in V , zz ∈ R and the quadratic algebra B = A~I with I being the ideal
generated by z, is given by the relations xx − yx and yy. Then A ∉ Ω.
63
Proof. By the assumptions, R is spanned by zz, xx − yx + f and yy + g with f, g ∈ span{xz, zx, yz, zy}.
Using a substitution x → x + sz, y → y + tz, z → z with appropriate s, t ∈ K, we can kill the xz and zx
coefficients of f . Then A is given by the generators x, y, z and the relations
xx − yx − ayz − bzy, yy − pxz − qzx − cyz − dzy and zz,
(9.9)
where a, b, c, d, p, q ∈ K.
Then Aopp, being A with the opposite multiplication, is isomorphic to the algebra C given by the
generators x, y, z and the relations xx − xy − azy − byz, yy − pzx − qxz − czy − dyz and zz. After the
substitution x → x − y, y → −y and z → z, the defining relations of C take the shape xx − yx + byz + ayz,
yy − qxz − pzx +(q + d)yz +(p + c)zy and zz. That is,
Aopp is isomorphic to an algebra given by (9.9) with the
parameters (−b, −a, −q − d, −p − c, q, p) in place of (a, b, c, d, p, q)
(9.10)
Throughout the proof we again use the order (8.1) on x, y, z monomials. Resolving the overlaps yyy
and xxx, we see that the degree 3 part of the Grobner basis of the ideal of relations of A consists of
two members
g1 = pyxz−pxzy+qyzx+(d−c)yzy−qzxy+(qc−pd)zxz,
g2 = xyx+axyz−pxzx+bxzy−(a + c)yzx−byzy−(b+d+q)zyx−aqzxz−a(q+d)zyz.
(9.11)
If p = q = 0, the result follows from Lemma 2.11. Thus we can assume that either p or q is non-zero.
According to (9.10), the cases p ≠ 0 and q ≠ 0 are reduced to each other by passing to the opposite
multiplication. Thus we can assume that p ≠ 0. Via scaling, we can turn p into 1. The leading
monomials of g1 and g2 are now yxz and xyx. One easily sees that dim A3 = 10 regardless what the
values of other parameters are. The degree 4 part of the Grobner basis of the ideal of relations is
spanned by 3 elements r1, r2 and r3 arising from the overlaps yxzz, xyxz and xyxx (other overlaps
resolve). The explicit formulae for rj (we assume p = 1) are as follows:
r1=xzyz−qyzxz+(c−d)yzyz+qzxyz;
r2=qxyzx−(c−d)xyzy−qxzxy+qyzxy+(qc−d)xzxz−(b+c)xzyz+(a+d)yzxz
+(b−c(c−d))yzyz−qczxyz−q2zxzx+(b+q(c−d))zxzy−q(b+d)zyzx+((b+d)(c−d)−b)zyzy;
r3=−(a+c)xyzx−bxyzy−(q+b+d)xzyx+(q+a+b+d)yzyx−qaxzxz+a(1−d−q)xzyz+a(q+a+c)yzyz
+qazxyz+(b+q(a+c))zxzx+qbzxzy+(d(a+c)+b(c−1))zyzx+b(b+d)zyzy.
Note that there are exactly 17 degree 4 monomials which do not contain the leading terms xx, yy, zz,
yxz and xyx of the members of the Grobner basis of degree up to 3.
Assume the contrary. That is, A ∈ Ω. Then we must have dim A4 = 15. This happens precisely
when the dimension of the space L spanned by r1, r2 and r3 is exactly 17 − 15 = 2.
Case 1: q ≠ 0.
In this case yzxy features in r2 with non-zero coefficient and does not feature in each of r1 or r3.
Since dim L = 2, we now have that the span of r1 and r3 is one-dimensional. Now yzxz features in r3
with non-zero coefficient and does not feature in r1. Hence we must have r1 = 0. This only happens if
a = b = c = 0 and d = −q. Then the defining relations of A acquire the form xx − yx, yy − xz + dzx − dzy
and zz with d ∈ K. Using Grobner basis technique, one easily sees that in this case for generic d,
dim A5 = 22. By Lemma 2.7, dim A5 ⩾ 22 for all q, contradicting A ∈ Ω.
Case 2: q = 0.
In this case it is easy to observe that r1, r2 and r3 are linearly independent if b ≠ 0 or if b = 0 and
d ≠ 0 or if b = d = 0 and a ≠ 0 or if b = d = a = 0 and c ≠ 0. This only leases us the case a = b = c = d = q = 0
in which case r1 = r2 = 0 and L is one-dimensional. This contradiction completes the proof.
64
9.8 Case R0 = span {xx − αxy − yy, yx} with α ∈ K, α2 + 1 ≠ 0
This is the most technically annoying case. To add an insult to injury, it turns out that there are no
algebras from Ω in it. The plan of action is the following. We represent algebras in this case as a multi-
parametric family of quadratic algebras and observe that we always have dim A3 = 10. Using Grobner
basis technique, we identify the algebras of the family satisfying dim A4 = 15. These form several
one-parametric families of quadratic algebras. However, members of each of them satisfy dim A5 > 21,
which is incompatible with A ∈ Ω. We shall use the order (8.1) on the monomials. Anyone curious
can try to do the same using the degree-lexicographical ordering to see how much messier things get.
Lemma 9.16. Let A = A(V, R) be a quadratic algebra such that dim V = dim R = 3 and with respect
to some basis x, y, z in V , zz ∈ R and the quadratic algebra B = A~I with I being the ideal generated
by z, is given by the relations xy − αyx − yy and yy with α ∈ K and α2 + 1 ≠ 0. Then A ∉ Ω.
Proof. Assume the contrary: A ∈ Ω. The assumptions imply that R is spanned by zz, xy − αyx − yy + f
and yy + g with f, g ∈ span{xz, zx, yz, zy}. Using a linear substitution, which leaves z intact and
replaces x and y by x + sz and y + tz respectively with appropriate s, t ∈ K, we can kill the xz and yz
coefficients of f . After this substitution, A is given by the generators x, y, z and the relations
xx − αxy − yy − azx − bzy, yx − pxz − qyz − czx − dzy and zz,
(9.12)
where α, a, b, c, d, p, q ∈ K and α2 + 1 ≠ 0.
Throughout the proof we will use the order (8.1) on x, y, z monomials. Resolving the overlaps yxx
and xxx, we see that the degree 3 part of the Grobner basis of the ideal of relations of A consists of
two members
g1 = yyy−pxzx+αpxzy−(q−a)yzx+(αq+b)yzy+(αd−c)zyy−pdzxz−qdzyz,
g2 =(α2 +1)xyy−α(q+αp)xyz−(q+αp)yyz+(αp+a−αc)xzx+(b−α2p−αd)xzy+(αq−αa−c)yzx
−(α2q+αb+d)yzy−(α2d−αb−αc+a)zyy−p(αa−αd+b+c)zxz−(αpb−αqd+pd+qb)zyz.
One easily sees that dim A3 = 10 regardless what the values of the parameters are. There are 6 degree
4 overlaps xyyx, xyyy, yyyy, yyyx, xxyy, yxyy. The members of the ideal of relations obtained from
the last two overlaps always belong to the linear span of the ones obtained from the first four overlaps.
We denote these r1, r2, r3 and r4 respectively. The explicit formulae for rj are as follows:
r1=(α(q+αp)−(α2
+1)c)xyzx−(α2
+1)dxyzy+(q+αp)yyzx+α(αc−αp−a)xzxy+(αc−αp−a)xzyy+α(αa−αq+c)yzxy
+(αa−αq+c)yzyy+(α2p2
+p(b+αa−αd+c)zxzx+(pd+αpb+qb−αqd+ac−αbc−αc2
+αpd−pb−(α2
+α2cd)zyzx+(ad−αbd−αcd+α2d2)zyzy,
+1)pc−αqc+αpq+qa)xzxz−(α2
+1)pdxzyz+(α2pq+αpb+pd−αqa+αq2
−qc)yzxz
r2=(α2
+1)(a−q−αp)xyzx+(α(α2
+2)(q+αp)+(α2
+1)b)xyzy−(α2
+1)pyyzx+(α(α2
+2)p+q)yyzy+(αc−αp−a)xzxy
+(α2p+αd−b−(α2
−αp(α−αd+b+c)zxzx+(α2
+(α(α2
+2)pb−αqd+pd+qb−(αq+b)(α2d−αb−αc+a))zyzy,
+1)(c−αd))xzyy+(αa−αq+c)yzxy+(α2q+αb+d)yzyy−(α2
+1)pdxzxz−(α2
+1)qdxzyz
+1)p(α−αd+b+c)+((q−a)(α2d−αb−αc+a)−(α2
+1)pb)zyzx
r3=(q−a)yyzx−(αq+b)yyzy−pxzxy+αpxzyy−(q−a)yzxy+(αq−αd+b+c)yzyy+pdyzxz+qdyzyz+αpdzxzx
−(1+α2)pdzxzy+(pd+(q−a)(αd−c))zyzx−(αpd+(α2
+1)qd+αbd−αqc−bc)zyzy,
r4=−cyyzx−dyyzy+αpxzxy+pxzyy+α(q−a)yzxy+(q−a)yzyy−p(q+αp)xzxz+(aq−q2
−pc−αpq−pb)yzxz
−(pd+2q(b+αq))yzyz+pdzxzx+(qd+c2
−αcd)zyzx+d(c−αd)zyzy.
Note that there are exactly 18 degree 4 monomials which do not contain the leading terms xx, yx,
zz, xyy and yyy of the members of the Grobner basis of degree up to 3. Since we have assumed that
A ∈ Ω, we must have dim A4 = 15. This happens precisely when the dimension of the space L spanned
by r1, r2, r3, r4 is exactly 18 − 15 = 3.
Unfortunately, there are values of parameters for which precisely this happens. Obviously,
S ={(α, a, b, c, d, p, q) ∈ K7
∶ dim L ⩽ 3}
65
is an affine variety, while
S0 ={(α, a, b, c, d, p, q) ∈ K7
∶ dim L ⩽ 2}
and α2 + 1 ≠ 0. Note also that if we scale z, then the relations retain their form, α stays put, while the
is a subvariety of S. By the above remark dim A4 = 15, which only happens if (α, a, b, c, d, p, q) ∈ S ∖ S0
vector(a, b, c, d, p, q) is scaled by the same constant as z. This allows us to treat S and S0 as subvarieties
plus a finite set. In order to deal with the case (α, a, b, c, d, p, q) ∈ S ∖ S0, we have to go one step further
of K × KP 5, reducing the number of free parameters by 1. We shall see that S, as a subvariety of
K × KP 5 has dimension one and splits into the union of several irreducible one-dimensional varieties
in computing the Grobner basis to see that dim A5 > 21, which is incompatible with A ∈ Ω.
Now we shall sketch the procedure of pinpointing the variety S.
Case 1: d ≠ 0. By scaling z, we can without loss of generality assume that d = 1.
Case 1a: p ≠ 0 (in addition to d = 1).
Set κ = c + b + αa − α. The matrix of zxzx, zxzy and xzyz coefficients of r1 − κr4, r2 − κr3, r3, r4
now is
⎛⎜⎜⎜⎝
0
0
p
0
0
0
0 −(α2 + 1)p
−(α2 + 1)p
−(α2 + 1)q
0
0
⎞⎟⎟⎟⎠
.
Since p ≠ 0 and α2 + 1 ≠ 0, it follows that the only way for rj to be linearly dependent is for the
equation
ρ = −qr1 + pr2 + pκr3 + qκr4 = 0
(9.13)
to be satisfied. This happens precisely when the coefficients of ρ front of all 16 monomials featuring in
rj vanish, giving a system of algebraic equations on the parameters involved. Next, we observe that
ακq ≠ 0. Indeed, if κ = 0, then vanishing of the yyzy coefficient of ρ yields q = −α(α2 + 2)p. Plugging
this into the equation provided by vanishing of the yyzx coefficient of ρ, we get p2(α2 + 1)2 = 0, which
is impossible. Thus κ ≠ 0. If q = 0, the equation ρ = 0 reads r2 + κr3 = 0. The yzxz coefficient of
r2 + κr3 is κp. Thus κp = 0, which is impossible. Finally, if α = 0, it is an easy exercise to see that
rj are linearly independent (the matrix of coefficients of rj simplifies dramatically if α = 0). Thus we
First, we consider the case p = αq. In this case, the equation ρ = 0 resolves rather smoothly. looking
can assume that ακq ≠ 0. Equating the yzyz coefficients of ρ and 0, we get 2κq2(αq + b) = 0. Since q
at xyzy coefficient, we get (α2 + 1)q + αpq + α2(α2 + 2)p2 = 0. Together with p = αq, this yields
α(α2 +1) (we use the fact that α ≠ 0 here). Solving the equations arising from xyzx
α(α2 +1) .
q = −
and xzyy of ρ and using the above expressions for p and q, we find a = −1 −
Finally, b = −αq =
and κ are non-zero, we get b = −αq.
α2 and c = α −
1
α2(α2 +1) , p = −
α(α2 +1) . Summarising, we get
1
1
1
1
a = −1 −
1
α2 , b =
1
α(α2 +1) , c = α −
1
α(α2 +1) , d = 1, p = −
1
α(α2 +1) and q = −
1
α2(α2 +1) .
(9.14)
For these values of parameters rj span a 3-dimensional space.
Now we consider the case p ≠ αq. The equations arising from xzxy and yzxy coefficients of ρ read
(p − αq)(αc − a − p(κ + α)) = 0 and (p − αq)(c −(q − a)(κ + α)) = 0. Since p ≠ αq, we get
αc − a − p(κ + α) = c −(q − a)(κ + α) = 0.
Now the yyzy coefficient yields p(α(α2 + 2)p + q) − qκ = 0, from which we have κ = p(1 + α(α2 + 2) p
q).
The equation arising from xyzy reads q(α2 + 1) + αpq + α2(α2 + 2)p2 = 0. This implies q = −
α2(α2
+2)p2
αp+α2 +1 .
Plugging this into the above expression for κ and simplifying, we get κ = − α2
+1
(delightfully, p cancels
α
α = q. Hence p = αq and
out). Hence κ + α = −
we have arrived to a contradiction. This concludes Case 1a.
α. Plugging this into the above display, we get a − αc = p
1
66
Case 1b: p = 0 and q ≠ 0 (in addition to d = 1).
Considering the xyzy and xzyz coefficients of rj, we see that in this case r1 and r2 are linearly
independent modulo the span of r3 and r4. Since rj are linearly dependent, it follows that r3 and r4
must be linearly dependent. If, additionally, a ≠ q, then the 2 × 2 matrix of yzxz and yzyz coefficients
of r3 and r4 is non-degenerate providing a contradiction. Thus a = q. Now the 2 × 2 matrix of yyzx
and yzyz coefficients of r3 and r4 is non-degenerate unless c = 0. Thus c = 0. Finally, now the 2 × 2
matrix of yzyz and zyzx coefficients of r3 and r4 is non-degenerate unless q = 0. Thus q = 0, which is
a contradiction.
Case 1c: p = q = 0 (in addition to d = 1).
Now the matrix of coefficients of rj becomes rather simple and it is an elementary linear algebra
exercise to see that rj span a 3-dimensional space only when either a = c = b − α = 0 or b = a = c − α = 0
or b = c = α and a = α2 (recall that we have p = q = 0 and d = 1). This provides 3 curves sitting in S:
a = 0, b = α, c = 0, d = 1, p = 0, q = 0;
a = 0, b = 0, c = α, d = 1, p = 0, q = 0;
a = α2, b = α, c = α, d = 1, p = 0, q = 0.
(9.15)
(9.16)
(9.17)
This concludes Case 1, in which we have identified 4 one-dimensional irreducible components of S.
Case 2: d = 0 and p ≠ 0. By scaling z, we can without loss of generality assume that p = 1.
If αa + b + c ≠ 0, then by looking at zxzx and zxzy coefficients, we see that r1 and r2 are linearly
independent modulo span of r3 and r4. By looking at xzxy and xzyy coefficients, we see that r3 and
r4 are linearly independent. Hence rj span a 4-dimensional space. Thus we must have αa + b + c = 0.
That is, c = −αa − b.
Case 2a: q ≠ 0 (on top of d = 0, p = 1 and c = −αa − b).
First, consider the case b +αq ≠ 0. Looking at the 4 ×3 matrix of the yzxy, yzyy and yzyz coefficients
of rj, we see that in this case r1 and r4 are linearly independent modulo the span of r2 and r3. Thus
r2 and r3 must be linearly dependent. Considering the determinants of four of 2 × 2 submatrices of
the coefficients of r2 and r3, we get
(α2 + 1)(q − a) − α(α2 + 1) = α(α2 + 2)q +(α2 + 1)b + α2(α2 + 2)
=(q − a)(b + αq + α2 + 1) =((α2 + 1)a + αb + α)(q − a) −(b + αq) = 0.
One of the equations yields that either q = a or b + αq = −1 − α2. If q ≠ a, then b + αq = −1 − α2. Plugging
this into the rest of the equations, we easily see that the system is incompatible. Thus q = a. Then
the last equation yields b + αq = 0, which contradicts the assumption.
Thus b + αq = 0. That is, b = −αq. First, assume that q ≠ a. After plugging this in, the shape of
yzxy and yzyy coefficients of rj tells us that r3 and r4 are linearly independent modulo the span of
r1 and r2. If q + α(α2 + 2) ≠ 0, yyzy features with non-zero coefficient only in r2. Thus we must have
r1 = 0, which is not the case under the assumption q ≠ 0. Thus the only option is q = −α(α2 + 2).
is to have (α2 + 1)a − α2q + α = 0. This together with q = −α(α2 + 2) yields a = −α(α2 + 1). Plugging
Looking at yzxy and yzyy coefficients, we see that the only way for r1 and r2 to be linear dependent
in the rest of the data we have
a = −α(α2 + 1), b = α2(α2 + 2), c = −α2, d = 0, p = 1, q = −α(α2 + 2).
(9.18)
For these values of parameters rj span a 3-dimensional space. This concludes the case q ≠ a. Now
we assume a = q. In this case after eliminating zero columns as well as ones being obviously linear
combinations of the ones present, the matrix of coefficients of rj reduces to the form we are finally not
embarrassed to present in full:
⎛⎜⎜⎜⎝
q + α
−(α2 + 1)
0
0
q + α(α2 + 2)
0
0
0
0
q + α
q + α
1
0
0
0
−1
⎞⎟⎟⎟⎠
.
67
The only way for it to have rank 3 is for the equality (q + α)(q + α(α2 + 2)) = 0 to be satisfied, which
yields two more one-parametric families of algebras:
a = −α, b = α2, c = 0, d = 0, p = 1, q = −α;
a = −α(α2
+ 2), b = α2(α2
+ 2), c = 0, d = 0, p = 1, q = −α(α2
This concludes Case 2a.
Case 2b: q = 0 (on top of d = 0, p = 1 and c = −αa − b).
It is easy to see that for three specific cases
α = 0, a = 0, b = 0, c = 0, d = 0, p = 1, q = 0,
α2 + 2 = 0, a = 0, b = 0, c = 0, d = 0, p = 1, q = 0,
α2 + 2 = 0, a = α, b = 0, c = 2, d = 0, p = 1, q = 0,
+ 2).
(9.19)
(9.20)
(9.21)
rj span a 3-dimensional space, thus we have specified 5 points in S (zero dimensional irreducible
components, actually, when S is interpreted as a subvariety of K × KP 5).
On the other hand, if our parameters are not the ones provided in (9.21), one easily sees that r2
and r3 are linearly independent modulo linear span of r1 and r4. Thus for rj to span a 3-dimensional
space, r1 and r4 must be linearly dependent. The latter is easily seen to never happen. This concludes
Case 2b and Case 2.
Case 3: p = d = 0 and q ≠ 0. By scaling z, we can without loss of generality assume that q = 1.
and yzxz coefficients of rj is invertible (with the proper ordering of rows and columns, it is triangular
If (a − αc)(a − 1) ≠ 0, rj are linearly independent. Indeed the 4 × 4 matrix of the yyzx, yyzy, xzxz
with non-zero diagonal entries). Hence we must have (a − αc)(a − 1) = 0. First, assume that a = αc.
rj are still always linearly independent. Thus we have a − αc ≠ 0. Since (a − αc)(a − 1) = 0, we must
In this case, treating the cases b + α = 0 and b + α ≠ 0 separately, it is straightforward to verify that
have a = 1. In this case, again, rj are linearly independent, unless c = 0 and b = −α. In the latter case
rj span a 3-dimensional space, providing yet another piece of S:
a = 1, b = −α, c = 0, d = 0, p = 0, q = 1.
(9.22)
This concludes Case 3.
Case 4: p = d = q = 0.
In this case, the matrix of coefficients of rj becomes so simple that we just give the answer.
If
a = b = c = 0, all rj vanish (and have no chance to span a 3-dimensional space). If b ≠ 0 and a = c = 0,
rj span a 2-dimensional space (yielding dim A4 = 16). In all other cases rj are linearly independent.
Thus this final case provides no contribution into S ∖ S0.
As a result, the equality dim A = 15 only happens if the parameters (after appropriate scaling) fall
into one of the sets described in (9.14–9.22). Now a direct computation of the degree 5 part of the
Grobner basis of the ideal of relations of A yields dim A5 > 21 for 5 algebras corresponding to (9.21)
and for generic α (with finitely many possible exceptions when the leading monomials differ from the
ones in generic case) for each of the one-parametric families. As a matter of fact, dim A5 = 22 for the
family (9.14) and dim A5 = 23 for all other families. Anyway, by Lemma 2.7, dim A5 ⩾ 22 for every α,
which is incompatible with A ∈ Ω.
Now Part I of Theorem 1.11 follows straight away from Lemmas 8.2, 8.4, 8.6, 8.9, 8.10, 9.2, 9.3,
9.5, 9.8, 9.10, 9.13, 9.14, 9.15 and 9.16.
10 Proof of Part XI of Theorem 1.11
Throughout this section we work with the left-to-right degree-lexicographical ordering assuming x > y.
We start by ruling out an annoying case with no algebras from Λ in it.
68
Lemma 10.1. Let A be the cubic algebra given by generators x and y and relations
y3, x3
− axxy +(1 − a)xyx −(a + b)yxx −(c − a2)xyy − dyxy −(1 − bc + a − a2)yyx
for some a, b, c, d ∈ K. Then A ∉ Λ.
Proof. Assume the contrary. That is A ∈ Λ for some a, b, c, d ∈ K. Members of the Grobner basis of
the ideal of relations of A of degrees up to 5 consist of the two degree three defining relations, one
degree four element
g=xxyx−cxxyy−(1+b)xyxx−(d−a(1−a))xyxy−(1+a−c(1+b))xyyx−((1−a)(a+b)−d)yxyx
−(ad−(c−a2)(a+b))yxyy+(a(1+b)+b(a+b)+(1−bc))yyxx−(a(1−bc)+a2 (1−a)−d(a+b))yyxy
and one degree 5 element
+b+1)(a+b−c)]xyyxx−[−d((b−c)(1+b)+2)−a2 (1−c(1+b))−ac]xyyxy−(1+b)[a(a+b)+d]yxyxx
h=xxyyx+(1+b)(d+a(a+b))xyxxy−[−bd+b(1−a)(1+b+a)]xyxyx−[−b(1+b)(c−a2 )−(d−a(1−a))(d+a2 )]xyxyy
−[−b−(b2
−[−((1−a)(a+b)−d)(d+a2 )+bd]yxyxy−[+(1+a)(a+b)−(1+b)a2 (a+b)+d−ad(1+b)]yxyxy
−[+(a(1+b)+b(a+b))(d+a(a+b))+(1−bc)(d+a2 )]yxyxy−[−b(1−a)(a+b)(a+b)−ab(1−a)−bd(a+b)−b(1−bc)]yyxyx
−[+(c−a2)b[+b(a+b)+a(1+b)]−(1−bc)[−bc+ad+a2 (a+b)]−a2(1−a)(d+a2 )+d(a+b)(d+a2 )]yyxyy.
Note that there are exactly 16 degree 6 monomials in x, y, which do not contain the leading monomials
x3, y3, x2yx and x2y2x of the members of the Grobner basis of degree up to 5. Since 16 is also the
t6-coefficient of the series (1 + t)−1(1 − t)−3 from the definition of Λ, the only way for A to fall into Λ
Now, this does not occur. The overlap x2y(x3) =(x2yx)x2 yields a degree 6 homogeneous element
is for all degree 6 overlaps of the above leading monomials to resolve (produce no degree 6 members
of the Grobner basis).
f of the free algebra with coefficients being polynomials in a, b, c, d. Since A ∈ Λ, we must have
f = 0. This gives a system of algebraic equations on a, b, c, d. This system has no solutions at all,
which is easier to confirm by hand than one might think. For starters, the xyx2y2-coefficient in f is
(1 + b)(d + a(a + b))(d − c2 + bc + 2ac). Thus we must have b = −1 or d = −a(a + b) or d = c2 − bc − 2ac.
Plugging these one at a time into other coefficients of f we not only reduce the number of parameters
by one but cause massive cancellations each time. In each case we get another coefficient which is a
product of low degree polynomials and repeat the procedure. Anyway, we arrive to a contradiction,
which completes the proof.
Lemma 10.2. Let V be a 2-dimensional vector space over K and R be a 2-dimensional subspace
of V 3 such that y3 ∈ R and R is not contained in the ideal generated by y. Then the cubic algebra
A = B(V, R) belongs to Λ if and only if there is x ∈ V such that x, y form a basis in V and R is spanned
by y3 and x3 − xy2 − ayxy − y2x for some a ∈ K∗.
Proof. We start by picking an arbitrary x ∈ V such that x, y form a basis in V . By assumptions, R is
spanned by y3 and f = x3 + b1x2y + b2xyx + b3yx2 + b4xy2 + b5yxy + b6y2x for some b ∈ K6. If b1 ≠ b2
it is a routine exercise to see that a substitution of the form x → sx + ry, y → ty with s, t ∈ K∗, r ∈ K
transforms f into x3 − axxy +(1 − a)xyx −(a + b)yxx −(c − a2)xyy − dyxy −(1 − bc + a − a2)yyx for
some a, b, c, d ∈ K up to a scalar multiple. By Lemma 10.1, A ∉ Λ. Thus A ∉ Λ if b1 ≠ b2. The case
b2 ≠ b3 transforms into the case b1 ≠ b2 when we pass to the opposite multiplication. Thus A ∉ Λ unless
b1 = b2 = b3. Next, a substitution of the same form as above kills b1, b2 and b3. After this substitution,
R is spanned by y3 and f = x3 − sxy2 − ayxy − by2x for some s, a, b ∈ K.
Case 1: s ≠ 0.
In this case an additional scaling turns s into 1. Thus we can assume that s = 1 and therefore R is
spanned by y3 and f = x3 − xy2 − ayxy − by2x for some a, b ∈ K.
Case 1a: a = 0 in addition to s = 1.
69
The elements of the Grobner basis of the ideal of relations of A of degree up to 5 turn out to be
y3, x3 − xy2 − by2x, x2y2 +(b − 1)xy2x − by2x2, (1 − b)xy2xy + by2x2y and (1 − b + b2)xy2x2 − b2y2xy2.
If b ≠ 1 and b2 − b + 1 ≠ 0, the leading monomials of the elements of the Grobner basis of degrees
up to 5 are y3, x3, x2y2, xy2xy and xy2x2. There are exactly 16 degree 6 monomials which do not
contain the above leading monomials as subwords. Thus in order for A to be a member of Λ all
degree 6 overlaps of leading monomials must resolve (same argument as in the proof of Lemma 10.1).
A direct computation shows that this happens if and only if either b = 0 or b2 + 1 = 0. Thus the
only values of the parameter b for which A has a chance to belong to Λ are solutions of the equation
b(b − 1)(b2 + 1)(b2 − b + 1) = 0. It remains to deal with finitely many specific algebras (6 to be precise).
A direct Grobner basis computation in each case shows that for all these algebras dim A7 ⩾ 21, which
is incompatible with the membership in Λ. Thus no algebras from Λ feature in this case.
Case 1b: a ≠ 0 in addition to s = 1.
The elements of the Grobner basis of the ideal of relations of A of degree up to 5 turn out to be
y3, x3 − xy2 − ayxy − by2x, x2y2 + axyxy +(b − 1)xy2x − ayxyx − by2x2,
a xy2xy − yxyxy − b
a y2x2y and x2yxy − bxyxyx −
xyxy2 + b−1
1−b+b2
a xy2x2 +(b − 1)yxyx2 + b2
a y2xy2.
Again, there are exactly 16 degree 6 monomials, which do not contain the leading monomials y3, x3,
x2y2, xyxy2 and x2yxy as subwords. Thus in order for A to be a member of Λ all degree 6 overlaps of
leading monomials must resolve. A direct computation shows that this happens precisely when b = 1.
Now in the case b = 1, the five elements of the Grobner basis from the above display constitute the
entire Grobner basis of the ideal of relations of A (all higher degree overlaps resolve as well). Now
the five monomials y3, x3, x2y2, xyxy2 and x2yxy are all leading monomials of members of a Grobner
basis. This allows us to compute the Hilbert series of A, which is HA = (1 + t)−1(1 − t)−3. Thus in
this case A ∈ Λ precisely when b = 1. In this case the defining relations of A take the shape y3 and
x3 − xy2 − ayxy − y2x for some a ∈ K∗. That is, A coincides with Aa.
Case 2: s = 0 and b ≠ 0.
In this case an additional scaling turns b into 1. Then R is spanned by y3 and x3 − ayxy − y2x for
some a ∈ K. Then Aopp being A with the opposite multiplication is isomorphic to the algebra given
by generators x and y and relations y3 and x3 − ayxy − xy2. Thus Aopp is under the jurisdiction of
Case 1a and therefore Aopp ∉ Λ. Hence A ∉ Λ. It remains to consider the final case.
Case 3: s = b = 0.
If a = 0, R is spanned by x3 and y3. Then dim A4 = 14 and A fails to be in Λ′, let alone Λ. If a ≠ 0,
after a scaling , which turns a into 1, R becomes spanned by y3 and x3 − yxy. For this single algebra
direct Grobner computation (easy in this case because of the simplicity of relations) yields dim A9 = 31,
the coefficients of the Hilbert series coincide with that of (1 + t)−1(1 − t)−3 up to t8 inclusive. Still a
which is greater by 1 than the t9-coefficient of (1 + t)−1(1 − t)−3. Thus we have no algebras from Λ in
this case.
Lemma 10.3. Let A be the cubic algebra given by generators x and y and relations y3 and f , where
f is a linear combination of xyx, xy2, yxy and y2x. Then A ∉ Λ.
Proof. Case 1: xyx features in f with non-zero coefficient.
By a substitution of the form x → ux + ty, y → y with u ∈ K∗, t ∈ K, we can turn the xyx coefficient
in f into 1 and kill the xy2 coefficient. Thus without loss of generality f = xyx − ayxy − by2x with
some a, b ∈ K. If a ≠ 0, an additional scaling turns a into 1. Then f = xyx − yxy − by2x with b ∈ K.
Computing the Grobner basis in the ideal of relations for degrees up to 6, we see that apart from
the defining relations only xy2xy − yxy2x and yxy2x2 turn up. Still this yields dim A6 = 15, which
is incompatible with A being in Λ. If a = 0, after scaling we are left with two options f = xyx − y2x
and f = xyx. In both cases a Grobner basis computation yields dim A5 = 13, again incompatible with
membership in Λ.
70
Case 2: xyx does not feature f .
Now, f = axy2 + byxy + cy2x with some a, b, c ∈ K. An easy Grobner basis calculation shows that
dim A5 ⩾ 14 no matter what the parameters are. Hence A ∉ Λ.
Lemma 10.4. Let A be the cubic algebra given by generators x and y and relations y3 and x2y −
axyx − byx2 − yxy − cy2x with a, b, c ∈ K. Then A ∈ Λ if and only if either (a, b, c) = (1, −1, 0) or
(a, b, c) =(−1, −1, 0) or a = 0 and b + b2 + . . . + bk + c ≠ 0 for all k ∈ N.
Proof. Case 1: a ≠ 0.
A direct computation shows that the only members of the reduced Grobner basis in the ideal of
relations of A of degree at most 5 are y3, x2y − axyx − byx2 − yxy − cy2x and xyxy2 + byxyxy + b2y2xyx +
b+c
a y2xy2. Since there are exactly 16 degree 6 monomials that do not contain any of x2y, y3 and xyxy2
as submonomials, A can only fall into Λ if all overlaps of degree 6 of these three leading monomials
resolve. However the overlap (x2y)xy2 = x(xyxy2) resolves precisely when a2 = 1, b = −1 and c = 0.
Now if (a, b, c) =(±1, −1, 0), y3, x2y ∓xyx +yx2 −yxy and xyxy2 −yxyxy +y2xyx ±y2xy2 form the entire
reduced Grobner basis in the ideal of relations of A. This allows to confirm that HA =(1 + t)−1(1 − t)−3
and therefore A ∈ Λ in these two cases.
Case 2: a = 0.
This time the only members of the reduced Grobner basis in the ideal of relations of A of degree
at most 5 are y3 and x2y − byx2 − yxy − cy2x if b + c = 0 and y3, x2y − byx2 − yxy − cy2x and y2xy2 if
b + c ≠ 0. If b + c = 0, we have dim A5 = 13 and therefore A ∉ Λ. Assume now that b + c ≠ 0. There are
no elements of degree 6 of the Grobner basis. The only non-zero degree 7 member can arise from the
this case the members of the Grobner basis of degrees up to 7 are y3, x2y −byx2 −yxy −cy2x, y2xy2 and
y2xyxy2. This pattern goes on. If b + b2 + . . . + bk + c ≠ 0 for 1 ⩽ k ⩽ m, then the members of the Grobner
overlap (x2y)yxy2 = x2(y2xy2), which produces (c + b + b2)y2xyxy2. If c + b + b2 = 0, then dim A7 = 21
exceeds by 1 the t7 coefficient of (1 + t)−1(1 − t)−3. Thus to keep A in Λ, we must have c + b + b2 ≠ 0. In
basis of degrees up to 2m + 4 are y3, x2y − byx2 − yxy − cy2x and y2(xy)j xy2 for 0 ⩽ j ⩽ m − 1 and HA
and (1 +t)−1(1 −t)−3 match up to t2m+4. The only non-zero degree 2m +5 member of the Grobner basis
can arise only from the overlap x2y2(xy)m−1xy2, which produces (c + b + b2 + . . . + bm+1)y2(xy)mxy2.
If c + b + b2 + . . . + bm+1 = 0, dim A2m+5 exceeds by 1 the t2m+5 coefficient of (1 + t)−1(1 − t)−3 and A ∉ Λ.
Otherwise it matches and show goes on. As a result, A belongs to Λ if and only if b+b2 +. . .+bk +c ≠ 0 for
all k ∈ N. Note that in this case the full reduced Grobner basis in the ideal of relations of A consists
of the defining relations y3 and x2y − byx2 − yxy − cy2x together with infinitely many monomials
y2(xy)j xy2 for j ∈ Z+.
Lemma 10.5. Let A be the cubic algebra given by generators x and y and relations y3 and f =
x2y − axyx − byx2 − y2x with a, b ∈ K. Then A ∈ Λ if and only if a = 0.
Proof. Case 1: a ≠ 0.
1
A direct computation shows that the only members of the reduced Grobner basis in the ideal of
a y2xy2.
relations of A of degree at most 5 are y3, x2y −axyx −byx2 −y2x and xyxy2 +byxyxy +b2y2xyx +
Since there are exactly 16 degree 6 monomials that do not contain any of x2y, y3 and xyxy2 as
submonomials, A can only fall into Λ if all overlaps of degree 6 of these three leading monomials
resolve. However, the overlap (x2y)xy2 = x(xyxy2) is easily seen to never resolve. Hence A ∈ Λ.
Case 2: a = 0.
This time the only members of the reduced Grobner basis in the ideal of relations of A of degree
at most 5 are y3, x2y − byx2 − y2x and y2xy2. The pattern of the further Grbner basis construction
1 only. The full reduced Grobner basis in the ideal of relations of A consists of the defining relations
here is the same as in Case 2 of Lemma 10.4, only the monomials y2(xy)j xy2 pop up with coefficients
y3 and x2y − byx2 − yxy − cy2x together with infinitely many monomials y2(xy)j xy2 for j ∈ Z+. This
yields HA =(1 + t)−1(1 − t)−3 and therefore A ∈ Λ.
71
Lemma 10.6. Let A be the cubic algebra given by generators x and y and relations y3 and f =
x2y − axyx − byx2 with a, b ∈ K. Then A ∈ Λ if and only if a ≠ 0 and b = −a2.
Proof. Case 1: a = 0.
The only members of the reduced Grobner basis in the ideal of relations of A of degree at most 5
are the defining relations y3 and x2y − byx2. This yields dim A5 = 13 and therefore A ∉ Λ.
Case 2: a ≠ 0.
A direct computation shows that the only members of the reduced Grobner basis in the ideal of
relations of A of degree at most 5 are y3, x2y − axyx − byx2 and xyxy2 + byxyxy + b2y2xyx. Since there
are exactly 16 degree 6 monomials that do not contain any of x2y, y3 and xyxy2 as submonomials,
A can only fall into Λ if all overlaps of degree 6 of these three leading monomials resolve. This is
easily seen to happen precisely when a2 + b = 0. If a ≠ 0 and a2 + b = 0, then y3, x2y − axyx + a2yx2
and xyxy2 − a2yxyxy + a4y2xyx form the full Grobner basis of the ideal of relations of A, yielding
HA =(1 + t)−1(1 − t)−3.
Thus A ∈ Λ precisely when a ≠ 0 and b = −a2.
Lemma 10.7. Let A be the cubic algebra given by generators x and y and relations y3 and f =
x2y − xyx − ayx2 − xy2 − byxy − cy2x with a, b, c ∈ K. Then A ∈ Λ if and only if a = c = −1.
Proof. Case 1: a ≠ 0 and a ≠ −1.
1
a xy2 + b
a yxy +
1
1
a xyx −
a yx2 + c
In this case Aopp is isomorphic to the algebra given by generators x and y and relations y3 and
x2y +
1+a y, we see that
Aopp is isomorphic to the algebra given by generators x and y and relations y3 and x2y + sxyx − syx2 +
tyxy + sy2x, where s = 1
. If t ≠ 0, a scaling turns the defining relations into y3 and
x2y + sxyx − syx2 − yxy − s
t sy2x, which falls under the jurisdiction of Lemma 10.4. If t = 0, a scaling
turns the defining relations into y3 and x2y + sxyx − syx2 − y2x, which falls under the jurisdiction of
Lemma 10.5. Applying these two lemmas, we see that A ∉ Λ.
a y2x. Applying the substitution y → y, x → x − c
a and t = c(1−a)+b(1+a)
a(1+a)
Case 2: a = 0.
A direct computation shows that the only members of the reduced Grobner basis in the ideal of
relations of A of degree at most 5 are y3, x2y − xyx − xy2 − byxy − cy2x and xyxy2 + cy2xy2. Since there
are exactly 16 degree 6 monomials that do not contain any of x2y, y3 and xyxy2 as submonomials, A
can only fall into Λ if all overlaps of degree 6 of these three leading monomials resolve. This does not
happen with the overlap x2yxy2 though. Hence A ∉ Λ.
Case 3: a = −1.
A direct computation shows that the only members of the reduced Grobner basis in the ideal of
relations of A of degree at most 5 are y3, x2y − xyx + yx2 − xy2 − byxy − cy2x and xyxy2 − yxyxy +
y2xyx +(c + 1 − b)y2xy2. Since there are exactly 16 degree 6 monomials that do not contain any of
x2y, y3 and xyxy2 as submonomials, A can only fall into Λ if all overlaps of degree 6 of these three
leading monomials resolve. This is easily seen to happen precisely when c + 1 = 0. If a = c = −1, then
y3, x2y − xyx + yx2 − xy2 − byxy + y2x and xyxy2 − yxyxy + y2xyx − by2xy2 form the full Grobner basis
of the ideal of relations of A, yielding HA =(1 + t)−1(1 − t)−3.
Thus A ∈ Λ precisely when a = c = −1.
Now we are ready to prove Part XI of Theorem 1.11. The fact that all algebras A in (Z1–Z10) are
in Λ follows from Lemmas 10.2 and 10.4–10.7. Indeed, either A itself or Aopp features in these lemmas
as an algebra from Λ. Next, all algebras in (Z1–Z10) are pairwise non-isomorphic. Indeed, since y3
is the only cube (up to a scalar multiple) among the cubic relations of any algebra in (Z1–Z10), a
linear substitution providing an isomorphism between two algebras in (Z1–Z10) must be of the form
x → ux + ty, y → vy with u, v ∈ K∗ and t ∈ K. Since the subs x → sx, y → sy with s ∈ K∗ do not change
any of the spaces of cubic relations, we can restrict ourselves to the case u = 1. Now it is a matter
of routine verification to see that for two algebras from (Z1–Z10), a substitution x → x + ty, y → vy
72
transforms the space of cubic relations of one algebra to the space of cubic relations of the other only
if we had the two copies of the same item from (Z1–Z10) to begin with. It remains to verify that if
A ∈ Λ and the quasipotential Q = QA is the fourth power of a degree 1 element, then A is isomorphic
to an algebra form (Z1–Z10). Note that the algebras in (Z1–Z10) put together form a class of algebras
closed under passing to the opposite multiplication. Thus it is enough to show that either A or Aopp
is isomorphic to an algebra from (Z1–Z10).
Without loss of generality, we can assume that y4 is the quasipotential of A, where y ∈ V . If the
space R of cubic relations of A is not contained in the ideal generated by y, Lemma 10.2 guarantees
that A is isomorphic to an algebra from (Z1). Thus we can assume that R is contained in the ideal
generated by y. Then A is given by generators x, y and relations y3 and f , which is a non-zero linear
combination of x2y, xyx, yx2, xy2, yxy and y2x. If both x2y and yx2 do not feature in f , Lemma 10.3
implies that A ∉ Λ yielding a contradiction. Hence either x2y or yx2 feature in f with a non-zero
coefficient. These two options transform to one another when we pass to the opposite multiplication.
Since it is enough to verify that either A or Aopp is isomorphic to an algebra from (Z1–Z10), we can
without loss of generality assume that x2y features in f with non-zero coefficient.
Case 1: xy2 does not feature in f but yxy does.
In this case a scaling turns x2y and yxy coefficients into 1 and −1 respectively. That is, f =
x2y − axyx − byx2 − yxy − cy2x with a, b, c ∈ K. By Lemma 10.4, A is isomorphic to an algebra from
(Z2–Z4) or (Z6).
Case 2: xy2 and yxy do not feature in f but y2x does.
In this case a scaling turns x2y and y2x coefficients into 1 and −1 respectively. That is, f =
x2y − axyx − byx2 − y2x with a, b ∈ K. By Lemma 10.5, A is isomorphic to an algebra from (Z7).
Case 3: None of xy2, yxy or y2x feature in f .
In this case a scaling turns x2y-coefficient into 1. That is, f = x2y − axyx − byx2 with a, b ∈ K. By
Lemma 10.6, A is isomorphic to an algebra from (Z9).
Note that Cases 1–3 cover all options when xy2 does not make an appearance in f .
Case 4: xy2 features in f with non-zero coefficient.
If the sum of x2y and xyx coefficients in f is non-zero, a substitution x → x + sy, y → y with an
appropriate s ∈ K kills xy2 in f , after which Cases 1–3 kick in. Thus we can assume that x2y and xyx
coefficients in f add up to zero. A scaling turns x2y and xy2 coefficients into 1 and −1 respectively.
Then f = x2y − xyx − ayx2 − xy2 − byxy − cy2x for some a, b, c ∈ K. By Lemma 10.7, A is isomorphic to
an algebra from (Z10). This completes the final case and the proof of Part XI of Theorem 1.11.
11 Proof of Part XII of Theorem 1.11
Lemma 11.1. The algebras from (Y1–Y8) of Theorem 1.11 with different labels are non-isomorphic.
There are no isomorphism between two algebras from (Y1–Y8) with the same label apart from the label
(Y1), where the isomorphic ones are precisely those specified in (Y1).
Proof. Note that every algebra from (Y1–Y8) is quasipotential with the corresponding quasipotential
being of the form Q = uf v, where u, v ∈ V are linearly independent and the non-zero f ∈ V 2 is not a
square of a degree one element. If we have two isomorphic algebras A and A′ from the list (Y1–Y8),
then a linear substitution facilitating the isomorphism must transform the quasipotential Q = uf v of
A to the quasipotential Q′ = u′f ′v′ of A′ up to a non-zero scalar multiple. It follows that u, v and f
are transformed to scalar multiples of u′, v′ and f ′ respectively. Now using Lemma 2.2 (including the
isomorphism part), we easily see that algebras in (Y1–Y8) with different labels can not be isomorphic
and that algebras in (Y2–Y8) are pairwise non-isomorphic. As for two algebras from (Y1), Lemma 2.2
yields that the only substitutions which can transform an algebra from (Y1) to an algebra from from
(Y1) are x → sx, y → sy or x → sy, y → sx with s ∈ K∗. The first type of substitutions do not
73
change the parameters a, b, while the second type (the swap) transforms an algebras from (Y1) with
Lemma 11.2. Let A ∈ Λ′ be such that the corresponding quasipotential Q is not a fourth power of
parameters a, b to the algebra from (Y1) with parameters (a−1, b−1).
degree 1 element and satisfies n1(Q) = n2(Q) = 1 and E1(Q) = E2(Q). Then A ∉ Λ.
Proof. Let y ∈ V be such that y spans the one-dimensional space E1(Q). Take x ∈ V such that x and y
are linearly independent. Since y spans E1(Q) = E2(Q), we have Q = yf y, where f ∈ V 2 (a quadratic
element). Clearly f is a linear combination of x2, xy, yx and y2. Since Q is not a fourth power, f can
not be a scalar multiple of y2.
Case 1: x2 features in f with non-zero coefficient.
A substitution of the form x → ux + sy, y → y with appropriate u ∈ K∗, s ∈ K kills xy in f . Then
f becomes a linear combination of x2, yx and y2.
If yx still features in f after the substitution,
we turn the x2 and yx coefficients in f into 1 and −1 respectively by an appropriate scaling. Then
f = x2 − yx − ay2 for some a ∈ K. Hence Q = y(x2 − yx − ay2)y and therefore RQ is spanned by
x2y − yxy − ay3 and yx2 − y2x − ay3. Since these two are linearly independent, A is presented by
generators x and y and relations x2y − yxy − ay3 and yx2 − y2x − ay3. If yx does not features in f after
the above substitution, we turn the x2 coefficient of f into 1 by a scaling and observe that f = x2 − ay2
for some a ∈ K. Hence Q = y(x2 − ay2)y and therefore RQ is spanned by x2y − ay3 and yx2 − ay3. Since
these two are linearly independent, A is presented by generators x and y and relations x2y − ay3 and
yx2 − ay3. In both cases, an easy Grobner basis computation (apart from the defining relations there
is only one other degree ⩽ 5 member) yields dim A5 ⩾ 13, which is incompatible with A being in Λ.
Case 2: x2 does not feature in f , but the sum of the xy and yx coefficients is non-zero.
Clearly either xy or yx (or both) feature in f . By passing to the opposite multiplication, if necessary,
we can without loss of generality assume that xy features in f . A substitution of the form x → ux + sy,
y → vy with appropriate u, v ∈ K∗, s ∈ K kills y2 in f and turns the xy coefficient into 1. Then
f = xy − ayx for some a ∈ K. Hence Q = y(xy − ayx)y and therefore RQ is spanned by xy2 − ayxy
and yxy − ay2x. Since these two are linearly independent, A is presented by generators x and y and
relations xy2 − ayxy and yxy − ay2x. Same argument as in Case 1 yields dim A5 ⩾ 13 and therefore
A ∉ Λ.
Case 3: x2 does not feature in f and the xy and yx coefficients in f sum up to zero.
Since f can not be a scalar multiple of y2, both xy and yx feature in f . An appropriate scaling
turns f into either xy − yx or xy − yx − y2. This gives Q = y(xy − yx)y or Q = y(xy − yx − y2)y. Same
argument as in the previous cases shows that A is presented by the generators x and y and either the
relations xy2 − yxy and yxy − y2x or the relations xy2 − yxy − y3 and yxy − y2x − y3. Again, a Grobner
basis calculation gives dim A5 = 13 in both cases and therefore A ∉ Λ.
Lemma 11.3. Let A ∈ Λ be such that the corresponding quasipotential Q is not a fourth power of degree
1 element and satisfies n1(Q) = n2(Q) = 1. Then there is a basis x, y in V with respect to which Q has
one of the following forms∶ (x − by)(xy − ayx)(x − y) with b ≠ 1, a ≠ 0 and a ≠ 1, (x − y)(xy − yx − ayy)x
with a ≠ 0, x(xy − yx)y, (x − ay)xy(x − y) with a ≠ 1, (x − y)(xy − ayx)y with a ≠ 1, x(xy − ayx)y with
a ≠ 1, x(xy − yx − yy)y or y(xy − yx − yy)x.
Proof. In this proof we only use the left-to-right degree-lexicographical ordering on x, y monomials
assuming x > y. By Lemma 11.2, Q = uf v, where u, v are linearly independent elements of V and
f ∈ V 2 is non-zero. Clearly, RQ is spanned by uf and f v. Using the fact that u and v are linearly
independent, it is easy to see that uf and f v are linearly independent. Hence RQ is the two-dimensional
space spanned by uf and f v and therefore the ideal of relations of A is generated by uf and f v. First,
observe that f can not be a square of an element of V . Indeed, if f = y2 for y being a non-zero element
of V , then the ideal of relations I of A is generated by uy2 and y2v. Clearly I is contained in the
ideal J generated by y2 and therefore B = F(V)~J is a quotient of A. On the other hand, B has
exponential growth and therefore A has exponential growth. Since the latter is incompatible with the
74
membership in Λ, we arrive to a contradiction. Thus f is not a square. By Lemma 2.2, there is a
basis x, y in V such that f = xy or f = xy − ayx with a ∈ K∗ or f = xy − yx − yy. We go through these
options case by case.
Case 1: f = xy.
If u is not a scalar multiple of y and v is not a scalar multiple of either x or y, a scaling (of x
linearly independent. If u is not a scalar multiple of y and v is a scalar multiple of y, a scaling turns
and y) turns Q = uxyv into (x − ay)xy(x − y) with a ∈ K. We also have a ≠ 1 since u and v are
Q into either (x − y)xyy or xxyy (depending on whether u is a scalar multiple of x or not), which
are (x − y)(xy − ayx)y or x(xy − ayx)y with a = 0. If u is not a scalar multiple of y and v is a scalar
multiple of x, a scaling turns Q into (x − ay)xyx with a ∈ K. Since u and v are linearly independent,
a ≠ 0. Then the defining relations of A are x2y − ayxy and xyx. A direct computation shows that there
are no other members of degrees up to 5 in the reduced Grobner basis of the ideal of relations of A
apart from x2y − ayxy, xyx and xy2xy. This yields dim A5 = 13, which is incompatible with A ∈ Λ. It
remains to consider the case when u is a scalar multiple of y. Now a scaling turns Q into yxy(x − ay)
with a ∈ K. Then the defining relations of A are xyx − axy2 and yxy. A direct computation shows that
there are no other members of degrees up to 5 in the reduced Grobner basis of the ideal of relations
of A apart from x2y − ayxy, xyx and xy3. This yields dim A5 = 13, which is incompatible with A ∈ Λ.
Thus the last two cases do not occur.
Case 2: f = xy − yx.
we have b ≠ 1. If u is not a scalar multiple of either x or y and v is a scalar multiple of y, a scaling
Case 3: f = xy − ayx with a ∈ K∗, a ≠ 1.
If u is not a scalar multiple of y and v is not a scalar multiple of either x or y, a scaling (of x and
In this case it is easy to see that Q is a scalar multiple of u(uv − vu)v. In the basis x = u, y = v,
this yields Q = x(xy − yx)y.
y) turns Q = uxyv into (x − by)(xy − ayx)(x − y) with b ∈ K. Since u and v are linearly independent,
turns Q into (x − y)(xy − ayx)y. If u is a scalar multiple of x and v is a scalar multiple of y, a scaling
turns Q into x(xy − ayx)y. If u is not a scalar multiple of either x or y and v is a scalar multiple of
x, a scaling turns Q into (x − y)(xy − ayx)x. Swapping x and y and a further scaling transforms Q
into (x − y)(xy − a−1yx)y. All cases of u not being a scalar multiple of y are taken care of (some do
multiple of either x or y, a scaling (of x and y) turns Q into y(xy − ayx)(x − y). Swapping x and y and
a further scaling transforms Q into x(xy − a−1yx)(x − y) =(x − by)(xy − a−1yx)(x − y) with b = 0. If u
is a scalar multiple of y and v is a scalar multiple of x, a scaling turns Q into y(xy − ayx)x. Swapping
x and y and a further scaling transforms Q into x(xy − a−1yx)y.
α, γ ∈ K∗ and β ∈ K transforms Q into (x − y)(xy − yx − ayy)x with a ≠ 0. If v is a scalar multiple of y,
then u is not a scalar multiple of y and a substitution of the same form turns Q into x(xy − yx − yy)y.
turns Q into y(xy − yx − yy)x.
being one of the following∶ (x − ay)xy(x − y) with a ≠ 1, (x − y)(xy − ayx)y with a ≠ 1, x(xy − ayx)y
with a ∈ K, x(xy − yx − yy)y or y(xy − yx − yy)x. Then A ∈ Λ.
Q = (x − y)(xy − ayx)y, the defining relations of A are x2y − axyx − yxy + ay2x and xy2 − ayxy. If
Q = x(xy − ayx)y, the defining relations of A are x2y − axyx and xy2 − ayxy. If Q = x(xy − yx − yy)y,
monomials of the members being x2y and xy2. This yields HA =(1 + t)−1(1 − t)−3 and therefore A ∈ Λ.
the defining relations of A are x2y − xyx − yxy − y3 and xy2 − yxy − y3. In all these cases, the defining
relations themselves form the reduced Grobner basis of the ideal of relations of A with the leading
Case 4: f = xy − yx − y2.
If neither u nor v is a scalar multiple of y, a substitution of the form x → αx + βy, y → γy with
Lemma 11.4. Let A ∈ Λ′ be the quasipotential algebra on generators x, y with the quasipotential Q
Proof. We use the left-to-right degree-lexicographical ordering on x, y monomials assuming x > y. If
not feature since u and v are linearly independent). If u is a scalar multiple of y and v is not a scalar
If u is a scalar multiple of y, then v is not a scalar multiple of y and a substitution of the same form
75
An easy computation shows that the reduced Grobner basis of the ideal of relations of A consists of
x2y − ayxy, xy2n−1x − xy2n for n ∈ N and xy2nxy for n ∈ N. The leading monomials of the members of
The algebra with Q = y(xy −yx−yy)x is isomorphic to Bopp, where B is the algebra with quasipotential
x(xy − yx − yy)y. Since B and Bopp have the same Hilbert series, A ∈ Λ as well. This leaves only the
case Q =(x − ay)xy(x − y) with a ≠ 0. Then the defining relations of A are x2y − ayxy and xyx − xy2.
the basis are x2y and xy2n−1x, xy2nxy for n ∈ N. Knowing these, we get HA = (1 + t)−1(1 − t)−3 and
The next two lemmas deal with quasipotentials (x − by)(xy − ayx)(x − y) and (x − y)(xy − yx − ayy)x.
We have to confess that we could not find any nice pattern in the leading monomials of the members of
the Grobner basis arising from the natural presentations of the corresponding algebras. However, we
found different presentations of the same algebras (with three generators rather than two) for which
we can compute the basis and hence the Hilbert series.
therefore A ∈ Λ.
Lemma 11.5. Let A ∈ Λ′ be the quasipotential algebra on generators x, y with the quasipotential
Q = (x − by)(xy − ayx)(x − y) with b ≠ 1, a ≠ 0 and a ≠ 1. Then A ∈ Λ if and only if anb ≠ 1 for all
n ∈ N.
Proof. Clearly, A can be presented by generators x, y, f and relations xy − ayx − f , xf − byf , f x − f y.
To make A graded in the same way as it was originally, we assign degree 1 to x and y and degree
2 to f . Now we order the monomials in x, y, f in the following way. A monomial of greater degree
is greater. For two monomials of the same total degree, the one of greater x-degree (with more x in
it) is greater. For two monomials of the same degree and the same x-degree, the one with greater
y-degree is greater. Finally, for two monomials of the same degree, same x-degree and same y-degree
(they have the same number of x, the same number of y and therefore the same number of f as well),
we break ties using the left-to-right lexicographical order assuming f > x > y.
As an illustration, we list low degree terms of the reduced Grobner basis in the ideal of relations of
A as presented above. The degrees up to 3 terms are the defining relations xy − ayx − f , xf − byf and
a f 2. Two degree
f x − f y. The only degree 4 term arises from the overlap f xy and is f yx −
1
a2 f 2y respectively.
5 terms arise from the overlaps f xf and f yxy and are f yf and f y2x −
The degree 6 overlap f yxf produces (1 − ab)f y2f − f 3. Now, who is the leading monomial of this
1−ab f 3.
a(1−ab) f 3. All other degree 6
a2(1−ab) f 3y. Other degree
7 overlaps resolve except for f y2xf , which produces (1 − a2b)f y3f . If a2b = 1 the latter disappears
one depends on whether ab equals 1 or not. If ab ≠ 1, we have the degree 6 element f y2f −
Another degree 6 overlap f y2xy now yields f y3x −
overlaps resolve. Degree 7 overlap f y3xy produces f y4x −
a f y2 +
a2 f y3 +
from the list. Otherwise we have the monomial f y3f .
a3 f 2y2 +
a4 f 2y3 +
a3 f y4 +
1
1
a4 f y5 +
1
1
1
1
1
1
1
1
Case 1: anb ≠ 1 for all n ∈ N.
An easy inductive argument shows that the reduced Grobner basis in the ideal of relations of A
a f 2 and two elements of each degree m ⩾ 5,
consists of xy − ayx − f , xf − byf , f x − f y, f yx −
which have the form f ym−3x − h with h being a linear combination of f ym−4, f 2ym−2, . . . and f ym−4f
if m is odd or f ym−4f − smf m~2 with sm ∈ K if m is even. These two terms come from the overlaps
f ym−4xy and f ym−5xf respectively, while all other overlaps of degree m resolve. As a result, the
complete list of leading monomials of the members of the basis is xy, xf , f yjx for j ∈ Z+ and f yjf
for j ∈ N. Thus the corresponding normal words are yj xk and yjf myk with j, k, m ∈ Z+. Counting the
number of these words of given degree, we easily confirm that HA = (1 + t)−1(1 − t)−3 and therefore
a f y2 +
1
1
A ∈ Λ.
Case 2: n ∈ N is the smallest positive integer for which anb = 1.
We look at the whole family of algebras corresponding to a ∈ K∗, a ≠ 1 and b = a−n. Up to degree
n + 4 inclusive the Grobner basis elements are the same as in Case 1.
Case 2a: n is even.
76
Then we have just one (as opposed to two in Case 1) degree n + 5 element of the Grobner basis:
f yn+1f disappears. We acquire one extra normal word f yn+1f of degree n + 5. Hence dim An+5 is
greater by 1 than in Case 1 and therefore A ∉ Λ.
Case 2b: n is odd.
In this case the degree n + 5 element of the Grobner basis coming from the overlap f ymxf is a scalar
multiple of f k where k = n+5
2 . It vanishes for finitely many exceptional values of a, but for a (Zarisski)
generic a this makes f k a member of the Grobner basis. At degree n +5, we have 'lost' one normal word
f k and 'acquired' one normal word f yn+1f when compared to Case 1. The changes balance themselves
yielding the same dim An+5 as in Case 1. Performing two more steps of the Grobner basis calculation,
we see that for a generic a, in degree n + 6 we lose two normal words yf k, f ky and acquire two normal
words yf yn+1f and f yn+1f y as compared with Case 1. Still the changes balance and dim An+6 is the
same as in Case 1. However in degree n + 7, we lose four normal words y2f k, yf ky, f ky2 and f k+1 and
acquire five y2f yn+1f , yf yn+1f y, f yn+1f y2, f 2yn+1f and f yn+1f 2. Thus for a generic a, dim An+7 is
greater by 1 than in Case 1. By Lemma 2.7, for an arbitrary a, dim An+7 is greater than in Case 1 by
at least 1. Hence A ∉ Λ.
Lemma 11.6. Let A ∈ Λ′ be the quasipotential algebra on generators x, y with the quasipotential
Q =(x − y)(xy − yx − ayy)x with a ≠ 0. Then A ∈ Λ if and only if na + 1 ≠ 0 for all n ∈ N.
Proof. Clearly, A can be presented by generators x, y, f and relations xy − yx − ayy − f , xf − yf , f x.
We use the same grading and the same order on x, y, f monomials as in the proof Lemma 11.5.
As an illustration, we list low degree terms of the reduced Grobner basis in the ideal of relations of
A as presented above. The degrees up to 3 terms are the defining relations xy − yx − ayy − f , xf − yf
and f x. The only degree 4 term arises from the overlap f xy and is f yx +af y2 +f 2. Two degree 5 terms
arise from the overlaps f xf and f yxy and are f yf and f y2x + 2af y3 + f 2y respectively. The degree
whether 1 + a equals 0 or not. If 1 + a ≠ 0, we have the degree 6 element f y2f +
6 overlap f y2xy now yields f y3x + 3af y4 + f 2y2 −
7 overlap f y3xy produces f y4x + 4af y5 + f 2y3 −
6 overlap f yxf produces (1 + a)f y2f + f 3. Now, who is the leading monomial of this one depends on
1+a f 3. Another degree
1+a f 3. All other degree 6 overlaps resolve. Degree
1+a f 3y. Other degree 7 overlaps resolve except for
f y2xf , which produces (1 + 2a)f y3f . If 1 + 2a = 0, the latter disappears from the list. Otherwise we
1
1
1
have the monomial f y3f .
Case 1: na + 1 ≠ 0 for all n ∈ N.
An easy inductive argument shows that the reduced Grobner basis in the ideal of relations of A
consists of xy − yx − ayy − f , xf − yf , f x, f yx + af y2 + f 2 and two elements of each degree m ⩾ 5, which
have the form f ym−3x − h with h being a linear combination of f ym−4, f 2ym−2, . . . and f ym−4f if m is
odd or f ym−4f − smf m~2 with sm ∈ K if m is even. These two terms come from the overlaps f ym−4xy
and f ym−5xf respectively, while all other overlaps of degree m resolve. The leading monomials of
memebrs of the Grobner basis as well as normal words are the same as in Case 1 in the proof of
Lemma 11.5. Hence A ∈ Λ.
1
Case 2: char K = 0 and a = −
Exactly the same argument as in Case 2 in the proof of Lemma 11.5 shows that dim An+5 is greater
n for some n ∈ N.
Case 3: char K = p (p is a prime) and a = −
The defining relations of A are nxy − nyx + yy − nf , xf − yf and f x. If we treat them as relations
by 1 than the tn+5 coefficient of (1 + t)−1(1 − t)−3 if n is even and that dim An+7 is greater by 1 than
the tn+7 coefficient of (1 + t)−1(1 − t)−3 if n is odd. In any case A ∉ Λ.
n for some n ∈ N, n < p.
defining a Q-algebra B, Case 2 yields that dim Bk is greater by 1 than the tk coefficient of (1 + t)−1(1 −
t)−3, where k = n + 5 if n is even and k = n + 7 if n is odd. If we consider nxy − nyx + yy − nf , xf − yf
greater by at least 1 than the tk coefficient of (1 + t)−1(1 − t)−3 and therefore A ∉ Λ.
and f x as defining relations of a Zp-algebra C, an argument similar to the one from the proof of
Lemma 2.7 shows that dim Cj ⩾ dim Bj for all j ∈ Z+. On the other hand, HA = HC. Hence dim Ak is
1
77
Now Part XII of Theorem 1.11 is just an amalgamation of Lemmas 11.1 and 11.3–11.6.
12 More remarks
Cases char K = 2 and char K = 3 differ from the case char K ∉ {2, 3} we considered (and differ from
each other) in a whole lot of ways. Some parts of our results hold in these cases, however more require
adjustment. It would be interesting to have similar classifications of Ω and Λ in the cases char K = 2
and char K = 3.
Some entries in the tables from Theorem 1.11 specify a single algebra. In many cases, this is an
artifact of parametrization: the algebra should actually be a member of, say, a variety of algebras
from Ω featuring in another row of the same table, only excluded for one reason or another. For
example, if all members of a variety except for one algebra are non-PBWB, while the exceptional
algebra is PBWB, the latter will occupy its own row, while the parameters corresponding to it will
feature as exceptional in the row describing the variety. Sometimes the reason is different. If a variety
of algebras is naturally parameterized by the projective plane KP2, then it will occupy three rows
in our table: one parameterized by an affine plane, one parameterized by an affine line and a single
point. Although soundness of such practice is debatable, we have decided to parameterize by numbers
rather than equivalence classes of any sort. However there are few single algebras in the tables from
Theorem 1.11, which a genuinely isolated points of the 'scheme' of isomorphism classes of algebras
from Ω or Λ. For example the only 'isolated' twisted potential algebras in Ω are (T12–T15). Note
also that the algebra in (T12) is isomorphic to the opposite of (T13) and the same holds for the pair
(T14) and (T15). Thus we essentially have twodifferent algebras here: (T12) and (T14). We wonder
if there is something special about them apart from being isolated in Ω.
12.1 Regular algebras in Ω and Λ
It turns out that regular (in the Artin–Schelter sense [1]) algebras in Ω and Λ are precisely the twisted
potential ones (including potential). This is not really surprising since it is an easy consequence of the
definition of regular algebras. However there is a surprising bit as well. Artin, Tate and Van den Bergh
[3] show that Sklyanin algebras (=algebras from (P1)) are all domains. One can use the same technique
(and/or elementary arguments in most cases) to show that all twisted potential algebras featuring in
Theorem 1.11 are domains: these are (P1–P8), (F1–F4), (T1–T18) and (G1–G10). Curiously, no
other domains occur in Theorem 1.11. All other algebras have zero divisors of degree 1 or in rare
cases 2. Thus for algebras in Ω or Λ, being a domain is the same as being twisted potential. We
believe there should be a way to prove this equivalence other than working through the tables in
Theorem 1.11. We also conjecture that this fact goes beyond Ω and Λ. Namely, we conjecture that
if (n, k) ∈ N2, n ⩾ 2, k ⩾ 2 and (n, k) ≠ (2, 2), then a quasipotential algebra A with the Hilbert series
HA =(1 − nt + ntk − tk+1)−1 is a domain if and only if it is twisted potential. By the way, we might as
well keep the case (n, k) =(2, 2) in. It was excluded because the class of algebras in question is empty
cases (n, k) =(3, 2) and (n, k) =(2, 3).
(it is non-empty in all other cases). The above comment shows that the answer is affirmative in the
12.2 Hilbert series of algebras in Ω′
In [13], the authors have shown that there are just finitely many series (11 to be precise), featuring as
Hilbert series of quadratic algebras A satisfying dim A1 = dim A2 = 3. This class of algebras coincides
with the class of duals of quadratic algebras A = A(V, R) with dim V = dim R = 3. However, along
the way we have stumbled upon enough algebras to conclude that infinitely many different Hilbert
series occur for A ∈ Ω′. This fact was already applied by the authors. For example, the exceptions for
the family (N1) exhibit infinitely many different Hilbert series. Furthermore, we have already applied
78
the classification of Theorem 1.11. Namely, we constructed [14] (found among algebras in (N1) to be
precise) an automaton algebra, which fails to have a finite Grobner basis for any choice of generators
and a compatible order on monomials.
Acknowledgements
We are grateful to IHES and MPIM for hospitality, support, and excellent research atmosphere.
This work was partially funded by the ERC grant 320974, EPSRC grant EP/M008460/1 and the ESC
Grant N9038.
References
[1] M. Artin and W. Schelter, Graded algebras of global dimension 3, Adv. in Math. 66 (1987), 171–216
[2] M. Artin, J. Tate and M. Van den Bergh, Modules over regular algebras of dimension 3, Invent. Math. 106
(1991), 335–388
[3] M. Artin, J. Tate and M. Van den Bergh, Some algebras associated to automorphisms of elliptic curves,
The Grothendieck Festschrift I, 33–85, Progr. Math. 86, Birkhauser, Boston 1990
[4] R. Bocklandt, Graded Calabi Yau algebras of dimension 3, J. Pure and Applied Algebra 212(2008), 14–32
[5] R. Bocklandt, T. Schedler and M. Wemyss, Superpotentials and higher order derivations. J. Pure Appl.
Algebra 214 (2010), no. 9, 1501–1522
[6] V. Drinfeld, On quadratic quasi-commutational relations in quasi-classical limit, Selecta Math. Sovietica
11 (1992), 317–326
[7] M. Dubois-Violette, Multilinear forms and graded algebras,J. Algebra 317 (2007), 198-225
[8] M. Dubois-Violette, Graded algebras and multilinear forms. C. R. Math. Acad. Sci. Paris 341 (2005),
719-724
[9] V. Ginzburg, Calabi Yau algebras, ArXiv:math/0612139v3, 2007
[10] G. Gurevich, Foundations of the theory of algebraic invariants, Noordhoff, 1964
[11] N. Iyudu and S. Shkarin, Sklyanin Algebras and qubic root of unity, Max-Planck-Institute fur Mathematic
preprint series, 49 (2017)
[12] N. Iyudu, S. Shkarin, Three dimensional Sklyanin algebras and Grobner bases, J.Algebra, 470 (2017),
p.378-419,
[13] N. Iyudu and S. Shkarin, Potential algebras with few generators, arXiv:math/2301025.
[14] N. Iyudu and S. Shkarin, Automaton algebras and intermediate growth, J.of Combinatorial algebra, V
2(2018), N2, 147-167
[15] M. Kontsevich, Formal (non)commutative symplectic geometry, The Gelfand Math. Seminars (Paris 1992)
97–121, Progr. Math. 120, Birkhauser, Basel, 1994
[16] K. Kraft, Geometrische Methoden in Invarianttheorie, Friedr. Vieweg&Sohn, Brauunschweig, 1985
[17] A. Odesskii, Elliptic algebras, Russian Math. Surveys 57 (2002), 1127-1162
[18] A. Polishchuk and L. Positselski, Quadratic algebras, University Lecture Series 37 American Mathematical
Society, Providence, RI, 2005
[19] I. Mori and S. P. Smith, The classification of 3-Calabi-Yau algberas with 3 generators and 3 quadratic
relations,Math. Z. 287 (2017), no. 1-2, 215241.
[20] I. Mori and K. Ueyama, The classification of 3-dimensional noetherian cubic Calabi-Yau algebras,
arXiv:1606.00183. 1606.00183.
[21] V. Ufnarovski, Combinatorial and Asymptotic Methods in Algebra, Encyclopaedia of Mathematical Sciences
57, Editors: A. Kostrikin and I. Shafarevich, Berlin, Heidelberg, New York: Springer-Verlag (1995), 1–196
[22] C. Walton, Representation theory of three-dimensional Sklyanin algebras, Nuclear Phys. B860 (2012),
167–185
79
|
1302.6291 | 1 | 1302 | 2013-02-26T01:31:48 | Cohomologies and Deformations of Generalized Left-symmetric Algebras | [
"math.RA",
"math-ph",
"math-ph"
] | The purpose of this paper is to develop a cohomology and deformation theories for generalized left-symmetric algebras.We introduce the notions of generalized left-symmetric cohomology and deformation. We also generalize a theorem of Dzhumadil'daev on connections between the right-symmetric cohomology and Chevalley-Eilenberg cohomology. As an application, we obtain a factorization theorem in left-symmetric superalgebras cohomology. Finally, we obtain all complex simple left-symmetric superalgebras of dimension 3 by the infinitesimal deformations of a given left-symmetric superalgebras. | math.RA | math |
COHOMOLOGIES AND DEFORMATIONS OF GENERALIZED LEFT-SYMMETRIC
ALGEBRAS
RUNXUAN ZHANG
Abstract. The purpose of this paper is to develop a cohomology and deformation theories for generalized
left-symmetric algebras. We introduce the notions of generalized left-symmetric cohomology and defor-
mation. We also generalize a theorem of Dzhumadil'daev on connections between the right-symmetric co-
homology and Chevalley-Eilenberg cohomology. As an application, we obtain a factorization theorem in
left-symmetric superalgebras cohomology. Finally, we obtain all complex simple left-symmetric superalge-
bras of dimension 3 by the infinitesimal deformations of a given left-symmetric superalgebras.
1. Introduction
The supersymmetry theory is an active and interesting research object of mathematics and mathematical
physics; its mathematical foundation is the theory of Lie superalgebras as well as of Lie supergroups and
of supermanifolds. The generalized Lie algebras, which include Lie algebras and Lie superalgebras as
special cases, have been described in [24, 26, 27, 16, 17] from algebraic point of view and has been
introduced into physics in 1978 ([25]).
Let Γ be an abelian group and k∗ denote the multiplicative group of a field k. Then a mapping ǫ :
Γ × Γ → k∗ is called a commutation factor on Γ if for all α, β, γ ∈ Γ,
(1.1)
(1.2)
(1.3)
ǫ(α, β)ǫ(β, α) = 1,
ǫ(α, β + γ) = ǫ(α, β)ǫ(α, γ),
ǫ(α + β, γ) = ǫ(α, γ)ǫ(β, γ).
A Γ-graded algebra L = ⊕γ∈ΓLγ with the multiplication (x, y) 7→ [x, y] is called a generalized Lie algebra
(or ǫ-Lie algebra) if the following identities are satisfied:
(1.4)
(1.5)
[x, y] + ǫ(α, β)[y, x] = 0 (ǫ-skew symmetry),
ǫ(α, γ)[x, [y, z]] + cyclic = 0 (ǫ-Jacobi identity),
for all x ∈ Lα, y ∈ Lβ, z ∈ Lγ, α, β, γ ∈ Γ. On the other hand, a non-associative algebra S with the
multiplication (x, y) 7→ x · y is called a left-symmetric algebra (or pre-Lie algebra) ([1, 4]) if the following
identity is satisfied:
(1.6)
(x · y) · z − x · (y · z) = (y · x) · z − y · (x · z)
for all x, y, z ∈ S . Since left-symmetric algebras are a kind of Lie admissible algebras, there are some close
connections between left-symmetric algebra and Lie algebra theories. For example, if G is a connected and
Date: November 21, 2018.
2010 Mathematics Subject Classification. 17B56,17B70,17D25.
Key words and phrases. cohomology; deformation; generalized left-symmetric algebras.
1
2
RUNXUAN ZHANG
simply connected Lie group with Lie algebra g, then the isomorphism classes of left-symmetric algebraic
structures on g correspond to left-invariant flat affine structures on G.
The purpose of this paper is to introduce the notion of generalized left-symmetric algebras and to
investigate their cohomology and deformation theories. The cohomology theory of Lie algebras as an
important tool originated in the works of Cartan and was developed by Chevalley and Eilenberg [5], Koszul
[18], and Hochschild and Serre [14]. Moreover, the cohomological constructions of Lie superalgebra
and generalized Lie algebra first appeared in [20] and [22] respectively.
(see [2, 10, 28, 19] for the
calculations and applications of generalized Lie algebra cohomology.) In 1999, Dzhumadil'daev [6] has
defined cohomology groups Hn
rsym(S , M) for a right-symmetric algebra S and an antisymmetric module
M and moreover, it was proved that the right-symmetric cohomology can be deduced from corresponding
Lie algebra cohomology. In this paper, we will generalize the result of Dzhumadil'daev to the generalized
left-symmetry algebra cohomology case. In particular, we show that
(1.7)
Hi+1
lsym(S , M) (cid:27) Hi
Lie(gS , C1(S , M)) for all i > 0,
where gs denotes the associated ǫ-Lie algebra of S . Applying the Hochschild-Serre factorization theorem
in Lie superalgebras ([2]), we obtain a factorization theorem (Theorem 4.3) in left-symmetric superalge-
bras cohomology.
The formal deformation theory was introduced by Gerstenhaber ([11]) for associative algebras in 1960'.
The fundamental results of Gerstenhaber's theory connect deformation theory with the suitable coho-
mology groups. Nowadays deformation-theoretic ideas penetrate most aspects of both mathematics and
physics and cut to the core of theoretical and computational problems ([3, 7, 8, 9, 12, 23, 21, 13, 32]). In
this paper, we also develop a formal deformation theory for generalized left-symmetric algebras. As an ap-
plication, we obtain all complex simple left-symmetric superalgebras of dimension 3 by the infinitesimal
deformations of a given 3-dimensional left-symmetric superalgebras.
The present paper is organized as follows. In Section 2, we introduce the notions of the generalized
left-symmetric algebra and its bimodule. A few examples are presented. We also define a generalized
left-symmetric bimodule structure on the tensor product M ⊗ N of two bimodules M and N. In partic-
ular, one can endow Ci+1(S , M) = Hom(∧i
ǫS ⊗ S , M), i ≥ 0 with an antisymmetric bimodule structure.
In Section 3, we prove that C(S , M) = ⊕i≥0Ci+1(S , M) has the structure of an ǫ-pre-simplicial cochain
complex. The notions of generalized left-symmetric coboundary, cocycle and cohomology spaces are de-
fined. Section 4 deals with a generalization of a theorem due to Dzhumadil'daev on connections between
the left-symmetric cohomology and Chevalley-Eilenberg cohomology. In Section 5, the generalized left-
symmetric deformations are introduced in the sense of Gerstenhaber. We determine all complex simple
left-symmetric superalgebras that can be obtained by the infinitesimal deformations of a given simple
left-symmetric superalgebra of dimension 3. They are just all simple left-symmetric superalgebras of
dimension 3 in [29] (or see [31]).
Throughout this paper all vector spaces and (super)algebras are assumed to be finite-dimensional over
a field k of characteristic zero.
2. Generalized Left-symmetric Algebras and Bimodules
We first introduce the notion of generalized left-symmetric algebra.
COHOMOLOGIES AND DEFORMATIONS OF GENERALIZED LEFT-SYMMETRIC ALGEBRAS
3
Definition 2.1. Let Γ be an abelian group and ǫ be a commutation factor on Γ. A Γ-graded nonassociative
algebra S = ⊕γ∈ΓS γ with the multiplication (x, y) 7→ x · y satisfying
(2.1)
S α · S β ⊆ S α+β,
is called a generalized left-symmetric algebra (or a generalized left-symmetric algebra) if the following
identity is satisfied:
(2.2)
(x · y) · z − x · (y · z) = ǫ(α, β)((y · x) · z − y · (x · z))
for all x ∈ S α, y ∈ S β, z ∈ S , α, β ∈ Γ. (x, y, z) := (x · y) · z − x · (y · z) is called the associator of S .
Example 2.2. 1. If we choose ǫ to be the trivial commutation factor, that is ǫ(α, β) = 1 for all α, β ∈ Γ.
Then a Γ-graded generalized left-symmetric algebra is nothing but a Γ-graded left-symmetric algebra.
2. If Γ = Z2 is the additive group of integers modulo 2 and ǫ(α, β) := (−1)αβ for all α, β ∈ Z2, then
Z2-graded generalized left-symmetric algebras are just left-symmetric superalgebras (see [30] or [15]).
3. Let V be a Γ-graded vector space and ǫ be an arbitrary commutation factor on Γ. Then the sum of
the subspaces of all homogeneous k-linear mappings from V to V with degree γ
(2.3)
Hom(V, V) := ⊕γ∈ΓHom(V, V)γ,
is a Γ-graded generalized left-symmetric algebra with respect to the usual composition of linear mappings.
In fact, it is a Γ-graded associative algebra and denoted by gl(V, ǫ).
A generalized left-symmetric algebra S is an ǫ-Lie admissible algebra, that is, for all x ∈ S α, y ∈ S β
and α, β ∈ Γ, the ǫ-commutator
(2.4)
[x, y] := x · y − ǫ(α, β)y · x
makes S to become an ǫ-Lie algebra, which is called the associated ǫ-Lie algebra of S and denoted by gS .
Remark 2.3. We notice that every generalized left-symmetric algebra S has a natural Z2-gradation. In
fact, it follows from (1.1) that the mapping
(2.5)
φ : Γ → k∗, α 7→ ǫ(α, α)
is a homomorphism of groups and for all α in Γ, we have ǫ(α, α) = ±1. Let us define
(2.6)
(2.7)
Γ¯0
Γ¯1
:= {α ∈ Γǫ(α, α) = 1},
:= {α ∈ Γǫ(α, α) = −1}.
Then Γ¯0 = Kerφ is a subgroup of Γ and it follows that the decomposition
(2.8)
is a Z2-gradation of S , where S i := ⊕α∈ΓiS α for i = ¯0, ¯1.
S = S ¯0 ⊕ S ¯1,
Definition 2.4. A Γ-graded vector space M is said to be a bimodule over a generalized left-symmetric
algebra S if it is endowed with a left action A × M → M, (x, m) 7→ x · m and a right action M × A →
M, (m, x) 7→ m · x such that
(2.9)
(2.10)
(x · y) · m − x · (y · m) = ǫ(α, β)((y · x) · m − y · (x · m))
(x · m) · y − x · (m · y) = ǫ(α, γ)((m · x) · y − m · (x · y))
4
RUNXUAN ZHANG
for all x ∈ S α, y ∈ S β, m ∈ Mγ and α, β, γ ∈ Γ.
Definition 2.5. A generalized left-symmetric S -bimodule M is said to be antisymmetric if the right action
of S is trivial, that is m · x = 0 for all x ∈ S and m ∈ M.
A generalized left-symmetric S -bimodule M is said to be special if the left action of S is associative,
that is, (x · y) · m − x · (y · m) = 0 for all x, y ∈ S and m ∈ M.
Example 2.6. Any generalized left-symmetric algebra S can be endowed with a natural S -bimodule
structure, (x, m) 7→ x · m, (m, x) 7→ m · x for all x, m ∈ S . In this case, S is called the regular bimodule.
Moreover, it is easy to check that the regular bimodule of gl(V, ǫ) is special.
Proposition 2.7. Let S be a generalized left-symmetric algebra and M be an S -bimodule. Then the graded
space of k-linear maps
(2.11)
C1(S , M) := Hom(S , M) = ⊕γ∈ΓHomγ(S , M)
is an antisymmetric S -bimodule if we define the left action as follows:
(2.12)
(x · f )(y) := x · f (y) − ǫ(α, γ) f (x · y) + ǫ(α, γ) f (x) · y,
where x ∈ S α, y ∈ S β, f ∈ Homγ(S , M) and α, β, γ ∈ Γ.
Proof. Let f ∈ Homγ(S , M) and x, y, z belong to S α, S β, S respectively, then
((x · y) · f )(z) = (x · y) · f (z) − ǫ(α + β, γ) f ((x · y) · z) + ǫ(α + β, γ) f (x · y) · z,
((y · x) · f )(z) = (y · x) · f (z) − ǫ(α + β, γ) f ((y · x) · z) + ǫ(α + β, γ) f (y · x) · z,
and
(x · (y · f ))(z)
= x · ((y · f )(z)) − ǫ(α, β + γ)(y · f )(x · z) + ǫ(α, β + γ)((y · f )(x)) · z,
= x ·(cid:0)y · f (z) − ǫ(β, γ) f (y · z) + ǫ(β, γ) f (y) · z(cid:1) −
ǫ(α, β + γ)(y · f (x · z) − ǫ(β, γ) f (y · (x · z)) + ǫ(β, γ) f (y) · (x · z)) +
ǫ(α, β + γ)(y · f (x) − ǫ(β, γ) f (y · x) + ǫ(β, γ) f (y) · x) · z.
Similarly, we have
(y · (x · f ))(z)
= y ·(cid:0)x · f (z) − ǫ(α, γ) f (x · z) + ǫ(α, γ) f (x) · z(cid:1) −
ǫ(β, α + γ)(x · f (y · z) − ǫ(α, γ) f (x · (y · z)) + ǫ(α, γ) f (x) · (y · z)) +
ǫ(β, α + γ)(x · f (y) − ǫ(α, γ) f (x · y) + ǫ(α, γ) f (x) · y) · z.
The direct computation shows that
((x · y) · f )(z) − (x · (y · f ))(z) = ǫ(α, β)(((y · x) · f )(z) − (y · (x · f ))(z)).
Hence, C1(S , M) is an antisymmetric S -bimodule.
(cid:3)
COHOMOLOGIES AND DEFORMATIONS OF GENERALIZED LEFT-SYMMETRIC ALGEBRAS
5
Definition 2.8. If g is an ǫ-Lie algebra over a field k, then a left g-module is a Γ-graded vector space M
together with a multiplication g × M → M, (x, m) 7→ [x, m] satisfying the axioms:
(1) (x, m) 7→ [x, m] is linear in x and in m;
(2) [[x, y], m] = [x, [y, m]] − ǫ(α, β)[y, [x, m]] for all x ∈ gα and y ∈ gβ.
A generalized left-symmetric S -bimodule M can be endowed with an ǫ-Lie module structure over gS
by the action
(2.13)
[x, m] := x · m − ǫ(β, α)m · x
for all x ∈ gα and m ∈ Mβ. In this situation, we denote the associated ǫ-Lie module by M. Thus if
one wants to define a generalized left-symmetric S -bimodule structure over a Γ-graded vector space M,
we should first give a left ǫ-Lie module on M over gS and endow it with a right action that satisfies the
conditions in the definition of S -bimodule.
Proposition 2.9. Let M and N be two generalized left-symmetric S -bimodules. Then the tensor product
M ⊗ N is an S -bimodule if we define the left and right actions as follows:
(2.14)
(2.15)
x · (m ⊗ n)
:= (x · m − ǫ(α, β)m · x) ⊗ n + ǫ(α, β)m ⊗ x · n
(m ⊗ n) · x
:= m ⊗ n · x,
where x, m, n belong to S α, Mβ, N respectively.
Proof. For arbitrary x ∈ S α, y ∈ S β, m ∈ Mγ and n ∈ Nδ, it is clear that
(x · (m ⊗ n)) · y − x · ((m ⊗ n) · y) = ǫ(α, γ + δ)(((m ⊗ n) · x) · y − (m ⊗ n) · (x · y)).
Next it suffices to check that
(x · y) · (m ⊗ n) − x · (y · (m ⊗ n)) = ǫ(α, β)((y · x) · (m ⊗ n) − y · (x · (m ⊗ n))).
In fact,
and
(x · y) · (m ⊗ n) = (x · y) · m ⊗ n − ǫ(α + β, γ)m · (x · y) ⊗ n + ǫ(α + β, γ)m ⊗ (x · y) · n
x · (y · (m ⊗ n))
= x · (y · m ⊗ n − ǫ(β, γ)m · y ⊗ n + ǫ(β, γ)m ⊗ y · n)
= x · (y · m) ⊗ n − ǫ(α, β + γ)(y · m) · x ⊗ n + ǫ(α, β + γ)y · m ⊗ x · n −
ǫ(β, γ)(x · (m · y) ⊗ n − ǫ(α, β + γ)(m · y) · x ⊗ n + ǫ(α, β + γ)m · y ⊗ x · n) +
ǫ(β, γ)(x · m ⊗ y · n − ǫ(α, γ)m · x ⊗ y · n + ǫ(α, γ)m ⊗ x · (y · n)).
Similarly, we have
(y · x) · (m ⊗ n) = (y · x) · m ⊗ n − ǫ(α + β, γ)m · (y · x) ⊗ n + ǫ(α + β, γ)m ⊗ (y · x) · n
and
y · (x · (m ⊗ n))
= y · (x · m) ⊗ n − ǫ(β, α + γ)(x · m) · y ⊗ n + ǫ(β, α + γ)x · m ⊗ y · n −
ǫ(α, γ)(y · (m · x) ⊗ n − ǫ(β, α + γ)(m · x) · y ⊗ n + ǫ(β, α + γ)m · x ⊗ y · n) +
6
RUNXUAN ZHANG
ǫ(α, γ)(y · m ⊗ x · n − ǫ(β, γ)m · y ⊗ x · n + ǫ(β, γ)m ⊗ y · (x · n)).
The direct computation yields the desired result.
(cid:3)
Let V be a Γ-graded vector space, then the tensor algebra (see [27])
(2.16)
T (V) := ⊕i∈ZTi(V)
is a Z × Γ-graded associative algebra, where Ti(V) = {0} for all i < 0, T0(V) = k and
for all i > 0. Let I(V, ǫ) denote the two-sided ideal of T (V) generated by the elements of the form
Ti(V) = V ⊗ · · · ⊗ V
{z }
i factors
(2.17)
m ⊗ n + ǫ(α, β)n ⊗ m, m ∈ Vα, n ∈ Vβ.
Then the quotient algebra ∧ǫV := T (V)/I(V, ǫ) is called the ǫ-exterior algebra of V and the multiplication
in ∧ǫV is denoted by ∧ǫ. On the other hand, I(V, ǫ) is a Z × Γ-graded ideal of T (V) and thus the quotient
algebra ∧ǫV inherits from T (V) a canonical Z × Γ-gradation. In particular, if we write ∧i
ǫV = V ∧ǫ · · · ∧ǫ
V(i factors) for the canonical image of Ti(V) in ∧ǫV, then
∧ǫ V = ⊕i∈Z ∧i
(2.18)
where ∧i
ǫV = 0 for all i < 0, ∧0
ǫ (V) = k and ∧1
ǫ V,
ǫ (V) = V. Define
(2.19)
Ci+1(S , M) := Hom(∧i
ǫS ⊗ S , M) = ⊕γ∈ΓHomγ(∧i
ǫS ⊗ S , M), i ≥ 0.
Proposition 2.10. For each i ≥ 0, Ci+1(S , M) is an antisymmetric S -bimodule if we endow it with the
following left action:
(2.20)
S × Ci+1(S , M) → Ci+1(S , M), (x, f ) 7→ x · f,
where f ∈ Homβ(∧i
ǫS ⊗ S , M), x ∈ S α, xs ∈ S αs(s = 1, · · · , i + 1) and
(x · f )(x1, · · · , xi, xi+1)
= x · f (x1, · · · , xi, xi+1) − ǫ(α, β + α1 + · · · + αi) f (x1, · · · , xi, x · xi+1)
+ǫ(α, β + α1 + · · · + αi) f (x1, · · · , xi, x) · xi+1
iX
−
ǫ(α, β + α1 + · · · + αs−1) f (x1, · · · , xs−1, [x, xs], · · · , xi, xi+1).
s=1
Proof. Let gS denote the associated ǫ-Lie algebra. Note that for each i ≥ 0, there is an isomorphism η
which is homogeneous of Γ-degree zero of vector spaces:
η : Ci(gS , k) ⊗ C1(S , M) → Ci+1(S , M), g ⊗ h 7→ η(g ⊗ h),
where
(2.21)
Ci(gS , k) := Hom(∧i
ǫ gS , k) = ⊕γ∈ΓHomγ(∧i
ǫ gS , k), i ≥ 0
is an ǫ-Lie module of gS (see [28] for details) and for each g ∈ Homγ(∧i
ǫ gS , k), h ∈ Homδ(S , M),
η(g ⊗ h)(x1, · · · , xi, xi+1) := ǫ(δ, α1 + · · · + αi)g(x1, · · · , xi)h(xi+1).
Thus the antisymmetric generalized left-symmetric bimodule structure on Ci(gS , k) ⊗C1(S , M) will give
an antisymmetric generalized left-symmetric bimodule structure on Ci+1(S , M) by the transformation of
COHOMOLOGIES AND DEFORMATIONS OF GENERALIZED LEFT-SYMMETRIC ALGEBRAS
7
η. On the other hand, the antisymmetric S -bimodule structure on C1(S , M) is constructed in Proposition
2.7 and the ǫ-Lie module structure on Ci(gS , k) over gS is well known. Hence for h ∈ Homδ(S , M) and
g ∈ Homγ(∧i
ǫ gS , k), we define a left action of S on Ci(gS , k) ⊗ C1(S , M) by:
η(x · (g ⊗ h))(x1, · · · , xi, xi+1)
= η([x, g] ⊗ h + ǫ(α, γ)g ⊗ x · h)(x1, · · · , xi, xi+1)
= −
ǫ(α, γ + α1 + · · · + αs−1)ǫ(δ, α1 + · · · + αi)g(x1, · · · , [x, xs], · · · , xi)h(xi+1) +
s=1
iX
ǫ(α, γ)ǫ(α + δ, α1 + · · · + αi)g(x1, · · · , xi) ×
(x · h(xi+1) − ǫ(α, δ)h(x · xi+1) + ǫ(α, δ)h(x) · xi+1).
The direct calculations show that η(x · (g ⊗ h)) = x · (η(g ⊗ h)). Therefore (x, f ) 7→ x · f is an antisymmetric
S -bimodule.
(cid:3)
3. Cohomology of Generalized Left-symmetric Algebras
Let (S , Γ, ǫ) be a generalized left-symmetric algebra and M be an S -bimodule. We define eC(S , M) :=
⊕i≥0Ci+1(S , M), where Ci+1(S , M) is defined as in (2.19). Now we present an ǫ-pre-simplicial cochain
complex on eC(S , M). We define homogeneous linear mappings Dt of degree zero with respect to the
Γ-gradation on eC(S , M) by Dt : Ci(S , M) → Ci+1(S , M), (t = 1, 2, 3, · · · .)
(3.1)
(Dt f )(x1, · · · , xi, xi+1)
= ǫ(β + α1 + · · · + αt−1, αt)xt · f (x1, · · · , xt, · · · , xi, xi+1)
−ǫ(αt, αt+1 + · · · + αi) f (x1, · · · , xt, · · · , xi, xt · xi+1)
+ǫ(αt, αt+1 + · · · + αi) f (x1, · · · , xt, · · · , xi, xt) · xi+1
−X
t< j
ǫ(αt, αt+1 + · · · + α j−1) f (x1, · · · , xt, · · · , x j−1, [xt, x j], · · · , xi, xi+1)
for t ≤ i, i ≥ 1 and
(3.2)
Dt f = 0 for t > i,
where f ∈ Homβ(∧i−1
ǫ S ⊗ S , M), xs ∈ S αs(s = 1, · · · , i + 1). Here x means that the element x is omitted.
Proposition 3.1. For all 1 ≤ s < t, we have DtDs = DsDt−1. That is, the set of Di endows eC(S , M) =
⊕i≥1Ci(S , M) with an ǫ-pre-simplicial structure.
Proof. With the notations as above, for s < t, t > i, DtDs = DsDt−1 = 0. For 1 ≤ s < t ≤ i, by (3.1), we
have
(DtDs f )(x1, · · · , xi, xi+1)
= ǫ(β + α1 + · · · + αt−1, αt)xt · (Ds f )(x1, · · · , xt, · · · , xi, xi+1) −
ǫ(αt, αt+1 + · · · + αi)(Ds f )(x1, · · · , xt, · · · , xi, xt · xi+1) +
ǫ(αt, αt+1 + · · · + αi)(Ds f )(x1, · · · , xt, · · · , xi, xt) · xi+1 −
X
ǫ(αt, αt+1 + · · · + α j−1)(Ds f )(x1, · · · , xt, · · · , x j−1, [xt, x j], · · · , xi, xi+1),
t< j
8
where
RUNXUAN ZHANG
xt · (Ds f )(x1, · · · , xt, · · · , xi, xi+1)
= ǫ(β + α1 + · · · + αs−1, αs)xt · (xs · f (x1, · · · , xs, · · · , xt, · · · , xi, xi+1))
−ǫ(αs, αs+1 + · · · + αt + · · · + αi)xt · f (x1, · · · , xs, · · · , xt, · · · , xi, xs · xi+1)
+ǫ(αs, αs+1 + · · · + αt + · · · + αi)xt · ( f (x1, · · · , xs, · · · , xt, · · · , xi, xs) · xi+1)
− X
− X
ǫ(αs, αs+1 + · · · + α j−1)xt · f (x1, · · · , xs, · · · , [xs, x j], · · · , xt, · · · , xi, xi+1)
ǫ(αs, αs+1 + · · · + αt + · · · + α j−1)xt · f (x1, · · · , xs, · · · , xt, · · · , [xs, x j], · · · , xi+1),
s< j<t
s<t< j
(Ds f )(x1, · · · , xt, · · · , xi, xt · xi+1)
= ǫ(β + α1 + · · · + αs−1, αs)xs · f (x1, · · · , xs, · · · , xt, · · · , xi, xt · xi+1)
−ǫ(αs, αs+1 + · · · + αt + · · · + αi) f (x1, · · · , xs, · · · , xt, · · · , xi, xs · (xt · xi+1))
+ǫ(αs, αs+1 + · · · + αt + · · · + αi) f (x1, · · · , xs, · · · , xt, · · · , xi, xs) · (xt · xi+1)
− X
− X
ǫ(αs, αs+1 + · · · + α j−1) f (x1, · · · , xs, · · · , [xs, x j], · · · , xt, · · · , xi, xt · xi+1)
ǫ(αs, αs+1 + · · · + αt + · · · + α j−1) f (x1, · · · , xs, · · · , xt, · · · , [xs, x j], · · · , xt · xi+1),
s< j<t
s<t< j
(Ds f )(x1, · · · , xt, · · · , xi, xt) · xi+1
= ǫ(β + α1 + · · · + αs−1, αs)(xs · f (x1, · · · , xs, · · · , xt, · · · , xi, xt)) · xi+1
ǫ(αs, αs+1 + · · · + αt + · · · + α j−1) f (x1, · · · , xs, · · · , xt, · · · , [xs, x j], · · · , xt) · xi+1,
ǫ(αs, αs+1 + · · · + α j−1) f (x1, · · · , xs, · · · , [xs, x j], · · · , xt, · · · , xi, xt) · xi+1
s< j<t
s<t< j
ǫ(β + α1 + · · · + αs−1, αs)xs · f (x1, · · · , xs, · · · , xt, · · · , [xt, x j], · · · , xi, xi+1)
(Ds f )(x1, · · · , xt, · · · , x j−1, [xt, x j], · · · , xi, xi+1)
−ǫ(αs, αs+1 + · · · + αt + · · · + αi) f (x1, · · · , xs, · · · , xt, · · · , xi, xs · xt) · xi+1
+ǫ(αs, αs+1 + · · · + αt + · · · + αi)( f (x1, · · · , xs, · · · , xt, · · · , xi, xs) · xt) · xi+1
− X
− X
X
= X
−X
+X
− X
− X
−X
ǫ(αs, αs+1 + · · · + αi) f (x1, · · · , xs, · · · , xt, · · · , [xt, x j], · · · , xi, xs · xi+1)
ǫ(αs, αs+1 + · · · + αi) f (x1, · · · , xs, · · · , xt, · · · , [xt, x j], · · · , xi, xs) · xi+1
s< j1,t< j< j1
s< j1<t< j
t< j
t< j
t< j
t< j
t< j
ǫ(αs, αs+1 + · · · + α j1−1) f (x1, · · · , xs, · · · , [xs, x j1], · · · , xt, · · · , [xt, x j], · · · , xi+1)
ǫ(αs, αs+1 + · · · + α j1−1) f (x1, · · · , xs, · · · , xt, · · · , [xt, x j], · · · , [xs, x j1], · · · , xi+1)
ǫ(αs, αs+1 + · · · + αt + · · · + α j−1) f (x1, · · · , xs, · · · , xt, · · · , [xs, [xt, x j]], · · · , xi+1)
COHOMOLOGIES AND DEFORMATIONS OF GENERALIZED LEFT-SYMMETRIC ALGEBRAS
9
ǫ(αs, αs+1 + · · · + αt + · · · + α j1−1)
− X
s< j1,t< j1< j
× f (x1, · · · , xs, · · · , xt, · · · , [xs, x j1], · · · , [xt, x j], · · · , xi+1).
In a similar way, we have
(DsDt−1 f )(x1, · · · , xi, xi+1)
= ǫ(β + α1 + · · · + αs−1, αs)xs · (Dt−1 f )(x1, · · · , xs, · · · , xi, xi+1)
−ǫ(αs, αs+1 + · · · + αi)(Dt−1 f )(x1, · · · , xs, · · · , xi, xs · xi+1)
+ǫ(αs, αs+1 + · · · + αi)(Dt−1 f )(x1, · · · , xs, · · · , xi, xs) · xi+1
−X
ǫ(αs, αs+1 + · · · + α j−1)(Dt−1 f )(x1, · · · , xs, · · · , x j−1, [xs, x j], · · · , xi, xi+1),
s< j
where
xs · (Dt−1 f )(x1, · · · , xs, · · · , xi, xi+1)
= ǫ(β + α1 + · · · + αs + · · · + αt−1, αt)xs · (xt · f (x1, · · · , xs, · · · , xt, · · · , xi, xi+1))
−ǫ(αt, αt+1 + · · · + αi)xs · f (x1, · · · , xs, · · · , xt, · · · , xi, xt · xi+1)
+ǫ(αt, αt+1 + · · · + αi)xs · ( f (x1, · · · , xs, · · · , xt, · · · , xi, xt) · xi+1)
−X
ǫ(αt, αt+1 + · · · + α j−1)xs · f (x1, · · · , xs, · · · , xt, · · · , [xt, x j], · · · , xi, xi+1),
t< j
(Dt−1 f )(x1, · · · , xs, · · · , xi, xs · xi+1)
= ǫ(β + α1 + · · · + αs + · · · + αt−1, αt)xt · f (x1, · · · , xs, · · · , xt, · · · , xi, xs · xi+1)
−ǫ(αt, αt+1 + · · · + αi) f (x1, · · · , xs, · · · , xt, · · · , xi, xt · (xs · xi+1))
+ǫ(αt, αt+1 + · · · + αi) f (x1, · · · , xs, · · · , xt, · · · , xi, xt) · (xs · xi+1)
−X
ǫ(αt, αt+1 + · · · + α j−1) f (x1, · · · , xs, · · · , xt, · · · , [xt, x j], · · · , xi, xs · xi+1),
t< j
(Dt−1 f )(x1, · · · , xs, · · · , xi, xs) · xi+1
= ǫ(β + α1 + · · · + αs + · · · + αt−1, αt)(xt · f (x1, · · · , xs, · · · , xt, · · · , xi, xs)) · xi+1
−ǫ(αt, αt+1 + · · · + αi) f (x1, · · · , xs, · · · , xt, · · · , xi, xt · xs) · xi+1
+ǫ(αt, αt+1 + · · · + αi)( f (x1, · · · , xs, · · · , xt, · · · , xi, xt) · xs) · xi+1
−X
ǫ(αt, αt+1 + · · · + α j−1) f (x1, · · · , xs, · · · , xt, · · · , [xt, x j], · · · , xi, xs) · xi+1.
t< j
We represent Ps< j(Dt−1 f )(x1, · · · , xs, · · · , x j−1, [xs, x j], · · · , xi, xi+1) as a sum of X1, X2 and X3, where
s< j<t
X1 = X
− X
+ X
s< j<t
s< j<t
ǫ(β + α1 + · · · + αt−1, αt)xt · f (x1, · · · , xs, · · · , [xs, x j], · · · , xt, · · · , xi, xi+1)
ǫ(αt, αt+1 + · · · + αi) f (x1, · · · , xs, · · · , [xs, x j], · · · , xt, · · · , xi, xt · xi+1)
ǫ(αt, αt+1 + · · · + αi) f (x1, · · · , xs, · · · , [xs, x j], · · · , xt, · · · , xi, xt) · xi+1
10
RUNXUAN ZHANG
ǫ(αt, αt+1 + · · · + α j1−1)
− X
s< j<t< j1
× f (x1, · · · , xs, · · · , [xs, x j], · · · , xt, · · · , [xt, x j1], · · · , xi, xi+1),
X2 = ǫ(β + α1 + · · · + αs + · · · + αt−1, αs + αt)[xs, xt] · f (x1, · · · , xs, · · · , xt, · · · , xi+1)
−ǫ(αs + αt, αt+1 + · · · + αi) f (x1, · · · , xs, · · · , xt, · · · , xi, [xs, xt] · xi+1)
+ǫ(αs + αt, αt+1 + · · · + αi) f (x1, · · · , xs, · · · , xt, · · · , xi, [xs, xt]) · xi+1
−X
ǫ(αs + αt, αt+1 + · · · + α j−1) f (x1, · · · , xs, · · · , xt, · · · , [[xs, xt], x j], · · · , xi+1),
t< j
ǫ(β + α1 + · · · + αs + · · · + αt−1, αt)xt · f (x1, · · · , xs, · · · , xt, · · · , [xs, x j], · · · , xi+1)
ǫ(αt, αt+1 + · · · + αi + αs) f (x1, · · · , xs, · · · , xt, · · · , [xs, x j], · · · , xi, xt · xi+1)
ǫ(αt, αt+1 + · · · + αi + αs) f (x1, · · · , xs, · · · , xt, · · · , [xs, x j], · · · , xi, xt) · xi+1
s<t< j
t< j
s<t< j
s<t< j
X3 = X
− X
+ X
− X
−X
− X
t< j< j1
× f (x1, · · · , xs, · · · , xt, · · · , [xs, x j], · · · , [xt, x j1], · · · , xi+1).
ǫ(αt, αt+1 + · · · + α j1−1 + αs)
t< j1< j
ǫ(αt, αt+1 + · · · + α j1−1) f (x1, · · · , xs, · · · , xt, · · · , [xt, x j1], · · · , [xs, x j], · · · , xi+1)
ǫ(αt, αt+1 + · · · + α j−1) f (x1, · · · , xs, · · · , xt, · · · , [xt, [xs, x j]], · · · , xi+1)
Using the identities in the definitions of a generalized left-symmetric algebra and a bimodule and by
tedious calculations, we obtain DtDs = DsDt−1, s < t.
Corollary 3.2. Let d = −Pi(−1)iDi, then d2 = 0.
Define C0(S , M) := {m ∈ M(ab)m = a(bm) for all a, b ∈ S }, it is easy to check that C0(S , M)
is an S -submodule of M. By Corollary 3.2, the cochain complex eC(S , M) = ⊕i>0Ci(S , M) with its
coboundary operator d can be extended to a new cochain complex C(S , M) = ⊕i≥0Ci(S , M) if we de-
fine d : C0(S , M) → C1(S , M) by
(cid:3)
for x ∈ S α and f ∈ C0(S , M) ∩ Mβ. Hence
d( f )(x) = ǫ(β, α)x · f − f · x
Proposition 3.3. The space C(S , M) = ⊕i≥0Ci(S , M) is a cochain complex under the following cobound-
ary operator d of degree zero with respect to the Γ-gradation of C(S , M),
where x ∈ S α, f ∈ C0(S , M) ∩ Mβ and
d( f )(x) = ǫ(β, α)x · f − f · x,
(d f )(x1, · · · , xi, xi+1)
iX
t=1
= −
(−1)tǫ(β + α1 + · · · + αt−1, αt)xt · f (x1, · · · , xt, · · · , xi, xi+1)
COHOMOLOGIES AND DEFORMATIONS OF GENERALIZED LEFT-SYMMETRIC ALGEBRAS
11
+
−
t=1
iX
iX
+ X
t=1
1≤t< j≤i
(−1)tǫ(αt, αt+1 + · · · + αi) f (x1, · · · , xt, · · · , xi, xt · xi+1)
(−1)tǫ(αt, αt+1 + · · · + αi) f (x1, · · · , xt, · · · , xi, xt) · xi+1
(−1)tǫ(αt, αt+1 + · · · + α j−1) f (x1, · · · , xt, · · · , x j−1, [xt, x j], · · · , xi, xi+1),
where f ∈ Homβ(∧i−1
ǫ S ⊗ S , M), xs ∈ S αs(s = 1, · · · , i + 1) and i ≥ 1.
We conclude this section with introducing the concept of the cohomology spaces of a generalized left-
symmetric algebra S with coefficients in an S -bimodule M.
Definition 3.4. Let Z(S , M) = ⊕i≥0Zi(S , M) with
Zi(S , M) := { f ∈ Ci(S , M)d( f ) = 0}
denote the spaces of generalized left-symmetric cocycles and B(S , M) = ⊕i≥0Bi(S , M) with
Bi(S , M) := {d(g)g ∈ Ci−1(S , M)}
denote the spaces of generalized left-symmetric coboundaries. Then H(S , M) = ⊕i≥0Hi(S , M) with
are called the generalized left-symmetric cohomology spaces with coefficients in M.
Hi(S , M) := Zi(S , M)/Bi(S , M)
One can show that d is a homomorphism of antisymmetric S -bimodule, that is, d(x · f ) = x · (d f ) for
x ∈ S , f ∈ Ci(S , M) and then Zi(S , M), Bi(S , M) are antisymmetric S -subbimodules, hence Hi(S , M) is an
antisymmetric S -bimodule.
4. Connections Between Generalized Left-symmetric and ǫ-Lie Cohomologies
Let g be an ǫ-Lie algebra, M be a left g-module and C(g, M) = ⊕i≥0Ci(g, M) be the cochain complex on
g with coboundary operator d, where
(4.1)
Ci(g, M) = Hom(∧i
ǫ g, M) = ⊕γ∈ΓHomγ(∧i
ǫ g, M).
Recall that the standard ǫ-Lie algebra representation ξ on Ci(g, M) (see [28]) is given by
(ξ(x) f )(x1, · · · , xi) = [x, f (x1, · · · , xi)]
−
iX
j=1
ǫ(α, β + α1 + · · · + α j−1) f (x1, · · · , x j−1, [x, x j], · · · , xi),
where f ∈ Homβ(∧i
ǫ g, M) and (x, m) 7→ [x, m] is a representation corresponding to the ǫ-Lie module M.
On the other hand, we endow Ci+1(S , M) as defined in (2.19) with a representation ρ of the associated
ǫ-Lie algebra gS corresponding to the antisymmetric representation constructed in Proposition 2.10:
(ρ(x) f )(x1, · · · , xi, xi+1)
= x · f (x1, · · · , xi, xi+1) − ǫ(α, β + α1 + · · · + αi) f (x1, · · · , xi, x · xi+1)
+ǫ(α, β + α1 + · · · + αi) f (x1, · · · , xi, x) · xi+1
12
RUNXUAN ZHANG
−
iX
s=1
ǫ(α, β + α1 + · · · + αs−1) f (x1, · · · , xs−1, [x, xs], · · · , xi, xi+1),
where f ∈ Homβ(∧i
structure on C1(S , M) is given by
ǫS ⊗ S , M), x ∈ (gS )α, xs ∈ S αs(s = 1, · · · , i, i + 1). In particular, the gS -module
[x, f ](x1) = x · f (x1) − ǫ(α, β) f (x · x1) + ǫ(α, β) f (x) · x1.
Theorem 4.1. Let S be a generalized left-symmetric algebra and M an S -bimodule. If we define
by
ψ : Ci(gS , C1(S , M)) → Ci+1(S , M), i > 0,
ψ( f )(x1, · · · , xi, xi+1) = f (x1, · · · , xi)(xi+1),
then ψ induces an isomorphism of gS -modules. Moreover, ψ induces an isomorphism of cochain complexes
⊕i≥1Ci(S , M) and ⊕i≥0Ci(gS , C1(S , M)). In particular, we have
Hi+1(S , M) (cid:27) Hi
Lie(gS , C1(S , M)) for all i > 0.
Proof. We first prove that for all x ∈ gS and i > 0, the following diagram is commutative:
yψ
ψy
ξ(x)
ρ(x)
Ci(gS , C1(S , M))
−−−−−→ Ci(gS , C1(S , M))
Ci+1(S , M)
−−−−−→ Ci+1(S , M),
where ρ and ξ are defined as above. In fact, for f ∈ Ci
β(gS , C1(S , M)), we have
ψ(ξ(x) f )(x1, · · · , xi, xi+1)
= [x, f (x1, · · · , xi)](xi+1)
−
iX
j=1
ǫ(α, β + α1 + · · · + α j−1) f (x1, · · · , x j−1, [x, x j], · · · , xi)(xi+1)
= x · f (x1, · · · , xi)(xi+1) − ǫ(α, β + α1 + · · · + αi) f (x1, · · · , xi)(x · xi+1)
+ǫ(α, β + α1 + · · · + αi) f (x1, · · · , xi)(x) · xi+1
iX
−
ǫ(α, β + α1 + · · · + α j−1) f (x1, · · · , x j−1, [x, x j], · · · , xi)(xi+1),
j=1
and
ρ(x)(ψ( f ))(x1, · · · , xi, xi+1)
= x · f (x1, · · · , xi)(xi+1) − ǫ(α, β + α1 + · · · + αi) f (x1, · · · , xi)(x · xi+1)
+ǫ(α, β + α1 + · · · + αi) f (x1, · · · , xi)(x) · xi+1
iX
−
ǫ(α, β + α1 + · · · + α j−1) f (x1, · · · , x j−1, [x, x j], · · · , xi)(xi+1).
j=1
Thus ψ : Ci(gS , C1(S , M)) → Ci+1(S , M) is a homomorphism of gS -modules. Moreover, ψ has no kernel
and it is an epimorphism. That is, ψ is an isomorphism.
COHOMOLOGIES AND DEFORMATIONS OF GENERALIZED LEFT-SYMMETRIC ALGEBRAS
13
With an analogous argument, we have that for all i ≥ 0, the following diagram is commutative:
Ci(gS , C1(S , M))
−−−−−→ Ci+1(gS , C1(S , M))
d
yψ
ψy
Ci+1(S , M)
d
−−−−−→
Ci+2(S , M).
Hence the cochain complexes ⊕i≥0Ci(gS , C1(S , M)) and ⊕i≥0Ci+1(S , M) are equivalent and in particular,
we have Hi+1(S , M) (cid:27) Hi
Lie(gS , C1(S , M)) for all i > 0.
(cid:3)
Remark 4.2. Note that H1(S , M) = Z1(S , M)/B1(S , M) and
Z1(S , M) = { f ∈ C1(S , M)d( f ) = 0} = Z0(gS , C1(S , M)),
B1(S , M) = {dmm ∈ M, (xy)m = x(ym) for all x, y ∈ S }.
Then it is not difficult to check that the following sequence is exact:
0 → Z0(S , M) → C0(S , M) → H0(gS , C1(S , M)) → H1(S , M) → 0.
Theorem 4.1 means that the cohomology spaces Hi(S , M)(i ≥ 2) can be calculated if we are able to find
a way to compute the corresponding ǫ-Lie cohomology spaces. In particular,
Theorem 4.3. For a left-symmetric superalgebra S and its bimodule M, we have
Hi+1(S , M) (cid:27) Hi
Lie(gS , C1(S , M)) (cid:27) X
H p(gS /I, k) ⊗ Hq(I, C1(S , M))gS ,
p+q=i
where I is a graded ideal of gS such that gS /I is a semisimple Lie algebra and Hq(I, C1(S , M))gS is a
submodule of Hq(I, C1(S , M)) annihilated by gS .
Proof. Note that the Hochschild-Serre factorization theorem in Lie superalgebras ([2]) and apply Theorem
4.1.
(cid:3)
5. Deformations of Generalized Left-symmetric Algebras
In this section, we extend Gerstenhaber's theory of formal deformation of algebras ([11]) to generalized
left-symmetric algebras.
Let S = ⊕γ∈ΓS γ be a finite-dimensional generalized left-symmetric algebra over a field k of character-
istic zero and k((λ)) denote the field of fractions for the formal power series ring k[[λ]]. We extend the
coefficient domain from k to k((λ)) and construct a bilinear map fλ on the vector space S λ = S ⊗ k((λ)) of
the form
(5.1)
fλ(x, y) = x · y + λF1(x, y) + λ2F2(x, y) + · · · ,
where Fi are bilinear maps and we may set F0(x, y) = x · y, the product in S . Suppose further that
(x, y) 7→ fλ(x, y) defines a generalized left-symmetric algebra on S λ. Then we say that (S λ, fλ) is a
one-parameter family of deformations of S . It is evident that (2.1) cannot be maintained unless fλ and
each Fi are homogenous elements of degree zero. This is to say, a deformation must leave the gradation
undisturbed.
On the other hand, the generalized left-symmetric identity (2.2) requires
(5.2)
fλ( fλ(x, y), z) − fλ(x, fλ(y, z)) = ǫ(α, β) ( fλ( fλ(y, x), z) − fλ(y, fλ(x, z)))
14
RUNXUAN ZHANG
for all x ∈ S α, y ∈ S β, z ∈ S and α, β ∈ Γ. In terms of Fi, we have for all non-negative integers p:
(5.3)
(Fr(Fs(x, y), z) − Fr(x, Fs(y, z)) − ǫ(α, β) (Fr(Fs(y, x), z) − Fr(y, Fs(x, z)))) = 0.
X
r+s=p,r,s≥0
(5.3) are known as integrability conditions. If γ = 1, we obtain an equation for F1:
F1(x, y) · z + F1(x · y, z) − x · F1(y, z) − F1(x, y · z)
= ǫ(α, β)(F1(y · x, z) + F1(y, x) · z − F1(y, x · z) − y · F1(x, z)).
Hence the first integrability condition states that F1 must be a two-cocycle. An element F ∈ Z2(S , S )
is said to be integrable if it is the first term of a one-parameter deformation series (5.1). We call F1 an
infinitesimal deformation. Putting p = 2 in (5.3), we get
where
dF2(x, y, z) = µ2(x, y, z),
µ2(x, y, z) = F1(F1(x, y), z) − F1(x, F1(y, z)) − ǫ(α, β)(F1(F1(y, x), z) − F1(y, F1(x, z)))
and d is the coboundary operator defined as (3.6). One can show that µ2 is an element of Z3(S , S ) if F1
is in Z2(S , S ). If F1 is integrable, then this three-cocycle must be a coboundary. Hence the cohomology
class of µ2 is the first obstruction to the integration of F1. In general, we have
(5.4)
where
dF p(x, y, z) = µp(x, y, z),
µp(x, y, z) = X (Fr(Fs(x, y), z) − Fr(x, Fs(y, z)) − ǫ(α, β) (Fr(Fs(y, x), z) − Fr(y, Fs(x, z))))
and r + s = p, r, s > 0. We can show that if F1, · · · , F p−1 are chosen such that the integrability condition
(5.3) holds, then µp is a three-cocycle. (This is done by appropriate modification of Dzhumadil'daev's
proof in [6] to take care of the Γ-gradation.) Thus, the obstruction to continuing the deformation to the
p-th term lies in the possibility that the cocycle µp might not be a pure coboundary. For this reason,
H3(S , S ) may be regarded as the space of obstruction to deformations of S and if H3(S , S ) = 0, then all
obstructions vanish and every two-cocycle is integrable.
Now suppose fλ = P λnFn and gλ = P λnGn are two one parameter families of deformations of a
generalized left-symmetric algebra S . We say that fλ and gλ are equivalent if there exists a nonsingular
linear automorphism Φλ of S λ of the form
where all the ϕi : S → S are homogenous linear maps of degree zero, such that for x, y ∈ S
Φλ(x) = x + λϕ1(x) + λ2ϕ2(x) + · · · ,
(5.5)
fλ(x, y) = Φ−1
λ gλ(Φλ(x), Φλ(y)).
Expanding both sides of (5.5) in powers of λ, we find
F1(x, y) − G1(x, y) = dϕ1(x, y).
Thus, if two deformations are to be equivalent, then their infinitesimal generators must belong to the
same cohomology class in Z2(S , S ). A deformation fλ is called trivial if it is equivalent to the identity
deformation, that is Fi = 0 for all i > 0 in (5.1).
COHOMOLOGIES AND DEFORMATIONS OF GENERALIZED LEFT-SYMMETRIC ALGEBRAS
15
Suppose now that fλ is a one-parameter family of deformations of S and Fn(n ≥ 1) is the first nonzero
term. Then it follows from (5.4) that dFn = 0. If further Fn is in B2(S , S ), then Fn = −dϕ for some ϕ in
C1(S , S ). Setting Φλ(x) = x + λϕ(x), we have
Φ−1
λ fλ(Φλ(x), Φλ(y)) = x · y + λn+1F′
n+1(x, y) + λn+2F′
n+2(x, y) + · · · ,
where F′
n+1, F′
fλ and again F′
n+2, · · · , are the two-cochains of degree zero defining a deformation f ′
n+1 is in Z2(S , S ). Then we obtain the following result.
λ that is equivalent to
Proposition 5.1. Let fλ be a one-parameter family of deformations of a generalized left-symmetric algebra
S . Then fλ is equivalent to gλ(x, y) = x · y + λnGn(x, y) + · · · , where the first non-vanishing cochain Gn
is in Z2(S , S ) and is not a pure coboundary. In particular, if H2(S , S ) = 0, then every deformation is
equivalent to the trivial deformation.
The study of deformations is a way to obtain new algebras. We end this paper with an example that il-
lustrates how to determine all simple left-symmetric superalgebras which can be obtained by infinitesimal
deformations of a given left-symmetric superalgebra.
Example 5.2. Consider the 3-dimensional complex left-symmetric superalgebra S with a homogeneous
basis {x, y1, y2 x ∈ S ¯0, y1, y2 ∈ S ¯1} satisfying
x · x = 2x, x · y1 = y1, x · y2 = y2, y1 · y2 = x, y2 · y1 = −x.
With respect to the basis of S the second cohomology is given by
H2(S , S ) = {F1 ∈ C2(S , S )F1(x, x) = ax, F1(x, y1) = by2,
F1(x, y2) = cy1 + ay2, others are zero, a, b, c ∈ C}.
These two-cocycles satisfy the integrability conditions (5.3). We have a first-order deformation with
fλ = F0 + λF1.
Next, we determine the simple left-symmetric superalgebra (S λ, fλ). Note that every nonzero ideal will
contain x. On the other hand, the ideal generated by x equals S λ if the vectors fλ(x, y1), fλ(x, y2) are linear
independent. Then it is not difficult to classify the simple left-symmetric superalgebras constructed by this
way: any one of them is isomorphic to one of the following left-symmetric superalgebras
(1) S λ1,t : fλ(x, x) = (t + 1)x, fλ(x, y1) = y1, fλ(x, y2) = ty2, fλ(y1, y2) = x, fλ(y2, y1) = −x, 0 < t <
1 or t = eiθ, 0 ≤ θ ≤ π;
(2) S λ2 : fλ(x, x) = 2x, fλ(x, y1) = y1, fλ(x, y2) = y1 + y2, fλ(y1, y2) = x, fλ(y2, y1) = −x.
Here S λ1,t and S λ2 contain all the 3-dimensional complex simple left-symmetric superalgebras which have
been obtained in [31].
This work was supported by NNSF of China (11226051) and the Fundamental Research Funds for the
Central Universities (11QNJJ001).
Acknowledgments
16
RUNXUAN ZHANG
References
[1] O. Baues, Left-symmetric algebras for gl(n). Trans. Amer. Math. Soc. 351 (1999) 2979-2996.
[2] B. Binegar, Cohomology and deformations of Lie superalgebras. Lett. Math. Phys. 12 (1986) 301-308.
[3] H. Bjar and O. Laudal, Deformation of Lie algebras and Lie algebras of deformations. Compos. Math. 75 (1990) 69-111.
[4] D. Burde, Left-symmetric algebras, or pre-Lie algebras in geometry and physics. Cent. Eur. J. Math. 4 (2006) 323-357.
[5] C. Chevalley and S. Eilenberg, Cohomology theory of Lie groups and Lie algebras. Trans. Amer. Math. Soc. 63 (1948)
85-124.
[6] A. Dzhumadil'daev, Cohomologies and deformations of right-symmetric algebras. J. Math. Sci. 93 (1999) 836-876.
[7] A. Fialovski, Deformation of Lie algebras. Math. USSR Sbornik 55(1986) 467-473.
[8] A. Fialowski and M. de Montigny, Deformations and contractions of Lie algebras. J. Phys. A 38 (2005) 6335-6349.
[9] A. Fialowski and M. Penkava, Deformations of four-dimensional Lie algebras. Commun. Contemp. Math. 9 (2007)
41-79.
[10] D. Fuch and D. Leıtes, Cohomology of Lie superalgebras. C. R. Acad. Bulgare Sci. 37 (1984) 1595-1596.
[11] M. Gerstenhaber, On the deformations of rings and algebras I-IV. I, Ann. Math. 79 (1964) 59-103; II, Ann. Math. 84
(1966) 1-19; III, Ann. Math. 88 (1968) 1-34; IV, Ann. Math. 99 (1974) 257-276.
[12] F. Grunewald and J. O'Halloran, Deformations of Lie algebras. J. Algebra 162 (1993) 210-224.
[13] A. Hegazi and M. Mansour, Two-parameter quantum deformation of Lie superalgebras. Chaos Solitons Fractals 12 (2001)
445-452.
[14] G. Hochschild and J. Serre, Cohomology of Lie algebras. Ann. Math. 57 (1953) 591-603.
[15] X. Kong and C. Bai, Left-symmetric superalgebraic structures on the super-virasoro algebras. Pacific J. Math. 235 (2008)
43-55.
[16] R. Kleeman, Commutation factors on generalized Lie algebras. J. Math. Phys. 26 (1985) 2405-2412.
[17] M. Kochetov, Generalized Lie algebras and cocycle twists. Comm. Algebra 36 (2008) 4032-4051.
[18] J. Koszul, Homologie et cohomologie des alg´ebres de Lie. Bull. Soc. Math. France 78 (1950) 65-127.
[19] P. Lecomte, Application of the cohomology of graded Lie algebras to formal deformations of Lie algebras. Lett. Math.
Phys. 13 (1987) 157-166.
[20] D. Leıtes, Cohomology of Lie superalgebras. Funkt. Anal. Pril. 9 (1975) 75-76.
[21] M. L´evy-Nahas, Two simple applications of the deformation of Lie algebras. Ann. Inst. H. Poincar´e Sect. A 13 (1970)
221-227.
[22] B. Mitra and K. Tripathy, The cohomology of the generalized Lie algebras. J. Math. Phys. 25 (1984) 2550-2556.
[23] A. Nijenhuis and R. Richardson, Deformations of Lie algebra structures. J. Math. Mech. 17 (1967) 89-105.
[24] R. Ree, Generalized Lie elements. Canad. J. Math. 12 (1960) 493-502.
[25] V. Rittenberg and D. Wyler, Generalized superalgebras. Nuclear Phys. B 139 (1978) 189-202.
[26] M. Scheunert, Generalized Lie algebras. J. Math. Phys. 20 (1979) 712-720.
[27] M. Scheunert, Graded tensor calculus. J. Math. Phys. 24 (1983) 2658-2670.
[28] M. Scheunert and R. Zhang, Cohomology of Lie superalgebras and their generalizations. J. Math. Phys. 39 (1998) 5024-
5061.
[29] R. Zhang, Left-symmetric superalgebras and some related superalgebrac structures. Ph.D thesis, Chern Inst. Math,
Nankai University, Tianjin (2011).
[30] R. Zhang, Left-symmetric structures on complex simple Lie superalgebras. arXiv:1302.5776 [math.RA].
[31] R. Zhang and C. Bai, The classification of left-symmetric superalgebras in low dimensions. J. Algebra Appl. 11 (2012)
1250097 (26 pages).
[32] R. Zhang, D. Hou and C. Bai, A Hom-version of the affinizations of Balinskii-Novikov and Novikov superalgebras. J.
Math. Phys. 52 (2011) 023505.
School of Mathematics and Statistics, Northeast Normal University, Changchun 130024, P.R. China
E-mail address: [email protected]
|
1811.08467 | 2 | 1811 | 2018-11-29T22:42:25 | Schur's Lemma for Coupled Reducibility and Coupled Normality | [
"math.RA"
] | Let $\mathcal A = \{A_{ij} \}_{i, j \in \mathcal I}$, where $\mathcal I$ is an index set, be a doubly indexed family of matrices, where $A_{ij}$ is $n_i \times n_j$. For each $i \in \mathcal I$, let $\mathcal V_i$ be an $n_i$-dimensional vector space. We say $\mathcal A$ is reducible in the coupled sense if there exist subspaces, $\mathcal U_i \subseteq \mathcal V_i$, with $\mathcal U_i \neq \{0\}$ for at least one $i \in \mathcal I$, and $\mathcal U_i \neq \mathcal V_i$ for at least one $i$, such that $A_{ij} (\mathcal U_j) \subseteq \mathcal U_i$ for all $i, j$. Let $\mathcal B = \{B_{ij} \}_{i, j \in \mathcal I}$ also be a doubly indexed family of matrices, where $B_{ij}$ is $m_i \times m_j$. For each $i \in \mathcal I$, let $X_i$ be a matrix of size $n_i \times m_i$. Suppose $A_{ij} X_j = X_i B_{ij}$ for all~$i, j$. We prove versions of Schur's Lemma for $\mathcal A, \mathcal B$ satisfying coupled irreducibility conditions. We also consider a refinement of Schur's Lemma for sets of normal matrices and prove corresponding versions for $\mathcal A, \mathcal B$ satisfying coupled normality and coupled irreducibility conditions. | math.RA | math | Schur's Lemma for Coupled Reducibility and
Coupled Normality
v
o
N
9
2
]
.
A
R
h
t
a
m
[
2
v
7
6
4
8
0
.
1
1
8
1
:
v
i
X
r
a
∗Dana Lahat†
Christian Jutten‡
Helene Shapiro§
December 3, 2018
Abstract
Let A = {Aij}i,j∈I, where I is an index set, be a doubly indexed
family of matrices, where Aij is ni × nj. For each i ∈ I, let Vi be
an ni-dimensional vector space. We say A is reducible in the coupled
sense if there exist subspaces, Ui ⊆ Vi, with Ui 6= {0} for at least
one i ∈ I, and Ui 6= Vi for at least one i, such that Aij(Uj) ⊆ Ui
for all i, j. Let B = {Bij}i,j∈I also be a doubly indexed family of
matrices, where Bij is mi × mj. For each i ∈ I, let Xi be a matrix of
size ni × mi. Suppose AijXj = XiBij for all i, j. We prove versions of
Schur's Lemma for A, B satisfying coupled irreducibility conditions.
We also consider a refinement of Schur's Lemma for sets of normal
matrices and prove corresponding versions for A, B satisfying coupled
normality and coupled irreducibility conditions.
∗Funding: The work of D. Lahat and C. Jutten was supported by the project CHESS,
2012- ERC-AdG-320684. GIPSA-lab is a partner of the LabEx PERSYVAL-Lab (ANR11-
LABX-0025).
†IRIT, Universit´e de Toulouse, CNRS, Toulouse, France ([email protected]). This
work was carried out when D. Lahat was with GIPSA-lab, Grenoble, France.
‡Univ. Grenoble Alpes, CNRS, Grenoble INP, GIPSA-lab, 38000 Grenoble, France. C.
Jutten is a senior member of Institut Univ. de France ([email protected]
inp.fr).
§Department of Mathematics and Statistics, Swarthmore College, Swarthmore, PA
19081 ([email protected]).
1
1
Introduction
Let K be a positive integer and let n1, . . . , nK and m1, . . . , mK be positive
integers. Consider two doubly indexed families of matrices, A = {Aij}K
i,j=1
i,j=1, where Aij is ni×nj and Bij is mi×mj. Put N = PK
and B = {Bij}K
i=1 ni
and M = PK
i=1 mi. Arrange the Aij's into an N × N matrix, A, with Aij in
block i, j of A.
A =
A11 A12
A21 A22
...
...
AK1 AK2
· · · A1K
· · · A2K
...
· · · AKK
.
Similarly, form an M × M matrix B with Bij in block i, j. Let Xi be an
ni × mi matrix and form an N × M matrix X with blocks X1, . . . , XK down
the diagonal and zero blocks elsewhere. Thus,
· · ·
· · ·
· · ·
X =
,
X1
0
0 X2
0
...
0
0
0
0 X3
...
...
0
0
0
0
0
...
· · · XK
AijXj = XiBij,
where the 0 in position i, j represents an ni × mi block of zeroes. Then
AX = XB if and only if
(1)
for all i, j = 1, . . . , K. We may rewrite AX = XB as AX − XB = 0, a
homogeneous Sylvester equation [13].
We define several versions of coupled reducibility and prove corresponding
versions of Schur's Lemma [12] for pairs A, B. Imposing coupled irreducibil-
ity constraints on A and B restricts the possible solutions, X1, . . . , XK, to
the equations (1). We also discuss a refinement of Schur's Lemma for nor-
mal matrices, and prove corresponding versions for A, B satisfying coupled
normality conditions.
The system of coupled matrix equations (1) arises in a recent model
for multiset data analysis [5, 8], called joint independent subspace analysis,
or JISA. This model consists of K datasets, where each dataset is an un-
known mixture of several latent stochastic multivariate signals. The blocks
of A and B represent statistical links among these datasets. More specifi-
cally, Aij and Bij represent statistical correlations among latent signals in the
2
ith and jth datasets. The multiset joint analysis framework, as opposed to
the analysis of K distinct unrelated datasets, arises when a sufficient number
of cross-correlations among datasets, i.e., Aij and Bij, i 6= j, are not zero.
It turns out [6, 9, 10] that the identifiability and uniqueness of this model,
in its simplest form, boil down to characterizing the set of solutions to the
system of matrix equations (1), when the cross-correlations among the latent
signals are subject to coupled (ir)reducibility. In other words, the coupled
reducibility conditions that we introduce in this paper can be attributed with
a physical meaning, and can be applied to real-world problems. We refer the
reader to [6, 10] for further details on the JISA model, and on the derivation
of (1). In this paper, we consider scenarios more general than those required
to address the signal processing problem in [6, 10]. The analysis in Section
6, which focuses on coupled normal matrices, is inspired by the JISA model
in [6, 10], in which the matrices A and B are Hermitian, and thus a spe-
cial case of coupled normal. A limited version of some of the results in this
manuscript was presented orally in, e.g., [7, 4, 1] and in a technical report [6];
however, they were never published.
While motivated by the matrix equation, AX = XB, the main defi-
nitions, theorems, and proofs do not require K to be finite. Thus, for a
general index set, I, we consider doubly indexed families, A = {Aij}i,j∈I
and B = {Bij}i,j∈I. For the situation described above, I = {1, 2, . . . , K}.
We use F to denote the field of scalars. For some results, F can be any
field. For results involving unitary and normal matrices, F = C, the field of
complex numbers. For each i ∈ I, let ni and mi be positive integers, let Vi
be an ni-dimensional vector space over F, and let Wi be an mi-dimensional
vector space over F. (Essentially, Vi = Fni and Wi = Fmi.) For all i, j ∈ I,
let Aij be an ni × nj matrix and let Bij be mi × mj. View Aij as a linear
transformation from Vj to Vi, and Bij as a linear transformation from Wj
to Wi. For each i ∈ I, let Xi be an ni × mi matrix; view Xi as a linear
transformation from Wi to Vi. We are interested in families A, B satisfying
the equations (1) for all i, j ∈ I.
For some results, all of the ni's will be equal, and all of the mi's will be
equal. In this case, we use n for the common value of the ni's and m for the
common value of the mi's. We then set V = Fn and W = Fm. Each Xi is
then n × m. For I = {1, . . . , K}, we have N = Kn, and M = Km. All of
the blocks Aij in A are n × n, while all of the blocks Bij in B are m × m.
Section 2 reviews the usual matrix version of Schur's Lemma and its
proof. Section 3 defines coupled reducibility and two restricted versions,
3
called proper and strong reducibility. Section 4 states and proves versions of
Schur's Lemma for coupled reducibility and proper reducibility, Theorem 4.2.
In section 5, we define some graphs associated with the pair A, B and use
them for versions of Schur's Lemma corresponding to strongly coupled re-
ducibility, Theorems 5.1 and 5.2. Section 6 deals with a refinement of Schur's
Lemma for sets of normal matrices and corresponding versions for pairs A,
B which are coupled normal, Theorem 6.2. The Appendix presents examples
to support some claims made in Section 3.
2 Reducibility and Schur's Lemma
We begin by reviewing the ordinary notion of reducibility for a set of linear
transformations and Schur's Lemma.
Definition 2.1. A set, T , of linear transformations, on an n-dimensional
vector space, V, is reducible if there is a proper, non-zero subspace U of V
such that T (U) ⊆ U for all T ∈ T .
If T is not reducible, we say it is
irreducible.
The subspace U is an invariant subspace for each transformation T in T .
We say T is fully reducible if it is possible to decompose V as a direct sum
V = U ⊕ U, where U and U are both nonzero, proper invariant subspaces
of T .
Alternatively, one can state this in matrix terms. The linear transforma-
tions in T may be represented as n× n matrices, relative to a choice of basis
for V. Let d be the dimension of U; choose a basis for V in which the first d
basis vectors are a basis for U. Since U is an invariant subspace of each T
in T , the matrices representing T relative to this basis are block upper tri-
angular with square diagonal blocks of sizes d× d and (n− d)× (n− d). The
first diagonal block, of size d× d, represents the action of the transformation
on the invariant subspace U. The (n − d) × d block in the lower left hand
corner consists of zeroes. Since changing basis is equivalent to applying a
matrix similarity, we have the following matrix version of Definition 2.1.
Definition 2.2. (Matrix version of reducibility.) A set M of n× n matrices
is reducible if, for some d, with 0 < d < n, there is a nonsingular matrix S
such that, for each A in M, the matrix S−1AS is block upper triangular with
square diagonal blocks of sizes d × d and (n − d) × (n − d).
4
When T is fully reducible, we can use a basis for V in which the first d
basis vectors are a basis for U, and the remaining n − d basis vectors are a
basis for U. The corresponding matrices for T are then block diagonal, with
diagonal blocks of sizes d × d and (n − d) × (n − d).
If F = C and M is reducible, the S in Definition 2.2 can be chosen to be
a unitary matrix, by using an orthonormal basis for Cn in which the first d
basis vectors are an orthonormal basis for U. If M is fully reducible, and
the subspaces U and U are orthogonal, use an orthonormal basis of U for the
first d columns of S and an orthonormal basis of U for the remaining n − d
columns. Then S is unitary, and S−1MS is block diagonal, with diagonal
blocks of sizes d × d and (n − d) × (n − d).
The following fact is well known; we include the proof because the idea is
used later. For this fact, we assume we are working over C, or at least over
a field that contains the eigenvalues of the transformations.
Proposition 2.1. Let M be an irreducible set of n × n complex matrices.
Suppose the n× n matrix C commutes with every element of M. Then C is
a scalar matrix.
Proof. Let λ be an eigenvalue of C, and let Uλ denote the corresponding
eigenspace. Let A ∈ M. For any v ∈ Uλ, we have C(Av) = A(Cv) = λ(Av).
Hence Av is in Uλ, and so Uλ is invariant under each element of M. Since
an eigenspace is nonzero, and M is irreducible, we must have Uλ = Cn.
Hence C = λIn.
Schur's Lemma plays a key role in group representation theory. It is used
to establish uniqueness of the decomposition of a representation of a finite
group into a sum of irreducible representations. However, one need not have
a matrix group; the result holds for irreducible sets of matrices. We include
the usual proof [2], because the same idea is used to prove our versions for
coupled reducibility.
Theorem 2.1 (Schur's Lemma). Let {Ai}i∈I be an irreducible family of
n × n matrices, and let {Bi}i∈I be an irreducible family of m × m matrices.
Suppose P is an n × m matrix such that AiP = P Bi for all i ∈ I. Then,
either P = 0, or P is nonsingular; in the latter case we must have m = n.
For matrices of complex numbers, if Ai = Bi for all i ∈ I, then P is a scalar
matrix.
5
Proof. View the Ai's as linear transformations on an n-dimensional vector
space V, and the Bi's as linear transformations on an m-dimensional vector
space W. The n × m matrix P represents a linear transformation from W
to V. So ker(P ) is a subspace of W and range(P ) is a subspace of V.
Let w ∈ ker(P ). Then P (Biw) = AiP w = 0. Hence, ker(P ) is invariant
under {Bi}i∈I. Since {Bi}i∈I is irreducible, ker(P ) is either {0} or W. In
If P 6= 0, then ker(P ) = {0}. Now consider the
the latter case, P = 0.
range space of P . For any w ∈ W, we have Ai(P w) = P (Biw) ∈ range(P ),
so the range space of P is invariant under Ai for each i. Since {Ai}i∈I
is irreducible, range(P ) is either {0} or V. But we are assuming P 6= 0
so range(P ) = V. Since we also have ker(P ) = {0}, the matrix P must be
nonsingular and m = n.
If each Ai
is a complex matrix, then, since {Ai}i∈I is irreducible, P must be a scalar
matrix.
For nonsingular P , we have P −1AiP = Bi for all i ∈ I, so {Ai}i∈I and {Bi}i∈I
are simultaneously similar.
If Ai = Bi for all i ∈ I, then P commutes with each Ai.
3 Coupled Reducibility
For simultaneous similarity of {Ai}i∈I and {Bi}i∈I, there is a nonsingular
matrix, P , such that P −1AiP = Bi for all i. We now define a "coupled"
version of similarity for two doubly indexed families, A and B, with ni = mi
for all i ∈ I. In this case, Aij and Bij are matrices of the same size, and in
the equations (1), each Xi is a square matrix.
Definition 3.1. Let A = {Aij}i,j∈I and B = {Bij}i,j∈I, where ni = mi for
all i ∈ I. We say A and B are similar in the coupled sense if there exist
nonsingular matrices {Ti}i∈I, where Ti is ni × ni, such that T −1
i AijTj = Bij
for all i, j ∈ I.
For a finite index set I = {1, . . . , K}, this can be stated in terms of the
matrices, A and B. Let T be the block diagonal matrix T1 ⊕ T2 ⊕ · · · ⊕ TK.
Then AT = T B if and only if AijTj = TiBij for all i, j. Hence, A and B are
similar in the coupled sense if and only if T is nonsingular and T −1AT = B.
We define several versions of "reducible in the coupled sense" for a doubly
indexed family A. The basic idea is that there are subspaces, {Ui}i∈I, where
6
Ui ⊆ Vi, such that Aij(Uj) ⊆ Ui for all i, j ∈ I. This holds trivially when Ui
is zero for all i, and when Ui = Vi for all i, so we shall insist that at least one
subspace is nonzero, and at least one is not Vi. We are also interested in two
more restrictive versions: the case where at least one Ui is a nonzero, proper
subspace, and the case where every Ui is a nonzero, proper subspace of Vi.
Definition 3.2. Let A = {Aij}i,j∈I where Aij is ni × nj. We say A is
reducible in the coupled sense if there exist subspaces {Ui}i∈I, where Ui ⊆ Vi,
such that the following hold.
1. For at least one i, we have Ui 6= {0}.
2. For at least one i, we have Ui 6= Vi.
3. For all i, j ∈ I we have Aij(Uj) ⊆ Ui.
We say {Ui}i∈I is a reducing set of subspaces for A, or that A is reduced by
{Ui}i∈I. If A is not reducible in the coupled sense, we say it is irreducible
in the coupled sense. We say A is properly reducible in the coupled sense if
at least one Ui is a nonzero, proper subspace of Vi. We say A is strongly
reducible in the coupled sense if every Ui is a nonzero, proper subspace of Vi.
Remark 3.1. If ni = 1 for all i, the one-dimensional spaces Vi have no
nonzero proper invariant subspaces, so A cannot be properly or strongly
irreducible. If K = 1, then A consists of a single n × n matrix.
Remark 3.2. If ni = n for all i, and the subspaces {Ui}i∈I are all the same
nonzero proper subspace, i.e, for all i, we have Ui = U where U is a nonzero,
proper subspace of V, then A is reducible in the ordinary sense, given in
Definition 2.1.
Note the following facts.
1. If Uj = {0}, then Aij(Uj) ⊆ Ui holds for any Aij and any Ui.
2. If Ui = Vi, then Aij(Uj) ⊆ Ui holds for any Aij and any Uj.
3. If Uj = Vj and Ui = {0}, then Aij = 0.
4. For i = j, we have Aii(Ui) ⊆ Ui, so Ui is an invariant subspace of Aii.
7
An equivalent matrix version of Definition 3.2 is obtained by choosing an
appropriate basis for each Vj. Let dj be the dimension of the subspace Uj.
We have 0 ≤ dj ≤ nj. If dj is positive, let vj,1, . . . , vj,dj be a basis for Uj and
let Tj be a nonsingular nj × nj matrix which has vj,1, . . . , vj,dj in the first dj
columns. If dj = 0, we may use any nonsingular nj × nj matrix for Tj. Set
Bij = T −1
i AijTj; equivalently,
AijTj = TiBij.
The first dj columns of Tj are a basis for Uj, so Aij(Uj) ⊆ Ui tells us the
first dj columns of AijTj are in Ui. Hence, the first dj columns of TiBij
are in Ui, so by the definition of Ti, each of the first dj columns of TiBij
is a linear combination of the first di columns of Ti. Therefore, each of the
first dj columns of Bij will have zeroes in all entries below row di, and the
lower left hand corner of Bij is a block of zeroes of size (ni − di) × dj. When
0 < di < ni and 0 < dj < nj, the matrix Bij has the form
Bij = (cid:18) C
0(n−di)×dj E (cid:19) ,
D
(2)
where C is size di × dj and represents the action of Aij on the subspace Uj.
The zero block in the lower left hand block has size (n − di) × dj, while D is
di × (nj − dj) and E is (ni − di) × (nj − dj).
Remark 3.3. The block matrix (2) has a block of zeroes in the lower left
hand corner of size (n − di) × dj. We also use this terminology for the cases
di = 0, di = ni, dj = 0, dj = nj, in which case we mean the following. If
di = 0, the first dj columns of Bij are zero. If di = ni there is no restriction
on the form of Bij. If dj = 0 there is no restriction on the form of Bij. If
dj = nj the last n − di rows of Bij are zero.
Conversely, if each Bij = T −1
i AijTj has the block form in (2), define Ui to
be the subspace spanned by the first di columns of Ti. The subspaces {Ui}i∈I
then satisfy Aij(Uj) ⊆ Ui. Hence, we have the following equivalent matrix
form of Definition (3.2).
Definition 3.3 (Matrix version of coupled reducibility). Let A = {Aij}i,j∈I.
We say A is reducible in the coupled sense, or, reducible by coupled similarity,
if there exist integers {di}i∈I, with 0 ≤ di ≤ ni, and nonsingular ni × ni
matrices Ti, such that the following hold.
8
1. At least one di is positive.
2. At least one di is less than ni.
3. Each matrix Bij = T −1
i AijTj has a block of zeroes in the lower left
hand corner of size (ni − di) × dj.
We say A is properly reducible in the coupled sense if 0 < di < ni for at
least one value of i. We say A is strongly reducible in the coupled sense if
0 < di < ni for every i.
Full reducibility by coupled similarity occurs when, for each i, there is also
a subspace Ui such that Vi = Ui ⊕ Ui, and Aij( Uj) ⊆ Ui for all i, j ∈ I. For
the corresponding matrix version, use a basis for Uj in the first dj columns
of Tj and a basis for Uj in the remaining nj − dj columns. For 0 < di < ni
and 0 < dj < nj, the matrix Bij = T −1
i AijTj has the block form
Bij = (cid:18) C
0(n−di)×dj
0di×(n−dj )
E
(cid:19) .
(3)
The di × dj matrix C represents the action of Aij on Uj and the (ni − di) ×
(nj − dj) matrix E represents the action of Aij on Uj.
For the field of complex numbers we have unitary versions.
Definition 3.4 (Unitary version of reducible in the coupled sense). Let A be
a family of complex matrices. We say A is unitarily reducible in the coupled
sense if the conditions of Definition 3.3 are satisfied with unitary matrices Ti.
For complex A, reducibility by coupled similarity implies reducibility by
coupled unitary similarity. Simply use an orthonormal basis for each Ui, and
extend it to an orthonormal basis for Vi to obtain a unitary matrix for Ti.
If A is fully reducible, and Ui and Ui are orthogonal subspaces, then, for
each Vi = Ui ⊕ Ui, we can form a unitary matrix Ti using an orthonormal
basis for Ui for the first di columns and an orthonormal basis for Ui for the
remaining ni − di columns. Each Bij then has the block form (3).
Unitary reducibility matters in the JISA model, because A and B are
correlation matrices, and the appropriate linear change of variable leads to
a congruence, rather than a similarity. When Ti is unitary, T −1
i = T ∗i . For
T = T1 ⊕ T2 ⊕ · · · ⊕ Tk we then have T −1AT = T ∗AT .
From the definition, it is clear that if A is strongly reducible, then it is
also properly reducible, and if it is properly reducible, it is reducible. We
9
introduce some notation. Fix an index set, I. Use I to denote the size of I;
when I is a finite set with K elements, we assume I = {1, 2, . . . , K}. Fix
a family {ni}i∈I of positive integers, and a field F. Consider the set of all
A = {Aij}i,j∈I, where Aij is an ni × nj matrix with entries from F. We use
Red(F,{ni}i∈I) to denote the set of all such families A which are reducible
in the coupled sense. We use P ropRed(F,{ni}i∈I) for the set of all such A
which are properly reducible in the coupled sense, and StrRed(F,{ni}i∈I)
for the set of all such A that are strongly reducible in the coupled sense.
When all ni's have the same value, n, and I = K, we use the notations
Red(F, n, K), P ropRed(F, n, K), and StrRed(F, n, K).
When ni = 1, the space Vi = F is one dimensional and has no nonzero
proper subspaces. Hence, StrRed(F,{ni}i∈I) is the empty set if ni = 1 for
some i, and P ropRed(F,{ni}i∈I) is the empty set when ni = 1 for all i.
From Definition 3.2 it is obvious that
StrRed(F,{ni}i∈I) ⊆ P ropRed(F,{ni}i∈I) ⊆ Red(F,{ni}i∈I).
(4)
Using the superscript "C" to indicate the complement of a set, we then have
RedC(F,{ni}i∈I) ⊆ P ropRedC(F,{ni}i∈I) ⊆ StrRedC(F,{ni}i∈I).
(5)
The symbol "⊆" means "subset of or equal to." We use "⊂" to indicate
"proper subset of." One might expect that "⊆" can generally be replaced
by "⊂" in (4) and (5). This is correct when I has at least four elements,
and ni ≥ 2 for at least one value of i. Furthermore, for I ≥ 2, we have
StrRed(F,{ni}i∈I) ⊂ P ropRed(F,{ni}i∈I), provided ni ≥ 2 for at least one i.
However, for I = 2 and I = 3, whether P ropRed(F,{ni}i∈I) is equal to, or
is a proper subset of, Red(F,{ni}i∈I) depends on the field F, and on the ni's.
The appendix treats this in more detail. Here is a summary of what is shown
there.
1. For any field F, if I ≥ 4 and ni ≥ 2 for at least one i,
StrRed(F,{ni}i∈I) ⊂ P ropRed(F,{ni}i∈I) ⊂ Red(F,{ni}i∈I).
Consequently,
RedC(F,{ni}i∈I) ⊂ P ropRedC(F,{ni}i∈I) ⊂ StrRedC(F,{ni}i∈I).
2. For any field F, if I ≥ 2 and ni ≥ 2 for at least one i,
StrRed(F,{ni}i∈I) ⊂ P ropRed(F,{ni}i∈I).
10
3. If F is algebraically closed and n ≥ 2, then
P ropRed(F, n, 2) = Red(F, n, 2) and P ropRed(F, n, 3) = Red(F, n, 3).
4. For the field, R, of real numbers, when n = 2, we have
P ropRed(R, 2, 2) ⊂ Red(R, 2, 2) and P ropRed(R, 2, 3) ⊂ Red(R, 2, 3).
For n ≥ 3,
P ropRed(R, n, 2) = Red(R, n, 2) and P ropRed(R, n, 3) = Red(R, n, 3).
4 A coupled version of Schur's Lemma
The main result of this section is Theorem 4.2, a coupled version of Schur's
Lemma for reducibility and proper reducibility. Section 5 deals with the more
complicated version for strong reducibility.
Consider families A = {Aij}i,j∈I and B = {Bij}i,j∈I, where Aij is ni × nj
and Bij is mi × mj, linked by equations AijXj = XiBij, where Xi is ni × mi.
Recall that Aij is a linear transformation from Vj to Vi, and Bij is a linear
transformation from Wj to Wi The matrix Xi is a linear transformation from
Wi to Vi. Note that ker(Xi) is a subspace of Wi and range(Xi) is a subspace
of Vi.
Reviewing the proof of Schur's Lemma (Theorem 2.1), the key facts are
that ker(P ) is an invariant subspace of {Bi}i∈I, and range(P ) is an invariant
subspace of {Ai}i∈I. For the case of complex matrices with Ai = Bi for
all i, any eigenspace of P is an invariant subspace of {Ai}i∈I. The following
"coupled" versions of these facts are used to prove coupled versions of Schur's
lemma for A, B. In the coupled versions, the Xi's play the role of the P .
Lemma 4.1. Let A = {Aij}i,j∈I and B = {Bij}i,j∈I, where Aij is ni × nj
and Bij is mi × mj. Let Xi be ni × mi and suppose for all i, j ∈ I, we
have AijXj = XiBij. If mi = ni for some i, then, for any scalar α, define
Ui(α) = {v Xiv = αv}. The following hold for all i, j ∈ I.
1. Bij(ker(Xj)) ⊆ ker(Xi).
2. Aij(range(Xj)) ⊆ range(Xi).
3. If A = B, then Aij(Uj(α)) ⊆ Ui(α).
11
Proof. For any w ∈ ker(Xj), we have Xi(Bijw) = AijXjw = 0. Hence,
Bijw ∈ ker(Xi). This proves 1.
For w ∈ W, we have Aij(Xjw) = Xi(Bijw) ∈ range(Xi), proving 2.
Finally, suppose A = B. Then AijXj = XiAij for all i, j. Let v ∈ Uj(α).
Then Xi(Aijv) = AijXjv = α(Aijv), showing Aijv ∈ Ui(α).
If mi = ni and α is an eigenvalue of Xi, with α ∈ F, then Ui(α) is the
corresponding eigenspace. If α is not an eigenvalue of Xi, then Ui(α) is the
zero subspace.
We now state a version of Schur's Lemma for families that are irreducible
in the coupled sense. The proofs simply extend the argument used to prove
the usual Schur Lemma.
Theorem 4.2. Let A = {Aij}i,j∈I and B = {Bij}i,j∈I, where Aij is ni × nj
and Bij is mi × mj. Let Xi be ni × mi and suppose for all i, j ∈ I, we have
AijXj = XiBij.
1. Suppose both A and B are irreducible in the coupled sense. Then either
Xi = 0 for all i, or Xi is nonsingular for all i. In the latter case, mi = ni
for all i. If A = B, and A is a family of complex matrices, then there
is a scalar α such that Xi = αIni for all i.
2. Suppose neither A nor B is properly reducible in the coupled sense.
Then for each i, either Xi = 0 or Xi is nonsingular. If Xi is nonzero we
must have mi = ni. If A = B and consists of complex matrices, then
any nonzero Xi is a scalar multiple of Ini.
Proof. For part 1, assume A and B are both coupled irreducible. Consider
the subspaces ker(Xi), i ∈ I. Since B is irreducible in the coupled sense,
statement 1 of Lemma 4.1 tells us there are only two possibilities: either
ker(Xi) = {0} for all i, or ker(Xi) = Wi for all i. In the latter case, Xi = 0
for all i and so we are done.
Suppose now that ker(Xi) = {0} for all i. We now use the subspaces
range(Xi), i ∈ I. Since A is irreducible in the coupled sense, part 2 of
Lemma 4.1 tells us the only possibilities are range(Xi) = {0} for all i or
range(Xi) = Vi for all i. If range(Xi) = {0} for all i, then Xi = 0 for all i.
Otherwise, we have both kerXi = {0} and range(Xi) = Vi for all i. Hence
each Xi is nonsingular and mi = ni.
Now suppose A = B and F = C. Let λ be an eigenvalue of Xp for
some fixed p ∈ I. Part 3 of Lemma 4.1 tells us Aij(Uj(λ)) ⊆ Ui(λ) for
12
all i, j. Since A is irreducible in the coupled sense, there are then only two
possibilities for the subspaces Ui(λ): either they are all zero, or Ui = Vi for
all i. Since λ was chosen to be an eigenvalue of Xp, we know Up(λ) is not
zero. Therefore, Ui(λ) = Vi for all i and hence Xi = λIni for all i.
For part 2, assume neither A nor B is properly reducible in the coupled
sense. Consider ker(Xi). Since B is not properly reducible in the coupled
sense, Lemma 4.1 tells us ker(Xi) cannot be a nonzero, proper subspace
of Wi. Hence, for each particular i, either ker(Xi) = {0} or ker(Xi) = Wi.
In the latter case, Xi = 0.
Suppose ker(Xi) = {0} for some i. Since A is not properly reducible
in the coupled sense, Lemma 4.1 tells us range(Xi) is either {0} or Vi. If
range(Xi) = {0} then Xi = 0 . Otherwise, we have both kerXi = {0} and
range(Xi) = Vi, so Xi is nonsingular and mi = ni.
Now suppose A = B is a family of complex matrices. Suppose Xp 6= 0
for some p. Let λp be an eigenvalue of Xp. Note λp 6= 0 because Xp is
nonsingular. By Lemma 4.1, we have Aij(Uj(λp)) ⊆ Ui(λp) for all i, j. Since A
is not properly reducible in the coupled sense, each Ui(λp) is either zero or
the full vector space Vi. Since λp is an eigenvalue of Xp, the space Up(λp) is
not zero. Therefore, Up(λj) = Vp and Xp = λpInp.
Remark 4.1. The ordinary version of Schur's Lemma, Theorem 2.1, applies
to the case where both A and B are irreducible in the sense of Definition 2.1,
and Xi = P for all i.
Note the different conclusions for the two parts of Theorem 4.2. For
part 1, either all Xi's are zero, or all are nonsingular. When A = B and
F = C, all the Xi's are the same scalar multiple of the identity matrix.
In part 2, there are more options for the Xi's. Each Xi is either zero or
nonsingular, but some can be zero and others nonsingular. For A = B and
F = C, the proof for part 2 gives Xp = λpInp for a particular value of p;
it does not show every nonzero Xi equals the same scalar multiple of the
identity matrix.
The broader range of options for the Xi's in part 2 makes sense when
we consider that, at least for I ≥ 4, we have P ropRed(F,{ni}i∈I) ⊂
Red(F,{ni}i∈I), and hence RedC(F,{ni}i∈I) ⊂ P ropRedC(F,{ni}i∈I). Part 2
applies to a broader set of pairs A, B than part 1.
Consider the situation in part 2 of Theorem 4.2. Suppose Xi = 0 and Xj
is nonsingular. The equation AijXj = XiBij then tells us Aij = 0, while
13
AjiXi = XjBji gives Bji = 0. Set
I0 = {i ∈ I Xi = 0}
and
Inon = {i ∈ I Xi is nonsingular}.
We have Aij = 0 and Bji = 0 whenever i ∈ I0 and j ∈ Inon. For ex-
ample, suppose I = {1, 2, . . . , K}, and, for some 0 < s < K, we have
I0 = {1, 2, . . . , s} and Inon = {s + 1, s + 2, . . . , K}. The N × N matrix A
then has only zero blocks in the upper right hand corner formed from the
first s rows and last K − s columns. The M × M matrix B has zero blocks in
the lower left hand corner formed by the last K − s rows and first s columns.
A =
A11
...
As1
A(s+1)1
...
AK1
· · ·
· · ·
A1s
...
Ass
...
AKs
0
...
0
...
· · ·
· · · A(s+1)s A(s+1)(s+1)
· · ·
· · · A(s+1)K
· · ·
0
...
0
...
AKK
.
AK(s+1)
· · ·
Returning to the case of general I, one can check that A is coupled reducible
via the subspaces Ui = {0} for i ∈ I0, and Ui = Vi for i ∈ Inon. The family B
is coupled reducible via the subspaces Ui = Wi when i ∈ I0, and Ui = {0}
when i ∈ Inon.
5 Strong reducibility and Schur's Lemma
We now consider strongly coupled reducibility. Our goal is a version of
Schur's lemma for families that are not strongly reducible in the coupled
sense, with a conclusion similar to that of Theorem 4.2: each Xi is either
zero or nonsingular. The next example shows that for such a conclusion, we
need some restrictions on Aij and Bij.
Example 5.1. Let n = m = 2 (so V = W = F2), and K = 2. Put
A12 = (cid:18) a b
d(cid:19)
0 0(cid:19) .
B12 = (cid:18) 0 0
A11 = A22 = (cid:18) 0
0(cid:19)
0 0(cid:19)
B11 = B22 = (cid:18) 0 1
1
0
A21 = (cid:18) 0 0
0 0(cid:19)
d(cid:19)
B21 = (cid:18) a
c
b
c
14
In terms of the matrices A, B:
A =
B =
.
a b
d
1
0
c
0
0
0 1
0 0
0 0
0 0
0 0
1
0
0 0
0
0
0 1
a b
c d 0 0
Let U be the subspace spanned by e1 = (cid:18) 1
0(cid:19). One may easily check that A is
properly reducible in the coupled sense with U1 = U and U2 = {0}, while B is
properly reducible in the coupled sense with U1 = {0} and U2 = U. However,
if c 6= 0, then neither A nor B is strongly reducible in the coupled sense.
The reason is that U is the only nonzero, proper invariant subspace for the
diagonal blocks, (cid:18) 0 1
0 0(cid:19), of A and B, so U1 = U2 = U is the only possible
choice for nonzero, proper subspaces U1 and U2. If c 6= 0, then A12(U) 6⊂ U,
and B21(U) 6⊂ U. So if c 6= 0, neither A nor B is strongly reducible in the
0 0(cid:19) and X = X1 ⊕ X2. One
coupled sense. Set X1 = (cid:18) 0
0(cid:19) , X2 = (cid:18) 0 0
1
0
may check that AX = XB = 0 and hence AijXj = XiBij for i, j = 1, 2. The
point is that the matrix X1 is neither zero nor nonsingular.
Our theorem for coupled pairs A, B that are not strongly reducible will
be for the case when ni = n and mi = m for all i. It will have a hypothesis
about graphs related to A and B; roughly speaking, this hypothesis will tell
us there are "enough" nonsingular Aij's and Bij's. Although our main result
assumes A is a family of n× n matrices and B is a family of m× m matrices,
we define the graphs for families with matrices of any size.
Recall that a matrix is said to have full column rank if the columns are
linearly independent; thus, the rank of the matrix equals the number of
columns. If A is a p × q matrix with full column rank, and U is a subspace
of Fq, then A(U) has the same dimension as U.
Consider A = {Aij}i,j∈I, and subspaces {Ui}i∈I, satisfying Aij(Uj) ⊆ Ui
for all i, j. Let di be the dimension of Ui. If Aij has full column rank, then
Aij(Uj) ⊆ Ui tells us dj ≤ di. If Aji also has full column rank, then we also
have di ≤ dj, and hence di = dj. When Aij and Aji both have full column
rank, nj ≤ ni and ni ≤ nj, so ni = nj; hence, Aij and Aji are actually
square, nonsingular matrices. If all of the Aij's have full column rank, then
all of the ni's have the same value, n, and all of the subspaces Uj have the
15
same dimension, d. However, we need not assume all of the matrices Aij
are nonsingular in order to show the Uj's all have the same dimension. To
explore this further, we introduce a directed graph in which directed edges
correspond to the Aij's of full column rank.
Definition 5.1. Let A = {Aij}i,j∈I, with Aij of size ni × nj. The directed
graph (digraph) of A, denoted D(A), is the graph on vertices {vi}i∈I, such
that there is a directed edge (vi, vj) from vi to vj if and only if Aij has full
column rank.
For a finite index set, I = {1, . . . , K}, there are K vertices. If ni = 1
for all i, our D(A) is just the usual directed graph associated with a K × K
matrix.
More generally, there is a vertex for each i ∈ I, so there could be infinitely
many vertices. We use the same definition for directed walk as for graphs with
a finite number of vertices. A directed walk is a finite sequence of vertices,
vi1, vi2, . . . , vip, such that (vij , vi(j+1)) is a directed edge for 1 ≤ j ≤ (p − 1).
In this case, we write vi1 → vi2 → · · · → vip. Vertices v and w in a directed
graph D are said to be strongly connected if there is a directed walk from v
to w and a directed walk from w to v. We say D is strongly connected if
each pair of vertices of D is strongly connected.
Proposition 5.1. Let A = {Aij}i,j∈I and suppose the subspaces {Ui}i∈I
satisfy Aij(Uj) ⊆ Ui, for all i, j. Then the following hold.
1. If there is a directed walk from vi to vj in D(A), then nj ≤ ni, and
dim(Uj) ≤ dim(Ui).
2. If the vertices vi and vj are strongly connected in D(A), then ni = nj,
and dim(Uj) = dim(Ui)
3. If D(A) is strongly connected, all of the ni's are equal, and all of the
subspaces Ui have the same dimension.
Proof. Let di = dim(Ui). If (vi, vj) is a directed edge of D(A), then Aij has
full column rank, so, as we have already observed, nj ≤ ni and dj ≤ di.
More generally, suppose vi = vi1 → vi2 → · · · → vip = vj is a directed
walk from vi to vj in D(A). Working from right to left, we have
nj = nip ≤ nip−1 ≤ · · · ≤ ni2 ≤ ni1 = ni
16
and
dj = dip ≤ dip−1 ≤ · · · ≤ di2 ≤ di1 = di.
Hence, nj ≤ ni and dj ≤ di.
and a directed walk from vj to vi. So ni = nj and di = dj.
If vi and vj are strongly connected, there is a directed walk from vi to vj
If D(A) is strongly connected, then, for all i, j, we have di = dj and
ni = nj, so all of the subspaces Uj have the same dimension and all of
the ni's have the same value.
As an example, suppose I = {1, . . . , K} and A12, A23, . . . AK−1,K, AK1 all
have full column rank. Then D(A) contains the directed cycle
v1 → v2 → · · · → vK−1 → vK → v1,
and is strongly connected.
If D(A) is not strongly connected, the strong components identify sets
of ni's which must be equal, and sets of subspaces Ui which must have the
same dimension. For each strong component, C, of D(A), all ni's corre-
sponding to vertices of C must be equal, and all subspaces Ui corresponding
to vertices of C must have the same dimension. For a finite I, we can use the
strong components to put the N × N matrix A into a block triangular form
in which none of the Aij's below the diagonal blocks has full column rank.
(See [3], section 3.2.)
For the proofs of coupled versions of Schur's Lemma, the subspaces Ui of
interest are the kernels and ranges of the matrices {Xi}i∈I.
Proposition 5.2. Let A = {Aij}i,j∈I and B = {Bij}i,j∈I. Let Xi be ni× mi,
and suppose AijXj = XiBij for all i, j ∈ I. Then the following hold.
1. If vi and vj are strongly connected in D(A), then range(Xi) and
range(Xj) have the same dimension, i.e., Xi and Xj have the same
rank.
2. If vi and vj are strongly connected in D(B), then ker(Xi) and ker(Xj)
have the same dimension, i.e., Xi and Xj have the same nullity.
3. If D(A) is strongly connected, all of the ni's have the same value, n,
and all of the Xi's have the same rank.
17
4. If D(B) is strongly connected, all of the mi's have the same value, m,
and all of the Xi's have the same nullity, d.
5. If vi and vj are strongly connected in D(B), then Xi and Xj have the
same rank.
6. If D(B) is strongly connected, all of the Xi's have the same rank.
Proof. The first four parts follow from Lemma 4.1 and Proposition 5.1. For
part 5, suppose vi and vj are strongly connected in D(B). Then mi = mj, so
the matrices Xi and Xj have the same number of columns. From part 2, we
know Xi and Xj have the same nullity. The rank plus nullity theorem then
tells us Xi and Xj have the same rank. Part 6 is an immediate consequence
of part 5.
We now have a version of Schur's lemma for families A, B when neither
is strongly reducible in the coupled sense.
Theorem 5.1. Assume neither A = {Aij}i,j∈I nor B = {Bij}i,j∈I is strongly
reducible in the coupled sense. Assume also that D(A) and D(B) are strongly
connected. Let Xi be n × m for all i ∈ I, and suppose AijXj = XiBij for
all i, j ∈ I. Then either Xi = 0 for all i, or Xi is nonsingular for all i. In
the latter case we must have m = n. If A = B and is a family of complex
matrices, then there is some scalar α such that Xi = αIn for all i.
Proof. Note first that since D(A) and D(B) are both strongly connected,
the Aij's are all square matrices of the same size, n, and the Bij's are all
square matrices of the same size, m.
By Proposition 5.2, the subspaces ker(Xi), for i ∈ I, all have the same
dimension, d. Since B is not strongly reducible in the coupled sense, either
d = 0 or d = m. If d = m, then Xi = 0 for all i and we are done.
Assume then that d = 0. Proposition 5.2 tells us the subspaces range(Xi)
all have the same dimension, r. Since A is not strongly reducible in the
coupled sense either r = 0 or r = n. If r = 0, then Xi = 0 for all j. If
r = n, then, since we also have d = 0, the Xi's are nonsingular; we then
have m = n.
If A = B, we have AijXj = XiAij for all i, j. Fix p and let λ be an
eigenvalue of Xp with corresponding eigenspace Up(λ); note the subspace
Up(λ) is nonzero, because λ is an eigenvalue of Xp. From part 3 of Lemma 4.1,
we have Aij(Uj(λ)) ⊆ Ui(λ) for all i, j. Since D(A) is strongly connected,
18
Proposition 5.1 tells us the spaces Ui(λ) all have the same dimension; call
it f . Since Up(λ) is nonzero, we know f > 0. Hence, since A is not strongly
reducible in the coupled sense, we must have f = n, and Xi = λIn for
all i.
Remark 5.1. Earlier work [6] gives a proof, using block matrix computation,
for the case where all Aij's and Bij's are assumed to be nonsingular.
The proof of Theorem 5.1 uses the assumption that both D(A) and D(B)
are strongly connected in two ways: to establish that ni = n and mi = m
for all i, and to show that the relevant subspaces (kernels and ranges of
the Xi's) have the same dimension. We now develop another version of
Theorem 5.1, in which we weaken the hypothesis about the graphs, but then
need to explicitly assume that ni = n and mi = m for all i. The key point
for this second version is that Xi and Xj have the same rank whenever vi
and vj are strongly connected in either of the digraphs D(A) or D(B).
We use A and B to define an undirected graph, G(A,B), as follows.
Definition 5.2. The undirected graph, G(A,B), is the graph on vertices
{vi}i∈I, such that {vi, vj} is an (undirected) edge of G(A,B) if and only if
the vertices vi and vj are either strongly connected in D(A), or in D(B) (or
both). We call this the linked graph of A and B.
Proposition 5.3. Let A = {Aij}i,j∈I and B = {Bij}i,j∈I. Let Xi be ni × mi
and suppose AijXj = XiBij for all i, j ∈ I. If vi and vj are connected in
G(A, B) then Xi and Xj have the same rank. If G(A, B) is connected, then
all of the matrices Xi have the same rank.
Proof. Suppose vi and vj are connected in G(A,B). Then there is a sequence
of vertices, vi = vi1, vi2, vi3, . . . , vip−1, vip = vj, such that {vik, vik+1} is an edge
of G(A,B) for k = 1, . . . , p − 1. This means vik and vik+1 are either strongly
connected in D(A) or strongly connected in D(B), or both. Therefore,
rank(Xik) = rank(Xik+1) for k = 1, . . . , p−1, and rank(Xi) = rank(Xj).
If either D(A) or D(B) is strongly connected, then G(A,B) will be con-
nected. However, G(A,B) can be a connected graph even if neither of the
digraphs D(A) or D(B) is strongly connected. For example, suppose K = 3
and
A =
0
A21
0
A12
0
0
0
0
0
B =
0
0
B31
0 B13
0
0
0
0
,
19
v3
r
v3
r
✡
✡✣
✡
✡
✡
✡
✡✢
r
v1
✲✛
D(A)
r
v2
v1
r✡
r
v2
D(B)
v3
r
✡✡
✡
✡
✡
r✡
v1
r
v2
G(A,B)
Figure 1: D(A), D(B) and G(A, B)
where A12, A21, B13 and B31 are all nonsingular. Neither D(A) nor D(B) is
connected, but G(A,B) is connected. (See Figure 1.)
As an example, suppose I1 and I2 are nonempty, disjoint subsets of I
such that I = I1 ∪ I2. Partition the vertices of G(A, B) into two sets
corresponding to I1 and I2, setting
S = {vi i ∈ I1}
and
T = {vi i ∈ I2}.
Suppose rank(Aij) < nj and rank(Bij) < mj, whenever i ∈ I1 and j ∈ I2.
Then neither D(A) nor D(B) has any directed edges from vertices in S to
vertices in T . The linked graph G(A,B) then has no edges from vertices in S
to vertices in T and hence is not connected.
Now suppose that, whenever i ∈ I1 and j ∈ I2, we have rank(Aij) < nj
and rank(Bji) < mi, (note the reversal of subscripts on Bji). In this case,
D(A) has no directed edges from vertices in S to vertices in T , while D(B)
has no directed edges from vertices in T to vertices in S. Consequently, if
v ∈ S and w ∈ T , then the pair v, w is not strongly connected in either
D(A) or D(B). Hence, G(A,B) has no edges between vertices in S and
vertices in T ; thus G(A,B) is not connected.
The following variation of Theorem 5.1 uses this linked graph, G(A, B).
Theorem 5.2. Assume neither A = {Aij}i,j∈I nor B = {Bij}i,j∈I is strongly
reducible in the coupled sense. Assume also that ni = n and mi = m for all i,
and that G(A,B) is connected. Let Xi be n× m, and suppose AijXj = XiBij
for all i, j ∈ I. Then either Xi = 0 for all i, or Xi is nonsingular for all i. In
the latter case we must have m = n. If A = B and is a family of complex
matrices, then there is some scalar α such that Xi = αIn for all i.
Proof. By Proposition 5.3, the subspaces range(Xi), for i in I, all have the
same dimension, r. Since all of the Xi's have the same number of columns,
the rank plus nullity theorem tells us the subspaces ker(Xi), for i ∈ I, must
also all have the same dimension, d. The remainder of the proof is the same
as that for Theorem 5.1.
20
Comparing Theorems 4.2, 5.1, and 5.2, the simplest version is part 1 of
Theorem 4.2. It is the closest to the usual Schur's Lemma. However, the
hypothesis that A, B be irreducible in the coupled sense is more restrictive
than the hypothesis of part 2 of Theorem 4.2. The conclusion of part 2 has
more options for the Xi's than part 1. Theorems 5.1 and 5.2 apply to the
larger class of pairs, A, B, which are not strongly reducible in the coupled
sense, but have additional restrictions about the connectivity of the graphs
D(A), D(B), and G(A, B) and the equality of the ni's and mi's.
6 Normality and coupled normality
We now consider a refinement of Schur's Lemma for irreducible sets of normal
matrices. This is closely related to Lemma A.4 of [11]. We obtain correspond-
ing results for sets A, B satisfying a "coupled normality" condition. For this
section we work over the field of complex numbers. We use * to denote the
transpose conjugate of a matrix. If U is a subspace of V, we use U⊥ for the
orthogonal complement of U. We will need the following facts.
Proposition 6.1. Let A be a normal matrix; let S be nonsingular and let
B = S−1AS. Then the following are equivalent.
1. The matrix B is normal.
2. S−1A∗S = B∗.
3. The matrix SS∗ commutes with A.
4. The matrix SS∗ commutes with A∗.
5. The matrix S∗S commutes with B.
6. The matrix S∗S commutes with B∗.
Proof. The equivalence of 2, 3 and 4 is easily shown. Using B = S−1AS,
S−1A∗S = B∗ ⇐⇒ S−1A∗S = S∗A∗S−∗
⇐⇒ A∗SS∗ = SS∗A∗
⇐⇒ SS∗A = ASS∗,
21
where the third line comes from taking the transpose conjugate of the equa-
tion in the second line. A similar calculation, starting with A = SBS−1,
shows 2, 5 and 6 are equivalent:
SB∗S−1 = A∗ ⇐⇒ SB∗S−1 = S−∗B∗S∗
⇐⇒ S∗SB∗ = B∗S∗S
⇐⇒ BS∗S = S∗SB.
The fact that 2 implies 1 is also easy. If S−1A∗S = B∗, use AA∗ = A∗A to
get
BB∗ = (S−1AS)(S−1A∗S) = (S−1A∗S)(S−1AS) = B∗B.
The only part needing any work at all is to show 1 implies 2. Let λ1, . . . , λn
be the eigenvalues of A and let D be the diagonal matrix with diagonal entries
λ1, . . . , λn. Since A is normal, A = U∗DU for some unitary matrix U, and
A∗ = U∗DU, where the bar denotes complex conjugation. Note λ1, . . . , λn
are the eigenvalues of A∗. Let p(x) be a polynomial such that p(λi) = λi for
each eigenvalue λi. Then D = p(D), and
A∗ = U∗p(D)U = p(U∗DU) = p(A).
Since B is similar to A, the matrix B also has eigenvalues λ1, . . . , λn and B∗
If B is normal, B = V ∗DV for some unitary
has eigenvalues λ1, . . . , λn.
matrix V . Hence,
B∗ = V ∗DV = V ∗p(D)V = p(V ∗DV ) = p(B).
But p(B) = p(S−1AS) = S−1p(A)S = S−1A∗S, so B∗ = S−1A∗S.
This gives an easy proof of the following.
Theorem 6.1. Suppose {Ai}i∈I and {Bi}i∈I are irreducible families of nor-
mal matrices, and S is a nonsingular matrix such that S−1AiS = Bi for
all i ∈ I. Then S is a scalar multiple of a unitary matrix.
Proof. By the preceding proposition, SS∗ commutes with each Ai. Since
{Ai}i∈I is an irreducible family, SS∗ must be a scalar matrix. Since S is
nonsingular, the Hermitian matrix SS∗ is positive definite; hence SS∗ = αI
where α is a positive real number. Set U = 1√α S. Then UU∗ = 1
α SS∗ = I.
So U is unitary and S = √αU.
22
Remark 6.1. This argument is essentially the proof of Lemma A.4 of [11],
which says that if two irreducible representations of a ∗-algebra of square
matrices are equivalent, then they are similar via a unitary similarity. Let S
be the algebra generated by {Ai}i∈I and let T be the algebra generated by
{Bi}i∈I. For any normal matrix, N, the matrix N∗ is a polynomial in N, so
the algebras S and T are ∗-algebras, (which means that whenever A is in the
algebra, so is A∗). Let S be a nonsingular matrix such that S−1AiS = Bi for
all i. Proposition 6.1 tells us S−1A∗i S = B∗i for all i, so S may be extended
to an isomorphism of the ∗-algebras S and T in the usual way.
We now introduce the idea of coupled normality.
Definition 6.1. The family A = {Aij}i,j∈I is normal in the coupled sense if
for all i, j ∈ I we have A∗ijAij = AjiA∗ji.
If A is normal in the coupled sense, setting i = j gives A∗iiAii = AiiA∗ii,
so Aii is normal for all i. Note also that if Aji = A∗ij for all i, j, then A is
coupled normal. When I = {1, . . . , K}, the condition Aji = A∗ij for all i, j
holds when A is a Hermitian matrix. In the JISA model, A is a covariance
matrix, and hence is a real, symmetric matrix, so it is Hermitian.
Recall that, for any matrix G, the four matrices G, G∗, GG∗, and G∗G
all have the same rank. Hence, when A is normal in the coupled sense, the
In particular, note that Aij is
matrices Aij and Aji have the same rank.
nonsingular if and only if Aji is nonsingular.
Let C be a q × p matrix, let D be a p × q matrix, and let M be the
(p + q) × (p + q) matrix
C 0 (cid:19) ,
where the zero blocks are p × p and q × q. Then
M = (cid:18) 0 D
MM∗ = (cid:18) DD∗
0
0
CC∗(cid:19) and M∗M = (cid:18) C∗C
0
0
D∗D(cid:19) .
Hence, M is normal if and only if C∗C = DD∗ and D∗D = CC∗. The
if we set Mij = (cid:18) 0 Aij
0 (cid:19),
connection with coupled normality is this:
then A = {Aij}i∈I is normal in the coupled sense if and only if Mij is normal
for all i, j ∈ I.
Suppose A is normal in the coupled sense and the subspaces {Ui}i∈I
satisfy Aij(Uj) ⊆ Ui for all i, j. Let di be the dimension of Ui. We use the
fact that Mij is normal to show that Aij(U⊥j ) ⊆ U⊥i
for all i, j.
Aji
23
Proposition 6.2. Let C, D be matrices of sizes q× p and p× q, respectively,
such that C∗C = DD∗ and D∗D = CC∗. Suppose there are subspaces U
of Cp, and W of Cq, such that C(U) ⊆ W and D(W) ⊆ U. Then C(U⊥) ⊆
W⊥ and D(W⊥) ⊆ U⊥.
Proof. Let M = (cid:18) 0 D
C 0 (cid:19). For any x ∈ Cp and y ∈ Cq,
M (cid:18) x
y(cid:19) = (cid:18) Dy
Cx(cid:19) .
If x ∈ U and y ∈ W, then Dy ∈ U and Cx ∈ W. So U ⊕ W, (which is a
subspace of Cp⊕Cq), is invariant under M. Since M is normal, the orthogonal
complement of U ⊕ W in Cp ⊕ Cq must also be invariant under M. Hence,
U⊥ ⊕ W⊥ is invariant under M. This means that, for x ∈ U⊥ and y ∈ W⊥,
we have Dy ∈ U⊥ and Cx ∈ W⊥. So C(U⊥) ⊆ W⊥ and D(W⊥) ⊆ U⊥.
Aji
Apply Proposition 6.2 to the normal matrix Mij = (cid:18) 0 Aij
0 (cid:19), to get
Aij(U⊥j ) ⊆ U⊥i
for all i, j. Hence, if A is normal in the coupled sense, and is
reducible in the coupled sense, then it is fully reducible in the coupled sense,
because we can form Tj using a basis for Uj for the first dj columns and a
basis for U⊥j
for the remaining n − dj columns. If we use orthonormal bases
for Uj and U⊥j , then Tj will be unitary. Hence A is fully reducible in the
coupled sense with a coupled unitary similarity.
We will give three versions of Theorem 6.1 for A, B which are normal in
the coupled sense, corresponding to the three types of reducibility. The proofs
depend on the following proposition. The first two statements are a "coupled"
version of Proposition 6.1. Part 4 uses the digraphs D(A) and D(B).
Proposition 6.3. Assume the families A = {Aij}i,j∈I and B = {Bij}i,j∈I,
where Aij and Bij are complex matrices, are normal in the coupled sense.
Suppose AijSj = SiBij for all i, j, where Si is ni × mi. For any i ∈ I, and
any scalar α, define
Ui(α) = {v SiS∗i v = αv} and Yi(α) = {w S∗i Siw = αw}.
Then the following hold.
1. If Si is nonsingular then SiS∗i commutes with Aii, and S∗i Si commutes
with Bii.
24
2. If Si and Sj are both nonsingular,
SiS∗i Aij = AijSjS∗j
and S∗i SiBij = BijS∗j Sj.
3. If Si and Sj are both nonsingular,
Aij(Uj(α)) ⊆ Ui(α)
and Bij(Yj(α)) ⊆ Yi(α).
If Aij is also nonsingular, then dim(Ui(α)) = dim(Uj(α)).
If Bij is also nonsingular, then dim(Yi(α)) = dim(Yj(α)).
4. Assume Si is nonsingular for all i ∈ I. Then the following hold.
If vi and vj are strongly connected in D(A), then Ui(α) and Uj(α) have
the same dimension.
If vi and vj are strongly connected in D(B), then Yi(α) and Yj(α) have
the same dimension.
5. If α 6= 0 then Ui(α) and Yi(α) have the same dimension.
6. Assume Si is nonsingular for all i ∈ I. Then if vi and vj are connected
in G(A, B), and α 6= 0, we have dim(Ui(α)) = dim(Uj(α)).
Proof. Suppose Si is nonsingular. Since Aii and Bii are both normal, and
S−1
i AiiSi = Bii, Proposition 6.1 tells us that SiS∗i commutes with Aii and
S∗i Si commutes with Bii.
Now suppose i 6= j, and Si and Sj are both nonsingular. Set Mij =
Aji
0 (cid:19). Then
(cid:18) 0 Aij
Mij (cid:18) Si
0 Sj (cid:19) = (cid:18) 0 Aij
Aji
0
0
0 Sj (cid:19)
AijSj
0 (cid:19)(cid:18) Si
0 (cid:19)
0 (cid:19)
SiBij
AjiSi
= (cid:18) 0
= (cid:18) 0
= (cid:18) Si
SjBji
0
0 Sj (cid:19)(cid:18) 0 Bij
0 (cid:19) .
Bji
25
So,
0 Sj (cid:19)−1
(cid:18) Si
0
Mij (cid:18) Si
0 Sj (cid:19) = (cid:18) 0 Bij
0 (cid:19) .
Bji
0
0 (cid:19)
Since A and B are both normal in the coupled sense, Mij and (cid:18) 0 Bij
are both normal. Set S = (cid:18) Si
0 Sj (cid:19). Proposition 6.1 tells us that SS∗
Bji
0
commutes with Mij. Hence,
(cid:18) SiS∗i
0
0
SjS∗j (cid:19)(cid:18) 0 Aij
0 (cid:19) = (cid:18) 0 Aij
0 (cid:19)(cid:18) SiS∗i
Aji
Aji
0
SjS∗j (cid:19) ,
0
and SiS∗i Aij = AijSjS∗j . Use the fact that S∗S commutes with (cid:18) 0 Bij
0 (cid:19)
to show S∗i SiBij = BijS∗j Sj for all i, j.
Bji
For part 3, assume Si and Sj are nonsingular. Let v ∈ Uj(α). By part 2,
SiS∗i (Aijv) = Aij(SjS∗j v) = α(Aijv). This shows Aij(Uj(α)) ⊆ Ui(α). If Aij
is nonsingular, dim(Aij(Uj(α))) = dim(Uj(α)), so dim(Uj(α)) ≤ dim(Ui(α)).
Since A is coupled normal, Aji is also nonsingular, giving the reverse inequal-
ity, so dim(Uj(α)) = dim(Ui(α)). The corresponding facts for B come from
the same argument, using S∗i SiBij = BijS∗j Sj.
For part 4, assume vi and vj are strongly connected in D(A). Proposi-
tion 5.1, together with part 3, gives dim(Ui(α)) = dim(Ui(α)). The same
argument applies when vi and vj are strongly connected in D(B)
Part 5 comes from the fact that S∗j Sj and SjS∗j have the same nonzero
eigenvalues with the same multiplicities.
For part 6, suppose vi and vj are connected in G(A, B). Then there is a
sequence of vertices, vi = vi1, vi2, vi3, . . . , vip−1, vip = vj, such that {vik , vik+1}
is an edge of G(A,B) for k = 1, . . . , p− 1. This means vik and vik+1 are either
strongly connected in D(A) or strongly connected in D(B) (or both). If vik
and vik+1 are strongly connected in D(A), then dim(Uik (α)) = dim(Uik+1(α))
by part 4. If vik and vik+1 are strongly connected in D(B), then part 4 tells
us dim(Yik(α)) = dim(Yik+1(α)). But, since α is nonzero, Ui(α) and Yi(α)
have the same dimension. So, in either case, dim(Uik(α)) = dim(Uik+1(α))
for 1 ≤ k ≤ p − 1, and hence dim(Ui(α)) = dim(Uj(α)).
With these preliminaries completed, we state and prove a version of
Schur's Lemma for A, B that are normal in the coupled sense. The three
cases correspond to the three types of coupled reducibility.
26
Theorem 6.2. Let A = {Aij}i,j∈I and B = {Bij}i,j∈I where Aij is ni × nj
and Bij is mi × mj. Assume A and B are normal in the coupled sense.
Suppose Si is ni × mi and AijSj = SiBij for all i, j.
1. If A and B are both irreducible in the coupled sense, then either Si = 0
for all i, or there is a scalar α such that every Si is α times a unitary
matrix; i.e., Si = αUi, where Ui is unitary. In the latter case, mi = ni
for all i. Furthermore, if A = B, then there is a scalar β such that
Si = βIni for all i.
2. If neither A nor B is properly reducible in the coupled sense, then, for
each i, either Si = 0 or Si is a scalar multiple of a unitary matrix. In
the latter case, mi = ni. Furthermore, if A = B, then every Si is a
scalar matrix.
3. Suppose neither A nor B is strongly reducible in the coupled sense.
Assume also that ni = n and mi = m for all i ∈ I, and that the graph
G(A,B) is connected. Then either Si = 0 for all i, or there is a scalar α
such that each Si is a α times a unitary matrix; i.e., Si = αUi, where Ui
is unitary. In the latter case we must have m = n. Furthermore, if
A = B, then there is some scalar β such that Si = βIn for all i.
Proof. The proofs are similar to those of Theorems 4.2 and 5.1.
Suppose A and B are both irreducible in the coupled sense. Part 1 of
Theorem 4.2 tells us that, either Si = 0 for all i, or Si is nonsingular for all i.
In the latter case we must have mi = ni for all i. Suppose Si is nonsingular
for all i. Fix p and let λ be an eigenvalue of SpS∗p . Proposition 6.3 gives
Aij(Uj(λ)) ⊆ Ui(λ) for all i, j. Since A is irreducible in the coupled sense,
either all of the subspaces Ui(λ) are zero, or Ui(λ) = Vi for all i ∈ I. Since λ
is an eigenvalue of SpS∗p, the space Up(λ) is nonzero. Therefore, Ui(λ) = Vi
for all i, and SiS∗i = λIni for all i. Since SiS∗i
is positive definite, λ is a
positive real number and Ui = 1√λ
Si is a unitary matrix.
For the second version, assume neither A nor B is properly reducible in
the coupled sense. From part 2 of Theorem 4.2, we know that, for each i,
either Si = 0 or Si is nonsingular. If Si is nonsingular we must have mi = ni.
Suppose Sp is nonsingular for some p. Let λp be an eigenvalue of SpS∗p .
Since Sp is nonsingular, λp 6= 0. Let N denote the set of all q such that Sq is
nonsingular. Consider the statement
Aij(Uj(λp)) ⊆ Ui(λp).
27
(6)
If i, j are both in N , then Si, Sj are both nonsingular and Proposition 6.3
tells us (6) holds. If j /∈ N , then Sj = 0, and hence, since λp is nonzero,
Uj(λp) = {0}, so (6) holds. Finally, if i /∈ N but j ∈ N , then Si = 0 and Sj
is nonsingular. In this case, AijSj = SiBij tells us Aij = 0, and (6) holds.
Hence, (6) holds for all i, j. Since A is not strongly reducible in the coupled
sense, there are only two possibilities for each Ui(λ): it is either zero or the
whole space Vi. Since λp is an eigenvalue of SpS∗p, we know Up(λp) is nonzero;
therefore it must be the whole space and SpS∗p = λpInp. Since SpS∗p is positive
definite, λp is a positive real number and Up = 1√λp
Sp is a unitary matrix.
Finally, consider the third version, where we assume neither A nor B
is strongly reducible in the coupled sense and G(A,B) is connected. From
Theorem 5.2, either Si = 0 for all i, or Si is nonsingular for all i. In the
latter case, m = n.
Suppose Si is nonsingular for all i. Fix p and let λ be an eigenvalue
of SpS∗p . Since Sp is nonsingular, λ 6= 0. From Proposition 6.3, we have
Aij(Uj(λ)) ⊆ Ui(λ), for all i, j, and the subspaces Ui(λ) all have the same
dimension. Let f be the dimension of these subspaces. Since λ is an eigen-
value of SpS∗p, the eigenspace Up(λ) is nonzero. Hence, f > 0. Since A is
not strongly reducible in the coupled sense we must have f = n. There-
fore SiS∗i = λIn for all i, the number λ must be a positive real number and
Ui = 1√λ
Si is a unitary matrix.
7 Appendix
We construct examples to establish the claims made in Section 3.
Let I be the index set; let {ni}i∈I be a family of positive integers.
If
ni = 1, set Ni = (0). If ni ≥ 2, let Ni be the ni × ni matrix with a 1 in each
superdiagonal entry and zeroes elsewhere. This is the standard nilpotent
matrix used in the blocks of the Jordan canonical form. For any x ∈ Fni,
Nix =
0 1
0 0
...
...
0 0
0 0
0 · · · 0
1 · · · 0
...
...
0 · · · 1
0 · · · 0
x1
x2
...
xni−1
xni
=
x2
x3
...
xni
0
.
Multiplying x on the left by Ni moves the coordinates up one position and
puts a 0 in the last entry. Let ei
j denote the vector with ni coordinates that
28
has a 1 in entry j and zeroes in all other positions. Thus, ei
ni are
the unit coordinate vectors for Fni. Then Niei
j−1. Henceforth, we omit
the superscript i on ej, as the number of coordinates will be clear from the
context. For example, if we write Aijv, then it is understood that v has nj
coordinates.
1, . . . , ei
j = ei
Here is the key fact used in the examples.
Proposition 7.1. For n ≥ 2, let N be the n × n matrix with a 1 in each
superdiagonal entry and zeroes elsewhere. Suppose U is a nonzero, proper
invariant subspace of N. Then e1 ∈ U and en /∈ U.
Proof. Let x be a nonzero vector in U, and let xk be the last nonzero coor-
dinate of x, i.e., xk+1 = · · · = xn = 0. Then N k−1x = xke1, so e1 ∈ U.
For the second part, note that N n−1en, N n−2en, . . . , N en, en are the unit
coordinate vectors e1, . . . , en. Hence, if en ∈ U, then U is the whole space V.
Since U is a proper subspace of V, the vector en cannot be in U.
Remark 7.1. Let Yj be the j-dimensional subspace spanned by e1, . . . , ej,
i.e., the set of all vectors with zeroes in the last n − j entries. A similar
argument shows that the nonzero invariant subspaces of N are the subspaces
Y1, . . . ,Yn.
We now construct some examples.
Example 7.1. Assume I ≥ 2 and that np ≥ 2 for some p ∈ I. Define A
as follows.
1. Aii = Ni for all i ∈ I.
2. If j 6= p, set Apj = 0.
3. If i 6= p let Aip be any matrix which has eni in the first column.
4. If i 6= p, and j 6= p, and i 6= j, then Aij can be any ni × nj matrix.
Set Ui = Vi for i 6= p, and let Up be the line spanned by e1. Since np ≥ 2, the
subspace Up is a nonzero, proper subspace of Vp. One can easily check that
the subspaces {Ui}i∈I properly reduce A.
We now show A is not strongly reducible in the coupled sense. Suppose
there were nonzero, proper subspaces {Ui}i∈I that reduced A. (Note we must
then have ni ≥ 2 for all i.) Each Ui is a nonzero, proper invariant subspace
29
of Ni, so e1 ∈ Ui and eni /∈ Ui. Choose i 6= p. Then Aip has eni in its first
column, so Aipe1 = eni. But e1 ∈ Up and eni /∈ Ui, so Aip(Up) 6⊆ Ui. Hence,
we have a contradiction, and A is not strongly reducible in the coupled sense.
Example 7.1 shows A can be properly reducible in the coupled sense
without being strongly reducible. Thus, for any field F, when I ≥ 2 and
ni ≥ 2 for at least one i, we have StrRed(F,{ni}i∈I) ⊂ P ropRed(F,{ni}i∈I).
The next example shows that if I ≥ 4, and ni ≥ 2 for at least one value
of i, we have P ropRed(F,{ni}i∈I) ⊂ Red(F,{ni}i∈I).
Example 7.2. Assume I ≥ 4 and that np ≥ 2 for some p ∈ I. Choose any
q ∈ I, with q 6= p, and define A as follows.
1. Aii = Ni for all i ∈ I.
2. For all i with i 6= p and i 6= q, set Aip = 0 and Aiq = 0.
3. For all other choices of i, j with i 6= j, let Aij be any matrix with en1
in the first column.
We illustrate for I = {1, 2, . . . , K}, with p = 1 and q = 2.
A =
∗
N1
∗ N2
0
0
0
...
0
0
...
0
(cid:12)(cid:12)(cid:12)
(cid:12)(cid:12)(cid:12)
(cid:12)(cid:12)(cid:12)
(cid:12)(cid:12)(cid:12)
(cid:12)(cid:12)(cid:12)(cid:12)
(cid:12)(cid:12)(cid:12)
∗
∗
∗
∗
∗
N3
∗ N4
...
...
· · ·
· · ·
· · ·
· · ·
. . .
∗
∗
∗
∗
...
∗
∗
· · · NK
,
where each asterisk (∗) represents an ni × nj matrix with eni in the first
column.
Set Up = Vp, and Uq = Vq. For all other values of i, set Ui = 0. One can
check that A is coupled reducible via {Ui}i∈I.
We now show A is not properly reducible. Suppose A could be properly
reduced by subspaces {Ui}i∈I. At least one Ui must be a nonzero, proper
subspace; we first show this holds for at most one value of i. Suppose Ui
and Uj were both nonzero, proper subspaces, with i 6= j. We must then
30
have ni ≥ 2 and nj ≥ 2. Since Ui is a nonzero, proper invariant subspace
of Ni, and Uj is a nonzero, proper invariant subspace of Nj, we know e1 is
in both Ui and Uj, and eni /∈ Ui and enj /∈ Uj. If i = p and j = q, use the
matrix Apq. Since Apqe1 = enp, we see Apq(Uq) 6⊆ Up. The same argument,
using Aqp, applies when i = q and j = p. Suppose then that at least one of
i, j is different from p and q. Without loss of generality, assume j 6∈ {p, q}.
Then use Aij, which has eni in its first column. So Aij(Uj) 6⊆ Ui.
So, at most one Ui is a proper, nonzero subspace; each of the other sub-
spaces is either the whole space Vi or the zero subspace. Assume Ui is the
nonzero, proper subspace; note ni ≥ 2. We claim we can then choose j 6= i
so that Aij has eni in the first column and Aji has enj in the first column. If
i = p, choose j = q, and if i = q, choose j = p. If i 6= p and i 6= q, choose
any j which is different from i, p and q. (This is where we use the fact that
I ≥ 4.) The subspace Uj is either the full space Vj, or it is the zero sub-
space. If Uj = Vj, then Aij(Vj) contains Aije1 = eni, which is not in Ui. So
Aij(Uj) 6⊆ Ui. If Uj = {0}, then, since e1 ∈ Ui, we have Ajie1 = enj ∈ Aij(Ui).
So Aji(Ui) 6⊆ Uj. Hence, A is not properly reducible in the coupled sense.
In the example above, we needed I ≥ 4. What can we say when K = 2
or K = 3? In these cases, the field F must be considered. The reason is, that
for A to be properly reducible in the coupled sense, at least one Aii must have
a nonzero, proper invariant subspace. If F is algebraically closed and n ≥ 2,
then any n × n matrix over F has an eigenvalue in F, and the line spanned
by a corresponding eigenvector is a nonzero, proper invariant subspace. But
if F is not algebraically closed, there may be n × n matrices over F which
have no proper invariant subspaces. We shall give an example for the real
numbers later, but first we show that if F is an algebraically closed field, then
P ropRed(F, n, 2) = Red(F, n, 2) and P ropRed(F, n, 3) = Red(F, n, 3) for all
n ≥ 2.
We use the following lemma to deal with the cases K = 2 and K = 3.
Lemma 7.1. Let A = {Aij}i,j∈I where Aij is ni × nj. Suppose A is coupled
reducible with {Ui}i∈I satisfying one of the following.
1. Up = Vp for exactly one index value p, and Ui = {0} when i 6= p.
2. Up = {0} for exactly one index value p and Ui = Vi when i 6= p.
Suppose Wp is a nonzero, proper invariant subspace of App. Then A is prop-
erly reducible by coupled similarity via the subspaces obtained by replacing
Up by Wp, and leaving the other Ui's unchanged.
31
Proof. Since Up is the only subspace that is changed, we continue to have
Aij(Uj) ⊆ Ui whenever i and j are both different from p. Also, Wp is chosen
to satisfy App(Wp) ⊆ Wp. It remains to consider Aip and Api for i 6= p.
In case 1, we have Ui = {0} for i 6= p, so Api(Ui) = {0} ⊆ Wp. We also
have Aip(Up) ⊆ Ui = {0}. Since Up = Vp, we must have Aip(Wp) = {0} = Ui.
In case 2, we have Ui = Vi for i 6= p, and Up = {0}. So Api(Ui) = {0} ⊆
Wp. We also have Aip(Wp) ⊆ Vi = Ui for i 6= p.
Now suppose F is algebraically closed, and n ≥ 2. Any n × n matrix
over F has a nonzero, proper invariant subspace. For K = 2, Lemma 7.1
immediately tells us that A is coupled reducible if and only if it is properly
reducible, i.e., P ropRed(F, n, 2) = Red(F, n, 2). For the case K = 3, suppose
A is reduced by U1,U2,U3. If none of the Ui's is a nonzero proper subspace,
then each is either V or 0, so either two of them are V, with the third being
zero, or vice versa, two of them are zero, with the third being V. Lemma 7.1
then tells us A is properly reducible. Hence, for algebraically closed F and
n ≥ 2 we have P ropRed(F, n, 3) = Red(F, n, 3).
If F is not algebraically closed, then a matrix over F need not have a
proper invariant subspace. Consider the case F = R, the field of real numbers.
Let A be a real n×n matrix, where n ≥ 2. The eigenvalues of A are in C, and
the non-real eigenvalues occur in conjugate pairs. If λ is a real eigenvalue of
A then there is a corresponding real eigenvector, v, and the line spanned by v
is a proper, nonzero invariant subspace of A. For a pair of complex conjugate,
non-real eigenvalues, λ, λ, there is a corresponding two dimensional invariant
subspace.
Consider the following example for K ≥ 2 and 2 × 2 real matrices.
Example 7.3. Choose an angle θ with 0 < θ < π. For 1 ≤ i ≤ K, set
Aii = (cid:18) cos θ − sin θ
cos θ (cid:19) .
sin θ
This is the matrix for rotation of the plane R2 by angle θ. Since no line
through the origin is mapped to itself by this rotation, this map has no
nonzero, proper invariant subspace. Hence, for any choice of the Aij's when
i 6= j, the set A is not properly reducible in the coupled sense. It is, however,
possible to find Aij's such that A is reducible in the coupled sense. Choose
a positive integer s with 1 ≤ s < K and set Aij = 0 whenever i > s and
j ≤ s. Set Ui = R2 for 1 ≤ i ≤ s and Ui = {0} for s + 1 ≤ i ≤ K. It is
32
easy to check that the subspaces U1, . . . ,UK reduce A. For, when i and j are
both less than or equal to s, we have Ui = Uj = R2, and hence Aij(Uj) ⊆ Ui.
If i and j are both greater than s, then Ui = Uj = {0}, so Aij(Uj) ⊆ Ui. If
i > r and j ≤ s, then Aij = 0; hence Aij(Uj) = {0} ⊆ Ui. Finally, if i ≤ s
and j > s, then Uj = {0} so Aij(Uj) = {0} ⊆ Ui. So A is reducible in the
coupled sense, but not properly reducible.
So for K ≥ 2, we have P ropRed(R, 2, K) ⊂ Red(R, 2, K). From Ex-
ample 7.2, we already knew this for K ≥ 4; the new information is that
P ropRed(R, 2, 2) ⊂ Red(R, 2, 2) and P ropRed(R, 2, 3) ⊂ Red(R, 2, 3).
However, for n ≥ 3, any n × n real matrix has a nonzero proper invari-
ant subspace. Lemma 7.1 then gives P ropRed(R, n, 2) = Red(R, n, 2) and
P ropRed(R, n, 3) = Red(R, n, 3) when n ≥ 3.
Acknowledgements. D. Lahat thanks Dr. Jean-Fran¸cois Cardoso for in-
sightful discussions during her Ph.D., which set the foundations for and mo-
tivated this work.
References
[1]
[2]
[3]
[4]
[5]
[6]
T. Adalı, D. Lahat, and C. Jutten. Data fusion: Benefits of fully
exploiting diversity. Tutorial presented at EUSIPCO, Aug. 2016.
Michael Artin. Algebra, 2nd edition, Pearson Prentice Hall, New
Jersey, 2011.
Brualdi and Ryser, Combinatorial Matrix Theory, Cambridge Uni-
versity Press, 1991.
D. Lahat. A data fusion perspective on source separation. Talk
presented at the Hausdorff School: Low-rank Tensor Techniques in
Numerical Analysis and Optimization, Apr. 2016.
D. Lahat and C. Jutten, Joint blind source separation of multidi-
mensional components: Model and algorithm,
in Proc. EUSIPCO,
Lisbon, Portugal, Sep. 2014, pp. 1417 -- 1421.
D. Lahat and C. Jutten, A generalization to Schurs lemma with
an application to joint independent subspace analysis, Technical
33
[7]
[8]
[9]
[10]
[11]
[12]
Report hal-01247899, GIPSA-Lab, Grenoble, France, Dec. 2015,
https://hal.archives-ouvertes.fr/hal-01247899.
D. Lahat and C. Jutten. On the uniqueness of coupled matrix block
diagonalization in the joint analysis of multiple datasets (talk). In
SIAM Conference on Applied Linear Algebra, Atlanta, GA, USA,
Oct. 2015, http://meetings.siam.org/sess/dsp talk.cfm?p=72077.
D. Lahat and C. Jutten, Joint independent subspace analysis using
second-order statistics, IEEE Trans. Signal Process., 64 (2016), pp.
4891 -- 4904, https://doi.org/10.1109/tsp.2016.
D. Lahat and C. Jutten. Joint independent subspace analysis by cou-
pled block decomposition: Non-identifiable cases. In Proc. ICASSP,
Calgary, Canada, Apr. 2018.
D. Lahat and C. Jutten,
Joint independent subspace analysis:
Uniqueness and identifiability IEEE Trans. Signal Process., 2018,
to appear, DOI 10.1109/TSP.2018.2880714
K. Murota, Y. Kanno, M. Kojima, and S. Kojima, A numerical
algorithm for block-diagonal decomposition of matrix *-algebras with
application to semidefinite programming, Japan J. Indust. Appl.
Math., 27 (2010), pp. 125 -- 160.
I. Schur,
Neue Begrundung der Theorie der Gruppencharak-
tere, Sitzungsberichte der Koniglich-Preussischen Akademie der Wis-
senschaften zu Berlin, Berlin, Germany, Jan. -- Jun. 1905, pp. 406 -- 432,
https://archive.org/details/sitzungsberichte1905deutsch.
[13]
J. J. Sylvester. Sur l'´equation en matrices px=xq. Comptes Rendus,
99: 67 -- 71, 1884.
34
|
0912.3635 | 3 | 0912 | 2011-03-30T14:42:28 | Algebraic Geometry of Topological Spaces I | [
"math.RA",
"math.FA",
"math.GN",
"math.KT"
] | We use techniques from both real and complex algebraic geometry to study K-theoretic and related invariants of the algebra C(X) of continuous complex-valued functions on a compact Hausdorff topological space X. For example, we prove a parametrized version of a theorem of Joseph Gubeladze; we show that if M is a countable, abelian, cancellative, torsion-free, seminormal monoid, and X is contractible, then every finitely generated projective module over C(X)[M] is free. The particular case when M=N^n gives a parametrized version of the celebrated theorem proved independently by Daniel Quillen and Andrei Suslin that finitely generated projective modules over a polynomial ring over a field are free. The conjecture of Jonathan Rosenberg which predicts the homotopy invariance of the negative algebraic K-theory of C(X) follows from the particular case when M=Z^n. We also give algebraic conditions for a functor from commutative algebras to abelian groups to be homotopy invariant on C*-algebras, and for a homology theory of commutative algebras to vanish on C*-algebras. These criteria have numerous applications. For example, the vanishing criterion applied to nil-K-theory implies that commutative C*-algebras are K-regular. As another application, we show that the familiar formulas of Hochschild-Kostant-Rosenberg and Loday-Quillen for the algebraic Hochschild and cyclic homology of the coordinate ring of a smooth algebraic variety remain valid for the algebraic Hochschild and cyclic homology of C(X). Applications to the conjectures of Beilinson-Soule and Farrell-Jones are also given. | math.RA | math |
ALGEBRAIC GEOMETRY OF TOPOLOGICAL SPACES I
GUILLERMO CORTI NAS AND ANDREAS THOM
Abstract. We use techniques from both real and complex algebraic geometry to study K-
theoretic and related invariants of the algebra C(X) of continuous complex-valued functions
on a compact Hausdorff topological space X. For example, we prove a parametrized version
of a theorem of Joseph Gubeladze; we show that if M is a countable, abelian, cancella-
tive, torsion-free, seminormal monoid, and X is contractible, then every finitely generated
projective module over C(X)[M ] is free. The particular case M = Nn
0 gives a parametrized
version of the celebrated theorem proved independently by Daniel Quillen and Andrei Suslin
that finitely generated projective modules over a polynomial ring over a field are free. The
conjecture of Jonathan Rosenberg which predicts the homotopy invariance of the negative
algebraic K-theory of C(X) follows from the particular case M = Zn. We also give alge-
braic conditions for a functor from commutative algebras to abelian groups to be homotopy
invariant on C ∗-algebras, and for a homology theory of commutative algebras to vanish
on C ∗-algebras. These criteria have numerous applications. For example, the vanishing cri-
terion applied to nil-K-theory implies that commutative C ∗-algebras are K-regular. As
another application, we show that the familiar formulas of Hochschild-Kostant-Rosenberg
and Loday-Quillen for the algebraic Hochschild and cyclic homology of the coordinate ring
of a smooth algebraic variety remain valid for the algebraic Hochschild and cyclic homology
of C(X). Applications to the conjectures of Beılinson-Soul´e and Farrell-Jones are also given.
Contents
Introduction.
1.
2. Split-exactness, homology theories, and excision.
2.1. Set-valued split-exact functors on the category of compact Hausdorff spaces.
2.2. Algebraic K-theory.
2.3. Homology theories and excision.
2.4. Milnor squares and excision.
3. Real algebraic geometry and split exact functors.
3.1. General results about semi-algebraic sets.
3.2. The theorem on split exact functors and proper maps.
4. Large semi-algebraic groups and the compact fibration theorem.
4.1. Large semi-algebraic structures.
4.2. Construction of quotients of large semi-algebraic groups.
2
8
8
10
11
13
13
14
16
17
17
20
Key words and phrases. algebraic K-theory, Serre's Conjecture, projective modules, rings of continuous
functions, algebraic approximation.
Cortinas' research was partly supported by grants PICT 2006-00836, UBACyT-X057, and MTM2007-
64704. Thom's research was partly supported by the DFG (GK Gruppen und Geometrie Gottingen).
1
2
GUILLERMO CORTI NAS AND ANDREAS THOM
4.3. Large semi-algebraic sets as compactly generated spaces.
4.4. The compact fibration theorem for quotients of large semi-algebraic groups.
5. Algebraic compactness, bounded sequences and algebraic approximation.
5.1. Algebraic compactness.
5.2. Bounded sequences, algebraic compactness and K0-triviality.
5.3. Algebraic approximation and bounded sequences.
5.4. The algebraic compactness theorem.
6. Applications: projective modules, lower K-theory, and bundle theory.
6.1. Parametrized Gubeladze's theorem and Rosenberg's conjecture.
6.2. Application to bundle theory: local triviality.
7. Homotopy invariance.
7.1. From compact polyhedra to compact spaces: a result of Calder-Siegel.
7.2. Second proof of Rosenberg's conjecture.
7.3. The homotopy invariance theorem.
7.4. A vanishing theorem for homology theories.
8. Applications of the homotopy invariance and vanishing homology theorems.
8.1. K-regularity for commutative C ∗-algebras.
8.2. Hochschild and cyclic homology of commutative C ∗-algebras.
8.3. The Farrell-Jones Isomorphism Conjecture.
8.4. Adams operations and the decomposition of rational K-theory.
References
21
22
23
23
24
25
26
27
27
29
29
29
31
32
34
34
34
35
37
38
39
1. Introduction.
In his foundational paper [39], Jean-Pierre Serre asked whether all finitely generated
projective modules over the polynomial ring k[t1, . . . , tn] over a field k are free. This ques-
tion, which became known as Serre's conjecture, remained open for about twenty years. An
affirmative answer was given independently by Daniel Quillen [35] and Andrei Suslin [42].
Richard G. Swan observed in [45] that the Quillen-Suslin theorem implies that all finitely
generated projective modules over the Laurent polynomial ring k[t1, t−1
n ] are free.
1
This was later generalized by Joseph Gubeladze [20], [19], who proved, among other things,
that if M is an abelian, cancellative, torsion-free, seminormal monoid, then every finitely
generated projective module over k[M] is free. Quillen-Suslin's theorem and Swan's theorem
are the special cases M = Nn
0 and M = Zn of Gubeladze's result. On the other hand, it
is classical that if X is a contractible compact Hausdorff space, then all finitely generated
projective modules over the algebra C(X) of complex-valued continuous functions on X --
which by another theorem of Swan, are the same thing as locally trivial complex vector
bundles on X -- are free. In this paper we prove (see 6.1.3):
. . . , tn, t−1
Theorem 1.1. Let X be a contractible compact space, and M a countable, cancellative,
torsion-free, seminormal, abelian monoid. Then every finitely generated projective module
over C(X)[M] is free.
ALGEBRAIC GEOMETRY OF TOPOLOGICAL SPACES I
3
Moreover we show (Theorem 6.2.1) that bundles of finitely generated free C[M]-modules
over a not necessarily contractible, compact Hausdorff X which are direct summands of
trivial bundles, are locally trivial. The case M = Nn
0 of Theorem 1.1 gives a parametrized
version of Quillen-Suslin's theorem. The case M = Zn is connected with a conjecture of
Jonathan Rosenberg [38] which predicts that the negative algebraic K-theory groups of
C(X) are homotopy invariant for compact Hausdorff X. Indeed, if R is any ring, then the
negative algebraic K-theory group K−n(R) is defined as a certain canonical direct summand
of K0(R[Zn]); the theorem above thus implies that K−n(C(X)) = 0 if X is contractible.
Using this and excision, we derive the following result (see Theorem 6.1.5).
Theorem 1.2. Let Comp be the category of compact Hausdorff spaces and let n > 0. Then
the functor Comp → Ab, X 7→ K−n(C(X)) is homotopy invariant.
A partial result in the direction of Theorem 1.2 was obtained by Eric Friedlander and
Mark E. Walker in [15]. They proved that K−n(C(∆p)) = 0 for p ≥ 0, n > 0. In Section 7.2,
we give a second proof of Theorem 1.2 which uses the Friedlander-Walker result. Elaborating
on their techniques, and combining them with our own methods, we obtain the following
general criterion for homotopy invariance (see Theorem 7.3.1).
Theorem 1.3. Let F be a functor on the category Comm/C of commutative C-algebras with
values in the category Ab of abelian groups. Assume that the following three conditions are
satisfied.
(i) F is split-exact on C ∗-algebras.
(ii) F vanishes on coordinate rings of smooth affine varieties.
(iii) F commutes with filtering colimits.
Then the functor
is homotopy invariant and F (C(X)) = 0 for X contractible.
Comp → Ab, X 7→ F (C(X)),
Observe that K−n satisfies all the hypothesis of the theorem above (n > 0). This gives
a third proof of Theorem 1.2. We also use Theorem 1.3 to prove the following vanishing
theorem for homology theories (see 7.4.1). In this paper a homology theory on a category C
of algebras is simply a functor E : C → Spt to the category of spectra which preserves finite
products up to homotopy.
Theorem 1.4. Let E : Comm/C → Spt be a homology theory of commutative C-algebras
and n0 ∈ Z. Assume that the following three conditions are satisfied.
(i) E satisfies excision on commutative C ∗-algebras.
(ii) En commutes with filtering colimits for n ≥ n0.
(iii) En(O(V )) = 0 for each smooth affine algebraic variety V for n ≥ n0.
Then En(A) = 0 for every commutative C ∗-algebra A for n ≥ n0.
4
GUILLERMO CORTI NAS AND ANDREAS THOM
Recall that a ring R is called K-regular if
coker (Kn(R) → Kn(R[t1, . . . , tp])) = 0
(p ≥ 1, n ∈ Z).
As an application of Theorem 1.4 to the homology theory F p(A) = hocofiber(K(A ⊗C
O(V )) → K(A ⊗C O(V )[t1, . . . , tp]) where V is a smooth algebraic variety, we obtain the
following (Theorem 8.1.1).
Theorem 1.5. Let V be a smooth affine algebraic variety over C, R = O(V ), and A a
commutative C ∗-algebra. Then A ⊗C R is K-regular.
The case R = C of the previous result was discovered by Jonathan Rosenberg, see
Remark 8.1.2.
We also give an application of Theorem 1.4 which concerns the algebraic Hochschild and
cyclic homology of C(X). We use the theorem in combination with the celebrated results of
Gerhard Hochschild, Bertram Kostant and Alex Rosenberg ([24]) and of Daniel Quillen and
Jean-Louis Loday ([30]) on the Hochschild and cyclic homology of smooth affine algebraic
varieties and the spectral sequence of Christian Kassel and Arne Sletsjøe ([27]), to prove
the following (see Theorem 8.2.6 for a full statement of our result and for the appropriate
definitions).
Theorem 1.6. Let k ⊂ C be a subfield. Write HH∗(/k), HC∗(/k), Ω∗
dR(/k)
for algebraic Hochschild and cyclic homology, algebraic Kahler differential forms, exterior
differentiation, and algebraic de Rham cohomology, all taken relative to the field k. Let X be
a compact Hausdorff space. Then
/k, d and H ∗
HCn(C(X)/k) = Ωn
dR (C(X)/k)
(n ∈ Z)
HHn(C(X)/k) = Ωn
C(X)/k ⊕ M2≤2p≤n
C(X)/k/dΩn−1
C(X)/k
H n−2p
We also apply Theorem 1.4 to the K-theoretic isomorphism conjecture of Farrell-Jones
and to the Beilison-Soul´e conjecture. The K-theoretic isomorphism conjecture for the group
Γ with coefficients in a ring R asserts that a certain assembly map
AΓ(R) : HΓ(EVC(Γ), K(R)) → K(R[Γ])
is an equivalence. Applying Theorem 1.4 to the cofiber of the assembly map, we obtain that if
AΓ(O(V )) is an equivalence for each smooth affine algebraic variety V over C, then AΓ(A) is
an equivalence for any commutative C ∗-algebra A. The (rational) Beılinson-Soul´e conjecture
concerns the decomposition of the rational K-theory of a commutative ring into the sum of
eigenspaces of the Adams operations
n (R)
The conjecture asserts that if R is regular noetherian, then
Kn(R) ⊗ Q = ⊕i≥0K (i)
K (i)
n (R) = 0 for n ≥ max{1, 2i}
It is well-known that the validity of the conjecture for R = C would imply that it also holds
for R = O(V ) whenever V is a smooth algebraic variety over C. We use Theorem 1.4 to
ALGEBRAIC GEOMETRY OF TOPOLOGICAL SPACES I
5
show that the validity of the conjecture for C would further imply that it holds for every
commutative C ∗-algebra.
Next we give an idea of the proofs of our main results, Theorem 1.1 and Theorem 1.3.
The basic idea of the proof of Theorem 1.1 goes back to Rosenberg's article [36] and
ultimately to the usual proof of the fact that locally trivial bundles over a contractible
compact Hausdorff space are trivial. It consists of translating the question of the freedom of
projective modules into a lifting problem:
(1.7)
X
e
Pn(C[M])
ι
GL(C[M])
π
GL(C[M ])
GL[1,n](C[M ])×GL[n+1,∞)(C[M ])
Here we think of a projective module of constant rank n over C(X) as a map e to the
set of all rank n idempotent matrices, which by Gubeladze's theorem is the same as the
set Pn(C[M]) of those matrices which are conjugate to the diagonal matrix 1n ⊕ 0∞. Thus
g 7→ g(1n ⊕ 0∞)g−1 defines a surjective map GL(C[M]) → Pn(C[M]) which identifies the
latter set with the quotient of GL(C[M]) by the stabilizer of 1n ⊕ 0∞, which is precisely
the subgroup GL[1,n](C[M]) × GL[n+1,∞)(C[M]). For this setup to make sense we need to
equip each set involved in (1.7) with a topology in such a way that all maps in the diagram
are continuous. Moreover for the lifting problem to have a solution, it will suffice to show
that ι is a homeomorphism and that π is a compact fibration, i.e. that it restricts to a
fibration over each compact subset of the base. In Section 4 we show that any countable
dimensional R-algebra R is equipped with a canonical compactly generated topology which
makes it into a topological algebra. A subset F ⊂ R is closed in this topology if and only if
F ∩ B ⊂ B is closed for every compact semi-algebraic subset B of every finite dimensional
subspace of R. In particular this applies to M∞R. The subset Pn(R) ⊂ M∞(R) carries
the induced topology, and the map e of (1.7) is continuous for this topology. The group
GL(R) also carries a topology, generated by the compact semi-algebraic subsets GLn(R)B.
Here B ⊂ MnR is any compact semi-algebraic subset as before, and GLn(R)B consists of
those n × n invertible matrices g such that both g and g−1 belong to B. The subgroup
GL[1,n](R) × GL[n+1,∞)(R) ⊂ GL(R) turns out to be closed, and we show in Section 4.2 --
with the aid of Gregory Brumfiel's theorem on quotients of semi-algebraic sets (see 3.1.4)
-- that for a topological group G of this kind, the quotient G/H by a closed subgroup H
is again compactly generated by the images of the compact semi-algebraic subsets defining
the topology of G, and these images are again compact, semi-algebraic subsets. Moreover
the restriction of the projection π : G → G/H over each compact semi-algebraic subset
S ⊂ G is semi-algebraic. We also show (Theorem 4.4.3) that π is a compact fibration.
This boils down to showing that if S ⊂ G is compact semi-algebraic, and T = f (S), then
we can find an open covering of T such that π has a section over each open set in the
covering. Next we observe that if U is any space, then the group map(U, G) acts on the set
map(U, G/H), and a map U → G/H lifts to U → G if and only if its class in the quotient
/
/
3
3
o
o
6
GUILLERMO CORTI NAS AND ANDREAS THOM
F (U) = map(U, G/H)/map(U, G) is the class of the trivial element: the constant map u 7→ H
(u ∈ U). For example the class of the composite of π with the inclusion S ⊂ G is the trivial
element of F (S). Hence if p = πS : S → T , then F (p) sends the inclusion T ⊂ G/H to
the trivial element of F (S). In Section 2 we introduce a notion of (weak) split exactness
for contravariant functors of topological spaces with values in pointed sets; for example the
functor F introduced above is split exact (Lemma 2.1.3). The key technical tool for proving
that π is a fibration is the following (see 3.2.1); its proof uses the good topological properties
of semi-algebraic sets and maps, especially Hardt's triviality theorem 3.1.10.
Theorem 1.8. Let T be a compact semi-algebraic subset of Rk. Let S be a semi-algebraic
set and let f : S → T be a proper continuous semi-algebraic surjection. Then, there exists a
semi-algebraic triangulation of T such that for every weakly split-exact contravariant functor
F from the category Pol of compact polyhedra to the category Set+ of pointed sets, and every
simplex ∆n in the triangulation, we have
ker(F (∆n) → F (f −1(∆n))) = ∗
Here ker is the kernel in the category of pointed sets, i.e. the fiber over the base point.
In our situation Theorem 1.8 applies to show that there is a triangulation of T ⊂ G/H
such that the projection π has a section over each simplex in the triangulation. A standard
argument now shows that T has an open covering (by open stars of a subdivision of the
previous triangulation) such that π has section over each open set in the covering. Thus
in diagram (1.7) we have that π is a compact fibration and that e is continuous. The map
ι : GL(R)/GL[1,n](R) ×GL[n+1,∞)(R) → Pn(R) is continuous for every countable dimensional
R-algebra R (see 5.1). We show in Proposition 5.2.2 that it is a homeomorphism whenever
the map
(1.9)
K0(ℓ∞(R)) → Yn≥1
K0(R)
is injective. Here ℓ∞(R) is the set of all sequences N 7→ R whose image is contained in one of
the compact semi-algebraic subsets B ⊂ R which define the topology of R; it is isomorphic
to ℓ∞(R) ⊗ R (Lemma 5.2.1). The algebraic compactness theorem (5.4.1) says that if R is a
countable dimensional C-algebra such that
(1.10)
K0(O(V )) ∼−→ K0(O(V ) ⊗C R)
(∀ smooth affine V ),
then (1.9) is injective. A theorem of Swan (see 6.1.2) implies that R = C[M] satisfies
(1.10). Thus the map ι of diagram (1.7) is a homeomorphism. This concludes the sketch of
the proof of Theorem 1.1.
The proof of the algebraic compactness theorem uses the following theorem (see 5.3.1).
Theorem 1.11. Let F and G be functors from commutative C-algebras to sets. Assume that
both F and G preserve filtering colimits. Let τ : F → G be a natural transformation. Assume
that τ (O(V )) is injective (resp. surjective) for each smooth affine algebraic variety V over
C. Then τ (ℓ∞(C)) is injective (resp. surjective).
ALGEBRAIC GEOMETRY OF TOPOLOGICAL SPACES I
7
The proof of 1.11 uses a technique which we call algebraic approximation, which we now
explain. Any commutative C-algebra is the colimit of its subalgebras of finite type, which
form a filtered system. If the algebra contains no nilpotent elements, then each of its subal-
gebras of finite type is of the form O(Y ) for affine variety Y , by which we mean a reduced
affine scheme of finite type over C. If C[f1, . . . , fn] ⊂ ℓ∞(C) is the subalgebra generated by
f1, . . . , fn, and C[f1, . . . , fn] ∼= O(Y ), then Y is isomorphic to a closed subvariety of Cn, and
f = (f1, . . . , fn) defines a map from N to a precompact subset of the space Yan of closed
points of Y equipped the topology inherited by the euclidean topology on Cn. The space Yan
is equipped with the structure of a (possibly singular) analytic variety, whence the subscript.
Summing up, we have
(1.12)
ℓ∞(C) = colimN→YanO(Y )
where the colimit runs over all affine varieties Y and all maps with precompact image. The
proof of Theorem 1.11 consists of showing that in (1.12) we can restrict to maps N → V
with V smooth. This uses Hironaka's desingularization [23] to lift a map f : N → V with
V affine and singular, to a map f ′ : N → V with V smooth and possibly non-affine, and
Jouanoulou's device [25] to further lift f ′ to a map f ′′ : N → W with W smooth and affine.
The idea of algebraic approximation appears in the work of Jonathan Rosenberg [36,
37, 38], and later in the article of Eric Friedlander and Mark E. Walker [15]. One source
of inspiration is the work of Andrei Suslin [41]. In [41], Suslin studies an inclusion of alge-
braically closed fields L ⊂ K and analyzes K successfully in terms of its finitely generated
L-subalgebras.
Next we sketch the proof of Theorem 1.3. The first step is to reduce to the polyhedral
case. For this we use Theorem 1.13 below, proved in 7.1.3. Its proof uses another algebraic
approximation argument together with a result of Allan Calder and Jerrold Siegel, which
says that the right Kan extension to Comp of a homotopy invariant functor defined on Pol
is homotopy invariant on Comp.
Theorem 1.13. Let F : Comm → Ab be a functor. Assume that F satisfies each of the
following conditions.
(i) F commutes with filtered colimits.
(iii) The functor Pol → Ab, D 7→ F (C(D)) is homotopy invariant.
Then the functor
is homotopy invariant.
Comp → Ab, X 7→ F (C(X))
Next, Proposition 2.1.5 says that we can restrict to showing that F vanishes on con-
tractible polyhedra. Since any contractible polyhedron is a retract of its cone, which is a
starlike polyhedron, we further reduce to showing that F vanishes on starlike polyhedra.
Using excision, we may restrict once more, to proving that F (∆p) = 0 for all p. For this we
follow the strategy used by Friedlander-Walker in [15]. To start, we use algebraic approxi-
mation again. We write
(1.14)
C(∆p) = colim∆p→YanO(Y )
8
GUILLERMO CORTI NAS AND ANDREAS THOM
where the colimit runs over all continuous maps from ∆p to affine algebraic varieties,
equipped with the euclidean topology. Since F is assumed to vanish on O(V ) for smooth
affine V , it would suffice to show that any map ∆p → Yan factors as ∆p → Van → Yan with
V smooth and affine. Actually using excision again we may restrict to showing this for each
simplex in a sufficiently fine triangulation of ∆p. As in the proof of Theorem 1.1, this is done
using Hironaka's desingularization, Jouanoulou's device and Theorem 1.8.
The rest of this paper is organized as follows. In Section 2 we give the appropriate
definitions and first properties of split exactness. We also recall some facts about algebraic
K-theory and cyclic homology, such as the key results of Andrei Suslin and Mariusz Wodz-
icki on excision for algebraic K-theory and algebraic cyclic homology. In Section 3, we recall
some facts from real algebraic geometry, and prove Theorem 1.8 (3.2.1). Large semi-algebraic
groups and their associated compactly generated topological groups are the subject of Sec-
tion 4. The main result of this section is the Fibration Theorem 4.4.3 which says that the
quotient map of such a group by a closed subgroup is a compact fibration. Section 5 is de-
voted to algebraic compactness, that is, to the problem of giving conditions on a countable
dimensional algebra R so that the map ι : GL(R)/GL[1,n](R) × GL[n+1,∞)(R) → Pn(R) be a
homeomorphism. The connection between this problem and the algebra ℓ∞(R) of bounded
sequences is established by Proposition 5.2.2. Theorem 1.11 is proved in 5.3.1. Theorem
5.4.1 establishes that the map (1.9) is injective whenever (1.10) holds. Section 6 contains
the proofs of Theorems 1.1 and 1.2 (6.1.3 and 6.1.5). We also show (Theorem 6.2.1) that if
M is a monoid as in Theorem 1.1 then any bundle of finitely generated free C[M] modules
over a compact Hausdorff space which is a direct summand of a trivial bundle is locally triv-
ial. Section 7 deals with homotopy invariance. Theorems 1.13, 1.3 and 1.4 (7.1.3, 7.3.1 and
7.4.1) are proved in this section, where also a second proof of Rosenberg's conjecture, using
a result of Friedlander and Walker, is given (see 7.2). Section 8 is devoted to applications
of the homotopy invariance and vanishing homology theorems, including Theorems 1.5 and
1.6 (8.1.1 and 8.2.6) and also to the applications to the conjectures of Farrell-Jones (8.3.2,
8.3.6) and of Beılinson-Soul´e (8.4.4).
2. Split-exactness, homology theories, and excision.
2.1. Set-valued split-exact functors on the category of compact Hausdorff spaces.
In this section we consider contravariant functors from the category of compact Hausdorff
topological spaces to the category Set+ of pointed sets. Recall that if T is a pointed set and
f : S → T is a map, then
ker f = {s ∈ S : f (s) = ∗}
We say that a functor F : Comp → Set+ is split exact if for each push-out square
(2.1.1)
X12
ι2
X2
ι1
X1
/ X
/
/
/
ALGEBRAIC GEOMETRY OF TOPOLOGICAL SPACES I
9
of topological spaces with ι1 or ι2 split injective, the map
F (X) → F (X1) ×F (X12) F (X2)
is a surjection with trivial kernel. We say that F is weakly split exact if the map above has
trivial kernel. In case the functor takes values in abelian groups, the notion of split exactness
above is equivalent to the usual one. For more details on split-exact functors taking values
in the category Ab of abelian groups, see Subsection 2.4.
In the next lemma and elsewhere, if X and Y are topological spaces, we write
for the set of continuous maps from X to Y .
map(X, Y ) = {f : X → Y continuous}
Lemma 2.1.2. Let (Y, y) be a pointed topological space. The contravariant functor
X 7→ map(X, Y )
from compact Hausdorff topological spaces to pointed sets is split-exact.
Proof. Note that map(X, Y ) is naturally pointed by the constant map taking the value y ∈ Y .
Let (2.1.1) be a push-out of compact Hausdorff topological spaces and assume that ι1 is a
split-injection. It is sufficient to show that the diagram
map(X12, Y )
map(X1, Y )
map(X2, Y )
map(X, Y )
is a pull-back. But this is immediate from the universal property of a push-out.
(cid:3)
Lemma 2.1.3. Let H ⊂ G be an inclusion of topological groups. Then the pointed set
map(X, G/H) carries a natural left action of the group map(X, G) and the functor
X 7→
map(X, G/H)
map(X, G)
is split exact.
Proof. We need to show that the map
(2.1.4)
map(X, G/H)
map(X, G)
→
map(X1, G/H)
map(X1, G)
× map(X12,G/H)
map(X12,G)
map(X2, G/H)
map(X2, G)
is a surjection with trivial kernel. Let f : X → G/H be such that its pull-backs fi : Xi →
G/H admit continuous lifts fi : Xi → G. Although the pull-backs of f1 and f2 to X12 might
not agree, we can fix this problem. Let σ be a continuous splitting of the inclusion X12 ֒→ X1.
Define a map
γ : X1 → H,
γ(x) = ( f1X12(σ(x))−1 · ( f2X12(σ(x))
o
o
O
O
O
O
o
o
10
GUILLERMO CORTI NAS AND ANDREAS THOM
Note f1 · γ is still a lift of f1 and agrees with f2 on X12; hence they define a map f : X → G
which lifts f . This proves that (2.1.4) has trivial kernel. Let now f1 : X1 → G/H and
f2 : X2 → G/H be such that there exists a function θ : X12 → G with θ(x) · f1(x) = f2(x)
for all x ∈ X12. Using the splitting σ of the inclusion X12 ֒→ X1 again, we can extend θ to
X1 to obtain f ′
1 is just another representative of the
class of f1. Since f ′
1 and f2 agree on X12, we conclude that there exists a continuous map
f : X → G/H, which pulls back to f ′
1 on X1 and to f2 on X2. This proves that (2.1.4) is
surjective.
(cid:3)
1(x) = θ(σ(x))f1(x), for x ∈ X1. Note f ′
Proposition 2.1.5. Let C be either the category Comp of compact Hausdorff spaces or the
full subcategory Pol of compact polyhedra. Let F : C → Ab be a split-exact functor. Assume
that F (X) = 0 for contractible X ∈ C. Then F is homotopy invariant.
Proof. We have to prove that if X ∈ C and 1X × 0 : X → X × [0, 1] is the inclusion, then
F (1X × 0) : F (X × [0, 1]) → F (X) is a bijection. Since it is obviously a split-surjection it
remains to show that this map is injective. Consider the pushout diagram
X
⋆
1X ×0
/ X × [0, 1]
/ cX
By split-exactness, the map F (cX) → P := F (∗) ×F (X) F (X × [0, 1]) is onto. Since we are
also assuming that F vanishes on contractible spaces, we further have F (∗) = F (cX) = 0,
whence P = ker(F (1X × 0)) = 0.
(cid:3)
2.2. Algebraic K-theory. In the previous subsection we considered contravariant functors
on spaces; now we turn our attention to the dual picture of covariant functors from categories
of algebras to pointed sets or abelian groups. The most important example for us is algebraic
K-theory. Before we go on, we want to quickly recall some definitions and results. Let R
be a unital ring. The abelian group K0(R) is defined to be the Grothendieck group of the
monoid of isomorphism classes of finitely generated projective R-modules with direct sum
as addition. We define
Kn(R) = πn(BGL(R)+), ⋆),
∀n ≥ 1,
where X 7→ X + denotes Quillen's plus-construction [34]. Bass' Nil K-groups of a ring are
defined as
(2.2.1)
NKn(R) = coker (Kn(R) → Kn(R[t])) .
The so-called fundamental theorem gives an isomorphism
(2.2.2)
Kn(R[t, t−1]) = Kn(R) ⊕ Kn−1(R) ⊕ NKn(R) ⊕ NKn(R),
∀n ≥ 1
which holds for all unital rings R. One can use this to define K-groups and Nil-groups in
negative degrees. Indeed, if one puts
Kn−1(R) = coker (cid:0)Kn(R[t]) ⊕ Kn(R[t−1]) → Kn(R[t, t−1])(cid:1) ,
/
/
ALGEBRAIC GEOMETRY OF TOPOLOGICAL SPACES I
11
negative K-groups can be defined inductively. There is a functorial spectrum K(R), such
that
(2.2.3)
Kn(R) = πnK(R)
(n ∈ Z).
This spectrum can be constructed in several equivalent ways (e.g. see [17], [32],[33, §5], [49,
§6], [50]). Functors from the category of algebras to spectra and their properties will be
studied in more detail in the next subsection.
A ring R is called Kn-regular if the map Kn(R) → Kn(R[t1, . . . , tm]) is an isomorphism
for all m; it is called K-regular if it is Kn-regular for all n. It is well-known that if R is
a regular noetherian ring then R is K-regular and KnR = 0 for n < 0. In particular this
applies when R is the coordinate ring of a smooth affine algebraic variety over a field. We
think of the Laurent polynomial ring R[t1, t−1
n ] as the group ring R[Zn] and use
the fact that if the natural map K0(R) → K0(R[Zn]) is an isomorphism for all n ∈ N, then
all negative algebraic K-groups and all (iterated ) nil-K-groups in negative degrees vanish.
This can be proved with an easy induction argument.
1 , . . . , tn, t−1
Remark 2.2.4. Iterating the nil-group construction, one obtains the following formula for the
K-theory of the polynomial ring in m-variables
m
p
(2.2.5)
Kn(R[t1, . . . , tm]) =
N pKn(R) ⊗
Mp=0
^ Zm
Here Vp is the exterior power and N pKn(R) denotes the iterated nil-group defined using the
analogue of formula (2.2.1). Thus a ring R is Kn-regular if and only if N pKn(R) = 0 for all
p ≥ 0. In [1] Hyman Bass raised the question of whether the condition that NKn(R) = 0
is already sufficient for Kn-regularity. This question was settled in the negative in [9, Thm.
4.1], where an example of a commutative algebra R of finite type over Q was given such that
NK0(R) = 0 but N 2K0(R) 6= 0. On the other hand it was proved [8, Cor. 6.7] (see also [21])
that if R is of finite type over a large field such as R or C, then NKn(R) = 0 does imply
that R is Kn-regular. This is already sufficient for our purposes, since the rings this paper
is concerned with are algebras over R. For completeness let us remark further that if R is
any ring such that NKn(R) = 0 for all n then R is K-regular, i.e. Kn-regular for all n ∈ Z.
As observed by Jim Davis in [11, Cor. 3] this follows from Frank Quinn's theorem that the
Farrell-Jones conjecture is valid for the group Zn (see also [8, 4.2]).
2.3. Homology theories and excision. We consider functors and homology theories of
associative, not necessarily unital algebras over a fixed field k of characteristic zero. In
what follows, C will denote either the category Ass/k of associative k-algebras or the full
subcategory Comm/k of commutative algebras. A homology theory on C is a functor E :
C → Spt to the category of spectra which preserves finite products up to homotopy. That
is, E(Qi∈I Ai) → Qi∈I E(Ai) is a weak equivalence for finite I. If A ∈ C and n ∈ Z, we write
En(A) = πnE(A) for the n-th stable homotopy group. Let E be a homology theory and let
(2.3.1)
0 → A → B → C → 0
12
GUILLERMO CORTI NAS AND ANDREAS THOM
be an exact sequence (or extension) in C. We say that E satisfies excision for (2.3.1), if
E(A) → E(B) → E(C) is a homotopy fibration. The algebra A is E-excisive if E satisfies
excision on any extension (2.3.1) with kernel A. If A ⊂ C is a subcategory, and E satisfies
excision for every sequence (2.3.1) in A, then we say that E satisfies excision on A.
Remark 2.3.2. If we have a functor E which is only defined on the subcategory C1 ⊂ C
of unital algebras and unital homomorphisms, and which preserves finite products up to
homotopy, then we can extend it to all of C by setting
E(A) = hofiber(E(A+
k ) → E(k))
Here A+
k denotes the unitalization of A as a k-algebra. The restriction of the new functor
E to unital algebras is not the same as the old one, but it is homotopy equivalent to it.
∼= A ⊕ k as k-algebras. Since E preserves finite products,
Indeed, for A unital, we have A+
k
this implies the claim. In this article, whenever we encounter a homology theory defined
only on unital algebras, we shall implicitly consider it extended to non-unital algebras by
the procedure just explained. Similarly, if F : C1 → Ab is a functor to abelian groups which
preserves finite products, it extends to all of C by
F (A) = ker(F (A+
k ) → F (k))
In particular this applies when F = En is the homology functor associated to a homology
theory as above.
The main examples of homology theories we are interested in are K-theory, Hochschild
homology and the various variants of cyclic homology. A milestone in understanding excision
in K-theory is the following result of Andrei Suslin and Mariusz Wodzicki, see [43, 44].
Theorem 2.3.3 (Suslin-Wodzicki). A Q-algebra R is K-excisive if and only if for the Q-
algebra unitalization R+
Q = R ⊕ Q we have
R+
Q
n (Q, R) = 0
Tor
for all n ≥ 0.
For example it was shown in [44, Thm. C] that any ring satisfying a certain "triple
factorization property" is K-excisive; since any C ∗-algebra has this property, ([44, Prop.
10.2]) we have
Theorem 2.3.4 (Suslin-Wodzicki). C ∗-algebras are K-excisive.
Excision for Hochschild and cyclic homology of k-algebras, denoted respectively HH(/k)
and HC(/k), has been studied in detail by Wodzicki in [52]; as a particular case of his results,
we cite the following:
Theorem 2.3.5 (Wodzicki). The following are equivalent for a k-algebra A.
(1) A is HH(/k)-excisive.
(2) A is HC(/k)-excisive.
(3) TorA+
∗ (k, A) = 0.
k
ALGEBRAIC GEOMETRY OF TOPOLOGICAL SPACES I
13
Note that it follows from (2.3.3) and (2.3.5) that a k-algebra A is K-excisive if and only
if it is HH(/Q)-excisive.
Remark 2.3.6. Wodzicki has proved (see [53, Theorems 1 and 4]) that if A is a C ∗-algebra
then A satisfies the conditions of Theorem 2.3.5 for any subfield k ⊂ C.
2.4. Milnor squares and excision. We now record some facts about Milnor squares of
k-algebras and excision.
Definition 2.4.1. A square of k-algebras
(2.4.2)
A
C
B
f
g
/ D
is said to be a Milnor square if it is a pull-back square and either f or g is surjective. It is
said to be split if either f or g has a section.
Let F be a functor from C to abelian groups and let
(2.4.3)
0
/ A
/ B
/ C
/ 0
be a split extension in C. We say that F is split exact on (2.4.3) if
0
/ F (A)
/ F (B)
/ F (C)
/ 0
is (split) exact. If A ⊂ C is a subcategory, and F is split exact on all split exact sequences
contained in A, then we say that F is split exact on A.
Lemma 2.4.4. Let E : C → Spt be a homology theory and (2.4.2) a Milnor square. Assume
that ker(f ) is E-excisive. Then E maps (2.4.2) to a homotopy cartesian square.
Lemma 2.4.5. Let F : C → Ab be a functor, A ⊂ C a subcategory closed under kernels, and
(2.4.2) a split Milnor square in A. Assume that F is split exact on A. Then the sequence
0 → F (A) → F (B) ⊕ F (C) → F (D) → 0
is split exact.
3. Real algebraic geometry and split exact functors.
In this section, we recall several results from real algebraic geometry and prove a the-
orem on the behavior of weakly split exact functors with respect to proper semi-algebraic
surjections (see Theorem 3.2.1). Recall that a semi-algebraic set is a priori a subset of Rn
which is described as the solution set of a finite number of polynomial equalities and inequal-
ities. A map between semi-algebraic sets is semi-algebraic if its graph is a semi-algebraic set.
For general background on semi-algebraic sets, consult [2].
/
/
/
/
/
/
u
u
/
/
/
/
/
14
GUILLERMO CORTI NAS AND ANDREAS THOM
3.1. General results about semi-algebraic sets. Let us start with recalling the following
two propositions.
Proposition 3.1.1 (see Proposition 3.1 in [2]). The closure of a semi-algebraic set is semi-
algebraic.
Proposition 3.1.2 (see Proposition 2.83 in [2]). Let S and T be semi-algebraic sets, S′ ⊂ S
and T ′ ⊂ T semi-algebraic subsets and f : S → T a semi-algebraic map. Then f (S′) and
f −1(T ′) are semi-algebraic.
Note that a semi-algebraic map does not need to be continuous. Moreover, within the
class of continuous maps, there are surjective maps f : S → T , for which the quotient
topology induced by S does not agree with the topology on T . An easy example is the
projection map from {(0, 0)} ∪ {(t, t−1) t > 0} to [0, ∞). This motivates the following
definition.
Definition 3.1.3. Let S, T be semi-algebraic sets. A continuous semi-algebraic surjection
f : S → T is said to be topological, if for every semi-algebraic map g : T → Q the composition
g ◦ f is continuous if and only if g is continuous.
Gregory Brumfiel proved the following result, which says that (under certain conditions)
semi-algebraic equivalence relations lead to good quotients.
Theorem 3.1.4 (see Theorem 1.4 in [5]). Let S be a semi-algebraic set and let R ⊂ S × S
be a closed semi-algebraic equivalence relation. If π1 : R → S is proper, then there exists a
semi-algebraic set T and a topological semi-algebraic surjection f : S → T such that
R = {(s1, s2) ∈ S × S f (s1) = f (s2)}.
Remark 3.1.5. Note that the properness assumption in the previous theorem is automatically
fulfilled if S is compact. This is the case we are interested in.
Corollary 3.1.6. Let S, S′ and T be compact semi-algebraic sets and
f : T → S and f ′ : T → S′
be continuous semi-algebraic maps. Then, the topological push-out S ∪T S′ carries a canonical
semi-algebraic structure such that the natural maps σ : S → S ∪T S′ and σ′ : S′ → S ∪T S′
are semi-algebraic.
For semi-algebraic sets, there is an intrinsic notion of connectedness, which is given by
the following definition.
Definition 3.1.7. A semi-algebraic set S ⊂ Rk is said to be semi-algebraically connected if
it is not a non-trivial union of semi-algebraic subsets which are both open and closed in S.
One of the first results on connectedness of semi-algebraic sets is the following theorem.
Theorem 3.1.8 (see Theorem 5.20 in [2]). Every semi-algebraic set S is the disjoint union
of a finite number of semi-algebraically connected semi-algebraic sets which are both open
and closed in S.
ALGEBRAIC GEOMETRY OF TOPOLOGICAL SPACES I
15
Next we come to aspects of semi-algebraic sets and continuous semi-algebraic maps
which differ drastically from the expected results for general continuous maps. In fact, there
is a far-reaching generalization of Ehresmann's theorem about local triviality of submersions.
Let us consider the following definition.
Definition 3.1.9. Let S and T be two semi-algebraic sets and f : S → T be a continuous
semi-algebraic function. We say that f is a semi-algebraically trivial fibration if there exist
a semi-algebraic set F and a semi-algebraic homeomorphism
such that f ◦ θ is the projection onto T .
θ : T × F → S
A seminal theorem is Hardt's triviality result, which says that away from a subset of T
of smaller dimension, every map f : S → T looks like a semi-algebraically trivial fibration.
Theorem 3.1.10 ([2] or §4 in [22]). Let S and T be two semi-algebraic sets and f : S → T a
continuous semi-algebraic function. Then there exists a closed semi-algebraic subset V ⊂ T
with dim V < dim T , such that f is a semi-algebraically trivial fibration over every semi-
algebraic connected component of T \ V .
We shall also need the following result about semi-algebraic triangulations.
Theorem 3.1.11 (Thm. 5.41 in [2]). Let S ⊂ Rk be a compact semi-algebraic set, and let
S1, . . . , Sq be semi-algebraic subsets. There exists a simplicial complex K and a semi-algebraic
homeomorphism h : K → S such that each Sj is the union of images of open simplices of
K.
Remark 3.1.12. In the preceding theorem, the case where the subsets Sj are closed is of special
interest. Indeed, if the subsets Sj are closed, the theorem implies that the triangulation of
S induces triangulations of Sj for each j ∈ {1, . . . , q}.
The following proposition is an application of Theorems 3.1.10 and 3.1.11
Proposition 3.1.13. Let T ⊂ Rm be a compact semi-algebraic subset, S a semi-algebraic
set and f : S → T a continuous semi-algebraic map. Then there exist a semi-algebraic
triangulation of T and a finite sequence of closed subcomplexes
∅ = Vr+1 ⊂ Vr ⊂ Vr−1 ⊂ · · · ⊂ V1 ⊂ V0 = T
such that the following conditions are satisfied
(i) For each k = 0, . . . , r, we have dim Vk+1 < dim Vk, and the map
f f −1(Vk\Vk+1) : f −1(Vk \ Vk+1) → Vk \ Vk+1
is a semi-algebraically trivial fibration over every semi-algebraic connected compo-
nent.
(ii) Each simplex in the triangulation lies in some Vk and has at most one face of codi-
mension one which intersects Vk+1.
16
GUILLERMO CORTI NAS AND ANDREAS THOM
Proof. We set n = dim T . By Theorem 3.1.10, there exists a closed semi-algebraic subset
V1 ⊂ T with dim V1 < n, such that f is a semi-algebraic trivial fibration over every semi-
algebraic connected component of T \ V1. Consider now f f −1(V1) : f −1(V1) → V1 and proceed
as before to find V2 ⊂ V1. By induction, we find a chain
∅ ⊂ Vr ⊂ Vr−1 ⊂ · · · ⊂ V1 ⊂ V0 = T
such that Vk ⊂ Vk−1 is a closed semi-algebraic subset and
f f −1(Vk−1\Vk) : f −1(Vk−1 \ Vk) → Vk−1 \ Vk
is a semi-algebraically trivial fibration over every semi-algebraic connected component, for
all k ∈ {1, . . . , r}. Using Theorem 3.1.11, we may now choose a semi-algebraic triangulation
of T such that the subsets Vk are sub-complexes. Taking a barycentric subdivision, each
simplex lies in Vk for some k ∈ {0, . . . , r} and has at most one face of codimension one which
intersects the set Vk+1.
(cid:3)
3.2. The theorem on split exact functors and proper maps. The following result is
our main technical result. It is key to the proofs of Theorems 4.4.3 and Theorem 7.3.1.
Theorem 3.2.1. Let T be a compact semi-algebraic subset of Rk. Let S be a semi-algebraic
set and let f : S → T be a proper continuous semi-algebraic surjection. Then, there exists a
semi-algebraic triangulation of T such that for every weakly split-exact contravariant functor
F : Pol → Set+ and every simplex ∆n in the triangulation, we have
ker(F (∆n) → F (f −1(∆n))) = ∗
Proof. Choose a triangulation of T and a sequence of subcomplexes Vk ⊂ T as in Proposition
3.1.13. We shall show that ker(F (∆n) → F (f −1(∆n))) = ∗ for each simplex in the chosen
triangulation. The proof is by induction on the dimension of the simplex. The statement is
clear for zero-dimensional simplices since f is surjective. Let ∆n be an n-dimensional simplex
in the triangulation. By assumption, f is a semi-algebraically trivial fibration over ∆n \ ∆n−1
for some face ∆n−1 ⊂ ∆n. Hence, there exists a semi-algebraic set K and a semi-algebraic
homeomorphism
θ : (∆n \ ∆n−1) × K → f −1(∆n \ ∆n−1)
over ∆n \∆n−1. Consider the inclusion f −1(∆n−1) ⊂ f −1(∆n). Since f −1(∆n−1) is an absolute
neighborhood retract, there exists a compact neighborhood N of f −1(∆n−1) in f −1(∆n) which
retracts onto f −1(∆n−1). We claim that the set f (N) contains some standard neighborhood
A of ∆n−1. Indeed, assume that f (N) does not contain standard neighborhoods. Then there
exists a sequence in the complement of f (N) converging to ∆n−1. Lifting this sequence
one can choose a convergent sequence in the complement of N converging to f −1(∆n−1).
This contradicts the fact that N is a neighborhood and hence, there exists a standard
compact neighborhood A of ∆n−1 in ∆n such that f −1(A) ⊂ N. Since (∆n \ ∆n−1) × K ∼=
f −1(∆n \ ∆n−1), any retraction of ∆n onto A yields a retraction of f −1(∆n) onto f −1(A). We
have that f −1(A) ⊂ N, and thus we can conclude that f −1(∆n−1) is a retract of f −1(∆n).
ALGEBRAIC GEOMETRY OF TOPOLOGICAL SPACES I
17
By Corollary 3.1.6, the topological push-out
f −1(∆n−1)
f −1(∆n)
∆n−1
/ Z
carries a semi-algebraic structure. Moreover, by weak split-exactness
(3.2.2)
ker(F (Z) → F (∆n−1) ×F (f −1(∆n−1)) F (f −1(∆n))) = ∗.
Note that f −1(∆n \∆n−1) ⊂ Z by definition of Z. We claim that the natural map σ : Z → ∆n
is a semi-algebraic split surjection. Indeed, identify f −1(∆n \ ∆n−1) = (∆n \ ∆n−1) × K, pick
k ∈ K, and consider
Gk = {(d, (d, k)) ∈ ∆n × Z d ∈ ∆n \ ∆n−1}.
Then Gk is semi-algebraic by Proposition 3.1.1, and therefore defines a continuous semi-
algebraic map ρk : ∆n → Z which splits σ. Thus F (σ) : F (∆n) → F (Z) is injective, whence
(3.2.3)
ker(F (∆n) → F (∆n−1) ×F (f −1(∆n−1)) F (f −1(∆n))) = ∗,
by (3.2.2). But the kernel of F (∆n−1) → F (f −1(∆n−1)) is trivial by induction, so the same
must be true of F (∆n) → F (f −1(∆n)), by (3.2.3).
(cid:3)
4. Large semi-algebraic groups and the compact fibration theorem.
4.1. Large semi-algebraic structures. Recall a partially ordered set Λ is filtered if for
any two λ, γ ∈ Λ there exists a µ such that µ ≥ λ and µ ≥ γ. We shall say that a filtered
poset Λ is archimedian if there exists a monotone map φ : N → Λ from the ordered set of
natural numbers which is cofinal, i.e. it is such that for every λ ∈ Λ there exists an n ∈ N
such that φ(n) ≥ λ. If X is a set, we write P (X) for the partially ordered set of all subsets
of X, ordered by inclusion.
A large semi-algebraic structure on a set X consists of
(i) An archimedian filtered partially ordered set Λ.
(ii) A monotone map X : Λ → P (X), λ 7→ Xλ, such that X = Sλ Xλ.
(iii) A compact semi-algebraic structure on each Xλ such that if λ ≤ µ then the inclusion
Xλ ⊂ Xµ is semi-algebraic and continuous.
We think of large a semi-algebraic structure on X as an exhaustive filtration {Xλ} by compact
semi-algebraic sets. A structure {Xγ : γ ∈ Γ} is finer than a structure {Xλ : λ ∈ Λ} if for
every γ ∈ Γ there exists a λ ∈ Λ so that Xγ ⊂ Xλ and the inclusion is continuous and
semi-algebraic. Two structures are equivalent if each of them is finer than the other. A large
semi-algebraic set is a set X together with an equivalence class of semi-algebraic structures
on X. If X is a large semi-algebraic set, then any large semi-algebraic structure {Xλ} in
the equivalence class defining X is called a defining structure for X. If X = (X, Λ) and
Y = (Y, Γ) are large semi-algebraic sets, then a set map f : X → Y is called a morphism
if for every λ ∈ Λ there exists a γ ∈ Γ such that f (Xλ) ⊂ Yγ, and such that the induced
/
/
/
18
GUILLERMO CORTI NAS AND ANDREAS THOM
map f : Xλ → Yγ is semi-algebraic and continuous. We write V∞ for the category of large
semi-algebraic sets.
Remark 4.1.1. If f : X → Y is a morphism of large semi-algebraic sets, then we may
choose structures {Xn} and {Yn} indexed by N, such that f strictly preserves filtrations, i.e.
f (Xn) ⊂ Yn for all n. However, if X = Y , then there may not exist a structure {Xλ} such
that f (Xλ) ⊂ Xλ.
Remark 4.1.2. Consider the category Vs,b of compact semi-algebraic sets with continuous
semi-algebraic mappings and its ind-category ind-Vs,b. The objects are functors
T : (XT , ≤) → Vs,b
where (XT , ≤) is a filtered partially ordered set. We set
hom(T, S) = lim
d∈XT
colim
e∈XS
homVs,b(T (d), S(e)).
The category of large semi-algebraic sets is equivalent to the subcategory of those ind-
objects whose structure maps are all injective and whose index posets are archimedean. In
particular, a filtering colimit of an archimedian system of injective homomorphisms of large
semi-algebraic sets is again a large semi-algebraic set.
If X and Y are large semi-algebraic sets with structures {Xλ : λ ∈ Λ} and {Yγ : γ ∈ Γ}
then the cartesian product X ×Y is again a large semi-algebraic set, with structure {Xλ×Yγ :
(λ, γ) ∈ Λ × Γ}. A large semi-algebraic group is a group object G in V∞. Thus G is a group
which is a large semi-algebraic set and each of the maps defining the multiplication, unit, and
inverse, are homomorphisms in V∞. We shall additionally assume that G admits a structure
{Gλ} such that G−1
λ ⊂ Gλ. This hypothesis, although not strictly necessary, is verified by
all the examples we shall consider, and makes proofs technically simpler. We shall also need
the notions of large semi-algebraic vectorspace and of large semi-algebraic ring, which are
defined similarly.
Example 4.1.3. Any semi-algebraic set S can be considered as a large semi-algebraic set, with
the structure defined by its compact semi-algebraic subsets, which is equivalent to the struc-
ture defined by any exhaustive filtration of S by compact semi-algebraic subsets. In particular
this applies to any finite dimensional real vectorspace V ; moreover the vectorspace opera-
tions are semi-algebraic and continuous, so that V is a (large) semi-algebraic vectorspace.
Any linear map between finite dimensional vectorspaces is semi-algebraic and continuous,
whence it is a homomorphism of semi-algebraic vectorspaces. Moreover, the same is true of
any multilinear map f : V1 × · · · × Vn → Vn+1 between finite dimensional vectorspaces.
Definition 4.1.4. Let V be a real vectorspace of countable dimension. The fine large semi-
algebraic structure F (V ) is that given by all the compact semi-algebraic subsets of all the
finite dimensional subspaces of V .
Remark 4.1.5. The fine large semi-algebraic structure is reminiscent of the fine locally convex
topology which makes every complex algebra of countable dimension into a locally convex
algebra. For details, see [4], chap. II, §2, Exercise 5.
ALGEBRAIC GEOMETRY OF TOPOLOGICAL SPACES I
19
Lemma 4.1.6. Let n ≥ 1, V1, . . . , Vn+1 be countable dimensional R-vectorspaces, and f :
V1 × · · · × Vn → Vn+1 a multilinear map. Equip each Vi with the fine large semi-algebraic
structure and V1 × · · · × Vn with the product large semi-algebraic structure. Then f is a
morphism of large semi-algebraic sets.
Proof. In view of the definition of the fine large semi-algebraic structure, the general case is
immediate from the finite dimensional case.
(cid:3)
Proposition 4.1.7. Let A be a countable dimensional R-algebra, equipped with the fine large
semi-algebraic structure. Then:
(i) A is a large semi-algebraic ring.
(ii) Assume A is unital. Then the group GLn(A) together with the structure
F (GLn(A)) = {{g ∈ GLn(A) : g, g−1 ∈ F } : F ∈ F (Mn(A))}
is a large semi-algebraic group, and if A → B is an algebra homomorphism, then
the induced group homomorphism GLn(A) → GLn(B) is a homomorphism of large
semi-algebraic sets.
Proof. Part (i) is immediate from Lemma 4.1.6. If F ∈ F (Mn(A)), write
(4.1.8)
GLn(A)F = {g ∈ GLn(A) : g, g−1 ∈ F }
We will show that GLn(A)F is a compact semi-algebraic set. Write m, πi : Mn(A)×Mn(A) →
Mn(A) for the multiplication and projection maps i = 1, 2, and τ : Mn(A) × Mn(A) →
Mn(A) × Mn(A) for the permutation of factors. If A is unital, then
GLn(A)F = π1(m−1
F ×F (1) ∩ τ m−1
F ×F (1))
which is compact semi-algebraic, by Proposition 3.1.2.
(cid:3)
We will also study the large semi-algebraic group GLS(R) for a subset S ⊂ Z. This is
understood to be the group of matrices g indexed by Z, where gi,j = δi,j if either i or j is
not in S.
Corollary 4.1.9. If A is unital and S ⊂ Z, then GLS(A) carries a natural large semi-
algebraic group structure, namely that of the colimit GLS(A) = ST GLT (A) where T runs
among the finite subsets of S.
Remark 4.1.10. If A is any, not necessarily unital ring, and a, b ∈ MnA, then a⋆b = a+b+ab
is an associative operation, with neutral element the zero matrix; the group GLn(A) is
defined as the set of all matrices which are invertible under ⋆. If A happens to be unital,
the resulting group is isomorphic to that of invertible matrices via g 7→ g + 1. If A is any
countable dimensional R-algebra, then part ii) of Proposition 4.1.7 still holds if we replace
g−1 by the inverse of g under the operation ⋆ in the definition of GLn(A)F . Corollary 4.1.9
also remains valid in the non-unital case, and the proof is the same.
20
GUILLERMO CORTI NAS AND ANDREAS THOM
4.2. Construction of quotients of large semi-algebraic groups.
Lemma 4.2.1. Let X ⊂ Y be an inclusion of large semi-algebraic sets. Then the following
are equivalent.
(i) There exists a defining structure {Yλ} of Y such that {Yλ ∩ X} is a defining structure
for X.
(ii) For every defining structure {Yλ} of Y , {Yλ ∩ X} is a defining structure for X.
Definition 4.2.2. Let X ⊂ Y be an inclusion of large semi-algebraic sets. We say that X
is compatible with Y if the equivalent conditions of Lemma 4.2.1 are satisfied.
Proposition 4.2.3. Let H ⊂ G be a inclusion of large semi-algebraic groups, {Gλ} a defining
structure for G and π : G → G/H the projection. Assume that the inclusion is compatible
in the sense of Definition 4.2.2. Then (G/H)λ = π(Gλ) is a large semi-algebraic structure,
and the resulting large semi-algebraic set G/H is the categorical quotient in V∞.
Proof. The map Gλ → (G/H)λ is the set-theoretical quotient modulo the relation Gλ ×Gλ ⊃
Rλ = {(g1, g2) : g−1
1 g2 ∈ H}. Let µ be such that the product map m sends Gλ × Gλ into Gµ;
write inv : G → G for the map inv(g) = g−1. Then
Rλ = (m ◦ (inv, id))−1(H ∩ Gµ)
Because H ⊂ G is compatible, H ∩Gµ ⊂ Gµ is closed and semi-algebraic, whence the same is
true of Rλ. By Theorem 3.1.4, (G/H)λ is semi-algebraic and Gλ → (G/H)λ is semi-algebraic
and continuous. It follows that the (G/H)λ define a large semi-algebraic structure on G/H
and that the projection is a morphism in V∞. The universal property of the quotient is
straightforward.
(cid:3)
Examples 4.2.4. Let R be a unital, countable dimensional R-algebra. The set
Pn(R) = {g(1n ⊕ 0∞)g−1 : g ∈ GL(R)}
of all finite idempotent matrices which are conjugate to the n × n identity matrix can be
written as a quotient of a compatible inclusion of large semi-algebraic groups. We have
Pn(R) =
GL(R)[1,∞)
GL[1,n](R) × GL[n+1,∞)(R)
.
On the other hand, since Pn(R) ⊂ M∞R, it also carries another large semi-algebraic struc-
ture, induced by the fine structure on M∞R. Since g 7→ g(1n ⊕ 0∞)g−1 is semi-algebraic, the
universal property of the quotient 4.2.3 implies that the quotient structure is finer than the
subspace structure. Similarly, we may write the set of those idempotent matrices which are
stably conjugate to 1n ⊕ 0∞ as
n (R) ={g(1∞ ⊕ 1n ⊕ 0∞)g−1 : g ∈ GLZ(R)}
P ∞
=
GL(−∞,+∞)(R)
GL(−∞,n](R) × GL[n+1,∞)(R)
ALGEBRAIC GEOMETRY OF TOPOLOGICAL SPACES I
21
Again this carries two large semi-algebraic structures; the quotient structure, and that coming
from the inclusion
into the 2 × 2 matrices of the unitalization of M∞(R).
P ∞
n (R) ⊂ M2(M∞(R)+)
4.3. Large semi-algebraic sets as compactly generated spaces. A topological space
X is said to be compactly generated if it carries the inductive topology with respect to its
compact subsets, i.e. a map f : X → Y is continuous if and only if its restriction to any
compact subset of X is continuous. In other words, a subset U ⊂ X is open (resp. closed)
if and only if U ∩ K is open (resp. closed) in K for every K ⊂ X compact. Observe that
any filtering colimit of compact spaces is compactly generated. In particular, if X is a large
semi-algebraic set, with defining structure {Xλ}, then X = Sλ Xλ equipped with the col-
imit topology is compactly generated, and this topology depends only on the equivalence
class of the structure {Xλ}. In what follows, whenever we regard a large semi-algebraic set
as a topological space, we will implicitly assume it equipped with the compactly generated
topology just defined. Note further that any morphism of large semi-algebraic sets is con-
tinuous for the compactly generated topology. Lemma 4.3.1 characterizes those inclusions of
large semi-algebraic sets which are closed subspaces, and Lemma 4.3.2 concerns quotients
of large semi-algebraic groups with the compactly generated topology. Both lemmas are
straightforward.
Lemma 4.3.1. An inclusion X ⊂ Y of large semi-algebraic sets is compatible if and only if
X is a closed subspace of Y with respect to the compactly generated topologies.
Lemma 4.3.2. Let H ⊂ G be a compatible inclusion of large semi-algebraic groups. View
G and H as topological groups equipped with the compactly generated topologies. Then the
compactly generated topology associated to the quotient large semi-algebraic set G/H is the
quotient topology.
We shall be concerned with large semi-algebraic groups which are Hausdorff for the
compactly generated topology. The main examples are countable dimensional R-vector spaces
and groups such as GLS(A), for some subset S ⊂ Z and some countably dimensional unital
R-algebra A.
Lemma 4.3.3. Let F = R or C, V a countable dimensional F-vectorspace, A a count-
able dimensional F-algebra, S ⊂ Z a subset, X a compact Hausdorff topological space, and
C(X) = map(X, F). Equip V , A, and GLS(A) with the compactly generated topologies. Then
the natural homomorphisms
C(X) ⊗F V → map(X, V ) and GLS(map(X, A)) → map(X, GLS(A))
are bijective.
Proof. It is clear that the both homomorphisms are injective. The image of the first one
consists of those continuous maps whose image is contained in a finitely generated subspace
of V . But since X is compact and V has the inductive topology of all closed balls in a
finitely generated subspace, every continuous map is of that form. Next note that GLS(A) =
22
GUILLERMO CORTI NAS AND ANDREAS THOM
SS ′,F GLS ′(A)F where the union runs among the finite subsets of S and the compact semi-
algebraic sets of the form F = MS ′(B) where B is a compact semi-algebraic subset of some
finitely generated subspace of A. Hence any continuous map f : X → GLS(A) sends X into
some MS ′B, and thus each of the entries f (x)i,j i, j ∈ S′ is a continuous function. Thus f (x)
comes from an element of GLS(map(X, A)).
(cid:3)
4.4. The compact fibration theorem for quotients of large semi-algebraic groups.
Recall that a continuous map f : X → Y of topological spaces is said to have the homotopy
lifting property with respect to a space Z if for any solid arrow diagram
Z
id×0
Z × [0, 1]
/ X
f
Y
of continuous maps a continuous dotted arrow exists and makes both triangles commute.
The map f is a (Hurewicz) fibration if it has the HLP with respect to any space Z, and is
a Serre fibration if it has the HLP with respect to all disks Dn n ≥ 0.
Definition 4.4.1. Let X, Y be topological space and f : X → Y a continuous map. We say
that f is a compact fibration if for every compact subspace K, the map f −1(K) → K is a
fibration.
Remark 4.4.2. Note that every compact fibration is a Serre fibration. Also, since every map
p : E → B with compact B which is a locally trivial bundle is a fibration, any map f : X → Y
such that the restriction f −1(K) → K to any compact subspace K ⊂ Y is a locally trivial
bundle, is a compact fibration. The notion of compact fibration comes up naturally in the
study of homogenous spaces of infinite dimensional topological groups.
Theorem 4.4.3. Let H ⊂ G be a compatible inclusion of large semi-algebraic groups. Then
the quotient map π : G → G/H is a compact fibration.
Proof. Choose defining structures {Hp} and {Gp} indexed over N and such that Hp = Gp∩H;
let {(G/H)p} be as in Proposition 4.2.3. Since any compact subspace K ⊂ G/H is contained
in some (G/H)p, it suffices to show that the projection πp = ππ−1((G/H)p) : π−1((G/H)p) →
(G/H)p is a locally trivial bundle. By a well-known argument (see e.g. [47, Thm. 4.13]) if
the quotient map of a group by a closed subgroup admits local sections then it is a locally
trivial bundle; the same argument applies in our case to show that if πp admits local sections
then it is a locally trivial bundle. Consider the functor
F : Comp → Set+, F (X) =
map(X, G/H)
map(X, G)
By Lemma 2.1.3, F is split exact. By Theorem 3.2.1 applied to F and to the proper semi-
algebraic surjection Gp → (G/H)p, there is a triangulation of (G/H)p such that
(4.4.4)
ker(F (∆n) → F (π−1
p (∆n))) = ∗
/
/
/
:
:
ALGEBRAIC GEOMETRY OF TOPOLOGICAL SPACES I
23
for each simplex ∆n in the triangulation. The diagram
π−1(∆n) ∩ Gp
Gp
πp
G
π
∆n
/ (G/H)p
/ G/H
shows that the class of the inclusion ∆n ⊂ G/H is an element of that kernel, and therefore it
can be lifted to a continuous map ∆n → G, by (4.4.4). Thus πp admits a continuous section
over every simplex in the triangulation. Therefore, using split-exactness of F , it admits
a continuous section over each of the open stars sto(x) of the vertices of the barycentric
subdivision. As the open stars of vertices form an open covering of (G/H)p, we conclude
that πp admits local sections. This finishes the proof.
(cid:3)
5. Algebraic compactness, bounded sequences and algebraic approximation.
5.1. Algebraic compactness. Let R be a countable dimensional unital R-algebra, equipped
with the fine large semi-algebraic structure. Consider the large semi-algebraic sets
(5.1.1)
(5.1.2)
M∞R ⊃ Pn(R) =
M2((M∞R)+) ⊃ P ∞
n (R) =
GL(R)[1,∞)
GL[1,n](R) × GL[n+1,∞)(R)
GL(−∞,+∞)(R)
GL(−∞,n](R) × GL[n+1,∞)(R)
introduced in 4.2.4. Recall that each of Pn(R), Pn(R)∞ carries two large semi-algebra struc-
tures: the homogenous ones, coming from the quotients, and those induced by the inclusions
above. As the homogeneous structures are finer than the induced ones, the same is true of the
corresponding compactly generated topologies; they agree if and only if the corresponding
large semi-algebraic structures are equivalent, or, what is the same, if every subset which is
compact in the homogeneous topology is also compact in the induced one. This motivates
the following definition.
Definition 5.1.3. Let R be a countable dimensional unital R-algebra, equipped with the
fine large semi-algebraic structure. We say that R has the algebraic compactness property if
for every n ≥ 1 the homogeneous and the induced large semi-algebraic structure of P ∞
n (R)
agree.
We show in Proposition 5.1.5 below that if R satisfies algebraic compactness, then the
two topologies in (5.1.1) also agree. For this we need some properties of compactly generated
topological groups. All topological groups under consideration are assumed to be Hausdorff.
Lemma 5.1.4. Let H and H ′ be closed subgroups of a Hausdorff compactly generated group
G. Then the quotient topology on H/H ∩ H ′ is the subspace topology inherited from the
quotient topology on G/H ′.
/
/
/
/
/
/
24
GUILLERMO CORTI NAS AND ANDREAS THOM
Proof. First of all, it is clear that the canonical inclusion map ι : H/H ∩ H ′ → G/H ′ is
continuous. Indeed, let π : G → G/H ′ be the projection; identify H → H/H ∩ H ′ with the
restriction of π. A subset A ⊂ G/H ′ is closed if and only if π−1(A) ∩ K is closed for K ⊂ G
compact. Hence, π−1(A) ∩ H ∩ K = π−1(A ∩ π(H)) ∩ K is closed and the claim follows
since compact subsets of H are also compact in G. Let now A ⊂ H/H ∩ H ′ be closed, i.e.
π−1(A) ∩ K ′ closed for every compact K ′ ⊂ H. For compact K ⊂ G, the set K ′ = K ∩ H
is compact in H and we get that π−1(A) ∩ K closed in H and hence in G. This finishes the
proof.
(cid:3)
Proposition 5.1.5. Let R be a unital, countable dimensional R-algebra, equipped with the
fine large semi-algebraic structure. Assume that R has the algebraic compactness property.
Then the homogeneous and induced large semi-algebraic structures of Pn(R) agree.
Proof. Consider the diagram
GL(−∞,n](R) × GL[n+1,∞)(R)
/ GL(−∞,∞)(R)
/ P ∞
n (R)
GL[1,n](R) × GL[n+1,∞)(R)
/ GL[1,∞)(R)
Pn(R).
Now apply Lemma 5.1.4.
(cid:3)
5.2. Bounded sequences, algebraic compactness and K0-triviality. Let X be a large
semi-algebraic set and let {Xλ} be a defining structure. The space of bounded sequences in
X is
ℓ∞(X) = ℓ∞(N, X) = {z : N → X ∃λ : z(N) ⊂ Xλ}
Note that with our definition, the objects ℓ∞(R) and ℓ∞C coincide with the well-known
spaces of bounded sequences.
Lemma 5.2.1. Let F = R or C and V a countable dimensional F-vectorspace equipped with
the fine large semi-algebraic structure. Then the natural map
is an isomorphism.
ℓ∞(F) ⊗F V → ℓ∞(V )
Proof. Choose a basis {vq} of V . Any element of ℓ∞(F) ⊗F V can be written uniquely as a
finite sum Pq λq ⊗ vq; this gets mapped to the sequence {n 7→ Pq λq(n) · vq}, which vanishes
if and only if all the λq are zero. This proves the injectivity statement. Let z ∈ ℓ∞(V ); by
definition, there is a finite dimensional subspace W ⊂ V and a bounded closed semi-algebraic
subset S ⊂ W such that z(N) ⊂ S. We may assume that S is a closed ball centered at zero,
and that W is the smallest subspace containing z(N). Hence there exist i1 < · · · < ip ∈ N
such that B = {zi1, . . . , zip} is a basis of W . The map W → Rp, w 7→ [w]B that sends a
vector w to the p-tuple of its coordinates with respect to B is linear and therefore bounded.
In particular there exists a C > 0 such that [w]B∞ < C for all w ∈ S. Thus we may write
(cid:3)
j=1 λizij with λi ∈ ℓ∞(F). This proves the surjectivity assertion of the lemma.
z = Pp
/
/
O
O
/
/
/
O
O
O
O
ALGEBRAIC GEOMETRY OF TOPOLOGICAL SPACES I
25
Proposition 5.2.2. Let R be a unital, countable dimensional R-algebra, equipped with the
fine large semi-algebraic structure. Then the following are equivalent
(i) R has the algebraic compactness property.
(ii) For every n the map
(5.2.3)
P ∞
n (ℓ∞(R)) → ℓ∞(P ∞
n (R))
is surjective.
(iii) The map K0(ℓ∞(R)) → Qr≥1 K0(R) is injective.
Proof. Choose countable indexed structures {Gn} on G = GLZ(R), and {Xn} on X =
M2((M∞R)+). Let π : G → G/H = P ∞
n (R) be the projection. We know already (see 4.2.4)
that the induced structure on P ∞
n (R) is coarser than the homogeneous one, i.e. that each
(G/H)r = π(Gr) is contained in some Xm. Assertion (i) is therefore equivalent to saying that
each P ∞
n (R) ∩ Xm is contained in some π(Gr). Negating this we obtain a bounded sequence
e = (er) of idempotent matrices, i.e. e ∈ ℓ∞(P ∞
n (R)) with respect to the induced large semi-
n (R), each er is equivalent to 1∞ ⊕ (1n ⊕ 0∞) in M2(M∞(R)+), but
algebraic structure on P ∞
there is no sequence (gr) of invertible matrices in GL2(M∞(R)+) such that grerg−1
r = 1∞ ⊕
(1n ⊕ 0∞) and both (gr) and (g−1
n (ℓ∞(R)). We
have shown that the negation of (i) is equivalent to that of (ii). Next note that every element
x ∈ K0(ℓ∞(R)) can be written as a difference x = [e] − [1∞ ⊕ 0∞] with e ∈ M2(M∞(R)+)
idempotent and e ≡ 1∞ ⊕ 0∞ modulo the ideal M2M∞(R). The idempotent e is determined
r ) are bounded. In other words, e is not in P ∞
if each er is conjugate to 1∞ ⊕ 0∞. Hence condition (iii) is satisfied if (ii) is. The converse
follows easily. Indeed, for any sequence (er) as above, we see that the image of the classes
up to conjugation by GL2(M∞(R)+). The element x goes to zero in Qr≥1 K0(R) if and only
[e]−[1∞ ⊕0∞] and [1∞ ⊕(1n ⊕0∞)]−[1∞ ⊕0∞] in Qp≥1 K0(R) coincide. Hence, by injectivity
of the comparison map, e is conjugate to 1∞ ⊕ (1n ⊕ 0∞) and we get a sequence of invertible
elements (gr) in GL2(M∞(R)+) such that gr conjugates er to 1∞⊕(1n⊕0∞) and the sequences
(gr) and (g−1
(cid:3)
r ) are bounded. This completes the proof.
Example 5.2.4. Both R and C have the algebraic compactness property since the third
condition is well-known to be satisfied. Indeed, ℓ∞(R) and ℓ∞(C) are (real) C ∗-algebras, and
one can easily compute that
K0(ℓ∞(C)) = ℓ∞(Z) ⊂ Yn≥1
Z = Yn≥1
K0(C).
The same computation applies to R in place of C.
5.3. Algebraic approximation and bounded sequences.
Theorem 5.3.1. Let F and G be functors from commutative C-algebras to sets. Assume
that both F and G preserve filtering colimits. Let τ : F → G be a natural transformation.
Assume that τ (O(V )) is injective (resp. surjective) for each smooth affine algebraic variety
V over C. Then τ (ℓ∞(C)) is injective (resp. surjective).
26
GUILLERMO CORTI NAS AND ANDREAS THOM
Proof. Let F ⊂ ℓ∞C be a finite subset. Put AF = ChF i ⊂ ℓ∞(C) for the unital subalgebra
generated by F . Because AF is reduced, it corresponds to an affine algebraic variety VF and
the inclusion AF ⊂ ℓ∞C is dual to a map with pre-compact image ιF : N → (VF )an to the
analytic variety associated to VF . Thus we may write ℓ∞(C) as the filtering colimit
(5.3.2)
ℓ∞(C) = colimN→VanO(V )
Here the colimit is taken over all maps ι : N → Van whose codomain is the associated
analytic variety of the closed points of some affine algebraic variety V , and which have
precompact image in the euclidean topology. We claim that every such map factors through
a map V ′
an → Van with V ′ smooth and affine. Note that the claim implies that we may write
(5.3.2) as a colimit of smooth algebras; the theorem is immediate from this. Recall Hironaka's
desingularization (see [23]) provides a proper surjective homomorphism of algebraic varieties
π : V → V from a smooth quasi-projective variety. Thus the induced map πan : Van → Van
between the associated analytic varieties is proper and surjective for the usual euclidean
topologies. It follows from this that we can lift ι along πan. Next, Jouanoulou's device (see
[25]) provides a smooth affine vector bundle torsor σ : V ′ → V ; the associated map σan is
also a bundle torsor, and in particular a fibration and weak equivalence. Because Van is a
CW -complex, σan admits a continuous section. Thus ιF finally factors through the smooth
affine variety V ′
F .
(cid:3)
Remark 5.3.3. The proof above does not work in the real case, since a desingularization
V → V of real algebraic varieties need not induce a surjective map between the corresponding
real analytic (or semi-algebraic) varieties. For example, consider R = R[x, y]/hx2 + y2 − x3i.
The homomorphism f : R → R[t], f (x, y) 7→ f (t2 + 1, t(t2 + 1)), is injective and R[t] is
integral over R. Thus the induced scheme homomorphism f# : A1
R = Spec R[t] → V =
Spec R[x, y]/hx2 + y2 − x3i is a desingularization; it is finite (whence proper) and surjective,
and an isomorphism outside of the point zero, represented by the maximal ideal M = hx, yi ∈
V . But note that the pre-image of M consists just of the maximal ideal ht2 + 1i, which has
residue field C; this means that the pre-image of zero has no real points. Therefore the
restriction of f# to real points is not surjective.
5.4. The algebraic compactness theorem.
Theorem 5.4.1. Let R be a countable dimensional unital C-algebra. Assume that the map
K0(O(V )) → K0(O(V ) ⊗ R) is an isomorphism for every affine smooth algebraic variety V
over C. Then R has the algebraic compactness property.
Proof. We have a commutative diagram
K0(ℓ∞(R))
K0(ℓ∞(C))
/ Qp≥1 K0(R)
/ Qp≥1 K0(C)
/
O
O
/
O
O
ALGEBRAIC GEOMETRY OF TOPOLOGICAL SPACES I
27
The bottom row is a monomorphism by Example 5.2.4. Our hypothesis on R together with
Theorem 5.3.1 applied to the natural transformation K0(−) → K0(− ⊗C R) imply that both
columns are isomorphisms. It follows that the top row is injective, which by Proposition
5.2.2 says that R satisfies algebraic compactness.
(cid:3)
6. Applications: projective modules, lower K-theory, and bundle theory.
6.1. Parametrized Gubeladze's theorem and Rosenberg's conjecture. All monoids
considered are commutative, cancellative and without nonzero torsion elements. If M is
cacellative then it embeds into its total quotient group G(M). A cancellative monoid M is
said to be seminormal if for every element x of the total quotient group G(M) for which 2x
and 3x are contained in the monoid M, it follows that x is contained in the monoid M.
The following is a particular case of a theorem of Joseph Gubeladze, which in turn
generalized the celebrated theorem of Daniel Quillen [35] and Andrei Suslin [42] which settled
Serre's conjecture: every finitely generated projective module over a polynomial ring over a
field is free.
Theorem 6.1.1 (see [20],[19]). Let D be a principal ideal domain, and M a commutative,
cancellative, torsion free, seminormal monoid. Then every finitely generated projective mod-
ule over the monoid algebra D[M] is free.
We shall also need the following generalization of Gubeladze's theorem, due to Richard
G. Swan. Recall that if R → S is a homomorphism of unital rings, and M is an S-module,
then we say that M is extended from R if there exists an R-module N such that M ∼= S ⊗R N
as S-modules.
Theorem 6.1.2 (see [46]). Let R = O(V ) be the coordinate ring of a smooth affine algebraic
variety over a field, and let d = dim V . Also let M be a torsion-free, seminormal, cancellative
monoid. Then all finitely generated projective R[M]-modules of rank n > d are extended from
R.
In the next theorem and elsewhere below, we shall consider only the complex case; thus
in what follows, C(X) shall always means map(X, C).
Theorem 6.1.3. Let X be a contractible compact space, and M an abelian, countable,
torsion-free, seminormal, cancellative monoid. Then every finitely generated projective mod-
ule over C(X)[M] is free.
Proof. The assertion of the theorem is equivalent to the assertion that every idempotent
matrix with coefficients in C(X)[M] is conjugate to a diagonal matrix with only zeroes and
ones in the diagonal. By Lemma 4.3.3, an idempotent matrix with coefficients in C(X)[M]
is the same as a continuous map from X to the space Idem∞(C[M]) of all idempotent
matrices in M∞(C[M]), equipped with the induced topology. Now observe that, since the
trace map M∞C[M] → C[M] is continuous, so is the rank map Idem∞(C[M]) → N0. Hence
by Theorem 6.1.1, the space Idem∞(C[M]) is the topological coproduct Idem∞(C[M]) =
`n Pn(C[M]), and thus any continuous map e : X → Idem∞(C[M]) factors through a
28
GUILLERMO CORTI NAS AND ANDREAS THOM
map e : X → Pn(C[M]). By Theorem 6.1.2 and Theorem 5.4.1, the induced topology
of Pn(C[M]) = GL(C[M])/GL[1,n](C[M]) × GL[n+1,∞)(C[M]) coincides with the quotient
topology. By Theorem 4.4.3, e lifts to a continuous map g : X → GL(C[M]). By Lemma
4.3.3, g ∈ GL(C(X)[M]) and conjugates e to 1n ⊕ 0∞. This concludes the proof.
(cid:3)
Theorem 6.1.4. The functor Comp → Ab, X 7→ K0(C(X)[M]) is homotopy invariant.
Proof. Immediate from Theorem 6.1.3 and Proposition 2.1.5.
(cid:3)
Theorem 6.1.5 (Rosenberg's conjecture). The functor Comp → Ab, X 7→ K−n(C(X)) is
homotopy invariant for n > 0.
Proof. By (2.2.2), K−n(C(X)) is naturally a direct summand of K0(C(X)[Zn]), whence it is
homotopy invariant by Theorem 6.1.4.
(cid:3)
Remark 6.1.6. Let X be a compact topological space, S1 the circle and j ≥ 0. By (2.2.2),
Theorem 6.1.5 and excision, we have
K−j(C(X × S1)) =K−j(C(X)) ⊕ K−j−1(C(X))
=K−j(C(X)[t, t−1])
Thus the effect on negative K-theory of the cartesian product of the maximal ideal spectrum
X = Max(C(X)) with S1 is the same as that of taking the product of the prime ideal spec-
trum Spec C(X) with the algebraic circle Spec (C[t, t−1]). More generally, for the C ∗-algebra
tensor product ⊗min and any commutative C ∗-algebra A, we have K−j(A ⊗min C(S1)) =
K−j(A ⊗C C[t, t−1]) (j > 0).
Remark 6.1.7. Theorem 6.1.5 was stated by Jonathan Rosenberg in [36, Thm. 2.4] and again
in [37, Thm. 2.3] for the real case. Later, in [38], Rosenberg acknowledges that the proof
was faulty, but conjectures the statement to be true. Indeed, a mistake was pointed out
by Mark E. Walker (see line 8 on page 799 in [15] or line 12 on page 26 in [38]). In their
work on semi-topological K-theory, Eric Friedlander and Mark E. Walker prove [15, Thm.
5.1] that the negative algebraic K-theory of the ring C(∆n) of complex-valued continuous
functions on the simplex vanishes for all n. We show in Subsection 7.2 how another proof of
Rosenberg's conjecture can be obtained using the Friedlander-Walker result.
Remark 6.1.8. The proof of Theorem 6.1.5 does not need the detour of the proof of our main
results in the case n = 1. Indeed, the ring of germs of continuous functions at a point in X
is a Hensel local ring with residue field C and Vladimir Drinfel'd proves that K−1 vanishes
for Hensel local rings with residue field C, see [12, Thm. 3.7]. This solves the problem locally
and reduces the remaining complications to bundle theory. (This was observed by the second
author in discussions with Charles Weibel at Institut Henri Poincar´e, Paris in 2004.) No direct
approach like this is known for K−2 or in lower dimensions.
Already in [36], Rosenberg computed the values of negative algebraic K-theory on com-
mutative unital C ∗-algebras, assuming the homotopy invariance result.
ALGEBRAIC GEOMETRY OF TOPOLOGICAL SPACES I
29
Corollary 6.1.9 (Rosenberg, [38]). Let X be a compact topological space. Let bu denote the
connective K-theory spectrum. Then,
K−i(C(X)) = bui(X) = [ΣiX, bu],
i ≥ 0.
The fact that connective K-theory shows up in this context was further explored and
clarified in the thesis of the second author [48], which was also partially built on the validity
of Theorem 6.1.5.
6.2. Application to bundle theory: local triviality. Let R be a countable dimensional
R-algebra. Any finitely generated R-module M is a countable dimensional vectorspace, and
thus it can be regarded as a compactly generated topological space. We consider not neces-
sarily locally trivial bundles of finitely generated free R-modules over compact spaces, such
that each fiber is equipped with the compactly generated topology just recalled. We call such
such a gadget a quasi-bundle of finitely generated free R-modules
Theorem 6.2.1. Let X be a compact space, M a countable, torsion-free, seminormal, can-
cellative monoid. Let E → X be a quasi-bundle of finitely generated free C[M]-modules.
Assume that there exist an n ≥ 1, another quasi-bundle E′, and a quasi-bundle isomorphism
E ⊕ E′ ∼= X × C[M]n. Then E is locally trivial.
Proof. Put R = C[M]. The isomorphism E ⊕ E′ ∼= X × Rn gives a continuous function
e : X → Pn(R); E is locally trivial if e is locally conjugate to an idempotent of the form
1r ⊕ 0∞, i.e. if it can be lifted locally along the projection GL(R) → Pn(R) to a continuous
map X → GL(R). Our hypothesis on M together with Theorems 6.1.3, 5.4.1, 6.1.2, and
4.4.3 imply that such local liftings exist.
(cid:3)
7. Homotopy invariance.
7.1. From compact polyhedra to compact spaces: a result of Calder-Siegel. Con-
sider the category Comp of compact Hausdorff topological spaces with continuous maps
and its full subcategory Pol ⊂ Comp formed by those spaces which are compact polyhe-
dra. In this subsection we show that for a functor which commutes with filtering colimits
and is split exact on C ∗-algebras, homotopy invariance on Pol implies homotopy invariance
on Comp. For this we shall need a particular case of a result of Allan Calder and Jerrold
Siegel [6, 7] that we shall presently recall. We point out that the Calder-Siegel results have
been further generalized by Armin Frei in [14]. For each object X ∈ Comp we consider
the comma category (X ↓ Pol) whose objects are morphisms f : X → cod(f ) where the
codomain cod(f ) is a compact polyhedron. Morphisms are commutative diagrams as usual.
Let G : Pol → Ab be a (contravariant) functor to the category of abelian groups. Its right
Kan extension GPol : Comp → Ab is defined by
GPol(X) = colimf ∈(X↓Pol)G(cod(f )),
∀X ∈ Comp.
The result of Calder-Siegel (see Corollary 2.7 and Theorem 2.8 in [7]) says that homotopy
invariance properties of G give rise to homotopy invariance properties of GPol. More precisely:
30
GUILLERMO CORTI NAS AND ANDREAS THOM
Theorem 7.1.1 (Calder-Siegel). If G : Pol → Ab is a (contravariant) homotopy invariant
functor, then the functor GPol : Comp → Ab is homotopy invariant.
We want to apply the theorem when G is of the form D 7→ E(C(D)), the functor E
commutes with (algebraic) filtering colimits, and is split exact on C ∗-algebras. For this we
have to compare E with the right Kan extension of G; we need some preliminaries. Let
X ∈ Comp, D ⊂ C the unit disk, and F the set of all finite subsets of C(X, D). Since X
is compact, for each f ∈ C(X) there is an n ∈ N such that (1/n)f ∈ C(X, D). Thus any
finitely generated subalgebra A ⊂ C(X) is generated by a finite subset F ⊂ C(X, D). Let F
be the set of all finite subsets of C(X, D); since C(X) is the colimit of its finitely generated
subalgebras ChF i, we have
colimF ∈F ChF i = C(X).
For F ∈ F , write YF ⊂ CF for the Zariski closure of the image of the map αF : X → CF ,
x 7→ (f (x))f ∈F . The image of αF is contained in the compact semi-algebraic set
PF = D ∩ YF
In particular, PF is a compact polyhedron. Note that αF induces an isomorphism between
the ring O(YF ) of regular polynomial functions and the subalgebra ChF i ⊂ C(X) generated
by F . Hence the inclusion PF ⊂ YF induces a homomorphism βF which makes the following
diagram commute
ChF i
/ C(X)
:uuuuuuuuu
$IIIIIIIII
βF
C(PF )
Taking colimits we obtain
(7.1.2)
C(X)
C(X)
'OOOOOOOOOOO
β
π
7ooooooooooo
colimF ∈F C(PF )
Thus the map π is a split surjection.
We can now draw the desired conclusion.
Theorem 7.1.3. Let E : Comm → Ab be a functor. Assume that F satisfies each of the
following conditions.
(1) E commutes with filtered colimits.
(2) Pol → Ab, D 7→ E(C(D)), is homotopy invariant.
Then the functor
is homotopy invariant on the category of compact topological spaces.
Comp → Ab, X 7→ E(C(X)),
$
/
:
'
7
ALGEBRAIC GEOMETRY OF TOPOLOGICAL SPACES I
31
Proof. By (7.1.2) and the first hypothesis, the map E(β) is a right inverse of the map
E(π) : E(colimF ∈F C(PF )) = colimF ∈F E(C(PF )) → E(C(X)).
On the other hand, we have a commutative diagram
(7.1.4)
colimF ∈FE(C(PF ))
θ
colimf ∈(X↓Pol)E(C(cod(f )))
*VVVVVVVVVVVVVVVVVVV
E(π)
π′
E(C(X))
Hence the map π′ in the diagram above is split by the composite θE(β), and therefore
E(C(−)) is naturally a direct summand of the functor
(7.1.5)
GPol : Comp → Ab, X 7→ colimf ∈(X↓Pol)E(C(cod(f )))
But (7.1.5) is the right Kan extension of the functor G : Pol → Ab, G(K) = E(C(K)),
and thus it is homotopy invariant by the second hypothesis and Calder-Siegel's theorem. It
follows that E(C(−)) is homotopy invariant, as we had to prove.
(cid:3)
Remark 7.1.6. If in Theorem 7.1.3 the functor E is split exact, then A 7→ E(A) is homotopy
invariant on the category of commutative C ∗-algebras. Indeed, since every commutative
unital C ∗-algebra is of the form C(X) for some compact space X, it follows that F is
homotopy invariant on unital commutative C ∗-algebras. Using this and split exactness, we
get that it is also homotopy invariant on all commutative C ∗-algebras.
Remark 7.1.7. In general, one cannot expect that the homomorphism π′ of (7.1.4) be an
isomorphism. For the injectivity one would need the following implication: If D is a compact
polyhedron, and f : X → D and s : D → C are continuous maps such that 0 = s◦f : X → C,
then there exist a compact polyhedron D′ and continuous maps g : X → D′, h : D′ → D
such that h ◦ g = f : X → D and 0 = h ◦ s : D′ → C. That is too strong if X is a
pathological space. To give a concrete example: let X be a Cantor set inside [0, 1], f the
natural inclusion, and s the distance function to the Cantor set, and suppose g and h as
above exist. If 0 = h ◦ s : D′ → C, then the image of D′ in [0, 1] has to be contained in
X. But the image has only finitely many connected components, since D′ has this property.
Hence, since X is totally disconnected, the image of D′ in [0, 1] cannot be all of X. This is
a contradiction.
7.2. Second proof of Rosenberg's conjecture. A second proof of Rosenberg's conjecture
can be obtained by combining Theorem 7.1.3 with the following theorem, which is due to
Eric Friedlander and Mark E. Walker.
Theorem 7.2.1 (Theorem 5.1 in [15]). If n > 0 and q ≥ 0, then
K−n(C(∆q)) = 0
Second proof of Rosenberg's conjecture. By Proposition 2.1.5 and Theorem 7.1.3 it suffices
to show that Kn(C(D)) = 0 for contractible D ∈ Pol. If D is contractible, then the identity
1D : D → D factors over the cone cD. Hence, it is sufficient to show that Kn(C(cD)) = 0.
/
/
*
32
GUILLERMO CORTI NAS AND ANDREAS THOM
The cone cD is a star-like simplicial complex and for any subcomplexes A, B ⊂ D with
A ∪ B = D, we get a Milnor square
C(cD)
C(cA)
C(cB)
/ C(c(A ∩ B)).
Since cA is contractible, it retracts onto c(A ∩ B) and therefore, the square above is split.
Using excision, we obtain split-exact sequence of abelian groups as follows:
0 → Kn(C(cD)) → Kn(C(cA)) ⊕ Kn(C(cB)) → Kn(C(c(A ∩ B))) → 0.
Decomposing cD like this, we see that the result of Theorem 7.2.1 is sufficient for the van-
ishing of Kn(C(cD)).
(cid:3)
7.3. The homotopy invariance theorem. The aim of this subsection is to prove the
following result.
Theorem 7.3.1. Let F be a functor on the category of commutative C-algebras with values
in abelian groups. Assume that the following three conditions are satisfied.
(i) F is split-exact on C ∗-algebras.
(ii) F vanishes on coordinate rings of smooth affine varieties.
(iii) F commutes with filtering colimits.
Then the functor
Comp → Ab, X 7→ F (C(X))
is homotopy invariant on the category of compact Hausdorff topological spaces and F (C(X)) =
0 for X contractible.
Proof. Note that, since a point is a smooth algebraic variety, our hypothesis imply that
F (C) = 0. Thus if F is homotopy invariant and X is contractible, then F (X) = F (C) = 0.
Let us prove then that X 7→ F (C(X)) is homotopy invariant on the category of compact
Hausdorff topological spaces. Proceeding as in the proof of Theorem 7.2.1, we see that it is
sufficient to show that F (C(∆n)) = 0, for all n ≥ 0. Any finitely generated subalgebra of
C(∆n) is reduced and hence corresponds to an algebraic variety over C. Since F commutes
with filtered colimits, we obtain:
F (C(∆n)) = colim∆n→Yan F (O(Y )),
where the colimit runs over all continuous maps from ∆n to the analytic variety Yan equipped
with the usual euclidean topology. For ease of notation, we will from now on just write Y
for both the algebraic variety and the analytic variety associated to it. Let ι : ∆n → Y be a
continuous map. As in the proof of the algebraic approximation theorem (5.3.1), we consider
Hironaka's desingularization π : Y → Y and Jouanoulou's affine bundle torsor σ : Y ′ → Y .
Let T ⊂ Y be a compact semi-algebraic subset such that ι(∆n) ⊂ T . Because π is a proper
morphism, T = π−1(T ) is compact and semi-algebraic. By definition of vector bundle torsor
/
/
/
ALGEBRAIC GEOMETRY OF TOPOLOGICAL SPACES I
33
([51]), there is a Zariski cover of Y such that the pull-back of σ over each open subscheme
U ⊂ Y of the covering is isomorphic, as a scheme over U, to an algebraic trivial vector
bundle. Thus σ is locally trivial fibration for the euclidean topologies, and the trivialization
maps are (semi)-algebraic. Hence because T , being compact, is locally compact, we may
find a finite covering { Ti} of T by closed semi-algebraic subsets such that σ is a trivial
fibration over each Ti, and compact semi-algebraic subsets Si ⊂ Y ′ such that σ(Si) = Ti.
Put S = Si Si; then S is compact semi-algebraic, and f = (π ◦ σ)S : S → T is a continuous
semi-algebraic surjection. By Theorem 3.2.1, there exists a semi-algebraic triangulation of T
such that ker(F (∆m) → F (f −1(∆m)) = 0 for each simplex ∆m in the triangulation. Consider
the diagram
F (C(f −1(∆m))
F (C(S))
F (O(Y ′))
F (C(∆m))
F (C(T ))
F (O(Y ))
If α ∈ F (O(Y )), then its image in F (C(f −1(∆m))) vanishes since f −1(∆m) → ∆m factors
through the smooth affine variety Y ′, and F (O(Y ′)) = 0. Hence, by Theorem 3.2.1, α∆m = 0,
for each simplex in the triangulation. Coming back to the map ι : ∆n → Y , we have a diagram
∆n
∆m
!CCCCCCCC
Y
/ T
ι
θ
Here θ : ∆m → T is the inclusion of a simplex in the triangulation, and is the correstriction
of ι. We need to conclude that ι∗(α) = 0, knowing only that θ∗(α) = 0 for each simplex in a
triangulation of T . This is done using split-exactness and barycentric subdivisions. Indeed, we
perform the barycentric subdivision of ∆n sufficiently many times so that each n-dimensional
simplex is mapped to the closed star st(x) of some some vertex x in the triangulation of T .
Since ∆n is star-like, the reduction argument of the proof of Theorem 5.3.1 shows that it is
enough to show the vanishing of ι∗(α) for the (top-dimensional) simplices in this subdivision
of ∆n. If ∆′n is one of these top dimensional simplices, and ∆m ⊂ st(x), we can complete
the diagram above to a diagram
∆′n
∆m
"FFFFFFFF
/ st(x)
ι
∆n
!CCCCCCCCC
Y
/ T
Hence it suffices to show that the pullback of α to st(x) vanishes. But since st(x) is star-like,
then by the same reduction argument as before, the vanishing of the pullback of α to each
of the top simplices ∆m ⊂ st(x) is sufficient to conclude that αst(x) = 0. This finishes the
proof.
(cid:3)
As an application, we obtain the following.
o
o
o
o
O
O
O
O
o
o
o
o
O
O
/
/
!
/
O
O
/
/
"
/
/
!
/
/
O
O
34
GUILLERMO CORTI NAS AND ANDREAS THOM
Third proof of Rosenberg's conjecture. If n < 0 then Kn is split-exact and vanishes
on coordinate rings of smooth affine algebraic varieties. By Theorem 7.3.1, this implies that
X 7→ Kn(C(X)) is homotopy invariant.
(cid:3)
7.4. A vanishing theorem for homology theories.
Theorem 7.4.1. Let E : Comm/C → Spt be a homology theory of commutative C-algebras
and n0 ∈ Z. Assume
(i) E is excisive on commutative C ∗-algebras.
(ii) En commutes with algebraic filtering colimits for n ≥ n0.
(iii) En(O(V )) = 0 for each smooth affine algebraic variety V for n ≥ n0.
Then En(A) = 0 for every commutative C ∗-algebra A for n ≥ n0.
Proof. Let n ≥ n0. We have to show that En(A) = 0 for every commutative C ∗-algebra A.
Because by i), each En is split exact on commutative C ∗-algebras, it suffices to show that
En(A) = 0 for unital A, i.e. for A = C(X), X ∈ Comp. Since by ii) En preserves filtering
colimits, the proof of Theorem 7.1.3 shows that En(C(X)) is a direct summand of
colimf ∈(X↓Pol)E(C(cod(f )))
Hence it suffices to show that En(C(D)) = 0 for every compact polyhedron D. By iii) and
excision, this is true if dim D = 0. Let m ≥ 1 and assume the assertion of the theorem
holds for compact polyhedra of dimension < m. By Theorem 7.3.1, D 7→ En(C(D)) is
homotopy invariant; in particular En(C(∆m)) = 0. If dim D = m and D is not a simplex,
write D = ∆m ∪ D′, as the union of an m-simplex and a subcomplex D′ which has fewer
m-dimensional simplices. Put L = ∆m ∩ D′; then dim L < m, and we have an exact sequence
En+1(C(L)) → En(C(D)) → En(C(∆m)) ⊕ E(C(D′)) → En(C(L))
We have seen above that En(C(∆m)) = 0; moreover En(C(L)) = En+1(C(L)) = 0 because
dim L < m, and En(C(D′)) = 0 because D′ has fewer m-dimensional simplices than D. This
concludes the proof.
(cid:3)
8. Applications of the homotopy invariance and vanishing homology
8.1. K-regularity for commutative C ∗-algebras.
theorems.
Theorem 8.1.1. Let V be a smooth affine algebraic variety over C, R = O(V ), and A a
commutative C ∗-algebra. Then A ⊗C R is K-regular.
Proof. For each fixed p ≥ 1 and i ∈ Z, write F p(A) = hocofiber(K(A⊗R) → K(A[t1, . . . , tp]⊗
R)) for the homotopy cofiber. It suffices to prove that the homology theory F p : Comm/C →
Spt satisfies the hypothesis of Theorem 7.4.1. By [52, Corollary 9.7], A[t1, . . . , tp] ⊗C R is
K-excisive for every C ∗-algebra A and every p ≥ 1. It follows that the homology theory
F p : Ass/C → Spt is excisive on C ∗-algebras. In particular its restriction to Comm/C is
excisive on commutative C ∗-algebras. Moreover, if W is any smooth affine algebraic variety,
ALGEBRAIC GEOMETRY OF TOPOLOGICAL SPACES I
35
then R ⊗C O(W ) = O(V × W ), is regular noetherian, and therefore K-regular. Finally, F p
∗
preserves filtering colimits because both K∗ and (−) ⊗ Z[t1, . . . , tp] ⊗ R do.
(cid:3)
Remark 8.1.2. The case R = C of the previous theorem was discovered by Jonathan Rosen-
berg. Unfortunately, the two proofs he has given, in [37, Thm. 3.1] and [38, p. 866] turned
out to be problematic. A version of Theorem 8.1.1 for A = C(D), D ∈ Pol, was given by
Friedlander and Walker in [15, Thm. 5.3]. Furthermore, Rosenberg acknowledges in [38, p.
24] that Walker also found a proof of this in the general case, but that he did not publish it.
Anyhow, as Rosenberg observed in [37, p. 91], in this situation, the polyhedral case implies
the general case by a short reduction argument (this also follows from Theorem 7.1.3 above).
Hence, an essentially complete argument for the proof of Theorem 8.1.1 existed already in
the literature, although it was scattered in various sources.
The following corollary compares Quillen's algebraic K-theory with Charles A. Weibel's
homotopy algebraic K-theory, KH, introduced in [51].
Corollary 8.1.3. If A is a commutative C ∗-algebra, then the map K∗(A) → KH∗(A) is an
isomorphism.
Proof. Weibel proved in [51, Proposition 1.5] that if A is a unital K-regular ring, then
K∗(A) ∼→ KH∗(A). Using excision, it follows that this is true for all commutative C ∗-
algebras. Now apply Theorem 8.1.1.
(cid:3)
8.2. Hochschild and cyclic homology of commutative C ∗-algebras. In the following
paragraph we recall some basic facts about Hochschild and cyclic homology which we shall
need; the standard reference for these topics is Jean-Louis Loday's book [29].
Let k be a field of characteristic zero. Recall a mixed complex of k-vectorspaces is a graded
vectorspace {Mn}n≥0 together with maps b : M∗ → M∗−1 and B : M∗ → M∗+1 satisfying
b2 = B2 = bB + Bb = 0. One can associate various chain complexes to a mixed complex
M, giving rise to the Hochschild, cyclic, negative cyclic and periodic cyclic homologies of
M, denoted respectively HH∗, HC∗, HN∗ and HP∗. For example HH∗(M) = H∗(M, b).
A map of mixed complexes is a homogeneous map which commutes with both b and B.
It is called a quasi-isomorphism if it induces an isomorphism at the level of Hochschild
homology; this automatically implies that it also induces an isomorphism for HC and all
the other homologies mentioned above. For a k-algebra A there is defined a mixed complex
(C(A/k), b, B) with
Cn(A/k) = (cid:26) Ak ⊗k A⊗kn n > 0
n = 0
A
We write HH∗(A/k), HC∗(A/k), etc. for HH∗(C(A/k)), HC∗(C(A/k)), etc. If furthermore A
is unital and ¯A = A/k, then there is also a mixed complex ¯C(A/k) with ¯Cn(A/k) = A⊗k ¯A⊗kn
and the natural surjection C(A/k) → ¯C(A/k) is a quasi-isomorphism. Note also that
(8.2.1)
ker( ¯C( Ak/k) → ¯C(k/k)) = C(A/k)
36
GUILLERMO CORTI NAS AND ANDREAS THOM
If A commutative and unital, we have a third mixed complex (ΩA/k, 0, d) given in degree n
by Ωn
A/k, the module of n-Kahler differential forms, and where d is the exterior derivation of
forms. A natural map of mixed complexes µ : ¯C(A/k) → ΩA/k is defined by
(8.2.2)
µ(a0 ⊗k ¯a1 ⊗k · · · ⊗k ¯an) =
1
n!
a0da1 ∧ · · · ∧ dan
It was shown by Loday and Quillen in [30] (using a classical result of Hochschild-Kostant-
Rosenberg [24]) that µ is a quasi-isomorphism if A is a smooth k algebra, i.e. A = O(V ) for
some smooth affine algebraic variety over k. It follows from this (see [29]) that for ZΩn
A/k =
ker(d : Ωn
dR(A/k) = H ∗(ΩA/k, d), we have (n ∈ Z)
A/k → Ωn+1
(8.2.3)
HCn(A/k) = Ωn
A/k ) and H ∗
HHn(A/k) = Ωn
A/k
HNn(A/k) = ZΩn
A/k ⊕Yp>0
H n+2p
dR (A/k)
A/k/dΩn−1
A/k ⊕ M0≤2i<n
HPn(A/k) = Yp∈Z
H n−2i
dR (A/k)
H 2p−n
dR (A/k)
For arbitrary commutative unital A, there is a decomposition
Cn(A/k) = ⊕n
p=0C (p)(A/k)
such that b maps C (p) to itself, while B(C (p)) ⊂ C (p+1) (see [29]). One defines
HH (p)
n (A/k) = HnC (p)(A/k)
p (A/k) = Ωp
q (A/k) = 0 for q < p, HH (p)
We have HH (p)
A/k, but in general for q > 0,
HH (p)
p+q(A/k) 6= 0. The map (8.2.2) is still a quasi-isomorphism if A is smooth over a field
F ⊃ k; this follows from the Loday-Quillen result using the base change spectral sequence
of Kassel-Sletsjøe, which we recall below.
Lemma 8.2.4. (Kassel-Sletsjøe, [27, 4.3a]) Let k ⊆ F be fields of characteristic zero. For
each p ≥ 1 there is a bounded second quadrant homological spectral sequence (0 ≤ i < p,
j ≥ 0):
pE1
−i,i+j = Ωi
F/k ⊗F HH (p−i)
p−i+j(R/F ) ⇒ HH (p)
p+j(R/k)
Corollary 8.2.5. If A is a smooth F -algebra, then (8.2.2) is a quasi-isomorphism.
Theorem 8.2.6. Let X be a compact topological space, A = C(X), and k ⊂ C a subfield.
Then the map (8.2.2) is a quasi-isomorphism, and we have the identities (8.2.3).
Proof. Extend C (p)(/k) (and HH (p)
n (/k)) to non-unital algebras by
C (p)(A/k) = ker(C (p)( Ak/k) → C (p)(k/k)).
Let E(p)(A/k) be the spectrum the Dold-Kan correspondence associates to C (p)(A/k). Regard
E(p) as a homology theory of C-algebras. Then E(p) is excisive on C ∗-algebras, by Remark
2.3.6 and naturality. Further E(p)
n (A/k) = 0 whenever n > p and A is smooth
over C, by Corollary 8.2.5. It is also clear that HH (p)
∗ (/k) preserves filtering colimits, since
HH∗(/k) does. Thus we may apply Theorem 7.4.1 to conclude the proof.
(cid:3)
n (A/k) = HH (p)
ALGEBRAIC GEOMETRY OF TOPOLOGICAL SPACES I
37
8.3. The Farrell-Jones Isomorphism Conjecture. Let A be a ring, Γ a group, and
q ≤ 0. Put
W hA
q (Γ) = coker (Kq(A) → Kq(A[Γ]))
for the cokernel of the map induced by the natural inclusion A ⊂ A[Γ]. Recall [31, Conjec-
ture 1, pp. 708] that the Farrell-Jones conjecture with coefficients for a torsion free group
implies that if Γ torsion free and A a noetherian regular unital ring, then
(8.3.1)
W hA
q (Γ) = 0
(q ≤ 0)
Note that the conjecture in particular implies that Kq(A[Γ]) = 0 for q < 0 if A is noetherian
regular, for in this case we have Kq(A) = 0 for q < 0.
Theorem 8.3.2. Let Γ be torsion free group which satisfies (8.3.1) for every commutative
smooth C-algebra A. Also let q ≤ 0. Then the functor W h?
q(Γ) is homotopy invariant on
commutative C ∗-algebras.
Proof. It follows from Theorem 7.3.1 applied to W h?
q(Γ).
(cid:3)
Corollary 8.3.3. Let Γ be as above and X a contractible compact space. Then K0(C(X)[Γ]) =
Z and Kq(C(X)[Γ]) = 0 for q < 0.
Corollary 8.3.4. Let Γ be as above. Then, the functor
X 7→ Kq(C(X)[Γ])
is homotopy invariant on the category of compact topological spaces for q ≤ 0.
Proof. This follows directly from Proposition 2.1.5 and the preceding corollary.
(cid:3)
The general case of the Farrell-Jones conjecture predicts that for any group Γ and any
unital ring R, the assembly map [31]
AΓ(R) : HΓ(EVC(Γ), K(R)) → K(R[Γ])
(8.3.5)
is an equivalence. Here HΓ(−, K(R)) is the equivariant homology theory associated to the
spectrum K(R) and EVC((Γ)) is the classifying space with respect to the class of virtually
cyclic subgroups (see [31] for definitions of these objects).
We also get:
Theorem 8.3.6. Let Γ be a group such that the map (8.3.5) is an equivalence for every
smooth commutative C-algebra A. Then (8.3.5) is an equivalence for every C ∗-algebra A.
Proof. It follows from Theorem 7.4.1 applied to E(R) = hocofiber(AΓ(R)).
(cid:3)
Remark 8.3.7. We have
Kq(C(X)[Γ]) = πS
q (cid:0)map+(X+, KD(C[Γ]))(cid:1) ,
(q ≤ 0)
where KD(C[Γ]) denotes the diffeotopy K-theory spectrum, see Definition 4.1.3 in [10].
This follows from the study of a suitable co-assembly map and is not carried out in detail
here. The homotopy groups of KD(C[Γ]) can be computed from the equivariant connective
K-homology of EΓ using the Farrell-Jones assembly map.
38
GUILLERMO CORTI NAS AND ANDREAS THOM
8.4. Adams operations and the decomposition of rational K-theory. The rational
K-theory of a unital commutative ring A carries a natural decomposition
Kn(A) ⊗ Q = ⊕i≥0Kn(A)(i)
Here Kn(A)(i) = Tk6=0{x ∈ Kn(A) ψk(x) = kix}, where ψk is the Adams operation. For
example, K (0)
6.8]). A conjecture of Alexander Beılinson and Christophe Soul´e (see [3], [40]) asserts that
0 (A) = H 0(Spec A, Q) is the rank component, and K (0)
n (A) = 0 for n > 0 ([28,
(8.4.1)
K (i)
n (A) = 0 for n ≥ max{1, 2i}.
The conjecture as stated was proved wrong for non-regular A (see [16] and [13, 7.5.6]) but
no regular counterexamples have been found. Moreover, the original statement has been
formulated in terms of motivic cohomology (with rational, torsion and integral coefficients)
and generalized to regular noetherian schemes [26, 4.3.4]. For example if X = SpecR is
smooth then K (i)
n (R) = H 2i−n(X, Q(i)) is the motivic cohomology of X with coefficients in
the twisted sheaf Q(i).
We shall need the well-known fact that the validity of (8.4.1) for C implies its validity for
all smooth C-algebras; this is Proposition 8.4.3 below. In turn this uses the also well-known
fact that rational K-theory sends field inclusions to monomorphisms. We include proofs of
both facts for completeness sake.
Lemma 8.4.2. Let F ⊂ E be fields. Then K∗(E) ⊗ Q → K∗(F ) ⊗ Q is injective.
Proof. Since K-theory commutes with filtering colimits, we may assume that E/F is a finitely
generated field extension, which we may write as a finite extension of a finitely generated
purely transcendental extension. If E/F is purely transcendental, then by induction we are
reduced to the case E = F (t), which follows from [18, Thm. 1.3]. If d = dimF E is finite, then
the transfer map K∗(E) → K∗(F ) [34, pp. 111] splits K∗(F ) → K∗(E) up to d-torsion. (cid:3)
Proposition 8.4.3. If (8.4.1) holds for C, then it holds for all smooth C-algebras.
Proof. The Gysin sequence argument at the beginning of [26, 4.3.4] shows that if (8.4.1) is
an isomorphism for all finitely generated field extensions of C then it is an isomorphism for
all smooth R. If E ⊃ C is a finitely generated field extension, then we may write E = F [α]
for some purely transcendental field extension F ∼= C(t1, . . . , tn) ⊃ C and some algebraic
element α. From this and the fact that C is algebraically closed and of infinite transcendence
degree over Q, we see that E is isomorphic to a subfield of C. Now apply Lemma 8.4.2. (cid:3)
Theorem 8.4.4. Assume that (8.4.1) holds for the field C. Then it also holds for all com-
mutative C ∗-algebras.
Proof. By Proposition 8.4.3, our current hypothesis imply that (8.4.1) holds for smooth A. In
particular the homology theory K (i) vanishes on smooth A for n ≥ n0 = max{2i, 1}. Because
K-theory satisfies excision for C ∗-algebras and commutes with algebraic filtering colimits,
the same is true of K (i). Hence we can apply Theorem 7.4.1, concluding the proof.
(cid:3)
ALGEBRAIC GEOMETRY OF TOPOLOGICAL SPACES I
39
Acknowledgements. Part of the research for this article was carried out during a visit of
the first named author to Universitat Gottingen. He is indebted to this institution for their
hospitality. He also wishes to thank Chuck Weibel for a useful e-mail discussion of Beılinson-
Soul´e's conjecture. A previous version of this article contained a technical mistake in the
proof of Theorem 7.1.3; we are thankful to Emanuel Rodr´ıguez Cirone for bringing this to
our attention.
References
[1] Bass, H., Some problems in "classical" algebraic K-theory. In Algebraic K-theory, II: "Classical" alge-
braic K-theory and connections with arithmetic (Proc. Conf., Battelle Memorial Inst., Seattle, Wash.,
1972), pages 3 -- 73. Lecture Notes in Math., Vol. 342. Springer, Berlin, 1973.
[2] Basu, S., Pollack, R., & Roy, M-F., Algorithms in real algebraic geometry, volume 10 of Algorithms and
Computation in Mathematics. Springer-Verlag, Berlin, 2003.
[3] Beılinson, A. A., Higher regulators and values of L-functions. In Current problems in mathematics, Vol.
24, Itogi Nauki i Tekhniki, pages 181 -- 238. Akad. Nauk SSSR Vsesoyuz. Inst. Nauchn. i Tekhn. Inform.,
Moscow, 1984.
[4] Bourbaki, N., Espaces vectoriels topologiques. Chapitres 1 `a 5. Masson, Paris, new edition, 1981.
´El´ements de math´ematique. [Elements of mathematics].
[5] Brumfiel, G. W., Quotient spaces for semialgebraic equivalence relations. Math. Z., 195(1):69 -- 78, 1987.
[6] Calder, A. & Siegel, J., Homotopy and Kan extensions. In Categorical topology (Proc. Conf., Mannheim,
1975), pages 152 -- 163. Lecture Notes in Math., Vol. 540. Springer, Berlin, 1976.
[7] Calder, A. & Siegel, J., Kan extensions of homotopy functors. J. Pure Appl. Algebra, 12(3):253 -- 269,
1978.
[8] Cortinas, G., Haesemeyer, C., Walker, M. E. & Weibel, C. A. Bass' N K-groups and cdh-fibrant
Hochschild homology. arXiv:0802.1928, to appear in Invent. Math.
[9] Cortinas, G., Haesemeyer, C., Walker, M. E. & Weibel, C. A., A negative answer to a question of Bass.
arXiv:1004.3829, to appear in Proc. Amer. Math. Soc.
[10] Cortinas, G. & Thom, A., Comparison between algebraic and topological K-theory of locally convex
algebras. Adv. Math., 218(1):266 -- 307, 2008.
[11] Davis, J.F., Some remarks on nil groups in algebraic K-theory. arXiv:0803.1641v2
[12] Drinfeld, V., Infinite-dimensional vector bundles in algebraic geometry: an introduction. In The unity
of mathematics, volume 244 of Progr. Math., pages 263 -- 304. Birkhauser Boston, Boston, MA, 2006.
[13] Feıgin, B. L. & Tsygan, B. L., Additive K-theory. In K-theory, arithmetic and geometry (Moscow,
1984 -- 1986), volume 1289 of Lecture Notes in Math., pages 67 -- 209. Springer, Berlin, 1987.
[14] Frei, A., Kan extensions along full functors: Kan and Cech extensions of homotopy invariant functors.
J. Pure Appl. Algebra, 17(3):285 -- 292, 1980.
[15] Friedlander, E. M. & Walker, M. E., Comparing K-theories for complex varieties. Amer. J. Math.,
123(5):779 -- 810, 2001.
[16] Geller, S. E. & Weibel, C. A., Hodge decompositions of Loday symbols in K-theory and cyclic homology.
K-Theory, 8(6):587 -- 632, 1994.
[17] Gersten, S. M., On the spectrum of algebraic K-theory. Bull. Amer. Math. Soc., 78:216 -- 219, 1972.
[18] Gersten, S. M., Some exact sequences in the higher K-theory of rings. In Algebraic K-theory, I: Higher
K-theories (Proc. Conf., Battelle Memorial Inst., Seattle, Wash., 1972), pages 211 -- 243. Lecture Notes
in Math., Vol. 341. Springer, Berlin, 1973.
[19] Gubeladze, J., The Anderson conjecture and projective modules over monoid algebras. Soobshch. Akad.
Nauk Gruzin. SSR, 125(2):289 -- 291, 1987.
40
GUILLERMO CORTI NAS AND ANDREAS THOM
[20] Gubeladze, J., The Anderson conjecture and a maximal class of monoids over which projective modules
are free. Mat. Sb. (N.S.), 135(177)(2):169 -- 185, 271, 1988.
[21] Gubeladze, J., On Bass' question for finitely generated algebras over large fields. Bull. Lond. Math. Soc.,
41(1):36 -- 40, 2009.
[22] Hardt, R. M., Semi-algebraic local-triviality in semi-algebraic mappings. Amer. J. Math., 102(2):291 --
302, 1980.
[23] Hironaka, H., Resolution of singularities of an algebraic variety over a field of characteristic zero. I, II.
Ann. of Math. (2) 79 (1964), 109 -- 203; ibid. (2), 79:205 -- 326, 1964.
[24] Hochschild, G. , Kostant, B. & Rosenberg, A., Differential forms on regular affine algebras. Trans. Amer.
Math. Soc., 102:383 -- 408, 1962.
[25] Jouanolou, J. P., Une suite exacte de Mayer-Vietoris en K-th´eorie alg´ebrique. In Algebraic K-theory, I:
Higher K-theories (Proc. Conf., Battelle Memorial Inst., Seattle, Wash., 1972), pages 293 -- 316. Lecture
Notes in Math., Vol. 341. Springer, Berlin, 1973.
[26] Kahn, B., Algebraic K-theory, algebraic cycles and arithmetic geometry. In Handbook of K-theory. Vol.
1, 2, pages 351 -- 428. Springer, Berlin, 2005.
[27] Kassel, C. & Sletsjøe, A. B., Base change, transitivity and Kunneth formulas for the Quillen decompo-
sition of Hochschild homology. Math. Scand., 70(2):186 -- 192, 1992.
[28] Kratzer, Ch., λ-structure en K-th´eorie alg´ebrique. Comment. Math. Helv., 55(2):233 -- 254, 1980.
[29] Loday, J-L., Cyclic homology, volume 301 of Grundlehren der Mathematischen Wissenschaften [Funda-
mental Principles of Mathematical Sciences]. Springer-Verlag, Berlin, second edition, 1998. Appendix
E by Mar´ıa O. Ronco, Chapter 13 by the author in collaboration with Teimuraz Pirashvili.
[30] Loday, J-L & Quillen, D., Cyclic homology and the Lie algebra homology of matrices. Comment. Math.
Helv., 59(4):569 -- 591, 1984.
[31] Luck, W. & Reich, H., The Baum-Connes and the Farrell-Jones conjectures in K- and L-theory. In
Handbook of K-theory. Vol. 1, 2, pages 703 -- 842. Springer, Berlin, 2005.
[32] Pedersen, E. K. & Weibel, C. A., A nonconnective delooping of algebraic K-theory. In Algebraic and
geometric topology (New Brunswick, N.J., 1983), volume 1126 of Lecture Notes in Math., pages 166 -- 181.
Springer, Berlin, 1985.
[33] Pedersen, E. K. & Weibel, C. A., K-theory homology of spaces. In Algebraic topology (Arcata, CA,
1986), volume 1370 of Lecture Notes in Math., pages 346 -- 361. Springer, Berlin, 1989.
[34] Quillen, D., Higher algebraic K-theory. I. In Algebraic K-theory, I: Higher K-theories (Proc. Conf.,
Battelle Memorial Inst., Seattle, Wash., 1972), pages 85 -- 147. Lecture Notes in Math., Vol. 341. Springer,
Berlin, 1973.
[35] Quillen, D., Projective modules over polynomial rings. Invent. Math., 36:167 -- 171, 1976.
[36] Rosenberg, J., K and KK: topology and operator algebras. In Operator theory: operator algebras and
applications, Part 1 (Durham, NH, 1988), volume 51 of Proc. Sympos. Pure Math., pages 445 -- 480.
Amer. Math. Soc., Providence, RI, 1990.
[37] Rosenberg, J., The algebraic K-theory of operator algebras. K-Theory, 12(1):75 -- 99, 1997.
[38] Rosenberg, J., Comparison between algebraic and topological K-theory for Banach algebras and C ∗-
algebras. In Handbook of K-theory. Vol. 1, 2, pages 843 -- 874. Springer, Berlin, 2005.
[39] Serre, J-P., Faisceaux alg´ebriques coh´erents. Ann. of Math. (2), 61:197 -- 278, 1955.
[40] Soul´e, Ch., Op´erations en K-th´eorie alg´ebrique. Canad. J. Math., 37(3):488 -- 550, 1985.
[41] Suslin, A. A., On the K-theory of algebraically closed fields. Invent. Math., 73(2):241 -- 245, 1983.
[42] Suslin, A. A., Projective modules over polynomial rings are free. Dokl. Akad. Nauk SSSR, 229(5):1063 --
1066, 1976.
[43] Suslin, A. A. & Wodzicki, M., Excision in algebraic K-theory and Karoubi's conjecture. Proc. Nat.
Acad. Sci. U.S.A., 87(24):9582 -- 9584, 1990.
[44] Suslin, A. A. & Wodzicki, M., Excision in algebraic K-theory. Ann. of Math. (2), 136(1):51 -- 122, 1992.
ALGEBRAIC GEOMETRY OF TOPOLOGICAL SPACES I
41
[45] Swan, R. G., Projective modules over Laurent polynomial rings. Trans. Amer. Math. Soc., 237:111 -- 120,
1978.
[46] Swan, R. G., Gubeladze's proof of Anderson's conjecture. In Azumaya algebras, actions, and modules
(Bloomington, IN, 1990), volume 124 of Contemp. Math., pages 215 -- 250. Amer. Math. Soc., Providence,
RI, 1992.
[47] Switzer, R. M., Algebraic topology -- homotopy and homology. Classics in Mathematics. Springer-Verlag,
Berlin, 2002. Reprint of the 1975 original [Springer, New York].
[48] Thom, A. B., Connective E-theory and bivariant homology. Thesis, Universitat Munster, Mathematisch-
Naturwissenschaftliche Fakultat, Preprint of the SFB 478, 289:127 p., 2003.
[49] Thomason, R. W. & Trobaugh, T., Higher algebraic K-theory of schemes and of derived categories. In
The Grothendieck Festschrift, Vol. III, volume 88 of Progr. Math., pages 247 -- 435. Birkhauser Boston,
Boston, MA, 1990.
[50] Wagoner, J. B., Delooping classifying spaces in algebraic K-theory. Topology, 11:349 -- 370, 1972.
[51] Weibel, C. A., Homotopy algebraic K-theory. In Algebraic K-theory and algebraic number theory (Hon-
olulu, HI, 1987), volume 83 of Contemp. Math., pages 461 -- 488. Amer. Math. Soc., Providence, RI,
1989.
[52] Wodzicki, M., Excision in cyclic homology and in rational algebraic K-theory. Ann. of Math. (2),
129(3):591 -- 639, 1989.
[53] Wodzicki, M., Homological properties of rings of functional-analytic type. Proc. Nat. Acad. Sci. U.S.A.,
87(13):4910 -- 4911, 1990.
Guillermo Cortinas, Departamento de Matem´atica, FCEN-Universidad de Buenos Aires,
Ciudad Universitaria Pab 1, (1428) Buenos Aires, Argentina
E-mail address: [email protected]
URL: http://mate.dm.uba.ar/~gcorti
Andreas Thom, Mathematisches Institut der Universitat Leipzig, Johannisgasse 26,
04103 Leipzig, Germany.
E-mail address: [email protected]
URL: http://www.math.uni-leipzig.de/MI/thom
|
1312.5008 | 1 | 1312 | 2013-12-18T00:21:50 | Symmetric matrices, orthogonal Lie algebras, and Lie-Yamaguti algebras | [
"math.RA",
"math.RT"
] | On the set H_n(K) of symmetric n by n matrices over the field K we can define various binary and ternary products which endow it with the structure of a Jordan algebra or a Lie or Jordan triple system. All these non-associative structures have the orthogonal Lie algebra so(n,K) as derivation algebra. This gives an embedding of so(n,K) into so(N,K) for N = n(n+1)/2 - 1. We obtain a sequence of reductive pairs (so(N,K), so(n,K)) that provides a family of irreducible Lie-Yamaguti algebras. In this paper we explain in detail the construction of these Lie-Yamaguti algebras. In the cases n < 5, we use computer algebra to determine the polynomial identities of degree < 7; we also study the identities relating the bilinear Lie-Yamaguti product with the trilinear product obtained from the Jordan triple product. | math.RA | math |
SYMMETRIC MATRICES, ORTHOGONAL LIE ALGEBRAS,
AND LIE-YAMAGUTI ALGEBRAS
PILAR BENITO, MURRAY BREMNER, AND SARA MADARIAGA
Abstract. On the set Hn (K) of symmetric n × n matrices over the field K
we can define various binary and ternary products which endow it with the
structure of a Jordan algebra or a Lie or Jordan triple system. All these non-
associative structures have the orthogonal Lie algebra so(n, K) as derivation
algebra. This gives an embedding so(n, K) ⊂ so(N , K) for N = (cid:0)n+1
2 (cid:1) − 1.
We obtain a sequence of reductive pairs (so(N , K), so(n, K)) that provides a
family of irreducible Lie-Yamaguti algebras. In this paper we explain in detail
the construction of these Lie-Yamaguti algebras. In the cases n ≤ 4, we use
computer algebra to determine the polynomial identities of degree ≤ 6; we
also study the identities relating the bilinear Lie-Yamaguti product with the
trilinear product obtained from the Jordan triple product.
1. Introduction
In this paper we use the representation theory of Lie algebras and computer
algebra to study a family of Lie-Yamaguti algebras whose standard envelopes are
orthogonal simple Lie algebras. Our main goal is to determine the polynomial
identities satisfied by these structures.
(LY2)
(LY3)
Definition 1.1. [17, 18, 26] A Lie-Yamaguti algebra (LY algebra for short) is a
vector space m with a bilinear product m × m → m, (x, y ) 7→ x · y , and a trilinear
product m × m × m → m, (x, y , z ) 7→ hx, y , z i, satisfying:
x · x ≡ 0
(LY1)
hx, x, y i ≡ 0
(cid:9)x,y ,z (hx, y , z i + (x · y ) · z ) ≡ 0
(cid:9)x,y ,z hx · y , z , ti ≡ 0
hx, y , u · vi ≡ hx, y , ui · v + u · hx, y , vi
hx, y , hu, v , wii ≡ hhx, y , ui, v , wi + hu, hx, y , vi, wi + hu, v , hx, y , wii
(LY6)
Here (cid:9)x,y ,z means the cyclic sum on x, y , z , and ≡ indicates that the equation holds
for all values of the variables.
(LY5)
(LY4)
LY algebras with trivial ternary product are Lie algebras. LY algebras with
trivial binary product are Lie triple systems (LTS), which appear in the study of
the symmetric spaces. They are the odd parts of Z2 -graded Lie algebras which
2010 Mathematics Subject Classification. Primary 17A30. Secondary 17A40, 17A60, 17B10,
17B60, 17C50.
Key words and phrases. Symmetric matrices, orthogonal Lie algebras, Lie-Yamaguti algebras,
representation theory, polynomial identities, computer algebra, Lie-Jordan-Yamaguti algebras.
1
2
BENITO, BREMNER, AND MADARIAGA
are simple as graded algebras, and were classified in [20] using involutive auto-
morphisms; see [21] for an approach using representation theory. A simple LTS is
either irreducible or 2-irreducible as a module over its derivation Lie algebra; in
the 2-irreducible case, the components are dual modules. For general simple LY
algebras, the structure is more complicated [2]. One goal of the present paper is to
develop interesting examples to clarify the general theory.
Many examples of LY algebras are provided by reductive pairs [23] which are
closely related to reductive homogeneous spaces; in particular, LY algebras related
to spheres have been studied in [10, 11]. The LY algebras whose standard envelope
is a Lie algebra of type G2 are studied in [2]; several new examples of binary-
ternary products are given in terms of the octonions, and all nonabelian reductive
subalgebras of G2 are classified up to conjugation over a field of characteristic 0. In
that paper, the associated reductive pairs and LY algebras were described in detail,
using the representation theory of the split 3-dimensional simple Lie algebra in the
spirit of [7].
The paper is organized as follows. Section 2 gives some basic definitions and
constructions. Section 3 introduces the reductive pairs related to orthogonal Lie al-
gebras, and the corresponding LY algebras. Section 4 provides some combinatorial
and algorithmic background on polynomial identities. Section 5 presents our com-
putational results for the LY algebra obtained from the embedding so(4) ⊂ so(9).
Section 6 studies the binary-ternary structure obtained by replacing the trilinear
LY operation with the Jordan triple product. Section 7 gives some suggestions for
further research.
2. Preliminaries
Following [3, 22], we note that two elements x, y in an LY algebra m define a
linear map dx,y : m → m, z 7→ hx, y , z i, which by (LY5-6) is a derivation of both the
binary and ternary products.
Definition 2.1. The linear maps dx,y are called inner derivations. We write
D(m) ⊆ End(m) for their linear span.
Remark 2.2. By (LY6), the space D(m) is closed under the Lie bracket.
Definition 2.3. The standard envelope of m is the direct sum g(m) = D(m) ⊕ m
endowed with the structure of a Lie algebra by
[D(x, y ), D(z , t)] = D([x, y , z ], t) + D(z , [x, y , t]),
[z , t] = D(z , t) + z · t.
Remark 2.4. D(m) is a subalgebra of g(m), and m is an D(m)-module.
[D(x, y ), z ] = D(x, y )(z ) = [x, y , z ],
Definition 2.5. Let g be a Lie algebra with product [x, y ]. A reductive decomposi-
tion of g is a vector space direct sum g = h ⊕ m satisfying [h, h] ⊆ h and [h, m] ⊆ m.
In this case, we call (g, h) a reductive pair.
For a reductive decomposition g = h ⊕ m, there exist natural binary and ternary
products on m defined by
[x, y , z ] = (cid:2)πh ([x, y ]), z ],
x · y = πm (cid:0)[x, y ](cid:1),
where πh and πm are the pro jections on h and m. These products endow m with the
structure of an LY algebra, and any LY algebra has this form since ( g(m), D(m) )
is a reductive pair.
(1)
LIE-YAMAGUTI ALGEBRAS
3
Definition 2.6. The LY algebra structure (1) is the standard LY algebra given by
the reductive pair (g, h). We call this LY algebra irreducible if m is an irreducible
h-module, and k -irreducible if m is the direct sum of k irreducible h-modules.
The classification of isotropically irreducible reductive homogeneous spaces by
Wolf [24, 25] was reconsidered in [3, 4] using an algebraic approach with reductive
pairs related to nonassociative structures, in particular Lie and Jordan algebras and
triples. A summary of the close connections between nonassociative structures and
isotropically irreducible reductive homogeneous spaces is given in [4, Tables 7–9].
We emphasize [24, Ch. I, §11] the two sequences of homogeneous spaces
SO(2k2+k−1)/ SO(2k)
SO(2k2+3k)/ SO(2k+1)
(k ≥ 4).
(k ≥ 2),
If we study the corresponding Lie algebras, we can combine these two sequences
into a single sequence of orthogonal reductive pairs,
N = (cid:18)n+1
2 (cid:19) − 1.
n ≥ 3, n 6= 4,
We have included two additional cases, one from each sequence of homogeneous
spaces. For n = 3 (first sequence, k = 1) we obtain SO(5)/ SO(3), and for n = 6
(second sequence, k = 3) we obtain SO(20)/ SO(6). The case n = 2 (second
sequence, k = 1) is trivial. The case n = 4 (second sequence, k = 2) is a 2-
irreducible LY algebra related to the homogeneous space SO(9)/(SO(3) × SO(3)).
(See Dickinson and Kerr [8] for general results on compact homogeneous spaces
with two isotropy summands.)
All structures in this paper are finite-dimensional over an algebraically closed
field K of characteristic 0.
(so(N ), so(n))
3. LY algebras related to symmetric matrices
The vector space Hn (K) of symmetric n× n matrices can be endowed with binary
and ternary products giving a Jordan algebra, a Lie triple system or a Jordan triple
system. Each of these non-associative structures has the orthogonal Lie algebra
son (K) of type B or D as its Lie algebra of derivations Der Hn (K); this Lie algebra
is simple except for n = 2, 4. We denote by Hn (K)0 the subspace of matrices with
trace 0; this is the orthogonal complement of the 1-dimensional subspace spanned by
the identity matrix In with respect to the trace form T (a, b) = tr(ab). We denote the
dimension of Hn (K)0 by N = (cid:0)n+1
2 (cid:1)−1. Every derivation d ∈ Der Hn (K) annihilates
In and satisfies d(Hn (K)) ⊂ Hn (K)0 . Since the restriction of T to Hn (K)0 is
nondegenerate, Der Hn (K) embeds into the Lie algebra soN (K) of endomorphisms
of Hn (K)0 which are orthogonal with respect to T . For n ≥ 2 we obtain a sequence
of reductive pairs (so(Hn (K)0 , T ), Der Hn (K)) which is the Lie algebraic analogue
of the reductive homogeneous spaces mentioned in the Introduction as quotients of
orthogonal Lie groups. (Since n = 2 gives N = 2, we ignore this trivial case.) The
components of these reductive pairs are simple Lie algebras for n ≥ 3, n 6= 4; this
implies some preliminary structural results on the associated LY algebras (see [22,
Th. 2] and [3, Pr. 1.3]):
Theorem 3.1. If g is a simple Lie algebra and h ⊂ g is a semisimple subalgebra,
then (g, h) is a reductive pair with m = h⊥ with respect to the Kil ling form, and
either [m, m] = {0} or m is simple.
4
BENITO, BREMNER, AND MADARIAGA
[x, x] = 0,
Proposition 3.2. Let g = h ⊕ m be a reductive decomposition of a simple Lie
algebra with m 6= {0}. Then for the LY algebra structure on m defined by (1), we
have h ∼= D(m) (the inner derivation algebra) and g ∼= D(m) ⊕ m (the standard
envelope). Furthermore, if h is semisimple and m is an irreducible h-module, then
either h ∼= m as adh -modules, or m = h⊥ with respect to the Kil ling form of g.
In the rest of this section we explain different constructions of the reductive pairs
(cid:0)so(Hn (K)0 , T ), Der Hn (K)(cid:1) and the associated LY algebras using representation
theory of Lie algebras and structure theory of Jordan algebras.
3.1. LY algebras from the viewpoint of representation theory. Let h be a
simple Lie algebra and let m be an irreducible h-module. If the exterior square Λ2m
contains both m and the adjoint h-module, then there are h-module morphisms
β : Λ2m −→ m,
τ : Λ2m −→ h,
which give bilinear and trilinear products on m defined by
hx, y , z i = τ (x, y ) · z ,
[x, y ] = β (x, y ),
where · is the action of h on m. Since β and τ have domain Λ2m, these products
satisfy the skew-symmetries (LY1) and (LY2) in the definition of an LY algebra:
hx, x, y i = 0.
Identities (LY5) and (LY6) hold because β and τ are h-module morphisms, so h
acts as derivations of both products. Thus a Lie algebra structure can be defined
on h ⊕ m if and only if identities (LY3) and (LY4) are satisfied. If h ⊕ m cannot
be made into a Lie algebra, then these products can still be defined, but we will
obtain a different class of algebras closely related to LY algebras.
We can refine this approach by considering a simple Lie algebra h and a faithful
representation ρ : h → EndK (V ). We have the embeddings
h ∼= ρ(h) ⊆ sl(V ) ⊆ gl(V ) ∼= V ⊗ V ∗ ,
which product reductive pairs (sl(V ), h) and the associated LY algebras.
If the
h-module V has an h-invariant symmetric or skew-symmetric bilinear form then
h embeds into either so(V ) or sp(V ), and we obtain reductive pairs (so(V ), h) or
(sp(V ), h). We will use this approach to understand the structure of the LY algebras
associated to the reductive pairs (soN (K), son (K)).
We recall basic facts about orthogonal Lie algebras. Let V be an n-dimensional
vector space (n ≥ 2) over K with a nondegenerate symmetric bilinear form ϕ. Fixing
an orthonormal basis of V , we may assume V = Kn and ϕ(x, y ) = Pn
i=1 xi yi . If
n = 2 then so2 (K) ∼= K is the 1-dimensional abelian Lie algebra. If n = 3 then
so3 (K) ∼= sl2 (K) (that is, B1 = A1 ), the 3-dimensional simple Lie algebra (recall
that K is algebraically closed). If n = 4 then so4 (K) ∼= so3 (K) ⊕ so3 (K) (that is,
D2 = A1 ∪ A1 ) is the direct sum of two simple ideals. If n = 5 then so5 (K) ∼= sp4 (K)
(that is, B2 = C2 ) can also be defined by a nondegenerate skew-symmetric bilinear
form in 4 dimensions. If n = 6 then so6 (K) ∼= sl4 (K) (that is, D3 = A3 ), the 4 × 4
matrices of trace 0. For n ≥ 7, the orthogonal Lie algebras are simple of types Bk
(n = 2k + 1) or Dk (n = 2k).
We write V (λ) for the simple son (K)-module of highest weight λ; we use the
labelling of the fundamental weights given by [16]. In the non-simple case so4 (K) ∼=
so3 (K)⊕so3 (K), we use λ and λ′ for the weights for the first and second components.
We record the structure of the adjoint and natural son (K)-modules in Table 1.
LIE-YAMAGUTI ALGEBRAS
5
Natural: dimension n
n = 5
son (K) Adjoint: dimension (cid:0)n
2 (cid:1)
V (2λ1 )
n = 3
V (2λ1 )
(cid:0)V (2λ1 ) ⊗ K(cid:1) ⊕ (cid:0)K ⊗ V (2λ′1 )(cid:1) V (λ1 ) ⊗ V (λ′1 )
n = 4
V (2λ2 )
V (λ1 )
V (λ1 )
V (λ2 + λ3 )
V (λ2 )
V (λ1 )
n = 6
n ≥ 7
Table 1. Representations of orthogonal Lie algebras
Lemma 3.3. For n ≥ 3, the symmetric square S 2 Kn of the natural son (K)-module
has dimension (cid:0)n+1
2 (cid:1) and decomposes as fol lows:
S 2 K3 = V (4λ1 ) ⊕ V (0)
n = 3 :
S 2 K4 = (cid:0)V (2λ1 ) ⊗ V (2λ′1 )(cid:1) ⊕ (cid:0)V (0) ⊗ V (0)(cid:1)
n = 4 :
S 2 Kn = V (2λ1 ) ⊕ V (0)
n ≥ 5 :
Proof. We consider the tensor square of the natural module,
Kn ⊗ Kn = S 2 Kn ⊕ Λ2 Kn .
Since Kn has an invariant nondegenerate symmetric bilinear form, V (0) is a sub-
module of S 2 Kn . For n 6= 4, the stated decompositions then follow from [1, Th. 5]
and a dimension count. For n = 4, since so4 (K) = so3 (K) ⊕ so3 (K), the simple
so4 (K)-modules are tensor products V ⊗ V ′ of simple so3 (K)-modules. Denoting
by λ and λ′ the fundamental weights of the two copies of so3 (K) inside so4 (K), the
result follows from the calculation given in detail in Example 3.4 below.
(cid:3)
For n ≥ 3 we write Qn for the quotient of the son (K)-module S 2 Kn by its trivial
1-dimensional submodule; we have dim Qn = N .
Lemma 3.4. For n ≥ 3, the exterior square Λ2 Qn has dimension (cid:0)N
2 (cid:1) and decom-
poses as fol lows:
Λ2 Q3 = V (2λ1 ) ⊕ V (6λ1 )
n = 3 :
Λ2 Q4 = (cid:0)V (2λ1 ) ⊗ K(cid:1) ⊕ (cid:0)V (2λ1 ) ⊗ V (4λ′1 )(cid:1) ⊕
(cid:0)K ⊗ V (2λ′1 )(cid:1) ⊕ (cid:0)V (4λ1 ) ⊗ V (2λ′1 )(cid:1)
Λ2 Q5 = V (2λ2 ) ⊕ V (2λ1 + 2λ3 )
Λ2 Q6 = V (λ2 + λ3 ) ⊕ V (2λ1 + λ2 + λ3 )
n = 6 :
Λ2 Qn = V (λ2 ) ⊕ V (2λ1 + λ2 )
n ≥ 7 :
Proof. Similar to the proof of Lemma 3.3.
n = 5 :
n = 4 :
(cid:3)
The decompositions in Lemma 3.4 have a natural multilinear interpretation. The
orthogonal Lie algebra son (K) consists of the n × n skew-symmetric matrices (with
respect to the symmetric bilinear form given by the identity matrix In ). The space
of all n × n matrices decomposes as the direct sum of symmetric matrices and
skew-symmetric matrices:
Mn (K) = son (K) ⊕ Hn (K) = son (K) ⊕ (cid:0)Hn (K)0 ⊕ KIn (cid:1).
(2)
6
BENITO, BREMNER, AND MADARIAGA
Using the commutator, Mn(K) becomes a Z2 -graded Lie algebra with the simple
Lie algebra son (K) as the even part and Hn (K) as the odd part. Using the anti-
commutator, Mn(K) becomes a Z2 -graded Jordan algebra with the simple Jordan
algebra Hn (K) as the even part and son (K) as the odd part. The generic trace
of the Jordan algebra Hn (K) is non-degenerate when restricted to the subspace
Hn (K)0 which has dimension N ; the skew-symmetric N × N matrices with respect
to this trace form define the orthogonal Lie algebra so(N , K). The Jordan algebra
Hn (K) is a realization of S 2 Kn , and Qn = Hn (K)0 is then a realization of the
n × n symmetric matrices with trace 0. In this way, the LY algebras associated
to this sequence of orthogonal reductive pairs are seen to be closely related to Lie
and Jordan algebras. From the classification of irreducible LY algebras [4], we see
that for n ≥ 7 there is only one way to define binary and ternary products on the
son (K)-module V (2λ1 + λ2 ) to obtain the structure of an LY algebra.
Lie bracket
Jordan product
(3)
3.2. From Jordan algebras to LY algebras. Recall that the field K is assumed
to be algebraically closed. We generalize the Z2 -grading of Mn (K) = son (K) ⊕
Hn (K) given by the trace form tr(a) which is symmetric and non-degenerate. We
consider an n-dimensional vector space V with a nondegenerate symmetric bilinear
form ϕ. Using ϕ we may identify V and its dual V ∗ by f ↔ f ∗ where f ∗ is defined
as usual by ϕ(f ∗ (x), y ) = ϕ(x, f (y )). We have EndK (V ) ∼= V ∗ ⊗K V ; interchanging
the tensor factors gives V ⊗K V ∗ , and using ϕ we can identify this with EndK (V ).
This gives us an involution on the Lie algebra gl(V ), which is EndK (V ) using the
commutator, and from this we obtain the decomposition
gl(V ) = Skew(V , ϕ) ⊕ Sym(V , ϕ)
where Sym(V , ϕ) and Skew(V , ϕ) are respectively the symmetric and skew-symmetric
endomorphisms defined by f ∗ = f and f ∗ = −f . If we fix an orthonormal basis of
V with respect to ϕ we obtain the matrix decomposition (2).
We consider the following binary and ternary products on Hn (K):
[a, b] = ab − ba
a ◦ b = ab + ba
ha, b, ci = (b, c, a)◦
{a, b, c} = abc + cba
Jordan triple product
where (b, c, a)◦ = (a ◦ b) ◦ c − a ◦ (b ◦ c) is the Jordan associator (cyclically permuted).
We recall some basic results, proofs of which may be found in [21]. The subspace
Hn (K) is a simple Jordan algebra using the ◦ product. The generic trace of this
Jordan algebra is the matrix trace tr(a), and T (a, b) = tr(ab) is a non-degenerate
bilinear form, which is associative in the sense that T (a ◦ b, c) = T (a, b ◦ c). The
subspace Hn (K) is a simple Jordan triple system using the brace product { , , }.
The linear map da,b ∈ End(Hn (K)) defined by da,b (c) = (b, c, a)◦ = [La , Lb ](c)
where La(b) = a ◦ b is a derivation of both the Jordan product and the Jordan triple
product; linear combinations of the maps da,b are called inner derivations. The full
derivation algebras of Hn (K) with respect to both products coincide and so we may
write Der Hn (K). Every derivation is an inner derivation: we have
Der Hn (K) = span( [La , Lb ] a, b ∈ Hn (K) ) = { ada a ∈ Mn (K), at = −a },
where ada (b) = [a, b]. In fact, as Lie algebras we have Der Hn (K) ∼= son (K), which
is simple for n 6= 4.
Lie triple product
LIE-YAMAGUTI ALGEBRAS
7
For any d ∈ Der Hn (K) we have T (d(a), b)+ T (a, d(b)) = 0; thus the trace form T
is invariant under derivations. The direct sum Hn (K) = Hn (K)0 ⊕KIn is orthogonal
with respect to T . If d ∈ Der Hn (K) then d(In ) = 0 and d(Hn (K)) ⊆ Hn (K)0 ; hence
Hn (K)0 is a Der Hn (K)-module isomorphic to V (2λ1 ) as an son (K)-module.
We have (b, c, a)◦ = [[a, b], c], and therefore [[La , Lb ], Lc ] = L(b,c,a)◦ The subspace
Hn (K) is a Lie triple system with respect to the product h , , i. The decomposition
Hn (K) = Hn (K)0 ⊕ KIn is a direct sum of ideals, and Hn (K)0 is a simple Lie triple
system. For any a, b ∈ Hn (K)0 , the linear map da,b = (b, −, a)◦ is a derivation of the
Lie triple product on Hn (K); the full derivation algebra coincides with Der Hn (K).
This derivation algebra is a subalgebra of so(Hn (K)0 , T ) ∼= soN (K), the Lie algebra
of endomorphisms d of Hn (K)0 which are orthogonal with respect to T , in the sense
that T (d(a), b) + T (a, d(b)) = 0 for all a, b ∈ Hn (K)0 . The pair
(cid:0) so(Hn (K)0 , T ), Der Hn (K) (cid:1) = (cid:0) soN (K), son (K) (cid:1)
is a reductive pair and the orthogonal complement LYn = (son (K))⊥ with respect
to the Killing form of soN (K) is the unique Der Hn (K)-invariant complement to
Der Hn (K) in so(Hn (K)0 , T ). Hence, LYn is the unique standard LY algebra asso-
ciated to the reductive pair (so(Hn (K)0 , T ), Der Hn (K)).
For n ≥ 3, the LY algebra LYn is simple with nontrivial binary and ternary
products, and has dimension
2(cid:19) = (cid:18)(cid:0)n+1
2 (cid:1) − 1
dim LYn = (cid:18)N
2 (cid:19) − (cid:18)n
(cid:19) − (cid:18)n
2(cid:19) =
1
(n − 1)(n + 1)(n − 2)(n + 4).
8
2
The derivation algebra Der LYn is isomorphic to son (K). For n 6= 4, LYn is irre-
ducible as a module over its derivation algebra; the highest weight of this module
is given in Lemma 3.4. For n = 4, the same lemma shows that
LY4 = U ⊕ U ′ = (cid:0)V (2λ1 ) ⊗ V (4λ′1 )(cid:1) ⊕ (cid:0)V (4λ1 ) ⊗ V (2λ′1 )(cid:1).
Thus LY4 is a 2-irreducible module over its derivation algebra and each summand is
15-dimensional. We now consider some other constructions of LY3 and LY4 which
have appeared in the literature.
(4)
3.3. Three different approaches to LY3 . In 1984, Dixmier [7] constructed some
simple nonassociative algebras using the 3-dimensional Lie algebra sl2 (K), its irre-
ducible modules, and transvections from classical invariant theory. In particular,
we recall from [7, §6.2] the 3-parameter family of simple Lie algebras of type B2
given by a reductive decomposition:
10λµ = 9ν 2 .
(5)
B3,λ,µ,ν = V (2λ1 ) ⊕ V (6λ1 ),
Since V (2λ1 ) is the adjoint module of sl2 (K) ∼= so3 (K) and B2 ∼= so5 (K), we obtain
(6)
g = so5 (K) = sl2 (K) ⊕ LY3 = h ⊕ m.
This leads to the following explicit construction of LY3 . For each k ≥ 0, we consider
the (k+1)-dimensional vector space
P (k) = span(xk−i y i 0 ≤ i ≤ k),
of homogeneous polynomials in x, y of degree n. We can identify P (k) with the
sl2 (K)-module V (kλ1 ) if we define the action by the differential operators,
δ
δ
δ
δ
δx − y
δx
δy
δy
h = x
f = y
.
,
e = x
,
8
BENITO, BREMNER, AND MADARIAGA
The following equation defines a Lie algebra structure on B3,6,15,10 :
[ℓ1 + m1 , ℓ2 + m2 ] = 6(ℓ1 ℓ2)1 + 15(m1m2 )5 + 6(ℓ1m2 )1 + 6(m1 ℓ2 )1 + 10(m1m2 )3 ,
where
(ℓ1 ℓ2 )1 =
(ℓ1m2 )1 =
(m1m2 )5 =
δx (cid:19) ,
4 (cid:18) δℓ1
1
δℓ2
δℓ2
δℓ1
δy −
δy
δx
δx (cid:19) ,
12 (cid:18) δℓ1
δm2
δm2
δℓ1
1
δy −
δy
δx
5
δ5m2
i (cid:19) δ5m1
(−1)i(cid:18)5
Xi=0
δxi δy 5−i
δx5−i δy i
3
i (cid:19) δ3m1
(−1)i(cid:18)3
Xi=0
δx3−i δy i
For ℓ1 = ℓ2 = 0 we obtain the Lie bracket of two elements in LY3 :
δ3m2
δxi δy 3−i
(m1m2 )3 =
(m1 ℓ2 )1 = −(ℓ2m1 )1
[m1 , m2 ] = 15(m1m2 )5 + 10(m1m2 )3 .
Since (m1m2 )5 ∈ h and (m1m2 )3 ∈ LY3 , we apply equations (1) to obtain the
binary and ternary products on LY3 :
{m1 , m2 , m3} = 90((m1m2 )5m3 )1 .
m1 · m2 = 10(m1m2 )3 ,
For a second approach using the representation theory of sl2 (K), see [5]. The
binary and ternary products of LY3 were obtained using the decomposition
Λ2V (6λ1 ) = V (2λ1 ) ⊕ V (6λ1 ) ⊕ V (10λ1 ).
The binary product is given (up to a scalar) by the pro jection Λ2V (6λ1 ) → V (6λ1 );
the ternary product is the composition of Λ2V (6λ1 ) → V (2λ1 ) ∼= sl2 (K) with the
action of sl2 (K) on V (6λ1 ). Using computer algebra, the polynomial identities in
low degree relating these products were studied.
The third approach to LY3 is given by the isomorphism so5 (K) ∼= sp4 (K). Fol-
lowing [4], we have the reductive decomposition:
sp4 (K) = Der TK ⊕ LY3 ,
where TK is the symplectic triple system associated to the Jordan algebra of a
nondegenerate cubic form with basepoint given by the norm n(α) = α3 on the
base field K. The triple TK can be identified with the vector space M2(K) of 2 × 2
matrices; for definitions and formulas for the triple product see [9, Th. 2.21] where a
survey of these triples is also included. Symplectic triple systems are closely related
to Freudenthal triple systems, Lie triple systems and 5-graded Lie algebras.
3.4. The 2-irreducible LY algebra LY4 . Using multilinear algebra we can ob-
tain a convenient model of this LY algebra. Let V1 = V2 = K3 and let ϕi be a
nondegenerate symmetric bilinear form on Vi . We define
Sym0 (Vi , ϕi ) = { f ∈ EndK (Vi ) ϕi (f (x), y ) = ϕi (x, f (y )), tr(f ) = 0 }.
On V1 ⊗ V2 we have the nondegenerate symmetric bilinear form ϕ = ϕ1 ⊗ ϕ2 defined
on simple tensors by ϕ(x1 ⊗ x2 , y1 ⊗ y2 ) = ϕ1 (x1 , y1 )ϕ2 (x2 , y2 ). Equation (3) gives
a decomposition of the Lie algebra gl(Vi ):
gl(Vi ) = so(Vi , ϕi ) ⊕ Sym0 (Vi , ϕi ) ⊕ K IdVi
(i = 1, 2).
LIE-YAMAGUTI ALGEBRAS
9
As vector spaces, gl(V1 ) ⊗ gl(V2 ) ∼= gl(V1 ⊗ V2 ). On this space we define a Lie
bracket by
[a ⊗ b, f ⊗ g ] = ab ⊗ f g − ba ⊗ gf ,
where ab and f g denote compositions of linear maps, and we obtain the Lie algebra
decomposition:
gl(V1 ⊗ V2 ) = so(V1 ⊗ V2 , ϕ) ⊕ Sym(V1 ⊗ V2 , ϕ).
From the last two displayed equations we get the reductive decomposition:
so(V1 ⊗ V2 , ϕ) ∼= (cid:0) so(V1 , ϕ1 ) ⊗ K IdV2 (cid:1) ⊕ (cid:0) K IdV1 ⊗ so(V2 , ϕ2 ) (cid:1)
⊕ (cid:0) so(V1 , ϕ1 ) ⊗ Sym0 (V2 , ϕ2 ) (cid:1) ⊕ (cid:0) Sym0 (V1 , ϕ1 ) ⊗ so(V2 , ϕ2 ) (cid:1)
∼= (cid:0) so3 (K) ⊕ so3 (K) (cid:1) ⊕ LY4
∼= so9 (K),
which corresponds to the reductive pair (so9 (K), so3 (K) ⊕ so3 (K)). In this way, we
obtain the 2-irreducible simple LY algebra LY4 = U ⊕ U ′ of equation (4). Using
the Lie bracket defined above, we obtain
3 tr(st)I3 ) + [a, b] ⊗ 1
2 [a, b] ⊗ (cid:0)st + ts − 2
[a ⊗ s, b ⊗ t] = 1
3 tr(st)I3
3 tr(ab)I3 (cid:1) ⊗ [s, t] + 1
2 (cid:0)ab + ba − 2
+ 1
3 tr(ab)I3 ⊗ [s, t],
3 tr(a′ b′ )I3 (cid:1) + [s′ , t′ ] ⊗ 1
2 [s′ , t′ ] ⊗ (cid:0)a′ b′ + b′a′ − 2
[s′ ⊗ a′ , t′ ⊗ b′ ] = 1
3 tr(a′ b′ )I3
+ 1
3 tr(s′ t′ )I3 (cid:1) ⊗ [a′ , b′ ] + 1
2 (cid:0)s′ t′ + t′s′ − 2
3 tr(s′ t′ )I3 ⊗ [a′ , b′ ],
2 (as′ + s′a) ⊗ [s, a′ ] + 1
[a ⊗ s, s′ ⊗ a′ ] = 1
2 [a, s′ ] ⊗ (sa′ + a′s),
where
for the summand U ,
for the summand U ′ .
a, b ∈ so(V1 , ϕ1 ),
s, t ∈ Sym0 (V2 , ϕ2 ),
s′ , t′ ∈ Sym0 (V1 , ϕ1 ),
a′ , b′ ∈ so(V2 , ϕ2 ),
The binary product on LY4 is given by the following equations:
(a ⊗ s) · (b ⊗ t) = 1
2 (ab + ba − 2
2 [a, b] ⊗ (st + ts − 2
3 tr(st)I3 ) + 1
3 tr(ab)I3 ) ⊗ [s, t],
(a ⊗ s) · (s′ ⊗ a′ ) = 1
2 (as′ + s′a) ⊗ [s, a′ ] + 1
2 [a, s′ ] ⊗ (sa′ + a′s),
2 [s′ , t′ ] ⊗ (a′ b′ + b′a′ − 2
(s′ ⊗ a′ ) · (t′ ⊗ b′ ) = 1
3 tr(a′ b′ )I3 )
+ 1
2 (s′ t′ + t′s′ − 2
3 tr(s′ t′ )I3 ) ⊗ [a′ , b′ ].
The ternary product is given by the following equations:
3 tr(st)[[a, b], c] ⊗ u + 1
{a ⊗ s, b ⊗ t, c ⊗ u} = 1
3 tr(ab)c ⊗ [[s, t], u],
{a ⊗ s, b ⊗ t, s′ ⊗ a′} = 1
3 tr(st)[[a, b], s′ ] ⊗ a′ + 1
3 tr(ab)s′ ⊗ [[s, t], a′ ],
{a ⊗ s, s′ ⊗ a′ , b ⊗ t} = 0,
{a ⊗ s, s′ ⊗ a′ , t′ ⊗ b′} = 0,
{s′ ⊗ a′ , t′ ⊗ b′ , a ⊗ s} = 1
3 tr(a′ b′ )[[s′ , t′ ], a] ⊗ s + 1
3 tr(s′ t′ )a ⊗ [[a′ , b′ ], s],
3 tr(a′ b′ )[[s′ , t′ ], u′ ] ⊗ c′ + 1
{s′ ⊗ a′ , t′ ⊗ b′ , u′ ⊗ c′} = 1
3 tr(s′ t′ )u′ ⊗ [[a′ , b′ ], c′ ].
10
BENITO, BREMNER, AND MADARIAGA
3.5. The reductive decomposition so (cid:0)(cid:0)n+1
2 (cid:1), R(cid:1) = so(n, R) ⊕ M . The Lie alge-
bra L = so(n, R) consists of the skew-symmetric n × n matrices A with real entries;
that is, At = −A. We have dim L = (cid:0)n
2 (cid:1); we consider the standard basis for L
consisting of the matrices Eij − Ej i for 1 ≤ i < j ≤ n.
We write Rn for the natural n-dimensional L-module with action given by A ·X =
AX for A ∈ L and X ∈ Rn , the usual product of a matrix and a column vector.
We identify the symmetric square of this natural module, S 2Rn , with the vector
space H = Hn (R) of symmetric n × n matrices. This becomes an L-module with
the action given by the Lie bracket of matrices
A · S = [A, S ] = AS − SA for A ∈ L, S ∈ H.
(This is the symmetrization of the left action of L on H , since AS+S tAt = AS−SA.)
We have gl(n, R) = L ⊕H as vector spaces. We write H 0 ⊂ H for the subspace of
symmetric matrices with trace zero, and so sl(n, R) = L ⊕ H 0 . This is a reductive
decomposition of sl(n, R): L is a Lie subalgebra, [L, L] ⊆ L, and H 0 is an L-
module for the restriction of the adjoint representation, [L, H 0 ] ⊆ H 0 . We have
that dim H = (cid:0)n+1
2 (cid:1) and so dim H 0 = (cid:0)n+1
2 (cid:1) − 1. We denote N = (cid:0)n+1
2 (cid:1).
Thus there is an embedding of so(n) into so(N ). We require an orthonormal basis
of H so that the N × N matrices representing the action of L ⊂ so(N ) on H are
skew-symmetric. Since sl(n, R) is a simple Lie algebra, its Killing form is a (nonzero)
scalar multiple of the trace form on n × n matrices: T r(A, B ) = trace(AB ). We
choose as standard basis for H the diagonal matrix units Eii for 1 ≤ i ≤ n together
1√2
1√2
are required
(Eij + Eij ) for 1 ≤ i < j ≤ n. The scalars
with the matrices
to make this basis orthonormal with respect to the trace form. So the action of
L ⊂ so(N ) on H is given as follows:
(Eij − Ej i ) · Ekk = δjk (Eik + Eki ) − δik (Ejk + Ekj ),
1√2
1√2
(Eij − Ej i ) · 1√2
(Eiℓ + Eℓi ) + δjℓ
(Ekℓ + Eℓk ) = δjk
(Eik + Eki )
1√2
1√2
(Ejℓ + Eℓj ) − δiℓ
− δik
The 1-dimensional subspace of H spanned by the identity matrix I is annihilated
by the action of so(n) so it also has to be annihilated by the action of its image in
so(N ). The subalgebra U ⊂ so(N ) which annihilates I is isomorphic to so(N −1),
so we have L = so(n) ⊂ so(N −1) = U . We have a reductive decomposition
U = L ⊕ M , where M = L⊥ , the orthogonal complement of L with respect to the
trace form on U . So we have [L, M ] ⊆ M . We provide M with the structure of a Lie-
Yamaguti algebra using the Lie bracket [A, B ]U in U together with the pro jections
pL and pM onto L and M . The bilinear operation [A, B ] and the trilinear operation
hA, B , C i are defined by
(7)
[A, B ] = pM ( [A, B ]U ),
hA, B , C i = pM ( [ pL ([A, B ]U ), C ]U ).
The dimension of this Lie-Yamaguti algebra is
2(cid:19) = (cid:18)(cid:0)n+1
2 (cid:1)−1
dim M = (cid:18)N −1
2 (cid:19) − (cid:18)n
(cid:19) − (cid:18)n
2(cid:19) =
(n−2)(n−1)(n+1)(n+4).
2
For n = 3, . . . , 9 we obtain dim M = 7, 30, 81, 175, 330, 567, 910.
(Ejk + Ekj ).
1
8
Lemma 3.5. For n ≥ 3 and n 6= 4 the L-module M is irreducible.
LIE-YAMAGUTI ALGEBRAS
11
4. Nonassociative polynomial identities
Before explaining our computational methods to find the polynomial identities
satisfied by the LY algebra M described below we need to introduce some of the
concepts involved. (For further details, see [5].)
4.1. Nonassociative polynomials. Let Ω be a set of multilinear operations de-
fined on a vector space A. A polynomial identity of degree d for the algebra A is
an element f ∈ F {Ω; X } of the free multioperator algebra with operations Ω and
generators X such that f (a1 , . . . , ad ) = 0 for all a1 , . . . , ad ∈ A. We denote it by
f ≡ 0. Over a field K of characteristic 0, every polynomial identity is equivalent
to a finite set of multilinear identities [27]. Given a set of operations Ω, the set of
association types of degree d for Ω is the set of all possible combinations of oper-
ations from Ω involving d variables. In our case, Ω = {[−, −], (−, −, −)} since LY
algebras have a binary and a ternary operation; in the rest of the paper we use
this notation, [a, b] instead of a · b, and (a, b, c) instead of ha, b, ci. Thus we can
consider the association types for [−, −], the association types for (−, −, −) and
the association types for [−, −] and (−, −, −). Owing to the skew-symmetry of the
operations in LY algebras, there are association types which are linearly dependent,
like [[−, −], −] and [−, [−, −]]. The next examples give the independent association
types in low degrees for [−, −] and (−, −, −) both separately and mixed.
Example 4.1. Binary types: association types in degrees d ≤ 7 for a skew-
symmetric binary operation (the commas within monomials are omitted):
d = 1 : − d = 2 : [−−] d = 3 : [[−−]−] d = 4 : [[[−−]−]−], [[−−][−−]]
d = 5 : [[[[−−]−]−]−]
[[[−−][−−]]−]
[[[−−]−][−−]]
[[[[−−]−][−−]]−]
[[[[−−][−−]]−]−]
d = 6 : [[[[−−]−]−]−]−]
[[[−−]−][[−−]−]]
[[[−−][−−]][−−]]
[[[[−−]−]−][−−]]
d = 7 : [[[[[[−−]−]−]−]−]−]
[[[[[−−][−−]]−]−]−]
[[[[[−−]−][−−]]−]−]
[[[[−−]−][[−−]−]]−]
[[[[−−][−−]][−−]]−]
[[[[[−−]−]−][−−]]−]
[[[[−−][−−]]−][−−]]
[[[[[−−]−]−]−][−−]]
[[[[−−]−][−−]][−−]]
[[[[−−]−]−][[−−]−]]
[[[−−][−−]][[−−]−]]
Example 4.2. Ternary types: association types in degrees d ≤ 7 for a ternary
operation skew-symmetric in its first two arguments (commas omitted):
d = 1 : −
d = 3 : (− − −)
d = 5 : ((− − −) − −),
(− − (− − −))
((− − −)(− − −)−)
((− − (− − −)) − −)
d = 7 : (((− − −) − −) − −)
((− − −) − (− − −))
(− − ((− − −) − −))
(− − (− − (− − −)))
Example 4.3. Mixed types: association types in degrees d ≤ 5 including those
involving both operations; the types from examples 4.1 and 4.2 also appear:
d = 3 : [[−−]−],
d = 2 : [−−]
(− − −)
d = 1 : −
([−−] − −)
[[−−][−−]]
[(− − −)−]
d = 4 : [[[−−]−]−]
(− − [−−])
[([−−] − −)−]
d = 5 : [[[[[−−]−]−]−]
[[(− − −)−]−]
[[[−−][−−]]−]
([[−−]−] − −)
[(− − −)[−−]]
[[[−−]−][−−]]
[(− − [−−])−]
12
BENITO, BREMNER, AND MADARIAGA
((− − −) − −)
(− − (− − −))
For degrees d = 6, 7 there are respectively 38 and 113 independent types.
([−−] − [−−])
([−−][−−]−)
(− − [[−−]−])
Given a set of operations Ω, the (multilinear) monomials of degree d are ob-
tained by applying all possible permutations of the d variables to the associa-
tion types of degree d. Since the LY operations have skew-symmetries, there
are monomials which are linearly dependent, such as [[x1 , x2 ], x3 ] and [[x2 , x1 ], x3 ],
or (x1 , x2 , [x3 , x4 ]) and (x2 , x1 , [x4 , x3 ]). Considering only independent types and
monomials reduces the memory needed for the computations. We consider the
equivalence classes of monomials determined by the skew-symmetries; each equiv-
alence class is represented by its normal form as defined in [5].
A multilinear polynomial f ∈ F {Ω; X } of degree d has consequences in higher
degrees, which are obtained using the operations ω ∈ Ω. To lift f using an operation
ω of arity r we need to add r − 1 new variables so the degree of the corresponding
liftings is d + r − 1. We obtain d liftings by replacing xi by ω (xi , xd+1 , · · · , xd+r−1 )
and another r liftings by making f (x1 , · · · , xd ) the j -th argument of ω . To lift a
given polynomial to a higher degree D we need to consider all the possible combi-
nations of liftings by operations from ω which increase degree d to degree D . The
liftings of f to degree D form a (not necessarily independent) set of generators for
the SD -module of all consequences of f in degree D .
For example, in the case Ω = {[−, −], (−, −, −)}, to lift a polynomial from degree
2 to degree 4 we can use the binary operation twice or the ternary operation once.
Owing to the skew-symmetries of the operations in a LY algebra some of the liftings
are redundant.
For each degree d the sets of binary, ternary and mixed monomials in normal
form are bases of the corresponding spaces of polynomials. These spaces are Sd -
modules, the action given by permuting the subscripts of the variables. We are
interested in finding module generators for the subspaces of multilinear identities
of the LY algebra M in each degree to determine whether they can be deduced
from the defining identities for LY algebras.
4.2. Computational methods. For each degree d, we perform the following steps:
(1) Generate the bases of binary, ternary and mixed monomials in normal form.
(2) Find the Sd -modules of all identities of degree d satisfied by M , considering
separately binary, ternary and mixed monomials.
(3) Obtain the corresponding Sd -submodules of lifted identities for M from
degrees < d.
(4) Compute the quotient modules (all identities modulo lifted identities) and
(if they are not trivial) find their minimal sets of generators; these genera-
tors are the new identities for M in degree d.
(5) Compare the generators for the quotient module of new identities with the
LY identities of degree d (the defining identities for LY algebras).
We use two algorithms: “fill and reduce” determines a basis for the vector space of
(binary, ternary or mixed) polynomial identities in degree d satisfied by M ; “module
generators” applies the representation theory of the symmetric group Sd to extract
a subset of the basis which generates the space of identities as an Sd -module. A
modification of “fill and reduce” also takes into account the liftings to degree d of
the known identities in lower degrees.
LIE-YAMAGUTI ALGEBRAS
13
4.2.1. Fil l and reduce. Let m be the dimension of M , and let q be the number of
normal (binary, ternary or mixed) monomials of degree d. We construct a zero
matrix E of size (q + m) × q whose columns are labeled by the normal monomials.
We divide it into an upper block of size q × q and a lower block of size m × q
whose rows are labeled by the elements of the chosen basis of M . We perform the
following steps until the rank of E has stabilized; that is, the rank has not increased
for some fixed number s of iterations:
(1) Generate d pseudorandom column vectors a1 , · · · , ad of dimension m, rep-
resenting elements of M with respect to the given basis.
(2) For each j = 1, · · · , q , evaluate the j -th monomial on the elements a1 , · · · , ad
using the LY operations [−, −] and (−, −, −) and store the coordinates of
the result in rows q+1 to q+m of column j . After this, each row of the
lower block of E represents a linear relation which must be satisfied by the
coefficients of any identity satisfied by M .
(3) Compute the row canonical form of E ; since it has size (q + m) × q , its rank
must be ≤ q , so the lower block of E becomes zero again.
After the rank has stabilized, the nullspace of E consists of the coefficient vectors of
the linear dependence relations between the monomials which are satisfied by many
pseudorandom choices of elements of M . We extract a basis of this nullspace, and
compute the row canonical form of the matrix whose rows are these basis vectors;
the rows of the reduced matrix represent nonassociative polynomials which are
“probably” polynomial identities satisfied by M . They still need to be proven
directly, or at least checked by another independent computation.
The construction of LY algebras by reductive pairs is defined over a field of
characteristic zero. To optimize computer time and memory we often use modular
instead of rational arithmetic, especially in higher degrees. When using modular
arithmetic the rank of the matrix E could be smaller than it would be using rational
arithmetic, and thus the nullspace will be too big. To avoid this we compute a
basis of the nullspace using modular arithmetic and reconstruct the most probable
corresponding integral vectors. We then check each of these identities using rational
arithmetic by generating pseudorandom integral elements of M and evaluating the
corresponding polynomial identity. If we obtain zero for a large number of choices,
we have obtained further confirmation of the hypothetical identity. For further
information, see [6, Lemma 8].
To reduce the number of identities we need to check, we first extract a set of
module generators from the linear basis of the nullspace.
4.2.2. Module generators. Let I1 , . . . , Iℓ be a linear basis for the Sd -module of iden-
tities in degree d satisfied by M ; that is, a basis for a certain subspace of the
q -dimensional vector space of nonassociative polynomials in degree d. We con-
struct a zero matrix G of size (q + d!) × q whose columns are labeled by the normal
monomials. We divide it into a q × q upper block and a d ! × q lower block whose
rows are labeled by the elements of the symmetric group Sd . We set oldrank ← 0
and then perform the following steps for k = 1, . . . , ℓ:
(1) Set i ← 0.
(2) For each permutation π in the symmetric group Sd do the following:
(a) Set i ← i + 1.
(b) For each j = 1, . . . , q do the following:
14
BENITO, BREMNER, AND MADARIAGA
• Let cj be the coefficient in Ik corresponding to monomial mj .
• If cj 6= 0 then apply π to mj obtaining πmj and replace πmj
by ±mj where mj is the normal form (the representative of the
equivalence class of πmj ).
• Store the corresponding coefficient ±cj in row q + i and column
j of G.
(3) Compute the row canonical form of G; the lower block of G is again zero.
(4) Set newrank ← rank(G).
(5) If oldrank < newrank then:
(a) Record Ik as a new module generator.
(b) Set oldrank ← newrank.
If we already know a set of generators for the lifted identities in degree d then we
apply the module generators algorithm to them. At the end of this computation
the row space of G contains a basis for the subspace of identities in degree d which
are consequences of identities of lower degree. We then apply the module genera-
tors algorithm to the linear basis produced by the fill and reduce algorithm: the
canonical basis of the nullspace of E which gives a basis for the subspace of all
identities in degree d.
In this way we obtain a set of module generators for the
space of all identities in degree d modulo the space of known identities in degree d;
that is, a set of module generators for the new identities in degree d.
5. Computational results for the LY algebra LY4
For the case so(3, R) ⊂ so(5, R), the bilinear operation gives the 7-dimensional
LY algebra LY3 the structure of a simple non-Lie Malcev algebra; the polynomial
identities for this LY algebra have been studied in [5].
In the rest of this section, we consider the 30-dimensional LY algebra LY4 ob-
tained from the embedding so(4, R) ⊂ so(9, R). We note that identities for the
trilinear operation occur only in odd degrees. We chose to perform s = 10 itera-
tions of the fill-and-reduce algorithm after the rank stabilizes.
5.1. Identities for LY4 in degree 3. We note that identities (LY1) and (LY2) in
the definition of LY algebras have already been taken into account in our definition
of normal monomials.
5.1.1. Binary identities. There is only one association type for [−, −] in degree
3, [[−, −], −], and three normal monomials, [[a, b], c], [[a, c], b], [[b, c], a]. Thus the
matrix for the fill-and-reduce algorithm has size 33 × 3, and it reaches full rank after
one iteration. Hence the binary product by itself satisfies no identities of degree 3
other than the consequences of anticommutativity.
5.1.2. Ternary identities. There is only one association type for (−, −, −) in degree
3, (−, −, −), and three normal monomials, (a, b, c), (a, c, b), (b, c, a). Thus the
matrix for the fill-and-reduce algorithm has size 33 × 3, and it reaches full rank
after one iteration. Hence the ternary product by itself satisfies no identities of
degree 3 other than the consequences of skew-symmetry in the first two arguments.
LIE-YAMAGUTI ALGEBRAS
15
5.1.3. Mixed identities. Using the normal monomials for both binary and ternary
products gives an ordered basis for the 6-dimensional space of multilinear LY poly-
nomials in degree 3; these monomials label the columns of the 36 × 6 matrix for the
fill-and-reduce algorithm:
[[a, b], c],
[[a, c], b],
[[b, c], a],
(a, b, c),
(a, c, b),
(b, c, a).
0 0
1 0
0 1
0 0
0 0
0
0
0
1
0
1
0
0
0
0
The rank reaches 5 after one iteration and remains unchanged for 10 iterations, so
there is an identity of degree 3 relating the binary and the ternary products. At
this point the row canonical form of the matrix is
A basis for the nullspace is the vector [1, 102, 1, 1, 102, 1]; using representatives from
−51 to 51 modulo 103 we obtain the vector [1, −1, 1, 1, −1, 1], which corresponds
to identity (LY3) in the definition of LY algebras:
[[a, b], c] − [[a, c], b] + [[b, c], a] + (a, b, c) − (a, c, b) + (b, c, a) ≡ 0,
5.1.4. Conclusion. There are no identities for LY4 in degree 3 which are not con-
sequences of the defining identities of LY algebras.
0 102
0
1
0 102
0 102
1
1
5.2. Identities for LY4 in degree 4. Using the module generators algorithm, we
find that the S4 -module of consequences in degree 4 of the identity LY 3(a, b, c) ≡
0 has dimension 10 and is generated by the identities LY 3([a, d], b, c) ≡ 0 and
[LY 3(a, b, c), d] ≡ 0.
5.2.1. Binary identities. The two association types for [−, −] in degree 4, namely
[[[−, −], −], −] and [[−, −], [−, −]], have respectively 12 and 3 normal monomials.
The matrix for the fill-and-reduce algorithm has size 45 × 15, and it reaches full
rank after one iteration. Hence the binary product by itself satisfies no identities
of degree 3 other than the consequences of anticommutativity.
5.2.2. Mixed identities. The five mixed association types in degree 4, namely
[[[−, −], −], −],
(−, −, [−, −]),
([−, −], −, −),
[[−, −], [−, −]],
[(−, −, −), −],
have respectively 12, 12, 3, 12, 6 normal monomials. The matrix for the fill-and-
reduce algorithm has size 75 × 45. The rank reaches 26 after one iteration and
remains unchanged for 10 iterations, so there is a 19-dimensional space of identities
of degree 4 relating the binary and the ternary products. We obtain a basis for
the nullspace by setting the free variables equal to the standard basis vectors and
solving for the leading variables, and then compute the row canonical form of the
nullspace basis. The coefficients of the canonical basis vectors are 0, 1 and 102 ≡ −1.
Interpreting the coefficients 0, ±1 as integers, we sort the basis vectors by increasing
square length (from 3 to 18). We use the module generators algorithm to compare
these new identities with the submodule generated by the lifted identities, and find
that there are two new identities which cause the rank to increase from 10 to 19.
These two identities are (LY4) and (LY5) in the definition of LY algebra.
16
BENITO, BREMNER, AND MADARIAGA
5.2.3. Conclusion. There are no identities for LY4 in degree 4 which are not con-
sequences of the defining identities of LY algebras.
5.3. Identities for LY4 in degree 5. Using the module generators algorithm, we
find the dimension of, and generators for, the S5 -module of consequences in degree
5 of LY 3(a, b, c) ≡ 0, LY 4(a, b, c, d) ≡ 0, LY 5(a, b, c, d) ≡ 0. The module has
dimension 280 and 11 generators:
LY 3([a, d], [b, e], c) ≡ 0,
LY 3([[a, e], d], b, c) ≡ 0,
[[LY 3(a, b, c), d], e] ≡ 0,
LY 3((a, d, e), b, c) ≡ 0,
LY 4(a, b, [c, e], d) ≡ 0,
LY 4([a, e], b, c, d) ≡ 0,
LY 5([a, e], b, c, d) ≡ 0,
[LY 5(a, b, c, d), e] ≡ 0.
5.3.1. Binary identities. There are three association types for [−, −] in degree 5
and 105 normal monomials. The matrix for the fill-and-reduce algorithm has size
135 × 105. After 4 iterations it reaches full rank, so the binary product by itself
satisfies no identities in degree 5 other than the consequences of anticommutativity.
[LY 3([a, d], b, c), e] ≡ 0,
(LY 3(a, b, c), d, e) ≡ 0,
[LY 4(a, b, c, d), e] ≡ 0,
5.3.2. Ternary identities. There are two association types for (−, −, −) in degree
5 and 90 normal monomials. The matrix for the fill-and-reduce algorithm has size
120 × 90. After two iterations, the rank reaches 60, and does not increase for
another 10 iterations. Hence the ternary product satisfies a 30-dimensional space
of identities in degree 5. We use the module generators algorithm to compare these
identities with the submodule generated by the lifted identities, and find that there
is one new identity which causes the rank to increase from 280 to 296; this identity
is (LY6). (The intersection of the space of lifted identities with the space of ternary
identities has dimension 280 + 30 − 296 = 14.)
5.3.3. Mixed identities. There are 13 association types involving both operations
in degree 5, and 510 normal monomials. Thus the matrix for the fill-and-reduce
algorithm has size 540 × 510. After 8 iterations, the rank reaches 214, and does
not increase for another 10 iterations. Hence there is a 296-dimensional space of
identities in degree 5 involving both operations. But this space contains the space
of the same dimension generated by the lifted identities and (LY6), so there are no
more new identities in degree 5.
5.3.4. Conclusion. Thus there are no identities of degree 5 for LY4 except those
which are consequences of the defining identities for LY algebras.
5.4. Identities for LY4 in degree 6. We extended these computations to degree 6
and found that every identity satisfied by LY4 is a consequence of the defining iden-
tities for LY algebras. The space of all multilinear LY polynomials has dimension
7245, and the liftings of the defining identities for LY algebras generates a sub-
module of dimension 5151. The fill-and-reduce algorithm (for the mixed identities)
stabilizes at rank 2094, indicating a nullspace of dimension 5151.
We summarize the computations of this section in the following theorem.
Theorem 5.1. Every multilinear polynomial identity of degree ≤ 6 satisfied by the
LY algebra LY4 is a consequence of the defining identities for LY algebras.
LIE-YAMAGUTI ALGEBRAS
17
6. The Jordan triple product as trilinear operation
Grishkov and Shestakov [14] define a Lie-Jordan algebra to be a vector space with
a bilinear product [−, −] and a trilinear product {−, −, −} satisfying the following
multilinear identities for all x, y , z , t, u:
[x, y ] ≡ −[y , x],
[[x, y ], z ] ≡ {x, y , z } − {y , x, z },
{x, y , z } ≡ {z , y , x},
[{x, y , z }, t] ≡ {[x, t], y , z } + {x, [y , t], z } + {x, y , [z , t]},
{{x, y , z }, t, u} ≡ {{x, t, u}, y , z } − {x, {y , u, t}, z } + {x, y , {z , t, u}}.
Thus a Lie-Jordan algebra is a Lie algebra with respect to the bilinear product and
a Jordan triple system with respect to the trilinear product. A Lie-Jordan algebra
is called special if it is isomorphic to a subspace of an associative algebra with the
standard Lie bracket and Jordan triple product:
{x, y , z } = xyz + z yx.
[x, y ] = xy − yx,
From the construction of universal enveloping algebras of Lie-Jordan algebras [14],
it follows that every Lie-Jordan algebra over a field of characteristic not 2 is special.
The same authors [15] show that a Lie algebra L with bilinear product [−, −] is
isomorphic to a Lie algebra of skew-symmetric elements of an associative algebra
with involution if and only if L admits an additional trilinear operation {−, −, −}
such that the resulting structure is a Lie-Jordan algebra.
Motivated by these considerations, we recall that the Lie algebra so(n, K) consists
of the n × n matrices which are skew-symmetric with respect to the transpose
involution, and so this subspace is also closed under the Jordan triple product. As
explained in Section 3, the underlying vector space of the Lie-Yamaguti algebra LYn
is the orthogonal complement of so(n, K) with respect to its natural embedding into
so(N , K). So we can retain the original bilinear operation on LYn but replace the
trilinear operation by the pro jection of the Jordan triple product. That is, for
U = L ⊕ M ,
U = so(N −1),
L = so(n),
M = LYn ,
we use the associative matrix product to replace equations (7) for A, B , C ∈ M by
hA, B , C i = pM ( ABC + CBA ).
[A, B ] = pM ( AB − BA ),
(8)
We call these new structures Lie-Jordan-Yamaguti algebras or LJY algebras for
short. We write LJ Yn for the subspace LYn with the same bilinear operation as
LYn but the new trilinear operation.
LJY algebras generalize Lie algebras and Jordan triple systems in a way some-
what analogous to the way that LY algebras generalize Lie algebras and Lie triple
systems. However, an LJY algebra with zero bilinear product is a triple system
with a completely symmetric trilinear operation satisfying the Jordan triple deriva-
tion identity, which provides a Jordan analogue of Filippov’s 3-Lie algebras [13].
On the other hand, an LJY algebra with zero trilinear product is a 2-step nilpotent
anticommutative (hence Lie) algebra.
In this section we study the polynomial identities satisfied by LJ Y3 and LJ Y4 .
6.1. The LJY algebra LJ Y3 . In this case the underlying vector space LJ Y3 with
the anticommutative bilinear product defines the 7-dimensional simple non-Lie Mal-
cev algebra; however, the symmetric trilinear product comes from the pro jection of
the Jordan triple product obtained from the orthogonal complement of the embed-
ding of so(3) into so(5). The computational methods are the same as before, but
18
BENITO, BREMNER, AND MADARIAGA
evaluating the products in the algebra is much easier since the tables of structure
constants are much smaller.
We recall the the multilinear form of the Malcev identity,
[[a, c], [b, d]] ≡ [[[a, b], c], d] + [[[b, c], d], a] + [[[c, d], a], b] + [[[d, a], b], c],
and Filippov’s h-identity [12] for the 7-dimensional simple non-Lie Malcev algebra,
[[[[a, b], c], d], e] + [[[[a, b], c], e], d] − [[[[a, b], d], c], e] − [[[[a, b], e], c], d]
+ [[[[a, d], b], e], c] − [[[[a, d], e], b], c] + [[[[a, e], b], d], c] − [[[[a, e], d], b], c]
+ 2[[[a, b], [c, d]], e] + 2[[[a, b], [c, e]], d] + 2[[[a, d], [b, e]], c] + 2[[[a, e], [b, d]], c] ≡ 0.
We introduce three independent identities in degree 5 relating the bilinear and
trilinear products:
(9)
2[[{a, c, d}, b], e] + [[{a, c, d}, e], b] − 2[[{a, e, d}, b], c] − [[{a, e, d}, c], b]
+[{a, c, d}, [b, e]] − [{a, e, d}, [b, c]] + {a, [[b, c], e], d} − {a, [[b, e], c], d}
−2{a, [[c, e], b], d} ≡ 0,
2{[[b, c], e], a, d} − 2{[[b, d], e], a, c} + {[[b, e], c], a, d} − {[[b, e], d], a, c}
+2{[[c, d], e], a, b} − {[[c, e], b], a, d} + {[[c, e], d], a, b} + {[[d, e], b], a, c}
−{[[d, e], c], a, b} ≡ 0,
{[[a, d], e], c, b} − {[[a, e], d], c, b} − {[[c, d], e], a, b} + {[[c, e], d], a, b}
−2{[[d, e], a], c, b} + 2{[[d, e], c], a, b} + 2{b, [[a, c], d], e} − 2{b, [[a, c], e], d}
+{b, [[a, d], c], e} − {b, [[a, e], c], d} − {b, [[c, d], a], e} + {b, [[c, e], a], d} ≡ 0.
Recall that the set Shn1 ···nk of (n1 , . . . , nk )-shuffles where n1 + · · · + nk = n consists
of all permutations σ of the ordered set X = {x1 , . . . , xn } (where xi ≺ xj ⇐⇒
n1+···+ni+1 ≺ · · · ≺ xσ
i < j ) satisfying xσ
n1+···+ni+1 for all i = 1, . . . , k−1. We
introduce three new independent identities in degree 6 relating the bilinear and
trilinear products; in each case the set X of permuted variables is indicated by the
superscript σ :
Xσ∈Sh212
{ [ [bσ , cσ ], dσ ], a, [eσ , f σ ]} ≡ 0,
Xσ∈Sh21(cid:16)3[{a, [cσ , dσ ], e}, [b, f σ ]] + 3[{a, [[cσ , dσ ], b], e}, f σ ]
−3[[cσ , dσ ], {a, [b, f σ ], e}] − 3[[[cσ , dσ ], b], {a, f σ , e}]
−[{a, [[cσ , dσ ], f σ ], e}, b] − [{a, b, e}, [[cσ , dσ ], f σ ]](cid:17) ≡ 0.
For the third new identity in degree 6, see Figure 1.
Altogether we found five new independent identities in degree 6 using modular
arithmetic (p = 103 and √2 ≡ 38). We then used the Maple function iratrecon to
determine the simplest rational numbers corresponding to each modular coefficient.
This worked for the first four identities, but failed for the fifth. We therefore ran the
Maple worksheet again with a much larger prime (p = 100049 and √2 ≡ 10948);
this time the rational reconstruction was successful for all five identities. Table
2 gives the number of terms in each identity, the integer coefficients, the squared
Euclidean length of the coefficient vector, and the dimension of the S6 -module
generated by the lifted identities and the new identities up to the current identity.
For each identity, we have multiplied the rational coefficients by the least common
(10)
LIE-YAMAGUTI ALGEBRAS
19
− 9{e,[[[b,d],a],c],f } − 9[[{c,b,d},e],[a,f ]] − 8{c,[[a,f ],[b,e]],d} − 8[[{e,d,f },b],[a,c]]
− 6[{c,b,d},[[e,f ],a]] − 6[[[{e,d,f },c],b],a] − 4{c,[[[a,f ],b],e],d} − 4[{e,d,f },[[b,c],a]]
− 4[{e,[[b,d],c],f },a] − 4[[{e,d,f },[a,c]],b] − 3{e,[[[b,d],c],a],f } − 3{e,[[[a,d],b],c],f }
− 3[{c,b,d},[[a,e],f ]] − 3[[{c,e,d},f ],[a,b]] − 3[{[[d,f ],c],b,e},a] − 3[{[[c,f ],d],b,e},a]
− 2{c,[[[e,f ],a],b],d} − 2{c,[[[b,f ],a],e],d} − 2[{e,d,f },[[a,b],c]] − 2[{c,f ,d},[[b,e],a]]
− 2[[{c,e,d},a],[b,f ]] − 2[{e,[[c,d],b],f },a] − 2[{e,[[b,c],d],f },a] − 2[{[[d,e],f ],b,c},a]
− 2[{[[c,e],f ],b,d},a] − 2[[{c,f ,d},[b,e]],a] −[{c,f ,d},[[a,b],e]] −[{c,e,d},[[a,f ],b]]
−[[{c,f ,d},b],[a,e]] −[{[[d,f ],e],b,c},a] −[{[[c,f ],e],b,d},a] +[{c,e,d},[[a,b],f ]]
+[{[[e,f ],d],b,c},a] +[{[[e,f ],c],b,d},a] + 2{c,[[[e,f ],b],a],d} + 2{c,[[[b,f ],e],a],d}
+ 2[{c,e,d},[[b,f ],a]] + 2[[{c,f ,d},a],[b,e]] + 2[[{e,d,f },[b,c]],a] + 2[[{c,e,d},[b,f ]],a]
+ 3{e,[[[c,d],b],a],f } + 3{e,[[[a,d],c],b],f } + 3[{c,f ,d},[[a,e],b]] + 3[{c,b,d},[[a,f ],e]]
+ 3[[{c,f ,d},e],[a,b]] + 3[[{c,e,d},b],[a,f ]] + 4{c,[[a,e],[b,f ]],d} + 4{c,[[a,b],[e,f ]],d}
+ 4{c,[[[a,f ],e],b],d} + 4[[{e,d,f },a],[b,c]] + 4[[{c,f ,d},[a,e]],b] + 6[[{e,d,f },c],[a,b]]
+ 6[[{c,b,d},a],[e,f ]] + 6[[[{c,b,d},e],f ],a] + 9{e,[[[c,d],a],b],f } + 9[[{c,b,d},f ],[a,e]]
+ 12{e,[[a,d],[b,c]],f } + 12{e,[[[b,c],d],a],f }
Figure 1. Third new multilinear identity in degree 6 for LJ Y3
multiple of their denominators, and divided the resulting integers by their greatest
common divisor. The new identities are sorted by increasing length. We used the
integer coefficients to check the new identities using arithmetic in characteristic
0 with √2:
for each identity, we performed ten trials by generating six random
elements of LJ Y3 with two-digit decimal coefficients, substituting these elements
into the identity, verifying that the result evaluated to 0 after simplification.
terms
30
18
58
333
635
coefficients
±1
±1, ±3
±1, ±2, ±3, ±4, ±6, −8, ±9, 12
±1, ±2, ±3, ±4, ±5, ±6, ±8, ±9, ±12
±3, ±9, ±18, ±27, ±36, ±45, ±104
Table 2. Five new identities in degree 6 for LJ Y3
30
114
1244
7468
1172619
length2 dimension
2632
2647
2701
2732
2733
Theorem 6.1. Every multilinear identity of degree ≤ 6 satisfied by LJ Y3 is a
consequence of the skew-symmetry of [−, −], the symmetry of {−, −, −}, the multi-
linear form of the Malcev identity, the Filippov h-identity, the three identities (9),
the two identities (10), the identity in Figure 1, and two more multilinear identities
in degree 6 with 333 and 635 terms.
We note that LJ Y3 does not satisfy the identity [[x, y ], z ] ≡ {x, y , z } − {y , x, z }
from the definition of Lie-Jordan algebras. On the other hand, the Malcev identity is
a consequence of the Lie-Jordan identities of degree ≤ 4; in fact, it is a consequence
of the liftings of [[x, y ], z ] ≡ {x, y , z } − {y , x, z }.
20
BENITO, BREMNER, AND MADARIAGA
6.2. The LJY algebra LJ Y4 . Similar computations to those described in Section
5 give the following result.
Theorem 6.2. Every multilinear polynomial identity of degree ≤ 6 satisfied by the
LJY algebra LJ Y4 is a consequence of the skew-symmetry of the bilinear operation
and the symmetry of the trilinear operation.
7. Conclusion
The class of Lie-Yamaguti algebras has not been broadly studied, probably be-
cause of its complex structure theory. It would be interesting to work out some more
explicit examples to help clarify the situation. Here are some ideas to consider.
We could extend the computations in this paper to higher degrees, study other
families of reductive pairs using the methods of this paper, including those produc-
ing k -irreducible Lie-Yamaguti algebras for k ≥ 2.
In the spirit of Section 6, we can replace the LY products by other operations
to see which other structures can be defined on the underlying vector spaces of
the representations of the Lie algebras. As mentioned in § 3.1, if h is a simple
Lie algebra and m is an irreducible h-module, and the exterior square Λ2m contains
both m and the adjoint h-module, then there are h-module morphisms β : Λ2m → m
and τ : Λ2m → h, which give bilinear and trilinear products on m defined by [x, y ] =
β (x, y ) and hx, y , z i = τ (x, y ) · z ; then a Lie algebra structure can be defined on
h ⊕ m if and only if identities (LY3) and (LY4) are satisfied. Thus we need to find
the polynomial identities of low degrees satisfied by these operations to see if we
get LY-structures or different classes of algebras.
We used the online LiE software [19] to study the Lie algebra of type A2 = sl3 (C).
The adjoint representation has weight λ1+λ2 , denoted [1, 1] in LiE. All the examples
we found of modules whose exterior squares contain both the original module and
the adjoint module satisfy a = b (and hence are self-dual), which suggests the
conjecture that Λ2 [a, b] contains [a, b] and [1, 1] if and only if a = b.
• For [1, 1], the pro jection onto [1, 1] gives the Lie bracket on the adjoint module,
so this is a Lie-Yamaguti algebra of adjoint type:
Λ2 [1, 1] = [3, 0] ⊕ [0, 3] ⊕ [1, 1]
• For [2, 2], all multiplicities are 1 in the decomposition:
Λ2 [2, 2] = [5, 2] ⊕ [2, 5] ⊕ [3, 3] ⊕ [4, 1] ⊕ [1, 4] ⊕ [2, 2] ⊕ [3, 0] ⊕ [0, 3] ⊕ [1, 1]
This produces an LY algebra that appears in references [3] and [4].
• For [3, 3], the adjoint module [1, 1] occurs with multiplicity 1, but [3, 3] occurs
with multiplicity 2, so there will be a one-parameter family of different bilinear
products to consider:
Λ2 [3, 3] = [7, 4] ⊕ [4, 7] ⊕ [5, 5] ⊕ [9, 0] ⊕ [6, 3] ⊕ [3, 6] ⊕ [0, 9] ⊕ [7, 1] ⊕ [4, 4] ⊕ [1, 7]
⊕ 2[5, 2] ⊕ 2[2, 5] ⊕ 2[3, 3] ⊕ [4, 1] ⊕ [1, 4] ⊕ [2, 2] ⊕ [3, 0] ⊕ [0, 3] ⊕ [1, 1]
LIE-YAMAGUTI ALGEBRAS
21
• For [4, 4] the multiplicities are the same as in the previous case:
Λ2 [4, 4] = [9, 6] ⊕ [6, 9] ⊕ [7, 7] ⊕ [11, 2] ⊕ [8, 5] ⊕ [5, 8] ⊕ [2, 11] ⊕ [9, 3] ⊕ [6, 6]
⊕ [3, 9] ⊕ [10, 1] ⊕ 2[7, 4] ⊕ 2[4, 7] ⊕ [1, 10] ⊕ [8, 2] ⊕ 2[5, 5] ⊕ [2, 8] ⊕ [9, 0]
⊕ 2[6, 3] ⊕ 2[3, 6] ⊕ [0, 9] ⊕ [7, 1] ⊕ 2[4, 4] ⊕ [1, 7] ⊕ 2[5, 2] ⊕ 2[2, 5]
⊕ 2[3, 3] ⊕ [4, 1] ⊕ [1, 4] ⊕ [2, 2] ⊕ [3, 0] ⊕ [0, 3] ⊕ [1, 1]
References
[1] H. Aslaksen: Determining summands in tensor products of Lie algebra representations. J.
Pure Appl. Algebra 93 (1994), no. 2, 135–146.
[2] P. Benito, C. Draper, A. Elduque: Lie-Yamaguti algebras related to g2 . J. Pure Appl.
Algebra 202 (2005), no. 1-3, 22–54.
[3] P. Benito, A. Elduque, F. Mart´ın-Herce: Irreducible Lie-Yamaguti algebras. J. Pure
Appl. Algebra 213 (2009), no. 5, 795–808.
[4] P. Benito, A. Elduque, F. Mart´ın-Herce: Irreducible Lie-Yamaguti algebras of generic
type. J. Pure Appl. Algebra 215 (2011), no. 2, 108–130.
[5] M. R. Bremner, A. F. Douglas: The simple non-Lie Malcev algebra as a Lie-Yamaguti
algebra. J. Algebra 358 (2012), 269–291.
[6] M. R. Bremner, L. A. Peresi: Nonhomogeneous subalgebras of Lie and special Jordan
superalgebras. J. Algebra 322 (2009), no. 6, 2000–2026.
[7] J. Dixmier: Certaines alg`ebres non associatives simples d´efinies par la transvection des formes
binaires. J. Reine Angew. Math. 446 (1984) 110–128.
[8] W. Dickinson, M. M. Kerr: The geometry of compact homogeneous spaces with two
isotropy summands. Ann. Global Anal. Geom. 34 (2008), no. 4, 329–350.
[9] A. Elduque: New simple Lie superalgebras on characteristic 3. J. Algebra 296 (2006), no. 1,
196–233.
[10] A. Elduque, H. Myung: The reductive pair (B3 , G2 ) and affine connections on S 7 . J. Pure
Appl. Algebra 86 (1993), no. 2, 155–171.
[11] A. Elduque, H. Myung: Octonions and affine connections on spheres. Nonassociative Alge-
bra and Its Applications (Sao Paulo, 1998), pages 43–54. Lecture Notes in Pure and Appl.
Math., 211, Dekker, New York, 2000.
[12] V. T. Filippov: On a variety of Malcev algebras. Algebra i Logika 20 (3) (1981) 300–314.
[13] V. T. Filippov: n-Lie algebras. Sibirsk. Mat. Zh. 26 (1985), no. 6, 126–140.
[14] A. N. Grishkov, I. P. Shestakov: Speciality of Lie-Jordan algebras. J. Algebra 237 (2001),
no. 2, 621–636.
[15] A. N. Grishkov, I. P. Shestakov: A characterization of Lie algebras of skew-symmetric
elements. Acta Appl. Math. 85 (2005), no. 1-3, 157–159.
[16] J. E. Humphreys: Introduction to Lie Algebras and Representation Theory. Second printing,
revised. Graduate Texts in Mathematics, 9. Springer-Verlag, New York-Berlin, 1978.
[17] M. Kikkawa: Geometry of homogeneous Lie loops. Hiroshima Math. J. 5 (1975), no. 2,
141–179.
[18] M. K. Kinyon, A. Weinstein: Leibniz algebras, Courant algebroids, and multiplications on
reductive homogeneous spaces. Amer. J. Math. 123 (2001), no. 3, 525–550.
[19] M. A. A. van Leeuwen, A. M. Cohen, B. Lisser: LiE, A Package for Lie Group Compu-
tations. Computer Algebra Nederland (1992). Software available online at:
http://young.sp2mi.univ-poitiers.fr/~marc/LiE/form.html
[20] W. G. Lister: A structure theory of Lie triple systems. Trans. Amer. Math. Soc. 72 (1952)
217–242.
[21] F. Mart´ın-Herce: ´Algebras de Lie-Yamaguti y sistemas algebraicos no asociativos. Ph. D.
thesis, University of La Rio ja, 2005.
[22] A. A. Sagle, D. J. Winter: On homogeneous spaces and reductive subalgebras of simple
Lie algebras. Trans. Amer. Math. Soc. 128 (1967) 142–147.
[23] A. A. Sagle: A note on simple anti-commutative algebras obtained from reductive homoge-
neous spaces. Nagoya Math. J. 31 (1968) 105–124.
[24] J. A. Wolf: The geometry and structure of isotropy irreducible homogeneous spaces. Acta
Math. 120 (1968) 59–148.
22
BENITO, BREMNER, AND MADARIAGA
[25] J. A. Wolf: Correction to: “The geometry and structure of isotropy irreducible homogeneous
spaces”. Acta Math. 152 (1984), nos. 1-2, 141–142.
[26] K. Yamaguti: On the Lie triple system and its generalization. J. Sci. Hiroshima Univ. Ser.
A 21 (1957/1958) 155–160.
[27] K. A. Zhevlakov, A. M. Slinko, I. P. Shestakov, A. I. Shirshov: Rings That Are Nearly
Associative. Translated from the Russian by Harry F. Smith. Pure and Applied Mathematics,
104. Academic Press, Inc., 1982.
Departamento de Matem´aticas y Computaci´on, Universidad de La Rioja, Espana
E-mail address : [email protected]
Department of Mathematics and Statistics, University of Saskatchewan, Canada
E-mail address : [email protected]
Department of Mathematics and Statistics, University of Saskatchewan, Canada
E-mail address : [email protected]
|
1805.01796 | 2 | 1805 | 2019-01-23T10:15:25 | Bounding the free spectrum of nilpotent algebras of prime power order | [
"math.RA"
] | Let $\mathbf{A}$ be a finite nilpotent algebra in a congruence modular variety with finitely many fundamental operations. If $\mathbf{A}$ is of prime power order, then it is known that there is a polynomial $p$ such that for every $n \in \mathbb{N}$, every $n$-generated algebra in the variety generated by $\mathbf{A}$ has at most $2^{p(n)}$ elements. We present a bound on the degree of this polynomial. | math.RA | math |
BOUNDING THE FREE SPECTRUM OF NILPOTENT
ALGEBRAS OF PRIME POWER ORDER
ERHARD AICHINGER
Abstract. Let A be a finite nilpotent algebra in a congruence modular vari-
ety with finitely many fundamental operations. If A is of prime power order,
then it is known that there is a polynomial p such that for every n ∈ N, every
n-generated algebra in the variety generated by A has at most 2p(n) elements.
We present a bound on the degree of this polynomial.
1. Introduction
The binary commutator operation defined by [Smi76] and studied in [FM87,
MMT87] has allowed to generalize concepts from group theory, such as solvability
or nilpotency, from groups to arbitrary universal algebras. For an algebra A in a
congruence modular variety, its lower central series is a series of its congruence
relations, and it is defined by λ1 := 1A and λk+1 := [1A, λk] for k ∈ N, where
[. , .] denotes the term condition commutator defined in [FM87, MMT87].
If
λk+1 = 0A, then A is called k-nilpotent. From [Hig67], we know that for a k-
nilpotent group G, there is a polynomial p of degree k such that for all n ∈ N, all
n-generated groups in the variety generated by G are of size at most 2p(n). This
property can be investigated for arbitrary algebraic structures, and we say that
a finite algebra A has small free spectrum if there is a polynomial p such that
for all n ∈ N, every n-generated algebra in the variety generated by A is of size
at most 2p(n). Straightforward generalizations of the group theoretic results fail:
In [VL83, p. 308, Example 2] Vaughan-Lee constructed a nilpotent loop of size
12, and [AM07, p. 283] exhibits a nilpotent expansion of the six element abelian
group with one unary operation, which both fail to have small free spectrum.
However, in a congruence modular variety, the following result is known:
Theorem 1.1 ([BB87, Theorem 2]). Let A be a finite nilpotent algebra of finite
type in a congruence modular variety. We assume that A is a direct product of
algebras of prime power order. Then A has small free spectrum.
Date: January 24, 2019.
2010 Mathematics Subject Classification. 08A40 (08B20,20N05).
Key words and phrases. nilpotent algebra, free spectrum, supernilpotent algebra, congruence
modular variety.
Supported by the Austrian Science Fund (FWF):P29931.
1
2
ERHARD AICHINGER
If A is a group, this is known from [Hig67]. The proof of the above theorem relies
on a generalization of Higman's combinatorial argument given in [BB87] and on
bounding the rank of the commutator terms of A. Such a bound was derived
in [VL83] and Chapter 14 of [FM87] in the course of proving that an algebra
satisfying the assumptions of Theorem 1.1 has a finite basis for its equational
laws. In other words, the above theorem by Berman and Blok tells that for each
algebra A satisfying the assumptions of Theorem 1.1, there is a polynomial p such
that every n-generated algebra in the variety generated by A has at most p(n)
elements. The contribution of the present work is an upper bound on the degree
of p. In deriving this upper bound, we obtain an alternative proof of Theorem 1.1.
We observe that for a finite algebra A, every n-generated algebra in the variety
generated by A is a homomorphic image of the free algebra in this variety, and
this free algebra is isomorphic to the algebra Clon(A) of n-ary term functions
on A, and the free spectrum fA of A is defined by fA(n) := Clon(A). We
also mention that Theorem 3.14 from [Kea99] provides some kind of a converse:
a finite algebra in a congruence modular variety with small free spectrum is a
direct product of algebras of prime power order.
The property of having small free spectrum is closely related to supernilpo-
tency, a notion introduced in [AE06, AM10]. We say that an algebra A is k-
supernilpotent if the higher commutator operation defined in [Bul01] and stud-
ied, e.g., in [AM10, Moo18] satisfies [1A, . . . , 1A]A = 0A (k + 1 repetitions of
1A); this condition is formulated without using higher commutators in Defini-
tion 2.1 below. The algebra A is called supernilpotent if there is k ∈ N such
that A is k-supernilpotent. For those classes of algebra that we will study here,
supernilpotency implies nilpotency: this implication holds in congruence per-
mutable varieties by [AM10], and more generally in congruence modular varieties
by [Wir19]. The connection between supernilpotency and small free spectrum is
stated in Lemma 2.4 below. From this Lemma, we see that a finite algebra A in a
congruence modular variety is k-supernilpotent if and only if there is a polynomial
p of degree k such that for its free spectrum, we have fA(n) ≤ 2p(n) for all n ∈ N;
hence A is supernilpotent if and only if A has small free spectrum. Using the
concept of supernilpotency, the theorem by Berman and Blok can be rephrased
as "every nilpotent algebra of finite type and prime power order in a congruence
modular variety is supernilpotent". However, although [BB87] yields the exis-
tence of a k such that the algebra is k-supernilpotent, no explicit upper bound
for k has been computed. For groups and rings, k can be chosen to be the nilpo-
tency degree, but this does not hold in general: for every k, m ∈ N with m ≥ 2,
[AM13] exhibits a k-nilpotent algebra of size 2k with fundamental operations of
arity at most m that is mk−1-supernilpotent, but not (mk−1 − 1)-supernilpotent.
These examples show that a bound on the supernilpotency degree cannot be a
function of k alone, but must contain more information on the algebra. For cer-
tain algebras (groups expanded with multilinear operations), an explicit bound
NILPOTENT ALGEBRAS OF PRIME POWER ORDER
3
was given in [AM13]. Our main theorem provides such a bound for all algebras
covered by the Berman-Blok-Theorem; in particular, it applies to nilpotent loops
of prime power order. One ingredient used in this bound is the height of the
congruence lattice of A, which we define as the maximal size of a linearly ordered
subset of the lattice minus one; hence the height of the 1-element lattice is 0 and
the height of a linearly ordered set with n elements is n − 1.
Theorem 1.2. Let q > 1 be a prime power, let m ∈ N, and let A be a nilpotent
algebra in a congruence modular variety with A = q such that all fundamental
operations of A are of arity at most m. Let h be the height of the congruence
lattice of A, and let
s := (cid:0)m(q − 1)(cid:1)h−1
.
Then A is s-supernilpotent, and there is a polynomial p ∈ R[x] of degree at most
s such that the free spectrum satisfies fA(n) = 2p(n) for all n ∈ N.
From this result, we obtain the following improvement of Theorem 1.1.
Corollary 1.3. Let A be a finite nilpotent algebra in a congruence modular va-
riety that is a direct product of algebras of prime power order, and let m ∈ N be
such that such that all fundamental operations of A are of arity at most m. We
assume A > 1. Let
s := (cid:0)m(A − 1)(cid:1)(log2(A)−1).
Then A is s-supernilpotent and there is a polynomial p ∈ R[x] of degree ≤ s such
that the free spectrum satisfies fA(n) = 2p(n) for all n ∈ N.
Combining this with [Kea99], we obtain:
Corollary 1.4. Let A be a finite algebra in a congruence modular variety with
A > 1, and let m ∈ N be such that such that all fundamental operations of A
are of arity at most m. Then we have:
(1) If A has small free spectrum, then there is a polynomial p ∈ R[x] of degree
at most (m(A − 1))(log2(A)−1) such that fA(n) = 2p(n) for all n ∈ N.
(2) If A is supernilpotent, then it is (cid:0)(m(A − 1))(log2(A)−1)(cid:1)-supernilpotent.
The proofs of these results will be given in Section 7. Our proof of Theorem 1.2
will proceed as follows: We define a binary operation + on A = (A, F ) such that
(A, +) is an elementary abelian group and A′ = (A, F ∪ {+}) is still nilpotent.
Since (A, +) is elementary abelian, we can expand it to a finite field (A, +, ·)
and represent all fundamental operations from A by polynomials over this field.
Using this representation, we show that A′ is s-supernilpotent, which implies
that its reduct A is also s-supernilpotent.
4
ERHARD AICHINGER
2. Preliminaries about supernilpotency
We use the definition of supernilpotency in [AM10, Definition 7.1]. This definition
can be stated as follows:
Definition 2.1 (Term condition for supernilpotency). Let A be an algebra and
k ∈ N. Then A is k-supernilpotent if for all n1, . . . , nk+1 ∈ N0 and for all
h(a(i)
i=1 ni-ary term
functions t of A the following holds: if for all f : {1, . . . , k} → {1, 2} such that f
is not constantly 2, we have
i=1 (Ani × Ani) and for all Pk+1
2 ) i ∈ {1, . . . , k + 1}i ∈ Qk+1
1 , a(i)
t(a(1)
f (1), . . . , a(k)
f (k), a(k+1)
1
then
t(a(1)
2 , . . . , a(k)
2 , a(k+1)
1
) = t(a(1)
f (1), . . . , a(k)
f (k), a(k+1)
2
),
) = t(a(1)
2 , . . . , a(k)
2 , a(k+1)
2
).
From this definition, we see immediately that reducts of supernilpotent algebras
are supernilpotent:
Lemma 2.2. Let s ∈ N, and let A, B be universal algebras with the same uni-
verse. If B is s-supernilpotent and the clones of term operations of these algebras
satisfy Clo(A) ⊆ Clo(B), then A is also s-supernilpotent.
We also see that s-supernilpotency is defined by an infinite set of quasi-identities,
and is therefore preserved under taking subalgebras and direct products.
If A = (A, +, −, 0, (fi)i∈I) is an expanded group, we can describe supernilpotency
more easily. For n ∈ N, we call a function f : An → A absorbing if for all
a1, . . . , an ∈ A with 0 ∈ {a1, . . . , an}, we have f (a1, . . . , an) = 0. The prototypes
of absorbing functions are the commutator (a1, a2) 7→ −a1 − a2 + a1 + a2 in any
group, (a1, a2) 7→ a1a2 in any ring, and, also on every ring, every function that
can be written as (a1, . . . , an) 7→ a1a2 · · · an · g(a1, . . . , an). The essential arity of
f : An → A is the number of arguments on which f depends. We note that the
essential arity of an absorbing function f : An → A is either n or 0.
Lemma 2.3. Let A = (A, +, −, 0, (fi)i∈I) be an expanded group, and let s ∈ N.
Then the following are equivalent:
(1) A is s-supernilpotent.
(2) All absorbing polynomial functions of A are of essential arity at most s.
If A is finite, then (1) and (2) are furthermore equivalent to
(3) There is a polynomial p ∈ R[x] of degree ≤ s such that fA(n) =
Clon(A) = 2p(n) for all n ∈ N.
Proof: The equivalence of (1) and (2) follows from Corollary 6.12 of [AM10] by
observing that s-supernilpotency is equivalent to the higher commutator property
NILPOTENT ALGEBRAS OF PRIME POWER ORDER
5
[1A, . . . , 1A] = 0A ((s + 1) times 1A). The equivalence of (3) and (1) follows from
Corollary 4.3 of [Aic14]; there it was proved using a modification of an argument
that goes back to [Hig67].
(cid:3)
The equivalence of (1) and (3) is actually true for all finite algebras in congruence
modular varieties. Following [FM87], we say that a term w(x1, . . . , xr+1) in the
language of A is a commutator term of rank r for A if A = w(z, x2, . . . , xr, z) ≈
w(x1, z, . . . , xr, z) ≈ · · · ≈ w(x1, x2, . . . , z, z) ≈ z. A commutator term
w(x1, . . . , xr+1) is called trivial if A = w(x1, . . . , xr, z) ≈ z. A part of the next
lemma has also been stated in [AMO18].
Lemma 2.4. Let A be a finite algebra in a congruence modular variety, and let
s ∈ N. Then the following are equivalent:
(1) A is s-supernilpotent.
(2) A is nilpotent, and all nontrivial commutator terms of A are of rank at
most s.
(3) There is a polynomial p ∈ R[x] of degree at most s such that fA(n) = 2p(n)
for all n ∈ N.
(4) limn→∞ (cid:0) log2(fA(n))/ns+1(cid:1) = 0.
Proof: (1)⇒(2): By [Wir19, Theorem 4.11], A has an (s + 1)-difference term d.
Since A is s-supernilpotent, d is a Mal'cev term. From [AM10, Lemma 7.5], we
obtain that A is nilpotent and all commutator terms have rank at most s. This
bound on the rank can also be seen directly from the term condition that defines
supernilpotency: to this end, let w(x1, . . . , xr+1) be a commutator term of A with
r > s. We want to show that w satisfies A = w(x1, . . . , xr, z) ≈ z. To this end,
let ξ1, . . . , ξr, ζ ∈ A. We apply the term condition from Definition 2.1 with the
following settings: t := wA, a(i)
:=
1
(ζ, . . . , ζ) (r − s + 1 times ζ), a(s)
:= (ξs+1, . . . , ξr, ζ). Then the term condition
2
implies t(ξ1, . . . , ξs, ζ, . . . , ζ) = t(ξ1, . . . , ξs, . . . , ξr, ζ). Since w is a commutator
term, ζ = t(ξ1, . . . , ξs, ζ, . . . , ζ). Thus A = w(x1, . . . , xr, z) ≈ z, and hence w is
trivial.
:= ξi for i ∈ {1, . . . , s}, a(s+1)
:= ζ and a(i)
2
1
(2)⇒(3): Under the additional assumption that A is a direct product of algebras
of prime power order, this is shown in the proof of Theorem 2 of [BB87]. However,
this additional assumption is only used to obtain a bound on the rank of nontrivial
commutator terms, which is claimed by (2).
(3)⇒(4): Obvious.
(4)⇒(1): The proof for this implication comes from [AMO18]; it is included for
easier reference.
Theorem 9.18 of [HM88] implies that the variety V (A) omits types 1 and 5.
From [HM88, Lemma 12.4], we obtain that A is right nilpotent, and since the
commutator operation in a congruence modular variety is commutative, A is
6
ERHARD AICHINGER
therefore nilpotent. Now [FM87, Theorem 6.2] yields that A has a Mal'cev term.
Let A∗ be the expansion of A with all its constants. Then A∗ is nilpotent and
generates a congruence permutable variety. The variety V (A∗) is nilpotent by
[FM87, Theorem 14.2], and hence congruence uniform by [FM87, Corollary 7.5].
Since for all n ∈ N, fA∗(n) ≤ fA(n + A) ≤ 2p(n+A), we obtain from the proof
of [BB87, Theorem 1] that all commutator terms of A∗ are of rank at most
s. Hence all commutator polynomials (in the sense of [AM10, Definition 7.2])
of A are of rank at most s, and then [AM10, Lemma 7.5] yields that A is s-
supernilpotent.
(cid:3)
It is worth noting that in proving Lemma 2.4, we needed to employ substantial
results from each of the sources [BB87, FM87, HM88, AM10, Wir19].
3. Preliminaries on commutators and nilpotency
In this section, we compile some well known facts on the relation between the
commutator operation and the Mal'cev term of an algebra. This is an extension
of [Aic06, p. 14]. Let A be an algebra with a Mal'cev term d. We fix an element
o ∈ A and define two binary operations +o and −o by
(3.1)
x +o y := d(x, o, y) and
x −o y := d(x, y, o) for x, y ∈ A.
Sometimes, we also use −o as a unary operation: then −o y stands for o −o y =
d(o, y, o).
In the following proposition, we compile those relations of +o and
−o with the commutator that we will need in the sequel. Such properties have
been established from the very beginning of modular commutator theory (cf.
[Her79, Gum83]), and the proofs of several of these properties are taken from
[Aic06]. The proofs given below rely only on the following fact that follows rather
directly from the definition of the term condition defining the binary commutator
operation (see Lemma 2.2 of [Aic06] or Exercise 4.156(2) from [MMT87]):
if α
and β are congruences of any algebra A, (a, b) ∈ α, (c, d) ∈ β, p ∈ Pol2(A), and
p(a, c) = p(a, d), then p(b, c) ≡ p(b, d) (mod [α, β]).
Lemma 3.1 (cf.
[Aic06, Proposition 2.7]). Let A be an algebra with a Mal'cev
term d, let a, b, c, o be elements of A, let +o and −o be defined as in (3.1), and
let α, β be congruences of A. Then we have:
(1) a +o o = o +o a = a −o o = a.
(2) a −o a = o.
(3) If a ≡α b ≡β o, then (a −o b) +o b ≡ a (mod [α, β]) .
(4) If a ≡α o ≡β b, then (a +o b) −o b ≡ a (mod [α, β]) .
(5) If a ≡α o ≡β b, then a +o b ≡ b +o a (mod [α, β]).
(6) If a ≡α o ≡β b, then (a +o b) +o c ≡ a +o (b +o c) (mod [α, β]).
(7) If a ≡α o ≡β b, then d(a +o b, b, c) ≡ a +o c (mod [α, β]).
NILPOTENT ALGEBRAS OF PRIME POWER ORDER
7
(8) If a ≡α o, then (−o a) +o a ≡ o (mod [α, α]).
Proof: Properties (1) and (2) follow from the properties of the Mal'cev term
d. For proving (3), we define a polynomial function t ∈ Pol2(A) by t(x, y) :=
d(d(a, x, y), y, b). We have t(a, o) = t(a, b) = b. Thus we obtain t(b, o) ≡
t(b, b) (mod [α, β]), which means (a−o b)+o b ≡ a (mod [α, β]) . For proving (4),
we define a polynomial function t ∈ Pol2(A) by t(x, y) := d(d(x, y, b), d(b, y, o), o).
We have t(o, o) = t(o, b) = o. Thus we obtain t(a, o) ≡ t(a, b) (mod [α, β]),
which means (a +o b) −o b ≡ a (mod [α, β]) . For proving (5), we define t(x, y) :=
d(y +o x, x +o y, a +o b) for x, y ∈ A. Then we have t(o, o) = t(o, b) = a +o b, and
therefore t(a, o) ≡ t(a, b) (mod [α, β]), which implies a+ob ≡ b+oa (mod [α, β]).
For proving (6), we define t(x, y) := d(cid:0)x+o (y +o c), (x+o y)+o c, (a+o b)+o c)(cid:1) and
have t(o, o) = t(o, b) = (a +o b) +o c, and therefore t(a, o) ≡ t(a, b) (mod [α, β]),
which implies (a +o b) +o c ≡ a +o (b +o c) (mod [α, β]). For proving (7), we
consider the polynomial function of A defined by t(x, y) := d(x +o y, y, c) for
x, y ∈ A. Then t(o, o) = d(o, o, c) = c and t(o, b) = d(b, b, c) = c. There-
fore, (t(a, o), t(a, b)) ∈ [α, β], and thus d(a, o, c) ≡ d(a +o b, b, c) (mod [α, β]).
Since d(a, o, c) = a +o c, the result follows. For property (8), we observe that
(−o a) +o a = (o −o a) +o a. By property (3), the last expression is congruent
modulo [α, α] to o.
(cid:3)
The following well known Lemma goes back to [Her79, Fre83].
Lemma 3.2. Let A be an algebra with Mal'cev term d, and let α be a congruence
of A with [α, α] = 0A. Let Q := o/α. Then Q := (Q, +o, −o, o) is an abelian
group. If α is furthermore a minimal congruence of A and Q is finite, then Q is
of prime exponent.
Proof: The first part follows from items (6),(1), and (8) of Lemma 3.1. For the
second part, we sketch an argument taken from [Fre83, p. 151]: From Proposi-
tion 2.8(2) of [Aic06], it is not hard to infer that the group Q can be seen as
a module over the finite ring ({pQ : p ∈ Pol1(A), p(o) = o}, +o, ◦). Since α is a
minimal congruence, this module has no submodules, and thus Q is the additive
group of a finite simple module, and has therefore prime exponent.
(cid:3)
We will also use the following relational description of centrality that goes back
to [Kis92]. We call a congruence relation ζ of A central in A if [ζ, 1A] = 0A (cf.
[BS81, Definition 13.1]).
Lemma 3.3 (Relational description of centrality, cf. Theorem 3.2(iii) of [Kis92]).
Let A be an algebra with a Mal'cev term d, and let ζ ∈ Con(A). Then ζ is
central in A if and only if all fundamental operations of A preserve the relation
ρ = {(a1, a2, a3, a4) ∈ A4 (a1, a2) ∈ ζ, d(a1, a2, a3) = a4}.
Proof: The result is a special case of [AM07, Proposition 2.3 and Lemma 2.4]. (cid:3)
8
ERHARD AICHINGER
In expanded groups, the commutator of two congruences can be calculated
from the associated 0-classes (ideals) and binary polynomial functions [AM07,
Lemma 2.9]. We will only use the following assertion:
Lemma 3.4. Let A = (A, +, −, 0, (fi)i∈N) be an expanded group, let ξ, η be con-
gruences of A, let X := 0/ξ and Y := 0/η be the ideals of A associated with these
congruences, and let p ∈ Pol2(A) such that p(a, 0) = p(0, a) = 0 for all a ∈ A.
Then for all x ∈ X and y ∈ Y , p(x, y) ≡ 0 (mod [ξ, η]), and therefore p(x, y) lies
in the ideal [X, Y ] := 0/[ξ, η] associated with [ξ, η].
Proof:
p(x, y) (mod [ξ, η]).
Since p(0, 0) = p(0, y),
the term condition yields p(x, 0) ≡
(cid:3)
4. Expanding an algebra with a group operation
Let A be an algebra in a congruence modular variety, let m ∈ N0, and let
0A = α0 ≤ α1 ≤ · · · ≤ αm = 1A be a linearly ordered sequence of equivalence
relations on A. L = hαi i ∈ {0, 1, . . . , m}i is a central series of A if for each
i ∈ {1, . . . , m}, αi is a congruence relation of A and αi/αi−1 is central in A/αi−1;
using the homomorphism property of the modular commutator, this centrality
can be expressed by [1A, αi] ≤ αi−1. An algebra is nilpotent if and only if it has
a finite central series. We fix an element o ∈ A and a Mal'cev term d of A. For
each i ∈ {1, . . . , m}, we let Gi ⊆ A/αi−1 be defined by
Gi := {x/αi−1 x ∈ A, (x, o) ∈ αi}.
In other words, Gi is the image of o/αi under the canonical projection from A
to A/αi−1. Let ¯o := o/αi−1. Since [αi, αi] ≤ [1, αi] ≤ αi−1, Lemma 3.2 and
the homomorphism property of the modular commutator tell that the operations
g +i h := d(g, ¯o, h) and −i g := d(¯o, g, ¯o) turn Gi into an abelian group. We call
m
G :=
(Gi, +i, −i, ¯o)
Y
i=1
the abelian group associated with the algebra A, its central ceries L, and zero o.
Lemma 4.1. Let A be a finite nilpotent algebra in a congruence modular variety,
let 0A = α0 ≺ α1 ≺ · · · ≺ αm = 1A be a maximal chain in the congruence lattice
of A, and let o ∈ A. Let G = Qm
i=1 Gi be the abelian group associated with A,
L := hαi i ∈ {0, 1, . . . , m}i, and zero o. Then G is a direct product of groups
of prime order, and G = A. If A is furthermore of prime power order, then
G is elementary abelian.
Proof: Since A is nilpotent, L is a central series of A. By Lemma 3.2, each
of the groups Gi is an abelian group of prime exponent. Furthermore, A is
congruence uniform [FM87, Corollary 7.5]. For proving G = A, we proceed
NILPOTENT ALGEBRAS OF PRIME POWER ORDER
9
by induction on m and have A = o/α1 · A/α1 = G1 · A/α1, which is equal
to to G1 · Qm
i=2 Gi = G by the induction hypothesis. Now assume that A is
of prime power order. Then G is of prime power order and squarefree exponent,
and therefore elementary abelian.
(cid:3)
The following theorem allows to expand a nilpotent algebra with a Mal'cev term
with group operations such that nilpotency is preserved.
Theorem 4.2. Let A = (A, F ) be a nilpotent algebra with a Mal'cev term d,
let m ∈ N, let o ∈ A, and let L = hαi i ∈ {0, . . . , m}i be a central series of
A. Then there exist a binary function + : A × A → A and a unary function
− : A → A such that
(1) (A, +, −, o) is isomorphic to the abelian group G associated with A, L,
and o.
(2) L = hαi i ∈ {0, . . . , m}i is a central series also of the expansion A′ =
(A, F ∪ {+, −, o}) of A, and therefore A′ is nilpotent of class at most m.
Proof: As in (3.1), we define x +o y := d(x, o, y), x −o y := d(x, y, o), and
−o y := d(o, y, o) for x, y ∈ A. We proceed by induction on m. We show that
there exist +, − such that
(i) for each i ∈ {0, 1, . . . , m}, both + and − preserve the congruence αi,
(ii) the algebra (A, +, −, o) is isomorphic to the abelian group G associated
with A, L and zero o,
(iii) for each i ∈ {1, . . . , m}, both + and − preserve the relation γi given by
γi = {(x1, x2, x3, x4) ∈ A4 (x1, x2) ∈ αi,(cid:0)d(x1, x2, x3), x4(cid:1) ∈ αi−1}.
If m = 0, then A = 1. Defining + as the only binary and − as the only unary
operation of this set, we see that (A, +, −, o) is a one element group, and hence
isomorphic to the one element group G.
Now we assume m ≥ 1. Let α := α1. Then A/α has a central series L1 =
h0A/α = α1/α, α2/α, . . . , αm/α = 1A/αi which is shorter than L, and so we may
apply the induction hypothesis on A/α to obtain ⊕ : A/α × A/α → A/α
and ⊖ : A/α → A/α such that (A/α, ⊕, ⊖, o/α) is isomorphic to the abelian
group associated to A/α, L1, and zero o/α. Furthermore, ⊕ and ⊖ preserve all
congruences in L1 and, for each i ∈ {2, 3, . . . , m}, the relation
(4.1) δi := {(y1, y2, y3, y4) ∈ (A/α)4 (y1, y2) ∈ αi/α,
(cid:0)d(y1, y2, y3), y4(cid:1) ∈ αi−1/α}.
Now let Q := o/α. We choose R to be a set of representatives of A modulo α
with o ∈ R, and we let r : A → R be the function that assigns to each a the
element r(a) ∈ R with (a, r(a)) ∈ α. We define the mapping
ψ : A → A/α × Q
10
by
ERHARD AICHINGER
for a ∈ A. Searching for its inverse, we define
ψ(a) = (ψ1(a), ψ2(a)) := (a/α, a −o r(a))
ϕ : A/α × Q → A
by ϕ(a/α, q) := q+o r(a). We will now prove that ψ is bijective and that ϕ = ψ−1.
To this end, we first compute ϕ(ψ(a)) = ϕ(a/α, a −o r(a)) = (a −o r(a)) +o r(a).
Since L is a central series, we have [α, 1A] = 0A, and therefore Lemma 3.1(3)
yields (a −o r(a)) +o r(a) = a. Second, we let a ∈ A and q ∈ Q and compute
ψ(ϕ(a/α, q)) = ψ(q +o r(a))
Since q +o r(a) ≡α o +o r(a) = r(a), we have
= (cid:0)(q +o r(a))/α, (q +o r(a)) −o r(q +o r(a))(cid:1).
(cid:0)(q +o r(a))/α, (q +o r(a)) − r(q +o r(a))(cid:1) = (cid:0)r(a)/α, (q +o r(a)) −o r(a)(cid:1).
Now applying Lemma 3.1(4), we obtain that the last expression is equal to
(a/α, q). Thus ψ and ϕ are mutually inverse to each other, and hence bijec-
tive. Now we define the functions + : A × A → A and − : A → A by
a + b := ϕ(cid:0)(a/α) ⊕ (b/α), ψ2(a) +o ψ2(b)(cid:1) and
−b := ϕ(cid:0) ⊖ (b/α), −o ψ2(b)(cid:1).
for a, b ∈ A. We now prove that + and − satisfy the required properties and
start with property (i). We consider the algebra
(B, ⊞, ⊟, o′) := (A/α, ⊕, ⊖, o/α) × (Q, +o, −o, o).
Since
ψ(a + b) = (cid:0)(a/α) ⊕ (b/α), ψ2(a) +o ψ2(b)(cid:1)
= (cid:0)ψ1(a) ⊕ ψ1(b), ψ2(a) +o ψ2(b)(cid:1)
= (cid:0)ψ1(a), ψ2(a)(cid:1) ⊞ (cid:0)ψ2(b), ψ2(b)(cid:1)
= ψ(a) ⊞ ψ(b)
for all a, b ∈ A, and since, similarly, ψ(−b) = ⊟(ψ(b)) and ψ(o) = o′, the mapping
ψ is an isomorphism from (A, +, −, o) to (B, ⊞, ⊟, o′) and ψ1 is an epimorphism
from (A, +, −, o) to (A/α, ⊕, ⊖, o/α). Since the kernel of ψ1 is α, we see that α
is a congruence relation of (A, +, −, o), and therefore + and − preserve α1. In
order to show that + and − preserve αi for i ≥ 2, we let i ∈ {2, . . . , m} and
observe that by the construction of ⊕ and ⊖ through the induction hypothesis,
αi/α is a congruence relation of (A/α, ⊕, ⊖, o/α), and therefore its pre-image β
under the homomorphism ψ1, given by
β := {(a, b) ∈ A (cid:0)ψ1(a), ψ1(b)(cid:1) ∈ αi/α},
is a congruence of (A, +, −, o). By its definition, β = {(a, b) ∈ A (a/α, b/α) ∈
αi/α} = αi. Hence, + and − preserve αi. This completes the proof of item (i).
NILPOTENT ALGEBRAS OF PRIME POWER ORDER
11
Next, we prove item (ii). The group (Q, +o, −o, o) is isomorphic to the component
G1 of the abelian group G := Qm
i=1 Gi associated with A, L, and o. The group
associated with A/α, L1 and o/α is isomorphic to Qm
i=2 Gi. Hence (B, ⊞, ⊟, o′) =
(A/α, ⊕, ⊖, o/α) × (Q, +o, −o, o) is isomorphic to G, and therefore (A, +, −, o)
is isomorphic to G. This completes the proof of (ii), and thus item (1) of the
statement of the theorem is proved.
the case i ≥ 2.
Let A′ be the expan-
For (iii), we first consider
Its homomorphic image A′/α is equal to
sion (A, F ∪ {+, −, o}) of A.
(A/α, F ∪ {⊕, ⊖, o/α}). By the construction of ⊕ and ⊖ as functions preserving
the relations in (4.1) and the relational description of centrality (Lemma 3.3), we
have [1A/α, αi/α]A′/α ≤ αi−1/α. Hence [1A, αi]A′ ≤ αi−1, and thus the relational
description of commutators implies that + and − preserve γi. Before proving
that + and − also preserve γ1, which encodes the centrality of α1 = α in A′, we
prove some connections between +, −, +o, −o and the Mal'cev term d. First, we
check that for all a ∈ A and q ∈ Q, we have
(4.2)
q + a = q +o a = a +o q.
For proving the first equality, we compute
q + a = ϕ(cid:0)(q/α) ⊕ (a/α), ψ2(q) +o ψ2(a)(cid:1)
= ϕ(cid:0)(o/α) ⊕ (a/α), (q −o r(q)) +o (a −o r(a))(cid:1)
= ϕ(cid:0)a/α, (q −o o) +o (a −o r(a))(cid:1)
= ϕ(cid:0)a/α, q +o (a −o r(a))(cid:1)
= (cid:0)q +o (a −o r(a))(cid:1) +o r(a)
= q +o (cid:0)(a −o r(a)) +o r(a)(cid:1) (by Lemma 3.1(6))
= q +o a (by Lemma 3.1(3)).
The second equality of (4.2) now follows from Lemma 3.1(5). We will also need
that for all q ∈ Q and a, b ∈ A, we have
(4.3)
d(q +o a, a, b) = q +o b.
This follows from Lemma 3.1(7). Next, we observe that if (v, w) ∈ α, then
w −o v ∈ Q and then v = (w −o v) +o v by Lemma 3.1(3).
With these preparations, we are ready to prove that + preserves γ1. To this end,
let (x1, x2, x3, x4) ∈ γ1 and (y1, y2, y3, y4) ∈ γ1. We have to prove
(4.4)
(cid:0)x1 + y1, x2 + y2, x3 + y3, x4 + y4(cid:1) ∈ γ1.
By the fact that + preserves α1, we obtain (x1 + y1, x2 + y2) ∈ α1. Hence for
completing the proof of (4.4), we have to show
d(x1 + y1, x2 + y2, x3 + y3) = x4 + y4.
12
We set
and compute
ERHARD AICHINGER
v := x2, y := x1 −o x2, w := y2, z := y1 −o y2.
d(cid:0)x1 + y1, x2 + y2, x3 + y3)(cid:1)
= d(cid:0)(y +o v) + (z +o w), v + w, x3 + y3)(cid:1)
= d(cid:0)(y + v) + (z + w)), v + w, x3 + y3(cid:1) (by (4.2))
= d(cid:0)(y + z) + (v + w), v + w, x3 + y3(cid:1) (because (A, +) is abelian)
= d(cid:0)(y +o z) +o (v + w), v + w, x3 + y3(cid:1)(by (4.2))
= (y +o z) +o (x3 + y3) (by (4.3))
= (y + z) + (x3 + y3) (by (4.2))
= (y + x3) + (z + y3) (because (A, +) is abelian)
= (y +o x3) + (z +o y3) (by (4.2))
= d(y +o v, v, x3) + d(z +o w, w, y3) (by (4.3))
= d(x1, x2, x3) + d(y1, y2, y3)
= x4 + y4 (because (x1, x2, x3, x4) ∈ γ1 and (y1, y2, y3, y4) ∈ γ1).
This completes the proof of (4.4), and therefore + preserves γ1. We now show
that − preserves γ1. As a first step, we show that for all a, b, c ∈ A with (a, b) ∈ α,
we have
(4.5)
a −o b = a + (−b) and d(a, b, c) = a + (−b) + c.
From Lemma 3.1(3), we obtain a = (a −o b) +o b, which is equal to (a −o b) + b
by (4.2). Since (A, +, −, o) is a group, we therefore have a + (−b) = a −o b,
establishing the first part of (4.5). For the second part, we observe that
d(a, b, c) = d(a −o b) +o b, b, c)
= (a −o b) +o c (by (4.3))
= (a −o b) + c (by (4.2))
= (a + (−b)) + c (by the first part of (4.5)).
We now take (x1, x2, x3, x4) ∈ γ1, and prove that (−x1, −x2, −x3, −x4) ∈ γ1.
Since − preserves α, (−x1, −x2) ∈ α, and thus it remains to show that
(4.6)
d(−x1, −x2, −x3) = −x4.
We have
d(−x1, −x2, −x3) = (−x1) + x2 + (−x3) (by (4.5))
= −(x1 + (−x2) + x3) (because (A, +, −, o) is an abelian group)
= −d(x1, x2, x3) (by (4.5))
= −x4.
NILPOTENT ALGEBRAS OF PRIME POWER ORDER
13
Hence (−x1, −x2, −x3, −x4) ∈ γ1, and therefore − preserves γ1, which completes
the proof of (ii).
Now to establish (2), we observe that for each i ∈ {1, . . . , m}, the nilpotency of
A implies [1, αi]A ≤ αi−1. Thus (by Lemma 3.3) each fundamental operation
of A preserves γi. Since also + and − preserve γi by item (iii), all fundamental
operations of A′ preserve γi, which implies [1, αi]A′ ≤ αi−1. Hence A′ is nilpotent
of class at most m.
(cid:3)
5. Clones of polynomials
All finitary functions on a finite field are induced by polynomials. When con-
sidering polynomials instead of functions, we can use notions such as degree or
monomial. Such an approach has been used, e.g., in [Kre18]. In this section, we
will study polynomials in the polynomial ring K[xi i ∈ N] = Sn∈N K[x1, . . . , xn]
over countably many variables over some (not necessarily finite) field K. Adapt-
ing [CF09], we define the product of A, B ⊆ K[xi i ∈ N] by
AB = {p(q1, . . . , qn) n ∈ N, p ∈ A ∩ K[x1, . . . , xn], q1, . . . , qn ∈ B}.
Here p(q1, . . . , qn) denotes the polynomial obtained from p by substituting simul-
taneously each variable xi with qi. We say that a subset C of K[xi i ∈ N] is
a clone of polynomials if for each i ∈ N, xi ∈ C and CC ⊆ C. Given a subset
F of K[xi i ∈ N], we use Clop(F ) to denote the clone of polynomials that is
generated by F . By L, we denote the set
L := {
n
X
i=1
aixi n ∈ N, ∀i ∈ {1, . . . , n} : ai ∈ Z}
Hence L = Clop({x1+x2, −x1, 0}), and if F is a nonempty subset of K[xi i ∈ N],
then LF is exactly the subgroup of (K[xi i ∈ N], +, −, 0) generated by F .
We notice that a clone of polynomials is not a clone in the usual sense, since
its elements are polynomials, and not finitary functions on some set. A bridge
between these concepts is provided in the following lemma. For a field K, f ∈
K[x1, . . . , xn] and m ≥ n, we let f K,m be the m-ary function that f induces on
K. For example xK,5
induces the projection (a1, . . . , a5) 7→ a3 on K.
3
Lemma 5.1. Let K be a field, and let C ⊆ K[xi i ∈ N] be a clone of polynomials.
Let
C ′ := {f K,m n ∈ N, f ∈ K[x1, . . . , xn], m ≥ n}.
Then C ′ is a clone on the set K.
Given this close connection, it is not suprising that we may transfer some results
from clone theory to clones of polynomials.
14
ERHARD AICHINGER
Lemma 5.2 (Associativity Lemma, cf.
[CF09]). Let A, B, C ⊆ K[xi i ∈ N],
let L := Clop({x1 + x2, −x1, 0}), and let P := Clop(∅) = {xi i ∈ N}. Then we
have:
(1) (AB)C ⊆ A(BC) ⊆ (A(BP ))C. In particular if BP ⊆ B, then (AB)C =
A(BC).
(2) L(AL) is closed under composition with polynomials in L from both sides;
in other words, L (L(AL)) ⊆ L(AL) and (L(AL)) L ⊆ L(AL).
Proof: The proof of item (1) is straightforward and can be developed along the
lines of [CF09]. Item (2) then follows by observing L(L(AL)) ⊆ (L(LP ))(AL) =
(LL)(AL) = L(AL) and (L(AL))L ⊆ L((AL)L) ⊆ L(A(LL)) = L(AL).
(cid:3)
For a set C ⊆ K[xi i ∈ N], let C (0) := {xi i ∈ N} and for n ∈ N0, C (n+1) =
C (n) ∪ (C C (n)). For f ∈ K[xi i ∈ N], the depth of f with respect to C, denoted
by δC(f ), is the smallest n ∈ N with f ∈ C (n), and undefined if no such n exists.
Lemma 5.3. Let K be a field, and let C ⊆ K[xi i ∈ N]. Then the clone
generated by C, Clop(C), is equal to S{C (n) n ∈ N0}. If M ⊆ K[xi i ∈ N] is
such that {xi i ∈ N} ⊆ M and CM ⊆ M, then Clop(C) ⊆ M.
Proof: Let U := S{C (n) n ∈ N0}. Then C ⊆ C (1) ⊆ U ⊆ Clop(C), and
hence it is sufficient to prove UU ⊆ U. To this end, we prove by induction on
n that C (n)U ⊆ U. This is obvious for n = 0. For the induction step, we let
n ∈ N0, m ∈ N, u ∈ (C (n+1) \ C (n)) ∩ K[x1, . . . , xm], and v1, . . . , vm ∈ U. Since
u ∈ CC (n), there are l ∈ N, r ∈ C and s1, . . . , sl ∈ C n with u = r(s1, . . . , sl). Now
u(v1, . . . , vm) = r(s1(v1, . . . , vm), . . . , sl(v1, . . . , vm)). Then each si(v1, . . . , vm) is
an element of U by the induction hypothesis. Thus there is k ∈ N such that
{si(v1, . . . , vm) i ∈ {1, . . . , l}} ⊆ C (k), and therefore u(v1, . . . , vm) ∈ C (k+1) ⊆ U.
This completes the induction step; therefore UU ⊆ U and U = Clop(C). We
show the second part by proving that for all n ∈ N, C (n) ⊆ M. The induction
basis n = 0 follows from the condition {xi i ∈ N} ⊆ M. For the induction
step, let n ∈ N0. Then C (n+1) = C (n) ∪ (CC (n)) ⊆ M ∪ CM by the induction
hypothesis. Applying the assumption CM ⊆ M, we obtain C (n+1) ⊆ M. Thus
Clop(C) ⊆ M.
(cid:3)
The total degree of of a monomial is defined by
n
n
deg(a
xαi
i ) :=
Y
i=1
αi
X
i=1
for n ∈ N and a ∈ K \{0} and the total degree of a polynomial is the maximum of
the total degrees of its monomials. A polynomial is called homovariate if all of its
monomials contain exactly the same variables. For example, over K = Z7, each
1x3
of the polynomials 5x1x3
2 and 2 is homovariate,
but none of the polynomials x1 + x2, 1 + 3x3
1 is homovariate. For a finite
2x20
1 + x5
4 , x2 + 6x4
2x4 − 2x17
1 x2x3
4 + x6
NILPOTENT ALGEBRAS OF PRIME POWER ORDER
15
subset I of N, the homovariate component HI(p) of the polynomial p with respect
to I is defined as the sum those monomials whose set of variables is I. As an
example, we compute
H{2,3,4}(5x2
and H{2,3}(x2
components. For a set of polynomials F ⊆ K[xi i ∈ N],
3x7
2x3 + 7x2
4
2+x3) = 0. Hence each polynomial is the sum of all of its homovariate
4x5 + 13x6
2x3x5
4 + x1x2x3x4 + 4x3
4) = 7x2
2x3x5
2x8
3x7
4 + 13x6
2x8
Hoc(F ) := {HI(f ) I ⊆ N, f ∈ F }
is the set of the homovariate components of elements of F . We note that by this
definition, for every polynomial f , 0 ∈ Hoc({f }): let j ∈ N be such that xj does
not occur in f . Then H{j}(f ) = 0. We also see that for every f 6= 0, the set
Hoc({f }) has at least two elements.
Lemma 5.4. Let F ⊆ K[xi i ∈ N], and let L := Clop({x1 + x2, −x1, 0}). Then
we have:
(1) F ⊆ L Hoc(F ).
(2) Hoc(F ) ⊆ L(F L).
Proof: For every f ∈ F , we have f = Ph∈Hoc(f ) h, and therefore F ⊆ L Hoc(F ).
For proving the second assertion, we show that for every n ∈ N and every
f ∈ K[x1, . . . , xn], every h ∈ Hoc({f }) satisfies h ∈ L({f }L). We pro-
ceed by induction on the number of homogeneous components of f , i.e., on
Hoc({f }). If Hoc({f }) = 1, then f = 0, therefore Hoc({f }) = {0} and thus
Hoc({f }) ⊆ L({f }L). For the induction step, we assume that Hoc({f }) ≥ 2.
We list all subsets of {1, . . . , n} as (I1, I2, . . . , I2n) in such a way that for all
i, j ∈ {1, 2, . . . , n}, we have Ii ⊆ Ij ⇒ i ≤ j. Now
f =
2n
X
j=1
HIj (f ).
Let k be minimal with HIk(f ) 6= 0. Then, for all j with j > k, Ij 6⊆ Ik, and
hence there is m ∈ Ij such that m 6∈ Ik. We produce f ′ from f by setting all
variables whose indices are not in Ik to 0. Clearly f ′ ∈ {f }L. Since all summands
of f for j > k become 0 by this setting, we have f ′ = HIk(f ). By the induction
hypothesis, every homogeneous component of f − f ′ = P2n
j=k+1 HIj (f ) lies in
L({f − f ′}L). Since f ′ ∈ {f }L, we have f − f ′ ∈ L({f }L), and therefore L({f −
f ′}L) ⊆ L((L({f }L))L) = L({f }L) and so we obtain that {HIk+1, . . . , HI2n } ⊆
L({f }L).
(cid:3)
The following theorem will help us to represent term functions of the algebra A as
sums of absorbing functions. Informally, the idea is the following: Suppose that
we have a universal algebra A = (A, +, −, 0, (fi)i∈I), and let F := {fi i ∈ I}.
To simplify the discussion, we assume that all fi have positive arity. Every term
16
ERHARD AICHINGER
function of A can be represented as by a tree whose leaves are variables or 0, and
whose other nodes are elements of F ∪ {+, −}. Our goal is to move + and − to
the top of the tree. To this end, we transform the tree into a tree whose nodes
are labelled by a new set of functions, H, and by + and −. All functions in H
will be absorbing, and in the new tree, no node labelled by + or − will appear
inside a subtree rooted by an element of H. Deviating from this explanation,
we will not work with the operations of the algebra A directly, but rather with
polynomials over a field whose universe is A. Given a set F of polynomials, we
will obtain a set H of homovariate polynomials such that each polynomial in
Clop(F ∪ {x1 + x2, −x1, 0}) is a sum of compositions of polynomials in H; this
set of sums of compositions is the just the product L C, where C = Clop(H).
Theorem 5.5. Let K be a field, let F ⊆ K[xi
i ∈ N], L := Clop({x1 +
x2, −x1, 0}), and let n ∈ N be such that the total degree of each f ∈ F is at most
n. Then there exists a set H ⊆ K[x1, . . . , xn] of homovariate polynomials such
that
(5.1)
L Clop(H) = Clop(F ∪ {x1 + x2, −x1, 0})
and the total degree of each h ∈ H is at most n.
Proof: In the case F = ∅, we choose H := F obtain that both sides are equal
to the subgroup of (K[xi i ∈ N], +, −, 0) generated by {xi i ∈ N}. Let us
now assume F 6= ∅. We consider the subgroup S = (cid:0)L(F L)(cid:1) ∩ K[x1, . . . , xn] of
(K[x1, . . . , xn], +, −, 0). Let
H := Hoc(S) = Hoc(cid:16)(cid:0)L(F L)(cid:1) ∩ K[x1, . . . , xn](cid:17).
Now each h ∈ H is a homovariate component of a polynomial in L(F L), and has
therefore total degree at most n. We now start to establish (5.1). By Lemma 5.4,
we have H = Hoc(S) ⊆ L(SL). Since S ⊆ L(F L), and since by Lemma 5.2(2),
L(F L) is closed under composition with polynomials from L from both sides, we
obtain L(SL) ⊆ L(F L), and therefore H ⊆ L(F L), and then also
(5.2)
HL ⊆ L(F L).
We will now prove
(5.3)
Clop(H ∪ {x1 + x2, −x1, 0}) = L Clop(H).
⊇: Both sets L and Clop(H) are subsets of Clop(H ∪ {x1 + x2, −x1, 0}). Since
Clop(H) is a clone, their product L Clop(H) is also a subset of Clop(H).
⊆: We use Lemma 5.3 with C := H ∪ {x1 + x2, −x1, 0} and M = L Clop(H), and
observe that {xi i ∈ N} ⊆ M. For proving CM ⊆ M, we observe that using the
Associativity Lemma (Lemma 5.2), we obtain CM = HM ∪{x1+x2, −x1, 0}M ⊆
HM ∪ L(L Clop(H)) = HM ∪ L Clop(H) = HM ∪ M. Hence what remains to
NILPOTENT ALGEBRAS OF PRIME POWER ORDER
17
prove is HM ⊆ M. To this end, we will show that for all t1, . . . , tn ∈ L Clop(H)
and for all g ∈ H, we have
(5.4)
g(t1, . . . , tn) ∈ L Clop(H).
We fix t1, . . . , tn ∈ L Clop(H) and g ∈ H. Each ti is a sum of elements in
Clop(H) ∪ {−p p ∈ Clop(H)}. We collect these summands and thereby find
N ∈ N0, s1, . . . , sN ∈ Clop(H), σ : {1, 2, . . . , N} → {0, 1}, and (mi)n
i=1 with
0 = m0 ≤ m1 ≤ m2 ≤ · · · ≤ mn = N such that for each i ∈ {1, 2, . . . , n}, we
have
mi
ti =
X
j=mi−1+1
(−1)σ(j) sj.
We define e ∈ K[x1, . . . , xN ] by
m1
N
e(x1, . . . , xN ) := g(
which implies
(−1)σ(j) xj, . . . ,
X
j=1
X
j=mn−1+1
(−1)σ(j) xj),
e(s1, . . . , sN ) = g(t1, . . . , tn).
Then e ∈ HL, and thus by (5.2), e ∈ L(F L). We decompose e into its homovari-
ate components and obtain
(5.5)
e = X
I⊆{1,...,N }
HI(e).
Let I ⊆ {1, . . . , N}. We first observe that HI(e) ∈ Hoc(L(F L)), which by
Lemma 5.4 is a subset of L(cid:16)(cid:0)L(F L)(cid:1)L(cid:17). Hence by Lemma 5.2(2), we obtain
(5.6)
HI(e) ∈ L(F L).
We will now show that for each I ⊆ {1, 2, . . . , N}, we have
(5.7)
HI(e) ∈ {0} ∪ Clop(H) ⊆ L Clop(H).
We first consider the case I > n. Since e ∈ L(F L) is obtained by adding and
substituting linear polynomials into polynomials from F , e has total degree at
most n. Hence HI(e) = 0.
In the case I ≤ n, we let π : {1, 2, . . . , N} →
{1, 2, . . . , N} be a bijection such that I ⊆ π[{1, 2, . . . , n}]. Then clearly π−1[I] ⊆
{1, 2, . . . , n}. We define
pI(x1, . . . , xN ) := HI(e) (xπ−1(1), xπ−1(2), . . . , xπ−1(N )).
Then by (5.6), pI ∈ (cid:0)L(F L)(cid:1)L ⊆ L(F L). Since HI(e) (x1, . . . , xN ) contains only
variables xi with i ∈ I, HI(e) (xπ−1(1), . . . , xπ−1(N )) contains only xπ−1(i) with
i ∈ I. Thus pI ∈ K[xπ−1(i)i ∈ I], which implies pI ∈ K[x1, . . . , xn]. Therefore, pI
18
ERHARD AICHINGER
is a homovariate polynomial in (L(F L)) ∩ K[x1, . . . , xn], and thus pI ∈ H. Now
we compute
pI(sπ(1), . . . , sπ(N )) = HI(e) (sπ(π−1(1)), . . . , sπ(π−1(N )))
= HI(e) (s1, . . . , sN ).
Therefore,
Since pI ∈ H, we have pI(sπ(1), . . . , sπ(N )) ∈ Clop(H).
HI(e) (s1, . . . , sN ) ∈ Clop(H), which completes the proof of (5.7). Using (5.5),
we obtain that e(s1, . . . , sN ) ∈ L Clop(H). This completes the proof of (5.4).
Now applying Lemma 5.3 we obtain the "⊆"-inclusion of (5.3).
We finish the proof by establishing that
(5.8)
Clop(F ∪ {x1 + x2, −x1, 0}) = Clop(H ∪ {x1 + x2, −x1, 0}).
For ⊆, we first observe that F ⊆ L Hoc(F ). Each g ∈ Hoc(F ) contains at most n
variables. Replacing these n variables by x1, . . . , xn and undoing this replacement
afterwards, we obtain
(5.9)
Hoc(F ) ⊆ (cid:0)(Hoc(F ) L) ∩ K[x1, . . . , xn](cid:1) L.
The next goal is to prove
(5.10)
(Hoc(F ) L) ∩ K[x1, . . . , xn] ⊆ LH.
By Lemma 5.4, Hoc(F ) ⊆ L(F L), thus Hoc(F ) L ⊆ L(F L). Therefore
(Hoc(F ) L) ∩ K[x1, . . . , xn] ⊆ (L(F L)) ∩ K[x1, . . . , xn].
Now by Lemma 5.4,
(L(F L)) ∩ K[x1, . . . , xn] ⊆ L Hoc(cid:0)L(F L) ∩ K[x1, . . . , xn](cid:1) = LH,
which completes the proof of (5.10). Combining (5.9) and (5.10), we get
F ⊆ L Hoc(F )
⊆ L(cid:16)(cid:0)(Hoc(F ) L) ∩ K[x1, . . . , xn](cid:1)L(cid:17) (by (5.9))
⊆ L ((LH)L) (by (5.10)).
Since both L and H are subsets of Clop(H ∪ {x1 + x2, −x1, 0}), we obtain F ⊆
Clop(H ∪ {x1 + x2, −x1, 0}). From this, the inclusion ⊆ of (5.8) immediately
follows. For the other inclusion in (5.8), we use (5.2) to obtain H ⊆ L(F L).
Hence H ⊆ Clop(F ∪ {x1 + x2, −x1, 0}). This proves (5.8); together with (5.3),
this establishes the claim in (5.1).
(cid:3)
6. Clones of finitary functions
We call a finite algebra A = (A, +, −, 0, (fi)i∈I) an expanded elementary abelian
group if (A, +, −, 0) is a finite abelian group of prime exponent. We call ∗ a field
multiplication on A if K := (A, +, −, 0, ∗) is a field; K is then a field associated
NILPOTENT ALGEBRAS OF PRIME POWER ORDER
19
with A. We do not claim that such a multiplication has any further connection
to the algebra A.
Lemma 6.1. Let K be a field, let n ∈ N, and let p ∈ K[x1, . . . , xn] be such that
pK,n is an absorbing function from K n to K. Then pK,n = (H{1,2,...,n}(p))K,n.
Proof: We proceed by induction on the number k := #{I ⊆ {1, 2, . . . , n}
HI(p) 6= 0} of non-zero homovariate components of p.
If k = 0, then p = 0
and H{1,2,...,n} = 0. If k ≥ 1, we let I be minimal with respect to ⊆ such that
HI(p) 6= 0. If I = {1, 2, . . . , n}, then p = HI(p). If I 6= {1, 2, . . . , n}, we write p
as the sum of its homovariate components, which means
p = X
J⊆{1,2,...,n}
HJ (p).
We set all xi with i
There-
fore, q := p − HI(p) satisfies pK,n = qK,n. By the induction hypothesis
qK,n = H{1,2,...,n}(q)K,n. Now since H{1,2,...,n}(q) = H{1,2,...,n}(p), we obtain
(H{1,2,...,n}(q))K,n = (H{1,2,...,n}(p))K,n.
(cid:3)
6∈ I to 0 and obtain 0 = HI(p)K,n.
Theorem 6.2. Let A = (A, +, −, 0, (fi)i∈I) an expanded elementary abelian
group, and let m, k ∈ N. We assume that for each i ∈ I, the arity of fi is at most
m, and that A is nilpotent of class at most k. Then all absorbing polynomial
functions of A are of essential arity at most (m(A − 1))k−1.
Proof. If A = 1, then all polynomial functions are of essential arity 0, and hence
the claim holds. We will now assume A > 1. We let (αi)i∈N0 be the lower central
series of A defined by α0 := 1A and αi = [1A, αi−1] for i ∈ N, and for i ∈ N0, we
define Ai := 0/αi to be the ideal of A associated with αi; hence A0 = A. Then
by k-nilpotency, Ak = 0. Let K be a field associated with A. For each i ∈ I, we
let mi be the arity of fi, and we choose f ′
i ∈ K[x1, . . . , xm] such that
(f ′
i )
K,n(a1, . . . , am) = fi(a1, . . . , ami)
for all a1, . . . , am ∈ A and degxj (f ′
Then the total degree of f ′
i ) < A for all j ∈ {1, 2, . . . , m}.
i is at most n := m(A − 1). Let
F := {f ′
i i ∈ I}.
Then F ⊆ K[x1, . . . , xm]. We use Theorem 5.5 to obtain a set H ⊆ K[x1, . . . , xn]
of homovariate polynomials such that
(6.1)
L Clop(H) = Clop(F ∪ {x1 + x2, −x1, 0})
and the total degree of each h ∈ H is at most n.
We will show next that for all l ∈ N, for all N ∈ N, and for all p ∈ Clop(H) ∩
K[x1, . . . , xN ], the following property holds:
(6.2)
if p contains at least nl−1 + 1 variables, then pK,N (AN ) ⊆ Al.
20
ERHARD AICHINGER
Seeking a contradiction, we let l ∈ N be minimal such that there is an N ∈ N
and a p ∈ Clop(H) ∩ K[x1, . . . , xN ] that contains at least nl−1 + 1 variables
and pK,N (AN ) 6⊆ Al. Among those p, we choose one of minimal depth δH(p) (as
defined before Lemma 5.3) with respect to H. Since l ∈ N, p contains at least two
variables, and thus p is not a variable and not a constant polynomial. Therefore,
with h ∈ H nonconstant and t1, . . . , tn ∈ Clop(H).
p = h(t1, . . . , tn)
In the case that h contains only one variable, we let xj be this variable. The
polynomial tj must then also contain at least nl−1 + 1 variables. By the
minimality of δH(p), tK,n
(An) ⊆ Al. Since h is homovariate, the function
g1 : aj 7→ hK,n(a1, . . . , an), which is formally defined by
j
g1 = {(aj, hK,n(a1, . . . , an)) a1, . . . , an ∈ A},
satisfies g1(0) = 0. Since hK,n is a term operation of A, we have g1(Al) ⊆ Al,
and therefore pK,N (AN ) ⊆ Al, contradicting the choice of p.
In the case that h contains exactly r variables with 2 ≤ r ≤ n, we let xj1, . . . , xjr
be these variables, and define g2 : Ar → A by
g2 : (aj1, aj2, . . . , ajr) 7→ hK,n(a1, . . . , an)
We first show that for all i1, . . . , ir ∈ N0,
g2(Ai1 × · · · × Air ) ⊆ Amax({i1,...,ir})+1.
(6.3)
To this end, we fix (a1, . . . , ar) ∈ Qr
s=1 Ais. The function g2 is a term function of
A. Let u be such that iu = max({i1, . . . , ir}),and let v ∈ {1, . . . , r} \ {u}. We
define
g3(x, y) := g2(a1, . . . , au−1, x, au+1, . . . , av−1, y, av+1, . . . , ar).
Then g3 is a polynomial function of A. Since h is homovariate, g3(a, 0) =
g3(0, a) = 0 for all a ∈ A. Denoting 0/[αd, αe] simply by [Ad, Ae], Lemma 3.4
implies g3(au, av) ∈ [Aiu, Aiv] ⊆ [Aiu, A] ⊆ Aiu+1. This completes the proof
of (6.3).
Continuing with the proof of (6.2), we first consider the case l = 1. Then by (6.3),
r) ⊆ A1, and therefore hK,n(An) ⊆ A1. Hence pK,N (AN ) ⊆ A1,
g2(Ar) = g2(A0
contradicting the choice of p.
In the case l ≥ 2, one of the polynomials tj1, . . . , tjr contains at least nl−2 + 1
if all contained at most nl−2 variables, also p = h(t1, . . . , tn) would
variables:
contain at most rnl−2 ≤ nl−1 variables, contradicting the choice of p. Let s ∈ N
be such that tjs contains at least nl−2 + 1 variables. By the minimality of l, we see
that tjs(An) ⊆ Al−1. By (6.3), g2(A × · · · × A × Al−1 × A × · · · × A) ⊆ Al, where
Al−1 occurs at place s. Thus pK,N (AN ) ⊆ Al, contradicting again the choice of
p. This completes the proof of (6.2).
NILPOTENT ALGEBRAS OF PRIME POWER ORDER
21
Setting l := k, we see that every p ∈ Clop(H) that contains at least nk−1 + 1
variables induces the constant 0 function on K.
We will now show that all absorbing polynomial functions of A depend on at
most nk−1 variables. To this end, let N > nk−1, and let q be an N-ary absorbing
polynomial function of A. Then there is M ∈ N and there are t ∈ CloM +N (A)
and b1, . . . , bM ∈ A such that
q(a1, . . . , aN ) = t(a1, . . . , aN , b1, . . . , bM )
for all a1, . . . , aN ∈ A. Since t ∈ CloN +M (A), there is a polynomial p ∈ Clop(F ∪
{x1 + x2, −x1, 0}) ∩ K[x1, . . . , xN +M ] such that t = pK,N +M . Then by (6.1),
p ∈ L Clop(H), and therefore, there is l ∈ N such that p = Pl
i=1 pi with pi ∈
Clop(H). We let
:= {i ∈ {1, . . . , l} : pi contains all the variables x1, . . . , xN },
I
J := {1, . . . , l} \ I.
For i ∈ I, pi contains at least nk−1 + 1 variables, and therefore pi induces the
0-function on A. Thus
pK,N +M = X
i∈J
pK,N +M
i
.
For i ∈ J, let
ri(x1, . . . , xN ) := pi(x1, . . . , xN , b1, . . . , bM ) ∈ K[x1, . . . , xN ].
Then we have
q(a1, . . . , aN ) = X
i∈J
rK,N
i
(a1, . . . , aN )
for all a1, . . . , aN ∈ A. Since q is absorbing, Pi∈J ri(x1, . . . , xN ) induces an ab-
sorbing function on K. By Lemma 6.1, Pi∈J ri(x1, . . . , xN ) induces the same
function as H{1,...,N }(Pi∈J ri(x1, . . . , xN )). For each i ∈ J, pi does not contain
all the variables x1, . . . , xN . Thus ri has no monomial that contains all the vari-
ables x1, . . . , xN , and therefore the sum Pi∈J ri(x1, . . . , xN ) does not contain
such a monomial, either. Hence H{1,...,N }(Pi∈J ri(x1, . . . , xN )) = 0. Therefore
Pi∈J ri(x1, . . . , xN ) induces the 0-function on K, which implies q = 0.
(cid:3)
7. Proofs of the main results
Proof of Theorem 1.2: As a nilpotent algebra in a congruence modular variety,
A has a Mal'cev term (see Theorem 6.2 of [FM87] and the remarks after the
proof of Corollary 7.2, cf. [Kea99, Theorem 2.7]). We choose an element o ∈ A
and let L = h0A = α0, α1, · · · , αh = 1Ai be a maximal chain in the congruence
lattice of A. By Lemma 4.1, the abelian group associated with A, L and o
is elementary abelian, and therefore we can use Theorem 4.2 to expand A =
(A, (fi)i∈I) with operations + and − and thereby obtain an h-nilpotent expanded
group V := (A, +, −, 0, (fi)i∈I) with elementary abelian group reduct. Then by
22
ERHARD AICHINGER
s = (cid:0)m(q − 1)(cid:1)h−1
Theorem 6.2, all nonzero absorbing polynomial functions of V are of arity at most
. Hence by Lemma 2.3, V is s-supernilpotent, and then by
Lemma 2.2, its reduct A is also s-supernilpotent. The claim on the free spectrum
now follows from Lemma 2.4.
(cid:3)
Proof of Corollary 1.3: As a nilpotent algebra in a congruence modular variety,
A has a Mal'cev term. We write A = Qn
i=1 Bi with each Bi of prime power
order. By Theorem 1.2, each Bi is si-supernilpotent with si = (m(Bi − 1))hi−1,
where hi is the height of the congruence lattice of Bi. As a nilpotent algebra in
a congruence modular variety, Bi is congruence uniform [FM87, Corollary 7.5],
which implies hi ≤ log2(Bi). Since Bi ≤ A, we have si ≤ s, and therefore
each factor Bi is s-supernilpotent. Hence A is s-supernilpotent. The claim on
the free spectrum again follows from Lemma 2.4.
(cid:3)
Proof of Corollary 1.4: For proving (1), we assume that A has small free
spectrum. Then from Lemma 2.4(4)⇒(2), we obtain that A is nilpotent. By
[Kea99, Theorem 3.14], A is isomorphic to a direct product of algebras of prime
power order. Now Corollary 1.3 yields that A is (cid:0)(m(A − 1))(log2(A)−1)(cid:1)-
supernilpotent and that the free spectrum fA is of the form fA(n) = 2p(n) with
deg(p) ≤ (m(A−1))(log2(A)−1). For proving (2), we assume that A is supernilpo-
tent. Then from Lemma 2.4(1)⇒(4), we obtain that A has small free spectrum.
Now we proceed as in (1).
(cid:3)
8. Acknowledgements
The author thanks Sebastian Kreinecker for numerous discussions on clones of
polynomials and Nebojsa Mudrinski and Jakub Oprsal for discussions on Sec-
tion 2.
References
[AE06] E. Aichinger and J. Ecker, Every (k + 1)-affine complete nilpotent group of class k is
affine complete, Internat. J. Algebra Comput. 16 (2006), no. 2, 259 -- 274.
[Aic06] E. Aichinger, The polynomial functions of certain algebras that are simple modulo their
center, Contributions to general algebra. 17, Heyn, Klagenfurt, 2006, pp. 9 -- 24.
[Aic14]
, On the direct decomposition of nilpotent expanded groups, Comm. Algebra 42
(2014), no. 6, 2651 -- 2662.
[AM07] E. Aichinger and P. Mayr, Polynomial clones on groups of order pq, Acta Math. Hun-
gar. 114 (2007), no. 3, 267 -- 285.
[AM10] E. Aichinger and N. Mudrinski, Some applications of higher commutators in Mal'cev
algebras, Algebra Universalis 63 (2010), no. 4, 367 -- 403.
[AM13]
, On various concepts of nilpotence for expansions of groups, Publ. Math. De-
brecen 83 (2013), no. 4, 583 -- 604.
[AMO18] E. Aichinger, N. Mudrinski, and J. Oprsal, Complexity of term representations of
finitary functions, Internat. J. Algebra Comput. 28 (2018), 1101 -- 1118.
NILPOTENT ALGEBRAS OF PRIME POWER ORDER
23
[BB87] J. Berman and W. J. Blok, Free spectra of nilpotent varieties, Algebra Universalis 24
(1987), no. 3, 279 -- 282.
[BS81] S. Burris and H. P. Sankappanavar, A course in universal algebra, Springer New York
Heidelberg Berlin, 1981.
[Bul01] A. Bulatov, On the number of finite Mal'tsev algebras, Contributions to general algebra,
13 (Velk´e Karlovice, 1999/Dresden, 2000), Heyn, Klagenfurt, 2001, pp. 41 -- 54.
[CF09] M. Couceiro and S. Foldes, Function classes and relational constraints stable under
compositions with clones, Discuss. Math. Gen. Algebra Appl. 29 (2009), no. 2, 109 --
121.
[FM87] R. Freese and R. N. McKenzie, Commutator theory for congruence modular varieties,
London Math. Soc. Lecture Note Ser., vol. 125, Cambridge University Press, 1987.
[Fre83] R. Freese, Subdirectly irreducible algebras in modular varieties, Universal algebra and
lattice theory (Puebla, 1982), Lecture Notes in Math., vol. 1004, Springer, Berlin, 1983,
pp. 142 -- 152.
[Gum83] H. P. Gumm, Geometrical methods in congruence modular algebras, vol. 45, Mem.
Amer. Math. Soc., no. 286, American Mathematical Society, 1983.
[Her79] C. Herrmann, Affine algebras in congruence modular varieties, Acta Sci. Math.
(Szeged) 41 (1979), no. 1-2, 119 -- 125.
[Hig67] G. Higman, The orders of relatively free groups, Proc. Internat. Conf. Theory of Groups
(Canberra, 1965), Gordon and Breach, New York, 1967, pp. 153 -- 165.
[HM88] D. Hobby and R. McKenzie, The structure of finite algebras, Contemporary mathe-
matics, vol. 76, American Mathematical Society, 1988.
[Kea99] K. A. Kearnes, Congruence modular varieties with small free spectra, Algebra Univer-
salis 42 (1999), no. 3, 165 -- 181.
[Kis92] E. W. Kiss, Three remarks on the modular commutator, Algebra Universalis 29 (1992),
no. 4, 455 -- 476.
[Kre18] S. Kreinecker, Closed function sets on groups of prime order, ArXiv e-prints (2018),
1810.09175.
[MMT87] R. N. McKenzie, G. F. McNulty, and W. F. Taylor, Algebras, lattices, varieties, vol-
ume I, Wadsworth & Brooks/Cole Advanced Books & Software, Monterey, California,
1987.
[Moo18] A. Moorhead, Higher commutator theory for congruence modular varieties, J. Algebra
513 (2018), 133-158.
[Smi76] J. D. H. Smith, Mal'cev varieties, Lecture Notes in Math., vol. 554, Springer Verlag
Berlin, 1976.
[VL83] M. R. Vaughan-Lee, Nilpotence in permutable varieties, Universal algebra and lat-
tice theory (Puebla, 1982), Lecture Notes in Math., vol. 1004, Springer, Berlin, 1983,
pp. 293 -- 308.
[Wir19] A. Wires, On Supernilpotent Algebras, Algebra Universalis 80:1 (2019).
Erhard Aichinger, Institut fur Algebra, Johannes Kepler Universitat Linz, 4040
Linz, Austria
E-mail address: [email protected]
URL: http://www.jku.at/algebra
|
1301.0828 | 1 | 1301 | 2013-01-04T16:16:27 | Semilattices of rectangular bands and groups of order two | [
"math.RA"
] | We prove that a semigroup S is a semilattice of rectangular bands and groups of order two if and only if it satisfies the identity x = xxx and for all x,y in S, xyx is in the set {xyyx,yyxxy}. | math.RA | math | Semilattices of Rectangular Bands and Groups of Order Two.
R. A. R. Monzo
Abstract. We prove that a semigroup S is a semilattice of rectangular bands and groups of order two if and only if it
.
x
3
2
2
2
xyx
xy x y x y
,
,
x y
S
x
and
satisfies the identity
AMS Mathematics Subject Classification (2010): 18B40
Key words and phrases: Semilattice of semigroups, Inclusion class, Rees matrix semigroup
1. Introduction
Semigroup theory has a certain symmetric elegance that links it to group theory. For example, a semigroup S
S , Sx
is a group if (and only if) for every x
. In addition, a semigroup S is a union of groups if (and
S
xS
S ,
2
2
x
x
x
S
S
only if) for every x
[3]. The powerful result that a semigroup is a union of groups if and only
if it is a semilattice of completely simple semigroups is so well known that its beauty can almost be overlooked
[3]. In this paper we apply this result to prove that the collection of all semigroups that are semilattices of
semigroups that are either rectangular bands or groups of order two is a semigroup inclusion class [4]. Precisely,
a semigroup is a semilattice of rectangular bands and groups of order two if and only if it satisfies the
S .
x
3
2
2
2
,x y
and
identity
xyx
,
xy x y x y
x
2. Notation, definitions and preliminary results
P , with
Definition. Let G be a group and
j denoted by
I a non-empty set. Let
,P
jp .
I G
:
Let S I G and define a product on S as follows:
. Then S endowed with this
,
,
i a
i ap b
,
,
,
,
j b
j
product is called a Rees I matrix semigroup (over the group G with the sandwich matrix P).
The well known definition above is repeated here because it will be used extensively throughout this paper.
All other terminology and notation can be found in [3]. Other well known results that will be used here follow.
Result 1. [5, Corollary IV.2.8] A semigroup is completely simple if and only if it is isomorphic to a Rees
matrix semigroup.
Result 2. [3,Theorem 4.3] A semigroup S is a union of groups if and only if for every
Sx ,
x
2
x
S S
x
2
.
Result 3. [3, Theorem 4.6] A semigroup is a union of groups if and only if it is a semilattice Y of completely
.
I G
S
Y
simple semigroups
Result 4. [2, Proposition 1] A semilattice Y of completely simple semigroups is a semilattice of rectangular
groups if and only if the product of two idempotents is an idempotent.
Definition. We will denote the collection of all left zero, right zero and rectangular bands by L 0 , R 0 and RB
Sx then 1 x will
respectively and that of all groups of order two by G 2 . If S is a union of groups and
denote the identity element of any group to which x belongs. If G is a group then the identity element of G is
1x or
1
x
denotes the inverse of x in
denoted by 1 or 1 G . Also if x is an element of a semigroup S then
G
any subgroup G of S. A rectangular group is the direct product of a rectangular band and a group.
2
Note that if an element x of a semigroup belongs to two groups, G and H, then
1
1
1
1
x x
x x
x
1
x
x
x
1
1
1 1
1
, and therefore1x is well-defined.
H
G H
G
H
G
H
H
G
G
Similarly,
1
1
1
1
x
1
x x
x
G
G
1x is well-defined.
1
H
1
H
1
G
x
x
x
x
x
1
1
1
H
G
H
H
H
G
G
1
and so
W
1
(1)
Definition. [4] An inclusion class of semigroups is a collection of all semigroups that satisfy a given set of
k number of inclusions as follows:
,
w w
t
1,1
1,2
1,1
t
(k)
w w
W
,
t
k
k
k
,1
some alphabet.
, where the w’s and the t’s are semigroup words over
;
T
1
,
w w
2 ,1
2 ,2
w
k n
,
1
; …;
(2)
, ...,
, ...,
, ...,
, ...,
, ...,
, ...,
W
2
w
2 ,
w
1,
t
1,2
T
2
T
k
,
k m
k
t
1,
k
,2
2 ,
m
2
,
t
,
t
t
t
2 ,1
2 ,2
n
1
n
1
,2
m
1
,
k
,1
] to denote the inclusion class determined by the set
Notation. We write [
W T W T
W T
; ...;
;
1
k
k
2
2
1
of inclusions. If S is a semilattice and
,
then we write if
W T i
1, 2, ...,
k
S
i
i
< if
and .
and
Definition. A semigroup S is a semilattice if S [
,
,
and implies
S
.
x
x
2 ;
xy
yx
]. A semilattice S is a chain if
Definition. A semigroup S is a semilattice Y of semigroups
and for every
,
, S S
Y
Y
S
S
.
if S is a disjoint union of the
Y
S
3. Some inclusion classes of semilattices of rectangular bands and groups of order two.
Theorem 1. The following statements are equivalent:
1. S [
xyx
,
x y
] and
2. S is a chain Y of semigroups
where
S
Y
(1)
S
RB G
2
Y
,
(2) for any <
Y and any
,
,
Sx
, xy
Sy
yx
and
x
(3) for any <
Y with
,
S G ,
2
S
1
.
Proof:
,
2 Let S [
x
]. Then clearly, S [
1
x y
xyx
] and so by Result 2, S is a union of groups and,
.
Y
S
therefore, by Result 3, S is a semilattice Y of completely simple semigroups
x
3
Y . Then either (a) xyx
and
x and yxy
y or (b) xyx
x and yxy
x or
,
Let
Sx
Sy
. So if then either
x . In cases (a) and (d) ,
y and yxy
y or (d) xyx
y and yxy
(c) xyx
case (b) or (c) holds. Therefore,
implies that either or . Hence, Y is a chain.
3
1 .
. Assume that either I
1 or
I G
Y
RB G
S
Y
S
We now show that
. Each
2
j h
i g
j or
,
,
y
x
,
,
and where g and h are arbitrary elements of
, where either i
Then let
and
j and . Thus, (a) holds.
G . Now (b), (c) and (d) all imply that i
1
1
,
,
i p
. Hence,
gp g
i
i
. Then
1
arbitrary element of G . Since g was arbitrary we can let
g p
g p
p
i
i
i
and hence
2g
G . Hence, G is abelian and
G
is the identity element of G and therefore
2
. Let g be an
p g p
i
i
1
p
i
x . So
and so 2
1x
,
i gp g
i
Now
x
,
i g
p
i
p
i
g
x
1
1
,
,
.
1
2
3
Now (a) holds and so xyx
Setting
h
G1g
G
1
Therefore
x . So
,
i gp hp g
i
i
gives G1
. But G is abelian and so
p p
i
i
.
. This implies S
RB
,
i g
,
,
g
and therefore
g
gp hp g
i
i
and1x
gp hp g
i
i
.
2
g p p h
h
i
i
p hp g
i
j
.
and
Now if
I
i
As shown two paragraphs above,
and so (1) is valid.
i g p
1
,
,
is an isomorphism between G and S .
i
S G . We have therefore shown that
2
RB G
2
S
Y
then the mapping
g
G
G
2
and so
Now assume that
1
x x
Then
1
x
and let
,
Y
1
xx
1
1 1
x yxy
and
x y
x
and
Sx
1
1
yxyx
. Then (b) holds and so xyx
Sy
1 1
.
y x
x
yxy
.
If
S G then S is commutative and so
2
2 1 1
2
xyx y
xy x y
x xy
y
x y
xy
x y
. Therefore
S
x
1
z
xy z
x
x yz
z
valid.)
1
x
z
2
yx
1
x
x
and so 1
x
y xyx
x
xy
yx
yx
.
. However, if
Sz
(We have shown that (3) is
then
RB
If S
Hence, x
then
xy
2
3
. Also, from (b),
x
1x
yxy
yx
yx
xy
xy
yxy
xyx
x
. We have shown that (2) is valid and this completes the proof of (1
yx
3
2
x
x
xy
.
xyx
2 ).
2
where (1), (2) and (3) in Theorem 1 are valid.
1 Assume that S is a chain Y of semigroups
Y
S
,
for any
,
x y . Since
, we can assume that .
S
RB G
S
xyx
x y
Y
We wish to show that
2
Case 1. Suppose that < and
S G . Then by (3),
x
S
2
Case 2. Suppose that < and S
xyx
xy x
. Then
RB
Case 3. If < and
G
S
y
. Therefore,
then, by (3),
S
2
Case 4. As in the proof of case 2, <and S
implies yxy
RB
2
.
xy y
y
x yxy x
x yx
,
xyx
x y
, completing the proof of ( 2
We have therefore shown that
Theorem 1.
xyx
xy
y
2
2
S
xyx
x
and so
2
2
x yx x
x
xy x
xyx
S
y
.
y . Then
1 ). This completes the proof of
4
.
.
x
,
Note that we have shown that any S [
x y
xyx
] is a chain Y of rectangular bands, except possibly if Y
has a maximal element and
S G
with S >1. The question arises as to whether the collection of chains
2
of semigroups that are either rectangular bands or groups of order 2 is an inclusion class. In Theorem 5 below
we prove this question in the affirmative when the word “chain” is replaced by “semilattice”.
Theorem 2. The following statements are equivalent:
1. S [
xyx
,
y yx
] and
2. S is a semilattice Y of semigroups
where
Y
S
(1)
2
S
R G
0
2
Y
,
(2) < and
2
S
G
2
implies
2S
1
,
(3)
S
S
2
implies
2S
1
,
(4) for each
x
S/
: S
Y and
2
S
there is a mapping
x such that for any
S
: S
Sx
is a homomorphism, satisfying
Y ,
(4.1) if
2
S
G
2
then for any
g
2S
and
,
x y
,
S
x
y
x
g
1g
on S ,
and
(4.2) if
Y
,
,
2
S
R
0
then
y
y
x
x
x
y
on S and
(4.3) for every
Sx
and
,
Sy
xy
y
y
x
.
x
,
]. Let
Proof:
3
2 Assume that S [
x y . Then
]. First we will prove that S [
1
xyx
y yx
,
S
xy
xy
yx
xy
y and yxy
x or (b) xyx
y and yxy
xy
yx
x or (d) xyx
and yxy
either (a) xyx
.
and yxy
or (c) xyx
] ,
. So in each case we can assume that
2
3
2
3
Note that since S [
,
y yx
xy xy
,
3
4
6
9
3
xyx
Case (a):
xy
yxy
xyx
xyx
yxy
xy
yx
xy
xy
xy
.
xy
xy
xy
.
3
.
3
xy
x yxy
2
x
. But then
x
xy
2
2
x
implies
x
xy
2
x
and so
5
xyx
yxy
yxy
x
xy
Case (b):
xy
Case (c):
xy
xy
Case (d):
xy
2
2
x y
y xy
xyx y
3
xy
yxy
.
yx y
xyx y
xy
2
xy
3
.
2S is a semilattice Y of completely simple semigroups
. We now show that the
Y
By Results 2 and 3,
product of two idempotents of S is an idempotent. Let
E . Then
f then
,e f
,
f
efe
fe
. If efe
S
. So we can assume that
then
2
2
. If efe
fe
f E
efe
ef
e efe
fef
ef
,
e ef
efe f
S
2
2
and this completes the proof that the product of two idempotents is idempotent.
ef
Hence
e .
2S
ef
ef
ef
f
Now by Result 4,
2S
L G R
2S is a semilattice Y of rectangular groups
. We want to show now that each
Y
is either a right-zero semigroup or a group of order two.
2S
2S
j . So
2
Let
j h
i g
,
y yx
,
,
,
,
,
x y
S
2S is a right group G R .
We have shown that an arbitrary rectangular group component of
and so i
. Then
with
and
xyx
x
y
.
G R
Note that, since we have already shown above that
g
g and
2
1
,
h g h
1
satisfies
ghg
G
g
,
.
xy
xy
3
,
G G
2
and so G is commutative and
,
g r
Now let
s . Then
g s , with r
G R
,
. So either R 1 or G 1 . Hence G R
g
2
g
1
,
. Let
x y with
S
x
. First note that S is a null extension of a union of groups. From the proof of Theorem 5 [1],
,
,
g s g r
,
is either a right-zero semigroup or a group of order 2.
. We show that
. Therefore
2
g r
,
S :
x
For
and
,
g s
,
g r
,
g r
S
2S
2S
xy
S
x
y
2
2
2
2
Y we define
2S
2
xy
S S
S
2
x y
xy
. It is then straightforward to show that
(1)
xy
1
xy
y
2
1
xy
y
2
1
xy
1
x
2
xy
1
x
2
xy
1
[ ], with
1
x
2
2
S
and
1
2
y
2
S
.
xy
Assume that
that
.
2S
. Then, since
2S is a semilattice of the semigroups
2S
, it follows from (1)
Y
3
3
x
2S
and
Note that
Therefore S is a semilattice of the semigroups
we have proved part (1) of Theorem 2.
. Therefore
2S
y
2
2
3
, which implies that
3
x y
xy y
x
. It is straightforward to show that
Y
S
S
.
2
S
2
. Thus,
Suppose now that ,
G
2
2
S
2
S
yx
x
S
and so
2
2S
and
G
y
2
1x . Therefore
2S
. Then
1
xyx
,
y yx
. This proves part (2) of Theorem 2.
and so xyx
. But
yx
Suppose that
y
of Theorem 2.
S
S
2
. Let
x
2S
. Then by hypothesis xyx
yx
, and so
1x . This proves part (3)
6
. Note that
xyx y
,
y z
S
. If
S
3
3
2
2
yx
xyx
S
S
, because
x as
we define
Sx
S
: S
y
2
. Let
2
2S
x yx
We proceed with the proof of part (4). For any
2
x yx
S/
: S
x
then
R
S
S
0
xyx
zx
xyxx zx
zx
zx
xyzx
zx
xy
yz
xy zx
x zx
x
. If then either< or< and so, by (2),
Suppose that
2
2
S
G
S
1
2
2
is a homomorphism.
S/x
and so
S
z
y
yz
,
,
that
1
x
x
x
and so
2
We can therefore assume that
S
S
. Then,
x
2
xyzx
y
xyx zx
hypothesis,
,
,
x
x x
x
1
x
x
homomorphism in any case. This proves (4).
S
z
2
2
2
3
2
2
3
2
2
4
. Since, by
G
2
is a
S/x
and so
x
yz
xyx
xzx
y
x
x
z
. This implies
g
and
Let
2
2S
. Then
,
,1 1g
g x
S
,
x y z
,
S
G
2
4
2
4
2
y z
yx yz
1 1
z
1
y z
z
yx xyz
xyzyx
y z
yx yz
x
2
1 1
x zx
1
x z
x x
xzx
1
x x z
x z x
1
and this proves (4.1).
Hence,
x
y
x
g
1g
2
1 1
x z
4
x z
1
z
gxg
z
1
xz
. Also,
x xz x
1 1
g x g
2
1 1 1 1
g x
x
1
x
and
2
2
x zx
2
1 1
z
1
z
.
and
Let
Y
,
,
z
x z
xyx
yxy
y
y
x
yxz
yxzxy
2
,
. Suppose
Sx
S
R
0
3
3
xyzxy
xyzxy
xy
z xy
z
yxzxy
. This proves (4.2).
x
y
. Then
and
Sz
Sy
2
zxy
xyzxyxy
zxy
yxzyx
zxy
Finally, (4.3) follows from the fact that
xy
xy
3
. This completes the proof of
1
2 .
1 Assume that the hypotheses of the “only if” part of Theorem 2 are valid. We first show that the product
2
,
. Using (4.3) we need to show that:
,
defined is associative. Let
Sx
Sy
Sz
. However, since by (4)
z
y
z
x
z
y
x
y
x
z
z
y
x
z
y
y
x
2
S and
2
homomorphisms on
S respectively, this equation becomes:
y
z
y
z
x
y
x
y
z
x
z
z
y
z
y
y
x
z and x are
(2)
,
x
y
x
x
z
x
y
y
7
with – by (4) again -- each of the 6 terms an element of
2
S . If
, equation (2) is valid in this case.
(4.2 ),
y x
x
y
z
z
z
y
2
then, since by hypothesis
R
S
0
So we can assume that
2
. We can therefore assume that
, or else there
S
G
2
such that <, which implies
. [This would imply that equation 2 is
2
,
,
S
exists
1
then, by (4.1 ), each side of equation 2 equals,
, so 2 is valid.
valid.]. But if
x
y
z
1
1
1
Now we need to prove that for
xyxx
xyx
xyx
yx
2
yx
. If
2
then, since
and
,
xyx
,
y yx
R
S
Sx
Sy
0
2
. We can assume, therefore, that
.
S
G
2
,
xyx xyxx
S
2
,
If then either<or< . In either case
and so xyx
xyx yx
S
,
1
, or else by (3),
2S
therefore that . We can also assume that
,
S
xyx yx
y
,
2S is an abelian group. Therefore,
2
y
G
S
y
2
y
x
y
x
2
. This completes the proof that S [
x
. Hence,
2
xyx
y
1
yx
xyx
y
1
x
Also,
since
xy
xy
x
x
2
2
yx
. We can assume
2
. Note that
1
x
x
x
y
,
y yx
].
yx
y
.
So the proof of Theorem 2 is complete.
Corollary 3. The following statements are equivalent:
1. S is a semilattice Y of semigroups
where
Y
S
(1.1)
S
R G
0
2
Y
,
(1.2)
,
, and
Y
S G
2
implies
S
1
(1.3) there is a collection of mappings
: S
x
S /
x
S
satisfying the following properties:
(a) for
,
each
Y
x
S/
: S
is a homomorphism,
S
(b)
,
Y
,
,
,
Sx
Sy
and
S
R
0
imply
y
y
x
x
y
x
on S and
(c) if
S G
2
and
,
x y
S
then 1
x
y
x
on S and
2. S [
xyx
,
y yx
;
x
3
x
].
Theorem 4. The following statements are equivalent:
1. S is a semilattice Y of semigroups
with
Y
S
S
R G
0
2
and
2. S [
xyx
2
2
,
yx y x y
;
x
3
x
].
8
and
Sx
Y .
,
Sy
2
,
xy
xyx
xyx
Proof:
2 For any
1
then
If
R
S
0
3
x yxy xy
xy
2 3
2 3
y y x
y x y
1
1
x R G
2
0
2
xyx
x xy yxy
. Also,
2
x y
3 3
y x
2
2
y x y
2
2
yx y x y
,
2
y
1
yx
2
1
x y
S [
] [
x
xyx
xxy
xyx
].
x
3
3
3
yx
yx
x
x
and therefore S [
x
]. Suppose that
x
G . Then,
S
2
3
2
3
2
1
x xyy
x y
. Suppose that
2
2
1
x y xy
2
. Then,
y
y
yy
1 1
y
1
1
1
y y
1
1
. We have proved therefore that S [
xyx
xy
x
y
2
2
2
2
3
x yy
2
1
2
3
1
y x y
2
2
,
yx y x y
]. Hence,
3
,
2
ef
ef
efe
,
fe fef
1 If
E then, since
,e f
ef efe f
ef
S
It then follows from Results 1, 2 and 3 that S is a semilatticeY of rectangular
1 and so
.Let S
. The fact that S [
2
2
] implies L
S
Y
L G R
groups
xyx
,
yx y x y
. Then
G . Let
], G [
x
x
. Since S [
3
3
S
,
,
g r
,
h s
y
G R
] and
G
x
2
1 implies G
1 . Thus,
2
2
. So R
,
,
h g h s
,
,
h s
,
ghg r
h r
,
hg r
,
1 or G
1 . So
either R
S
R G
, which is what we needed to prove. This completes the proof of
2
0
Theorem 4.
,
fe f ef
,
hg r
xyx
S
fef
ef
x
x
f
2
.
Theorem 5: The following statements are equivalent:
1. S is a semilattice Y of semigroups
with
Y
S
S
RB G
2
Y
and
2. S [
xyx
2
2
2
,
xy x y x y
;
x
3
x
].
and
Sx
.
Sy
3
2
x
xyx
Proof:
2 Clearly,
1
then
If S
RB
So we can assume that
1
1
xyx
xyx
Then,
x
4
Sx . Let
,
with
x
Y
for any
3
2
2
2
yx
xy x
xyyx
xyx
xy x yx
G . Then
1
xy
S
2
2
2
2
2
x y
x
1 1
x
y
y
1
2
2
2
x yx y
x yx
yx y x
2
2
2
2
x
yx
1 1
y y x
1
1
y x
2
2
2
xy x y x y
,
1
xyx y
yx
yx
.
y yx
1
. Therefore,
2
xy
x yx
2
3 2
y y x
y x
2
x y
y
3 4
y x
Hence, S [
xyx
;
yx
yx
].
1
1
1
1
x
x
x
y
x
x
3
2
2
2
2
3
4
2
2
2
2
2
2
2
y x
2
1
2
2
2
2
1
y x x
1
x y x
2
2
y x
1 x y
x xy
yx x
1 1 1 1
y
y
x
3
2
2
.
3
y
2
2
x y
2
y x
yx
2
y
4
.
2
1 Since S [
x
3
x
], by Results 2 and 3, S is a semilattice of completely simple semigroups.
efefe
If efefe
We now show that the product ef of the idempotents e and f is idempotent. We have
6
5
2
2
2
efe fef
,
efe
efe fef
efe
efe
then
3
efe
. So we can assume that
efe f
2
2
fe
efefe
f e
efefe f
fe
fe
f
Then,
Therefore,
. So by Result 3, S is a semilattice Y of rectangular
ef
e ef
e fe f
.
,where
Y
RB
G E
S
Y
groups
fef
efefe f
3
efe
,
f
.
2
fe
ef
2
ef
.
e
,
fe
fe
ef
efefe
ef
2
ef
2
ef
f
f
fe
f
E
fef
fe
2
ef
f
3
9
f
.
,
Y and let
with
h j
i
y
,
,
x
,
1,
S
x y
and
Fix
of G and
,i and
,j are arbitrary elements of E . Then,
h
2
2
2
2
i
h
xyx
,
,
,
,
,
,
,
,
h i
xy x y x y
h j
. So either E has only one element or
h
G .
x
. Hence S is isomorphic to G or to E . In the former case, since S [
3
every
G
],
Gh
x
2
This completes the proof of Theorem 5.
where h is an arbitrary element
2
for
References
[1] Clarke, G.T., Monzo, R.A.R.,: A Generalisation of the Concept of an Inflation of a Semigroup.
Semigroup Forum 60, 172-186 (2000)
[2] Clifford, A.H.: The Structure of Orthodox Unions of Groups. Semigroup Forum 3, 283-337 (1971).
[3] Clifford, A.H., Preston, G.B.: The Algebraic Theory of Semigroups. Math. Surveys of the American Math
Soc., vol. 1. Am. Math. Soc., Providence (1961).
[4] Monzo, R.A.R.: Further results in the theory of generalised inflations of semigroups. Semigroup Forum 76
540-560 (2008).
[5] Petrich, M.: Introduction to Semigroups. Charles E. Merrill Publishing Company. Columbus, Ohio. (1973).
|
1012.5745 | 3 | 1012 | 2011-03-15T01:40:08 | Linear groups over a locally linear division ring | [
"math.RA",
"math.GR"
] | In this paper, in the first we give definitions of some classes of division rings which strictly contain the class of centrally finite division rings. One of our main purpose is to construct non-trivial examples of rings of new defined classes. Further, we study linear groups over division rings of these classes. Our new obtained results generalize precedent results for centrally finite division rings. | math.RA | math | Linear groups over a locally linear division
ring
Bui Xuan Hai∗, Mai Hoang Bien†, and Trinh Thanh Deo‡
August 21, 2018
Abstract
In this paper, in the first we give definitions of some classes of division rings
which strictly contain the class of centrally finite division rings. One of our main
purpose is to construct non-trivial examples of rings of new defined classes. Further,
we study linear groups over division rings of these classes. Our new obtained results
generalize precedent results for centrally finite division rings.
Key words: Division ring; algebraic; strongly algebraic; locally linear; linear groups.
Mathematics Subject Classification 2010: 16K20, 16K40
1
1
0
2
r
a
M
5
1
]
.
A
R
h
t
a
m
[
3
v
5
4
7
5
.
2
1
0
1
:
v
i
X
r
a
∗Faculty of Mathematics and Computer Science, University of Science, VNU-HCM, 227 Nguyen Van
Cu Str., Dist. 5, HCM-City, Vietnam, e-mail: [email protected]
†Department of Basic Sciences, University of Architecture, 196 Pasteur Str., Dist. 1, HCM-City,
Vietnam, e-mail: [email protected]
‡Faculty of Mathematics and Computer Science, University of Science, VNU-HCM, 227 Nguyen Van
Cu Str., Dist. 5, HCM-City, Vietnam, e-mail: [email protected]
1
1
Introduction
Let D be a division ring and F be its center. Recall that D is centrally finite if D is a finite
dimensional vector space over F ; D is locally centrally finite if for every finite subset S of
D, the division subring F (S) of D generated by S over F is a finite dimensional vector
space over F . If a is an element from D, then we have the field extension F ⊆ F (a).
Obviously, a is algebraic over F if and only if this extension is finite. We say that a
non-empty subset S of D is algebraic over F if every element of S is algebraic over F .
A division ring D is algebraic over the center F (briefly, D is algebraic), if every element
of D is algebraic over F . Clearly, the class of algebraic division rings contains the class
of locally centrally finite division rings and the last class contains the class of centrally
It is not difficult to give examples showing that these classes are
finite division rings.
different. In this paper we give the definition of the class of so called strongly algebraic
division rings, which lies between the class of centrally finite division rings and the class
of algebraic division rings. Also, we define the classes of linear and locally linear division
rings. The relation between these classes is explained in Section 2. One of our main
purposes is to construct in Section 2 the non-trivial examples of rings belonging to our
new defined classes. Section 3 is devoted to the study of subgroups in locally linear
division rings.
In Section 4 we shall investigate some properties of linear groups over
division rings of these new defined classes. Our new obtained results generalize precedent
results for centrally finite division rings. The symbols and notation we use in this paper
are standard and they should be found in the literature on subgroups in division rings
and on skew linear groups.
2 Definitions and examples
Definition 2.1. Let D be a division ring.
i) We say that D is a linear division ring if D can be embedded in some centrally finite
division ring.
ii) We say that D is a locally linear division ring if for every finite subset S of D, the
division subring of D generated by S is linear.
Definition 2.2. Let D be a division ring which is algebraic over its center F . We say
that D is strongly algebraic over F if D contains a maximal subfield K satisfying the
following conditions:
i) there exists a subset S of K such that K = F (S),
ii) for each x ∈ D, there are at most finitely many elements y in S such that xy 6= yx.
2
Proposition 2.3. Every centrally finite division ring is strongly algebraic.
Proof. Suppose that D is a centrally finite division ring with center F . By [[3], §7, Th.
4, p.45], there exists a maximal subfield K of D containing F . Since [K : F ] < ∞, there
exists a finite subset S such that K = F (S). Now, by definition we see that D is strongly
algebraic over F .
Theorem 2.4. Let D be a strongly algebraic division ring over its center F and T be a
finite subset of D. Then, there exists some division subring of D containing F which is
centrally finite by itself and contains T .
Proof. Suppose that K = F (S) is a maximal subfield of D such that for every element x
in D, there are at most finitely many elements y in S such that xy 6= yx. Denote by U
the set of elements s in S such that s does not commute at least with one element of T .
Then, U is a finite subset of S and every element of S \ U commutes with all elements
from T . Therefore, F (S \ U) ⊆ Z(K(T )). Note that, since K is a maximal subfield of
D, K is also a maximal subfield of K(T ). It follows that Z(K(T )) ⊆ K. Further, since
K = F (S) = F ((S \ U) ∪ U) = F (S \ U)(U) and U is a finite subset algebraic over F , we
have [K : F (S \ U)] < ∞. Consequently, [K : Z(K(T ))] < ∞. Now, by [[6],(15.8), p.255],
K(T ) is centrally finite and obviously, this fact completes the proof of the theorem.
Corollary 2.5. If a division ring D is strongly algebraic over its center, then D is locally
linear.
Our next purpose in this section is to construct some examples showing the difference
between new defined classes and the precedent classes of division rings. In order to do so,
first, we shall construct division subrings of the ring D = K((G, Φ)), which was introduced
in [2].
∞
Namely, if we denote by G =
Z the direct sum of infinitely many of copies of the
Li=1
additive group (Z, +) of all integers , then G is the set of all infinite sequences of integers
of the form (n1, n2, n3, . . .) with only finitely many non-zeros ni. For any positive integer
i, denote by xi = (0, . . . , 0, 1, 0, . . .) the element of G with 1 in the i-th position and 0
elsewhere. Then G is a free abelian group generated by all xi and every element x in G
is written uniquely in the form
x = Xi∈I
nixi,
with ni ∈ Z and some finite set I.
Now, we define an order in G as follows:
For elements x = (n1, n2, n3, . . .) and y = (m1, m2, m3, . . .) in G, define x < y if either
n1 < m1 or there exists k ∈ N such that n1 = m1, . . . , nk = mk and nk+1 < mk+1. Clearly,
with this order G is a totally ordered set.
3
Suppose that p1 < p2 < . . . < pn < . . . is a sequence of prime numbers and
K = Q(√p1,√p2, . . .) is the subfield of the field R of real numbers generated by Q
and √p1,√p2, . . ., where Q is the field of rational numbers. For any i ∈ N, suppose that
fi : K → K is Q-isomorphism satisfying the following condition:
fi(√pj) = (cid:26) √pj,
−√pi,
if j 6= i;
if j = i.
It is easy to verify that fifj = fjfi,∀i, j ∈ N. Moreover, we have the following lemma,
whose proof can be found in [2]:
Lemma 2.6. Suppose that x ∈ K. Then, fi(x) = x for any i ∈ N if and only if x ∈ Q.
. Clearly Φx ∈
Gal(K/Q) and the map Φ : G → Gal(K/Q), defined by Φ(x) = Φx is a group homomor-
phism. It is easy to prove the following proposition:
For an element x = (n1, n2, ...) = Pi∈I
nixi ∈ G, define Φx := Qi∈I
f ni
i
Proposition 2.7.
i) Φ(xi) = fi for any i ∈ N.
ii) If x = (n1, n2, . . .) ∈ G, then Φx(√pi) = (−1)ni√pi.
For the convenience, from now on we write the operation in G multiplicatively. For G
and K as above, consider formal sums of the form
α = Xx∈G
axx,
ax ∈ K.
For such an α, define the support of α by supp(α) = {x ∈ G : ax 6= 0}. Put
axx, ax ∈ K supp(α) is well-ordered o.
D = K((G, Φ)) := nα = Xx∈G
For α = Px∈G
axx and β = Px∈G
bxx from D, define
(ax + bx)x;
α + β = Xx∈G
α.β = Xz∈G(cid:16) Xxy=z
axΦx(by)(cid:17)z.
In [[6], p.243], it is proved that these operations are well-defined. Moreover, the
following theorem holds:
Theorem 2.8 ([6], Th.(14.21), p.244). D = K((G, Φ)) with the operations, defined as
above is a division ring.
4
Put H := {x2 : x ∈ G} and
Q((H)) := nα = Xx∈G
axx, ax ∈ Q supp(α) is well-ordered o.
that in [[2], Theorem 2.2] it was proved that Q((H)) is the center of D.
It is easy to check that H is a subgroup of G and for every x ∈ H, Φx = IdK. Note
Now, for n ≥ 1, denote by Ln := F (√p1, . . . ,√pn, x1, . . . , xn). Then, Ln ⊆ Ln+1 and
Ln is the division subring generated by all √pi and all xi over F .
∞
L∞ :=
Sn=1
Proposition 2.9. The division ring L∞ satisfies the following conditions:
i) L∞ is locally centrally finite.
ii) L∞ is strongly algebraic over its center.
iii) L∞ is not linear.
Proof. i) For a finite subset S ⊆ L∞, since L∞ =
Ln, there exists some n such that
S ⊆ Ln. By Lemma 3.1 and Theorem 3.3 in [2], Z(Ln) = Z(L∞) = F and Ln is
centrally finite. Consequently, the division subring of L∞ generated by S over F is finite
dimensional over F . Hence L∞ is locally centrally finite.
Sn=1
∞
ii) By i), L∞ is locally centrally finite with the center F . So, L∞ is algebraic over F .
Denote by K∞ = F (√p1,√p2, . . .) the subfield of L∞ generated by √p1,√p2, . . . over F
and suppose that α ∈ CL∞(K∞) \ K∞. Then, there exists some i such that xi appears
in the expression of α as a formal sum. Since x2
i ∈ F , α can be expressed in the form
α = βxi + γ, where β 6= 0 and xi does not appear in the formal expressions of β and
γ. Therefore, √piα − α√pi = 2β√pixi 6= 0. It follows that α does not commute with
√pi ∈ K∞, that is a contradiction. Hence, CL∞(K∞) = K∞.
In view of [[6], Prop.
(15.7),p.254], we have K∞ is a maximal subfield of L∞.
∞
Moreover, we have K∞ = F (S), where S = {√p1, ...,√pn, ...}. Since L∞ =
Ln, for
α ∈ L∞, there exists n such that α ∈ Ln. From the proof of [[2], Lemma 3.1 i)], we see
that every element of Ln can be expressed in the form
a(ε1,...,εn,µ1,...,µn)(√p1)ε1 . . . (√pn)εnxµ1
Sn=1
1 . . . xµn
n ,
a(ε1,...,εn,µ1,...,µn) ∈ F.
α = X0≤εi,µi≤1
Therefore, α commutes with each from elements √pn+1, √pn+2, . . . . By definition we see
that L∞ is strongly algebraic over F .
iii) Suppose that L∞ is linear. This means that, there exists some centrally finite
division ring L such that L∞ ⊆ L. Since xi does not commute with √pi, xi
6∈ Z(L).
We claim that the set B = {xi : i = 1, 2, ...,} is linearly independent over Z(L). Thus,
5
suppose that B is linearly dependent over Z(L). Then, there exists some n such that xn is a
linear combination of x1, . . . , xn−1 over Z(L). Since √pn commutes with x1, . . . , xn−1, √pn
commutes with xn, which is a contradiction. Thus, B is an infinite linearly independent
set over Z(L) that is impossible.
Suppose that α = x−1
1 + x−1
2 + . . . is an infinite sum. Since x−1
1 < x−1
2 < . . . , supp(α)
is well-ordered. Hence α ∈ D. Put
Rn = Ln(α) = F (√p1,√p2,· · · ,√pn, x1, x2,· · · xn, α),∀n ≥ 1;
and
The following theorem holds:
R∞ =
∞
[n=1
Rn.
Proposition 2.10. The division ring R∞ is locally linear and it is not algebraic over its
center.
Proof. First, we prove that Rn is centrally finite for each positive integer n. Consider the
element
Since αn = α − (x−1
1 + x−1
n+1 + x−1
αn = x−1
2 + · · · + x−1
n+2 + · · ·
n ), αn ∈ Rn and
(infinite sum).
F (√p1,√p2, . . . ,√pn, x1, x2, . . . xn, α) = F (√p1,√p2, . . . ,√pn, x1, x2, . . . xn, αn).
Note that αn commutes with all √pi and all xi (for i = 1, 2, ..., n). Therefore
Rn = F (√p1,√p2,· · · ,√pn, x1, x2,· · · xn, αn)
= F (αn)(√p1,√p2,· · · ,√pn, x1, x2,· · · xn).
In combination with the equalities
(√pi)2 = pi, x2
i ∈ F,√pixj = xj√pi, i 6= j,√pixi = −xi√pi,
it follows that every element β of Rn can be written in the form
a(ε1,...,εn,µ1,...,µn)(√p1)ε1 . . . (√pn)εnxµ1
1 . . . xµn
n ,
β = X0≤εi,µi≤1
where
a(ε1,...,εnµ1,...,µn) ∈ F (αn).
Hence Rn is a vector space over F (αn) having the finite set Bn which consists of the
products
(√p1)ε1 . . . (√pn)εnxµ1
1 . . . xµn
n , 0 ≤ εi, µi ≤ 1
6
as a base. Thus, Rn is a finite dimensional vector space over F (αn). Since αn commutes
with all √pi and all xi, F (αn) ⊆ Z(Rn). It follows that dimZ(Rn) Rn ≤ dimF (αn) Rn < ∞
and consequently, Rn is centrally finite.
For any finite subset S ⊆ R∞, there exists n such that S ⊆ Rn. Therefore, the division
subring of R∞, generated by S over F is contained in Rn, which is centrally finite. Thus,
R∞ is locally linear.
On the other hand, by [[2], Theorem 3.3], we have Z(L∞) = F . So, Z(R∞) =
Z(L∞(α)) = F . The proof of [[2], Theorem 3.1] shows that α is not algebraic over F ,
hence R∞ is not algebraic over F .
According to the theorem above, the division ring R∞ is locally linear. However, it is
not algebraic and consequently it is not strongly algebraic over its center.
Finally, we strongly believe that there exists a division ring which is algebraic but it is
not strongly algebraic over its center. However, at the present time, we do not have any
counterexample to this problem. On the other hand, we believe that any division subring
of a centrally finite division ring is itself centrally finite. In fact, we propose the following
conjecture:
Conjecture. Any division subring of a centrally finite division ring is itself centrally
finite.
3 Locally linear division rings
Theorem 3.1. If D is a locally linear division ring, then Z(D′) is a torsion group.
Proof. By [[4], Proposition 2.1], Z(D′) = D′ ∩ F . For any x ∈ Z(D′), there exists some
positive integer n and some ai, bi ∈ D∗, 1 ≤ i ≤ n, such that
· · · anbna−1
x = a1b1a−1
1 a2b2a−1
n b−1
n .
1 b−1
2 b−1
2
Set S := {ai, bi : 1 ≤ i ≤ n}. Since D is locally linear, there exists a centrally finite
division ring L such that P (S) is embedded in L. Clearly, we can suppose that S ⊆ L.
Put K = Z(L), L1 = K(S) and F1 = Z(L1). Since K ⊆ L1 ⊆ L and dimKL < ∞, it
follows that K ⊆ F1 and dimKL1 < ∞. Hence n = dimF1 L1 < ∞. On the other hand,
since x ∈ F, x commutes with every element of S. Therefore, x commutes with every
element of L1 = K(S), and consequently, x ∈ F1. So,
1 a2b2a−1
xn = NL1/F1(x) = NL1/F1(a1b1a−1
n ) = 1.
n b−1
2 b−1
2
1 b−1
· · · anbna−1
Thus, x is torsion.
The following corollary carries over one of Herstein's result [[5], Theorem 2] to the
case of locally linear division rings.
7
Corollary 3.2. Let D be a locally linear division ring with the center F . If for any a, b ∈
D∗, there exists a positive integer n = nab depending on a and b, such that (aba−1b−1)n ∈
F , then D is commutative.
Proof. By Theorem 3.1, for any a, b ∈ D∗, aba−1b−1 is torsion . By [[5], Theorem 1], D is
commutative.
Using Theorem 3.1, it is easy to prove that Conjecture 3 in [5] is true for locally linear
division rings. In fact, we have the following result:
Theorem 3.3. Let D be a locally linear division ring with the center F and N be a
subnormal subgroup of D∗. If N is radical over F , then N is central, i.e. N is contained
in F .
Proof. Consider the subgroup N ′ = [N, N] ⊆ D′ and suppose that x ∈ N ′. Since N is
radical over F , there exists some positive integer n such that xn ∈ F . Hence xn ∈ F ∩D′ =
Z(D′). By Theorem 3.1, xn is torsion, and consequently, x is torsion too. Moreover, since
N is subnormal in D∗, so is N ′. Hence, by [[5], Th. 8], N ′ ⊆ F . Thus, N is solvable, and
by [[9], 14.4.4, p. 440], N ⊆ F .
Now, we study subgroups of D∗, that are radical over some subring of D. To prove
the next theorem we need the following useful property of locally linear division rings.
Lemma 3.4. Let D be a locally linear division ring with the center F and N be a sub-
normal subgroup of D∗. If for every elements x, y ∈ N , there exists some positive integer
nxy such that xnxy y = yxnxy , then N ⊆ F .
Proof. Replacing K := F (x, y) by K := P (x, y) (P is the prime subfield of F ) in the
proof of [[4], Lem. 3.1], we can obtain similar proof of this lemma for the case of locally
linear division ring instead of the case of division rings of type 2.
Using this lemma, by the same way as in [4], we can prove the following non-trivial
theorem whose proof is identified with the proof of Theorem 3.1 in [4] for division rings
of type 2.
Theorem 3.5. Let D be a locally linear division ring with the center F, K be a proper
division subring of D and suppose that N is a normal subgroup of D∗. If N is radical
over K, then N ⊆ F .
4 Finitely generated skew linear groups
The following result generalizes Theorem 1 in [7].
8
Theorem 4.1. Let D be a division ring strongly algebraic over its center F and N be a
subnormal subgroup of D∗. If N is finitely generated, then N is central.
Proof. Since N is finitely generated, by Theorem 2.4, there exists some centrally finite
division subring L of D such that N ⊆ L. By [[7], Th.1], N ⊆ Z(L). Consequently, N is
abelian. Now, by [[9], 14.4.4, p. 440], N ⊆ F .
In the following we identify F ∗ with F ∗I := {αI α ∈ F ∗}, where I denotes the identity
matrix in GLn(D).
Theorem 4.2. Let D be a division ring strongly algebraic over its center F and N be
a infinite subnormal subgroup of GLn(D) with n ≥ 1. If N is finitely generated, then
N ⊆ F .
Proof. If n = 1, then the result follows from Theorem 4.1.
Suppose that n > 1 and N is non-central. Then, by [[8], Th.4], SLn(D) ⊆ N. So, N
is normal in GLn(D). Suppose that N is generated by matrices A1, A2, ..., Ak in GLn(D)
and T is the set of all coefficients of all Aj. By Theorem 2.4, there exists some centrally
finite division subring L of D containing T . It follows that N is a normal finitely generated
subgroup of GLn(L). By [[1], Th.5], N ⊆ Z(GLn(L)). In particular, N is abelian and
consequently, SLn(D) is abelian, that is a contradiction.
Lemma 4.3. Let D be a division ring with center F and n ≥ 1. Then, Z(SLn(D)) is a
torsion group if and only if Z(D′) is a torsion group.
Proof. The case n = 1 is clear. So, we can assume that n ≥ 2. By [[3], §21, Th.1, p.140],
Z(SLn(D)) = (cid:8)dId ∈ F ∗ and dn ∈ D′(cid:9).
If Z(SLn(D)) is a torsion group, then, for any d ∈ Z(D′) = D′ ∩ F , dI ∈ Z(SLn(D)).
It follows that d is periodic. Conversely, if Z(D′) is a torsion group, then, for any A ∈
Z(SLn(D)), A = dI for some d ∈ F ∗ such that dn ∈ D′. It follows that dn is periodic.
Therefore, A is periodic.
Now we can prove the following theorem, which shows that if D is a non-commutative
division ring which strongly algebraic over its center, then there are no finitely generated
subgroups of GLn(D), containing F ∗.
Theorem 4.4. Let D be a non-commutative division ring which is strongly algebraic over
its center F and N be a subgroup of GLn(D) containing F ∗, n ≥ 1. Then N is not finitely
generated.
9
Proof. Recall that if D is strongly algebraic over its center, then Z(D′) is a torsion group
(see Corollary 2.5 and Theorem 3.1). Therefore, by Lemma 4.3, Z(SLn(D)) is a torsion
group.
Suppose that there is a finitely generated subgroup N of GLn(D) containing F ∗. Then,
in virtue of [[9], 5.5.8, p. 113], F ∗N ′/N ′ is a finitely generated abelian group, where N ′
denotes the derived subgroup of N.
Case 1: char(D) = 0.
Then, F contains the field Q of rational numbers and it follows that Q∗I/(Q∗I ∩N ′) ≃
Q∗N ′/N ′. Since F ∗N ′/N ′ is finitely generated, Q∗N ′/N ′ is finitely generated and conse-
quently Q∗I/(Q∗I ∩ N ′) is finitely generated. Considering an arbitrary A ∈ Q∗I ∩ N ′.
Then A ∈ F ∗I ∩ SLn(D) ⊆ Z(SLn(D)). Therefore A is periodic. Since A ∈ Q∗I, we
have A = dI for some d ∈ Q∗. It follows that d = ±1. Thus, Q∗I ∩ N ′ is finite. Since
Q∗I/(Q∗I ∩ N ′) is finitely generated, Q∗I is finitely generated. Therefore Q∗ is finitely
generated, that is impossible.
Case 2: char(D) = p > 0.
Denote by Fp the prime subfield of F , we shall prove that F is algebraic over Fp.
In fact, suppose that u ∈ F and u is transcendental over Fp. Put K := Fp(u), then
the group K ∗I/(K ∗I ∩ N ′) considered as a subgroup of F ∗N ′/N ′ is finitely generated.
Considering an arbitrary A ∈ K ∗I∩N ′, we have A = (f (u)/g(u))I for some f (X), g(X) ∈
Fp[X], ((f (X), g(X)) = 1 and g(u) 6= 0. As mentioned above, we have f (u)s/g(u)s = 1
for some positive integer s. Since u is transcendental over Fp, f (u)/g(u) ∈ Fp. Therefore,
It follows that K ∗ is
K ∗I ∩ N ′ is finite and consequently, K ∗I is finitely generated.
finitely generated. This contradicts to [[4], Lem. 2.2], stating that K is finite. Hence F is
algebraic over Fp and it follows that D is algebraic over Fp. Now, in virtue of Jacobson's
Theorem [8, (13.11), p. 219], D is commutative, which completes the proof.
From Theorem 4.4 we get the following result, which generalizes Theorem 1 in [1]:
Corollary 4.5. Let D be a division ring which strongly algebraic over its center. If the
group GLn(D) is finitely generated, then D is commutative.
If M is a maximal finitely generated subgroup of GLn(D), then GLn(D) is finitely
generated. So, the next result follows immediately from Corollary 4.5.
Corollary 4.6. Let D be a division ring which strongly algebraic over its center. If the
group GLn(D) has a maximal finitely generated subgroup, then D is commutative.
By the same way as in the proof of Theorem 4.4, we obtain the following corollary.
10
Corollary 4.7. Let D be a non-commutative division ring which strongly algebraic over
its center F and S is a subgroup of GLn(D). If N = F ∗S, then N/N ′ is not finitely
generated.
Proof. Suppose that N/N ′ is finitely generated. Since N ′ = S ′ and F ∗I/(F ∗I ∩ S ′) ≃
F ∗S ′/S ′, F ∗I/(F ∗I∩S ′) is a finitely generated abelian group. Now, by the same arguments
as in the proof of Theorem 4.4, we conclude that D is commutative.
The following result follows immediately from Corollary 4.7.
Corollary 4.8. Let D be a non-commutative division ring which strongly algebraic over
its center. Then, GLn(D)/SLn(D) is not finitely generated.
References
[1] S. Akbari and M. Mahdavi-Hezavehi, Normal subgroups of GLn(D) are not finitely
generated, Proc. of the Amer. Math. Soc., 128 (2000), No.6, p. 1627-1632.
[2] Trinh Thanh Deo, Mai Hoang Bien, Bui Xuan Hai, On one Laurent series ring over
an extension of Q, arXiv:1009.4537.
[3] P. Draxl, Skew fields, London Math. Soc., Lecture Note Series 81 (1983), Cambridge
Univ. Press.
[4] Bui Xuan Hai, Trinh Thanh Deo, Mai Hoang Bien, On subgroups in division rings
of type 2, arXiv:1009.4537v2.
[5] I. N. Herstein, Multiplicative commutators in division rings, Israel Journal of Math.,
Vol. 31, No. 2 (1978), 180-188.
[6] T.Y. Lam, A First course in non-commutative rings, GTM 131 (1991), Springer-
Verlag.
[7] M. Mahdavi-Hezavehi, M.G. Mahmudi, and S. Yasamin, Finitely generated subnor-
mal subgroups of GLn(D) are central, Journal of Algebra 255, 517-521 (2000).
[8] M. Mahdavi-Hezavehi and S. Akbari, Some special subgroups of GLn(D), Algebra
Colloq., 5, No.4, 1998, 361-370.
[9] W. R. Scott, Group theory, Dover Publication, INC, 1988.
11
|
1709.04717 | 4 | 1709 | 2018-12-31T06:20:17 | Singular Degenerations of Lie Supergroups of Type $D(2,1;a)$ | [
"math.RA",
"math.DG"
] | The complex Lie superalgebras $\mathfrak{g}$ of type $D(2,1;a)$ - also denoted by $\mathfrak{osp}(4,2;a) $ - are usually considered for "non-singular" values of the parameter $a$, for which they are simple. In this paper we introduce five suitable integral forms of $\mathfrak{g}$, that are well-defined at singular values too, giving rise to "singular specializations" that are no longer simple: this extends the family of simple objects of type $D(2,1;a)$ in five different ways. The resulting five families coincide for general values of $a$, but are different at "singular" ones: here they provide non-simple Lie superalgebras, whose structure we describe explicitly. We also perform the parallel construction for complex Lie supergroups and describe their singular specializations (or "degenerations") at singular values of $a$. Although one may work with a single complex parameter $a$, in order to stress the overall $\mathfrak{S}_3$-symmetry of the whole situation, we shall work (following Kaplansky) with a two-dimensional parameter $\boldsymbol{\sigma} = (\sigma_1,\sigma_2,\sigma_3)$ ranging in the complex affine plane $\sigma_1 + \sigma_2 + \sigma_3 = 0$. | math.RA | math |
Symmetry, Integrability and Geometry: Methods and Applications
SIGMA 14 (2018), 137, 36 pages
Singular Degenerations of
Lie Supergroups of Type D(2, 1; a)
Kenji IOHARA † and Fabio GAVARINI ‡
† Univ Lyon, Universit´e Claude Bernard Lyon 1, CNRS UMR 5208, Institut Camille Jordan,
43 Boulevard du 11 Novembre 1918, F 69622 Villeurbanne Cedex, France
E-mail: [email protected]
URL: http://math.univ-lyon1.fr/~iohara/
‡ Dipartimento di Matematica, Universit`a di Roma "Tor Vergata",
Via della ricerca scientifica 1, I-00133 Roma, Italy
E-mail: [email protected]
Received October 31, 2017, in final form December 11, 2018; Published online December 31, 2018
https://doi.org/10.3842/SIGMA.2018.137
Abstract. The complex Lie superalgebras g of type D(2, 1; a) -- also denoted by osp(4, 2; a) --
are usually considered for "non-singular" values of the parameter a, for which they are sim-
ple. In this paper we introduce five suitable integral forms of g, that are well-defined at
singular values too, giving rise to "singular specializations" that are no longer simple: this
extends the family of simple objects of type D(2, 1; a) in five different ways. The result-
ing five families coincide for general values of a, but are different at "singular" ones: here
they provide non-simple Lie superalgebras, whose structure we describe explicitly. We also
perform the parallel construction for complex Lie supergroups and describe their singular
specializations (or "degenerations") at singular values of a. Although one may work with
a single complex parameter a, in order to stress the overall S3-symmetry of the whole situa-
tion, we shall work (following Kaplansky) with a two-dimensional parameter σ = (σ1, σ2, σ3)
ranging in the complex affine plane σ1 + σ2 + σ3 = 0.
Key words: Lie superalgebras; Lie supergroups; singular degenerations; contractions
2010 Mathematics Subject Classification: 14A22; 17B20; 13D10
Contents
1 Introduction
2 Preliminaries
2.1 Basic superobjects . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
2.2 Lie superalgebras . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
2.3 Lie supergroups . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
2.4 Super Harish-Chandra pairs and Lie supergroups . . . . . . . . . . . . . . . . . .
3 Lie superalgebras of type D(2, 1; σ)
3.1 Construction of gσ (after Kaplansky) . . . . . . . . . . . . . . . . . . . . . . . . .
3.2 Kac' realization . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
3.3 Bases of gσ
. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
4 Integral forms & degenerations for Lie superalgebras of type D(2, 1; σ)
4.1 First family: the Lie superalgebras g(σ)
. . . . . . . . . . . . . . . . . . . . . . .
4.2 Second family: the Lie superalgebras g(cid:48)(σ) . . . . . . . . . . . . . . . . . . . . . .
4.3 Third family: the Lie superalgebras g(cid:48)(cid:48)(σ) . . . . . . . . . . . . . . . . . . . . . .
2
3
4
4
4
6
7
8
9
12
14
15
16
18
2
4.4 Degenerations from contractions: the(cid:98)g(σ)'s and the(cid:98)g(cid:48)(σ)'s . . . . . . . . . . . .
K. Iohara and F. Gavarini
5 Lie supergroups of type D(2, 1; σ): presentations and degenerations
5.1 First family: the Lie supergroups Gσ . . . . . . . . . . . . . . . . . . . . . . . . .
5.2 Second family: the Lie supergroups G(cid:48)
. . . . . . . . . . . . . . . . . . . . . . .
σ
5.3 Third family: the Lie supergroups G(cid:48)(cid:48)
σ . . . . . . . . . . . . . . . . . . . . . . . .
5.4 Lie supergroups from contractions: the family of the (cid:98)Gσ's . . . . . . . . . . . . .
5.5 Lie supergroups from contractions: the family of the (cid:98)G(cid:48)
σ's . . . . . . . . . . . . .
References
1
Introduction
20
23
23
25
27
30
32
35
In the classification of simple finite-dimensional complex Lie superalgebras -- due to Kac
(cf. [12]) -- a special one-parameter family occurs, whose elements ga depend on a parameter
a ∈ C \ {0,−1}. These are "generically non-isomorphic", and all isomorphisms between them
are encoded in a free action of the symmetric group S3 on the family {ga}a∈C\{0,−1}. It was
used to define this Lie superalgebra
pointed out in [12] that the Cartan matrix A =
(cid:16) 2 −1 0
(cid:17)
−1 0 a
0 −1 2
For any a ∈ {1,−2,−1/2} one has g1
had already appeared in [18], as a Cartan matrix of a one-parameter family of 16-dimensional
simple Lie algebras over a field k of characteristic 2 with a ∈ k \ {0, 1}.
∼= osp(4, 2), which is of type D(2, 1): thus Kac called
each ga to be "of type D(2, 1; a)" -- while D(m, n) is the type of the orthosymplectic Lie su-
peralgebra osp(2m, 2n). For the same reason, some authors, for example [3] -- cf. also [2] --
use instead notation osp(4, 2; a). By general theory, one can complete each of the (simple)
Lie superalgebras ga and form a so-called super Harish-Chandra pair: and then one associates
to the latter a corresponding complex Lie supergroup, say Ga, whose tangent Lie superalge-
bra is ga -- as prescribed in Kac' classification of simple algebraic supergroups, cf. [11]. All
these Ga's form a family {Ga}a∈C\{0,−1}, which bears a free S3-action that induces the S3-
action on {ga}a∈C\{0,−1}. The starting point of the present paper is the following question: can
we "take the limit" (in some sense) of ga for a approaching to the "singular values" a = 0 and
a = −1? And if yes, what is the structure of the resulting "limit" Lie superalgebra? Similarly,
we raise the same questions for the family of the supergroups Ga.
In this article, we show that there are several ways to answer, in the positive, these questions.
In fact, we present five possible ways to complete the family of simple Lie superalgebras D(2, 1; a)
with additional Lie superalgebras for the "singular values" a ∈ {0,−1}. Each one of these new,
extra objects can be thought of as a "limit" of the older ones; however, the existence of different
options show that such "limits" have no intrinsic meaning, but strongly depend on some choice --
roughly, on "how you approach the singular point". For each of these choices, the corresponding
new objects that are "limits" of the (original) simple Lie superalgebras D(2, 1; a) happen to be
non-simple, and we describe explicitly their structure, which is different for the different choices.
Therefore, we extend the old family {ga = osp(4, 2; a)}a∈C\{0,−1} of simple Lie superalgebras
to five larger families, indexed by the points of P1(C) ∪ {∗}, whose elements at "non-singular
values" a ∈ {0,−1,∞,∗} are non-simple -- which is why we call them "degenerations" -- and
(when comparing one family with a different one) non-isomorphic.
By the way, our analysis is by no means exhaustive: one can still provide further ways to
complete the family of the simple ga's (for non-singular values of a) by adding some extra
objects at singular values of a, right in the same spirit but with different outcomes. Our goal
here is only to explain the existence and non-uniqueness of such constructions. A few words
about our construction. First, instead of working with Lie superalgebras ga indexed by a single
parameter a ∈ C \ {0,−1} -- later extended to a ∈ C -- we rather deal with a multiparameter
Singular Degenerations of Lie Supergroups of Type D(2, 1; a)
3
σ ∈ V := (cid:8)(σ1, σ2, σ3) ∈ C3(cid:12)(cid:12)(cid:80)
yields a full family of Lie superalgebras(cid:8)gσ
the "general locus" V × := V \(cid:0)(cid:83)3
in addition, the gσ's are well-defined also at singular values σ ∈ V ∩(cid:0)(cid:83)3
Kaplansky, cf. [13]; see also [15] -- that for each σ ∈ V provides a Lie superalgebra gσ: this
σ∈V , forming a bundle over V , naturally endowed
with an action of the group G := C× × S3 via Lie superalgebra isomorphisms. For each σ in
∼= ga for some a ∈ C \ {0,−1} so the
original family {ga = osp(4, 2; a)}a∈C\{0,−1} of simple Lie superalgebras is taken into account;
i σi = 0(cid:9). The starting point is a construction -- due to
i=1{σi = 0}(cid:1) we have gσ
i=1{σi = 0}(cid:1), but there
(cid:9)
they are non-simple instead.
Thus Kaplansky's family of Lie superalgebras provides a first solution to our problem. In
addition, we re-visit this construction and devise five recipes to construct similar families, as
follows. For σ ∈ V ×, we fix in gσ a particular C-basis, call it B, in such a way that the structure
constants are polynomials in σ. When we replace σ = (σ1, σ2, σ3) with a formal parameter
x = (x1, x2, x3), the previous multiplication table defines a Lie superalgebra structure on the
free C[x]-module with basis B, denoted by gB (x). Then for each σ = (σ1, σ2, σ3) ∈ V the
quotient gB (σ) := gB (x)/(xi − σi)i=1,2,3gB (x) is a well-defined complex Lie superalgebra, such
that gB (σ) ∼= gσ for σ ∈ V ×; thus we get a whole family {gB (σ)}σ∈V as requested, that actually
i=1{σi = 0}(cid:1) these families present different (non-isomorphic) non-simple Lie
locus" V ∩(cid:0)(cid:83)3
depends on the choice of the basis B. We present five explicit examples that give rise to five
different outcomes -- one being Kaplansky's family.
Indeed, at each point of the "singular
superalgebras, that we describe in detail. As a second contribution, we perform a parallel
construction at the level of Lie supergroups: namely, for each σ ∈ V we "complete" each Lie
superalgebra gB (σ) to form a super Harish-Chandra pair, and then take the corresponding
(complex holomorphic) Lie supergroup. This yields a family {GB (σ)}σ∈V of Lie supergroups,
with Gσ isomorphic to Ga for a suitable a ∈ C\{0,−1} for non-singular values of σ, while GB (σ)
is not simple for singular values instead; moreover, the group G := C× × S3 freely acts on this
family via Lie supergroup isomorphisms. In other words, we complete the "old" family of the
simple Lie supergroups Ga's (isomorphic to suitable Gσ's) by suitably adding new, non-simple
Lie supergroups at singular values of σ. The construction depends on B, and with our five,
previously fixed choices we find five different families: for each of them, we describe explicitly
the non-simple supergroups Gσ at singular values of σ -- which are referred to as "degenerations"
of the (previously known, simple) Ga's.
This analysis might be reformulated in the language of deformation theory of supermanifold --
e.g., as treated in [17]. However, this goes beyond the scope of the present article. This article is
organized as follows. In Section 2, we briefly recall the basic algebraic background necessary for
this work, in particular, some language about supermathematics. In Section 3, we introduce our
Lie superalgebras gσ = osp(4, 2; σ). Several integral forms of the Lie superalgebra gσ are intro-
duced in Section 4. In particular, as an application, the structure of their singular degenerations
is studied in detail (Theorems 4.1, 4.2, 4.3, 4.4 and 4.5). Section 5 is the last highlight of this
paper: we introduce and analyze the Lie supergroups whose Lie superalgebras are studied in
Section 4, and we describe the (non-simple) structure of their degenerations -- i.e., the member
of the families at singular values of σ (Theorems 5.1, 5.2, 5.3, 5.4 and 5.5).
As the main objects treated in this article have many special features, most of the above
descriptions are given in a down-to-earth manner, so that even the readers who are not familiar
with the subject could follow easily our exposition.
2 Preliminaries
In this section, we recall the notions and language of Lie superalgebras and Lie supergroups.
Our purpose is to fix the terminology, but everything indeed is standard matter.
4
K. Iohara and F. Gavarini
2.1 Basic superobjects
All throughout the paper, we work over the field C of complex numbers (nevertheless, immedi-
ate generalizations are possible), unless otherwise stated. By C-supermodule, or C-super vector
space, any C-module V endowed with a Z2-grading V = V¯0 ⊕ V¯1, where Z2 = {¯0, ¯1} is the group
with two elements. Then V¯0 and its elements are called even, while V¯1 and its elements odd. By
x(∈ Z2) we denote the parity of any non-zero homogeneous element, defined by the condition
x ∈ Vx.
We call C-superalgebra any associative, unital C-algebra A which is Z2-graded: so A has
a Z2-grading A = A¯0 ⊕ A¯1, and AaAb ⊆ Aa+b. Any such A is said to be commutative if
xy = (−1)xyyx for all homogeneous x, y ∈ A; so, in particular, z2 = 0 for all z ∈ A¯1. All
C-superalgebras form a category, whose morphisms are those of unital C-algebras preserving
the Z2-grading; inside it, commutative C-superalgebras form a subcategory, that we denote
by (salg). We denote by (alg) the category of (associative, unital) commutative C-algebras, and
by (mod) that of C-modules. Note also that there is an obvious functor ( )¯0 : (salg) −→ (alg)
given on objects by A (cid:55)→ A¯0. We call Weil superalgebra any finite-dimensional commutative
C-superalgebra A such that A = C ⊕ N(A) where C is even and N(A) = N(A)¯0 ⊕ N(A)¯1
is a Z2-graded nilpotent ideal (the nilradical of A). Every Weil superalgebra A is endowed
with a canonical epimorphisms pA : A −−−(cid:16) C and an embedding uA : C (cid:44)−→ A, such that
pA ◦ uA = id. Weil superalgebras over C form a full subcategory of (salg), denoted by (Wsalg).
Finally, let (Walg) := (Wsalg) ∩ (alg) be the category of Weil algebras (over C), i.e., the full
subcategory of all totally even objects in (Wsalg) -- namely, those whose odd part is trivial.
Then the functor ( )¯0 : (salg) −→ (alg) obviously restricts to a similar functor ( )¯0 : (Wsalg) −→
(Walg) given again by A (cid:55)→ A¯0.
xy
zy
[z, [x, y]] = 0 (Jacobi identity).
xz
[x, [y, z]] + (−1)
yx
[y, x] = 0 (anti-symmetry);
[y, [z, x]] + (−1)
2.2 Lie superalgebras
By definition, a Lie superalgebra is a C-supermodule g = g¯0 ⊕ g¯1 with a (Lie super)bracket
[·,·] : g × g −→ g, (x, y) (cid:55)→ [x, y], which is C-bilinear, preserving the Z2-grading and satisfies the
following (for all homogenenous x, y, z ∈ g):
(a) [x, y] + (−1)
(b) (−1)
In this situation, we write Y (cid:104)2(cid:105) := 2−1[Y, Y ] (∈ g¯0) for all Y ∈ g¯1. All Lie C-superalgebras form
a category, denoted by (sLie), whose morphisms are C-linear, preserving the Z2-grading and
the bracket. Note that if g is a Lie C-superalgebra, then its even part g¯0 is automatically a Lie
C-algebra.
category of Lie C-algebras. Then every Lie C-superalgebra g ∈ (sLie) defines a functor
A (cid:55)→ Lg(A) := (A ⊗ g)¯0 = (A¯0 ⊗ g¯0) ⊕ (A¯1 ⊗ g¯1).
Lie superalgebras can also be described in functorial language.
Lg : (Wsalg) −−−→ (Lie),
Indeed, let (Lie) be the
Indeed, A ⊗ g is a Lie superalgebra (in a suitable, more general sense, over A) on its own, with
aa(cid:48) ⊗ [X, X(cid:48)]; now Lg(A) is the even part of A ⊗ g,
Lie bracket [a ⊗ X, a(cid:48) ⊗ X(cid:48)] := (−1)
hence it is a Lie algebra on its own.
Xa(cid:48)
2.3 Lie supergroups
We shall now recall, in steps, the notion of complex holomorphic "Lie supergroups", as a special
kind of "supermanifold". The following is a very concise summary of a long, detailed theory:
further details are, for instance, in [1, 4, 17].
Singular Degenerations of Lie Supergroups of Type D(2, 1; a)
2.3.1. Supermanifolds. By superspace we mean a pair S =(cid:0)S,OS
(cid:1) of a topological space S
5
As basic model, the linear supervariety Cpq
and a sheaf of commutative superalgebras OS on it such that the stalk OS,x of OS at each point
x ∈ S is a local superalgebra. A morphism φ : S −→ T between superspaces S and T is
a pair (φ, φ∗) where φ : S −→ T is a continuous map of topological spaces and the induced
morphism φ∗ : OT −→ φ∗(OS) of sheaves on T is such that φ∗
x(mφ(x)) ⊆ mx, where mφ(x)
and mx denote the maximal ideals in the stalks OT,φ(x) and OS,x, respectively.
C (in Leites' terminology) is, by definition, the topo-
logical space Cp endowed with the following sheaf of commutative superalgebras: OCpqC
(U ) :=
HCp(U ) ⊗C ΛC(ξ1, . . . , ξq) for any open set U ⊆ Cp, where HCp is the sheaf of holomorphic
functions on Cp and ΛC(ξ1, . . . , ξq) is the complex Grassmann algebra on q variables ξ1, . . . , ξq
of odd parity. A (complex holomorphic) supermanifold of (super)dimension pq is a superspace
M = (M,OM ) such that M is Hausdorff and second-countable and M is locally isomorphic
to Cpq
C , i.e., for each x ∈ M there is an open set Vx ⊆ M with x ∈ Vx and U ⊆ Cp such
that OMVx
C ). A morphism between
holomorphic supermanifolds is just a morphism (between them) as superspaces.
U (in particular, it is locally isomorphic to Cpq
∼= OCpqC
We denote the category of (complex holomorphic) supermanifolds by (hsmfd).
Let now M be a holomorphic supermanifold and U an open subset in M. Let IM (U ) be the
(nilpotent) ideal of OM (U ) generated by the odd part of the latter: then OM
of purely even superalgebras over M, locally isomorphic to HCp. Then Mrd := (M,OM /IM )
is a classical holomorphic manifold, called the underlying holomorphic (sub)manifold of M ; the
standard projection s (cid:55)→ s := s + IM (U ) (for all s ∈ OM (U )) at the sheaf level yields an
embedding Mrd −→ M , so Mrd can be seen as an embedded sub(super)manifold of M . The
whole construction is clearly functorial in M .
(cid:14)IM defines a sheaf
Finally, each "classical" manifold can be seen as a "supermanifold", just regarding its struc-
ture sheaf as one of superalgebras that are actually totally even, i.e., with trivial odd part. Con-
versely, any supermanifold enjoying the latter property is actually a manifold, nothing more. In
other words, every manifold identify with a supermanifold M that actually coincides with its
underlying (sub)manifold Mrd.
2.3.2. Lie supergroups and the functorial approach. A group object in the category
(hsmfd) is called (complex holomorphic) Lie supergroup. These objects, together with the obvi-
ous morphisms, form a subcategory among supermanifolds, denoted (Lsgrp).
x ∈ M and every A ∈ (Wsalg) we set MA,x = Hom(salg)(OM,x, A) and MA = (cid:70)
Lie supergroups -- as well as supermanifolds -- can also be conveniently studied via a functorial
approach that we now briefly recall (cf. [1] or [9] for details). Let M be a supermanifold. For every
x∈M MA,x;
then we define WM : (Wsalg) −→ (set) to be the functor given by A (cid:55)→ MA and ρ (cid:55)→ ρ(M ) with
ρ(M ) : MA −→ MB, xA (cid:55)→ ρ◦xA. Overall, this provides a functor B : (hsmfd) −→ [(Wsalg), (set)]
given on objects by M (cid:55)→ WM ; we can now refine still more.
Given a finite dimensional commutative algebra A¯0 over C, a (complex holomorphic) A¯0-
manifold is any manifold that is locally modelled on some open subset of some finite dimensional
A¯0-module, so that the differential of every change of charts is an A¯0-module isomorphism. An
A¯0-morphism between two A¯0-manifolds is any smooth morphism whose differential is every-
where A¯0-linear (we then say that "it is A¯0-smooth"). Gathering all A¯0-manifolds (for all
possible A), and suitably defining morphisms among them, one defines the category (A¯0-hmfd)
of all "A¯0-manifolds".
A key point now is that each WM turns out to be a functor from (Wsalg) into (A¯0-hmfd).
same objects but whose morphisms are all natural transformations φ : G −−→ H such that
for every A ∈ (Wsalg) the induced φA : G(A) −−→ H(A) is A¯0-smooth. Then the second key
Furthermore, let [[(Wsalg), (A¯0-hmfd)]] be the subcategory of (cid:2)(Wsalg), (A¯0-hmfd)(cid:3) with the
6
K. Iohara and F. Gavarini
point is that if φ : M −−→ N is a morphism of supermanifolds, then φA is a morphism in
[[(Wsalg), (A¯0-hmfd)]]. The final outcome is that we have a functor S : (hsmfd) −→ [[(Wsalg),
(A¯0-hmfd)]], given on objects by M (cid:55)→ WM ; the key result is that this embedding is full and
faithful, so that for any two supermanifolds M and N one has M ∼= N if and only if S(M ) ∼=
S(N ), i.e., WM
Still relevant to us, is that the embedding S preserves products, hence also group objects.
Therefore, a supermanifold M is a Lie supergroup if and only if S(M ) := WM takes values in
the subcategory (among A¯0-manifolds) of group objects -- thus each WM (A) is a group.
∼= WN .
Finally, in the functorial approach the "classical" manifolds (i.e., totally even supermanifolds)
can be recovered as follows: in the previous construction one simply has to replace the words
"Weil superalgebras" with "Weil algebras" everywhere. It then follows, in particular, that the
functor of points WM of any holomorphic, manifold M is actually a functor from (Walg) to
(A¯0-hmfd); one can still see it as (the functor of points of) a supermanifold -- that is totally even,
though -- by composing it with the natural functor ( )¯0 : (Wsalg) −→ (Walg). On the other hand,
given any supermanifold M , say holomorphic, the functor of points of its underlying submanifold
Mrd is given by WMrd(A) = WM (A) for each A ∈ (Walg), or in short WMrd = WM
(cid:12)(cid:12)(Walg).
Finally, it is worth stressing that the functorial point of view on supermanifolds was originally
developed -- by Leites, Berezin, Deligne, Molotkov, Voronov and many others -- in a slightly
different way. Namely, they considered functors defined, rather than on Weil superalgebras, on
Grassmann (super)algebras. Actually, the two approaches are equivalent: see [1] for a detailed,
critical analysis of the matter.
There are some advantages in restricting the focus onto Grassmann algebras. For instance,
they are the superalgebras of global sections onto the superdomains of dimension 0q -- i.e.,
"super-points". Therefore, if M is a supermanifold considered as a super-ringed space, its
description via a functor defined on Grassmann algebras (only) can be really seen as the true
restriction of the functor of points of M , considered as a super-ringed space.
On the other hand, the use of Weil superalgebras has the advantage that one can use it to
perform differential calculus on functors WM , much in the spirit of Weil's approach to differential
calculus in algebraic geometry. Note also that some peculiar properties for Grassmann algebras
are still available for every Weil superalgebra A: e.g., the existence of the maps pA : A −−(cid:16) C
and uA : C (cid:44)−−→ A, key tools in the theory (for instance, for any Lie supergroup G this implies
the existence of a semidirect product splitting of the group G(A) of A-points of G). See [1] for
further details.
2.4 Super Harish-Chandra pairs and Lie supergroups
A different way to deal with Lie supergroups (or algebraic supergroups) is via the notion of
"super Harish-Chandra pair", that gathers together the infinitesimal counterpart -- that of Lie
superalgebra -- and the classical (i.e., "non-super") counterpart -- that of Lie group -- of the
notion of Lie supergroup. We recall it shortly, referring to [9] (and [8]) for further details.
2.4.1. Super Harish-Chandra pairs. We call super Harish-Chandra pair -- or just "sHCp"
in short -- any pair (G, g) such that G is a (complex holomorphic) Lie group, g a complex
Lie superalgebra such that g¯0 = Lie(G), and there is a (holomorphic) G-action on g by Lie
superalgebra automorphisms, denoted by Ad : G −−−→ Aut(g), such that its restriction to g¯0
is the adjoint action of G on Lie(G) = g¯0 and the differential of this action is the restriction to
Lie(G) × g = g¯0 × g of the adjoint action of g on itself. Then a morphism (Ω, ω) : (G(cid:48), g(cid:48)) −−→
(G(cid:48)(cid:48), g(cid:48)(cid:48)) between sHCp's is given by a morphism of Lie groups Ω : G(cid:48) −→ G(cid:48)(cid:48) and a morphism
of Lie superalgebras ω : g(cid:48) −→ g(cid:48)(cid:48) such that ωg¯0 = dΩ and ω ◦ Adg = AdΩ+(g) ◦ω for all g ∈ G.
We denote the category of all super Harish-Chandra pairs by (sHCp).
Singular Degenerations of Lie Supergroups of Type D(2, 1; a)
2.4.2. From Lie supergroups to sHCp's. For any A ∈ (Wsalg), let A[ε] := A[x](cid:14)(cid:0)x2(cid:1),
with ε := x mod(cid:0)x2(cid:1) being even. Then A[ε] = A ⊕ Aε ∈ (Wsalg), and there exists a natural
morphism pA : A[ε] −→ A given by (a + a(cid:48)ε)
pA(cid:55)→a. For a Lie supergroup G, thought of as
a functor G : (Wsalg) −−→ (groups) -- i.e., identifying G ∼= WG -- let G(pA) : G(A[ε]) −−→
(cid:1). The key fact
Lie(G) : (Wsalg) −−→ (groups) given on objects by Lie(G)(A) := Ker(cid:0)G(p)A
G(A) be the morphism associated with pA : A[ε] −−→ A. Then there exists a unique functor
7
now is that Lie(G) is actually valued in the category (Lie) of Lie algebras, i.e., it is a functor
Lie(G) : (Wsalg) −→ (Lie). Furthermore, there exists a Lie superalgebra g -- identified with the
tangent superspace to G at the unit point -- such that Lie(G) = Lg (cf. Section 2.2). Moreover,
for any A ∈ (Wsalg) one has Lie(G)(A) = Lie(G(A)), the latter being the Lie algebra of the
Lie group G(A).
Finally, the construction G (cid:55)→ Lie(G) for Lie supergroups is actually natural, i.e., provides
a functor Lie : (Lsgrp) −−−→ (sLie) from Lie supergroups to Lie superalgebras.
On the other hand, each Lie supergroup G is a group object in the category of (holomor-
phic) supermanifolds: therefore, its underlying submanifold Grd is in turn a group object in
the category of (holomorphic) manifolds, i.e., it is a Lie group. Indeed, the naturality of the
construction G (cid:55)→ Grd provides a functor from Lie supergroups to (complex) Lie groups.
On top of this analysis, if G is any Lie supergroup, then (Grd, Lie(G)) is a super Harish-
Chandra pair; more precisely, we have a functor Φ : (Lsgrp) −−→ (sHCp) given on objects by
G (cid:55)→ (Grd, Lie(G)) and on morphisms by φ (cid:55)→ (φrd, Lie(φ)).
2.4.3. From sHCp's to Lie supergroups. The functor Φ : (Lsgrp) −−→ (sHCp) has a quasi-
inverse Ψ : (sHCp) −−→ (Lsgrp) that we can describe explicitly (see [8, 9]).
of g¯1. For any A ∈ (Wsalg), we define GP (A) as being the group with generators the elements
of the set ΓB
Indeed, let P :=(cid:0)G, g(cid:1) be a super Harish-Chandra pair, and let B := {Yi}i∈I be a C-basis
A := G(A)(cid:83){(1 + ηiYi)}(i,ηi)∈I×A¯1
and relations
g(cid:48) · g(cid:48)(cid:48) = g(cid:48) ·G g(cid:48)(cid:48),
1G = 1,
(1 + ηiYi) · g = g · (1 + cj1ηiYj1)··· (1 + cjk ηiYjk )
with Ad(cid:0)g−1(cid:1)(Yi) = cj1Yj1 + ··· + cjk Yjk ,
i Yi) =(cid:0)1G + η(cid:48)(cid:48)
· (1 + (η(cid:48)
(cid:104)2(cid:105)
i η(cid:48)
iY
i
(cid:1)
iYi) · (1 + η(cid:48)(cid:48)
(1 + η(cid:48)
(1 + ηiYi) · (1 + ηjYj) = (1G + ηjηi[Yi, Yj])G · (1 + ηjYj) · (1 + ηiYi)
i + η(cid:48)(cid:48)
i )Yi),
G
i, η(cid:48)(cid:48)
(cid:0)A(cid:48)(cid:48)(cid:1) be the group morphism uniquely defined on generators by GP (ϕ)(g(cid:48)) := G(ϕ)(g(cid:48)),
for g, g(cid:48), g(cid:48)(cid:48) ∈ G(A), ηi, η(cid:48)
i , ηj ∈ A¯1, i, j ∈ I. This defines the functor GP on objects, and one
then defines it on morphisms as follows: for any ϕ : A(cid:48) −→ A(cid:48)(cid:48) in (Wsalg) we let GP (ϕ) : GP (A(cid:48))
−→ GP
GP (ϕ)(1 + η(cid:48)Yi) := (1 + ϕ(η(cid:48))Yi).
One proves (see [8, 9]) that every such GP is in fact a Lie supergroup -- thought of as a special
functor, i.e., identified with its associated Weil -- Berezin functor. In addition, the construction
P (cid:55)→ GP is natural in P, i.e., it yields a functor Ψ : (sHCp) −−→ (Lsgrp); moreover, the latter
is a quasi-inverse to Φ : (Lsgrp) −−→ (sHCp).
3 Lie superalgebras of type D(2, 1; σ)
In this section, we present the complex Lie superalgebras of type D(2, 1; a). On the one hand,
one can construct them directly, through Kaplansky's representation (cf.
[14]; a widely acces-
sible account of it is also in Scheunert's book [15, Chapter I, Section 1.5]), which depend on
parameters. On the other hand, for non-singular values of the parameters one can realize them
via Kac' method, choosing a suitable Cartan matrix, which still depends on parameters.
8
K. Iohara and F. Gavarini
3.1 Construction of gσ (after Kaplansky)
To fix notation, we set V := (cid:8)(σ1, σ2, σ3) ∈ C3(cid:12)(cid:12)(cid:80)
i σi = 0(cid:9) -- a plane in C3 -- and V × :=
V ∩ (C×)3, where C× := C\{0}; also, 0 := (0, 0, 0) ∈ V \ V ×. The Lie superalgebras gσ we deal
with will depend on a parameter σ := (σ1, σ2, σ3) ∈ V .
3.1.1. Kaplansky's realization. We recall hereafter the construction of Lie superalgebras
introduced by Kaplansky (cf. [14]) who denoted them by Γ (A, B, C), with A, B, C ∈ C. With
a suitable normalization, and different terminology, they form the family nowadays called "of
type D(2, 1; a)", with a ∈ C and a (cid:54)∈ {0,−1} to ensure simplicity; we shall stick to Kaplansky's
point of view, but using the parameter σ ∈ C3 (and later in V ) and adopting the convention
of denoting by osp(4, 2; σ) the Lie superalgebra of type D(2, 1; σ). A detailed account of Kap-
lansky's realization of these Lie algebras can be found in Scheunert's book (cf. [15, Chapter I,
Section 1, Example 5]). Recall that having a Lie superalgebra g = g¯0 ⊕ g¯1 amounts to having:
(a) a Lie algebra g¯0; (b) a g¯0-module g¯1; (c) a g¯0-valued symmetric product on g¯1, such that the
g¯0-action is by derivations of the (symmetric) product.
Indeed, in the family of Lie superalgebras gσ := osp(4, 2; σ) parts (a) and (b) above will stand
the same for any σ ∈ V , while the dependence (of the Lie superalgebra structure) on σ will
actually occur only for part (c).
Step (a): Let sli(2) := Cei ⊕ Chi ⊕ Cfi (i = 1, 2, 3) be three isomorphic copy of sl(2), in its
standard realization. Then we consider their direct sum sl1(2) ⊕ sl2(2) ⊕ sl3(2) with its natural
Step (b): Let 2 := C+(cid:105) ⊕ C−(cid:105) be the (natural) tautological 2-dimensional sl(2)-module,
¯1 of gσ is (isomorphic
for every 1, 2, 3 ∈ {+,−}, we set
structure of Lie algebra: this will be the even part(cid:0)gσ
and 2i (for all i = 1, 2, 3) its i-th copy for sli(2). The odd part (cid:0)gσ
to) 21 (cid:2) 22 (cid:2) 23, with its natural structure of(cid:0)sl1(2) ⊕ sl2(2) ⊕ sl3(2)(cid:1)-module. We describe
a C-basis of it with the following shorthand notation:
v1,2,3 := 1(cid:105) ⊗ 2(cid:105) ⊗ 3(cid:105) ∈ 21 (cid:2) 22 (cid:2) 23.
Step (c): We define a projection ψ : 2⊗2 ∼= S22 ⊕ ∧22 −−(cid:16) ∧22 ∼= C by
¯0 of our gσ.
(cid:1)
(cid:1)
±(cid:105) ⊗ ±(cid:105) (cid:55)−→ 0,
±(cid:105) ⊗ ∓(cid:105) (cid:55)−→ ±2−1.
Then for σ(cid:48) ∈ C, we define the linear map p : 2⊗2 ∼= S22 ⊕ ∧22 −−(cid:16) S22 ∼= sl2 by
p(u, v).w := σ(cid:48)(cid:0)ψ(v, w).u − ψ(w, u).v(cid:1)
∀ u, v, w ∈ 2
that more explicitly reads
+(cid:105) ⊗ +(cid:105) (cid:55)−→ σ(cid:48)e,
±(cid:105) ⊗ ∓(cid:105) (cid:55)−→ −2−1σ(cid:48)h,
−(cid:105) ⊗ −(cid:105) (cid:55)−→ −σ(cid:48)f
Now, for each triple σ := (σ1, σ2, σ3) ∈ C3 and each i ∈ {1, 2, 3}, let pi : 2⊗2
with {e, h, f} being the standard sl2-triple, i.e., [e, f ] = h, [h, e] = 2e, [h, f ] = −2f .
i −−(cid:16) sli(2) be
the above map with scalar factor σ(cid:48) := −2σi ∈ C. The Lie superbracket [ ] on g¯1 × g¯1 can be
expressed as
(cid:88)
(cid:3) =
i=1ui,⊗3
(cid:2)⊗3
Tidying everything up, we can define a bracket on gσ =(cid:0)gσ
τ∈S3
i=1vi
ψ(uτ (1), vτ (1))ψ(uτ (2), vτ (2))pτ (3)(uτ (3), vτ (3)).
(cid:1)
∀ x, y ∈ (gσ)¯0,
¯0 ⊕(cid:0)gσ
(cid:1)
v, w ∈ (gσ)¯1.
¯1 by the formula
[x + v, y + w] := [x, y] + x.w − y.v + [v, w]
The following proposition resumes the outcome of this construction (see also Theorem 4.1
later on for what happens for singular values of σ):
Singular Degenerations of Lie Supergroups of Type D(2, 1; a)
Proposition 3.1 (cf. [15, Chapter II, Section 4.5, p. 135]). Let σ ∈ C3.
(a) The bracket given above defines a structure of Lie superalgebra on gσ if and only if the
given σ ∈ C3 satisfies the condition σ1 + σ2 + σ3 = 0, that is, σ ∈ V := C3(cid:84)(cid:8)(cid:80)
i σi = 0(cid:9).
(b) Let σ ∈ V . Then the Lie superalgebra gσ is simple if and only if σ ∈ V ×.
(c) Let σ(cid:48), σ(cid:48)(cid:48) ∈ V . Then the Lie superalgebras gσ(cid:48) and gσ(cid:48)(cid:48) are isomorphic if and only if there
exists τ ∈ S3 such that σ(cid:48)(cid:48) and τ.σ(cid:48) are proportional.
onto P(V )(cid:83){0 := (0, 0, 0)} ∼= P1C(cid:83){∗}, a complex projective line plus an extra point.
Thus, the isomorphism classes of our gσ's are in bijection with the orbits of the S3-action
9
3.1.2. The multiplicative table of gσ. For later use, we record hereafter the complete
multiplication table of the Lie superalgebra gσ -- for every σ = (σ1, σ2, σ3) ∈ V -- with respect to
the C-basis {hi, ej, fj, v1,2,3 i, j ∈ {1, 2, 3}, 1, 2, 3 ∈ {+,−}} given from scratch. In addition,
in the formulas below we also take into account the following coroots:
hβ1 := +2−1σ1h1 − 2−1σ2h2 − 2−1σ3h3,
hβ2 := −2−1σ1h1 + 2−1σ2h2 − 2−1σ3h3,
hβ3 := −2−1σ1h1 − 2−1σ2h2 + 2−1σ3h3,
hθ := +2−1σ1h1 + 2−1σ2h2 + 2−1σ3h3.
In short, the table is the following (∀ i, j ∈ {1, 2, 3}, 1, 2, 3 ∈ {+,−}):
[ei, fj] = δijhj,
,
,
[ei, ej] = 0,
[fi, fj] = 0,
[hi, fj] = −2δijfj,
[hi, hj] = 0,
[hi, ej] = 2δijej,
[ei, v1,2,3] = δi,−v(−1)δ1,i 1,(−1)δ2,i 2,(−1)δ3,i 3
[hi, v1,2,3] = iv1,2,3,
[fi, v1,2,3] = δi,+v(−1)δ1,i 1,(−1)δ2,i 2,(−1)δ3,i 3
[v+,+,+, v+,−,−] = −σ1e1,
[v+,+,+, v−,+,−] = −σ2e2,
[v+,+,+, v−,−,+] = −σ3e3,
[v+,+,−, v+,−,+] = +σ1e1,
[v+,+,−, v−,+,+] = +σ2e2,
[v+,−,+, v−,+,+] = +σ3e3,
[v+,+,+, v−,−,−] = +2−1σ1h1 + 2−1σ2h2 + 2−1σ3h3 = hθ,
[v+,+,−, v−,−,+] = −2−1σ1h1 − 2−1σ2h2 + 2−1σ3h3 = hβ3,
[v+,−,+, v−,+,−] = −2−1σ1h1 + 2−1σ2h2 − 2−1σ3h3 = hβ2,
[v−,+,+, v+,−,−] = +2−1σ1h1 − 2−1σ2h2 − 2−1σ3h3 = hβ1,
[v+,−,+, v−,−,−] = +σ2f2,
[v−,+,+, v−,−,−] = +σ1f1,
[v−,+,−, v−,−,+] = −σ1f1,
[v+,+,−, v−,−,−] = +σ3f3,
[v+,−,−, v−,−,+] = −σ2f2,
[v+,−,−, v−,+,−] = −σ3f3.
3.2 Kac' realization
We show now how the Lie superalgebras gσ of Section 3.1, for non-singular values σ ∈ V ×, can
be also realized as contragredient Lie superalgebras (via Kac' method).
3.2.1. First construction. For any σ ∈ V , we consider the Cartan matrix
0 −σ3 −σ2
−σ3
−σ2 −σ1
0 −σ1
0
Aσ = (ai,j)j=1,2,3
i=1,2,3 =
10
K. Iohara and F. Gavarini
(see [3] for a far-reaching analysis of what else is associated with such a Cartan matrix). We
associate to it a Dynkin diagram for which the nodes are defined as in [12, Section 2.5.5], and
we join any two vertices i and j with an edge labeled by aij: the resulting diagram then is
h1
−σ3
h
−σ2
h
3
2
−σ1
Following Kac, we consider a realization(cid:0)h, Π∨ = {Hβi}i=1,2,3, Π = {βi}i=1,2,3
(cid:1) of Aσ, that is
1) h is a C-vector space,
2) Π∨ is the set of simple coroots, a basis of h,
3) Π is the set of simple roots, a basis of h∗,
4) βj(Hβi) = ai,j for all 1 ≤ i, j ≤ 3.
The (contragredient) Lie superalgebra g(Aσ) is, by definition, the simple Lie superalgebra defined
as follows. First, let g(Aσ) be the Lie superalgebra with the nine generators Hβi, X±βi (i =
1, 2, 3), and relations (for 1 ≤ i, j ≤ 3)
[Hβi, Hβj ] = 0,
[Xβi, X−βj ] = δi,jHβi,
[Hβi, X±βj ] = ±βj(Hβi)X±βj ,
[X±βi, X±βi] = 0
with parity Hβi = ¯0 and X±βi = ¯1 for all i. Then one considers the maximal homogeneous
ideal Iσ of g(A) which meets trivially the C-span of the generators, and finally defines g(Aσ) :=
g(Aσ)/Iσ.
A straightforward (and easy) analysis shows that g(Aσ) is finite-dimensional. Then, by
general results (cf. [12, Section 2.5.1]), there exists an epimorphism Φσ : gσ −−(cid:16) g(Aσ) of Lie
superalgebras uniquely determined by
hβ2 (cid:55)→ Hβ2,
v+,−,+ (cid:55)→ X+β2,
v−,+,− (cid:55)→ X−β2,
hβ1 (cid:55)→ Hβ1,
v−,+,+ (cid:55)→ X+β1,
v+,−,− (cid:55)→ X−β1,
hβ3 (cid:55)→ Hβ3,
v+,+,− (cid:55)→ X+β3,
v−,−,+ (cid:55)→ X−β3.
When σ is non-singular, that is σ ∈ V ×, this epimorphism Φσ is actually an isomorphism, so
∼= g(Aσ). One can see it in two ways: first, since gσ is simple for σ ∈ V ×,
that osp(4, 2; σ) =: gσ
the kernel of Φσ is then necessarily trivial; second, direct inspection shows that for σ ∈ V × the
ideal Iσ has the "correct" codimension in g(Aσ) so that Φσ be injective.
We assume now, for the rest of the present subsection, that σ ∈ V × (non-singular case).
Then the set ∆+ of positive roots of g(Aσ) has the following description:
∆+ =(cid:8)β1, β2, β3, β1 + β2, β2 + β3, β3 + β1, β1 + β2 + β3
(cid:9).
∼= g(Aσ), or more directly by inspection (namely,
The dual h∗ of the Cartan subalgebra has the following description. Let {εi}i=1,2,3 ⊂ h∗ be an
2 σi (i = 1, 2, 3); then one can verify
This can be seen as a consequence of gσ
describing Iσ explicitly). The set of roots then is ∆ = ∆+ ∪ ∆−, with ∆− := −∆+.
orthogonal basis normalized by the conditions (εi, εi) = − 1
that (βi, βj) = −σk with {i, j, k} = {1, 2, 3}, where the simple roots are
β3 = ε1 + ε2 − ε3.
β1 = −ε1 + ε2 + ε3,
β2 = ε1 − ε2 + ε3,
Singular Degenerations of Lie Supergroups of Type D(2, 1; a)
11
Now let ∆+
¯0 and ∆+
¯0 =(cid:8)β1 + β2, β2 + β3, β3 + β1
∆+
(cid:9) =(cid:8)2εi 1 ≤ i ≤ 3(cid:9),
¯1 be the set of even (resp. odd) positive roots. One has
¯1 =(cid:8)β1, β2, β3, θ(cid:9),
∆+
where θ := β1 + β2 + β3 = ε1 + ε2 + ε3 is the highest root.
We introduce now further root vectors and coroots, defined by
X2ε1 := [Xβ2, Xβ3],
X−2ε1 := −[X−β2, X−β3],
H2ε1 := −(Hβ2 + Hβ3),
X2ε2 := [Xβ3, Xβ1],
X−2ε2 := −[X−β3, X−β1],
H2ε2 := −(Hβ3 + Hβ1),
X2ε3 := [Xβ1, Xβ2],
X−2ε3 := −[X−β1, X−β2],
H2ε3 := −(Hβ1 + Hβ2).
It can be checked that, for i, j ∈ {1, 2, 3}, one has
[H2εi, X±2εj ] = ±2σiδi,jX±2εj ,
[X2εi, X−2εj ] = σiδi,jH2εi,
(3.1)
which implies that each ai := CX2εi⊕CH2εi⊕CX−2εi (for 1 ≤ i ≤ 3) is a Lie sub-(super)algebra,
with [aj, ak] = 0 for j (cid:54)= k, and ai is isomorphic to sl(2) since σi (cid:54)= 0. In particular, the even
part of the Lie superalgebra g(Aσ) is nothing but g(Aσ)¯0 =(cid:76)3
(cid:3) ∈ g(Aσ)−θ; then
θ =(cid:2)X2εi, Xβi
(cid:3) = −σjσk
(cid:2)X j
θ , X k−θ
(cid:3) ∈ g(Aσ)θ and X i−θ =(cid:2)X−2εi, X−βi
3(cid:88)
i=1 ai.
Hβi.
For i = 1, 2, 3 we set X i
the following identities hold:
3(cid:88)
X i±θ = 0,
(3.2)
i=1
i=1
These formulas imply that there exists Xθ ∈ g(Aσ)θ and X−θ ∈ g(Aσ)−θ such that
X i
X i−θ = σiX−θ,
θ = σiXθ,
(3.3)
for any 1 ≤ i ≤ 3. Hence, setting Hθ = −(Hβ1 + Hβ2 + Hβ3), one has also [Xθ, X−θ] = Hθ.
Moreover, it also follows that H2ε1 + H2ε2 + H2ε3 = 2Hθ. Eventually, from all this we see that
the odd part g(Aσ)¯1 of the Lie superalgebra g(Aσ) is the C-span of {X±βi}i=1,2,3 ∪ {X±θ}.
Finally, from the previous description we see that the isomorphism osp(4, 2; σ) ∼= g(Aσ) can
be described on basis elements by
ei (cid:55)→ σ−1
i X2εi,
v+,−,+ (cid:55)→ Xβ2 ,
v−,+,− (cid:55)→ X−β2,
hi (cid:55)→ σ−1
i H2εi,
v−,+,+ (cid:55)→ Xβ1 ,
v+,−,− (cid:55)→ X−β1,
i = 1, 2, 3,
v+,+,+ (cid:55)→ Xθ,
v−,−,− (cid:55)→ X−θ.
Note that in first line of (3.4) the non-singularity of σ ∈ V × plays a key role!
fi (cid:55)→ σ−1
i X−2εi,
v+,+,− (cid:55)→ Xβ3 ,
v−,−,+ (cid:55)→ X−β3,
(3.4)
3.2.2. Second construction. For the reader's convenience, we present now a second construc-
tion based upon a different, more familiar Cartan matrix (and associated Dynkin diagram).
The link with the previous construction of Section 3.2.1 is through the application of the odd
reflection with respect to the root β2, following V. Serganova (see [16]).
To begin with, set α1 = β2 + β3, α2 = −β2, α3 = β1 + β2. Then Π(cid:48) := {αi}i=1,2,3 is another
set of simple roots of g(Aσ), which is not Weyl-group conjugate to Π; the corresponding set of
i=1,2,3 should be taken as hα1 = H2ε1, hα2 = Hβ2, hα3 = H2ε3. With such
a choice, the associated Cartan matrix A(cid:48)
i,j := αj(hαi))j=1,2,3
i=1,2,3 is given by
(cid:9)
coroots (Π(cid:48))∨ =(cid:8)hαi
2σ1 −σ1
A(cid:48)
σ =
−σ1
0 −σ3
= σ1
2
σ := (a(cid:48)
−1
−1
0
0 − σ3
σ1
0
0 −σ3
2σ3
,
0
− σ3
σ1
2 σ3
σ1
12
K. Iohara and F. Gavarini
where the second equality is available only if σ1 (cid:54)= 0. In particular, for σ1 (cid:54)= 0 and σ3 (cid:54)= 0, our
original g(Aσ) can be also defined via the Cartan matrix
1
a
0 −1 2
. When a (cid:54)= −1, this in turn corresponds -- following Kac' conventions, up to
with a := σ3
σ1
renumbering the vertices (cf. [12, Section 2.5]) -- to the simple Lie superalgebra attached to the
following distinguished (i.e., with just one odd vertex) Dynkin diagram of type D(2, 1; a)
2 −1 0
0
A(cid:48)
a =
h h h
α2
α3
α1
1
a
(instead, a = −1 corresponds to σ2 = 0, that is a singular case). Kac' results tell us that, for
all a ∈ C \ {−1, 0}, the set of positive roots with respect to Π(cid:48) is given by
∆(cid:48),+ =(cid:8)α1, α2, α3, α1 + α2, α2 + α3, α1 + α2 + α3, α1 + 2α2 + α3
hα3 = Hβ2+β1 = −(cid:0)Hβ2 + Hβ1
while the coroots can be expressed as
hα1 = Hβ2+β3 = −(Hβ2 + Hβ3),
hα2 = H−β2 = Hβ2,
(cid:1),
hα1+α2 = −Hβ3 = hα1 + hα2,
hα1+α2+α3 = Hβ1+β2+β3 = −Hβ1 − Hβ2 − Hβ3 = hα1 + hα2 + hα3,
hα1+2α2+α3 = Hβ1+β3 = −Hβ1 − Hβ3 = hα1 + 2hα2 + hα3.
hα3+α2 = −Hβ1 = hα3 + hα2,
(cid:9),
3.2.3. The singular case. We saw in Section 3.2.1 that for every non-singular parameter
σ ∈ V ×, we have a Lie superalgebra isomorphism Φσ : osp(4, 2; σ) = gσ
(cid:44)−−−(cid:16)g(Aσ). For singular
values σ ∈ V \V ×, instead, the epimorphism Φσ is no longer an isomorphism: indeed, in this case,
the Lie ideal Iσ in g(Aσ) is bigger than in the non-singular case, for instance (as a straightforward
calculation shows), we have
∼=
(cid:2)Xβi, Xβj
(cid:3) ∈ Iσ ⇐⇒ σk = 0
∀{i, j, k} = {1, 2, 3}.
Similarly, the contragredient Lie superalgebra g(A(cid:48)
a) of Section 3.2.2 can be defined also for
the "singular value" a = −1 (corresponding to σ1 + σ3 = 0, which is equivalent to σ2 = 0).
However, in this case the corresponding Lie ideal I(cid:48)
a) is bigger, as we have, for instance,
a ⇐⇒ a = −1,
therefore g(cid:0)A(cid:48)
[[[Xα1, Xα2], Xα3], Xα2] ∈ I(cid:48)
a) ∼= gσ := osp(4, 2; σ) via Φσ for all σ = (1, a + 1, a) (cid:54)= (1, 0,−1).
(cid:1) has strictly smaller dimension than osp(4, 2; σ) for σ := (1, 0,−1), whereas
a in g(A(cid:48)
g(A(cid:48)
a=−1
This shows that, in a sense, describing our objects gσ = osp(4, 2; σ) of Section 3.1 as con-
tragredient Lie superalgebras is problematic, so to say, at singular values of σ, in that the
contragredient construction yields not the outcome we are looking for.
3.3 Bases of gσ
In this subsection, for any given σ ∈ V × we sort out three special bases of gσ. Later on (in
Section 4) we use them to construct three different families, indexed by V , of Lie superalgebras:
by construction these families will coincide on all non-singular parameters σ ∈ V × but will
actually differ instead on singular values σ ∈ V \ V ×.
Singular Degenerations of Lie Supergroups of Type D(2, 1; a)
3.3.1. First basis. Let B := {H2ε1, H2ε2, H2ε3}(cid:83){Xα}α∈∆ and B+ := B∪{Hθ} be the subsets
13
of gσ whose elements are defined by
H2εi := σihi,
X 2εi := σiei,
X−2εi := σifi,
i = 1, 2, 3,
Hθ := 2−1
3(cid:88)
i=1
σihi,
:= v−,+,+,
Xβ1
X−β1 := v+,−,−,
:= v+,−,+,
Xβ2
X−β2 := v−,+,−,
:= v+,+,−,
Xβ3
X−β3 := v−,−,+,
(3.5)
(cf. Section 3.1.1); the analysis in Section 3.1.1 tells us that, for every σ ∈ V ×, the set B is a C-
basis of gσ, hence B+ is a spanning set. Again from Section 3.1 -- in particular Section 3.1.2 --
for the Lie brackets among the elements of B+ we find the following multiplication table
Xθ := v+,+,+,
X−θ := v−,−,−,
[H2εi, X±2εj ] = ±2δi,jσiX±2εj ,
[X−2εi, X−2εj ] = 0,
[X2εi, X−2εj ] = δi,jσiH2εi,
[H2εi, X±θ] = ±σiX±θ,
[Hθ, X±βi] = ∓σiX±βi,
[X2εi, X−βj ] = (1 − δi,j)σiXβk ,
[Hθ, X±θ] = 0,
[X2εi, X−θ] = σiX−βi,
[X−2εi, X−βj ] = δi,jσiX−θ,
[X−2εi, Xθ] = σiXβi,
[X−βi, X−βj ] = −(1 − δi,j)X−2εk ,
[H2εi, H2εj ] = 0,
[X2εi, X2εj ] = 0,
[H2εi, X±βj ] = ±(−1)δi,j σiX±βj ,
[Hθ, X±2εi] = ±σiX2εi,
[X2εi, Xβj ] = δi,jσiXθ,
[X−2εi, Xβj ] = (1 − δi,j)σiX−βk ,
[X2εi, Xθ] = 0,
[Xβi, Xβj ] = (1 − δi,j)X2εk ,
[Xβi, X−βj ] = δi,j(H2εi − Hθ),
[Xβi, Xθ] = 0,
[Xθ, Xθ] = 0,
[Xβi, X−θ] = X−2εi,
[Xθ, X−θ] = Hθ,
[X−θ, X−θ] = 0
[X−βi, Xθ] = −X2εi,
[X−βi, X−θ] = 0,
for all i, j ∈ {1, 2, 3}, with k ∈ {1, 2, 3} \ {i, j}.
3.3.2. Second basis. Let now B(cid:48) := {H(cid:48)
the subsets of gσ with elements
2ε1
, H(cid:48)
2ε2
, H(cid:48)
2ε3
}(cid:83){X(cid:48)
α}α∈∆ and B(cid:48)
+ := B(cid:48) ∪ {H(cid:48)
θ} be
[X−2εi, X−θ] = 0,
3(cid:88)
i=1
σihi,
H(cid:48)
2εi := hi,
X(cid:48)
β1 := v−,+,+,
X(cid:48)
−β1 := v+,−,−,
X(cid:48)
+2εi := ei,
X(cid:48)
β2 := v+,−,+,
X(cid:48)
−β2 := v−,+,−,
X(cid:48)
−2εi := fi,
X(cid:48)
β3 := v+,+,−,
X(cid:48)
−β3 := v−,−,+,
θ := 2−1
H(cid:48)
i = 1, 2, 3,
X(cid:48)
θ := v+,+,+,
X(cid:48)
−θ := v−,−,−,
(3.6)
(cf. Section 3.1.1). Again from Section 3.1.1 we see that, for every σ ∈ V (including also the
singular locus), B(cid:48) is a C-basis of gσ, so B(cid:48)
+ is a spanning set: indeed, B(cid:48) is nothing but a different
notation for the natural, built-in C-basis of gσ in Kaplansky's realization (cf. Section 3.1.1). Then
from Section 3.1.2 we get the following multiplication table for Lie brackets among elements
of B(cid:48)
+
[H(cid:48)
[X(cid:48)
[H(cid:48)
[H(cid:48)
[X(cid:48)
[X(cid:48)
[H(cid:48)
[X(cid:48)
] = ±(−1)δi,j X(cid:48)
2εi, H(cid:48)
2εi, X(cid:48)
2εi, X(cid:48)
θ, X(cid:48)
2εi, X(cid:48)
−2εi, X(cid:48)
±2εj ] = ±2δi,jX(cid:48)
2εi, X(cid:48)
2εj ] = 0,
−2εi, X(cid:48)
−2εj ] = 0,
2εj ] = 0,
2εi, X(cid:48)
[H(cid:48)
±βj
±βj
,
±2εi] = ±σiX(cid:48)
] = ∓σiX(cid:48)
θ, X(cid:48)
[H(cid:48)
±βi
±βi
2εi,
,
] = (1 − δi,j)X(cid:48)
2εi, X(cid:48)
] = δi,jX(cid:48)
[X(cid:48)
−βj
θ,
] = (1 − δi,j)X(cid:48)
−2εi, X(cid:48)
[X(cid:48)
−βk
−βj
βj
βj
,
±2εj ,
[X(cid:48)
−2εj ] = δi,jH(cid:48)
2εi, X(cid:48)
±θ] = ±X(cid:48)
±θ,
2εi,
[H(cid:48)
θ, X(cid:48)
±θ] = 0,
,
βk
] = δi,jX(cid:48)
−θ,
14
K. Iohara and F. Gavarini
θ] = X(cid:48)
[X(cid:48)
−2εi, X(cid:48)
] = −(1 − δi,j)σiX(cid:48)
−βj
βi
,
, X(cid:48)
−2εk
[X(cid:48)
,
−2εi, X(cid:48)
−θ] = 0,
βi
[X(cid:48)
[X(cid:48)
[X(cid:48)
[X(cid:48)
[X(cid:48)
[X(cid:48)
2εi, X(cid:48)
, X(cid:48)
, X(cid:48)
, X(cid:48)
θ, X(cid:48)
−θ] = X(cid:48)
−βi
,
[X(cid:48)
−βi
,
2εi, X(cid:48)
θ] = 0,
] = (1 − δi,j)σiX(cid:48)
βj
2εi − H(cid:48)
] = δi,j(σiH(cid:48)
−βj
θ),
−θ] = σiX(cid:48)
, X(cid:48)
[X(cid:48)
θ] = 0,
−θ] = H(cid:48)
θ, X(cid:48)
[X(cid:48)
θ,
θ] = 0,
2εk
βi
βi
βi
[X(cid:48)
, X(cid:48)
, X(cid:48)
[X(cid:48)
−βi
−θ, X(cid:48)
−θ] = 0,
θ] = −σiX(cid:48)
−βi
2εi,
−θ] = 0,
−2εi,
[X(cid:48)
for all i, j ∈ {1, 2, 3}, with k ∈ {1, 2, 3} \ {i, j}.
3.3.3. Third basis. Let now σ ∈ V × (generic case again!). We fix as third basis (and spanning
set) of gσ a suitable blending of the two ones in Sections 3.3.1 and 3.3.2 above. Namely, we
set B(cid:48)(cid:48) := {H(cid:48)
and (3.5). Then B(cid:48)(cid:48) is another C-basis, and B(cid:48)(cid:48)
which can be argued at once from those for B+ and B(cid:48)
2εi, X±2εj ] = ±2δi,jX±2εj ,
For later use we record hereafter the complete table of Lie brackets among elements of B(cid:48)(cid:48)
+,
+ in Sections 3.5 and 3.6, respectively:
(cid:9), with elements defined by (3.6)
}(cid:83){Xα}α∈∆ and B(cid:48)(cid:48)
+ := B(cid:48)(cid:48) ∪(cid:8)H(cid:48)
+ another spanning set, of gσ.
2εi, H(cid:48)
, H(cid:48)
, H(cid:48)
2ε1
2ε3
2ε2
θ
[X2εi, X−2εj ] = δi,jσ2
i H(cid:48)
2εi,
2εi, X±θ] = ±X±θ,
θ, X±βi] = ∓σiX±βi,
[H(cid:48)
[X−2εi, X−2εj ] = 0,
[H(cid:48)
[H(cid:48)
2εj ] = 0,
[X2εi, X2εj ] = 0,
2εi, X±βj ] = ±(−1)δi,j X±βj ,
[H(cid:48)
θ, X±2εi] = ±σiX2εi,
[H(cid:48)
[H(cid:48)
[H(cid:48)
[X2εi, X−βj ] = (1 − δi,j)σiXβk ,
[X2εi, Xβj ] = δi,jσiXθ,
[X−2εi, Xβj ] = (1 − δi,j)σiX−βk ,
[X2εi, Xθ] = 0,
[X−2εi, X−θ] = 0,
[X−βi, X−βj ] = −(1 − δi,j)X−2εk ,
[Xβi, Xθ] = 0,
[Xθ, Xθ] = 0,
[Xβi, X−θ] = X−2εi,
[Xθ, X−θ] = H(cid:48)
θ,
[Xβi, Xβj ] = (1 − δi,j)X2εk ,
[X2εi, X−θ] = σiX−βi,
[X−θ, X−θ] = 0,
[Xβi, X−βj ] = δi,j(σiH(cid:48)
[X−2εi, X−βj ] = δi,jσiX−θ,
[X−βi, Xθ] = −X2εi,
[X−2εi, Xθ] = σiXβi,
θ, X±θ] = 0,
2εi − H(cid:48)
θ),
[X−βi, X−θ] = 0,
for all i, j ∈ {1, 2, 3}, with k ∈ {1, 2, 3} \ {i, j}.
4
Integral forms & degenerations for Lie superalgebras
of type D(2, 1; σ)
Let l be any Lie (super)algebra over a field K, and R any subring of K. By integral form
of l over R, or (integral) R-form of l, we mean by definition any Lie R-sub(super)algebra tR
∼= l as Lie (super)algebras
of l whose scalar extension to K is l itself: in other words K ⊗R tR
over K. In this subsection we introduce five particular integral forms of l = gσ, and study some
remarkable specializations of them. Let ∆ := ∆+ ∪ (−∆+) be the root system of gσ.
From now on, for any σ := (σ1, σ2, σ3) ∈ V :=(cid:8)σ ∈ C3 3(cid:80)
σi = 0(cid:9) we denote by Z[σ] the
(unital) subring of C generated by {σ1, σ2, σ3}.
i=1
We warn the reader that the choice of a Z[σ]-form becomes very important when one considers
a singular degeneration: one cannot speak instead of the singular degeneration, in that any
degeneration actually depends not only on the specific specialization value taken by σ but also
on the previous choice of a specific Z[σ]-form, that must be fixed in advance. Some specific
features of this phenomenon are presented in Theorems 4.1, 4.2, 4.3 etc.
Singular Degenerations of Lie Supergroups of Type D(2, 1; a)
15
4.1 First family: the Lie superalgebras g(σ)
4.1.1. Construction of the g(σ)'s. For any σ ∈ V (cf. Section 3.1), let gZ(σ) be the Lie
superalgebra over Z[σ] defined as follows. As a Z[σ]-module, gZ(σ) is spanned by the formal
set of Z[σ]-linear generators Bg := {H2ε1, H2ε2, H2ε3, Hθ}(cid:83){Xα}α∈∆ subject only to the single
relation H2ε1 + H2ε2 + H2ε3 = 2Hθ. Then it follows, in particular, that gZ(σ) is clearly a free
Z[σ]-module, with basis Bg\{H2εi} for any 1 ≤ i ≤ 3. The Lie superalgebra structure of gZ(σ) is
defined by the formulas in Section 3.3.1, now taken as definitions of the Lie brackets among the
(linear) generators of gZ(σ) itself. Overall, all these gZ(σ)'s form a family of Lie superalgebras
indexed by the points of the complex plane V . Moreover, taking g(σ) := C ⊗Z[σ] gZ(σ) for all
σ ∈ V we find a more regular situation, as now these (extended) Lie superalgebras g(σ) all
share C as their common ground ring. Moreover,
in the non-singular case, i.e., for σi ∈ V ×, we have g(σ) ∼= gσ
(4.1)
by Section 3.3.1 and the very definition of g(σ) itself. Indeed, the analysis in Section 3.3.1 --
describing a C-basis and its multiplication table for gσ with σ ∈ V × -- prove that for all σ ∈ V ×,
the Lie superalgebra gZ(σ) over Z[σ] identifies to an integral Z[σ]-form of gσ. In order to for-
malize the description of the family {g(σ)}σ∈V , we proceed as follows. Let Z[x] := Z[V ] ∼=
Z[x1, x2, x3]/(x1 + x2 + x3) be the ring of global sections of the Z-scheme associated with V . In
the construction of gZ(σ), formally replace x to σ (hence the xi's to the σi's): this makes sense,
provides a meaningful definition of a Lie superalgebra over Z[x], denoted by gZ(x), and then
g(x) := C[x]⊗Z[x] gZ(x) by scalar extension, which is a Lie superalgebra over C[x] := C⊗Z Z[x].
Definitions imply that, for any σ ∈ V , we have a Lie Z[σ]-superalgebra isomorphism
gZ(σ) ∼= Z[σ] ⊗
Z[x]
gZ(x)
-- through the ring isomorphism Z[σ] ∼= Z[x]
(cid:46)(cid:0)xi − σi
(cid:1)
i=1,2,3 -- and similarly
g(σ) ∼= C ⊗
C[x]
g(x)
on Spec(C[x]) = V ∪{(cid:63)} are, by definition, given by (LgC[x]
as Lie C-superalgebras, through the ring isomorphism C ∼= C[x]/(xi − σi)i=1,2,3. In geometrical
language, all this can be formulated as follows. The Lie superalgebra g(x) -- being a free,
finite rank C[x]-module -- defines a coherent sheaf LgC[x]
of Lie superalgebras over Spec(C[x]).
Moreover, there exists a unique fibre bundle over Spec(C[x]), say LgC[x]
, whose sheaf of sections
is exactly LgC[x]
. This fibre bundle can be thought of as a (total) deformation space over the base
space Spec(C[x]), in which every fibre can be seen as a "deformation" of any other one, and also
any single fibre can be seen as a degeneration of the original Lie superalgebra g(x). Moreover, the
g(x) ∼= g(σ)
fibres of LgC[x]
for any closed point σ ∈ V ⊆ Spec(C[x]), while for the generic point (cid:63) ∈ Spec(C[x]) we have
(LgC[x]
g(x)(=: gC(x)). Finally, it follows from our construction that these sheaf
and fibre bundle do admit an action of C× × S3, that on the base space Spec(C[x]) = V ∪ {(cid:63)}
∼= g(σ) for all σ ∈ V . The outcome is that in the "regular"
structure of these fibres (LgC[x]
locus V × they are simple (as Lie superalgebras), while in the "singular" locus V \ V × they are
not, and we can describe explicitly their structure.
Theorem 4.1. Let σ ∈ V as above, and set ai := CX2εi ⊕ CH2εi ⊕ CX−2εi with X2εi , H2εi and
X−2εi as defined in Section 3.3.1, for all i = 1, 2, 3.
(cid:1)-action on V . In the next result we describe the
)σ = C ⊗
C[x]
)(cid:63) = C(x) ⊗
C[x]
simply fixes {(cid:63)} and is the standard(cid:0)C× × S3
)σ
16
(1) If σ ∈ V ×, then the Lie superalgebra g(σ) is simple.
(2) If σ ∈ V \ V ×, with σi = 0 and σj (cid:54)= 0 (cid:54)= σk for {i, j, k} = {1, 2, 3}, then ai is a central Lie
ideal of g(σ), isomorphic to C30, and g(σ) is the universal central extension of psl(22)
by ai (cf. [10, Theorem 4.7]); in other words, there exists a short exact sequence of Lie
superalgebras
K. Iohara and F. Gavarini
0 −−−−−→ C30 ∼= ai −−−−−→ g(σ) −−−−−→ psl(22) −−−−−→ 0.
A parallel result also holds true when working with gZ(σ) over the ground ring Z[σ].
(3) If σ = 0 (∈ V \ V ×), i.e., σh = 0 for all h ∈ {1, 2, 3}, then g(0)¯0
∼= C90 is the center of
∼= C08 is Abelian; in particular, g(0) is a non-trivial,
g(0), and the quotient g(0)
non-Abelian central extension of C08 by C90, i.e., there exists a short exact sequence of
Lie superalgebras, with non-Abelian middle term,
(cid:46)
g(0)¯0
0 −−−−−→ C90 ∼= g(0)¯0 −−−−−→ g(0) −−−−−→ C08 −−−−−→ 0.
A parallel result holds true when working with gZ(0) over the ground ring Z[0] = Z.
Proof . Part (1) is a direct consequence of (4.1) and Proposition 3.1. Claim (2) instead follows
at once by direct inspection of the formulas in Section 3.3.1. For instance, reading the lines
in the first and second line in the table of formulas for Lie brackets therein, we see that ai is
∼= C3 as claimed. Moreover, the formulas from the third to
Abelian when σi = 0, so that ai
the seventh in the same table tell us also that all brackets of the generators of ai with all other
generators turn to zero when σi = 0: thus ai is central in g(σ) -- hence, in particular, it is a Lie
ideal -- as claimed. Similarly, a direct verification (setting σi = 0 in those formulas) shows that
the quotient g(σ)/ai is isomorphic to psl(22).
Claim (3) follows again by straightforward analysis of the table of formulas in Section 3.3.1.
Indeed, the first two lines in the table describes the structure of the Lie algebra g(σ)¯0: when
σ = 0 they simply tell that this structure is trivial, that is g(0) is Abelian, hence isomorphic
to C9 with trivial Lie bracket. Moreover, the lines from third to seventh in the same table
describe the adjoint action of g(σ)¯0 onto g(σ)¯1: when σ = 0, all the Lie brackets therein turn to
zero, so the g(0)¯0-action is trivial, which means exactly that g(0)¯0 is central. Finally, the lines
from eighth to eleventh describe the Lie brackets among linear generators of g(σ)¯1: all these
brackets are independent of σ, and prove that none of these generators is central. Therefore we
can conclude that the center of g(0) is exactly g(0)¯0.
As g(0)¯0 is the center of g(0), the space g(0)/g(0)¯0 bears the quotient Lie superalgebra
structure, entirely odd. Since g(0) = g(0)¯0 ⊕ g(0)¯1, this Lie structure is automatically trivial
(no need of looking at formulas whatsoever . . . ) because [g(0)¯0, g(0)¯0] ⊆ g(0)¯0. Therefore
g(0)/g(0)¯0 is Abelian, and isomorphic to C08 because dimC(g(0)¯1) = 8.
Finally, we stress the point that g(0) has non-trivial structure, as the lines from eighth to
(cid:4)
eleventh (in the table) display non-zero Lie brackets among some of its generators.
4.2 Second family: the Lie superalgebras g(cid:48)(σ)
4.2.1. Construction of the g(cid:48)(σ)'s. For any σ ∈ V (cf. Section 3.1), we define the Lie
Z(σ) over Z[σ] as follows. As a Z[σ]-module, g(cid:48)
superalgebra g(cid:48)
Z(σ) is spanned by the formal
α}α∈∆ subject only to the single
set of Z[σ]-linear generators B(cid:48)
relation σ1H(cid:48)
+ σ2H(cid:48)
+ σ3H(cid:48)
Z(σ) is defined by
the formulas in Section 3.3.2, which now we read as definitions of the Lie brackets among the
(linear) generators of g(cid:48)
, H(cid:48)
θ. The Lie superalgebra structure of g(cid:48)
θ}(cid:83){X(cid:48)
g := {H(cid:48)
= 2H(cid:48)
, H(cid:48)
, H(cid:48)
2ε3
2ε1
2ε2
2ε3
2ε1
2ε2
Z(σ) itself.
Singular Degenerations of Lie Supergroups of Type D(2, 1; a)
17
Altogether, the g(cid:48)
taking g(cid:48)(σ) := C ⊗Z[σ] g(cid:48)
superalgebras g(cid:48)(σ) now share C as their common ground ring.
manner of presenting the family of Kaplansky's Lie superalgebras gσ, in that
Z(σ)'s form a family of Lie superalgebras indexed by the points of V . Then
Z(σ) for all σ ∈ V we find a more regular situation, as these Lie
In fact, this is just another
for all σi ∈ V , we have g(cid:48)(σ) ∼= gσ
(4.2)
by Sections 3.1.1 and 3.3.2 -- cf. (3.6) -- and the very construction of g(cid:48)(σ). In fact, the analysis
in Section 3.3.2 (describing a C-basis of gσ, for any σ ∈ V , and its multiplication table) prove
that for all σ ∈ V , the Lie Z[σ]-superalgebra g(cid:48)
Z(σ) identifies to an integral Z[σ]-form of gσ.
We can formalize the description of the family {g(cid:48)(σ)}σ∈V proceeding like in Section 4.1.1; in
particular we keep the same notation, such as Z[x] := Z[V ] ∼= Z[x1, x2, x3]/(x1 + x2 + x3), etc.
In the construction of g(cid:48)
Z(σ), formally replace x to σ: this yields a meaningful definition of a Lie
superalgebra over Z[x], denoted by g(cid:48)
Z(x) by scalar extension.
Now definitions imply that, for any σ ∈ V , we have a Lie Z[σ]-superalgebra isomorphism
Z(x), and also g(cid:48)(x) := C[x]⊗Z[x] g(cid:48)
Z(σ) ∼= Z[σ] ⊗
g(cid:48)
Z[x]
g(cid:48)
Z(x)
-- through the ring isomorphism Z[σ] ∼= Z[x]/(xi − σi)i=1,2,3 -- and similarly
g(cid:48)(σ) ∼= C ⊗
C[x]
g(cid:48)(x)
C[x]
as Lie C-superalgebras, through the ring isomorphism C ∼= C[x]/(xi − σi)i=1,2,3.
One can argue similarly as in Section 4.1.1 to have a geometric picture of the above descrip-
tion: this amounts to literally replacing g(σ) with g(cid:48)(σ), which eventually provide a coherent
sheaf Lg(cid:48)
of complex Lie superalgebras over V , with a (C× × S3)-action on it, and a corre-
, whose fibres are the g(cid:48)(σ)'s; details are left to the reader. In the
sponding fibre bundle Lg(cid:48)
next result we describe these Lie superalgebras g(cid:48)(σ) (∼=gσ), for all σ ∈ V : like for the g(σ)'s,
the outcome is again that in the "regular" locus V × they are simple, while in the "singular"
locus V \ V × we can describe explicitly their non-simple structure.
Theorem 4.2. Given σ ∈ V , let a(cid:48)
(1) If σ ∈ V ×, then the Lie superalgebra g(cid:48)(σ) is simple.
(2) If σ ∈ V \ V ×, with σi = 0 and σj (cid:54)= 0 (cid:54)= σk for {i, j, k} = {1, 2, 3}, then if
, for all i ∈ {1, 2, 3}.
⊕ CX(cid:48)−2εi
i := CX(cid:48)
⊕ CH(cid:48)
C[x]
2εi
2εi
(cid:18) (cid:88)
b(cid:48)
i :=
(cid:19)
(cid:18)(cid:88)
(cid:19)
CX(cid:48)
α
⊕
CH(cid:48)
2εj
α(cid:54)=±2εi
(cid:69) g(cid:48)(σ) (a Lie ideal), a(cid:48)
j(cid:54)=i
we have b(cid:48)
isomorphisms b(cid:48)
of Lie superalgebras. In short, there exists a split short exact sequence
i ≤ g(cid:48)(σ) (a Lie subsuperalgebra), and there exist
∼= sl(2) and g(cid:48)(σ) ∼= sl(2) (cid:110) psl(22) -- a semidirect product
∼= psl(22), a(cid:48)
i
i
i
0 −−−−→ psl(22) ∼= b(cid:48)
(cid:76)(cid:57)(cid:57)−−−
−−−−−→a(cid:48)
A parallel result also holds true when dealing with g(cid:48)
i −−−−→ g(cid:48)(σ)
i
∼= sl(2) −−−−→ 0.
Z(σ) over the ground ring Z[σ].
(3) If σ = 0 (∈ V \ V ×), i.e., σh = 0 for all h ∈ {1, 2, 3}, then g(cid:48)(0)¯0
(super)algebras, the Lie (super)bracket is trivial on g(cid:48)(0)¯1 and g(cid:48)(0)¯1
over g(cid:48)(0)¯0
∼= sl(2)
g(cid:48)(0) ∼= g(cid:48)(0)¯0
⊕3. Finally, we have
(cid:110) g(cid:48)(0)¯1
∼= sl(2)
⊕3 (cid:110) 2(cid:2)3
∼= sl(2)
⊕3 as Lie
∼= 2(cid:2)3 as modules
18
K. Iohara and F. Gavarini
-- a semidirect product of Lie superalgebras. In other words, there is a split short exact
sequence
0 −−−−−→ 2(cid:2)3 ∼= g(cid:48)(0)¯1 −−−−−→ g(cid:48)(0)
⊕3 −−−−−→ 0.
A parallel result holds true when working with gZ(0) over the ground ring Z[0] = Z.
∼= sl(2)
(cid:76)(cid:57)(cid:57)−−−
−−−−−→g(cid:48)(0)¯0
Proof . Part (1) is a direct consequence of (4.2) and Proposition 3.1. Claim (2) instead follows
∼= sl(2) follows from the
easily from direct inspection of the formulas in Section 3.3.2. Indeed, a(cid:48)
first two lines of the table of formulas for Lie brackets in Section 3.3.2, which also show that a(cid:48)
is a Lie subsuperalgebra of g(cid:48)(σ) -- for all σ ∈ V , indeed.
i
i
Similarly, those formulas show that b(cid:48)
i is stable by the adjoint a(cid:48)
i-action exactly if and only
∈ b(cid:48)
i ⇐⇒ σi = 0, and also
σkH(cid:48)
2εk
θ = 2−1
3(cid:80)
(cid:3) = (1 − δj,k)σiX(cid:48)
k=1
in fact, the critical point is that H(cid:48)
if σi = 0:
, H(cid:48)
[X(cid:48)
2εi
θ] = ∓σiX(cid:48)
if σi = 0 -- for instance, because(cid:2)X(cid:48)
i ⇐⇒ σi = 0.
Moreover, the same formulas altogether show also that b(cid:48)
∈ b(cid:48)
, X(cid:48)
2εi
βj
βk
i
∼= psl(22) as claimed.
i is a direct sum complement of b(cid:48)
σi = 0, looking closely at the specific form of the Lie (sub)superalgebra b(cid:48)
show that b(cid:48)
is a(cid:48)
Finally, as a(cid:48)
i-stable (when σi = 0) we conclude that it is also a Lie ideal, q.e.d.
Claim (3) follows again by straightforward analysis of the table of formulas in Section 3.3.2 --
(cid:4)
i is a Lie subsuperalgebra and
i in g(cid:48)(σ), since b(cid:48)
just a matter of sheer bookkeeping.
∈ b(cid:48)
i is a Lie subsuperalgebra if and only
i ⇐⇒ σi = 0. Then, when
i these formulas also
2εi
Z (σ) over Z[σ] as follows. By definition, g(cid:48)(cid:48)
, H(cid:48)
4.3 Third family: the Lie superalgebras g(cid:48)(cid:48)(σ)
4.3.1. Construction of the g(cid:48)(cid:48)(σ)'s. For any σ ∈ V (cf. Section 3.1), we define the Lie
superalgebra g(cid:48)(cid:48)
Z (σ) is the Z[σ]-module spanned by the
set of formal (linear) generators Bg(cid:48)(cid:48) := {H(cid:48)
single relation σ1H(cid:48)
Z (σ) is
defined by the formulas in Section 3.3.3, which now must be read as definitions for Lie brackets
among the (linear) generators of g(cid:48)(cid:48)
ting g(cid:48)(cid:48)(σ) := C ⊗Z[σ] g(cid:48)(cid:48)
superalgebras g(cid:48)(cid:48)(σ) share C as their common ground ring. In addition
in the non-singular case, i.e., for σi ∈ V ×, we have g(cid:48)(cid:48)(σ) ∼= gσ
Z (σ)'s altogether form a family of Lie superalgebras indexed by the points of V . Set-
Z (σ) for all σ ∈ V we find a more regular situation, in that these Lie
, H(cid:48)
θ. The Lie superalgebra structure of g(cid:48)(cid:48)
= 2H(cid:48)
θ}(cid:83){Xα}α∈∆ subject only to the
+ σ2H(cid:48)
+ σ3H(cid:48)
The g(cid:48)(cid:48)
, H(cid:48)
Z (σ).
(4.3)
2ε3
2ε1
2ε2
2ε2
2ε1
2ε3
by Section 3.3.3 and the very definition of g(cid:48)(cid:48)(σ) itself.
Indeed, from Section 3.3.3 -- where
a C-basis and its multiplication table for gσ, with σ ∈ V ×, are described -- we see that for all
σ ∈ V ×, the Lie superalgebra g(cid:48)(cid:48)
Keeping notation as before, we can describe the family {g(cid:48)(cid:48)(σ)}σ∈V in a formal way, taking its
"version over Z[x]", denoted by g(cid:48)(cid:48)Z(x) -- just replacing the complex parameters (σ1, σ2, σ3) =: σ
with a triple of formal parameters (x1, x2, x3) =: x adding to zero -- and its complex-based
counterpart g(cid:48)(cid:48)(x) := C[x]⊗Z[x] g(cid:48)(cid:48)Z(x). Then the very construction implies that, for any σ ∈ V ,
one has a Lie Z[σ]-superalgebra isomorphism
Z (σ) over Z[σ] identifies to an integral Z[σ]-form of gσ.
-- through Z[σ] ∼= Z[x]/(xi − σi)i=1,2,3 -- and similarly
Z (σ) ∼= Z[σ] ⊗
g(cid:48)(cid:48)
Z[x]
g(cid:48)(cid:48)Z(x)
g(cid:48)(cid:48)(σ) ∼= C ⊗
C[x]
g(cid:48)(cid:48)(x)
Singular Degenerations of Lie Supergroups of Type D(2, 1; a)
19
as Lie C-superalgebras, through C ∼= C[x]/(xi − σi)i=1,2,3. Finally, the reader can easily mimick
what is done in Section 4.1.1 and find a geometric description of the family of the g(cid:48)(cid:48)(σ)'s.
sheaf Lg(cid:48)(cid:48)
C[x]
responding fibre bundle Lg(cid:48)(cid:48)
C[x]
Like in Section 4.1.1, we can re-cast all this in geometrical terms, defining a coherent
of complex Lie superalgebras over V , with a (C× × S3)-action on it, and a cor-
with the g(cid:48)(cid:48)(σ)'s as fibres; the reader can easily fill in the details.
In the next result we describe these Lie superalgebras g(cid:48)(cid:48)(σ), for all σ ∈ V :
like for the
previous two families, the outcome is that in the "regular" locus V × they are simple, while in
the "singular" locus V \ V × we can describe explicitly their non-simple structure.
Theorem 4.3. Let σ ∈ V , and set c(cid:48)(cid:48)
and a quotient superspace of g(cid:48)(cid:48)(σ), in general -- for all i ∈ {1, 2, 3}.
(1) If σ ∈ V ×, then the Lie superalgebra g(cid:48)(cid:48)(σ) is simple.
(2) If σ ∈ V \ V ×, with σi = 0 and σj (cid:54)= 0 (cid:54)= σk for {i, j, k} = {1, 2, 3}, then c(cid:48)(cid:48)
:= (CX2εi ⊕ CX−2εi), b(cid:48)(cid:48)
i -- a subsuperspace
:= g(σ)/c(cid:48)(cid:48)
i
i
i := (CX2εi ⊕
i is a quotient Lie superalgebra of g(cid:48)(cid:48)(σ);
CX−2εi) is an Abelian Lie ideal of g(cid:48)(cid:48)(σ), hence b(cid:48)(cid:48)
therefore, there exists a short exact sequence
i :=(cid:0)CX2εi ⊕ CX−2εi
(cid:1) −−−−→ g(cid:48)(cid:48)(σ) −−−−→ b(cid:48)(cid:48)
(cid:0)CH(cid:48)
:= (cid:0)(cid:76)
(cid:1)(cid:1)(cid:76)(cid:0)(cid:76)
i −−−−→ 0.
Furthermore, setting d(cid:48)(cid:48)
the quotient Lie superalgebra b(cid:48)(cid:48)
superalgebra of b(cid:48)(cid:48)
product of Lie superalgebras. In short, there exists a split short exact sequence
i is a Lie ideal and CH(cid:48)
2εi) (cid:110) d(cid:48)(cid:48)
j(cid:54)=i
i -- we have that d(cid:48)(cid:48)
⊕ CX+2εj ⊕ CX−2εj
∼= psl(22), CH(cid:48)
∼= C, and b(cid:48)(cid:48)
i , with d(cid:48)(cid:48)
∼= (CH(cid:48)
2εj
2εi
i
i
i
γ∈∆¯1
2εi is a Lie sub-
i -- a semidirect
CXγ
(cid:1) -- in
0 −−−−→ c(cid:48)(cid:48)
0 −−−−−→ psl(22) ∼= d(cid:48)(cid:48)
i −−−−−→ b(cid:48)(cid:48)
i
(cid:76)(cid:57)(cid:57)−−−
−−−−−→CH(cid:48)
2εi
∼= C −−−−−→ 0.
A parallel result holds true when working with g(cid:48)(cid:48)
Z (σ) over the ground ring Z[σ].
(3) If σ = 0(∈ V \ V ×), i.e., σh = 0 for all h ∈ {1, 2, 3}, then c(cid:48)(cid:48) :=
c(cid:48)(cid:48)
i is an Abelian Lie
ideal, hence b(cid:48)(cid:48) := g(cid:48)(cid:48)(σ)/c(cid:48)(cid:48) is a quotient Lie superalgebra (of g(cid:48)(cid:48)(σ)); therefore, there exists
a short exact sequence
i=1
3(cid:76)
0 −−−−−→ c(cid:48)(cid:48) :=(cid:76)3
0 −−−−−→(cid:76)
i=1c(cid:48)(cid:48)
i −−−−−→ g(cid:48)(cid:48)(0) −−−−−→ b(cid:48)(cid:48) −−−−−→ 0.
Moreover, there exists a second, split short exact sequence
−−−−−→(cid:76)3
CX α −−−−−→ b(cid:48)(cid:48) (cid:76)(cid:57)(cid:57)−−−
CX α (cid:69) b(cid:48)(cid:48) is an Abelian Lie ideal and (cid:76)3
(cid:1)(cid:110)(cid:0)(cid:76)
where (cid:76)
subsuperalgebra (of b(cid:48)(cid:48)), so that b(cid:48)(cid:48) ∼=(cid:0)(cid:76)3
CH(cid:48)
α∈∆¯1
α∈∆¯1
i=1
i=1
2εi
CH(cid:48)
CH(cid:48)
i=1
α∈∆¯1
2εi −−−−−→ 0,
2εi ≤ b(cid:48)(cid:48) is an Abelian Lie
CX α
(cid:1) -- a semidirect product
of Lie superalgebras.
A parallel result holds true when working with g(cid:48)(cid:48)(0) over the ground ring Z[0] = Z.
Proof . Claim (1) is a direct consequence of (4.3) along with Proposition 3.1. Claims (2)
and (3), like for the previous, parallel results, both follow as direct outcome of the formulas for
Lie brackets among linear generators of g(cid:48)(cid:48)(σ), which we read from Section 3.3.3.
is zero when σi = 0, so that c(cid:48)(cid:48)
i
is then an Abelian Lie subsuperalgebra. Moreover, c(cid:48)(cid:48)
i is stable for the adjoint action by elements
of {H2ε1, H2ε2, H2ε3, Hθ} by construction (this holds true for every σ ∈ V indeed). Finally, c(cid:48)(cid:48)
i
is also stable for the adjoint action by odd root vectors such as X±βj and X±θ because the
Indeed, for claim (2) we notice that [X2εi, X−2εj ] = δi,jσ2
i H(cid:48)
2εi
20
K. Iohara and F. Gavarini
3(cid:76)
i is an Abelian Lie ideal of g(cid:48)(cid:48)(σ).
formulas in fifth, sixth and seventh line of the table in Section 3.3.3 all give zero Lie brackets
when σi = 0. Overall, this means that c(cid:48)(cid:48)
The claim about b(cid:48)(cid:48) := g(cid:48)(cid:48)(σ)/c(cid:48)(cid:48) and the short exact sequence now are obvious consequences
of c(cid:48)(cid:48)
i being a Lie ideal.
As to the rest of claim (1), one sees again that everything follows from straightforward
bookkeeping, nothing more. For claim (3) one has again to carry out a similar analysis. For
is an Abelian Lie ideal because of claim (2) and the fact that the c(cid:48)(cid:48)
instance, c(cid:48)(cid:48) :=
i 's
commute with each other.
Something less obvious only occurs with the analysis of(cid:76)
are trivial instead in its subquotient(cid:76)
CX α. Indeed, the fact that
this is an Abelian Lie ideal of b(cid:48)(cid:48) follows once more from the formulas in Section 3.3.3 but one
also has to pay attention to some detail.
Indeed, a first bunch of Lie brackets among odd root vectors which are non-zero in g(0) but
α∈∆¯1
c(cid:48)(cid:48)
i=1
i
[Xβi, Xβj ] = (1 − δi,j)X2εk = −0,
[Xβi, X−θ] = X−2εi = −0,
[X−βi, Xθ] = −X2εi = −0.
α∈∆¯1
CX α are the following:
[X−βi, X−βj ] = −(1 − δi,j)X−2εk = −0,
Second, the remaining non-obvious Lie brackets are described by the two formulas
[Xβi, X−βj ] = δi,j(σiH(cid:48)
2εi − H(cid:48)
θ),
+ σ3H(cid:48)
[Xθ, X−θ] = H(cid:48)
= 2H(cid:48)
θ
but the relation σ1H(cid:48)
from the last formulas above we get [Xβi, X−βj ] = 0 and [Xθ, X−θ] = 0 in b(cid:48)(cid:48).
θ in g(cid:48)(cid:48)(σ) reads now H(cid:48)
+ σ2H(cid:48)
2ε1
2ε2
2ε3
θ = 0 in g(cid:48)(cid:48)(0), hence
All other parts of claim (3) follow equally from a similar analysis -- a sheer matter of book-
(cid:4)
keeping -- so we leave them to the reader.
4.4 Degenerations from contractions: the (cid:98)g(σ)'s and the (cid:98)g(cid:48)(σ)'s
We finish our study of remarkable integral forms of gσ by introducing some further ones, that all
are obtained through a general construction; when specializing these forms, one obtains again
degenerations, now of the kind that is often referred to as "contraction" (see, e.g., [5]). We
start with a very general construction. Let R be a (commutative, unital) ring, and let A be an
"algebra" (not necessarily associative, nor unitary), in some category of R-bimodules, for some
"product" denoted by "·": we assume in addition that
F · C ⊆ C,
F · F ⊆ F,
A = F ⊕ C
C · F ⊆ C,
C · C ⊆ F
(4.4)
with
Choose now τ be a non-unit in R, and correspondingly consider in A the R-submodules
Fτ := F,
Cτ := τC,
Aτ := Fτ + Cτ = F ⊕ (τC).
(4.5)
Fix also a (strict) ideal I (cid:69) R; then set RI := R/I for the corresponding quotient ring, and use
∼= (R/I) ⊗R F and Cτ,I := Cτ /ICτ =
notation Aτ,I := Aτ /IAτ
(τC)/(IτC) ∼= (R/I) ⊗R Cτ
∼= Fτ,I ⊕ Cτ,I as an
RI -module; moreover,
Fτ,I · Fτ,I ⊆ Fτ,I ,
∼= (R/I) ⊗R Aτ , Fτ,I := Fτ /IFτ
∼= (R/I) ⊗R (τC). By construction we have Aτ,I
Cτ,I · Cτ,I ⊆ τ 2Fτ,I ,
Cτ,I · Fτ,I ⊆ Cτ,I ,
Fτ,I · Cτ,I ⊆ Cτ,I ,
where the last identity comes from Cτ · Cτ = τ 2(C · C) ⊆ τ 2F = τ 2Fτ and we write τ :=
(τ mod I) ∈ R/I. In particular, if τ ∈ I, then Cτ,I · Cτ,I = {0} and we get
Aτ,I = Fτ,I (cid:110) Cτ,I ,
(4.6)
Singular Degenerations of Lie Supergroups of Type D(2, 1; a)
21
where Cτ,I bears the Fτ,I -bimodule structure induced from A and is given a trivial product, so that
it sits inside Aτ,I as a two-sided Abelian ideal, with (4.6) being a semidirect product splitting.
In fact, Aτ,I is what is called a "central extension of Fτ,I by Cτ,I ".
In short, for τ ∈ I this process leads us from the initial object A, that splits into A = F ⊕ C
as R-module, to the final object Aτ,I = Fτ,I (cid:110)Cτ,I , now split as a semidirect product. Following
[5, Section 2 and references therein], we shall refer to this process as "contraction", and also
refer to Aτ,I as to a "contraction of A". Note, however, that these are contractions of a very
special type, in that only the odd part is "contracted":
in the general theory of contractions
of Lie superalgebras, instead, one has to do with a richer variety of objects -- cf., e.g., [17].
For our purpose however we do not need the general theory in its full extent. We apply now
the above contraction procedure to a couple of integral forms of gσ. First consider the case
A := gZ(x), F := gZ(x)¯0 and C := gZ(x)¯1; here the ground ring is R := Z[x], and we choose in
it τ := x1x2x3 and the ideal I = Iσ generated by x1 − σ1, x2 − σ2 and x3 − σ3. In this case,
also with the simpler notation (cid:98)gZ(x) := gZ(x)τ ; similarly we write (cid:98)gZ(σ) := gZ(x)τ,Iσ
the "blown-up" Lie superalgebras in (4.5) reads gZ(x)τ = gZ(x)¯0 ⊕ (τ gZ(x)¯1), that we write
that each(cid:98)gZ(σ) for non-singular σ ∈ V × is yet another Z[σ]-integral form of our initial complex
. Note
Lie superalgebra gσ. Similarly occurs if we work over C, i.e., when we consider A := g(x),
contraction (cid:98)g(σ) := g(x)τ,Iσ
F := g(x)¯0, C := g(x)¯1 and the blown-up algebra g(x)τ with ground ring R := C[x], and the
(cid:1)-action on it, and an associated
over C. This gives (as in Section 4.1.1) a new coherent sheaf of
complex Lie superalgebras over V , say L(cid:98)gC[x]
with the(cid:98)g(σ)'s as fibres; details are left to the reader. Next result describes
fibre bundle L(cid:98)gC[x]
the structure of the(cid:98)g(σ)'s.
Theorem 4.4. Let σ ∈ V , i ∈ {1, 2, 3},(cid:98)ai := CX2εi ⊕ CH2εi ⊕ CX−2εi (= ai of Section 3.2.1).
(1) If σ ∈ V ×, then the Lie superalgebra(cid:98)g(σ) is simple.
(2) If σ ∈ V \ V ×, with σi = 0 and σj (cid:54)= 0 (cid:54)= σk for {i, j, k} = {1, 2, 3}, then (cid:98)ai (cid:69) (cid:98)g is
a central Lie ideal in(cid:98)g(σ), with(cid:98)ai
∼= C30, while(cid:98)aj
∼=(cid:98)ak
∼= sl(2) for {j, k} = {1, 2, 3}\{i}.
(cid:98)g(σ) ∼=(cid:98)g(σ)¯0
(cid:110)(cid:98)g(σ)¯1
∼= C30 ⊕ sl(2) ⊕ sl(2) while(cid:98)g(σ)¯1 is endowed with trivial Lie bracket
(cid:96)=1(cid:98)a(cid:96)
with(cid:98)g(σ)¯0 = ⊕3
and (cid:98)g(σ)¯1
(cid:98)g(σ)¯0
∼= ((cid:4) ⊕(cid:4)) (cid:2) 2 (cid:2) 2 -- where (cid:4) is the trivial representation -- as a module over
∼= C30 ⊕ sl(2)⊕ sl(2); so there exists a split short exact sequence of Lie superalgebras
∼= ((cid:4) ⊕(cid:4)) (cid:2) 2 (cid:2) 2 −−→(cid:98)g(σ)
0 −−→(cid:98)g(σ)¯1
∼= C30 ⊕ sl(2) ⊕ sl(2) −−→ 0.
A parallel result also holds true when working with(cid:98)gZ(σ) over the ground ring Z[σ].
(3) If σ = 0(∈ V \ V ×), i.e., σh = 0 for all h ∈ {1, 2, 3}, then(cid:98)g(0) is the Abelian complex Lie
superalgebra of superdimension 98, that is(cid:98)g(0) ∼= C98 with trivial bracket.
A parallel result holds true when working with(cid:98)gZ(0) over the ground ring Z[0] = Z.
, with a(cid:0)C× × S3
Moreover, we have a semidirect product splitting
(cid:76)(cid:57)(cid:57)−−−→(cid:98)g(σ)¯0
(cid:1)
Proof . The claim follows directly from Theorem 4.1 once we take also into account the fact
that the(cid:98)g(σ)'s are specializations of g(x)τ , and for singular values σ ∈ V × any such speciali-
zation is a contraction of (cid:98)g(σ), of the form (cid:98)g(σ) = g(x)τ,I for the element τ := x1x2x3 and
explicit formulas for (linear) generators of(cid:98)g(x): indeed, the latter are easily obtained as slight
i=1,2,3}. Otherwise, one can deduce the statement directly from the
the ideal I := {(xi − σi
modification -- taking into account that odd generators must be "rescaled" by the coefficient
τ := x1x2x3 -- of the similar formulas in Section 3.3.1 for g(σ), which read as formulas for g(x)
(cid:4)
just switching the σ(cid:96)'s into x(cid:96)'s.
22
K. Iohara and F. Gavarini
C[x]
i=1
2i
i=1
τ,Iσ
C[x]
Z(x)
τ
Z(x)
τ
2εi
2εi
= g(cid:48)
Z(x)¯0
Z(x)¯0
Z(x)¯1
Z(x)¯1
Z(x)
and C := g(cid:48)
⊕ (τ g(cid:48)
Z(σ) := g(cid:48)
As a second instance, we consider the case A := g(cid:48)
Z(x), F := g(cid:48)
. Again, each(cid:98)g(cid:48)
like in Section 4.1.1: details are left to the reader). Next result describes the structure of these
; similarly we write also(cid:98)g(cid:48)
; the
ground ring is again R := Z[x], and again we choose in it τ := x1x2x3 and the ideal I generated
by x1−σ1, x2−σ2 and x3−σ3. In this second case, we have again a "blown-up" Lie superalgebra
(cid:98)g(cid:48)
as in (4.5), that now reads g(cid:48)
), for which we use the simpler notation
Z(x) := g(cid:48)
Z(σ) for non-singular
σ ∈ V × is another Z[σ]-integral form of the complex Lie superalgebra gσ we started with.
algebra g(cid:48)(x)τ with ground ring R := C[x], and the contraction(cid:98)g(cid:48)(σ) := g(cid:48)(x)τ,Iσ
Similarly, working over C we consider A := g(cid:48)(x), F := g(cid:48)(x)¯0, C := g(cid:48)(x)¯1 and the blown-up
over C. This
provides one more coherent sheaf of complex Lie superalgebras over V , denoted L(cid:98)g(cid:48)
having the(cid:98)g(cid:48)(σ)'s as fibres (just
, with
a (C× × S3)-action on it, and an associated fibre bundle L(cid:98)g(cid:48)
fibres(cid:98)g(cid:48)(σ)'s:
Theorem 4.5. Let σ ∈ V , and(cid:98)a(cid:48)
(1) If σ ∈ V ×, then the Lie superalgebra(cid:98)g(cid:48)(σ) is simple.
(2) If σ ∈ V \ V ×, then (cid:98)g(cid:48)(σ)¯0
⊕3 and the Lie bracket is trivial on(cid:98)g(cid:48)(σ)¯1; finally, we have semidirect
over(cid:98)g(cid:48)(σ)¯0
(cid:1).
(cid:110)(cid:98)g(cid:48)(σ)¯1
−−−−−→(cid:98)g(cid:48)(σ)¯0
⊕3 as Lie superalgebras, (cid:98)g(cid:48)(σ)¯1
∼= sl(2)
⊕3 (cid:110)(cid:0)(cid:2)3
(cid:98)g(cid:48)(σ) ∼=(cid:98)g(cid:48)(σ)¯0
0 −−−−−→ 2(cid:2)3 ∼=(cid:98)g(cid:48)(σ)¯1 −−−−−→(cid:98)g(cid:48)(σ)
In other words, there exists a split short exact sequence
(cid:76)(cid:57)(cid:57)−−−
⊕ CX(cid:48)−2εi
product splittings
for all i = 1, 2, 3.
2i as modules
i := CX(cid:48)
∼= sl(2)
∼= sl(2)
∼= sl(2)
⊕ CH(cid:48)
∼= (cid:2)3
⊕3 −−−−−→ 0.
Z(σ) over the ground ring Z[σ].
A parallel result also holds true when working with(cid:98)g(cid:48)
Proof . One can easily deduce the claim from Theorem 4.2 along with the fact that each(cid:98)g(cid:48)(σ)
is indeed a contraction of(cid:98)g(cid:48)(σ), namely of the form(cid:98)g(cid:48)(σ) = g(cid:48)(x)τ,I for the element τ := x1x2x3
is a specialization of g(cid:48)(x)τ , and in particular for singular values σ ∈ V × any such specialization
means of a direct analysis of the explicit formulas for (linear) generators of(cid:98)g(cid:48)(x): in fact, one
and the ideal I := {(xi − σi)i=1,2,3}. As alternative method, one can obtain the statement by
easily obtains such formulas as slight modifications -- taking into account the "rescaling" of odd
generators by the coefficient τ := x1x2x3 -- of the formulas in Section 3.3.2. We leave details to
(cid:4)
the interested reader.
Remark 4.6. We considered five families of Lie superalgebras, denoted by {g(σ)}σ∈V ,
{g(cid:48)(σ)}σ∈V , {g(cid:48)(cid:48)(σ)}σ∈V , {(cid:98)g(σ)}σ∈V and {(cid:98)g(cid:48)(σ)}σ∈V , all being indexed by the points of the
V ∩(cid:0)(cid:83)
Lie superalgebras over Spec(C[x]) ∼= V ∪{(cid:63)}(cid:0)∼=A2C ∪{(cid:63)}(cid:1) share the same stalks on all "general"
complex plane V . Now, our analysis shows that these five families share most of their elements,
namely all those indexed by "general points" σ ∈ V × := V ∩ (C×)3. On the other hand,
the five families are drastically different at all points in the "singular locus" S := V \ V × =
of
i=1,2,3{σi = 0}(cid:1). In other words, the five sheaves LgC[x]
points (i.e., those outside S), and have different stalks instead on "singular" points (i.e., those
in S). Likewise, the five fibre bundles LgC[x]
over Spec(C[x])
share the same fibres on all general points and have different fibres on singular points. Let
us also stress that the second family {g(cid:48)(σ)}σ∈V is just Kaplansky's one, as g(cid:48)(σ) ∼= gσ -- cf.
Section 4.2. The outcome of the previous discussion is, loosely speaking, that our construction
provides five different "completions" of the family {gσ}σ∈V \S of simple Lie superalgebras, by
adding -- in five different ways -- some new non-simple extra elements on top of each point of
and L(cid:98)g(cid:48)
and L(cid:98)g(cid:48)
, L(cid:98)gC[x]
, L(cid:98)gC[x]
, Lg(cid:48)(cid:48)
C[x]
, Lg(cid:48)(cid:48)
C[x]
, Lg(cid:48)
, Lg(cid:48)
C[x]
C[x]
C[x]
C[x]
Singular Degenerations of Lie Supergroups of Type D(2, 1; a)
23
the "singular locus" S. In particular, this shows that it makes no sense to speak of "taking the
limit for σ going to S" of the simple Lie superalgebras gσ (for σ ∈ V ), unless one states exactly
what is the total family -- namely, indexed over all of V -- of Lie superalgebras one has chosen
to complete the family {gσ}σ∈V \S. Indeed, as our results show, depending on such a choice one
finds very different, non-isomorphic "limits".
5 Lie supergroups of type D(2, 1; σ):
presentations and degenerations
In this section, we introduce (complex) Lie supergroups of type D(cid:0)2, 1; σ), basing on the five
families of Lie superalgebras introduced in Section 4 and following the approach of Section 2.4.3.
For simplicity, we formulate everything over C, but the reader may see some subtleties to discuss
about the Chevalley groups over a Z[σ]-algebra. The latter had been discussed in [6] and [7] for
some basis, i.e., for one particular choice of Z[σ]-integral form (though with slightly different
formalism); in the present case everything works similarly, up to paying attention to the σ-
dependence of the commutation relations of the Z[σ]-form one chooses (cf. Section 4). The
details are left to the reader.
5.1 First family: the Lie supergroups Gσ
Given σ = (σ1, σ2, σ3) ∈ V , let g = g(σ) be the complex Lie superalgebra associated with σ as in
Section 4.1, and g¯0 its even part. We recall that g is spanned over C by {H2ε1, H2ε2, H2ε3, Hθ}∪
{Xα}∆. Like in Section 4.1, we set ai := CX2εi ⊕ CH2εi ⊕ CX−2εi for each i -- all these being Lie
i=1ai. When σi (cid:54)= 0, the Lie algebra ai is isomorphic to sl(2):
subalgebras of g(σ), with g(σ)¯0 = ⊕3
an explicit isomorphism is realized by mapping X2εi (cid:55)→ σie, H2εi (cid:55)→ σih and X−2εi (cid:55)→ σif , where
∼= C⊕3 becomes the 3-dimensional
{e, h, f} is the standard basis sl(2). When σi = 0 instead, ai
Abelian Lie algebra.
Let us now set Ai := SL2 if σi (cid:54)= 0 and Ai := C× C× × C if σi = 0, and define G := ×3
i=1Ai --
a complex Lie group such that Lie(G) = g(σ)¯0. One sees that the adjoint action of g(σ)¯0
onto g(σ) integrates to a Lie group action of G onto g(σ) again, so that the pair Pσ :=(cid:0)G, g(σ)(cid:1) --
endowed with that action -- is a super Harish-Chandra pair (cf. Section 2.4.1); note that its
dependence on σ lies within all its constituents: the structure of G, the Lie superalgebra g(σ),
and the action of the former onto the latter.
Finally, we let
Gσ := GPσ
be the complex Lie supergroup associated with the super Harish-Chandra pair Pσ trough the
category equivalence given in Section 2.4.3.
5.1.1. A presentation of Gσ. We shall now provide an explicit presentation by generators
and relations for the supergroups Gσ, i.e., for the abstract groups Gσ(A), A ∈ (Wsalg).
To begin with, inside each subgroup Ai we consider the elements
x2εi(c) := exp(cX2εi),
x−2εi(c) := exp(cX−2εi)
for every c ∈ C; then Γi := {x2εi(c), h2εi(c), x−2εi(c)}c∈C is a generating set for Ai.
We define also elements hθ(c) := exp(cid:0)cHθ
(cid:1) for all c ∈ C: then the commutation relations
h2εi(c) := exp(cH2εi),
[H2εr , H2εs] = 0 and H2ε1 + H2ε2 + H2ε3 = 2Hθ inside g(σ) together imply the group relations
h2ε1(c)h2ε2(c)h2ε3(c) = hθ(c)2 for all c ∈ C.
24
K. Iohara and F. Gavarini
The complex Lie group G is clearly generated by
Γ¯0 := {x2εi(c), h2εi(c), hθ(c), x−2εi(c)}i∈{1,2,3}
c∈C
(the hθ(c)'s might be dropped, but we prefer to add them too as generators).
In addition, when we consider G as a (totally even) supergroup and we look at it as a group-
valued functor G : (Wsalg) −→ (grps), the abstract group G(A) of its A-points -- for A ∈
(Wsalg) -- is generated by the set
.
a∈A¯0
Γ¯0(A) := {x2εi(a), h2εi(a), hθ(a), x−2εi(a)}i∈{1,2,3}
Following the recipe in Section 2.4.3, in order to generate the group Gσ(A) := GPσ
(5.1)
Note that here the generators do make sense -- as operators in GL(A ⊗ g(σ)), but also
formally -- since A = C ⊕ N(A) (cf. Section 2.1), so each a ∈ A reads as a = c + na for some
c ∈ C and a nilpotent na ∈ N(A), hence exp(aX2εi) = exp(cX2εi) exp(naX2εi), etc., are all
well-defined.
(A), beside
the subgroup G(A) we need also all the elements of the form (1 + ηiYi) with (i, ηi) ∈ I × A¯1 -- cf.
Section 2.4.3 -- where now the C-basis {Yi}i∈I of g¯1 is {Yi}i∈I = {X±θ, X±βi}i=1,2,3. Therefore,
we introduce notation x±θ(η) := (1 + ηX±θ
and we consider the set Γ¯1(A) := {x±θ(η), x±βi(η) η ∈ A¯1}.
Now, taking into account that G(A) is generated by Γ¯0(A), we can modify the set of relations
given in Section 2.4.3 by letting g ∈ G(A) range inside the set Γ¯0(A): then we can find the
following full set of relations (where hereafter we freely use notation eZ := exp(Z)):
(cid:1), x±βi(η) := (1 + ηX±βi) for all η ∈ A¯1, i ∈ {1, 2, 3},
hθ(a)x±θ(η)hθ(a)
−1 = x±θ(η),
∀ g(cid:48), g(cid:48)(cid:48) ∈ G(A),
−δi,j σiaη(cid:1),
(cid:0)e±(−1)
(cid:0)e±σiaη(cid:1),
(cid:0)e∓σiaη(cid:1),
g(cid:48) · g(cid:48)(cid:48) = g(cid:48) ·G g(cid:48)(cid:48),
−1 = x±βj
−1 = x±θ
−1 = x±βi
−1 = xβj (η)xθ(δi,jσiaη),
−1 = x−βj (η)xβk ((1 − δi,j)σiaη),
−1 = xβj (η)x−βk ((1 − δi,j)σiaη),
−1 = x−βj (η)x−θ(δi,jσiaη),
1G = 1,
h2εi(a)x±βj (η)h2εi(a)
h2εi(a)x±θ(η)h2εi(a)
hθ(a)x±βi(η)hθ(a)
x2εi(a)xβj (η)x2εi(a)
x2εi(a)x−βj (η)x2εi(a)
x−2εi(a)xβj (η)x−2εi(a)
x−2εi(a)x−βj (η)x−2εi(a)
x2εi(a)xθ(η)x2εi(a)
x2εi(a)x−θ(η)x2εi(a)
x−2εi(a)xθ(η)x−2εi(a)
x−2εi(a)x−θ(η)x−2εi(a)
xβi(ηi)xβj (η(cid:48)
j) = x−2εk (−(1 − δi,j)η(cid:48)
x−βi(ηi)x−βj (η(cid:48)
j) = h2εi(δi,jη(cid:48)
xβi(ηi)x−βj (η(cid:48)
xβi(ηi)xθ(η) = xθ(η)xβi(ηi),
x−βi(ηi)xθ(η) = x2εi(−ηηi)xθ(η)x−βi(ηi),
xθ(η+)x−θ(η−) = hθ(η−η+)x−θ(η−)xθ(η+),
x±βi(η(cid:48))x±βi(η(cid:48)(cid:48)) = x±βi(η(cid:48) + η(cid:48)(cid:48)),
−1 = xθ(η),
−1 = x−θ(η)x−βi(σiaη),
−1 = xθ(η)xβi(σiaη),
−1 = x−θ(η)
j)xβi(ηi),
jηi)x−βj (η(cid:48)
j) = x2εk ((1 − δi,j)η(cid:48)
jηi)xβj (η(cid:48)
jηi)hθ(−δi,jη(cid:48)
jηi)x−βj (η(cid:48)
j)x−βi(ηi),
j)xβi(ηi),
xβi(ηi)x−θ(η) = x−2εi(ηηi)x−θ(η)xβi(ηi),
x−βi(ηi)x−θ(η) = x−θ(η)x−βi(ηi),
x±θ(η(cid:48))x±θ(η(cid:48)(cid:48)) = x±θ(η(cid:48) + η(cid:48)(cid:48))
with {i, j, k} ∈ {1, 2, 3}.
Singular Degenerations of Lie Supergroups of Type D(2, 1; a)
25
5.1.2. Singular specializations of the supergroup(s) Gσ. From the very construction of
the supergroups Gσ, we get that
Gσ is simple (as a Lie supergroup) for all σ = (σ1, σ2, σ3) ∈ V ×,
where we recall that a Lie supergroup is said to be simple if it has no non-trivial normal closed
connected Lie sub-supergroup. This follows from the presentation of Gσ in Section 5.1.1 above,
or it can be seen as a direct consequence of the relation Lie(Gσ) = g(σ) = gσ and of Proposi-
tion 3.1.
On the other hand, the situation is different at "singular values" of the parameter σ: the
following records the whole situation.
Theorem 5.1. Let σ ∈ V as usual, and keep notation as above.
(1) If σ ∈ V ×, then the Lie supergroup Gσ is simple.
(2) If σ ∈ V \ V ×, with σi = 0 and σj (cid:54)= 0 (cid:54)= σk for {i, j, k} = {1, 2, 3}, then Ai is a central
subgroup of Gσ, isomorphic to C × C× × C, and Gσ is the universal central extension of
PSL(22) by Ai; in other words, there exists a short exact sequence of Lie supergroups
1 −−−−−→ C × C× × C ∼= Ai −−−−−→ Gσ −−−−−→ PSL(22) −−−−−→ 1.
(3) If σ = 0(∈ V \ V ×), i.e., σh = 0 for all h ∈ {1, 2, 3}, then (Gσ)rd
∼= (cid:0)C × C× × C(cid:1)×3
∼= C8 is Abelian; in particular, Gσ is
is the center of Gσ, and the quotient Gσ/(Gσ
a central extension of C8 by (C × C× × C)×3, i.e., there exists a short exact sequence of
Lie supergroups, with non-Abelian middle term,
(cid:1)
rd
1 −−−−−→(cid:0)C × C× × C(cid:1)×3 ∼=(cid:0)Gσ
(cid:1)
rd −−−−−→ Gσ −−−−−→ C8 −−−−−→ 1.
Proof . The claim follows directly from the presentation of Gσ given in Section 5.1.1 above, or
(cid:4)
also from the relation Lie(Gσ) = g(σ) along with Theorem 4.1.
5.2 Second family: the Lie supergroups G(cid:48)
Given σ = (σ1, σ2, σ3) ∈ V , let g(cid:48) := g(cid:48)(σ) be the complex Lie superalgebra associated with σ
}i=1,2,3 of g(cid:48)
as in Section 4.2.1, and let g(cid:48)
¯0
as in Section 3.3, and set a(cid:48)
for each i: each one of these is a Lie
2 ⊕ a(cid:48)
subalgebra of g(cid:48)
¯0 = a(cid:48)
i is isomorphic to sl(2),
(cid:55)→ f , where {e, h, f} is
an explicit isomorphism being given by X(cid:48)
⊕3. For each i ∈ {1, 2, 3}, let A(cid:48)
the standard basis sl(2). It follows that g(cid:48)
1 × A(cid:48)
be a copy of SL2, and set G(cid:48) := A(cid:48)
3. By the previous analysis, Lie(G(cid:48)) is isomorphic
to g(cid:48)
¯0 and the Lie(G(cid:48))-action lifts to a holomorphic G(cid:48)-action on g(cid:48) again: in fact, one easily sees
that this action is faithful too. With this action, P(cid:48)
(cf. Section 2.4.1), which overall depends on P(cid:48)
σ (although G(cid:48) alone does not). Finally, we define
, H(cid:48)
⊕ CH(cid:48)
3. Moreover, each Lie algebra a(cid:48)
(cid:55)→ h and X(cid:48)−2εi
σ :=(cid:0)G(cid:48), g(cid:48)(cid:1) is a super Harish-Chandra pair
¯0 be its even part. Fix the C-basis {X(cid:48)
i := CX(cid:48)
1 ⊕ a(cid:48)
2εi
¯0 is isomorphic to sl(2)
2 × A(cid:48)
⊕ CX(cid:48)−2εi
(cid:55)→ e, H(cid:48)
¯0, with g(cid:48)
, X(cid:48)−2εi
2εi
2εi
2εi
2εi
i
σ
2εi
G(cid:48)
σ := GP(cid:48)
σ
to be the complex Lie supergroup associated with the super Harish-Chandra pair P(cid:48)
equivalence of categories given in Section 2.4.3.
5.2.1. A presentation of G(cid:48)
explicit presentation by generators and relations of all the abstract groups G(cid:48)
in (Wsalg). To this end, we first consider the Lie group G(cid:48) = A(cid:48)
σ can be described in concrete terms via an
σ(A), with A ranging
∼= SL2. Letting
σ. The supergroups G(cid:48)
σ via the
3 with A(cid:48)
2×A(cid:48)
1×A(cid:48)
i
∼= Lie(cid:0)G(cid:48)(cid:1) −−−−→ G(cid:48) be the exponential map, we consider x(cid:48)
26
exp : g(cid:48)
¯0
θ(c) := exp(cH(cid:48)
h(cid:48)
that the commutation relations [H(cid:48)
inside g(cid:48)(σ)C together imply, inside G(cid:48)
(σ1c)h(cid:48)
for all c ∈ C. The complex Lie group G(cid:48) is clearly generated by the set
2εr , H(cid:48)
+, the group relations h(cid:48)
2εs] = 0 along with σ1H(cid:48)
(c) := exp(cX(cid:48)−2εi
(c) := exp(cH(cid:48)
), x(cid:48)−2εi
) and h(cid:48)
2ε1
2ε2
2ε1
2εi
2εi
+ σ2H(cid:48)
K. Iohara and F. Gavarini
2εi
(c) := exp(cX(cid:48)
),
θ) for all c ∈ C. Note
= 2H(cid:48)
(σ2c)h(cid:48)
θ(c)2
(σ3c) = h(cid:48)
+ σ3H(cid:48)
2ε3
2ε2
2εi
2ε3
θ
θ(c), x(cid:48)
2εi(c), h(cid:48)
¯0 := {x(cid:48)
2εi(c), h(cid:48)
Γ (cid:48)
(where the h(cid:48)
θ(c)'s might be discarded, but we prefer to keep them). Then, looking at G(cid:48) as
a (totally even) supergroup thought of as a group-valued functor G(cid:48) : (Wsalg) −→ (grps), each
abstract group G(cid:48)(A) of its A-points -- for A ∈ (Wsalg) -- is generated by the set
−2εi(c) c ∈ C}
2εi(a), h(cid:48)
2εi(a), h(cid:48)
¯0(A) := {x(cid:48)
Γ (cid:48)
−2εi(a) a ∈ A¯0}.
θ(a), x(cid:48)
Following Section 2.4.3, we need as generators of G(cid:48)
(cid:9)
σ(A) := GP(cid:48)
and all those of the form x(cid:48)
±θ) or x(cid:48)
±θ(η) := (1 + ηX(cid:48)
±βi
}i=1,2,3 as our C-basis of g(cid:48)
i∈I = {X(cid:48)
±θ, X(cid:48)
±βi
(η) η ∈ A¯1}.
¯1(A) := {x(cid:48)
±θ(η), x(cid:48)
±βi
i ∈ {1, 2, 3} -- since now we fix(cid:8)Y (cid:48)
(5.2)
(A) all the elements of G(cid:48)(A)
) with η ∈ A¯1 and
¯1; we denote the
Implementing the recipe in Section 2.4.3, and recalling that G(cid:48)(A) is generated by Γ (cid:48)
¯0(A),
we can now slightly modify the relations presented in Section 2.4.3 and consider instead the
following, alternative full set of relations among the generators of G(cid:48)
set of all the latter by Γ (cid:48)
(η) := (1 + ηX(cid:48)
±βi
σ
i
σ(A):
∀ g(cid:48), g(cid:48)(cid:48) ∈ G(cid:48)(A),
−δi,j aη(cid:1),
θ(δi,jaη),
(η)x(cid:48)
(η)x(cid:48)
βj
−βj
((1 − δi,j)aη),
βk
((1 − δi,j)aη),
−βk
(η)x(cid:48)
−θ(δi,jaη),
θ(a)x(cid:48)
h(cid:48)
±θ(η)h(cid:48)
θ(a)
−1 = x(cid:48)
±θ(η,
(cid:0)e±(−1)
(cid:0)e±aη(cid:1),
(cid:0)e∓σiaη(cid:1),
= 1,
g(cid:48) · g(cid:48)(cid:48) = g(cid:48) ·G(cid:48) g(cid:48)(cid:48),
1(cid:48)
G
−1 = x(cid:48)
(η)h(cid:48)
h(cid:48)
2εi(a)x(cid:48)
±βj
±βj
2εi(a)
−1 = x(cid:48)
±θ(η)h(cid:48)
h(cid:48)
2εi(a)x(cid:48)
±θ
2εi(a)
−1 = x(cid:48)
(η)h(cid:48)
θ(a)x(cid:48)
h(cid:48)
±βi
±βi
θ(a)
−1 = x(cid:48)
(η)x(cid:48)
(η)x(cid:48)
x(cid:48)
2εi(a)x(cid:48)
2εi(a)
βj
βj
−1 = x(cid:48)
(η)x(cid:48)
x(cid:48)
2εi(a)x(cid:48)
−βj
−βj
2εi(a)
−1 = x(cid:48)
(η)x(cid:48)
−2εi(a)x(cid:48)
x(cid:48)
−2εi(a)
βj
−1 = x(cid:48)
(η)x(cid:48)
x(cid:48)
−2εi(a)x(cid:48)
−2εi(a)
−βj
−1 = x(cid:48)
θ(η)x(cid:48)
x(cid:48)
2εi(a)x(cid:48)
θ(η),
2εi(a)
−1 = x(cid:48)
−θ(η)x(cid:48)
−θ(η)x(cid:48)
x(cid:48)
2εi(a)x(cid:48)
2εi(a)
−1 = x(cid:48)
θ(η)x(cid:48)
θ(η)x(cid:48)
x(cid:48)
−2εi(a)x(cid:48)
−2εi(a)
βi
−1 = x(cid:48)
−2εi(a)x(cid:48)
x(cid:48)
−θ(η)x(cid:48)
−2εi(a)
−θ(η),
(+(1 − δi,j)σiη(cid:48)
j) = x(cid:48)
(η(cid:48)
(ηi)x(cid:48)
x(cid:48)
βj
βi
(η(cid:48)
x(cid:48)
(ηi)x(cid:48)
j) = x(cid:48)
−2εk
−βi
−βj
j) = h(cid:48)
(ηi)x(cid:48)
x(cid:48)
2εi(δi,jσiη(cid:48)
(η(cid:48)
−βj
θ(η)x(cid:48)
θ(η) = x(cid:48)
(ηi)x(cid:48)
x(cid:48)
(ηi),
βi
2εi(−σiηηi)x(cid:48)
x(cid:48)
(ηi)x(cid:48)
θ(η) = x(cid:48)
−βi
θ(η−η+)x(cid:48)
−θ(η−) = vh(cid:48)
θ(η+)x(cid:48)
x(cid:48)
x(cid:48)
(η(cid:48))x(cid:48)
(η(cid:48)(cid:48)) = x(cid:48)
±βi
±βi
±βi
(η(cid:48) + η(cid:48)(cid:48)),
2εk
βi
βi
−βi
(aη),
(aη),
with {i, j, k} = {1, 2, 3}.
βj
(ηi),
jηi)x(cid:48)
(−(1 − δi,j)σiη(cid:48)
j)x(cid:48)
(η(cid:48)
(ηi),
βi
(η(cid:48)
j)x(cid:48)
jηi)x(cid:48)
−βj
−βi
θ(−δi,jη(cid:48)
(η(cid:48)
jηi)x(cid:48)
j)x(cid:48)
jηi)h(cid:48)
−βj
−2εi(+σiηηi)x(cid:48)
−θ(η) = x(cid:48)
(ηi)x(cid:48)
x(cid:48)
βi
θ(η)x(cid:48)
x(cid:48)
(ηi)x(cid:48)
−βi
−βi
(ηi),
−θ(η−)x(cid:48)
θ(η+),
x(cid:48)
±θ(η(cid:48))x(cid:48)
±θ(η(cid:48)(cid:48)) = x(cid:48)
(ηi),
βi
±θ(η(cid:48) + η(cid:48)(cid:48))
−θ(η)x(cid:48)
+βi
−θ(η) = x(cid:48)
−θ(η)x(cid:48)
(ηi),
−βi
(ηi),
Singular Degenerations of Lie Supergroups of Type D(2, 1; a)
27
5.2.2. Singular specializations of the supergroup(s) G(cid:48)
groups G(cid:48)
σ we have that
σ is simple (as a Lie supergroup) for all σ = (σ1, σ2, σ3) ∈ V ×.
G(cid:48)
Indeed, this follows from the presentation of G(cid:48)
the relation Lie(cid:0)G(cid:48)
(cid:1) = g(cid:48)(σ)C = gσ along with Proposition 3.1. The situation is different at
σ in Section 5.2.1 above, but also as a fallout of
σ
σ. By construction, for the super-
"singular values" of the parameter σ; the complete result is
Theorem 5.2. Given σ ∈ V , keep notation as above.
(1) If σ ∈ V ×, then the Lie supergroup G(cid:48)
(2) If σ ∈ V \ V ×, with σi = 0 and σj (cid:54)= 0 (cid:54)= σk for {i, j, k} = {1, 2, 3}, then letting B(cid:48)
σ is simple.
i be the
Lie subsupergroup of G(cid:48)
i(A) :=(cid:10){h(cid:48)
B(cid:48)
σ defined on every A ∈ (Wsalg) by
±β(η)}t(cid:54)=i,α∈∆¯0\{2εi},β∈∆¯1
±α(b), x(cid:48)
(cid:11)
2εt(a), x(cid:48)
a∈A¯0,b∈A¯0,η∈A¯1
i ≤ G(cid:48)
∼= SL2 and G(cid:48)
(cid:69) G(cid:48)
we have B(cid:48)
exist isomorphisms B(cid:48)
product of Lie supergroups. In short, there exists a split short exact sequence
σ (a normal Lie subsupergroup), A(cid:48)
∼= PSL(22), A(cid:48)
σ (a Lie subsupergroup), and there
∼= SL2 (cid:110) PSL(22) -- a semidirect
σ
i
i
i
1 −−−−→ PSL(22) ∼= B(cid:48)
(3) If σ = 0(∈ V \ V ×), i.e., σh = 0 for all h ∈ {1, 2, 3}, then letting (cid:0)G(cid:48)
∼= SL2 −−−−→ 1.
(cid:76)(cid:57)(cid:57)−−−
−−−−−→A(cid:48)
σ
i
(cid:1)
¯1 be the Lie
σ
σ defined on every A ∈ (Wsalg) by
i −−−−→ G(cid:48)
(cid:11)
(G(cid:48)
subsupergroup of G(cid:48)
σ)¯1(A) :=(cid:10){x(cid:48)
(cid:0)∼=C8(cid:1) as modules over (G(cid:48)
we have (G(cid:48)
α(η)}α∈∆¯1
η∈A¯1
2 and (G(cid:48)
σ)¯0
∼= SL3
σ)rd
G(cid:48)
σ
∼= (G(cid:48)
σ)rd (cid:110) (G(cid:48)
σ)¯1
∼= SL3
2
2-dimensional module over the i-th copy SL(i)
2 of SL2 (for i = 1, 2, 3). Finally, we have
∼= 2(cid:2)3
∼= 2 := C+(cid:105) ⊕ C−(cid:105) is the tautological
∼= (cid:2)3
σ)¯1
2i
i=1
∼= C8 as Lie (super)groups, (G(cid:48)
σ)¯1
∼= SL3
2 -- where 2i
(cid:110)(cid:0)(cid:2)3
(cid:1) ∼= SL3
(cid:110) 2(cid:2)3
2i
i=1
2
-- a semidirect product of Lie supergroups.
In other words, there is a split short exact sequence
1 −−−−−→ 2(cid:2)3 ∼=(cid:0)G(cid:48)
(cid:1)
σ
−−−−−→(cid:0)G(cid:48)
(cid:76)(cid:57)(cid:57)−−−
(cid:1)
σ
rd
¯1 −−−−−→ G(cid:48)
σ
∼= SL×3
2 −−−−−→ 1.
in Section 5.2.1, or otherwise from the relation Lie(cid:0)G(cid:48)
Proof . Like for Theorem 5.1, the present claim can be obtained from the presentation of G(cid:48)
(cid:4)
(cid:1) = g(cid:48)(σ) along with Theorem 4.2.
σ
σ
σ
5.3 Third family: the Lie supergroups G(cid:48)(cid:48)
Given σ = (σ1, σ2, σ3) ∈ V , let g(cid:48)(cid:48) := g(cid:48)(cid:48)(σ) be the complex Lie superalgebra associated with σ
θ (with α ∈ ∆,
as in Section 4.3.1, and let g(cid:48)(cid:48)
⊕ CX−2εi for each i: the
i ∈ {1, 2, 3}) of g as in Section 3.3.3 and set a(cid:48)(cid:48)
latter are Lie subalgebras of g(cid:48)(cid:48)
¯0 = a(cid:48)(cid:48)
is isomorphic
to sl(2) when σi (cid:54)= 0 -- an explicit isomorphism being given by X2εi
(cid:55)→ h and
X−2εi (cid:55)→ σif , where {e, h, f} is the standard basis sl(2) -- while for σi = 0 it is isomorphic to
the Lie subalgebra of b+ ⊕ b−, with b+ := Ce + Ch and b− := Ch + Cf being the standard Borel
subalgebras inside sl(2), with C-basis {(e, 0), (h, h), (0, f )}.
¯0 be its even part. Fix the elements Xα, H(cid:48)
¯0 such that g(cid:48)(cid:48)
, H(cid:48)
3. Moreover, every a(cid:48)(cid:48)
(cid:55)→ σie, H(cid:48)
:= CX2εi ⊕ CH(cid:48)
1 ⊕ a(cid:48)(cid:48)
2 ⊕ a(cid:48)(cid:48)
2εi
2εi
2εi
i
i
28
K. Iohara and F. Gavarini
Let B± be the Borel subgroup of SL2 of all upper, resp. lower, triangular matrices, and let S
be the subgroup of B+× B− whose elements are all the pairs of matrices (X+, X−) such that the
diagonal parts of X+ and of X− are the same. For each i ∈ {1, 2, 3}, let A(cid:48)(cid:48)
i (depending on σi)
respectively be a copy of SL2 if σi (cid:54)= 0 and a copy of S otherwise; then set G(cid:48)(cid:48) := A(cid:48)(cid:48)
2 × A(cid:48)(cid:48)
3.
The adjoint action of g(cid:48)(cid:48)
¯0
faithful again; then the pair P(cid:48)(cid:48)
the sense of Section 2.4.1. At last, we can define
∼= Lie(cid:0)G(cid:48)(cid:48)(cid:1) on g(cid:48)(cid:48) lifts to a holomorphic G(cid:48)(cid:48)-action on g(cid:48)(cid:48), which is
σ :=(cid:0)G(cid:48)(cid:48), g(cid:48)(cid:48)(cid:1) with this action is a super Harish-Chandra pair, in
1 × A(cid:48)(cid:48)
G(cid:48)(cid:48)
σ := GP(cid:48)(cid:48)
σ
as being the complex Lie supergroup associated with the super Harish-Chandra pair P(cid:48)(cid:48)
the equivalence of categories given in Section 2.4.3.
5.3.1. A presentation of G(cid:48)(cid:48)
σ. In order to describe the supergroups G(cid:48)(cid:48)
plicit presentation by generators and relations of the abstract groups G(cid:48)(cid:48)
1 × A(cid:48)(cid:48)
To start with, let G(cid:48)(cid:48) = A(cid:48)(cid:48)
exp : g(cid:48)(cid:48)
¯0
h(cid:48)
that G(cid:48)(cid:48) is generated by the set
2εi(c), h(cid:48)
∼= Lie(cid:0)G(cid:48)(cid:48)(cid:1) −−−−→ G(cid:48)(cid:48) be the exponential map: then consider x2εi(c) := exp(cid:0)cX2εi
(c) := exp(cid:0)cH(cid:48)
σ, we aim now for an ex-
σ(A), for all A ∈ (Wsalg).
3 be the complex Lie group considered above, and let
θ) for all c ∈ C. It is clear
(cid:1), x−2εi(c) := exp(cid:0)cX−2εi
θ(c), x−2εi(c) c ∈ C}
¯0 := {x2εi(c), h(cid:48)
Γ (cid:48)(cid:48)
(cid:1) and h(cid:48)
θ(c) := exp(cH(cid:48)
2 × A(cid:48)(cid:48)
σ through
(cid:1),
2εi
2εi
θ(c)'s might be discarded, but we choose to keep them); therefore, looking at G(cid:48)(cid:48)
(actually the h(cid:48)
as a supergroup, thought of as a group-valued functor G(cid:48)(cid:48) : (Wsalg) −→ (grps), every abstract
group G(cid:48)(cid:48)(A) of its A-points -- for A ∈ (Wsalg) -- is generated by the set
2εi(a), h(cid:48)
¯0 (A) := {x2εi(a), h(cid:48)
Γ (cid:48)(cid:48)
According to Section 2.4.3, the group G(cid:48)(cid:48)
elements of the form x±θ(η) := (cid:0)1 + ηX±θ
θ(a), x−2εi(a) a ∈ A¯0}.
(cid:1) or x±βi(η) := (cid:0)1 + ηX±βi
(cid:9)
¯1 is(cid:8)Y (cid:48)
i∈I =(cid:8)X±θ, X±βi
(cid:9)
σ(A) := GP(cid:48)(cid:48)
(cid:9) -- coinciding with Γ (cid:48)
¯1 (A) :=(cid:8)x±θ(η), x±βi(η) η ∈ A¯1
i ∈ {1, 2, 3} -- as now the chosen C-basis of g(cid:48)(cid:48)
latter is denoted Γ (cid:48)(cid:48)
From the recipe in Section 2.4.3, and the fact that G(cid:48)(cid:48)(A) is generated by Γ (cid:48)(cid:48)
i=1,2,3; the set of all the
¯1(A) in Section 5.2.1.
¯0 (A), with a slight
modification of the relations in Section 2.4.3 we can find the following full set of relations among
generators of G(cid:48)(cid:48)
σ(A) (for all {i, j, k} = {1, 2, 3}):
(5.3)
(A) is generated by G(cid:48)(cid:48)(A) and all
(cid:1) with η ∈ A¯1 and
σ
i
∀ g(cid:48), g(cid:48)(cid:48) ∈ G(cid:48)(cid:48)(A),
(cid:0)e±(−1)δi,j aη(cid:1),
(cid:0)e±aη(cid:1),
(cid:0)e∓σiaη(cid:1),
g(cid:48) · g(cid:48)(cid:48) = g(cid:48) ·G(cid:48)(cid:48) g(cid:48)(cid:48),
−1 = x±βj
−1 = x±θ
−1 = x±βi
−1 = xβj (η)xθ(δi,jσiaη),
−1 = x−βj (η)xβk ((1 − δi,j)σiaη),
−1 = xβj (η)x−βk ((1 − δi,j)σiaη),
−1 = x−βj (η)x−θ(δi,jσiaη),
2εi(a)
2εi(a)
1G(cid:48)(cid:48) = 1,
h(cid:48)
2εi(a)x±βj (η)h(cid:48)
h(cid:48)
2εi(a)x±θ(η)h(cid:48)
hθ(a)x±βi(η)hθ(a)
x2εi(a)xβj (η)x2εi(a)
x2εi(a)x−βj (η)x2εi(a)
x−2εi(a)xβj (η)x−2εi(a)
x−2εi(a)x−βj (η)x−2εi(a)
x2εi(a)xθ(η)x2εi(a)
x2εi(a)x−θ(η)x2εi(a)
x−2εi(a)xθ(η)x−2εi(a)
x−2εi(a)x−θ(η)x−2εi(a)
hθ(a)x±θ(η)hθ(a)
−1 = x±θ(η),
−1 = xθ(η),
−1 = x−θ(η)x−βi(σiaη),
−1 = xθ(η)xβi(σiaη),
−1 = x−θ(η),
Singular Degenerations of Lie Supergroups of Type D(2, 1; a)
29
jηi)xβj (η(cid:48)
j) = x2εk ((1 − δi,j)η(cid:48)
j) = x−2εk (−(1 − δi,j)η(cid:48)
j) = h(cid:48)
xβi(ηi)xβj (η(cid:48)
x−βi(ηi)x−βj (η(cid:48)
2εi(δi,jσiη(cid:48)
xβi(ηi)x−βj (η(cid:48)
xβi(ηi)xθ(η) = xθ(η)xβi(ηi),
x−βi(ηi)xθ(η) = x2εi(−ηηi)xθ(η)x−βi(ηi),
xθ(η+)x−θ(η−) = h(cid:48)
θ(η−η+)x−θ(η−)xθ(η+),
x±βi(η(cid:48))x±βi(η(cid:48)(cid:48)) = x±βi(η(cid:48) + η(cid:48)(cid:48)),
jηi)h(cid:48)
j)xβi(ηi),
jηi)x−βj (η(cid:48)
j)x−βi(ηi),
jηi)x−βj (η(cid:48)
θ(−δi,jη(cid:48)
xβi(ηi)x−θ(η) = x−2εi(ηηi)x−θ(η)x+βi(ηi),
j)xβi(ηi),
x−βi(ηi)x−θ(η) = x−θ(η)x−βi(ηi),
x±θ(η(cid:48))x±θ(η(cid:48)(cid:48)) = x±θ(η(cid:48) + η(cid:48)(cid:48)).
5.3.2. Singular specializations of the supergroup(s) G(cid:48)(cid:48)
supergroups G(cid:48)(cid:48)
σ we have
σ is simple (as a Lie supergroup) for all σ = (σ1, σ2, σ3) ∈ V ×.
G(cid:48)(cid:48)
of the relation Lie(cid:0)G(cid:48)(cid:48)
(cid:1) = g(cid:48)(cid:48)(σ) = gσ along with Proposition 3.1.
This follows from the presentation of G(cid:48)(cid:48)
σ
σ in Section 5.3.1 above, and also as a consequence
σ. One sees easily that for the
Things change, instead, for "singular values" of σ: hereafter is the general result.
Theorem 5.3. Given σ ∈ V , keep notation as above.
(1) If σ ∈ V ×, then the Lie supergroup G(cid:48)(cid:48)
σ is simple.
(2) If σ ∈ V \ V ×, with σi = 0 and σj (cid:54)= 0 (cid:54)= σk for {i, j, k} = {1, 2, 3}, consider the
i of G(cid:48)(cid:48)
σ which is given by
subsupergroup K(cid:48)(cid:48)
i (A) :=(cid:10){x2εi(a+), x−2εi(a−)}a+,a−∈A¯0
K(cid:48)(cid:48)
Then K(cid:48)(cid:48)
the quotient supergroup -- there exists a short exact sequence
i is an Abelian normal Lie subsupergroup of G(cid:48)(cid:48)
(cid:11)
∀ A ∈ (Wsalg).
σ, hence -- letting B(cid:48)(cid:48)
i := G(cid:48)(cid:48)
σ/K(cid:48)(cid:48)
i be
1 −−−−→ K(cid:48)(cid:48)
i −−−−→ 1.
i (A) := (cid:10){h(cid:48)
σ −−−−→ B(cid:48)(cid:48)
i (A) the two subgroups H(cid:48)(cid:48)
i −−−−→ G(cid:48)(cid:48)
Furthermore, defining inside B(cid:48)(cid:48)
D(cid:48)(cid:48)
we overall find two Lie subsupergroups D(cid:48)(cid:48)
subsupergroup with D(cid:48)(cid:48)
of Lie supergroups. In short, there exists a split short exact sequence
(aj), x+2εj (c+), x−2εj (c−), xβ(η)}aj ,c±∈A¯0,η∈A¯1
j(cid:54)=i,β∈∆¯1
i of B(cid:48)(cid:48)
i such that D(cid:48)(cid:48)
∼= H(cid:48)(cid:48)
i and H(cid:48)(cid:48)
∼= C×, and B(cid:48)(cid:48)
∼= PSL(22), H(cid:48)(cid:48)
(cid:110) D(cid:48)(cid:48)
2εj
i
i
i
i
i (A) := (cid:10){h(cid:48)
(cid:11) and
(cid:11) -- for all A ∈ (Wsalg) --
(ai)}ai∈A¯0
2εi
is a normal Lie
i -- a semidirect product
i
1 −−−−−→ PSL(22) ∼= D(cid:48)(cid:48)
i −−−−−→ B(cid:48)(cid:48)
i
(cid:76)(cid:57)(cid:57)−−−
−−−−−→H(cid:48)(cid:48)
i
∼= C× −−−−−→ 1.
(3) If σ = 0 (∈ V \ V ×), i.e., σh = 0 for all h ∈ {1, 2, 3}, then K(cid:48)(cid:48) :=
3×
K(cid:48)(cid:48)
i is an Abelian
σ/K(cid:48)(cid:48) is a quotient Lie supergroup (of G(cid:48)(cid:48)
σ);
i=1
normal Lie subsupergroup, hence B(cid:48)(cid:48) := G(cid:48)(cid:48)
therefore, there exists a short exact sequence
i −−−−−→ G(cid:48)(cid:48)
K(cid:48)(cid:48)
Moreover, setting O(cid:48)(cid:48)(A) :=(cid:10){xβ(η)}η∈A¯1
1 −−−−−→ K(cid:48)(cid:48) :=
3×
i=1
β∈∆¯1
(cid:11) and T(cid:48)(cid:48)(A) :=(cid:10){h(cid:48)
σ −−−−−→ B(cid:48)(cid:48) −−−−−→ 1.
(a)}a∈A¯0
2εi
i∈{1,2,3}
(cid:11) inside B(cid:48)(cid:48)(A)
for all A ∈ (Wsalg) we overall find two subsupergroups O(cid:48)(cid:48) and T(cid:48)(cid:48) of B(cid:48)(cid:48) such that O(cid:48)(cid:48) is
normal Abelian, isomorphic to A08
C -- the (totally odd) complex affine Abelian supergroup
of superdimension (08) -- and T(cid:48)(cid:48) is Abelian, isomorphic to T3
C -- the (totally even) 3-
dimensional complex torus -- with B(cid:48)(cid:48) ∼= T(cid:48)(cid:48) (cid:110)O(cid:48)(cid:48) -- a semidirect product of Lie supergroups.
In other words, there exists a second, split short exact sequence
1 −−−−−→ O(cid:48)(cid:48) ∼= A08
C −−−−−→ B(cid:48)(cid:48) (cid:76)(cid:57)(cid:57)−−−
−−−−−→T3
C ∼= T(cid:48)(cid:48) −−−−−→ 1.
(cid:9).
:= Ai
σ
c∈C
algebra -- if σi = 0 (see also Section 5.1).
∼= sli(2) when σi (cid:54)= 0 and ai
K. Iohara and F. Gavarini
σ in Sec-
(cid:4)
Given σ = (σ1, σ2, σ3) ∈ V , following Section 4.4 we fix the element τ := x1x2x3 ∈ C[x] and the
30
Proof . Like for Theorem 5.1, one can deduce the claim from the presentation of G(cid:48)(cid:48)
tion 5.3.1, or also from the relation Lie(cid:0)G(cid:48)(cid:48)
(cid:1) = g(cid:48)(cid:48)(σ) along with Theorem 4.3.
5.4 Lie supergroups from contractions: the family of the (cid:98)Gσ's
ideal I = Iσ := ({xi − σi}i=1,2,3), and we consider the corresponding complex Lie algebra(cid:98)g(σ),
with(cid:98)g(σ)¯0 and(cid:98)g(σ)¯1 as its even and odd part, respectively. With a slight abuse of notation, for
any element Z ∈(cid:98)g(x) we denote again by Z its corresponding coset in(cid:98)g(σ) = C[σ] ⊗C[x](cid:98)g(x) ∼=
(cid:98)g(x)/Iσ(cid:98)g(x) (see Section 4.4 for notation). By construction,(cid:98)g(σ) admits as C-basis the set
(cid:98)B := {Xα, H2εi α ∈ ∆, i ∈ {1, 2, 3}} ∪(cid:8)(cid:98)Xβ := τ Xβ β ∈ ∆¯1
We consider also(cid:98)ai := ai (:= CX2εi ⊕ CH2εi ⊕ CX−2εi) for all 1 = 1, 2, 3, that all are Lie
subalgebras of(cid:98)g¯0, with(cid:98)ai
∼= C⊕3 -- the 3-dimensional Abelian Lie
Recalling the construction of Gσ in Section 5.1, for each i ∈ {1, 2, 3} we set (cid:98)Ai
i=1(cid:98)Ai = G,
(isomorphic to either SL2 or C × C× × C depending on σi (cid:54)= 0 or σi = 0) and (cid:98)G := ×3
a complex Lie group such that Lie(cid:0)(cid:98)G(cid:1) =(cid:98)g(σ)¯0. Like in Section 5.1, the adjoint action of(cid:98)g(σ)¯0
onto(cid:98)g(σ) integrates to a Lie group action of (cid:98)G onto(cid:98)g(σ): endowed with this action, the pair
(cid:98)Pσ :=(cid:0)(cid:98)G,(cid:98)g(σ)(cid:1) is a super Harish-Chandra pair (cf. Section 2.4.1). Eventually, we can define
to be the complex Lie supergroup associated with (cid:98)Pσ following Section 2.4.3.
5.4.1. A presentation of (cid:98)Gσ. To describe the supergroups (cid:98)Gσ, we provide hereafter an
explicit presentation by generators and relations of all the abstract groups (cid:98)Gσ(A), with A ∈
∼= Lie((cid:98)G) −−−−→ (cid:98)G be the exponential map. Like we did
(Wsalg). To begin with, let exp :(cid:98)g¯0
for every c ∈ C; then (cid:98)Γi := {x2εi(c), h2εi(c), x−2εi(c)}c∈C is a generating set for (cid:98)Ai; also, we
(cid:98)G = (cid:98)A1 × (cid:98)A2 × (cid:98)A3 is generated by
(cid:98)Γ¯0 := {x2εi(c), h2εi(c), hθ(c), x−2εi(c)}i∈{1,2,3}
When we think of (cid:98)G as a (totally even) supergroup, looking at it as a group-valued functor
(cid:98)G : (Wsalg) −→ (grps), the abstract group (cid:98)G(A) of its A-points -- for A ∈ (Wsalg) -- is generated
by the set(cid:98)Γ¯0(A) := {x2εi(a), h2εi(a), hθ(a), x−2εi(a)}i∈{1,2,3}
isomorphism (cid:98)G ∼= G (see Section 5.1.1 for the definition of G) as complex Lie groups.
To generate the group (cid:98)Gσ(A) := G(cid:98)Pσ
(cid:98)g(σ)¯1 the C-basis {Yi}i∈I =(cid:8)(cid:98)Xβ := τ Xβ β ∈ ∆¯1 = {±θ,±β1,±β2,±β3}(cid:9). Thus, besides the
generating elements from (cid:98)G(A), we take as generators also those of the set
(cid:98)Γ¯1(A) :=(cid:8)(cid:98)x±θ(η) :=(cid:0)1 + η(cid:98)X±θ
(cid:1)(cid:9)i∈{1,2,3}
(cid:1),(cid:98)x±βi(η) :=(cid:0)1 + η(cid:98)X±βi
consider elements hθ(c) := exp(cHθ) for all c ∈ C.
In fact, we would better stress that, by construction (cf. Section 5.1), we have an obvious
x2εi(c) := exp(cX2εi),
h2εi(c) := exp(cH2εi),
x−2εi(c) := exp(cX−2εi)
(we could drop the hθ(c)'s, but we prefer to keep them among the generators).
a∈A¯0
.
(5.4)
(A) applying the recipe in Section 2.4.3, we fix in
η∈A¯1
.
(cid:98)Gσ := G(cid:98)Pσ
in Section 5.1.1 for the supergroup Gσ, inside each subgroup Ai we consider
It follows that the complex Lie group
= 1,
g(cid:48)(cid:48),
−1 =(cid:98)x±θ(η),
∀ g(cid:48), g(cid:48)(cid:48) ∈ (cid:98)G(A),
−δi,j σiaη(cid:1),
(cid:0)e±(−1)
g(cid:48) · g(cid:48)(cid:48) = g(cid:48) ·(cid:98)G
1(cid:98)G
h2εi(a)(cid:98)x±βj (η)h2εi(a)
−1 =(cid:98)x±βj
(cid:0)e±σiaη(cid:1),
−1 =(cid:98)x±θ
h2εi(a)(cid:98)x±θ(η)h2εi(a)
(cid:0)e∓σiaη(cid:1),
hθ(a)(cid:98)x±βi(η)hθ(a)
−1 =(cid:98)x±βi
hθ(a)(cid:98)x±θ(η)hθ(a)
x2εi(a)(cid:98)xβj (η)x2εi(a)
−1 =(cid:98)xβj (η)(cid:98)xθ(δi,jσiaη),
x2εi(a)(cid:98)x−βj (η)x2εi(a)
−1 =(cid:98)x−βj (η)(cid:98)xβk ((1 − δi,j)σiaη),
x−2εi(a)(cid:98)xβj (η)x−2εi(a)
−1 =(cid:98)xβj (η)(cid:98)x−βk ((1 − δi,j)σiaη),
x−2εi(a)(cid:98)x−βj (η)x−2εi(a)
−1 =(cid:98)x−βj (η)(cid:98)x−θ(δi,jσiaη),
x2εi(a)(cid:98)xθ(η)x2εi(a)
−1 =(cid:98)xθ(η),
x2εi(a)(cid:98)x−θ(η)x2εi(a)
−1 =(cid:98)x−θ(η)(cid:98)x−βi(σiaη),
−1 =(cid:98)xθ(η)(cid:98)xβi(σiaη),
x−2εi(a)(cid:98)xθ(η)x−2εi(a)
x−2εi(a)(cid:98)x−θ(η)x−2εi(a)
−1 =(cid:98)x−θ(η),
(cid:0)(1 − δi,j)τ 2
(cid:1)(cid:98)xβj (η(cid:48)
(cid:98)xβi(ηi)(cid:98)xβj (η(cid:48)
j)(cid:98)xβi(ηi),
(cid:0)−(1 − δi,j)τ 2
(cid:1)(cid:98)x−βj (η(cid:48)
σ η(cid:48)
j)(cid:98)x−βi(ηi),
(cid:98)x−βi(ηi)(cid:98)x−βj (η(cid:48)
jηi
(cid:0)δi,jτ 2
(cid:1)hθ
(cid:1)(cid:98)x−βj (η(cid:48)
(cid:0)−δi,jτ 2
(cid:98)xβi(ηi)(cid:98)x−βj (η(cid:48)
j)(cid:98)xβi(ηi),
(cid:0)τ 2
σ η(cid:48)
σ η(cid:48)
(cid:98)xβi(ηi)(cid:98)x−θ(η) = x−2εi
(cid:98)xβi(ηi)(cid:98)xθ(η) =(cid:98)xθ(η)(cid:98)xβi(ηi),
jηi
jηi
(cid:1)(cid:98)xθ(η)(cid:98)x−βi(ηi),
(cid:0)−τ 2
(cid:98)x−βi(ηi)(cid:98)x−θ(η) =(cid:98)x−θ(η)(cid:98)x−βi(ηi),
(cid:98)x−βi(ηi)(cid:98)xθ(η) = x2εi
(cid:0)τ 2
(cid:1)(cid:98)x−θ(η−)(cid:98)xθ(η+),
(cid:98)xθ(η+)(cid:98)x−θ(η−) = hθ
(cid:98)x±βi(η(cid:48))(cid:98)x±βi(η(cid:48)(cid:48)) =(cid:98)x±βi(η(cid:48) + η(cid:48)(cid:48)),
(cid:98)x±θ(η(cid:48))(cid:98)x±θ(η(cid:48)(cid:48)) =(cid:98)x±θ(η(cid:48) + η(cid:48)(cid:48))
(cid:1)(cid:98)x−θ(η)(cid:98)xβi(ηi),
j) = x−2εk
j) = h2εi
σ ηηi
σ η−η+
j) = x2εk
σ η(cid:48)
jηi
σ ηηi
Singular Degenerations of Lie Supergroups of Type D(2, 1; a)
Taking into account that (cid:98)G(A) is generated by (cid:98)Γ¯0(A), we can modify the set of relations in
Section 2.4.3 by letting g+ ∈ (cid:98)G(A) range inside the set (cid:98)Γ¯0(A): then we can find the following
31
full set of relations (freely using notation eZ := exp(Z)):
with {i, j, k} ∈ {1, 2, 3}.
(cid:98)Gσ is simple (as a Lie supergroup) for all σ = (σ1, σ2, σ3) ∈ V ×.
5.4.2. Singular specializations of the supergroup(s) (cid:98)Gσ. The very construction of the
supergroups (cid:98)Gσ implies that
This also follows from the presentation of (cid:98)Gσ in Section 5.4.1 above, or as a direct consequence
(cid:1) =(cid:98)g(σ) =(cid:98)gσ and of the fact that(cid:98)gσ
of the relation Lie(cid:0)(cid:98)Gσ
(1) If σ ∈ V ×, then the Lie supergroup (cid:98)Gσ is simple.
(2) If σ ∈ V \ V ×, with σi = 0 and σj (cid:54)= 0 (cid:54)= σk for {i, j, k} = {1, 2, 3}, then (cid:98)Ai (cid:69) (cid:98)Gσ is
a central Lie subsupergroup of (cid:98)Gσ, with (cid:98)Ai
∼= SL(2) for
∼= C × C× × C, while (cid:98)Aj
Theorem 5.4. Given σ ∈ V , keep notation as above.
Things change instead at "singular values" of σ. The complete result is the following:
∼= gσ when σi (cid:54)= 0 for all i.
∼= (cid:98)Ak
(cid:98)G(σ) ∼= (cid:98)G(σ)rd
(cid:98)A(cid:96)
{j, k} = {1, 2, 3} \ {i}. Also, we have a semidirect product splitting
(cid:110) (cid:98)G(σ)¯1
with (cid:98)G(σ)rd =
∼= (C × C× × C) × SL(2) × SL(2) while (cid:98)G(σ)¯1 is the subsupergroup
of (cid:98)Gσ generated by the(cid:98)x±θ's and the(cid:98)x±βi's (for all i), which is normal Abelian, isomorphic
3×
(cid:96)=1
32
to A08
such that (cid:98)G(σ)¯1
over (cid:98)G(σ)rd
K. Iohara and F. Gavarini
C -- the (totally odd) complex affine Abelian supergroup of superdimension (08) -- and
∼= ((cid:4) ⊕(cid:4)) (cid:2) 2 (cid:2) 2 -- where (cid:4) is the trivial representation -- as a module
∼=(cid:0)C × C× × C(cid:1) × SL(2) × SL(2). In other words, there exists a split short
1 −−→ ((cid:4) ⊕(cid:4)) (cid:2) 2 (cid:2) 2 ∼= (cid:98)G(σ)¯1 −−→ (cid:98)G(σ)
∼=(cid:0)C × C× × C(cid:1) × SL(2)2 −−→ 1.
(cid:76)(cid:57)(cid:57)−−−→(cid:98)G(σ)rd
exact sequence of Lie supergroups
β := τ X(cid:48)
β β ∈ ∆¯1
i (:= CX(cid:48)
2εi
⊕ CH(cid:48)
2εi
⊕ CX(cid:48)−2εi
σ, we can deduce the claim from the
∼= (C × C× × C)3 × ((cid:4) ⊕(cid:4))
(cid:2)3.
Proof . As for the parallel results for Gσ, G(cid:48)
, cf. Section 4.2.1) for all 1 = 1, 2, 3: all
i. The faithful
2εi α ∈ ∆, i ∈ {1, 2, 3}} ∪(cid:8)(cid:98)X(cid:48)
Given σ = (σ1, σ2, σ3) ∈ V , we follow again Section 4.4 and set τ := x1x2x3 ∈ C[x] and
(3) If σ = 0 (∈ V \V ×), i.e., σh = 0 for all h ∈ {1, 2, 3}, then (cid:98)Gσ is the Abelian Lie supergroup
(cid:98)Gσ
presentation of (cid:98)Gσ in Section 5.4.1, or from the link Lie((cid:98)Gσ) =(cid:98)g(σ) along with Theorem 4.4. (cid:4)
5.5 Lie supergroups from contractions: the family of the (cid:98)G(cid:48)
I = Iσ := ({xi − σi}i=1,2,3); but now we consider the corresponding complex Lie algebra(cid:98)g(cid:48)(σ),
with (cid:98)g(cid:48)(σ)¯0 and (cid:98)g(cid:48)(σ)¯1 as its even and odd part, respectively (and we still make use of some
abuse of notation as in Section 5.4). By construction, a C-basis of(cid:98)g(cid:48)(σ) is
(cid:98)B(cid:48) := {X(cid:48)
α, H(cid:48)
Consider also (cid:98)a(cid:48)
i=1(cid:98)a(cid:48)
these are Lie subalgebras in (cid:98)g(cid:48)(σ), isomorphic to sl(2), and (cid:98)g(cid:48)(σ)¯0 = ⊕3
i := a(cid:48)
adjoint action of(cid:98)g(cid:48)(σ)¯0 onto(cid:98)g(cid:48)(σ) gives a Lie algebra monomorphism(cid:98)g(cid:48)(σ)¯0 (cid:44)−→ gl((cid:98)g(cid:48)(σ)), by
which we identify(cid:98)g(cid:48)(σ)¯0 with its image in gl((cid:98)g(cid:48)(σ)). Then exp : gl(cid:0)(cid:98)g(cid:48)(σ)(cid:1) −→ GL((cid:98)g(cid:48)(σ)) yields
a Lie subgroup (cid:98)G(cid:48) := exp((cid:98)g(cid:48)(σ)¯0) in GL((cid:98)g(cid:48)(σ)) which faithfully acts onto(cid:98)g(cid:48)(σ) and is such that
Lie((cid:98)G(cid:48)) = ((cid:98)g(cid:48)(σ))¯0; finally, the pair (cid:98)P(cid:48)
σ := ((cid:98)G(cid:48),(cid:98)g(cid:48)(σ)) with this action is then a super Harish-
As alternative method, we might also construct the super Harish-Chandra pair (cid:98)P(cid:48)
Once we have the super Harish-Chandra pair (cid:98)P(cid:48)
(cid:98)G(cid:48)
σ := G(cid:98)P(cid:48)
that is the complex Lie supergroup associated with (cid:98)P(cid:48)
σ. We shall presently describe the supergroups (cid:98)G(cid:48)
5.5.1. A presentation of (cid:98)G(cid:48)
an explicit presentation by generators and relations of all the abstract groups (cid:98)G(cid:48)
A ∈ (Wsalg). To begin with, let exp :(cid:98)g(cid:48)
2εi(c) := exp(cX(cid:48)
x(cid:48)
for every c ∈ C; then (cid:98)Γ (cid:48)
(cid:98)G(cid:48) = (cid:98)A(cid:48)
2 × (cid:98)A(cid:48)
1 × (cid:98)A(cid:48)
(cid:98)Γ (cid:48)
¯0 := {x(cid:48)
2εi(c), h(cid:48)
(cid:1) for all c ∈ C. It follows that the complex Lie group
−2εi(c)}i∈{1,2,3}
Chandra pair (cf. Section 2.4.1).
same procedure, up to the obvious, minimal changes, adopted for P(cid:48)
can also do the converse, namely use the present method to construct P(cid:48)
∼= Lie((cid:98)G(cid:48)) −−−→ (cid:98)G(cid:48) be the exponential map. Just like
(c)}c∈C is a generating set for (cid:98)A(cid:48)
for the supergroup Gσ in Section 5.1.1, inside each subgroup A(cid:48)
σ via the
σ in Section 5.2; indeed, one
i we consider
x(cid:48)
−2εi(c) := exp(cX(cid:48)
2εi(c) := exp(cH(cid:48)
h(cid:48)
(c), h(cid:48)
(c), x(cid:48)−2εi
2εi),
i := {x(cid:48)
θ(c) := exp(cid:0)cH(cid:48)
2εi
2εi
3 is generated by
θ(c), x(cid:48)
2εi(c), h(cid:48)
σ after the recipe in Section 2.4.3.
we consider elements h(cid:48)
σ by means of
σ(A), for all
2εi),
−2εi)
σ, it makes sense to define
σ as well.
¯0
θ
c∈C
i = A(cid:48)
i; also,
σ and G(cid:48)(cid:48)
σ's
(cid:9).
σ
Singular Degenerations of Lie Supergroups of Type D(2, 1; a)
33
σ
±βi
±θ(η),
a∈A¯0
.
±θ
±βi
(cid:1)(cid:9)i∈{1,2,3}
η∈A¯1
.
generators also those of the set
(as before, we could drop the h(cid:48)
θ(c)'s, but we prefer to keep them among the generators).
When thinking of (cid:98)G(cid:48) as a (totally even) supergroup, considered as a group-valued functor
(cid:98)G(cid:48) : (Wsalg) −→ (grps), the abstract group (cid:98)G(cid:48)(A) of its A-points -- for A ∈ (Wsalg) -- is gene-
(cid:98)Γ (cid:48)
¯0(A) := {x(cid:48)
−2εi(a)}i∈{1,2,3}
2εi(a), hθ(a), x(cid:48)
2εi(a), h(cid:48)
rated by the set
(5.5)
σ(A) := G(cid:98)P(cid:48)
β := τ X(cid:48)
¯1(A) :=(cid:8)(cid:98)x(cid:48)
(cid:98)Γ (cid:48)
(η) :=(cid:0)1 + η(cid:98)X(cid:48)
2εi(a)
2εi(a)
θ(a)
2εi(a)
2εi(a)
−2εi(a)
−2εi(a)
¯0(A), we can modify the set of relations in Section 2.4.3
¯0(A); eventually, we can find the following full set of relations
Indeed, we can also stress that, by construction (cf. Section 5.1), there exists an obvious
(A)
β β ∈
isomorphism (cid:98)G(cid:48) ∼= G(cid:48) as complex Lie groups. Now, to generate the group (cid:98)G(cid:48)
following the recipe in Section 2.4.3, we fix in ((cid:98)g(cid:48)(σ)C)¯1 the C-basis {Yi}i∈I =(cid:8)(cid:98)X(cid:48)
∆¯1 = {±θ,±β1,±β2,±β3}(cid:9). Then, beside the generating elements from (cid:98)G(cid:48)(A) we take as
(cid:1),(cid:98)x(cid:48)
±θ(η) :=(cid:0)1 + η(cid:98)X(cid:48)
Knowing that (cid:98)G(cid:48)(A) is generated by (cid:98)Γ (cid:48)
by letting g ∈ (cid:98)G(cid:48)(A) range inside (cid:98)Γ (cid:48)
∀ g(cid:48), g(cid:48)(cid:48) ∈ (cid:98)G(cid:48)(A),
(freely using notation eZ := exp(Z)):
−δi,j aη(cid:1),
(cid:0)e±(−1)
g(cid:48) · g(cid:48)(cid:48) = g(cid:48) ·(cid:98)G(cid:48) g(cid:48)(cid:48),
1(cid:98)G(cid:48) = 1,
−1 =(cid:98)x(cid:48)
2εi(a)(cid:98)x(cid:48)
(cid:0)e±aη(cid:1),
(η)h(cid:48)
h(cid:48)
−1 =(cid:98)x(cid:48)
2εi(a)(cid:98)x(cid:48)
±βj
±βj
(cid:0)e∓σiaη(cid:1),
±θ(η)h(cid:48)
h(cid:48)
θ(a)(cid:98)x(cid:48)
−1 =(cid:98)x(cid:48)
θ(a)(cid:98)x(cid:48)
±θ
(η)h(cid:48)
h(cid:48)
(η)(cid:98)x(cid:48)
−1 =(cid:98)x(cid:48)
2εi(a)(cid:98)x(cid:48)
±βi
±βi
(η)x(cid:48)
x(cid:48)
(η)(cid:98)x(cid:48)
−1 =(cid:98)x(cid:48)
2εi(a)(cid:98)x(cid:48)
βj
(η)x(cid:48)
x(cid:48)
(η)(cid:98)x(cid:48)
−1 =(cid:98)x(cid:48)
−2εi(a)(cid:98)x(cid:48)
−βj
(η)x(cid:48)
x(cid:48)
(η)(cid:98)x(cid:48)
−1 =(cid:98)x(cid:48)
−2εi(a)(cid:98)x(cid:48)
βj
(η)x(cid:48)
x(cid:48)
−1 =(cid:98)x(cid:48)
2εi(a)(cid:98)x(cid:48)
−βj
θ(η)x(cid:48)
x(cid:48)
−θ(η)(cid:98)x(cid:48)
−1 =(cid:98)x(cid:48)
2εi(a)(cid:98)x(cid:48)
2εi(a)
−θ(η)x(cid:48)
x(cid:48)
−2εi(a)(cid:98)x(cid:48)
−1 =(cid:98)x(cid:48)
θ(η)(cid:98)x(cid:48)
2εi(a)
θ(η)x(cid:48)
x(cid:48)
−2εi(a)(cid:98)x(cid:48)
−1 =(cid:98)x(cid:48)
−2εi(a)
(cid:0)(1 − δi,j)τ 2
(cid:1)(cid:98)x(cid:48)
βi
x(cid:48)
−θ(η)x(cid:48)
(ηi)(cid:98)x(cid:48)
(cid:98)x(cid:48)
j)(cid:98)x(cid:48)
−2εi(a)
−θ(η),
(cid:1)(cid:98)x(cid:48)
(cid:0)−(1 − δi,j)τ 2
(η(cid:48)
j) = x(cid:48)
(η(cid:48)
σ σiη(cid:48)
j)(cid:98)x(cid:48)
(cid:98)x(cid:48)
(ηi)(cid:98)x(cid:48)
(ηi),
jηi
(cid:0)δi,jτ 2
(cid:1)h(cid:48)
(cid:1)(cid:98)x(cid:48)
(cid:0)−δi,jτ 2
βj
βi
βj
(η(cid:48)
(η(cid:48)
j) = x(cid:48)
σ σiη(cid:48)
(ηi)(cid:98)x(cid:48)
(cid:98)x(cid:48)
j)(cid:98)x(cid:48)
−βj
−βj
jηi
(cid:1)(cid:98)x(cid:48)
(cid:0)τ 2
j) = h(cid:48)
(η(cid:48)
σ σiη(cid:48)
σ η(cid:48)
(cid:98)x(cid:48)
(cid:98)x(cid:48)
(ηi)(cid:98)x(cid:48)
θ(η)(cid:98)x(cid:48)
θ(η) =(cid:98)x(cid:48)
(ηi)(cid:98)x(cid:48)
−θ(η)(cid:98)x(cid:48)
−βj
−βj
jηi
jηi
(cid:0)−τ 2
(cid:1)(cid:98)x(cid:48)
−θ(η) = x(cid:48)
(cid:98)x(cid:48)
(ηi)(cid:98)x(cid:48)
−θ(η) =(cid:98)x(cid:48)
(ηi)(cid:98)x(cid:48)
−θ(η)(cid:98)x(cid:48)
(cid:98)x(cid:48)
θ(η)(cid:98)x(cid:48)
−2εi
(cid:1)(cid:98)x(cid:48)
(cid:0)τ 2
θ(η) = x(cid:48)
−θ(η−)(cid:98)x(cid:48)
θ(η+)(cid:98)x(cid:48)
(cid:98)x(cid:48)
−βi
−βi
(ηi),
±θ(η(cid:48)(cid:48)) =(cid:98)x(cid:48)
(cid:98)x(cid:48)
(cid:98)x(cid:48)
(η(cid:48))(cid:98)x(cid:48)
(η(cid:48)(cid:48)) =(cid:98)x(cid:48)
±θ(η(cid:48))(cid:98)x(cid:48)
−θ(η−) = h(cid:48)
θ(η+),
±θ(η(cid:48) + η(cid:48)(cid:48))
±βi
5.5.2. Singular specializations of the supergroup(s) (cid:98)G(cid:48)
the supergroups (cid:98)G(cid:48)
((1 − δi,j)aη),
βk
((1 − δi,j)aη),
−βk
−θ(δi,jaη),
(cid:98)G(cid:48)
σ is simple (as a Lie supergroup) for all σ = (σ1, σ2, σ3) ∈ V ×.
σ. From the very construction of
with {i, j, k} ∈ {1, 2, 3}.
σ η−η+
(η(cid:48) + η(cid:48)(cid:48)),
±βi
(ηi),
−βi
(η(cid:48)
βi
σ σiηηi
−1 =(cid:98)x(cid:48)
βi
(ηi),
σ σiηηi
(ηi),
−βi
(ηi),
βi
−βi
βi
βi
−βi
±θ(η)h(cid:48)
θ(a)
−βi
(aη),
(aη),
βj
−βj
βj
−βj
2εk
−2εk
2εi
h(cid:48)
θ(δi,jaη),
(ηi),
+βi
±βi
2εi
θ
σ we get
θ(η),
θ
βi
K. Iohara and F. Gavarini
σ in Section 5.5.1 above, but also as direct consequence
∼= gσ when σi (cid:54)= 0 for all i.
±θ's and the(cid:98)x(cid:48)
σ generated by the(cid:98)x(cid:48)
σ)rd-module; in particular, ((cid:98)G(cid:48)
's
σ, and there exist isomorphisms
σ)¯1 is
±βi
σ
34
σ) =(cid:98)g(cid:48)(σ) =(cid:98)g(cid:48)
Theorem 5.5. Given σ ∈ V , keep notation as above.
σ is simple.
This follows from the presentation of (cid:98)G(cid:48)
of the relation Lie((cid:98)G(cid:48)
σ and of(cid:98)g(cid:48)
(1) If σ ∈ V ×, then the Lie supergroup (cid:98)G(cid:48)
σ)¯1 be the subsupergroup of (cid:98)G(cid:48)
(2) If σ ∈ V \ V ×, let ((cid:98)G(cid:48)
σ)¯1 is Abelian and normal in (cid:98)G(cid:48)
(for all i). Then ((cid:98)G(cid:48)
×3 and ((cid:98)G(cid:48)
2i as a ((cid:98)G(cid:48)
((cid:98)G(cid:48)
∼= SL(2)
(cid:1)
(cid:110)(cid:0)(cid:98)G(cid:48)
(cid:1)
∼=(cid:0)(cid:98)G(cid:48)
(cid:98)G(cid:48)
1 −−−−−→ 2(cid:2)3 ∼=(cid:0)(cid:98)G(cid:48)
−−−−−→(cid:0)(cid:98)G(cid:48)
σ in Section 5.5.1, or from the relation Lie((cid:98)G(cid:48)
σ, (cid:98)Gσ and (cid:98)G(cid:48)
In other words, there exists a split short exact sequence
(cid:76)(cid:57)(cid:57)−−−
σ)¯1
∼= SL(2)
(cid:1)
¯1 −−−−−→ (cid:98)G(cid:48)
Abelian; moreover, there exists an isomorphism
×3 (cid:110) 2(cid:2)3.
σ, G(cid:48)(cid:48)
∼= (cid:2)3
σ)rd
i=1
rd
¯1
σ
σ
σ
(cid:1)
∼= SL(2)
×3 −−−−−→ 1.
σ
σ
σ
rd
Proof . Much like the similar result for (cid:98)Gσ, we can deduce the present claim from the presen-
tation of (cid:98)G(cid:48)
σ) =(cid:98)g(cid:48)(σ) along with Theorem 4.5. (cid:4)
σ as algebraic supergroups.
5.5.3. The integral case: Gσ, G(cid:48)
In
the integral case, i.e., when σ ∈ Z3, the Lie supergroups we have introduced above are, in
indeed, this follows as a consequence of an alternative
fact, complex algebraic supergroups:
presentation of them, that makes sense if and only if σ ∈ Z3.
Let us look at Gσ, for some fixed σ ∈ Z3. Consider the generating set (5.1) for the groups
a ∈ A¯0 -- therein with(cid:101)hα(u) -- for all u ∈ U (A¯0), the group of units of A¯0. Every such(cid:101)hα(u) is
Γ¯0(A), and for each α ∈ {2ε1, 2ε2, 2ε3, θ} replace the generators hα(a) := exp(aHα) -- for all
the toral element in G(A) whose adjoint action on gσ is given by Ad((cid:101)hα(u))(Xγ) = uγ(Hα)Xγ
the set(cid:101)Γ¯0(A) :=(cid:8)x2εi(a),(cid:101)h2εi(u),(cid:101)hθ(u), x−2εi(a) i ∈ {1, 2, 3}, a ∈ A¯0, u ∈ U (A¯0)(cid:9)
generated by (cid:101)Γ (A) := G(A) ∪ Γ¯1(A) with the same relations as in Section 5.1.1 up to the
for all γ ∈ ∆; note that this makes sense, since we have γ(Hα) ∈ Z just because σ ∈ Z3. Now,
still generates G(A). A moment's thought then shows that Gσ(A) can be realized as the group
following changes: all relations that involve no generators of type hα(a) are kept the same, while
the others are replaced by the following ones (with {i, j, k} ∈ {1, 2, 3}):
At "singular values" of σ instead things are quite different; the precise claim is as follows:
−1
(cid:0)u±(−1)
(cid:101)h2εi(u)x±βj (η)(cid:101)h2εi(u)
(cid:0)u±σiη(cid:1),
(cid:101)h2εi(u)x±θ(η)(cid:101)h2εi(u)
(cid:0)u∓σiη(cid:1),
(cid:101)hθ(u)x±βi(η)(cid:101)hθ(u)
xθ(η+)x−θ(η−) =(cid:101)hθ(η−η+)x−θ(η−)xθ(η+).
= x±βj
= x±θ
= x±βi
−1
−1
−δi,j σiη(cid:1),
(cid:101)hθ(u)x±θ(η)(cid:101)hθ(u)
−1
= x±θ(η),
In fact, the key point here is that if (and only if) σ ∈ Z3, then all our construction does make sense
in the framework of algebraic supergeometry, namely Pσ := (G, g(σ)) is a super Harish-Chandra
in Section 2.4.3 -- cf. [8] again. If we present the groups G(A) using (cid:101)Γ¯0(A) as generating set,
is nothing but the corresponding
pair in the algebraic sense -- like in [8] -- and Gσ := GPσ
algebraic supergroup associated with Pσ trough the algebraic version of category equivalence
σ, (cid:98)Gσ
we can also extend such a description -- as σ ∈ Z3 -- to a presentation of the groups Gσ(A) as
or (cid:98)G(cid:48)
above. Leaving details to the reader, the same analysis applies when we look at G(cid:48)
σ instead of Gσ: whenever σ ∈ Z3, all of them are in fact complex algebraic supergroups.
σ, G(cid:48)(cid:48)
Singular Degenerations of Lie Supergroups of Type D(2, 1; a)
35
and {(cid:98)G(cid:48)
σ}σ∈V , {(cid:98)Gσ}σ∈V
5.5.4. A geometrical interpretation. In the previous discussion we considered five families
of Lie supergroups indexed by the points of V , namely {Gσ}σ∈V , {G(cid:48)
σ}σ∈V . Our analysis shows that these families have in common all the elements indexed
{σi = 0}. On the other hand,
, respectively, over the base space Spec(C[x]) ∼= V ∪ {(cid:63)}(cid:0)∼=A2C ∪ {(cid:63)}(cid:1), whose fibres
, L(cid:98)GC[x]
by "general" points, i.e., elements σ ∈ V × := V \ S with S :=
these families are entirely different at all points in the "singular locus" S.
In geometrical terms, each family forms a fibre space, say LGC[x], LG(cid:48)
and L(cid:98)G(cid:48)
σ}σ∈V , {G(cid:48)(cid:48)
3(cid:83)
i=1
, LG(cid:48)(cid:48)
C[x]
C[x]
,
C[x]
are Lie supergroups. Our results show that the fibres in any two of these fibre spaces do coincide
at general points -- where they are simple Lie supergroups -- and do differ instead at singular
points -- where they are non-simple indeed.
As an outcome, loosely speaking we can say that our construction provides five different
"completions" of the family {Gσ}σ∈V \S of simple Lie supergroups, by adding -- in five different
ways -- some new non-simple extra elements. Note also that, a priori, many other such "com-
pletions", more or less similar, could be devised: we just presented these ones as significant,
interesting examples, with no claims whatsoever of being exhaustive.
Acknowledgements
The first author is partially supported by the French Agence Nationale de la Recherche (ANR
GeoLie project ANR-15-CE40-0012). The second author acknowledges the MIUR Excellence
Department Project awarded to the Department of Mathematics, University of Rome "Tor Ver-
gata", CUP E83C18000100006. The authors also would like to thank the anonymous referees
for their useful comments and suggestions to improve the presentation of this article.
References
[1] Balduzzi L., Carmeli C., Fioresi R., A comparison of the functors of points of supermanifolds, J. Algebra
Appl. 12 (2013), 1250152, 41 pages, arXiv:0902.1824.
[2] Bouarroudj S., Grozman P., Leites D., Classification of finite dimensional modular Lie superalgebras with
indecomposable Cartan matrix, SIGMA 5 (2009), 060, 63 pages, arXiv:0710.5149.
[3] Chapovalov D., Chapovalov M., Lebedev A., Leites D., The classification of almost affine (hyperbolic) Lie
superalgebras, J. Nonlinear Math. Phys. 17 (2010), suppl. 1, 103 -- 161, arXiv:0906.1860.
[4] Deligne P., Morgan J.W., Notes on supersymmetry (following Joseph Bernstein), in Quantum Fields and
Strings: a Course for Mathematicians, Vols. 1, 2 (Princeton, NJ, 1996/1997), Amer. Math. Soc., Providence,
RI, 1999, 41 -- 97.
[5] Dooley A.H., Rice J.W., On contractions of semisimple Lie groups, Trans. Amer. Math. Soc. 289 (1985),
185 -- 202.
[6] Fioresi R., Gavarini F., Chevalley supergroups, Mem. Amer. Math. Soc. 215 (2012), vi+64 pages,
arXiv:0808.0785.
[7] Gavarini F., Chevalley supergroups of type D(2, 1; a), Proc. Edinb. Math. Soc. 57 (2014), 465 -- 491,
arXiv:1006.0464.
[8] Gavarini F., Global splittings and super Harish-Chandra pairs for affine supergroups, Trans. Amer. Math.
Soc. 368 (2016), 3973 -- 4026, arXiv:1308.0462.
[9] Gavarini F., Lie supergroups vs. super Harish-Chandra pairs: a new equivalence, arXiv:1609.02844.
[10] Iohara K., Koga Y., Central extensions of Lie superalgebras, Comment. Math. Helv. 76 (2001), 110 -- 154.
[11] Kac V.G., Classification of simple algebraic supergroups, Russian Math. Surveys 32 (1977), no. 3, 214 -- 215,
in Russian, available at http://www.mathnet.ru/php/archive.phtml?wshow=paper&jrnid=rm&paperid=
3212&option_lang=eng.
[12] Kac V.G., Lie superalgebras, Adv. Math. 26 (1977), 8 -- 96.
36
K. Iohara and F. Gavarini
[13] Kaplansky I., Graded Lie algebras I, University of Chicago Report, 1976, available at http://www1.osu.
cz/~zusmanovich/links/files/kaplansky/.
[14] Kaplansky I., Graded Lie algebras II, University of Chicago Report, 1976, available at http://www1.osu.
cz/~zusmanovich/links/files/kaplansky/.
[15] Scheunert M., The theory of Lie superalgebras. An introduction, Lecture Notes in Math., Vol. 716, Springer,
Berlin, 1979.
[16] Serganova V., On generalizations of root systems, Comm. Algebra 24 (1996), 4281 -- 4299.
[17] Vaintrob A.Yu., Deformation of complex superspaces and coherent sheaves on them, J. Sov. Math. 51 (1990),
2140 -- 2188.
[18] Veisfeiler B.Ju., Kac V.G., Exponentials in Lie algebras of characteristic p, Math. USSR Izv. 5 (1971),
777 -- 803.
|
1607.00336 | 1 | 1607 | 2016-06-28T16:47:48 | Two problems from the Polishchuk and Positselski book on Quadratic algebras | [
"math.RA",
"math.AC",
"math.CO"
] | In the book 'Quadratic algebras' by Polishchuk and Positselski [23] algebras with a small number of generators (n=2,3) are considered. For some number r of relations possible Hilbert series are listed, and those appearing as series of Koszul algebras are specified. The first case, where it was not possible to do, namely the case of three generators n=3 and six relations r=6 is formulated as an open problem. We give here a complete answer to this question, namely for quadratic algebras with dim A_1=dim A_2=3, we list all possible Hilbert series, and find out which of them can come from Koszul algebras, and which can not.
As a consequence of this classification, we found an algebra, which serves as a counterexample to another problem from the same book [23] (Chapter 7, Sec. 1, Conjecture 2), saying that Koszul algebra of finite global homological dimension d has dim A_1 >= d. Namely, the 3-generated algebra A given by relations xx+yx=xz=zy=0 is Koszul and its Koszul dual algebra A^! has Hilbert series of degree 4: H_{A^!}(t)= 1+3t+3t^2+2t^3+t^4, hence A has global homological dimension 4. | math.RA | math |
Two problems from the Polishchuk and Positselski book on Quadratic
algebras
Natalia Iyudu and Stanislav Shkarin
January 17, 2018
Abstract
In the book 'Quadratic algebras' by Polishchuk and Positselski [23] algebras with a small number
of generators (n = 2, 3) are considered. For some number r of relations possible Hilbert series are
listed, and those appearing as series of Koszul algebras are specified. The first case, where it was
not possible to do, namely the case of three generators n = 3 and six relations r = 6 is formulated
as an open problem. We give here a complete answer to this question, namely for quadratic algebras
with dim A1 = dim A2 = 3, we list all possible Hilbert series, and find out which of them can come
from Koszul algebras, and which can not.
As a consequence of this classification, we found an algebra, which serves as a counterexample
to another problem from the same book [23] (Chapter 7, Sec. 1, Conjecture 2), saying that Koszul
algebra of finite global homological dimension d has dim A1 > d. Namely, the 3-generated algebra
A given by relations xx + yx = xz = zy = 0 is Koszul and its Koszul dual algebra A! has Hilbert
series of degree 4: HA! (t) = 1 + 3t + 3t2 + 2t3 + t4, hence A has global homological dimension 4.
MSC: 16A22, 16S37, 14A22
Keywords: Quadratic algebras, Koszul algebras, Hilbert series, Grobner basis
1
Introduction
Quadratic algebras have been studied intensely during the past several decades. Being interesting in
their own right, they have many important applications in various parts of mathematics and physics
including algebraic geometry, algebraic topology and group theory as well as in mathematical physics.
They frequently originate in physics. One example is that the 3-dimensional Sklyanin algebras were
introduced and used in order to integrate a wide class of quantum systems on a lattice. These algebras
have their Koszul duals in the class of quadratic algebras with dim A1 = dim A2 = 3, the very class we
study in this paper.
Quadratic algebras are noncommutative objects which lie in foundation of many noncommutative
theories, for example, in work of A.Connes and M.Dubois-Violette [8], notions of noncommutative
differential geometry obtain their purely algebraic counterpart through introducing the appropriate
quadratic form on quadratic algebras. The big area of research generalizing notions of algebraic ge-
ometry to noncommutative spaces, due to Artin, Tate, Van den Bergh, Stafford etc.
[3, 4, 5, 25]
contains a great deal of studying structural and homological properties of quadratic algebras and their
representations. Certain quadratic algebras serve as important examples for the notions of noncommu-
tative (symplectic) spaces introduced by Kontsevich [18, 19], so information about general rules on the
structure of such algebras makes it possible to describe examples explicitly.
We find it very important to study fundamental, most general properties of quadratic algebras,
their Hilbert series, Koszulity, other homological properties, PBW type properties, etc. and to develop
appropriate tools for that, which is a goal of present paper.
For further information on quadratic algebras, their Hilbert series and various aspects of their appli-
cations, we refer to [1, 2, 7, 14, 15, 17, 16, 23, 9, 21, 24, 13, 26, 22, 10, 11, 6, 20, 27, 12] and references
therein, however it will give still the list which is far from being exhaustive.
Throughout this paper, K is an arbitrary field. For a Z+-graded vector space B, Bm always stands
for the mth component of B and HB(t) =
∞
dim Bj tj is the Hilbert series of B.
Pj=0
If V is an n-dimensional vector space over K, then F = F (V ) is the tensor algebra of V , which is
naturally identified with the free algebra Khx1, . . . , xni for any choice of a basis x1, . . . , xn in V . We
often use the juxtaposition notation for the operation in F (V ) (for instance, we write xjxk instead of
xj ⊗ xk). We always use the degree grading on F : the mth graded component Fm of F is V m. A
degree graded algebra A is a quotient of F by a proper graded ideal I (I is graded if it is the direct
sum of I ∩ Fm). This ideal is called the ideal of relations of A. If x1, . . . , xn is a fixed basis in V and
the monomials in xj carry an ordering compatible with the multiplication in A, we can speak of the
Grobner basis of the ideal of relations of A. If Λ is the set of leading monomials of the elements of
such a basis, then the normal words for A are the monomials in xj featuring no element of Λ as a
submonomial. Normal words form a basis in A as a K-vector space. Thus, knowing the normal words
implies knowing the Hilbert series.
If R is a subspace of the n2-dimensional space V 2, then the quotient of F by the ideal I(V, R)
generated by R is called a quadratic algebra and denoted A(V, R). Following [23], we say that a quadratic
algebra A = A(V, R) is a PBW-algebra if there are linear bases x1, . . . , xn and g1, . . . , gm in V and R
respectively such that g1, . . . , gm is a Grobner basis of the ideal I = I(V, R) with respect to some well-
ordering on the monomials in x1, . . . , xn, compatible with multiplication. For a given basis x1, . . . , xn in
V , we get a bilinear form on Khx1, . . . , xni by setting [u, v] = δu,v for every pair of monomials u and v in
x1, . . . , xn. The quadratic algebra A! = A(V, R⊥), where R⊥ = {u ∈ V 2 : [r, u] = 0 for each r ∈ R}, is
called the Koszul dual algebra of A, we will call it also just dual algebra. Note that up to an isomorphism,
the graded algebra A! does not depend on the choice of a basis in V . It is well-known that A is PBW
if and only if A! is PBW.
A degree graded algebra A is called Koszul if the graded left A-module K (the structure is provided
by the augmentation map) has a free resolution · · · → Mm → · · · → M1 → A → K → 0 with the second
last arrow being the augmentation map, and with each Mk generated in degree k. The latter means
that the matrices of the maps Mm → Mm−1 with respect to some free bases consist of homogeneous
elements of degree 1. Replacing left modules by the right ones leads to the same class of algebras. We
use the following well-known properties of Koszul algebras:
every Koszul algebra is quadratic; every PBW-algebra is Koszul;
A is Koszul ⇐⇒ A! is Koszul;
HA(−t)HA!(t) = 1 if A is Koszul.
(1.1)
In [23, Chapter 6, Section 5] possible Hilbert series of Koszul algebras A with small values of dim A1
and dim A2 are listed. The first case not covered there is dim A1 = dim A2 = 3. In this case, only the
Hilbert series of PBW algebras are given. It is stated in [23] that the complete list of Hilbert series of
quadratic algebras satisfying dim A1 = dim A2 = 3 as well as the complete list of the Hilbert series of
Koszul algebras in this case are unknown. We fill this gap by proving the following results.
Theorem 1.1. For quadratic algebras A satisfying dim A1 = dim A2 = 3, the complete list of possible
Hilbert series is {H1, . . . , H11}, where
H1(t)=1+3t+3t2;
H3(t)=1+3t+3t2+t3+t4+t5+ · · · = 1+2t−2t3
H5(t)=1+3t+3t2+2t3+t4;
H7(t)=1+3t+3t2+2t3+2t4+2t5+ · · · = 1+2t−t3
H9(t)=1+3t+3t2+4t3+4t4+4t5+ · · · = 1+2t+t3
H11(t)=1+3t+3t2+5t3+8t4+13t5+ · · · = 1+2t−t2−t3
1−t
1−t
1−t
;
1−t−t2
,
H2(t)=1+3t+3t2+t3 = (1 + t)3;
H4(t)=1+3t+3t2+2t3;
H6(t)=1+3t+3t2+2t3+t4+t5+ · · · = 1+2t−t3−t4
1−t
; H8(t)=1+3t+3t2+3t3+3t4+3t5+ · · · = 1+2t
1−t ;
; H10(t)=1+3t+3t2+4t3+5t4+6t5+ · · · = 1+t−2t2+t3
;
(1−t)2
;
where the last series, starting from the third term, is formed by consecutive Fibonacci numbers.
As we have mentioned above, the complete list of Hilbert series of PBW algebras A satisfying
dim A1 = dim A2 = 3 can be found in [23]. It consists of Hj with j ∈ {2, 7, 8, 9, 10, 11}.
Theorem 1.2. For Koszul algebras A satisfying dim A1 = dim A2 = 3, the complete list of Hilbert
series consists of Hj with j ∈ {2, 5, 6, 7, 8, 9, 10, 11}. That is, for each j ∈ {2, 5, 6, 7, 8, 9, 10, 11}, there
is a Koszul algebra A satisfying HA = Hj, while for j ∈ {1, 3, 4}, every quadratic algebra A satisfying
HA = Hj is non-Koszul.
As a consequence of these classification Theorems 1.1 and 1.2, we can find an algebra, which serves
as a counterexample to another problem formulated in the Polishchuk and Positselski book [23].
Conjecture 1.3. ([23], Chapter 7, Sec. 1, Conjecture 2) Any Koszul algebra of finite global homological
dimension d has dim A1 > d. By duality this is equivalent to the following statement:
for a Koszul
algebra B with Bd+1 = 0 and Bd 6= 0 one has dim B1 > d.
We can consider algebra A5 from the Table 1. It is given by relations xx−yx = xy = yy = yx = zx =
zz = 0. The Hilbert series of this algebra is HA(t) = H5 = 1 + 3t + 3t2 + 2t3 + t4, so it is a polynomial
of degree 4. Then quite straightforward arguments in Proposition 4.2 ensure that the Koszul dual
algebra A!
5, given by relations xx + yx = xz = zy = 0, is Koszul. This provides a counterexample to
Conjecture 1.3. Namely, the algebra A! serves as a counterexample to the first part of the conjecture: it
has a global homological dimension 4. The algebra A itself is a counterexample to the second statement
of the conjecture, since its Hilbert series is a polynomial of degree 4.
Note that the list of series acquires two extra members H5 and H6 when the PBW condition is
relaxed to Koszulity.
The key lemma, allowing to manage all possibilities, is the following linear algebra statement.
Lemma 1.4. Let V be a 3-dimensional vector space over an infinite field K and R be a 6-dimensional
subspace of V ⊗ V . Then at least one of the following statements is true:
(P1) there is a 1-dimensional subspace L ⊂ V such that (V ⊗ L) ⊕ R = V ⊗ V or (L ⊗ V ) ⊕ R = V ⊗ V ;
(P2) there is 1-dimensional subspace L ⊂ V such that V ⊗ L ⊂ R or L ⊗ V ⊂ R;
(P3) there is an invertible linear map T : V → V such that R = span {x ⊗ T x : x ∈ V }.
While (P1) and (P2) are not mutually exclusive, (P3) is incompatible with each of (P1) and (P2).
In Section 2 we show that the Hilbert series of any quadratic algebra A, satisfying dim A1 = dim A2 =
3 belongs to {H1, . . . , H11}, applying Lemma 1.4 and Grobner basis techniques. Then in Section 3 we
give a proof of Lemma 1.4. In Section 4 we show that Hj is the Hilbert series of a quadratic algebra A
for 1 6 j 6 11. We also observe that A can be chosen Koszul if j ∈ {2, 5, 6, 7, 8, 9, 10, 11} and that every
algebra A with HA = Hj for j ∈ {1, 3, 4} is non-Koszul, thus completing the proofs of Theorems 1.1
and 1.2.
Throughout the paper, when talking of Grobner bases, assume that the monomials carry the left-
to-right degree lexicographical ordering with the variables ordered by x > y > z or x1 > x2 > x3
(depending on how the variables are called in each case).
2 Admissible series
In this section we apply Lemma 1.4 and Grobner basis arguments to prove the following result. Next
section will be dedicated to the proof of the Lemma 1.4 itself.
Proposition 2.1. Let A be a quadratic K-algebra satisfying dim A1 = dim A2 = 3. Then HA ∈
{H1, . . . , H11}.
Since replacing the ground field K by a field extension does not change the Hilbert series of an algebra
given by generators and relations, for the purpose of proving Proposition 2.1, we can without loss of
generality assume that K is algebraically closed. Then K is infinite. By Lemma 1.4, Proposition 2.1 is
an immediate corollary of the following three lemmas. We essentially consider three possibilities given
by Lemma 1.4, and in each case find out which series are possible, looking mainly at the shape of the
Grobner basis.
Lemma 2.2. Let V be a 3-dimensional vector space over K and R be a 6-dimensional subspace of
V ⊗ V such that condition (P1) of Lemma 1.4 is satisfied. Then for the quadratic algebra A = A(V, R),
HA ∈ {H1, . . . , H8}.
Lemma 2.3. Let V be a 3-dimensional vector space over an algebraically closed field K and R be a
6-dimensional subspace of V ⊗ V such that condition (P2) of Lemma 1.4 is satisfied, while (P1) fails.
Then for the quadratic algebra A = A(V, R), HA ∈ {H1, H2, H3, H7, H9, H10, H11}.
Lemma 2.4. Let V be a 3-dimensional vector space over K and R ⊂ V ⊗ V be a subspace satisfying
condition (P3) of Lemma 1.4. Then for the quadratic algebra A = A(V, R), HA = H2.
Proof of Lemma 2.2. Since (P1) is satisfied, there is a 1-dimensional subspace L ⊂ V such that (V ⊗
L) ⊕ R = V ⊗ V or (L ⊗ V ) ⊕ R = V ⊗ V . These two cases reduce to each other by passing to the
algebra with the opposite multiplication. Thus we can assume that (L ⊗ V ) ⊕ R = V ⊗ V . Pick a basis
x1, x2, x3 in V such that x3 spans L. Since (L ⊗ V ) ⊕ R = V ⊗ V , there is a linear basis in R of the
form (we skip the symbol ⊗ for the rest of the proof):
rj,k = xjxk − x3uj,k for 1 6 j 6 2 and 1 6 k 6 3, where uj,k ∈ V .
(2.1)
It follows that in the algebra A, A2 = x3V = x3A1. Then A3 = A2V = x3V V = x3A2 = x2
Iterating, we get An = x3An−1 = xn−1
each n > 2. We also know that dim A1 = dim A2 = 3.
3V .
3 V for each n > 2. In particular, dim An 6 dim An−1 6 3 for
Case 1: dim A3 = 3. This can only happen if rj,k form a Grobner basis of the ideal I = I(V, R).
Since the leading monomials of these relations are xjxk for 1 6 j 6 2 and 1 6 k 6 3, the normal words
of degree n > 3 are xn−1
3 xj for 1 6 j 6 3. Hence dim An = 3 for n > 3 and HA = H8.
3xj with 1 6 j 6 3. If j = 3, we have g = x3
Case 2: dim A3 = 2. This happens when there is exactly one degree 3 element g of the Grobner basis
of I. The leading monomial of g must have the shape x2
3 and
3 = 0 in A. Hence for n > 4, An = xn−1
3 V = {0}. Thus HA = 1 + 3t + 3t2 + 2t3 = H4. It remains to
x3
consider the case j ∈ {1, 2}. Swapping x1 and x2, if necessary, we can without loss of generality assume
that j = 1. We know that dim A4 6 dim A3 = 2. The case dim A4 = 2 can only happen if the relations
rj,k together with g form a Grobner basis of I. In this case the normal words of degree n > 3 are
xn−1
3 xk with k ∈ {2, 3}. This gives HA = 1 + 3t + 3t2 + 2t3 + 2t4 + 2t5 + · · · = H7. It remains to consider
the case dim A4 = 1. This happens when there is exactly one degree 4 element h in the Grobner basis
3xk with 2 6 k 6 3. If k = 3, we have h = x4
of I. The leading monomial of h must have the shape x3
3
and x4
3 V = {0}. Thus HA = 1 + 3t + 3t2 + 2t3 + t4 = H5.
Assume now that k = 2. If the relations rj,k together with g and h do not form the Grobner basis of
I, there is a degree 5 element q of this Grobner basis. By looking at the leading terms of rj,k, g and
h, we see that the only possibility is for q to be equal x5
3 up to a non-zero scalar multiple. Again, this
gives HA = H5. On the other hand, if rj,k together with g and h do form the Grobner basis of I, the
only normal word of degree n > 4 is xn
3 = 0 in A. Hence for n > 5, An = xn−1
3 . Hence HA = 1 + 3t + 3t2 + 2t3 + t4 + t5 + · · · = H6.
3xj and x2
3xk respectively with 1 6 j < k 6 3. If k = 3, we have h = x3
Case 3: dim A3 = 1. This happens when there are exactly two degree 3 elements g and h of the
Grobner basis of I. By swapping g and h, if necessary, we can assume that the leading terms of g and
h are x2
3 = 0 in A. Hence
for n > 4, An = xn−1
3 V = {0}. Thus HA = 1 + 3t + 3t2 + t3 = H2. It remains to consider the case
j = 1, k = 2. If the relations rj,k together with g and h do not form the Grobner basis of I, there is a
degree 4 element q in this Grobner basis. By looking at the leading terms of rj,k, g and h, we see that
the only possibility is for q to be equal x4
3 up to a non-zero scalar multiple. Again, this gives HA = H2.
If rj,k together with g and h do form the Grobner basis of I, the only normal word of degree n > 3 is
3 . Hence HA = 1 + 3t + 3t2 + t3 + t4 + t5 + · · · = H3.
xn
Case 4: dim A3 = 0. Obviously, HA = 1 + 3t + 3t2 = H1.
3 and x3
Proof of Lemma 2.3. Since (P2) holds, there is a 1-dimensional subspace L of V such that V L ⊂ R or
LV ⊂ R (we skip the symbol ⊗ throughout the proof). The cases V L ⊂ R and LV ⊂ R reduce to each
other by passing to the algebra with the opposite multiplication. Thus we can assume that LV ⊂ R.
Pick x in V , which spans L. Since (P1) fails,
for each u ∈ V \ {0}, there is v = v(u) ∈ V \ {0} such that uv ∈ R.
(2.2)
We shall verify that there are y, z ∈ V such that x, y, z is a basis in V and at least one of the following
conditions holds:
R = span {xx, xy, xz, yx, zx, h} with h ∈ {yy, yz − azy, yz − zy + zz} (a ∈ K);
R = span {xx, xy, xz, yy, zy, h} with h ∈ {yx − zz, yz − zx, yx, yz, zx, zz};
R = span {xx, xy, xz, yy, zz, h} with h ∈ {yx + zx, yz + zy, yx + zx − yz − zy}.
(2.3)
(2.4)
(2.5)
Case 1: V L ⊂ R. Pick arbitrary u, v ∈ V such that x, u, v is a basis in V . Then the 5-dimensional
space LV + V L spanned by S0 = {xx, xu, xv, ux, vx} is contained in R. Since R is 6-dimensional, it is
spanned by S0 ∪ {f }, where f = auu + buv + cvu + dvv with (a, b, c, d) ∈ K4, (a, b, c, d) 6= (0, 0, 0, 0).
Since K is algebraically closed, there is a non-zero (p, s) ∈ K2 such that ap2 + (b + c)ps + ds2 = 0.
Next, pick (q, t) ∈ K2 such that (p, s) and (q, t) are linearly independent. The non-degenerate linear
substitution, in which old u and v are replaced by pu + qv and su + tv respectively, transforms f into
g = αuv + βvu + γvv with non-zero (α, β, γ) ∈ K3. If α = 0 and β 6= 0, we set y = v and z = βu + γv,
while if α 6= 0 and β = 0, we set y = αu + γv and z = u. This substitution transforms g into a
(non-zero) scalar multiple of yz. Now, with respect to the basis x, y, z, R is spanned by S ∪ {yz} with
S = {xx, xy, xz, yx, zx}. If α = β = 0, we set y = v and z = u and observe that R is spanned by
S ∪ {yy}. If αβ 6= 0 and α + β 6= 0, we set y = u + γv
α+β , z = v and observe that R is spanned by
S ∪ (cid:8)yz + β
α zy(cid:9). If αβ 6= 0 and α + β = γ = 0, with respect to y = u and z = v, R is spanned by
S ∪ {yz − zy}. Finally, if αβγ 6= 0 and α + β = 0, we set y = αu
γ and z = v, with respect to which R is
spanned by S ∪ {yz − zy + zz}. Thus (2.3) is satisfied provided V L ⊂ R.
Case 2: V M ⊂ R for a 1-dimensional subspace M of V such that M 6= L. Pick u ∈ M \ {0}.
Since L 6= M , x and u are linearly independent. For w ∈ V such that x, u, w is a basis in V , R =
span (S0 ∪ {f }), where S0 = {xx, xu, xw, uu, wu} and f = aux + buw + cwx + dww with (a, b, c, d) ∈ K4,
(a, b, c, d) 6= (0, 0, 0, 0). For α ∈ K∗ and p, q ∈ K we can consider the basis x, y, z in V defined by u = αy
and w = z + px + qy. A direct computation shows that with respect to this basis, R = span (S ∪ {g}),
where S = {xx, xy, xz, yy, zy} and
g = (aα + bpα + cq + dpq)yx + (bα + dq)yz + (c + dp)zx + dzz.
d and p = − c
If d 6= 0 and ad = bc, by choosing α = 1, q = − b
d , we turn g into a (non-zero) scalar
multiple of zz. If d = 0 and ad 6= bc, by choosing α = d2
d , we turn g into a
scalar multiple of yx − zz. If d = 0 and bc 6= 0, by choosing α = − c
b , we turn g into
a scalar multiple of yz − zx. If b = d = 0 and c 6= 0, by choosing α = 1, p = 0 and q = − a
c , we turn g
into a scalar multiple of zx. If c = d = 0 and b 6= 0, by choosing α = 1, q = 0 and p = − a
b , we turn
g into a scalar multiple of yz. Finally, if b = c = d = 0, by choosing α = 1 and p = q = 0, we turn g
into a scalar multiple of yx. Thus (2.4) is satisfied provided V M ⊂ R for a 1-dimensional subspace M
different from L.
b , p = 0 and q = a
bc−ad , q = − b
d and p = − c
Case 3: V M 6⊆ R for every 1-dimensional subspace M of V . This is precisely the negation of the
assumptions of Cases 1 and 2. First, we shall verify that in this case
yz /∈ R whenever x, y, z is a basis in V .
(2.6)
We argue by contradiction. Assume that (2.6) fails. Then there are y, z ∈ V such that x, y, z is a basis in
V and yz ∈ R. By (2.2), there are non-zero (a, b, c), (p, q, r) ∈ K3 such that z(ax + by + cz), (y + z)(px +
qy + rz) ∈ R. The assumption of Case 3 implies linear independence of z, ax + by + cz and px + qy + rz.
Indeed, if they were linearly dependent, using the inclusions yz, z(ax+by +cz), (y +z)(px+qy +rz) ∈ R,
one easily finds a non-zero u ∈ V such that yu, zu ∈ R. Since xu ∈ R, this implies V M ⊂ R with M
being the linear span of u. Linear independence of z, ax + by + cz and px + qy + rz implies that aq 6= bp
and that
R = span {xx, xy, xz, yz, z(ax + by + cz), (y + z)(px + qy + rz)}.
(2.7)
Since K is infinite, we can pick θ ∈ K \ {0, 1}. By (2.2), there is a non-zero (α, β, γ) ∈ K3 such that
(y + θz)(αx + βy + γz) ∈ R. By (2.7), there exist c1, c2, c3 ∈ K such that
(y + θz)(αx + βy + γz) = c1yz + c2z(ax + by + cz) + c3(y + z)(px + qy + rz),
where the equality holds in Khx, y, zi. Opening up the brackets in the above display, we obtain
α − pc3 = β − qc3 = γ − c1 − rc3 = αθ − ac2 − qc3 = βθ − bc2 − qc3 = γθ − cc2 − rc3 = 0.
Plugging α = pc3 and β = qc3 into αθ − ac2 − qc3 = βθ − bc2 − qc3 = 0, we get bc2 + (1 − θ)qc3 =
ac2 + (1 − θ)pc3 = 0. Since θ 6= 1 and aq 6= bp, the determinant (1 − θ)(bp − aq) of this system of
two linear equations on c2 and c3 is non-zero. Hence c2 = c3 = 0. Since θ 6= 0, the above display
implies α = pc3 = 0, β = qc3 = 0 and γ = cc2+rc3
= 0, which contradicts (α, β, γ) 6= (0, 0, 0). This
contradiction proves (2.6).
θ
Now (2.6) together with (2.2) imply that
for each u ∈ V \ L, there is (au, bu) ∈ K2 \ {(0, 0)} such that u(auu + bux) ∈ R.
(2.8)
Indeed, otherwise V L ⊂ R. Next, bu 6= 0 for
Observe that au 6= 0 for a Zarisski generic u ∈ V .
a Zarisski generic u ∈ V .
Indeed, otherwise R contains the 6-dimensional space S of symmetric
elements of V 2. Since R also contains LV and LV ∩ S is one-dimensional, dim R > 8 > 6, which is a
contradiction. Thus we can pick s, t ∈ V such that x, s, t is a basis in V and asbsatbt 6= 0. Now, using
the inclusions s(ass + bss), t(att + btx) ∈ R, we can pick p, q ∈ K∗ such that for u = ps and v = qt,
u(x − u), v(x − v) ∈ R. For a = au+v and b = bu+v, according to (2.8), we have (a, b) 6= (0, 0) and
(u + v)(au + av + bx) ∈ R. Then R = span {xx, xu, xv, u(x − u), v(x − v), (u + v)(au + av + bx)}. Now
set y = x − u and z = x − v. With respect to the basis x, y, z, the last equality can be rewritten as
R = span (S ∪ {c(y + z)x − a(yz + zy)}, where S = {xx, xy, xz, yy, zz} with c = b + 2a. If c = 0, then
R = span (S ∪ {yz + zy}). If a = 0, then R = span (S ∪ {yx + zx}). If ac 6= 0, then replacing y and
z by αy and αz for an appropriate α ∈ K∗, we get R = span (S ∪ {yx + zx − yz − zy}). Thus (2.5) is
satisfied provided V M 6⊆ R for every 1-dimensional subspace M of V .
It remains to determine the Hilbert series of A = A(V, R) when R satisfies one of the conditions
(2.3), (2.4) or (2.5). If (2.3) is satisfied, the defining relations xx, xy, xz, yx, zx, and h of A form a
Grobner basis of the ideal I = I(V, R). If h = yz − azy or h = yz − zy + zz, then the normal words of
degree n > 2 are zkyn−k for 0 6 k 6 n. Since there are n + 1 of them, we have HA = H10. If h = yy,
then the normal words of degree n > 2 are all monomials in y and z, which do not contain yy as a
submonomial. It is easy to see that the number an of such monomials satisfies the recurrent relation
an+2 = an+1 + an, which together with a2 = 3 and a3 = 5 implies HA = H11.
Next, assume that (2.4) is satisfied. That is, A is given by the relations xx, xy, xz, yy, zy and h
with h ∈ {yx − zz, yz − zx, yx, yz, zx, zz}. If h = yx − zz, the defining relations together with yzz,
zzx, zzy and zzz form a Grobner basis of I. The only normal word of degree > 3 is yzx, which gives
HA = H2. If h = yz − zx, then the defining relations together with zzx form a Grobner basis of I. The
only normal word of degree n > 3 is zn, which implies HA = H3. If h ∈ {yx, yz, zx, zz}, A is monomial
and therefore the defining relations form a Grobner basis of I. If h = zz, the only normal word of
degree > 3 is yzx and HA = H2. If h = zx, the only normal words of degree n > 3 are zn and yzn−1,
which gives HA = H7. If h = yz, the only normal words of degree n > 3 are zn−1x and zn, yielding
HA = H7. If h = yx, the only normal words of degree n > 3 are yzn−2x yzn−1, zn−1x and zn. Hence
HA = H9.
Finally, assume that (2.5) is satisfied. That is, A is given by the relations xx, xy, xz, yy, zz and
h with h ∈ {yx + zx, yz + zy, yx + zx − yz − zy}. If h = yx + zx − yz − zy, the defining relations
together with yzx, yzy and zyz form a Grobner basis of I. There are no normal words of degree > 3
and therefore HA = H1. If h = yz + zy, the defining relations form a Grobner basis of I. The only
normal word of degree > 3 is zyx and HA = H2. Finally, if h = yx + zx, the defining relations together
with yzx form a Grobner basis of I. For n > 3 there are exactly 2 normal words of degree n being the
monomials in y and z in which y and z alternate: yzyz . . . and zyzy . . . Hence HA = H7. Since we
have exhausted all the options, the proof is complete.
Proof of Lemma 2.4. The fact that R is 6-dimensional is straightforward. Indeed, R is the image of the
6-dimensional space of the symmetric elements of V ⊗ V under the invertible linear map I ⊗ T . Now,
replacing the ground field by a field extension does not change the Hilbert series of an algebra given
by generators and relations. Hence, without loss of generality we can assume that K is algebraically
closed. This allows us to pick a basis x1, x2, x3 in V with respect to which the matrix of T has the
Jordan normal form.
If T has 3 Jordan blocks, T has the diagonal matrix with respect to the basis x1, x2, x3 with the
non-zero numbers (eigenvalues) λ1, λ2 and λ3 on the diagonal. One easily sees that in this case R is
spanned by x2
j with 1 6 j 6 3 and λkxjxk + λjxkxj with 1 6 j < k 6 3. If T has 2 Jordan blocks,
we can assume that the size two block is in the left upper corner and corresponds to the eigenvalue λ,
while the size one block is in the right lower corner and corresponds to the eigenvalue µ. In this case
1, λx2
x2
3, x1x2 + x2x1, µx1x3 + λx3x1 and µx2x3 + x3x1 + λx3x2 forms a linear basis in R.
Finally, if T has just one Jordan block corresponding to the eigenvalue λ, a linear basis in R is formed
by x2
3, x1x2 + x2x1, x1x3 + x3x1 − x2x1 and x2x3 + x3x2 + x3x1. In
any case, this linear basis in R is also a Grobner basis in I(V, R) with the only normal word of degree
> 3 being x3x2x1. This gives HA = H2.
2 + x2x1, λx3x1 + λx3x2 + x2
2 + x2x1, x2
1, λx2
This completes the proof of Proposition 2.1. Note that if K is algebraically closed and A is a
quadratic algebra satisfying HA = Hj for j ∈ {8, 9, 10, 11}, Lemma 1.4 can be applied and A falls into
one of the cases considered in the proofs of Lemmas 2.2 and 2.3. Scanning the proofs, one sees that
whenever HA = Hj for j ∈ {8, 9, 10, 11}, A is actually PBW and therefore Koszul. Since the Hilbert
series or Koszulity do not notice an extension of the ground field, we can drop the condition that K is
algebraically closed. This observation automatically implies the following Koszulity result.
Proposition 2.5. If A is a quadratic algebra satisfying HA = Hj for j ∈ {8, 9, 10, 11}, then A is
Koszul. Moreover, A is PBW provided K is algebraically closed.
3 Proof of Lemma 1.4
We start by reformulating Lemma 1.4. First, if we take a pairing on V ⊗ V as in the definition of a
dual algebra, then in terms of S = R⊥, Lemma 1.4 reads in the following way.
Lemma 3.1. Let V be a 3-dimensional vector space over an infinite field K and S be a 3-dimensional
subspace of V ⊗ V . Then at least one of the following statements is true:
(P1′) there is a 2-dimensional subspace M ⊂ V such that (V ⊗M )⊕S=V ⊗V or (M ⊗V )⊕S=V ⊗V ;
(P2′) there is a 2-dimensional subspace M ⊂ V such that V ⊗ M ⊃ S or M ⊗ V ⊃ S;
(P3′) there is an invertible linear map T : V → V such that S = span {x ⊗ T y − y ⊗ T x : x, y ∈ V }.
For two vector spaces V1 and V2 over K, L(V1, V2) stands for the vector space of all linear maps from
V1 to V2. Using the natural isomorphism between V ⊗ V and L(V ∗, V ) together with the fact that a
two-dimensional subspace of V is exactly the kernel of a non-zero linear functional, we can reformulate
Lemma 3.1 in the following way.
Lemma 3.2. Let V be a 3-dimensional vector space over an infinite field K and S be a 3-dimensional
subspace of L(V ∗, V ). Then at least one of the following statements is true:
(P1′′) there is f ∈ V ∗ such that the map A 7→ Af or the map A 7→ A∗f from S to V is injective;
(P2′′) there is a non-zero f ∈ V ∗ such that Af = 0 for all A ∈ S or A∗f = 0 for all A ∈ S;
(P3′′) there is an invertible T ∈ L(V, V ) such that g(T Af ) = −f (T Ag) for all f, g ∈ V ∗ and A ∈ S.
In other words, we have to show that (P3′′) holds if both (P1′′) and (P2′′) fail. Hence, Lemma 3.2
and therefore Lemma 1.4 will follow if we prove the following result.
Lemma 3.3. Let V1 and V2 be a 3-dimensional vector spaces over an infinite field K and S be a
3-dimensional subspace of L(V1, V2). Assume also that
(L1) TA∈S
(L2) {Au : A ∈ S} 6= V2 for each u ∈ V1 and {A∗f : A ∈ S} 6= V ∗
ker A = TA∈S
ker A∗ = {0};
1 for each f ∈ V ∗
2 .
Then there exist linear bases in V1 and V2 such that S in the corresponding matrix form is exactly the
space of 3 × 3 antisymmetric matrices.
Proof. First, we shall show that
for every non-zero x ∈ V1, the space Sx = {Ax : A ∈ S} is two-dimensional.
(3.1)
By (L1), Sx 6= {0} for each x ∈ V1 \ {0}. By (L2), Sx 6= V2 for each x ∈ V1. Thus Sx for x ∈ V1 \ {0}
is either one-dimensional or two-dimensional. Assume that (3.1) fails. Then there is x1 ∈ V1 such
that Sx1 is one-dimensional. Then we can pick a basis A1, A2, A3 in S such that A1x1 = y1 6= 0 and
A2x1 = A3x1 = 0. By (L1), the linear span of the images of all A ∈ S is V2. Hence we can pick
x2, x3 ∈ V1 such that x1, x2, x3 is a basis in V1, while y1, y2, y3 is a basis in V2, where yj = Ajxj. With
respect to the bases x1, x2, x3 and y1, y2, y3, the matrices of A1, A2 and A3 have the shape
1 ∗ ∗
0 ∗ ∗
0 ∗ ∗
,
0 0 ∗
0 1 ∗
0 0 ∗
and
0 ∗ 0
0 ∗ 0
0 ∗ 1
,
respectively.
Keeping the basis in V2 as well as x1 and replacing x2 and x3 by x2 + αx1 and x3 + βx1 respectively
with appropriate α, β ∈ K, we can kill the second and the third entries in the first row of the first
matrix. With respect to the new basis, the matrices of A1, A2 and A3 are
0
a8
a9
,
1
0
0 a1 a2
0 a3 a4
with aj ∈ K.
By (L2), for every u = (x, y, z) ∈ K3, A1u, A2u, A3u are linearly dependent and AT
3 u are
linearly dependent as well (here Aj stand for the matrices of the linear maps Aj), where T denotes the
transpose matrix. Computing these vectors, we see that these conditions read
and
0
0
0
0
0 a10 1
0 0 a5
0 1 a6
0 0 a7
1 u, AT
2 u, AT
det
x
a5z
a8y
a1y + a2z a3y + a4z
y + a6z
a7z
a9y
a10y + z
= det
a1y + a3z
x
0
0 a8x + a9y + a10z
y
a2y + a4z
a5x + a6y + a7z
z
= 0
for all x, y, z ∈ K. Since K is infinite, the above two determinants must be zero as polynomials in x, y, z.
The first determinant has the shape x(a10y2 + a6z2 + (1 + a6a10 − a7a9)yz) + g with g ∈ K[y, z]. Hence
a10 = a6 = 0 and a7a9 = 1. Taking into account that a10 = a6 = 0, we see that the xyz-coefficient of
the second determinant is a7a9. Hence a7a9 = 0, which contradicts a7a9 = 1. This contradiction proves
(3.1).
Now we pick a non-zero u ∈ V1. By (L2), there is a non-zero A1 ∈ S such that A1u = 0. Since
A1 6= 0, there is x ∈ V1 such that A1x 6= 0. Since a Zarisski generic x will do, we can assure the
extra condition Su 6= Sx (otherwise (L1) is violated). Obviously, u and x are linearly independent.
By (3.1), we can find A2 ∈ S such that A1x and A2x are linearly independent. Again suppose A2 is
Zarisski generic and therefore we can achieve the extra condition that A1x, A2x and A2u are linearly
independent (otherwise Su = Sx). Now y1 = −A2u, y2 = A1x and y3 = A2x form a basis in V2. By
(L2), there is a non-zero A3 ∈ S such that A3x = 0. Clearly, Aj are linearly independent (=they form
a basis in S). Pick a basis x1, x2, x3 in V1 such that x1 = x and x3 = u. With respect to the bases
x1, x2, x3 and y1, y2, y3, the matrices of A1, A2 and A3 have the form
0 ∗ 0
1 ∗ 0
0 ∗ 0
,
0 ∗ −1
0 ∗
0
1 ∗
0
and
0 ∗ ∗
0 ∗ ∗
0 ∗ ∗
,
respectively.
Keeping the basis in V2 as well as x1 and x3 and replacing x2 by x2 + αx1 + βx3 with appropriate
α, β ∈ K, we can kill the middle entry in the first matrix and the second entry of the first row of the
second matrix. With respect to the new basis, the matrices of A1, A2 and A3 have shape:
0 a1 0
1
0
0 a2 0
0
with aj ∈ K.
By (L2), for every vector u = (x, y, z) ∈ K3, A1u, A2u, A3u are linearly dependent and AT
are linearly dependent. Computing these vectors, we see that these conditions mean:
and
0 a6
a9
0 a7 a10
0 a8 a11
0
0 a4
1 a5
0 −1
0
0
,
1 u, AT
2 u, AT
3 u
det
a1y
−z
x
a4y
a2y
x + a5y
a6y + a9z a7y + a10z a8y + a11z
= det
y
−z
0
a1x + a2z
a4y + a5z
0
x
a6x + a7y + a8z a9x + a10y + a11z
= 0
for all x, y, z ∈ K. Since K is infinite, these two determinants are zero as polynomials in x, y, z. The
terms containing x2 in the first determinant amount to x2(a6y + a9z), which implies a6 = a9 = 0. The
yz2-coefficient of the same polynomial is −a2a10. Hence a2a10 = 0. First, we show that a2 = 0. Indeed,
assume the contrary. Then a2 6= 0 and therefore a10 = 0. Now z3-coefficient in the second determinant
is −a2a11. Hence a11 = 0. Taking into account that a6 = a9 = a10 = a11 = 0, we see that in the
first determinant the terms containing z amount to z(a8xy − a2a7y2).
It follows that a7 = a8 = 0
and therefore A3 = 0, which is a contradiction. Hence a2 = 0. Recall that we already know that
a6 = a9 = 0. Next, we show that a1 6= 0.
Indeed, assume the contrary: a1 = 0. Then the first
determinant simplifies to zx(a8y + a11z). Hence a8 = a11 = 0. Now the second determinant simplifies
to a4a10y3 + a5a10y2z − a7xy2. Hence a7 = 0 and a4a10 = a5a10 = 0. If a10 = 0, we have A3 = 0, which
is a contradiction. If a10 6= 0, we have a4 = a5 = 0. In this case the second column of each Aj is zero.
Thus the second basic vector in V1 is in the common kernel of all elements of S, which contradicts (L1).
These contradictions prove that a1 6= 0. By normalizing the second basic vector in V1 appropriately,
we can assume that a1 = −1. Taking this into account together with a2 = a6 = a9 = 0, we see that
the xy2 and xz2 coefficients in the first determinant are −a7 and a11 respectively. Hence a7 = a11 = 0.
Now the determinants in the above display simplify to
−a4a8y3 + (a8 + a10)xyz + a5a10y2z and a4a10y3 + a5a10y2z − (a8 + a10)xyz.
Since they vanish, a10 = −a8. If a8 = 0, then a10 = 0 and A3 = 0, which is a contradiction. Thus
a8 6= 0. Now vanishing of the polynomials in the above display implies a4 = a5 = 0. Hence, the matrices
A1, A2 and A3 acquire the shape
0 −1 0
0
1
0
0
0
0
,
0 0 −1
0
0 0
1 0
0
and a8
0 0
0
0 0 −1
0 1
0
with a8 6= 0
and S becomes the space of all antisymmetric matrices.
Since Lemma 3.3 is equivalent to Lemma 1.4, the proof of Lemma 1.4 is now complete.
Table 1: Algebras Aj for 1 6 j 6 11
j
1
2
3
4
5
6
7
8
9
defining relations of Aj
xx−zx, xy−zz, xz, yx, yy−zy, yz
xx, yx, yy, zx, zy, zz
xx−zx, xy, xz, yx, yy−zy, yz
xx, xy, xz−zz, yx, yy, yz−zz
xx−yx, xy, yy, yz, zx, zz
xz−yz, xy, yx, yy, zx, zz
xx, xy, xz, yy, zx, zy
xy, xz, yx, yz, zx, zy
xx, xz, yx, zx, zy, zz
other elements of
the Grobner basis
zzx, zzy, zzz
none
zzx, zzy
zzz
zyx
zyz
none
none
none
10 xx, xy, xz, yx, yz, zx
11 xx, xy, xz, yx, yy, zx
none
none
normal words
of degrees > 3
none
xyz
zn for n > 3
zzx, zzy
xzy, yxz, yxzy
yzy, xn for n > 3
zn, yzn−1 for n > 3
xn, yn, zn for n > 3
xyn−1, xyn−2z, yn,
yn−1z for n > 3
zmyn−m for n > 3,
0 6 m 6 n
all monomials in y, z
without yy as a subword
Hilbert
series
H1
H2
H3
H4
H5
H6
H7
H8
H9
H10
H11
4 Specific algebras satisfying HA = Hj
For each j ∈ {1, . . . , 11}, we provide a quadratic algebra Aj (generated by degree 1 elements x, y and
z) satisfying HAj = Hj. In each case the last equality is an easy exercise since we supply the finite
Grobner basis in the ideal of relations and describe the normal words. These 11 examples are presented
in Table 1.
Counting normal words is trivial for all Aj except the last one, where the normal words of degree
n > 3 are exactly monomials in z and y, which do not contain yy as a submonomial. In this case, as
it was already observed in the proof of Lemma 2.3, the numbers an of such monomials of degree n are
consecutive Fibonacci numbers with a3 = 5, yielding HA11 = H11.
Proof of Theorem 1.1. By Proposition 2.1, HA ∈ {H1, . . . , H11} for every quadratic K-algebra A satis-
fying dim A1 = dim A2 = 3. The examples in Table 1 show that each Hj with 1 6 j 6 11 is indeed the
Hilbert series of a quadratic K-algebra.
It remains to deal with Koszulity. We need the following elementary observation.
Lemma 4.1. Assume that A is a degree graded algebra on generators x1, . . . , xn, that the monomials in
xj carry a well-ordering compatible with the multiplication and that Λ is the set of the leading monomials
of all members of the corresponding Grobner basis of the ideal of relations of A. Let also 1 6 j, k 6 n
be such that xj 6= 0, xk 6= 0 and xjxk = 0 in A. Finally, assume that Λ contains no monomial ending
with xsxk for s 6= j. Then for u ∈ A, uxk = 0 ⇐⇒ u = vxj for some v ∈ A.
Proof. Since xjxk = 0 in A, uxk = vxjxk = 0 if u = vxj for some v ∈ A.
Assume now that u ∈ A and uxk = 0.
It remains to show that u = vxj for some v ∈ A. Let
N be the set of all normal words for A. That is, N is the set of monomials containing no member
of Λ as a submonomial. Since N is a linear basis in A, we can write u as a linear combination of
elements of N . We also separate those words in this combination ending with xj from the rest of them:
u = Pα
dβvβ, where both sums are finite (an empty sum is supposed to be zero), wαxj, vβ
are pairwise distinct normal words, none of vβ ends with xj and cα, dβ ∈ K∗. Then
cαwαxj + Pβ
0 = uxk = Pα
cαwαxjxk + Pβ
dβvβxk = Pβ
dβvβxk in A,
where the last equality is due to xjxk = 0. Since vβ does not end with xj and Λ contains neither xj nor
xk nor any monomial ending with xsxk with s 6= j, we easily see that each vβxk is a normal word. Since
the set of normal words is linearly independent in A, the above display implies that the sum Pβ
dβvβ is
empty and therefore u = Pα
Proposition 4.2. The algebras Aj with j ∈ {2, 5, 6, 7, 8, 9, 10, 11} are Koszul.
cαwαxj = vxj with v = Pα
cαwα.
Proof. For j ∈ {2, 7, 8, 9, 10, 11}, Aj is a monomial algebra, hence it is Koszul. It remains to verify
Koszulity of A5 and A6. Consider the algebra B given by the generators x, y, z and the relations xx+yx,
xz, zy and the algebra C given by the generators x, y, z and the relations xx, xz + yz, zy. A direct
computation shows that B! = A5 and C ! = A6. Hence Koszulity of A5 and A6 is equivalent to Koszulity
of B and C respectively. It remains to prove that B and C are Koszul, which is our objective now.
Consider the following sequences of free graded left B-modules and C-modules:
0 → B
δ5−→ C
d4−→ B2 d3−→ B3 d2−→ B3 d1−→ B
d0−→ K → 0,
δ4−→ C 2 δ3−→ C 3 δ2−→ C 3 δ1−→ C
δ0−→ K → 0,
. . .
δ5−→ C
δ5−→ C
(4.1)
(4.2)
where d0 and δ0 are the augmentation maps,
d1(u, v, w) = ux + vy + wz, d2(u, v, w) = (u(x + y), vz, wx), d3(u, v) = (0, ux, v(x + y)),
d4(u) = (u(x + y), 0), δ1(u, v, w) = ux + vy + wz, δ2(u, v, w) = (ux, vz, w(x + y)),
δ3(u, v) = (ux, v(x + y), 0), δ4(u) = (ux, 0) and δ5(u) = ux.
Using the relations of B and C, one easily sees that the composition of any two consecutive arrows
in both sequences is indeed zero. By definition of Koszulity, the proof will be complete if we show
that these sequences are exact. The exactness of (4.1) and (4.2) boils down to verifying the following
statements:
for u ∈ B, u(x + y) = 0 ⇐⇒ u = 0,
for u ∈ B, ux = 0 ⇐⇒ u = v(x + y) for some v ∈ B,
for u ∈ B, uz = 0 ⇐⇒ u = vx for some v ∈ B,
for u ∈ C, ux = 0 ⇐⇒ u = vx for some v ∈ C,
for u ∈ C, u(x + y) = 0 ⇐⇒ u = 0,
for u ∈ C, uz = 0 ⇐⇒ u = v(x + y) for some v ∈ C.
(4.3)
(4.4)
(4.5)
(4.6)
(4.7)
(4.8)
Indeed, the exactness of (4.1) at the leftmost B is equivalent to (4.3), its exactness at B2 is equivalent to
(4.3) and (4.4) and its exactness at the leftmost B3 is equivalent to (4.3), (4.4) and (4.5). The exactness
of (4.2) at each C which is to the left of C 2 is equivalent to (4.6), its exactness at C 2 is equivalent to
(4.6) and (4.7), while its exactness at the leftmost C 3 is equivalent to (4.6), (4.7) and (4.8). Checking
the exactness of both complexes at three terms on the right is a straightforward exercise. Alternatively,
one can notice that (4.1) and (4.2) are the Koszul complexes of B and C respectively, and that the
Koszul complex happens to be exact at three right terms for every quadratic algebra, see [23] (exactness
of the Koszul complex at the two right terms holds for every graded algebra generated in degree 1,
while the exactness at the term third from the right holds for all quadratic algebras).
Thus it remains to prove (4.3 - 4.8). Observe that the defining relations xx + yx, xz, zy of B
together with xykx + yk+1x, k > 1 form the Grobner basis of the ideal of relations of B, while the
defining relations xx, xz + yz, zy of C together with xyz form the Grobner basis of the ideal of
relations of C. Now a direct application of Lemma 4.1 justifies (4.5) and (4.6). In order to prove the
rest, we perform the following linear substitution. Keeping x and z as they were, we set the new y to
be x + y in the old variables. This substitution provides an isomorphism of B and the algebra A given
by the generators x, y, z and the relations xz, yx, zx − zy. These relations together with zyz form
the Grobner basis of the ideal of relations of A. The same substitution provides an isomorphism of C
and the algebra D given by the generators x, y, z and the relations xx, yz, zx − zy. These relations
together with zyx form the Grobner basis of the ideal of relations of D.
Now we can rewrite (4.3), (4.4), (4.7) and (4.8) in terms of multiplication in A and D. Namely, they
are equivalent to
for u ∈ A, uy = 0 ⇐⇒ u = 0,
for u ∈ A, ux = 0 ⇐⇒ u = vy for some v ∈ A,
for u ∈ D, uy = 0 ⇐⇒ u = 0,
for u ∈ D, uz = 0 ⇐⇒ u = vy for some v ∈ D,
(4.9)
(4.10)
(4.11)
(4.12)
respectively. Again, Lemma 4.1 justifies (4.10) and (4.12). Next, one easily sees that the sets of normal
words for both A and D are closed under the multiplication by y on the right. This implies (4.9) and
(4.11). Hence (4.3 - 4.8) hold and therefore B and C are Koszul.
Proposition 4.3. Let A be a quadratic algebra such that HA ∈ {H1, H3, H4}. Then A is non-Koszul.
Proof. Assume the contrary. Then A is Koszul and by (1.1), HA!(t) =
coefficients of the series
HA(−t) must be non-negative. On the other hand,
1
1
HA(−t) .
In particular, all
1
H1(−t) = 1+3t+6t3 +9t3 +9t4 −27t6 +· · · and
1
H3(−t) = 1+3t+6t3 +10t3 +14t4 +16t5 +12t6 −4t7 +· · ·
Hence H1 and H3 can not be Hilbert series of a Koszul algebra.
It remains to consider the case HA = H4. Since replacing the ground field by a field extension
does not effect the Hilbert series or Koszulity, we can without loss of generality assume that K is
algebraically closed. By Lemmas 1.4, 2.3 and 2.4, A = A(V, R) with R satisfying condition (P1) of
Lemma 1.4. Thus, by passing to the algebra with the opposite multiplication, if necessary, we can
assume that R ⊕ (L ⊗ V ) = V ⊗ V for a 1-dimensional subspace L of V . Now choose a basis x, y, z
in V such that x spans L. Then R⊥ ⊕ (M ⊗ V ) = V ⊗ V , where M = span {y, z}. It follows that we
can choose a basis f , g, h in R⊥ such that the leading monomials of f , g and h are xx, xy and xz
respectively. Since HA = H4 = 1 + 3t + 3t2 + 2t3, a direct computation shows that
HA! = 1
HA(−t) = 1 + 3t + 6t2 + 11t3 + 21t4 + 42t5 + 85t6 + · · ·
(we need few first coefficients). Since dim A!
3 = 11, there should be exactly one degree 3 element q of
the Grobner basis of the ideal of relations of A!. The leading monomial q of q can have either the shape
u1u2x or the shape u1u2u3, where uj ∈ {y, z}. First, assume that q = u1u2u3. Then A!
4 is spanned by
v1v2v3x with vj ∈ {y, z} and v1v2v3 6= u1u2u3 and by v1v2v3v4 with vj ∈ {y, z} and v1v2v3 6= u1u2u3,
v2v3v4 6= u1u2u3. The number of these monomials is 20 if u1 = u2 = u3 and is 19 otherwise. Thus
dim A!
4 = 21, we have arrived to a contradiction, which proves
that q can not be of the shape u1u2u3.
4 6 20. Since by the above display dim A!
4 would have been 22. Since dim A!
Hence q = u1u2x with u1, u2 ∈ {y, z}. In this case, were the relations f , g, h together with q is
the Grobner basis, the dimension of A!
4 = 21, there is exactly one
degree 4 element p of the Grobner basis of the ideal of relations of A!. The leading monomial p of p can
have either the shape w1w2w3x or the shape w1w2w3w4, where wj ∈ {y, z}. Again, first, assume that
p = w1w2w3w4. Then A!
5 is spanned by v1v2v3v4x with vj ∈ {y, z} and v3v4 6= u1u2 and by v1v2v3v4v5
with vj ∈ {y, z} and v1v2v3v4 6= w1w2w3w4, v2v3v4v5 6= w1w2w3w4. It easily follows that dim A!
5 6 41.
Since by the above display dim A!
5 = 42, we have arrived to a contradiction, which proves that p can
not be of the shape w1w2w3w4.
Hence p = w1w2w3x with wj ∈ {y, z} and w2w3 6= u1u2.
6 is spanned by 64 ele-
ments v1v2v3v4v5v6 with vj ∈ {y, z} and by 20 elements v1v2v3v4v5x with vj ∈ {y, z}, v4v5 6= u1u2,
v3v4v5 6= w1w2w3. Hence dim A!
6 = 85, we have arrived to a
contradiction. Thus H4 is not the Hilbert series of a Koszul algebra.
6 6 84. Since by the above display dim A!
In this case, A!
Proof of Theorem 1.2. By Proposition 4.2, for j ∈ {2, 5, 6, 7, 8, 9, 10, 11}, there is a Koszul algebra A
satisfying HA = Hj. By Proposition 4.3, every quadratic algebra A satisfying HA = Hj with j ∈ {1, 3, 4}
is non-Koszul.
4.1 Some remarks
1. The condition of K being infinite in Lemma 1.4 can be relaxed to K having sufficiently many
elements. More precisely, examining closely the idea behind the proof, one gets that Lemma 1.4 holds
if the condition of K being infinite is relaxed to K > 4. On the other hand, the following example
shows that the conclusion of Lemma 1.4 fails if K = 2.
Example 4.4. Let x, y, z be a basis of a 3-dimensional vector space V over the 2-element field K = Z2.
Let also R ⊂ V ⊗ V be the linear span of x ⊗ x, y ⊗ y, z ⊗ z, y ⊗ z + z ⊗ y, x ⊗ y + z ⊗ x + z ⊗ y and
x ⊗ z + y ⊗ x + z ⊗ y. Then R is a 6-dimensional subspace of V ⊗ V for which each of the conditions
(P1–P3) of Lemma 1.4 fails.
We leave the verification to the reader. It can be done by brute force since V + = V \ {0} has just
7 elements. For example, to show that (P1) fails, one has to find for every u ∈ V +, v, w ∈ V + such
that u ⊗ v, w ⊗ u ∈ R. Note that extending K to a 4-element field forces R from the above example to
satisfy (P1). We do not know whether the conclusion of Lemma 1.4 holds if K = 3.
2. By Proposition 2.5, A is automatically Koszul if HA = Hj for j ∈ {8, 9, 10, 11}. Note that if A is
a quadratic algebra satisfying HA = Hj with j ∈ {2, 5, 6, 7}, this does not necessarily mean that A is
Koszul. We construct the following examples to illustrate this. In these examples we assume K > 2
and α ∈ K is an arbitrary element different from 0 or 1. Table 2 is completed by computing the Grobner
bases of ideals of relations of algebras Bj.
Table 2: Algebras Bj for j ∈ {2, 5, 6, 7}
j
relations of Bj
2 xx+yz, xz, yx, yy+zx, zy, zz
5
xx−zx, xy−zx, yx−zx, yy−zx,
1−α zz, yz+αzx− 1
xz+αzx− 1
1−α zz
other elements of
the Grobner basis
yzx−zxy, xyz−zxy H2
HBj
zzx, zzzz
relations of B!
j
xx−yz, xy, yy−zx
zy, xz+yz+(1−α)zz,
xx+xy+yx+yy+zx+α(1−α)zz
xx + xz + 1
xx + αxz + zx, zy, zz
α zx, xy + zy, zz
H5
H6
H7
6 xx−αzx, xy−zy, yx, xz−αzx, yz, yy
7 xx−zx, xy, yx, xz−αzx, yz, yy
zzx, zzzy
zzx
Computing the Grobner bases of ideals of relations of algebras B!
j up to degree 4, we easily obtain
the data presented in Table 3.
(t) up to degree 4
Table 3: The series (HBj (−t))−1 and HB!
(HBj (−t))−1 up to t4
1 + 3t + 6t2 + 10t3 + 15t4 + · · ·
1 + 3t + 6t2 + 11t3 + 20t4 + · · ·
1 + 3t + 6t2 + 11t3 + 20t4 + · · ·
1 + 3t + 6t2 + 11t3 + 19t4 + · · ·
HB!
1 + 3t + 6t2 + 10t3 + 17t4 + · · ·
1 + 3t + 6t2 + 11t3 + 21t4 + · · ·
1 + 3t + 6t2 + 11t3 + 21t4 + · · ·
1 + 3t + 6t2 + 11t3 + 20t4 + · · ·
j
(t) up to t4
j
j
2
5
6
7
Table 3 ensures that each Bj fails to satisfy HBj (−t)HB!
j
(t) = 1 and therefore each Bj is non-Koszul.
Thus the following statement holds true.
Proposition 4.5. Assuming K > 2, for each j ∈ {2, 5, 6, 7}, there is a non-Koszul quadratic algebra
B satisfying HB = Hj.
3. By the duality formula (1.1), Theorem 1.2 implies the list of all Hilbert series of Koszul algebras
A satisfying dim A1 = 3 and dim A2 = 6. They are the series 1/H(−t) for H from the list specified in
Theorem 1.2. Thus we have the following corollary.
Corollary 4.6. For Koszul algebras A satisfying dim A1 = 3 and dim A2 = 6, the complete list of
Hilbert series consists of
Hj (−t) with j ∈ {2, 5, 6, 7, 8, 9, 10, 11}.
1
4. A quadratic algebra A satisfying HA(t)HA!(−t) = 1 is called numerically Koszul. There are examples
of numerically Koszul quadratic algebras which are not Koszul, see [23]. While proving Proposition 4.3,
we have actually shown that Hj with j ∈ {1, 3, 4} can not be the Hilbert series of a numerically Koszul
quadratic algebra. We do not know an answer to the following question.
Question 4.7. Let A be a numerically Koszul quadratic algebra satisfying dim A1 = dim A2 = 3. Is it
true that A is Koszul?
We are especially interested in the following particular case.
Question 4.8. 1 Let A be a quadratic algebra satisfying HA(t) = (1 − t)−3 and HA!(t) = (1 + t)3. Is
it true that A is Koszul?
Note that (1−t)−3 is the Hilbert series of K[x, y, z]. The following example shows that for a quadratic
algebra A, the equality HA = (1 − t)−3 alone does not guarantee numeric Koszulity.
Example 4.9. Let A be the quadratic algebra given by the generators x, y, z and the relations xx,
xz + yy + zx, xy + yx + zz. Then HA = (1 − t)−3, while HA! = H3. In particular, A is not numerically
Koszul and therefore is non-Koszul.
Proof. A direct computation shows that the defining relations of A together with yyz−zyy and yzz−zyy
form a Grobner basis for the ideal of relations of A. Now one easily sees that the normal words for A
are zk(yz)lymxε with k, l, m ∈ Z+ and ε ∈ {0, 1} and that the number of normal words of degree n is
(n+1)(n+2)
. Hence HA(t) = (1 − t)−3. The dual A! is given by the relations yz, zy, yy − zx, xz − zx,
xy − zz and yx − zz, which together with zxx, zzx and zzz form a Grobner basis for the ideal of
relations of A!. The only normal word of degree n > 3 is xn, which gives HA!(t) = H3.
2
5. The following question remains open.
Question 4.10. Which series feature as the Hilbert series of quadratic algebras satisfying dim A1 = 3
and dim A2 = 4? Which of these occur for Koszul A?
The answer to the above question would complete the list of Hilbert series of Koszul algebras A
satisfying dim A1 = 3. In [23], it is mentioned that it is unknown whether there is a Koszul algebra A
satisfying dim A1 = 3, dim A2 = 4 and dim A3 = 3. It is important to answer also because if such an
algebra exists, it would provide a counterexample to the conjecture on rationality of the Hilbert series
of Koszul modules over Koszul algebras. However if such an algebra does not exist, nothing can be
derived about rationality.
Acknowledgements:
We are grateful to IHES and MPIM for hospitality, support, and excellent research atmosphere. We
would like to thank anonymous referees for careful reading and useful comments. This work is funded
by the ERC grant 320974, and partially supported by the project PUT9038.
1In the meantime we have acquired an affirmative solution of this question
References
[1] D. Anick, Generic algebras and CW complexes, Algebraic topology and algebraic K-theory (Princeton, N.J.,
1983), 247–321, Ann. of Math. Stud. 113, Princeton Univ. Press, Princeton, NJ, 1987
[2] D. Anick, Noncommutative graded algebras and their Hilbert series, J. Algebra 78 (1982), 120–140
[3] M. Artin and W. Shelter, Graded algebras of global dimension 3, Adv. in Math. 66 (1987), 171-216.
[4] Artin, M.; Tate, J.; Van den Bergh, M. Modules over regular algebras of dimension 3 Invent.Math. 106
(1991), 335-388.
[5] Artin, M.; Tate, J.; Van den Bergh, M. Some algebras associated to automorphisms of elliptic curves. The
Grothendieck Festschrift, Vol. I, 33 - 85, Progr. Math., 86, Birkhauser Boston, Boston, MA, 1990.
[6] Bocklandt, Raf; Schedler, Travis; Wemyss, Michael Superpotentials and higher order derivations. J. Pure
Appl. Algebra 214 (2010), no. 9, 1501-1522.
[7] P. Cameron and N. Iyudu, Graphs of relations and Hilbert series, J. Symbolic Comput. 42 (2007), 1066–1078
[8] Alain Connes, Michel Dubois-Violette, Non commutative finite dimensional manifolds II. Moduli space and
structure of non commutative 3-spheres, Communications in Mathematical Physics 281 (2008), 23-127
[9] V. Drinfeld, On quadratic quasi-commutational relations in quasi-classical limit, Selecta Math. Sovietica 11
(1992), 317–326.
[10] Dubois-Violette, Michel Graded algebras and multilinear forms. C. R. Math. Acad. Sci. Paris 341 2005, no.
12, 719-724.
[11] Dubois-Violette, Michel Multilinear forms and graded algebras. J. Algebra 3172007, no. 1, 198-225.
[12] Ershov, M., Golod - Shafarevich groups: A survey. Int. J. Algebra Comput. 22(2012), N5, 1-68
[13] E. Golod and I. Shafarevich, On the class field tower (Russian), Izv. Akad. Nauk SSSR Ser. Mat. 28 (1964),
261–272.
[14] N. Iyudu and S. Shkarin, The Golod-Shafarevich inequality for Hilbert series of quadratic algebras and the
Anick conjecture, Proc. Roy. Soc. Edinburgh A141 (2011), 609–629
[15] N. Iyudu and S. Shkarin, Finite dimensional semigroup quadratic algebras with minimal number of relations,
Monatsh. Math. 168 (2012), 239–252
[16] N. Iyudu and S. Shkarin, Asymptotically optimal k-step nilpotency of quadratic algebras and the Fibonacci
numbers, Combinatorica [to appear]
[17] N. Iyudu and S. Shkarin, Optimal 5-step nilpotent quadratic algebras, J. Algebra 412 (2014), 1-14
[18] Kontsevich, Maxim, Formal (non)commutative symplectic geometry. The Gel'fand Mathematical Seminars,
1990-1992, 173-187, Birkhuser Boston, Boston, MA, 1993.
[19] Kontsevich, Maxim, Rosenberg, Alexander Noncommutative smooth spaces. arXivmath/9812158
[20] T. H. Lenagan and Agata Smoktunowicz, An infinite dimensional affine nil algebra with finite Gelfand-
Kirillov dimension, J. Amer. Math. Soc. 20 (2007), 989–1001.
[21] Manin Yu.I. Some remarks on Koszul algebras and quantum groups Annales de l'institut Fourier 37 (1987)
n.4, p.191-205
[22] Odesskii, A. V.; Feigin, B. L. Sklyanin's elliptic algebras. (Russian) Funktsional. Anal. i Prilozhen. 23 (1989),
no. 3, 45–54; translation in Funct. Anal. Appl. 23 (1990), no. 3, 207-214
[23] A. Polishchuk and L. Positselski, Quadratic algebras, University Lecture Series 37 American Mathematical
Society, Providence, RI, 2005
[24] Priddy, Stewart Koszul resolutions and the Steenrod algebra. Bull. Amer. Math. Soc. 76 (1970), no. 4,
834–839
[25] S.Sierra, D. Rogalski, J. T. Stafford, Noncommutative blowups of elliptic algebras Algebras and Representa-
tion Theory, 18 (2015), 491–529.
[26] Sklyanin, E. K. Some algebraic structures connected with the Yang-Baxter equation. Representations of a
quantum algebra. (Russian)Funktsional. Anal. i Prilozhen, 17 (1983), no. 4, 34-48.
[27] Zelmanov, E. Some open problems in the theory of infinite dimensional algebras, J. Korean Math. Soc.
44(2007), N5, 1185-1195.
Natalia Iyudu
School of Mathematics
The University of Edinburgh
James Clerk Maxwell Building
The King's Buildings
Peter Guthrie Tait Road
Edinburgh
Scotland EH9 3FD
E-mail address:
[email protected]
Stanislav Shkarin
Queens's University Belfast
Department of Pure Mathematics
University road, Belfast, BT7 1NN, UK
E-mail addresses:
[email protected]
|
1009.4506 | 1 | 1009 | 2010-09-23T00:26:40 | A Localization in MV-algebras | [
"math.RA"
] | In this document we consider a way of localizing an MV-algebra. Given any prime filter $F$ we find a local MV-algebra which has the same poset of prime filters as the poset of prime filters comparable to $F$. | math.RA | math | A LOCALIZATION IN MV-ALGEBRAS
COLIN G. BAILEY
Abstract. In this document we consider a way of localizing an MV-
algebra. Given any prime filter F we find a local MV-algebra which has
the same poset of prime filters as the poset of prime filters comparable
to F.
0
1
0
2
p
e
S
3
2
]
.
A
R
h
t
a
m
[
1
v
6
0
5
4
.
9
0
0
1
:
v
i
X
r
a
1. Introduction
A local MV-algebra is one with a single maximal implication filter. Such
algebras are of interest in the representation theory of MV-algebras (see [7]
for example).
The set of prime implication filters of an MV-algebra forms a spectral
root system, ordered by set-inclusion. The existence of a unique maximal
filter is equivalent to the stem of this root system being nonempty. (The stem
is the set Stem = (cid:8)P (cid:12)(cid:12)(cid:12)
P is a prime filter comparable to every other prime filter(cid:9).)
Whenever the stem is non-empty it has a least element, the Conrad filter (de-
fined below). This filter can be characterized in several ways, as we show in
section 2 below. This work is heavily based on work of Conrad on lattice-
ordered groups (see [5]), recasting his material in terms of implication filters
in MV-algebras.
In the last section we consider how to invert this characterization to get a
prime filter into the stem of an MV-algebra. This localization takes a prime
implication filter P and finds a quotient in which the maximal filter over P
is the unique maximal filter, and the prime filter structure of the quotient is
isomorphic to the set of prime filters comparable to P.
In most of what follows the filters are taken to be implication filters rather
than lattice filters. We recall that an implication filter is a lattice filter closed
under powers.
Date: November 5, 2018.
1
2
COLIN G. BAILEY
Given an MV-algebra L, there are several sets of filters that we are inter-
ested in:
P is a prime implication filter of L(cid:9)
= the prime spectrum;
P is a prime implication filter of L comparable to F(cid:9) ;
P is a minimal prime filter of L(cid:9)
= the minimal spectrum;
PSpec = (cid:8)P (cid:12)(cid:12)(cid:12)
PSpec(F) = (cid:8)P (cid:12)(cid:12)(cid:12)
µS = (cid:8)P (cid:12)(cid:12)(cid:12)
µS(F) = (cid:8)P (cid:12)(cid:12)(cid:12)
P is a minimal prime filter of L comparable to F(cid:9) .
Our notation usually follows that of [3] with the exception that we use ⊗
instead of ⊙.
2. Counits
Definition 2.1. u ∈ L is a counit iff u < 1 and there exists some v < 1 with
u ∨ v = 1.
Definition 2.2. The Conrad filter of an MV-algebra is the implication filter
generated by the counits.
We usually denote it by N (L) or N.
If N = N (L) then N is prime as a ∨ b = 1, a, b < 1 implies a and b are
counits and so in N.
All implication filters that contain N form a chain. The following lemma
provides an alternative characterization of the prime filters in this chain.
Lemma 2.3. Let P be a prime implication filter. Then P contains all counits
iff for all x < P and all p ∈ P p ≥ x.
Proof. Suppose that x < P and y ∈ P with x 6≤ y. We know that (x →
y) ∨ (y → x) = 1.
As x 6≤ y we have x → y < 1, and y 6≤ x implies y → x < 1 and so y → x
is a co-unit.
If it is in P then so is x ∧ y = (y → x) ⊗ y, contradicting x < P. Thus P
cannot contain all co-units.
Conversely if a is a co-unit and a ∨ b = 1 for some b > 0. One of a or b is
in P (as P is prime). If a < P then a ≤ b which is impossible, so a ∈ P. (cid:3)
A slight variation of this proof lets us see that filters are incomparable
because of counits.
Lemma 2.4. Let P and Q be incomparable implication filters. Then there
is a counit in Q \ P.
A LOCALIZATION IN MV-ALGEBRAS
3
Proof. Suppose not, ie every counit in Q is also in P. As P and Q are
incomparable we can find x ∈ Q \ P and y ∈ P \ Q. Thus x 6≤ y and y 6≤ x
and so x → y < 1 and y → x < 1 and (x → y) ∨ (y → x) = 1. So y → x
is a counit in Q and (by assumption) must be in P. As y ∈ P we now have
x ∧ y = y ⊗ (y → x) ∈ P contradicting x < P.
(cid:3)
The next two results show that N is actually the minimum prime filter
comparable to all prime filters.
Proposition 2.5. Let P be any prime implication filter that does not contain
all counits. Then there is a prime implication filter incomparable to P.
Proof. As P does not contain all counits we know that there is some g < P
that is not below P, ie there is some p ∈ P with p (cid:3) g. Of course g (cid:3) p.
Thus g → p < 1 and p → g < 1 and (g → p) ∨ (p → g) = 1.
As (p → g) ⊗ (p ∨ g) = g we must have p → g < P.
Let Q be maximal avoiding g → p. Then Q is prime and as (g → p) ∨
(p → g) = 1 ∈ Q we have p → g ∈ Q \ P. By construction g → p ∈ P \ Q
and so these two ideals are incomparable.
(cid:3)
Proposition 2.6. If P is a prime implication filter then either N ⊆ P or
P ⊆ N.
Proof. If P is not a subset of N then we can find p ∈ P \ N. p < N implies
p is below N and so N ⊆ [p, 1] ⊆ P.
(cid:3)
Thus N is the minimum prime implication filter comparable to all others.
The existence of such a filter implies that N is a proper filter, as if we have
a minimal prime implication filter F comparable to all others then it must
contain all counits -- by proposition 2.5 and so N exists and so F equals N.
Since any desired root system is the root system of an MV-algebra ([4]),
we see that it is possible to have non-trivial N.
Proposition 2.7. N is a minimal prime implication filter iff N = {1}.
Proof. The right to left implication is immediate.
If N is minimal then it is the unique minimal implication filter and so
must equal {1} -- as we know the intersection of all minimal implication
filters is {1}.
Or just notice that L embeds into Qm∈µS L/m = L/N is linearly ordered,
and so L is linearly ordered which implies N = {1}.
(cid:3)
We also note that if N is proper then there is a unique maximal implica-
tion filter -- the one that contains N. We also have the converse.
Proposition 2.8. If there is exactly one maximal proper implication filter
then it contains all counits.
4
COLIN G. BAILEY
Proof. Let M be the maximum implication filter. Let a, b < 1 with a∨b = 1.
Let Fb = {x x ∨ b = 1}. Then 0 < Fb, a ∈ Fb and it is easy to see that Fb is
a lattice filter. Also, x ∈ Fb implies xn ∨ b ≥ xn ∨ bn
= 1 and so
Fb is an implication filter. Hence a ∈ Fb ⊆ M.
= (x ∨ b)n
(cid:3)
Thus, if there is a maximum implication filter M then N ⊆ M and N is
proper.
3. Localization
Let P be a prime implication filter. We seek a quotient of L in which P
contains the Conrad filter. The construction we give below also preserves
the structure of PSpec(P).
Definition 3.1. Let P be a prime implication filter. Then
ℓ(P) = [[ {x → p p ∈ P and x < P} ]].
Because of lemma 2.4 we need to quotient out by at least ℓ(P) in order to
make P contain all counits in a quotient.
It is clear that ℓ(P) ⊆ P as x → p ≥ p for any p ∈ P. In general this
inclusion is strict, with the only exception being minimal prime filters.
Lemma 3.2. P is minimal prime iff ℓ(P) = P.
Proof. If P is minimal prime and p ∈ P then there is some t < P with
t ∨ p = 1. Therefore t → p = 1 → p = p ∈ ℓ(P).
If ℓ(P) = P and p ∈ P then p ≥ x → p′ for some x < P and p′ ∈ P.
Now p′ → x < P else we would have p′ ⊗ (p′ → x) = p′ ∧ x ∈ P and so
x ∈ P. Also p ∨ (p′ → x) ≥ (x → p′) ∨ (p′ → x) = 1. Thus P must be
minimal.
(cid:3)
The next few lemmas show the relationship of ℓ(P) to the minimal filters
below P.
Lemma 3.3. If m ⊆ P is minimal prime then ℓ(P) ⊆ m.
Proof. Let x < P and p ∈ P. Then p ⊗ (p → x) = p ∧ x implies p → x < P
and so is not in m. But (x → p) ∨ (p → x) = 1 ∈ m and m is prime, so
x → p ∈ m.
(cid:3)
Lemma 3.4. Let p ∈ P \ ℓ(P). Then there is some minimal prime filter
m ⊆ P with p < m.
Proof. Look in L/ℓ(P). Then [p] , 1 and is in P/ℓ(P). We also know that
the Conrad filter of L/ℓ(P) is contained in P/ℓ(P) -- since x < P and p ∈ P
implies x → p ∈ ℓ(P) and so x ≤ p mod ℓ(P). All minimal filters must
be subsets of the Conrad filter and so take M to be a minimal prime filter
A LOCALIZATION IN MV-ALGEBRAS
5
of L/ℓ(P) that avoids [p] < [1]. Then M ⊆ P/ℓ(P) and so the preimage M′
gives a prime subfilter of P that avoids p.
Any minimal filter of L contained in M′ works.
(cid:3)
Theorem 3.5.
ℓ(P) = \ {m m ∈ µS and m ⊆ P} .
Proof. By lemma 3.3 we know that LHS⊆RHS.
From lemma 3.4 we know that p <LHS implies p <RHS, i.e. RHS⊆LHS.
(cid:3)
We can now define the localization of an MV-algebra at a prime implica-
tion filter.
Definition 3.6. Let P be a prime implication filter of an MV-algebra L.
Then the localization of L at P is the MV-algebra L/ℓ(P).
If Q ⊆ P are two prime implication filters then we have {m m ∈ µS and m ⊆ Q} ⊆
{m m ∈ µS and m ⊆ P} and so ℓ(P) ⊆ ℓ(Q) (from the theorem). Hence
there is a natural MV-morphism L/ℓ(P) → L/ℓ(Q).
And finally a universal property of this localization.
We recall that if f : L → M is an MV-morphism then the shell of f is
sh( f ) = f −1[1] = {x f (x) = 1}
is an implication filter in L.
Theorem 3.7. Let P be any filter and f : L → M such that sh( f ) ⊆ P and
N (M) ⊆ f [P] ↑.
Then ℓ(P) ⊆ sh( f ).
Proof. Let x < P and p ∈ P.
f (x → p) = 1, i.e. x → p ∈ sh( f ).
If f (x) < f [P] then f (x) ≤ f (p) and so
If f (x) ∈ f [P] then for some p ∈ P we have x → p and p → x both
in the shell of f and hence in P. But then x ∧ p = p ⊗ (p → x) ∈ P --
contradiction.
(cid:3)
From the theorem we see that if f takes P to a filter containing all counits
then f factorizes through L/ℓ(P), and so, in some sense, L/ℓ(P) is the
largest quotient in which P contains all counits (or dominates its comple-
ment).
The assumption that sh( f ) ⊆ P is essential, else the theorem yields only
that the smaller set ℓ(P ∨ sh( f )) ⊆ sh( f ). Indeed if P, Q are incomparable
prime filters then N (L/Q) = {1} ⊆ P/Q but if q ∈ Q \ P and p ∈ P \ Q then
q → p ∈ ℓ(P) \ Q -- else p ∧ q = q ⊗ (q → p) ∈ Q, contradicting p < Q.
Lemma 3.8. Let F be a prime filter. Then ℓ(P) ⊆ F iff F is comparable to
P.
6
COLIN G. BAILEY
Proof. If P ⊆ F then ℓ(P) ⊆ P ⊆ F. If F ⊆ P then ℓ(P) ⊆ ℓ(F) ⊆ F.
Conversely, if ℓ(P) ⊆ F then F/ℓ(P) is prime in L/ℓ(P) and so compa-
rable to P/ℓ(P). Hence F = η−1[F/ℓ(P) is comparable to η−1[P/ℓ(P)] =
P.
(cid:3)
Theorem 3.9. PSpec(P) is order-isomorphic to PSpec(L/ℓ(P)).
Proof. We know that PSpec(L/ℓ(P)) is order-isomorphic to
F is a prime filter with ℓ(P) ⊆ F(cid:9) and from the lemma the latter set is
(cid:8)F (cid:12)(cid:12)(cid:12)
PSpec(P).
(cid:3)
References
[1] C. G. Bailey, The Prime Filters of an MV-Algebra, in preparation, available at
arXiv:0907.3328v1[math.RA].
[2] Dumitru Bus¸neag and Dana Piciu, Localization of MV -algebras and lu-groups, Alge-
bra Universalis 50 (2003), 359 -- 380.
[3] R. Cignoli, I. M. D'Ottaviano, and D. Mundici, Algebraic Foundations of Many-valued
Reasoning, Kluwer, 2000.
[4] R. Cignoli and A. Torrens, The Poset of Prime ℓ-ideals of an Abelian ℓ-group with a
Strong Unit, Journal of Algebra 184 (1996), 604 -- 612.
[5] Paul Conrad, The lattice of all convex ℓ-subgroups of a lattice-ordered group,
Czechoslovak Mathematical Journal 15 (1965), 101 -- 123.
[6]
, Some Structure Theorems for Lattice-Ordered Groups, Transactions of the
American Mathematical Society 99 (1961), 212 -- 240.
[7] A. Di Nola, I. Esposito, and B. Gerla, Local algebras in the representation of MV-
algebras, Algebra Universalis 56 (2007), 133 -- 164.
School of Mathematics, Statistics & Operations Research, Victoria University of
Wellington, Wellington, New Zealand,
E-mail address: [email protected]
|
1309.6170 | 3 | 1309 | 2015-06-18T21:34:53 | Graded cluster algebras | [
"math.RA",
"math.RT"
] | In the cluster algebra literature, the notion of a graded cluster algebra has been implicit since the origin of the subject. In this work, we wish to bring this aspect of cluster algebra theory to the foreground and promote its study.
We transfer a definition of Gekhtman, Shapiro and Vainshtein to the algebraic setting, yielding the notion of a multi-graded cluster algebra. We then study gradings for finite type cluster algebras without coefficients, giving a full classification.
Translating the definition suitably again, we obtain a notion of multi-grading for (generalised) cluster categories. This setting allows us to prove additional properties of graded cluster algebras in a wider range of cases. We also obtain interesting combinatorics---namely tropical frieze patterns---on the Auslander--Reiten quivers of the categories. | math.RA | math | Graded cluster algebras
Jan E. Grabowski†
Department of Mathematics and Statistics, Lancaster University,
Lancaster, LA1 4YF, United Kingdom
18th June 2015
Abstract
In the cluster algebra literature, the notion of a graded cluster algebra has been implicit
since the origin of the subject. In this work, we wish to bring this aspect of cluster algebra
theory to the foreground and promote its study.
We transfer a definition of Gekhtman, Shapiro and Vainshtein to the algebraic setting,
yielding the notion of a multi-graded cluster algebra. We then study gradings for finite type
cluster algebras without coefficients, giving a full classification.
Translating the definition suitably again, we obtain a notion of multi-grading for (gen-
eralised) cluster categories. This setting allows us to prove additional properties of graded
cluster algebras in a wider range of cases. We also obtain interesting combinatorics -- namely
tropical frieze patterns -- on the Auslander -- Reiten quivers of the categories.
MSC (2010): 13F60 (Primary), 18E30, 16G70 (Secondary)
5
1
0
2
n
u
J
8
1
]
.
A
R
h
t
a
m
[
3
v
0
7
1
6
.
9
0
3
1
:
v
i
X
r
a
Contents
1 Introduction
2 Preliminaries
3 Multi-graded seeds and cluster algebras
4 Gradings in finite type with no coefficients
5 Graded cluster categories
6 Tropical friezes
7 Homogenisation
2
3
4
7
13
16
20
†Email: [email protected]. Website: http://www.maths.lancs.ac.uk/~grabowsj/
1
1
Introduction
Gradings for cluster algebras have been introduced in various ways by a number of authors and
for a number of purposes. The evolution of the notion started with the foundational work of
Fomin and Zelevinsky ([12]), who consider Zn-gradings where n is precisely the rank of the cluster
algebra.
Gekhtman, Shapiro and Vainshtein have also given a definition of a multi-graded cluster
algebra more generally, in the dual language of toric actions ([14, Section 5.2]). In the case that
the underlying field F is R or C, they discuss a toric action of (F∗)r determined by a choice of
integer weight. Their Lemma 5.3 states a necessary and sufficient condition that a local toric
action on a seed extends to a global one on the associated cluster algebra.
Berenstein and Zelevinsky ([3, Definition 6.5]) have given a definition of graded quantum
seeds, which give rise to module gradings but not algebra gradings. Then in work with St´ephane
Launois ([15]), in which we proved that the quantum versions of homogeneous coordinate rings
of Grassmannians are quantum cluster algebras, we independently introduced the notion of a
Z-grading for a quantum cluster algebra and noted that the definition also applies to the classical
commutative case.
In the first part of this note, we briefly set out the definition of a Zd-graded seed and cluster
algebra, in the algebraic setting. Since the proofs reduce to the Z-graded case, given in detail
in this language in [15], we omit these. These definitions and results are equivalent to those
of Gekhtman -- Shapiro -- Vainshtein, though the toric action setting suggests a different set of
associated questions and we recommend Chapter 5 of [14] to readers interested in the more
geometrical aspects. We will concentrate only on algebraic and combinatorial aspects here.
We wish to promote the use of gradings in cluster algebra theory and to show that there
are interesting questions and especially combinatorial phenomena associated with gradings. To
do this, we consider the usual starting case of finite type cluster algebras without coefficients.
In this case, we can give a complete classification of the gradings that occur. In particular we
observe that the gradings we obtain are all balanced, that is, there is a bijection between the set
of variables of degree d and those of degree −d.
Next we introduce the notion of a graded (generalised) cluster category. The idea of the
definition is the same as previously: to associate an integer vector (the multi-degree) to an
object in the category in such a way that the vectors are additive on distinguished triangles
and transform naturally under mutation. This is done via the key fact that every object in a
generalised cluster category has a well-defined index; in order to satisfy the aforementioned two
properties, degrees are necessarily linear functions of the index.
In finite types, indices are very closely related to both dimension vectors and almost positive
roots -- it is not surprising that the combinatorics of cluster algebra gradings should be closely
related to the latter. However the theory of generalised cluster categories goes very far beyond
finite type.
The categorical approach has the advantage that it encapsulates the global cluster combi-
natorics, or more accurately the set of indices does. Another consequence is an explanation for
the observed balanced gradings in finite type: we show that the auto-equivalence of the cluster
category given by the shift functor induces an automorphism of the set of cluster variables that
reverses signs of degrees. Hence any cluster algebra admitting a cluster categorification necessar-
ily has all its gradings being balanced, for example finite type or, more generally, acyclic cluster
algebras having no coefficients.
Two of the highlights of the resulting analysis are firstly the close link between the gradings
on the cluster algebra and the representation theory encoded in the associated cluster category,
2
and secondly the emergence of a combinatorial pattern called a tropical frieze on the Auslander --
Reiten quiver of the cluster category, arising from the degrees of the cluster variables. For us,
this illustrates how deeply integrated into the theory of cluster algebras gradings are.
The structure of this paper is as follows. We begin with a brief exposition of the definition of
a (multi-)grading for a cluster algebra and some associated lemmas (Section 3). We then classify
gradings for coefficient-free cluster algebras of finite type (Section 4), using a result of Fomin
and Zelevinsky on Laurent expressions for cluster variables in finite type. (This encompasses all
finite type cases, not just the simply laced ones.)
We then turn to cluster categories and show that we can introduce the multi-gradings at
the categorical level (Section 5). As a consequence of the graded cluster category setting, we
obtain so-called tropical friezes on these categories and, when the generalized cluster category
comes from a derived category, on that derived category too. This is explained and illustrated
in Section 6.
We conclude with some remarks in Section 7 on how one may add coefficients to an initial
seed that does not admit a grading, so that the new seed does -- that is, how to homogenise a
cluster algebra.
Acknowledgements
The author would particularly like to thank St´ephane Launois for many helpful discussions
throughout the collaboration that this work originated from, Robert Marsh for useful references
regarding cluster categories and Thomas Booker-Price for permission to reproduce the results of
his calculations in Sections 4.2 and 4.3. We also thank numerous colleagues with whom prelimi-
nary versions of these results were discussed following their presentation at various institutions.
2 Preliminaries
The construction of a cluster algebra of geometric type from an initial seed (x, B), due to Fomin
and Zelevinsky ([12]), is now well-known. Here x is a transcendence base for a certain field
of fractions of a polynomial ring and B is a skew-symmetrizable integer matrix; in the skew-
symmetric case B is often replaced by the quiver Q = Q(B) it defines in the natural way. For
simplicity, we consider our base field to be Q.
We refer the reader who is unfamiliar with this construction to the survey of Keller ([19])
and the book of Gekhtman, Shapiro and Vainshtein ([14]) for an introduction to the topic and
summaries of the main related results in this area.
We set some notation for later use. For k a mutable index, set
(cid:88)
k = −ek − (cid:88)
k = −ek +
b+
b−
bik>0
bikei
and
bikei
bik<0
where the vector ei ∈ Zr (r being the number of rows of B) is the ith standard basis vector.
Note that the kth row of B may be recovered as Bk = b+
k − b−
k .
Then given a cluster x = (X1, . . . , Xr) and exchange matrix B, the exchange relation for
mutation in the direction k is given by
X(cid:48)
k = xb+
k + xb−
k
3
where for a = (a1, . . . , ar) we set
r(cid:89)
i=1
X ai
i
.
xa =
Later we will briefly discuss cluster algebras with coefficients (also called frozen variables).
That is, we designate some of the elements of the initial cluster to be mutable (i.e. we are allowed
to mutate these) and the remainder to be non-mutable. We will also talk about the corresponding
indices for the variables as being mutable or not; in [3] the former are referred to as exchangeable
indices. The rank of the cluster algebra is the number of mutable variables in a cluster; we will
refer to the total number of variables, mutable and not, as the cardinality of the cluster.
We will retain the usual convention that B will be a matrix with rows indexed by the initial
cluster variables and columns indexed by the mutable initial cluster variables. The matrix Bmut
obtained by taking only the rows of B corresponding to mutable variables is the principal part
of B.
3 Multi-graded seeds and cluster algebras
The natural definition for a multi-graded seed is as follows.
Definition 3.1. A multi-graded seed is a triple (x, B, G) such that
(a) (x = (X1, . . . , Xr), B) is a seed of cardinality r and
(b) G is an r × d integer matrix such that BT G = 0.
From now on, we use the term "graded" to encompass multi-graded; if the context is unclear,
we will say Zd-graded.
The above data defines deg
(Xi) = Gi (the ith row of G) and this can be extended to
rational expressions in the generators Xi in the obvious way. We also need to be able to mutate
our grading, which we do via the matrix E (denoted E+ in [3]) that encodes mutation of B:
G
Then we have that B(cid:48) = EBET . Setting G(cid:48) = ET G, it is straightforward to verify that the ith
row of G(cid:48) is given by
Note that we have that if Y ∈ x(cid:48), deg
and furthermore (B(cid:48))T G(cid:48) = 0 so that (x(cid:48), B(cid:48), G(cid:48)) is again a graded seed.
G(cid:48)(Y ) by definition: the degree of Y with
respect to G is precisely the kth row of G(cid:48), which is also the degree of Y viewed as an element
of the graded seed (x(cid:48), B(cid:48), G(cid:48)).
(Y ) = deg
G
Then we see that repeated mutation propagates a grading on an initial seed to every cluster
variable and hence to the associated cluster algebra, as every exchange relation is homogeneous.
Proposition 3.2. The cluster algebra A(x, B, G) associated to an initial graded seed (x, B, G),
with G an r × d integer matrix, is a Zd-graded algebra. Every cluster variable of A(x, B, G) is
homogeneous with respect to this grading.
4
δrs
Ers =
if s (cid:54)= k;
if r = s = k;
if r (cid:54)= s = k.
−1
max(0,−brk)
(cid:40)
if i (cid:54)= k
Gi
(b−
k )T G if i = k
.
G(cid:48)
i =
Remark 3.3. If we have a grading G, each column of G is itself a cluster algebra grading. Similarly,
every J ⊆ {1, . . . , d} gives rise to a ZJ-grading, by taking the submatrix of G on the column
subset J.
Also notice that d is independent of r: every cluster algebra admits a Zd-grading for any
d (cid:62) 1, namely taking G to be the r × d zero matrix. Equally, if H ∈ Zr defines a non-zero
Z-grading, so does H (d), the matrix with d columns all equal to H, again for any d.
Remark 3.4. From the definition of a grading, we see that the existence of a grading is a linear
algebra problem: if B has rank equal to the number of mutable indices1, the only solution is the
zero grading 0, assigning degree 0 to every cluster variable.
Classification of gradings for a particular B is also a linear algebra problem, of finding a basis
for the kernel in the case that the rank is not maximal. (Here, and below, "kernel" refers to the
kernel of the map of free abelian groups induced by B. We will also say e.g. dim ker B rather
than the more usual group-theoretic term "rank", to avoid further overloading that word.)
However it will in general be difficult to find the degrees of every cluster variable, especially
in infinite types. In finite types, one can reasonably expect to solve this problem and we will do
so in two ways, one algebraic and one categorical.
Remark 3.5. For some cluster algebra problems, the presence or absence of coefficients does not
play a large part and the phenomena seen are determined by the cluster algebra type. This
is not the case for gradings, though. As the examples later will illustrate, adding or removing
coefficients can radically change the gradings that can exist. This is to be expected: adding
coefficients increases the number of rows of the associated exchange matrix and this can impact
on the rank and hence the solutions G that are the grading vectors.
We conclude this section by recording some elementary results on a particular class of gradings
which, as we see, essentially contain information about every possible grading.
Definition 3.6. Let (x, B) be a seed. We call a multi-grading G whose columns are a basis for
the kernel of B a standard multi-grading, and call (x, B, G) a standard graded seed.
Lemma 3.7. Let Σ = (x, B, G) be a standard graded seed. Then any mutation of Σ, say
Σ(cid:48) = (x(cid:48), B(cid:48), G(cid:48)), is again a standard graded seed. Hence any graded seed that is mutation
equivalent to a standard graded seed is itself standard.
Proof: Recall that we have G(cid:48) = ET G and (B(cid:48))T G(cid:48) = 0. So the columns of G(cid:48) certainly belong
to ker B(cid:48) and since E is invertible, the column rank of G(cid:48) is equal to that of G. (Indeed E2 is
the identity, corresponding to matrix mutation being an involution.) Noting that mutation also
preserves rank, since B(cid:48) = EBET , so that dim ker B = dim ker B(cid:48), the columns of G(cid:48) form a
basis for ker B(cid:48) as required.
The final claim is immediate.
Lemma 3.8. Let (x, B, G) be a standard graded seed and let H be any grading for (x, B). Then
there exists an integer matrix M = M (G, H) such that for any cluster variable Y in A(x, B, H)
we have
deg
H
(Y ) = deg
(Y )M
G
where on the right-hand side we regard Y as a cluster variable of A(x, B, G) in the obvious way.
1That is, if the (row) rank of the matrix B equals the rank of the cluster algebra -- an unfortunate coincidence
of terminology.
5
Proof: Since G is standard, its columns are a basis for the kernel of B. Furthermore every
column of H belongs to ker B since H is a grading, and so there exists a matrix M encoding the
expression of the columns of H in the basis of columns of G, i.e. H = GM . Hence if Y is an
initial cluster variable, i.e. Y ∈ x, we have the result.
will follow by induction. Let X(cid:48)
k and let E be the associated matrix as above. Then
It then suffices to show that the result remains true under mutation, and the full statement
k = µk(x, B) be the mutation of the seed (x, B) in the direction
deg
H
(X(cid:48)
k) = (H(cid:48))k = (ET H)k = (ET GM )k = (ET G)kM = (G(cid:48))kM = deg
(X(cid:48)
k)M
G
as required.
Therefore, to describe the degree of a cluster variable of a graded cluster algebra A(x, B, H), it
suffices to know its degree with respect to some standard grading G and the matrix M = M (G, H)
transforming G to H.
In particular, to understand the distribution of the degrees of cluster
variables, it suffices to know this for standard gradings.
Since the lemma applies in the particular case when G and H are both standard, we see that
from one choice of basis for the kernel of B, we obtain complete information. For if we chose a
second basis, the change of basis matrix tells us how to transform the degrees. Hence up to a
change of basis, there is essentially only one standard grading for each seed.
Corollary 3.9. Let (x, B) and (y, C) be mutation equivalent seeds in a cluster algebra A and
denote by µ• a composition of (seed) mutations such that µ•(x, B) = (y, C). Let G be a standard
grading for (x, B) and H any grading for (y, C).
Then there exists an integer matrix M such that for every cluster variable Y of A we have
deg
H
(Y ) = deg
(Y )M.
G
Proof: Let E• be the product of the matrices E associated to µ•. Then E•BET• = C and
(y, C, ET• G) is a grading for the seed (y, C). Now (y, C, ET• G) is standard by Lemma 3.7 and so
we may apply Lemma 3.8 to (y, C, ET• G) and (y, C, H).
(Y ) is equal to deg
That is, there exists M such that deg
(Y )M . But as we noted earlier
(Y ): the degree of Y in the graded seed (y, C, ET• G) is by definition
deg
the degree of Y with respect to the grading propagated from the initial graded seed (x, B, G).
Hence we have that deg
(Y ) = deg
(Y )M as required.
E•G
(Y ) = deg
H
ET• G
G
H
G
Notice that M is mutation invariant: once we know the respective grading matrices H(cid:48) and
G(cid:48) for the same seed (after some mutations from (x, B, G) and (y, C, H)), M is easily calculated
from H(cid:48) = G(cid:48)M . This same M then compares the respective grading matrices at any seed, or
indeed we can compare gradings at different seeds via the matrix E•.
In the next section, our goal is to classify gradings in finite types (with no coefficients) in the
following sense: for an initial seed (x, B) of finite type, we find a standard grading and establish
the number of cluster variables in each degree for this G. The main consequence of the above
results is that the resulting distribution is essentially independent of the choices of seed and
standard grading.
6
4 Gradings in finite type with no coefficients
In finite types, we have not only the Laurent phenomenon but a rather stronger statement: for
certain choices of initial seed, every cluster variable is a Laurent expression in the variables from
that initial seed having a special form. This result, due to Fomin and Zelevinsky, is holds in
the case that the cluster algebra has coefficients but in this section, we will concentrate only on
cluster algebras of finite type without coefficients (for reasons that will become apparent later).
As such, we state a slightly simplified version of that theorem, for the no coefficients case.
Recall that to a square integer matrix B, we may associate a Cartan companion A = A(B),
by setting aii = 2 and aij = −bij if i (cid:54)= j. Then in the no coefficients case, given a square
skew-symmetrizable initial exchange matrix B, we have a Cartan companion associated to B
and so can associate to B the root system Φ of A(B). We lose no generality by assuming B
yields an irreducible root system, or equivalently that the Dynkin diagram associated to A(B)
is connected2
Theorem 4.1 ([13, Theorem 1.9]). Let (x, B) be an initial seed for a cluster algebra A = A(x, B)
of finite type, such that bijbik (cid:62) 0 for all i, j, k. Then there is a unique bijection α (cid:55)→ X[α] between
the almost positive roots Φ(cid:62)1 in Φ and the cluster variables in A such that, for each α ∈ Φ(cid:62)1,
the cluster variable X[α] is expressed in terms of the initial cluster x as
X[α] =
Pα(x)
xα ,
As before, for α =(cid:80)r
we have xα def= (cid:81)r
where Pα is a polynomial over Z with non-zero constant term. Under this bijection, X[−αi] = Xi.
i=1 aiαi (the expansion of α in the basis of simple roots {αi 1 (cid:54) i (cid:54) r})
.
We note that the condition on B, namely that bijbik (cid:62) 0, corresponds in the simply laced case
to choosing a quiver with a source-sink orientation on the underlying Dynkin diagram associated
to A(B). Such an orientation exists because A(B) is of finite type, but this condition is satisfied
by a larger class of matrices than just these. A seed with B satisfying this condition is called
bipartite.
i=1 X ai
i
It is well-known that all orientations of a Dynkin diagram are mutation equivalent but not
all orientations yield the particular form of Laurent polynomial in the above theorem: taking the
linear orientation of the Dynkin diagram of type A3, the exchange relation at the middle node
is of the form X2X(cid:48)
2 = X1 + X3 and we do not have a non-zero constant term.
But to classify gradings on a cluster algebra, we may choose any convenient initial seed and
so we choose a bipartite seed. The following is then immediate from the above theorem.
Corollary 4.2. Let A(x, B, G) be a graded cluster algebra of finite type such that the seed (x, B)
i=1 aiαi, we
is bipartite. Then for each α ∈ Φ(cid:62)0, expressed in the basis of simple roots as α =(cid:80)r
i=1
i=1
2This reduction is universal in the literature, if rarely explicit. It is justified by the observation that a finite
type cluster algebra associated to a reducible root system is the tensor product of the cluster algebras associated
to each irreducible component and when considering gradings we obtain complete information from studying the
constituent cluster algebras.
7
have
deg
G
(X[α]) = −αG = (− r(cid:88)
aiGi1, . . . ,− r(cid:88)
aiGid).
Proof: For α a negative simple root, the statement is immediate: the degree of Xi = X[−αi] is
by definition Gi (the ith row of G).
For the positive roots, we have from the theorem of Fomin -- Zelevinsky above that X[α] =
Pα(x)/xα. Then X[α] and xα are homogeneous with respect to G, the latter obviously so and the
former by the fundamental property of graded cluster algebras (Proposition 3.2). So it follows
that Pα(x) is homogeneous also and must in fact have degree 0 since it has non-zero constant
term.
(xα) = −αG, the negative sign arising because α is
expressed in terms of the simple positive roots, whereas Gi is the degree of the corresponding
negative simple root. Note that in particular X[αi] = −Gi = −X[−αi] = −deg
We deduce that deg
(X[α]) = deg
(Xi).
G
G
G
That is, in order to know the degree of a cluster variable in finite type with no coefficients,
we need only know the almost positive root α to which it is associated and the initial grading
G. Then the degree in question is simply a very natural linear function in these, −αG. In the
subsequent section, we will see a generalisation of this result, using cluster categories.
Furthermore, the results of the previous section show us that we may infer everything we
wish to know from a standard grading and that then we may in fact use any initial seed we like,
at the cost of altering the formula in the above Corollary by post-multiplication by some integer
matrix. Note though that in any case we certainly still obtain a linear expansion of the degrees
in terms of the components of the positive root α.
To complete our classification programme then, we calculate a basis for the kernel of an initial
exchange matrix B for some bipartite seed, form a standard multi-grading G and calculate −αG
for each positive root α, using the well-known description of the sets of positive roots as sums of
simple ones.
4.1 Type A
We take as initial cluster the set x = {X1, . . . , Xn} ⊆ Q(X1, . . . , Xn) and initial exchange quiver
Q as follows:
1
2
3
4
···
n
More precisely, we orient the Dynkin diagram of type An with every odd-numbered vertex being
a source and every even-numbered vertex a sink.
n − 1 if n is odd. Hence if n is even, the only grading is the zero grading.
The exchange matrix B(Q) associated to Q is easily seen to have rank n if n is even and rank
Assume now that n is odd. Then the kernel of B(Q) is spanned by the n × 1 vector G ∈ Zn
with
i=1 aiαi is a positive root expressed in the
1
−1
0
Gi =
if i ≡ 1 mod 4
if i ≡ 3 mod 4
otherwise
We deduce from our previous results that if α =(cid:80)n
2(cid:88)
(X[α]) = −αG =
deg
n+1
basis of simple roots, then
G
(−1)ja2j−1
j=1
(Since G induced a Z-grading, this is of course an integer rather than a vector.)
8
α =(cid:80)l
n odd:
Any positive root α in type An is a sum of consecutive simple roots with multiplicity 1, i.e.
j=1 αi+j−1. So we see that deg
G
(X[α]) ∈ {−1, 0, 1} for all α ∈ Φ(cid:62)0.
Indeed it is straightforward to calculate the following distribution of degrees in type An for
• the number of cluster variables of degree 1 is equal to (n+1)(n+3)
• the number of cluster variables of degree 0 is equal to (n−1)(n+3)
• the number of cluster variables of degree −1 is equal to (n+1)(n+3)
4
8
,
.
and
8
These counts accord with the total number of cluster variables being the number of almost
positive roots in a root system of type An, which is n2+3n
Let us say that a Zd-grading is balanced if for all degrees d ∈ Zd there is a bijection between
the variables of degree d and those of degree −d. (Note that this definition is valid in infinite
types also.)
2
.
Then we observe that every grading for a type An cluster algebra with no coefficients is
balanced. For even n, the zero grading is certainly balanced and for odd n, any initial grading
vector is an integer multiple of G and hence gives rise to a balanced grading. Subsequently, we
will explain why these gradings are balanced in terms of a property of the associated cluster
category.
4.2 Type B
In type B, we have the following Dynkin diagram and associated Cartan matrix:
•
•
•
•
•
•
2 −1
−1
0 −1
0
0
2 −1
0
2 −1
. . .
2 −1
−1
0 −2
0
2 −1
2
We choose an exchange matrix B whose Cartan companion is the above and is bipartite,
specifically:
0
1
0
0
−1 0 −1 0
1
0
1
0
. . .
1
0
0
−1 0 −1
0
0
2
in the case that n is odd, and the same but with the signs of the final rows reversed when n is
even.
grading) and rank n − 1 when n is odd.
Then this exchange matrix has rank n when n is even (and again we only have the zero
9
Assume now that n is odd. Then the kernel of BT is spanned by the n × 1 vector G ∈ Zn
with
Gi =
2
−2
1
0
if i ≡ 1 mod 4, i < n
if i ≡ 3 mod 4, i < n
if i = n
otherwise
As above, we may deduce a (linear) formula for the degree of an arbitrary cluster variable
and from this easily infer that the degrees in this case belong to the set {−2,−1, 0, 1, 2}. The
distribution of the degrees for this grading in type Bn (n odd) is ([4])
• the number of cluster variables of degree 2 is equal to (n+1)(n−1)
• the number of cluster variables of degree 1 is equal to n+1
2 ,
• the number of cluster variables of degree 0 is equal to (n+1)(n−1)
• the number of cluster variables of degree −1 is equal to n+1
2 and
• the number of cluster variables of degree −2 is equal to (n+1)(n−1)
4
4
,
,
.
4
These counts accord with the total number of cluster variables being the number of almost
positive roots in a root system of type Bn, which is n2 + n. We observe that this grading (and
hence all gradings in type B) is again balanced.
4.3 Type C
In type C, we have the following Dynkin diagram and associated Cartan matrix:
•
•
•
•
•
•
2 −1
−1
0 −1
0
0
2 −1
0
2 −1
. . .
2 −1
−1
0 −1
0
2 −2
2
We choose an exchange matrix B whose Cartan companion is the above and is bipartite,
specifically:
0
1
0
0
−1 0 −1 0
1
0
1
0
. . .
1
0
0
−1 0 −2
0
0
1
in the case that n is odd, and the same but with the signs of the final rows reversed when n is
even.
and rank n − 1 when n is odd.
Again this exchange matrix has rank n when n is even (so we only have the zero grading)
10
Ä
2
Ä
ä
2
2
ä
Assume now that n is odd. Then the kernel of BT is spanned by the n × 1 vector G ∈ Zn
with
Gi =
1
−1
1
0
if i ≡ 1 mod 4, i < n
if i ≡ 3 mod 4, i < n
if i = n
otherwise
As above, we easily infer a formula for the degree of an arbitrary cluster variable and that the
degrees in this case belong to the set {−1, 0, 1}. The distribution of the degrees for this grading
in type Cn (n odd) is ([4])
• the number of cluster variables of degree 1 is equal to
,
• the number of cluster variables of degree 0 is equal to (n+1)(n−1)
• the number of cluster variables of degree −1 is equal to
n+1
n+1
2
.
2
and
We observe that this grading is again balanced, and thus so are all gradings in type C.
4.4 Type D
We now turn our attention to type D, taking as our initial quiver
1
2
3
···
n − 2
n − 1
n
when n is even; the vertex n − 2 is instead a source for n odd. The corresponding exchange
matrix B has rank n − 1 if n is odd and rank n − 2 if n is even, so we have non-zero gradings in
all cases. Indeed, the even n case gives our first example of a Z2-grading.
For odd n, the kernel of BT is spanned by the vector G with
0
1
−1
Gi =
if i (cid:54) n − 2
if i = n − 1
if i = n
Clearly the degree of any cluster variable X[α] with α =(cid:80)n
i=1 aiαi is −αG = an − an−1. From
this and the well-known description of the positive roots in type D, we deduce that the degrees
in this case belong to the set {−1, 0, 1} and
• the number of cluster variables of degree 1 is equal to n,
• the number of cluster variables of degree 0 is equal to n(n − 2) and
• the number of cluster variables of degree −1 is equal to n.
11
−1
0
n
2
n
2
n
−1
0
1
Total
0
n2−2n
4
n2−2n
2
n
2
n
2
n2−2n
4
0
n2 − 2n n
1 Total
n2
4
n2
2
n2
4
n2
Table 1: Distribution of degrees for type Dn, n even: the (a, b)-entry of the table gives the
number of cluster variables of degree (a, b).
For even n, we know that it suffices to study the standard grading G. This case breaks down
further according to whether n is congruent to 0 or 2 modulo 4. If n ≡ 0 mod 4, G is given by
If n ≡ 2 mod 4, we have
Gi =
Gi =
(1, 0)
(0, 0)
(−1, 0)
(−1, 1)
(0,−1)
if i ≡ 1 mod 4, i < n − 1
if i even, i < n − 1
if i ≡ 3 mod 4, i < n − 1
if i = n − 1
if i = n
(1, 0)
(0, 0)
(−1, 0)
(1,−1)
(0, 1)
if i ≡ 1 mod 4, i < n − 1
if i even, i < n − 1
if i ≡ 3 mod 4, i < n − 1
if i = n − 1
if i = n
Analysing the resulting formula deg
(X[α]) = −αG, we see that the (multi-)degrees in this
case belong to the set ({−1, 0, 1}×{−1, 0, 1})\{(1, 1), (−1,−1)}. The corresponding distribution
of degrees is given in Table 1. We note that the distribution does not in fact depend on the
congruence of n modulo 4 and also that this bi-grading is also balanced: the number of variables
of degree d is equal to the number of degree −d.
G
4.5 Types E, F and G
One easily checks that the exchange matrices of type G2, F4, E6 and E8 have maximal rank, so
that the coefficient-free cluster algebras of these types admit no non-zero gradings.
However, exchange matrices of type E7 have rank 6, so we do have a grading in this case.
For the quiver
1
3
2
4
12
5
6
7
we have the grading
1
0
0
0
−1
0
1
By computer-aided calculation of the cluster variables in this case, we find that this grading
has
• 15 cluster variables in degree 1,
• 40 in degree 0 and
• 15 in degree −1.
This concludes our analysis of the finite type cases with no coefficients. We now turn our
attention to cluster categories and gradings on these, in order to mirror and extend the above
results in the categorical setting.
In particular we will explain the repeated occurrences of
balanced gradings.
5 Graded cluster categories
We wish to lift the notion of a multi-graded cluster algebra to the setting of (generalised) cluster
categories. Doing so, we obtain categorical versions of the results of the previous sections and
indeed gain further insight and generalisations. A significant advantage to working with a cluster
category is that it gives a global picture of the cluster combinatorics, which is particularly helpful
when examining gradings.
We make use of recent results of Dominguez and Geiss ([10]), generalising earlier work of
Caldero -- Chapoton ([7]), Palu ([20]) and others. The following constructions are in the skew-
symmetric and no coefficients settings but we are no longer assuming finite type. We recap the
necessary setup from [10] and adopt the conventions there.
Definition 5.1. Let K be an algebraically closed field. Let C be a triangulated 2-Calabi -- Yau
K-category with suspension functor Σ and let T ∈ C be a basic cluster-tilting object. We will
call the pair (C, T ) a generalised cluster category.
Following the nomenclature of Assem -- Dupont -- Schiffler ([1]), we might rather call (C, T ) a
generalised rooted cluster category, as the analogue of the initial seed is required to be part of
the data, but for brevity we shall not.
Write T = T1 ⊕ ··· ⊕ Tr. Setting Λ = EndC(T )op, the functor E = C(T,−) : C → Λ-mod
induces an equivalence C/add(T ) → Λ-mod. We may also define an exchange matrix associated
to T by
(BT )ij = dim Ext1
Λ(Si, Sj) − dim Ext1
Λ(Sj, Si)
Here the Si = EΣ−1Ti/rad EΣ−1Ti, i = 1, . . . , r are the simple Λ-modules.
(cid:33)
Then for each X ∈ C there exists a distinguished triangle
(cid:32) r(cid:77)
r(cid:77)
T m(i,X)
i
→ r(cid:77)
T p(i,X)
i
→ X → Σ
T m(i,X)
i
i=1
i=1
i=1
13
Define the index of X with respect to T , indT (X), to be the integer vector with indT (X)i =
p(i, X) − m(i, X). By [20, §2.1], indT (X) is well-defined and taking K = C we have a cluster
character
? : Obj(C) → Q[X±1
CT
X (cid:55)→ xindT (X)(cid:88)
1 , . . . , X±1
r
]
χ(Gre(EX))xBT ·e
e
Here Gre(EX) is the quiver Grassmannian of Λ-submodules of EX of dimension vector e and χ is
the topological Euler characteristic. We also use the same monomial notation (xa) as previously.
We note that this formula generalises that of Fomin -- Zelevinsky recalled in Theorem 4.1.
We also recall that for any cluster-tilting object U of C and for any Uk an indecomposable
k (cid:54)∼= Uk such that U∗ = (U/Uk)⊕U∗
summand of U , there exists a unique indecomposable object U∗
is again cluster-tilting and there exist non-split triangles
k
k → M → Uk → ΣU∗
U∗
k
and
Uk → M(cid:48) → U∗
k → ΣUk
with M, M(cid:48) ∈ add(U/Uk). In the generality of our setting, this is due to Iyama and Yoshino
([18]).
The obvious definition of a graded generalised cluster category is then the following.
Definition 5.2. Let (C, T ) be a generalised cluster category and let G be an r× d integer matrix
such that BT G = 0. We call the tuple (C, T, G) a graded generalised cluster category.
Definition 5.3. Let (C, T, G) be a graded generalised cluster category. For any X ∈ C, we define
deg
(X) = indT (X)G.
G
These definitions are justified by the following proposition.
Proposition 5.4. Let (C, T, G) be a graded generalised cluster category.
X ∈ Q[X±1
1 , . . . , X±1
r
] where the latter
(i) For all X ∈ C, deg
is Zd-graded by deg
(X) is equal to the degree of CT
(Xi) = Gi (the ith row of G).
G
G
(ii) For any distinguished triangle in C, X → Y → Z → ΣX, we have
deg
G
(Y ) = deg
(X) + deg
(Z)
G
G
(iii) The degree deg
is compatible with mutation in the sense that for every cluster-tilting object
U of C with indecomposable summand Uk we have
G
(U∗
k ) = deg
G
deg
G
(M ) − deg
(Uk) = deg
G
G
(M(cid:48)) − deg
(Uk)
G
where U∗
k , M and M(cid:48) are as in the above description of mutation in C.
Proof:
(i) For α = (a1, . . . , ar) ∈ Zr the degree of xα is (cid:80)r
(xBT ·e) = (BT · e)G = 0 since BT G = 0. It follows that
deg
G
i=1 aiGi = αG. Hence for each e,
deg
G
(CT
X ) = deg
G
(xindT (X)) = indT (X)G = deg
(X).
G
14
(ii) By [10, Proposition 2.5(b)], for any distinguished triangle in C, X α→ Y
indT (Y ) = indT (X) + indT (Z) + BT · dimΛ(ker Eα)
β→ Z
γ→ ΣX,
Multiplying by G on the right we immediately deduce that
since (BT · dimΛ(ker Eα))G = 0, similarly to (i).
G
G
deg
(Y ) = deg
(X) + deg
(Z)
G
(iii) This is immediate from the above description of mutation of cluster-tilting objects in C and
(M(cid:48)), which is the categorical version of the claim
(ii). We remark that deg
that all exchange relations in a graded cluster algebra are homogeneous.
(M ) = deg
G
G
Observe that the degree of an object X in a graded generalised cluster category is a linear
function of indT (X), namely indT (X)G. This is a generalisation of Corollary 4.2 to arbitrary
types, where the collection of vectors {indT (X) X ∈ C, X indecomposable} replaces the set
of almost positive roots. Indeed, we may deduce Corollary 4.2 from well-known properties of
cluster categories of finite type.
Given a triangulated category A, we may form its Grothendieck group K0(A) as the group
generated by isoclasses of objects, [A] for A ∈ A, modulo relations [X] − [Y ] + [Z] = 0 for every
distinguished triangle X → Y → Z → ΣX.
We have the following remarkable characterisation of gradings for generalised cluster cate-
gories:
Proposition 5.5. The space of gradings for a generalised cluster category (C, T ) may be identified
with the Grothendieck group K0(C).
Proof: Let T = proj Λ be the category of finitely generated projective Λ-modules. Then K0(T )
is free abelian on the basis {[Ti]}. By work of Palu ([21]), K0(C) is isomorphic to the quotient of
K0(T ) by all relations [M ] = [M(cid:48)] where
k → M → Uk → ΣU∗
U∗
Uk → M(cid:48) → U∗
k → ΣUk
and
k
are the non-split triangles associated to mutation of cluster-tilting objects in (C, T ). But the
latter is easily identified with Im BT . That is, K0(C) may be identified with Ker BT , the space
of gradings.
Notice that this implies, for example, that the Grothendieck group of the cluster category of
type A2m is trivial. Similarly K0(CA2m+1) ∼= Z. These calculations of the Grothendieck groups
in finite types had been found by Barot -- Kussin -- Lenzing ([2]).
However we obtain even more from the categorical approach. Recall that C has a suspension
(or shift) functor Σ that is an automorphism of C. This additional symmetry of C induces the
following property of deg
Lemma 5.6. For each X ∈ C, deg
That is, for each d ∈ Zd, Σ induces a bijection between the objects of C of degree d and those
(ΣX) = −deg
(X).
G
G
.
G
of degree −d.
Proof: By [10, Proposition 2.5(a)] we have that
−BT · dimΛ(EX) = indT (X) + indT (ΣX)
from which the claim follows similarly to above, by post-multiplication by G.
15
1 , . . . , X±1
r
1 , . . . , X±1
Note that as a consequence any object X for which Σ2m+1X = X must have degree 0.
Let us say that a cluster algebra A = A(Q) ⊆ Q[X±1
] arising from a quiver
Q (with no coefficients) admits a cluster categorification (C(A), T ) if the cluster character
? : C(A) → Q[X±1
] is a bijection between indecomposable objects of C(A) and the
CT
cluster variables of A. In particular, by work of Palu ([20]), if Q is a finite connected3 acyclic
quiver (that is, Q is mutation equivalent to a quiver having no oriented cycles) then its associated
cluster algebra A(Q) admits a cluster categorification, with C(A) = CQ the usual cluster category
introduced in [5].
Corollary 5.7. Let A(Q) be a cluster algebra admitting a cluster categorification. Then every
grading for A(Q) is balanced.
r
In particular, every grading in types A, D and E is balanced. For type B, Buan, Marsh and
Vatne ([6]) have introduced a categorification, the cluster category associated to a tube, so this
case is also explained. Since by [13], the cluster variables in type Bn−1 and Cn−1 are both in
bijection with centrally-symmetric pairs of diagonals of a regular 2n-gon and the aforementioned
work of [6] goes via this bijection also, we see that type C is covered also. That is, for every
exchange matrix of finite type, every grading is balanced, explaining our previous observations
of this fact.
6 Tropical friezes
We recall the notion of a frieze pattern, as introduced by Conway and Coxeter ([9], [8]). A frieze
pattern of order n consists of n − 1 infinite rows of positive integers, with the first and last rows
consisting of only the integer 1 and such that in a diamond of adjacent integers
a
b
c
d
we have ad − bc = 1. The latter property is called the unimodular rule.
Following Fock -- Goncharov ([11]) and Propp ([22]), Guo ([16]) has studied tropical friezes on
generalised cluster categories. A tropical frieze on C is a map f : C → Z which is constant on iso-
morphism classes, additive with respect to direct sums and where the unimodular rule is replaced
by a + d = max(b + c, 0). More precisely, for all objects U and V of C with dim Ext1C(U, V ) = 1,
f (U ) + f (V ) = max{f (M ), f (M(cid:48))} where we have non-split triangles
U → M → V → ΣU
and
V → M(cid:48) → U → ΣV
Note that it is clear how to extend this definition to that of a "multi-frieze", that is when f takes
values in Zd for d (cid:62) 1. We will continue to use the term "frieze" to encompass multi-friezes also.
In [16], Guo shows that the sum of two tropical friezes need not be again a tropical frieze and
gives a necessary and sufficient condition for this to be the case. Furthermore, she shows that if
Q is a Dynkin quiver with n vertices and if C = CQ is its cluster category, then tropical friezes
on C are in bijection with elements of Zn, by showing that each tropical frieze is determined by
its values on the indecomposable summands of a basic cluster-tilting object and all choices are
permitted.
3This restriction is again mild, similarly to before.
16
1
2
3
4
5
Σ(
5 )
4
2
3
1
2
4
3
5
4
Σ(
4 )
3
2
4
1
5
3
4
2
3
Σ( 2
3
4 )
4
3
4
2
5
1
3
2
Σ( 2
)
5
4
3
2
1
Σ(1
2
)
Figure 1: The Auslander -- Reiten quiver for the cluster category of type A5. Each quiver repre-
sentation is given in terms of its Loewy series, writing just "i" for the simple module Si. (Note
that the left- and right-hand edges are to be identified at the dotted line, with a twist.)
The connection with the previous section of this work is as follows. Let us say that a tropical
frieze f is exact if for all objects U and V with dim Ext1C(U, V ) = 1 we have f (U ) + f (V ) =
f (M ) = f (M(cid:48)) where we have non-split triangles involving U , V , M and M(cid:48) as above.
: Obj(C) → Zd
Lemma 6.1. Let (C, T, G) be a graded generalised cluster category. Then deg
is an exact tropical frieze on C. Conversely, if f is an exact tropical frieze on (C, T ) then
(C, T, (f (T1), . . . , f (Tr))) is a graded generalised cluster category.
G
Proof: The first claim is immediate from Proposition 5.4. The converse follows from the result
of [10] recalled in part (ii) of the proof of that Proposition.
We note that exact tropical friezes are better behaved than tropical friezes generally:
in
particular, the sum of any two exact tropical friezes is again a tropical frieze. Furthermore the
exact tropical friezes are classified by means of linear algebra, by finding a basis for the kernel
of the associated exchange matrix BT . We note that Guo has classified all tropical friezes when
Q is Dynkin ([16, Theorem 5.1]) in categorical terms; the classification for exact tropical friezes
is elementary but this applies to a restricted class of friezes.
We conclude this section with two examples of exact tropical friezes that illustrate the re-
sults of this and the two previous sections in types A5 and D4.
In both cases we give the
Auslander -- Reiten quiver of the associated cluster category (Figures 1 and 3) and the grading --
or equivalently exact tropical frieze -- corresponding to the choices of bipartite quiver Q and
grading G in Sections 4.1 and 4.4 respectively (Figures 2 and 4). We see not only the tropical
unimodular rule in the additivity on meshes but also the sign-changing bijection induced by Σ
(which is perhaps more familiar as the Auslander -- Reiten translation τ ).
We have concentrated on the cluster category, for obvious reasons. However we observe
that the combinatorics described above can be extended to the bounded derived category DQ
associated to the category of finite dimensional CQ-modules. By work of Happel ([17]), the full
17
0
0
−1
1
−1
0
0
1
−1
1
0
0
−1
1
−1
0
0
1
−1
1
Figure 2: The Auslander -- Reiten quiver for the cluster category of type A5 with degrees replacing
modules.
2
3
2
1 2
2
4
1 22 3
4
1 2
2
4
3
4
3
1 2
1
1 2
3
4
3
4
Σ( 2
)
3
Σ( 2
Σ(1 2
)
)
Σ( 2
)
4
Figure 3: The Auslander -- Reiten quiver for the cluster category of type D4. As usual, we have
arranged the diagram so that modules in the same orbit under the shift are aligned on the same
horizontal line. (Note that the left- and right-hand edges are to be identified, in the sense that
the arrows at the right edge should be regarded as pointing to the representation on the far left.)
(1,−1)
(−1, 0)
(−1, 1)
(1, 0)
(1,−1)
(−1, 0)
(−1, 1)
(1, 0)
(0, 0)
(0, 0)
(0, 0)
(0, 0)
(0, 1)
(0,−1)
(0, 1)
(0,−1)
Figure 4: The Auslander -- Reiten quiver for the cluster category of type D4, with bi-degrees
corresponding to the grading bivector given in Section 4.4.
18
Σ−1(
5 )
1
2
3
4
5
Σ(
5 )
4
3
Σ( 2
)
Σ( 1
)
Σ−1( 3
5)
4
2
1
3
4
2
5
3
4
Σ(
4 )
Σ( 2
3
5)
4
Σ( 1
3
2
)
Σ2( 2
)
···
Σ−1( 3
)
3
2
4
1
5
3
4
2
3
Σ( 2
3
4 )
Σ( 1
3
5)
4
2
Σ(
3
)
···
Σ−1(1
3
2
)
4
3
2
4
5
3
1
2
Σ( 2
)
Σ( 1
3
2
4 )
Σ(
3
5)
4
Σ2(
4 )
Σ−1(1
)
5
4
3
2
1
Σ( 1
2
)
Σ(
3
4 )
Σ(
5)
Figure 5: Part of the Auslander -- Reiten quiver for the bounded derived category of type A5.
(The morphisms going between the shifts of the CA5-module category are indicated by dashed
lines, to highlight the repetitive structure.)
···
0
0
1
−1
1
0
0
−1
1
−1
0
0
1
−1
1
0
0
−1
1
−1
0
0
1
−1
1
0
0
−1
1
−1
0
0
1
−1
1
0
0
···
Figure 6: Part of the Auslander -- Reiten quiver for the bounded derived category of type A5, with
degrees replacing modules.
subcategory of indecomposable objects of DQ is equivalent to the mesh category of the repetition
quiver ZQ (see [19, Section 5] for detailed definitions). The Auslander -- Reiten quiver of DQ then
takes the form of an infinite strip and in the case of type A, we may describe it as the quiver given
by taking the finite strip corresponding to CAn-modules repeated and reflected, as in Figure 5
for type A5.
Each repeated segment corresponds to an application of the suspension (or shift) functor of
the mesh category, which we also denote by Σ. The cluster category is constructed from the
derived category by taking a certain quotient and inherits its own shift functor. In the case at
hand, this construction yields the Mobius strip of Figure 1.
We notice that if n is odd, when we have the non-zero grading, the degree pattern above
extends to give an exact tropical frieze pattern on the derived category -- the shift functor reverses
the parity of the degree, so it suffices to know the frieze pattern on CQ. This is illustrated for
n = 5 in Figure 6.
We recall that we have no non-zero grading in the case of An for n even. We observe that
the derived category does admit a tropical frieze pattern similarly to the odd case, but this does
not descend to the cluster category when n is even: going from the derived category in type A4
3 ) being given degrees −1 and 0, an
to the cluster category would see the objects 4 and Σ( 2
inconsistency since these objects are isomorphic in the cluster category.
19
7 Homogenisation
Given an initial seed (x = (X1, . . . , Xr), B) and an arbitrary r × d integer matrix G, we will not
typically have a multi-graded seed. However, we can make a modification to the initial data in
order to homogenise this and produce a related cluster algebra of the same cluster algebra type
that is multi-graded.
Lemma 7.1. Let A = A(x, B) be a cluster algebra and let G be an r × d integer matrix. Then
there exists a multi-graded cluster algebra structure on a subalgebra Ahom of the field of rational
functions Q(X1, . . . , Xr, h1, . . . , hd) in which the elements hi are coefficients. Furthermore the
quotient cluster algebra obtained by setting hi = 1 for all i is isomorphic (as a cluster algebra)
to A.
Proof: We claim that the following is valid initial data for a multi-graded cluster algebra struc-
ture:
Ä
• Ghom =(cid:0) G
I
We have
• xhom = (X1, . . . , Xr, h1, . . . , hd);
• Bhom =
B−GT B
ä
(cid:1), with I = Id×d the identity matrix.
(Bhom)T Ghom = ( BT (−GT B)T )(cid:0) G
;
I
Here we are using the natural block matrix notation.
(cid:1) = BT G − BT G = 0
so that we have a multi-graded seed. Hence we may construct Ahom within the field of rational
functions Q(X1, . . . , Xr, h1, . . . , hd).
It is clear that taking the quotient setting all hi to 1 recovers A.
Example 7.2. A simple example is as follows, observing that we could fix the lack of a grading
in type An for even n. Let us take the quiver An, n even, with a linear orientation
1
2
3
4
···
n
We can easily check that this quiver admits no non-zero grading. Let G = (0 1 0 1 ··· 0 1)T .
Following the above homogenisation procedure, we see that we should add one coefficient corre-
sponding to an additional vertex 0 to the quiver to give the ice quiver
0
1
2
3
4
···
n
This admits the grading Ghom = (1 0 1 0 1 ··· 0 1)T (where we use the natural ordering, writing
the degree at vertex 0 first, rather than at the end as per the definition of Ghom). We obtain a
(graded) cluster algebra A(cid:48) = A(xhom, Bhom, Ghom) of type An.
Remark 7.3. In general, this does not yield a multi-graded cluster algebra structure on the
polynomial extension A[h1, . . . , hd] of A, since the new coefficients are involved in the exchange
relations. (Clearly it does if G was in fact a multi-grading to begin with, for then we are simply
adding d degree 0 disconnected coefficients. But then we have no need to homogenise.)
20
There is an second, equally natural, way to homogenise a seed. We may take principal
coefficients, that is, extend B by the n × n identity matrix (or indeed any diagonal matrix with
diagonal entries ±1, as we prefer), adding one row per mutable variable. Then we simply set the
degree of the ith new variable to be whatever is required to homogenise at the ith position --
namely the sum of the degrees over arrows leaving the ith vertex minus the sum of the degrees
over arrows entering that vertex, if we are in the quiver setting. This corrects inhomogeneity
one mutable variable at a time, whereas the above lemma fixes inhomogeneity one coordinate of
the multi-degree at a time. Each construction might be appropriate in certain circumstances.
Example 7.4. We extend the cluster algebra without coefficients of type A2 by adding principal
coefficients, so that the initial exchange quiver becomes
1
1(cid:48)
2
2(cid:48)
This quiver admits non-zero gradings, in contrast to type A2 with no coefficients. The space of
solutions to BkG = 0 for mutable indices k is 2-dimensional, with basis
{G = (1, 0, 0,−1), H = (0, 1, 1, 0)}.
The associated standard Z2-grading is represented by the following diagram:
(1, 0)
(0, 1)
(0, 1)
(−1, 0)
Note that the frozen vertices 1(cid:48) and 2(cid:48) are exempted from the condition that the sums of the
degrees at arrows entering and at arrows leaving the vertex are equal -- we only require this at
mutable vertices.
Let us take as our initial cluster (X1, X2, X3, X4) (writing X3 for X1(cid:48) etc., for clarity). Then
the cluster variables and their degrees for the standard grading (G, H), the two gradings G and
H individually and their sum G + H are as follows.
X1
X2
X3
(G, H)
(1, 0)
(0, 1)
(0, 1)
G
H
G + H
1
0
1
0
1
1
0
1
1
X4
(−1, 0)
−1
0
−1
X2+X3
X1
(−1, 1)
−1
1
0
X1X4+1
X2
(0,−1)
0
−1
−1
X2+X3+X1X3X4
X1X2
(−1, 0)
−1
0
−1
While we do again have the property of variables being concentrated in degrees −1, 0 and 1
for both gradings, as we had previously in type A, none of (G, H), G or H is balanced. There
does exist a balanced grading however, namely G + H.
21
We conclude by observing that in both homogenisation constructions, i.e. that of Lemma 7.1
and by taking principal coefficients, we may extend any choice of r× d matrix G to a grading. So
once one fixes a coefficient pattern, the space of associated gradings is fixed but if one varies the
coefficient pattern, one may obtain any grading one wants. This justifies our earlier comment that
properties of graded cluster algebras, e.g. classification, depend strongly on fixing a particular
coefficient pattern, as opposed to being determined by the cluster algebra type alone.
References
[1] Ibrahim Assem, Gr´egoire Dupont, and Ralf Schiffler, On a category of cluster algebras, J. Pure Appl.
Algebra 218 (2014), no. 3, 553 -- 582. MR 3124219
[2] M. Barot, D. Kussin, and H. Lenzing, The Grothendieck group of a cluster category, J. Pure Appl.
Algebra 212 (2008), no. 1, 33 -- 46. MR 2355032 (2008j:18010)
[3] Arkady Berenstein and Andrei Zelevinsky, Quantum cluster algebras, Adv. Math. 195 (2005), no. 2,
405 -- 455.
[4] Thomas Booker-Price, Personal communication, 2014.
[5] Aslak Bakke Buan, Robert Marsh, Markus Reineke, Idun Reiten, and Gordana Todorov, Tilting
theory and cluster combinatorics, Adv. Math. 204 (2006), no. 2, 572 -- 618.
[6] Aslak Bakke Buan, Robert J. Marsh, and Dagfinn F. Vatne, Cluster structures from 2-Calabi-Yau
categories with loops, Math. Z. 265 (2010), no. 4, 951 -- 970. MR 2652543 (2011h:18014)
[7] Philippe Caldero and Fr´ed´eric Chapoton, Cluster algebras as Hall algebras of quiver representations,
Comment. Math. Helv. 81 (2006), no. 3, 595 -- 616. MR 2250855 (2008b:16015)
[8] J. H. Conway and H. S. M. Coxeter, Triangulated polygons and frieze patterns, Math. Gaz. 57 (1973),
no. 400, 87 -- 94. MR 0461269 (57 #1254)
[9] H. S. M. Coxeter, Frieze patterns, Acta Arith. 18 (1971), 297 -- 310. MR 0286771 (44 #3980)
[10] Salom´on Dominguez and Christof Geiss, A Caldero-Chapoton formula for generalized cluster cate-
gories, J. Algebra 399 (2014), 887 -- 893. MR 3144617
[11] Vladimir V. Fock and Alexander B. Goncharov, Cluster ensembles, quantization and the dilogarithm,
Ann. Sci. ´Ec. Norm. Sup´er. (4) 42 (2009), no. 6, 865 -- 930. MR 2567745 (2011f:53202)
[12] Sergey Fomin and Andrei Zelevinsky, Cluster algebras. I. Foundations, J. Amer. Math. Soc. 15
(2002), no. 2, 497 -- 529 (electronic).
[13]
, Cluster algebras. II. Finite type classification, Invent. Math. 154 (2003), no. 1, 63 -- 121.
[14] Michael Gekhtman, Michael Shapiro, and Alek Vainshtein, Cluster algebras and Poisson geometry,
Mathematical Surveys and Monographs, vol. 167, American Mathematical Society, Providence, RI,
2010. MR 2683456 (2011k:13037)
[15] Jan E. Grabowski and St´ephane Launois, Graded quantum cluster algebras and an application to
quantum Grassmannians, preprint, 2013.
[16] Lingyan Guo, On tropical friezes associated with Dynkin diagrams, preprint, 2012.
[17] Dieter Happel, On the derived category of a finite-dimensional algebra, Comment. Math. Helv. 62
(1987), no. 3, 339 -- 389. MR 910167 (89c:16029)
[18] Osamu Iyama and Yuji Yoshino, Mutation in triangulated categories and rigid Cohen-Macaulay mod-
ules, Invent. Math. 172 (2008), no. 1, 117 -- 168. MR 2385669 (2008k:16028)
[19] Bernhard Keller, Cluster algebras, quiver representations and triangulated categories, Triangulated
categories, London Math. Soc. Lecture Note Ser., vol. 375, Cambridge Univ. Press, Cambridge, 2010,
pp. 76 -- 160. MR 2681708 (2011h:13033)
22
[20] Yann Palu, Cluster characters for 2-Calabi-Yau triangulated categories, Ann. Inst. Fourier (Grenoble)
58 (2008), no. 6, 2221 -- 2248. MR 2473635 (2009k:18013)
[21]
, Grothendieck group and generalized mutation rule for 2-Calabi-Yau triangulated categories,
J. Pure Appl. Algebra 213 (2009), no. 7, 1438 -- 1449. MR 2497588 (2010i:18014)
[22] James Propp, The combinatorics of frieze patterns and Markoff numbers, preprint, 2005.
23
|
1610.00088 | 2 | 1610 | 2017-06-28T19:29:29 | Malcev algebras corresponding to smooth Almost Left Automorphic Moufang Loops | [
"math.RA"
] | In this note we introduce the concept of an almost left automorphic Moufang loop and study the properties of tangent algebras of smooth loops of this class. | math.RA | math |
Malcev algebras corresponding to smooth almost
left automorphic Moufang loops
Ramiro Carrillo-Catal´an
CONACyT - Universidad Pedag´ogica Nacional
Unidad 201 Oaxaca
[email protected],
Marina Rasskazova
Omsk Technic University
Omsk,644050, pr.Mira 15
[email protected],
Liudmila Sabinina
Centro de Investigacion en Ciencias
UAEM, Cuernavaca
[email protected]
June 27, 2017
Abstract
In this note we introduce the concept of an almost left automorphic
Moufang loop and study the properties of tangent algebras of smooth
loops of this class.
Key words: Malcev algebras, Moufang loops, Alternative loops, Au-
tomorphic loops, Local almost left automorphic Moufang loops.
2010 Mathematics Subject Classification: 17D10, 20N05.
1 Preliminaries
In what follows, a loop structure is a universal algebra hQ, ·, \, /, 1i of type
(3, 0, 1) such that the identities
(x · y)/y = x = (x/y) · y,
x · (x\y) = y = x\(x · y),
x · 1 = x = 1 · x.
are satisfied for any two elements in the loop. For simplicity we will write xy
instead of x·y. A Moufang loop is a loop in which any of the following equivalent
identities
((xy)x)z = x(y(xz));
((xy)z)y = x(y(zy));
(xy)(zx) = (x(yz))x.
1
hold for every three elements of the loop. Moufang loops are diassociative: the
subloop generated by every two elements is a group.
For an element a of Q the bijections La : Q 7−→ Q and Ra : Q 7−→ Q given
by Lax = ax and Rax = xa will be called, respectively, the left and the right
translation by a. The left and right translations generate the multiplication
group, Mlt(Q). The stabilizer of the neutral element in Q defines the inner
mapping group denoted by Inn(Q). This group is actually a subgroup of the
multiplication group generated by three different types of elements of Mlt(Q),
namely
ℓx,y = L−1
xy ◦ Lx ◦ Ly,
rx,y = R−1
xy ◦ Ry ◦ Rx
and
Tx = L−1
x ◦ Rx
for every two elements x and y in Q. If the group Inn(Q) acts on Q by au-
tomorphisms, the loop Q is said to be automorphic. Equivalently, the loop is
automorphic if the mappings ℓx,y, rx,y and Tx are automorphisms of Q. All the
left translations by elements in Q generate the left multiplication subgroup of
Mlt(Q), denoted by LMlt(Q). The stabilizer of the neutral element in LMlt(Q)
defines the left inner mapping group, LInn(Q), generated by the mappings ℓx,y.
If LInn(Q) is a group of automorphisms of Q
then Q will be called a left automorphic loop.
Definition 1. A Moufang loop L with the property that every three elements of
L generate a left automorphic subloop will be called an almost left automorphic
Moufang loop.
In the differential geometry framework the notion of a loop being left auto-
morphic is important: a loop with the left automorphic and left power alterna-
tivity properties completely defines a reductive homogeneous space. [K3], [SLV].
Following ideas of M. Kikkawa [K1] we have studied [CS1] the so-called reduc-
tive Kikkawa spaces. There it is shown that smooth left automorphic power
alternative loops are Moufang.
The corresponding tangent algebra A of a reductive Kikkawa space is char-
acterized as an anticommutative algebra satisfying the relations
J(xy, z, u) + J(yz, x, u) + J(zx, y, u) = 0,
J(x, y, uv) = J(x, y, u)v − J(x, y, v)u,
∀x, y, z, u, v ∈ A.
where J(x, y, z) := (xy)z + (yz)x + (zx)y.
In ([CS1], p.6, T.2,13 ) it was also proved that tangent algebras of reductive
Kikkawa spaces are a certain type of Malcev algebras (anticommutative algebras
satisfying J(x, y, xz) = J(x, y, z)x, see [Sa]). More precisely, we have
2
Theorem 1. [Carrillo-Catal´an, Sabinina] Let A be an anticommutative algebra
for which the following identities hold for every five elements x, y, z, u, v of A:
J(xy, z, u) + J(yz, x, u) + J(zx, y, u) = 0,
J(x, y, uv) = J(x, y, u)v − J(x, y, v)u,
(1)
(2)
If the characteristic of A is not 2 then A is contained in the variety of Malcev
algebras defined by the identity:
J(x, y, z)x = J(x, y, xz) = 0,
∀x, y, z ∈ A.
(3)
It turns out that the converse of last theorem is not true. In what follows,
let us call Malcev algebras of the first type those Malcev algebras which satisfy
the identities (1) and (2). In the same way, those Malcev algebras which satisfy
(3) will be called Malcev algebras of the second type.
Malcev algebras of the second type form a larger variety of algebras than
Malcev algebras of the first type. In 2015 I. Shestakov constructed an algebra
of dimension 29 by using the Non Associative Algebra System [Albert] (Jacobs,
Muddana and Offutt, 1993), which is a Malcev algebra of the second type, but
not a Malcev algebra of the first type, [SIP].
In this paper we construct a new example of Malcev algebra of the second type,
but not a Malcev algebra of the first type of dimension 23, moreover, we think
that this example has the minimal dimension.
Conjecture 1. Every Malcev algebra the second type of dimension < 23 over
a field of characteristic 0 is an algebra of the first type.
In this work we try to answer the question: What kind of smooth loop
corresponds to a Malcev algebra of the second type?
2 Malcev algebras of the first type
It is possible to get equivalent characterizations of Malcev algebras of the first
type by using Sagle's properties of Malcev algebras. The following result gives
the equivalences.
Lemma 1. A is an anticommutative algebra for which the identities (1) and
(2) hold for every five elements of A if and only if A is a Malcev algebra such
that for every four elements of A the identity
J(x, y, uv) = 0
(4)
holds, provided the characteristic of A is not 2. In the special case where the
characteristic of A is neither 2 nor 3, the former identity is equivalent to having
J(x, y, z)u = 0,
(5)
for every four elements of A.
3
Proof. From the last result we know that A is a Malcev algebra for which (3)
holds. Now, since the characteristic of A is not 2, the identity (1) can be
expressed, by using Sagle's properties for Malcev algebras ([Sa], p.429, L. 2.10,
2.14) and some odd permutations, as:
−2uJ(x, y, z) = −J(u, x, yz) − J(u, y, zx) − J(u, z, xy)
= J(yz, x, u) + J(zx, y, u) + J(xy, z, u)
= 0.
Therefore the identity (1) can be replaced by the expression J(x, y, z)u = 0.
Moreover, this identity can be used to rewrite (2). Namely, If J(x, y, z)u = 0
holds for any four elements of A then J(x, y, u)v = J(x, y, v)u = 0 and therefore
J(x, y, uv) = 0.
Conversely, suppose A is a Malcev algebra such that for every four elements
(4) holds. Then, being a Malcev algebra, the linearized form of the Malcev
property ([Sa], p.429, 2.7):
J(x, y, wz) = J(x, y, z)w + J(w, y, z)x − J(w, y, xz)
is satisfied for any x, y, z, w in A. In particular, if (4) holds, then
J(x, y, v)u = −J(u, y, v)x
for any four elements in A. Then, after an odd permutation, for any four
elements in the algebra:
J(x, y, u)v − J(x, y, v)u = −J(v, y, u)x − (−J(u, y, v))x = 2J(u, y, v)x.
If A is a Malcev algebra with characteristic different from 2, then 2J(u, y, v)x =
0 ([Sa], p.429, 2.14). Hence the identity (2) holds trivially:
J(x, y, uv) − J(x, y, u)v + J(x, y, v)u = 0.
For a Malcev algebra of characteristic different from 2 or 3 the identities
J(x, y, uv) = 0 and J(x, y, z)u = 0 are equivalent. To see this, suppose that
J(x, y, uv) = 0 for every four elements of the algebra. Since the characteristic
of the algebra is not 2 and due to Malcev algebra properties ([Sa], p.429, 2.14)
we should have:
−2uJ(x, y, z) = J(yz, x, u) + J(zx, y, u) + J(xy, z, u) = 0,
hence J(x, y, z)u = 0. Conversely, assume the characteristic of the algebra is
different from 3, then once again due to Malcev algebra properties ([Sa], p.430,
2.15) we know that
3J(wx, y, z) = J(x, y, z)w − J(y, z, w)x − 2J(z, w, x)y + 2J(w, x, y)z.
4
Therefore, if J(x, y, z)u = 0 then J(wx, y, z) = 0 for any four elements of
the algebra. Hence J(y, z, wx) = 0.
Finally, to prove (5), suppose that for every four elements in A the identities
(1) and (2) hold. Then by Theorem 1 A is a Malcev algebra. As before, the
identity (1) and Sagle's properties for Malcev algebras ([Sa], p.429, 2.14) can be
combined to get (5). Conversely, we know that (5) holds in A if and only if (4)
is satisfied for every four elements of the algebra. But the identity (4) directly
implies (1) and (2) as shown above.
From all of the above, we conclude that an anticommutative algebra with
characteristic different from 2 and 3 and satisfying the identities (1) and (2) is
equivalent to a Malcev algebra for which (4) or (5) holds for every four elements
in the algebra. In other words, Malcev algebras of the first type can be charac-
terized as algebras for which (1) and (2) hold or as Malcev algebras satisfying
(4) or as Malcev algebras satisfying (5) or as anticommutative algebra satisfying
(4) and (5), provided the characteristic of the algebra is not 2 nor 3.
In what follows we will use the definition of Malcev algebra of the first type
as a Malcev algebra with the identity (4). It is known that the tangent algebra
of a smooth diassociative loop is a Binary Lie algebra, which can be defined by
the identity J(x, y, xy) = 0 (see [GA])
3 Malcev algebras of second type
Let A be a Malcev algebras of the second type. In this section we will prove
that such algebras turn out to be tangent of a local almost left automorphic
Moufang loops.
We begin with statements which are direct consequences of Malcev algebras
properties:
Proposition 1. If y and z are two elements of A such that J(y, z, A) = 0 then
yz belongs to the Lie kernel of A, that is, yz ∈ N (A) = {x ∈ A J(x, A, A) = 0}.
Proof. Since A is a Malcev algebra, it follows from Sagle ([Sa], p.431, 2.26) that
the identity
J(wx, y, z) = wJ(x, y, z) + J(w, y, z)x + 2J(yz, w, x),
(6)
holds for any four elements of the algebra. Now, if J(y, z, A) = 0 then 2J(yz, A, A) =
0.
Proposition 2. A4 ⊆ N (A).
5
Proof. By the last result, since the identity (3) holds in A, we know that x(xz) ∈
N (A) for any two elements of the algebra. Then the quotient algebra A/N (A)
must satisfy x(xy) = 0. As proved in ([EA], Prop 3.1, p.1-5) we can conclude
that (A/N (A))4 = 0 or, equivalently, A4 ⊆ N (A).
The above ideas can be generalized by using the fact that the Jacobian is a
skew-symmetric multilinear map.
Lemma 2. The multilinear maps ξ, ζ, ς : A × A × A × A −→ A given by
ξ(x1, x2, x3, x4) = J(x1, x2, x3x4)
ζ(x1, x2, x3, x4) = J(x1, x2, x3)x4
ς(x1, x2, x3, x4, x5) = J(x1x2, x3x4, x5)
are skew-symmetric.
Proof. Since A is anticommutative and due to the properties of the Jacobian it
is clear that ξ(x1, x2, x3, x4) = J(x1, x2, x3x4) is skew-symmetric in x1 and x2
as well as in x3 and x4. If (3) holds for A, the map ξ is also skew-symmetric for
x1 and x3 since J((x1 + x3), x2, (x1 + x3)x4) = 0 can be rewritten by linearity
as
J(x1, x2, x1x4) + J(x3, x2, x3x4) + J(x1, x2, x3x4) + J(x3, x2, x3x4) = 0.
Using (3) we obtain:
J(x1, x2, x3x4) = −J(x3, x2, x1x4).
Notice that if ξ is skew-symmetric in x1 and x3 it is also skew-symmetric in x1
and x4:
J(x1, x2, x3x4) = −J(x1, x2, x4x3) = J(x4, x2, x3x1) = −J(x4, x2, x1x3).
The proof that the map ζ is skew-symmetric is similar: due to the properties
of the Jacobian ζ is skew-symmetric in x1 and x2 as well as in x1 and x3. Now
by linearity and applying (3), the fact that J((x1 + x4), x2, x3)(x1 + x4) = 0
implies the skew-symmetry in x1 and x4.
As above, ς is skew-symmetric in x1 and x2 as well as in x1 and x4. The
skew-symmetry of ξ can be used to prove that ς is skew-symmetric in xi and
x5 for i = 1, 2, 3, 4. Notice that J(x1x2, x3x4, x5) = J(x5, x1x2, x3x4) due to an
even permutation. Since ξ is skew-symmetric in x1 and x3, ς is skew-symmetric
in x5 and x3: J(x5, x1x2, x3x4) = −J(x3, x1x2, x5x4), as well as in x5 and x4:
J(x5, x1x2, x3x4) = −J(x4, x1x2, x3x5). To verify that ς is skew-symmetric in
x5 and x2 we also use the skew-symmetry of ξ, suitable permutations and the
properties of the Jacobian:
J(x5, x1x2, x3x4) = −J(x5, x3x4, x1x2) = −J(x1, x5x2, x3x4),
6
and similarly for x5 and x1. The result follows since the permutations (i, 5) for
i = 1, 2, 3, 4 generate the symmetric group in {1, 2, 3, 4, 5}.
Theorem 2. Let A be an anticommutative algebra satisfying (3). Then
A. A3 belongs to the Lie Kernel N (A). In particular A/N (A) is a nilpotent
Lie algebra.
B. A2 is a Lie algebra.
C. A satisfies the identity:
2wJ(x, y, z) = 3J(w, x, yz).
D. The following hold:
i. J(Ai, Aj, Ak) = 0
ii. J(Ai, Aj, Ak)Ar = 0
if
if
i + j + k ≥ 5.
i + j + k + r ≥ 5.
iii. (J(A, A, A)A)A = 0.
E. J(A, A, A)2 = 0.
Proof. Notice that due to anticommutativity we know that
J(x1x2, x3x4, x5) = −J(x3x4, x1x2, x5)
for any elements x1, x2, x3, x4, x5 of A. In the previous Lemma, the map ς was
proved to be skew-symmetric. Then we can write
J(x1x2, x3x4, x5) = −J(x3x2, x1x4, x5) = J(x3x4, x1x2, x5).
We conclude that J(A2, A2, A) = 0. Now, due to the skew-symmetry of the map
ξ, for every J(xy, ab, c) ∈ J(A2, A2, A) we have
J(xy, ab, c) = −J(x(ab), y, c).
In the same way, by using the skew-symmetry of ς followed by the skew-
symmetry of the map ξ we get:
J(xy, ab, c) = −J(xy, ac, b) = J(x(ac), y, b).
Hence
0 = J(A2, A2, A) = J(A3, A, A),
which implies that A3 ⊆ N (A). On the other hand, A/N (A) is a Lie algebra
since J(A, A, A) ⊂ A3 ⊆ N (A) and N (A) is zero in A/N (A). This proves (A),
7
(B ) and part (i) of (D ) of the Theorem.
On the other hand, since A is a Malcev algebra, once again according to
Sagle's properties for Malcev algebras ([Sa], p.429, L. 2.10, 2.14), for every
x, y, z, w of A we should have
2wJ(x, y, z) = J(w, x, yz) + J(w, y, zx) + J(w, z, xy).
It is enough to use this identity and the skew-symmetry of the map ξ to
prove (C ): each of the terms can be written as J(w, y, zx) = −J(w, y, xz) =
J(w, x, yz) and J(w, z, xy) = −J(w, x, zy) = J(w, x, yz).
The last part of (A) follows from (C ) and (i):
from (C ) we know that
2AJ(A, A, A) = 3J(A, A, A2). This implies that 2AJ(A, A, A2) = 3J(A, A, A3) =
0 due to (i). We conclude that AJ(A, A, A2) = 0. Therefore
(2AJ(A, A, A))A = (cid:0)3J(A, A, A2)(cid:1) A = 3J(A, A, A2)A = 0.
Since (2AJ(A, A, A))A = −2(J(A, A, A)A)A we conclude that (J(A, A, A)A)A =
0.
Finally (C ) implies that J(A, A, A)A3 = 0. Since J(A, A, A) ⊆ A3, we get:
J(A, A, A)2 ⊆ J(A, A, A)A3 = 0.
Corollary 1. If A is a semiprime algebra, then A is a Lie algebra.
Proof. Suppose that A is a semiprime algebra. Since, by the previous Theorem,
the ideal J(A, A, A) satisfies J(A, A, A)2 = 0 it follows that J(A, A, A) = 0,
namely, A is a Lie algebra.
Theorem 3. A 3-generated Malcev algebra A of the second type is a Malcev
algebra of the first type.
Proof. Let A be a Malcev algebra of the second type generated by the elements
a, b, c. Let ω be an arbitrary element of A. Since A is a 3- generated algebra, ω
is a formal linear combination of products of the generators a, b, c.
Let us show that the identity (4) holds in A. First, consider J(x, y, cω),
where c is an arbitrary generator of A. Since the Jacobian is a multilinear
operator, J(x, y, cω) can be
simplified using (4, i) of Theorem 2:
J(x, y, cω) = ℓ1J(a, b, ca) + ℓ2J(a, b, cb) + ℓ4J(a, b, c(ab)) + ℓ5J(a, b, c(ac))+
ℓ6J(a, b, c(bc)) + ℓ7J (a, b, c((ab)c)) + ℓ8J (a, b, c(a(bc))) + · · ·
Using again the fact that A3 ⊆ N (A) and J(a, b, cb) = 0, we conclude that
8
J(x, y, cω) = 0 for any ω in A.
The general case, J(x, y, uv) = 0 where x, y, u and v are arbitrary elements
of A, can be handled by analogous considerations.
Due to Malcev -Kuzmin theory [Kuz], we have
Corollary 2. A Malcev algebra of the second type is, in fact, a tangent algebra
of a local almost left automorphic Moufang loop. The tangent algebra of every
smooth almost left automorphic Moufang loop is a Malcev algebra of the second
type.
Remark The question of the existence of global almost left automorphic
Moufang loop, which corresponds to the the given Malcev algebra of the second
type is solved positevly in the article [CGRS].
4 Example
In this section we discuss an example of a 23-dimensional algebras of second
type which is not an algebra of first type.
Let F be a free anti commutative algebra generated by X = {x1, x2, x3, x4},
nilpotent of class 4, it means F 4 = 0. Let I be a subspace with a basis of all
X-words, w = w(x1, x2, x3, x4), such that some letter xi appears in w two or
more times. It is clear that I is an ideal of F . Let's denote by A the factor
algebra F/I. Then a basis of A has 22 elements: B = ∪3
i=1Bi with
B1 =X,
B2 ={[xi, xj] 1 ≤ i < j ≤ 4},
B3 ={[xi, xj, xk] 1 ≤ i < j ≤ 4, 1 ≤ k ≤ 4, k 6= i, j}.
The algebra A is a Malcev algebra.
Let's define an antisymmetric bilinear function ψ : A × A 7−→ k given by the
following values:
ψ([x1, x2], [x3, x4]) = 2, ψ([x1, x3], [x2, x4]) = −2,
ψ([x1, x4], [x2, x3]) = 2, ψ([x2, x3, x1], x4) = −3,
ψ([x2, x4, x1], x3) = 3,
ψ([x2, x4, x3], x1) = −1,
ψ([x3, x4, x1], x2) = −3, ψ([x3, x4, x2], x1) = 1,
and ψ(v, w) = 0 for all other values.
9
Consider a space A = A ⊕ kv and define a product on A as follows:
[(a, αv), (b, βv)] = ([a, b], ψ(a, b))
(7)
Direct computations show that A is a second type Malcev algebra. On the
other hand, if we set xi = (xi, 0), then
[J(x1, x2, x3), x4] = [x1, x2, x3, x4] + [x2, x3, x1, x4] − [x1, x3, x2, x4]
= (0, −3v) 6= 0.
Hence A is not an algebra of the first type.
Acknowledgments
The authors thank Alberto Elduque, Alexander Grishkov and Ivan Shestakov
for useful comments.
Funding
The first author thanks CONACYT and Universidad Pedag´ogica Nacional Unidad
201 Oaxaca for supporting the C´atedras CONACYT project 1522. The second
author thanks CNPq (Brasil), grant 308221/2012-5 and the third author thanks
SNI and FAPESP grant process 2015/07245-4 for support.
10
References
[CS1]
[CPS]
[EA]
[Ga]
Carrillo-Catal´an R., Sabinina L. On smooth power-alternative loops,
Communications in algebra, Vol. 32, No. 8, pp. 2969 - 2976, (2004).
Chein O., Pflugfelder H.O. and Smith J.D.H., Quasigroups and loops:
Theory and applications, Sigma Series in Pure Mathematics, 8, Hel-
dermann Verlag, (1990).
Elduque A., Quadratic Alternative Algebras, J. Math. Phys.31, 1, 1-5,
17A45 (17D05), (1990).
Gainov A.T. Binary Lie algebras of characteristic two. Algebra and
Logic , 8 : 5 (1969) pp. 287 - 297 Algebra i Logika , 8 : 5 pp. 505 -
522, (1969).
[CGRS] Grishkov A.,Carrillo Catalan R.,Rasskazova M.,Sabinina L. Nilpotent
by Lie center Malcev algebras and corresponding analytic Moufang
loops. Preprint.
[Ker]
[K1]
[K3]
[Kuz]
[SLV]
[Sa]
Kerdman F.S. Analytic Moufang loops in the large, Algebra and Logic
(1979) 18: 325. Translated from Algebra i Logika, Vol. 18, No. 5, pp.
523 - 555, September -October, (1979).
Kikkawa M., On local loops in affine manifolds, J.Sci. Hiroshima Univ.
Ser. A-I, 28 , 199 - 207. (1964).
Kikkawa M., Geometry of homogeneous Lie loops, Hiroshima Math. J.
5:141-179, (1975).
Kuz'min, E.N. On the relation between Mal'tsev algebras and analytic
Mufang groups. Algebra and Logic (1971) 10: 1. Translated from
Algebra i Logika, Vol. 10, No. 1, pp. 3 -22, January -February, (1971).
Sabinin Lev V. Smooth Quasigroups and Loops Kluver Academic Pub-
lishers. Dordrecht/Boston/London 1999.
Sagle A. Malcev Algebras. Trans. Amer. Math. Soc.,101 (1961), 3,
426 - 458 MR 26 1343.(1963).
[SIP]
Shestakov I.P. Private Communication.
11
|
1610.03977 | 1 | 1610 | 2016-10-13T08:37:17 | G-Graded Central Polynomials and G-Graded Posner's Theorem | [
"math.RA"
] | Let F be characteristic zero field, G a residually finite group and W a G-prime and PI F-algebra. By constructing G-graded central polynomials for W, we prove the G-graded version of Posner's theorem. More precisely, if S denotes all non-zero degree e central elements of W, the algebra S^{-1}W is G-graded simple and finite dimensional over its center.
Furthermore, we show how to use this theorem in order to recapture the result of Aljadeff and Haile stating that two G-simple algebras of finite dimension are isomorphix iff their ideals of graded identities coincide. | math.RA | math |
G-Graded Central Polynomials and G-Graded Posner's
Theorem
Yakov Karasik
November 15, 2018
Abstract
Let F be characteristic zero field, G a residually finite group and W a G-prime and PI
F-algebra. By constructing G-graded central polynomials for W , we prove the G-graded
version of Posner's theorem. More precisely, if S denotes all non-zero degree e central
elements of W , the algebra S−1W is G-graded simple and finite dimensional over its
center.
Furthermore, we show how to use this theorem in order to recapture the result of
Aljadeff and Haile stating that two G-simple algebras of finite dimension are isomorphix
iff their ideals of graded identities coincide.
1 Introduction
Let W be an associative algebra over a field F. The most basic setup of the theory of associative
algebras satisfying a polynomial identity (PI in short) assumes the existence of a non-zero
polynomial f = f (x1, ..., xn) with non-commuting variables x1, ..., xn, such that f (a1, ..., an) =
0 for every a1, ..., an ∈ W . In that case we say that W is a PI algebra and f is a polynomial
identity of A. The class of PI algebras includes all commutative algebras (since clearly [x1, x2]
is an identity), finite dimensional algebras, Grassmann algebras and Grassmann envelopes of
finite dimensional algebras. This family is rigid in the sense of being closed to major algebraic
operations. For instance, the homomorphic image of a PI algebra is easily seen to be also PI.
If A ⊆ B and B is PI so does A. The direct product and even the tensor product of two PI
algebras is, by a celebrated theorem of Regev [9], also PI.
PI algebras possesses properties similar to finite dimensional algebras. The Jacobson rad-
ical of an affine PI algebra is nilpotent [1].
If A is a PI algebra containing no nilpotent
ideals, then it also does not contain nil ideals. The primitive PI algebras are simple and finite
dimensional over their center of dimension bounded by the degree of a multilinear identity.
Furthermore, if A is a semiprime (i.e. the intersection of its prime ideals is zero which is equiv-
alent to A having no nilpotent ideals) PI algebra, then every non-zero ideal of A intersects the
center of A non-trivially. The last statement yields the famous theorem of Posner (see all in
chapter 1.11 of [7]):
Theorem 1.1 (Posner's Theorem). Suppose A is semiprime and PI algebra having a field as
its center. Then, A is a simple finite dimensional algebra over its center.
A more general version of this theorem is
Theorem 1.2. Suppose A is semiprime and PI algebra with a center C. Then, every non-zero
ideal of A intersects non-trivially C.
1
The main idea of the proof involves the existence of a central polynomial for Mn(F) -
n × n matrices over the field F and the fact, due to Amitsur, that if A is semiprime, then
A[x] = A ⊗F F[x] is semisimple.
In this paper we consider the group graded analogue of Posner's Theorem. To be more
precise, suppose G is a residually finite group. We prove:
Theorem 1.3 (G-graded Posner's Theorem). Suppose A is a G-graded algebra over a field F
of characteristic zero. Suppose further that A is G-semiprime and (ordinary) PI. Then, every
G-graded ideal {0} 6= I of A intersects the e-part of the center of A (denoted by Ce) in a
non-trivial fashion.
As a result, if Ce is a field, then A is a G-simple algebra. Moreover, A is finite dimensional
over Ce.
Remark 1.4. In a (English translation of) paper of Balaba [4] it seems that the author have
proved a version of theorem 1.3 for every G. However, the proof given there is valid only for
finite abelian groups G. Indeed, the author starts his proof (Proposition 2) by using Corollary
4.5 from [5] which is valid for finite groups (it is false for infinite groups, see 4.6). The end of the
proof makes sense only for abelian groups. Indeed, the auther claims that given an ungraded
central polynomial q(x1, ..., xn) of a G-graded algebra A, it is possible to find a0
n ∈ A
- a homogenous evaluation of q such that q(a0
n) is non-zero and homogenous. If G is
abelian, then every monomial a0
σ(n) is of the same homogenous degree, so the above
statment holds. However, if G is not abelian there is no reason for it to hold.
σ(1) ··· a0
1, ..., a0
1, ..., a0
In fact, the above difficulties are the main focus of this paper. They require a new technic
to construct G-graded central polynomials of G-simple and finite dimensional algebras.
Finally, in the final section, we show how to use the above theorem to recover a result of
Aljadeff and Haile [2] which states
Theorem 1.5. Suppose G is a finite group and A1, A2 are G-simple algebras of finite dimen-
sion over an algebraically closed field F of characteristic zero. Then, idG(A1) = idG(A2) if
and only if A1 and A2 are G-isomorphic over F.
Our technique uses very little from the structure theory of G-simple algebras and will be
used in consequent papers in order to investigate the connection between G-simple structures
and their graded identities.
2 G-graded structure theorems
Throughout this section F will denote a field of any characteristic and G will denote any
group. The goal of this section is to prove the G-graded analogs of the basic structure theorems
of primitive and simple PI algebras. More precisely, we will prove the G-graded Wedderburn's
theorem, G-graded Jacobson's density theorem and Kaplansky's theorem on primitive PI
algebras. These theorems even in the G-graded case should be well known, however the
author has failed to find a full account of their proofs in the literature.
We note that the work of Bakhturin, Segal and Zaicev (see [3]) treats the G-graded Wed-
derburn's theorem in the case where G is finite and the characteristic of F is zero or coprime
to G.
We follow the exposition of the ungraded case in [8].
Assumption 2.1. Throughout the chapter A denotes a G-graded F-algebra.
2
Definition 2.2. Let M be a left A-module. We say that M is G-graded A-module if
Mg
M =Mg∈G
(Mg is an F-vector space) and AgMh ⊆ Mgh for all g, h ∈ G.
graded A-submodules.
A G-graded A-module M is called G-graded irreducible if it contains no non-trivial G-
Let M and N be G-graded A-modules. We say that the A-module map φ : M → N is a
G-graded homomorphism of degree g ∈ G, if for every h ∈ G: φ(Mh) ⊆ Ngh.
Notation 2.3. Since G and A are fixed we will use the abbreviated phrases "graded module",
"graded irreducible" and "graded homomorphisms".
Note 2.4. The kernel of a graded module homomorphism φ : M → N is a graded module.
Indeed, suppose a = Pg∈G ag ∈ M (ag ∈ Mg) is mapped to zero. Since each φ(ag) is an
element of a distinct homogeneous component of N , the ag must be mapped to zero.
Moreover, the image of such a homomorphism is a graded submodule of N .
Suppose M is a graded module. Let EndG
A(M ) denote the F-algebra generated by all
A(M )g consists of all
A(M ): It is
graded endomorphisms of M . This is a G-graded algebra where EndG
graded homomorphisms of degree g. This indeed defines a G-grading on EndG
trivial that EndG
Remark 2.5. If G is a finite group, then EndG
A(M ) = EndA(M ). Indeed, if φ ∈ EndA(M ),
write φg =Ph∈G PghφPh, where Ph is the projection onto Mh, and notice that φh sits inside
A(M )g · EndG
A(M )g′ ⊆ EndG
EndA(M )h. Now,
A(M )gg′.
Xg∈G
φg = Xh,g∈G
PghφPh = Xh∈G
Ph! · φ ·
Xg∈G
Pg
= φ.
Definition 2.6. A G-graded F-algebra D is called G-division algebra if every non-zero homo-
geneous element of D is invertible.
Remark 2.7. It is clear that 1 ∈ De, so in particular De 6= 0.
Lemma 2.8 (Schur's Lemma). If M is graded irreducible, then EndG
A(M ) is a G-division
algebra. Moreover, if N is an irreducible graded module which is non-graded isomorphic to M ,
then there are no graded homomorphisms from M to N but the zero homomorphism.
Proof. Let 0 6= φ be a homogeneous element in EndG
A(M ). Since ker φ is a graded submodule
of M , we have ker φ = 0. So φ is injective. From the same reason its image must be equal to
M . So φ is also surjective. All in all, φ is isomorphism, thus invertible.
For the second part, let φ be a graded homomorphism between M and N . As before, the
kernel and the image must be zero or everything. If the kernel is zero and the image is N , we
get that φ is a graded isomorphism - contradicting the assumption on M and N . In any other
case φ must be the zero homomorphism.
Lemma 2.9. If D is a G-division F-algebra, then D = ⊕h∈H(Dh = Dbh), where D is an
F-division algebra, H is a subgroup of G and bh is an invertible element of Dh. Moreover,
if F is algebraically closed and dimFDe < ∞, then D is a twisted group algebra. That is,
D = FαH, where α ∈ Z 2(H, F×) (H acts trivially on F×).
3
Proof. Only the second part requires a proof. We first remark that we will rely heavily on the
algebraic closeness of F.
De is a division algebra over F. Moreover, the center of De must be a finite extension of
F. Hence it is equal to F. Thus De is a central simple F-algebra. So De = F.
Let H be the subset of G, such that Dh 6= 0 for h ∈ H. If 0 6= bh ∈ Dh, then multiplication
by bh from (say) the right is an F-linear homomorphism from De to Dh. Since bh is invertible,
this homomorphism is invertible. Thus Dh = Fbh.
Since the product of two invertible elements is also invertible, we deduce that H is a
subgroup of G.
Definition 2.10. A G-graded algebra A is called G-primitive if there is a G-graded module
V of A such that the action of A on V is faithful and G-irreducible.
By lemma 2.9, EndG
A(V ) = ⊕g∈HDbg =: D, where H is a subgroup of G, D is a division
F-algebra and bg is an invertible element of D. Notice that H = suppD (the support of a
G-graded algebra A is the subgroup of G generated by all h ∈ G for which Ag 6= {0}).
Theorem 2.11 (G-graded density). Suppose A is G-primitive F-algebra acting on the G-
graded module V irreducibly and faithfully. Let D := EndG
A(V ), D = De, H = suppD and
VT =Lt∈T Vt, where T is a set of representatives of right cosets of H inside G.
Then, for every homogeneous elements v1, ...., vm ∈ VT which are D-independent and for
every homogeneous u1, ..., um ∈ V , such that g0 := (deg ui)−1 · deg vi is the same for i = 1...m,
there is an a ∈ A such that avi = ui (i = 1...m).
Note 2.12. It is clear that if a = Pg∈G ag is the decomposition of a into its homogeneous
components, then agvi = 0 for i = 1...m and g 6= g0. Thus, we may replace a by ag0 and thus
assume that a is homogeneous.
Proof. We claim that it is enough to establish the following: Suppose U is a finite dimensional
(over D) G-graded subspace of VT and v′ /∈ U is a homogeneous element of VT , then there is
a homogeneous a ∈ A such that aU = 0 but av′ 6= 0.
To see this we proceed by induction on m. Let U = SpanD{v1, ..., vm} and v′ = vm+1. By
induction we can find a1 ∈ A such that aivi = ui for i = 1...m. Moreover, let a2 ∈ A be a
homogeneous element such that a2U = 0 and a2v′ 6= 0. Since A acts G-irreducibly on V and
a2v′is homogeneous, it is possible to find a3 ∈ A for which a3 (a2v′) = um+1 − a1v′. Thus,
a = a3a2 + a1 moves each vi to ui.
We prove the claim by induction on m = dimD U . The case m = 0 follows from faithfulness
of the action of A on V . Suppose m > 0. Hence, there is a homogeneous v ∈ U and a subspace
U0 of U such that U = U0 + Dv0 and v0 /∈ U0.
By induction, the annihilator of V0, annA(U0), does not annihilate any element of VT not
in U0. Moreover, annA(U0) is a left G-graded ideal of A, hence by G-irreducibility, for every
homogeneous v /∈ U0, annA(U0)v = V .
Assume to the contrary that annA(U )v′ = 0.We define a map τ : V → V by the following
procedure: For a homogeneous 0 6= v ∈ V we can always find a homogeneous a ∈ annA(U0)
such that av0 = v. Using a we declare τ (v) = av′. This is well defined, since if av0 = 0,
then a must be in annA(U ). Hence, by our assumption, a annihilates v′, that is av′ = 0.
Moreover, for every homogeneous v, (deg τ (v))−1 deg v = (deg v′)−1 deg v0 =: h0. Hence τ is
a homogeneous element of D.
4
The map τ is clearly F-linear. Moreover, if b ∈ A is homogeneous, bav0 = bv, hence
τ (bv) = bav′ = bτ (v).
Thus, τ ∈ D and therefore equals to αbh0, where α ∈ D. Furthermore, for a ∈ A:
av′ = τ (av0) = aτ (v0) ⇒ a(v′ − τ (v0)) = 0.
Thus,
Since both v0 and v′ are in VT , we deduce that v′ = αev0, which is absurd.
v′ = τ (v0) ⇒ v′ = αbh0v0.
We turn to consider G-graded G-simple algebras.
Definition 2.13. A is G-simple if A2 6= 0 and every G-graded two-sided ideal of A is trivial.
It is easy to check that every G-simple algebra is also G-primitive. Indeed, Choose V = A.
The action is, by definition, is G-graded irreducible. Moreover, if there was a homogeneous
a ∈ A such that aA = 0, then Aa was a two-sided G-graded ideal of A. Therefore, aA = 0 or
A. The latter is impossible since then A2 = AaA = 0. So F a is a two-sided G-graded ideal
of A. As before, the conclusion is that Fa = 0 ⇒ a = 0. Thus, if a ∈ A is not necessary a
homogeneous element it is impossible that aA = 0. All in all, the action is also faithful.
In order to give an example of a G-simple algebra, we define:
Definition 2.14. Let B be any G-graded F-algebra, a natural number n and g = (g1, ..., gn) ∈
Gn. Denote by Mg(B) the F-algebra B ⊗ Mn(F), G-graded by
Mg(B)g = SpanF{b ⊗ ei,j b ∈ Bh, g = g−1
i hgj}.
Note 2.15. It is easy to check that if B is G-simple, then so does Mg(B).
Definition 2.16. Let A be a G-graded F-algebra. Denote by Aop the opposite G-graded F-
algebra of A defined by Aop = A as F-vector spaces; the multiplication is given by a1a2 = a2·a1,
where · is the multiplication in A; the G-grading is given via:
Aop
g = Ag−1
for all g ∈ G.
Lemma 2.17. Let A be G-simple F-algebra. Then, Aop is also G-simple. Furthermore, if A
is a G-division F-algebra, so does Aop.
Proof. The proof is omitted.
Notation 2.18. For an algebra A, we denote by Z(A) the center of A. Moreover, If A is G-
graded, the center Z(A) will (usually) not be graded by G, unless G is abelian. Nevertheless,
we denote by Z(A)e all elements of degree e inside Z(A).
Definition 2.19. A finite dimensional G-simple algebra A over F is called G-central-simple
algebra if F = Z(A)e.
5
Proposition 2.20. Suppose A is a G1-central-simple algebra over F. If B is G2-simple unital
algebra over F, then A⊗FB is G1×G2-central simple algebra over Z(B)e (here, (A⊗B)(g1,g2) =
Ag1 ⊗ Bg2). In particular, if B is a field (considered graded by {e}), then A ⊗F B is G-simple
algebra over Z(A ⊗F B)e = B.
Proof. If 0 6= I is a G1 × G2-graded ideal of A ⊗ B, and 0 6= x = Pt
shortest element in I (that is, for every 0 6= x′ =Pt′
can write Pp c′
It is easy to see that b1, ..., bt are independent over F. Furthermore, since Aa1A = A, one
j ∈ I, one has t′ ≥ t).
p = e for all p. Hence,
p = 1. We can assume that deg c′
j=1 aj ⊗ bj ∈ I(g1,g2) is
p · g1 · deg c′′
j=1 a′
j ⊗ b′
pa1c′′
p ⊗ 1 = 1 ⊗ b1 + d2 ⊗ b2 + ··· + dt ⊗ bt,
c′
p ⊗ 1 · x′ · c′′
Xp
p ∈ Ae. Thus, for every a ∈ A,
pajc′′
where dj =Pp c′
is of length shorter than t, hence [a, dj] = 0 for j = 2, ..., t. As a result, dj ∈ F. Hence,
[a, 1 ⊗ b1 + d2 ⊗ b2 + ··· + dt ⊗ bt] ∈ I
z = 1 ⊗ (b1 + d2b2 + ··· + dtbt).
Since b1, ..., bt are linearly independent over F, we get that z 6= 0. As a result, t = 1 and
(may assume) x = 1 ⊗ b1. The G-simplicity of B and the fact that b1 is homogenous forces
A ⊗ B · x · A ⊗ B. All in all, I = A ⊗ B.
Next, it is clear that Z(B)e ⊆ Z(A ⊗ B)e. For the other direction,
Z(A ⊗ B)e ∋ 0 6= z =
t
Xi=1
ai ⊗ bi,
where t is minimal. As before, b1, ..., bt, are independent over F. For a ∈ A,
0 = [z, a] =
t
Xi=1
[ai, a] ⊗ bi.
Thus, [ai, a] = 0 for i = 1, ..., t. Hence, ai ∈ Z(A)e = F. As a result, t = 1 and z = 1 ⊗ b1 ∈ B
As a result, z ∈ Z(B).
Lemma 2.21. Suppose A is G-primitive F-algebra acting on the G-graded module V irreducibly
and faithfully. Let D := EndG
of representatives of right cosets of H inside G. Suppose further that VT is finite dimensional
over D.
A(V ) D = De, H = suppD and VT =Lt∈T Vt, where T is a set
Then, A is a G-simple algebra over the field K = Z(A)e and dimK Ae < ∞. Furthermore,
A = Mg(D) for some g = (g1, ..., gn) ∈ Gn and a finite dimensional G-division K-algebra D.
Proof. Consider the map φ : Aop → EndG
D(V ) given by right multiplication. It is easy to check
that this is a G-graded homomorphism. Since the action is faithful the map is an embedding.
By lemma 2.17 Aop is G-simple and thus also G-primitive. Moreover, suppose v1, ..., vm is
a homogeneous basis of VT . Then any homogeneous f ∈ EndD(V ) is completely determined
by its action on v1, ..., vm. Write u1 = f (v1), ..., ul = f (vm). Then clearly (deg ui)−1 · deg vi
6
is the same for i = 1...m. Hence, by theorem 2.11, there is an a ∈ Aop such that φ(a) = f .
Therefore, φ is a G-graded isomorphism.
It is easy to check that
EndD(V ) ⋍ D ⊗D EndD(VT ) ⋍ D ⊗D Mg(D) ⋍ D ⊗Z(D) Mg(Z(D)),
where g = (t1, ..., t1, ..., tl, ..., tl), where {t1, ..., tl} = T and each ti appears exactly dimD Vti
times in g. Now, notice that if e = g−1
i hgj (here h ∈ H and gi, gj ∈ T ), then gi = hgj.
Therefore, gi = gj and h = e (recall that T is a transveral of H in G). As a result, Aop
e ⊆
D ⊗D EndD(VT ). So K = Z(Aop)e = Z(D), which is a field (since D is a division ring). All in
all, A is G-isomorphic to Mg(Dop) which is G-simple due to lemma 2.17 and 2.15.
For the next Corollary we will need the following definition.
Definition 2.22. A G-graded F-algebra A is called G-Artinian if every descending sequence
of G-graded ideals of A: I1 ⊇ I2 ⊇ ··· eventually stabilizes.
Note 2.23. If dimF A < ∞, it is clear that A is G-Artinian.
Theorem 2.24 (G-graded Wedderburn's theorem). Suppose A is G-simple and (left) G-
Artinian F-algebra. Then, A = Mg(D), where g = (g1, ..., gn) ∈ Gn and D is a finite dimen-
sional G-division F-algebra. If moreover K = Z(A)e is algebraically closed, then D = FαH,
where H = suppD and α ∈ Z 2(H, F×).
Proof. We want to use 2.21 with V = A (as a G-graded left A-module). Suppose VT is of
infinite dimension over F = D (see lemma 2.9). The set
Rn = {a ∈ A aUn = 0},
where Un is an n-dimensional G-vector space inside VT , is clearly a left G-graded ideal of A.
By choosing U1 ⊆ U2 ⊆ ··· we get R1 ⊇ R2 ⊇ ··· . Moreover, by irreducibility we know that
these are strict inclusions. Hence, we reached a contradiction.
The statement now follows from 2.21 and lemma 2.9.
Corollary 2.25 (G-graded Kapalansky). If A is G-primitive F-algebra and moreover satisfies
an ordinary PI of degree d, then the conclusion of the previous corollary holds. Furthermore,
K = Z(A)e is a field and dimK AT ≤ d
2 .
Proof. We use the notation of theorem 2.11. It is enough to show that dimD VT ≤ d/2. Indeed,
otherwise there is an d/2 < n-dimensional G-graded D-subspace U of VT and consider the
subring of A given by
R = {a ∈ A aU ⊆ U}.
By the previous theorem there is an epimorphism from R to EndD(U ) = Mn(D). However,
Mn(F) satisfies a PI of degree d < 2n or higher. This contradicts the assumption that A
satisfies a PI of degree d.
Theorem 2.26. Suppose A is a G-simple PI F-algebra, where G is finite. Then, there is a
field F ⊆ K and a G-graded K-algebra B = KαH ⊗ Mg(K) such that
idG,F(A) = idG,F(B).
7
Proof. By 2.25, for L = Z(A)e, we get that A is finite dimensional over L and A = Mg(D) for
a G-division (finite dimensional) L-algebra D. As a result, for K = L, K ⊗L A ⋍ Mg(K ⊗L D)
(notice that L being e-homogeneous implies that the above isomorphism preserves the G-
grading). Furthermore, by lemma 2.9, K ⊗L D is of the form KαH, where H is a subgroup of
G and α ∈ Z 2(H, K∗). Finally the theorem follows since
idG,F(A) = idG,F(K ⊗L A).
Remark 2.27. Notice that K ⊗F idG,F(B) ⊆ idG,K(B). However, the other inclusion is not
assured. For instance, let F = Q, K = C, G = C3 × C3 = hσi × hτi and B = KαG, where
uσuτ = ξuτ uσ (here ξ = 3√1) and u3
τ = 1, we obtain that x1.σx2,τ − ξx2.τ x1,σ ∈ idG,K(B)
but not in K ⊗F idG,F(B).
Definition 2.28. For a G-graded algebra A we denote by JG(A) the intersection of all of its
G-primitive ideals. We call JG(A) the G-graded Jacobson radical of A.
σ = u3
If JG(A) = {0}, we say that A is G-semisimple.
Theorem 2.29 (see [5]). If G is a finite group and F is of zero characteristic, then JG(A) =
J(A) for every G-graded F-algebra A. In particular, A is G-semisimple if and only if A is
semisimple.
Lemma 2.30. Suppose A is a G-semisimple PI F-algebra and I is a G-graded ideal of A.
Then I is also G-semsimple algebra. Furthermore, Z(I) ⊆ Z(A).
Proof. By assumption A is a subdirect product of G-primitive algebras Ai = A/Mi, where Mi
is a G-primitive ideal (i ∈ J - an index set). That is, there is a G-graded embedding
φ : A →Yi∈J
Ai
By 2.25, we get that each Ai is, in fact, G-simple algebra.
such that pi◦ φ is surjective for every i ∈ J, where pi :Qj∈J Aj → Ai is the natural projection.
Since every Ai is G-simple, for every i ∈ J pi◦ φ(I) = 0 or Ai. Denote by J0 all the indexes
j ∈ J for which the result is Aj. Hence, the sequance
0 → I → Yi∈J0
Ai → Ai → 0
is exact. This shows that I is a G-semisimple algebra.
If a ∈ Z(I), then pi ◦ φ(a) is, in any case, inside Z(Ai) (for every i ∈ J). As a result
a ∈ Z(cid:8)Qi∈J Ai(cid:9) ∩ A ⊆ Z(A).
3 G-graded PI
We will use from now on the standard notations of G-graded PI theory.
A G-graded polynomial is an element of the non-commutative free algebra F hXGi gener-
ated by the variables from XG = {xi,gi ∈ N, g ∈ G}. It is clear that FhXGi is itself G-graded,
where FhXGig is the span of all monomials xi1,g1 ··· xin,gn, where g1 ··· gn = g.
8
We say that a G-graded polynomial f (xi1,g1, ..., xin,gn) is a G-graded identity of a G-graded
algebra A, if f (a1, ..., an) = 0 for all a1 ∈ Ag1, ..., an ∈ Agn. The set of all such polynomials
is denoted by idG(A) and it is evident that it is an ideal of FhXGi. In fact, it is a G-graded
T -ideal, that is an ideal which is closed to G-graded endomorphisms of FhXGi.
For G finite. it is often convenient to view any ungraded polynomial as G-graded via the
embedding
FhXi → FhXGi ,
where xi 7→Pg∈G xi,g. This identification respects the G-graded identities, in the sense that
if f ∈ id(A) (the T -ideal of ungraded identities of A), then f ∈ idG(A).
As in the ungraded case, the set of (G-graded) multilinear identities of A
(Xσ∈Sn
ασxiσ(1),gσ(1) ··· xiσ(n),gσ(n) ∈ idG(A) ασ ∈ F; i1, ..., in are all distinct)
T -generates idG(A).
Suppose f ∈ FhXGi, A is a G-graded algebra and S ⊆ A . We write f (S) for the subset of
A consisting of all the graded evaluations of f by elements of S. Moreover, if B is a subset of
A consisting of homogenous elements such that spB = A, then spf (B) = A. Thus, f ∈ idG(A)
if and only if f vanishes under graded substitutions by elements from B.
3.1 Central polynomials
In this section we introduce the main tool for proving Posner's theorem (see section §4).
Definition 3.1. Let A be an F-algebra. We say that f ∈ FhXi is a central polynomial of A
if f (A) ⊆ Z(A) and f is not a PI of A.
If A is also graded by a group G, then we say that f ∈ FhXGi is a G-graded central
polynomial of A if f (A) ⊆ Z(A) and f is not a G-graded PI of A. However, we will most of
the time use the phrase "central polynomials" also for G-graded central polynomials.
Suppose F is a field of characteristic zero. Then, the following polynomial (called the Regev
polynomial)
F hXi ∋ Ld(x1, ..., xd, y1, ..., yd) = Xσ,τ ∈Sd
(−1)στ xσ(1)yτ (1)xσ(2)xσ(3)xσ(4)yτ (2)yτ (3)yτ (4) ···
··· xσ((n−1)2+1) ··· xσ(d)yτ ((n−1)2+1) ··· yτ (d)
is a central polynomial of Mn(F), where exp(A) = d = n2. That is, the values of Ld when
evaluated on Mn(F) are in the center of Mn(F) and there are non-zero values (see [6]).
Remark 3.2. A basic property of Ld polynomial is that for a1, ..., ad, b1, ..., bd ∈ Mn(F ) the
value of Ld(a1, ..., ad, b1, ..., bd) is non-zero if and only if each one of the sets {a1, ..., ad} and
{b1, ..., bd} is F -independent.
Lemma 3.3. Suppose A is a semiprime F-algebra (see 4.1) of exponent exp(A) = d, where
F is a characteristic zero field. Then, d = n2 is a square and A is PI equivalent to Mn(F).
Furthermore, if f is a central polynomial of A it is an identity of any semiprime algebra A′ of
lower exponent exp(A′) = m2.
9
Proof. The first part is a consequence of the ungraded Posner's Theorem (see Theorem 1.12.1
in [7]).
Furthermore, let f be any central polynomial of Mn(F). Since for m < n we may view
Mm(F) as a subalgebra of Mn(F) via the embedding A 7→
(notice that the
0
A 0 ···
0
...
. . .
0
0
0
in embedding does not preserve the unit element of Mm(F)), so f is an identity of Mm(F).
We turn to G-graded polynomials.
Definition 3.4. For f ∈ FhXGi and g ∈ G, denote by ρg(f ) the g-component of f . When
g = e, we write ρ for ρe.
In order to deal with the infinite group case we will need a stronger notion of ungraded
central polynomials.
Definition 3.5. A polynomial f = f (x1, ..., xl) ∈ FhXi is called strong central polynomial
for a G-graded F-algebra A (or G-strong central polynomial), if:
1. f is a central polynomial of A.
2. For every g1, ..., gl ∈ G and ag1 ∈ Ag1, ..., agl ∈ Agl such that f1 = f (x1,g1, ..., xl,gl) is
non-zero when evaluated by x1,g1 = ag1, ..., xl,gl = agl, then ρ(f1)(ag1, ..., agl) 6= 0.
The following Lemma shows that if f is a strong central polynomial, then we can construct
from it e-homogenous G-graded central polynomials.
Lemma 3.6. Suppose f is G-graded central polynomial of a G-graded F-algebra A. Then,
ρ(f ) is either central or a G-graded identity of A.
Proof. Decompose f into a sum of its G-graded components: f = Pg∈G ρg(f ). Fix h ∈ G
and consider yh ∈ XG which does not appear in f . Since f is central, it is clear that [yh, f ] ∈
idG(A). Hence, Pg∈G[yh, ρg(f )] = 0. Since [yh, ρg(f )] ∈ FhXGih if and only if g = e (notice
that [yh, ρg(f )] ∈ FhXGigh ⊕ FhXGihg), we get [yh, ρe(f )] ∈ idG(A). Because this holds for
all h ∈ G, we conclude that ρ(f )(A) lies inside the center of A.
Corollary 3.7. Suppose f = f (x1, ..., xl) is a strong central polynomial for the G-graded F-
algebra A. Then, if f1 = f (x1,g1, ..., xl,gl) is not an identity of A (here g1, ..., gl ∈ G), ρ(f1) is
a central polynomial of A.
We will be interested in graded central polynomials for the G-graded algebra A = FαH ⊗
Mg(F) (as in 2.14). We introduce two methods of constructing them.
3.1.1 Strong central polynomials using involution.
Consider the following involution on the free algebra FhXi given by: (αxi1 ··· xim)t = αxin ··· xi1.
We use this action to construct new (ungraded) central polynomials: Suppose f = f (x1, ..., xm) ∈
FhXi is a central polynomial of Mn(F). Since the transpose action on matrices is an in-
volution on Mn(F), we conclude that f t is also a central polynomial of Mn(F). Hence,
f = f (x1, ..., xm)f (y1, ..., ym)t is also a central polynomial.
10
Lemma 3.8. Consider the G-graded algebra A = FαH ⊗ Mg(F), where F = C and the values
α takes are roots of unity. Let f = Ld(x1, ..., xd, y1, ..., yd) ∈ FhXi ⊆ FhXGi be the Regev
polynomial, where d = exp(A). Then, for every a1, ..., ad ∈ A, the e part of
f (a1, ..., ad, a1, ..., ad; a∗
1, ..., a∗
d, a∗
1, ..., a∗
d)
is not zero if and only if f (a1, ..., ad, a1, ..., ad) 6= 0, where (cuh ⊗ M )∗ = ¯cu−1
denotes the complex conjugate of c ∈ C and M ∗ = M
Proof. First, we claim that ()∗ is an anti-automorphism of A. Indeed,
for M ∈ Mm(C).)
t
h ⊗ M ∗. (Here ¯c
whereas
Finally,
(cid:0)cuh ⊗ M · c′uh′ ⊗ M ′(cid:1)∗ = cc′α(h, h′)u−1
(cuh ⊗ M )∗(cid:0)c′uh′ ⊗ M ′(cid:1)∗ = cc′u−1
h u−1
u−1
h u−1
h′ = (α(h, h′)uhh′)−1 = α(h, h′)u−1
hh′.
hh′ ⊗ M ′∗M ∗
h′ ⊗ M ′∗M ∗.
The last equality follows from α(h, h′) = 1.
Now we are ready to prove the Lemma. If f (a1, ..., ad, a1, ..., ad) = 0, it is obvious that also
f (a1, ..., ad, a1, ..., ad; a∗
1, ..., a∗
d, a∗
1, ..., a∗
d) = 0,
in particular its e component must be zero.
If, on the other hand, f (a1, ..., ad, a1, ..., ad) is non-zero, since f is a central polynomial of
A, it must be equal to
where at least one of the αi 6= 0. Since ()∗ is an anti-automorphism,
αgug ⊗ I ∈ A,
Xg∈H
d, a∗
1, ..., a∗
1, ..., a∗
f (a1, ..., ad, a1, ..., ad; a∗
d) = Xg
The e part of the last expression is equal to Pg∈H αg2 > 0.
Theorem 3.9. Suppose that F is any field of characteristic zero, G is a finite group and
A = FαH ⊗ Mg(F) is G-graded as in 2.14. Suppose further that f is a central polynomial of
A. Then, f is strong central polynomial of A.
αgug! Xg
¯αgug−1! ⊗ I.
Proof. By extending scalars, we may assume that F is an algebraically closed field containing
C. Hence, since every α ∈ Z 2(H, F∗) is cohomologous to α′ (over the algebraically closed field
F) which satisfies the assumption in the previous theorem. Moreover, by Lemma 1.3 in [2],
the G-graded algebras FαH ⊗ Mg(F) and Fα′
is a strong central polynomial of AC, hence also if A.
As a result AC = CαH ⊗ Mg(C) is well defined and F ⊗C AC. By the previous Lemma, f
H ⊗ Mg(F) are G-graded isomorphic.
11
3.1.2 Strong central polynomials using projective representations of groups.
The following method is due to Eli Aljadeff.
We start with a preliminary Lemma:
Lemma 3.10. Suppose A and B are semisimple and finite dimensional F-algebras. If φ : A →
B is an epimorphism, then every central minimal idempotent in B can be lifted in A.
Proof. The kernel of φ, denoted by I, must be of the form (as any ideal of A) Ai1 × ··· × Ais.
Without loss of generality assume that I = Aq′+1 ×···× Ar. Thus, the restriction φA1×···×Aq′
is an isomorphism. From Corollary 7.50 in ???, it follows that there is q = q′ and there is
σ ∈ Sq such that φ(Ai) = Bσ(i). The Lemma is now clear.
Lemma 3.11. Assume F is a characteristic zero algebraically closed field and 0 6= ρ =
Pg∈G µgug ∈ FαG is a central idempotent. Then µe 6= 0.
Proof. By a theorem of Schur (see Theorem 11.17 in ???) we know that there exist a finite
central extension
0
/ C
/ Γ π /
/ G
/ 0
and λ ∈ hom(C, F∗) such that [λ ◦ β] = [α] ∈ H 2(G, F∗), where [β] ∈ H 2(G, C) corresponds
to the above extension. Hence, by choosing a transversal π(γ1) = g1, ..., π(γm) = gm (here
G = {e = g1, g2, ..., gm}), we get an epimorphism
φ : FΓ → FαG
given by φ(ucγi) = λ(c)ugi.
Since both algebras are finite dimensional and semisimple, by lemma 3.10, there is a central
idempotent ρ′ ∈ FΓ which projects (via π) onto ρ. Thus, there exists a character χ of Γ such
that:
ρ′ =
χ(e)
Γ Xγ∈Γ
χ(γ)uγ.
Since C ≤ Z(Γ) and χ is a character of an irreducible representation V of Γ we know, by
Schur's Lemma, that c acts on V by multiplication by
dim V χ(c). Hence, χ(cγ) = χ(c)χ(γ).
This all boils down to:
1
ρ = π(ρ′) = Xc∈C
χ(c)λ(c)! ·
χ(e)
Γ
m
Xi=1
χ(γi)ugi.
Since e 6= 0, χ(e)Pc∈C χ(c)λ(c) 6= 0. Hence also µe = µg1 =(cid:16)Pc∈C χ(c)λ(c)(cid:17) · χ(e)
Theorem 3.12. Let A = FαH ⊗ Mg(F), where F is a field of zero characteristic. Suppose
F = F (x1, ..., xl) is a central polynomial of A. Then, F is a strong central polynomial of A.
Proof. Since FαH is semisimple (since F is of zero characteristic) and finite dimensional, we
can write it as the product Mk1(F) × ··· × Mkl(F), where assuming k1 ≥ k2 ≥ ··· ≥ kl. So A
as an ungraded algebra is isomorphic to
6= 0.
Γ
(Mk1(F) × ··· × Mkl(F)) ⊗ Mg(F) = Mn=k1+g(F) × ··· × Mkl+g(F).
As a result, A is PI equivalent to Mn(F). So, F is a central polynomial of Mn(F). Thus,
W = F (cid:0)Mk1(F) ⊗ Mg(F) = Mn(F)(cid:1) = Fρ ⊗ I\, where ρ = 1Mk1 (F) is a central idempotent of
FαH. By the previous Lemma, the e-component of W is not zero, hence f is a non-identity.
Thus, by lemma 3.6, f is a central polynomial of B of degree e.
12
/
/
/
3.1.3 G-graded central polynomials for lower exponent algebras.
We start with a definition:
Definition 3.13. A G-graded polynomial f is called d-sharp, if for every G-simple algebra A
the following holds:
• exp(A) = d =⇒ f /∈ idG(A).
• exp(A) < d =⇒ f ∈ idG(A).
Theorem 3.14. Suppose A is a G-simple F-algebra, where F is a characteristic zero field,
which is ungraded PI. Every ungraded central polynomial f of A is G-strong and ρ(f ) is
exp(A)-sharp.
Proof. By theorem 2.26, there is a field F ⊆ K and a G-graded K-algebra B = KαH ⊗ Mg(K)
such that idG,F(A) = idG,F(B). Hence, by theorem 3.12, f is G-strong central polynomial of
A.
The second part follows from lemma 3.3 since every G-simple algebra over a characteristic
zero field is semisimple (hence semiprime).
4 G-graded Posner's theorem
In order to introduce the main theorem we need the following definitions.
Definition 4.1. Let G be any group and suppose that A is a G-graded F -algebra. We say
that A is a G-prime algebra if for every two graded ideals I and J such that IJ is equal to
the zero ideal, one of them is already the zero ideal. A graded ideal P of A is called G-prime
if the graded algebra A/P is G-prime.
Furthermore, A is called G-graded semiprime if the intersection of all G-graded prime
ideals of A is zero.
Remark 4.2. It is a simple task to verify that every G-primitive algebra is also G-prime.
Moreover, by [5], every G-semiprime is also (ungraded) semiprime.
In the previous section we established that in the case where A is G-primitive and satisfies
an ordinary PI, A is, in fact, of the form Mg(F αH) given that the field F is algebraically
closed. The G-graded version of Posner's theorem asserts a similar "rigidity" result for G-
prime algebras (which satisfy a PI). Here is the precise statement:
Conjecture 4.3. Suppose G is any group and A is a G-prime F -algebra which also satisfies a
PI, then S = Z(A)e is a domain and (A ⊆)S−1A satisfies the conclusion of 2.21 for the field
K = S−1S (In particular, this algebra is G-simple and finite dimensional over K).
In this section we prove this conjecture for G residually finite. The first step is to
establish the claim for G finite. For this we follow the footsteps of Rowen ([10]) and prove the
following key theorem.
Theorem 4.4. Suppose A is a G-graded F-algebra, where G is a finite group and F is of
characteristic zero. Suppose further that A is G-semiprime and ungraded PI. Then, every
G-graded ideal I intersects non-trivially Z(A)e.
13
Proof. By [5] A[x] = A ⊗F F[x] is G-semisimple, where the G-grading is given by A[x]g =
Ag[x]. Since Z(A[x]) = (Z(A))[x], it is suffice to prove the theorem for A[x] and I[x], so we
may assume from the beginning that A is G-semisimple. By lemma 2.30, I is G-semisimple
and Z(I) ⊆ Z(A). So it is enough to show that every G-semisimple PI algebra A satisfies
Z(A)e 6= {0}. This will follow if we will show that A has a degree e central polynomial.
By the proof of lemma 2.30, A is a subdirect product of G-simple F-algebras Aj, where
j ∈ J. In other words, the map
φ : A →Yj∈J
Aj
natural map.
is a G-graded embedding and πi : A → Ai is onto, where πi : A → Qj∈J Aj → Ai is the
Notice that for every i ∈ J, id(A) ⊆ id(Aj). So exp(Aj) ≤ exp(A). Choose j0 ∈ J0
such that expF(Aj0) is maximal among the exponents of all other algebras corresponding
to elements in J0. Write B for Aj0 and d for exp(B). By theorem 3.14, B has a strong
polynomial F such that f = ρ(F ) is d-sharp. That is, f is central for Aj0 and for every
other j ∈ J f is either an identity of Aj or a central polynomial of Aj.
In other words,
f (A) ∈ Z(cid:16)Qj∈J Aj(cid:17)e ∩ A ⊆ Z(A)e.
Remark 4.5. The proof above shows that if A is G-semiprime (G is a finite group), then every
ungraded central polynomial f of A is G-strong central polynomial of A.
Remark 4.6. Notice that for G infinite, it is no longer true that A[x] is G-semisimple (take
G = Z and consider A = F[x] where deg xn = n ∈ Z). As a result, in order to generalize the
previous theorem to infinite groups one should introduce a d idea.
As a result we obtain 4.3 for finite groups. Here is the precise statement.
Corollary 4.7. Let G be a finite group and F a field of characteristic zero. Suppose A is
G-prime over F and satisfies an ordinary PI. Then, S = Z(A)×
e does not contain zero divisors
of A. As a result A is G-embedded in A1 = S−1A which is a G-graded K = S−1Z(A)e-algebra.
Moreover, A1 is finite dimensional G-simple K-algebra.
Proof. . Regarding the first part, it is enough to prove that no element c ∈ Z(A)∗
e annihilates
a non-zero homogenous element a ∈ A. Suppose otherwise: ca = 0. Hence, AcA · AaA is a
product of graded ideals which is equal to A3caA = 0. Thus, AcA = 0 or AaA = 0. It is
therefore enough to show that AaA = 0 is impossible (since it is more general than AcA = 0).
Indeed, in that case Aa is a graded ideal, thus we obtain from AaA = 0 that Aa = 0 (since
A 6= 0). But now Fa is an ideal and so Fa = 0, which yields a = 0 - a contradiction.
that I must be equal to A1. By 2.25 we are done.
Suppose I is a non-zero G-graded ideal of A1. Since I ∩ K = I ∩ Z(A1) 6= 0, we conclude
4.0.4 Grading by a quotient group
Definition 4.8. Let A be an F-algebra graded by G and let Q = G/N be any quotient group
of G. We define the induced Q-grading on A by setting
for every g ∈ G.
AgN = ⊕h∈N Agn,
14
Example 4.9. Two notable special cases are when N = {e} and N = G. In the first case,
we obtain the given G-grading on A and in the second we are in the situation of having no
grading at all
Definition 4.10. Let G be a group and Q = G/N be any quotient group of G. Denote by
ψG,Q the F-algebra map ψG.Q : FhXGi → FhXQi induced by ψG,Q(xi,g) = xi,gN .
Furthermore. f ∈ FhXGi is said to be Q-stable if the diagram:
FhXGi
ρg
FhXQie
ψG,Q
FhXQi
ρgN
ψG,Q
/ FhXQieN
commutes for every g ∈ G on f .
Example 4.11. Let G = C4 = hτi and N =(cid:10)τ 2(cid:11). The polynomial f1 = x2,τ 2 is not Q-stable
but f2 = 2x1,e is.
4.0.5 Residually finite case.
Theorem 4.12. Let G be a residually finite group and F a characteristic zero field. Suppose A
is G-semiprime and ungraded PI. Then, every G-graded ideal I intersects non-trivially Z(A)e.
Proof. By [5] A is a semiprime F-algebra. Thus, A[x] is semisimple. As before we replace
A and I by A[x] and I[x] what allows us to assume that I (and A) is a semisimple algebra
and Z(I) ⊆ Z(A). So, it will be suffice to show that I has a strong central polynomial. In
other words, we need to show that if A is G-graded and semisimple, it possess a strong central
polynomial.
By lemma 3.3, A is PI equivalent to Mn(F), where n2 = exp(A). Hence Regev's polynomial
F = Ln2(x1, ..., x2n2 ) is a central polynomial for A (in fact, we could as well chosen any other
central polynomial of Mn(F)). As a result, there are g1, ..., g2n2 ∈ G for which there are
ag1 ∈ Ag1, ..., ag2n2 ∈ Ag2n2 such that F (ag1, ..., ag2n2 ) 6= 0. Denote FG = F (x1,g1, ..., x2n2,g2n2 )
and f = ρ(FG). We will show that f is central polynomial of A.
Let N be a normal finite index subgroup of G such that all the distinct products of
{g1, ..., g2n2} of length 2n2 remain distinct in Q = G/N . Therefore, FG is Q-stable. Regard A
as a Q-graded algebra. Hence, by the proof of 4.7 (see also 4.5), ρeN (ψG,Q(FG)) is Q-central
polynomial of degree eN for A.
Since ψG,Q (f ) = ρeN (ψG,Q(FG)), it is clear that
f (A) ⊆ (ρeN (ψG,Q(FG)) (A)) ∩ Ae ⊆ Z(A)e.
Finally, ψG,Q(FG)(ag1, ..., ag2n2 ) = FG(ag1, ..., ag2n2 ) 6= 0. Hence, the fact that F is Q-strong
for A, forces
0 6= ρeN (ψG,Q(FG)) (ag1, ..., ag2n2 ) = f (ag1, ..., ag2n2 ).
All in all, we shown that f is indeed a G-graded central polynomial of A.
Corollary 4.13. theorem 4.4 holds also when G is a residually finite group and F is of char-
acteristic zero.
15
/
/
/
5 A consequence of the G-graded Posner's Theorem.
In [2] Aljadeff and Haile proved the following theorem:
Theorem 5.1. Suppose G is a finite group and A1, A2 are G-simple algebras of finite dimen-
sion over an algebraically closed field F of characteristic zero. Then, idG(A1) = idG(A2) if
and only if A1 and A2 are G-isomorphic over F.
Note 5.2. It is clear that if A1 is G-isomorphic to A2, then A1 and A2 share the same G-graded
identities. The other direction requires all the work.
The proof they presented is based heavily on an elaborate analysis of the structure of a
G-simple algebra given in theorem 2.24. Here we present an alternative proof based on the
G-graded Posner theorem proven in the previous section.
To start, recall the construction of the (G-graded) generic algebra over F. Given a finite
dimensional G-graded F-algebra, chose a G-graded F-basis of A: B = ∪g∈GBg (Bg consists of
elements of a basis of Ag) and consider the commutative variables
Λ = {ti,g,b b ∈ Bg, g ∈ G, i ∈ N} .
Finally denote by UA the F-subalgebra of A ⊗F F(Λ) generated by the elements yi,g =
Pb∈Bg ti,g,bb. It is easy to verify that the G-homomorphism
φ : FhXGi /idG(A) → U
given by φ(xi,g) = yi,g is well defined and is G-isomorphism. As a result, we may identify
FhXGi /idG(A) with U .
In our case we have Ui ≃ FhXGi /idG(Ai) (i = 1, 2) and since idG(A1) = idG(A2), we
must have U1 = U2. As a result, we will denote this algebra (also) by U . Notice that
U ⊆ Ai ⊗F F(Λi).
Lemma 5.3. U is G-semisimple.
Proof. If JG(U ) 6= 0, then there is a G-graded substitution of the variables xi,g ∈ XG by
elements from A1 (of course, we could as well use A2) for which JG(U ) is mapped to a
non-zero ideal I in A1 and the corresponding map is surjective. However, it is well known
that Jacobson radical is mapped to Jacobson radical when the map under consideration is
surjective. We got a contradiction to A1 being G-semisimple.
Lemma 5.4. Z(U )e ⊆ F(Λi) · 1Ai ⊗ 1 for i = 1, 2. As a result, Z(Ue) does not contain any
zero-divisors of U .
Proof. First observe that Z(Ai)e = F1Ai. This can be verified easily using theorem 2.24.
As a result, if f (y1,g1, ..., yn,gn) ∈ Z(Ue), then φ(f ) ∈ Z (Ai ⊗ F(Λi))e = Z(Ai)e ⊗ F(Λi) =
F(Λi) · 1Ai ⊗ 1.
Remark 5.5. It is possible to prove that Z(Ai) = F1Ai by only using the classical Wedderburn
theorem for (ungraded) semisimple algebras. Here is a sketch: Set A = A1. Since the Jacobson
radical is nilpotent for commutative algebras, one shows that the semisimplicity of A forces
Z(Ae) to be also semisimple. Next, if e1, e2 are two nonzero orthogonal idempotents in Z(A)e,
then Aei is G-graded ideal of A, so equals A. Hence, there is a ∈ A for which ae1 = e2. But
this is an absurd since 0 = e1 · e2 = a · e2
1 = ae1 = e1. Thus Z(Ue) must be a field. Since
surely, F1 ⊆ Z(Ue) and F is algebraically closed, we must have equality.
16
Notation 5.6. S = Z(U )e − {0}.
Due to the previous Lemma,
U ⊆ S−1U ⊆ Ai ⊗ F(Λi).
Now by theorem 4.4, we know that S−1U is G-simple and f.d. over K := Z(S−1U )e ⊆ F(Λi)
(which is a field). Furthermore, since the graded identities of a G-simple algebra are defined
over an algebraically closed field (this follows from theorem 2.26 and the proof of theorem 3.9)
and the fact that idG,F(Ai) = idG,F(S−1U ) forces
idG,F(Λi)(S−1U ⊗K F(Λi)) = idG,F(Λi)(Ai ⊗ F(Λi)).
Lemma 5.7. The F(Λi)-algebra S−1U ⊗K F(Λi) is G-simple. Furthermore, the map
ν : S−1U ⊗K F(Λi) → S−1U · F(Λi)
is a G-graded isomorphism.
Proof. The first statement is clear by 2.20 Moreover, it is also clear that ν is a G-graded
epimomorphism. Finally, the kernel of ν (which is a G-graded ideal of S−1U ⊗K F(Λi)) is zero,
thus proving that ν is a G-graded isomorphism.
Lemma 5.8. S−1U · F(Λi) = Ai ⊗ F(Λi).
Proof. We know by now that S−1U · F(Λi) is a G-simple F(Λi)-subalgebra of Ai ⊗ F(Λ) having
the same G-graded identities. So
dimF(Λi) S−1U · F(Λi) = expG S−1U · F(Λi) = expG Ai ⊗ F(Λ) = dimF(Λi) Ai ⊗ F(Λ).
Thus the statement follows.
We have shown that for a field L containing both F(Λ1) and F(Λ2), one has
A1 ⊗F L = A2 ⊗F L.
To conclude that A1 and A2 are G-isomorphic we only need to use the well known fact
that if X is any algebraic variety over a field (F ⊆)L, then since F is algebraically closed,
X(F) (the Fpoints of X) is dense in X. Indeed, we take X to be HomG,L(A1 ⊗F L, A2 ⊗F L).
Notice that X is defined over F. The subset V = IsoG,L(A1 ⊗F L, A2 ⊗F L) is open (defined
by non-vanshing of a certain determinant) and non-empty. Hence, V contains a point defined
over F - finishing the proof.
References
[1] Louis Rowen Alexei Belov-Kanel. The braun-kemer-razmyslov theorem for affine-pi alge-
bras. see http://arxiv.org/pdf/1405.0730.pdf.
[2] Eli Aljadeff and Darrell Haile. Simple G-graded algebras and their polynomial identities.
Trans. Amer. Math. Soc., 366(4):1749 -- 1771, 2014.
[3] Yu. A. Bakhturin, M. V. Zaıtsev, and S. K. Segal. Finite-dimensional simple graded
algebras. Mat. Sb., 199(7):21 -- 40, 2008.
17
[4] I. N. Balaba. Graded prime PI-algebras. Fundam. Prikl. Mat., 9(1):19 -- 26, 2003.
[5] M. Cohen and S. Montgomery. Group-graded rings, smash products, and group actions.
Trans. Amer. Math. Soc., 282(1):237 -- 258, 1984.
[6] Edward Formanek. A conjecture of Regev about the Capelli polynomial. J. Algebra,
109(1):93 -- 114, 1987.
[7] Antonio Giambruno and Mikhail Zaicev. Polynomial identities and asymptotic methods,
volume 122 of Mathematical Surveys and Monographs. American Mathematical Society,
Providence, RI, 2005.
[8] I. N. Herstein. Noncommutative rings, volume 15 of Carus Mathematical Monographs.
Mathematical Association of America, Washington, DC, 1994. Reprint of the 1968 origi-
nal, With an afterword by Lance W. Small.
[9] Amitai Regev. Existence of polynomial identities in A ⊗F B. Bull. Amer. Math. Soc.,
77:1067 -- 1069, 1971.
[10] Louis Rowen. Some results on the center of a ring with polynomial identity. Bull. Amer.
Math. Soc., 79:219 -- 223, 1973.
18
|
1305.5217 | 2 | 1305 | 2015-07-03T15:29:21 | Period and index, symbol lengths, and generic splittings in Galois cohomology | [
"math.RA",
"math.AG"
] | We use constructions of versal cohomology classes based on a new notion of "presentable functors," to describe a relationship between the problems of bounding symbol length in cohomology and of finding the minimal degree of a splitting field. The constructions involved are then also used to describe generic splitting varieties for degree 2 cohomology with coefficients in a commutative algebraic group of multiplicative type. | math.RA | math |
PERIOD AND INDEX, SYMBOL LENGTHS, AND GENERIC SPLITTINGS
IN GALOIS COHOMOLOGY
DANIEL KRASHEN
Abstract. We use constructions of versal cohomology classes based on a new
notion of "presentable functors," to describe a relationship between the problems
of bounding symbol length in cohomology and of finding the minimal degree of
a splitting field. The constructions involved are then also used to describe generic
splitting varieties for degree 2 cohomology with coefficients in a commutative
algebraic group of multiplicative type.
1. Introduction
For a field F containing an ℓ'th root of unity, and a central simple F-algebra A of
period ℓ, it follows from the seminal result of Merkurjev and Suslin [MS82], that
A is similar to a product of symbol algebras of degree ℓ. The minimal number
of symbols needed in such a presentation is called the symbol length of A. The
question then becomes, for a given field F, what is the minimal number of symbols
which are necessary to express any central simple F-algebra of period ℓ, or in other
words, what is the maximum symbol length of such an algebra? One may think of
this measure as an attempt to bound the potential "complexity" of the structure of
central simple F-algebras.
This question may naturally be generalized to higher degree cohomology classes
as well. Thanks to the Bloch-Kato conjecture/norm residue isomorphism Theorem,
proved in [Voe11] (see also [Wei09]), no longer assuming that µℓ ⊂ F, if we are
given a cohomology class α ∈ Hn(F, µ⊗n
ℓ ), we know that α may be written as a sum
of some number of symbols, where a symbol in this case refers to an n-fold cup
product of classes in H1(F, µℓ). We may then ask, how many symbols are necessary
to express our given class α -- this is the symbol length of α. As before, we ask:
"what the maximum value of the length over all possible classes in Hn(F, µ⊗n
One may also ask a similar question for quadratic forms. Namely, we know that
the powers of the fundamental ideal In(F) are generated by n-fold Pfister forms as
Abelian groups. We can then ask how many n-fold Pfister forms are necessary to
write a given element of In(F)?
ℓ )?"
Conjecturally, these questions should be closely connected with other measure-
ments of complexity of the arithmetic of the field F, such as the cohomological
dimension, the Diophantine dimension, the u-invariant of F, as well as its Brauer
dimension (defined as the minimum d such that ind α(per α)d for all Brauer classes
α ∈ Br(F)). The question of determining the Brauer dimension is a special case of
the period-index problem in cohomology.
This research was partially supported by NSF grants DMS-1007462 and DMS-1151252.
1
2
DANIEL KRASHEN
In this paper, we draw connections between these various measures of com-
plexity and describe new relationships between them.
In part, our method is
to construct "generic cohomology classes," of various types using objects which
we call "presentable functors," a concept somewhat related to O'Neil's notion of
"sampling spaces" [O'N12, O'N05]. In addition, the new technique of presentable
functors allows for the writing down of new generic splitting varieties for coho-
mology classes, which we do at the end of the paper.
Acknowledgments
The material in this paper came as a direct result of discussions and questions
which arose at the workshop "Deformation theory, patching, quadratic forms, and
the Brauer group," organized by the author and Max Lieblich, held at the Amer-
ican Institute of Mathematics in 2011, and the author benefited from a number of
conversations from the participants of that conference. The author would like to
thank Parimala and A. Merkurjev for useful conversations during the writing of
this paper, and Suresh and Parimala for pointing out inconsistencies in a prelim-
inary version of the manuscript. In addition, the author would like to thank the
anonymous referee for a large number of helpful comments and corrections.
The work presented is closely related to the work of Saltman [Sal11], which also
had its origins in the same workshop, although these works were done indepen-
dently and using different methods. The paper [Sal11], written prior to this one,
considers the problem of symbol length in mod-2 cohomology based on a finite
u-invariant assumption and in this context give not only the boundedness results
obtained in this paper, but also more detailed information concerning solvable
lengths of Galois groups which split cohomology classes. The present paper pro-
vides results concerning cohomology with more general coefficient groups, and
also addresses related topics such as Pfister length, the period-index problem, and
generic splitting varieties.
2. Overview and general remarks
In this section we briefly give an overview of the results of the paper.
Symbol length. Suppose for a field F, we may bound the degrees of splitting fields
for cohomology classes up to a certain degree n. In this case, we we may bound
the symbol length of cohomology classes in degree n. More precisely:
Definition 2.1. For α ∈ Hi(F, µ⊗i
ℓ ), we define the effective index of α, denoted eind(α)
to be the minimum degree of a finite separable field extension E/F such that αE = 0.
Definition 2.2. For α ∈ Hi(F, µ⊗i
ℓ ), we define the length of α, denoted λ(α) to be the
smallest m such that α may be written as a sum of m symbols.
Theorem (4.2). Let F be a field and fix a positive integer ℓ prime to the characteristic of F.
Fix d > 0. Suppose that for every m > 0, there is some Nm such that for every cohomology
class β ∈ Hd′(L, µ⊗d′
) with [L : F] ≤ m and d′ < d we have eind β ≤ Nm. Then for
α ∈ Hd(F, µ⊗d
ℓ ), we may bound λ(α) in terms of the effective index of α.
ℓ
We note that in particular, the bound doesn't depend on the particular field
F except in the determination of the constant N. In section 5, we present some
evidence that N might be bounded in terms of the Diophantine dimension of the
field F and its finite extensions, and in fact prove that this is the case for ℓ = 2.
PERIOD-INDEX, SYMBOL LENGTH, AND GENERIC SPLITTINGS
3
Remark 2.3. We note that it is immediate that if α ∈ Hd(F, ℓ⊗d) then we may bound
the effective index of α in terms of its symbol length. To see this, suppose that
i). One easily finds that the field
1, . . . , aj
we may write α = Pm
L = F(cid:18) ℓqa1
1, . . . , ℓpam
1, ℓqa2
j=1 aj with aj = (aj
1(cid:19) splits α.
Generic splittings.
Definition 2.4. Let A be a commutative group scheme and α ∈ Hd(F, A). We say
that a scheme X/F is a generic splitting variety for α if for every field extension L/F
we have X(L) , ∅ if and only if αL = 0.
ℓ
We note that for elements α of Hd(F, µ⊗d−1
) which are symbols (i.e. cup products
of a class in H1(F, Z/ℓ) and d − 1 classes in H1(F, µℓ)), Rost's norm varieties, as con-
structed in [SJ06, HW09] are known to be generic splitting varieties for α under the
assumption that ℓ is a prime and that F has no prime-to-ℓ extensions. Specific con-
structions for symbols give the existence of generic splitting varieties for symbols
in small degree, such as the Merkurjev-Suslin variety for a symbol in H3(F, µ⊗2
ℓ ). For
non-symbols in degree at least 2, the Severi-Brauer variety gives the only known
example of a generic splitting variety for a general class α ∈ H2(F, µℓ) = Br(F)ℓ.
We show that one may construct such generic splitting varieties for a general
class in H2(F, A) for any commutative group scheme A of multiplicative type:
Theorem (6.3). Let A be any commutative group scheme over F of multiplicative type
and suppose α ∈ H2(F, A). Then there exists a smooth generic splitting variety X/F for α.
Pfister numbers. Recall that the fundamental ideal I(F) of the Witt ring W(F) of
equivalence classes of quadratic forms over F is defined as the ideal generated by
even dimensional forms. One may show that the m'th power of this ideal Im(F),
is generated as an Abelian group by the set of m-fold Pfister forms: that is, those
forms which may be written as an m-fold tensor product of those of the form h1, −ai.
For such a form q with [q] ∈ Im(F), the m-Pfister number of q, denoted Pf m(q), is
defined to be the minimum number n such that q is equivalent to a sum of n m-fold
Pfister forms.
Theorem (5.5). Suppose that F is a field of characteristic not 2 in which −1 ∈ (F∗)2. Then
the following are equivalent:
(1) u(F) < ∞.
(2) For all m > 0, there exists Nm > 0 such that for all n > 0 and for all α ∈ Hn(L, µ2)
with [L : F] ≤ m, we have eind α < Nm.
(3) There exists N > 0 such that for all n > 0 and for all α ∈ Hn(F, µ2), we have
λ(α) < N.
(4) There exists N > 0 such that for all n > 0 and for all q ∈ In(F), we have Pf n(q) < N.
Cohomological dimension. It turns out that bounds on the symbol length and
related notions in Galois cohomology has striking consequences for cohomological
dimension. This can be seen using the reduced power operations, as described in
[Kah00, Rev73]. The result below is stated in [Kah00, Proposition 1(8)] in the case
ℓ = 2, but is easily extended to general ℓ as well.
Proposition 2.5. Suppose that F is a field such that the length of all classes in H2n(F, µ⊗2n
is bounded by m. Then H2n(m+1)(F, µ⊗2n(m+1)
) = 0 if ℓ is odd or −1 ∈ (F∗)2.
)
ℓ
ℓ
4
DANIEL KRASHEN
Remark 2.6. In particular, it follows from this result combined with Theorem 4.2
that a bound for the index of Brauer classes of period ℓ gives a bound on the
ℓ-cohomological dimension of F.
Proof. Write a symbol α ∈ H2n(m+1)(cid:16)F, µ⊗2n(m+1)
ℓ
(cid:17) as α = α1 ∪ · · · ∪ αm+1 where αi ∈
H2n(F, µ⊗2n
ℓ
). We note that we have
where [m] is the reduced power operation. But sincePi αi can also be written as a
sum of at most m symbols, it follows that α = 0.
(cid:3)
α = (α1 + · · · + αm+1)[m+1]
3. Compactness of functors and bounding symbol length
3.1. Functors. Let T be a commutative ring. By a T-algebra, we mean a ring R
together with a ring homomorphism T → R. By convention, all rings and ring
homomorphisms are assumed to be unital.
Definition 3.1. A T-functor is a functor from the category of commutative T-
algebras to the category of sets. We denote the category of T-functors by Fun/T.
Definition 3.2. An Abelian T-functor is a functor from the category of commutative
T-algebras to the category of Abelian groups. We denote the category of Abelian
T-functors by Ab/T.
Remark 3.3. We will generally abuse notation and consider the (faithful) forgetful
functor Ab/T → Fun/T as an inclusion.
Notation 3.4. Let X be a T-scheme. We let hX denote the functor represented by
X. Similarly, for a commutative T-algebra R, we will abbreviate hSpec(R) by hR.
We will often abuse notation and write X in place of hX.
Definition 3.5. We say that a morphism F : X → Y for X , Y ∈ Fun/T is
surjective (resp. injective) if for every commutative T-algebra R, the set map F (R) :
X (R) → Y (R) is surjective (resp. injective). Similarly, we say that a sequence of
Abelian T-functors is exact if it yields an exact sequence when evaluated at every
commutative T-algebra R.
Definition 3.6. Suppose that φ : T → R is a ring homomorphism, and H ∈ Fun/T.
We let HR ∈ Fun/R be defined by HR(S) = H (S) where a R-algebra S is considered
as a commutative T-algebra by composition of the structure map of S with φ.
3.2. Presentations. In this section we will develop the concept of "presentable
functors." In fact, we will not so much be interested in this particular property
as much as a weaker condition for a functor to satisfy, "pointwise coverability,"
which we describe in Section 3.3. The usefulness of presentability is that the set
of presentable functors forms an Abelian category (see 3.9), and is therefore closed
under useful constructions such as sums, kernels and cokernels. These provide us
with tools therefore to show that functors are pointwise coverable.
Definition 3.7. Let H ∈ Ab/T. We say that H is presentable if there exist affine
T-group schemes A1 and A2 and an exact sequence A1 → A0 → H → 0. In this
case we say that A• is a presentation of H .
PERIOD-INDEX, SYMBOL LENGTH, AND GENERIC SPLITTINGS
5
Remark 3.8. Here we are abusing notation and writing Ai, A′
algebraic groups and the Abelian T-functors which they represent.
i both for the linear
Note that if H , H ′ ∈ Ab/T are both presentable, then so is H × H ′ by taking a
term by term product of their two presentations. We write Pres to be the additive
subcategory of Fun/T consisting of presentable functors.
Lemma 3.9. Pres is an Abelian subcategory of Ab/T.
Proof. We need to show that Pres is closed under taking kernels and cokernels in
Ab/T. Suppose that φ : H → H ′ is a morphism between presentable functors,
and let K and C be the kernel and cokernel respectively, computed in the category
Ab/T. Suppose that we are given presentations A1 → A0 → H and A′
0 →
H ′. We may extend φ to a commutative diagram of maps
1 → A′
A1
φ1
A′
1
A0
φ0
/ A′
0
H
φ
/ H ′
0
/ 0
To see that we can do this, consider the element of a ∈ H ′(A0) corresponding to
the composition A0 → H → H ′. Since the map A′
0(A0) → H ′(A0) is surjective,
0(A0) mapping to a. But this element
it follows that we may find an element of A′
is exactly the data of a morphism φ0 : A0 → A′
0 making the right-hand square
commute. Now, given the existence of φ0, we consider the element b ∈ A′
0(A1)
0. It follows from the commu-
corresponding to the composition A1 → A0
tativity of the right-hand square and the exactness of the top row that b maps to
0 in H (A1). By exactness of the bottom row, it follows that there is an element of
0(A1). But, this element is exactly the data of a morphism
1(A1) mapping to b in A′
A′
φ1 : A1 → A′
1 making the diagram commute as desired.
φ0
−→ A′
We now turn to constructing presentations for K and C . Consider the commu-
tative diagram of morphisms of functors:
A1 × A′
1
A0 ×A′
0 A′
1
A1
φ1
A′
1
A′
1 × A0
A0
φ0
A′
0
/ A′
0
K
H
φ
H ′
0
0
0
/ C
/ 0
It follows from a diagram chase that the top row is a presentation for K and the
bottom row is a presentation for C .
(cid:3)
3.3. Coverings and compactness.
Definition 3.10. A morphism φ : H ′ → H with H ′, H ∈ Fun/T is called a cover
if for every T′ ∈ CAlg/T, H ′(T′) → H (T′) is surjective. It is called a pointwise
/
/
/
/
/
/
/
/
/
/
/
/
/
/
/
/
/
/
/
/
/
/
/
/
/
/
/
/
/
/
6
DANIEL KRASHEN
cover if for every field L with T-algebra structure T → L, the map H ′(L) → H (L)
is surjective.
Definition 3.11. We say that H is coverable (respectively pointwise coverable) if there
exists a cover (respectively pointwise cover) X → H with X represented by a
T-scheme of finite type.
Remark 3.12. We note that this idea of pointwise cover is close to the notion
of the sampling space for a functor introduced by O'Neil [O'N12, O'N05]. In our
definition of pointwise coverability however, we consider arbitrary field extensions
as opposed to only finite extensions as is done with sampling spaces.
Definition 3.13. Suppose we have T-functors H ′, H ′′ ⊂ H . We say that H ′ is
pointwise contained in H ′′ if for every field L with T-algebra structure T → L we
have H ′(L) ⊂ H ′′(L) as subsets of H (L).
Definition 3.14. Let H ∈ Fun/T and H ′ ⊂ H a subfunctor. We say that H ′
is pointwise quasi-compact in H if for every collection of coverable subfunctors
Ri ⊂ H , i ∈ I such that H ′ is pointwise contained in Si∈I Ri, there exists a finite
subset I′ ⊂ I such that H ′ is pointwise contained inSi∈I′ Ri.
Definition 3.15. Let H ∈ Fun/T. We say that H is localizable if given a domain
R ∈ CAlg/T with fraction field L, and a, b ∈ H (R) with aL = bL, there exists s ∈ R
with aR[s−1] = bR[s−1].
Theorem 3.16. Let T be a Noetherian ring, let X ∈ Fun/T be pointwise coverable, and
let H ∈ Fun/T be localizable. Suppose that ψ : X → H is a morphism. Then ψ(X ) is
a pointwise quasi-compact subfunctor of H .
Proof. Let φi : Ui → H , i ∈ I be a collection of morphisms from coverable functors
such that ψ(X ) is pointwise contained in Si∈I φi(Ui). We wish to show that we
may find a finite subset I′ ⊂ I with ψ(X ) also pointwise contained inSi∈I′ φi(Ui).
Since each Ui is coverable, we may replace each by an affine T-scheme Ui =
Spec(Si) (a cover), and preserve the hypothesis.
Choose a morphism from a finite type affine scheme X → X giving a pointwise
cover of X . Since X is finite type, it has a finite number of components Xi, and it
suffices to check that the image of each Xi in H is pointwise contained in a union
of a finite number of subfunctors φi(Ui) of H . In particular, we may consider each
Xi individually, and hence reduce to the case where X is irreducible. Further, since
our hypothesis only involved the field points of X, we may also assume that X is
reduced by replacing it with its reduced subscheme. In particular, we may reduce
to the case where X = X = Spec(R) for an integral domain R.
We therefore replace X by an integral affine T-scheme X = Spec(R), and proceed
by induction on the Krull dimension of X.
In the case that dim(X) = 0, we have R is a field. The hypothesis on surjectivity
on R points gives us for some i, a commutative diagram
Spec(R)
❱
❱
❱
❱
❱
❱
❱
❱
❱
❱
❱
❱
❱
❱
❱
❱
❱
❱
❱
❱
❱
❱
❱
❱
❱
❱
Ui
X
❯
❯
❯
❯
❯
❯
❯
❯
❯
❯
❯
*❯
/ H
which gives a morphism X → Ui making the diagram commute. It follows that
for every commutative T-algebra T′ (and in particular for T′ a field), we have
X(T′) ⊂ Ui(T′). We therefore obtain our result in this case by choosing I′ = {i}.
/
/
*
/
PERIOD-INDEX, SYMBOL LENGTH, AND GENERIC SPLITTINGS
7
In the general case, assume that we have shown that the conclusion holds for
affine integral schemes X when dim(X) < n, and suppose we are given an affine
integral scheme of dimension n. Let η ∈ X be the generic point, with residue
field κ(η) = frac(R). By the hypothesis, we may find an index i ∈ I and a point
q ∈ Ui(frac(R)) such that we have a commutative diagram:
Spec(frac(R))
q
❲
❲
❲
❲
❲
❲
❲
❲
❲
❲
❲
❲
❲
❲
❲
❲
❲
❲
❲
❲
❲
❲
❲
❲
❲
❲
Ui
η
,❩❩❩❩❩❩❩❩❩❩❩❩❩❩❩❩❩❩❩❩❩❩❩❩❩❩
φi
/ X
/ H
ψ
This gives a morphism η → Ui and hence a rational map from X to Ui. Choose
an open set V = Spec(R[s−1]) ⊂ X on which this map is defined. We then have
two elements of H (R[s−1]), given by V → X ψ
φi
→ H . By
the commutativity of the original diagram, these points coincide when restricted
to frac(R). Since H is localizable, it follows that we may perform an additional
localization of R[s−1] so that these two elements agree. Therefore, by adjusting our
choice of s, we may assume that we have a map V → Ui, commuting with the
morphisms to H . Now, let Z = X \ V. It follows that ψ(V) is pointwise contained
in φi(Ui).
→ H and V → Ui
Since dim Z < dim X, it follows by induction that we may find a finite subset
one easily sees that we have a pointwise containment of ψ(X) into ψ(Z) ∪ ψ(V),
I′′ ⊂ I such that ψ(Z) is pointwise contained in Sj∈I′′ φj(Uj). But therefore, since
that ψ(X) is pointwise contained in Sj∈I′′∪{i} φj(Uj), and hence setting I′ = I′′ ∪ {i},
(cid:3)
we are done.
3.4. Relative Galois cohomology functors. Let G be a finite group, T′/T a G-Galois
extension of commutative rings (as in [Sal80, DI71]), and X ∈ Fun/T′. Consider
the Weil restriction RT′/T X. By definition, this is the object in Fun/T defined by1
RT′/T(R) = X(T′ ⊗T R).
The same definition applies to Abelian functors as well, associating to an object
A ∈ Ab/T′, an object RT′/T A ∈ Ab/T.
One may check that the Weil restriction of a presentable (Abelian) functor is
presentable (by taking Weil restrictions of its presentation). It It also follows from
the definition, that the functor RT′/T A admits an action by the Galois group G
induced by the action on T′ ⊗T R.
Definition 3.17. Let GH i
Hi (G, A(T′ ⊗T R)).
Lemma 3.18. Let A ∈ Pres be a presentable functor. Then the functor GH i
presentable.
T′/T,A ∈ Ab/T denote the functor defined by GH i
T′/T,A is also
T′/T(R) =
Proof. Consider the presentable functor Cj = (cid:0)RT′/T A(cid:1)Gj
. We think of elements of
this group as functions from Gj to RT′/T A -- i.e. for a commutative T-algebra R, we
have a natural bijection
Cj(R) ↔ Map(Gj, RT′/T A(R)) = Map(Gj, A(T′ ⊗T R)),
1This discussion make sense for more general extensions T′/T, though we will not need this.
/
/
,
/
/
8
DANIEL KRASHEN
which we consider as an identification. Note that these Abelian groups carry a
natural action of the Galois group G = Gal(T′/T). We may obtain morphisms of
group schemes δ : Cj → Cj+1 by the standard formula:
(δφ)(g0, . . . , gj+1) = g0(φ(g1, . . . , gj)) +
nX
(−1)iφ(g0, . . . ,bgi, . . . , gj).
i=1
We thus obtain, for every choice of j + 1-tuple (g0, . . . , gj), a homomorphism of
group schemes Cj → RT′/T A defined by taking φ to (δφ)(g0, . . . , gj+1). Let Zj be
the intersection of the kernels of all these morphisms. Denote the corresponding
representable functors by C j, Z j. We then have, by definition of cohomology, a
short exact sequence of functors:
C i−1 → Z i → GH i
T′/T,A → 0,
showing that GH i
T′/T,A is presentable as claimed.
(cid:3)
Definition 3.19. For a commutative algebraic group A (or more generally, Abelian
functor) over T, let H i
A(R) =
Hi
A ∈ Ab/T denote the functor defined by H i
A/T = H i
´et (R, AR).
Definition 3.20. Let T be a commutative ring and R a commutative T-algebra.
Suppose we are given a finite group G, and a R-algebra R′ with G-action such that
R′ is a G-Galois extension of R. We say that R′/R is pointwise G-Galois versal if for
every T-algebra L which is a field, and for every G-Galois extension E/L, there
exists a homomorphism of T-algebras R → L such that E (cid:27) R′ ⊗R L as G-algebras.
Lemma 3.21. Let T be a commutative ring. Then there exist pointwise G-Galois versal
extensions of commutative T-algebras R′/R for every finite group G.
Proof. Let M be the free Z-module with generators xg in bijection with the elements
of G, let R′ = T[M][s−1] be the localization of the group algebra of M, where
s = Qg,h(xg − xh), endowed by the natural action of G, and R = (R′)G. This is a
G-Galois extension, by [DI71, Ch.III, Prop. 1.2(5)].
To see that R′/R is in fact pointwise G-Galois versal, let E/L be a G-Galois exten-
sion of T-algebras with L a field. By the normal basis theorem [Jac85, Theorem 4.30],
we may find a basis for E/L consisting of distinct elements yg, permuted by the
Galois group in the obvious way. Consequently, by sending xg to yg, we obtain a
homomorphism R′ → E preserving the Galois action, and hence a homomorphism
R → L. It follows immediately that L ⊗R R′ (cid:27) E as desired.
(cid:3)
Remark 3.22. It follows from the construction above that if T is Noetherian, so is
R. To see this, we note first that R′ is Noetherian by the Hilbert Basis Theorem,
since it is finitely generated over T. On the other hand, since R′/R is Galois, it
follows from [DI71, Ch.III, Prop. 1.2(3)] that R′ is finitely generated over R. By
the Eakin-Nagata theorem (see, for example [Mat89, Theorem 3.7]), we may then
conclude that R is Noetherian.
Let T be a commutative ring, G a finite group, and A ∈ Ab/T. For S′/S a
G-Galois extension of commutative T-algebras, the Hochschild-Serre spectral se-
quence ([Mil80, Theorem 2.20]):
Hp(G, Hq(S′, A)) =⇒ Hp+q(S, A)
PERIOD-INDEX, SYMBOL LENGTH, AND GENERIC SPLITTINGS
9
yields an "edge morphism" Hi(G, A) → Hi(S, A), which in the case that S′/S is
an extension of fields, coincides with the standard inflation map in Galois co-
homology. Define GH i
A/T whose S-points
consists of those classes in Hi(S, A) which may be represented as the image of a
cocycle in Hi(Gal(S′/S), A) for some Galois extension S′/S with group G. We note
that although H i
A is valued in sets and not Abelian
groups.
A ∈ Fun/T to be the subfunctor of H i
A/T is an Abelian functor, GH i
Proposition 3.23. Let T be a commutative ring, G a finite group, and A ∈ Ab/T. Then
GH i
A is pointwise coverable.
Proof. Via Lemma 3.21, we may choose a pointwise G-Galois versal extension of
commutative T-algebras R′/R. Suppose H is an affine scheme which is a pointwise
cover of GH i
R′/R,A as in Lemma 3.18. H is a R-scheme, but may also be considered
as an T scheme by virtue of the T-algebra structure of R. We obtain a morphism:
φ : H → GH i
A
as follows. If S is a commutative T-algebra, a element of H(S) yields a morphisms
Spec(S) →G H i
R′/R,A which is equivalent to giving a R-algebra structure to S (via the
structure morphism GH i
R′/R,A → Spec(R)) together with an class in Hi(G, A(R′⊗RS)).
Since R′ ⊗R S/S is a G-Galois extension of rings, and in particular an étale G-cover
of the corresponding schemes, we obtain a natural morphism Hi(G, A(R′ ⊗R S)) →
Hi(S, A) as the edge maps of the Hochschild-Serre spectral sequence Hp(G, Hq(R′ ⊗R
S, A)) =⇒ Hp+q(S, A) giving finally an element of Hi(S, A) determined by our element
of H(S).
To see that H is a pointwise cover of GH i
A, suppose that E/L is any G-Galois
extension of fields with T-algebra structure, and that α ∈ Hi(L, A) is the image of
β ∈ Hi(E/L, A(E)). By definition of R′/R, we may find a homomorphism of rings
ψ : R → L such that E/L is the pullback of R′/R along ψ -- i.e. we have an
isomorphism of G-algebras R′ ⊗R,ψ L (cid:27) E. In particular, ψ gives L the structure
of a R-algebra, and by definition, β corresponds to an L/R-point of GH i
R′/R,A. But,
therefore, by definition of H this comes from a point of H(L) which in turn maps
to α, as desired.
(cid:3)
3.5. Bounding symbol length in terms of the representation index.
Definition 3.24. Let K i
M(R) =
Z(R∗)/ ha ⊗ (1 − a) a, (1 − a) ∈ R∗i is the "naive" Milnor K-theory of R (as in [Ker10]),
T•
ℓ (R) = Ki
and let K i
ℓ denote
the subfunctors of K i
ℓ respectively consisting of those classes which may
be expressed as a sum of at most m symbols.
ℓ be defined by K i
M and K i
M be the Abelian functor K i
M(R). We let mK i
M and mK i
M(R) = Ki
M(R), where K•
M(R)/ℓKi
Remark 3.25. It follows immediately from the definition that each of the functors
mK i
M is coverable by a finite number of copies of Gm.
Remark 3.26. We have a morphism of Abelian functors Gm = K 1
µℓ induced
ℓ
by the boundary map in the long exact sequence of étale cohomology induced by
the short exact sequence
→ H 1
1 → µℓ → Gm → Gm → 1
10
DANIEL KRASHEN
Using the cup product in cohomology, this induces morphisms: K i
ℓ
i.
→ H i
µ⊗i
ℓ
for all
Definition 3.27. Let α ∈ Hd(F, µ⊗d
ℓ ), we define the representation index of α, denoted
rind(α) to be the minimum degree of a finite Galois extension E/F such that α is in
the image of the inflation map Hd(Gal(E/F), µ⊗d
ℓ (E)) → Hd(F, µ⊗d
ℓ ).
Theorem 3.28. Let α ∈ Hd(F, µ⊗d
of the representation index of α.
ℓ ). Then we may bound the length of α only as a function
is pointwise contained in the image of the unionSm>0
Proof. Fix a finite group G, and let T′/T be a pointwise G-Galois versal extension
of Z-algebras. By Proposition 3.23, the functor GH d
is a pointwise coverable
µ⊗d
ℓ
subfunctor of H d
. It follows from the Bloch-Kato conjecture that this subfunctor
µ⊗d
ℓ
ℓ . By Remark 3.25, these
are coverable, and so by Theorem 3.16 (using the fact that T is Noetherian as in
mK d
Remark 3.22), it follows that GH d
ℓ .
µ⊗d
ℓ
But this precisely says that for any cohomology class α ∈ Hd(F, µ⊗d
ℓ ) for any field
F of characteristic not divisible by ℓ, we may write α as a sum of no more than M
symbols. In particular, M depends only on the group G.
is contained in a finite union S0<m<M
mK d
In particular, given any bound on the representation index of α, we may come up
with an exhaustive finite list of groups G such that α arises as the inflation of a class
Hd(Gal(E/F), µd
ℓ) for a Galois extension E/F with group G. For each such G, since
the previous paragraph gives the existence of a bound MG for the symbol length
for such classes, we may write any such class as a sum of at most M = max{MG}
symbols.
(cid:3)
4. Bounding representation index in terms of effective indices
Theorem 4.1. Pick a positive integer n, and suppose that for every m > 0 and q ≤ n − 1
there is an integer Nq,m such that for every finite field extension L/F with [L : F] ≤ m,
and α ∈ Hq(L, µℓ), eind α ≤ Nq,m. Then every class in Hn(F, µℓ) has representation index
bounded in terms of its effective index.
We note here that since we are not focusing on constructing an optimal bound,
we are not concerned with the twist on the roots of unity; after a extension of
degree at most Φ(ℓ) we may assume µℓ ⊂ F in any case.
Before proceeding to the proof of this theorem, we first note that it quickly
implies the following:
Theorem 4.2. Let F be a field and fix a positive integer ℓ prime to the characteristic of F.
Fix d > 0. Suppose that for every m > 0, there is some Nm such that for every cohomology
class β ∈ Hd′(L, µ⊗d′
) with [L : F] ≤ m and d′ < d we have eind β ≤ Nm. Then for
α ∈ Hd(F, µ⊗d
ℓ ), we may bound λ(α) in terms of the effective index of α.
ℓ
Proof. Under these hypotheses, by Theorem 4.1, we may bound the representation
index. But by Theorem 3.28, we may therefore bound the length of α.
(cid:3)
To prove Theorem 4.1, we will need to make detailed use of the Hochschild-Serre
spectral sequence, which we now recall:
PERIOD-INDEX, SYMBOL LENGTH, AND GENERIC SPLITTINGS
11
Suppose that K/F is a G-Galois extension, G = Gal(F), N = Gal(K) (so that
G/N = G). Then we have a convergent spectral sequence
Hp(G, Hq(N, µℓ)) = Ep,q
2 =⇒ Ep+q = Hp+q(G, µℓ)
This comes with a filtration on En which we denote by F•En, with Fi+1En ⊂ FiEn,
and FiEn/Fi+1En (cid:27) Ei,n−i
∞ .
We will have to make use of this spectral sequence for different choices of
N⊳G, and the spectral
N⊳G. Note that if we have normal subgroups N′ ⊂ N
subgroups N ⊳ G. In this case, we will write the filtrations as F•
sequence terms as Ep,q
of G, we obtain a morphism of spectral sequences
N⊳G,j and En
Ep,q
N,⊳G
Ep,q
N′,⊳G
Hp(G/N, Hq(N, µℓ))
3 Hp+q(G, µℓ)
N⊳GEp+q
F•
N⊳G
Hp(G/N′, Hq(N′, µℓ))
3 Hp+q(G, µℓ)
N′⊳GEp+q
F•
N′⊳G
that is, a map on each page which commutes with the differentials, and induces a
corresponding map on the filtered parts. In this case, the maps on the E2-page are
induced by a combination of inflation and restriction: the inclusion N′ ⊂ N gives
a restriction map Hq(N, µℓ) → Hq(N′, µℓ) which then gives
Map((G/N)p+1, Hq(N, µℓ))
Map((G/N)p+1, Hq(N′, µℓ))
❱
❱
❱
❱
❱
❱
❱
❱
*❱
Map((G/N′)p+1, Hq(N′, µℓ)),
where the latter map is obtained by pre-composition. Together the resulting dashed
arrow above give homomorphisms
Hp(G/N, Hq(N, µℓ)) → Hp(G/N′, Hq(N′, µℓ)),
which gives the map on the E2 page.
proof of Theorem 4.1. Let α ∈ ker Hn(F, µℓ) → Hn(K, µℓ). Without loss of generality,
we may assume that K/F is Galois, say with Galois group G. LeteK/K be some field
extension with eK/F Galois with group G′ = G/N′, and consider the Hochschild-
N′⊳GHn(F, µℓ) (since the restriction map to Hn(eK, µℓ)G coincides
Serre spectral sequence for N′ ⊳ G. By definition, we know that our element α is
in the filtered part F1
with the map En
N′⊳G → E0,n
En,0
N′⊳G,2 = Hn(G′, µℓ) → Fn
N′⊳G,2). Since we have a surjection
N′⊳GHn(F, µℓ) = Fn
N′⊳GEn
N′⊳G,
N′⊳G.
it follows that α is in the image of a class α′ ∈ Hn(G′, µℓ) exactly when it lies in the
n'th filtered part of En
bounded in terms of [K : F] and n such that the class α is in the correct filtered part
with respect to the corresponding Hochschild-Serre spectral sequence.
To proceed, we will inductively construct our field extension eK whose degree is
For the sake of notation, let us begin with the case eK = K and let us assume that
N⊳GHn(F, µℓ), with 0 < i < n. We will show that we may find eK with
we have α ∈ Fi
+
+
/
/
*
12
DANIEL KRASHEN
α ∈ Fi+1
Since the number of such i's we must consider is at most n − 1, this will prove the
claim.
N′⊳GHn(F, µℓ) where the degree of eK/K is bounded in terms of n and [K : F].
N⊳GHn(F, µℓ) =
Consider the image α of α in the subquotient Fi
Ei,n−i
∞ . This group is the image of a subgroup
N⊳GHn(F, µℓ)/Fi+1
Zi,n−i ⊂ Ei,n−i
2
= Hi(G, Hn−i(N, µℓ))
consisting of those elements such that all the successive differentials vanish. In
particular, we may choose a representativeeα ∈ Zi,n−i ⊂ Hi(G, Hn−i(N, µℓ)) for α, and
a representative cochain foreα represented as a function from Gi+1 to Hn−i(N, µℓ) =
Hn−i(K, µℓ). By hypothesis, the effective indices of the images of this function are
bounded by Nn−i,[K:F] since n − i < n. Since we need only consider Gi+1 such
classes, we may find an extension eK/K of degree at most NGi+1
n−i,[K:F] which splits
these classes. In particular, replacing eK with its Galois closure over F, also of size
found an extension eK/K of bounded size with the image of α in the filtered part
Fi+1
N′⊳GHn(F, µℓ) as desired.
bounded inductively in terms of the previous extension degree [K : F], we have
(cid:3)
5. Cohomology at 2, Pfister numbers and index bounds
For Galois cohomology with coefficients in µ2, we may use the tools of quadratic
form theory to obtain, in certain cases, bounds on the indices of cohomology classes,
in particular, in terms of the u-invariant. These results are fairly direct applications
of the Milnor conjectures and other results in the theory of quadratic forms.
We will have use of the following theorem proved by Karpenko, originally
conjectured by Vishik:
Theorem 5.1 ([Kar04], conjecture 1.1). If φ is an anisotropic quadratic form such that
φ ∈ In ⊂ W(F), where I is the fundamental ideal of the Witt ring, and dim(φ) < 2n+1, then
dim(φ) = 2n+1 − 2i for 1 ≤ i ≤ n + 1.
Note that by convention, we consider the 0 form to be anisotropic.
Proposition 5.2. Suppose F is a field of characteristic not equal to 2 and with u invariant
u(F) ≤ 2n (for example, a Cn field), and α ∈ Hm(F, µ2). Then
(1) If m > n then α = 0. In other words, cd2(F) ≤ n.
(2) If m = n then eind(α)2.
(3) If m < n, then eind(α)22n−1−2m+m+1.
Proof. In each case, by the Milnor conjecture ([OVV07] for characteristic 0, or
Merkurjev's article in [Mil10] for the case of general characteristic not 2), we may
choose a quadratic form φ ∈ Im such that em(φ) = α. By our assumption on the u(F),
we may represent φ by an anisotropic form of dimension at most 2n. By theorem
5.1, if dim(φ) < 2m+1 it must be of the form 2m+1 − 2i for some 1 ≤ i ≤ m + 1.
In case 1, we know that φ may be represented by an anisotropic form of dimen-
sion less than 2n, which is in turn less than 2m. However, 2m+1 − 2i is either 0 or
greater than or equal to 2n. Therefore φ = 0 and α = em(φ) = 0 as well.
Alternately, since u(F) ≤ 2n, every m-fold Pfister form is trivial. Since Im is
generated by m-fold Pfister forms, it must be trivial. Therefore, φ = 0 and α = 0.
PERIOD-INDEX, SYMBOL LENGTH, AND GENERIC SPLITTINGS
13
In case 2, the dimension of an anisotropic representation of φ has dimension at
most 2n, and must be of the form 2n+1 − 2i. Therefore it is either 0 or of dimension
exactly 2n+1 − 2n = 2n. If we then choose a quadratic extension L/F such that φL is
isotropic, then its anisotropic part will be a strictly smaller dimensional form in In,
which again by theorem 5.1 immediately forces it to be 0. Therefore eind(α)2.
In case 3, note that we may always find a tower of quadratic extensions such
that each successive extension lowers the dimension of the anisotropic part of φ
by at least 2. Since the anisotropic part has dimension at most 2n, we may find a
field extension L/F which is a tower of
2n − (2m+1 − 2)
2
= 2n−1 − 2m + 1
many quadratic extensions such that φL has anisotropic part at most 2m+1 − 2. By
lemma 5.3, φL is split by an extension of degree dividing 2m. Therefore, φ is split
by an extension of degree
2m22n−1−(2m−1) = 22n−1−2m+m+1.
(cid:3)
Lemma 5.3. Suppose ψ ∈ Im has dimension at most 2m+1 − 2i. Then there is an extension
K/F of degree dividing 2m+1−i which splits ψ.
Proof. Choose any quadratic extension E/F such that ψE is isotropic. By theorem
5.1, its anisotropic part must have dimension at most 2m+1 −2i+1 and so by induction
it is split by an extension K/E of degree dividing 2m+1−(i+1) = 2m−i. But this implies
that [K : F] divides 2m−i+1 and we are done.
(cid:3)
It seems extremely unlikely that the bounds obtained above on the effective
index are the optimal ones. Rather, it seems more reasonable to expect the more
optimistic conjecture, which came out of discussions at the AIM conference "De-
formation theory, patching, quadratic forms, and the Brauer group":
Conjecture 5.4. Suppose α ∈ Hn(F, µ⊗n
by d. Then
ℓ ), and the Diophantine dimension of F is bounded
eind(α)ℓ(d−1
n−1).
Let us describe briefly some circumstantial evidence for this. The main support
for this conjecture comes from the case ℓ = 2. If we suppose that F with Diophantine
dimension bounded by d, then it follows that u(F) = 2d. An anisotropic form would
then be at most dimension 2d, and in this case, it would be Witt equivalent to an
orthogonal sum q (cid:27) b1 ⊥ · · · ⊥ b2n−1 for binary forms bi. At worst, this could
be split by individually splitting each of the binary forms in successive quadratic
extensions, yielding a splitting field of degree 22d−1. On the other hand, another
way to split the form would be to split each of its cohomological invariants, starting
from e1 all the way to ed. If we assume that 2ni is the effective index of the ei invariant
obtained at the i'th stage, then we should obtain a splitting field of degree 2Pd
i=1 ni.
Assuming that the ei invariants are "independent," and "general," we expect then
that 2ni should represent the maximal effective index of a class in Hi(F, µ2) (implicitly
assuming that this bound should stay the same over finite extensions of F). We
14
DANIEL KRASHEN
therefore expect that if 2ni is the maximal effective index of a class in Hi(F, µ2), then
Qd
i=1 2ni = 22d−1. In other words
n1 + n2 + · · · + nd = 2d−1
Further, we have n1 = 1 = nd from Theorem 5.2, and conjecturally n2 = d − 1 (see
[Lie11, page 2], where it is attributed to a question of Colliot-Thélène). Besides
this, the conjecture that ni should at worst be (cid:0)d−1
i−1(cid:1) is simply based on the above
numerology and the feeling that in some sense the cohomology ring "feels a lot
like an exterior algebra."
Recall that the m-Pfister number of a quadratic form q, denoted Pf m(q) is the
smallest number j such that q is Witt equivalent to a sum of φ1 ⊥ φ2 ⊥ · · · ⊥ φj
of m-fold Pfister forms. This terminology is due to Brosnan, Reichstein and Vistoli
[BRV10, Section 4].
Theorem 5.5. Suppose that F is a field of characteristic not 2 in which −1 ∈ (F∗)2. Then
the following are equivalent:
(1) u(F) < ∞.
(2) For all m > 0, there exists Nm > 0 such that for all n > 0 and for all α ∈ Hn(L, µ2)
with [L : F] ≤ m, we have eind α < Nm.
(3) There exists N > 0 such that for all n > 0 and for all α ∈ Hn(F, µ2), we have
λ(α) < N.
(4) There exists N > 0 such that for all n > 0 and for all q ∈ In(F), we have Pf n(q) < N.
Proof.
(1) =⇒ (2): By [Lee84], u(F) < ∞ implies that for every finite extension L/F, u(L) is
bounded in terms of the extension degree. In particular, without loss of generality
we may assume L = F, and so we must show that if u(F) < ∞ then eind(α) is
universally bounded for α ∈ Hn(F, µ2) for all n. But this follows directly from
Proposition 5.2.
(2) =⇒ (3): Our hypothesis implies that, in particular, eind α < N1 for α ∈ H2(F, µ2).
It follows from the remark just after Proposition 2.5, that we may bound the 2-
cohomological dimension of F. But then, Theorem 4.2 applies in each of the finite
number of relevant degrees to give a bound on the length of any mod-2 cohomology
class.
(3) =⇒ (4): It follows from Proposition 2.5 that we may bound the 2-cohomological
dimension of F. Say Hm(F, µ2) = 0 for some m > 0. Let
Pf n(F) = max
q∈In(F)
{Pf n(q)}.
We proceed to show that Pf n < ∞ for all n by descending induction on n. To begin,
we know that since Hm(F, µ2) = 0, it follows from the Milnor Conjectures (or the
mod-2 case of the Bloch-Kato Conjecture), proved in [Voe03b, Voe03a] (see also
Merkurjev's article in [Mil10] for general characteristic not 2), or more generally
the Bloch-Kato conjecture/Norm residue isomorphism theorem [Voe11, Wei09],
that Hm′(F, µ2) = 0 and therefore, that Im′(F) = 0 for all m′ ≥ m since the canonical
map from Milnor K-theory to the graded Witt ring is surjective (see [Mil70]). Now,
suppose that we have shown Pf n < ∞. Consider q ∈ In−1(F). Since λ(en−1(q)) < N,
we may write q as q = q′ + r where r is a sum of at most N (n − 1)-fold Pfister forms
and where q′ ∈ In(q). By induction, we know that Pf n(q′) < Pf n(F) < ∞, and so we
ri. But if n > 1, each ri is the sum of three (n − 1)-fold Pfister
may write r = PPf n(F)
i=1
PERIOD-INDEX, SYMBOL LENGTH, AND GENERIC SPLITTINGS
15
forms: if we write ri =< 1, −a >< 1, −b > φ, where φ is an (n − 2)-fold Pfister form
(or 1 if n = 2), then since we have (using the fact that < 1, 1 >=< 1, −1 >, since
−1 ∈ (F∗)2):
< 1, −a >< 1, −b >=< 1, −a >⊥< −b, ab >∼< 1, −a >⊥< 1, −b >⊥< 1, ab >
and so ri =< 1, −a > φ ⊥< 1, −b > φ ⊥< 1, ab > φ. Therefore, we have Pf n(F) <
N + 3 Pf n−1(F) < ∞, as desired.
In the case n = 1, it is immediate that every form in I(F) is the sum of a binary
form and one in I2(F). Using the argument of the preceding paragraph, one has
Pf 1(F) ≤ 1 + 3 Pf 2.
(4) =⇒ (1): This follows immediately upon considering the case n = 1, since every
form differs from a form in I(F) by at most a 1-dimensional form.
(cid:3)
6. Generic splitting varieties in degree 2
Suppose H is a functor from the category of F-algebras to the category of
Abelian groups, and let α ∈ H (F).
Definition 6.1. We say that a variety (resp. scheme) X/F is a generic splitting
variety (resp. scheme) for α if for every field extension L/F, we have X(L) , ∅ if
and only if αL = 0.
We follow [CTS87, 0.1] and use the term multiplicative type to mean a finite
type scheme over F which is diagonalizable after extending scalars to a suitable
separable extension L/F (a finite type isotrivial group scheme in the language of
[ABD+64, X]). In particular, this includes the cases of smooth finite group schemes
as well as tori.
In the case that a functor is (almost) presentable, we may in fact construct such
generic splitting schemes easily, and this will be the basis of our construction:
Proposition 6.2. Suppose H ′ ∈ Ab/F is presentable, and let H be any Abelian functor
together with a map φ : H ′ → H which is an injection on L-points for every L/F a field
extension. Let α ∈ H (F) be the image of α′ ∈ H ′(F) under φ. Let
A1
g
/ A0
f
/ H ′
/ 0
be a presentation andeα ∈ A0(F) be a preimage of α′. Then X = g−1(eα) is a generic splitting
scheme for α′.
Proof. This follows essentially immediately from the definitions of presentability
and the injectivity of φ. Since the map f is surjective as a map of functors, it follows
that the preimage exists, and further, for a given field extension L/F, we will have
α′
exactly when X(L) is nonempty. Since α′
φ, we are done.
L = f (eαL) = 0 exactly when eαL is the image of a class in A1(L), which is to say,
L = 0 if and only if αL = 0 by injectivity of
(cid:3)
Theorem 6.3. Let A be any commutative group scheme over F of multiplicative type and
suppose α ∈ H2(F, A). Then there exists a smooth generic splitting variety X/F for α.
Proof. By the definition of multiplicative type and the Weil restriction, we may find
an separable extension L/F and an embedding of A into RL/F Gn
m.
/
/
/
16
DANIEL KRASHEN
Now, choose a field extension E/F such that αE = 0, and consider the composition
We then find that, using Shapiro's lemma as in [KMRT98, VII(29.6)], that the image
lemma, also in the context of étale algebras (see [KMRT98, VII(29.7)])
of embeddings
A → RL/F Gn
of α in H2(F, T) = H2(cid:16)E,(cid:0)RL/F Gn
H1(F, T) = H1(cid:16)E,(cid:0)RL/F Gn
m(cid:1)E(cid:17) .
m → T = RE/F(cid:16)(cid:0)RL/F Gn
m(cid:1)E(cid:17) is trivial. Further we have, again by Shapiro's
m(cid:1)E(cid:17) = H1(cid:0)E, RL⊗E/E Gn
m(cid:1) = H1(L ⊗ E, Gn
m) = 0
using Hilbert's theorem 90. In particular, if we let T′ = T/A be the quotient torus,
then using the exact sequence in Galois cohomology, we obtain:
0 → H1(F, T′) → H2(F, A) → H2(F, T)
Which implies, in particular, that the image of H1(F, T′) in H2(F, A) contains the
element α.
Using [CTS87, Proposition 1.3(1.3.1)], we may find an exact sequence of tori
0 → T′ → P → Q → 0
where P = RK/F Gm for K/F an étale algebra is a quasitrivial torus and Q is coflasque.
In particular, we obtain for every F-algebra T, an exact sequence:
P(T)
f
→ Q(T) → H1(T, T′) → H1(T, P)
in particular, if we let H ′ be the quotient cokernel functor (not sheaf!) of P → Q,
and H ∈ Ab/F be defined by H (T) = H2(T, A), then we obtain a morphism
H ′ → H via the composition
Q(T) → H1(T, T′) → H2(T, A)
which, using the fact that P is quasitrivial, induces a injection H ′(L) → H (L) for
every field extension L/F. Since α is in the image of this map on F-points, say
of a point α′ ∈ H ′(F), it follows that φ−1(eα) is a generic splitting scheme for α,
whereeα is a preimage of α′ in Q(F). But from the definition of H ′, it follows from
Proposition 6.2 that X = φ−1(eα) is a generic splitting scheme for α. But since X is a
homogeneous variety under the action of the torus P, it follows that it is a smooth
variety as desired.
(cid:3)
References
[CTS87]
[BRV10]
[ABD+64] M. Artin, J. E. Bertin, M. Demazure, P. Gabriel, A. Grothendieck, M. Raynaud, and J.-P. Serre.
Schémas en groupes. Fasc. 3: Exposés 8 à 11, volume 1963/64 of Séminaire de Géométrie Algébrique
de l'Institut des Hautes Études Scientifiques. Institut des Hautes Études Scientifiques, Paris,
1964.
Patrick Brosnan, Zinovy Reichstein, and Angelo Vistoli. Essential dimension, spinor groups,
and quadratic forms. Ann. of Math. (2), 171(1):533 -- 544, 2010.
Jean-Louis Colliot-Thélène and Jean-Jacques Sansuc. Principal homogeneous spaces under
flasque tori: applications. J. Algebra, 106(1):148 -- 205, 1987.
Frank DeMeyer and Edward Ingraham. Separable algebras over commutative rings. Lecture
Notes in Mathematics, Vol. 181. Springer-Verlag, Berlin, 1971.
Christian Haesemeyer and Chuck Weibel. Norm varieties and the chain lemma (after
Markus Rost). In Algebraic topology, volume 4 of Abel Symp., pages 95 -- 130. Springer, Berlin,
2009.
Nathan Jacobson. Basic algebra. I. W. H. Freeman and Company, New York, second edition,
1985.
Bruno Kahn. Comparison of some field invariants. J. Algebra, 232(2):485 -- 492, 2000.
[HW09]
[Kah00]
[DI71]
[Jac85]
PERIOD-INDEX, SYMBOL LENGTH, AND GENERIC SPLITTINGS
17
[Kar04]
[Ker10] Moritz Kerz. Milnor K-theory of local rings with finite residue fields. J. Algebraic Geom.,
Nikita A. Karpenko. Holes in In. Ann. Sci. École Norm. Sup. (4), 37(6):973 -- 1002, 2004.
19(1):173 -- 191, 2010.
[KMRT98] Max-Albert Knus, Alexander Merkurjev, Markus Rost, and Jean-Pierre Tignol. The book of
involutions. American Mathematical Society, Providence, RI, 1998. With a preface in French
by J. Tits.
David B. Leep. Systems of quadratic forms. J. Reine Angew. Math., 350:109 -- 116, 1984.
Max Lieblich. Period and index in the Brauer group of an arithmetic surface. J. Reine Angew.
Math., 659:1 -- 41, 2011. With an appendix by Daniel Krashen.
[Lee84]
[Lie11]
[Mil70]
[Mil80]
[Mat89] Hideyuki Matsumura. Commutative ring theory, volume 8 of Cambridge Studies in Advanced
Mathematics. Cambridge University Press, Cambridge, second edition, 1989. Translated from
the Japanese by M. Reid.
John Milnor. Algebraic K-theory and quadratic forms. Invent. Math., 9:318 -- 344, 1969/1970.
James S. Milne. Étale cohomology, volume 33 of Princeton Mathematical Series. Princeton
University Press, Princeton, N.J., 1980.
John Milnor. Collected papers of John Milnor. V. Algebra. American Mathematical Society,
Providence, RI, 2010. Edited by Hyman Bass and T. Y. Lam.
A. S. Merkur′ev and A. A. Suslin. K-cohomology of Severi-Brauer varieties and the norm
residue homomorphism. Izv. Akad. Nauk SSSR Ser. Mat., 46(5):1011 -- 1046, 1135 -- 1136, 1982.
Catherine O'Neil. Models of some genus one curves with applications to descent. J. Number
Theory, 112(2):369 -- 385, 2005.
Catherine O'Neil. Sampling spaces and arithmetic dimension. In Number theory, analysis and
geometry, pages 499 -- 518. Springer, New York, 2012.
[O'N12]
[O'N05]
[Mil10]
[MS82]
[OVV07] D. Orlov, A. Vishik, and V. Voevodsky. An exact sequence for KM
∗ /2 with applications to
[Rev73]
[Sal80]
[Sal11]
[SJ06]
quadratic forms. Ann. of Math. (2), 165(1):1 -- 13, 2007.
Philippe Revoy. Formes alternées et puissances divisées. In Séminaire P. Dubreil, 26e année
(1972/73), Algèbre, Exp. No. 8, page 10. Secrétariat Mathématique, Paris, 1973.
David J. Saltman. Generic Galois extensions. Proc. Nat. Acad. Sci. U.S.A., 77(3, part 1):1250 --
1251, 1980.
David J Saltman. Finite u invariant and bounds on cohomology symbol lengths. 2011.
Andrei Suslin and Seva Joukhovitski. Norm varieties. J. Pure Appl. Algebra, 206(1-2):245 -- 276,
2006.
[Voe03a] Vladimir Voevodsky. Motivic cohomology with Z/2-coefficients. Publ. Math. Inst. Hautes
Études Sci., (98):59 -- 104, 2003.
[Voe03b] Vladimir Voevodsky. Reduced power operations in motivic cohomology. Publ. Math. Inst.
[Voe11]
[Wei09]
Hautes Études Sci., (98):1 -- 57, 2003.
Vladimir Voevodsky. On motivic cohomology with Z/l-coefficients. Ann. of Math. (2),
174(1):401 -- 438, 2011.
C. Weibel. The norm residue isomorphism theorem. J. Topol., 2(2):346 -- 372, 2009.
|
1908.03235 | 1 | 1908 | 2019-08-08T18:33:03 | Bioperational Multisets in Various Semi-rings | [
"math.RA",
"math.NT"
] | One can find lists of whole numbers having equal sum and product. We call such a creature a bioperational multiset. No one seems to have seriously studied them in areas outside whole numbers such as the rationals, Gaussian integers, or semi-rings. We enumerate all possible sum-products for a bioperational multiset over whole numbers and six additional domains. | math.RA | math |
Bioperational Multisets in Various Semi-rings
O.M. Cain
[email protected]
July 2019
Abstract. One can find lists of whole numbers having equal sum and prod-
uct. We call such a creature a bioperational multiset. No one seems to have
seriously studied them in areas outside whole numbers such as the rationals,
Gaussian integers, or semi-rings. We enumerate all possible sum-products for a
bioperational multiset over whole numbers and six additional domains.
1 Introduction
The numbers 1, 2, and 3 have the property that their sum is also their product.
That is, 1 + 2 + 3 = 1 · 2 · 3 = 6. As Matt Parker [1] has pointed out, this gives
us as a strange sort of byproduct:
log(1 + 2 + 3) = log 1 + log 2 + log 3,
due to the identity
log(ab) = log a + log b.
We coin the term bioperational here to refer to any such multiset {ai}n
having an equal sum and product. That is to say,
i=1
n
X
i=1
ai =
n
Y
i=1
ai.
A multiset, as can be guessed, is a set in which a number can occur multiple
times (instead of just once or not at all).
There are some open conjectures about the number of bioperational multisets
over N of length n as n gets bigger and bigger [2]. There is also a smattering
of analysis available on math.stackexchange including uniqueness of solutions
and connections with trigonometry [3][4][5][6]. One can also find a surprisingly
complicated solution algorithm [7]. But it seems no one has formally categorized
bioperational multisets over N (unless we count Matt's "pseudoproof" and the
passing comments of others).
In addition to N, it is interesting to explore bioperational multisets in other
environments. They are well-defined anywhere addition and multiplication are
1
well-defined (hence in any semi-ring). We would suspect therefore to find these
creatures lurking in Z, C, Fp, GLn(R), and many other places (even in non-
Abelian rings!).
In this paper, we limit ourselves to analyzing bioperational
multisets over
• non-negative integers (N) in Section 3,
• integers (Z) in Section 4,
• general fields (Q, Fp, etc) in Section 5,
• lunar integers (L) in Section 6,
• Gaussian integers (Z[i]) in Section 7,
• Eisenstein integers (Z[ω]) in Section 8,
• and integers with √2 appended (Z[√2]) in Section 9.
2 Some Definitions
Firstly, we have to blow some dead leaves out of the way to see clearly. To do so
requires the leafblower of vocabulary. Suppose we have a bioperational multiset
S = {ai}n
i=1. We say
• S is trivial if it contains only one element,
• S vanishes if the sum-product is zero, and
• S is minimal if no proper subset of its terms forms a bioperational set of
the same sum-product.
Note that we are using 'trivial' differently than in [2].
All the examples we are considering are 1) integral domains and 2) Abelian.
That means 1) if any two numbers have a product of zero then one or both of
them must also be zero (ab = 0 implies a = 0 or b = 0) And 2) the order we
multiply stuff doesn't matter (so ab = ba). This has some consequences on our
analysis.
1) In an integral domain, any vanishing bioperational multisets must contain
zero -- which is rather boring. An analysis of vanishing bioperational multisets
over non-integral domains might be interesting. In fact, the first example (1, 2, 3)
vanishes if we place it in Z/6Z. But the adventure of non-integral domains in
general will be neglected here.
2) Since the order of multiplication doesn't matter in our examples, we call
our subjects bioperational multisets. However, the author greatly hopes that
bioperational multisets will be explored in non-Abelian rings (in which case they
would be bioperational sequences). The author would have loved to explore
these objects in the quaternions (H) themselves but was too ignorant for the
attempt.
2
For two multisets A and B we let "A + B" denote their multiset sum which
is best explained with an example:
{2, 7, 2, 2, 3} + {1, 2, 7, 7} = {1, 2, 2, 2, 2, 3, 7, 7, 7}.
In technical terms, we are summing the multiplicities of all elements involved.
Similarly "A − B" will denote subtracting multiplicites:
{2, 7, 2, 2, 3} − {2, 2, 3} = {2, 7}.
We also use coefficients of multisets to denote scaling multiplicities. Or in other
words
3{2, 5, 5} = {2, 2, 2, 5, 5, 5, 5, 5, 5}.
Lastly, to keep our equations less messy, for a multiset S we will write σ(S)
for the sum of its elements and π(S) for the product. They look nicer than
Pn
i=1 ai and Qn
i=1 ai.
3 Non-negative integers (N)
We begin with
Theorem 3.1. There is exactly one non-vanishing bioperational multiset over
N of length n for n = 2, 3, 4 with constructions
2 + 2 = 2 · 2 = 4,
1 + 2 + 3 = 1 · 2 · 3 = 6,
1 + 1 + 2 + 4 = 1 · 1 · 2 · 4 = 8.
and
This confirms Matt's conjecture for n = 2 and is stated without proof in [2].
Proof. We first take n = 2. Suppose we have ab = a + b. Rearrangement yields
ab − a − b + 1 = (a − 1)(b − 1) = 1. Since the only way to factor 1 as two
positive integers is 1 = 1 · 1 it follows that a − 1 = b − 1 = 1 or equivalently,
that a = b = 2.
A phenomenal proof of n = 3 was given by Mark Bennet in [3]. We repeat
it here. Suppose we have a + b + c = abc. At least one term must be 1 since if
otherwise a ≥ b ≥ c ≥ 2 and we would have
3a ≥ a + b + c = abc ≥ 4a
which is true only if a ≤ 0. But that is a contractiction since we are assuming
a ≥ 2..
Okay, so let c = 1. Then we have a + b + 1 = ab. Rearranging yields
ab − a − b + 1 = (a − 1)(b − 1) = 2. It follows that {a − 1, b − 1} = {1, 2} or
equivalently, that {a, b} = {2, 3}.
3
Similarly for n = 4, suppose a + b + c + d = abcd. Similar to the case n = 3,
we know at least one term must be 1 since a ≥ b ≥ c ≥ d ≥ 2 would imply
4a ≥ a + b + c = abc ≥ 8a.
So let d = 1. But we can play the same trick again. If we have a ≥ b ≥ c ≥ 2
then
3a + 1 ≥ a + b + c + 1 = abc ≥ 4a
from which it follows that a ≤ 1 which again contradicts our assumption a ≥ 2.
Let c = 1 also.
We have a + b + 2 = ab. Rearranging yields (a − 1)(b − 1) = 3 from which
it follows {a, b} = {4, 2}.
One may be tempted to generalize this proof technique and keep tackling
larger and larger n (In fact, we wrote a program to do exactly this [7]. See [8]
for another such solution algorithm). For example, n = 5 yields
Theorem 3.2. There are 3 non-vanishing bioperational multisets over N of
length n = 5 with constructions
1 + 1 + 2 + 2 + 2 = 1 · 1 · 2 · 2 · 2 = 8,
1 + 1 + 1 + 3 + 3 = 1 · 1 · 1 · 3 · 3 = 9,
and 1 + 1 + 1 + 2 + 5 = 1 · 1 · 1 · 2 · 5 = 10.
Proof. From computation.
But there turns out to be an easier way to catalog all bioperational multisets.
We first need a lower foothold (or "lemma" as they're called).
Lemma 3.3. The product of one or more real numbers, all greater than or equal
to 2, is greater than or equal to their sum. That is, if ai ≥ 2, i = 1, .., n then
n
Y
i=1
ai ≥
n
X
i=1
ai.
Next suppose we have some multiset S = {ai}n
Proof. Induction will be used on n. The base case, n = 1, of a single number is
clearly true since every number is equal to itself -- and therefore is greater than
or equal to itself (a1 ≥ a1).
i=1 for which the theorem
statement is true. So π(S) ≥ σ(S). We have to show the theorem true for a new
multiset S′ formed by appending a new element an+1 ≥ 2 since any multiset
can be built up one element at a time.
Let k be the largest integer such that an+1 > π(S)k. From this we grab two
crimps,
an+1 − 1 ≥ π(S)k
and an+1 < π(S)k+1,
4
with which the last bit of the proof can be shown easily.
π(S′) = an+1π(S) = (an+1 − 1)π(S) + π(S) ≥ π(S)kπ(S) + σ(S)
= π(S)k+1 + σ(S) > an+1 + σ(S) = σ(S′)
Technically this is a bit overkill since we've shown π(S′) > σ(S′) when all we
needed was π(S′) ≥ σ(S′). Oh well.
With that Lemma we can catalog all bioperational multisets over N by their
sum-product.
Theorem 3.4. For every composite integer m ∈ N there exists a non-trivial
bioperational multiset over N with a sum-product of m.
Proof. Suppose a composite integer m = a1a2...ak with k > 1 and ai ≥ 2 for
i = 1, ..., k. Let S = {ai}k
i=1. By Lemma 3.3, we know π(S) ≥ σ(S). So let the
non-negative integer d = π(S) − σ(S) be their difference. The multiset
S′ = {a1, a2, ..., ak,
d times
z } {
1, ..., 1}
is bioperational with sum-product m since
d times
σ(S′) = a1 + ... + ak +
z
1 + ... + 1 = σ(S) + (π(S) − σ(S)) = π(S) = π(S′) = m.
}
{
From the proof of the previous theorem we can also make a statement about
the lengths of bioperational multisets.
Corollary 3.4.1. For every factorization of a composite integer m = a1a2...ak
there exists a non-vanishing bioperational multiset over N of length m+k−P ai.
Proof. Let S and S′ denote the same multisets as in the proof of Theorem 3.4.
S′ is bioperational and contains k + d = k + (π(S) − σ(S)) = m + k − P ai
elements.
Starting at n = 2 the number of non-vanishing bioperational multisets over
N of length n is
1, 1, 1, 3, 1, 2, 2, 2, 2, 3, 2, 4, 2, ...
(sequence A033178 in OEIS [8]). The positions of record in this list occur at
n = 2, 5, 13, 25, 37, 41, 61, 85, 113, 181, 361, 421, 433, ...
(sequence A309230 in OEIS). The terms all appear to have fewer prime factors
than their neighbors.
5
4 Integers (Z)
An interesting thing happens once negatives are on the playing field. A multiset
can be extended without changing either sum or product. Consider
S = {1, 2, 3,−1,−1, 1, 1}
which is bioperational with sum-product 6. This is the first example of a non-
minimal bioperational multiset. In N every non-vanishing bioperational multiset
is also minimal. Not so in Z!
Accordingly, we now use bioperation as a verb as well. We say a multiset
has been bioperated if it has been made bioperational by means of changing its
sum with appendages. For example, we may bioperate S = {3,−5} which has
a sum σ(S) = −2 and product π(S) = −15. Since appending T = {−1,−1, 1}
decrements σ(S) and fixes π(S), bioperation is accomplished by just repeatedly
appending T . In particular,
S′ = S + 13T
is bioperational. Note however S′ is not minimal. We can trim it down to
minimality by shaving off groups of {−1,−1, 1, 1} which have no effect on the
sum-product obtaining
14 times
S′′ = S + 13T − 6{−1,−1, 1, 1, 1} = {3,−5,
z
−1, ...,−1, 1}
}
{
which is, in fact, minimal.
There are three important appendages in Z which fix the product.
label
T1
T0
T−1
appendage
∆σ(S)
{1}
{1, 1,−1,−1}
{1,−1,−1}
+1
0
−1
We now give the parallel of Theorem 3.4 for Z.
Theorem 4.1. For every composite integer m ∈ Z there exists a non-trivial
minimal bioperational multiset over Z with a sum-product of m.
Proof. Choose a factorization m = a1...an with n ≥ 2 where each ai may be
positive or negative and ai ≥ 2 for i = 1, ..., n. The multiset S = {ai}n
i=1 has
the desired product. Bioperate S producing S′ such that σ(S′) = π(S′) = π(S).
This is done by appending T±1 as needed. To be
Finally, if S′ is not minimal we may take a minimal bioperational multiset,
S′′, from it. S′′ must include the non-units a1, ..., an (a "unit" by the way is a
fancy name for a number with an inverse in its same ring; in this case 1 and
−1). Since n ≥ 2 we are assured that S′′ is non-trivial.
6
5 Fields
Bioperational multisets turn out disappointingly abundant in fields.
Lemma 5.1. Given any multiset S = {ai}n
i=1 whose elements are in a field
F and such that π(S) 6= 1, one can bioperate S into a unique multiset S′ by
appending a single element,
an+1 =
σ(S)
π(S) − 1
.
This was stated for F = Q and n = 4 by Robert Israel in [6].
Proof. Any element an+1 ∈ F which might bioperate S must satisfy σ(S) +
an+1 = π(S)an+1. Rearranging yields an+1 = σ(S)
π(S)−1 which exists if π(S) 6=
1.
The lemma turns out to be an exhaustive description.
Theorem 5.2. In any field, all non-trivial bioperational multisets can be pro-
duced with Lemma 5.1.
Proof. Suppose we have some bioperational multiset S = {ai}n
i=1 which cannot
be produced by the lemma. Let S′i be the multiset formed by removing ai. It
follows from the lemma π(S′i) = π(S)
= 1 for all i = 1, ..., n. This in turn implies
π(S) = ai for i = 1, ..., n and we see that all ai are equal. We therefore have a
solution to
ai
an
1 = na1.
But dividing out an a1 from both sides gives us n = an−1
S is trivial.
1 = π(S′1) = 1 showing
Before leaving this territory, we note that there are solutions to an−1 = n
leading to bioperational sets of a single value. Take for instance {2, 2, 2, 2, 2}
which is bioperational in F11.
6 Lunar Integers (L)
The Lunar Integers are the only strictly semi-ring to be considered. Their
arithmetic is well analyzed in [10] (there called "Dismal" Arithmetic) and Neil
Sloane gives a wonderful introduction in a Numberphile interview [11].
We neglect to explain the arithmetic here ourselves. We need only note some
properties of the number of digits. If we let D(a) denote the number of digits
of a lunar integer a ∈ L, then
D(ab) = D(a) + D(b) − 1
and D(a + b) = max{D(a), D(b)}.
These give us
7
Lemma 6.1. In any Lunar Bioperational Set, there is at most one element with
2 or more digits.
Proof. Suppose S = {ai}n
i = 2, ..., n. From the aforementioned identities
i=1 ⊂ L is bioperational and that D(a1) ≥ D(ai) for
D(σ(S)) = max{D(ai)}n
i=1 = D(a1)
and
D(π(S)) = 1 +
n
X
i=1
(D(ai) − 1) = D(a1) +
n
X
i=2
(D(ai) − 1).
Since D(π(S)) = D(σ(S)), it follows that Pn
D(ai) = 1 for i = 2, ..., n.
i=2(D(ai) − 1) = 0 and hence that
Apparently bioperational multisets can't breathe well on the moon:
Theorem 6.2. Every minimal bioperational multiset of Lunar integers is triv-
ial.
Proof. We prove the contrapositive. Suppose S = {ai}n
i=1 ⊂ L bioperational
and non-trivial. From Lemma 6.1 we may assume D(ai) = 1 for i = 2, ..., n.
For a ∈ L, let F (a) ∈ L denote be the last digit of a. From the definitions of
addition and multiplication over L
max{F (a1), a2, ..., an} = F (σ(S)) = F (π(S)) = min{F (ai), a2, ..., an}.
But this implies F (a1) = a2 = ... = an. In which case the multiset S′ = {a1}
is trivially bioperational with the same sum-product as S and hence S is not
minimal.
So there are bioperational multisets in L, like {17, 7} and {2, 2, 2}, but they
aren't very interesting.
7 Gaussian Integers (Z[i])
Gaussian integers are numbers of the form a + bi where a and b are integers and
i2 = −1 (so like 2 + 3i or −1 − 19i for example). In addition to the appendages
T−1, T0, and T1 given in Section 4, two more appear in Z[i],
T±2i = {±i,±i,−1, 1},
which perturb the sum by ±2i and fix the product. So sometimes bioperate the
imaginary part of a multiset sum.
Take, for instance, S = {1 + 2i, 2 + 3i}. We have
σ(S) = 3 + 5i and π(S) = −4 + 7i.
8
The difference is π(S) − σ(S) = −7 + 2i. We bioperate by 1) appending T−1
seven times, 2) appending T2i once, and 3) shaving off T0 until minimality is
reached. The result is
S′ = {1 + 2i, 2 + 3i, i, i,−1,−1,−1,−1,−1,−1,−1}
which is bioperational with π(S′) = σ(S′) = −4 + 7i.
We need a couple lemmas before the result analogous to Theorem 3.4.
Lemma 7.1. A Gaussian integer a + bi is a multiple of 1 + i if and only if a
and b have the same parity (that is, are both odd or both even).
Proof. Firstly, suppose a + bi = (1 + i)(c + di) is a multiple of 1 + i. Then
a = c− d and b = c + d. a and b therefore have the same parity since b = a + 2d.
Conversely, suppose a and b have the same parity. If both even, then we
may write
a + bi = 2(cid:16) a
2
+
b
2
i(cid:17) = (1 + i)(1 − i)(cid:16) a
2
+
i(cid:17)
b
2
and are done. If both odd, then we may write
a + bi = (1 + i)(cid:16) a + b
2
+
b − a
2
i(cid:17).
Lemma 7.2. For any Gaussian integers α1, ..., αn ∈ Z[i] such that 1 + i does
not divide any αi,
Im(cid:16)Y αi(cid:17) ≡ Im(cid:16)X αi(cid:17) mod 2.
Proof. Let ϕ(a + bi) = b % 2 ∈ F2. It follows that ϕ(α + β) = ϕ(α) + ϕ(β). But
more interestingly, it turns out that when neither of α nor β are multiples of
1 + i we have also ϕ(αβ) = ϕ(α) + ϕ(β). From lemma 7.1 it follows that the
residues of α and β in Z[i]/(2) ∼= F2[i] are in {1, i}. It is enought to check that
ϕ has the desired property on {1, i} :
0 = ϕ(1) = ϕ(1 · 1) = ϕ(1) + ϕ(1) = 0 + 0 = 0
1 = ϕ(i) = ϕ(1 · i) = ϕ(1) + ϕ(i) = 0 + 1 = 1
0 = ϕ(−1) = ϕ(i · i) = ϕ(i) + ϕ(i) = 1 + 1 = 0
The lemma follows noting
Im(cid:16)Y αi(cid:17) % 2 = ϕ(cid:16)Y αi(cid:17) = X ϕ(αi) = ϕ(cid:16)X αi(cid:17) = Im(cid:16)X αi(cid:17) % 2.
The enzymes of Z[i] have been assembled. We are ready to digest the theo-
rem.
9
Theorem 7.3. For every µ ∈ Z[i] which factors into non-units (i.e. µ = αβ
with α, β 6∈ {1,−1, i,−i}) there exists a non-trivial minimal bioperationl multiset
over Z[i] with a sum-product of µ.
Proof. Pick some factorization µ = a1...an and let S = {ai}n
two ai non-units. We break into two cases.
i=1 with at least
Case 1) if Im(π(S)) and Im(σ(S)) have the same parity, we may bioper-
ate S by appending T±1 and T±2i as needed. The result is S′; bioperational
with sum-product µ. If S′ is not minimal, we may take a minimal subset S′′.
And we are assured S′′ is non-trivial since otherwise S′′ = {µ} which implies
ai = µ for some. And ai = µ implies all other αj for j 6= i are units since
α1...αi−1αi+1...αn = 1 (note we are using in this last step the fact that Z[i] is
an integral domain).
Case 2) if Im(π(S)) and Im(σ(S)) have different parities, we may suppose
from Lemma 7.2 that some αj is divisible by 1 + i. We create a new multiset
by removing αj from S and appending {iαj, i,−1}. In notation,
The product remains unchanged since
S′ = {αi}i6=j + {iαj, i,−1}.
π(S′) =
π(S)
αj
(i2αj)(−1) = π(S).
More importantly, it is claimed that Im(σ(S′)) and Im(σ(S)) have different
parity. Their difference is
Im(σ(S′)) − Im(σ(S)) = Im(σ(S′) − σ(S)) = Im(iαj + i − 1 − αj).
Let αj = a + bi for some integers a and b. Substitution gives
Im(σ(S′)) − Im(σ(S)) = Im(ai − b + i − 1 − a − bi) = a − b + 1.
From Lemma 7.1 we may suppose that a and b have the same parity since
1 + iαj. It follows that a − b + 1 is odd, that Im(σ(S′)) and Im(σ(S)) have
different parity, and therefore that Im(σ(S′)) and Im(π(S)) = Im(π(S′)) have
the same parity. And so we return to the first case to bioperate S′.
8 Eisenstein Integers (Z[ω])
Eisenstein integers are similar to the Gaussians in that they are all of the form
a + bω where a and b are integers and ω is a strictly complex number such that
ω3 = 1. Right away this gives us our first appendage,
One can show further show that ω2 = −1 − ω from which we get
T3ω = {ω, ω, ω}.
T−2ω = {−ω,−1 − ω,−1, 1, 1}.
10
Thus we have Tω = T3ω + T−2ω and T−ω = T3ω + 2T−2ω at our disposal. Supris-
ingly, we can therefore bioperate any multiset over Z[ω]. Our main theorem in
this section will therefore run almost identically to its analog over Z.
Theorem 8.1. For µ ∈ Z[ω] which factors into non-units there exists a non-
trivial minimal bioperational multiset over Z[ω] with a sum-product of µ.
Proof. Choose a factorization µ = α1...αn with 2 non-units and let S = {αi}n
i=1.
Bioperate S with T±1 and T±ω. The resulting bioperational multiset S′ can be
shaved down to minimality without becoming trivial since the non-units cannot
be trimmed off.
9 Integers √2 Appended (Z[√2])
Lastly, we consider the integer ring of a real quadratic number field. Numbers
in Z[√2] are of the form a + b√2 where a and b are integers (again, very similar
to Z[i] and Z[ω]). We use the fact that
to create
T
(1 + √2)(−1 + √2) = 1
√2. ∓ 1 ±
±2√2 = {±1 ±
√2}.
There's good reason to think that this is the best we can do (I.e. that T
±√2
doesn't exist over Z[√2]). But the best proof the author could come up with for
such a fact uses difficult results about quadratic number fields and complicated
induction. Instead, we take a route similar to that taken through Z[i].
Lemma 9.1. The number a + b√2 ∈ Z[√2] is a multiple of √2 if and only if a
is even.
Proof. Note (c + d√2)√2 = 2d + c√2.
Lemma 9.2. For numbers α1, ..., αn ∈ Z[√2] let
a + b√2 = X αi
and c + d√2 = Y αi.
If no αi is divisible by √2 then b ≡ d mod 2.
Proof. We again create a strange homomorphism. Let ϕ(a+b√2) = b % 2 ∈ F2.
It follows that ϕ(α + β) = ϕ(α) + ϕ(β). We claim if √2 divides neither α nor β
then ϕ(αβ) = ϕ(α) + ϕ(β) as well. From the previous lemma, we see that the
residues of such α and β with coefficients in F2 are in {1, 1 + √2}. We check ϕ
by hand:
0 = ϕ(1) = ϕ(1 · 1) = ϕ(1) + ϕ(1) = 0 + 0 = 0
1 = ϕ(1 + √2) = ϕ(1 · (1 + √2)) = ϕ(1) + ϕ(1 + √2) = 0 + 1 = 1
0 = ϕ(1) = ϕ((1 + √2) · (1 + √2)) = ϕ(1 + √2) + ϕ(1 + √2) = 1 + 1 = 0
11
We end noting
d % 2 = ϕ(cid:16)Y αi(cid:17) = X ϕ(αi) = ϕ(cid:16)X αi(cid:17) = b % 2.
It's probable that if the author knew more about ring isomorphisms, the
results of this section and those of Section 7 could have been demonstrated
simultaneously.
Theorem 9.3. For every µ ∈ Z[√2] which factors into non-units there exists a
non-trivial minimal bioperationl multiset over Z[√2] with a sum-product of µ.
i=1, a + b√2 = σ(S),
Proof. Pick a factorization µ = α1...αn and let S = {αi}n
and c + d√2 = π(S). If b and d have the same parity, S can be bioperated into
the desired result. If not, we may pick some αj a multiple of √2. Letting
S′ = {αi}i6=j + {(1 + √2)αj ,−1 + √2}
Letting αj = x + y√2, the change σ(S) is
σ(S′)−σ(S) = (x+y√2)(1+√2)+(−1+√2)−(x+y√2) = (2y−1)+(x+1)√2.
But from Lemma 9.1, we may suppose that x is even and that therefore σ(S′)
and σ(S) have √2 coefficients of different parity.
It follows that σ(S′) and
π(S′) = π(S) have √2 coefficients of the same parity and that S′ can therefore
be bioperated into the desired result.
10 Generalization and Open Problems
Let's start this section by bundling up our main theorems into a single statement
Theorem 10.1. If R is one of N, Z, Z[i], Z[ω], or Z[√2] then for every µ ∈ R
which factors into non-units, there exists a non-trivial minimal bioperational
multiset over R with a sum-product of µ.
Proof. Theorems 3.4, 4.1, 7.3, 8.1, 9.3.
Some open problems of interest:
• Does Theorem 10.1 hold over the quaternions?
Order of multiplication now matters. We at least have T±2i, T±2j, and
T±2k at our disposal since
T±2v = (v, v,−1, 1)
has a product of 1 for v ∈ {i, j, k}.
12
• Does Theorem 10.1 hold for all integer rings of real quadratic number
fields?
There are families of such rings that admit easy attack. Take for instance
Z[pt2 ± 1]. From
(t + pt2 ± 1)(t − pt2 ± 1) = ∓1
±2√t2±1 which gives us pretty good flexi-
we can construct appendages T
bility for bioperation. And in general, for d = t(b2t ± 2) we can construct
±2b√d. The first values not covered by these parametriza-
appendages T
tions are
13, 19, 21, 22, 28, 29, 31, 33, 39, 41, 43, 44, 45, 46, 52, 53, 54, 55, 57, 58, 59, 61, 67, 69, ...
Perhaps Z[√13], which has a relatively large fundamental unit, is our
first example for which Theorem 10.1 fails. One would think it easy to
construct a counter-example ring to the theorem. However the handful of
examples the author toyed with proved dead ends.
References
[1] Matt Parker, What's the story with log(1 + 2 + 3)?.
https://www.youtube.com/watch?v=OcTMBrUutfk
[2] L. Kurlandchik and A. Nowicki, When the sum equals the product,
The Mathematical Gazette, 84(499), 91-94. doi:10.2307/3621488.
[3] Mark Bennet, https://math.stackexchange.com/questions/1176875
[4] Michael Rozenberg, https://math.stackexchange.com/questions/2640531
[5] mathlove, https://math.stackexchange.com/questions/929564
[6] Robert Israel, https://math.stackexchange.com/questions/111040
[7] On Github: https://github.com/onnomc/bioperational-multisets
On Repl.it: https://repl.it/@onnomc/BioperationalMultisets
[8] M.A. Nyblom, C.D. Evans, An Algorithm to Solve the Equal-Sum-Product
Problem. https://arxiv.org/abs/1311.3874
[9] The Online Encyclopedia of Integer Sequences, https://oeis.org/
[10] David Applegate, Marc LeBrun, N. J. A. Sloane, Dismal Arithmetic,
https://arxiv.org/abs/1107.1130
[11] Niel Sloane, Primes on the Moon (Lunar Arithmetic),
Numberphile, [interview by Brady Haran],
https://www.youtube.com/watch?v=cZkGeR9CWbk
13
|
1508.07287 | 2 | 1508 | 2016-02-08T16:27:25 | Zeta functions for tensor products of locally coprime integral adjacency algebras of association schemes | [
"math.RA"
] | The zeta function of an integral lattice $\Lambda$ is the generating function $\zeta_{\Lambda}(s) = \sum\limits_{n=0}^{\infty} a_n n^{-s}$, whose coefficients count the number of left ideals of $\Lambda$ of index $n$. We derive a formula for the zeta function of $\Lambda_1 \otimes \Lambda_2$, where $\Lambda_1$ and $\Lambda_2$ are $\mathbb{Z}$-orders contained in finite-dimensional semisimple $\mathbb{Q}$-algebras that satisfy a "locally coprime" condition. We apply the formula obtained above to $\mathbb{Z}S \otimes \mathbb{Z}T$ and obtain the zeta function of the adjacency algebra of the direct product of two finite association schemes $(X,S)$ and $(Y,T)$ in several cases where the $\mathbb{Z}$-orders $\mathbb{Z}S$ and $\mathbb{Z}T$ are locally coprime and their zeta functions are known. | math.RA | math |
Zeta Functions for tensor products of locally
coprime integral adjacency algebras of
association schemes
Allen Herman∗, Mitsugu Hirasaka†, and Semin Oh†
February 1, 2016
Abstract
Pn=0
The zeta function of an integral lattice Λ is the generating function
ann−s, whose coefficients count the number of left ideals
ζΛ(s) =
∞
of Λ of index n. We derive a formula for the zeta function of Λ1 ⊗ Λ2,
where Λ1 and Λ2 are Z-orders contained in finite-dimensional semisim-
ple Q-algebras that satisfy a "locally coprime" condition. We apply
the formula obtained above to ZS ⊗ ZT and obtain the zeta function
of the adjacency algebra of the direct product of two finite association
schemes (X, S) and (Y, T ) in several cases where the Z-orders ZS and
ZT are locally coprime and their zeta functions are known.
Key words : Association schemes, adjacency algebras, zeta functions, integral
representation theory.
AMS Classification: Primary: 16G30; Secondary: 05E30, 20C15.
1. Introduction
Suppose Λ is a Z-order in a finite-dimensional Q-algebra A, V is a finite-
dimensional A-module, and L is a full Λ-lattice in V ; i.e. a free Z-submodule
∗The first authour acknowledges the support of an NSERC Discovery Grant.
†This research was supported by Basic Research Program through the National Re-
search Foundation of Korea(NRF) funded by the Ministry of Education, Science and
Technology (grant number NRF-2013R1A1A2012532).
1
of V with QL = V . Solomon's zeta function is the function of a complex
variable s defined by
ζΛ(L, s) = XN ⊆L
(L : N)−s
where N runs over all full Λ-sublattices of L. (The reader should be mindful
that this was denoted ζL(s; Λ) in [7].) For each positive integer n, let an be
the number of full Λ-sublattices of L with index n. Then
ζΛ(L, s) =
ann−s.
∞
Xn=1
We will not deal here with issues of convergence or continuation of these
series, except to say it is well-known that they do converge uniformly on a
region in the complex plane where the real part of s is sufficiently large. In
[7], Louis Solomon establishes several results for ζΛ(L, s) when Λ is an order
in a finite-dimensional semisimple Q-algebra, the first of these being an Euler
product formula
ζΛ(L, s) =Yp
ζΛp(Lp, s).
Here p runs through the rational primes p, and Λp and Lp are the p-adic
completions of Λ and L, respectively, obtained by tensoring with the ring of
p-adic integers Zp. (Although Solomon uses this notation for p-localizations,
this does not make a difference to the zeta function - see [7, Lemma 9].)
If the semisimple algebra QpΛp decomposes as ⊕h
i=1Mri(Di), where the p-
adic division algebras Di occurring in this decomposition have index mi,
the center of Di has ring of integers Ri, and each simple component has
maximal order Γp,i, then we can compute the p-adic zeta function ζΓp(Lp, s).
To do so, first decompose the QpΛ-module Qp ⊗Q V into its homogeneous
components. This expresses it as a direct sum of modules Vp,i, each of which
is isomorphic to a direct sum of ki-copies of an irreducible Mri(Di)-module,
for i = 1, . . . , h. The lattice Lp similarly decomposes as the direct sum of
Lp,i's, each a sublattice of Vp,i. According to Hey's formula (see [1, pg. 145])
Here ζRi(s) is the usual Dedekind zeta function; i.e.
ζΓp(Lp, s) = Qh
= Qh
ζR(s) =YP
i=1 ζΓp,i(Lp,i, s)
i=1Qki−1
j=0 ζRi(rimis − jmi).
(1 − N (P)−s)−1,
2
where P runs over all maximal ideals of the Dedekind domain R, and N (P)
is the least power of the positive integral prime p ∈ P that generates the
norm of the ideal P. For all but finitely many primes p, the Zp-order Λp is
a maximal order of QpΛ, so the problem reduces to calculating ζΛp(Lp, s) for
these particular primes.
In the particular case of the regular left A-module V = AA, we write
In this case the left Λ-sublattices of Λ of index n are
ζΛ(s) for ζΛ(Λ, s).
the Z-free left ideals of Λ with index n. In [7], Louis Solomon established
several results for the zeta function of the integral group ring ZG treated
as an order in QG. Up to now explicit formulas expressing zeta functions
of integral group rings in terms of products of Dedekind zeta functions of
number fields have been found in a few instances. The first was the case of
the integral group ring ZCp of a cyclic group of prime order p, whose zeta
function is (see [7, Theorem 1])
ζZCp(s) = (1 − p−s + p1−2s)ζZ(s)ζZ[εp](s),
where εp denotes a primitive root of unity of order p.
We will be interested in Z-orders in the Q-adjacency algebras of finite
association schemes. An association scheme is a purely combinatorial object
generalizing the set of orbitals of a transitive permutation group. Let X be
a finite set and let S be a partition of X × X which does not contain the
empty set. We say that the pair (X, S) is an association scheme if it satisfies
the following conditions (see [9]):
(i) {(x, x) x ∈ X} is a member of S;
(ii) For each s ∈ S, s∗ = {(x, y) (y, x) ∈ s} is a member of S;
(iii) For all s, t, u ∈ S, the size of {z ∈ X (x, z) ∈ s, (z, y) ∈ t} is constant
whenever (x, y) ∈ u.
Each relation s ∈ S has a corresponding adjacency matrix σs whose rows
and columns are indexed by the elements of X as follows:
(σs)x,y =(1 if (x, y) ∈ s
0 if (x, y) /∈ s.
The pair (X, S) is an association scheme if and only if
3
(i) The identity matrix is a member of {σs s ∈ S};
(ii) Ps∈S σs = J, the all 1's matrix whose rows and columns are indexed
by X;
(iii) for all s ∈ S, σs∗ = σ⊤
s , where σ⊤
s
is the transpose of the matrix σs;
and
(iv) For all s, t ∈ S the matrix product σsσt is a nonnegative integer linear
combination of {σs s ∈ S}.
For an association scheme (X, S) we denote the set of Z-linear combinations
of {σs s ∈ S} by ZS, and for a ring R with unity we denote the tensor
product R⊗Z ZS by RS. The above conditions imply that RS is a subalgebra
of the full matrix algebra of degree X over R, which is called the adjacency
algebra of (X, S) over R. It is well-known that RS is semisimple if R is an
extension field of Q (see [2]). In particular, ZS is naturally a Z-order in the
semisimple algebra QS.
Recently formulas for the zeta function for ZS were produced for some
schemes of small rank [3]. We will consider the problem of expressing the
zeta function of a tensor product of two integral adjacency algebras. The
approach of the present paper is based on ideas used by Hironaka [4] to
extend the calculation of ζZCp(s) to the calculation of ζZCn(s) for n a square-
free integer. The methods of this paper will allow us to extend the cases
where the zeta function is known to certain direct products of these known
cases.
2. Zeta functions for rank 2 orders over rings of p-adic integers
Let R be the ring of integers of a p-adic number field F . This means the
field F is a finite extension of the usual p-adic field Qp, and its ring of integers
is a local PID with maximal ideal πR for some π ∈ R. The degree of the
extension F/Qp is ef , where f is its residue degree and e is its ramification
index. This means [R : πR] = pf and πeR = pR for the unique integer prime
p for which p ∈ πR.
Let Λ be a Z-lattice of rank 2. This means Λ is an R-free subring of QΛ,
and Λ = R1 + Rx for some x ∈ Λ \ R1. In this case the minimal polynomial
of x has degree 2 in Z[x]. We will assume the (monic) minimal polynomial of
x is reducible in Z[x], so it factors as the product of (not necessarily distinct)
linear factors. By a shift of variable we may assume the minimal polynomial
4
of x has the form x(x − n) for some integer n. We wish to calculate the zeta
function of RΛ under these assumptions, for our p-adic ring of integers R.
For a nonzero integer n we define [n] to be the maximal positive integer
[n] for which p[n] divides n.
Theorem 1. Let F be an extension of Qp with residue degree f and ramifi-
cation index e. Let R be the ring of integers of F and let π ∈ R be chosen
so that πR is the maximal ideal of R.
Let RΛ = R[x]/x(x − n)R[x].
(i) If n = 0, then ζRΛ(s) = (1 − pf (1−2s))−1(1 − pf (−s))−1.
(ii) If n 6= 0, then
ζRΛ(s) =
e[n]−1
Xr2=0
pf r2(1−2s)(1 − pf (−s))−1 + pf e[n](1−2s)(1 − pf (−s))−2.
Proof. Note that {1, x} is an R-basis of RΛ. Let I be an R-free ideal of RΛ
with finite index; i.e. I is a free R-submodule that is an ideal of RΛ. Since
I is a free R-submodule, it has a generating set of the form {πr1 + ax, πr2x},
where a ∈ R is given up to addition modulo πr2R. For such a submodule,
we have [RΛ : I] = [R : πR](r1+r2) = pf (r1+r2).
Since I is an ideal of RΛ, it must be closed under multiplication by x,
and this imposes extra conditions on our generating set. In particular, using
the fact that x2 = nx in RΛ, we must have that
x(πr1 + ax) = (πr1 + an)x = c1(πr1 + ax) + c2(πr2x), ∃c1, c2 ∈ R, and
x(πr2x) =
(πr2n)x
= d1(πr1 + ax) + d2(πr2x), ∃d1, d2 ∈ R.
The second equation has a unique solution: d1 = 0 and d2 = n. In the first
equation, c1 = 0, so it reduces to πr1 + an = c2πr2. The number of distinct
ideals for a fixed pair of nonnegative integers r1 and r2 is
N(r1, r2) = {a ∈ R/πr2R : πr1 + an ∈ πr2R}.
When n = 0 (which is allowed), the condition reduces to r1 ≥ r2 and we
have N(r1, r2) = pf r2. On the other hand, when n 6= 0, we have that nR =
p[n]R = πe[n]R. When 0 ≤ r2 < e[n], then we will have r2 ≤ r1, so every a ∈ R
satisfies πr1 + an ∈ πr2R. This means that N(r1, r2) = R : πr2R = pf r2 in
5
this case. However, when e[n] ≤ r2 we will have πr2/n ∈ πr2R. The condition
then reduces to a ∈ πr1/n + (πr2/n)R. The number of equivalence classes in
R/πr2R satisfying this condition is pf r2/pf r2−e[n] = pe[n].
We can now compute the zeta function of RΛ directly in each case. If
n = 0, then
ζRΛ(s) = Pr2≥0Pr1≥r2 pf r2pf (r1+r2)(−s)
= Pr2≥0 pf r2(1−s)Pr1≥r2 pf r1(−s)
= Pr2≥0 pf r2(1−s)pf r2(−s)(1 − pf (−s))−1
= (1 − pf (1−2s))−1(1 − pf (−s))−1.
When n 6= 0, we get
ζRΛ(s) = Pr2Pr1 N(r1, r2)pf (r1+r2)(−s)
= P0≤r2<e[n]Pr1≥r2 pf r2pf (r1+r2)(−s)
+Pr2≥e[n]Pr1≥e[n] pe[n]pf (r1+r2)(−s)
= Pe[n]−1
r2=0 pf r2(1−s)Pr1≥r2 pf r1(−s)
+pe[n]Pr2≥e[n] pf r2(−s)Pr1≥e[n] pf r1(−s)
= Pe[n]−1
r2=0 pf r2(1−2s)(1 − pf (−s))−1 + pf e[n](1−2s)(1 − pf (−s))−2.
We remark that the above approach is valid when n = 0 because it does
not require that Λ be contained in a maximal order of QpΛ.
If R is the
ring of integers of an algebraic number field K, then in order to apply this
approach we need to establish an Euler product formula in the absence of
the semisimplicity condition. Let spec(R) be the set of nonzero prime ideals
of R. Suppose Λ is an R-order in a nonsemisimple K-algebra A, and L is a
full Λ-lattice in an A-module V . By [6, Theorem I.8.8] if M is a full Λ-lattice
with finite index in L, then
L/M ≃ MP∈spec(R)
LP/MP,
and LP/MP can be nontrivial for only finitely many primes P in spec(R). It
follows that [L : M] =QP[LP : MP], and this product has only finitely many
factors larger than 1. Furthermore, for any selection of full ΛP-lattices {MP}
with all but finitely many equal to LP, it follows from [6, Lemma I.8.8] that
there exists a full Λ-lattice M with the selected P-localization at every prime
6
P. Therefore, keeping in mind these products are finite and the indices are
all powers of individual rational primes p, we get
ζΛ(L, s) = P[L:M ]<∞[L : M]−s
= PM(cid:0)QP[LP : MP](cid:1)−s
= QP(cid:0)PMP [LP : MP]−s(cid:1)
= QP ζΛP (LP, s).
Therefore, the factors in this Euler product can be computed using the above
Lemma even in the case where QpΛ is not semisimple.
As a corollary we can generalize Hanaki and Hirasaka's calculation of the
zeta function of an integral adjacency algebra of an association scheme of
rank 2 to larger coefficient rings.
Corollary 2. Let (Y, T ) be an association scheme of rank 2 and order n.
Let p be a prime divisor of n. Suppose F is an extension of Qp with residue
degree f and ramification index e, and let R be its ring of integers. Then
ζRT (s) =
e[n]−1
Xr2=0
pf r2(1−2s)(1 − pf (−s))−1 + pf e[n](1−2s)(1 − pf (−s))−2.
Proof. Under these assumptions T = {σ0, σ1}, where σ1 is the adjacency
matrix of the ordinary complete graph on n vertices Kn, whose minimal
polynomial is µ(x) = (x − (n − 1))(x + 1). It is easy to see that RT = R[σ1].
By the change of variable x 7→ x − 1 we have RT ≃ R[x]/x(x − n)R[x]. So
Theorem 1 can be applied to calculate the zeta function of RT directly.
3. Zeta functions for tensor products of locally coprime orders
In this section we consider the calculation of the zeta function of the
tensor product of two Z-orders Λ1 and Λ2 under certain conditions. Since Λ1
and Λ2 contain Z-bases b1 and b2, we have that ZΛ1 = Zb1 and ZΛ2 = Zb2,
where this notation denotes the integer span of the given set. This means
that for any extension R of Z, we will have
R ⊗Z (Λ1 ⊗Z Λ2) = R ⊗Z Z[b1 × b2]
= R[b1 × b2]
= Rb1 ⊗R Rb2
= RΛ1 ⊗R RΛ2.
7
Now for the conditions on the pair of orders Λ1 and Λ2, first we require
that QΛ1 and QΛ2 are finite-dimensional commutative semisimple algebras.
This implies that Λ1 and Λ2 are contained in maximal orders Γ1 and Γ2,
respectively, and we can write
Γi ≃ ⊕hi
j=1Rij, i = 1, 2,
where Rij is the ring of integers in the algebraic number field Fij that appears
as the center of the j-th simple component of QΛi. For a given rational prime
p, we will assume Fij has splitting degree gij, that Pij1, . . . , Pijgij are the
distinct primes of Fij lying above p, and that Rijk is the completion (Rij)Pijk
for k = 1, . . . , gij.
Second, we will require that the two orders Λ1 and Λ2 are locally coprime,
which means that for all primes p, either ZpΛ1 is a maximal order in QpΛ1 or
ZpΛ2 is a maximal order in QpΛ2. Under these two conditions we will obtain
an expression for ζZp[Λ1 ⊗ Λ2] for all primes p.
The zeta function for Λ1 ⊗Z Λ2 will be expressed using the Euler product
formula: if Λ is an order in a semisimple algebra and Γp is a maximal order
of QpΛ containing ZpΛ for all rational primes p, then
ζΛ(s) = ζΓ(s)Yp∈B
ζZpΛ(s)
ζΓp(s)
,
where B is the set of primes p for which ZpΛ is not a maximal order of QpΛ.
j=1R1j and Γ2 = ⊕h2
Theorem 3. Let Λ1 and Λ2 be orders in commutative semisimple Q-algebras
QΛ1 and QΛ2, respectively. Suppose that Λ1 and Λ2 are locally coprime. Let
Γ1 = ⊕h1
j=1R2j be maximal orders containing Λ1 and Λ2.
For each rational prime p, we have gij, Rijk, fijk, and eijk as defined above, so
in particular Rijk is a direct summand of the maximal order QpΓi of QpΛi for
all primes p and i = 1, 2. Let Bi be the set of primes p for which QpΛi 6= QpΓi
for i = 1, 2. Then
where
ζZ[Λ1⊗Λ2](s) = ζΓ1⊗Γ2(s)δB1(s)δB2(s),
δB1(s) = Yp∈B1
h2
Yj=1
g2j
Yk=1
ζR2jkΛ1(s)
ζR2jkΓ1(s)
,
8
and
δB2(s) = Yp∈B2
h1
Yj=1
g1j
Yk=1
ζR1jkΛ2(s)
ζR1jkΓ2(s)
.
Proof. By the Euler product formula, it suffices to verify the formula for
δB1(s). If p ∈ B1, then ZpΛ2 is a maximal order in QpΛ2, so our notation
says that
ZpΛ2 ≃ ⊕h2
j=1 ⊕g2j
k=1 R2jk.
Now, we have that
Zp ⊗ [Λ1 ⊗ Λ2] ≃ ZpΛ1 ⊗Zp ZpΛ2
≃ ZpΛ1 ⊗Zp (⊕h2
= ⊕h2
j=1 ⊕g2j
k=1 R2jkΛ1.
j=1 ⊕g2j
k=1 R2jk)
It follows that
h2
g2j
ζZp[Λ1⊗Λ2](s) =
ζR2jkΛ1(s).
Yj=1
Yk=1
The formula for δB1(s) is now a consequence of the Euler product formula.
It is straightforward to extend this idea to the tensor product of finitely
many orders that are locally coprime, where this is taken to mean that for
any rational prime p, at most one of the p-adic completions of the orders is
not a maximal order in its overlying Qp-algebra.
The direct product of (X, S) and (Y, T ) is the scheme (X × Y, S × T )
whose adjacency matrices are the pairwise tensor products of the adjacency
matrices of S with those of T . (see [9, §7]). This produces a Q-algebra with
a predictable Wedderburn decomposition since
Q[S × T ] ≃ QS ⊗Q QT.
Since ZS and ZT are orders with
Z[S × T ] ≃ ZS ⊗Z ZT,
it is clear that the approach of section 3 can be directly applied in cases where
we have two commutative association schemes whose integral scheme rings
are locally coprime. We will give several examples where this does occur in
the next section.
9
4. Examples of explicit zeta functions
In this section we apply the results and methods of section 2 and 3 to
produce explicit zeta functions in several new instances. Our first application
combines the notation of section 3 with Corollary 2.
Proposition 4. Let (Y, T ) be an association scheme of rank 2 and order n.
Let R be the ring of integers in an algebraic number field F . For each prime
p, let gp be the splitting degree of F at p, let Rp1, . . . , Rpgp be the distinct
p-adic completions of R at the primes of F lying over p, and suppose the
quotient field of each Rpj has residue degree fpj and ramification index epj.
Then
ζRT (s) = ζR(s)2Ypn
Yj=1
gp
ζRpjT (s)
ζRpj (s)2 ,
where ζRpjT (s) is calculated for the p-adic ring of integers Rpj using the for-
mula of Corollary 2 in terms of n, epj, and fpj.
Proof. This is a consequence of the Euler product formula. The set of rational
primes B where ZT is not a maximal order of QT is the set of divisors of n.
This means the set of primes P of F for which RT is not a maximal order
of F T is contained in the set of primes of F lying over a prime dividing n.
The square of the Dedekind zeta function occurs because the maximal order
of F T is the direct sum of two copies of R. The square in the denominator
occurs because the maximal order containing RpjT is a direct sum of two
copies of Rpj.
For our main applications we give several examples of zeta functions for
locally coprime pairs of integral adjacency algebras of association schemes
among those whose zeta functions were known previously.
Example 5. Consider the special case of the zeta function of the integral
adjacency algebra of the direct product S × T of two association schemes
in which the second scheme is the group C2. In this case in the notation of
section 3 we have B2 = {2} and Γ2 = Z ⊕ Z. Let (X, S) be any commutative
scheme for which Z2S is a maximal order of Q2S. Using the notation of
section 3, write the Wedderburn decomposition of Q2S as ⊕h1
k=1 F1jk.
Then in the notation of Theorem 3 we have
j=1 ⊕g1j
δB2(s) =
h1
Yj=1
g1j
(1 − 2−f1jks + 2−f1jk(1−2s)).
Yk=1
10
For an easy explicit example, we will give the zeta function of ZC6 ≃
Z[C3 × C2] (though it should be noted that Hironaka's results will give this
by essentially the same method). In this case B1 = {3} and B2 = {2} so ZC3
and ZC2 are locally coprime. The Wedderburn decomposition of Q[C3 × C2]
is Q ⊕ Q ⊕ Q(ε3) ⊕ Q(ε3). Q2(ε3) is unramified of degree f = 2. Therefore,
from Theorem 3 and Corollary 2, we have
ζZ[C3×C2](s) = ζZ(s)2ζZ[ε3](s)2 × (1 − 3−s + 3(1−2s))(1 − 3−2s + 32(1−2s))
× (1 − 2−s + 2(1−2s))2.
We can even say something about the 2-adic zeta function of S ×C2 when
the association scheme S is not commutative, as long as Z2S is a maximal
order of Q2S. If the Wedderburn decomposition of Q2S is
Q2S ≃
h2
g1j
Mj=1
Mk=1
Mr1jk (D1jk),
and the 2-adic divison algebras D1jk have index m1jk and center F1jk, then
the fact that the maximal order containing Z2T is the direct sum of two
copies of Z2 implies that
ζZ2[S×T ](s) =
h1
Yj=1
g1j
(1 − 2−f1jks + 2−f1jk(1−2s))ζΓ1jk (s)2,
Yk=1
where Γ1jk is a maximal order of the local central simple algebra Mr1jk (D1jk),
whose zeta function can be calculated using Hey's formula.
Example 6. Next we give the zeta function of Z[S × T ], where S and T are
association schemes of rank 2 having coprime orders m and n, respectively.
These scheme rings are the integral adjacency algebras of the ordinary com-
plete graphs Km and Kn and it is convenient to use this notation. In this case
B1 is the set of primes dividing m, and B2 is the set of primes dividing n, so
the assumption (m, n) = 1 implies that ZKm and ZKn are locally coprime.
The Wedderburn decomposition of Q[Km × Kn] consists of the direct sum of
4 copies of Q, so by a direct application of Theorem 3, we find
11
r2=0 pr2(1−2s)(1 − p(−s))−1) + p[m]p(1−2s)(1 − p(−s))−2
r2=0 qr2(1−2s)(1 − q(−s))−1) + q[n]q(1−2s)(1 − q(−s))−2
(1 − p−s)−2
(1 − q−s)−2
(cid:19)2
(cid:19)2
ζZ[Km×Kn](s) = ζZ(s)4
×Qpm(cid:18)(P[m]p−1
×Qqn(cid:18)(P[n]q−1
= ζZ(s)4 ×Qpm(cid:18)(P[m]p−1
×Qqn(cid:18)(P[n]q−1
r2=0 pr2(1−2s)(1 − p−s)) + p[m]p(1−2s)(cid:19)2
r2=0 qr2(1−2s)(1 − q−s)) + q[n]q(1−2s)(cid:19)2
.
where p runs over prime divisors of m and q runs over prime divisors of n.
This formula generalizes immediately to the product of any number of
rank 2 schemes of pairwise coprime orders.
Example 7. Finally, we give the zeta function of Z[Cp × Kn], where Z[Kn] is
the adjacency algebra of the rank 2 scheme of order n and p is a prime that
does not divide n. In this case the Wedderburn decomposition of Q[Cp × Kn]
is Q[Cp × Kn] ≃ Q ⊕ Q ⊕ Q(εp) ⊕ Q(εp). Note that, in our notation we will
have h1 = h2 = 2. Suppose that for each prime divisor q of n, Qq(εp) has
residue degree fq. Then there are gq = (p − 1)/fq primes of Zq[εp] lying over
12
q. By Theorem 3, we have
ζZ[Cp×Kn](s) = ζZ(s)2ζZ[εp](s)2(1 − p−s + p1−2s)2QqnQh1
j=1Qg1j
k=1
= ζZ(s)2ζZ[εp](s)2(1 − p−s + p1−2s)2
ζR1jkKn(s)
ζR1jk(s)2
r2=0 qr2(1−2s))(1 − q−s)−1 + q[n]q(1−2s)(1 − q−s)−2
(1 − q−s)−2
×Qqn(cid:18)(P[n]q−1
(P[n]q−1
×Q(p−1)/fq
k=1
= ζZ(s)2ζZ[εp](s)2(1 − p−s + p1−2s)2
r2=0 qfqr2(1−2s))(1 − q−fqs)−1 + qfq[n]q(1−2s)(1 − q−fq−s)−2)
(1 − q−fqs)−2
(cid:19)
×Qqn(cid:18)(cid:0)(P[n]q−1
×(cid:0)Q(p−1)/fq
r2=0 qr2(1−2s))(1 − q−s) + q[n]q(1−2s)(cid:1)
r2=0 qfqr2(1−2s))(1 − q−fqs) + qfq[n]q(1−2s)(cid:1)(cid:19).
(P[n]q−1
k=1
Again it should be straightforward to generalize this formula to give the zeta
function of Z[Cm × T ], where T is a rank 2 scheme of order n and m is a
square-free positive integer that is relatively prime to n.
Zeta functions have been described for Cp × Cp for a prime p [8], Cp2 for
a prime p [5] and nonabelian metacyclic groups of type Cp ⋊ Cq for a prime
p and a prime q dividing p − 1 [4]. The formulas for these are a bit more
complicated, but in all of these cases the sets of "bad" primes are precisely
the primes dividing the order of the group. The direct product of any of these
groups with a cyclic group of coprime square-free coprime order or a rank
2 scheme of coprime order can be handled using the present methods. For
example, the zeta functions of the integral scheme rings of (C2 × C2) × K3,
C4 × C3, or S3 × K5 can be established from their results using an application
of Theorem 3.
The explicit formulas for the zeta function of C2 × C2 and C3 × C3 pre-
sented by Takagehara [8] indicate that it will be considerably more difficult
to calculate the zeta function of the tensor product of two integral adjacency
algebras that are not locally coprime.
13
References
[1] C. Bushnell and I. Reiner, Zeta functions of arithmetic orders and
Solomon's conjectures, Math. Z., 173 (1980), 135-161.
[2] A. Hanaki, Semisimplicity of adjacency algebras of association schemes,
J. Algebra, 225 (2000), no. 1, 124-129.
[3] A. Hanaki and M. Hirasaka, The zeta function of the integral adjacency
algebra of association scheme of rank 2, Hokkaido Math. J., to appear.
[4] Y. Hironaka, Zeta functions of integral group rings of metacyclic groups,
Tskuba Math. J, 5 (2), (1981), 267-283.
[5] I. Reiner, Zeta functions of integral representations, Comm. Algebra, 8
(10), (1980), 911-925.
[6] K. W. Roggenkamp and V. Huber-Dyson, Lattices over Orders, I, Lec-
ture Notes in Mathematics, No. 115, Springer-Verlag, 1970.
[7] L. Solomon, Zeta functions and integral representation theory, Adv.
Math. 26 (1977), 306-326.
[8] Y. Takegahara, Zeta functions of integral group rings of abelian (p,p)-
groups, Comm. Algebra, 15 (12), (1992), 2565-2615.
[9] P.-H. Zieschang, Theory of Association Schemes, Springer-Verlag, 2005.
14
|
1702.03532 | 1 | 1702 | 2017-02-12T15:18:56 | Omni $n$-Lie algebras and linearization of higher analogues of Courant algebroids | [
"math.RA",
"math-ph",
"math-ph"
] | In this paper, we introduce the notion of an omni $n$-Lie algebra and show that they are linearization of higher analogues of standard Courant algebroids. We also introduce the notion of a nonabelian omni $n$-Lie algebra and show that they are linearization of higher analogues of Courant algebroids associated to Nambu-Poisson manifolds. | math.RA | math |
Omni n-Lie algebras and linearization of higher analogues of
Courant algebroids ∗
Jiefeng Liu1, Yunhe Sheng1,2 and Chunyue Wang2,3
1Department of Mathematics, Xinyang Normal University,
Xinyang 464000, Henan, China
2Department of Mathematics, Jilin University,
Changchun 130012, Jilin, China
3 Department of Mathematics, Jilin Engineering Normal University,
Changchun 130052, Jilin, China
Email:[email protected]; [email protected]; [email protected]
Abstract
In this paper, we introduce the notion of an omni n-Lie algebra and show that they are linearization
of higher analogues of standard Courant algebroids. We also introduce the notion of a nonabelian
omni n-Lie algebra and show that they are linearization of higher analogues of Courant algebroids
associated to Nambu-Poisson manifolds.
1
Introduction
Courant algebroids were introduced in [20] (see also [22]), and have many applications. See [17] and
references therein for more details. On T n−1M , T M ⊕ ∧n−1T ∗M , define a symmetric nondegenerate
∧n−2T ∗M -valued pairing (·, ·)+ : T n−1M × T n−1M −→ ∧n−2T ∗M by
(X + α, Y + β)+ = iX β + iY α,
∀X + α, Y + β ∈ X(M ) ⊕ Ωn−1(M ),
and define a bracket operation J·, ·K : Γ(T n−1M ) × Γ(T n−1M ) −→ Γ(T n−1M ) by
JX + α, Y + βK = [X, Y ] + LXβ − iY dα.
(1)
(2)
The quadruple (T M ⊕∧n−1T ∗M, (·, ·)+, {·, ·}, prT M ) is called the higher analogue of the standard Courant
algebroid. In particular, if n = 2, we obtain the standard Courant algebroid. Recently, due to applications
in multisymplectic geometry, Nambu-Poisson geometry, L∞-algebra theory and topological field theory,
higher analogues of Courant algebroids are widely studied. See [2, 3, 9, 10, 11, 12, 14, 27] for more details.
0Keyword: omni n-Lie algebra, higher analogue of the standard Courant algebroid, nonabelian omni n-Lie algebra,
Nambu-Poisson structure
0MSC: 53D17, 17B99.
∗Research supported by NSFC (11471139), NSF of Jilin Province (20140520054JH,20170101050JC) and Nan Hu Scholar
Development Program of XYNU.
1
The notion of an omni-Lie algebra was introduced by Weinstein in [26] to study the linearization of
the standard Courant algebroid. Then it was studied from several aspects [15, 23, 25]. An omni-Lie
algebra associated to a vector space V is a triple (gl(V ) ⊕ V, (·, ·)+, {·, ·}), where (·, ·)+ is the V -valued
pairing given by
(A + u, B + v)+ = Av + Bu,
∀ A + u, B + v ∈ gl(V ) ⊕ V,
and {·, ·} is the bilinear bracket operation given by
{A + u, B + v} = [A, B] + Av.
(3)
(4)
Even though (gl(V ) ⊕ V, {·, ·}) is not a Lie algebra, its Dirac structures characterize all Lie algebra
structures on V . We can construct a Lie 2-algebra from an omni-Lie algebra. See [23] for more details.
In [19], the authors introduced the notion of a nonabelian omni-Lie algebra (gl(g) ⊕ g, (·, ·)+, {·, ·}g)
associated to a Lie algebra (g, [·, ·]g), which originally comes from the study of homotopy Poisson manifolds
g∗ associated
[18]. In particular, they showed that it is the linearization of the Courant algebroid T g∗⊕T ∗
πg
to the linear Poisson manifold (g∗, πg), where πg is the Lie-Poisson structure on g∗.
The purpose of this paper is to extend the above results to the n-ary case. First we introduce the no-
tion of an omni n-Lie algebra, which is a triple (gl(V ) ⊕ ∧n−1V, (·, ·)+, {·, ·}) including a bracket operation
{·, ·} and a (V ⊗ ∧n−2V )-valued pairing (·, ·)+. Similar to the classical case, (gl(V ) ⊕ ∧n−1V, {·, ·}) is a
Leibniz algebra. We show that a linear map F : ∧nV −→ V defines an n-Lie algebra structure on V if and
only if the graph of F ♯ : ∧n−1V −→ gl(V ) is a sub-Leibniz algebra of (gl(V ) ⊕ ∧n−1V, {·, ·}). Note that
this result is not totally parallel the classical case. Namely the condition that F being skew-symmetric
can not be simply described by being isotropic with respect to the (V ⊗ ∧n−2V )-valued pairing (·, ·)+. We
further show that an omni n-Lie algebra (gl(V ) ⊕ ∧n−1V, (·, ·)+, {·, ·}) can be viewed as the linearization
of the higher analogue of the standard Courant algebroid (T M ⊕∧n−1T ∗M, (·, ·)+,J·, ·K , prT M ) via letting
M = V ∗. Then we introduce the notion of a nonabelian omni n-Lie algebra (gl(g) ⊕ ∧n−1g, (·, ·)+, {·, ·}g)
associated to an n-Lie algebra g and study its algebraic properties. Finally, we give the notion of higher
analogues of Courant algebroids associated to Nambu-Poisson manifolds and study their properties. Fur-
thermore, we show that nonabelian omni n-Lie algebras are linearization of higher analogues of Courant
algebroids associated to Nambu-Poisson manifolds.
The paper is organized as follows. In Section 2, we recall n-Lie algebras and Nambu-Poisson manifolds.
In Section 3, we introduce the notion of an omni n-Lie algebra associated to a vector space V and
characterize n-Lie algebra structures on V via sub-Leibniz algebra structures of the omni n-Lie algebra.
In Section 4, we show that an omni n-Lie algebra is the linearization of the higher analogue of the
standard Courant algebroid. In Section 5, we introduce the notion of a nonabelian omni n-Lie algebra
and study its algebraic properties. In Section 6, we introduce the notion of higher analogues of Courant
algebroids associated to Nambu-Poisson manifolds and show that nonabelian omni n-Lie algebras are
their linearization.
2 Preliminaries
In this section, we briefly recall the notions of n-Lie algebras and Nambu-Poisson manifolds. The notion of
an n-Lie algebra, or a Filippov algebra, was introduced in [8] and have many applications in mathematical
physics. See the review article [6] for more details. Nambu-Poisson structures were introduced in [24]
by Takhtajan in order to find an axiomatic formalism for Nambu-mechanics which is a generalization of
Hamiltonian mechanics
2
Definition 2.1. An n-Lie algebra is a vector space g together with an n-multilinear skew-symmetric
bracket [·, · · · , ·]g : ∧ng −→ g such that for all ui, vi ∈ g, the following Fundamental Identity is satisfied:
[u1, u2, · · · , un−1, [v1, v2, · · · , vn]g]g =
nX
i=1
[v1, v2, · · · , [u1, u2, · · · , un−1, vi]g, · · · , vn]g.
(5)
For u1, u2, · · · , un−1 ∈ g, define ad : ∧n−1g −→ gl(g) by
adu1,u2,··· ,un−1v = [u1, u2, · · · , un−1, v]g,
∀ v ∈ g.
Then Eq. (5) is equivalent to that adu1,u2,··· ,un−1 is a derivation, i.e.
adu[v1, v2, · · · , vn]g =
nX
i=1
[v1, v2, · · · , aduvi, · · · , vn]g,
∀ u = u1 ∧ u2 ∧ · · · ∧ un−1 ∈ ∧n−1g.
(6)
Elements in ∧n−1g are called fundamental objects of the n-Lie algebra (g, [·, · · · , ·]g). In the sequel,
we will denote aduv simply by u ◦ v.
Define a bilinear operation on the set of fundamental objects ◦ : (∧n−1g) ⊗ (∧n−1g) −→ ∧n−1g by
u ◦ v =
n−1X
i=1
v1 ∧ · · · ∧ vi−1 ∧ u ◦ vi ∧ vi+1 ∧ · · · ∧ vn−1,
(7)
for all u = u1 ∧ u2 ∧ · · · ∧ un−1 and v = v1 ∧ v2 ∧ · · · ∧ vn−1. In [5], the authors proved that (∧n−1g, ◦)
is a Leibniz algebra. See [21] for details about Leibniz algebras, which are also called Loday algebras.
Moreover, the Fundamental Identity (5) is equivalent to
u ◦ (v ◦ w) − v ◦ (u ◦ w) = (u ◦ v) ◦ w,
∀ u, v ∈ ∧n−1g, w ∈ g.
(8)
Definition 2.2. [24] A Nambu-Poisson structure of order n − 1 on M is an n-linear map {·, · · · , ·} :
C∞(M ) × · · · × C∞(M ) −→ C∞(M ) satisfying the following properties:
(1) skewsymmetry, i.e. for all f1, · · · , fn ∈ C∞(M ) and σ ∈ Sym(n),
{f1, · · · , fn} = (−1)σ{fσ(1), · · · , fσ(n)};
(2) the Leibniz rule, i.e. for all g ∈ C∞(M ), we have
{f1g, · · · , fn} = f1{g, · · · , fn} + g{f1, · · · , fn};
(3) integrability condition, i.e. for all f1, · · · , fn−1, g1, · · · , gn ∈ C∞(M ),
{f1, · · · , fn−1, {g1, · · · , gn}} =
nX
i=1
{g1, · · · , {f1, · · · , fn−1, gi}, · · · , gn}.
In particular, a Nambu-Poisson structure of order 1 is exactly a usual Poisson structure. Similar to
the fact that a Poisson structure is determined by a bivector field, a Nambu-Poisson structure of order
n − 1 is determined by an n-vector field π ∈ Xn(M ) such that
{f1, · · · , fn} = π(df1, · · · , dfn).
3
An n-vector field π ∈ Xn(M ) is a Nambu-Poisson structure if and only if for all f1, · · · , fn−1 ∈ C∞(M ),
we have
where π♯ : ∧n−1T ∗M −→ T M is defined by
Lπ♯(df1∧···∧dfn−1)π = 0,
hπ♯(ξ1 ∧ · · · ∧ ξn−1), ξni = π(ξ1 ∧ · · · ∧ ξn−1 ∧ ξn),
∀ξ1, · · · , ξn ∈ Ω1(M ).
Let π be a Nambu-Poisson structure on a manifold M . Then there is a natural operation [·, ·]π on
Ωn−1(M ) given by
[α, β]π = Lπ♯(α)β − Lπ♯(β)α + diπ♯(β)α,
∀ α, β ∈ Ωn−1(M )
(9)
such that (∧n−1T ∗M, [·, ·]π, π♯) is a Leibniz algebroid. See [2, 13] for more details.
3 Omni n-Lie algebras
Let V be a finite dimensional vector space. For all A ∈ gl(V ), define LA : ⊗n−1V −→ ⊗n−1V by
LA(v1 ⊗ · · · ⊗ vn−1) =
n−1X
i=1
v1 ⊗ · · · ⊗ Avi ⊗ · · · ⊗ vn−1.
Definition 3.1. An omni n-Lie algebra associated to a vector space V is a triple (gl(V )⊕∧n−1V, (·, ·)+, {·, ·}),
where {·, ·} is the bilinear bracket operation given by
{A + u, B + v} = [A, B] + LAv,
(10)
and (·, ·)+ is the (V ⊗ ∧n−2V )-valued pairing given by
(A + u, B + v)+ =
n−1X
(−1)i+1(cid:0)Avi ⊗ v1 ∧ · · · ∧ vi ∧ · · · ∧ vn−1 + Bui ⊗ u1 ∧ · · · ∧ ui ∧ · · · ∧ un−1(cid:1), (11)
i=1
where u = u1 ∧ u2 ∧ · · · ∧ un−1 and v = v1 ∧ v2 ∧ · · · ∧ vn−1.
Remark 3.2. When n = 2, we recover Weinstein's omni-Lie algebras [26].
Proposition 3.3. With the above notations, (gl(V ) ⊕ ∧n−1V, {·, ·}) is a Leibniz algebra. Furthermore,
the pairing (·, ·)+ and the bracket {·, ·} are compatible in the sense that
({e1, e2}, e3)+ + (e2, {e1, e3})+ = ρV (e1)(e2, e3)+,
where ei ∈ gl(V ) ⊕ ∧n−1V, i = 1, 2, 3, and ρV : gl(V ) ⊕ ∧n−1V −→ gl(V ⊗ ∧n−2V ) is given by
ρV (A + u)(w) = LAw,
∀ w ∈ V ⊗ ∧n−2V.
(12)
(13)
Proof. Since L[A,B] = LALB − LBLA, we can deduce that (gl(V ) ⊕ ∧n−1V, {·, ·}) is a Leibniz algebra
directly.
4
For all A + u, B + v, C + w ∈ gl(V ) ⊕ ∧n−1V , on one hand, we have
({A + u, B + v}, C + w)+ + (B + v, {A + u, C + w})+
= ([A, B] + LAv, C + w)+ + (B + v, [A, C] + LAw)+
n−1X
(−1)j+1(cid:0)ABwj ⊗ w1 ∧ · · · ∧ wj ∧ · · · ∧ wn−1 + ACvj ⊗ v1 ∧ · · · ∧ vj ∧ · · · ∧ vn−1(cid:1)
=
j=1
n−1X
(−1)i+1(cid:0)Bwi ⊗ w1 ∧ · · · ∧ Awj ∧ · · · ∧ wn−1) + Cvi ⊗ v1 ∧ · · · ∧ Avj ∧ · · · ∧ vn−1(cid:1).
+
i6=j
On the other hand, we have
ρV (A + u)(B + v, C + w)+
=
=
n−1X
(−1)j+1ρV (A + u)(cid:0)Bwj ⊗ w1 ∧ · · · ∧ wj ∧ · · · ∧ wn−1 + Cvj ⊗ v1 ∧ · · · ∧ vj ∧ · · · ∧ vn−1(cid:1)
j=1
n−1X
(−1)j+1(cid:0)ABwj ⊗ w1 ∧ · · · ∧ wj ∧ · · · ∧ wn−1 + ACvj ⊗ v1 ∧ · · · ∧ vj ∧ · · · ∧ vn−1(cid:1)
j=1
n−1X
(−1)i+1(cid:0)Bwi ⊗ w1 ∧ · · · ∧ Awj ∧ · · · ∧ wn−1 + Cvi ⊗ v1 ∧ · · · ∧ Avj ∧ · · · ∧ vn−1(cid:1),
+
i6=j
which implies that (12) holds.
Let F : ∧nV −→ V be a linear map. Then F induces a linear map F ♯ : ∧n−1V −→ gl(V ) by
Denote by GF ⊂ gl(V ) ⊕ ∧n−1V the graph of F ♯.
F ♯(u)(u) = F (u, u),
∀u ∈ ∧n−1V, u ∈ V.
Theorem 3.4. Let F : ∧nV −→ V be a linear map. Then (V, F ) is an n-Lie algebra if and only if GF
is a Leibniz subalgebra of the Leibniz algebra (gl(V ) ⊕ ∧n−1V, {·, ·}).
Proof. GF is a Leibniz subalgebra of the Leibniz algebra (gl(V ) ⊕ ∧n−1V, {·, ·}) if and only if for all
u, v ∈ ∧n−1V , {F ♯(u) + u, F ♯(v) + v} ∈ GF , which is equivalent to
Since LF ♯(u)v = Pn−1
i=1 v1 ∧ · · · ∧ F (u, vi) ∧ · · · ∧ vn−1, thus the above equality can be written as
F ♯(LF ♯(u)v) = [F ♯(u), F ♯(v)].
which is equivalent to that (V, F ) is an n-Lie algebra.
F ♯(u ◦ v) = [F ♯(u), F ♯(v)],
4 Linearization of the higher analogue of the standard Courant
algebroid
Let V be an m-dimensional vector space and V ∗ its dual space. We consider the direct sum bundle
T n−1V ∗ = T V ∗ ⊕ ∧n−1T ∗V ∗. Denote the vector spaces of linear vector fields and constant (n − 1)-forms
5
con (V ∗) respectively. It is obvious that Xlin(V ∗) ⊕ Ωn−1
con (V ∗) ∼= gl(V ) ⊕ ∧n−1V .
on V ∗ by Xlin(V ∗) and Ωn−1
To make this explicit, for any x ∈ V , denote by lx the corresponding linear function on V ∗. Let {xi} be
a basis of the vector space V . Then {lxi} forms a coordinate chart for V ∗. So { ∂
} constitutes a basis
∂lxi
of vector fields on V ∗ and {dlxi} constitutes a basis of 1-forms on V ∗. For A = (aij ) ∈ gl(V ), we get a
on V ∗. Also u = P1≤i1<···<in−1≤m ξi1···in−1xi1 ∧ · · · ∧ xin−1 defines
linear vector field bA = Pi lA(xi)
a constant (n − 1)-form bu = P1≤i1<···<in−1≤m ξi1···in−1 dlxi1 ∧ · · · ∧ dlxin−1 .
Define Φ : gl(V ) ⊕ ∧n−1V −→ Xlin(V ∗) ⊕ Ωn−1
con (V ∗) by
∂
∂lxi
Φ(A + u) = bA +bu.
Obviously, Φ is an isomorphism between vector spaces.
Any element v ⊗ u ∈ V ⊗ ∧n−2V will give rise to a linear (n − 2)-form v ⊗ u defined by
We give some useful formulas first.
v ⊗ u = lvbu.
Lemma 4.1. With the above notations, for all A ∈ gl(V ) and u ∈ ∧n−1V , we have
( bA,bu)+ = (A, u)+,
d( bA,bu)+ = dLAu,
LbAbu = dLAu,
[ bA, bB] = \[A, B].
Proof. On one hand, for u = P1≤i1<···<in−1≤m ξi1···in xi1 ∧ · · · ∧ xin−1 ∈ ∧n−1V , we have
(14)
(15)
(16)
(17)
( bA,bu)+ = X
1≤i1<···<in−1≤m
nX
j=1
On the other hand, we have
(−1)j+1ξi1···ij ···in lAxij dlxi1 ∧ · · · ∧ dlxij ∧ · · · ∧ dlxin−1 .
(A, u)+ =
X
n−1X
(−1)j+1ξi1···in (Axij ⊗ xi1 ∧ · · · ∧ xij ∧ · · · ∧ xin−1 )
1≤i1<···<in−1≤m
j=1
=
X
n−1X
(−1)j+1ξi1···ij ···in−1lAxij dlxi1 ∧ · · · ∧ dlxij ∧ · · · ∧ dlxin−1 ,
1≤i1<···<in−1≤m
j=1
which implies that (14) holds.
By direct calculation, we have
d( bA,bu)+ =
X
n−1X
(−1)j+1ξi1···ij ···in dlAxij ∧ dlxi1 ∧ · · · ∧ dlxij ∧ · · · ∧ dlxin−1
1≤i1<···<in≤m
j=1
=
X
nX
mX
1≤i1<···<in−1≤m
j=1
k=1
(−1)j−1ak,ij ξi1···ij ···in−1dlxk ∧ dlxi1 ∧ · · · ∧ dlxj ∧ · · · ∧ dlxin−1 .
6
On the other hand, we have
LAu = A( X
ξi1···ip xi1 ∧ · · · ∧ xin−1 )
1≤i1<···<in−1≤m
X
1≤i1<···<in−1≤m
n−1X
j=1
ξi1···in−1xi1 ∧ · · · ∧ Axij ∧ · · · ∧ xin−1
X
n−1X
mX
1≤i1<···<in−1≤m
j=1
k=1
ak,ij ξi1···in−1 xi1 ∧ · · · ∧ xk ∧ · · · ∧ xin−1 .
=
=
Thus (15) follows immediately.
(16) follows from
(17) is direct. We omit the details.
LbAbu = ιbA
dbu + dιbAbu = d( bA,bu)+ = dLAu.
Now we are ready to show that an omni n-Lie algebra can be viewed as linearization of the higher
analogue of the standard Courant algebroid.
Theorem 4.2. The omni n-Lie algebra (gl(V ) ⊕ ∧n−1V, (·, ·)+, {·, ·}) is induced from the higher analogue
con (V ∗).
of the standard Courant algebroid (T n−1(V ∗), (·, ·)+ ,J·, ·K , prT V ∗ ) via restriction to Xlin(V ∗) ⊕ Ωn−1
More precisely, we have
(Φ(A + u), Φ(B + v))+ = (A + u, B + v)+,
JΦ(A + u), Φ(B + v)K = Φ{A + u, B + v},
LprT V ∗ Φ(A+u)w = ρV (A + u)(w).
(18)
(19)
(20)
Proof. By (14), we have
(Φ(A + u), Φ(B + v))+ = ( bA,bv)+ + (bu, bB)+ = (A, u)+ + (v, B)+ = (A + u, B + v)+.
By (16) and (17), we have
JΦ(A + u), Φ(B + v)K = r bA +bu, bB +bvz = [ bA, bB] + LbAbv
= \[A, B] + dLAv = Φ{A + u, B + v}.
Finally, for all w = w1 ⊗ w2 ∧ · · · ∧ wn−1 ∈ V ⊗ ∧n−2V , we have
LprT V ∗ Φ(A+u)w = LbA(lw1dw2 ∧ · · · ∧ dwn−1)
= LbA(lw1)dw2 ∧ · · · ∧ dwn−1 + lw1(
dw2 ∧ · · · ∧ LbA
dwi ∧ · · · ∧ dwn−1)
= lAw1dw2 ∧ · · · ∧ dwn−1 +
lw1dw2 ∧ · · · ∧ d(Awi) ∧ · · · ∧ dwn−1)
n−1X
i=2
n−1X
= Aw1 ⊗ w2 ∧ · · · ∧ wn−1 +
= ρV (A + u)(w).
i=2
n−1X
i=2
w1 ⊗ w2 ∧ · · · ∧ Awi ∧ · · · ∧ wn−1
The proof is finished.
7
5 Nonabelian omni n-Lie algebras
Definition 5.1. A nonabelian omni n-Lie algebra associated to an n-Lie algebra (g, [·, · · · , ·]g) is
a triple (gl(g) ⊕ ∧n−1g, (·, ·)+, {·, ·}g), where (·, ·)+ is the (g ⊗ ∧n−1g)-valued pairing given by (11) and
{·, ·}g is the bilinear bracket operation given by
{A + u, B + v}g = [A, B] + [A, adv] + [adu, B] − adLAv + LAv + u ◦ v,
∀ A, B ∈ gl(g), u, v ∈ ∧n−1g. (21)
Theorem 5.2. Let (gl(g) ⊕ ∧n−1g, (·, ·)+, {·, ·}g) be a nonabelian omni n-Lie algebra. Then we have
(i) (gl(g) ⊕ ∧n−1g, {·, ·}g) is a Leibniz algebra;
(ii) {A + u, A + u}g = −adLAu + LAu + u ◦ u;
(iii) the pairing (·, ·)+ and the bracket {·, ·}g are compatible in the sense that
ρg(e1)(e2, e3)+ = ({e1, e2}g − ([A, adv] − adLAv), e3)+ + (e2, {e1, e3}g − ([A, adw] − adLAw))+, (22)
where e1 = A + u, e2 = B + v, e3 = C + w ∈ gl(g) ⊕ ∧n−1g and ρg : gl(g) ⊕ ∧n−1g −→ gl(g ⊗ ∧n−2g)
is given by
ρg(A + u)(w) = LA+adu w,
∀ w ∈ g ⊗ ∧n−2g.
(23)
Proof. (i) We can prove that (gl(g) ⊕ ∧n−1g, {·, ·}g) is a Leibniz algebra directly by a complicated
computation. In the sequel, we will show that (gl(g) ⊕ ∧n−1g, {·, ·}g) is a trivial deformation of the omni
n-Lie algebra (gl(g) ⊕ ∧n−1g, {·, ·}). Thus, we omit details here.
(ii) It follows from (21) directly.
(iii) By straightforward computation, we have
([A, B] + [adu, B] + LAv + u ◦ v, C + w)+
n−1X
+X
i=1
=
(−1)i+1([A, B] + [adu, B])wi ⊗ w1 ∧ · · · ∧ wi ∧ · · · ∧ wn−1
(−1)j+1Cvj ⊗ v1 ∧ · · · ∧ vj ∧ · · · ∧ (Avi + u ◦ vi) ∧ · · · ∧ vn−1
i6=j
n−1X
+
On the other hand, we have
i=1
(−1)i+1(CAvi + C(u ◦ vi)) ⊗ v1 ∧ · · · ∧ bvi ∧ · · · ∧ vn−1.
(B + v, [A, C] + [adu, C] + LAw + u ◦ w)+
=
n−1X
+X
i=1
(−1)i+1([A, C] + [adu, C])vi ⊗ v1 ∧ · · · ∧ bvi ∧ · · · ∧ vn−1
(−1)j+1(Bwj ⊗ w1 ∧ · · · ∧ wj ∧ · · · ∧ (Awi + u ◦ wi) ∧ · · · ∧ wn−1)
i6=j
n−1X
(−1)i+1(BAwi + B(u ◦ wj)) ⊗ w1 ∧ · · · ∧ wi ∧ · · · ∧ wn−1.
+
i=1
8
Thus we have
([A, B] + [adu, B] + LAv + u ◦ v, C + w)+ + (B + v, [A, C] + [adu, C] + LAw + u ◦ w)+
n−1X
(−1)i+1(A ◦ B + adu ◦ B)wi ⊗ w1 ∧ · · · ∧ wi ∧ · · · ∧ wn−1
=
i=1
+
(−1)i+1(A ◦ C + adu ◦ C)vi ⊗ v1 ∧ · · · ∧ vi ∧ · · · ∧ vn−1
(−1)j+1Bvj ⊗ v1 ∧ · · · ∧ vj ∧ · · · ∧ (Avi + u ◦ vi) ∧ · · · ∧ vn−1
i=1
n−1X
+X
+X
i6=j
(−1)j+1Bwj ⊗ w1 ∧ · · · ∧ wj ∧ · · · ∧ (Awi + u ◦ wi) ∧ · · · ∧ wn−1
i6=j
= ρg(A + u)(B + v, C + w)+.
The proof is finished.
Obviously, for all A ∈ Der(g), we have [A, adv] − adLAv = 0. Thus, we have
Corollary 5.3. For all e1, e2, e3 ∈ Der(g) ⊕ ∧n−1g, we have
ρg(e1)(e2, e3)+ = ({e1, e2}g, e3)+ + (e2, {e1, e3}g)+.
In the sequel we show that a nonabelian omni n-Lie algebra (gl(g) ⊕ ∧n−1g, {·, ·}g) can be viewed
as a trivial deformation of the omni n-Lie algebra (gl(g) ⊕ ∧n−1g, {·, ·}). For details of deformations of
Leibniz algebras, see [4, 16].
Let (L, [·, ·]L) be a Leibniz algebra. For an endomorphism N of L, define
[e1, e2]N = [N e1, e2]L + [e1, N e2]L − N [e1, e2]L,
and set
T N (e1, e2) = [N e1, N e2]L − N [e1, e2]N .
The endomorphism N is called a Nijenhuis operator if T N = 0.
A Nijenhuis operator gives a trivial deformation of the Leibniz algebra (L, [·, ·]L).
Proposition 5.4. [4] Let N be a Nijenhuis operator on the Leibniz algebra (L, [·, ·]L). Then we have
(1) (L, [·, ·]N ) is a Leibniz algebra;
(2) N is a morphism of Leibniz algebras from (L, [·, ·]N ) to (L, [·, ·]L);
(3) (L, [·, ·]L + [·, ·]N ) is a Leibniz algebra.
Let (g, [·, · · · , ·]g) be an n-Lie algebra. Then we can define a linear map N : gl(g) ⊕ ∧n−1g −→
gl(g) ⊕ ∧n−1g by
N (A + u) = adu.
(24)
Lemma 5.5. The linear map N given by (24) is a Nijenhuis operator on the Leibniz algebra (gl(g) ⊕
∧n−1g, {·, ·}), where the Leibniz bracket {·, ·} is given by (10).
9
Proof. First by definition, we have
{A + u, B + v}N = {N (A + u), B + v} + {A + u, N (B + v)} − N {A + u, B + v}
= [adu, B] + Ladu v + [A, adv] − adLAv
= [adu, B] + u ◦ v + [A, adv] − adLAv.
Hence it is clear that
N {A + u, B + v}N = adu◦v = [adu, adv] = {N (A + u), N (B + v)},
which says that N is a Nijenhuis operator.
It is straightforward to see that
{A + u, B + v}g = {A + u, B + v} + {A + u, B + v}N .
Therefore, by Proposition 5.4 and Lemma 5.5, we have
Theorem 5.6. Let (g, [·, · · · , ·]g) be an n-Lie algebra. Then the bracket {·, ·}g is a trivial deformation of
the Leibniz bracket {·, ·}. In particular, (gl(g) ⊕ ∧n−1g, {·, ·}g) is a Leibniz algebra.
Remark 5.7. If we view (gl(g), [·, ·]) as a Leibniz algebra, then (gl(g), [·, ·]) and (∧n−1g, ◦) form a matched
pair of Leibniz algebras and the Leibniz algebra (gl(g) ⊕ ∧n−1g, {·, ·}g) is exactly their double. See [1] for
more details about matched pairs of Leibniz algebras.
6 Linearization of higher analogues of Courant algebroids asso-
ciated to Nambu-Poisson structures
Γ(T n−1M ) by
Let (M, π) be a Nambu-Poisson manifold. We introduce a bracket J·, ·Kπ : Γ(T n−1M ) × Γ(T n−1M ) −→
JX + α, Y + βKπ = [X, Y ] + [X, π♯(β)] + [π♯(α), Y ] − π♯(LX β) + π♯(iY dα) + LX β − iY dα + [α, β]π, (25)
where X, Y ∈ X(M ), α, β ∈ Ωn−1(M ) and [·, ·]π is given by (9).
Let ρπ : T n−1M → T M be the bundle map defined by
ρπ(X + α) = X + π♯(α),
∀ X ∈ X(M ), α ∈ Ωn−1(M ).
(26)
We call the quadruple (T n−1M, (·, ·)+,J·, ·Kπ , ρπ) the higher analogue of the Courant algebroid
associated to a Nambu-Poisson manifold and denote it by T n−1
π M . In the sequel, we will see that
even though we call it the higher analogue of a Courant algebroid, some important properties for Courant
algebroids do not hold anymore.
to a Nambu-Poisson manifold. Then we have
Theorem 6.1. Let (T n−1M, (·, ·)+,J·, ·Kπ , ρπ) be the higher analogue of the Courant algebroid associated
(i) (T n−1M,J·, ·Kπ , ρπ) is a Leibniz algebroid.
(ii) The bracket J·, ·Kπ is not skew-symmetric. Instead, we have
JX + α, X + αKπ = d(X, α)+ + d(π♯(α), α)+ − π♯(d(X, α)+).
(27)
10
(iii) The pairing (1) and the bracket J·, ·Kπ are compatible in the following sense:
Lρ(e1)(e2, e3)+ = (cid:0)Je1, e2Kπ − ([X, π♯(β)] − π♯(LX β) + π♯(iY dα)) + iπ♯(β)dα, e3(cid:1)+
+(cid:0)e2,Je1, e3Kπ − ([X, π♯(γ)] − π♯(LX γ) + π♯(iZ dα)) + iπ♯(γ)dα(cid:1)+
where e1 = X + α, e2 = Y + β, e3 = Z + γ.
Proof. (i) Let Ψ : T n−1M −→ T n−1M be the invertible bundle map defined by
Ψ(X + α) = X + α + π♯(α),
∀ X ∈ X(M ), α ∈ Ωn−1(M ).
By direct calculation, we have
.
(28)
(29)
ΨJX + α, Y + βKπ = JΨ(X + α), Ψ(Y + β)K ,
where the bracket J·, ·K is given by (2). Since (Γ(T n−1M ),J·, ·K) is a Leibniz algebra, we deduce that
(Γ(T n−1M ),J·, ·Kπ) is also a Leibniz algebra.
For all f ∈ C∞(M ), X, Y ∈ X(M ) and α, β ∈ Ωn−1(M ), we have
JX + α, f (Y + β)Kπ = [X, f Y ] + [X, f π♯(β)] + [π♯(α), f Y ] − π♯(LX f β)
+π♯(if Y dα) + LXf β − if Y dα + [α, f β]π
= f [X, Y ] + X(f )Y + f [X, π♯(β)] + X(f )π♯(β) + f [π♯(α), Y ]
+π♯(α)(f )Y − f π♯(LX β) − X(f )π♯(β) + f π♯(iY dα) + f LXβ
+X(f )β − f iY dα + f [α, β]π + π♯(α)(f )β
= f JX + α, Y + βKπ + (X + π♯(α))(f )(Y + β).
Thus, (T n−1M ),J·, ·Kπ , ρπ) is a Leibniz algebroid.
(ii) It is straightforward to obtain (27) by (25).
(iii) The left hand side of the above equality is equal to
The right hand side is equal to
LXiY γ + LXiZ β + Lπ♯(α)iY γ + Lπ♯(α)iZ β.
iZ LX β + iZ Lπ♯(α)β − iZ Lπ♯(β)α + iZ diπ♯(β)α + i[X,Y ]γ
+i[X,π♯(β)]γ + i[π♯(α),Y ]γ − iπ♯(LX β)γ + iπ♯(iY dα)γ
+iY LX γ + iY Lπ♯(α)γ − iY Lπ♯(γ)α + iY diπ♯(γ)α + i[X,Z]β
+i[X,π♯(γ)]β + i[π♯(α),Z]β − iπ♯(LX γ)β + iπ♯(iZ dα)β.
The conclusion follows from
This finishes the proof.
i[X,Y ] = LXiY − iY LX.
Let XH (M ) and Ωn−1
cl
(M ) denote the set of the Hamiltonian vector fields and closed (n − 1)-forms
respectively.
Corollary 6.2. For all e1, e2, e3 ∈ XH (M ) ⊕ Ωn−1
cl
(M ), we have
Lρπ (e1)(e2, e3)+ = (Je1, e2Kπ , e3)+ + (e2,Je1, e3Kπ)+ .
(30)
11
Proof. For all e1 = X + α, e2 = Y + β, e3 = Z + γ ∈ XH (M ) ⊕ Ωn−1
cl
(M ), since α is closed, we have
iπ♯(iY dα)γ − iY iπ♯(γ)dα + iπ♯(iZ dα)β − iZ iπ♯(β)dα = 0.
For all ξ, η ∈ Ωn−1(M ), we have the following formula
π♯(Lπ♯(ξ)η) − [π♯(ξ), π♯(η)] = (−1)n−1(idξπ)π♯(η).
Without loss of generality, let X = π♯(df1 ∧ · · · ∧ dfn−1), then we have
i[X,π♯(β)]γ − iπ♯(LX β)γ + i[X,π♯(γ)]β − iπ♯(LX γ)β
= (−1)ni(id(df1 ∧···∧dfn−1)π)π♯(β)γ + (−1)ni(id(df1 ∧···∧dfn−1)π)π♯(γ)β = 0.
We finishes the proof.
In the following, we show that the nonabelian omni n-Lie algebra is a linearization of the higher
analogue of the Courant algebroid (T n−1M, (·, ·)+,J·, ·Kπ , ρπ) associated to a Nambu-Poisson manifold
(M, π).
Let (g, [·, · · · , ·]g) be an m-dimensional n-Lie algebra such that it induces a linear Nambu-Poisson
g∗. Let {xi} be
structure1 πg on g∗. Then we obtain the higher analogue of the Courant algebroid T n−1
a basis of the vector space g. Using the same notations as in Section 4, we have
πg
πg = X
1≤i1<···<in≤m
l[xi1 ,··· ,xin ]g
∂
∂lxi1
∧ · · · ∧
∂
∂lxin
.
Lemma 6.3. For all A ∈ gl(g) and u, v ∈ ∧n−1g, we have
g(bu) = dadu,
π♯
= du ◦ v,
[bu,bv]πg
π♯(LbAbu) = \adLAu.
(31)
(32)
(33)
Proof. For u = P1≤j1<···<jn−1≤m ξj1···jn−1 xj1 ∧ · · · ∧ xjn−1 ∈ ∧n−1g with the corresponding constant
(n − 1)-form bu = P1≤j1<···<jn−1≤m ξj1···jn−1 dlxj1 ∧ · · · ∧ dlxjn−1 , we have
π♯
g(bu) =
=
=
X
nX
1≤i1<···<in≤m
k=1
X
nX
1≤i1<···<in≤m
k=1
(−1)n−kξi1··· ik···in
l[xi1 ,··· ,xik ,··· ,xin ]g
∂
∂lxik
ξi1··· ik···in
l[xi1 ,··· ,xin ,xik ]g
∂
∂lxik
X
mX
1≤j1<···<jn−1≤m
l=1
ξj1···jn−1 l[xj1 ,··· ,xjn−1 ,xl]g
∂
∂lxl
= dadu,
which implies that (31) holds.
1Not all n-Lie algebras can give linear Nambu-Poisson structures on dual spaces, see [7] for details.
12
Since bu is a constant (n − 1)-form, by (16) and (31), we have
[bu,bv]πg
= Lπ♯(bu)bv − Lπ♯(bv)bu + diπ♯(bv)bu
= Lπ♯(bu)bv − iπ♯(bv)dbu
= Lπ♯(bu)bv = Lcadubv
= \Ladu v = du ◦ v,
which implies that (32) holds.
By (16) and (31), we have
This ends the proof.
π♯(LbAbu) = π♯(dLAu) = \adLAu.
Theorem 6.4. Let (g, [·, · · · , ·]g) be an m-dimensional n-Lie algebra such that it induces a linear Nambu-
Poisson structure πg on g∗. Then the nonabelian omni n-Lie algebra (gl(g) ⊕ ∧n−1g, (·, ·)+, {·, ·}g) is
, ρπg) associated to the
induced from the higher analogue of the Courant algebroid (T n−1g∗, (·, ·)+,J·, ·Kπg
Nambu-Poisson manifold (g∗, πg) via restriction to Xlin(g∗) ⊕ Ωn−1
con (g∗). More precisely, we have
(Φ(A + u), Φ(B + v))+ = (A + u, B + v)+,
JΦ(A + u), Φ(B + v)Kπg
= Φ{A + u, B + v}g,
Lρπg Φ(A+u)w = ρg(A + u)(w).
(34)
(35)
(36)
Proof. (34) has been proved in Theorem 4.2. By (15)-(17) and (31)-(33), we deduce that
r bA +bu, bB +bvzπg
= [ bA, bB] + [ bA, π♯(bv)] + [π♯(bu), bB] − π♯(LbAbv) + π♯(ibB
= [ bA, bB] + [ bA,dadv] + [dadu, bB] − \adLAv + dLAv + du ◦ v
= \[A, B] + \[A, adv] + \[adu, B] − \adLAv + dLAv + du ◦ v
= Φ{A + u, B + v}g,
dbu) + LbAbv − ibB
dbu + [bu,bv]π
which implies that (35) holds. By (17), we have LbA
LρπΦ(A+u)w = LbA+π♯(bu)
w = LbA
w + Lcadu
w
w = ρg(A)(w). Thus
= ρg(A)(w) + ρg(adu)(w) = ρg(A + u)(w),
which says that (36) holds. This ends the proof.
References
[1] A. L. Agore and G. Militaru, Unified products for Leibniz algebras. Applications., Linear Algebra
Appl. 2013, 439(9): 2609-2633.
[2] Y. Bi and Y. Sheng, On higher analogues of Courant algebroids, Sci. China Math. Vol. 54 (3) (2011),
437-447.
13
[3] P. Bouwknegt and B. Jurcč, AKSZ construction of topological open p-brane action and Nambu
brackets, Rev. Math. Phys. 25 (2013), 1330004, 31 pages.
[4] J. Cariñena, J. Grabowski and G. Marmo, Courant algebroid and Lie bialgebroid contractions, J.
Phys. A: Math. Gen. 2004, 37(19): 5189-5202.
[5] Y. Daletskii and L. Takhtajan, Leibniz and Lie algebra structures for Nambu algebra, Lett. Math.
Phys. 39 (1997), 127-141.
[6] J. A. de Azc´arraga and J. M. Izquierdo, n-ary algebras: a review with applications, J. Phys. A: Math.
Theor. 43 (2010), 293001.
[7] J. P. Dufour and N. T. Zung, Linearization of Nambu structures, Compositio Math. 117 (1999), no.
1, 77 -- 98.
[8] V. T. Filippov, n-Lie algebras, Sib. Mat. Zh. 26 (1985) 126-140.
[9] J. Grabowski, Brackets, Int. J. Geom. Methods Mod. Phys. 10 (2013), 1360001, 45 pages.
[10] M. Grützmann and T. Strobl, General Yang-Mills type gauge theories for p-form gauge fields: from
physics-based ideas to a mathematical framework or from Bianchi identities to twisted Courant alge-
broids, Int. J. Geom. Methods Mod. Phys. 12 (2015), no. 1, 1550009, 80 pp.
[11] Y. Hagiwara, Nambu -- Dirac manifolds, J. Phys. A 35(5) (2002) 1263-1281.
[12] C. M. Hull, Generalised geometry for M-theory, J. High Energy Phys. 07 (2007) 079.
[13] R. Ibáñez, M. de León, J. Marrero and E. Padrón, Leibniz algebroid associated with a Nambu-
Poisson structure, J. Phys. A 32 (46): 8129-8144, 1999
[14] B. Jurčo and J. Vysoký, Leibniz algebroids, generalized Bismut connections and Einstein -- Hilbert
actions, J. Geom. Phys. 97 (2015), 25-33.
[15] M. K. Kinyon and A. Weinstein, Leibniz algebras, Courant algebroids, and multiplications on
reductive homogeneous spaces, Amer. J. Math. 123 (3): 525-550, 2001.
[16] Y. Kosmann-Schwarzbach, Nijenhuis structures on Courant algebroids, Bull. Braz. Math. Soc. (N.S.)
2011, 42(4): 625-649.
[17] Y. Kosmann-Schwarzbach, Courant algebroids. A short history, SIGMA Symmetry Integrability
Geom. Methods Appl. 9 (2013), Paper 014, 8 pp.
[18] H. Lang, Y. Sheng and X. Xu, Strong homotopy Lie algebras, homotopy Poisson manifolds and
Courant algebroids, Lett. Math. Phys. (2016), doi:10.1007/s11005-016-0925-8.
[19] H. Lang, Y. Sheng and X. Xu, Nonabelian omni-Lie algebras, Banach Center Publications, 110
(2016), 167-176.
[20] Z. Liu, A. Weinstein and P. Xu, Manin triples for Lie bialgebroids, J. Diff. Geom. 1997, 45(3):
547-574.
[21] J. L. Loday and T. Pirashvilo, Universal enveloping algebras of Leibniz algebras and (co)homology,
Math. Ann. 1993, 296 (1): 139-158.
14
[22] D. Roytenberg, Courant algebroids, derived brackets and even symplectic supermanifolds, PhD thesis,
UC Berkeley, 1999, arXiv: math.DG/9910078.
[23] Y. Sheng and C. Zhu, Semidirect products of representations up to homotopy, Pacific J. Math. 2011,
249(1): 211-236.
[24] L. Takhtajan, On foundation of the generalized Nambu mechanics, Comm. Math. Phys. 160(2):295-
315, 1994
[25] K. Uchino. Courant brackets on noncommutative algebras and omni-Lie algebras, Tokyo J. Math.
30(1):239-255, 2007.
[26] A. Weinstein, Omni-Lie algebras, Microlocal analysis of the Schrödinger equation and related topics
(Japanese) (Kyoto, 1999), S¯urikaisekikenky¯usho K¯uky¯uroku, 2000, 1176: 95-102.
[27] M. Zambon, L∞-algebras and higher analogues of Dirac structures and Courant algebroids, J. Sym-
plectic Geom. 10(4) (2012), 563-599.
15
|
1602.02195 | 2 | 1602 | 2016-02-23T02:00:03 | Semiclassical limits of Ore extensions and a Poisson generalized Weyl algebra | [
"math.RA"
] | We observe \cite[Proposition 4.1]{LaLe} that Poisson polynomial extensions appear as semiclassical limits of a class of Ore extensions. As an application, a Poisson generalized Weyl algebra $A_1$ considered as a Poisson version of the quantum generalized Weyl algebra is constructed and its Poisson structures are studied. In particular, it is obtained a necessary and sufficient condition such that $A_1$ is Poisson simple and established that the Poisson endomorphisms of $A_1$ are Poisson analogues of the endomorphisms of the quantum generalized Weyl algebra. | math.RA | math |
SEMICLASSICAL LIMITS OF ORE EXTENSIONS AND A POISSON
GENERALIZED WEYL ALGEBRA
EUN-HEE CHO AND SEI-QWON OH
Abstract. We observe [8, Proposition 4.1] that Poisson polynomial extensions appear as semi-
classical limits of a class of Ore extensions. As an application, a Poisson generalized Weyl algebra
A1 considered as a Poisson version of the quantum generalized Weyl algebra is constructed and
its Poisson structures are studied. In particular, it is obtained a necessary and sufficient condi-
tion such that A1 is Poisson simple and established that the Poisson endomorphisms of A1 are
Poisson analogues of the endomorphisms of the quantum generalized Weyl algebra.
1. Introduction
Let h be a nonzero, nonunit, non-zero-divisor and central element of an algebra R such that
R/hR is commutative. Then R/hR becomes a Poisson algebra with Poisson bracket
{a, b} = h−1(ab − ba)
for all a, b ∈ R/hR. In such case, R/hR is called a semiclassical limit of R. Since the Pois-
son bracket of R/hR is induced by the commutation rule of R, it is expected that the Poisson
structures of R/hR are heavily related to the algebraic structures of deformation algebras of R
since they are induced by the same algebra R. In fact, algebraic structures of quantized algebras
are analogues to Poisson structures of their semiclassical limits as seen in many cases [3], [4],
[6], [9], [11] and [12]. A main aim of this paper is to give a method how to construct Poisson
algebras considered as a Poisson version of algebras related to a class of Ore extensions and an
example illustrating this method. Namely, we observe [8, Proposition 4.1] that Poisson polyno-
mial extensions appear as semiclassical limits of a class of Ore extensions, construct a Poisson
generalized Weyl algebra A1 as an application and we verify that the Poisson endomorphisms
of A1 are Poisson analogues of the endomorphisms of quantum generalized Weyl algebra.
The Poisson polynomial extensions were constructed as a Poisson version of Ore extensions
by the second author in [10] and the quantum generalized Weyl algebra A(a(h), q) over a Lau-
rent polynomial ring in one variable was constructed by Bavula [1] and the endomorphisms of
A(a(h), q) were completely classified by Kitchin and Launois [7] in the case when a(h) is not
invertible and q is not a root of unity. Here we find a natural map Γ from Ore extensions onto
their semiclassical limits and then, as an application, we construct a Poisson generalized Weyl
algebra A1 induced from A(a(h), q) by using Γ. Next we find a necessary and sufficient condition
such that A1 is Poisson simple and establish that the Poisson endomorphisms of A1 are Poisson
analogues of the endomorphisms of A(a(h), q) by classifying the Poisson endomorphisms of A1.
2010 Mathematics Subject Classification. 16S36, 16W35, 17B63.
Key words and phrases. Poisson polynomial extension, Ore extension, Semiclassical limit, Quantum generalized
Weyl algebra.
1
2
EUN-HEE CHO AND SEI-QWON OH
Assume throughout the paper that all algebras have unity and that the base field is the
complex number field C.
Let us begin with recalling the following basic terminologies.
Definition 1.1. (1) Let F be a commutative C-algebra. Given an F-automorphism α on an
F-algebra R, an F-linear map δ is said to be a left α-derivation on R if
δ(ab) = δ(a)b + α(a)δ(b)
for all a, b ∈ R. For such pair (α, δ), there exists a skew polynomial F-algebra (or Ore extension)
R[z; α, δ]. Refer to [5, Chapter 2] for details of the skew polynomial ring.
(2) A commutative C-algebra A is said to be a Poisson algebra if there exists a bilinear
product {−, −} on A, called a Poisson bracket, such that (A, {−, −}) is a Lie algebra and
{ab, c} = a{b, c} + {a, c}b for all a, b, c ∈ A.
We recall [10, 1.1]. A derivation α on A is said to be a Poisson derivation if
α({a, b}) = {α(a), b} + {a, α(b)}
for all a, b ∈ A. Let α be a Poisson derivation and let δ be a derivation on A. If the pair (α, δ)
satisfies the following condition
(1.1)
δ({a, b}) − {δ(a), b} − {a, δ(b)} = α(a)δ(b) − δ(a)α(b)
for all a, b ∈ A, then the commutative polynomial algebra A[z] becomes a Poisson algebra with
Poisson bracket
{z, a} = α(a)z + δ(a)
for all a ∈ A. Such Poisson algebra A[z] is called a Poisson polynomial extension (or Poisson Ore
extension) and denoted by A[z; α, δ]p. (In [10, 1.1], {z, a} is defined by {z, a} = −α(a)z − δ(a)
for all a ∈ A.) If α = 0 then we write A[z; δ]p for A[z; 0, δ]p and if δ = 0 then we write A[z; α]p
for A[z; α, 0]p.
(3) An ideal I of a Poisson algebra A is said to be a Poisson ideal if {I, A} ⊆ I. A Poisson
ideal P is said to be Poisson prime if, for all Poisson ideals I and J, IJ ⊆ P implies I ⊆ P or
J ⊆ P . If A is noetherian then a Poisson prime ideal of A is a prime ideal by [3, Lemma 1.1(d)].
Let t be an indeterminate.
2. Polynomial extensions
Notation 2.1. Let a 5-tuple (K, F, A, t − 1, (α, δ)) satisfy the following conditions (1)-(5):
(1) Assume that K is an infinite subset of the set C \ {0, 1}.
(2) Assume that F is a subring of the ring of regular functions on K ∪ {1} containing C[t, t−1].
That is,
(2.1)
C[t, t−1] ⊆ F ⊆ {f /g ∈ C(t)f, g ∈ C[t] such that g(1) 6= 0, g(λ) 6= 0 ∀λ ∈ K}.
(3) Assume that A is an F-algebra generated by x1, . . . , xn. Note that A is also a C-algebra
since C ⊆ F.
(4) Assume that t − 1 is a nonzero, nonunit and non-zero-divisor of A such that the factor
A1 = A/(t − 1)A is commutative. (Note that t − 1 ∈ F is a central element of A and thus (t − 1)A
is an ideal of A.)
A POISSON GENERALIZED WEYL ALGEBRA
3
(5) Assume that α and δ are F-linear maps from A into itself such that α is an automorphism,
δ is a left α-derivation and the pair (α, δ) satisfies the condition
(2.2)
(α − id)(A) ⊆ (t − 1)A,
δ(A) ⊆ (t − 1)A,
where id is the identity map on A.
By Notation 2.1(5), there exists the skew polynomial F-algebra
B := A[z; α, δ].
For each λ ∈ K ∪ {1}, (t − λ)A and (t − λ)B are ideals of A and B respectively since t − λ is a
central element of A and B. Set
Aλ := A/(t − λ)A, Bλ := B/(t − λ)B.
For an element a of A or B, denote by a the canonical image of a in Aλ and Bλ. For each λ ∈ K,
define
αλ : Aλ −→ Aλ,
αλ(a) = α(a),
δλ : Aλ −→ Aλ,
δλ(a) = δ(a).
Lemma 2.2. (1) For each λ ∈ K, αλ is an automorphism and δλ is a left αλ-derivation.
(2) For each λ ∈ K, Bλ
∼= Aλ[z; αλ, δλ] as C-algebras.
Proof. (1) Since α and δ are F-linear maps and t − λ ∈ F, αλ and δλ are well-defined. Since α
is an automorphism on A, αλ is an automorphism on Aλ. Moreover δλ is a left αλ-derivation
because δ is a left α-derivation.
(2) Let ψ : A −→ Aλ[z; αλ, δλ] be the map defined by ψ(a) = a. Then, by [5, Proposition
2.4], there is an extension ψ of ψ to B such that ψ(z) = z since za = α(a)z + δ(a) for all a ∈ A.
Clearly ker ψ = (t − λ)B and ψ is surjective.
(cid:3)
Since t − 1 is a nonzero, nonunit and non-zero-divisor of A, it is also a nonzero, nonunit and
non-zero-divisor of B. Moreover B1 is a commutative C-algebra by (2.2). Hence A1 and B1 are
Poisson algebras with Poisson brackets
(2.3)
{a, b} = (t − 1)−1(ab − ba)
for all a, b ∈ A and a, b ∈ B by [2, III.5.4]. The C-algebras A1 and B1 are said to be semiclassical
limits of A and B respectively. For each λ ∈ K, the C-algebras Aλ and Bλ are said to be
deformations of A and B respectively.
Lemma 2.3. If b ∈ B is a central element then b ∈ B1 is a Poisson central element.
Proof. It is clear by (2.3).
(cid:3)
In the following theorem, note that the maps α1, δ1 are constructed in a different way from
αλ, δλ for λ ∈ K.
Theorem 2.4. [8, Proposition 4.1] B1 ∼= A1[z; α1, δ1]p as Poisson C-algebras, where α1 and δ1
are defined by
for all a ∈ A.
α1(a) = (t − 1)−1(α(a) − a),
δ1(a) = (t − 1)−1δ(a)
4
EUN-HEE CHO AND SEI-QWON OH
For each λ ∈ K, let Cλ = C. Note that, for f (t) ∈ F, the complex number f (λ) is well-defined
by (2.1). Define a C-algebra homomorphism
γF : F −→ Y
λ∈K
Cλ,
γF(f (t)) = (f (λ))λ∈K.
Lemma 2.5. The C-algebra homomorphism γF is injective.
Proof. Suppose that γF(f (t)) = 0. Then f (λ) = 0 for all λ ∈ K. Since K is an infinite set and
every nonzero polynomial has only finite zeros, f (t) = 0. Hence γF is injective.
(cid:3)
Set
Let
(2.4)
λ∈K
Aλ,
bA = Y
πλ : bA −→ Aλ,
λ∈K
Bλ.
bB = Y
πλ : bB −→ Bλ
γA : A −→ bA,
γB : B −→ bB,
γ(a) = (a)λ∈K,
γ(b) = (b)λ∈K.
be the canonical projections onto Aλ and Bλ for each λ ∈ K and let γA and γB be the C-algebra
homomorphisms
Thus πλγA(a) = a and πλγB(b) = b for each λ ∈ K.
Lemma 2.6. γA is injective if and only if γB is injective.
Proof. If γB is injective then γA is also injective since γA is the restriction of γB to A. Suppose
that γA is injective and that γB(Pi aizi) = 0, where ai ∈ A. Then ai = 0 for each λ ∈ K. It
follows γA(ai) = 0. Hence ai = 0 for all i since γA(ai) = 0 and γA is injective.
(cid:3)
Theorem 2.7. Suppose that γA is injective. Set
ΓA : γA(A) −→ A1,
ΓA(x) = γ−1
A (x),
ΓB(x) = γ−1
Then the C-algebra homomorphisms ΓA and ΓB are surjective.
ΓB : γB(B) −→ B1,
B (x).
Proof. By Lemma 2.6, γA and γB are injective and thus ΓA and ΓB are well-defined C-algebra
homomorphisms. Since the canonical maps from A into A1 is surjective, ΓA is surjective clearly.
Similarly ΓB is surjective.
(cid:3)
3. Poisson generalized Weyl algebra
The following quantum generalized Weyl algebra A(a(h), q) is a special case of the generalized
Weyl algebra introduced by Bavula in [1].
Definition 3.1. Let 0 6= q ∈ C be not a root of unity and let 0 6= a(h) ∈ C[h±1]. The
quantum generalized Weyl algebra A(a(h), q) is the C-algebra generated by h±1, x, y subject to
the relations
xh = qhx, yh = q−1hy, xy = a(qh), yx = a(h), h±1h∓1 = 1.
A POISSON GENERALIZED WEYL ALGEBRA
5
Set
F = C[t, t−1], K = C \ ({0, 1} ∪ {roots of unity}).
Assume throughout the section that 0 6= q ∈ C is not a root of unity and that 0 6= a(h) ∈
C[h±1] is not invertible. Hence q ∈ K and a(h) has at least two nonzero terms.
Set
where
(3.1)
B := F[h±1][x; α][y; β, δ],
α(h) = th,
β(h) = t−1h, β(x) = x,
δ(h) = 0,
δ(x) = a(h) − a(th).
Lemma 3.2. The element xy − a(th) ∈ B is a central element.
Proof. It is proved routinely by (3.1).
(cid:3)
Denote by B(a(h), q) the C-algebra generated by h±1, x, y subject to the relations
(3.2)
xh = qhx, yh = q−1hy, yx = xy + a(h) − a(qh), h±1h∓1 = 1,
which is obtained from B by substituting q for t. The C-algebra B(a(h), q) is an iterated skew
polynomial algebra
B(a(h), q) = C[h±1][x; αq][y; βq, δq],
where αq, βq, δq are the maps induced by α, β, δ respectively. Moreover
(3.3)
B(a(h), q) ∼= B/(t − q)B = Bq
as C-algebras by Lemma 2.2(2). For each λ ∈ K, λ is not a root of unity and thus there exists
the C-algebra B(a(h), λ) ∼= Bλ which is obtained from (3.3) by substituting λ for q. Observe
that q is not only a nonzero and non-root of unity but also plays a role as a parameter taking
values in K.
Observation 3.3. In B(a(h), q), q plays a role as a parameter taking values in K and thus, for
each λ ∈ K, there exists an evaluation map
(3.4)
eλ : B(a(h), q) −→ Bλ,
f (q) 7→ f (λ).
Note that the 5-tuples (K, F, F[h±1], t − 1, (α, 0)) and (K, F, F[h±1][x; α], t − 1, (β, δ)) satisfy
Notation 2.1(1)-(5). Applying Theorem 2.4 to F[h±1][x; α] and B, there exists the Poisson
C-algebra
B1 = C[h±1][x; α1]p[y; β1, δ1]p,
where
(3.5)
α1(h) = h,
β1(h) = −h, β1(x) = 0,
δ1(h) = 0,
here a′(h) is the formal derivative of a(h).
δ1(x) = −a′(h)h,
Define a C-algebra homomorphism γB by
γB : B −→ bB = Y
λ∈K
Bλ,
γB(z) = (z)λ∈K.
6
EUN-HEE CHO AND SEI-QWON OH
Applying Lemma 2.6 to γB inductively, γB is injective by Lemma 2.5 and thus, by Theorem 2.7,
there exists the C-algebra homomorphism
where γ1 : B −→ B1 = B/(t − 1)B is the canonical projection.
ΓB : γB(B) −→ B1, ΓB = γ1γ−1
B ,
Define a map (not C-linear map)
b : B(a(h), q) −→ bB = Y
λ∈K
Bλ,
f (q) 7→ df (q) := (f (λ))λ∈K
which exists by (3.4). Note that the image of b is equal to the image of γB. Hence there exists
the composition Γ := ΓB ◦b . Roughly speaking, Γ is a map defined by plugging 1 to q.
Bλ ←− B −→ B1.
(3.6)
Γ : B(a(h), q)−→bB = Y
λ∈K
Note that Γ is surjective. By Lemma 2.3 and Lemma 3.2, Γ(xy − a(qh)) = xy − a(h) is a Poisson
central element of B1. Thus (xy − a(h))B1 is a Poisson ideal. Set
Henceforth, we simply write w for w ∈ A1.
A1 := B1/(xy − a(h))B1.
Theorem 3.4. (1) The quantum generalized Weyl algebra A(a(h), q) is equal to the factored
C-algebra B(a(h), q)/(xy − a(qh))B(a(h), q).
(2) The Poisson algebra A1 is induced from A(a(h), q) by Γ in (3.6).
(3) The Poisson algebra A1 is the C-algebra C[h±1, x, y]/hxy − a(h)i with Poisson bracket
(3.7)
{x, h} = hx, {y, h} = −hy, {y, x} = −a′(h)h.
We will call A1 a Poisson generalized Weyl algebra.
Proof. (1) It follows by (3.2).
(2) It follows by the facts that Γ(B(a(h), q)) = B1 and Γ(xy − a(qh)) = xy − a(h).
(3) Since B1 = C[h±1, x, y], the result follows by (3.5).
(cid:3)
Let R be a Poisson algebra. If there exists subspaces Rk, k ∈ Z, such that R = Lk∈Z Rk and
RkRℓ ⊆ Rk+ℓ, {Rk, Rℓ} ⊆ Rk+ℓ
for all k, ℓ ∈ Z then R is said to be a Z-graded Poisson algebra.
Give degrees on the generators h, x, y of B1 by deg(h) = 0, deg(x) = 1 and deg(y) = −1.
Then B1 is a Z-graded Poisson algebra. Moreover the Poisson ideal hxy − a(h)i is graded. Thus
A1 is also a Z-graded Poisson algebra
(3.8)
where
Wk =
Wk,
k∈Z
A1 = M
C[h±1]xk
C[h±1]
C[h±1]y−k
if k > 0,
if k = 0,
if k < 0.
A POISSON GENERALIZED WEYL ALGEBRA
7
Lemma 3.5. Define a C-linear map f by
f : A1 −→ A1,
f (a) = {a, h}h−1.
Then, for each k ∈ Z, every nonzero element of Wk is an eigenvector of f with eigenvalue k.
Note that C[h±1] is a unique factorization domain since it is a principal ideal domain.
Theorem 3.6. The Poisson algebra A1 is Poisson simple if and only if every root of a(h) is a
simple root.
Proof. Let b(h) ∈ C[h±1] be the greatest common divisor of a(h) and a′(h). Note that a(h) has
a root with multiplicity > 1 if and only if b(h) is a nonunit.
(⇒) Suppose that a(h) has a root with multiplicity > 1. Thus b(h) is a nonunit. Let I be the
ideal of A1 generated by x, y and b(h). Then A1/I is isomorphic to the algebra C[h±1, x, y]/J,
where J is the ideal of C[h±1, x, y] generated by x, y and b(h), since a(h) is divided by b(h).
Moreover C[h±1, x, y]/J is isomorphic to C[h±1]/b(h)C[h±1]. Thus I is a nontrivial ideal of A1.
Observe that
{x, b(h)} = b′(h){x, h} = b′(h)a′(h)hx ∈ I,
{y, b(h)} = b′(h){y, h} = −b′(h)a′(h)hy ∈ I
{y, x} = −a′(h)h ∈ I
by the chain rule and (3.7). Hence I is a nontrivial Poisson ideal of A1 and thus A1 is not
Poisson simple.
(⇐) Suppose that A1 is not Poisson simple. Then there exists a nontrivial Poisson ideal I of
A1. Let P be a minimal prime ideal over I. By [4, 6.2], P is a prime Poisson ideal. Applying f
to P , P contains a nonzero element wk ∈ Wk for some k ∈ Z by Lemma 3.5. If k ≥ 0 then x ∈ P
or c(h) ∈ P for some 0 6= c(h) ∈ C[h±1] since P is prime. If x ∈ P then a′(h)h = {x, y} ∈ P
and thus a(h) ∈ P and a′(h) ∈ P . It follows that the greatest common divisor b(h) of a(h) and
a′(h) is an element of P and thus P contains a unit, a contradiction. Similarly, if k < 0 then
repeating the argument gives that y /∈ P and that P contains an element 0 6= c(h) ∈ C[h±1].
Let 0 6= c(h) ∈ P and x, y /∈ P . We may assume c(h) ∈ C[h] by multiplying hn to c(h) for
sufficiently large n. If the degree of c(h) is greater than 1 then 0 6= c′(h) ∈ P since
P ∋ {x, c(h)} = c′(h)hx.
Repeating this process, we get that P contains a unit, a contradiction. Hence A1 is Poisson
simple.
(cid:3)
Let us find the Poisson endomorphisms of A1. The following arguments are Poisson analogues
of those in [7]. For completion, we repeat them. Note that the unit group of A1 is {γhiγ ∈
C∗, i ∈ Z}. Let ψ be a Poisson endomorphism of A1. Then ψ(h) = γhi for some γ ∈ C∗ and
i ∈ Z since Poisson endomorphisms preserve units. Hence there are possible three types of
Poisson endomorphisms of A1 as in [7]. Positive-type Poisson endomorphisms, that is, Poisson
endomorphisms ψ such that ψ(h) = γhi (i > 0); Zero-type Poisson endomorphisms, that is,
Poisson endomorphisms ψ such that ψ(h) = γ; Negative-type Poisson endomorphisms, that is,
Poisson endomorphisms ψ such that ψ(h) = γhi (i < 0).
In the following theorem, we see that the positive-type Poisson endomorphisms and the
negative-type Poisson endomorphisms of A1 are the same forms as those of endomorphisms
of A(a(h), q) but the zero-type Poisson endomorphisms of A1 are slightly different from those
8
EUN-HEE CHO AND SEI-QWON OH
of A(a(h), q). (One should compare the following theorem with [7, Proposition 3.1, Proposition
4.1 and Proposition 5.3].)
Theorem 3.7. Let d be the maximal degree of a(h). Write a(h) by
a(h) = ai1hi1 + ai2hi2 + . . . + aimhim,
where i1 < i2 < . . . < im = d and aij ∈ C∗ for all j = 1, 2, . . . , m. Denote by k the greatest
common divisor of d − i1, d − i2, . . . , d − im−1.
(1) Let ψ be a positive-type Poisson endomorphism of A1. Then
(3.9)
where γ is a k-th root of unity, b ∈ C∗ and n ∈ Z. Conversely, a map ψ satisfying (3.9)
determines a unique Poisson automorphism of A1.
ψ(h) = γh, ψ(x) = bhnx, ψ(y) = γdb−1h−ny,
(2) Let ψ be a zero-type Poisson endomorphism of A1. Then
(3.10)
where γ ∈ C∗ is a root of a(h) with multiplicity > 1. If a(h) has a root with multiplicity > 1
then a map ψ satisfying (3.10) determines a unique Poisson endomorphism. If every root of
a(h) has multiplicity 1 then there are no zero-type Poisson endomorphisms.
ψ(h) = γ, ψ(x) = 0, ψ(y) = 0,
(3) Let ψ be a negative-type Poisson endomorphism of A1. Then
(3.11)
where γ, b, c ∈ C∗ and u, v ∈ Z satisfy the relation
ψ(h) = γh−1, ψ(x) = chvy, ψ(y) = bhux,
(3.12)
bchu+va(h) = a(γh−1).
Conversely, a map ψ satisfying (3.11) determines a unique Poisson automorphism of A1.
Proof. Note that every Poisson endomorphism ψ of A1 preserves the following equations
(3.13)
{x, h} = hx, {y, h} = −hy, {y, x} = −a′(h)h, xy = a(h).
(1) Let ψ be a Poisson endomorphism such that ψ(h) = γhj for γ ∈ C∗ and j > 0. Suppose
that ψ(x) = 0. Applying ψ to the third equation of (3.13), the left hand side is zero and the
right hand side is −γa′(γhj)hj which is nonzero, a contradiction. Hence ψ(x) 6= 0. Similarly
ψ(y) 6= 0. We can set ψ(x) = Pk∈Z wk by (3.8), where wk ∈ Wk for each k. Applying ψ to
The left hand side of the above equation is Pk jkγhj wk and thus jk = 1 for all k such that
{ψ(x), γhj } = γhjψ(x).
{x, h} = hx, we have
wk 6= 0. Thus we have that
for some 0 6= b(h) ∈ C[h±1]. Repeating this argument on {y, h} = −hy, we get
ψ(h) = γh, ψ(x) = b(h)x
ψ(y) = c(h)y
for some 0 6= c(h) ∈ C[h±1].
Applying ψ to the last equation of (3.13), we get b(h)c(h)a(h) = a(γh). Comparing the
maximal and the minimal degrees for h on both sides and then coefficients, we get
where γ is a k-th root of unity, b ∈ C∗ and n ∈ Z.
ψ(x) = bhnx, ψ(y) = γdb−1h−ny,
A POISSON GENERALIZED WEYL ALGEBRA
9
Conversely, let ψ be a map satisfying (3.9). Then ψ determines a unique algebra endomor-
phism since B1 is the commutative polynomial ring C[h±1, x, y] and it preserves the last equation
of (3.13). It is checked routinely that ψ preserves the other equations of (3.13). Thus ψ is a
Poisson endomorphism. Such ψ is a Poisson automorphism since there exists ψ−1 defined by
ψ−1(h) = γ−1h, ψ−1(x) = γnb−1h−nx, ψ−1(y) = γ−n−dbhny.
(2) Let ψ be a Poisson endomorphism such that ψ(h) = γ, where γ ∈ C∗. Then ψ(x) =
ψ(y) = 0 by the first and the second equations of (3.13). Applying ψ to the third and the last
equations of (3.13), we get that γ is a common root of a(h) and a′(h). Thus γ is a root of a(h)
with multiplicity > 1.
Conversely, let ψ be a map satisfying (3.10). Then ψ determines a unique algebra endomor-
phism since ψ preserves the last equation of (3.13). Moreover ψ preserves the other equations
of (3.13) and thus ψ is a Poisson endomorphism. Now the other statements are trivial by
Theorem 3.6.
(3) Let ψ be a Poisson endomorphism such that ψ(h) = γhj for γ ∈ C∗ and j < 0. Since ψ2
is of positive type, ψ(x) 6= 0 and ψ(y) 6= 0 by (1). We can set ψ(x) = Pk∈Z wk by (3.8), where
wk ∈ Wk for each k. Applying ψ to {x, h} = hx, we have
The left hand side of the above equation is Pk jkγhj wk and thus jk = 1 for all k such that
wk 6= 0. Thus we have that
{ψ(x), γhj } = γhjψ(x).
for some 0 6= c(h) ∈ C[h±1]. Repeating this argument on {y, h} = −hy, we get
ψ(h) = γh−1, ψ(x) = c(h)y
ψ(y) = b(h)x
for some 0 6= b(h) ∈ C[h±1]. Moreover we have that b(h) = bhu, c(h) = chv for some b, c ∈ C∗
and u, v ∈ Z by (1) since ψ2 is of positive type. Applying ψ to the last equation of (3.13), we
get the relation bchu+va(h) = a(γh−1).
Conversely, let ψ be a map satisfying (3.11). Then ψ preserves the last equation of (3.13) and
thus ψ determines a unique algebra endomorphism. It is easy to check that ψ preserves the first
and the second equations of (3.13). Moreover it is shown that ψ preserves the third equation
of (3.13) by differentiating the equation (3.12) by h. Hence ψ satisfying (3.11) determines a
unique Poisson endomorphism. Since there exists a map ψ−1 of the negative type, ψ determines
a unique Poisson automorphism.
(cid:3)
Acknowledgments The authors would like to appreciate S. Launois for informing a reference
[8, Proposition 4.1] about semiclassical limits of Ore extensions, that is used for constructing a
Poisson generalized Weyl algebra, after the first version of the paper is posted.
The second author is supported by Chungnam Nationality University Grant and thanks the
Korea Institute for Advanced Study for the warm hospitality during a part of the preparation
of this paper.
1. V.V. Bavula, Generalized Weyl algebras and their representations, Translation in St. Petersburg Math. J.
4(1) (1993), 71–92.
References
10
EUN-HEE CHO AND SEI-QWON OH
2. K. A. Brown and K. R. Goodearl, Lectures on algebraic quantum groups, Advanced courses in mathematics-
CRM Barcelona, Birkhauser Verlag, Basel·Boston·Berlin, 2002.
3. K. R. Goodearl, A Dixmier-Moeglin equivalence for Poisson algebras with torus actions, in Algebra and Its
Applications (D. V. Huynh, S. K. Jain, and S. R. Lopez-Permouth, Eds.) Contemp. Math. 419 (2006),
131–154.
4.
, Semiclassical limits of quantized coordinate rings, in Advances in Ring Theory (D. V. Huynh and S.
R. Lopez-Permouth, Eds.) Basel Birkhauser (2009), 165–204.
5. K. R. Goodearl and R. B. Warfield, An introduction to noncommutative noetherian rings, Second ed., London
Mathematical Society Student Text 61, Cambridge University Press, 2004.
6. David A. Jordan and Sei-Qwon Oh, Poisson brackets and Poisson spectra in polynomial algebras, New Trends
in Noncommutative Algebra, Contemp. Math. 562 (2012), 169–187.
7. Andrew P. Kitchin and St´ephane Launois, Endomorphisms of quantum generalized Weyl algebras, Letters in
Math. Phys. 104 (2014), 837–848.
8. St´ephane Launois and C. Lecoutre, A quadratic Poisson Gel'fand-Kirillov problem in prime characteristic,
arXiv:1302.2046v2.
9. Sei-Qwon Oh, A natural map from a quantized space onto its semiclassical limit and a multi-parameter Poisson
Weyl algebra, arXiv:1511.00802v1(to appear in Comm. Algebra)
10.
11.
, Poisson polynomial rings, Comm. Algebra 34 (2006), 1265–1277.
, Quantum and Poisson structures of multi-parameter symplectic and Euclidean spaces, J. Algebra 319
(2008), 4485–4535.
12. Sei-Qwon Oh and Mi-Yeon Park, Relationship between quantum and Poisson structures of odd dimensional
Euclidean spaces, Comm. Algebra 38(9) (2010), 3333–3346.
Department of Mathematics, Chungnam National University, 99 Daehak-ro, Yuseong-gu, Dae-
jeon 34134, Korea
E-mail address: [email protected]
E-mail address: [email protected]
|
1611.00932 | 1 | 1611 | 2016-11-03T09:43:53 | Natural Partial Order on Rings with Involution | [
"math.RA",
"math.CO"
] | In this paper, we introduce a partial order on rings with involution, which is a generalization of the partial order on the set of projections in a Rickart *-ring. We prove that a *-ring with the natural partial order form a sectionally semi-complemented poset. It is proved that every interval [0,x] forms an orthomodular lattice in case of abelian Rickart *-rings. The concepts of generalized comparability (GC) and partial comparability (PC) are extended to involve all the elements of a *-ring. Further, it is proved that these concepts are equivalent in finite abelian Rickart *-rings. | math.RA | math |
NATURAL PARTIAL ORDER ON RINGS WITH INVOLUTION
AVINASH PATIL1 AND B. N. WAPHARE2
Abstract. In this paper, we introduce a partial order on rings with involution,
which is a generalization of the partial order on the set of projections in a Rickart ∗-
ring. We prove that, a ∗-ring with the natural partial order form a sectionally semi-
complemented poset. It is proved that every interval [0, x] forms an orthomodular
lattice in case of abelian Rickart ∗-rings. The concepts of generalized comparability
(GC) and partial comparability (P C) are extended to involve all the elements of
a ∗-ring. Further, it is proved that these concepts are equivalent in finite abelian
Rickart ∗-rings.
Keywords : ∗-ring, partial order, generalized comparability, partial comaparbility.
MSC(2010): Primary: 16W10, Secondary: 06F25
1. introduction
An involution ∗ on an associative ring R is a mapping such that (a + b)∗ = a∗ + b∗,
(ab)∗ = b∗a∗ and (a∗)∗ = a, for all a, b ∈ R. A ring with involution ∗ is called a ∗-ring.
Clearly, identity mapping is an involution if and only if the ring is commutative. An
element e of a ∗-ring R is a projection if e = e2 and e = e∗. For a nonempty subset
B of R, we write r(B) = {x ∈ R : bx = 0, ∀b ∈ B}, and call a right annihilator of B
in R. A Rickart ∗-ring R is a ∗-ring in which right annihilator of every element is
generated, as a right ideal, by a projection in R. Every Rickart ∗-ring contains unity.
For each element a in a Rickart ∗-ring, there is unique projection e such that ae = a
and ax = 0 if and only if ex = 0, called a right projection of a denoted by RP (a).
In fact, r({a}) = (1 − RP (a))R. Similarly, the left annihilator l({a}) and the left
projection LP (a) are defined for each element a in a Rickart ∗-ring R. The set of
projections P (R) in a Rickart ∗-ring R forms a lattice, denoted by L(P (R)), under
the partial order 'e ≤ f if and only if e = f e = ef '. In fact, e ∨ f = f + RP (e(1 − f ))
and e ∧ f = e − LP (e(1 − f )). This lattice is extensively studied by I. Kaplanski [4],
S. K. Berberian [1], S. Maeda in [5, 6] and others.
Drazin [2], was the first to introduce "∗-order" to involve all elements, where ∗-
b if and only if a∗a = a∗b and aa∗ = ba∗, which is a partial order
order is given by, a 6
∗
on a semigroup with proper involution(i.e., aa∗ = 0 implies a = 0). In particular,
with the obvious choices for ∗-rings with proper involution, all commutative rings
with no nonzero nilpotent elements, all Boolean rings, the ring B(H) of all bounded
linear operators on any complex Hilbert space H, the Rickart ∗-ring. Janowitz [3]
studies ∗-order on Rickart ∗-ring. Thakare and Nimbhorkar [9] used ∗-order on
Rickart ∗-ring and generalized the comparability axioms to involve all elements of
∗-ring. Mitsch [8] defined a partial order on a semigroup. We modify that order to
have partial order on a ∗-ring.
1
NATURAL PARTIAL ORDER ON RINGS WITH INVOLUTION
2
In this paper, we introduce a partial order on a ∗-ring which is a generalization
of the partial order on the set of projections in a Rickart ∗-ring. For a ∗-ring R,
it is proved that the poset (R, ≤) is an sectionally semi-complemented (SSC) poset.
For an abelian Rickart ∗-ring, we prove that every interval [0, x] is an orthomodular
poset, in fact, an orthomodular lattice. In the last section, comparability axioms are
introduced to involve all elements of the ∗-ring.
2. natural partial order and its properties
We introduce an order on a ∗-ring with unity.
Definition 2.1. Let R be a ∗-ring with unity. Define a relation '≤' on R by 'a ≤ b
if and only if a = xa = xb = ax∗ = bx∗, for some x ∈ R'.
Proposition 2.2. Let R be a ∗-ring with unity. Then the relation '≤' given in
Definition 2.1 is a partial order on R.
Proof. Reflexive: for x = 1, we have a = xa = ax∗. Hence a ≤ a, ∀a ∈ R.
Antisymmetric: Let a ≤ b and b ≤ a. Then there exist x, y ∈ R such that a = xa =
xb = ax∗ = bx∗ and b = yb = ya = by∗ = ay∗, hence b = ya = y(bx∗) = bx∗ = a.
Transitive: Let a ≤ b and b ≤ c. Hence there exist x, y ∈ R such that a = xa =
xb = ax∗ = bx∗ and b = yb = yc = by∗ = cy∗. Then (xy)a = (xy)(bx∗) = x(yb)x∗ =
xbx∗ = ax∗ = a, (xy)c = x(yc) = xb = a, a(xy)∗ = (xb)(y∗x∗) = x(by∗)x∗ = xbx∗ =
ax∗ = a and c(xy)∗ = c(y∗x∗) = (cy∗)x∗ = bx∗ = a. Hence a ≤ c.
(cid:3)
Henceforth R denotes a ∗-ring with unity and we say that a ≤ b through
x whenever a = xa = xb = ax∗ = bx∗.
Note 2.3. If we restrict this partial order to the set of projections in a Rickart ∗-ring,
then it coincides with the usual partial order for projections given in Berberian [1].
Remark 2.4. This order is different from ∗-order.
2 2 (cid:21) = ba∗, hence a 6
∗
b.
For, let a = (cid:20) 1 2
Then a∗a = (cid:20) 2 1
Next let x = (cid:20) x1 x2
(cid:20) 1 2
1 2 (cid:21) = (cid:20) x1 x2
(cid:20) 1 2
1 2 (cid:21) = (cid:20) 1 2
0 1 (cid:21) ∈ R = M2(Z3) with transpose as an involution.
1 2 (cid:21) , b = (cid:20) 2 0
1 2 (cid:21) = a∗b, aa∗ = (cid:20) 2 2
x3 x4 (cid:21) be such that a = xa = xb = ax∗ = bx∗. Then a = xa gives
x3 x4 (cid:21)(cid:20) 1 2
1 2 (cid:21)(cid:20) x1 x3
1 2 (cid:21) = (cid:20) x1 + x2 2(x1 + x2)
x3 + x4 2(x3 + x4) (cid:21) and a = ax∗ gives
x2 x4 (cid:21) = (cid:20) x1 + 2x2 x3 + 2x4
x1 + 2x2 x3 + 2x4 (cid:21).
On comparing, we get x1 +x2 = 1 = x1 +2x2, which gives x2 = 0, x1 = 1. Similarly
x3 + x4 = 1, x3 + 2x4 = 2, giving x3 = 1, x4 = 0, i.e., x = (cid:20) 1 0
1 0 (cid:21). But xb 6= a.
NATURAL PARTIAL ORDER ON RINGS WITH INVOLUTION
3
Hence a (cid:2) b. On the Other hand, if c = (cid:20) 1 1
1 1 (cid:21) , d = (cid:20) 0 1
1 1 (cid:21) and y = (cid:20) 0 1
0 1 (cid:21).
Then c ≤ d through y. While c∗c 6= c∗d, hence c (cid:10)
d. Thus these two partial orders
(natural partial order and ∗-order) are distinct.
∗
Proposition 2.5. Let R be a commutative ∗-ring. Then a ≤ b implies a 6
b.
∗
Proof. Let a ≤ b. Then there exists x ∈ R such that a = xa = xb = ax∗ = bx∗. This
yields a∗b = (xa)∗b = a∗x∗b = a∗(bx∗) = a∗a and since R is commutative, we get
aa∗ = ba∗. Hence a 6
(cid:3)
b.
∗
Note that the converse of the above statement is not true in general. Since ∗-order
is not a partial order on Z12 with identity mapping as an involution(as 6∗6 = 6∗0 =
0∗0).
In the next result, we provide properties of the natural partial order.
Theorem 2.6. Let R be a ∗-ring with unity. Then the following statements hold.
(1) 0 is the least element of the poset R.
(2) a ≤ e, a ∈ R, e ∈ P (R)(set of projections in R) implies a ∈ P (R).
(3) a ≤ b if and only if a∗ ≤ b∗.
(4) If a ≤ b, then r(b) ⊆ r(a) and l(b) ⊆ l(a).
(5) a ≤ b, b regular (i.e., bb′b = b, for some b′ ∈ R) implies a is regular.
(6) a ≤ b and a has right(resp.
left) inverse imply a = b, i.e., every element
having right(resp. left) inverse is maximal.
(7) If a ≤ b. Then ac ≤ bc if and only if ca ≤ cb, ∀c ∈ R.
Proof. (1) Obvious.
(2) Suppose a ≤ e, e ∈ P (R). Let a ≤ e through x, for some x ∈ R, i.e., a = xa =
xe = ax∗ = ex∗. This yields a2 = xe.ex∗ = xex∗ = ax∗ = a. Also, a∗ = (xe)∗ =
ex∗ = a, hence a ∈ P (R).
(3) Let a ≤ b. Then a = xa = xb = ax∗ = bx∗, for some x ∈ R. Hence a∗ = xa∗ =
a∗x∗ = b∗x∗ = xb∗ which gives a∗ ≤ b∗. The Converse follows from the fact that
(a∗)∗ = a.
(4) Obvious.
(5) Suppose a ≤ b and b is regular, i.e., bb′b = b, for some b′ ∈ R. Let a = xa = xb =
ax∗ = bx∗, for some x ∈ R. Then a = ax∗ = xbx∗ = xbb′bx∗ = (xb)b′(bx∗) = ab′a.
Hence a is regular.
(6) Let c ∈ R be such that ac = 1(resp. ca = 1) and a ≤ b. Let a = xa = xb = ax∗ =
bx∗, for some x ∈ R. Then a = xa(resp. a = ax∗) gives ac = xac(resp. ca = cax∗).
Thus 1 = x(resp. 1 = x∗). Hence a = b, i.e, a is a maximal element.
(7) Suppose a ≤ b and ac ≤ bc, ∀c ∈ R. Since a ≤ b, by (3) above, we have
a∗ ≤ b∗ giving a∗c∗ ≤ b∗c∗, which further yields (a∗c∗)∗ ≤ (b∗c∗)∗, i.e., ca ≤ cb and
conversely.
(cid:3)
In a poset P , the principal ideal generated by a ∈ P is given by (a] = {x ∈ P : x ≤
a}.
NATURAL PARTIAL ORDER ON RINGS WITH INVOLUTION
4
Proposition 2.7. If a and b are central elements of a ∗-ring R which generate the
same ideals of a ring R, then there is a order isomorphism between the set of elements
≤ a and the set of elements ≤ b.
1t = x1xt = xt, x1b = x1at = ax1t = xt and bx∗
Proof. Let a and b are central elements with Ra = Rb. Then a = bs, b = at, for
some s, t ∈ R. Denote (a] = {x ∈ R : x ≤ a}. Define φ : (a] → (b] by φ(x) = xt.
1 = xx∗
We claim that xt ≤ b, ∀x ∈ (a]. As x ≤ a, we have x = x1x = x1a = ax∗
1,
for some x1 ∈ R. Then x1xt = xt, xtx∗
1at =
x1ax∗
1at = ax∗
1t = xt.
Hence xt ≤ b. Now, let x, y ∈ (a] be such that x = x1x = x1a = ax∗
1 = xx∗
and y = y1y = y1a = ay∗
1, for some x1, y1 ∈ R. Then φ(x) = φ(y) if and
only if xt = yt if and only if x1at = y1at if and only if x1b = y1b if and only if
x1a = y1a if and only if x = y. Hence φ is well defined and one-one. Let z ∈ (b].
Then as above zs ∈ (a] and z = z1b = z1z = bz∗
1 = zz∗
1, for some z1 ∈ R. Also
φ(zs) = zst = z1bst = z1at = z1b = z, i.e., φ is a bijection.
1 = x1bx∗
1 = x∗
1 = x1x∗
1b = x∗
1 = x1atx∗
1b = x1x∗
1 = yy∗
1
1 = yy∗
2 = xx∗
1 and x = x2x = x2y = yx∗
Now, suppose that x, y ∈ (a] with x ≤ y. Then x = x1x = x1a = ax∗
y = y1y = y1a = ay∗
R. Next, (x1x2)xt = x1xt = xt, (x1x2)yt = x1xt = xt, xt(x1x2)∗ = xtx∗
x1atx∗
1 = x1bx∗
yt(x1x2)∗ = ytx∗
yx∗
In fact, ψ : (b] → (a] defined by ψ(y) = ys, works as an inverse of φ.
1 = xx∗
1,
2, for some x1, x2, y1 ∈
1 =
1t = xt and
1t =
1t = xt. Consequently φ(x) ≤ φ(y). Hence φ is an order isomorphism.
(cid:3)
1at = ax1x∗
2x∗
1 = y1x∗
1 = x1x∗
2x∗
1 = y1atx∗
1b = x1x∗
2x∗
2x∗
1 = y1bx∗
1t = xx∗
2x∗
1b = y1x∗
1t = xx∗
2x∗
1t = xx∗
1at = y1ax∗
2x∗
2x∗
2x∗
2x∗
2x∗
2x∗
2x∗
2x∗
Theorem 2.8 (Condition of Compatability). If xa = ax∗, ∀a, x ∈ R, then the
natural partial order is compatible with multiplication, i.e., a ≤ b implies ca ≤ cb,
for all c ∈ R.
Proof. In view of Theorem 2.6 (7), it is enough to show that a ≤ b implies ac ≤
bc, ∀c ∈ R. Let a ≤ b, then there exists x ∈ R such that a = xa = xb = ax∗ = bx∗.
Hence ac = xac = xbc = ax∗c = bx∗c, i.e., ac = xac, acx∗ = ca∗x∗ = c(xa)∗ = ca∗ =
ac. Also bcx∗ = cb∗x∗ = c(xb)∗ca∗ = ac, hence ac ≤ bc.
(cid:3)
Definition 2.9. Two elements a and b in a ∗-ring R are orthogonal, denoted by
a ⊥ b, if there exists x ∈ R such that xa = a = ax∗ and xb = 0 = bx∗.
The orthogonality relation in a ∗-ring has the following properties.
Theorem 2.10. For elements a, b, c in a ∗-ring R , the following statements hold.
(1) a ⊥ a implies a = 0.
(2) a ⊥ b if and only if b ⊥ a if and only if a ⊥ (−b).
(3) a ⊥ b, c ≤ a imply c ⊥ b.
(4) a ⊥ b if and only if a ≤ a − b.
(5) a ≤ b implies b − a ≤ b and b − a ⊥ a.
(6) If a ⊥ b, then a ∧ b = 0 and a + b is an upper bound of both a, b.
(7) a ⊥ b, (a + b) ⊥ c imply a ⊥ (b + c).
NATURAL PARTIAL ORDER ON RINGS WITH INVOLUTION
5
Proof. (1), (2) Obvious.
(3) Suppose that a ⊥ b and c ≤ a. Let x, y ∈ R be such that a = xa = ax∗,
xb = 0 = bx∗ and c = yc = cy∗ = ya = ay∗. Then (yx)c = yxay∗ = yay∗ = cy∗ = c.
Similarly, c(yx)∗ = c. On the other hand, (yx)b = 0 and b(yx)∗ = 0. Consequently,
c ⊥ b.
(4) Suppose a and b are orthogonal. Let x ∈ R be such that a = xa = ax∗ and
xb = 0 = bx∗. Then a = x(a−b) = (a−b)x∗ = xa = ax∗, hence a ≤ a−b. Conversely,
suppose that a ≤ a − b. Let x ∈ R be such that a = x(a − b) = (a − b)x∗ = xa = ax∗.
Then a = x(a − b) and a = xa gives xb = 0. Similarly, bx∗ = 0. Hence a ⊥ b.
(5) Let x ∈ R be such that a = xa = xb = ax∗ = bx∗. Then (1 − x)(b − a) =
b−a−xb+xa = b−a−a+a = b−a, (1−x)b = b−xb = b−a, b(1−x)∗ = b−bx∗ = b−a
and (b − a)(1 − x)∗ = b − a − bx∗ − ax∗ = b − a − a + a = b − a. Hence b − a ≤ b.
Also (1 − x)(b − a) = b − a = (b − a)(1 − x)∗ and (1 − x)a = 0 = a(1 − x)∗. Hence
b − a ⊥ a.
(6) Suppose a ⊥ b and x be such that xa = a = ax∗, xb = 0 = bx∗. Let c ≤ a and
c ≤ b i.e. c = x1c = x1a = cx∗
2, for some
x1, x2 ∈ R. Then x1a = c = x2b gives c = x1a = x2b = x1ax∗ = x2bx∗ = 0. Hence
a ∧ b = 0. From (2) and (4), we have a ≤ a + b and b = (a + b) − a ≤ a + b.
(7) Suppose that a ⊥ b, (a+ b) ⊥ c. From (6), we have a ≤ a+ b and a+ b ≤ a+ b+ c.
This gives a ≤ a + b + c. Then from (5), we get b + c = (a + b + c) − a ≤ a + b + c
and (b + c) ⊥ a, as required.
(cid:3)
1 and c = x2c = x2b = cx∗
1 = ax∗
2 = bx∗
A poset P with 0 is called sectionally semi-complemented (in brief SSC) if, for
a, b ∈ P , a < b, there exists an element c ∈ P such that 0 < c < b and {a, c}l = {0},
where {a, c}l = {x ∈ P : x ≤ a and x ≤ c}. Thus from (5) and (6) of Theorem 2.10
, we have the following result.
Theorem 2.11. Let R be a ∗-ring. Then the poset (R, ≤) is an SSC poset.
A ring is called an abelian ring if all of its idempotents are central.
Proposition 2.12. In an abelian Rickart ∗-ring R, the following statements are
equivalent.
i) a ≤ b.
ii) There exists a projection e such that a = ae = be.
iii) ab = a2 = ba.
Proof. i) =⇒ ii) Suppose a ≤ b, then there exists x ∈ R such that a = xa = xb =
ax∗ = bx∗. Since a = xa, we have (1 − x)a = 0. This gives a ∈ r{1 − x} = eR, for
some projection e ∈ R. Then ea = a = ae and (1 − x)e = 0, i.e., e = xe = ex∗.
Also, a = xb implies ea = exb = xeb = eb. Thus a = ae = be.
ii) =⇒ iii) Obvious.
iii) =⇒ i) Let ab = a2, i.e., a(b − a) = 0. Then there exists a projection e such that
a = ea = ea and e(b − a) = 0, i.e., eb = ea = a, hence a ≤ b.
(cid:3)
NATURAL PARTIAL ORDER ON RINGS WITH INVOLUTION
6
Lemma 2.13. If R is an abelian Rickart ∗-ring, then a ⊥ b implies a ∧ b = 0 and
a ∨ b = a + b.
Proof. Suppose a ⊥ b. By Theorem 2.10 (6), a ∧ b = 0 and a + b is an upper bound of
a and b. Let a ≤ c and b ≤ c, then there exist projections e, f such that a = ea = ec
and b = f b = f c. Since a ⊥ b, there exists x ∈ R such that xa = a = ax∗ and xb =
0 = bx∗. Let y = ex+f (1−x). Then y(a+b) = exa+exb+f (1−x)a+f (1−x)b = a+b,
(a + b)y∗ = a + b, yc = exc + f (1 − x)c = a + b and cy∗ = a + b, i.e., a + b ≤ c. Thus
a ∨ b = a + b.
(cid:3)
Before proceeding further, we need the definition of orthomodular poset. An
orthomodular poset is a partially ordered set P with 0 and 1 equipped with a mapping
x → x⊥ (called the orthocomplementation) satisfying the conditions.
i) a ≤ b ⇒ b⊥ ≤ a⊥,
ii) (a⊥)⊥ = a for all a ∈ P ,
iii) a ∨ a⊥ = 1 and a ∧ a⊥ = 0, for all a ∈ P ,
iv) a ≤ b⊥ implies that a ∨ b exists in P ,
v) a ≤ b ⇒ b = a ∨ (a ∨ b⊥)⊥.
The following result is essentially due to Marovt et al. [7, Theorem 1].
Theorem 2.14. Let R be a Rickart ∗-ring. Then a 6
b if and only if there exist
∗
projections p and q such that a = pb = bq.
Thus, from Proposition 2.12 and Theorem 2.14, the natural partial order and
∗-order are equivalent on abelian Rickart ∗-rings. This leads to the following two
results which are also proved independently by Janowitz [3].
Theorem 2.15. Let R is an abelian Rickart ∗-ring. Then every interval [0, x] is an
orthomodular poset.
We know that, if R is a Rickart ∗-ring, then the set of projection P (R) forms a
lattice and the set {e ∈ P (R) : e ≤ x′′} is sub lattice of P (R), where x′ is a projection
which generates the right annihilator of x.
Theorem 2.16. In an abelian Rickart ∗-ring R every interval [0, x] is ortho-isomorphic
to {e ∈ P (R) : e ≤ x′′}. Hence every interval [0, x] is an orthomodular lattice.
3. comparability axioms
Two projections e and f are said to be equivalent, written e ∼ f , if there is
w ∈ eRf such that e = ww∗ and f = w∗w, which is an equivalence relation on
the set of projections in a Rickart ∗-ring. A projection e is said to be dominated
by the projection f , denoted by e . f , if e ∼ g ≤ f , for some projection g in
R. Two projections e and f are said to be generalized comparable if there exists a
central projection h such that he . hf and (1 − h)f . (1 − h)e. A ∗-ring is said
to satisfy the generalized comparability (GC) if any two projections are generalized
comparable. Two projections e and f are said to be partially comparable if there
NATURAL PARTIAL ORDER ON RINGS WITH INVOLUTION
7
exist non zero projections e0, f0 in R such that e0 ≤ e, f0 ≤ f and e0 ∼ f0. If for
any pair of projections in R, eRf 6= 0 implies e and f are partially comparable, then
R is said to satisfy partial comparability (P C). More about comparability axioms
on the set of projections in a Rickart ∗-ring can be found in Berberian [1].
Drazin [2] extended the relation of equivalence of two projections to general ele-
ments of a ∗-ring as follows.
Definition 3.1 ([2, Definition 2*]). Let R be a ∗-ring with unity. We say that a ∼ b
if and only if there exists x ∈ aRb, y ∈ bRa such that aa∗ = xx∗, bb∗ = yy∗, a∗a =
y∗y, b∗b = x∗x.
This relation is symmetric on a ∗-ring. Thakare and Nimbhorkar [9] extended the
comparability axioms using the above relation and ∗-order to involve all elements of
Rickart ∗-ring.
We provide a relation which is symmetric and transitive on general elements of
∗-ring as an extension of the relation of equivalence of two projections.
Definition 3.2. Let R be a ∗-ring with unity. We say that a ∼ b if and only if
there exists x, y ∈ R such that aa∗ = xx∗, bb∗ = yy∗, a∗a = y∗y, b∗b = x∗x with
x = ax = xb and y = by = ya.
Now, we extend the concepts of dominance, GC, P C etc. from the set of projec-
tions in a Rickart ∗-ring to general elements in a ∗-ring.
Definition 3.3.
(1) Let R be a ∗-ring with unity. We say that a is dominated
by b if a ∼ c ≤ b for some c ∈ R. In notation a . b.
(2) A ∗-ring R is said to satisfy the generalized comparability for elements (GC)
for elements, if for any a, b ∈ R there exists a central projection h such that
ha . hb and (1 − h)b . (1 − h)a.
(3) Two elements a, b in a ∗-ring R are said to be partially comparable if there
exists two non-zero elements c, d in R such that c ≤ a, d ≤ b with c ∼ d. If
for any a, b ∈ R, aRb 6= 0 implies a and b are partially comparable then we
say that R has partial comparability for elements (P C).
Clearly, if a ≤ b or a ∼ b, then a . b.
Lemma 3.4. If a . b and h is a central projection, then ha . hb.
Definition 3.5. Two elements a and b in a ∗-ring R are said to be very orthogonal
if there exists a central projection h such that ha = a and hb = 0.
The relevance of very orthogonality to generalized comparability is as follows:
Theorem 3.6. If a and b are elements of a ∗-ring R. Then the following statements
are equivalent.
i) a and b are generalized comparable.
ii) There exists orthogonal decompositions a = x + y, b = z + w with x ∼ z, y
and w are very orthogonal.
NATURAL PARTIAL ORDER ON RINGS WITH INVOLUTION
8
2 = ha(1−h)k∗
Proof. i) ⇒ ii) Suppose a and b are generalized comparable. Let h be a central
projection such that ha . hb and (1 −h)b . (1 −h)a. Then ha ∼ k1 ≤ hb, (1 −h)b ∼
k2 ≤ (1 − h)a, for some k1, k2 ∈ R. Hence k1 = m1k1 = m1hb = k1m∗
1 = hbm∗
1,
for some m1 ∈ R. Then k1 = m1hb gives k1h = m1hbh = m1hb = k1. Similarly,
k2 = (1−h)k2. Also hak∗
1 = [(1−h)b]∗k1 = 0.
We claim that ha + k2 ∼ k1 + (1 − h)b. Since ha ∼ k1, there exist x1, y1 ∈ R
such that (ha)(ha)∗ = x1x∗
1x1 with
x1 = hax1 = x1k1 and y1 = k1y1 = y1ha. Clearly, x1 = hx1 and y1 = hy1,
since k1h = k1. Similarly, Since (1 − h)b ∼ k2, there exist x2, y2 ∈ R such that
k2k∗
2x2
with x2 = k2x2 = x2(1 − h)b and y2 = (1 − h)by2 = y2k2. Clearly, x2 = (1 − h)x2
and y2 = (1 − h)y2, since k2(1 − h) = k2.
2y2 and [(1 − h)b]∗[(1 − h)b] = x∗
2, [(1 − h)b][(1 − h)b]∗ = y2y∗
2 = 0 = (ha)∗k2, (1−h)bk∗
1, (ha)∗(ha) = y∗
1y1 and k∗
1k1 = x∗
1 = y1y∗
2 = x2x∗
2k2 = y∗
1, k1k∗
2, k∗
Let x = x1 + x2 and y = y1 + y2. Since hk2 = 0 and (1 − h)k1 = 0, we have
(ha + k2)x = (ha + k2)(x1 + x2) = hax1 + hax2 + k2x1 + k2x2 = x1 + 0 + 0 + x2 = x,
x[k1 + (1 − h)b] = (x1 + x2)[k1 + (1 − h)b] = x1k1 + x1(1 − h)b + x2k1 + x2(1 − h)b =
x1 + 0 + 0 + x2 = x. Similarly, we have y = [k1 + (1 − h)b]y = y(ha + k2).
2 = x1x∗
Also, xx∗ = (x1 + x2)(x1 + x2)∗ = x1x∗
2 = (ha)(ha)∗ + k2k∗
2x1 + x∗
1 + 0 + 0 +
2 = [ha + k2][ha + k2]∗ and x∗x = (x1 + x2)∗(x1 + x2) =
x2x∗
x∗
2x2 = x∗
1k1 + [(1 − h)b]∗[(1 − h)b] =
[k1 + (1 − h)b]∗[k1 + (1 − h)b]. On the other hand, yy∗ = (y1 + y2)(y1 + y2)∗ =
1+0+0+[(1−h)b][(1−h)b]∗ = [k1+(1−h)b][k1+(1−h)b]∗
y1y∗
and y∗y = (y1 + y2)∗(y1 + y2) = y∗
2y2 =
(ha)∗(ha) + k∗
2k2 = [ha + k2]∗[ha + k2]. Therefore ha + k2 ∼ k1 + (1 − h)b.
1x1 + 0 + 0 + x∗
1y1 + 0 + 0 + y∗
2x2 = k∗
2y2 = y∗
2 = k1k∗
1x1 + x∗
1x2 + x∗
1 + x1x∗
1 + x2x∗
2 + x2x∗
1y1 + y∗
1y2 + y∗
2y1 + y∗
1+y1y∗
2+y2y∗
1+y2y∗
1
2 = ax∗
2h+k2x∗
1 = hax∗
2h = ha+k2 and y1a = a(x1+hx2)∗ = ha+k2 = ay∗
Next, we claim that ha + k2 ≤ a and k1 + (1 − h)b ≤ b. Since h is a central
projection, k2 ≤ (1 − h)a ≤ a and ha ≤ a, implies k2 = x1k2 = x1a = k2x∗
1 = ax∗
and ha = x2ha = x2a = hax∗
2, for some x1, x2 ∈ R. Let y1 = x1 + hx2, then
y1(ha + k2) = x1ha + x1k2 + hx2ha + hx2k2 = ha + k2, (ha + k2)y∗
1 +
hax∗
1, therefore ha+k2 ≤ a.
Similarly, (1 − h)b + k1 ≤ b. Now put ha + k2 = x, (1 − h)b + k1 = z, y = a − x and
w = b − z implies hb − k1 = b − z = w. Then hw = h(hb − k1) = hb − k1 = w and
hy = h(a − x) = ha − hx = ha − ha − hk2 = 0 (since hk2 = 0), i.e., y and w are very
orthogonal. Thus a = x + y, b = z + w where x ⊥ y, z ⊥ w such that we get x ∼ z
with y and w are very orthogonal.
ii) ⇒ i) Let h be a central projection such that hw = w and hy = 0. Then ha =
hx+hy = hx and (1−h)b = (1−h)z +(1−h)w = (1−h)z, where ha = hx ∼ hz ≤ hb
and (1 − h)b = (1 − h)z ∼ (1 − h)x ≤ (1 − h)a. Thus ha . hb and (1 − h)b . (1 − h)a.
Hence a, b are generalized comparable.
(cid:3)
1 + k2x∗
Next result implies that GC for elements is stronger than P C for elements.
Theorem 3.7. If R is a ∗-ring with GC for elements then it has P C for elements.
Proof. Let a, b are elements of R which are not partially comparable. We will show
that aRb = 0. Applying GC to the pair a, b we get orthogonal decompositions
If x 6= 0
a = x + y and b = z + w, where x ∼ z and y, w are very orthogonal.
NATURAL PARTIAL ORDER ON RINGS WITH INVOLUTION
9
and w 6= 0 then a and b are partially comparable, which is a contradiction to the
assumption. Hence x = 0 = w, i.e., a, b are very orthogonal. Let h be a central
projection such that ha = a and hb = 0. Then aRb = haRb = aRhb = 0. Thus R
has P C for elements.
(cid:3)
Lemma 3.8. In an abelian Rickart ∗-ring a ⊥ b if and only if RP (a)RP (b) = 0.
Proof. First we show that ab = 0 if and only if RP (a)RP (b) = 0. Suppose that
ab = 0 which gives b ∈ r({a}) = (1 − RP (a))R. Hence (1 − RP (a))b = b giving
RP (a)b = 0. Since all projections in R are central, we get RP (a) ∈ r({b}) =
(1 − RP (b))R. Which yields RP (b)RP (a) = 0. Conversely, if RP (a)RP (b) = 0,
then ab = (aRP (a))(bRP (b)) = aRP (a)RP (b)b = 0.
Next, Suppose that a ⊥ b. Then there exists x ∈ R such that xa = a = ax∗
and xb = 0 = bx∗, i.e., a(1 − x∗) = 0. Hence RP (a)RP (1 − x∗) = 0. Since
R is abelian, we have RP (1 − x∗) = 1 − RP (x∗) = 1 − RP (x). Consequently,
RP (a)RP (x) = RP (a). On the other hand, xb = 0 implies RP (x)RP (b) = 0. Then
RP (a)RP (b) = RP (a)RP (x)RP (b) = 0, hence ab = 0. Conversely, if ab = 0, then
RP (a)RP (b) = 0. Thus RP (a)a = a = aRP (a) and RP (a)b = 0 = bRP (a). Hence
a ⊥ b.
(cid:3)
The next result shows that the relation ∼ is finitely additive.
Theorem 3.9. Let R be an abelian Rickart ∗-ring. If a1 ⊥ a2, b1 ⊥ b2 with a1 ∼ b1
and a2 ∼ b2, then a1 + a2 ∼ b1 + b2, i.e., the relation ∼ is finitely additive.
i , b∗
i = xix∗
i , a∗
i ai = y∗
i yi, bib∗
Proof. Since a1 ⊥ a2, b1 ⊥ b2, we have RP (a1)RP (a2) = 0 = RP (b1)RP (b2). Also,
a1 ∼ b1 and a2 ∼ b2 there exists xi, yi ∈ R such that aia∗
i =
yiy∗
i bi = xixi with xi = aixi = xibi and yi = biyi = yiai for i = 1, 2. This gives
xi(1 − ai) = 0(since in an abelian Rickart ∗-ring RP (x) = LP (x)), hence RP (xi) =
RP (xi)RP (ai), for i = 1, 2. Then for i 6= j, we have xiaj = xiRP (xi)ajRP (aj) =
xiRP (xi)RP (ai)ajRP (aj) = xiRP (xi)RP (a1)RP (aj)aj = 0. Moreover xix∗
j = 0 =
x∗
i xj for i 6= j. Similarly, we have bjxi = 0 for i 6= j.
Let x = x1 + x2 and y = y1 + y2. Then (a1 + a2)x = a1x1 + a2x1 + a1x2 + a2x2 =
x1 + 0 + 0 + x2 and (b1 + b2)x = b1x1 + b1x2 + b2x1 + b2x2 = x1 + 0 + 0 + x2 = x.
Consider xx∗ = x1x∗
1 + 0 + 0 + a2a∗
2 = (a1 + a2)(a1 + a2)∗
and x∗x = x∗
1x1 + x∗
1b2 = (b1 + b2)∗(b1 + b2). Similarly,
y = (b1 + b2)y = y(a1 + a2), yy∗ = (b1 + b2)(b1 + b2)∗ and y∗y = (a1 + a2)∗(a1 + a2).
Therefore a1 + a2 ∼ b1 + b2.
(cid:3)
2 + x2x∗
2x2 = b∗
2 = a1a∗
1b1 + b∗
1 + x2x∗
2x1 + x∗
1 + x1x∗
1x2 + x∗
Above result ensures that the converse of Theorem 3.7 is true for finite abelian
Rickart ∗-rings.
Theorem 3.10. Let R be a finite abelian Rickart ∗-ring. Then GC for elements
and P C for elements are equivalent.
It is enough to show that, P C for
Proof. Suppose that R has P C for elements.
elements implies GC for elements. Let a, b ∈ R. If aRb = 0, then ab = 0. This
NATURAL PARTIAL ORDER ON RINGS WITH INVOLUTION
10
gives RP (a)b = 0. Since R is an abelian ring, we get a and b are very orthogonal.
Hence we are done. Suppose aRb 6= 0. Hence there exist a0 ≤ a and b0 ≤ b such
that a0 ∼ b0. Let a1, b1 be the largest elements such that a1 ≤ a, b1 ≤ b and a1 ∼ b1.
Then a2 = a − a1 and b2 = b − b1 are such that a2 ≤ a, b2 ≤ b, a1 ⊥ a2 and
b1 ⊥ b2. By the maximality of a1 and b1, we get a2Rb2 = 0, which gives a2 and b2
very orthogonal. Thus we get an orthogonal decompositions a = a1 + a2, b = b1 + b2
such that a1 ∼ b1, a2 and b2 very orthogonal. By Theorem 3.6 we have a and b are
generalized comparable.
(cid:3)
Proposition 3.11. Let R be a ∗-ring with GC for elements and e is any projection
in R. Then eRe also has GC for elements.
Proof. Let a, b ∈ eRe ⊆ R. Then there exists a central projection h in R such
that ha . hb, (1 − h)b . (1 − h)a. Let g = ehe = he ∈ eRe and x be any
element in eRe. Then gx = hex = hx = xh = xeh = xhe = xg. Hence g is a
central projection in eRe with ga = hea = ha, gb = heb = hb, i.e., ga . gb and
(e − g)b = ab = hab = b − hb = (1 − h)b, (e − g)a = ea − hea = a − ha = (1 − h)a,
i.e., (e − g)b . (e − g)a. Thus a and b are generalized comparable in eRe.
(cid:3)
Corollary 3.12. If the matrix ring Mn(R) has GC for elements, then R has GC
for elements.
An ideal I of a ∗-ring R is a ∗-ideal if a∗ ∈ I whenever a ∈ I.
Proposition 3.13. Let I be a ∗-ideal of R. If R has GC for elements, then R/I
has GC for elements.
Proof. Let a + I, b + I ∈ R/I. Applying GC to a, b ∈ R, there exists a central
projection h ∈ R such that ha . hb and (1 − h)b . (1 − h)a. Then passing to
cosets, h + I is central projection in R/I such that (h + I)(a + I) . (h + I)(b + I)
and [(1 + I) − (h + I)](b + I) . [(1 + I) − (h + I)](a + I). Hence R/I has GC for
elements.
(cid:3)
Remark 3.14. The converse of above statement is not true. For, let R = Z10 with
identity map as an involution and I = {0, 2, 4, 6, 8}. Then R/I = {0+I, 1+I} which
has GC for elements trivially. The poset R with natural partial order is depicted in
Figure 1.
7
1
9
3
4
8
2
6
5
0
Figure 1
Here R does not have GC for elements. On the contrary, if R has GC for elements,
then by Theorem 3.7, R has P C for elements. Let a = 2 and b = 4. Then aRb 6= 0
NATURAL PARTIAL ORDER ON RINGS WITH INVOLUTION
11
and 2 ≁ 4, Since 22∗ = 4 and 4∗4 = 6 and R being commutative there is no x ∈ R
such that xx∗ = 4 and x∗x = 6. Hence 2 and 4 are not partially comparable in R, a
contradiction.
References
[1] Berberian S.K., Baer ∗-rings, Springer-Verlag, Berlin and New York, 1972.
[2] Drazin M.P., Natural structure on semigroup with involution, Bull. Amer. Math. Sco.,
84(1),(1978), 139-141.
[3] Janowitz M.F. , On the ∗-order for Rickart ∗-rings, Algebra Universalis, 16(1983), 360-369.
[4] Kaplansky I., Rings of Operators, Benjamin, New York, 1968.
[5] Maeda S., On the lattice of projections of a Baer ∗-ring, J. Sci. Hiroshima Univ.Ser. A, 22
(1958), 75-88.
[6] Maeda S.,On ∗-rings satisfying the square root axiom , Proc. Amer. Math. Soc., 52(1975),
188-190.
[7] Marovt J., Dragan S.R., Dragan S.D., Star, left-star, and right-star partial orders in Rickart
-rings. Linear Multilinear Algebra, 63(2) (2015), 343-365.
[8] Mitsch H., A Natural partial order for semigroups, Proc. Amer. Math. Soc., 97(3),(1986),
384-388.
[9] Thakare N.K. and Nimbhorkar S.K., Modular pairs, compatible pairs and comparability axioms
in Rickart ∗-rings, J. Indian Math. Soc., 59(1993), 179-190.
1 Department of Mathematics, Garware College of Commerce, Pune-411004, In-
dia.
2 Center For Advanced Studies In Mathematics, Department of Mathematics,
Savitribai Phule Pune University, Pune-411007, India.
E-mail address: [email protected]
E-mail address: [email protected]; [email protected]
|
1302.1769 | 3 | 1302 | 2013-10-14T13:59:54 | Examples of polynomial identities distinguishing the Galois objects over finite-dimensional Hopf algebras | [
"math.RA",
"math.QA"
] | We define polynomial H-identities for comodule algebras over a Hopf algebra H and establish general properties for the corresponding T-ideals. In the case H is a Taft algebra or the Hopf algebra E(n), we exhibit a finite set of polynomial H-identities which distinguish the Galois objects over H up to isomorphism. | math.RA | math |
EXAMPLES OF POLYNOMIAL IDENTITIES
DISTINGUISHING THE GALOIS OBJECTS OVER
FINITE-DIMENSIONAL HOPF ALGEBRAS
CHRISTIAN KASSEL
Abstract. We define polynomial H-identities for comodule algebras
over a Hopf algebra H and establish general properties for the corre-
sponding T -ideals.
In the case H is a Taft algebra or the Hopf al-
gebra E(n), we exhibit a finite set of polynomial H-identities which
distinguish the Galois objects over H up to isomorphism.
1. Introduction
By the celebrated Amitsur-Levitzki theorem [4], the standard polynomial
of degree 2n
S2n = X
σ∈Sn
sign(σ) Xσ(1)Xσ(2) · · · Xσ(2n)
is a polynomial identity for the algebra Mn(C) of n×n-matrices with complex
entries, and Mn(C) has no non-zero polynomial identity of degree < 2n.
It follows that the identities S2n distinguish the finite-dimensional simple
associative algebras over C up to isomorphism.
When G is an abelian group, Koshlukov and Zaicev [13] established that
any finite-dimensional G-graded G-simple associative algebra over an alge-
braically closed field of characteristic zero is determined up to G-graded
isomorphism by its G-graded polynomial identities. Aljadeff and Haile [1]
extended their result to non-abelian groups. Similar results exist for other
classes of algebras.
Let now H be a Hopf algebra over a field k. Consider the class of H-
comodule algebras. This class contains the G-graded k-algebras; indeed,
such a algebra is nothing but a comodule algebra over the group algebra kG
equipped with its standard Hopf algebra structure. Similarly, a comodule
algebra over the Hopf algebra of k-valued functions on a finite group G is
the same as a G-algebra, i.e., an associative k-algebra equipped with a left
G-action by algebra automorphisms.
In this context we may wonder whether the following assertion holds: if
H is a Hopf algebra over an algebraically closed field of characteristic zero,
then any finite-dimensional simple H-comodule algebra is determined up to
H-comodule algebra isomorphism by its polynomial H-identities.
2010 Mathematics Subject Classification. 16R50, 16T05, 16T15.
Key words and phrases. Hopf algebra, comodule algebra, polynomial identity.
1
2
CHRISTIAN KASSEL
In this note we provide evidence in support of this assertion by means
of examples. When H is the n2-dimensional Taft algebra Hn2 or the 2n+1-
dimensional Hopf algebra E(n), we exhibit (finitely many) polynomial H-
identities that distinguish the H-Galois objects over an algebraically closed
field. Denoting the T -ideal of polynomial H-identities for a comodule al-
gebra A by IdH(A), we deduce that IdH(A) = IdH (A′) implies that A and
A′ are isomorphic Galois objects. Since each of our finite sets of identities
determines the Galois object A up to isomorphism, it also determine the T -
ideal IdH(A) completely; in a sense which we shall not make precise, these
identities generate the T -ideal.
Before giving the explicit identities, we have to define the concept of a
polynomial H-identity for a comodule algebra A over a Hopf algebra H;
this is done in full generality in § 2. When A is obtained from H by twisting
its product with the help of a two-cocycle, we produce in § 2.3 a universal
map detecting all polynomial H-identities for A, i.e., a map whose kernel is
exactly the T -ideal IdH(A).
In § 3 we deal with the Taft algebra Hn2. After recalling the classification
of its Galois objects, we show that the degree 2n polynomial
(Y X − qXY )n − (1 − q)nX nY n + (1 − q)nc EnX n
is a polynomial Hn2-identity and that it distinguishes the isomorphism
classes of the Galois objects over an algebraically closed field. In § 3.3 we
extend this to certain monomial Hopf algebras.
We prove a similar result for the Hopf algebra E(n) in § 4, exhibiting a fi-
nite set of polynomial E(n)-identities which distinguishes the Galois objects
over E(n).
2. Polynomial identities for comodule algebras
This is a general section in which we define polynomial identities for co-
module algebras and state general properties of the corresponding T -ideals.
We fix a ground field k over which all our constructions will be defined.
In particular, all linear maps are supposed to be k-linear and unadorned
tensor product symbols ⊗ mean tensor products over k. Throughout the
paper we assume that k is infinite.
2.1. Reminder on comodule algebras. We suppose the reader familiar
with the language of Hopf algebra, as presented for instance in [15, 19]. As
is customary, we denote the coproduct of a Hopf algebra by ∆, its counit
by ε, and its antipode by S. We also make use of a Heyneman-Sweedler-type
notation for the image
∆(x) = x1 ⊗ x2
of an element x of a Hopf algebra H under its coproduct.
Recall that a (right) H-comodule algebra over a Hopf k-algebra H is an
associative unital k-algebra A equipped with a right H-comodule structure
whose (coassociative, counital) coaction
δ : A → A ⊗ H
POLYNOMIAL IDENTITIES DISTINGUISHING GALOIS OBJECTS
3
is an algebra map. The subalgebra AH of coinvariants of an H-comodule
algebra A is defined by
AH = {a ∈ A δ(a) = a ⊗ 1} .
A Galois object over H is an H-comodule algebra A such that AH = k 1A
and the map β : A ⊗ A → A ⊗ H given by a ⊗ a′ 7→ (a ⊗ 1) δ(a′) (a, a′ ∈ A)
is a linear isomorphism. For more on Galois objects, see [15, Chap. 8].
Let us now concentrate on a special class of Galois objects, which we call
twisted comodule algebras. Recall that a two-cocycle α on H is a bilinear
form α : H × H → k satisfying the cocycle condition
α(x1, y1) α(x2y2, z) = α(y1, z1) α(x, y2z2)
for all x, y, z ∈ H. We assume that α is invertible (with respect to the con-
volution product) and normalized; the latter means that α(x, 1) = α(1, x) =
ε(x) for all x ∈ H.
Let uH be a copy of the underlying vector space of H. Denote the identity
map u from H to uH by x 7→ ux (x ∈ H). We define the algebra αH as the
vector space uH equipped with the product given by
ux uy = α(x1, y1) ux2y2
for all x, y ∈ H. This product is associative thanks to the cocycle condition;
the two-cocycle α being normalized, u1 is the unit of αH.
The algebra αH is an H-comodule algebra with coaction δ : αH → αH ⊗H
given for all x ∈ H by
δ(ux) = ux1 ⊗ x2 .
It is easy to check that the subalgebra of coinvariants of αH coincides
with k u1 and that the map β : αH ⊗ αH → αH ⊗ H is bijective, turning αH
into a Galois object over H. Conversely, when H is finite-dimensional, any
Galois object over H is isomorphic to a comodule algebra of the form αH.
2.2. Polynomial H-identities. Let us now define the notion of a polyno-
mial H-identity for an H-comodule algebra A. (Polynomial identities for
module algebras over a Hopf algebra have been defined e.g. in [5, 10].)
For each i = 1, 2, . . . consider a copy X H
sends an element x ∈ H to the symbol X x
to X H
i
linear and is determined by its values on a linear basis of H.
i of H; the identity map from H
is
i . Each map x 7→ X x
i
Now take the tensor algebra on the direct sum XH = Li≥1 X H
i :
T = T (XH) = T
M
i≥1
X H
i
.
This algebra is isomorphic to the algebra of non-commutative polynomials in
the indeterminates X xr
, where i = 1, 2, . . . and {xr}r is a linear basis of H.
i
The algebra T is graded with all generators X x
i homogeneous of degree 1.
There is a natural H-comodule algebra structure on T whose coaction
δ : T → T ⊗ H is given by
The coaction obviously preserves the grading.
δ(X x
i ) = X x1
i ⊗ x2 .
4
CHRISTIAN KASSEL
Definition 2.1. An element P ∈ T is a polynomial H-identity for the H-
comodule algebra A if µ(P ) = 0 for all H-comodule algebra maps µ : T → A.
Examples 2.2. (a) When H is the trivial one-dimensional Hopf algebra k,
then a H-comodule algebra A is nothing but an associative unital algebra. In
this case a polynomial H-identity for A is a classical polynomial identity, i.e.,
a non-commutative polynomial P (X1, X2, . . .) such that P (a1, a2, . . .) = 0
for all a1, a2, . . . ∈ A.
(b) When H = kG is a group algebra, a polynomial H-identity is the
same as a G-graded polynomial identity, as defined for instance in [6].
(c) Let H be an arbitrary Hopf algebra and A an H-comodule algebra.
Assume that the subalgebra AH of coinvariants is central in A (such a con-
dition is satisfied e.g. when A = αH is a twisted comodule algebra). For
x, y ∈ H consider the following elements of T :
1
Px = X x1
1 X S(x2)
1
Then the commutators Px X z
2 Qx,y are polynomial
H-identities for A for all x, y, z ∈ H. Indeed, Px and Qx,y are coinvariant
elements of T by [3, Lemma 2.1]. Thus for any H-comodule algebra map
µ : T → A, the elements µ(Px) and µ(Qx,y) are coinvariant, hence central,
in A.
and Qx,y = X x1
2 Px and Qx,y X z
2 −X z
1 X y1
1 X S(x2y2)
.
2 −X z
Denote the set of all polynomial H-identities for A by IdH(A). By defi-
nition,
IH(A) = \
µ
ker µ ,
where µ runs over all H-comodule algebra maps T → A.
Proposition 2.3. The set IH(A) has the following properties:
(a) it is a graded two-sided ideal of T = T (XH ), i.e.,
and
IH(A) T ⊂ IH(A) ⊃ T IH (A)
IH(A) = M
r≥0
(cid:16)IH(A) \ T r(XH )(cid:17) ;
(b) it is a right H-coideal of T , i.e.,
(c) any endomorphism f : T → T of H-comodule algebras preserves IH(A):
δ(cid:0)IH (A)(cid:1) ⊂ IH(A) ⊗ H ;
f (IH(A)) ⊂ IH(A) .
The proof follows the same lines as the proof of [3, Prop. 2.2]. Note that
the assumption that k is infinite is needed to establish that the ideal IH(A)
is graded. We can summarize Property (c) by saying that IH (A) is a T -ideal,
a standard concept in the theory of polynomial identities (see [18]).
It is also clear that, if A → A′ is a map of H-comodule algebras, then
In particular, if A and A′ are isomorphic H-comodule algebras, then
IH(A) ⊂ IH(A′) .
IH(A) = IH(A′) .
POLYNOMIAL IDENTITIES DISTINGUISHING GALOIS OBJECTS
5
In §§ 3 -- 4 we will consider certain finite-dimensional Hopf algebras H such
that the equality IH(A) = IH(A′) for twisted comodule algebras A, A′
implies that A and A′ are isomorphic.
To this end, we next show how to detect polynomial H-identities for
twisted comodule algebras.
2.3. Detecting polynomial identities. Let αH be a twisted comodule
algebra for some normalized convolution invertible two-cocycle α, as defined
in § 2.1. We claim that the polynomial H-identities for αH can be detected
by a "universal" comodule algebra map
which we now define.
µα : T → S ⊗ αH ,
For each i = 1, 2, . . ., consider a copy tH
i of H, identifying x ∈ H linearly
i ∈ tH
with the symbol tx
i , and define S to be the symmetric algebra on the
direct sum tH = Li≥1 tH
i :
S = S(tH ) = S
tH
i
M
i≥1
.
The algebra S is isomorphic to the algebra of commutative polynomials in
the indeterminates txr
, where i = 1, 2, . . . and {xr}r is a linear basis of H.
i
The map µα : T → S ⊗ αH is given by
i ) = tx1
µα(X x
(2.1)
i ⊗ ux2 .
The algebra S ⊗ αH is generated by the symbols tx
i uy (x, y ∈ H; i ≥ 1) as
a k-algebra (we drop the tensor product sign ⊗ between the t-symbols and
the u-symbols). It is an H-comodule algebra whose S(tH)-linear coaction
extends the coaction of αH:
δ(tx
i uy) = tx
i uy1 ⊗ y2 .
It is easy to check that µα : T → S ⊗ αH is an H-comodule algebra map.
Its raison d'etre becomes clear in the following statement.
Theorem 2.4. An element P ∈ T is a polynomial H-identity for αH if and
only if µα(P ) = 0; equivalently, IH(αH) = ker µα.
To prove Theorem 2.4 we need the following proposition.
Proposition 2.5. For every H-comodule algebra map µ : T → αH, there
is a unique algebra map χ : S → k such that
µ = (χ ⊗ id) ◦ µα .
Proof. By the universal property of the tensor algebra T , the set of H-
comodule algebra maps T → αH is naturally in bijection with the vector
space HomH (XH, αH) of H-colinear maps from XH to αH. Now, since the
isomorphism u : H → uH = αH is a comodule map, we have a natural
identification HomH (XH, αH) ∼= HomH(XH , H).
On the other hand, by the universal property of the symmetric alge-
bra S, the set of algebra maps S → k is in bijection with the vector space
Hom(tH , k) of linear maps from tH to k.
6
CHRISTIAN KASSEL
Recall a basic fact from "colinear algebra": for any right H-comodule M
with coaction δ : M → M ⊗ H, the linear map HomH(M, H) → Hom(M, k)
given by µ 7→ ε ◦ µ is an isomorphism with inverse χ 7→ (χ ⊗ id) ◦ δ. As a
consequence, any H-comodule map µ : M → H is necessarily of the form
µ = (χ ⊗ id) ◦ δ, where χ = ε ◦ µ.
Combining these observations yields a proof of the proposition. We can be
even more precise: the algebra map χ : S → k uniquely associated to µ in the
statement is determined on the generators tx
i )). (cid:3)
i ) = ε(u−1(µ(X x
i by χ(tx
Proof of Theorem 2.4. Let P ∈ T be in the kernel of µα. Since by Propo-
sition 2.5 any H-comodule algebra map µ : T → αH is of the form µ =
(χ ⊗ id) ◦ µα, it follows that µ(P ) = 0. This implies ker µα ⊂ IH (αH).
To prove the converse inclusion, start from a polynomial H-identity P
and observe that for every algebra map χ : S → k, the composite map
µ = (χ ⊗ id) ◦ µα from T to αH is a comodule algebra map. By definition of
a polynomial H-identity, we thus have µ(P ) = 0. Now choose a basis {xr}r
of H and expand µα(P ) as µα(P ) = Pr µ(r)
α (P ) belongs
χ(µ(r)
α (P )⊗uxr , where µ(r)
to S. Then
α (P )) uxr .
0 = µ(P ) = X
r
Since the elements uxr are linearly independent, we have χ(µ(r)
for all r. This means that µ(r)
S → k; in other words, the polynomial µ(r)
ground field k being infinite, this implies µ(r)
Therefore, IH(αH) ⊂ ker µα.
α (P )) = 0
α (P ) ∈ S vanishes under any evaluation χ :
α (P ) takes only zero values. The
α (P ) = 0, and hence µα(P ) = 0.
(cid:3)
3. Taft algebras
Let n be an integer ≥ 2 and k a field whose characteristic does not di-
vide n. We assume that k contains a primitive n-th root of unity, which we
denote by q.
3.1. Galois objects over a Taft algebra. The Taft algebra Hn2 has the
following presentation as a k-algebra:
Hn2 = k h x, y xn = 1 , yx = qxy , yn = 0 i
(see [20]). The set {xiyj}0≤i,j≤n−1 is a basis of the vector space Hn2, which
therefore is of dimension n2.
The algebra Hn2 is a Hopf algebra with coproduct ∆, counit ε and an-
tipode S defined by
(3.1)
(3.2)
(3.3)
∆(x) = x ⊗ x ,
ε(x) = 1 ,
S(x) = x−1 = xn−1 ,
∆(y) = 1 ⊗ y + y ⊗ x ,
ε(y) = 0 ,
S(y) = −yx−1 = −q−1xn−1y .
When n = 2, this is the four-dimensional Sweedler algebra.
Given scalars a, c ∈ k such that a 6= 0, we consider the algebra Aa,c with
the following presentation:
Aa,c = k h x, y xn = a , yx = qxy , yn = c i .
POLYNOMIAL IDENTITIES DISTINGUISHING GALOIS OBJECTS
7
This is a right Hn2-comodule algebra with coaction given by the same for-
mulas as (3.1).
By [14, Prop. 2.17 and Prop. 2.22] (see also [12]), any Galois object over Hn2
is isomorphic to Aa,c for some scalars a, c with a 6= 0. Moreover, Aa,c is
isomorphic to Aa′,c′ as a comodule algebra if and only if there is v ∈ k× =
k − {0} such that a′ = vna and c′ = c. It follows that (a, c) 7→ Aa,c induces a
bijection between k×/(k×)n × k and the set of isomorphism classes of Galois
objects over Hn2. (Note that k×/(k×)n is isomorphic to the cohomology
group H 2(G, k×).)
If k is algebraically closed, then k× = (k×)n and any Galois object
over Hn2 is isomorphic to A1,c for a unique scalar c.
3.2. A polynomial identity distinguishing the Galois objects. Let
A = Aa,c be a Galois object as defined in § 3.1. Such a comodule algebra is
a twisted comodule algebra αHn2 for some normalized convolution invertible
two-cocycle α. It can be checked that the map u : Hn2 → Aa,c is such that
u1 = 1, ux = x and uy = y. This allows us to compute the corresponding
universal comodule algebra map µα : T → S ⊗ Aa,c on certain elements of T .
1 for the X-symbols, and
1 for the t-symbols. In view of (2.1) and (3.1), we
For simplicity, we set E = X 1
1 , Y = X y
1 , X = X x
1, tx = tx
1, ty = ty
t1 = t1
have
(3.4)
µα(E) = t1 , µα(X) = txx , µα(Y ) = t1y + tyx .
(In the previous formulas we consider the commuting t-variables as extended
scalars; this allows us to drop the unit u1 of Aa,c and the tensor symbols
between the t-variables and the u-variables.)
Proposition 3.1. The degree 2n polynomial
Pc = (Y X − qXY )n − (1 − q)nX nY n + (1 − q)nc EnX n
is a polynomial Hn2-identity for the Galois object Aa,c.
This polynomial is a generalization of the degree 4 identity
(XY + Y X)2 − 4X 2Y 2 + 4c E2X 2
obtained for the Sweedler algebra in [3, Cor. 10.4] (in the special case b = 0).
Proof. It suffices to check that µα(Pc) = 0 using (3.4) and the defining rela-
tions of Aa,c.
Since yx = qxy, we have
µα(Y X − qXY ) = (t1y + tyx)txx − qtxx(t1y + tyx)
= (1 − q)txtyx2 + t1tx(yx − qxy)
= (1 − q)txtyx2 .
Therefore, in view of xn = a, we obtain
µα ((Y X − qXY )n) = (1 − q)ntn
xtn
1 and µα(X n) = tn
y x2n = a2(1 − q)ntn
xxn = atn
x.
xtn
y .
We also have µα(En) = tn
8
CHRISTIAN KASSEL
To compute µα(Y n), we need the following well-known fact (see [14, Lem-
ma 2.2]): if u and v satisfy the relation vu = quv for some primitive n-root
of unity q, then
(3.5)
(u + v)n = un + vn .
Since yx = qxy, we may apply (3.5) to u = tyx and v = t1y. We thus obtain
µα(Y n) = tn
y xn + tn
1 yn = atn
y + ctn
1 .
Combining the previous equalities, we obtain µα(Pc) = 0.
(cid:3)
The following result shows that the polynomial identity of Proposition 3.1
distinguishes the Galois objects of the Taft algebra.
Theorem 3.2. If k is algebraically closed, then IdHn2 (Aa,c) = IdHn2 (Aa′,c′)
implies that Aa,c and Aa′,c′ are isomorphic comodule algebras.
Proof. By the last remark in § 3.1, we may assume a = a′ = 1. Consider the
elements Pc ∈ IdHn2 (A1,c) and Pc′ ∈ IdHn2 (A1,c′) given by Proposition 3.1.
By the equality of T -ideals, both Pc and Pc′ are polynomial Hn2-identities
for A1,c. Hence, so is the difference Pc − Pc′. Therefore, µα(Pc − Pc′) = 0.
Now,
µα(Pc − Pc′) = µα (cid:0)(c − c′)(1 − q)nEnX n(cid:1) = (c − c′)(1 − q)ntn
1 tn
x .
x 6= 0, we have c − c′ = 0. This implies A1,c = A1,c′.
Since (1 − q)ntn
1 tn
(cid:3)
The previous proof shows that the theorem holds if we only assume an
inclusion IdHn2 (Aa′,c′) ⊂ IdHn2 (Aa,c) of T -ideals. Note also that the single
polynomial Hn2-identity Pc determines the full T -ideal IdHn2 (Aa,c) over an
algebraically closed field.
Remark 3.3. If k is not algebraically closed, then the equality of T -ideals
of Theorem 3.2 imply that Aa′,c′ is a form of Aa,c. Recall that an Hn2-
comodule algebra A is a form of Aa,c if there is an algebraic extension k′
of k such that k′ ⊗k A and k′ ⊗k Aa,c are isomorphic comodule algebras over
the Hopf algebra k′ ⊗k Hn2.
3.3. Monomial Hopf algebras. Taft algebras can be generalized as fol-
lows. Let G be a finite group and x a central element of G of order n. We
also assume that there exists a homomorphism χ : G → k× such that χn = 1
and χ(x) = q is the fixed primitive n-root of unity.
To these data we associate a Hopf algebra H as follows: as an algebra, H is
the quotient of the free product kG ∗ k[y] by the two-sided ideal generated
by the relations
yn = 0 and yg = χ(g) gy .
(g ∈ G)
The elements gyi, where g runs over the elements of G and i = 0, 1, . . . , n−1,
form a basis of H, whose dimension is equal to nG.
The algebra H has a Hopf algebra structure such that kG is a Hopf
subalgebra of H and
∆(y) = 1 ⊗ y + y ⊗ x ,
ε(y) = 0 , S(y) = −yx−1 .
In the literature this Hopf algebra is called a monomial Hopf algebra of
type I (see [9, Sect. 7], [11]).
POLYNOMIAL IDENTITIES DISTINGUISHING GALOIS OBJECTS
9
When G = Z/n and x is a generator of G, then H is the Taft algebra Hn2.
Note that for an arbitrary finite group G the inclusion Z/n x ⊂ G induces a
natural inclusion Hn2 ⊂ H of Hopf algebras.
Given a two-cocycle σ ∈ Z 2(G, k×) of the group G and a scalar c ∈ k, we
define Aσ,c as the algebra generated by the symbols uy and ug for all g ∈ G
and the relations
(3.6)
(3.7)
u1 = 1 ,
uguh = σ(g, h) ugh ,
un
y = c ,
uyug = χ(g) uguy
for all g, h ∈ G. The algebra Aσ,c is an H-comodule algebra with coaction
given by
(3.8)
δ(uy) = 1 ⊗ y + uy ⊗ x and δ(ug) = ug ⊗ g .
(g ∈ G)
Bichon proved that any Galois object over H is isomorphic to one of the
form Aσ,c. Moreover, Aσ,c and Aσ′,c′ are isomorphic comodule algebras if
and only c = c′ and the two-cocycles σ and σ′ represent the same element
of the cohomology group H 2(G, k×). In other words, the map (σ, c) 7→ Aσ,c
induces a bijection between H 2(G, k×)×k and the set of isomorphism classes
of Galois objects over H (see [8, Th. 2.1]).
Let now introduce the same X-symbols E = X 1
1 as
in § 3.2. Since Hn2 is a Hopf subalgebra of H, we can reproduce the same
computation as in the proof of Proposition 3.1. It allows us to conclude that
1 , X = X x
1 , Y = X y
(3.9)
(Y X − qXY )n − (1 − q)nX nY n + (1 − q)nc EnX n
is a polynomial H-identity for the Galois object Aσ,c.
Theorem 3.4. Suppose that k is algebraically closed. If
then Aσ,c and Aσ′,c′ are isomorphic comodule algebras.
IdH(Aσ,c) = IdH(Aσ′,c′) ,
Proof. Proceeding as in the proof of Theorem 3.2, we deduce c = c′ from (3.9).
It remains to check that σ and σ′ represent the same element of H 2(G, k×).
Consider the following diagram:
0 → IdkG(kσG) −→ T (XkG)
µ
−→ S(tkG) ⊗ kσG
ι ↓
ιT ↓
0 → IdH(Aσ,c) −→ T (XH )
ιS ↓
µ
−→ S(tH) ⊗ Aσ,c
Here kσG is the twisted group algebra generated by the symbols ug (g ∈ G)
and Relations (3.6); it is the subalgebra of Aσ,c generated by the elements ug,
where g runs over all elements of G.
The vertical map ιT : T (XkG) → T (XH ) is induced by the natural inclu-
sion kG → H; it is injective. The map ιS : S(tkG) ⊗ kσG → S(tH) ⊗ Aσ,c is
induced by the previous natural inclusion and the comodule algebra inclu-
sion kσG ⊂ Aσ,c; it sends a typical generator tg
i ug′ of S(tkG) ⊗ kσG to the
same expression viewed as an element of S(tH ) ⊗ Aσ,c. The maps µ are the
corresponding universal comodule maps; the horizontal sequences are exact
in view of Theorem 2.4. The diagram is obviously commutative. Hence, the
10
CHRISTIAN KASSEL
restriction ι of ιT to IdkG(kσG) send the latter to IdH(Aσ,c) and is injective.
Since ιS is injective, we have
IdkG(kσG) = T (XkG) ∩ IdH(Aσ,c) .
Consequently, the equality of the theorem implies the equality
IdkG(kσG) = IdkG(kσ′
G)
of T -ideals of graded identities. We now appeal to [2, Sect. 1], from which it
follows that σ and σ′ are cohomologous two-cocycles.
(cid:3)
4. The Hopf algebras E(n)
We now deal with the Hopf algebras E(n) considered in [7, 9, 16, 17].
When k is an algebraically closed field of characteristic zero, E(n) is up to
isomorphism the only 2n+1-dimensional pointed Hopf algebra with coradi-
cal kZ/2.
4.1. Galois objects over E(n). Fix an integer n ≥ 1. Assume that the
field k is of characteristic 6= 2. The algebra E(n) is generated by elements
x, y1, . . . , yn subject to the relations
x2 = 1 ,
y2
i = 0 ,
yix + xyi = 0 ,
yiyj + yjyi = 0
for all i, j = 1, . . . , n. As a vector space, E(n) is of dimension 2n+1.
The algebra E(n) is a Hopf algebra with coproduct ∆, counit ε and
antipode S determined for all i = 1, . . . , n by
(4.1)
(4.2)
(4.3)
∆(x) = x ⊗ x ,
ε(x) = 1 ,
S(x) = x ,
∆(yi) = 1 ⊗ yi + yi ⊗ x ,
ε(yi) = 0 ,
S(yi) = −yix .
When n = 1, the Hopf algebra E(n) coincides with the Sweedler algebra.
The Galois objects over E(n) can be described as follows. Let a 6= k×,
c = (c1, . . . , cn) ∈ kn, and d = (di,j)i,j=1,...,n be a symmetric matrix with
entries in k. To this collection of scalars we associate the algebra A(a, c, d)
generated by the symbols u, u1, . . . , un and the relations
(4.4)
u2 = a ,
u2
i = ci ,
uui + uiu = 0 ,
uiuj + ujui = di,j
for all i, j = 1, . . . , n. It is a comodule algebra with coaction δ : A(a, c, d) →
A(a, c, d) ⊗ E(n) given for all i = 1, . . . , n by
(4.5)
δ(u) = u ⊗ x ,
δ(ui) = 1 ⊗ yi + ui ⊗ x .
It follows from [17, Sect. 4] completed by [16, Sect. 2] that any Galois ob-
ject over E(n) is isomorphic to a comodule algebra of the form A(a, c, d).
Moreover, A(a, c, d) and A(a′, c′, d′) are isomorphic Galois objects if and
only if c = c′, d = d′ and a′ = v2a for some nonzero scalar v.
Consequently, if k is algebraically closed, then A(a, c, d) is isomorphic to
A(1, c, d) for any a 6= 0, and the Galois objects A(1, c, d) and A(1, c′, d′) are
isomorphic if and only if c = c′ and d = d′.
POLYNOMIAL IDENTITIES DISTINGUISHING GALOIS OBJECTS
11
4.2. Two families of polynomial identities. Let us now compute the
universal comodule algebra map µα : T → S ⊗ A(a, c, d) corresponding to
the comodule algebra A(a, c, d).
We set E = X 1
1 , X = X x
for the X-symbols, and t0 = t1
1,
1 for the corresponding t-symbols. In view of (2.1) and (4.5),
1 , Yi = X yi
1
1, ti = tyi
tx = tx
we have
(4.6)
µα(E) = t0 , µα(X) = tx u , µα(Yi) = t0 ui + ti u
for all i = 1, . . . , n.
Proposition 4.1. The degree 4 polynomials
(XYi + YiX)2 − 4X 2Y 2
i + 4ci E2X 2
(1 ≤ i ≤ n)
and
2(YiYj + YjYi)X 2 − (XYi + YiX)(XYj + YjX) − 2di,j E2X 2
(1 ≤ i ≤ j ≤ n)
are polynomial E(n)-identities for the Galois object A(a, c, d).
Proof. In view of (4.6) and of the defining relations of A(a, c, d), we obtain
µα(E2) = t2
0 , µα(X 2) = at2
x , µα(Y 2
i ) = at2
i + cit2
0 ,
µα(XYi + YiX) = 2atxti , µα((YiYj + YjYi) = 2atitj + di,jt2
0 .
From these equalities, it is easy to check that the above polynomials belong
to the kernel of µα, hence are polynomial E(n)-identities.
(cid:3)
Theorem 4.2. Suppose that k is algebraically closed. If
IdE(n)(A(a, c, d)) = IdE(n)(A(a′, c′, d′)) ,
then A(a, c, d) and A(a′, c′, d′) are isomorphic comodule algebras.
Proof. We proceed as in the proof of Theorem 3.2 by using the identities
of Proposition 4.1. Note that there is such an identity for each scalar used
to parametrize the Galois objects, and each such scalar appears as the co-
efficient of the monomial E2X 2; the latter cannot be an identity since its
image under the universal comodule map, being equal to at2
x, does not
vanish.
(cid:3)
0t2
We finally note that the set of n(n + 3)/2 polynomial E(n)-identities of
Proposition 4.1 determines the T -ideal IdE(n)(A(a, c, d)).
Acknowledgment
My warmest thanks go to Eli Aljadeff who initiated me to the theory of
polynomial algebras (classical and graded) and drew my attention to the
question addressed here.
12
CHRISTIAN KASSEL
References
[1] E. Aljadeff, D. Haile, Simple G-graded algebras and their polynomial identities,
arXiv:1107.4713, to appear in Trans. Amer. Math. Soc.
[2] E. Aljadeff, D. Haile, M. Natapov, Graded identities of matrix algebras and the uni-
versal graded algebra, Trans. Amer. Math. Soc. 362 (2010), 3125 -- 3147.
[3] E. Aljadeff, C. Kassel, Polynomial identities and noncommutative versal torsors, Adv.
Math. 218 (2008), 1453 -- 1495.
[4] A. S. Amitsur, J. Levitzki, Minimal identities for algebras, Proc. Amer. Math. Soc. 1
(1950), 449 -- 463.
[5] Y. A. Bahturin, V. Linchenko, Identities of algebras with actions of Hopf algebras,
J. Algebra 202 (1998), 634 -- 654.
[6] Y. A. Bahturin, M. Zaicev, Identities of graded algebras, J. Algebra 205 (1998), 1 -- 12.
[7] M. Beattie, S. Dascalescu, L. Grunenfelder, Constructing pointed Hopf algebras by
Ore extensions, J. Algebra 225 (2000), 743 -- 770.
[8] J. Bichon, Galois and bigalois objects over monomial non-semisimple Hopf algebras,
J. Algebra Appl. 5 (2006), 653 -- 680.
[9] J. Bichon, G. Carnovale, Lazy cohomology: an analogue of the Schur multiplier for
arbitrary Hopf algebras, J. Pure Appl. Algebra 204 (2006), 627 -- 665.
[10] A. Berele, Cocharacter sequences for algebras with Hopf algebra actions, J. Algebra
185 (1996), 869 -- 885.
[11] X.-W. Chen, H.-L. Huang, Y. Ye, P. Zhang, Monomial Hopf algebras, J. Algebra 275
(2004), 212 -- 232.
[12] Y. Doi, M. Takeuchi, Quaternion algebras and Hopf crossed products, Comm. Algebra
23 (1995), 3291 -- 3325.
[13] P. Koshlukov, M. Zaicev, Identities and isomorphisms of graded simple algebras, Lin-
ear Algebra Appl. 432 (2010), 3141 -- 3148.
[14] A. Masuoka, Cleft extensions for a Hopf algebra generated by a nearly primitive ele-
ment, Comm. Algebra 22 (1994), 4537 -- 4559.
[15] S. Montgomery, Hopf algebras and their actions on rings, CBMS Conf. Series in
Math., vol. 82, Amer. Math. Soc., Providence, RI, 1993.
[16] A. Nenciu, Cleft extensions for a class of pointed Hopf algebras constructed by Ore
extensions, Comm. Algebra 29 (2001), 1959 -- 1981.
[17] F. Panaite, F. van Oystaeyen, Quasitriangular structures for some pointed Hopf al-
gebras of dimension 2n, Comm. Algebra 27 (1999), 4929 -- 4942.
[18] L. Rowen, Polynomial identities in ring theory, Pure and Applied Mathematics, 84,
Academic Press, Inc., New York-London, 1980.
[19] M. Sweedler, Hopf algebras, W. A. Benjamin, Inc., New York, 1969.
[20] E. J. Taft, The order of the antipode of finite-dimensional Hopf algebra, Proc. Nat.
Acad. Sci. U.S.A. 68 (1971), 2631 -- 2633.
Christian Kassel: Institut de Recherche Math´ematique Avanc´ee, CNRS &
Universit´e de Strasbourg, 7 rue Ren´e Descartes, 67084 Strasbourg, France
E-mail address: [email protected]
URL: www-irma.u-strasbg.fr/~kassel/
|
1510.00747 | 1 | 1510 | 2015-10-02T22:18:05 | Elementary triangular matrices and inverses of $k$-Hessenberg and triangular matrices | [
"math.RA"
] | We use elementary triangular matrices to obtain some factorization, multiplication, and inversion properties of triangular matrices. We also obtain explicit expressions for the inverses of strict $k$-Hessenberg matrices and banded matrices. Our results can be extended to the cases of block triangular and block Hessenberg matrices. | math.RA | math |
Elementary triangular matrices and inverses of
k-Hessenberg and triangular matrices
Luis Verde-Star
Oct. 2, 2015
Abstract
We use elementary triangular matrices to obtain some factoriza-
tion, multiplication, and inversion properties of triangular matrices.
We also obtain explicit expressions for the inverses of strict k-Hessenberg
matrices and banded matrices. Our results can be extended to the
cases of block triangular and block Hessenberg matrices. An n × n
lower triangular matrix is called elementary if it is of the form I + C,
where I is the identity matrix and C is lower triangular and has all of
its nonzero entries in the k-th column, where 1 ≤ k ≤ n.
Keywords: Triangular matrices, factorization, k-Hessenberg matrices,
matrix inversion.
1
Introduction
The importance of triangular, Hessenberg, and banded matrices is well-
known. Many problems in linear algebra and matrix theory are solved by
some kind of reduction to problems involving such types of matrices. This
occurs, for example, with the LU factorizations and the QR algorithms.
In this paper, we study first some simple properties of triangular matrices
using a particular class of such matrices that we call elementary. Using
elementary matrices we obtain factorization and inversion properties and a
formula for powers of triangular matrices.
Some of our results may be useful to develop parallel algorithms to com-
pute powers and inverses of triangular matrices and also of block-triangular
matrices.
1
In the second part of the paper we obtain an explicit formula for the
inverse of a strict k-Hessenberg matrix in terms of the inverse of an associated
triangular matrix. Our formula is obtained by extending n × n k-Hessenberg
matrices to (n + k) × (n + k) invertible triangular matrices and using some
natural block decompositions. Our formula can be applied to find the inverses
of tridiagonal and banded matrices.
The problem of finding the inverses of Hessenberg and banded matri-
ces has been studied by several authors, using different approaches, such as
determinants and recursive algorithms. See [1], [2], [3], [8], and [9].
2 Elementary triangular matrices
In this section n denotes a fixed positive integer, N = {1, 2, . . . , n}, and T
denotes the set of lower triangular n × n matrices with complex entries. An
element of T is called elementary if it is of the form I + Ck, for some k ≤ n,
where I is the identity matrix and Ck is lower triangular and has all of its
nonzero entries in the k-th column.
Let A = [aj,k] ∈ T . For each k ∈ N we define Ek as the matrix obtained
from the identity matrix In by replacing its k-th column with the k-th column
of A, that is, (Ek)j,k = aj,k for j = k, k + 1, . . . , n, (Ek)j,j = 1 for j 6= k, and
all the other entries of Ek are zero. The matrices Ek are called the elementary
factors of A because
A = E1E2 · · · En.
(2.1)
Let us note that performing the multiplications in (2.1) does not require any
arithmetical operations. It is just putting the columns of A in their proper
places.
If for some k we have ak,k 6= 0 then Ek is invertible and it is easy to verify
that
(Ek)−1 = I −
1
ak,k
(Ek − I).
(2.2)
Note that (Ek)−1 is also an elementary lower triangular matrix.
If A is
invertible then all of its elementary factors are invertible and from (2.1) we
obtain
A−1 = (En)−1(En−1)−1 · · · (E2)−1(E1)−1.
(2.3)
Therefore A−1 is the product of the elementary factors of the matrix B =
(E1)−1(E2)−1 · · · (En)−1, but in reverse order.
2
Notice that B = I + (I − A)D−1, where D = Diag(a1,1, a2,2, . . . , an,n).
Therefore
B−1 = EnEn−1 · · · E2E1.
(2.4)
If we are only interested in computing the inverse of A we can find the
inverse of A = AD−1, which has all its diagonal entries equal to one, and then
we have A−1 = D−1 A−1. This means that it would be enough to consider
matrices with diagonal entries equal to one to study the construction of
inverses of triangular matrices. If we consider general triangular matrices we
can also obtain results about the computation of powers of such matrices.
We will obtain next some results about products of elementary factors.
We define Ck = Ek − I for k ∈ N. Note that (Ck)k,k = ak,k − 1, (Ck)j,k =
aj,k for j = k + 1, . . . , n, and all the other entries are zero, that is, all the
nonzero entries of Ck lie on the k-th column.
It is clear that A has the
following additive decomposition
A = I + C1 + C2 + · · · + Cn.
(2.5)
Let L be the n × n matrix such that Lk+1,k = 1 for k = 1, 2, . . . , n − 1
and all its other entries are zero. We call L the shift matrix, since M L is
M shifted one column to the left. Note that T is invariant under the maps
M → M L and M → LM.
In order to simplify the notation we will write the entries in the main
diagonal of A as ak,k = xk for k = 1, 2, . . . , n.
The matrices Ck have simple multiplication properties that we list in the
following proposition.
Theorem 2.1.
1. If j < k then CjCk = 0.
2. For m ≥ 1 we have C m
k = (xk − 1)m−1Ck.
3. If k > j then CkCj = ak,jCkLk−j.
The proofs are straightforward computations. Notice that all the nonzero
entries of CkLk−j are in the j-th column.
From part 3 of the previous theorem we obtain immediately the following
multiplication formula. If r ≥ 2 and k1 > k2 > · · · > kr then
Ck1Ck2 · · · Ckr = ak1,k2ak2,k3 · · · akr−1,kr Ck1Lk1−kr .
(2.6)
3
If K is a subset of N with at least two elements we define the matrix
G(K) = ak1,k2ak2,k3 · · · akr−1,kr Ck1Lk1−kr ,
(2.7)
where K = {k1, k2, . . . , kr} and k1 > k2 > · · · > kr. Let us note that all
the nonzero entries of G(K) are in its kr-th column. If K contains only one
element, that is, K = {j} then we put G(K) = Cj.
Since Ek = I + Ck, the multiplication properties of the Ck can be used
to obtain some corresponding properties of the Ek.
Theorem 2.2.
1. If k > j then EkEj = I + Ck + Cj + G({k, j}).
2. For m ≥ 1 we have
Em
k = I +
xm
k − 1
xk − 1
Ck.
3. If K = {k1, k2, . . . , kr} and k1 > k2 > · · · > kr then
Ek1Ek2 · · · Ekr = I + XJ⊆K
G(J).
Proof: The proof of the first part is trivial. For the second part, use part
2 of Theorem 2.1 and the binomial formula. For part 3, write each factor
in the form Ekj = I + Ckj , expand the product, collect terms and use the
definition of the function G. Alternatively, we can use part 1 repeatedly and
then use the definition of G.
(cid:3)
Observe that the number of summands in part 3 is at most equal to the
number of subsets of K. If some of the entries aj,k are equal to zero then
some of the matrices G(J) are zero.
Taking K = N in part 3 of Theorem 2.2 we obtain
EnEn−1 · · · E2E1 = I + XJ⊆N
G(J).
(2.8)
Let k ∈ N. Then the summands in (2.8) that may have nonzero entries
in the k-th column (other than I) are the matrices G({k} ∪ J) where J ⊆
{k + 1, k + 2, . . . , n}, and thus the number of such matrices is at most equal
to 2n−k. For k = n there is only one, which is Cn, for k = n − 1 they are
Cn−1 and G({n − 1, n}), and so on. Since G({k} ∪ J) is a scalar multiple of
CmLk−m where m is the largest element of J, we can group together all the
4
terms having the same largest element m and therefore the k-th column of the
product in (2.8) is a linear combination of Ck, Ck+1L, Ck+2L2, . . . , CnLn−k,
and the k-th column of the identity matrix.
Therefore, if we have computed EnEn−1 · · · Ek then the columns with
indices k, k +1, . . . , n are determined and do not change when we multiply by
Ek−1, Ek−2, etc. Thus EnEn−1 · · · Ek+1 and EnEn−1 · · · Ek+1Ek only differ in
their k-th columns. This means that computing the sequence EnEn−1 · · · Ek,
for k = n, n − 1, . . . , 2, 1 we obtain B−1 column by column, starting from the
last one. This procedure may be useful to develop parallel algorithms for the
computation of inverses of triangular matrices.
We consider now explicit expressions for the positive powers of a trian-
gular matrix A in terms of its elementary factors. We start with A2. Using
(2.5) and part 1 of Theorem 2.1 we get
A2 = (I + C1 + C2 + · · · + Cn)2 = I + 2
Therefore, by Theorem 2.1 we have
n
Xk=1
n
Ck +
Xk=1
C 2
k +Xj<k
CkCj. (2.9)
n
A2 = I +
Xk=1
(1 + xk)Ck +Xj<k
ak,jCkLk−j,
(2.10)
where the last sum runs over all pairs (j, k) such that 1 ≤ j < k ≤ n.
Let K = {k1, k2, . . . , kr} ⊆ N, where k1 > k2 > · · · > kr, and let m be a
positive integer. We define the scalar valued function g(K, m) as follows
g(K, m) = ∆[1, xk1, xk2, . . . , xkr ]tm+r,
(2.11)
where ∆[1, xk1, xk2, . . . , xkr ] denotes the divided differences functional with
respect to the numbers 1, xk1, xk2, . . . , xkr. g(K, m) is a symmetric polyno-
mial in the xj. For the properties of divided differences see [6].
Using induction and some basic properties of divided differences we can
obtain an expression for Am in terms of matrices of the form G(J).
Theorem 2.3. For m ≥ 1 we have
n
Am = I +
Xj=1
g({j}, m − 1)Cj + XJ⊂N, J=2
XJ⊂N, J=3
g(J, m−3)G(J)+· · ·+ XJ⊂N, J=m
g(J, m − 2)G(J)+
g(J, 0)G(J).
(2.12)
5
Note that the numbers g(J, 0) in the last sum are all equal to 1. A similar
result for triangular matrices with distinct diagonal entries xj was obtained
by Shur [4]. Other formulas for powers of general square matrices appear in
[7].
3
Inverses of Hessenberg matrices
In this section we use the fact that a k-Hessenberg matrix H is a submatrix
of a larger triangular matrix to obtain a formula for the inverse of H, in case
it exists. We also characterize the invertible k-Hessenberg matrices in terms
of properties of the blocks in a natural block decomposition.
We call an n × n matrix H = [hi,j] lower k-Hessenberg if hi,j = 0 for
i < j − k. We say that H is strict lower k-Hessenberg if hi,j 6= 0 for i = j − k.
Any lower k-Hessenberg n × n matrix H has the following block decom-
position. Let m = n − k. Then
D C(cid:21) ,
H = (cid:20)B A
(3.1)
where A is m × m, B is m × k, C is k × m, and D is k × k. Note that A is
lower triangular and it is invertible if H is strict k-Hessenberg.
Extending the block decomposition of equation (3.1) to form a lower
triangular matrix we obtain the following result.
Theorem 3.1. Let H be strict k-Hessenberg with the block decomposition
given in (3.1). Then H is invertible if and only if CA−1B − D is invertible
and if H is invertible we have
H −1 = (cid:20) 0
A−1 0(cid:21) −(cid:20)Ik
0
E(cid:21) G−1(cid:2)F Ik(cid:3) ,
(3.2)
where Ik is the k × k identity matrix, E = −A−1B,
G = CA−1B − D.
F = −CA−1, and
Proof: Suppose that G = CA−1B − D is invertible. Define the lower
triangular (n + k) × (n + k) matrix T by
T =
0
0
Ik
B A 0
D C Ik
.
6
(3.3)
It is easy to verify that
T −1 =
0
Ik
E A−1
G F
,
0
0
Ik
(3.4)
where E, F, and G are as previously defined. Consider now the block decom-
position
H S(cid:21) ,
T = (cid:20)R 0
where
R = (cid:2)Ik 0(cid:3) ,
S = (cid:20) 0
Ik(cid:21) .
From T T −1 = In+k we obtain the equations
and
H (cid:20) 0
0
A−1 0(cid:21) + S(cid:2)F Ik(cid:3) = In,
H (cid:20)Ik
E(cid:21) + SG = 0.
(3.5)
(3.6)
Since G is invertible by hypothesis, we can solve for S in the last equation
and substitute the resulting expression for S in equation (3.5). In this way
we obtain
H (cid:20) 0
A−1 0(cid:21) − H (cid:20)Ik
0
E(cid:21) G−1(cid:2)F Ik(cid:3) = In,
which implies that H is invertible and (3.2) holds.
Now suppose that H is invertible and let
H −1 = (cid:20) U V
W Y(cid:21) ,
where the block decomposition is compatible with the decomposition of H
given in (3.1). Then, from the equation H −1H = In we obtain
U B + V D = Ik,
U A + V C = 0.
(3.7)
Since A is invertible the second equation in (3.7) yields U + V CA−1 = 0, and
multiplying by B on the right we obtain U B + V CA−1B = 0. Combining
7
this last equation with the first one in (3.7) we get V (CA−1B − D) = −Ik
and therefore G is invertible and G−1 = −V .
(cid:3)
Let us observe that an important part of the computation of H −1 is the
computation of the inverse of the triangular matrix A. Note also that any
given strict k-Hessenberg matrix can be modified to become invertible by
changing the k × k block D of (3.1) in a suitable way.
In the case of k = 1 the matrix G reduces to a number and then the
second term in the right-hand side of (3.2) is the product of a column vector
times a row vector. See [2].
Note that Theorem 3.1 holds for tridiagonal matrices and also for banded
matrices. Suppose that k = 1, H is tridiagonal, and n ≥ 3. Then the
matrices in the block decomposition of H are B = [h1,1 h2,1 0 0 · · · 0]T,
C = [0 0 · · · 0 hn,n−1 hn,n], and D = 0. In this case A is lower triangular
and tridiagonal. Using the row version of the theory of elementary triangular
matrices, that we describe in the next section, it is easy to construct a re-
cursive algorithm to compute the inverses of tridiagonal matrices as the size
n increases.
The proof of Theorem 3.1 only uses the hypothesis that the block A is
invertible. Therefore the theorem holds also for other types of matrices, such
as block Hessenberg matrices.
4 Row and block versions of the theory of
elementary triangular matrices
In this section we present a brief description of two variations on the theory
of elementary triangular matrices presented in section 2. The first one is
obtained when we consider lower triangular matrices that are the sum of the
identity matrix plus a matrix that has all of its nonzero entries in a single
row.
In the second one we consider block lower triangular matrices with
square blocks along the diagonal that may have different sizes.
Recall that T denotes the n × n lower triangular matrices. An element of
T is called row elementary if it is of the form I + Rk, for some k ≤ n, where
I is the identity matrix and Rk is lower triangular and has all of its nonzero
entries in the k-th row.
Let A = [ak,j] ∈ T . For k ∈ N we define Fk as the matrix obtained from
the identity matrix I by replacing its k-th row with the k-th row of A, that
8
is, (Fk)k,j = ak,j for j = 1, 2, . . . , k, (Fk)j,j = 1 for j 6= k, and all the other
entries of Fk are zero. The matrices Fk are called the elementary factors by
rows of A because
If for some k we have ak,k 6= 0 then Fk is invertible and
A = F1F2 · · · Fn.
(Fk)−1 = I −
1
ak,k
(Fk − I).
Therefore, if A is invertible then
A−1 = (Fn)−1(Fn−1)−1 · · · (F2)−1(F1)−1.
The matrices (Fj)−1 are the elementary factors by rows of the matrix
(4.1)
(4.2)
(4.3)
B = I −
1
ak,k
(A − I),
and
B−1 = FnFn−1 · · · F2F1.
(4.4)
Define Rk = Fk − I for k ∈ N. Then A = I + R1 + R2 + · · · + Rn. It
is easy to see that FkFk−1 · · · F2F1 and Fk−1 · · · F2F1 only differ in the k-th
row, and the difference is a linear combination of translates of R1, R2, . . . , Rk.
Note that FkFk−1 · · · F2F1 is the inverse of the submatrix of B obtained by
deleting the rows and columns with indices k + 1, k + 2, . . . , n, which is often
called the k × k section of B. Therefore, computing the sequence of matrices
FkFk−1 · · · F2F1 for k = 1, 2, 3, . . . yields a recursive algorithm that gives
the inverse of B row by row. That algorithm can also be used to find the
inverses of the sections of infinite lower triangular matrices such as the ones
considered in [5]. The inversion algorithm introduced in [5] can be combined
with the computation of inverses of diagonal blocks of a triangular matrix,
using multiplication of elementary matrices, by rows or by columns.
The concept of elementary triangular matrices (by colummns of rows) can
be generalized to the case of block triangular matrices in a natural way. We
describe next how it is done in the case of column block elementary matrices.
Let k1, k2, . . . , kr be positive integers such that n = k1 + k2 + · · · + kr. Let
Xj be a kj × kj matrix for 1 ≤ j ≤ r and let A be an n × n block matrix
that has the matrices Xj along the diagonal and all its other nonzero entries
below the diagonal blocks.
9
For j ∈ {1, 2, . . . , r} let Ej, called the block elementary factor of A by
columns, be the matrix that coincides with A in all the columns correspond-
ing to the diagonal block Xj, that is, the columns with indices between
k1 + k2 + · · · + kj−1 + 1 and k1 + k2 + · · · + kj, and coincides with the identity
matrix in the rest of the columns. Then we have A = E1E2 · · · Er.
If the block Xj is invertible then Ej is also invertible and
(Ej)−1 = I − (Ej − I) Diag(Imj , (Xj)−1, Ipj ),
(4.5)
where Diag(Imj , (Xj)−1, Ipj) is the block diagonal matrix that coincides with
the block diagonal of A in the j-th block and with the identity matrix in the
rest of the blocks. Note that mj = k1 + k2 + · · · + kj−1 and pj = n − mj − kj.
If all the diagonal blocks Xj are invertible then
A−1 = (Er)−1(Er−1)−1 · · · (E2)−1(E1)−1.
(4.6)
References
[1] M. Elouafi, A. Driss Aiat Hadj, A new recursive algorithm for inverting
Hessenberg matrices, Appl. Math. Comp., 214 (2009) 497–499.
[2] Y. Ikebe, On inverses of Hessenberg matrices, Linear Algebra Appl., 24
(1979) 93–97.
[3] M. J. Piff, Inverses of banded and k-Hessenberg matrices, Linear Algebra
Appl., 85 (1987) 9–15.
[4] W. Shur, A simple closed form for triangular matrix powers, Electr. J.
Linear Algebra, 22 (2011) 1000–1003.
[5] L. Verde-Star, Infinite triangular matrices, q-Pascal matrices, and de-
terminantal representations, Linear Algebra Appl., 434 (2011) 307–318.
[6] L. Verde-Star, Divided differences and linearly recurrent sequences,
Stud. Appl. Math. 95 (1995) 433–456.
[7] L. Verde-Star, Functions of matrices, Linear Algebra Appl., 406 (2005)
285–300.
[8] Z. Xu, On inverses and generalized inverses of Hessenberg matrices,
Linear Algebra Appl., 101 (1988) 167–180.
10
[9] T. Yamamoto, Y. Ikebe, Inversion of band matrices, Linear Algebra
Appl., 24 (1979) 105–111.
Department of Mathematics, Universidad Aut´onoma Metropolitana,
Iztapalapa, Apartado 55–534, Mexico D.F., Mexico,
e-mail: [email protected].
11
|
1912.06696 | 1 | 1912 | 2019-12-13T20:34:51 | On moduli spaces of K\"ahler-Poisson algebra over rational functions in two variables | [
"math.RA"
] | K\"ahler-Poisson algebras were introduced as algebraic analogues of function algebras on K\"ahler manifolds, and it turns out that one can develop geometry for these algebras in a purely algebraic way. A K\"ahler-Poisson algebra consists of a Poisson algebra together with the choice of a metric structure, and a natural question arises: For a given Poisson algebra, how many different metric structures are there, such that the resulting K\"ahler-Poisson algebras are non-isomorphic? In this paper we initiate a study of such moduli spaces of K\"ahler-Poisson algebras defined over rational functions in two variables. | math.RA | math | ON MODULI SPACES OF K AHLER-POISSON ALGEBRAS OVER
RATIONAL FUNCTIONS IN TWO VARIABLES
AHMED AL-SHUJARY
c
e
D
3
1
]
.
A
R
h
t
a
m
[
1
v
6
9
6
6
0
.
2
1
9
1
:
v
i
X
r
a
Abstract. Kahler-Poisson algebras were introduced as algebraic analogues of
function algebras on Kahler manifolds, and it turns out that one can develop
geometry for these algebras in a purely algebraic way. A Kahler-Poisson alge-
bra consists of a Poisson algebra together with the choice of a metric structure,
and a natural question arises: For a given Poisson algebra, how many different
metric structures are there, such that the resulting Kahler-Poisson algebras
are non-isomorphic? In this paper we initiate a study of such moduli spaces
of Kahler-Poisson algebras defined over rational functions in two variables.
1. Introduction
In [3] we initiated the study of Kahler-Poisson algebras as algebraic analogues
of algebras of functions on Kahler manifolds. Kahler-Poisson algebras consist of
a Poisson algebra together with a metric structure. This study was motivated
by the results in [1, 2], where many aspects of the differential geometry of an
embedded almost Kahler manifold Σ can be formulated in terms of the Poisson
structure of the algebra of functions of Σ. In [3] we showed that "the Kahler -- Poisson
condition", being the crucial identity in the definition of Kahler-Poisson algebras,
allows for an identification of geometric objects in the Poisson algebra which share
important properties with their classical counterparts. For instance, we proved the
existence of a unique Levi-Civita connection on the module generated by the inner
derivations of the Kahler-Poisson algebra, and that the curvature operator has all
the classical symmetries. In [5] we explore further algebraic properties of Kahler-
Poisson algebras.
In particular, we find appropriate definitions of morphisms of
Kahler-Poisson algebras as well as subalgebras, direct sums and tensor products.
Starting from a Poisson algebra A, it is interesting to ask the following ques-
tion: How many non-isomorphic Kahler-Poisson algebras can one construct from
A? This amounts to the study of a "moduli space" for Kahler-Poisson algebras,
in analogy with the corresponding problem for Riemannian manifolds, where one
consider metrics giving rise to non-isometric Riemannian manifolds.
In general,
this is a hard problem and, in this note, we will focus on approaching this problem
in a particular setting. To this end, we start from the polynomial algebra C[x, y].
In order to understand isomorphism classes of Kahler-Poisson algebras based on
this algebra, one needs to study automorphisms of C[x, y]. In [10] Jung shows that
every automorphism of C[x, y] is a composition of so called elementary automor-
phisms. This result was later extended to fields of arbitrary characteristics and free
associative algebras ([6, 7, 11, 13, 15]). In the notation of [12] (which also provides
2010 Mathematics Subject Classification. 17B63.
1
2
AHMED AL-SHUJARY
an elementary proof of the automorphism theorem), every k-algebra automorphism
of k[x, y] is a finite composition of automorphisms of the type:
(1) x 7→ x, y 7→ y + h(x) with h(x) ∈ k[x]
(2) x 7→ a11x + a12y + a13, y 7→ a21x + a22y + a23 with a11a22 6= a21a12
for aij ∈ k. Using these automorphisms, we shall initiate a study of isomorphism
classes of Kahler-Poisson algebras over C[x, y]. However, as shown in [3], the con-
struction of a Kahler-Poisson algebra over C[x, y] will in most cases involve a lo-
calization of the algebra. Therefore, we shall rather start from C(x, y), the algebra
of rational functions of two variables, together with an appropriate linear Poisson
structure. Although we do not solve the problem in its full generality, the classifi-
cation results for certain classes of metrics obtained below, give an insight into the
complexity of the general problem.
2. Kahler-poisson algebras
We begin this section by recalling the main object of our investigation. Let us
consider a Poisson algebra (A, {·, ·}) over C and let {x1, ..., xm} be a set of dis-
tinguished elements of A. These elements play the role of functions providing an
isometric embedding for Kahler manifolds (cf.
[2]). Let us recall the definition of
Kahler-Poisson algebras together with a few basic results (cf. [3]).
Definition 2.1. Let (A, {·, ·}) be a Poisson algebra over C and let x1, ..., xm ∈ A.
Given a symmetric m × m matrix g = (gij) with entries gij ∈ A, for i, j = 1, ..., m,
we say that the triple K = (A, g, {x1, ..., xm}) is a Kahler -- Poisson algebra if there
exists η ∈ A such that
(2.1)
m
Xi,j,k,l=1
η{a, xi}gij{xj, xk}gkl{xl, b} = −{a, b}
for all a, b ∈ A. We call equation (2.1) "the Kahler -- Poisson condition" .
Given a Kahler-Poisson algebra K = (A, g, {x1, ..., xm}), let g denote the A-
module generated by all inner derivations , i.e.
g = {a1{c1, ·} + ... + aN {cN , ·} : ai, ci ∈ A and N ∈ N}.
It is a standard fact that g is a Lie algebra with respect to the bracket
[α, β](a) = α(β(a)) − β(α(a)),
where α, β ∈ g and a ∈ A (see e.g. [8]).
Moreover, it was shown in [3] that g is a projective module and that every
Kahler -- Poisson algebra is a Lie-Rinehart algebra. For more details on Lie-Rinehart
algebras, we refer to [9, 14]. It was also proven that the matrix g defines a metric
(in the context of metric Lie-Rinehart algebras [3]) on g via
g(α, β) =
α(xi)gij β(xj ),
m
Pi,j=1
for α, β ∈ g. Denoting by P the matrix with entries P ij = {xi, xj}, the Kahler-
Poisson condition (2.1) can be written in matrix notation as
if the algebra A is generated by {x1, ..., xm}. Let us now recall the concept of a
morphism of Kahler-Poisson algebras [4, 5], starting from the following definition.
ηPgPgP = −P,
ON MODULI SPACES OF K AHLER-POISSON ALGEBRAS OVER RATIONAL FUNCTIONS IN TWO VARIABLES3
Definition 2.2. For a Kahler-Poisson algebra K = (A, g, {x1, ..., xm}), let Afin ⊆ A
denote the subalgebra generated by {x1, ..., xm}.
Clearly, if A is generated by {x1, . . . , xm}, which is often the case in particular
examples, then Afin = A. Note that Afin is not necessarily a Poisson subalgebra of
A in the general case.
Definition 2.3. Let K = (A, g, {x1, ..., xm}) and K′ = (A′, g ′, {y1, ..., ym ′}) be
Kahler-Poisson algebras together with their modules of inner derivations g and
′, respectively. A morphism of Kahler-Poisson algebras is a pair of maps (φ, ψ),
g
with φ : A → A′ a Poisson algebra homomorphism and ψ : g → g
′ a Lie algebra
homomorphism, such that
(1) ψ(aα) = φ(a)ψ(α),
(2) φ(α(a)) = ψ(α)(φ(a)),
(3) φ(g(α, β)) = g ′(ψ(α), ψ(β)),
(4) φ(Afin) ⊆ A′
fin,
for all a ∈ A and α, β ∈ g.
Let us also recall the following result from [4, 5], where a condition for two Kahler-
Poisson algebras to be isomorphic is formulated. In this paper we shall repeatedly
make use of this result to understand when two Kahler-Poisson algebras are iso-
morphic for different choices of metrics.
Proposition 2.4. Let K = (A, g, {x1, ..., xm}) and K′ = (A′, g ′, {y1, ..., ym ′}) be
Kahler-Poisson algebras. Then K and K′ are isomorphic if and only if there exists
a Poisson algebra isomorphism φ : A → A′ such that φ(Afin) = A′
fin, and
(2.2)
P ′g ′P ′ = P ′AT φ(g)AP ′,
where Ai
α = ∂φ(xi)
∂yα and (P ′)αβ = {yα, yβ}′.
In what follows, the matrix P ′ will be invertible, implying that (2.2) is equivalent
to g ′ = AT φ(g)A. We shall also need the following result [4, 5].
Proposition 2.5. Let K = (A, g, {x1, ..., xm}) and K′ = (A′, g ′, {y1, ..., ym ′}) be
Kahler-Poisson algebras and let (φ, ψ) : K → K′ be an isomorphism of Kahler-
Poisson algebras. If
ηPgPgP = −P and η′P ′g ′P ′g ′P ′ = −P ′
then (φ(η) − η′)P ′ = 0.
Note that, in the current situation, Proposition 2.5 implies that φ(η) = η′ since P ′
is invertible.
3. Kahler-poisson algebras over rational functions
Let us start by considering C[x, y] together with a (non-zero) linear Poisson struc-
ture; i.e a Poisson bracket determined by
{x, y} = λx + µy
for λ, µ ∈ C such that at least one of λ, µ is non-zero (note that the Jacobi identity
is satisfied for all choices of λ and µ). The corresponding Poisson algebras are
isomorphic for different choices of λ and µ, and for definiteness we shall choose a
particular presentation.
4
AHMED AL-SHUJARY
Proposition 3.1. Let A1 = (C[x, y], {·, ·}1) denote the Poisson algebra defined by
{x, y}1 = λx + µy for λ, µ ∈ C such that {x, y}1 6= 0. Then A1 is isomorphic to
the Poisson algebra A = (C[x, y], {·, ·}) with {x, y} = x.
Proof. We will show that for every choice of λ, µ ∈ C (with at least one of them
being non-zero), there exists a Poisson algebra automorphism φ : A1 → A of the
form
φ(x) = ax + by
φ(y) = cx + dy
with a, b, c, d ∈ C and ad − bc 6= 0. Thus, one needs to prove that
(3.1)
{φ(x), φ(y)} = φ({x, y}1)
for every allowed choice of λ, µ ∈ C. Starting from the left hand side we get
{φ(x), φ(y)} = {ax + by, cx + dy} = ad{x, y} + bc{y, x}
= ad{x, y} − bc{x, y} = (ad − bc)x.
From the right hand side, we get
φ({x, y}1) = φ(λx + µy) = λφ(x) + µφ(y) = λ(ax + by) + µ(cx + dy)
= λax + λby + µcx + µdy
That is, (3.1) is equivalent to
(3.2)
(3.3)
λa + µc = ad − bc
λb + µd = 0
If λ = 0 then µ 6= 0, since {x, y}1 6= 0. From (3.3) one gets that d = 0 and inserting
this into (3.2) one obtains µc = bc implying that b = −µ (note that c 6= 0, since
ad − bc 6= 0). Hence, choosing e.g. c = 1 and a = 0, gives φ(x) = −µy and φ(y) = x
defining a Poisson algebra isomorphism.
If λ 6= 0 then from (3.3) one gets b = −µd
λ
obtains
and inserting this into (3.2) one
λa + µc = ad +
µcd
λ
⇔ a(λ − d) =
µc
λ
(d − λ),
and choosing d = λ we get b = −µ (note that a 6= 0 since ad − bc 6= 0). Hence,
choosing e.g. a = 1 and c = 0 gives φ(x) = x − µy and φ(y) = λy defining a Poisson
algebra isomorphism. Thus, we have shown that for every choice of λ, µ ∈ C such
that λx+µy 6= 0, one can construct a Poisson algebra isomorphism φ : A1 → A. (cid:3)
Now, let C(x, y) denote the rational functions in x, y and let C(x) denote the rational
functions in x. Any Poisson structure on C[x, y] extends to a Poisson structure on
C(x, y) via Leibniz rule
{p, q−1} = −{p, q}q−2
for p, q ∈ C[x, y]. Thus, in the following, we let A(x, y) denote the Poisson algebra
(C(x, y), {·, ·}) with {x, y} = x. Given the Poisson algebra A(x, y) we set out to
study the possible Kahler-Poisson algebra structures arising from A(x, y); that is,
finding gij such that (A(x, y), g, {x, y}) is a Kahler-Poisson algebra.
It is easy to check that for an arbitrary symmetric matrix g one obtains
PgPgP = −{x, y}2 det(g)P = −x2 det(g)P
ON MODULI SPACES OF K AHLER-POISSON ALGEBRAS OVER RATIONAL FUNCTIONS IN TWO VARIABLES5
giving η = (x2 det(g))−1, implying that (A(x, y), g, {x, y}) is a Kahler-Poisson alge-
bra as long as det(g) 6= 0. Hence, any non-degenerate (2 × 2)-matrix g, with entries
in A(x, y), gives rise to a Kahler-Poisson algebra over A(x, y).
Next, let us recall that all automorphisms of C[x, y] (see [10, 12]) are given by
compositions of
φ(x) = α1x + β1y + γ1 and φ(y) = α2x + β2y + γ2
for α1, β1, γ1, α2, β2, γ2 ∈ C, with α1β2 6= α2β1 and
φ(x) = x and φ(y) = y + p(x)
for all p(x) ∈ C[x].
morphisms, we need to check which ones that are Poisson algebra morphisms.
In order to use these to construct Kahler-Poisson algebra
Lemma 3.2. Let A(x, y) = C(x, y) be the rational functions in x, y with a Poisson
structure given by {x, y} = x. Then:
(A) φ(x) = α1x + β1y + γ1 and φ(y) = α2x + β2y + γ2, for α1, β1, γ1, α2, β2, γ2 ∈ C
with α1β2 6= α2β1, is a Poisson algebra automorphism of A(x, y) if β1 = γ1 = 0
and β2 = 1, giving φ(x) = α1x and φ(y) = α2x + y + γ2.
(B) φ(x) = αx and φ(y) = y + p(x) is a Poisson algebra automorphism for all
p(x) ∈ C[x] and α ∈ C\{0}.
Proof. (A) For φ to be a Poisson algebra automorphism we need to check that
φ({x, y}) = {φ(x), φ(y)} and since {x, y} = x this is equivalent to φ(x) = {φ(x), φ(y)}.
We start from
{φ(x), φ(y)} = {α1x + β1y + γ1, α2x + β2y + γ2}
= α1α2{x, x} + α1β2{x, y} + β1β2{y, y} + β1α2{y, x}
= α1β2{x, y} + β1α2{y, x} = (α1β2 − β1α2)x.
Now, φ(x) = {φ(x), φ(y)} gives α1x + β1y + γ1 = (α1β2 − β1α2)x. and one obtains
α1 = α1β2 − β1α2,
β1 = 0,
γ1 = 0,
implying that β2 = 1 since α1 6= 0 (by the assumption α1β2 − α2β1 6= 0). Hence,
we get φ(x) = α1x and φ(y) = α2x + y + γ2.
(B) For φ to be a Poisson algebra automorphism we need to check that φ({x, y}) =
{φ(x), φ(y)} and since {x, y} = x we show that φ(x) = {φ(x), φ(y)}. We start from
the right side
{φ(x), φ(y)} = {αx, y + p(x)} = α{x, y} + α{x, p(x)} = αx = φ(x)
since {x, p(x)} = 0 for arbitrary p(x). Hence, φ is a Poisson algebra automorphism
for an arbitrary p(x) ∈ C[x] and α ∈ C\{0}.
(cid:3)
Next, let us show that compositions of Poisson algebra automorphisms in Lemma 3.2
may be written in a simple form.
Proposition 3.3. Let φ = φ1 ◦ φ2 ◦ ... ◦ φn be an arbitrary composition of automor-
phisms, where each φk can be written as: either φ(x) = α1x, φ(y) = α2x + y + γ2
or φ(x) = αx, φ(y) = y + p(x). Then there exists α ∈ C and p(x) ∈ C[x] such that
φ(x) = αx and φ(y) = y + p(x).
Proof. We show that every composition of all Poisson algebra automorphisms in
Lemma 3.2 can be written with the form φ(x) = αx and φ(y) = y + p(x). Clearly,
both type (A) and (B) Poisson algebra automorphisms in Lemma 3.2 can be written
6
AHMED AL-SHUJARY
in this form. Thus, let φ1(x) = α1x, φ1(y) = y + p1(x), φ2(x) = α2x and φ2(y) =
y + p2(x). Then
(1) φ1(φ2(x)) = φ1(α2x) = α2φ1(x) = α2α1x = αx.
(2) φ1(φ2(y)) = φ1(y + p2(x)) = y + p1(x) + p2(α1x) = y + p(x).
which are again of the form φ(x) = αx and φ(y) = y + p(x), implying that an
arbitrary composition of automorphisms is of the same form.
(cid:3)
In order to prove results related to arbitrary automorphisms of A(x, y), we will
need to consider the case when φ(x) ∈ C(x). In this case, the possible types of
automorphisms can be explicitly described.
Proposition 3.4. Let φ : A(x, y) → A(x, y) be a Poisson algebra automorphism
such that φ(x) ∈ C(x). Then there exist α, β, γ, δ ∈ C and r(x) ∈ C(x) such that
αδ − βγ 6= 0 and
φ(x) =
φ(y) =
αx + β
γx + δ
(αx + β)(γx + δ)y
(αδ − βγ)x
+ r(x).
Proof. If φ is invertible then, since φ(x) ∈ C(x), φ(x) has to be an invertible rational
function of x. It is well known that such a function is of the form φ(x) = αx+β
γx+δ for
some α, β, γ, δ ∈ C with αδ − βγ 6= 0. Since φ is a Poisson algebra automorphism,
one can determine the possible φ(y) via
We start from the left hand side
{φ(x), φ(y)} = φ({x, y}).
{φ(x), φ(y)} =n αx + β
=(cid:16) αδ − βγ
γx + δ
(γx + δ)2(cid:17)xφ(y)′
, φ(y)o =n αx + β
γx + δ
and from the right hand side we get
y,
, yoφ(y)′
y =(cid:16) αx + β
γx + δ(cid:17)′
x
{x, y}φ(y)′
y
φ({x, y}) = φ(x) =
αx + β
γx + δ
.
Therefore, we obtain
αδ − βγ
γx + δ
xφ(y)′
y = αx + β ⇒
(αδ − βγ)xφ(y)′
y = (αx + β)(γx + δ) ⇒
φ(y) =
(αx + β)(γx + δ)y
(αδ − βγ)x
+ r(x).
for some r(x) ∈ C(x).
(cid:3)
Let us now start to investigate isomorphism classes of metrics for Kahler-Poisson
algebras over A(x, y). The simplest case is when the metrics are constant; i.e
g =(cid:18)a b
c(cid:19)
b
and
g =(cid:18)a b
c(cid:19)
b
for a, b, c, a, b, c ∈ C.
ON MODULI SPACES OF K AHLER-POISSON ALGEBRAS OVER RATIONAL FUNCTIONS IN TWO VARIABLES7
Proposition 3.5. Let K = (A(x, y), g, {x, y}) and K = (A(x, y), g, {x, y}) be
Kahler-Poisson algebras, with
g =(cid:18)a b
c(cid:19) and g =(cid:18)a b
c(cid:19)
b
b
where a, b, c, a, b, c ∈ C such that det(g) 6= 0 and det(g) 6= 0. Then K ∼= K if and
only if c = c.
Proof. First, we assume that c = c and show that K ∼= K by using an automorphism
of the form φ(x) = αx and φ(y) = y + p(x) (cf. Lemma 3.2). We will do this by
applying Proposition 2.4 to show that there exists α ∈ C and p(x) ∈ C[x] such that
(3.4)
g = AT φ(g)A.
First, note that φ(g) = g, since φ is unital. From Ai
∂yα one computes
A = ∂φ(x)
∂φ(y)
∂x
∂x
giving (3.4) as
α = ∂φ(xi)
1(cid:19),
0
∂φ(x)
∂y
∂y ! =(cid:18) α
p′(x)
∂φ(y)
(cid:18)a b
b
0
1 (cid:19)(cid:18)a b
c(cid:19) =(cid:18)α p′(x)
=(cid:18)α(αa + p′(x)b) + p′(x)(αb + p′(x)c) αb + p′(x)c
1(cid:19) =(cid:18)αa + p′(x)b αb + p′(x)c
c(cid:19)(cid:18) α
αb + P ′(x)c
(cid:19) ,
p′(x)
0
c
c
b
b
(cid:19)(cid:18) α
p′(x)
0
1(cid:19)
showing that (3.4) is equivalent to
(3.5)
(3.6)
a = α2a + 2αp′(x)b + (p′(x))2c
b = αb + p′(x)c
First, assume that c 6= 0. From (3.6) we get p′(x) =
and inserting p′(x) in (3.5) we obtain
b−αb
c
giving p(x) = (
b−αb
c
)x,
a = α2a + 2α(cid:16)
b − αb
c
(cid:17)b +(cid:16)
Multiplying both sides by c2, we get
(cid:17)2
b − αb
c
c = α2a +
2αbb − 2α2b2
c
+
b2c − 2αbbc + α2b2c
ac2 = α2ac2 + 2αbbc − 2α2b2c + b2c − 2αbbc + α2b2c ⇒ α2 =
where ac − b2 = det(g) 6= 0 by assumption. Hence, for c = c 6= 0, we have
constructed an isomorphism between K and K. If c = c = 0 we get from (3.6) that
b
α =
b
b
b , where b 6= 0 since det(g) 6= 0. Now, we find p′(x) from (3.5) by using α =
a = a(cid:16)
b
b(cid:17)2
b
+ 2(cid:16)
b(cid:17)p′(x)b +(cid:0)p′(x)(cid:1)2
c ⇒ p′(x) =
ab2 − ab2
2b2b
(note that b 6= 0 since det(g) 6= 0), giving p(x) = ( ab2
this gives an isomorphism between K and K. We conclude that K ∼= K if c = c.
)x. Hence, for c = c = 0,
−ab2
2b2b
Vice versa, assume that K ∼= K. We have
η = {x, y}2 det(g) = x2 det(g) and η = {x, y}2 det(g) = x2 det(g)
c2
ac − b2
ac − b2
,
8
AHMED AL-SHUJARY
with det(g), det(g) ∈ C. By using Proposition 2.5, stating that φ(η) = η, one
obtains
φ(x2) det(g) = x2 det(g) ⇒
φ(x2)
x2 =
det g
det(g)
∈ C
implying that φ(x) = αx for some α ∈ C. Furthermore, using Proposition 3.4 with
β = 0, γ = 0, δ = 1 one obtains φ(y) = y + r(x). Hence, any isomorphism have to
be of the form φ(x) = αx and φ(y) = y + r(x), for some α ∈ C and r(x) ∈ C(x).
Using the above form of the isomorphism in Proposition 2.4 one gets
(cid:18)a b
b
0
1 (cid:19)(cid:18)a b
c(cid:19) =(cid:18)α r′(x)
=(cid:18)α(αa + r′(x)b) + r′(x)(αb + r′(x)c) αb + r′(x)c
c(cid:19)(cid:18) α
αb + r′(x)c
1(cid:19)
r′(x)
0
c
b
giving c = c.
(cid:19) ,
(cid:3)
The above result shows that the isomorphism classes of Kahler-Poisson algebras
with constant metrics can be parametrized by one (complex) parameter. In the
next result, we study the case when the metric only depends on x, and start by
giving sufficient conditions for the Kahler-Poisson algebras to be isomorphic.
Proposition 3.6. Let K = (A(x, y), g, {x, y}) and K = (A(x, y), g, {x, y}) be
Kahler-Poisson algebras, with
g =(cid:18)a(x)
b(x)
b(x)
c(x)(cid:19) and g =(cid:18)a(x) b(x)
c(x)(cid:19)
b(x)
where a(x), b(x), c(x), a(x), b(x), c(x) ∈ C[x] such that det(g) 6= 0 and det(g) 6= 0.
If c(x) 6= 0 and there exists α ∈ C such that:
(1) (cid:0)a(x) − α2a(αx)(cid:1)c(αx) = b(x)2 − α2b(αx)2
b(x)−αb(αx)
∈ C[x]
(2)
(3) c(x) = c(αx)
c(αx)
then K ∼= K. If c(x) = c(x) = 0 and there exists α ∈ C such that:
(a) b(x) = αb(αx)
(b) a(x)−α2a(αx)
∈ C[x]
2αb(αx)
then K ∼= K.
Proof. Let φ be an automorphism of φ(x) = αx and φ(y) = y + p(x). We will show
that one may find α ∈ C and p(x) ∈ C[x] such that
(3.7)
g = AT φ(g)A,
implying, via Proposition 2.4, that K ∼= K. From Ai
α = ∂φ(xi)
∂yα one computes
A = ∂φ(x)
∂φ(y)
∂x
∂x
∂φ(x)
∂y
∂y ! =(cid:18) α
p′(x)
∂φ(y)
0
1(cid:19),
ON MODULI SPACES OF K AHLER-POISSON ALGEBRAS OVER RATIONAL FUNCTIONS IN TWO VARIABLES9
giving (3.7) as
(cid:18)a(x) b(x)
c(x)(cid:19) =(cid:18)α p′(x)
1 (cid:19)(cid:18)a(αx)
b(αx)
b(x)
0
b(αx)
c(αx)(cid:19)(cid:18) α
p′(x)
0
1(cid:19)
=(cid:18)α2a(αx) + 2αp′(x)b(αx) + p′(x)2c(αx) αb(αx) + p′(x)c(αx)
αb(αx) + p′(x)c(αx)
c(αx)
(cid:19) .
Hence, (3.7) is equivalent to
(3.8)
(3.9)
(3.10)
a(x) = α2a(αx) + 2αp′(x)b(αx) + p′(x)2c(αx)
b(x) = αb(αx) + p′(x)c(αx)
c(x) = c(αx)
First, assume that c(x) 6= 0 together with the assumptions (1) -- (3). From (3.9) we
get
p′(x) =
b(x) − αb(αx)
c(αx)
.
and, by assumption, this is in C[x] which implies that one may integrate to get
p(x). Inserting p′(x) in (3.8) we obtain
a(x)c(αx) = α2a(αx)c(αx) + 2αb(αx)p′(x)c(αx) + p′(x)2c(αx)2
= α2a(αx)c(αx) + 2αb(αx)(cid:0)b(x) − αb(αx)(cid:1) +(cid:0)b(x) − αb(αx)(cid:1)2
that is, (cid:0)a(x) − α2a(αx)(cid:1)c(αx) = b(x)2 − α2b(αx)2 which is true by assumption.
If c(x) = c(x) = 0, we assume that a(x)−α2a(αx)
= α2a(αx)c(αx) − α2b(αx)2 + b(x)2
∈ C[x] and b(x) = αb(αx).
2αb(αx)
Then (3.9) is immediately satisfied and from (3.8) we get p′(x) = a(x)−α2a(αx)
. By
assumption this is in C[x] which implies that one may integrate to get p(x). This
shows that one may explicitly construct an isomorphism between K and K, given
the assumptions in the statement.
(cid:3)
2αb(αx)
For the sake of illustration, let us use the above result to give a simple example,
and construct two seemingly different metrics that give rise to isomorphic Kahler-
Poisson algebras.
Example 3.7. Let K = (A(x, y), g, {x, y}) and K = (A(x, y), g, {x, y}) be Kahler-
Poisson algebras, with
g =(cid:18)a(x)
0
0
c(x)(cid:19) and g =(cid:18)a(x) + q(x)2c(x)
q(x)c(x)
q(x)c(x)
c(x) (cid:19)
for a(x), c(x), q(x) ∈ C[x]. We will use Proposition 3.6 to show that K ∼= K. Let us
check conditions (1) -- (3) with α = 1, b(x) = 0, c(x) = c(x), b(x) = q(x)c(x) and
a(x) = a(x) + q(x)2c(x).
(1) (cid:0)a(x) − α2a(αx)(cid:1)c(αx) = b(x)2 − α2b(αx)2:
(cid:0)a(x) − α2a(αx)(cid:1)c(αx) = q(x)2c(x)2
b(x)2 − α2b(αx)2 = q(x)2c(x)2,
10
AHMED AL-SHUJARY
(2)
b(x) − αb(αx)
c(αx)
= q(x) ∈ C[x],
(3) c(x) = c(αx) since α = 1 and c(x) = c(x).
For instance, with a(x) = x, c(x) = x2 and q(x) = x3 one concludes that
0
g =(cid:18)x
0 x2(cid:19) and (cid:18)x + x8 x5
x2(cid:19)
x5
define isomorphic Kahler-Poisson algebras.
The following result shows that, in the more restricted situation where the metrics
are assumed to be diagonal, one describe all isomorphism classes.
Proposition 3.8. Let K = (A(x, y), g, {x, y}) and K = (A(x, y), g, {x, y}) be
Kahler-Poisson algebras, with
g =(cid:18)a(x)
0
0
c(x)(cid:19) and g =(cid:18)a(x)
0
0
c(x)(cid:19)
where a(x), c(x), a(x), c(x) ∈ C[x] such that det(g) 6= 0 and det(g) 6= 0. Then K ∼= K
if and only if there exists α ∈ C such that:
(1) c(x) = c(αx)
(2) a(x) = α2a(αx)
Proof. To show that K ∼= K if (1) and (2) are satisfied, we use Proposition 3.6.
Namley,
for b = 0 becomes
(cid:0)a(x) − α2a(αx)(cid:1)c(αx) = b(x)2 − α2b(αx)2
which is satisfied since a(x) = α2a(αx). (Condition (2) in Proposition 3.6 is trivially
satisfied since b = b = 0.)
(cid:0)a(x) − α2a(αx)(cid:1)c(αx) = 0
Vice versa, assume that K ∼= K. We have
η = {x, y}2 det(g) = x2 det(g) and η = {x, y}2 det(g) = x2 det(g)
with det(g), det(g) ∈ C[x]. By using Proposition 2.5, which gives that φ(η) = η, it
follows that
φ(x2) det(g) = x2 det(g)
implying that φ(x) ∈ C(x). By Proposition 3.4, if φ(x) ∈ C(x), then the automor-
phism has to be of the form
φ(x) = αx+β
γx+δ and φ(y) = (αx+β)(γx+δ)y
(αδ−βγ)x + r(x)
where, α, β, γ, δ ∈ C, r(x) ∈ C(x) and αδ − βγ 6= 0. Moreover, by Proposition 2.4,
one also has
(3.11)
with
g = AT φ(g)A.
A =(cid:18)φ(x)′
φ(y)′
x φ(x)′
y
x φ(y)′
y(cid:19) =(cid:18)φ(x)′
x
x φ(y)′
φ(y)′
y(cid:19)
0
ON MODULI SPACES OF K AHLER-POISSON ALGEBRAS OVER RATIONAL FUNCTIONS IN TWO VARIABLES11
since φ(x)′
y = 0, giving
(cid:18)a(x)
0
0
0
c(x)(cid:19) =(cid:18)φ(x)′
x φ(y)′
x
φ(y)′
y(cid:19)(cid:18)φ(a(x))
x(cid:1)2
= φ(a(x))(cid:0)φ(x)′
φ(c(x))φ(y)′
0
x(cid:1)2
+ φ(c(x))(cid:0)φ(y)′
xφ(y)′
y
0
φ(c(x))(cid:19)(cid:18)φ(x)′
x
φ(y)′
x φ(y)′
0
y(cid:19)
y(cid:1)2 ! .
φ(c(x))φ(y)′
xφ(y)′
y
φ(c(x))(cid:0)φ(y)′
Hence, (3.11) is equivalent to
(3.12)
(3.13)
(3.14)
a(x) = φ(a(x))(φ(x)′
c(x) = φ(c(x))(φ(y)′
φ(c(x))φ(y)′
xφ(y)′
x)2
x )2 + φ(c(x))(φ(y)′
y )2
y = 0.
Now, φ(c(x)) 6= 0 (since det(g) 6= 0) and φ(y)′
implies, by (3.14), that
y 6= 0 (since αδ − βγ 6= 0) which
φ(y)′
(αδ−βγ)x y(cid:17)′
x =(cid:16) (αx+β)(γx+δ)
x
+ r′(x) = 0.
It follows that r(x) = r0 ∈ C and
(cid:16) (αx + β)(γx + δ)
(αδ − βγ)x
y(cid:17)′
x
= 0 ⇒
(αx + β)(γx + δ)
(αδ − βγ)x
= λ ∈ C,
yielding
(αx + β)(γx + δ) = λx(αδ − βγ)
and consequently
αγ = 0, βδ = 0 and αδ + βγ = λ(αδ − βγ).
If α = 0, then β 6= 0, γ 6= 0 (since αδ − βγ 6= 0) implying that δ = 0 and therefore,
βγ = −λβγ which implies that λ = −1. Hence, the automorphism have to be of
the form φ(x) = β
γx and φ(y) = −y + r0. Using Proposition 2.4 we get
(cid:18)a(x)
0
0
0
0
0
γx
0
−1(cid:19)(cid:18)φ(a(x))
c(x)(cid:19) =(cid:18)− β
=(cid:18)− β
−φ(c(x))(cid:19)(cid:18)− β
γx(cid:17)2
= φ(a(x))(cid:16)− β
φ(c(x))!
φ(c(x))(cid:19)(cid:18)− β
−1(cid:19)
γx φ(a(x))
γx
0
γx
0
0
0
0
0
0
0
−1(cid:19)
giving that
a(x) = φ(a(x))(cid:18)−
β
γx(cid:19)2
= a(φ(x))(cid:18)−
β
γx(cid:19)2
=(cid:18)−
β
γx(cid:19)2
γx(cid:19)
a(cid:18) β
However, the above form of the automorphism is not possible since we have assumed
that a(x), a(x) are non-zero polynomials (and not rational functions).
12
AHMED AL-SHUJARY
If α 6= 0, then γ = 0, δ 6= 0 implying that β = 0 and αδ = λαδ, which implies
δ = αx and
that λ = 1. Hence, the automorphism have to be of the form φ(x) = αx
φ(y) = λy = y + r0. Using Proposition 2.4 we get
(cid:18)a(x)
0
giving that
0
c(x)(cid:19) =(cid:18)α
0
0 −1(cid:19)(cid:18)φ(a(x))
φ(c(x))(cid:19)
=(cid:18)α2φ(a(x))
0
0
0
0
φ(c(x))(cid:19)(cid:18) α
0 −1(cid:19)
0
a(x) = α2φ(a(x)) = α2a(φ(x)) = α2a(αx)
c(x) = φ(c(x)) = c(φ(x)) = c(αx)
which concludes the proof of the statement.
(cid:3)
Let us give another simple example of isomorphic Kahler-Poisson algebras.
Example 3.9. Let K = (A(x, y), g, {x, y}) and K = (A(x, y), g, {x, y}) be Kahler-
Poisson algebras, with
g =(cid:18)a(x)
0
0
c(x)(cid:19) and g =(cid:18)α2a(αx)
0
0
c(αx)(cid:19)
for a(x), c(x) ∈ C[x] and α ∈ C. Proposition 3.8 shows that K ∼= K. For instance,
with a(x) = x, c(x) = 1 + x + x2 and α = −2 one finds that
g =(cid:18)x
0
0
1 + x + x2(cid:19) and g =(cid:18)−8x
0
0
1 − 2x + 4x2(cid:19)
give isomorphic Kahler-Poisson algebras.
For general metrics, the situation becomes much more complicated. However, let
us finish by giving a sufficient condition for diagonal metrics depending on y.
Proposition 3.10. Let K = (A(x, y), g, {x, y}) and K = (A(x, y), g, {x, y}) be
Kahler-Poisson algebras, with
g(y) =(cid:18)a(y)
0
0
c(y)(cid:19) and g(y) =(cid:18)a(y)
0
0
c(y)(cid:19)
for a(y), c(y), a(y), c(y) ∈ C[x]. If there exists α, λ ∈ C such that
(1) a(y) = α2a(y + λ)
(2) c(y) = c(y + λ)
then K ∼= K.
Proof. Assume a(y) = α2a(y + λ) and c(y) = c(y + λ). To prove that K ∼= K, we
will use Proposition 2.4 and show that
(3.15)
g = AT φ(g)A,
for an automorphism of the type φ(x) = αx and φ(y) = y + p(x). Let p(x) = λ,
then p′(x) = 0 and we compute
A =(cid:18) α
p′(x)
0
1(cid:19) =(cid:18)α 0
1(cid:19)
0
ON MODULI SPACES OF K AHLER-POISSON ALGEBRAS OVER RATIONAL FUNCTIONS IN TWO VARIABLES13
giving (3.15) as
(cid:18)a(y)
0
0
c(y)(cid:19) =(cid:18)α 0
1(cid:19)(cid:18)φ(a(y))
0
0
0
φ(c(y))(cid:19)(cid:18)α 0
1(cid:19)
0
=(cid:18)αa(y + λ)
0
0
c(y + λ)(cid:19)(cid:18)α 0
1(cid:19) =(cid:18)α2a(y + λ)
0
0
0
c(y + λ)(cid:19)
which is true by assumption. From Proposition 2.4 we conclude that K ∼= K.
For instance, with a(y) = y, c(y) = 1 + y2 , λ = 2 and α = 1 finds that
(cid:3)
0
give rise to isomorphic Kahler-Poisson algebras.
1 + y2(cid:19) and g =(cid:18)y + 2
g =(cid:18)y
0
0
0
5 + y2 + 4y(cid:19)
4. Summary
In this paper, we have started to investigate isomorphism classes of Kahler-Poisson
algebras for rational functions in two variables. Although a complete classifica-
tion has not been obtained there are several subclasses of metrics which can be
explicitly described. Proposition 3.5 shows that the class of constant metrics can
be parametrized by one (complex) parameter, and Proposition 3.8 describes the
isomorphism classes of diagonal metrics only depending on x. Furthermore, several
sufficient conditions are derived (Proposition 3.6 and Proposition 3.10) and a num-
ber of examples are given illustrating that seemingly different metrics may give rise
to isomorphic Kahler-Poisson algebras.
References
[1] J. Arnlind and G. Huisken. Pseudo-Riemannian geometry in terms of multi-linear brackets.
Lett. Math. Phys., 104(12):1507-1521, 2014.
[2] J. Arnlind, J. Hoppe, and G. Huisken. Multi-linear formulation of differential geometry and
matrix regularizations. J. Differential Geom., 91(1):1-39, 2012.
[3] J. Arnlind. A. Al-Shujary. Kahler-Poisson Algebras. J. Geometry and Physics., 136:156-172,
2019.
[4] A. Al-Shujary. Kahler-Poisson Algebras. Linkping Studies in Science and Technology. No.1813,
2018.
[5] A. Al-Shujary. Morphisms, direct sums and tensor products of Kahler-Poisson Algebras.
arXiv:1906.04519.
[6] A. G. Czerniakiewicz, Automorphisms of a free associative algebra of rank 2, I, II, Trans.
Amer. Math. Sec. 160 (1971), 393-401; 171 (1972), 309-315.
[7] W. Dicks, Automorphisms of the polynomial ring in two variables, Publ. Sec. Mat. Univ.
Autonoma Barcelona 27 (1983) 155-162.
[8] S. Helgason. Differential geometry, Lie groups, and symmetric spaces, volume 34 of Graduate
Studies in Mathematics. American Mathematical Society, Providence, RI, 2001.
[9] J. Huebschmann. Extensions of Lie-Rinehart algebras and the Chern-Weil construction. In
Higher homotopy structures in topology and mathematical physics, volume 227 of Contemp.
Math., Amer. Math. Soc., Providence, RI, pages 145-176, 1999.
[10] H.W.E. Jung, Uber ganze birationale Transformationen der Ebene, J. Reine Angew. Math.
(184), pp. 161-174. 1942.
[11] L. Makar-Limanov, The automorphisms of the free algebra of two generators, Funksional.
Anal. i Prilozhen. 4 (1970), no. 3, 107-108.
[12] J. H. McKay and S. S.-S. Wang, Stuart. An elementary proof of the automorphism theorem
for the polynomial ring in two variables. Journal of Pure and Applied Algebra - J PURE APPL
ALG. 52. 91-102.(1988).
[13] M. Nagata, On automorphism group of k[x, y], Lectures in Mathematics 5 (Kinokuniya Book
Store, Tokyo, 1972).
14
AHMED AL-SHUJARY
[14] G. S. Rinehart. Differential forms on general commutative algebras. Trans. Amer. Math. Soc.,
108:195-222, 1963.
[15] W. Van der Kulk, On polynomial rings in two variables, Nieuw Arch. Wisk. (3) 1 (1953)
33-41.
(Ahmed Al-Shujary) Dept. of Math., Linkoping University, 581 83 Linkoping, Sweden
E-mail address: [email protected]
|
1504.00373 | 1 | 1504 | 2015-04-01T20:07:19 | The Asymptotic Behavior of the Codimension Sequence of Affine G - Graded Algebras | [
"math.RA"
] | Let W be an affine PI algebra over a field of characteristic zero graded by a finite group G. We show that there exist $\alpha_{1},\alpha_{2}\in\mathbb{R}, \beta\in\frac{1}{2}\mathbb{Z}$, and $l\in\mathbb{N}$ such that $\alpha_{1}n^{\beta}l^{n}\leq c_{n}^{G}(W)\leq\alpha_{2}n^{\beta}l^{n}$. Furthermore, if W has a unit then the asymptotic behavior of $c_{n}^{G}(W)$ is $\alpha n^{\beta}l^{n}$ where $\alpha\in\mathbb{R}, \beta\in\frac{1}{2}\mathbb{Z}, l\in\mathbb{N}$. | math.RA | math |
THE ASYMPTOTIC BEHAVIOR OF THE CODIMENSION
SEQUENCE OF AFFINE G-GRADED ALGEBRAS
YUVAL SHPIGELMAN
Abstract. Let W be an affine PI algebra over a field of characteristic zero graded
by a finite group G. We show that there exist α1, α2 ∈ R, β ∈ 1
Z, and l ∈ N such
that α1nβln ≤ cG
n (W ) ≤ α2nβln. Furthermore, if W has a unit then the asymptotic
behavior of cG
n (W ) is αnβln where α ∈ R, β ∈ 1
Z, l ∈ N.
2
2
Introduction
Throughout this article F is an algebraically closed field of characteristic zero and W
is an affine associative F - algebra graded by a finite group G. We also assume that W
is a PI-algebra; i.e., it satisfies an ordinary polynomial identity. In this article we study
G-graded polynomial identities of W , and in particular the corresponding G-graded
codimension sequence. Let us briefly recall the basic setup. Let X G be a countable
set of variables {xi,g : g ∈ G; i ∈ N} and let F(cid:10)X G(cid:11) be the free algebra on the set X G.
Given a polynomial in F(cid:10)X G(cid:11) we say that it is a G-graded identity of W if it vanishes
an ideal in the free algebra F(cid:10)X G(cid:11) which we denote by IdG(W ). Moreover, IdG(W )
is a G-graded T -ideal, namely, it is closed under G-graded endomorphisms of F(cid:10)X G(cid:11).
Let RG(W ) denote the relatively free algebra F(cid:10)X G(cid:11) /IdG(W ), and C G
upon any admissible evaluation on W . That is, a variable xi,g assumes values only from
the corresponding homogeneous component Wg. The set of all G-graded identities is
n (W ) of the codimension sequence of W by cG
n (W ) denote
n is the Gn · n! dimensional F -space spanned by
the space P G
all permutations of the (multilinear) monomials x1,g1, x2,g2 · · · xn,gn where gi ∈ G. We
also define the n-th coefficient cG
n (W ) =
In [AB2] Aljadeff and Belov showed that for every affine G-graded
dimF
algebra there exists a finite dimensional G-graded algebra with the same ideal of graded
identities, thus with the same codimension sequence. We denote such algebra by AW .
In the last few years several papers have been written which generalize results from
the theory of ordinary polynomial identities to the G -graded case (e.g. see [Sv, AB1,
AB2]). One of these papers was by E.Aljadeff, A.Giambruno and D.La Matina who
showed, in different collaborations (see [GL, AG, AGL]), that, as in the nongraded case,
n (W ) exists, and it is a nonnegative integer called the G - graded exponent
of W , and denoted by expG(W ). In this paper we expand this result and prove that,
limn→∞ n(cid:112)cG
(cid:0)C G
n (W )(cid:1).
n ∩ IdG(A) where P G
n /P G
Key words and phrases. graded algebras, polynomial identities, representation theory, Hilbert series,
codimension.
1
in addition to the "exponential part", there is a "polynomial part" to the asymptotics
of cG
n (W ). More precisely, we prove :
Theorem (A). Let G be a finite group, and W an affine G- graded F -algebra where F
is a field of characteristic 0. Suppose that W satisfies an ordinary polynomial identity.
Then there exist α1, α2 > 0, β ∈ 1
2
Z, and l ∈ N such that
n (W ) ≤ α2nβln.
α1nβln ≤ cG
(cid:16) cn(W )
exp(W )n
(cid:17)
A conclusion of this theorem is that limn→∞ logn
nomial part") exists, and is an integer or a half-integer.
(the power of the "poly-
Furthermore, if W has a unit, we have found the structure of the codimension se-
quence's asymptotics.
Theorem (B). Let G be a finite group, and W an unitary affine G- graded F -algebra
where F is a field of characteristic 0. Suppose that W satisfies an ordinary polynomial
identity. Then there exist β ∈ 1
Z, l ∈ N, and α ∈ R such that
2
n (W ) ∼ αnβln.
cG
Theorems A and B generalize results from the theory of ordinary polynomial iden-
tities.
Indeed, in [BR] Berele and Regev proved the nongraded version of theorems
A and B for PI algebras satisfying a Capelli identity (e.g affine algebras, see [GZ2]).
However, in Theorem A (for ungraded algebras) Berele and Regev made an additional
assumption, namely, that the codimension sequence of the PI algebra is eventually non-
decreasing. Recently, in [GZ3], Giambruno and Zaicev showed that the codimension
sequence of any PI algebra is eventually nondecreasing and therefore the additional
assumption mentioned above can be removed. Here we prove a G-graded version of Gi-
ambruno and Zaicev's result and so the assumption on the monotonicity of the G-greded
codimension sequence is also unnecessary.
We begin our article (Section 1) by mentioning some notations from the represen-
tation theory of GLn(F ) and Sn. Section 2 is dealing with the case of affine (not
necessarily unitary) G - graded algebras (Theorem A). And Section 3 is about the case
of unitary affine G-graded algebras (Theorem B).
1. Preliminaries
In this section we recall some notations and definitions from the representation theory
of GLn(F ) and Sn.
Definition 1.1. A partition is a finite sequence of integers λ = (λ1, ..., λk) such that
λ1 ≥ ··· ≥ λk > 0. The integer k denoted by h(λ), and called the height of λ. The set
of all partitions is denoted by Λ, and the set of all partitions of height less or equal to
k is denoted by Λk .
2
We say that λ is a partition of n ∈ N if(cid:80)k
λ = n.
i=1 λi = n. In this case we write λ (cid:96) n or
It is known (e.g. see [Bru]) that the irreducible Sn representations are indexed by
partitions of n. Denote by Dλ the Sn irreducible representation indexed by λ, and by
χλ the corresponding character. By Maschke's theorem every Sn - representation V is
completely reducible, so we can write V =(cid:76)
(cid:88)
λ(cid:96)n mλDλ, and its character
mλχλ.
χ(V ) =
A GLn(F ) - representation Y = SpF{y1, y2, ...} is called a polynomial if there is a set
of polynomials {fi,j(zs,t)i, j ∈ N} ⊂ F [zs,t1 ≤ s, t ≤ n] such that for every i only finite
number of fi,j(zs,t)'s are nonzero, and the action of GLn(F ) on Y is given by:
λ(cid:96)n
(cid:88)
P · yi =
fi,j(ps,t)yj
for every P = (ps,t) ∈ GLn(F ). We say that Y is q−homogeneous if all the nonzero
fi,j's are homogeneous polynomials of (total) degree q. For α = (α1, ..., αn) ∈ Nn, one
defines the weight space of Y associated to α by
j
Y α = {y ∈ Y diag(z1, ..., zn) · y = zα1
1 ··· zαn
n y} .
It is known that
The series HY (t1, ..., tn) =(cid:80)
Y =
(cid:77)
Y α.
1 ··· tαn
α
α (dimF Y α) tα1
is called the Hilbert (or Poincare)
series of Y . The Hilbert series is known to be symmetric in t1, ..., tn, so we recall
an important basis of the space of symmetric series, namely the Schur functions. For
convenience we use the following combinatorial definition: (Note that although this
definition is not the classical one , it is equivalent to it e.g. see [Bru] )
n
Let λ = (λ1, .., λk) be a partition. The Young diagram associated with λ is the finite
subset of Z×Z defined as Dλ = {(i, j) ∈ Z × Zi = 1, .., k , j = 1, .., λi}. We may regard
Dλ as k arrays of boxes where the top one is of length λ1, the second of length λ2, etc.
For example
D(4,3,3,1) =
A Schur function sλ ∈ Z[t1, ..., tn]Sn is a polynomial such that the coefficient of
1 ··· tan
, and an n's
ta1
in Dλ such that in every row the numbers are non-decreasing, and in any column the
numbers are strictly increasing. Note that sλ is homogenous of degree λ.
n is equal to the number of ways to insert a1 ones, a2 twos, ...
3
Example.
(1) If λ is a partition of height one, i.e λ = (λ1), then the corresponding Schur
function is
s(λ1)(t1, ..., tn) =
(2) If λ = (2, 1) and n = 2 we have
(cid:88)
1 ··· tan
ta1
n .
a1+···+an=λ1
s(2,1)(t1, t2) = t2
1t2 + t1t2
2
since the only two ways to set ones and twos in D(2,1) is
1 1
2
and
1 2
2
.
The next lemma is easily implied from the definition of the Schur functions.
Lemma 1.2.
(1) sλ(t1, ..., tn) = 0 if and only if h(λ) > n.
(2) sλ(t1, ...., tn, 0, ..., 0) = sλ(t1, ..., tn).
As {sλ(t1, ..., tn)}λ∈Λn inform a basis of the space of symmetric series in n variables
there exist integers {mλ}λ∈Λn such that
HY (t1, ..., tn) =
mλsλ(t1, ..., tn)
(cid:88)
λ∈Λn
On the other hand, the group Sn is embed in GLn(F ) via permutation matrices, hence
Y is also an Sn representation. Note that if P is a permutation matrix there is σ ∈ Sn
such that diag(z1, ..., zn) · P = P · diag(zσ(1), ..., zσ(n)). Hence for every y in the weight
space Y (1n) (1n means 1, ..., 1 n times) we have
diag(z1, ..., zn) · P y = P · diag(zσ(1), ..., zσ(n))y = z1 ··· znP y
Denote X G
same coefficients mλ as in the Hilbert series of Y .
Next we recall some terminology and facts about G - graded relatively free algebras.
so P y ∈ Y (1n), and we obtain that Y (1n) is a sub Sn representation of Y . Moreover, by
λ(cid:96)n mλχλ with the
the representation theory of GLn(F ) the Sn character of Y (1n) is(cid:80)
n = {xi,gg ∈ G , 1 ≤ i ≤ n} ⊂ X G, and consider the free algebra F(cid:10)X G
(cid:11)
which is a subalgebra of F(cid:10)X G(cid:11). We equip F(cid:10)X G
(cid:11) with a G-grading by setting the
that I is a G - graded T -ideal of the free algebra F(cid:10)X G
(cid:11), then it is easy to see
to be gi1 ··· gil ∈ G. If we suppose
(cid:11) induces (naturally) a grading on F(cid:104)X G
that the grading on F(cid:10)X G
corresponding T -ideal of G-graded identities IdG(cid:0)F(cid:104)X G
n (cid:105)/I(cid:1) ⊂ F(cid:10)X G(cid:11).
n (cid:105)/I, giving us the
We say that I is PI if IdG(cid:0)F(cid:104)X G
n (cid:105)/I(cid:1) (cid:54)= {0} i.e.F(cid:104)X G
n (cid:105)/I(cid:1) is exactly I.
(cid:11) ∩ IdG(cid:0)F(cid:104)X G
T -ideal F(cid:10)X G
n (cid:105)/I is a PI algebra. Note that the
homogenous degree of a monomial xi1,gi1
n
For every g ∈ G the group GLn(F ) acts linearly on the n-dimensional vector space
SpF{x1,g, ..., xn,g}. Given these actions, we define a polynomial GLn(F ) representation
n
··· xil,gil
n
n
n
4
n
n(cid:105)/J1+J2
.
HF(cid:104)XG
n(cid:105)/J1
+ HF(cid:104)XG
n(cid:105)/J2
= HF(cid:104)XG
n(cid:105)/J1
(cid:11), then
(1) HF(cid:104)XG
(2) If J1 ⊂ J2 then
n(cid:105)/J1∩J2
structure on F(cid:10)X G
n
(cid:11) via P (xi1,g1 ··· xil,gl) = P (xi1,g1)··· P (xil,gl) for every P ∈ GLn(F ).
n (cid:105)/I, giving us the corresponding Hilbert series HF(cid:104)XG
This structure induces (naturally) a structure of GLn(F ) representation on the rela-
tively free algebra F(cid:104)X G
n(cid:105)/I(t1, ..., tn).
We recall from [AB1] (lemma 2.10 and 2.11) the following properties of that kind of
Hilbert series:
Lemma 1.3. Let J1,J2 be G - graded T -ideals of F(cid:10)X G
− HF(cid:104)XG
(cid:11)∩ IdG(W ) is G-graded T -ideal which is PI
be a collection of affine relatively free G-graded algebras of W , and consider the corre-
λ sλ(t1, ..., tn). We can show that
sponding Hilbert series HRG
we need only one set of coefficients {mλ}λ∈Λ to describe HRG
n (W )(t1, ..., tn) for ev-
Indeed, it is easy to see that by definition diag(z1, ..., zn) · xi1,g1 ··· xil,gl =
ery n.
n (W ) := F(cid:10)X G
(cid:8)RG
n (W )(t1, ..., tn) = (cid:80)
zi1 ··· zilxi1,g1 ··· xil,gl, hence the weight space(cid:0)RG
n (W )(cid:1)α is the span of all the mono-
m(W )(cid:1)α for every r < m, where
r (W ) =(cid:76)(cid:0)RG
(cid:88)
mials in RG
i = 1, ..., n. Therefore, we can write RG
the summing is over all the α ∈ Nm such that α = (α1, ..., αr, 0, ..., 0). Thus, we get the
m(W ),
Hilbert series of RG
i.e.
n (W ) with αi appearances of variables from the set {xi,gg ∈ G} for every
r (W ) by putting tr+1 = ··· = tm = 0 in the Hilbert series of RG
for every n ∈ N, and let
In our case recall that IdG
n (W ) = F(cid:104)X G
= HF(cid:104)XG
n(cid:105)/J2
(cid:88)
+ HJ2/J1.
n (cid:105)/IdG
n (W )
n∈N
λ m(n)
(cid:9)
n
m(r)
λ sλ(t1, ..., tr) =
m(m)
λ sλ(t1, ..., tr, 0, ...0).
HRG
r (W )(t1, ..., tr) =
λ∈Λr
Moreover, by Lemma 1.2 we get
m(r)
λ sλ(t1, ..., tr) =
m(m)
λ sλ(t1, ..., tr) +
m(m)
λ
· 0,
(cid:88)
λ∈Λr
λ∈Λm
(cid:88)
λ∈Λm\Λr
(cid:88)
λ∈Λr
hence for every r < m the coefficients m(r)
Making it possible to denote m(n)
λ = mλ, and write for every n ∈ N
λ and m(m)
λ
are the same for every λ ∈ Λr.
HRG
n (W )(t1, ..., tn) =
mλsλ(t1, ..., tn).
(cid:88)
λ∈Λn
Remember that AW is a finite dimensional G-graded algebra such that IdG(AW ) =
IdG(W ), thus with the same mλ. By [Go] (Lemma 1 and Lemma 7) mλ = 0 if
λ /∈ Λk where k = dimF (AW ), so all the nonzero coefficients mλ appear in the series
n (W ),
HRG
λ(cid:96)n mλdλ where dλ = dimF Dλ. The conclusion is that the G - graded
thus cG
k (W )(t1, ..., tk). Notice that for every n the weight space RG
n (W ) =(cid:80)
n (W )(1n) is exactly C G
5
codimension sequence of W is determined by a single Hilbert series HRG
We denote this series by HW (t1, ..., tk).
k (W )(t1, ..., tk).
As mention in the introduction, Berele and Regev proved versions of Theorems A
and B for (nongraded) affine PI algebras. Their proofs rely on a result about symmetric
series which we will also use. Before stating this result we need a definition.
Definition 1.4. A series H (t1, ..., tk) is called rational function if we can write it as
g(t1,...,tk) where f, g are polynomials, moreover a rational function f (t1,...,tk)
g(t1,...,tk) is called nice
f (t1,...,tk)
k ) where ∆ is some subset of Nk.
if g(t1, ..., tk) =(cid:81)
Theorem 1.5 ([BR]). Let H(t1, ..., tk) =(cid:80)
n =(cid:80)
α∈∆(1 − tα1
1 ··· tαk
yH
and aλ ∈ N. We assume that
λ∈Λ aλsλ(t1, ..., tk) be a series, and denote
λ(cid:96)n aλdλ where dλ is the dimension of the λ's irreducible representation of Sn,
(1) There is an integer r such that aλ = 0 for every partition λ satisfying λr+1 ≥ M
where M is a constant which depends only on r.
(2) H(t1, ..., tk) is nice rational function.
Then there is an integer d such that for every 0 ≤ m ≤ d − 1 there are constants
βm ∈ 1
Z, and αm ≥ 0. such that
2
nm,t ∼ αmnβm
yH
m,tlnm,t
t → ∞
;
where nm,t = m + td, and l is the minimal integer satisfying condition 1.
2. Affine G-graded Algebras
(cid:80)
n = cG
In this section we prove Theorem A. Our first step is to show that HW (t1, ..., tk) =
λ∈Λk mλsλ satisfies the conditions of Theorem 1.5, hence yHW
n (W ) satisfies the
conclusion of the theorem. Recall that according to the first condition there is an integer
r such that mλ = 0 for every partition λ satisfying λr+1 ≥ M where M is a constant
which depends only on r. In [Go] A.S Gordienko (see Lemma 1 and Lemma 7) showed
that this condition holds for finite dimensional G-graded PI F -algebras, however we
know (see the introduction) that there is a finite dimensional G-graded algebra AW
such that HW = HRG
k (AW ), so indeed HW satisfies the first condition of Theorem 1.5.
Let us show that the second condition of Theorem 1.5 holds, namely that HW is nice
rational. In [AB1] Aljadeff and Belov proved that the Hilbert series of F(cid:104)X G
k (cid:105)/I is rational
for every G - graded T - ideal I of F(cid:10)X G
Theorem 2.1. Let I be a G - graded T - ideal of F(cid:10)X G
set of all PI G - graded T -ideal of F(cid:10)X G
series of F(cid:104)X G
k (cid:105)/I is a nice rational function.
Proof. Assume by contradiction that HF(cid:104)XG
(cid:11) which is PI. In fact, with slight changes in
(cid:11) which is PI then the Hilbert
(cid:11) whose Hilbert series is not nice rational, is
k (cid:105)/I(t1, ..., tk) is not nice rational. Then the
their proof, we can prove that the Hilbert series is also nice.
k
k
k
6
k
not empty. Furthermore, we may assume that there exists a maximal element in this
set, because otherwise (by Zorn's lemma) there is an infinite ascending sequence in this
set. This sequence does not stabilize, and hence its union is a nonfinite generated T -
ideal. This of course contradicts the Specht property of G- graded T -ideals (see [AB2]
section 12).
So let J be a maximal PI G-graded T -ideal of F(cid:10)X G
k (cid:105)/J(cid:1) = IdG(A1 ⊕ ··· ⊕ Al)
nice rational. From [AB2] (Theorem 11.2) we know that
IdG(cid:0)F(cid:104)X G
(cid:11) such that HF(cid:104)XG
k (cid:105)/J is not
k (A1)∩IdG
(2.1)
where Ai is a G-graded basic F -algebra for every i, and idG(Ai) (cid:42) idG(Aj) for any
1 ≤ i, j ≤ m with i (cid:54)= j. We refer the reader to [AB2] for more details - in particular,
the definition of basic algebra. For our purpose, we just have to use one property of basic
algebras which we will present later. Now, from 2.1 we get that J = IdG
IdG
··· ⊕ Al) then we get from Lemma 1.3
k (cid:105)/J = HF(cid:104)XG
k (cid:105)/J1
k (A2⊕···⊕Al). Therefore, if we denote J1 = IdG
(cid:0)F(cid:104)X G
k (cid:105)/J(cid:1) =
k (A1) and J2 = IdG
k (cid:105)/J1+J2
Now, if l > 1 then J (cid:40) J1,J2 from the maximality of J the series
HF(cid:104)XG
HF(cid:104)XG
k (cid:105)/J1
k (cid:105)/J is nice rational which is a contradiction. Hence, l = 1, and
are all nice rational functions. Therefore,
and HF(cid:104)XG
− HF(cid:104)XG
, HF(cid:104)XG
k (cid:105)/J2
+ HF(cid:104)XG
k (cid:105)/J2
k (A2⊕
HF(cid:104)XG
.
k
k (cid:105)/J1+J2
IdG(cid:0)F(cid:104)X G
k (cid:105)/J(cid:1) = IdG(A)
where A is a basic F -algebra.
We recall from [AB1] (see section 2) that for any basic algebra A there is a G-graded
T -ideal K strictly containing J = IdG
k (A) such that K/J is a finite module over some
affine commutative F -algebra, hence by [St] (Chapter 1 Theorem 2.3) H(K/J ) is a nice
rational function. Now, by the maximality of J the Hilbert series of F(cid:104)X G
k (cid:105)/K is nice
rational. Moreover, by Lemma 1.3 we get
HF(cid:104)XG
k (cid:105)/J = HF(cid:104)XG
k (cid:105)/K + HK/J
hence HF(cid:104)XG
k (cid:105)/J is nice rational which contradicts our assumption.
(cid:3)
So indeed we can apply Theorem 1.5 on HW , and we arrive at the following corollary
for affine G-graded algebra with eventually non-decreasing codimension sequence (that
is cG
n+1(W ) for every n > N for some N ∈ N).
n (W ) ≤ cG
Corollary 2.2. If the sequence cG
α1, α2 > 0, β ∈ 1
2
Z, and l ∈ N such that
α1nβln ≤ cG
n (W ) ≤ α2nβln.
7
n (W ) is eventually non-decreasing then there exists
Proof. We need to show that all the βm's form Theorem 1.5 are equal. Suppose that
there are 0 ≤ m, l ≤ d− 1 such that βm < βl, and denote s = l − m (mod d). Note that
s = m − l + qd thus
nm,t + s = m + td + s = m + td + l − m + qd = l + (t + q)d = nl,t+q
therefore
lim
t→∞
cG
nm,t+s(W )
cG
nm,t(W )
which is a contradiction since cG
= lim
t→∞
cG
(W )
nl,t+q
cG
nm,t(W )
βllnl,t
m,tlnm,t
n (W ) is eventually non-decreasing.
αlnl,t
αmnβm
= lim
t→∞
= 0
(cid:3)
The final stage of Theorem A's proof is removing the non-decreasing assumption from
2.2.
Proposition 2.3. Let G be a finite group, and W an affine G- graded F -algebra where
F is a field of characteristic 0. Suppose that W satisfying an ordinary polynomial
identity. Then the G- graded codimension sequence of W is eventually non-decreasing.
Proof. Recall that there exists a finite dimensional algebra AW having the same codi-
mension sequence as W , so we may prove the proposition for AW . Let AW = S + J be
the Wedderburn-Malcev decomposition of AW , and we may assume that the semi-simple
part S, and the Jacobson radical J are G-graded. It is known (see [BRo]) that J is
nilpotent, i.e. there is an integer t such that J t = 0. We claim that cG
n+1(AW )
for every n > t.
n (AW ) ≤ cG
If S = 0 then AW is nilponent of degree t, thus cG
n (AW ) = 0 for every n > t. So assume
that S (cid:54)= 0, and let 1S be its unit element 1S. Because 12
S = 1S, the linear operator
T : AW → AW ; T (x) = x · 1S has eigenvalues 1 or 0, so we have the decomposition
AW = A0 ⊕ A1 where A0, A1 are the eigenspaces of 0 and 1 respectively. Now, let
B = B0 ∪ B1 be a basis of AW where B0 ⊂ A0 and B1 ⊂ A1. We may assume that
all the elements in the basis are G- homogenous. Since F is of characteristic 0, it is
enough to substitute the variables in a polynomial f ∈ F(cid:10)X G(cid:11) by element form B in
order to determine if f is an identity or not.
Denote G = {g1, ..., gs = e}, and for any h =(h1, ..., hn) ∈ Gn denote
Ph = SpF{x1,h1, ..., xn,hn} and Ch = Ph/Ph∩IdG(AW ). Note that C G
and Ch
P(gn1
write
h∈Gn Ch,
∼= Ck if the vector k is a permutation of the vector h. Therefore, if we denote
s ) = Pn1,...,ns and C(gn1
i means gi, ..., gi ni times), we can
s ) = Cn1,...,ns ( gni
1 ,...,gns
1 ,...,gns
n (AW ) =(cid:76)
(cid:88)
(cid:18)
(cid:19)
cn1,...,ns = dimF Cn1,...,ns.
8
(2.2)
where(cid:0)
n
n1,...,ns
(cid:1) = n!
cG
n (AW ) =
n
cn1,...,ns
n1+···+ns=n
n1, ..., ns
n1!···ns! is the generalized binomial coefficient and
Given t < n = n1 + ··· + ns, write cn1,...,ns = k and let
f1(x1,h1, ..., xn,hn), ..., fk(x1,h1, ..., xn,hn) be a set of linear independent polynomials in
Pn1,...,ns/Pn1,...,ns∩IdG(W ) (note that h1 = ··· = hn1 = g1, hn1+1 = ··· = hn1+n2 = g2 and so
on). For every i = 1, ...k we define the following polynomials in Pn1,...,ns+1 (recall that
gs = e, so Pn1,...,ns+1 = P(h1,...,hn,e)).
pi(x1,h1, ..., xn,hn, xn+1,e) =
fi(x1,h1, ..., xj,hj xn+1,e, ..., xn,hn).
so let(cid:80)k
We want to show that p1, ..., pk ∈ Pn1,...,ns+1/Pn1,...,ns+1∩IdG(W ) are linearly independent,
i=1 αipi be a nontrivial linear combination of the pi's. Since f1, ..., fk are linear
independent, there are bj,hj ∈ B ∩ (AW )hj
i=1 αifi(b1,h1, ..., bn,hn) (cid:54)= 0.
Since n is greater than the nilpotency index of J, one of the bi,hi's is necessarily in
S, and S ⊂ A1 so q = B1 ∩ {b1,h1, ..., bn,hn} is a nonzero integer. Let us substitute
b1,h1, ..., bn,hn, 1S in pi. Note that bj,hj · 1S = bj,hj if bj,hj ∈ B1, and bj,hj · 1S = 0 if
bj,hj ∈ B0, therefore
such that (cid:80)k
k(cid:88)
j=1
pi(b1,h1, ..., bn,hn, 1S) = q · fi(b1,h1, ..., bn,hn)
and
k(cid:88)
αipi(b1,h1, ..., bn,hn, 1S) = q ·
Thus the polynomial(cid:80)k
i=1
k(cid:88)
i=1
αifi(b1,h1, ..., bn,hn) (cid:54)= 0.
i=1 αipi(x1,h1, ..., xn,hn, xn+1,e) is not an identity, so
n1,...,ns+1
(cid:19)
(cid:88)
cn1,...,ns = k ≤ cG
(cid:18)
(cid:19)
n1+···+ns=n
n + 1
n
for every n1, ..., ns. Hence by 2.2
where(cid:0)
n
cn1,...,ns
n1, ..., ns
cG
n (W ) =
(cid:18)
≤ (cid:88)
(cid:1) is due to the combinatorial identity
(cid:1) ≤(cid:0)
(cid:18)
(cid:18)
(cid:19)
n1,...,ns+1 ≤ cG
cG
(cid:18)
s−1(cid:88)
n1+···+ns=n
n+1
n1, ..., ns + 1
(cid:19)
n + 1
n
n
n+1(W )
n1, ..., ns + 1
n1, ..., ns
=
+
i=1
n1, ..., ni − 1, ..., ns
n1,...,ns
n1,...,ns+1
(cid:19)
.
(cid:3)
From 2.2 and 2.3 we conclude:
Theorem (A). Let G be a finite group, and W an affine G- graded F -algebra where
F is a field of characteristic 0. Suppose W satisfies an ordinary polynomial identity.
9
Then there exists α1, α2 > 0, β ∈ 1
2
Z, and l ∈ N such that
n (W ) ≤ α2nβln.
α1nβln ≤ cG
A conclusion drawn from theorem A is that the codimension sequence's asymptotics
has a "polynomial part" of degree β ∈ 1
Z. More precisely,
Corollary 2.4. The following limit exists and it is a half integer.
2
(cid:18) cG
(cid:19)
lim
n→∞ logn
n (W )
expG(W )n
3. Unitary Affine G-graded Algebras
In this section we assume that W has a unit. We show that, as in the nongraded
case, the asymptotics of the codimension sequence of W is a constant times polynomial
part times exponential part (Theorem B).
In [D] V.Drensky showed that in the non-graded case (or in another words for
G = {e}) there is a subspace Bn of R
n (W ) = F [z1, ..., zn] ⊗ Bn
{e}
for every n ∈ N. We start this section by proving this result for any finite group
G. In order to keep the notations as light as possible, whenever there is an element
f + I in some quotient space A/I where A ⊂ F(cid:10)X G
(cid:11) we omit the I and leave only
{e}
n (W ) such that R
n
the polynomial f when the convention is that all the operations on such polynomi-
n (W ) spanned by all the poly-
als are modulo I. So let Bn be the subspace of RG
nomials xi1,h1 ··· xir,hr[xir+1,g1, ..., xir+l,gl]··· [..., xis,gs] where hj
(cid:54)= e for i = 1, ..., r.
n (W ), so the
Note that Bn is clearly a sub polynomial GLn(F ) representation of RG
λ sλ(t1, ..., tn) is well defined. Moreover, we
m for every r < m, where the sum is over all the α ∈ Nm such
that α = (α1, ..., αr, 0, ..., 0). Thus, we can get the Hilbert series of Br by putting
tr+1 = ··· = tm = 0 in the Hilbert series of Bm, i.e.
Hilbert series HBn(t1, ..., tn) = (cid:80)
can write Br = (cid:76) Bα
(cid:88)
(cid:88)
λ∈Λn a(n)
a(r)
λ sλ(t1, ..., tr) =
a(m)
λ sλ(t1, ..., tr, 0, ...0).
HBr(t1, ..., tr) =
λ∈Λr
λ∈Λm
Moreover, from Lemma 1.2
(cid:88)
λ∈Λr
a(r)
λ sλ(t1, ..., tr) =
a(m)
λ sλ(t1, ..., tr) +
(cid:88)
λ∈Λm\Λr
· 0.
a(m)
λ
(cid:88)
λ∈Λr
λ∈Λn
Hence, for every r < m the coefficients a(r)
Making it possible to denote a(n)
λ and a(m)
λ
(cid:88)
λ = aλ, and write for every n ∈ N
HBn =
aλsλ(t1, ..., tn).
are the same for every λ ∈ Λr.
We wish to show that, as in RG
n (W ), a single Hilbert series determines all the coeffi-
cients {aλ}λ∈Λ, and that Theorem 1.5 can be applied to this series. Fix an integer t,
10
Lemma 3.1.
p=0
n
Qs = R(t−s)
n,e B(s)
n ⊂ (cid:0)RG
are both polynomial GLn(F ) representations for every s and p.
Mp/Mp+1 as GLn(F ) modules.
n (W )(cid:1)(t) = M0
n (W )(cid:1)(t) ∼=(cid:76)t
(cid:11) is a universal enveloping of the free Lie algebra on X G
n (W )(cid:1)(t). For the other inclusion, note that the free
(cid:11) has a basis
(cid:111)
r βi ≥ 0
1 ··· lβr
lβ1
(cid:110)
and recall that for every polynomial GLn(F ) representation Y the t homogenous sub
s=p Qs for p = 0, ..t (and Mt+1 = {0}). It is easy to see that Qs and Mp
n . Hence, ac-
cording to Poincare-Birkhoff-Witt theorem, if l1, l2, ... is a totally ordered basis of the
. In our case we choose
li = xi,e for i = 1, ..., n, and the other basis elements are arranged in an order such that
representation of Y is Y (t) =(cid:76)α=t Y α. Let us define for every s = 0, ..., t the spaces
n (W )(cid:1)(t) where Rn,e = F(cid:104)x1,e,...,xn,e(cid:105)/IdG(W )∩F(cid:104)x1,e,...,xn,e(cid:105), and the
spaces Mp =(cid:80)t
(1) (cid:0)RG
(2) (cid:0)RG
Proof. It is clear that M0 ⊂ (cid:0)RG
algebra F(cid:10)X G
free Lie algebra, then F(cid:10)X G
deg(lj) ≤ deg(lj+1) for every j > n. We conclude that there is a basis K of F(cid:10)X G
(cid:11)
n,exi1,h1 ··· xir,hr[xir+1,g1, ...]··· [..., xis,gs]
n (W )(cid:1)(t) is spanned
G-graded homomorphism from F(cid:10)X G
(cid:11)(t) onto(cid:0)RG
n since hi (cid:54)= e for every i = 1, ..., r. Moreover, the quotient map is a
i.e. (cid:0)RG
representation for every i ∈ I. Moreover, Qs =(cid:76)
∼= (cid:76)
Mp = (cid:76)
p = 0, ..., t. The inclusion(cid:83)t
part one I =(cid:83)t
i belongs to(cid:83)t
I = (cid:83)t
Jp ∩ Jp(cid:48) for p > p(cid:48). The set Jp is a subset of Ip, and p ≥ p(cid:48) + 1, thus i ∈(cid:83)t
For the second part recall that every GLn(F ) representation is completely reducible,
i∈I Vi where I is a finite set of indexes and Vi is irreducible GLn(F )
Vi where Is is a subset of I, and
s=p+1 Is.
So to prove the second part we need to show that I is a disjoint union of Jp where
p=0 Jp ⊂ I is trivial. For the other direction, recall that by
s=0 Is, so for every i ∈ I there is a minimal p such that i ∈ Ip. Hence
s=p+1 Is. By definition, i ∈ Jp, and we conclude that
p=0 Jp. To show that the Jp's are disjoint suppose that there is an index i in
s=p(cid:48)+1 Is.
(cid:3)
consisting of polynomials of the form xα1
which is in R(α)
s=p Is and not to(cid:83)t
n (W )(cid:1)(t), thus(cid:0)RG
Vi where Jp = (cid:83)t
i∈Is
s=p Is −(cid:83)t
by the t-homogenous elements of K (modulo the identities).
Hence, by definition, i /∈ Jp(cid:48), which is a contradiction.
n (W )(cid:1)(t) =(cid:76)
i∈∪t
s=pIs
Vi. Therefore, Mp/Mp+1
i∈Jp
1,e ··· xαn
n
n
n,e B(s)
n
The next step is to find a basis of Mp/Mp+1 for every p ≤ t. In order to achieve this,
we need some preparations:
Lemma 3.2. Let {y1, ..., yk} be a set of non commutative variables where k ≥ 2, then
we have the following identity
(3.1)
y1 ··· yk = y2y1y3 ··· yk +
yj1 ··· yji[yl1, ..., ylk−i]
k−2(cid:88)
(cid:88)
i=0
11
where the inner sum is over some subset of
(cid:8)(j1, ..., ji, l1, ..., lk−i) ∈ Nk {j1, ..., ji, l1, ..., lk−i} = {1, ..., k}(cid:9) .
Proof. By induction on k. The base of the induction, k = 2, is true by the definition
of the commutator y1y2 = y2y1 + [y1, y2]. For the induction step let us assume that the
claim is true for k. So, by multiplying the identity 3.1 from the right by yk+1 we get:
k−2(cid:88)
(cid:88)
i=0
y1 ··· yk+1 = y2y1y3 ··· yk+1 +
yj1 ··· yji[yl1, ..., ylk−i]yk+1.
We complete the proof by using the commutator definition zw = wz + [z, w] with
z = [y1, ..., ylk−i] and w = yk+1 for every element in the sum.
Definition 3.3. We say that f ∈ F(cid:10)X G
(cid:32) n(cid:89)
n
(cid:11) is g multi-homogenous of degree γ =
(cid:33)
(cid:3)
(γ1, ..., γn), if substituting the variables xi,g by tixi,g in f gives
tγi
i
f.
i=1
(cid:16)
(cid:17)
γ
(cid:16)
(cid:110)
In other words for every i = 1, .., n the variable xi,g appears γi times in every monomial
of f. We denote the g multi-homogenous degree of polynomial f by degg(f ).
It is known that for any g ∈ G if f = (cid:80)
B(p)
n
γ
B(p)
n
where
n =(cid:76)
γ fγ is a G-graded identity, where fγ is g
multi-homogenous of degree γ, then fγ is also G-graded identity for every γ. Therefore,
the decomposition B(p)
n consisting
of g multi-homogenous polynomials of degree γ. We may therefore assume that for
every g ∈ G there is a basis
consisting of g
multi-homogenous polynomials. Now we are ready to present the basis of Mp/Mp+1.
Proposition 3.4. For every p ≤ t let
B(p)
n
(xl,h) j = 1, ..., dimF B(p)
(xl,h) j = 1, ..., dimF B(p)
is the subspace of B(p)
consisting of e multi-homogenous polynomials. Then:
(1) For every 1 ≤ j ≤ dimF B(p)
n and σ ∈ St−p
be a basis of
of B(p)
n
(cid:111)
(cid:110)
(cid:111)
f (p)
j
f (p)
j
n
(cid:17)
γ
n
xi1,e ··· xit−p,ef (p)
j
(xl,h) ≡ xiσ(1),e ··· xiσ(t−p),ef (p)
j
(xl,h) (mod Mp+1)
(2) The set
(cid:110)
Up =
1,e ··· xαn
xα1
is a basis of Mp/Mp+1.
n,ef (p)
j
(xl,h) α = t − p, j = 1, ..., dimF B(p)
n
(cid:111)
Proof. To prove the first part it is enough to show that for every 1 ≤ r ≤ t − p − 1
xi1,e ··· xit−p,ef (p)
j
(xl,h) ≡ xi1,e ··· xr+1,exr,e ··· xit−p,ef (p)
j
(xl,h) (mod Mp+1).
is a linear combination of polynomials of the form
Recall that f (p)
xi1,h1 ··· xiq,hq [xl1,g1, ..., ][..., xlp−q,gp−q ] where hi
12
j
(cid:54)= e. Hence, by using Lemma 3.2 on
xi1,e ··· xit−p,ef (p)
xr,e ··· xit−p,exi1,h1 ··· xiq,hq for every such monomial, we obtain
(xl,h) = xi1,e ··· xir+1,exir,e ··· xit−p,ef (p)
where P (xl,h) is a linear combination of polynomials of the form
xj1e, ..., xjt−q−k,exi1,h1 ··· xiq,hq [xl1,g1, ..., ][..., xlk,gk] with k + q > p. Thus P (xl,h) ∈ Mp+1,
and the first part is proven.
(xl,h) + P (xl,h)
j
j
Obviously Up spans Mp/Mp+1. It remains to be show that Up is linearly independent,
so let
T =
βj,αxα1
1,e ··· xαn
n,ef (p)
j
(xl,h)
(cid:88)
j,α
be a nontrivial linear combination of the elements in Up. Assume by contradiction that
T is an identity of W . Therefore Tγ, the e multi-homogenous component of degree γ of
T , is also an identity for every γ ∈ Nn. We choose γ such that Tγ is a nontrivial linear
combination, and write
Tγ =
(cid:17) βj,αxα1
where the first sum is over every 1 ≤ j ≤ dimF B(p)
every i = 1, ..., n (note that every index j is appearing at most once in the sum).
such that dege
α=γ−dege
n,ef (p)
(cid:16)
(cid:17)
f (p)
j
f (p)
j
n
1,e ··· xαn
j
j
i
≤ γi for
(cid:88)
(cid:88)
(cid:16)
Now, let substitute xi,e by xi,e + 1 in Tγ and denote the obtained identity by S.
are stay the same under this
there are no e-
Notice that since [y + 1, z] = [y, z] the polynomials f (p)
substitution (in the part which is not commutator products in f (p)
homogenous variables). Therefore, the p - homogenous component of S is
j
j
(cid:88)
βj,αf (p)
j
which is also an identity of W . But f (p)
and not all the βj,α are zero, and this is a contradiction .
are linear independent (modulo the identities)
(cid:3)
j
j
Corollary 3.5. .
n (W ) =(cid:80)n
n (W ) ∼= F [z1, ..., zn] ⊗ Bn as GLn(F ) representation.
(cid:1)δs where δs = dimF B(1s)
(cid:0)n
.
s
s=0
s
(1) RG
(2) cG
i=1 zαi
i ) ⊗ f (p)
to ((cid:81)n
Proof. By 3.4 the mapping from Mp/Mp+1 to F [z1, ..., zn](t−p)⊗B(p)
n (W )(t) ∼= (F [z1, ..., zn] ⊗ Bn)(t) for every t.
For the proof of the second part, note that
n (W ) ∼= (F [z1, ..., zn] ⊗ Bn)(1n) =
C G
(cid:77)
RG
j
F [z1, ..., zn](1n)−β ⊗ Bβ
n
β∈{0,1}n
13
is a GLn(F ) representation isomorphism, thus by Lemma 3.1
n taking(cid:0)(cid:81)n
i=1 xαi
i,e
(cid:1) f (p)
j
The dimension of F [z1, ..., zn](1n)−β is 1 for every β ∈ {0, 1}n. Furthermore, it is easy to
see that Bβ
n
that β = s. Thus dimF C G
for every β ∈ {0, 1}n, and there are(cid:0)n
(cid:0)n
(cid:1)δs.
(cid:1) vectors β ∈ {0, 1}n such
n (W ) =(cid:80)n
∼= B(1β)
β
n (W ) = cG
(cid:3)
s
s=0
s
With this corollary we can write
HRG
Recall that HRG
Moreover, the Hilbert series of F [z1, ..., zn] is known to be
n (W )(t1, ..., tn) =(cid:80)
(cid:88)
λ mλsλ(t1, ..., tn) and HBn(t1, ..., tn) =(cid:80)
n (W ) = HF [z1,...,zn] · HBn.
(cid:88)
n(cid:89)
α∈Nn
thus we have the identity
(3.2)
(3.3)
s(k)(t1, ..., tn),
k∈N
(1 − ti)−1 =
n =
1 ··· tαn
tα1
(cid:88)
i=1
mλsλ =
(cid:88)
(cid:88)
µ∈Λ
k∈N
λ
aµsµs(k)
λ aλsλ(t1, ..., tn).
This identity gives us the following arithmetic connection between mλ and aµ.
Proposition 3.6. For every partition λ
mλ =
(cid:88)
µ∈Lλ
aµ
where Lλ is the set of all partition µ such that λi+1 ≤ µi ≤ λi for every i = 1, ..., h(µ).
Proof. By Pieri's formula (see [M]) sµs(k) =(cid:80) sν where the sum is over all partitions
sµs(λ−µ) =(cid:80) sν if and only if λi+1 ≤ µi ≤ λi for every i.
Suppose that sλ appearing in the sum on the right hand side of sµs(λ−µ) =(cid:80) sν.
ν obtained from µ by adding k boxes to its Young's diagram, no two of them in the
same column. We have to show that sλ appearing in the sum on right hand side of
It is obvious that µi ≤ λi for every i, since we are adding boxes to µ in order to get λ.
Let us assume that there is an index i such that µi < λi+1 ≤ λi. In theµi + 1 column,
then, we necessarily added two boxes which is a contradiction. (see Figure 3.1 )
Figure 3.1.
For the other direction, let λ and µ be partitions such that λi+1 ≤ µi ≤ λi for every
i. Suppose that in the j's column we added two boxes in the k 's and i's row where
k > i. This implies that µi < j ≤ λk ≤ λi+1 and this is a contradiction.
(cid:3)
14
The conclusion of the above proposition is that the coefficients aλ inherit two prop-
erties from the coefficients mλ. Here is the first one.
Corollary 3.7. There exists an integer q such that aλ = 0 if λ /∈ Λq.
Proof. Recall from Section 1 that mλ = 0 if λ /∈ ΛdimF (AW ) where AW is the finite
dimensional algebra with the same G-graded identities as W .
for every i thus µ ∈ Lλ.
Let µ be a partition such that h(µ) > dimF (AW ) − 1 (i.e. µ /∈ ΛdimF AW −1), and
define λ = (µ1, ..., µh(µ), 1). Since h(λ) = h(µ) + 1 > dimF (AW ) we know that mλ =
aν = 0, so aν = 0 for any ν ∈ Lλ . Moreover, it is easy to see that λi+1 ≤ µi ≤ λi
(cid:3)
(cid:80)
By this corollary, and since HBn(t1, ..., tn) = (cid:80)
λ∈Λn aλsλ(t1, ..., tn) for every n, we
conclude that there is a single Hilbert series (here it is HBq (t1, ..., tq)) which determines
all the coefficients aλ. The second corollary of 3.6 is that, as in RG
n (W ) (see Section 1),
the series HBq (t1, .., tq) satisfies the first assumption of Theorem 1.5.
ν∈Lλ
Corollary 3.8. Let HBq (t1, ..., tq) =(cid:80)
λ aλsλ(t1, ..., tq). Then there is an integer r such
that aλ = 0 for every partition λ satisfying λr+1 ≥ M where M is a constant depend
only on r.
Moreover the integer r = expG(W ) − 1 is the minimal satisfying this condition.
ν∈Lλ
λi+1 ≤ µi ≤ λi for every i so µ ∈ Lλ.
Proof. From [Go] there exists an integer M such that mλ = 0 for every partition λ
satisfies λl+1 ≥ M where l = expG(W ). Suppose that µ is a partition such that µl ≥ M.
We define a partition λ = (µ1, ..., µl, µl, µl+2, ..., µh(µ)). Obviously, λl+1 = µl ≥ M, so
aν = 0, so aν = 0 for any ν ∈ Lλ. Moreover, one can check that
mλ = (cid:80)
p there is a partition λ with λl ≥ p such that mλ = (cid:80)
To show the minimality of l − 1, we have to show that for every integer p there is a
partition µ with µl−1 ≥ p such that aµ (cid:54)= 0. We know from [Go] that for every integer
aν (cid:54)= 0. Hence, there is
(cid:3)
µ ∈ Lλ, which satisfies µl−1 ≥ λl ≥ p, such that aµ (cid:54)= 0.
The Hilbert series HBq is also nice rational because HRG
q (W ) is a nice rational function,
ν∈Lλ
and since 3.2 imply that
HBq (t1, ..., tq) =
(1 − ti)HRG
q (W )(t1, ..., tq).
We get that the conclusion of Theorem 1.5 holds for the series {δs}s∈N, and together
with 3.5
(cid:18)n
(cid:19)
αmsβm(l − 1)s
n (W ) ∼ d−1(cid:88)
cG
m=0
q(cid:89)
i=1
(cid:88)
s
0 ≤ s ≤ n
s = m (mod d)
15
where βm ∈ 1
and note that(cid:80)d−1
2
Z for every m. To estimate this sum, let ω be a primitive dth root of unit
t=0 ω(s−m)t is zero if s (cid:54)= m (mod d), and is d if s = m (mod d). Thus,
(cid:88)
0 ≤ s ≤ n
(cid:18)n
(cid:19)
s
αmsβm(l − 1)s
cG
m=0
n (W ) ∼ d−1(cid:88)
d−1(cid:88)
n(cid:88)
d−1(cid:88)
d−1(cid:88)
1
d
m=0
s=0
=
=
d−1(cid:88)
s = m (mod d)
(cid:18)n
(cid:19)
(cid:19)
(cid:18)n
sβm(cid:0)ωt(l − 1)(cid:1)s
By lemma 1.1 in [BeR], for every m and t the expression (cid:80)n
αmsβm(l − 1)s
n(cid:88)
ω(s−m)t
ω−mt
αm
d
m=0
s=0
t=0
t=0
s
s
asymptotically equal to:
µm,tnβm(cid:0)ωt(l − 1) + 1(cid:1)n
(cid:0)n
s
(cid:1)sβm (ωt(l − 1))s
s=0
where µm,t is a constant which does not depend on n. Moreover, the absolute value of
the expression ωt(l − 1) + 1 is maximal when t = 0, so
where β = max(βm) ∈ 1
2
Z. And theorem B is proven.
n (W ) ∼ αnβln
cG
References
[AB1] E. Aljadeff and A. Kanel-Belov, Hilbert series of PI relatively free G-graded algebras are rational
functions. Bull. Lond. Math. Soc. 44 (2012), no. 3, 520 -- 532.
[AB2] E. Aljadeff and A. Kanel-Belov, 'Representability and Specht problem for G-graded algebras',
Adv. Math. 225 (2010) 2391 -- 2428
[AG] E. Aljadeff and A. Giambruno, Multialternating graded polynomials and growth of polynomial
identities, to appear in Proc. Amer. Math. Soc.
[AGL] E. Aljadeff, A. Giambruno and D. La Mattina, Graded polynomial identities and exponential
growth, J. Reine Angew. Math. 650 (2011), 83 -- 100.
[BeR] W.Beckner and A.Regev, Asymptotic estimates using probability. Adv. Math. 138 (1998), no.
1, 1 -- 14.
[BRo] A. Kanel-Belov and L.H. Rowen, Computational aspects of polynomial identities. Research
Notes in Mathematics, 9. A K Peters, Ltd., Wellesley, MA, 2005.
[BR] A.Berele, and A.Regev, Asymptotic behaviour of codimensions of p. i. algebras satisfying Capelli
identities. Trans. Amer. Math. Soc. 360 (2008), no. 10, 5155 -- 5172.
[Bru] Bruce .E. Sagan, The Symmetric Group: Representations, Combinatorial Algorithms and Sym-
[D]
metric Functions, Section 4.4, 155-157
V. Drensky, Codimensions of T-ideals and Hilbert series of relatively free algebras, J. of Alg.
89 (1984), 178 -- 223. MR765766 (86b:16010)
[GL] A. Giambruno and D. La Mattina, Graded polynomial identities and codimensions: computing
the exponential growth. Adv. Math. 225 (2010), no. 2, 859 -- 881.
16
[GZ1] A. Giambruno and M. V. Zaicev, On codimension growth of finitely generated associative
algebras, Adv. Math. 140 (1998), 145 -- 155.
[GZ2] Giambruno, Antonio; Zaicev, Mikhail, Polynomial identities and asymptotic methods. Mathe-
matical Surveys and Monographs, 122. American Mathematical Society, Providence, RI, 2005.
[GZ3] A.Giambruno; M.V.Zaicev, Growth of polynomial identities: Is the sequence of codimensions
eventually nondecreasing? Bull. Lond. Math. Soc. 46 (2014), no. 4, 771 -- 778.
[Go] A. S.Gordienko, Amitsur's conjecture for associative algebras with a generalized Hopf action.
[M]
J. Pure Appl. Algebra 217 (2013), no. 8, 1395 -- 1411.
I. G.Macdonald, (1995), Symmetric functions and Hall polynomials, Oxford Mathematical
Monographs (2nd ed.), The Clarendon Press Oxford University Press
R. Stanley, Combinatorics and Commutative Algebra, Birkhäuser, Boston, 1983.
[St]
[Stu] B. Sturmfels, On vector partition functions, J. Combin. Theory Ser. A 72 (1995) 302 -- 308.
[Sv]
I.Sviridova, Identities of pi-algebras graded by a finite abelian group. (English summary) Comm.
Algebra 39 (2011), no. 9, 3462 -- 3490.
Department of Mathematics, Technion - Israel Institute of Technology, Haifa 32000,
Israel.
E-mail address: yuvalshp 'at' tx.technion.ac.il
17
|
1711.02576 | 1 | 1711 | 2017-11-07T16:00:07 | Bounds on polynomial roots using intercyclic companion matrices | [
"math.RA"
] | The Frobenius companion matrix, and more recently the Fiedler companion matrices, have been used to provide lower and upper bounds on the modulus of any root of a polynomial $p(x)$. In this paper we explore new bounds obtained from taking the $1$-norm and $\infty$-norm of a matrix in the wider class of intercyclic companion matrices. As is the case with Fiedler matrices, we observe that the new bounds from intercyclic companion matrices can improve those from the Frobenius matrix by at most a factor of two. By using the Hessenberg form of an intercyclic companion matrix, we describe how to determine the best upper bound when restricted to Fiedler companion matrices using the $\infty$-norm. We also obtain a new general bound by considering the polynomial $x^qp(x)$ for $q>0$. We end by considering upper bounds obtained from inverses of monic reversal polynomials of intercyclic companion matrices, noting that these can make more significant improvements on the bounds from a Frobenius companion matrix for certain polynomials. | math.RA | math | BOUNDS ON POLYNOMIAL ROOTS
USING INTERCYCLIC COMPANION MATRICES
KEVIN N. VANDER MEULEN, TREVOR VANDERWOERD
Abstract. The Frobenius companion matrix, and more recently the Fiedler companion matrices, have been
used to provide lower and upper bounds on the modulus of any root of a polynomial p(x). In this paper
we explore new bounds obtained from taking the 1-norm and ∞-norm of a matrix in the wider class of
intercyclic companion matrices. As is the case with Fiedler matrices, we observe that the new bounds from
intercyclic companion matrices can improve those from the Frobenius matrix by at most a factor of two. By
using the Hessenberg form of an intercyclic companion matrix, we describe how to determine the best upper
bound when restricted to Fiedler companion matrices using the ∞-norm. We also obtain a new general
bound by considering the polynomial xq p(x) for q > 0. We end by considering upper bounds obtained from
inverses of monic reversal polynomials of intercyclic companion matrices, noting that these can make more
significant improvements on the bounds from a Frobenius companion matrix for certain polynomials.
1. Introduction
There are various techniques for approximating the roots of a polynomial (see for example [7, 8]). Some
algorithms for determining the roots of a polynomial rely on a good first approximation (see e.g. [1]) based
on the coefficients of the polynomial. One method [9] for finding the roots of a polynomial p(x) is to
find the eigenvalues of a companion matrix, since a companion matrix has characteristic polynomial p(x).
To approximate the roots of p(x), one can apply Gershgorin's Theorem [9] or use matrix norms [3] on a
companion matrix to find regions in the complex plane to locate the eigenvalues. For example, using these
methods, one can obtain Cauchy's bound: if λ is a root of
(1)
then
(2)
p(x) = xn + an−1xn−1 + an−2xn−2 + · · · + a1x + a0
λ ≤ max{a0, 1 + a1, 1 + a2, . . . , 1 + an−1}.
Typically one uses the classical Frobenius companion matrix,
7
1
0
2
v
o
N
7
]
.
A
R
h
t
a
m
[
1
v
6
7
5
2
0
.
1
1
7
1
:
v
i
X
r
a
for these purposes but recently other companion matrices have been discovered. In particular, the Fiedler
companion matrices, introduced in [5], were explored in [3] to provide upper and lower bounds on the
modulus of a root of p(x). The Frobenius matrix is itself a Fiedler matrix. More recently, the sparse
companion matrices (also known as intercyclic companion matrices) were characterized in [2]. This class of
matrices includes the Fiedler matrices as a special case. In this paper we develop new bounds on the modulus
Date: Preprint. November 1, 2017.
2010 Mathematics Subject Classification. 15A18, 15A42, 15B99, 26C10, 65F15, 65H04.
Key words and phrases. roots of polynomials, bounds, eigenvalues, Fiedler companion matrix, sparse companion matrix.
c(cid:13) 2017. This manuscript version is made available under the CC-BY-NC-ND 4.0 license
http://creativecommons.org/licenses/by-nc-nd/4.0/
1
(3)
F =
0
0
...
0
1
0
...
0
0
1
...
0
−a0 −a1 −a2
0
0
...
1
· · ·
· · ·
. . .
· · ·
· · · −an−1
,
2
KEVIN N. VANDER MEULEN, TREVOR VANDERWOERD
of a root of p(x) using the larger class of sparse companion matrices, comparing them with other known
bounds, especially those in [3]. We also show how some of the previous bounds are more easily obtained by
using the Hessenberg structure of sparse companion matrices.
We start by providing formal definitions of our terms and describing the Fiedler companion matrix in their
Hessenberg form in Section 2. In Section 3, using the 1-norm and ∞-norm we develop upper bounds based on
sparse companion matrices, extending results that are known [3] for the smaller class of Fiedler matrices. In
Section 4 we use the Hessenberg structure of sparse companion matrices to help choose the Fiedler companion
matrix that provides the best upper bound on the modulus of a root of p(x). In Section 5 we discuss the
usefulness of applying the techniques to a polynomial xqp(x) with q > 0 to obtain bounds on the roots
of p(x). Using monic reversal polynomials, we develop lower bounds on the modulus of a root of p(x) in
Section 6. Then in Section 7 we consider lower and upper bounds by using the inverse of a sparse companion
matrix.
2. Formal Definitions
Formally, as in [2], we say a companion matrix to p(x) = xn + an−1xn−1 + · · · + a0 is an n × n matrix
C = C(p) over a field F[a0, a1, . . . , an−1] with n2 − n entries constant in F and the remaining entries variables
−a0, −a1, . . . , −an−1 such that the characteristic polynomial of C is p. For example, Figures 1 and 3 display
companion matrices with characteristic polynomial p = x5+a4x4+a3x3+a2x2+a1x+a0, and Figure 2 displays
companion matrices of order 6 with characteristic polynomial p = x6 + a5x5 + a4x4 + a3x3 + a2x2 + a1x + a0.
0
0
0
0
1
0
0
0
0
1
0
0
0
0
1
0
0
0
0
1
−a0 −a1 −a2 −a3 −a4
0
0
0
1
0
,
0 −a3 −a2
1
8 −a1
0 −a0
1
0
0
0
0 −8 −a4
0
0
1
0
0
−a4
−a3
−a2
−a1
−a0
,
1 0
0 1
0 0
0 0
0 0
0 0
0 0
1 0
0 1
0 0
Figure 1. Various companion matrices
The first matrix in Figure 1 is a Frobenius companion matrix. The Frobenius companion matrix F is
sometimes represented in other Hessenberg forms, for example the third matrix in Figure 1 has also been
referred to as a Frobenius companion matrix. In particular, F T has the variables in the last column, RF R
has the variables in the first row, and RF T R has the variables in the first column (e.g. the last matrix in
Figure 1), using the reverse permutation matrix R =h 0
1
1
·
·
·
0i. We will say that two companion matrices are
equivalent if one can be obtained from the other by permutation similarity and/or transposition. As such,
each Frobenius companion matrix of order n is equivalent to F in (3).
0
0
0
0
1
0
0
1
−a4 −a5
−a3
−a1
−a0
0
0
0
0
−a2
0 0
0 0
1 0
0 1
0 0
0 0
0
0
0
0
1
0
,
0
1
−a4 −a5
0
0
−a2 −a3
0
0
−a0 −a1
0
1
0
0
0
0
0 0
0 0
1 0
0 1
0 0
0 0
0
0
0
−a2
−a1
−a0
0
0
0
0
1
0
,
Figure 2. Sparse companion matrices
0
1
0
0
0
1
1
0
0
0 −a4 −a5
0
0
−a3
0
0
0
0
0
0
1
0
0
0
0
0
0
1
0
The Frobenius companion matrix has exactly n − 1 of the constant entries set to 1, the remaining (n − 1)2
constant entries are zero. As noted in [2], while a companion matrix must have at least 2n − 1 nonzero
BOUNDS ON POLYNOMIAL ROOTS
USING INTERCYCLIC COMPANION MATRICES1
3
entries, there are companion matrices that have more than 2n − 1 nonzero entries. As such, we will say
that a companion matrix is sparse if it has exactly 2n − 1 nonzero entries. Sparse companion matrices have
also been called intercyclic companion matrices because of an associated digraph structure [6]. The second
companion matrix in Figure 1 is not sparse. If the n − 1 nonzero constant entries of a sparse companion
are 1, then we call the companion matrix unit sparse. Each unit sparse companion matrix is equivalent to
a unit lower triangular Hessenberg matrix, as will be noted in Theorem 2.1. For 0 ≤ k ≤ n − 1, we say
the k-th subdiagonal of a matrix A consists of the entries {ak+1,1, ak+2,2, . . . , an,n−k}. Note that the 0-th
subdiagonal is usually called the main diagonal of a matrix.
Theorem 2.1. [2, Corollary 4.3] A is an n × n unit sparse companion matrix if and only if A is equivalent
to a unit lower Hessenberg matrix
(4)
C =
O Im
O
K
In−m−1
O
for some (n − m) × (m + 1) matrix K with m(n − 1 − m) zero entries, such that C has −an−1−k on its kth
subdiagonal, for 0 ≤ k ≤ n − 1.
For example, Figures 2 and 3 give companion matrices with the structure specified in (4). Note that the
rectangular matrix K in the Hessenberg form of a sparse companion matrix has −an−1 in the top right
corner and −a0 in the lower left corner.
In 2003, Fiedler [5] introduced some companion matrices via products of some block diagonal matrices. Each
of the matrices introduced by Fiedler is unit sparse and is equivalent to a unit lower Hessenberg matrix, as
was noted in [2, Corollary 4.4]. We will define the Fiedler matrices using this Hessenberg form. A Fiedler
companion matrix is a sparse companion matrix which is unit lower Hessenberg with −a0 in position (n, 1),
and if ak−1 is in position (i, j) then ak is in position (i − 1, j) or (i, j + 1) for 1 ≤ k ≤ n − 1. Examples
of Fiedler matrices are given in Figure 3. Note that the class of Fiedler companion matrices includes the
0
1
1
0
0
−a4
−a2 −a3
−a0 −a1
0
0
0
0
0
0
0
1
0
0
0
0
0
1
0
,
0
0
0
0
1
0
0
0
0
1
0
0
0
1
−a3 −a4
−a0 −a1 −a2
0
0
0
0
1
0
,
Figure 3. Fiedler companion matrices
1
0
0
0
0
−a4
−a3
−a2
−a0 −a1
0
1
0
0
0
0 0
0 0
1 0
0 1
0 0
Frobenius companion matrix.
In [3], bounds for the roots of a polynomial were determined using the Fiedler companion matrices based
on the products of block diagonal matrices introduced by Fiedler. Here we will take advantage of the
Hessenberg form of the Fiedler companion matrices to provide insight into bounds on polynomial roots and
also compare the bounds developed for Fiedler matrices with new bounds based on the larger class of unit
sparse companion matrices.
The bounds will be obtained from matrix norms. A matrix norm is submultiplicative if kABk ≤ kAkkBk for
all matrices A, B ∈ Fn×n. As noted in [7], if a matrix norm is submultiplicative and λ is an eigenvalue of C
then
(5)
λ ≤ kCk.
Since the eigenvalues of a companion matrix C(p) are the same as the roots of p, this technique can be
used to find a bound on the roots of p. However, there are many submultiplicative matrix norms and many
4
KEVIN N. VANDER MEULEN, TREVOR VANDERWOERD
nonequivalent unit sparse companion matrices to consider. As was done in [3], for a matrix A = [aij] of
order n, we will focus on two norms that have simple expressions, the ∞-norm and the 1-norm:
kAk∞ = max
1≤i≤n
n
Xj=1
aij
and
kAk1 = max
1≤j≤n
aij.
n
Xi=1
We will let N (A) be the minimum of the ∞-norm and the 1-norm,
(6)
N (A) = min{kAk∞, kAk1},
so that λ ≤ N (A) for any matrix A with eigenvalue λ. Note that if A and B are equivalent matrices then
N (A) = N (B).
3. Upper Bounds Based on Sparse Companion Matrices
Let p be a polynomial as in (1) and C = C(p) be an arbitrary unit sparse companion matrix. Recall F is the
Frobenius companion matrix of p. Let Si = {k −ak is in row i of C} and let Ti = {k −ak is in column i of C}
for each i ∈ {1, 2, . . . , n}. Note that 0 ∈ Sn and 0 ∈ T1. Using the norms kCk∞, kCk1, kF k∞, and kF k1
respectively, the following known bounds on a root λ of p are obtained from (5):
(7)
λ ≤ max
Xi∈Sn
λ ≤ max(Xi∈T1
ai, 1 + Xi∈Sn−1
ai, 1 + Xi∈T2
ai, . . . , 1 + Xi∈S1
ai, . . . , 1 + Xi∈Tn
,
ai
ai) ,
λ ≤ max {1, a0 + a1 + · · · + an−1} ,
λ ≤ max {a0, 1 + a1, 1 + a2, . . . , 1 + an−1} .
and
We first note that, in order for N (C) to be an improvement over N (F ) as a bound on the roots of p, it is
necessary that the constant term of p be less than 1:
Theorem 3.1. Let F be the Frobenius companion matrix and let C 6= F be a unit sparse companion matrix,
both based on the polynomial p in (1). If N (C) < N (F ) then a0 < 1.
Proof . Suppose a0 ≥ 1. Let M = max{ak 1 ≤ k ≤ n − 1}.
Case 1: Suppose a0 ≥ 1 + M . Then a0 ≥ 1 + ai for all i ∈ {1, 2, . . . , n − 1}. Then N (F ) = a0. Further
ai ≥ a0, so kCk∞ ≥ N (F ) and kCk1 ≥ N (F ). Therefore N (C) ≥ N (F ).
Pi∈Sn
ai ≥ a0 and Pi∈T1
Case 2: Suppose 1 + M > a0. Then kF k1 = 1 + M and N (F ) = 1 + M . Note that every row and column
of C contains either 1 or a0 ≥ 1. Therefore N (C) ≥ 1 + M = N (F ).
The next theorem provides necessary conditions on the shape of the companion matrix C when N (C) provides
an improved bound on N (F ).
Theorem 3.2. Let F be the Frobenius companion matrix and let C 6= F be a unit sparse companion matrix,
both based on the polynomial p in (1). Let M = max{ak 1 ≤ k ≤ n − 1}. If N (C) < N (F ), then either
1. All coefficients with ai = M , i ∈ {1, 2, . . . , n − 1}, are in the n-th row of C and kCk∞ < N (F ) ≤
kCk1, or
2. All coefficients with ai = M , i ∈ {1, 2, . . . , n − 1}, are in the 1st column of C and kCk1 < N (F ) ≤
kCk∞.
BOUNDS ON POLYNOMIAL ROOTS
USING INTERCYCLIC COMPANION MATRICES2
5
Proof . Suppose N (C) < N (F ). By Theorem 3.1, a0 < 1 and so kF k1 = 1 + M and N (F ) ≤ 1 + M .
Suppose there exists a coefficient with ai = M , i ∈ {1, 2, . . . , n − 1}, that is not in the n-th row or the
1st column of C. Then kCk∞ ≥ 1 + M , kCk1 ≥ 1 + M , and N (C) ≥ 1 + M . Since N (F ) ≤ 1 + M ,
N (C) ≥ N (F ), which is a contradiction to N (C) < N (F ).
Likewise, there would be a contradiction if there exist two coefficients (not including a0) with modulus M
such that one is in the first column of C and one is in the last row of C.
Suppose all coefficients with ai = M , i ∈ {1, 2, . . . , n − 1}, are in the n-th row. Then kCk1 ≥ 1 + M.
Therefore kCk1 ≥ N (F ). Since N (C) < N (F ) ≤ kCk1, kCk∞ < N (F ) ≤ kCk1.
The proof of part 2 is similar with the 1-norm and ∞-norm reversed.
The next result demonstrates that if N (C) provides a better bound than N (F ) for a polynomial p, then
either p does not have many coefficients with maximum modulus, or the maximum modulus is small.
Corollary 3.3. Suppose M = max{ak 1 ≤ k ≤ n − 1} and u = {k M = ak}. If N (C) < N (F ) for
some unit sparse companion matrix C, then (u − 1)M < 1 − a0.
Proof . Suppose N (C) < N (F ). Then by Theorem 3.2 all coefficients with ai = M , i ∈ {1, 2, . . . , n − 1},
must be in the n-th row of C and Pj Cij < 1 + M for each row i of C. Working with the n-th row,
M + M + · · · + M + a0 < 1 + M.
Hence (u − 1)M < 1 − a0.
The last results have described various conditions under which a unit sparse companion matrix might produce
a better bound than the Frobenius companion matrix. When the unit sparse companion matrix provides a
better bound, the Frobenius bound N (F ) is at most twice as large as N (C) and can exceed N (C) by no
more than one. This mimics what was discovered about Fiedler matrices in [3, Theorem 4.3].
Theorem 3.4. Let F be the Frobenius companion matrix and let C 6= F be a unit sparse companion matrix,
both based on the polynomial p in (1). Then N (F ) < 2N (C) and N (F ) − N (C) ≤ 1. The latter inequality
is strict if a0 6= 0.
Proof . Suppose N (C) < N (F ). Let M = max{ak 1 ≤ k ≤ n − 1}. Then
N (F ) = min {max {a0, 1 + M } , max {1, a0 + a1 + · · · + an−1}} .
But since a0 < 1,
Hence N (F ) ≤ 1 + M . Similarly, with u = {k M = ak},
N (F ) = min {1 + M, max {1, a0 + a1 + · · · + an−1}} .
Thus
Solving for the ratio gives
since N (C) ≥ 1.
N (C) ≥ a0 + uM.
N (F ) − N (C) ≤ 1 + M − a0 − uM
≤ 1 − a0 − (u − 1)M ≤ 1.
N (F )
N (C)
≤
1
N (C)
+ 1 ≤ 2
Suppose N (F ) = 2N (C). Given that N (F ) ≥ N (C)+1, and N (C) ≥ 1, it follows that N (C) = 1. According
to Theorem 3.2, only one of kCk∞ and kCk1 can be sharper than N (F ). Thus by equivalence, we may assume
kCk∞. In this case kCk∞ = 1. Then all coefficients not in the n-th row of C are 0, otherwise kCk∞ ≥ 1 + ai
6
KEVIN N. VANDER MEULEN, TREVOR VANDERWOERD
where ai is any nonzero coefficient not in the n-th row of C. However, if the nonzero coefficients are all
in the n-th row of C, C is a Frobenius matrix. Then N (C) = N (F ) = 1, contradicting N (F ) = 2N (C).
Therefore N (F ) < 2N (C).
When seeking an optimal upper bound N (A) over all unit sparse companion matrices A using the Hessenberg
structure, then one can restrict attention to the ∞-norm. In particular, if A is a sparse companion matrix
in Hessenberg form with N (A) = kAk1 then B = RAT R (obtained from A by a reflection across the
antidiagonal) is also Hessenberg and B is equivalent to A with N (B) = kBk∞ = N (A). In the next section
we restrict our attention to the ∞-norm.
4. Upper Bounds Based on Fiedler Companion Matrices
In this section, using the Hessenberg structure, for a given polynomial, we note that there is a particular
Fiedler matrix which provides the best upper bound for N (C) over all Fiedler matrices C. Given 0 ≤ b ≤ n−1,
let
Lb =
O
Ib
O
−an−1
...
−ab+1
−ab
In−b−1
O
O
−a0
· · ·
.
Note that Lb is a Fiedler companion matrix and Ln−1 = F .
Theorem 4.1. Let Fb be any Fiedler companion matrix (in Hessenberg form) with ab appearing in row n
for some b ∈ {0, 1, 2, . . . , n − 2} but ab+1 is in row n − 1. Then kLbk∞ ≤ kFbk∞.
Proof . Let K = max{ai i > b} and g = max{i ai = K}. Then
kLbk∞ = max
1 + K,
b
Xj=0
.
aj
Note that there are no other coefficients of p in the same row as ag in Lb but there may be more coefficients
in the same row as ag in Fb. Therefore kFbk∞ ≥ 1 + K and kFbk∞ ≥
aj. Thus kFbk∞ ≥ kLbk∞.
b
Pj=0
Combined with Theorem 3.2, Theorem 4.1 implies that when seeking an upper bound with a Fiedler matrix
that improves on the Frobenius bounds, one can restrict their attention to the ∞-norm of an Lb Fiedler
matrix. Furthermore, the next theorem indicates the choice of b such that Lb gives the best upper bound of
all Fiedler matrices.
Theorem 4.2. Let r = max(cid:26)k (cid:12)(cid:12)(cid:12)(cid:12)
b ∈ {0, 1, 2, . . . , n − 2}.
k−1
Pi=0
ai < 1(cid:27) with r = 0 if a0 ≥ 1. Then kLrk∞ ≤ kLbk∞ for all
BOUNDS ON POLYNOMIAL ROOTS
USING INTERCYCLIC COMPANION MATRICES3
7
Proof . Suppose b > r. Then
kLbk∞ = max( b
Xi=0
= max( r
Xi=0
≥ max( r
Xi=0
kLbk∞ = max( b
Xi=0
Suppose b < r. Then
ai, 1 + ab+1, 1 + ab+2, . . . , 1 + an−1)
b
ai +
Xi=r+1
ai, 1 + ab+1, 1 + ab+2, . . . , 1 + an−1)
ai, 1 + ar+1, 1 + ar+2, . . . , 1 + an−1) = kLrk∞.
ai, 1 + ab+1, 1 + ab+2, . . . , 1 + an−1)
= max(cid:26)1 + ab+1, 1 + ab+2, . . . , 1 + an−1(cid:27)
≥ max(cid:26)1 + ar, 1 + ar+1, . . . , 1 + an−1(cid:27)
≥ max( r
Xi=0
ai, 1 + ar+1, 1 + ar+2, . . . , 1 + an−1) = kLrk∞.
Example 4.3. Consider the polynomial
p = x8 − 0.1x7 − 0.1x6 − 0.3x5 − 0.1x4 − 0.5x3 − 0.1x2 − 0.1x − 0.1.
To determine the Fiedler matrix that provides the best bound on the roots of p, we simply find the first partial
sum
b
Pi=0
but
ai which is more than one. In particular,
ai = − 0.1 + − 0.1 + − 0.1 + − 0.5 + − 0.1 < 1,
4
Xi=0
5
Thus, by Theorems 4.1 and 4.2, L5 gives the best bound of all Fiedler matrices. For comparison, we calculated
kLbk∞ for each b ∈ {0, 1, . . . , 7}:
ai = 1.2 > 1.
Xi=0
b
4
kLbk∞ 1.5 1.5 1.5 1.3 1.3
0
1
2
3
6
7
5
1.2 1.3 1.4
.
5. Upper Bounds obtained from extended polynomials
Sections 3 and 4 illustrate some necessary conditions for a unit sparse companion matrix of p to provide a
sharper bound on the roots of p than the Frobenius matrix. By multiplying the polynomial p by the factor
xq, for some q > 0, some of these restrictions can be removed. In particular, if q > 0, then a root of p will
also be a root of xqp. Thus if λ is a root of p, then λ ≤ N (C(xqp)). Using a unit sparse companion matrix
of the polynomial xqp for some q > 0, instead of the polynomial p, can provide sharper bounds on a root of
p than the Frobenius bounds and the companion matrix bounds developed in Section 3:
8
KEVIN N. VANDER MEULEN, TREVOR VANDERWOERD
Example 5.1. Consider the polynomial p = x6 − x5 − 2x4 − x3 − 4x2 − 2x − 3. By Theorem 3.1, N (C) ≥
N (F (p)) = 5 if C = C(p) is a companion matrix of p. Consider a companion matrix of the polynomial x3p:
(8)
C(x3p) =
0
0
0
0
0
0
0
0
0
1 0
0 1
0 0
0 0
0 0
2 0
0 0
0 3
0 0
0 0
0 0
1 0
0 1
0 0
0 0
0 1
0 0
0 0
0 0
0 0
0 0
0 0
1 0
1 1
2 0
0 0
4 0
0 0
0 0
0 0
0 0
0 0
0 0
1 0
0 1
0 0
.
The solid lines in (8) indicate the partition of the companion matrix C(x3p) as outlined in (4) and the
dashed lines illustrate the fact that C(x3p) has extra columns (and rows) compared to C(p). Note that
kC(x3p)k∞ = 4 and so kC(x3p)k∞ < N (F (p)). Thus, using the polynomial x3p we obtain a tighter upper
bound on a root of p than what can be obtained by using a companion matrix of p itself.
The companion matrices of xqp can be used to generalize the bound found in [3, Theorem 3.3]. Note that
given any partition P = {P1, P2, . . . , Pt} of {0, 1, . . . , n − 1}, by the pigeonhole principle, there exists a Pℓ
such that max{i i ∈ Pℓ} ≤ n − t. By relabeling, we can assume ℓ = 1.
Theorem 5.2. Let λ be a root of p as in (1). Let P = {P1, P2, . . . , Pt} be a partition of {0, 1, . . . , n − 1}
into t nonempty sets, with max{i i ∈ P1} ≤ n − t. Then
λ ≤ BP = max(Xi∈P1
ai, 1 + Xi∈P2
ai, . . . , 1 + Xi∈Pt
ai) .
Proof . Relabel the partition P = {P1, P2, . . . , Pt} of {0, 1, . . . , n − 1} with max{i i ∈ P1} ≤ n − t so that
max{i i ∈ Pv} > max{i i ∈ Pu} for all v > u > 1. Then n − 1 ∈ Pt, n − 2 ∈ Pt ∪ Pt−1, and inductively,
n − r ∈ Pt ∪ Pt−1 ∪ · · · ∪ Pt−r+1 for 1 ≤ r ≤ n − 1. In particular, if i ∈ Pj then i ≤ n + j − t − 1.
We can construct an order n + t − 1 companion matrix C = C(xt−1p) in the form (4) with K a t × n
submatrix. In particular, the order of C is sufficiently large so that if i ∈ Pj then ai can be in row n + t − j
of C: place ai in position (n + t − j, t + i − j + 1) so that ai is on the (n − i − 1)th subdiagonal of C. Note
that n ≤ n + t − j ≤ n + t − 1 since 1 ≤ j ≤ t, and 1 ≤ t + i − j + 1 ≤ n since 1 ≤ j ≤ t and i ≤ n + j − t − 1,
so that ai is in submatrix K. Now Bp = kCk∞.
With the partition relabeling as noted in the above proof, we can observe that an−1 ∈ Pt. Consequently the
proof used C = C(xqp) with q = t − 1 in order to ensure there are sufficient columns so that a0 could appear
in the same row as an−1 while satisfying the diagonal conditions of the matrix K in (4).
Theorem 5.2 is a generalization of the four bounds in (7), including the Frobenius bound N (F ), as well as
the bounds in [3, Theorem 3.3]. Particularly, the two portions of the Frobenius bound are obtained from
kCk∞ with C = C(xt−1p), using t = n and t = 1 respectively.
Example 5.3. Let p = x8 − x7 − 3x6 − 2x4 − 2x4 − 4x3 − 3x2 − 5x − 3, and let λ be a root of p. Then
N (F (p)) = 6, and by Theorem 3.1, N (F (p)) ≤ N (C(p)) for all unit sparse companion matrices of p.
Consider the partition: P = {{0, 7}, {1}, {2}, {3}, {4}, {5}, {6}}. Using the notation in the proof of Theo-
rem 5.2, t = 7 and n − t = 1. Since max{{1}} ≤ 1, P1 = {1}. Then
BP = max {a1, 1 + a2, 1 + a3, 1 + a4, 1 + a5, 1 + a6, 1 + a0 + a7}
= max {5, 4, 5, 3, 3, 4, 5} = 5.
BOUNDS ON POLYNOMIAL ROOTS
USING INTERCYCLIC COMPANION MATRICES4
9
Equivalently, BP = kC(x6p)k∞ for
C(x6p) =
0 1
0 0
0 0
0 0
0 0
0 0
0 0
3 0
0 0
0 0
0 0
0 0
0 0
0 0
0 0
1 0
0 1
0 0
0 0
0 0
0 0
0 0
0 0
0 0
0 0
0 0
0 0
0 0
0 0
0 0
0 0
1 0
0 1
0 0
0 0
0 0
0 0
0 0
0 0
0 0
0 0
0 0
0
0
0
0
0
1
0
0
0
0
0
0
0
0
0 0
0 0
0 0
0 0
0 0
0 0
1 0
1 1
3 0
2 0
2 0
4 0
3 0
5 0
0 0
0 0
0 0
0 0
0 0
0 0
0 0
0 0
1 0
0 1
0 0
0 0
0 0
0 0
0 0
0 0
0 0
0 0
0 0
0 0
0 0
0 0
0 0
0 0
1 0
0 1
0 0
0 0
0
0
0
0
0
0
0
0
0
0
0
0
1
0
.
In Example 5.3, t is large relative to n. In general, it is good to take t as large as possible in order to reduce
the size of the parts in the partition, since the sums in BP are minimized if the parts are small. However,
if we intend for the bound BP to be sharper than a Frobenius bound, then the coefficients with maximum
modulus need to be in P1, which limits t. In particular, if M = max{ak 1 ≤ k ≤ n − 1} and ai = M ,
then we would take t ≤ n − i.
Example 5.4. Let p = x8 − x7 − 3x6 − x5 − 6x4 − x3 − 5x2 − 3x − 1, and let λ be a root of p. Then
λ ≤ N (F (p)) = 7, and by Theorem 3.1, N (F (p)) ≤ N (C(p)) for all unit sparse companion matrices C(p).
Consider the partition: P = {{4}, {2}, {0, 5, 6}, {1, 3, 7}}. With this partition, t = P = 4 and n − t = 4.
Thus, by Theorem 5.2, we could choose P1 to be {2} or {4}. With P1 = {4}, we have
BP = max {a4, 1 + a2, 1 + a0 + a5 + a6, 1 + a1 + a3 + a7} = max {6, 6, 6, 6} = 6 < 7,
and λ ≤ 6.
The restrictions on a0 provided by Theorem 3.1 and Corollary 3.3 do not apply to C(xqp), q ≥ 1, as
they do to C(p). For instance, Example 5.4 illustrates that there is a unit sparse companion matrix C with
N (C(x4p)) < N (F (p)) = N (F (x4p)) but a0 = 1. However, C(p) can be replaced by C(xqp) in Theorems 3.2
and 3.4 without changing the conclusion. In particular, N (F (p)) < 2N (C(xqp)) and N (F (p))−N (C(xqp)) ≤
1.
6. Lower Bounds Using Monic Reversal Polynomials
One tool for obtaining lower bounds on the roots of a polynomial is to apply the techniques previously
developed to the monic reversal polynomial, as shown in [7]. Assuming that a0 6= 0, the monic reversal
polynomial of p is
(9)
p♯(x) =
xn
a0
a1
a0
xn−1 +
a2
a0
xn−2 + · · · +
an−1
a0
x +
1
a0
.
p(cid:0)x−1(cid:1) = xn +
In the case that a0 = 0, one could consider the monic reversal of the lower degree polynomial p/x if a1 6= 0.
Note that (p♯)♯ = p. Since the roots of p♯ are the reciprocals of the roots of p, the eigenvalues of a companion
matrix C(p♯) are the reciprocals of the roots of p. Therefore, if λ is a root of p,
(10)
1
kC(p♯)k
≤ λ
10
KEVIN N. VANDER MEULEN, TREVOR VANDERWOERD
for any submultiplicative matrix norm. Including (5), we have
(11)
1
N (C(p♯))
≤ λ ≤ N (C(p)).
Thus, the theorems from Sections 3 and 4 can be applied to companion matrices of monic reversal polynomials
to obtain lower bounds. For example, by Theorem 3.1, if N (C(p♯)) < N (F (p♯)) then a0 > 1.
Example 6.1. Let λ be a root of
Note that
p = x6 − 3x5 − 10x4 − 4x3 − 5x2 − x − 5.
p♯ = x6 +
1
5
x5 + x4 +
4
5
x3 + 2x2 +
3
5
x −
1
5
.
The Frobenius matrix provides a lower bound, λ ≥ N (F (p♯))−1 = 1/3, but the companion matrix
C(p♯) =
0
0
0
0
−3/5
1/5
1
0
1
−4/5
0
0
0
1
0
0
−1/5 1
0
0
0
0
0
−2
0 0
0 0
0 0
1 0
0 1
0 0
,
provides a lower bound of λ ≥ N (C(p♯))−1 = 5/11.
The next theorem implies that of all the unit sparse companion matrices of p, the Frobenius matrix will
provide either the sharpest lower bound, the sharpest upper bound, or both.
Theorem 6.2. Let C(p) and C(p♯) be unit sparse companion matrices for polynomials of the form (1) and
(9) respectively. Let F (p) and F (p♯) be the corresponding Frobenius companion matrices.
(i) If a0 ≥ 1 then N (F (p)) ≤ N (C(p)).
(ii) If a0 ≤ 1 then N (F (p♯)) ≤ N (C(p♯)).
Proof . Part (i) is the contrapositive of Theorem 3.1. Part (ii) is an application of Theorem 3.1 to the lower
bound in (11).
Example 6.3. Let λ be a root of
The Frobenius matrix will give the upper bound λ ≤ N (F ) = 7. Using the companion matrix
p = x7 − 4x6 − x5 − 2x4 − 6x3 − 5x2 − 2x − 0.2.
C =
0
0
0
0
5
2
0.2
1
0
0
0
0
0
0
0
1
0
1
0
0
0
0
0
1
4
0
2
6
0
0
0
1
0
0
0
0
0
0
0
1
0
0
0
0
0
0
0
1
0
,
we can improve the Frobenius bound to λ ≤ N (C) = 6.2. Since a0 = 0.2 < 1, Theorem 6.2(ii) implies that
the best lower bound will be obtained from the Frobenius matrix of the monic reversal polynomial
In particular, using unit sparse companion matrices of p, the best lower bound we can obtain from (11) is
p♯ = x7 + 10x6 + 25x5 + 30x4 + 10x3 + 5x2 + 20x − 5.
λ ≥ N (F (p♯)) =
1
31
.
BOUNDS ON POLYNOMIAL ROOTS
USING INTERCYCLIC COMPANION MATRICES5
11
Corollary 6.4. Let λ be a root of p. If a0 = 1, of all the unit sparse companion matrices of p, the Frobenius
companion matrix provides both the sharpest lower bound and the sharpest upper bound on λ in (11).
Note that the converse of Corollary 6.4 is not true:
Example 6.5. Consider the polynomial
p6 = x6 − x5 − x4 − x3 − 2x2 − 2x − 0.2.
By Corollary 3.3, N (F ) ≤ N (C) for any unit sparse companion matrix of p since (u − 1)M = 2 ≥ 0.8 =
1 − a0. Since a0 = 0.2 < 1, by Theorem 6.2 (ii), the Frobenius matrix also provides the sharpest lower
bound of all the unit sparse companion matrices of p.
As in the previous section, we note that Theorem 3.4 still applies: in particular, while N (C(p♯)) can be a
better bound than N (F (p♯)), it will not be an improvement of more than a factor of two.
7. Bounds Using Inverse Matrices and Monic Reversal Polynomials
If λ is a root of p and C is a unit sparse companion matrix of p, it follows that
1
λ
≤ kC−1k
for any submultiplicative matrix norm. Using this observation with the monic reversal polynomial, we obtain
the upper and lower bounds:
1
N (C−1(p))
≤ λ ≤ N (C−1(p♯)).
Given 0 ≤ c ≤ n − 1 and a0 6= 0, then a unit sparse companion matrix can be partitioned as
(12)
C = C(p) =
for some vectors u and y, and some (n − c − 1) × c matrix H, whose nonzero entries are coefficients of −p.
In this case, the inverse of C will be
O
u
Ic
O
H In−c−1
−a0 yT
O
,
.
(13)
C−1 =
1
a0
yT
Ic
O
O
− 1
a0
O
1
−a0
uyT − H In−c−1
1
a0
u
As noted in [3], for the Frobenius matrix F , F (p♯)−1 is equivalent to F (p). Hence we will continue to
compare new bounds to N (F ).
Due to the number of nonequivalent unit sparse companion matrices, determining a simple expression for
the 1-norm and the ∞-norm of the inverse of a companion matrix is not straightforward. In order to simplify
matters, we will restrict our attention to those with u = 0. In particular, for 1 ≤ c ≤ n − 2, we say a matrix
is of type Ec = Ec(p) if it has the form (12) with u = 0. In this case, y1 = −a1. Further, we say a matrix
A is of type Ec(p♯)−1 if A−1 is of type Ec(p♯). In this case, y1 = −an−1.
12
KEVIN N. VANDER MEULEN, TREVOR VANDERWOERD
There is only one matrix of type E1(p♯)−1, namely
(14)
W =
O
O
−a0
0
In−2
O
−an−1
1
a1/a0
a2/a0
...
an−2/a0
.
(The matrix W is equivalent to a matrix labeled F (p♯)−1 in [3].) When restricting to matrices A of type
Ec(p♯)−1 obtained from Fiedler matrices, [3, Theorem 5.4] compares kW k∞ and kF k∞ to kAk∞ under
various conditions. Theorem 7.1 expands the comparison to all matrices of type Ec(p♯)−1, including those
obtained from the larger class of unit sparse companion matrices, and once again highlights W when a0 > 1.
Later, in Theorem 7.3, we characterize when bounds derived from W improve those derived from F .
Theorem 7.1. Let M = max{ak 1 ≤ k ≤ n−1}. Suppose A is of type Ec(p♯)−1 for some c, 1 ≤ c ≤ n−2.
a) If a0 = 1 then N (F ) ≤ N (A).
b) If a0 > 1 then kW k∞ ≤ kAk∞.
c) If a0 > 1 and an−1 = M then N (F ) ≤ N (A).
d) If a0 < 1 then
i) N (F ) − kAk∞ ≤ 1,
ii) N (F ) ≤ 2kAk∞, and
iii) N (F ) ≤ kAk1.
Proof . Suppose 1 ≤ c ≤ n − 2 and A is of type Ec(p♯)−1. Then
a0yT
Ic
O
O
−H In−c−1
−a0
O
O
A =
.
for some y, and some (n − c − 1) × c matrix H, whose nonzero entries are in n a1
a0y1 = −an−1.
a0
, a2
a0
, . . . , an−2
a0 o, with
a) Suppose a0 = 1. Then there is a 1 in every row and column of A. Also, the nonzero values in H
are coefficients of p. Therefore N (A) ≥ 1 + M . Since a0 = 1,
N (F ) = min {max {a0, 1 + M } , max {1, a0 + a1 + · · · + an−1}}
= 1 + M.
Therefore N (F ) = 1 + M ≤ N (A).
b) Suppose a0 > 1. Note that kW k∞ = max(cid:26)a0 + an−1, 1 +
M
a0(cid:27). If a0 + an−1 ≥ 1 +
M
a0
then kW k∞ = a0 + an−1 ≤ kAk∞. Suppose kW k∞ = 1 +
. If there is a coefficient of p with
M
a0
modulus M in the first row of A then kAk∞ ≥ 1 + M > kW k∞. If there is a coefficient of p with
modulus M in H, then kAk∞ ≥ 1 + M
a0 = kW k∞. Therefore kW k∞ ≤ kAk∞.
c) Suppose a0 > 1 and an−1 = M . Note that kAk1 ≥ a0 and kAk∞ ≥ a0 + an−1. Since
an−1 = M , kAk1 ≥ 1 + M and kAk∞ ≥ a0 + M . Since a0 > 1, N (F ) = max {a0, 1 + M }.
Therefore N (F ) ≤ N (A).
d) Suppose a0 < 1.
BOUNDS ON POLYNOMIAL ROOTS
USING INTERCYCLIC COMPANION MATRICES6
13
i) Suppose kAk∞ < N (F ). Since a0 < 1,
N (F ) = min {max {a0, 1 + M } , max {1, a0 + a1 + · · · + an−1}}
= min {1 + M, max {1, a0 + a1 + · · · + an−1}} .
However, if 1 > a0 + a1 + · · · + an−1 then N (F ) = 1 ≤ kAk∞, so
N (F ) = min {1 + M, a0 + a1 + · · · + an−1} .
In order for kAk∞ < N (F ), all coefficients of p with modulus M must be in the first row of A,
M
a0
since 1 + M < 1 +
when a0 < 1. This gives kAk∞ ≥ a0 + uM , with u = {k M = ak}.
Therefore, since N (F ) ≤ 1 + M , N (F ) − kAk∞ ≤ 1 + M − a0 − uM ≤ 1.
ii) Using part (i), we have
N (F )
kAk∞
≤ 1 +
1
kAk∞
.
However, kAk∞ ≥ 1, so N (F ) ≤ 2kAk∞.
iii) Since a0 < 1, kAk1 ≥ 1 + M . Recall that
N (F ) = min {1 + M, a0 + a1 + · · · + an−1} ≤ 1 + M.
Therefore N (F ) ≤ kAk1.
Observe that the conditions given in Theorem 7.1d)i) and ii) mimic conditions given in Theorem 3.4. These
conditions can also be applied to the polynomials xqp in Section 5 and the lower bounds in Section 6.
However, when a0 > 1, it is possible that N (F ) > 2kW k∞ and N (F ) − kW k∞ > 1. In fact, as we note in
Remark 8.4, the ratio can be made arbitrarily large.
Example 7.2. Let λ be a root of p = x6 − x5 + 12x4 + 6x3 + 36x2 + 18x − 6. kW k∞ = 7 and N (F ) = 37,
giving
N (F ) − kW k∞ = 30 and
N (F )
kW k∞
=
37
7
.
= 5.3.
In the next theorem, we characterize when W provides a better upper bound on the roots of a polynomial
than the Frobenius matrix F .
Theorem 7.3. Given a polynomial p with M = max{ak 1 ≤ k ≤ n − 1}. Then kW k∞ < N (F ) if and
only if 1 < a0 < 1 + M − an−1.
Proof . Suppose kW k∞ < N (F ). Then
max(cid:26)a0 + an−1, 1 +
M
a0(cid:27) < min {max {a0, 1 + M } , max {1, a0 + a1 + . . . + an−1}} .
Suppose a0 ≤ 1. Then kW k∞ ≥ 1 + M ≥ N (F ). Thus a0 > 1.
Suppose a0 ≥ 1 + M . Then N (F ) = a0. But kW k∞ ≥ a0 + an−1. Therefore a0 < 1 + M and
N (F ) = 1 + M . Hence a0 + an−1 < 1 + M. Therefore 1 < a0 < 1 + M − an−1.
Suppose 1 < a0 < 1 + M − an−1. Since 1 < a0, 1 < a0 + a1 + · · · + an−1 and, in fact, 1 + M <
a0 + a1 + · · · + an−1. Thus N (F ) = 1 + M , since a0 < 1 + M . Therefore kW k∞ < N (F ).
Theorem 7.4. Given a polynomial p with a0 > 1. If N (F ) < kW k∞ then kW k∞ < 2N (F ).
14
KEVIN N. VANDER MEULEN, TREVOR VANDERWOERD
Proof . Suppose a0 > 1 and N (F ) < kW k∞. Let M = max{ak 1 ≤ k ≤ n − 1}. By Theorem 7.3,
a0 ≥ 1 + M − an−1. Then
Thus
Further,
a0 + an−1 ≥ 1 + M > 1 +
M
a0
.
kW k∞ = max(cid:26)a0 + an−1, 1 +
M
a0(cid:27) = a0 + an−1.
N (F ) = min {max {a0, 1 + M } , max {1, a0 + a1 + . . . + an−1}} = max {a0, 1 + M } .
Suppose N (F ) = a0. Then an−1 + 1 ≤ 1 + M ≤ a0. Thus kW k∞ < 2a0. And hence kW k∞ < 2N (F ).
Suppose N (F ) = 1 + M . Then a0 ≤ 1 + M . Further an−1 ≤ M < 1 + M . Thus a0 + an−1 < 2(1 + M ).
Therefore kW k∞ < 2(1 + M ) = 2N (F ).
In this section we compared upper bounds using the inverse of certain unit sparse companion matrices
corresponding to the monic reversal polynomial. Future work could explore other cases, as we have only
explored the case when u = 0 in line (13). In this case, as was observed in [3], significant improvement can
be made compared to the Frobenius bounds. In particular, we observed that kW k∞ can give a better bound
than N (F ), and that when N (F ) < kW k∞, it will not be an improvement by more than a factor of two.
Further work could be done on exploring the lower bounds using the inverse of an intercyclic companion
matrix. As with the upper bounds, there will be many options to consider, given the structure of the inverse
companion matrix in line (13). De Ter´an, Dopico, and P´erez [3, Example 5.8] give an example to illustrate
that the lower bound using a Fiedler companion matrix can be significantly better than using a Frobenius
companion matrix.
8. Concluding comments
It should be noted that there is a gap in the conditions of Theorem 7.1, namely the case when a0 > 1
and an−1 < M . Under these conditions, we suspect that there is no single matrix whose 1-norm provides
a sharper bound than those provided by the 1-norms of other matrices of type Ec(p♯)−1. However, the
following matrices of type En−2(p♯)−1 are useful. In particular, if 1 ≤ b ≤ n − 2, define
−an−1
· · · −ab+1
O
0 −a0
In−2
O O
O
ab/a0
· · ·
a1/a0
1
0
(15)
Xb =
Note that Xb is the inverse of the matrix
(16)
In−2
O
− 1
a0
− an−1
a0
− ab
a0
· · · − ab+1
a0
· · · − a1
a0
O
O
1
0
which is a Fiedler matrix of the monic reversal polynomial p♯. In the following result, we consider a particular
Xβ whose 1-norm is derived from the first column, or one of the last β + 1 columns.
.
,
BOUNDS ON POLYNOMIAL ROOTS
USING INTERCYCLIC COMPANION MATRICES7
15
Theorem 8.1. Let M = max{ak 1 ≤ k ≤ n − 1} and 1 ≤ c ≤ n − 2. Suppose 1 < a0, an−1 < M and
kXβk1 ≤ kXbk1 for all b, 1 ≤ b ≤ n − 2. If
then kXβk1 ≤ kAk1 for every matrix A of type Ec(p♯)−1.
kXβk1 ∈(cid:26)1 + an−1, 1 +(cid:12)(cid:12)(cid:12)(cid:12)
aβ−1
a0 (cid:12)(cid:12)(cid:12)(cid:12)
, . . . , 1 +(cid:12)(cid:12)(cid:12)(cid:12)
a1
a0(cid:12)(cid:12)(cid:12)(cid:12)
, a0(cid:27) ,
Proof . Suppose kXβk1 = 1 + an−1. Observe that kAk1 ≥ 1 + an−1 as an−1 must be in the location (1, 1).
Suppose kXβk1 = a0. Observe that kAk1 ≥ a0 as a0 must be in the location (1, n).
or kAk1 ≥ 1 + ak > 1 +(cid:12)(cid:12)(cid:12)
.
ak
a0(cid:12)(cid:12)(cid:12)
Theorem 8.2. Given M = max{ak 1 ≤ k ≤ n − 1} and m = max{k M = ak}. Suppose b ≤ n − 2.
ak
ak
a0(cid:12)(cid:12)(cid:12)
a0(cid:12)(cid:12)(cid:12)
where 1 ≤ k ≤ β − 1. Observe that kAk1 ≥ 1 +(cid:12)(cid:12)(cid:12)
Suppose kXβk1 = 1 +(cid:12)(cid:12)(cid:12)
Then kXbk1 < N (F ) if and only if 1 < a0 < 1 + M and, for some b ≥ m, ab+1 +(cid:12)(cid:12)(cid:12)(cid:12)
a0 (cid:12)(cid:12)(cid:12)(cid:12)
kXbk1 = max(cid:26)a0, 1 + an−1, . . . , 1 + ab+2, 1 + ab+1 +(cid:12)(cid:12)(cid:12)(cid:12)
, 1 +(cid:12)(cid:12)(cid:12)(cid:12)
Proof . Suppose b ≤ n − 2.
a0(cid:12)(cid:12)(cid:12)(cid:12)
Suppose kXbk1 < N (F ).
ab−1
ab
< M.
ab
a0(cid:12)(cid:12)(cid:12)(cid:12)
, . . . , 1 +(cid:12)(cid:12)(cid:12)(cid:12)
a1
a0(cid:12)(cid:12)(cid:12)(cid:12)
(cid:27) .
Suppose 1 + M ≤ a0. Then 1 < a0 + a1 + · · · + an−1 and N (F ) = a0. Since kXbk1 ≥ a0 = N (F ),
this a contradiction. Therefore a0 < 1 + M . This also implies that N (F ) ≤ 1 + M .
Suppose a0 ≤ 1. Then (cid:12)(cid:12)(cid:12)(cid:12)
ai
a0(cid:12)(cid:12)(cid:12)(cid:12)
≥ ai for all i, so that
kXbk1 ≥ 1 + M ≥ N (F )
which is a contradiction. Therefore a0 > 1.
With these two conditions, we have N (F ) = 1 + M . Since kXbk1 < N (F ), it follows that m ≤ b.
ab
For the converse, suppose ab+1 +(cid:12)(cid:12)(cid:12)(cid:12)
implies N (F ) = 1 + M . Since m ≤ b, 1 + ai < 1 + M for all i ≥ b + 2. Further, 1 +(cid:12)(cid:12)(cid:12)(cid:12)
i ≤ b − 1 since a0 > 1. It follows that kXbk1 < 1 + M = N (F ).
a0(cid:12)(cid:12)(cid:12)(cid:12)
< 1 + M for all
ai
a0(cid:12)(cid:12)(cid:12)(cid:12)
< M for some b ≥ m and 1 < a0 < 1 + M . The latter condition
Example 8.3. Consider the polynomial
p = x8 + 4x7 − x6 + x5 + 17x4 + 20x3 − 10x2 − 5x + 5.
Observe that a0 = 5 > 1 and an−1 = 4 < 20 = M , so we know that Theorem 7.1 part b) applies.
One can check that kX5k1 = a0 and kX5k1 ≤ kXbk1 for all b, 1 ≤ b ≤ n − 2. Hence, by Theorem 8.1
kX5k1 ≤ kAk1 for every matrix A of type Ec(p♯)−1. Further, since 1 < a0 = 5 < 1 + M − a0 = 16 the
= 1.2 < 20 = M for
b = 5 ≥ 4, the conditions of Theorem 8.2 are satisfied. Therefore, we should expect that W and X5 provide
sharper bounds than all other inverse companion matrices in (13) with u = 0, including the Frobenius
condition in Theorem 7.3 is satisfied. And since a0 = 5 < 21 = 1 + M and a6 +(cid:12)(cid:12)(cid:12)
a0(cid:12)(cid:12)(cid:12)
a5
16
KEVIN N. VANDER MEULEN, TREVOR VANDERWOERD
companion matrix. We can verify this directly: N (F ) = min {max {5, 21} , max {1, 63}} = 21,
W =
and X5 =
1
−4 0 0
0 0
−1 1 0
−2 0 1
0 0
3.4 0 0
0.2 0 0
−0.2 0 0
4
0 0
0 0
0 0
0 0
1 0
0 1
0 0
0 0
0
0
0
0
0
0
1
0
0 −5
0
0
0
0
0
0
0
0
0
0
0
0
1
0
−4
1
0
0
0
0
0
0
1
0
1
0
0
0
0
0 0
0 0
0 0
1 0
0 1
0 0
0 0
0 −5
0
0
0
0
0
0
0
0
0
0
0
0
0.2 3.4 4 −2 −1 1
0
0
0
0
0
0
0
1
0
0
0
0
0
1
0
.
Both W and X5 provide sharper bounds than N (F ), and the ratios are greater than two:
In this example, kX5k1 < kW k∞ but this may not always be the case.
N (F )
kW k∞
=
21
9
.
= 2.3
and
N(F )
kX5k1
=
21
5
= 4.2
Remark 8.4. Example 8.3 illustrates that kW k∞ and kXbk1 can be significantly smaller than N (F ). In
fact, the ratios can be made arbitrarily large. In particular, De Ter´an, Dopico, and P´erez [3, Example 5.6]
provide an example with kX1k1 = kW k∞ = 1 + 10m and N (F ) = 1 + 102m for any integer m > 0.
Since Xb is the inverse of a Fiedler matrix of a monic reversal polynomial, Theorem 5.4(b) of [3] implies that
if kXbk1 < kW k∞ for a given polynomial p, then kW k∞ is at most twice kXbk1. The following theorem is
a slight refinement for Xb.
Theorem 8.5. Let p be a polynomial as in (1). If 1 ≤ b ≤ n − 2 and a0 > 1, then
kW k∞
kXbk1
≤ 2 −
1
a0
.
Proof . Suppose kXbk1 < kW k∞. Then kW k∞ = a0 + an−1 since otherwise kW k∞ = 1 + M/a0 but
kXbk1 ≥ 1 + M/a0. Let U = max{1 + M/a0, 1 + an−1, a0}. Observe that kXbk1 ≥ U .
Suppose that U = a0. Then
kW k∞
kXbk1
≤
a0 + an−1
a0
≤
a0 + a0 − 1
a0
= 2 −
1
a0
.
Suppose that U = 1 + an−1. Then
kW k∞
kXbk1
≤
a0 + an−1
1 + an−1
= 1 +
a0 − 1
1 + an−1
≤ 1 +
a0 − 1
a0
= 2 −
1
a0
.
Suppose that U = 1 + M/a0. If a0 ≥ 1 + an−1 then
kW k∞
kXbk1
≤
a0 + an−1
1 + M/a0
≤
a0 + an−1
a0
≤ 2 −
1
a0
.
Otherwise, a0 < 1 + an−1, and
kW k∞
kXbk1
≤
a0 + an−1
1 + M/a0
≤
a0 + an−1
1 + an−1
= 1 +
a0 − 1
1 + an−1
< 1 +
a0 − 1
a0
= 2 −
1
a0
.
As a final note, we would like to point out that there is close connection between the unit sparse companion
matrices and the set of block Kronecker pencils described in [4].
In particular, if C(p) is a unit sparse
companion matrix associated with a monic polynomial p(x), then the matrix pencil λIn − C(p) is of the form
λIn − C(p) =(cid:20) Lm(λ)
K(λ) Ln−m−1(λ)T (cid:21) ,
O
BOUNDS ON POLYNOMIAL ROOTS
USING INTERCYCLIC COMPANION MATRICES9
17
with Lk(λ) a k × (k + 1) matrix of the form
Lk(λ) =
λ −1
0
0
...
0
λ −1
. . .
. . .
0
· · ·
0
· · ·
...
. . .
. . .
0
λ −1
.
This matrix pencil is permutationally similar to a block Kronecker pencil. The connection may provide
opportunities to extend this work to (monic) matrix polynomials.
Acknowledgement. Research supported in part by NSERC Discovery Grant 203336 and an NSERC USRA.
We thank a referee for a very careful reading of our paper and for pointing out the connection to [4] in the
conclusion.
References
[1] D.A. Bini and G. Fiorentino. Design, analysis, and implementation of a multiprecision polynomial rootfinder. Numerical
Algorithms 23.2-3 (2000) 127–173.
[2] B. Eastman, I.-J. Kim, B.L. Shader, and K.N. Vander Meulen. Companion matrix patterns. Linear Algebra Appl. 463
(2014) 255–272. See also, Corrigendum to 'Companion matrix patterns'. Linear Algebra Appl. (2017) to appear.
[3] F. De Ter´an, F.M. Dopico, and J. P´erez. New bounds for roots of polynomials based on Fiedler companion matrices. Linear
Algebra Appl. 451 (2014) 197–230.
[4] F.M. Dopico, P. Lawrence, J. P´erez, P. Van Dooren. Block Kronecker linearizations of matrix polynomials and their
backward errors. Preprint. Available as MIMS EPrint 2016.34, School of Mathematics, The Univeristy of Manchaster, UK.
[5] M. Fiedler. A note on companion matrices. Linear Algebra Appl. 372 (2003) 325–331.
[6] C. Garnett, B.L. Shader, C.L. Shader, and P. van den Driessche. Characterization of a family of generalized companion
matrices. Linear Algebra Appl. 498 (2016) 360–365.
[7] R.A. Horn and C.R. Johnson. Matrix Analysis. Cambridge University Press, Cambridge 1985.
[8] M. Marden. Geometry of Polynomials. 2nd ed. American Mathematical Society, Providence 1985.
[9] A. Melman. Modified Gershgorin disks for companion matrices. SIAM Review 54.2 (2012) 355–373.
Department of Mathematics, Redeemer University College, Ancaster, ON, L9K 1J4, Canada.
E-mail address: [email protected], [email protected]
Current address (T. Vanderwoerd): Department of Civil and Environmental Engineering, University of Waterloo, ON, N2L
3G1, Canada.
|
1306.4286 | 2 | 1306 | 2016-02-23T15:40:23 | Rings associated to coverings of finite p-groups | [
"math.RA"
] | In general the endomorphisms of a non-abelian group do not form a ring under the operations of addition and composition of functions. Several papers have dealt with the ring of functions defined on a group which are endomorphisms when restricted to the elements of a cover of the group by abelian subgroups. We give an algorithm which allows us to determine the elements of the ring of functions of a finite $p$-group which arises in this manner when the elements of the cover are required to be either cyclic or elementary abelian of rank $2$. This enables us to determine the actual structure of such a ring as a subdirect product. A key part of the argument is the construction of a graph whose vertices are the subgroups of order $p$ and whose edges are determined by the covering. | math.RA | math |
1 2
Rings associated to coverings of finite p-groups
1Department of Mathematics, Southeastern Louisiana University, Hammond, LA
Gary WALLS1, Linhong WANG2
2Department of Mathematics, University of Pittsburgh, Pittsburgh, PA
70402 USA
15260 USA
Abstract. In general the endomorphisms of a non-abelian group do not form a
ring under the operations of addition and composition of functions. Several papers
have dealt with the ring of functions defined on a group which are endomorphisms
when restricted to the elements of a cover of the group by abelian subgroups. We give
an algorithm which allows us to determine the elements of the ring of functions of a
finite p-group which arises in this manner when the elements of the cover are required
to be either cyclic or elementary abelian of rank 2. This enables us to determine the
actual structure of such a ring as a subdirect product.
A key part of the argument is the construction of a graph whose vertices are the
subgroups of order p and whose edges are determined by the covering.
1. Introduction
Covers of groups by subgroups and rings of functions that act as endomorphisms
on each subgroups are studied in many papers including [1, 2, 4, 5].
1.1. Definition. Suppose that G is a group and C is a collection of subgroups of G.
We say that C is a cover of G provided SC∈C C = G.
If all the elements of C have a certain property γ, we say that C is a γ-covering of G.
It is well known, e.g. [3], that the endomorphisms of a non-abelian group G do not
necessarily form a ring under the operations of function addition and composition.
Coverings by abelian subgroups are used to obtain rings of functions on G.
1.2. Definition. Let G be a group and C be an abelian-covering of G. Define
RC(G) = {f : G → G
for each C ∈ C, f C ∈ End(C)}.
1 Correspondence: [email protected]
2010 Mathematics Subject Classification 16S60
Key words and phrases: finite p-group, covers of groups, rings of functions
2Research of the second author supported by the Louisiana BoR [LEQSF(2012-15)-RD-A-20]
2
Note that RC(G) does form a ring under the natural operations on functions, since
functions in RC(G) are endomorphisms when restricted to the subgroups of the cover
C. The rings RC(G) are used in [5] to classify the maximal subrings of the nearring
M0(G) of the zero preserving functions defined on G.
Let p be a prime and G be a finite p-group. In this paper we consider the particular
case (∗) where all the subgroups in C are either maximal cyclic p-groups of G or are
elementary abelian of order p2. Let C be a ∗-covering of a finite p-group G. We prove
the following structure theorem to describe the rings arising as RC(G)'s.
Theorem Let G be a finite p-group, C be a ∗-covering of G. Then RC(G) is isomor-
phic to a direct product of rings isomorphic to M2(Zp) or rings of the form of 2.3.
A key part of our approach is a graph defined in 3.1. The vertices of the graph
are the subgroups of G of order p and the edges are determined by the particular
covering used. Each function in RC(G) is defined on the cyclic subgroups of G. This
definition is determined using a specific matrix and associated vector of tuples, even
though f may not be linear. In 3.8-3.10, a few examples are provided to illustrate the
theorem. We use the structure theorem to determine conditions for rings arising as
RC(G)'s to be of special types. In particular, when the rings are simple we see that
the ring RC(G) must be isomorphic to either Zp or M2(Zp). A similar result using a
different technique appears in [1], where covers by subgroups of order p2 are used for
finite p-groups of exponent p.
Throughout this paper, we always assume that G is a finite p-group and C is a
∗-covering of G. We refer to the subgroups in C as elements of C or cells in C.
2. Some particular rings
In this section we present some particular rings needed in order to state the conclu-
sion of our main result. For any positive integer n, the endomorphism ring End(Zpn) is
ring-isomorphic to Zpn, and is a simple ring if and only if n = 1. Further, End(Zp×Zp)
is isomorphic to M2(Zp), the ring of 2 × 2 matrices over Zp, and so is always simple.
In addition, we will need a subdirect product ring in 2.3, which is developed by first
constructing the matrix ring N m+n
i1, i2, ..., in in 2.1 and the ring RΛ(K) in 2.2.
2.1. Given integers m > 0 and n ≥ 0, we define a ring of (m + n) × (m + n)- matrices
as follows
N m+n
i1, i2, ..., in = (cid:26)(cid:20) λIm J(ν1, · · · , νn)
0 D(µ1, · · · , µn) (cid:21) λ, ν1, . . . , νn, µ1, . . . , µn ∈ Zp(cid:27)
where
3
D(µ1, · · · , µn) =
µ1
...
0
· · ·
0
...
. . .
· · · µn
and J(ν1, · · · , νn) is an m × n matrix which has value νj at the (ij, j) entry, for
1 ≤ j ≤ n and 1 ≤ i1, . . . , in ≤ m, and zeroes elsewhere. Note that there is at most
one nonzero entry in each column of J(ν1, · · · , νn). It is easy to see that N m+n
is a ring and that
i1, i2, ..., in
I m+n
m = (cid:26)(cid:20) λIm 0
0 (cid:21) λ ∈ Zp(cid:27)
0
is an ideal of N m+n
i1, i2,..., in, and so N m+n
i1, i2, ..., in is never a simple ring, if n > 0.
2.2. Let K be a subgroup of order p in G. There are maximal cyclic subgroups
of G that contain K. Assuming there is more than one, let Λ(K) be the directed
downward lattice of these cyclic subgroups containing K. Define S(K) to be the set of
functions on Λ(K) that are endomorphisms when restricted to the vertices of Λ(K).
On each maximal subgroup in Λ(K) of order pn, these functions are multiplications
by elements of Zpn. As we move down from one vertex to the one below, these
functions are multiplied by p. Obviously, these functions will agree on the vertices of
Λ(K). Each function in S(K) can be defined by beginning with an element λ ∈ Zp,
which determines an endomorphism on K, and then pulling it back up the vertices in
Λ(K). Thus, for a fixed λ ∈ Zp, such a function can be represented as an appropriate
tuple x = (λ1, · · · , λφ(K)) where φ(K) is the number of the maximal cyclic subgroups
in Λ(K) and where each entry λi ∈ Zpni determines the endomorphism on a maximal
cyclic subgroup of order pni in Λ(K) such that the properties discussed above hold.
The set of these functions, associated with K and λ, is denoted as RΛ(K), λ. By
this notation, we allow the trivial case when Λ(K) is a singleton and RΛ(K), λ is the
same as {(λ)}. For each subgroup K of G of order p contained in a maximal cyclic
subgroup, the set RΛ(K) = {RΛ(K), λ for λ ∈ Zp} does form a ring.
2.3. A subdirect product of rings can be formed from rings discussed in 2.1 and 2.2.
For any matrix in the ring N m+n
i1, i2, ..., in of 2.1 and certain selected subgroups K1, . . . , Km
of G of order p, we associate to the diagonal entries λ, . . . , λ, µ1, . . . , µn some tuples
4
from RΛ(Ki), λ for i = 1, . . . , m and tuples (µ1), . . . , (µn), respectively. That is
(cid:20) λIm J(ν1, · · · , νn)
0 D(µ1, · · · , µn) (cid:21) ,
x1
...
xm
(µ1)
...
(µn)
,
where each xi ∈ RΛ(Ki),λ. The arrays constructed in this way form a subdirect product
of rings N m+n
i1, i2, ..., in and RΛ(K1), . . . , RΛ(Km). In particular, if m = 1 and n = 0, the
subdirect product is isomorphic to a ring of the form of RΛ(K), which is isomorphic
to a direct product of Zpn for various integers n.
3. Determining the elements of the ring RC(G)
One of the main concerns in determining functions of RC(G) is to make sure that
they are well-defined. We introduce the following graph. The purpose of using such
a graph is reflected in Corollary 3.3, which is a direct consequence of Lemma 3.2.
This lemma appeared in [1]. For completeness, we include its proof.
3.1. Definition. Let Tp(G) denote the set of subgroups of G of order p. Let G be the
graph whose set of vertices is Tp(G). Two vertices A, B are joined in G by an edge
provided that there is a cell C ∈ C such that A, B ⊂ C and there exist C1, C2, C3 ∈ C
with intersections C ∩ C1, C ∩ C2, and C ∩ C3 all distinct subgroups of order p. We
call this graph the 3-intersecting graph of G. For A ∈ Tp(G), we let [A] denote the
G-connected component of G which contains A, and let
denote the set of connected components of G.
Con(G) = (cid:8)[A] A ∈ Tp(G)(cid:9)
3.2. Lemma. ( cf. [1, Lemma 6.2]) Suppose A and B are two distinct subgroups in
Tp(G) connected by an edge in the 3-intersecting graph G. Then for any f ∈ RC(G)
there is a λ ∈ Zp such that f (x) = λx for any x in the cell C = A × B.
Proof. Since A and B are connected by an edge in G, there is a C ∈ C so that
C = A × B ∼= Zp × Zp and there exist C1, C2, C3 ∈ C so that C ∩ C1 = he1i, C ∩ C2 =
he2i, and C ∩ C3 = he3i are three distinct subgroups of order p. For any f ∈ RC(G),
it is clear that he1i, he2i, he3i must be f -invariant. Hence f (ei) = λiei, for some
λ1, λ2, and λ3 ∈ Zp. Note that C = he1i × he2i and so e3 = µ1e1 + µ2e2 for some
nonzero µ1, µ2 ∈ Zp. It follows that f (e3) = f (µ1e1 + µ2e2) = µ1f (e1) + µ2f (e2) =
µ1λ1e1 + µ2λ2e2. This must equal λ3e3 = λ3(µ1e1 + µ2e2) = λ3µ1e1 + λ3µ2e2. Since
5
{e1, e2} is a basis of C, we get λ3µ1 = λ1µ1 and λ3µ2 = λ2µ2. It follows that λ1 = λ2,
and so f C is scalar multiplication for any f ∈ RC(G).
(cid:3)
3.3. Corollary. Suppose A ∈ Tp(G) and [A] > 1. Then f ∪[A] is multiplication by a
scalar λ in Zp.
3.4. Partition. Note that some of the connected components in Con(G) may be
singletons, as cells may not have three distinct intersections. We partition the cells
in C based on their intersections with other cells in C. Set C =
3
Gi=0
Ci, where
C0 = {C ∈ C C ∩ C ′ = {0} for any C ′ ∈ C and C ′ 6= C}
C3 = {C ∈ C C1 ∩ C, C2 ∩ C, and C3 ∩ C are all distinct subgroups of order
p for some C1, C2, C3 ∈ C}
C2 = {C ∈ C \ C3 C1 ∩ C and C2 ∩ C are distinct subgroups of order p for
some C1, C2 ∈ C}
C1 = C \ (C0 ∪ C2 ∪ C3).
Note that a cell C ∈ C2 may have more than two cells intersecting with it, for example,
C1 ∩ C and C2 ∩ C = C3 ∩ C are distinct subgroups of order p for some C1, C2, C3 ∈ C.
The above partition reveals the structure of a cover C that we will use to prove the
main result.
3.5. We will show constructively how a function f ∈ RC(G) can be defined on G
w.r.t. a chosen cover C. First denote a function from Con(G) to Zp by F . Let x be
an element of order p in G. If x belongs to a cyclic cell, then hxi is f -invariant and
we define f (x) = F ([hxi])x. It is clear that any cyclic cell can only belong to either
C0 or C1. If x belongs to a non-cyclic cell, there are several cases.
Case 1.
F ([hxi])x.
If x ∈ C for some C ∈ C3, following Corollary 3.3, we define f (x) =
Case 2. If x ∈ C for some C ∈ C2, then there are C1, C2 ∈ C such that C ∩ C1 =
2(C) of order p. Thus
2(C) for some α, β ∈ Zp. Note that both
2(C)i for some element e2(C) and e′
2(C)i and x = αe2(C) + βe′
he2(C)i, C ∩ C2 = he′
C = he2(C)i × he′
he2(C)i and he′
2(C)i are f -invariant. Hence we have
f : (e2(C), e′
2(C)) 7→ (e2(C), e′
2(C))(cid:18) F ([he2(C)i])
0
0
F ([he′
2(C)i]) (cid:19)
and f (x) can be defined accordingly. Note that e2(C) and e′
each C ∈ C2.
2(C) are symmetric for
Case 3. If x ∈ C for some noncyclic cell C in C1, then C ∩ C1 = he1(C)i for some
element e1(C) ∈ C and C1 ∈ C. Pick b1(C) ∈ C such that C = he1(C)i × hb1(C)i.
The choice of b1(C) is not unique, but C is the unique cell in C which contains b1(C).
6
Suppose x = αe1(C) + βb1(C) for some α, β ∈ Zp. Note that f (b1(C)) ∈ C and
he1(C)i is f -invariant. Hence we have
f : (e1(C), b1(C)) 7→ (e1(C), b1(C))(cid:18) F ([he1(C)i]) H(b1(C))
F ([hb1(C)i]) (cid:19)
0
where H(b1(C)) is a scalar in Zp. Then we define
f (x) = (cid:16)αF(cid:0)[he1(C)i](cid:1) + βH(b1(C))(cid:17)e1(C) + F ([hb1(C)i])b1(C).
We point out here that if a different choice had been made for b1(C), it is possible to
get the same value for f (x) by choosing a different scalar H(b1(C)).
Case 4.
If x ∈ C for some noncyclic cell C in C0, then C = hb0(C)i × hb′
for some b0(C), b′
Suppose x = αb0(C) + βb′
must be in C. Hence,
0(C) ∈ C. The choice of the basis {b0(C), b′
0(C) for some α, β ∈ Zp. Note that f (b0(C)) and f (b′
0(C)i
0(C)} is not unique.
0(C))
f : (b0(C), b′
0(C)) 7→ (b0(C), b′
0(C))(cid:18) F ([hb0(C)i]) B(b′
A(b0(C))
F ([hb′
0(C))
0(C)i]) (cid:19)
where A(b0(C)) and B(b′
0(C)) are scalars in Zp. Then f (x) can be defined accordingly.
After setting the image of f on any element of order p in G, now we extend f to
the elements of order bigger than p, if there are any. Let hyi be a maximal cyclic
subgroup of G and y = pn with n > 1. Note that hyi must be a cell in C. Then
f hyi ∈ End(hyi) ∼= End(Zpn) ∼= Zpn, and so f (y) = λy for some λ ∈ Zpn. Recall we
have already defined f (ypn−1
is an element of order
p. Simply working our way up the lattice of the cyclic subgroup hyi, we can choose a
proper scalar λ such that f (y) = λy and f (ypn−1
. Notice that as
we work our way up the lattice we have choices, but each choice leads to a different
function f ∈ RC(G).
, as x = ypn−1
) = F ([hypn−1
) = F ([hypn−1
i])ypn−1
i])ypn−1
It is clear that every function f in RC(G) arises in the above fashion, subject to the
cover C (mainly the intersections of the cells in C such as e2(C), e′
2(C), e1(C)), the
elements of the form of b0(C), b′
0(C), b1(C) as described above, and the choices of the
the function F and scalars H(b1(C)), A(b0(C)), B(b′
0(C)). In terms of notation, ei(C)
for some integer i is always an intersection or contained in an intersection of at least
two cells. Note that, by our notation, it may occur that he1(C1)i = he2(C2)i = K
when C1 is a cell in C1, C2 is a cell in C2, and K = C1 ∩ C2. Therefore, to determine
f , we need a set of elements of order p (indeed subgroups of order p) which includes
generators of any noncyclic cell in C and the unique element of order p (indeed the
p-socle) of any cyclic cell.
3.6. Setup. Given a cover C of G, we setup the following sets of subgroups of order
p. The union of these sets is denoted by B(C).
7
2(C)i C ∈ C2}
B3(C) = {hgi g ∈ C for some C ∈ C3 and hgi = C ∩ C ′ for some C ′ ∈ C}
B2(C) = {he2(C)i, he′
B1
1(C) = {he1(C)i C ∈ C1}
B2
1(C) = {hb1(C)i C ∈ C1 and C = he1(C)i × hb1(C)i for e1(C) ∈ B1
B0(C) = {hb0(C)i, hb′
0(C)i C ∈ C0}
1(C)}
As illustrated in 3.5, we also need the following functions.
F : Con(G) → Zp
H : {b1(C) hb1(C)i ∈ B2
1(C)} → Zp
A : {b0(C) hb0(C)i ∈ B0(C)} → Zp
B : {b′
0(C)i ∈ B0(C)} → Zp
0(C) hb′
3.7. Theorem. The ring RC(G) with a chosen covering C is isomorphic to a direct
product of matrix rings isomorphic to M2(Zp) or rings of the form of 2.3.
Proof. With the Zp-valued functions and sets of subgroups of order p described in 3.6,
any function f ∈ RC(G) can be defined as illustrated in 3.5. To prove the theorem,
we represent the way f is defined on the elements of B(C) in terms of a matrix with
blocks, which will be denoted by [f ]B(C). For this purpose, the subgroups in B(C)
need to be put in a certain order. The resulting ordered set will be denoted by A(C).
We start with hgi ∈ B3(C) if B3(C) is not empty. Let Dhgi be the set of the elements
hb1(C)i such that hyi × hb1(C)i is a noncyclic cell C ∈ C1 for some hyi ∈ [hgi] ∩ B3(C).
Suppose hbi1i, . . . , hbilii are from Dhgi associated to hyii ∈ [hgi] ∩ B3(C) for integers li
and i = 1, . . . , m. The rest of the subgroups hym+1i, . . . hyki ∈ [hgi] ∩ B3(C), if there
are any (i.e., k ≥ m), are either contained in a cyclic cell or in a noncyclic cell C ∈ C2.
Let Ag = {hy1i, . . . , hymi, hz11i, . . . , hz1 l1i, . . . hzm1i, . . . , hzmlmi, hym+1i, . . . hyki}, an
ordered set of subgroups from B(C).
The matrix block of [f ]B(C) corresponding to Ag can be determined as shown in 3.5
case 1 and case 3. Set λ = F ([hyii]) for i = 1, . . . k, µt = F ([hziji]) and νt = H(zij)
i=1 li. Following the notation in 2.1, we see that
for i = 1, . . . m and t = 1, . . . , n = Pm
the matrix block having the form of
(cid:20) λIm J(ν1, · · · , νn)
0 D(µ1, · · · , µn) (cid:21)
0
0
λIk−m
.
It is clear that Ag is the first part of the ordered set A(C). Of course, before we
pursue further, the set B(C) should be updated by removing the subgroups from the
8
set A(C) = Ag. That is, subsets B3(C), B1
1(C), and B0(C) of B(C) are updated
accordingly. Then we exhaust the set B3(C) by repeating the same process with other
subgroups hg′i ∈ B3(C). Clearly, [hg′i] 6= [hgi]. Each time, the ordered set A(C) is
expanded by sets Ag′, . . ., while the subsets of B(C) are updated accordingly.
1(C), B2
Next, we move on to the set B2(C). If a subgroup he2(C)i from B2(C) also belongs
to other noncyclic cells in C1, then we add he2(C)i and all hb1(C)i such that he2(C)i×
hb1(C)i is a noncyclic cell in C1. If not, we simply add he2(C)i. Again, each time
B2(C) needs to be updated. In terms of [f ]B(C), as shown in 3.5 and 2.1, these two
cases correspond to a matrix block, with n being zero or positive, having the form of
(cid:20) λ J(ν1, · · · , νn)
0 D(µ1, · · · , µn) (cid:21) .
Then we continue with subgroups in B1
he1(C)i from B1
in a matrix block of the form
1(C) and those hb1(C)i from B2
1 and B2
1 in a similar way such that adding
1(C) for a fixed noncyclic cell C results
(cid:20) λ J(ν1, · · · , νn)
0 D(µ1, · · · , µn) (cid:21) .
We finish the process by adding the subgroups hb0(C)i and hb′
0(C)i from B0 in pairs
or just hb0(C)i if it is from a cyclic cell and not paired with any other subgroup of
order p in B0. Each pair corresponds to a 2 × 2 block in M2(Zp) as shown in 3.5 case
4. Any single subgroup of order p corresponds to a 1 × 1 block [λ] for some λ ∈ Zp.
To summarize, we have an ordered set A(C) of subgroups of G of order p, under
which the definition of any function f ∈ RC(G) on elements of G of order p is
determined and represented in terms of a matrix fB(C) with blocks as described above,
i.e., blocks from M2(Zp) and blocks of the form described in 2.1 with choices of λ's,
µi's and νi's from Zp.
It is not hard to see that the ordered set A(C) may not
be unique, but the number and shape of the matrix blocks in fB(C) must be fixed,
corresponding to the chosen cover C. Note that any such block matrix with nonzero
scalar entries from Zp contains enough information to define a function in RC(G) on
elements of G of order p. The collection of these matrices, with respect to a chosen
cover C, does form a ring, which is a direct product of rings isomorphic to M2(Zp) or
N m+n
i1, i2, ..., in as in 2.1.
To fully represent a function f ∈ RC(G), additional information needs to be at-
tached to the matrix fB(C) so that f is defined for elements of order pn with n ≥ 2.
We have discussed the definition of f on these elements in the last part of 3.5. The
p-socle of these cyclic cells must appear in the list A(C). Following the discussion
and notation in 2.2, we associate the diagonal elements of each matrix block of the
9
form, allowing n = 0,
0 D(µ1, · · · , µn) (cid:21)
(cid:20) λIm J(ν1, · · · , νn)
an (m+n)×1 vector (x1, . . . , xm, xm+1, . . . , xm+n)T , where xi are tuples from RΛ(Ki), λ
for i = 1, . . . m and for the corresponding subgroup Ki of order p in A(C).
If a
subgroup Ki does not belong to a cyclic cell of order greater than p, then RΛ(Ki), λ =
{(λ)} as pointed in 2.2. Note that each xm+j = (µj), since the subgroups hb1(Ci)i
of order p corresponding to µ1, · · · , µn can only belong to the noncyclic cells Ci as
shown in 3.5 case 3. There is no need to associate any vector of tuples to the diagonal
of matrix blocks from M2(Zp) because these blocks are corresponding to noncyclic
cells in C0.
Finally, each extended matrix contains enough information to define a function in
RC(G). Therefore, RC(G) is isomorphic to a direct product of rings isomorphic to
M2(Zp) or rings of the form of 2.3. The proof is complete.
(cid:3)
3.8. Example. Let G = Q8 = hx, y x2 = y2, x4 = 1, y−1xy = x−1i, the quaternion
group of order 8. Then G has only one subgroup of order 2. The only ∗-covering of
G is C = {hxi, hyi, hxyi}. Hence
Note that RC(G) = 16.
RC(G) ∼= n(a, b, c) (cid:12)(cid:12)(cid:12)
a, b, c ∈ Z4, 2a = 2b = 2co .
3.9. Example. Let G be Q8 × Z2 = hx, yi × hwi. Now G has only one non-cyclic
subgroup of order 4 and exactly 6 cyclic subgroups of order 4. Consider two ∗-covers
C = {hxi, hyi, hxyi, hxwi, hywi, hxywi, hx2, wi}
and
D = {hxi, hyi, hxyi, hxwi, hywi, hxywi, hwi, hx2, wi}.
Then we have
RC(G) = (cid:26)(cid:18)(cid:20) λ1
0
d
λ2 (cid:21) ,
and
RD(G) =
λ1
0
0
0
λ2
0
0
0
λ3
(a, b, c)
(λ2) (cid:19) (cid:12)(cid:12)(cid:12)
,
(λ2)
(λ3)
(a, b, c)
λ1, λ2, d ∈ 2Z4, a, b, c ∈ Z4, 2a = 2b = 2c = λ1(cid:27)
(cid:12)(cid:12)(cid:12)
a, b, c ∈ Z4, λ1, λ2, λ3 ∈ 2Z4, 2a = 2b = 2c = λ1
.
Notice that RC(G) = RD(G) = 64.
3.10. Example. Let G = D8 × Z2, where D8 = hx, y x4 = 1, y2 = 1, (xy)2 = 1i is
the dihedral group of order 8 and Z2 = hwi. Take the ∗-cover
C = {hxi, hxwi, hxy, wi, hw, yi, hx2, xyi, hx2y, wi, hx2w, xyi}
10
Then
RC(G) = n(M, X) a, b ∈ Z4, λ1, b1, d1, c1, e1, h1, i1, f1, g1 ∈ 2Z4, 2a = 2b = λ1o,
where M =
and X =
λ1
0
0
0
0
0
0
0
b1 d1
0
c1
0
0
0
0
0
0
0
0
0
0
0
0
e1 h1
0
f1
0
0
0
0
0
i1
0
g1
(a, b)
(b1)
(c1)
(e1)
(f1)
(g1)
3.11. Remark. If RC(G) is a simple ring, it follows from Theorem 3.7 that RC(G)
must be isomorphic to either Zp or M2(Zp). A similar result appears in [1], where
covers by subgroups of order p2 are used for finite p-groups G of exponent p. An
intersection condition on the subgroups in the cover that is equivalent to both RC(G)
being simple and RC(G) ∼= Zp is developed, see [1, Theorem 6.10].
It is clear that RC(G) ∼= M2(Zp) only occurs when G ∼= Zp × Zp and C = {G}. Now
we derive what it takes for RC(G) to be isomorphic to Zp. Suppose that RC(G) ∼= Zp.
It then follows that G must have exponent p and that the 3-intersecting graph G
of G must be connected, as having more than one connected components leads to
non-trivial ideals. Take a nonzero element a ∈ G and let C be the cell containing a.
Since G > p2, there is an element b ∈ G \ C such that hai and hbi are adjacent in
G. Hence CG(a) ≥ p3. This motivates the following theorem.
3.12. Theorem. Suppose that G is a finite p-group of exponent p and CG(a) ≥ p3
for any element a ∈ G. Then there is a ∗-covering C of G such that RC(G) ∼= Zp.
Conversely, if G ≥ p3 and there is a ∈ G with CG(a) = p2 then RC(G) is not
simple for any ∗-covering C of G.
Proof. Suppose that b ∈ Z(G) and a is an element of G with a /∈ hbi. Since CG(a) ≥
p3, there is ca ∈ CG(a) \ ha, bi. Consider the cover
C = [a∈G\hbinha, cai, ha, bi, hb, cai, hab, caio.
Now we have ha, b, cai ∼= Zp × Zp × Zp. It follows that ha, bi ∩ ha, cai = hai, ha, bi ∩
hb, cai = hbi, and ha, bi ∩ hab, cai = habi are three distinct subgroups of order p.
Hence for all a ∈ G \ hbi, the subgroups hai and hbi are connected by an edge in G.
Therefore, RC(G) ∼= Zp.
On the other hand, if G ≥ p3 and there is an element a ∈ G with CG(a) = p2,
then hai and CG(a) are the only abelian subgroups of G which can contain a.
It
follows that any ∗-covering of G must contain one or the other. In either case RC(G)
is not simple.
(cid:3)
References
11
[1] Cannon G, Kabza L, Maxson C, Neuerberg K. Rings and covered groups II. Commun Algebra
2012; 40; 2808 -- 2824.
[2] Cannon G, Maxson C, Neuerberg K. Rings and covered groups. J Algebra 2008; 320; 1586 -- 1598.
[3] Jacobson N. Basic Algebra I, second ed. San Francisco, CA, USA; W. H. Freeman; 1985.
[4] Kreuzer A, Maxson C. E-locally cyclic abelian groups and maximal nearrings of mappings.
Forum Math 2006; 18; 107 -- 114.
[5] Maxson C, Smith K. Maximal rings associated with covers of abelian groups. J Algebra 2007;
315; 541 -- 554.
E-mail address: [email protected], [email protected]
Department of Mathematics, Southeastern Louisiana University, Hammond, LA
70402, Department of Mathematics, University of Pittsburgh, Pittsburgh, PA 15260
|
math/0612478 | 2 | 0612 | 2011-09-14T06:59:11 | The finite Rat-splitting for coalgebras | [
"math.RA",
"math.RT"
] | Let $C$ be a coalgebra. We investigate the problem of when the rational part of every finitely generated $C^*$-module $M$ is a direct summand $M$. We show that such a coalgebra must have at most countable dimension, $C$ must be artinian as right $C^*$-module and injective as left $C^*$-module. Also in this case $C^*$ is a left Noetherian ring. Following the classic example of the divided power coalgebra where this property holds, we investigate a more general type of coalgebras, the chain coalgebras, which are coalgebras whose lattice of left (or equivalently, right, two-sided) coideals form a chain. We show that this is a left-right symmetric concept and that these coalgebras have the above stated splitting property. Moreover, we show that this type of coalgebras are the only infinite dimensional colocal coalgebras for which the rational part of every finitely generated left $C^*$-module $M$ splits off in $M$, so this property is also left-right symmetric and characterizes the chain coalgebras among the colocal coalgebras. | math.RA | math |
THE FINITE RAT-SPLITTING FOR COALGEBRAS
MIODRAG CRISTIAN IOVANOV
Abstract. Let C be a coalgebra. We investigate the problem of when the rational
part of every finitely generated C ∗-module M is a direct summand M . We show that
such a coalgebra must have at most countable dimension, C must be artinian as right
C ∗-module and injective as left C ∗-module. Also in this case C ∗ is a left Noetherian
ring. Following the classic example of the divided power coalgebra where this property
holds, we investigate a more general type of coalgebras, the chain coalgebras, which are
coalgebras whose lattice of left (or equivalently, right, two-sided) coideals form a chain.
We show that this is a left-right symmetric concept and that these coalgebras have the
above stated splitting property. Moreover, we show that this type of coalgebras are the
only infinite dimensional colocal coalgebras for which the rational part of every finitely
generated left C ∗-module M splits off in M , so this property is also left-right symmetric
and characterizes the chain coalgebras among the colocal coalgebras.
Introduction
Let R be a ring and T be a torsion preradical on the category of left R-modules RM. Then
R is said to have splitting property provided that T (M ), the torsion submodule of M , is a
direct summand of M for any M ∈ RM. More generally, if C is a Grothendieck category
and A is a subcategory of C, then A is called closed if it is closed under subobjects, quotient
objects and direct sums. To every such subcategory we can associate a preradical t (also
called torsion functor) if for every M ∈ C we denote by t(M ) the sum of all subobjects of
M that belong to A. We say that C has the splitting property with respect to A if t(M ) is a
direct summand of M for all M ∈ C. In the case of the category of R-modules, the splitting
property with respect to some closed subcategory is a classical problem which has been
considered by many authors. In particular, when R is a commutative ring, the question
of when the (classical) torsion part of an R module splits off is a well known problem. J.
Rotman has shown in [Rot] that for a commutative domain the torsion submodule splits
off in every R-module if and only if R is a field. I. Kaplansky proved in [K1], [K2] that for
a commutative integral domain R the torsion part of every finitely generated R-module M
splits in M if and only if R is a Prufer domain. While complete results have been obtained
for commutative rings, the problem still remains wide open for the non-commutative case.
In this paper we investigate the situation when the ring R arises as the dual algebra of a
K-coalgebra C, R = C ∗. Then the category of the left R-modules naturally contains the
category MC of all right C-comodules as a full subcategory. In fact, MC identifies with
the subcategory Rat(C ∗M) of all rational left C ∗-modules, which is generally a closed
subcategory of C ∗M. Then two questions regarding the splitting property with respect
to Rat(C ∗M) naturally arise: first when is the rational part of every left C ∗-module
M a direct summand of M and when does the rational part of every finitely generated
C ∗-module M split in M . The first problem, the splitting of C ∗M with respect with the
Key words and phrases. Torsion Theory, Splitting, Coalgebra.
2000 Mathematics Subject Classification. Primary 16W30; Secondary 16S90, 16Lxx, 16Nxx, 18E40.
∗ This paper was partially supported by a CNCSIS BD-type grant and was written within the frame of
the bilateral Flemish-Romanian project "New Techniques in Hopf Algebra Theory and Graded Rings".
1
2
MIODRAG CRISTIAN IOVANOV
closed subcategory Rat(C ∗M) has been treated by C. Nasasescu and B. Torrecillas in [NT]
where it is proved that if all C ∗-modules split with respect to Rat then the coalgebra C
must be finite dimensional. The techniques used involve some amount of category theory
(localization in categories) and strongly relies on some general results of M.L.Teply from
[T1], [T2], [T3].
We consider the more general problem of when C has the splitting property only for
finitely generated modules, that is, the problem of when is the rational part Rat(M ) of
M a direct summand in M for all finitely generated left C ∗-modules M . We call these
coalgebras left finite Rat-splitting coalgebras (or we say that they have the left finite Rat-
splitting property). If the coalgebra C is finite dimensional, then every left C ∗-module
is rational so MC is equivalent to C ∗M and Rat(M ) = M for all C ∗-modules M and in
this case Rat(M ) trivially splits in any C ∗-module. Therefore we will deal with infinite
dimensional coalgebras, as generally the infinite dimensional coalgebras produce examples
essentially different from the ones in algebra theory. We first prove some general properties
for left finite Rat-splitting coalgebras, namely such a coalgebra C is artinian as right C ∗-
module and injective as left C ∗-module, it has at most countable dimension and has
finite dimensional coradical. Also C ∗ is a left Noetherian ring. We look at a very simple
example of a coalgebra where this property holds, namely, the divided power coalgebra
(see [DNR], Example 1.1.4), which has K[[X]] as its dual algebra. This is in some sense the
simplest possible example of infinite dimensional coalgebra that has the left (and right)
finite Rat-splitting property. We introduce and study (left) chain coalgebras to be the
coalgebras for which every two left subcomodules M, N satisfy either M ⊆ N or N ⊆ M .
We see that this is a left-right symmetric concept and we give a simple characterization
of these coalgebras as being exactly those having each factor of the coradical series a
simple comodule. Moreover, this gives a complete characterization of these coalgebras in
the case when the base field is algebraically closed: the divided power coagebra and its
subcoalgebras are the only ones of this type. We show that chain coalgebras have the (left
and right) finite Rat-splitting property. Then we investigate the colocal finite Rat-splitting
coalgebras. In the main result of the paper we show that a colocal coalgebra satisfying the
left finite Rat-splitting property must be a chain coalgebra, and therefore it also has the
right finite Rat-splitting property. This provides a characterization of the divided power
coalgebra over an algebraically closed field (or more generally of chain coalgebras) among
local coalgebras, namely they are exactly those coalgebras C for which the rational part
of every finitely generated left (or right) C ∗-module splits off.
1. Splitting Problem
Let C be a coalgebra with counit ε and comultimplication ∆. We use the Sweedler
convention ∆(c) = c1 ⊗ c2 where we omit the summation symbol. For a vector space V
and a subspace W of V denote by W ⊥ = {f ∈ V ∗ f (x) = 0, ∀ x ∈ W } and for a subspace
X ∈ V ∗ denote by X ⊥ = {x ∈ V f (x) = 0, ∀ f ∈ X}. If M is a right (or left) R-module
denote by LR(M ) (or RL(M )) the lattice of the submodules of M . Also, if S is another
ring and Q is a fixed R-S-bimodule, for any left R-module M we have applications
RL(M ) ∋ N → N ⊥ = {f ∈ HomR(M, Q) f (x) = 0, ∀ x ∈ N } ∈ LS(Hom(M, Q))
LS(Hom(M, Q)) ∋ X → X ⊥ = {x ∈ M f (x) = 0, ∀ f ∈ X} ∈ RL(M )
forming a Galois pair (see [AN]).
Lemma 1.1. Let C be a coalgebra. Then for any finitely generated right (or left, or
two-sided) submodule X of C ∗, (X ⊥)⊥ = X.
THE FINITE RAT-SPLITTING FOR COALGEBRAS
3
Proof. Put R = End(C C, C C). Then R is a ring with multiplication "·" equal to op-
posite composition of morphisms. Let M = C ∗C and Q = C ∗CR where the right R-
module structure on C is c · f = f (c).
It is not difficult to see that the isomorphism
of rings C ∗ ≃ End(C C, C C) = R, c∗ → (c 7→ c∗(c1)c2), transposes the problem to the
Galois correspondence X → X ⊥ between the left R module C and the right R module
EndC(C, C) = Hom(C ∗C, C ∗ C). That is, it is enough to prove the statement for finitely
generated right ideals of R. Suppose X ⊆ R is a right ideal generated by f1, . . . , fn as
right R module and let f ∈ R such that f X ⊥ = 0. Then we have X =
fi · R so
n
Pi=1
C
Kerfi
n
T
i=1
Kerfi. Then f induces a morphism f :
→ C and as C C
X ⊥ =
(fi · R)⊥ =
n
Ti=1
n
Ti=1
sequence
is injective, the canonic monomorphism 0 → C
C
Kerfi
gives rise to the exact
→
n
Li=1
n
T
i=1
Kerfi
n
Mi=1
HomC(
C
Kerfi
n
, C) ≃ HomC(
Mi=1
C
Kerfi
, C)
ϕ
→ HomC(
, C) → 0
C
Kerfi
n
Ti=1
n
Li=1
Let (gi)i=1,n ∈
HomC( C
Kerfi
monomorphism fi : C
any the diagram
Kerfi
, C) be such that ϕ( Pi=1,n
gi) = f . As for any i we have a
→ C induced by fi and as the right C-comodule C is injective,
0
✲
✲ C
C
fi
Kerfi
❄✛................................
h i
gi
C
can be completed commutatively by a morphism of right C-comodules hi. Then we have
fi · hi) = ϕ(
hi ◦ fi) = f and composing this with the cannonical projection
ϕ(
n
Pi=1
p : C → C
n
T
i=1
Kerfi
n
Pi=1
it is not difficult to see that we get f =
fi · hi so f ∈ X.
(cid:3)
n
Pi=1
Proposition 1.2. Let C be a coalgebra such that Rat(M ) splits off in any finitely generated
left C ∗-module M . Then any indecomposable injective left C-comodule E contains only
finite dimensional proper subcomodules.
Proof. Let T be the socle of E; then T is simple and E = E(T ) is the injective envelope
of T . We show that if K ⊆ E(T ) is an infinite dimensional subcomodule then K = E(T ).
Suppose K ( E(T ). Then there is a left C-subcomodule (right C ∗-submodule) K ( L ⊂
E(T ) such that L/K is finite dimensional. We have an exact sequence of left C ∗-modules:
0 → (L/K)∗ → L∗ → K ∗ → 0
As L/K is a finite dimensional left C-comodule, we have that (L/K)∗ is a rational left
C ∗-module; thus Rat(L∗) 6= 0. Also L∗ is finitely generated as it is a quotient of E(T )∗
which is a direct summand of C ∗. We have L∗ = Rat(L∗) ⊕ X for some left C ∗-submodule
X of L∗. Then Rat(L∗) is finitely generated because L∗ is, so it is finite dimensional.
4
MIODRAG CRISTIAN IOVANOV
As L is infinite dimensional by our assumption, we have X 6= 0. This shows that L∗ is
decomposable and finitely generated, thus it has at least two maximal submodules, say
M, N . We have an epimorphism E(T )∗ f
→ L∗ → 0 and then f −1(M ) and f −1(N ) are
distinct maximal C ∗-submodules of E(T )∗. But by [I], Lemma 1.4, E(T )∗ has only one
maximal C ∗-submodule which is T ⊥, so we have obtained a contradiction.
(cid:3)
Let C0 be the coradical of C, the sum of all simple subcomodules of C. By [DNR], Section
Ci and
3.1, C0 semisimple coalgebra that is a direct sum of simple subcoalgebras C0 = Li∈I
each simple subcoalgebra Ci contains only one type of simple left (or right) C-comodule;
moreover, any simple left (or right) C-comodule is isomorphic to one contained in a Ci.
Proposition 1.3. Let C be a coalgebra such that the rational part of every finitely gener-
ated left C ∗ module splits off. Then there is only a finite number of isomorphism types of
simple left C-comodules, equivalently, C0 is finite dimensional.
E(Ci)∗ as left C ∗-modules. As Si ⊆ E(Ci), we have epimorphisms of left C ∗-
Proof. By the above considerations, if Si is a simple left C-subcomodule of Ci, we
have that (Si)i∈I forms a set of representatives for the isomorphism types of simple left
C-comodules. Let S be a set of representatives for the simple right C-comodules. Let
E(Ci) be an injective envelope of the left C-comodule Ci included in C; then as C0 is
E(Ci) as left C-comodules or right C ∗-modules. Then
modules E(Ci)∗ → S ∗
essential in C we have C = Li∈I
C ∗ = Qi∈I
C ∗ → Qi∈I
C ∗ → QS∈S
S ∗
i → 0 and therefore we have an epimorphism of left C ∗-modules
i → 0. But there is a one-to-one correspondence between left and right simple
C-comodules given by {Si i ∈ I} ∋ S 7→ S ∗ ∈ S. Hence there is an epimorphism
S is finitely generated
S → 0, which shows that the left C ∗-module P = QS∈S
(actually generated by a single element). But then as Rat(C ∗ P ) is a direct summand in P ,
we must have that Rat(C ∗ P ) is finitely generated, so it is finite dimensional. Therefore,
S is a rational left C ∗-module which is naturally included in P , we have
as Σ = LS∈S
Σ ⊆ Rat(P ). This shows that LS∈S
S is finite dimensional so I must be finite. This is
equivalent to the fact that C0 is finite dimensional, because each Ci is a simple coalgebra,
thus a finite dimensional one.
(cid:3)
We shall say that a coalgebra is left (right) finite Rat-splitting if the rational part of any
finitely generated left (right) C ∗-module splits off.
Proposition 1.4. Let C be a left finite Rat-splitting coalgebra. Then the following asser-
tions hold:
(i) C is artinian as left C-comodule (equivalently, as right C ∗-module).
(ii) C ∗ is left Noetherian.
(iii) C has at most countable dimension.
(iv) C is injective as left C ∗-module.
Proof. (i) We have a direct sum decomposition C = Li∈F
E(Si) where C0 = Li∈F
decomposition of C0 into simple left C-comodules and E(Si) are injective envelopes of Si
contained in C. Then F is finite as C0 is finite dimensional. Also, by Proposition 1.2 the
E(Si) 's are artinian as they contain only finite dimensional proper subcomodules, thus C
is an artinian left C-comodule.
Si is the
THE FINITE RAT-SPLITTING FOR COALGEBRAS
5
(ii) Take I a left ideal of C ∗ and suppose it is not finitely generated; then we can find a
sequence (xk)k of elements of I such that denoting Ik = C ∗ < x1, x2, . . . , xk >, xk+1 /∈ Ik.
Then, corresponding to the ascending chain of left C ∗ submodules of C ∗, I1 ( I2 (
1 ⊇ I ⊥
· · · ( Ik ( . . . we have a descending chain of right C ∗ submodules of C, I ⊥
2 ⊇
k = I ⊥
· · · ⊇ I ⊥
k+1 = . . . .
k+1)⊥ = . . . and then by Lemma 1.1 we get that Ik = Ik+1, and then
Then (I ⊥
xk+1 ∈ Ik+1 = Ik which is a contradiction.
(iii) For any i ∈ F , if E(Si) is infinite dimensional, we may inductively build a sequence
k ⊇ . . . , which must be stationary as CC ∗ is artinian, so I ⊥
k )⊥ = (I ⊥
(xk)k of elements of E(Si) such that xk+1 /∈
xj · C ∗ for any k because
xk · C ∗, because Pk
always a finite dimensional comodule. Then E(Si) = Pk
infinite dimensional subcomodule of E(Si) and one can apply Lemma 1.2. Now, as each
xk · C ∗ is finite dimensional, the conclusion follows.
(iv) As C is a finite coproduct of E(Si)'s it is enough to prove that each E(Si) is injective
and by [I0] Lemma 2, it is enough to prove that E = E(Si) splits off in any left C ∗ module
M such as M/E is 1-generated, that is, it is generated by an element x ∈ M/E. Let
H = Rat(C ∗ · x); then there is X < C ∗ · x such that H ⊕ X = C ∗ · x. Then E + H is
a rational C ∗ module so (E + H) ∩ X = 0; also M = C ∗ · x + E so (E + H) + X = M ,
showing that E + H is a direct summand in M . But as E is an injective comodule, we
have that E splits off in E + H, thus E must split in M and the proof is finished.
(cid:3)
k
Pj=1
k
xj · C ∗ is
Pj=1
xk · C ∗ is an
2. Chain Coalgebras
Let C be a coalgebra and denote by C0 ⊆ C1 ⊆ C2 ⊆ . . . the coradical filtration of C,
that is, C0 is the coradical of C, and Cn+1 ⊆ C such that Cn+1/Cn is the socle of the right
(or left) C-comodule C/Cn for all n ∈ N. Then Cn is a subcoalgebra of C for all n, and
the same Cn is obtained whether we take the socle of the left C-comodule C/Cn or of the
right C-comodule C/Cn. Put C−1 = 0 and R = C ∗. By [DNR] we have Sn∈N
Cn = C.
Definition 2.1. We say that a coalgebra C is a left (right) chain coalgebra if and only if
the lattice of the left (right) subcomodules of C is a chain, that is, any two subcomodules
M, N of C are comparable (given two subsets A, B of a set X we say that A and B are
comparable if either A ⊆ B or B ⊆ A).
The following result shows that this definition is left-right symmetric and also characterizes
all chain coalgebras.
Proposition 2.2. The following assertions are equivalent for a coalgebra C:
(i) C is a right chain coalgebra.
(ii) Cn+1/Cn is either 0 or a simple (right) comodule for all n ≥ −1.
(iii) C is a left chain coalgebra. In this case C and Cn, n ≥ −1 are the only subcomodules
(left, right, two-sided) of C and Cn is finite dimensional for all n.
Proof. (ii)⇒(i) We prove that any subcomodule of C must be equal either to one of the
Cn's or to C. Let M be a right subcomodule of C and suppose M 6= C and M 6= 0.
Then there is n ≥ 0 such that Cn * M and let n the minimal natural number with this
property. Then we must have Cn−1 ⊆ M by the minimality of M and we show that
Cn−1 = M . Indeed, if Cn−1 ( M we can find a simple subcomodule of M/Cn−1. But
then Cn−1 6= C so Cn−1 6= Cn and as Cn/Cn−1 is the only simple subcomodule of C/Cn
we find Cn/Cn−1 ⊆ M/Cn−1, that is Cn ⊆ M , a contradiction. This also proves the last
6
MIODRAG CRISTIAN IOVANOV
statement of the proposition.
(i)⇒(ii) If Cn+1/Cn is nonzero and it is not simple then we can find S1 and S2 two distinct
simple modules contained in C/Cn. Then S1 = M1/Cn, S2 = M2/Cn and M1 ∩ M2 = Cn,
M1 6= Cn, M2 6= Cn because S1 ∩ S2 = 0 and S1 and S2 are distinct simple subcomodules
of C/Cn. But this shows that neither M1 ( M2 nor M2 ( M1 which is a contradiction.
(ii)⇔(iii) is proved similarly.
(cid:3)
0 ; by [DNR] Lemma 2.5.7 and Corollary 3.1.10 we have that J = J(C ∗) (the
Cn = C,
Denote J = C ⊥
Jacobson radical of C ∗) and (J n)⊥ = Cn−1, so J n ⊆ ((J n)⊥)⊥ = C ⊥
we see that Tn∈N
J n = 0.
n−1. As Sn∈N
Definition 2.3. We say that a coalgebra C is left almost finite if the left regular comodule
CC has only finite dimensional proper subcomodules.
By Proposition 2.2 a chain coalgebra is left and right almost finite.
Proposition 2.4. Let C be an left almost finite coalgebra. Then C ∗ is left Noetherian;
moreover all nonzero left ideals of C ∗ have finite codimension.
Proof. Then if I is a nonzero left ideal of C ∗ take f 6= 0, f ∈ I. Then F := (C ∗ · f )⊥
is a left coideal of C so F is finite dimensional. Then F ⊥ has finite codimension; but
C ∗ · f = F ⊥ by Proposition 1.1. This shows that I has finite codimension. Consequently,
C ∗ is left Noetherian.
(cid:3)
Proposition 2.5. If C is a chain coalgebra, then J n = C ⊥
are the only ideals (left, right, two-sided) of C ∗. Consequently, C ∗ is a chain algebra.
n−1 for all n and 0 and J n, n ≥ 0
Proof. If I is a left ideal of C ∗, then by Lemma 1.1 and Proposition 2.4 we have (I ⊥)⊥ = I.
By Proposition 2.2, I ⊥ = C or I ⊥ = Cn−1 for some n ≥ 1 and therefore I = (I ⊥)⊥ = C ⊥
or I = 0. We prove by induction on n that J n = C ⊥
and assume J n = C ⊥
(J n+1)⊥ = Cn, we get J n+1 = C ⊥
n−1
n−1, for all n ≥ 1. For n = 1, J = C ⊥
0
n−1 for some n ≥ 2. We again have J n+1 = ((J n+1)⊥)⊥ and as
(cid:3)
n and the proof is finished.
For a left C ∗-module M denote by T (M ) the set of all torsion elements of M , that is,
T (M ) = {x ∈ M annC ∗x 6= 0}. If C is a finite dimensional coalgebra, then obviously
any right C-comodule is rational and the category of right comodules coincides to that
of the left C ∗-modules. Then it is interesting to investigate the infinite dimensional case.
We first consider a special kind of coalgebra:
Proposition 2.6. Let C be an infinite dimensional left almost finite coalgebra and let
R = C ∗. Then for any left R module M we have Rat(M ) = T (M ); moreover, x ∈ Rat(M )
if and only if R · x is finite dimensional.
Proof. If x ∈ Rat(M ) then R · x is finite dimensional and then annRx must be of finite
codimension, thus nonzero as R is infinite dimensional. Conversely, if x ∈ T (M ) and
x 6= 0 then I = annRx is a nonzero left ideal of R so it must have finite codimension by
Proposition 2.4. Then as R/I ≃ R · x we get that R · x is finite dimensional. Also I ⊥ ⊂ C
is a finite dimensional left subcomodule of C and thus the subcoalgebra C ′ of C generated
by I ⊥ is finite dimensional. Taking Proposition 1.1 into account, I = (I ⊥)⊥ ⊇ C ′⊥. Then
C ′⊥ · x = 0, and R · x becomes a left R/C ′⊥-module. Now note that as C ′ is a finite
dimensional coalgebra with C ′∗ ≃ C ∗/C ′⊥, R · x has a structure of right C ′-module. So
we have a map ρ : R · x → R · x ⊗ C ′, ρ(c) = c0 ⊗ c1 such that h · c = h(c1)c0 for h ∈ C ′∗.
THE FINITE RAT-SPLITTING FOR COALGEBRAS
7
But then if π : C ∗ → C ′∗ is the cannonical projection π(r) = rC ′, for r ∈ R and c ∈ R · x
we have r · c = π(r) · c = π(r)(c1)c0 = r(c1)c0, so we may regard ρ as a C comultiplication
of R · x, thus x ∈ Rat(M ).
(cid:3)
We show that a chain coalgebra is a (left and right) finite splitting coalgebra. The proof
of this can be done by a standard extension to the noncommutative case of the proof of
the theorem on the structure of finitely generated modules over a PID, and obtain as a
consequence the fact that T (M ) is a direct summand in M for every finitely generated
module M . However we can do this in a more direct way.
Theorem 2.7. If C is a chain coalgebra, then C a left and right finite splitting coalgebra.
Proof. First notice that every torsion-free R finitely generated module M is free: indeed
if x1, . . . , xn is a minimal system of generators, then if λ1x1 + · · · + λnxn = 0 with λi
not all zero, we may assume that λ1 6= 0. Without loss of generality we may also assume
that λ1R ⊇ λiR, ∀i as any two ideals of R are comparable by Proposition 2.5. Therefore
we have λi = λ1si for some si ∈ R. Then λ1x1 + λ1s2x2 + · · · + λ1snxn = 0 implies
x1 + s2x2 + · · · + snxn = 0 as M is torsionfree and λ1 6= 0. Hence x1 ∈ R < x2, . . . , xn >,
contradicting the minimality of n.
Now if M is any left R module and T = T (M ) = Rat(M ) then T (M/T (M )) = 0.
Indeed take x ∈ T (M/T (M )) and put I = annC ∗ x 6= 0 so I has finite codimension
and I is a two-sided ideal by Proposition 2.5. By Proposition 2.2 I is generated by
some h1, . . . , hk ∈ I. Then if y ∈ Ix we have y = f · x, f ∈ I so f =
ri · hi and
n
Pi=1
n
Pi=1
y = f · x =
ri · hix. Therefore Ix is generated by h1x, . . . , hnx. Because I = annR x we
have Ix ⊆ T = Rat(M ) (we use Proposition 2.6) and as Ix is finitely generated rational
we get that Ix has finite dimension. We obviously have an epimorphism R
Ix which
shows that Rx/Ix is finite dimensional because I has finite codimension in R. Therefore
we get that dim(Rx) = dim(Rx/Ix) + dim(Ix) < ∞ so then by Proposition 2.6 we have
that Rx is rational, thus x ∈ T so x = 0.
Now as M/T is torsion-free, there are x1, . . . , xn ∈ M whose images x1, . . . , xn in M/T
form a basis. Then it is easy to see that x1, . . . , xn are linearly independent in M . Then if
X = Rx1 +· · ·+Rxn we have X +T = M and X ∩T = 0, because if a1x1 +· · ·+anxn ∈ T }
we get a1 x1 + · · · + an xn = 0 so ai = 0, ∀i because x1, . . . , xn are independent in M/T .
Thus T (M ) splits off in M and the theorem is proved, as T (M ) = RatR(M ) by 2.6. (cid:3)
I → Rx
ck 7→ Pi+j=k
We will denote by Kn the coalgebra with a basis c0, c1, . . . , cn−1 and comultiplication
Kn having a basis cn, n ∈ N
ci ⊗ cj and counit ε(ci) = δ0,i. The coalgebra Sn∈N
and comultiplication and counit given by these equations is called the divided power
coalgebra (see [DNR]).
Lemma 2.8. Let C be a finite dimensional chain coalgebra over an algebraically closed
field. Then C is isomorphic to Kn for some n ∈ N.
Proof. Let A = C ∗; we have dim C0 = 1 because K is algebraically closed (thus EndAC0
is a skewfield containing K). Thus dim Ck = k for all k for which Ck 6= C. As C ∗ is finite
dimensional J n = 0 for some n and let n be minimal with this property. By Proposition
k−1. Then J k/J k+1 has dimension equal to the dimension of Ck/Ck−1 which
2.5 J k = C ⊥
is 1 for k < n, because Ck+1/Ck it is a simple comodule isomorphic to C0. We then
have that J k/J k+1 is generated by any of its nonzero elements. Choose x ∈ J \ J 2.
8
MIODRAG CRISTIAN IOVANOV
We prove that xn−1 6= 0. Suppose the contrary holds and take y1, . . . , yn−1 ∈ J. As x
generates J/J 2, there is λ ∈ K such that y1 − λx ∈ J 2 and then y1xn−2 − λxn−1 ∈ J n,
so y1xn−2 ∈ J n = 0 because xn−1 = 0. Again, there is µ ∈ K such that y2 − µx ∈ J 2
and then y1y2 − µy1x ∈ J 3 so y1y2xn−3 ∈ J n (y1xn−2 = 0). By continuing this procedure,
one gets that y1y2 . . . yn−2x = 0 and then we again find α ∈ K with yn−1 − αx ∈ J 2, thus
y1 . . . yn−1 − αy1 . . . yn−2x ∈ J n = 0. This shows that y1 . . . yn−1 = 0 for all 1 . . . yn−1 = 0.
Thus J n−1 = 0, a contradiction.
As xn−1 6= 0 we see that xk ∈ J k \ J k+1 for all k = 0, . . . , n − 1, so J k/J k+1 is generated
by the class of xk. Now if y ∈ A, there is λ0 ∈ K such that y − λ0 · 1A ∈ J (either y ∈ J
or y generates A/J). As J/J 2 is 1 dimensional and generated by the image of x, there
is λ1 ∈ K such that y − λ0 − λ1x ∈ J 2. Again, as J 2/J 3 is 1 dimensional generated by
the image of x2, there is λ2 ∈ K such that y − λ0 − λ1x − λ2x2 ∈ J 3. By continuing this
procedure we find λ0, . . . , λn−1 ∈ K such that y − λ0 − λ1x − · · · − λn−1xn−1 ∈ J n = 0,
so y = λ0 + λ1x + · · · + λn−1xn−1. This obviously gives an isomorphism between A
and K[X]/(X n). Therefore C is isomorphic to Kn, because there is an isomorphism of
K-algebras K ∗
(cid:3)
n ≃ K[X]/(X n).
Theorem 2.9. If K is an algebraically closed field and C is an infinite dimensional chain
coalgebra, then C is isomorphic to the divided power coalgebra.
1 6= 0 and En+1
1 ∈ C ⊥
Proof. By the previous Lemma we have that Cn ≃ Kn for all n.
If e ∈ C0, ∆(e) =
λe ⊗ e, λ ∈ K, then for c0 = λe we get ∆(c0) = c0 ⊗ c0. Suppose we constructed
ci ⊗ cj, ε(ci) = δ0,i. Denote by
a basis c0, c1, . . . , cn−1 for Cn−1 with ∆(ck) = Pi+j=k
An = C ∗
n the dual of Cn; for the rest of this proof, if V ⊆ Cn is a subspace of Cn we
0 \ C ⊥
write V ⊥ for the set of the functions of An which are 0 on V . Choose E1 ∈ C ⊥
1 ;
then En
= 0 as in the proof of Lemma 2.8 (E1 ∈ An). This shows
that Ek
1 exhibits a basis for An and that there is an
isomorphism of algebras An ≃ K[X]/(X n+1) taking E1 to X. We can easily see that
Ei
1(cj) = δij, ∀k = 0, 1, . . . , n − 1 and then by a standard linear algebra result we can find
cn ∈ Cn such that En
1 (ci) = 0 for i < n. Then by dualization, the relations
Ei
1(cj) = δij, ∀i, j = 0, 1, . . . , n become ∆(ck) = Pi+j=k
ci ⊗ cj, ∀k = 0, 1, . . . , n. Therefore
we may inductively build the basis (cn)n∈N with ε(ck) = δ0k and ∆(cn) = Pi+j=n
1
k , that εCn , E1, . . . , En
k−1 \ C ⊥
1 (cn) = 1 and En
ci ⊗ cj, ∀n.
(cid:3)
In the following we construct an example of a chain coalgebra that is not cocommutative
and thus different of the divided power coalgebra over K. Recall that if A is a k algebra,
ϕ : A → A is a morphism and δ : A → A is a ϕ-derivation (that is a linear map such that
δ(ab) = δ(a)b + ϕ(a)δ(b) for all a, b ∈ A), we may consider the Ore extension A[X, ϕ, δ]
which is A[X] as a vector space and with multiplication induced by Xa = ϕ(a)X + δ(a).
Let K be a subfield of R, the field of real numbers. Let D be the subalgebra of Hamilton's
quaternion algebra having the set B = {1, i, j, k} as a vector space basis over K. Recall
that multiplication is given by the rules i · j = −j · i = k; j · k = −k · j = i; k · i = −i · k = j;
2
i
= −1. Denote by σ : D → D the linear map defined on the basis of D by
= k
= j
2
2
σ = (cid:18) 1 i
1 j k
j k
i (cid:19)
It is not difficult to see then that σ is an algebra automorphism, and that D is a division
algebra (skewfield). Our example will be such an Ore extension constructed with a trivial
THE FINITE RAT-SPLITTING FOR COALGEBRAS
9
derivation: denote by Dσ[X] = D[X, σ, 0] the Ore extension of D constructed by σ with
the derivation ϕ equal to 0 everywhere. Then a basis for Dσ[X] over K consists of the
elements uX k, with u ∈ B and k ∈ N. Also denote by An = Dσ[X]/ < X n > the algebra
obtained by factoring out the two-sided ideal generated by X n from Dσ[X].
Proposition 2.10. The two sided ideal < X n > of Dσ[X] consists of elements of the
alX l. Moreover, the only (left, right, two-sided) ideals containing < X n >
form f =
n+m
are the ideals < X l >, l = 0, . . . , n and consequently An is a chain K algebra.
Pl=n
N
Pl=0
Proof.
It is clear by the multiplication rule Xa = σ(a)X for a ∈ B that elemens of
Dσ[X] are of the type
alX l and that every element of An is a "polynomial" of the
form f = a0 + a1x + · · · + an−1xn−1, with al ∈ D and where x represents the class of X.
Such an element f is invertible if and only if a0 6= 0. To see this, first note that if a0 = 0
then f is nilpotent, as x is nilpotent and one has f l ∈< xl > by successively using the
relation xa = σ(a)x. Conversely write f = a0 · (1 + a−1
0 an−1xn−1) and note
that the element g = a−1
0 an−1xn−1 is nilpotent as before, so 1 + g must
be invertible in An and therefore f must be invertible. Thus we may write every element
f = alxl + ...an−1xn−1 of An as the product f = (al + al+1x + · · · + an−1xn−1−l) · xl = g · xl
with invertible g. Then if I is a left ideal of An and f ∈ I, we have f = g · xl for an
invertible element g and some l ≤ n. Hence it follows that xl ∈ I. Taking the smallest
number l with the property xl ∈ I, we obviously have that I =< xl >.
(cid:3)
0 a1x + · · · + a−1
0 a1x + · · · + a−1
Let Cn denote the coalgebra dual to An. Note that An has a K basis B = {axl a ∈ B, l ∈
0, 1, . . . , n − 1} and we have the relations (axi)(bxj ) = aσi(b)xi+j. Let (Ea
i )a∈B,i∈0,n−1 the
basis of Cn which is dual to B, that is, Ea
i (bxj) = δijδab for all a, b ∈ B and i, j ∈ N. Also,
for i ∈ N and a ∈ B denote by i · a = σi(a), the action of N on B induced by σ.
Proposition 2.11. With the above notations, denoting by ∆n and εn the comultiplication
and respectively, the counit of Cn we have
∆n(Ec
p) = Xi+j=p; a(i·b)=±c
c−1a(i · b)Ea
i ⊗ Eb
j
and
εn(Ec
p) = δp,0δc,1.
p(u(k · v)xk+l) and as k · v ∈ B
Proof. For u, v ∈ B and k, l ∈ N we have Ec
by the formulas defining D we have that if d = u(k · v) then either d ∈ B or −d ∈ B.
p(dxk+l) = δk+l,pδu(k·v),±cc−1u(k · v) as the sign of this expression
Then Ec
must be 1 if d ∈ B and −1 if d /∈ B, and this is exactly c−1u(k · v) when u(k · v) = ±c.
We also have
p(uxk · vxl) = Ec
p(uxk · vxl) = Ec
Xi+j=p; a(i·b)=±c
c−1a(i · b)Ea
i (uxk)Eb
j (vxl) =
Xi+j=p; a(i·b)=±c
δk,iδu,aδl,jδv,bc−1a(i · b)
= δk+l,pδu(k·v),±cc−1u(k · v)
and therefore we get
Xi+j=p; a(i·b)=±c
c−1a(i · b)Ea
i (uxk)Eb
j (vxl) = Ec
p(uxk · vxl)
10
MIODRAG CRISTIAN IOVANOV
As this is true for all uxk, vxl ∈ B, by the definition of the comultiplication of the coalgebra
dual to an algebra, we get the first equality in the statement of the proposition. The second
one is obvious, as εn(Ec
(cid:3)
p(1 · X 0) = δp,0δc,1.
p) = Ec
Now notice that there is an injective map Cn ⊂ Cn+1 taking Ec
i from Cn to Ec
Therefore we can regard Cn as subcoalgebra of Cn+1. Denote by C = Sn∈N
i from Cn+1.
Cn; it has a
basis formed by the elements Ec
by
n, n ∈ N, c ∈ B and comultiplication ∆ and counit ε given
and
∆(Ec
n) = Xi+j=n; a(i·b)=±c
c−1a(i · b)Ea
i ⊗ Eb
j
ε(Ec
n) = δn,0δc,1.
By Proposition 2.10 we have that An is a chain algebra and therefore Cn = A∗
n is a chain
coalgebra. Therefore, we get that the coradical filtration of C is C0 ⊆ C1 ⊆ C2 ⊆ . . . and
that this is a chain coalgebra which is obviously non-cocommutative.
3. The co-local case
Throughout this section we will assume (unless otherwise specified) that C is left finite
Rat-splitting and that it is a colocal coalgebra, that is, C0 is a simple left (and consequently
0 , C ∗ is a local algebra. We will also assume
simple right) C ∗-module. Then as J = C ⊥
that C is not finite dimensional, thus by Proposition 1.4 C has a countable basis. We
have that C is the injective envelope of C0 as left comodules, thus by Proposition 1.2 we
have that every left subcomodule of C is finite dimensional (all Cn are finite dimensional).
Then if I is a left ideal of C ∗ different from C ∗, by Proposition 1.4 and Proposition 2.4
I is finitely generated and of finite codimension. Denote again R = C ∗. Also for a left
R-module M denote by J(M ) the Jacobson radical of M .
Proposition 3.1. With the above notations, R is a domain.
Proof. Let S = End(CC, C C). Note that S is a ring with multiplication equal to the
composition of morphisms and that S is isomorphic to R by an isomorphism that takes
every morphism of left C-comodules f ∈ S to the element ε ◦ f ∈ R. Then it is enough
to show that S is a domain. If f : C → C is a nonzero morphism of left C comodules,
then Ker(f ) ( C is a proper left subcomodule of C so it must be finite dimensional. Then
as C is not finite dimensional we see that Im(f ) ≃ C/Ker(f ) is an infinite dimensional
subcomodule of C. Thus Im(f ) = C, and therefore every nonzero morphism of left
comodules from C to C must be surjective. Now if f, g ∈ S are nonzero then they are
surjective so f ◦ g is surjective and thus f ◦ g 6= 0.
(cid:3)
Proposition 3.2. R satisfies ACCP on right ideals and also on left ideals.
Proof. Suppose there is an ascending chain of right ideals x0 · R ( x1 · R ( x2 · R ( . . .
that is not stationary. Then there are (λn)n∈N in R such that xn = xn+1 · λn+1. Note
that λn+1 ∈ J, because otherwise λn+1 would be invertible in R as R is local and then we
would have xn+1 = xn · λ−1
n . This would yield xn · R = xn+1 · R, a contradiction. Then
J n = 0. Thus
x1 = xn+1 · λn+1λn . . . λ2, so x1 ∈ J n for all n ∈ N, showing that x1 ∈ Tn∈N
we obtain a contradiction: x0 · R ( x1 · R = 0. The statement is obvious for left ideals as
RR is Noetherian.
(cid:3)
The next proposition contains the main idea of the result.
THE FINITE RAT-SPLITTING FOR COALGEBRAS
11
Proposition 3.3. Suppose αR and βR are two right ideals that are not comparable. Then
any two principal right ideals of R contained in αR ∩ βR are comparable.
Proof. Take aR, bR ⊆ αR ∩ βR, so a = αx = βy and b = αu = βv; we may obviously
assume that a, b 6= 0 as otherwise the assertion is obvious. Then α, β, x, y, u, v are nonzero.
Denote by L the left submodule of R×R generated by (x, u) and by M the quotient module
R×R
L . We write (s, t) for the image of the element (s, t) through the canonical projection
π : R × R → M . We have (y, v) 6= (0, 0) as otherwise (y, v) = λ(x, u) for some λ ∈ R;
then we would have y = λx, v = λu so βy = βλx = αx and then βλ = α (because R is a
domain), a contradiction to αR ( βR. Also β · (y, v) = α · (x, u) = (0, 0) with β 6= 0. This
shows that (0, 0) 6= (y, v) ∈ T = T (M ), so T (M ) 6= 0. Take X < M such that M = T ⊕ X.
We must have X 6= 0, as otherwise (1, 0) ∈ T so there would be a nonzero λ ∈ R and a
µ ∈ R such that λ · (1, 0) = µ · (x, u) ∈ L. But then λ = µx, 0 = µu, so µ = 0 (u 6= 0)
showing that λ = 0, a contradiction.
Now note that x and u are not invertible, as otherwise, for x invertible, αx = βy implies
α ∈ βR so αR ⊆ βR; the same can be inferred if u is invertible. Therefore x, u ∈ J as
R is local so L ⊆ J × J. Hence J(M ) = J × J/L so M/J(M ) = R×R/L
J ×J/L ≃ R × R/J × J
which has dimension 2 as a module over the skewfield R/J. Since M = T ⊕ X and M is
finitely generated, then so are T and X and therefore J(X) 6= X and J(T ) 6= T . Then as
J(M ) = T
J(X) has dimension 2 over R/J, it follows that both T /J(T ) and X/J(X)
are simple. Hence T and X are local, and as they are finitely generated, it follows that they
are generated by any element not belonging to their Jacobson radical. Let T ′ (respectively
X ′) be the inverse images of T (and X respectively) in R × R and t ∈ T ′ and s ∈ X ′ be
such that Rt + L = T ′ and Rs + L = X ′. We have R × R = T ′ + X ′ = Rt + L + Rs + L =
(Rt + Rs) + L ⊆ (Rt + Rs) + J × J ⊆ R × R so (Rt + Rs) + J × J = R × R. Therefore
we obtain Rt + Rs = R × R because J × J is small in R × R.
Write t = (p, q) ∈ T ′. Then t = t + L ∈ T implies that there is λ 6= 0 in R such that
λt = 0 ∈ M and therefore there is µ ∈ R with λ(p, q) = µ(x, u). We show that either
p /∈ J or q /∈ J. Indeed assume otherwise: t = (p, q) ∈ J × J. Then we get Rt ⊆ J × J.
Because Rt + Rs = R × R we see that R × R/J × J must be generated over R by the
image of s. This shows that the R/J module R × R/J × J = (R/J)2 has dimension 1 and
this is obviously a contradiction.
Finally, suppose p /∈ J so p is invertible; then the equations λp = µx, λq = µu imply
λ = µxp−1 and µxp−1q = µu. But µ 6= 0 because p is invertible and λ 6= 0. Therefore we
obtain u = xp−1q; thus b = αu = αxp−1q = ap−1q showing that b ∈ aR i.e. bR ⊆ aR.
Similarly if q is invertible, we get aR ⊆ bR.
(cid:3)
J(T ) ⊕ X
M
Theorem 3.4. If C is a left finite splitting (infinite dimensional) local coalgebra, then C
is a chain coalgebra.
Proof. We first show that every two principal left ideals of R are comparable. Suppose
there are two left ideals of R, R · x0 and R · y0 that are not comparable. Then as they
have finite codimension and C ∗ is infinite dimensional, we have Rx0 ∩ Ry0 6= 0 and take
0 6= αx0 = βy0 ∈ Rx0 ∩ Ry0. Then the right ideals αR and βR are not comparable, as
otherwise, if for example αR ⊆ βR, we would have a relation α = βλ so αx0 = βλx0 = βy0.
As β 6= 0 we get λx0 = y0 because R is a domain, and then Ry0 ⊆ Rx0, a contradiction.
By Proposition 3.2 the set {λR λR ⊆ αR ∩ βR} is Noetherian (relative to inclusion)
and let λR be a maximal element. If x ∈ αR ∩ βR then by Proposition 3.3 we have that
xR and λR are comparable and by the maximality of λR it follows that xR ⊆ λR, so
x ∈ λR. Therefore αR ∩ βR = λR. Note that λ 6= 0, because αR and βR are nonzero
12
MIODRAG CRISTIAN IOVANOV
ideals of finite codimension. Then we see that λR ≃ R as right R modules, because R is
a domain, and again by Proposition 3.3 any two principal right ideals of λR = αR ∩ βR
are comparable, so the same must hold in RR. But this is in contradiction with the fact
that αR and βR are not comparable, and therefore the initial assertion is proved.
Now we prove that J n/J n+1 is a simple right module for all n. As R/J is semisimple
(it is a skewfield) and J n/J n+1 has an R/J module structure, it follows that J n/J n+1
is a semisimple left R/J-module and then J n/J n+1 is semisimple also as R-module. If
we assume that it is not simple, then there are f, g ∈ J n \ J n+1 such that R f = (Rf +
J n+1)/J n+1 and Rg = (Rg+J n+1)/J n+1 are different simple R-modules, so R f ∩Rg = 0 in
J n/J n+1. Then (Rf + J n+1) ∩ (Rg + J n+1) = J n+1 which shows that Rf and Rg cannot
be comparable, a contradiction. As J n = C ⊥
n−1, we see that dim(Cn−1) = codim(J n).
Then for n ≥ 1, dim(Cn/Cn−1) = dim(Cn) − dim(Cn−1) = codimR(J n) − codimR(J n+1) =
dim(J n/J n+1) = dim(C0). Because C0 is the only type of simple right C-comodule, this
last relation shows that the right C-comodule Cn/Cn−1 must be simple. Therefore C must
be a chain coalgebra.
(cid:3)
We may now combine the results of Sections 2 and 3 and obtain
Corollary 3.5. Let C be a co-local coalgebra. Then C is a left (right) finite splitting
coalgebra if and only if C is a chain coalgebra. Moreover, if the base field K is algebraically
closed then this is further equivalent to the fact that C is isomorphic to the divided power
coalgebra.
Proof. This follows from Theorems 2.7, 2.9 and 3.4.
(cid:3)
The author wishes to thank his Ph.D. adviser C. Nastasescu for very useful remarks on
the subject as well as for his continuous support throughout the past years.
Acknowledgment
THE FINITE RAT-SPLITTING FOR COALGEBRAS
13
References
[AN] T. Albu, C. Nastasescu, Relative Finiteness in Module Theory, Monogr. Textbooks Pure Appl. Math.,
vol. 84, Dekker, New York 1984.
[AF] D. Anderson, K.Fuller, Rings and Categories of Modules, Grad. Texts in Math., Springer, Berlin-
Heidelberg-New York, 1974.
[D1] Y. Doi, Homological Coalgebra, J. Math. Soc. Japan 33(1981), 31-50.
[BW] T. Brzezi´nski and R. Wisbauer, Corings and comodules, London Math. Soc. Lect. Notes Ser. 309,
Cambridge University Press, Cambridge, 2003.
[CIDN] F. Castano Iglesias, S. Dascalescu, C. Nastasescu, Symmetric Coalgebras, J. Algebra 279 (2004)
326-344.
[DNR] S. Dascalescu, C. Nastasescu, S¸. Raianu, Hopf Algebras: an introduction. Vol. 235. Lecture Notes
in Pure and Applied Math. Vol.235, Marcel Dekker, New York, 2001.
[GTN] J. Gomez-Torrecillas, C. Nastasescu, Quasi-co-Frobenius coalgebras, J. Algebra 174 (1995), 909-923.
[GMN] J. Gomez-Torrecillas, C. Manu, C. Nastasescu, Quasi-co-Frobenius coalgebras II, Comm. Algebra
Vol 31, No. 10, pp. 5169-5177, 2003.
[GNT] J. Gomez-Torrecillas, C. Nastasescu, B. Torrecillas, Localization in coalgebras. Applications to
finiteness conditions, eprint arXiv:math/0403248, http://arxiv.org/abs/math/0403248.
[I] M.C.Iovanov, Co-Frobenius Coalgebras, to appear, J. Algebra; eprint arXiv:math/0604251
http://xxx.lanl.gov/abs/math.QA/0604251.
[I0] M.C.Iovanov, Characterization of PF rings by the finite topology on duals of R modules, An. Univ.
Bucure¸sti Mat. 52 (2003), no. 2, 189-200.
[K1] I. Kaplansky, Modules over Dedekind rings and valuation rings, Trans. Amer. Math. Soc. 72 (1952)
327-340.
[K2] I. Kaplansky, A characterization of Prufer domains, J. Indian Math. Soc. 24 (1960) 279-281.
[L] B.I-Peng Lin, Semiperfect coalgebras, J. Algebra 30 (1974), 559-601.
[McL] S. Mac Lane, Categories for the Working Mathematician, Second Edition, Springer-Verlag, New
York, 1971.
[Mc1] S. Mac Lane, Duality for groups, Bull. Am. Math. Soc. 56, 485-516 (1950).
[NT] C. Nastasescu, B. Torrecillas, The splitting problem for coalgebras, J. Algebra 281 (2004), 144-149.
[NTZ] Nastasescu, B. Torrecillas, Y. Zhang, Hereditary Coalgebras, Comm. Algebra 24 (1996), 1521-1528.
[Rot] J. Rotman, A characterization of fields among integral domains, An. Acad. Brasil Cienc. 32 (1960)
193-194.
[T1] M.L. Teply, The torsion submodule of a cyclic module splits off, Canad. J. Math. XXIV (1972) 450-
464.
[T2] M.L. Teply, A history of the progress on the singular splitting problem, Universidad de Murcia, De-
partamento de ´Algebra y Fundamentos, Murcia, 1984, 46pp.
[T3] M.L. Teply, Generalizations of the simple torsion class and the splitting properties, Canad. J. Math.
27 (1975) 1056-1074.
Miodrag Cristian Iovanov
University of Bucharest, Faculty of Mathematics, Str. Academiei 14
RO-010014, Bucharest, Romania
E -- mail address: [email protected]
1
1
0
2
p
e
S
4
1
]
.
A
R
h
t
a
m
[
2
v
8
7
4
2
1
6
0
/
h
t
a
m
:
v
i
X
r
a
WHEN DOES THE RATIONAL TORSION SPLIT OFF FOR FINITELY
GENERATED MODULES
MIODRAG CRISTIAN IOVANOV
Dedicated to Fred Van Oystaeyen for his sixtieth birthday
Abstract. It is well known that the torsion part of any finitely generated module over
the formal power series ring K[[X]] is a direct summand. In fact, K[[X]] is an algebra
dual to the divided power coalgebra over K and the torsion part of any K[[X]]-module
actually identifies with the rational part of that module. More generally, for a certain
general enough class of coalgebras - those having only finite dimensional subcomodules
- we see that the above phenomenon is preserved: the set of torsion elements of any
C ∗-module is exactly the rational submodule. With this starting point in mind, given
a coalgebra C we investigate when the rational submodule of any finitely generated left
C ∗-module is a direct summand. We prove various properties of coalgebras C having
this splitting property. Just like in the K[[X]] case, we see that standard examples of
coalgebras with this property are the chain coalgebras which are coalgebras whose lattice
of left (or equivalently, right, two-sided) coideals form a chain. We give some representa-
tion theoretic characterizations of chain coalgebras, which turn out to make a left-right
symmetric concept. In fact, in the main result of this paper we characterize the colo-
cal coalgebras where this splitting property holds non-trivially (i.e. infinite dimensional
coalgebras) as being exactly the chain coalgebras. This characterizes the cocommutative
coalgebras of this kind. Furthermore, we give characterizations of chain coalgebras in
particular cases and construct various and general classes of examples of coalgebras with
this splitting property.
Introduction
Let R be a ring and T be a torsion preradical on the category of left R-modules RM. Then
R is said to have splitting property provided that T (M ), the torsion submodule of M , is a
direct summand of M for any M ∈ RM. More generally, if C is a Grothendieck category
and A is a subcategory of C, then A is called closed if it is closed under subobjects, quotient
objects and direct sums. To every such subcategory we can associate a preradical t (also
called torsion functor) if for every M ∈ C we denote by t(M ) the sum of all subobjects of
M that belong to A. We say that C has the splitting property with respect to A if t(M ) is a
direct summand of M for all M ∈ C. In the case of the category of R-modules, the splitting
property with respect to some closed subcategory is a classical problem which has been
considered by many authors. In particular, when R is a commutative domain, the question
of when the (classical) torsion part of an R module splits off is a well known problem. J.
Rotman has shown in [Rot] that for a commutative domain the torsion submodule splits
off in every R-module if and only if R is a field. I. Kaplansky proved in [K1], [K2] that for
Key words and phrases. Torsion Theory, Splitting, Coalgebra, Rational Module.
2000 Mathematics Subject Classification. Primary 16W30; Secondary 16S90, 16Lxx, 16Nxx, 18E40.
∗ This paper was partially supported by the contract nr. 24/28.09.07 with UEFISCU "Groups, quantum
groups, corings and representation theory" of CNCIS, PN II (ID 1002) and was also partially written within
the frame of the bilateral Flemish-Romanian project "New Techniques in Hopf Algebra Theory and Graded
Rings".
1
2
MIODRAG CRISTIAN IOVANOV
a commutative integral domain R the torsion part of every finitely generated R-module
M splits in M if and only if R is a Prufer domain. While complete or partial results have
been obtained for different cases of subcategories of RM - such as the Dickson subcategory
- or for commutative rings (see also [T1], [T2], [T3]), the general problem remains open
for the non-commutative case and the general categorical setting.
In this paper we investigate a special and important case of rings (algebras) R arising as
the dual algebra of a K-coalgebra C, R = C ∗. We are thus situated in the realm of the
theory of coalgebras and their dual algebras, a theory intensely studied over the last two
decades. Then the category of the left R-modules naturally contains the category MC of
all right C-comodules as a full subcategory. In fact, MC identifies with the subcategory
Rat(C ∗M) of all rational left C ∗-modules, which is generally a closed subcategory of C ∗M.
Then it is natural to study splitting properties with respect to this subcategory, and two
questions regarding this splitting property with respect to Rat(C ∗M) naturally arise: first,
when is the rational part of every left C ∗-module M a direct summand of M and second,
when does the rational part of every finitely generated C ∗-module M split in M . The
first problem, the splitting of C ∗M with respect to the closed subcategory Rat(C ∗M) has
been treated by C. Nastasescu and B. Torrecillas in [NT] where it is proved that if all
C ∗-modules split with respect to Rat then the coalgebra C must be finite dimensional.
The techniques used involve some amount of category theory (localization in categories)
and strongly rely on some general results of M.L.Teply from [T1], [T2], [T3]; another proof
of this fact also based on the general results of Teply is found in [C]; see also [I1] for a
direct approach.
We consider the more general problem of when C has the splitting property only for finitely
generated modules, that is, the problem of when is the rational part Rat(M ) of M a direct
summand in M for all finitely generated left C ∗-modules M . We say that such a coalgebra
has the left f.g. Rat-splitting property (or we say that it has the Rat-splitting property for
finitely generated left modules). If the coalgebra C is finite dimensional, then every left
C ∗-module is rational so MC is equivalent to C ∗M and Rat(M ) = M for all C ∗-modules
M and in this case Rat(M ) trivially splits in any C ∗-module. Therefore we will deal with
infinite dimensional coalgebras, as generally the infinite dimensional coalgebras produce
examples essentially different from the ones in algebra theory.
The starting and motivating point of our research is the fact that over the ring of formal
power series over a field R = K[[X]] (or a division algebra), any finitely generated module
splits into its torsion part and a complementary module. In this case, R is the dual of the
so called divided power coalgebra, and the torsion part of any module identifies with the
rational submodule. Here the analogue with classical torsion splitting problems becomes
obvious. In fact, what turns out to be essential in this example is the structure of ideals
of K[[X]], and that is, they are linearly ordered. This suggests the consideration of more
general coalgebras, those whose left subcomodules form a chain. This turns out to be a
left-right symmetric concept, and the most basic example of infinite dimensional coalge-
bra having the f.g. Rat-splitting property (Proposition 2.3 and Theorem 2.5). One key
observation in this study is that if C has the f.g. Rat-splitting property, then the indecom-
posable left injectives have only finite dimensional proper subcomodules, and this motivates
the introduction of comodules and coalgebras C having only finite dimensional proper left
subcomodules, which we call almost finite (or almost finite dimensional) comodules. This
proves to be the proper generalization of the phenomenon found in the case of K[[X]],
WHEN DOES THE RATIONAL TORSION SPLIT OFF FOR FINITELY GENERATED MODULES
3
i.e. the set of torsion elements of a left C ∗-module M forms a submodule which coincides
exactly with the rational submodule of M (Proposition 1.5). Before turning to the study
of chain comodules and coalgebras, we give several general results for coalgebras C with
the f.g. Rat-splitting property: they are artinian as right C ∗-module and injective as left
C ∗-module, have at most countable dimension and C ∗ is a left Noetherian ring. Moreover,
such coalgebras have finite dimensional coradical and the f.g. Rat-splitting property is
preserved by subcoalgebras.
The f.g. Rat-splitting property has been studied before in [C] where the last two of the
above statements were proven, but with the use of very strong results of M.L. Teply;
we also include alternate direct proofs. Chain coalgebras were also studied recently in
[LS] and also briefly in [C] and [CGT]. However, our interest in chain coalgebras is of
a different nature; it is a representation theoretic one and is directed towards our main
result of this paper, that generalizes a result previously obtained [C] in the commutative
case: we characterize the coalgebras having the f.g. Rat-splitting property and that are
colocal, and show that they are exactly the chain coalgebras (Section 3, Theorem 3.4), a
result that will involve quite technical arguments. In fact, our characterizations of chain
coalgebras are done as a consequence of more general discussions such as the study chain
of comodules and more generally almost finite comodules and coalgebras. For example,
we show that almost finite coalgebras are reflexive, and that chain coalgebras are almost
finite, and thus obtain the fact that chain coalgebras are reflexive (a result also found in
the recent [LS]) from our more general framework.
We provide several nontrivial examples. One will be the construction of a noncocommu-
tative chain coalgebra with coradical isomorphic to the dual of the Hamilton algebra of
quaternions. However, we see that when the base field K is algebraically closed or the
coalgebra is pointed, then a chain coalgebra is isomorphic to the divided power coalgebra
if it is infinite dimensional or to one of its subcoalgebras otherwise. This also characterizes
the divided power coalgebra over an algebraically closed field as the only local coalgebra
having the above mentioned splitting property. As an application of the main result, we
obtain the structure of cocommutative coalgebras having the f.g. Rat-splitting property
from [C] in a more precise form: they are finite coproducts of finite dimensional coalge-
bras and infinite dimensional chain coalgebras. Moreover, following this model, our results
allow us to generalize to the noncommutative case and show that a coalgebra that is a
finite direct sum of infinite dimensional left chain comodules (serial coalgebra) has the
left f.g. Rat-splitting property; moreover, this is again a left-right symmetric concept.
More generally, a coproduct of such a coalgebra and a finite dimensional one again has
the f.g. Rat-splitting property. We conclude by constructing a class of explicit examples
of noncocommutative coalgebras of this type over an arbitrary field, which will depend on
a positive integer q and a permutation σ of q elements.
1. General Considerations
Let C be a coalgebra with counit ε and comultiplication ∆. We use the Sweedler convention
∆(c) = c1 ⊗ c2 where we omit the summation symbol. For general facts about coalgebras
and comodules we refer to [A], [DNR] or [Sw]. For a vector space V and a subspace W
of V denote by W ⊥ = {f ∈ V ∗ f (x) = 0, ∀ x ∈ W } and for a subspace X ∈ V ∗ denote
by X ⊥ = {x ∈ V f (x) = 0, ∀ f ∈ X} (it will be understood from the context what is
the space V with respect to which the orthogonal is considered). Various properties of
this correspondence between subspaces of V and V ∗ are well known and studied in more
4
MIODRAG CRISTIAN IOVANOV
general settings in [DNR] (Chapter 1), [AF], [AN], [I0]. Related to that, we recall the
finite topology on the dual V ∗ of a vector space V : a basis of 0 for this linear topology
is given by the sets W ⊥ with W a finite dimensional subspace of V . Any topological
consideration will refer to this topology. We often use the following: a subspace X of V ∗
is closed (in the finite topology) if and only if (X ⊥)⊥ = X; also, if W is a subspace of V ,
then (W ⊥)⊥ = W . (see [DNR], Chapter 1)
For a coalgebra C denote by C0 ⊆ C1 ⊆ C2 ⊆ . . . the coradical filtration of C, that is, C0
is the coradical of C, and Cn+1 ⊆ C such that Cn+1/Cn is the socle of the right (or left)
C-comodule C/Cn for all n ∈ N. Then Cn is a subcoalgebra of C for all n, and the same
Cn is obtained whether we take the socle of the left C-comodule C/Cn or of the right C-
comodule C/Cn. Put C−1 = 0 and R = C ∗. Denote J = J(C ∗) the Jacobson radical of C ∗.
By [DNR] we have Sn∈N
n−1
and since Sn∈N
0 and (J n+1)⊥ = Cn. Then J n ⊆ ((J n)⊥)⊥ = C ⊥
J n = 0.
Cn = C, J = C ⊥
Cn = C, we see that Tn∈N
For a left (right) C-comodule M with comultiplication ρ : M → C ⊗ M (ρ : M → M ⊗ C),
the Sweedler notation writes ρ(m) = m−1 ⊗ m0 (respectively ρ(m) = m0 ⊗ m1). Moreover,
the dual M ∗ of M becomes a left (right) C ∗-module by the action induced by the right
(left) C ∗-action on M by duality:
for m∗ ∈ M ∗, m ∈ M and c∗ ∈ C ∗, (c∗ · m∗)(m) =
m∗(m · c∗) = c∗(m−1)m∗(m0) (respectively (m∗ · c∗)(m) = m∗(m0)c∗(m1)).
Lemma 1.1. Let C be a coalgebra over a field K and M be a left C-comodule. Then for
any finitely generated left submodule X of M ∗, (X ⊥)⊥ = X, that is, X is closed in the
finite topology on M ∗.
Proof. It is enough to prove this for cyclic submodules: if (C ∗f ⊥)⊥ = C ∗f for all f ∈ M ∗
and X = C ∗ · f1 + . . . C ∗ · fn then (X ⊥)⊥ = (
C ∗fi = X
(C ∗fi)⊥)⊥ =
(C ∗f ⊥
n
n
i )⊥ =
n
Ti=1
Pi=1
Pi=1
n
Ti=1
M ⊥
i
n
Pi=1
(since (
Mi)⊥ =
for Mi ⊆ M ; see, for example [I0], Proposition 3 or [DNR],
Chapter 1; also [AN] and [AF]).
Let X = C ∗f and u : M → C, u(m) = m−1f (m0), where for m ∈ M , m−1 ⊗ m0 ∈ C ⊗ M
denotes the comultiplication of m ∈ M ; then L = (C ∗f )⊥ = {m ∈ M (hf )(m) = 0, ∀h ∈
C ∗} = {m ∈ M f (h(m−1)m0) = h(m−1f (m0)) = 0, ∀h ∈ C ∗} = {m ∈ M m−1f (m0) =
0}, so L = ker(u) (the left C ∗-module structure on M ∗ is induced from the right C ∗-
module structure on M by duality). If g ∈ L⊥ ⊆ M ∗, then ker(u) ⊆ ker(g) we can factor
g as g = p ◦ u with p : Im(u) → K, and then defining h ∈ C ∗ as h = p on Im(u) ⊆ C
and 0 on some complement of Im(u) we get (hf )(m) = f (m · h) = f (h(m−1)m0) =
h(m−1)f (m0) = h(m−1f (m0)) = h(u(m)) = p ◦ u(m) = g(m), i.e. g ∈ C ∗f . This shows
that (C ∗f ⊥)⊥ = C ∗f .
(cid:3)
1.1. "Almost finite" coalgebras and comodules.
Definition 1.2. A C-comodule M will be called almost finite (or almost finite dimen-
sional) if it has only finite dimensional proper subcomodules. Call a coalgebra C left
almost finite if CC is almost finite.
Proposition 1.3. Let M be a left almost finite (dimensional) C-comodule. Then:
(i) M is artinian as left C-comodule (equivalently, as right C ∗-module).
(ii) Any nonzero submodule of M ∗ has finite codimension; consequently M ∗ is (left) Noe-
therian. Moreover, all submodules of M ∗ are closed in the finite topology of M ∗.
(iii) M has at most countable dimension.
WHEN DOES THE RATIONAL TORSION SPLIT OFF FOR FINITELY GENERATED MODULES
5
Proof. (i) Obvious.
(ii) Let 0 6= I < M ∗ be a submodule, 0 6= f ∈ M ∗. Then X = (C ∗ · f )⊥ is a subcomodule
of M which is finite dimensional and C ∗ · f = X ⊥ from Lemma 1.1, so C ∗ · f has finite
codimension, and so does I ⊇ C ∗ · f . Thus M ∗ is Noetherian and the last assertion of (ii)
follows now from Lemma 1.1.
(iii) Assume M is infinite dimensional and define inductively a sequence (mk)k≥0 such
that mk+1 /∈ Mk = m1 · C ∗ + ... + mk · C ∗. This can be done since the Mk's are finite
dimensional, and then Sk≥0
Mk ⊆ M is infinite-countable dimensional, and thus cannot be
a proper submodule of M . Thus Sk≥0
Mk = M , and the proof is finished.
(cid:3)
The above Proposition shows that a left almost finite coalgebra C is coreflexive by [DNR],
Exercise 1.5.14, since every ideal of finite codimension in C is closed (Also, by a result of
Radford, C is coreflexive if and only any finite dimensional C ∗-module is rational). Thus
we have:
Corollary 1.4. Let C be a left almost finite coalgebra. Then any nonzero left ideal of C ∗
is closed in the finite topology on C ∗ and has finite codimension, C ∗ is Noetherian and
J n = C ⊥
n−1. Moreover, C is coreflexive.
For a left C ∗-module M denote by T (M ) the set of all torsion elements of M , that
is, T (M ) = {x ∈ M annC ∗x 6= 0}.
it is
well known that the categories of right C-comodules and left C ∗-modules are equiva-
lent. Thus, we are interested in the infinite dimensional case. As mentioned above, for
coreflexive coalgebras, Rat(M ) = {x ∈ M C ∗ · x is finite dimensional} = {x ∈ M
annC ∗(x) has finite codimension}. For (infinite dimensional) almost finite coalgebras, we
see that the rational submodule of a C ∗-module has an even more special form:
If C is a finite dimensional coalgebra,
Proposition 1.5. Let C be an infinite dimensional left almost finite coalgebra and let
R = C ∗. Then for any left R-module M we have Rat(M ) = T (M ); moreover, x ∈ Rat(M )
if and only if R · x is finite dimensional.
Proof. If x ∈ Rat(M ) then R · x is finite dimensional and then annR(x) must be of finite
codimension, thus nonzero as R is infinite dimensional. Conversely, if x ∈ T (M ) and
x 6= 0 then I = annR(x) is a nonzero left ideal of R so it is closed by Corollary 1.4; thus
I = X ⊥ with X 6= C a finite dimensional subcomodule of C. Then R · x ≃ R/annR(x) =
C ∗/X ⊥ ≃ X ∗ which is a rational left C ∗-module, being the dual of a finite dimensional
subcomodule of C.
(cid:3)
1.2. The Splitting Property.
Definition 1.6. We shall say that a coalgebra C has the left (right) f.g. Rat-splitting
property, or that it has the left (right) Rat-splitting property for all finitely generated
modules if the rational part of any finitely generated left (right) C ∗-module splits off.
The following key observation, together with the succeding study of chain coalgebras,
motivates our previous introduction of almost finite comodules and coalgebras.
Proposition 1.7. Let C be a coalgebra such that Rat(M ) splits off in any finitely generated
left C ∗-module M . Then any indecomposable injective left C-comodule E is an almost
finite comodule.
Proof. Let T be the socle of E; then T is simple and E = E(T ) is the injective envelope
of T . We show that if K ⊆ E(T ) is an infinite dimensional subcomodule then K = E(T ).
6
MIODRAG CRISTIAN IOVANOV
Suppose K ( E(T ). Then there is a left C-subcomodule (right C ∗-submodule) K ( L ⊂
E(T ) such that L/K is finite dimensional. We have an exact sequence of left C ∗-modules:
0 → (L/K)∗ → L∗ → K ∗ → 0
As L/K is a finite dimensional left C-comodule, we have that (L/K)∗ is a rational left
C ∗-module; thus Rat(L∗) 6= 0. Also L∗ is finitely generated as it is a quotient of E(T )∗
which is a direct summand of C ∗. We have L∗ = Rat(L∗) ⊕ X for some left C ∗-submodule
X of L∗. Then Rat(L∗) is finitely generated because L∗ is, so it is finite dimensional.
As L is infinite dimensional by our assumption, we have X 6= 0. This shows that L∗ is
decomposable and finitely generated, thus it has at least two maximal submodules, say
M, N . We have an epimorphism E(T )∗ f
→ L∗ → 0 and then f −1(M ) and f −1(N ) are
distinct maximal C ∗-submodules of E(T )∗. But by [I], Lemma 1.4, E(T )∗ has only one
maximal C ∗-submodule which is T ⊥, so we have obtained a contradiction.
(cid:3)
C0 = Li∈I
Let C0 be the coradical of C, the sum of all simple subcomodules of C. By [DNR],
Section 3.1, C0 is a cosemisimple coalgebra that is a direct sum of simple subcoalgebras
Ci and each simple subcoalgebra Ci contains only one type of simple left (or
right) C-comodule; moreover, any simple left (or right) C-comodule is isomorphic to one
contained in some Ci. A coalgebra C with C0 finite dimensional is called almost connected
coalgebra.
The following two Propositions have also been observed in [C] (Lemma 3.2 and Lemma
3.3), but general powerful techniques from [T3] are used there. We provide here direct
simple arguments.
Proposition 1.8. Let C be a coalgebra with the left f.g. Rat-splitting property. Then there
is only a finite number of isomorphism types of simple left C-comodules, equivalently, C0
is finite dimensional.
Proof. By the above considerations, if Si is a simple left C-subcomodule of Ci, we
have that (Si)i∈I forms a set of representatives for the isomorphism types of simple left
C-comodules. Let S be a set of representatives for the simple right C-comodules. Let
E(Ci) be an injective envelope of the left C-comodule Ci included in C; then as C0 is
E(Ci) as left C-comodules or right C ∗-modules. Then
E(Ci)∗ as left C ∗-modules. As Si ⊆ E(Ci), we have epimorphisms of left C ∗-
modules E(Ci)∗ → S∗
essential in C we have C = Li∈I
C ∗ = Qi∈I
C ∗ → Qi∈I
C ∗ → QS∈S
S∗
i → 0 and therefore we have an epimorphism of left C ∗-modules
i → 0. But there is a one-to-one correspondence between left and right simple
C-comodules given by {Si i ∈ I} ∋ S 7→ S∗ ∈ S. Hence there is an epimorphism
S is finitely generated
S → 0, which shows that the left C ∗-module P = QS∈S
(actually generated by a single element). But then as Rat(C ∗ P ) is a direct summand in P ,
we must have that Rat(C ∗ P ) is finitely generated, so it is finite dimensional. Therefore,
S is a rational left C ∗-module which is naturally included in P , we have
as Σ = LS∈S
Σ ⊆ Rat(P ). This shows that LS∈S
S is finite dimensional, so S (and also I) must be finite.
This is equivalent to the fact that C0 is finite dimensional, because each Ci is a simple
coalgebra, thus a finite dimensional one.
(cid:3)
Proposition 1.9. If C is has the left f.g. Rat-splitting property then so does any subcoal-
gebra D of C.
WHEN DOES THE RATIONAL TORSION SPLIT OFF FOR FINITELY GENERATED MODULES
7
Proof. Let M be a finitely generated left D∗-module. Since C ∗/D⊥ ≃ D∗, M has an
induced left C ∗-module structure and is annihilated by D⊥ (that is, D⊥ · x = 0 for all
x ∈ M ). Then a subspace of M is a C ∗-submodule if and only if it is a D∗-submodule.
There is M = T ⊕ X a direct sum of C ∗-modules (equivalently D∗-submodules, since D⊥
annihilates the elements in both T and X) with T the rational C ∗-submodule of M . It
will suffice to show that a submodule of M is rational as C ∗-module if and only if it is
rational as D∗-module. Indeed, let m ∈ T = RatC ∗(M ); then there is Pi mi ⊗ ci ∈ T ⊗ C
such that c∗ m = Pi c∗(ci)mi; we may assume that the mi's are linearly independent.
Then for c∗ ∈ D⊥ ⊆ C ∗ we get 0 = c∗ · m = Pi c∗(ci)mi and so c∗(ci) = 0 since the
mi's are independent, showing that ci ∈ (D⊥)⊥ = D. Therefore ρ(m) = Pi
T ⊗ D, where ρ is the comultiplication of T , and thus m ∈ RatD∗(M ). The converse
inclusion RatD∗(M ) ⊆ RatC ∗(M ) is obvious, since the D-comultiplication RatD∗(M ) →
RatD∗(M ) ⊗ D ⊆ RatD∗(M ) ⊗ C induces a C-comultiplication through the canonical
inclusion D ⊆ C, compatible with the C ∗-multiplication of M .
(cid:3)
mi ⊗ ci ∈
Proposition 1.10. Let C be a coalgebra that has the left f.g. Rat-splitting property. Then
the following assertions hold:
(i) C is artinian as left C-comodule (equivalently, as right C ∗-module).
(ii) C ∗ is left Noetherian.
(iii) C has at most countable dimension.
(iv) C is injective as left C ∗-module.
Proof. (i) We have a direct sum decomposition C = Li∈F
E(Si) where C0 = Li∈F
decomposition of C0 into simple left C-comodules and E(Si) are injective envelopes of
Si contained in C. Since F is finite as C0 is finite dimensional, the result follows from
Propositions 1.7 and 1.3
Si is the
(ii) Since C ∗ = Li∈F
E(Si)∗, this also follows from 1.3.
(iii) Similar to (i).
(iv) By [I0] Lemma 2, it is enough to prove that E = C C splits off in any left C ∗-module
M in which it embeds (E ⊆ M ) and such that M/E is cyclic generated by an element
x ∈ M/E. Let H = Rat(C ∗ · x) ⊆ M ; then there is X < C ∗ · x such that H ⊕ X = C ∗ · x.
Then E + H is a rational C ∗-module so (E + H) ∩ X = 0; also M = C ∗ · x + E so
(E + H) + X = M , showing that E + H is a direct summand in M . But as E is an
injective comodule, we have that E splits off in E + H, thus E must split in M and the
proof is finished.
(cid:3)
2. Chain Coalgebras
Definition 2.1. We say that a left (right) C-comodule M is a chain (or uniserial) co-
module if and only if the lattice of the left (right) subcomodules of C is a chain, that is,
for any two subcomodules X, Y of M either X ⊆ Y or Y ⊆ X. We say a coalgebra C is a
left (right) chain coalgebra (or uniserial coalgebra) if C is a left (right) chain C-comodule.
In other words, a left C-comodule M is a chain comodule if M is uniserial as a right
C ∗-module. Part of the following proposition is a somewhat different form of Lemma 2.1
from [CGT]. However, we will need to use some of the other equivalent statements bellow.
Proposition 2.2. Let M be a left (right) C-comodule. The following assertions are equiv-
alent:
(i) M is a chain comodule.
8
MIODRAG CRISTIAN IOVANOV
(ii) M ∗ is a chain (uniserial) left (right) C ∗-module.
(iii) M and Mn = "the n'th Loewy term in the Loewy series of M for n ≥ −1", are the
only subcomodules of M (M−1 = 0).
(iv) M ⊥
M ∗.
(v) Mn/Mn−1 is either simple or 0 for all n ≥ −1. (If Mn/Mn−1 is 0 for some n then
Mk/Mk−1 is 0 for all k ≥ n.)
n = {u ∈ M ∗ u(x) = 0, ∀ x ∈ Mn} for n ≥ −1 and 0 are the only submodules of
Proof. (iv)⇒(ii) is obvious.
(ii)⇒(i) If M ∗ is uniserial, then for any two subcomodules X, Y of C we have X ⊥ ⊆ Y ⊥,
say. Thus we get X = (X ⊥)⊥ ⊇ (Y ⊥)⊥ = Y .
(i)⇔(iii) is obvious (note that (iii) this does not exclude the possibility that M = Mn from
some n onward)
(i)⇒(iv) If M is a chain comodule, it is enough to assume that M is infinite dimensional,
because of the duality of categories between finite dimensional left comodules and finite
dimensional right comodules. We note that each M ⊥
n+1.
Let f ∈ M ⊥
n and denote un, f : M → C, un(m) = m−1un(m0) (f (m) = m−1un(m0)).
Then un ∈ M ⊥
n+1 shows that un is a morphism of left C-comodules that factors to a
morphism M/Mn → C which does not cancel on Mn+1/Mn - the only simple subcomodule
of M/Mn. Therefore Ker (un : M/Mn → C) = 0 and we have a diagram
n is generated by any un ∈ M ⊥
n \ M ⊥
n \M ⊥
0
un
g
C
/ M
Mn
f
C
that is completed commutatively by g (as CC is injective), so that we get g ◦ un = f and
then if g = ε ◦ g we have, for m ∈ M , g(m−1)un(m0) = g(m−1un(m1)) = ε(g(un(m))) =
ε(f (m)) = ε(m−1)f (m0) = f (m). Thus g · un = f . This shows that any cyclic submodule
n , because for any 0 6= f ∈ M ∗ there is some n such that
of M ∗ coincides to one of the M ⊥
f ∈ M ⊥
Mn. It therefore follows that for any nonzero submodule I
n \ M ⊥
n+1, since M = Sn
k ⊆ I.
n ⊆ I; since the Mn's are (obviously) finite dimensional, M ⊥
of M ∗ there is M ⊥
n and I have
finite codimension and it now easily follow from the above considerations that I = M ⊥
k ,
where k is the smallest number such that M ⊥
(v)⇒(iii) Let X be a right subcomodule of M and suppose X 6= M and X 6= 0. Then there
is n ≥ 0 such that Mn * X and let n be minimal with this property. Then we must have
Mn−1 ⊆ X by the minimality of n and we show that Mn−1 = X. Indeed, if Mn−1 ( X we
can find a simple subcomodule of X/Mn−1. But then Mn−1 6= M , so Mn−1 6= Mn and as
Mn/Mn−1 is the only simple subcomodule of M/Mn, we find Mn/Mn−1 ⊆ X/Mn−1, that
is Mn ⊆ X, a contradiction.
(i)⇒(v) If Mn+1/Mn is nonzero and it is not simple then we can find S1 = X1/Mn and
S2 = X2/Mn (X1, X2 ⊆ M ) two distinct simple modules contained in M/Mn. Then
X1 ∩ X2 = Mn, X1 6= Mn, X2 6= Mn. But this shows that neither X1 ( X2 nor X2 ( X1
which is a contradiction.
(cid:3)
The following result shows that chain coalgebra is a left-right symmetric notion and also
characterizes chain coalgebras.
Proposition 2.3. The following assertions are equivalent for a coalgebra C:
(i) C is a right chain coalgebra.
/
/
/
WHEN DOES THE RATIONAL TORSION SPLIT OFF FOR FINITELY GENERATED MODULES
9
(ii) Cn+1/Cn is either 0 or a simple (right) comodule for all n ≥ −1.
(iii) Cn, n ≥ −1 and C are the only right subcomodules of C.
(iv) J n, n ≥ 0 and 0 are the only right ideals of C ∗.
(v) C ∗ is a right (or left) uniserial ring (chain algebra).
(vi) The left hand side version of (i)-(iv).
(vii) C1 has length less or equal to 2.
Proof. The equivalence of (i)-(vi) follows from Proposition 2.2 and also by Corollary 1.4
(i)⇒(vii) is obvious and (vii)⇒(i) is a result from [C]. We note a direct argument for
this case: it is enough to deal with the case when C1 has length 2; by induction, assume
Ck/Ck−1 is simple or 0 for k ≤ n. Assume Cn 6= Cn−1 and note that since Cn/Cn−1 is
the socle of C/Cn−1, then C/Cn−1 embeds in C and therefore if Cn+1/Cn−1 has length at
most 2, since it embeds in C1. Thus Cn+1/Cn is simple or 0.
(cid:3)
Remark 2.4. The above Proposition includes many of the results in [LS] sections 5.1-5.3.
By Proposition 2.2 a chain module is almost finite and by 2.3 a chain coalgebra is left and
right almost finite, so the results of the first section apply here. Therefore we also obtain
that a chain coalgebra is coreflexive.
Next we show that a chain coalgebra is both a left and a right f.g. Rat-splitting property
coalgebra. Although this follows in a more general setting as in Section 4, we also provide a
direct proof that does not involve the tools used in there, but makes use of the interesting
fact that for a left almost finite coalgebra C and any left C ∗-comodule M , T (M ) is a
submodule of M and is exactly the rational submodule of M .
Theorem 2.5. If C is a chain coalgebra, then C has the left and right f.g. Rat-splitting
property.
Proof. Of course, we only need to consider the case when C is infinite dimensional. First
notice that every torsion-free R-finitely generated module M is free: indeed if x1, . . . , xn is
a minimal system of generators, then if λ1x1 + · · · + λnxn = 0 with λi not all zero, we may
assume that λ1 6= 0. Without loss of generality we may also assume that λ1R ⊇ λiR, ∀i
as any two ideals of R are comparable by Proposition 2.3. Therefore we have λi = λ1si for
some si ∈ R. Then λ1x1 + λ1s2x2 + · · · + λ1snxn = 0 implies x1 + s2x2 + · · · + snxn = 0 as
M is torsionfree and λ1 6= 0. Hence x1 ∈ R < x2, . . . , xn >, contradicting the minimality
of n.
Now if M is any left R-module and T = T (M ) = Rat(M ) (by Proposition 1.5) then
Indeed take x ∈ T (M/T (M )) and put I = annC ∗ x 6= 0 so I has
T (M/T (M )) = 0.
finite codimension and I is a two-sided ideal by Proposition 2.3. By Corollary 1.4 and
Remark 2.4, I is finitely generated and therefore Ix is also finitely generated. Also, since
I = annC ∗ x, we get Ix ⊆ T = Rat(M ). Thus Ix is finitely generated rational, so Ix
has finite dimension. We obviously have an epimorphism R
Ix which shows that
Rx/Ix is finite dimensional because I has finite codimension in R. Therefore we get that
dim(Rx) = dim(Rx/Ix) + dim(Ix) < ∞, so then by Proposition 1.5 we have that Rx is
rational, thus x ∈ T so x = 0.
Now as M/T is torsion-free, there are x1, . . . , xn ∈ M whose images x1, . . . , xn in M/T
form a basis. Then it is easy to see that x1, . . . , xn are linearly independent in M . Then if
X = Rx1 + · · · + Rxn we have X + T = M and X ∩ T = 0, because if a1x1 + · · · + anxn ∈ T
we get a1 x1 + · · · + an xn = 0 so ai = 0, ∀i because x1, . . . , xn are independent in M/T .
Thus T (M ) splits off in M and the theorem is proved, as T (M ) = RatR(M ) by 1.5. (cid:3)
I → Rx
10
MIODRAG CRISTIAN IOVANOV
We will denote by Kn the coalgebra with a basis c0, c1, . . . , cn−1 and comultiplication
Kn having a basis cn, n ∈ N
ci ⊗ cj and counit ε(ci) = δ0,i. The coalgebra Sn∈N
ck 7→ Pi+j=k
and comultiplication and counit given by these equations is called the divided power
coalgebra (see [DNR]). Part of the following Lemma is discussed in [CGT] Theorem 3.2;
also part of it in the cocommutative case is observed in [C], 3.5 and 3.6. The same result
appears in [LS], but with a different proof. Also Theorem 2.7 below can be obtained as a
consequence of the general theory of serial coalgebras developed in [CGT] (Theorem 2.10
(iii) and Remark 2.12); in this respect, Lemma 2.6 could then be obtained as a consequence
of Theorem 2.7. We provide here a direct argument.
Lemma 2.6. Let C be a finite dimensional chain coalgebra over a field K and suppose
that either K is algebraically closed or C is pointed. Then C is isomorphic to Kn for some
n ∈ N.
Proof. Let A = C ∗; we have dim C0 = 1 because K is algebraically closed (thus EndAC0
is a skewfield containing K). Thus dim Ck = k for all k for which Ck 6= C. As C ∗ is finite
dimensional J n = 0 for some n and let n be minimal with this property. By Corollary
k−1. Then J k/J k+1 has dimension equal to the dimension of Ck/Ck−1 which
1.4 J k = C ⊥
is 1 for k < n, because Ck+1/Ck it is a simple comodule isomorphic to C0. We then
have that J k/J k+1 is generated by any of its nonzero elements. Choose x ∈ J \ J 2.
We prove that xn−1 6= 0. Suppose the contrary holds and take y1, . . . , yn−1 ∈ J. As x
generates J/J 2, there is λ ∈ K such that y1 − λx ∈ J 2 and then y1xn−2 − λxn−1 ∈ J n,
so y1xn−2 ∈ J n = 0 because xn−1 = 0. Again, there is µ ∈ K such that y2 − µx ∈ J 2
and then y1y2 − µy1x ∈ J 3 so y1y2xn−3 ∈ J n (y1xn−2 = 0). By continuing this procedure,
one gets that y1y2 . . . yn−2x = 0 and then we again find α ∈ K with yn−1 − αx ∈ J 2, thus
y1 . . . yn−1 − αy1 . . . yn−2x ∈ J n = 0. This shows that y1 . . . yn−1 = 0 for all y1, . . . , yn−1 ∈
J. Thus J n−1 = 0, a contradiction.
As xn−1 6= 0 we see that xk ∈ J k \ J k+1 for all k = 0, . . . , n − 1, so J k/J k+1 is generated
by the class of xk. Now if y ∈ A, there is λ0 ∈ K such that y − λ0 · 1A ∈ J (either y ∈ J
or y generates A/J). As J/J 2 is 1 dimensional and generated by the image of x, there
is λ1 ∈ K such that y − λ0 − λ1x ∈ J 2. Again, as J 2/J 3 is 1 dimensional generated by
the image of x2, there is λ2 ∈ K such that y − λ0 − λ1x − λ2x2 ∈ J 3. By continuing this
procedure we find λ0, . . . , λn−1 ∈ K such that y − λ0 − λ1x − · · · − λn−1xn−1 ∈ J n = 0,
so y = λ0 + λ1x + · · · + λn−1xn−1. This obviously gives an isomorphism between A
and K[X]/(X n). Therefore C is isomorphic to Kn, because there is an isomorphism of
K-algebras K ∗
(cid:3)
n ≃ K[X]/(X n).
Theorem 2.7. If K is an algebraically closed field and C is an infinite dimensional chain
coalgebra, then C is isomorphic to the divided power coalgebra. The same conclusion holds
provided the infinite dimensional chain coalgebra C is pointed.
If e ∈ C0, ∆(e) =
Proof. By the previous Lemma we have that Cn ≃ Kn for all n.
λe ⊗ e, λ ∈ K, then for c0 = λe we get ∆(c0) = c0 ⊗ c0. Suppose we constructed
ci ⊗ cj, ε(ci) = δ0,i. Denote by
a basis c0, c1, . . . , cn−1 for Cn−1 with ∆(ck) = Pi+j=k
An = C ∗
n the dual of Cn; for the rest of this proof, if V ⊆ Cn is a subspace of Cn we
0 \ C ⊥
write V ⊥ for the set of the functions of An which are 0 on V . Choose E1 ∈ C ⊥
1 ;
then En
= 0 as in the proof of Lemma 2.6 (E1 ∈ An). This shows
that Ek
1 exhibits a basis for An and that there is an
isomorphism of algebras An ≃ K[X]/(X n+1) taking E1 to X. We can easily see that
Ei
1(cj) = δij, ∀k = 0, 1, . . . , n − 1 and then by a standard linear algebra result we can find
1 6= 0 and En+1
1 ∈ C ⊥
1
k , that εCn , E1, . . . , En
k−1 \ C ⊥
WHEN DOES THE RATIONAL TORSION SPLIT OFF FOR FINITELY GENERATED MODULES 11
1 (cn) = 1 and En
cn ∈ Cn such that En
Ei
1(cj) = δij, ∀i, j = 0, 1, . . . , n become ∆(ck) = Pi+j=k
we may inductively build the basis (cn)n∈N with ε(ck) = δ0k and ∆(cn) = Pi+j=n
1 (ci) = 0 for i < n. Then by dualization, the relations
ci ⊗ cj, ∀k = 0, 1, . . . , n. Therefore
ci ⊗ cj, ∀n.
(cid:3)
A non-trivial example. In the following we construct an example of a chain coalgebra
that is not cocommutative and thus different of the divided power coalgebra over K. Recall
that if A is a k algebra, ϕ : A → A is a morphism and δ : A → A is a ϕ-derivation (that
is a linear map such that δ(ab) = δ(a)b + ϕ(a)δ(b) for all a, b ∈ A), we may consider the
Ore extension A[X, ϕ, δ] which is A[X] as a vector space and with multiplication induced
by Xa = ϕ(a)X + δ(a). Let K be a subfield of R, the field of real numbers. Let D be
the subalgebra of Hamilton's quaternion algebra having the set B = {1, i, j, k} as a vector
space basis over K. Recall that multiplication is given by the rules i · j = −j · i = k;
2
j · k = −k · j = i; k · i = −i · k = j; i
= −1. Denote by σ : D → D the linear
map defined on the basis of D by
= k
= j
2
2
σ = (cid:18) 1 i
1 j k
j k
i (cid:19)
It is not difficult to see then that σ is an algebra automorphism, and that D is a division
algebra (skewfield). Our example will be constructed with the aid of such an Ore extension
constructed with a trivial derivation: denote by Dσ[X] = D[X, σ, 0] the Ore extension
of D constructed by σ with the derivation ϕ equal to 0 everywhere. Then a basis for
Dσ[X] over K consists of the elements uX k, with u ∈ B and k ∈ N. Also denote by
An = Dσ[X]/ < X n > the algebra obtained by factoring out the two-sided ideal generated
by X n from Dσ[X].
Proposition 2.8. The two sided ideal < X n > of Dσ[X] consists of elements of the form
alX l. Moreover, the only (left, right, two-sided) ideals containing < X n > are
f =
n+m
the ideals < X l >, l = 0, . . . , n and consequently An is a chain K-algebra.
Pl=n
Proof.
It is clear by the multiplication rule Xa = σ(a)X for a ∈ B that elements of
N
Pl=0
Dσ[X] are of the type
alX l and that every element of An is a "polynomial" of the
form f = a0 + a1x + · · · + an−1xn−1, with al ∈ D and where x represents the class of X.
Such an element f is invertible if and only if a0 6= 0. To see this, first note that if a0 = 0
then f is nilpotent, as x is nilpotent and one has f l ∈< xl > by successively using the
relation xa = σ(a)x. Conversely write f = a0 · (1 + a−1
0 an−1xn−1) and note
that the element g = a−1
0 an−1xn−1 is nilpotent as before, so 1 + g must
be invertible in An and therefore f must be invertible. Thus we may write every element
f = alxl + ...an−1xn−1 of An as the product f = (al + al+1x + · · · + an−1xn−1−l) · xl = g · xl
with invertible g. Then if I is a left ideal of An and f ∈ I, we have f = g · xl for an
invertible element g and some l ≤ n. Hence it follows that xl ∈ I. Taking the smallest
number l with the property xl ∈ I, we obviously have that I =< xl >.
(cid:3)
0 a1x + · · · + a−1
0 a1x + · · · + a−1
Let Cn denote the coalgebra dual to An. Note that An has a K basis B = {axl a ∈ B, l ∈
0, 1, . . . , n − 1} and we have the relations (axi)(bxj) = aσi(b)xi+j. Let (Ea
i )a∈B,i∈0,n−1 be
i (bxj) = δijδab for all a, b ∈ B and i, j ∈ N.
the basis of Cn which is dual to B, that is, Ea
Also, for i ∈ N and a ∈ B denote by i · a = σi(a) the action of N on B induced by σ.
12
MIODRAG CRISTIAN IOVANOV
Proposition 2.9. With the above notations, denoting by ∆n and εn the comultiplication
and respectively, the counit of Cn we have
∆n(Ec
p) = Xi+j=p; a(i·b)=±c
c−1a(i · b)Ea
i ⊗ Eb
j
and
εn(Ec
p) = δp,0δc,1.
p(u(k · v)xk+l) and as k · v ∈ B
Proof. For u, v ∈ B and k, l ∈ N we have Ec
by the formulas defining D we have that if d = u(k · v) then either d ∈ B or −d ∈ B.
p(dxk+l) = δk+l,pδu(k·v),±cc−1u(k · v) as the sign of this expression
Then Ec
must be 1 if d ∈ B and −1 if d /∈ B, and this is exactly c−1u(k · v) when u(k · v) = ±c.
We also have
p(uxk · vxl) = Ec
p(uxk · vxl) = Ec
Xi+j=p; a(i·b)=±c
c−1a(i · b)Ea
i (uxk)Eb
j (vxl) =
Xi+j=p; a(i·b)=±c
δk,iδu,aδl,jδv,bc−1a(i · b)
= δk+l,pδu(k·v),±cc−1u(k · v)
and therefore we get
Xi+j=p; a(i·b)=±c
c−1a(i · b)Ea
i (uxk)Eb
j (vxl) = Ec
p(uxk · vxl)
As this is true for all uxk, vxl ∈ B, by the definition of the comultiplication of the coalgebra
dual to an algebra, we get the first equality in the statement of the proposition. The second
one is obvious, as εn(Ec
(cid:3)
p(1 · X 0) = δp,0δc,1.
p) = Ec
Now notice that there is an injective map Cn ⊂ Cn+1 taking Ec
i from Cn to Ec
Therefore we can regard Cn as subcoalgebra of Cn+1. Denote by C = Sn∈N
i from Cn+1.
Cn; it has a
basis formed by the elements Ec
by
n, n ∈ N, c ∈ B and comultiplication ∆ and counit ε given
and
∆(Ec
n) = Xi+j=n; a(i·b)=±c
c−1a(i · b)Ea
i ⊗ Eb
j
ε(Ec
n) = δn,0δc,1.
By Proposition 2.8 we have that An is a chain algebra and therefore Cn = A∗
n is a chain
coalgebra. Therefore, we get that the coradical filtration of C is C0 ⊆ C1 ⊆ C2 ⊆ . . . and
that this is a chain coalgebra which is obviously non-cocommutative.
3. The co-local case
Throughout this section we will assume (unless otherwise specified) that C has the left
f.g. Rat-splitting property and that it is a colocal coalgebra, that is, C0 is a simple left
0 , C ∗ is a local algebra.
(and consequently simple right) C ∗-module. Then as J = C ⊥
We will also assume that C is not finite dimensional, thus by Proposition 1.10 C has a
countable basis. We have that C is the injective envelope of C0 as left comodules, thus by
Proposition 1.7 we have that every left subcomodule of C is finite dimensional (all Cn are
finite dimensional). Then if I is a left nonzero ideal of C ∗ different from C ∗, by Corollary
1.4 I is finitely generated and of finite codimension. Denote again R = C ∗. Also for a left
R-module M denote by J(M ) the Jacobson radical of M .
Proposition 3.1. With the above notations, R is a domain.
WHEN DOES THE RATIONAL TORSION SPLIT OFF FOR FINITELY GENERATED MODULES 13
Proof. Let S = End(CC, C C). Note that S is a ring with multiplication equal to the
composition of morphisms and that S is isomorphic to R by an isomorphism that takes
every morphism of left C-comodules f ∈ S to the element ε ◦ f ∈ R. Then it is enough
to show that S is a domain. If f : C → C is a nonzero morphism of left C comodules,
then Ker(f ) ( C is a proper left subcomodule of C so it must be finite dimensional. Then
as C is not finite dimensional we see that Im(f ) ≃ C/Ker(f ) is an infinite dimensional
subcomodule of C. Thus Im(f ) = C, and therefore every nonzero morphism of left
comodules from C to C must be surjective. Now if f, g ∈ S are nonzero then they are
surjective so f ◦ g is surjective and thus f ◦ g 6= 0.
(cid:3)
Proposition 3.2. R satisfies ACCP on right ideals and also on left ideals.
Proof. Suppose there is an ascending chain of right ideals x0 · R ( x1 · R ( x2 · R ( . . .
that is not stationary. Then there are (λn)n∈N in R such that xn = xn+1 · λn+1. Note
that λn+1 ∈ J, because otherwise λn+1 would be invertible in R as R is local and then we
would have xn+1 = xn · λ−1
n . This would yield xn · R = xn+1 · R, a contradiction. Then
J n = 0. Thus
x1 = xn+1 · λn+1λn . . . λ2, so x1 ∈ J n for all n ∈ N, showing that x1 ∈ Tn∈N
we obtain a contradiction: x0 · R ( x1 · R = 0. The statement is obvious for left ideals as
RR is Noetherian.
(cid:3)
The next proposition together with the following theorem contain the main ideas of the
result.
Proposition 3.3. Suppose αR and βR are two right ideals that are not comparable, i.e.
neither one is a subset of the other. Then any two principal right ideals of R contained in
αR ∩ βR are comparable.
Proof. Take aR, bR ⊆ αR ∩ βR, so a = αx = βy and b = αu = βv; we may obviously
assume that a, b 6= 0 as otherwise the assertion is obvious. Then α, β, x, y, u, v are nonzero.
Denote by L the left submodule of R×R generated by (x, u) and by M the quotient module
R×R
L . We write (s, t) for the image of the element (s, t) through the canonical projection
π : R × R → M . We have (y, v) 6= (0, 0) as otherwise (y, v) = λ(x, u) for some λ ∈ R;
then we would have y = λx, v = λu so βy = βλx = αx and then βλ = α (because R is a
domain), a contradiction to αR ( βR. Also β · (y, v) = α · (x, u) = (0, 0) with β 6= 0. This
shows that (0, 0) 6= (y, v) ∈ T = T (M ), so T (M ) 6= 0. Take X < M such that M = T ⊕ X.
We must have X 6= 0, as otherwise (1, 0) ∈ T so there would be a nonzero λ ∈ R and a
µ ∈ R such that λ · (1, 0) = µ · (x, u) ∈ L. But then λ = µx, 0 = µu, so µ = 0 (u 6= 0)
showing that λ = 0, a contradiction.
Now note that x and u are not invertible, as otherwise, for x invertible, αx = βy implies
α ∈ βR so αR ⊆ βR; the same can be inferred if u is invertible. Therefore x, u ∈ J as
R is local so L ⊆ J × J. Hence J(M ) = J × J/L so M/J(M ) = R×R/L
J ×J/L ≃ R × R/J × J
which has dimension 2 as a module over the skewfield R/J. Since M = T ⊕ X and M is
finitely generated, then so are T and X and therefore J(X) 6= X and J(T ) 6= T . Then as
J(M ) = T
J(X) has dimension 2 over R/J, it follows that both T /J(T ) and X/J(X)
are simple. Hence T and X are local, and as they are finitely generated, it follows that they
are generated by any element not belonging to their Jacobson radical. Let T ′ (respectively
X ′) be the inverse images of T (and X respectively) in R × R and t ∈ T ′ and s ∈ X ′ be
such that Rt + L = T ′ and Rs + L = X ′. We have R × R = T ′ + X ′ = Rt + L + Rs + L =
(Rt + Rs) + L ⊆ (Rt + Rs) + J × J ⊆ R × R so (Rt + Rs) + J × J = R × R. Therefore
we obtain Rt + Rs = R × R because J × J is small in R × R.
J(T ) ⊕ X
M
14
MIODRAG CRISTIAN IOVANOV
Write t = (p, q) ∈ T ′. Then t = t + L ∈ T implies that there is λ 6= 0 in R such that
λt = 0 ∈ M and therefore there is µ ∈ R with λ(p, q) = µ(x, u). We show that either
p /∈ J or q /∈ J. Indeed assume otherwise: t = (p, q) ∈ J × J. Then we get Rt ⊆ J × J.
Because Rt + Rs = R × R we see that R × R/J × J must be generated over R by the
image of s. This shows that the R/J module R × R/J × J = (R/J)2 has dimension 1 and
this is obviously a contradiction.
Finally, suppose p /∈ J so p is invertible; then the equations λp = µx, λq = µu imply
λ = µxp−1 and µxp−1q = µu. But µ 6= 0 because p is invertible and λ 6= 0. Therefore we
obtain u = xp−1q; thus b = αu = αxp−1q = ap−1q showing that b ∈ aR i.e. bR ⊆ aR.
Similarly if q is invertible, we get aR ⊆ bR.
(cid:3)
Theorem 3.4. If C is an (infinite dimensional) local coalgebra with the left f.g. Rat-
splitting property, then C is a chain coalgebra.
Proof. We first show that every two principal left ideals of R are comparable. Suppose
there are two left ideals of R, R · x0 and R · y0 that are not comparable. Then as they
have finite codimension and C ∗ is infinite dimensional, we have Rx0 ∩ Ry0 6= 0 and take
0 6= αx0 = βy0 ∈ Rx0 ∩ Ry0. Then the right ideals αR and βR are not comparable, as
otherwise, if for example αR ⊆ βR, we would have a relation α = βλ so αx0 = βλx0 = βy0.
As β 6= 0 we get λx0 = y0 because R is a domain, and then Ry0 ⊆ Rx0, a contradiction.
By Proposition 3.2 the set {λR λR ⊆ αR ∩ βR} is Noetherian (relative to inclusion)
and let λR be a maximal element. If x ∈ αR ∩ βR then by Proposition 3.3 we have that
xR and λR are comparable and by the maximality of λR it follows that xR ⊆ λR, so
x ∈ λR. Therefore αR ∩ βR = λR. Note that λ 6= 0, because αR and βR are nonzero
ideals of finite codimension. Then we see that λR ≃ R as right R modules, because R is
a domain, and again by Proposition 3.3 any two principal right ideals of λR = αR ∩ βR
are comparable, so the same must hold in RR. But this is in contradiction with the fact
that αR and βR are not comparable, and therefore the initial assertion is proved.
Now we prove that J n/J n+1 is a simple right module for all n. As R/J is semisimple
(it is a skewfield) and J n/J n+1 has an R/J module structure, it follows that J n/J n+1
is a semisimple left R/J-module and then J n/J n+1 is semisimple also as R-module. If
we assume that it is not simple, then there are f, g ∈ J n \ J n+1 such that R f = (Rf +
J n+1)/J n+1 and Rg = (Rg+J n+1)/J n+1 are different simple R-modules, so R f ∩Rg = 0 in
J n/J n+1. Then (Rf + J n+1) ∩ (Rg + J n+1) = J n+1 which shows that Rf and Rg cannot
be comparable, a contradiction. As J n = C ⊥
n−1, we see that dim(Cn−1) = codim(J n).
Then for n ≥ 1, dim(Cn/Cn−1) = dim(Cn) − dim(Cn−1) = codimR(J n) − codimR(J n+1) =
dim(J n/J n+1) = dim(C0). Because C0 is the only type of simple right C-comodule, this
last relation shows that the right C-comodule Cn/Cn−1 must be simple. Therefore C must
be a chain coalgebra.
(cid:3)
We may now combine the results of Sections 2 and 3 and obtain
Corollary 3.5. Let C be a co-local (infinite dimensional) coalgebra. Then C is a left
(right) finite splitting coalgebra if and only if C is a chain coalgebra. Moreover, if the base
field K is algebraically closed or the coalgebra C is pointed, then this is further equivalent
to the fact that C is isomorphic to the divided power coalgebra.
Proof. This follows from Theorems 2.5, 2.7 and 3.4.
(cid:3)
WHEN DOES THE RATIONAL TORSION SPLIT OFF FOR FINITELY GENERATED MODULES 15
4. Serial coalgebras and General Examples
In this section we provide some nontrivial general examples of non-colocal coalgebras for
which this splitting property holds.
Lemma 4.1. Let C = D ⊕ E be coproduct of two coalgebras D and E. Then C has the
left f.g. Rat-splitting property if and only if D and E have the Rat-splitting property.
Proof. Assume C has the left f.g. Rat-splitting property.
It is well known that the
category of modules over C ∗ ≃ D∗ ×E∗ is isomorphic to the product of the category of D∗-
modules with that of E∗-modules; in this respect, if M is a left C ∗-module, then M = N ⊕P
where N = E⊥ · M , P = D⊥ · M are C ∗ submodules that have an induced D∗ = C ∗/D⊥-
and respectively E∗ = C ∗/E⊥-module structure (since D⊥ · N = 0 = E⊥ · P ). Also, one
can check that a D∗-module X is rational if and only if it is rational as C ∗-module with
its induced C ∗-module structure: if ρ : X → X ⊗ C is a C-comultiplication then we must
have ρ(X) ⊆ X ⊗ D since D⊥ cancels X, and ρ becomes a D-comultiplication. Indeed, if
j ∈ X assumed linearly independent, yi ∈ D and
ρ(x) = Pi
j ∈ E, then for any e∗ ∈ C ∗ such that e∗D = 0, we have 0 = e∗ · x = Pj
j ∈ (D⊥)⊥ = D so x′
xi ⊗ yi +Pj
j) = 0 by linearly independence. This shows that x′
e∗(y′
j = 0 for all
j. Thus, we obtain that Rat(D∗N ) = Rat(C ∗N ) and Rat(E ∗P ) = Rat(C ∗P ), and we have
direct sums N = Rat(N )⊕N ′ and P = Rat(P )⊕P ′ in D∗M and E ∗M; but N ′ and P ′ also
have an induced C ∗-module structure with E∗ = D⊥ acting as 0, and we finally observe
that this yields a direct sum of C ∗ modules M = Rat(C ∗N ) ⊕ N ′ ⊕ Rat(C ∗P ) ⊕ P ′ =
Rat(C ∗M ) ⊕ (N ′ ⊕ P ′).
The other implication follows from Proposition 1.9.
(cid:3)
x′
j ⊗ y′
j with xi, x′
y′
e∗(y′
j)x′
j, so
We note now the following proposition which was also proved in [C], but with techniques
involving general results of M. Teply from [T1] and [T3].
Proposition 4.2. Assume C is a cocommutative coalgebra. Then C is is a f.g. Rat-
splitting coalgebra if and only if it is a finite coproduct of finite dimensional coalgebras and
infinite dimensional chain coalgebras. Moreover, these chain coalgebras are isomorphic to
the divided power coalgebra in any of the cases:
(i) the base field is algebraically closed;
(ii) C is pointed.
Proof. Since C is cocommutative, C =
Ci, where Ci are colocal subcoalgebras of C.
n
Li=1
Now each of the Ci must have the splitting property for finitely generated modules by
Proposition 1.9, and therefore they must be either finite dimensional or be chain coalge-
bras. The converse follows from the previous Lemma and the results of Section 2. The
final assertion comes from Theorem 2.7.
(cid:3)
Recall, for example from [F], 25.1.12 that a module M is called serial if it is a direct
sum of uniserial (chain) modules; a ring R is said to be left (right) serial if R is serial
when regarded as left (right) R-module, and serial when R is both left and right serial.
In analogy to these definitions, for a C-comodule M we say that M is serial if it is
serial when regarded as C ∗-module (so it is a direct sum of serial -or chain- comodules).
A coalgebra will be called left (right) serial if and only if it is serial as a right (left) C ∗-
module, i.e. as a left (right) C-comodule, and serial if it is both left and right serial. These
definitions coincide with those in [CGT]. We note at this point that in our definitions, a
16
MIODRAG CRISTIAN IOVANOV
uniserial coalgebra is the same as a chain coalgebra, while a uniserial coalgebra in [CGT]
is understood as a homogeneous uniserial coalgebra, that is, a coalgebra C that is serial
and the composition factors of each indecomposable injective comodule are isomorphic
(see Definition 1.3 [CGT]). The following is a generalization of Proposition 1.6, [CGT].
Proposition 4.3. Let C be a coalgebra. Then the following are equivalent:
(i) C is a right serial coalgebra and C0 is finite dimensional.
(ii) C ∗ is a right serial algebra.
Consequently C ∗ is serial if an only if C is serial and C0 is finite dimensional, equivalently,
C is serial and C ∗ is semilocal.
Proof. (i)⇒(ii) Let C0 =
k
Li=1
Si be a decomposition of C0 into simple right comodules,
k
Li=1
E(Si) be an injective envelope of Si contained in C; then C =
E(Si) in MC and
Li=1
C ∗M. Since any other decomposition of C in MC is equivalent to this one, we have that
E(Si) are chain comodules and then E(Si)∗ are chain modules by Proposition 2.2. As
C ∗ =
E(Si)∗ in MC ∗ we get that C ∗ is right serial.
n
(ii)⇒(i) If C ∗ is right serial, it is a direct sum of uniserial modules C ∗ = Li
which has to be cyclic; then we easily see that these modules have to be local (for example
by [F], 25.4.1B) and indecomposable (a finitely generated local module is indecomposable).
Since there can be only a finite number of Mi's in a decomposition of C ∗, and each of the
Mi's are local we get that C ∗ is semilocal, and then C ∗/J is semisimple (J = C ⊥
0 ). But
Li=1
0 and thus C0 is cosemisimple finite dimensional. Then C ∗ =
C ∗/J = C ∗/C ⊥
Mi, each of
0 = C ∗
Mi
k
Mj is finitely generated, it is closed in
with Mi local uniserial. Let Ei = (Lj6=i
the finite topology of C ∗ and therefore E⊥
Mj)⊥; since Lj6=i
i = Lj6=i
Mj, so E∗
i ≃ C ∗/E⊥
i = C ∗/(Lj6=i
Mj) ≃ Mi.
Then by Proposition 2.2 we get that Ei is a right chain C-comodule; also because of the
anti-isomorphism of latices between the right subcomodules of C and closed right C ∗-
modules of C ∗ (see [DNR] or [I0], Theorem 1), we get that C =
comodules. Thus C is a left serial coalgebra.
k
Li=1
Ei, with Ei right chain
(cid:3)
We say that a coalgebra C is purely infinite dimensional serial if it is serial and the
uniserial left (and also the uniserial right) comodules into which it decomposes are infinite
dimensional. Equivalently, one can say that injective envelopes of any left (and also
every right) simple C-comodule is infinite dimensional. It is not difficult to see that for an
almost connected coalgebra it is enough to ask that only left injective envelopes are infinite
k
Li=1
dimensional: let C =
E(Si) be a decomposition of C with Si simple left comodules and
E(Si) an injective envelope for each Si. Assume C is serial; then each E(Si) is uniserial.
Then if LnE(Si) is the n-th term in the Loewy series of E(Si) then Cn =
LnE(Si)
and E(Si) is infinite dimensional for all i if and only if Ln−1E(Si) 6= LnE(Si) for all i
and all n ≥ 0 (L−1 = 0), equivalently, Cn/Cn−1 ≃
LnE(Si)/Ln−1E(Si) has length k
k
Li=1
k
Li=1
WHEN DOES THE RATIONAL TORSION SPLIT OFF FOR FINITELY GENERATED MODULES 17
(as a module) for all n. Since this last condition is a left-right symmetric condition, the
assertion follows. The next proposition provides the general example of this section:
Proposition 4.4. Let C be a purely infinite dimensional serial coalgebra which is almost
connected. Then C has the left (and also the right) f.g. Rat-splitting property.
Proof. By the previous proposition, C ∗ is serial. Let M be a left finitely generated
k
C ∗-module. Let C =
Li=1
comodules; then C ∗ = Li∈I
E(Si) be a decomposition as above, in CM, with E(Si) chain
E(Si)∗ in C ∗M. By Remark 2.4 and Proposition 1.3 the E(Si)∗'s
Lj=1
are noetherian. Hence C ∗ is Noetherian (both left and right, since C is left and right serial).
This shows that every finitely generated C ∗-module is also finitely presented. Then, by
Mj with Mj cyclic uniserial left C ∗-modules. For each j
[F], Corollary 25.3.4, M =
n
there are two possibilities:
• Mj is finite dimensional. Let mj be a generator of the left C ∗-module Mj, and then let
I = annC ∗(mj). Then I is a left ideal of C ∗ and it is finitely generated (C ∗ is Noetherian),
so I = X ⊥, X ⊆ C (Lemma 1.1). Moreover, C ∗/I ≃ C ∗ · mj = Mj and so I has finite
codimension since Mj is finite dimensional. Hence X is finite dimensional and is a left
subcomodule of C. Then Mj ≃ C ∗/X ⊥ ≃ X ∗, following that Mj is rational as a dual of
the rational right C ∗-module X. So Rat(Mj) = Mj.
• Mj is infinite dimensional. Let mj be a generator of Mj as before, and S = Mj/J(Mj )
which is a simple module because Mj is local since it is cyclic and uniserial. Let Pi =
E(Si)∗; since C ∗/J =
Then we have a diagram
k
Li=1
Pi/JPi and all Pi are local, there is some i such that Pi/JPi ≃ S.
Pi
p
}
}
}
u
}
}
}
}
~}
Mj
/ S
π
/ 0
completed commutatively by u since Pi is projective, and p, π are the canonical maps.
Note that u is surjective, since otherwise Im(u) ⊆ Ker (π) because Ker (π) is the only
maximal submodule of the finitely generated module Mi. This cannot happen since πu =
p 6= 0. By Remark 2.4 and Proposition 1.3, we see that any nonzero submodule of
Pi = E(Si)∗ has finite codimension. Then if Ker (u) 6= 0, Mj = Im(u) ≃ Pi/Ker (u) would
be finite dimensional, which is excluded by the hypothesis on Mj. This shows that u is an
isomorphism so Mj ≃ E(Si)∗ and we now get that Mj has no finite dimensional submodules
besides 0 (again by Remark 2.4 and Proposition 1.3). This shows that Rat(Mj) = 0
Finally, if we set F = {j Mj finite dimensional}, we see that Rat(M ) =
Mj, and this shows that Rat(M ) is a direct summand in M =
Lj∈F
Mj.
n
Lj=1
n
Lj=1
Rat(Mj) =
(cid:3)
Example 4.5. Let K be a field, q ≥ 1 and σ ∈ Sq be a permutation of {1, 2, . . . , q}.
Denote by K q
σ[X] the vector space with basis xp,n with p ∈ {1, 2 . . . , q} and n ≥ 0. Define
~
/
/
18
MIODRAG CRISTIAN IOVANOV
a comultiplication ∆ and a counit ε on K q
∆(xp,n) = Xi+j=n
σ[X] as follows:
xp,i ⊗ xσi(p),j
ε(xp,n) = δn,0, ∀ p ∈ {1, 2, . . . , q}, n ≥ 0
It is easy to see that ∆ is coassociative and ε becomes a counit, so K q
coalgebra:
σ[X] becomes a
(∆ ⊗ I)∆(xp,n) = (∆ ⊗ I)( Xi+j=n
xp,i ⊗ xσi(p),j)
xp,s ⊗ xσs(p),t ⊗ xσi(p),j
xp,s ⊗ xσs(p),t ⊗ xσs+t(p),j
= Xi+j=n Xs+t=i
= Xs+t+j=n
= Xs+u=n
= (I ⊗ ∆)( Xs+u=n
xp,s ⊗ Xt+j=u
= (I ⊗ ∆)∆(xp,n)
xσs(p),t ⊗ xσt(σs(p)),j
xp,s ⊗ xσs(p),u)
n
xp,iε(xσi(p),j) =
Pi=0
δi,0xσi(p),n−i = xp,n and Pi+j=n
σ[X] together with these morphisms is a coalgebra.
ε(xp,i)xσi(p),j =
xp,iδj,0 = xp,n, showing that K q
Also, we have Pi+j=n
Pi+j=n
Let Ep be the vector subspace of K q
subcomodules of K q
comodules in several steps:
(i) Let Ep,n =< xp,0, xp,1, . . . , xp,n > be the space with basis {xp,0, xp,1, . . . , xp,n}; it is
actually a right subcomodule of Ep. We note that Ep/Ep,n ≃ Eσn+1(p). Indeed, if x denotes
the image of x ∈ Ep in Ep/Ep,n, we have the following formulas for the comultiplication
of Ep/Ep,n
σ[X] with basis xp,n, n ≥ 0. Note that the Ep's are right
σ[X] (obviously by the definition of ∆ and ε). We show Ep are chain
xp,m 7−→ Xi+j=m,i≥n+1
xp,i ⊗ xσi(p),j = Xi+j=m−n−1
xp,i+n+1 ⊗ xσi(σn+1(p)),j
for m ≥ n + 1. The comultiplication of Eσn+1(p) is given by the formulas:
xσn+1(p),s 7−→ Xi+j=s
xσn+1(p),i ⊗ xσi(σn+1(p)),j
σ[X]-comodules.
These relations show that the correspondence xp,i+n+1 7−→ xσn+1(p),i is an isomorphism of
K q
(ii) Let x = λ0xp,0 + λ1xp,1 + · · · + λp,nxp,n ∈ Ep and assume λn 6= 0. Let f ∈ K q
σ[X]∗
be equal to 1 on xp,n and 0 on the rest of the elements of the basis xt,i. Then one easily
λi+jxp,if (xσi(p),j) = λnxp,0 (the only terms remaining are the one
sees that f · x = Pi+j≤n
having j = n, i = 0, and such a term occurs only once in this sum). Since λn 6= 0, we get
that xp,0 belongs to the subcomodule generated by x. This shows that Ep,0 is contained in
any submodule of Ep. This shows that that Ep is colocal and Ep,0 is its socle (which is a
simple comodule).
(iii) An inductive argument now sows that Ep,n are chain comodules for all n. Indeed, by
WHEN DOES THE RATIONAL TORSION SPLIT OFF FOR FINITELY GENERATED MODULES 19
the isomorphism in (i) and by (ii), we have that Ep,n+1/Ep,n ≃ Eσn+1(p),0. This shows
that Ep is a chain comodule by Proposition 2.2.
Since K q
σ[X]-comodules, we see that K q
σ[X] is right serial, so it
Ep as right K q
σ[X] =
q
Lp=1
is serial by Proposition 4.3 and even purely infinite dimensional, and thus constitutes an
example of a left and right f.g. Rat-splitting coalgebra by Proposition 4.4.
More examples can be obtainted by
Corollary 4.6. If C = D ⊕ E where D is a finite dimensional coalgebra and E is a
purely infinite serial dimensional coalgebra, then C has the both the left and the right f.g.
Rat-splitting property.
Remark 4.7. The fact that K q
σ[X] is also left serial (and then purely infinite dimensional)
can also follow by noting that K q
σ−1 [X] as coalgebras. It is also interesting to
note that if σ = σ1 . . . σr is a decomposition of σ into disjoint cycles of respective lengths
q1, . . . , qr (or, more generally, into mutually commuting permutations), then there is an
isomorphism of coalgebras
σ[X]op ≃ K q
r
K q
σ[X] ≃
K qi
σi[X]
Mi=1
We omit the proofs here. As a final comment, we note that by the above results, some
natural questions arise:
is the concept of f.g. Rat-splitting left-right symmetric? That
is, does the left f.g. Rat-splitting property of a coalgebra also imply the right f.g. Rat-
splitting property? One should note that all the above examples have both the left and
the right Rat-splitting property. Also, it would be interesting to know whether a general-
ization of the results in the local case hold in the general non-cocommutative case as the
cocommutative case of this section and the above non-cocommutative examples seem to
suggest:
if C has the left f.g. Rat-splitting property, can it be written as a direct sum
of finite dimensional injectives and infinite dimensional chain injectives (likely in CM),
or maybe a decomposition of coalgebras as in Corollary 4.6, and to what extent such a
decomposition would characterize this property?
The author wishes to thank C. Nastasescu for useful remarks on the subject as well as for
his support throughout the past years.
Acknowledgment
References
[A] E. Abe, Hopf Algebras, Cambridge Univ. Press, 1977.
[AN] T. Albu, C. Nastasescu, Relative Finiteness in Module Theory, Monogr. Textbooks Pure Appl. Math.,
vol. 84, Dekker, New York 1984.
[AF] D. Anderson, K. Fuller, Rings and Categories of Modules, Grad. Texts in Math., Springer, Berlin-
Heidelberg-New York, 1974.
[BW] T. Brzezi´nski and R. Wisbauer, Corings and comodules, London Math. Soc. Lect. Notes Ser. 309,
Cambridge University Press, Cambridge, 2003.
[C] J. Cuadra, When does the rational submodule split off ?, Ann. Univ. Ferrarra -Sez. VII- Sc. Mat. Vol.
LI (2005), 291-298.
[CGT] J. Cuadra, J. Gomez-Torrecillas, Serial Coalgebras, J. Pure App. Algebra 189 (2004), 89-107.
20
MIODRAG CRISTIAN IOVANOV
[DNR] S. Dascalescu, C. Nastasescu, S¸. Raianu, Hopf Algebras: an introduction. Vol. 235. Lecture Notes
in Pure and Applied Math. Vol.235, Marcel Dekker, New York, 2001.
[F] C. Faith, Algebra II: Ring Theory. Vol 191, Springer-Verlag, Berlin-Heidelberg-New York, 1976.
[GTN] J. G´omez-Torrecillas, C. Nastasescu, Quasi-co-Frobenius coalgebras, J. Algebra 174 (1995), 909-923.
[GMN] J. G´omez-Torrecillas, C. Manu, C. Nastasescu, Quasi-co-Frobenius coalgebras II, Comm. Algebra
Vol 31, No. 10, pp. 5169-5177, 2003.
[GNT] J. G´omez-Torrecillas, C. Nastasescu, B. Torrecillas, Localization in coalgebras. Applications to
finiteness conditions, J. Algebra Appl. 6 (2007), no. 2, 233 -- 243. eprint arXiv:math/0403248,
http://arxiv.org/abs/math/0403248.
[I] M.C. Iovanov, Co-Frobenius Coalgebras, J. Algebra 303 (2006), no. 1, 146 -- 153;
eprint arXiv:math/0604251, http://xxx.lanl.gov/abs/math.QA/0604251.
[I0] M.C. Iovanov, Characterization of PF rings by the finite topology on duals of R modules, An. Univ.
Bucure¸sti Mat. 52 (2003), no. 2, 189-200.
[I1] M.C. Iovanov, The Splitting Problem for Coalgebras: A Direct Approach, Applied Categorical Struc-
tures 14 (2006) - Categorical Methods in Hopf Algebras - no. 5-6, 599-604.
[IO] .C. Iovanov, The finite Rat-splitting for coalgebras, preprint arXiv:math/0612478, old version of this
paper.
[K1] I. Kaplansky, Modules over Dedekind rings and valuation rings, Trans. Amer. Math. Soc. 72 (1952)
327-340.
[K2] I. Kaplansky, A characterization of Prufer domains, J. Indian Math. Soc. 24 (1960) 279-281.
[LS] C. Lomp, A. Sant'ana, Chain Coalgebras and Distributivity, J. Pure Appl. Algebra 211 (2007), no. 3,
581 -- 595. eprint arXiv:math.RA/0610135.
[NT] C. Nastasescu, B. Torrecillas, The splitting problem for coalgebras, J. Algebra 281 (2004), 144-149.
[NTZ] Nastasescu, B. Torrecillas, Y. Zhang, Hereditary Coalgebras, Comm. Algebra 24 (1996), 1521-1528.
[Rot] J. Rotman, A characterization of fields among integral domains, An. Acad. Brasil Cienc. 32 (1960)
193-194.
[Sw] M.E. Sweedler, Hopf Algebras, Benjamin, New York, 1969.
[T1] M.L. Teply, The torsion submodule of a cyclic module splits off, Canad. J. Math. XXIV (1972) 450-
464.
[T2] M.L. Teply, A history of the progress on the singular splitting problem, Universidad de Murcia, De-
partamento de ´Algebra y Fundamentos, Murcia, 1984, 46pp.
[T3] M.L. Teply, Generalizations of the simple torsion class and the splitting properties, Canad. J. Math.
27 (1975) 1056-1074.
Miodrag Cristian Iovanov
University of Bucharest, Faculty of Mathematics, Str. Academiei 14
RO-010014, Bucharest, Romania
and
State University of New York @ Buffalo
Department of Mathematics, 244 Mathematics Building
Buffalo, NY 14260-2900, USA
E -- mail address: [email protected]; [email protected]
|
1503.00740 | 1 | 1503 | 2015-03-02T21:05:00 | The Interplay of Algebra and Geometry in the Setting of Regular Algebras | [
"math.RA"
] | This article is based on a talk given by the author at MSRI in the workshop "Connections for Women" in January 2013, while being a part of the program "Noncommutative Algebraic Geometry and Representation Theory" at MSRI. One purpose of the exposition is to motivate and describe the geometric techniques introduced by M. Artin, J. Tate and M. Van den Bergh in the 1980s at a level accessible to graduate students. Additionally, some advances in the subject since the early 1990s are discussed, including a recent generalization of complete intersection to the noncommutative setting, and the notion of graded skew Clifford algebra and its application to classifying quadratic regular algebras of global dimension at most three. The article concludes by listing some open problems. | math.RA | math | THE INTERPLAY OF ALGEBRA AND GEOMETRY IN THE
SETTING OF REGULAR ALGEBRAS
5
1
0
2
r
a
M
2
Michaela Vancliff∗
Department of Mathematics, P.O. Box 19408,
University of Texas at Arlington,
Arlington, TX 76019-0408
[email protected]
uta.edu/math/vancliff
]
.
A
R
h
t
a
m
[
1
v
0
4
7
0
0
.
3
0
5
1
:
v
i
X
r
a
Abstract. This article is based on a talk given by the author at MSRI in the workshop
Connections for Women in January 2013, while being a part of the program Noncommutative
Algebraic Geometry and Representation Theory at MSRI. One purpose of the exposition is to
motivate and describe the geometric techniques introduced by M. Artin, J. Tate and M. Van
den Bergh in the 1980s at a level accessible to graduate students. Additionally, some advances
in the subject since the early 1990s are discussed, including a recent generalization of complete
intersection to the noncommutative setting, and the notion of graded skew Clifford algebra
and its application to classifying quadratic regular algebras of global dimension at most three.
The article concludes by listing some open problems.
Introduction
Many non-commutative algebraists in the 1980s were aware of the successful marriage of
algebra and algebraic geometry in the commutative setting and wished to duplicate that
relationship in the non-commutative setting. One such line of study was the search for a
subclass of non-commutative algebras that "behave" enough like polynomial rings that a
geometric theory could be developed for them. One proposal for such a class of algebras are
the regular algebras, introduced in [2], that were investigated using new geometric techniques
in the pivotal papers of M. Artin, J. Tate and M. Van den Bergh ([3, 4]).
About the same time, advances in quantum mechanics in the 20th century had produced
many new non-commutative algebras on which traditional techniques had only yielded limited
success, so a need had arisen to find new techniques to study such algebras (c.f., [13, 16, 31,
32, 43]). One such algebra was the Sklyanin algebra, which had emerged from the study
2010 Mathematics Subject Classification. 16S38, 14A22, 16E65, 16S37, 14M10.
Key words and phrases. regular algebra, Clifford algebra, skew polynomial ring, base point, point module,
quadric, complete intersection.
∗M. Vancliff was supported in part by NSF grants DMS-0900239 and DMS-1302050 and by general mem-
bership at MSRI.
2
ALGEBRA & GEOMETRY IN THE SETTING OF REGULAR ALGEBRAS
of quantum statistical mechanics ([31, 32]). By the early 1990s, T. Levasseur, S. P. Smith,
J. T. Stafford and others had solved the 10-year old open problem of completely classifying all
the finite-dimensional irreducible representations (simple modules) over the Sklyanin algebra,
and their methods were the geometric techniques developed by Artin, Tate and Van den Bergh
([22, 33, 34]).
Concurrent with the above developments, another approach was considered via differential
geometry and deformation theory to study the algebras produced by quantum physics. That
approach is the study of certain non-commutative algebras via Poisson geometry (c.f., [12]).
At the heart of both approaches are homological and categorical techniques, so it is perhaps
no surprise that the two approaches have much overlap; often, certain geometric objects from
one approach are in one-to-one correspondence with various geometric objects from the other
approach (depending on the algebra being studied -- c.f., [44, 45, 46]). A survey of recent
advances in Poisson geometry may be found in [15].
Given the above developments, the early 1990s welcomed a new era in the field of non-
commutative algebra in which geometric techniques took center stage. Since that time, the
subject has spawned many new ideas and directions, as demonstrated by the MSRI programs
in 2000 and 2013.
This article is based on a talk given by the author in the Connections for Women workshop
held at MSRI in January 2013 and it has two objectives. The first is to motivate and describe
the geometric techniques of Artin, Tate and Van den Bergh at a level accessible to graduate
students, and the second is to discuss some developments towards the attempted classification
of quadratic regular algebras of global dimension four, while listing open problems. An outline
of the article is as follows.
Section 1 concerns the motivation and development of the subject, with emphasis on qua-
dratic regular algebras of global dimension four. Section 2 discusses constructions of certain
types of quadratic regular algebras of arbitrary finite global dimension, with focus on graded
Clifford algebras and graded skew Clifford algebras. This section also discusses a new type of
symmetry for square matrices called µ-symmetry. We conclude this section by revisiting the
classification of quadratic regular algebras of global dimension at most three, since almost all
such algebras may be formed from regular graded skew Clifford algebras. In Section 3, we
discuss geometric techniques that apply to graded Clifford algebras and graded skew Clifford
algebras in order to determine when those algebras are regular. This section also considers the
issue of complete intersection in the non-commutative setting. We conclude with Section 4
which lists some open problems and related topics.
Although the main objects of study from [3, 4] are discussed in this article, several topics
from [3, 4] are omitted; for surveys of those topics, the reader is referred to [35, 36] and to
D. Rogalski's lecture notes, [26], from the graduate workshop "Noncommutative Algebraic
Geometry" at MSRI in June 2012.
ALGEBRA & GEOMETRY IN THE SETTING OF REGULAR ALGEBRAS
3
Acknowledgments. The author is grateful to MSRI for general membership in 2013 and
for providing an office during her visits. Additionally, special thanks are extended to the
organizers of the workshop Connections for Women in January 2013 for their invitation to
give a talk in the workshop. The environment at MSRI was particularly stimulating during the
program Noncommutative Algebraic Geometry and Representation Theory in Spring 2013, and
for that the author warmly thanks the organizers and participants of the program, especially
T. Cassidy, M. Van den Bergh and J. T. Stafford.
1. The Geometric Objects
In this section, we discuss the motivation and development of the subject, with emphasis
on quadratic regular algebras of global dimension at most four.
Throughout this section, k denotes an algebraically closed field and, for any graded alge-
bra B, the span of the homogeneous elements of degree i will be denoted by Bi.
1.1. Motivation.
Consider the k-algebra, S, on generators z1, . . ., zn with defining relations:
zjzi = µijzizj,
for all distinct i, j,
where 0 6= µij ∈ k for all i, j, and µijµji = 1 for all distinct i, j. If µij = 1, for all i, j, then S
is the commutative polynomial ring and has a rich subject of algebraic geometry associated
with it; in particular, by the (projective) Nullstellensatz, the points of P(S1
) are in one-to-
one correspondence with certain ideals of S via (α1, . . . , αn) ↔ hαiz1 − α1zi, . . . , αizn − αnzii,
where αi 6= 0. Before continuing, we first observe that for such an ideal I, the graded module
S/I has the property that its Hilbert series is H(t) = 1/(1 − t) and that S/I is a 1-critical
(with respect to GK-dimension) graded cyclic module over S.
∗
However, if µij 6= 1 for any i, j, then S still "feels" close to commutative, and one would
expect there to be a way to relate algebraic geometry to it. The geometric objects in [3] are
modelled on the module S/I above; instead of using actual points or lines etc, certain graded
modules are used as follows.
1.2. Points, Lines, etc.
Definition 1.1. [3] Let A = L∞
k-algebra generated by A1 where dim(A1) = n < ∞. A graded right A-module M = L∞
is called a right point module (respectively, line module) if:
i=0 Ai denote an N-graded, connected (meaning A0 = k)
i=0 Mi
(a) M is cyclic with M = M0A, and
(b) dimk(Mi) = 1 for all i (respectively, dimk(Mi) = i + 1) for all i.
4
ALGEBRA & GEOMETRY IN THE SETTING OF REGULAR ALGEBRAS
If A is the polynomial ring S, then the module S/I from §1.1 is a point module. In general,
one may associate some geometry to point and line modules as follows. Condition (a) implies
that A maps onto M via a 7→ ma, for all a ∈ A, where {m} is a k-basis for M0, and this
map restricts via the grading to a linear map θ : A1 → M1. Let K ⊂ A1 denote the kernel
of θ. Condition (b) implies that dimk(K) = n − 1 (respectively, n − 2), so that K ⊥ ⊂ A1
has dimension one (respectively, two). Thus, P(K ⊥) is a point (respectively, a line) in the
geometric space P(A1
).
∗
∗
The Hilbert series of a point module is H(t) = 1/(1 − t), whereas the Hilbert series of a line
module is 1/(1 − t)2. Hence, a plane module is defined as in Definition 1.1 but condition (b) is
replaced by the requirement that the module have Hilbert series 1/(1−t)3 (c.f., [3]). Similarly,
one may define d-linear modules, where the definition is modelled on Definition 1.1, but the
module has Hilbert series 1/(1 − t)d+1 (c.f., [29]).
For many algebras, d-linear modules are (d+1)-critical with respect to GK-dimension. This
leads to the following generalization of a point module.
i=0 Mi = M0A and M has Hilbert series HM (t) = c/(1 − t) for some c ∈ N.
Definition 1.2. [10] With A as in Definition 1.1, we define a right base-point module over A
to be a graded 1-critical (with respect to GK-dimension) right A-module M such that M =
L∞
If c = 1 in Definition 1.2, then the module is a point module; whereas if c ≥ 2, then the
module is called a fat point module ([1]). The only base-point modules over the polynomial
ring are point modules. On the other hand, in general, the algebra S from §1.1 can have fat
point modules, so fat point modules are viewed as generalizations of points, and this is made
more precise in [1].
In [3], Artin, Tate and Van den Bergh proved that, under certain conditions, the point
modules are parametrized by a scheme; that is, there is a scheme that represents the functor
of point modules. Later, in [48], this scheme was called the point scheme. A decade later,
in [29], it was proved by B. Shelton and the author that (under certain conditions) d-linear
modules are parametrized by a scheme; that is, there is a scheme that represents the functor
of d-linear modules. If d = 0, then this scheme is isomorphic to the point scheme; if d = 1,
the scheme is called the line scheme.
By factoring out a nonzero graded submodule from a point module, one obtains a truncated
point module as follows.
Definition 1.3. [3] With A as in Definition 1.1, we define a truncated right point module of
length m to be a graded right A-module M = Lm−1
i=0 Mi such that M is cyclic, M = M0A and
dimk(Mi) = 1 for all i = 0, . . . , m − 1.
For many quadratic algebras A, there exists a one-to-one correspondence between the trun-
cated point modules over A of length three and the point modules over A. Moreover, if the
ALGEBRA & GEOMETRY IN THE SETTING OF REGULAR ALGEBRAS
5
algebra A in Definition 1.3 is quadratic, then the truncated point modules of length three are
in one-to-one correspondence with the zero locus in P(A1
) of the defining relations
of A. To see this, we fix a k-basis {x1, . . . , xn} for A1, and use T to denote the free k-algebra
on x1, . . . , xn, and let Z ⊂ P(A1
) denote the zero locus of the defining relations
of A. Viewing each xi as the i'th coordinate function on A1
) and
), where αi, βi ∈ k for all i = 1, . . . , n. Let M = kv0 ⊕ kv1 ⊕ kv2 denote a
r = (βi) ∈ P(A1
three-dimensional vector space that is a T -module via the action determined by
, let p = (αi) ∈ P(A1
) × P(A1
∗
∗
∗
) × P(A1
∗
∗
∗
∗
v0xi = αiv1,
v1xi = βiv2,
v2xi = 0,
for all i. It follows that M is a truncated point module over T of length three. If g ∈ T2,
then v1g = 0 = v2g and v0g = g(p, r)v2. In particular, if f ∈ T2 is a defining relation of A,
then Mf = 0 if and only if f (p, r) = 0. Hence, M is an A-module if and only if (p, r) ∈ Z.
This one-to-one correspondence between Z and truncated point modules of length three also
exists at the level of schemes; the reason being that the scheme Z represents the functor
of truncated point modules of length three. The method of proof of this is to repeat the
preceding argument for a truncated point module of length three over R ⊗k T and R ⊗k A,
where R is a commutative k-algebra, together with localization techniques; for details the
reader is referred to [3, Proposition 3.9], its proof, and the paragraph preceding that result.
This correspondence will be revisited in §1.4.
For completeness, we finish this subsection with some technical definitions that play mi-
nor roles throughout the text. The reader is referred to [21, 22] for details and for results
concerning algebras satisfying these definitions.
Definition 1.4. [21, Definition 2.1] A noetherian ring B is called Auslander-regular (respec-
tively, Auslander-Gorenstein) if
(a) the global homological dimension (respectively, (left and right) injective dimension)
of B is finite, and
(b) every finitely generated B-module M satisfies the Auslander condition, namely, for
B(M, B), we have j(N) ≥ i, where
every i ≥ 0 and for every B-submodule N of Exti
j(N) = inf{ℓ : Extℓ
B(N, B) 6= 0}.
Definition 1.5. [21, Definition 5.8] A noetherian k-algebra B of integral GK-dimension n
satisfies the Cohen-Macaulay property if GKdim(M) + j(M) = n for all nonzero finitely
generated B-modules M.
6
ALGEBRA & GEOMETRY IN THE SETTING OF REGULAR ALGEBRAS
1.3. Regular Algebras.
The goal of [3] was to classify, in a user-friendly way, the generic regular algebras of global
dimension three that were first analysed in [2]. In [3], such algebras were shown to be noe-
therian by using the geometric techniques developed in [3]. Regular algebras are often viewed
as non-commutative analogues of polynomial rings and are defined as follows.
Definition 1.6. [2] A finitely generated, N-graded, connected k-algebra A = L∞
erated by A1, is regular (or AS-regular) of global dimension r if
i=0 Ai, gen-
(a) it has global homological dimension r < ∞, and
(b) it has polynomial growth (i.e., there exist positive real numbers c and δ such that
dimk(Ai) ≤ ciδ for all i), and
(c) it satisfies the Gorenstein condition, namely, a minimal projective resolution of the left
trivial module Ak consists of finitely generated modules and dualizing this resolution
yields a minimal projective resolution of the right trivial module kA[e], shifted by some
degree e.
Although all three conditions in Definition 1.6 are satisfied by the polynomial ring, the main
reason a regular algebra is viewed as a non-commutative analogue of a polynomial ring is
due to condition (c), since it imposes a symmetry condition on the algebra that replaces
the symmetry condition of commutativity. The reader should note that, in the literature,
(c) is sometimes replaced by an equivalent condition that makes the symmetry property less
r is the Kronecker-delta symbol. An N-graded
obvious; namely, Exti
connected k-algebra that is generated by degree-1 elements and which is Auslander-regular
with polynomial growth is AS-regular ([21]). For a notion of regular algebra where the algebra
is not generated by degree-1 elements, see [6, 7, 37, 38, 39, 40].
A(Ak, A) ∼= δi
kA[e], where δi
r
Examples 1.7.
(a) The algebra S from §1.1 is regular.
(b) If k = C, then many algebras from physics are regular. In particular, homogenizations
of universal enveloping algebras of finite-dimensional Lie algebras, the coordinate ring of
quantum affine n-space, the coordinate ring of quantum m × n matrices, and the coordinate
ring of quantum symplectic n-space are all regular ([19, 20, 23])
(c) If the global dimension of a regular algebra is one, then the algebra is the polynomial
ring on one variable. However, by [2], if the global dimension is two, then there are two types
of such algebra as follows. For both types, the algebra has two generators, x, y, of degree one
and one defining relation f , where either f = xy − yx − x2 (Jordan plane) or f = xy − qyx
(quantum affine plane), where q ∈ k can be any nonzero scalar.
However, if the global dimension is three, then the situation is much richer; some of the
algebras are quadratic with three generators and three defining relations, whereas the rest
ALGEBRA & GEOMETRY IN THE SETTING OF REGULAR ALGEBRAS
7
have two generators and two cubic relations ([2]). Such algebras that are generic are classified
in [3] according to their point schemes, and in all cases, the point scheme is the graph of an
automorphism σ. Moreover, the algebra is a finite module over its center if and only if σ has
finite order.
1.4. Global Dimension Four.
Although many regular algebras of global dimension four have been extensively studied,
there is no classification yet. Recently, the progress towards classifying non-quadratic regular
algebras of global dimension four made good headway via the work in [24, 27]. However,
quadratic regular algebras of global dimension four constitute most of the regular algebras
of global dimension four, so their attempted classification is one of the motivating problems
that drives the subject forward. We end this section by summarizing some key results for this
latter case; in this setting, the algebra has four generators and six relations.
∗
∗
) × P(A1
∗
) × P(A1
In unpublished work, Van den Bergh proved in the mid-1990s that any quadratic (not nec-
essarily regular) algebra A on four generators with six generic defining relations has twenty
(counted with multiplicity) nonisomorphic truncated point modules of length three. Hence,
A has at most twenty nonisomorphic point modules. He also proved that if, additionally, A is
Auslander-regular of global dimension four, then A has a 1-parameter family of line modules.
For lack of a suitable reference, we outline the proof of these results. Let M(4, k) denote
the space of 4 × 4 matrices with entries in k. For the first result, we write points of P(A1
)
) to the matrix abT ∈ M(4, k), we have
as columns and, by mapping (a, b) ∈ P(A1
that P(A1
) is isomorphic to the scheme Ω1 of rank-1 elements in P(M(4, k)). Corre-
spondingly, the defining relations of A map to homogeneous degree-1 polynomial functions on
M(4, k), and their zero locus Z ′ ⊂ P(M(4, k)) can be identified with a P9. With these iden-
tifications, the zero locus Z ⊂ P(A1
) of the defining relations of A is isomorphic to
Ω1 ∩Z ′ ⊂ P(M(4, k)). Since Ω1 has dimension six and degree twenty, dim(Z) ≥ 6+9−15 = 0,
and, by B´ezout's Theorem, deg(Z) = 20. Hence, generically, Z is finite with twenty points,
so the first result follows by using the discussion after Definition 1.3. For the second result,
we identify A1 ⊗k A1 with M(4, k), and the assumption on regularity allows the application
of [22, Proposition 2.8], so that the line modules are in one-to-one correspondence with the
elements in the span of the defining relations of A that have rank at most two. In particular,
we compute dim(Ω2 ∩ ∆) in P(M(4, k)), where Ω2 denotes the elements in P(M(4, k)) of rank
at most two and ∆ denotes the projectivization of the image in P(M(4, k)) of the span of the
defining relations of A. Since ∆ ∼= P5 and dim(Ω2) = 11, the dimension is thus at least equal
to 11 + 5 − 15 = 1, so, generically, A has a 1-parameter family of line modules.
) × P(A1
∗
∗
∗
∗
In spite of Van den Bergh's work, it was still not clear that a regular algebra satisfying the
hypotheses from the preceding paragraph could have both a finite point scheme (especially one
of cardinality twenty) and a 1-dimensional line scheme simultaneously. However, in [49], the
8
ALGEBRA & GEOMETRY IN THE SETTING OF REGULAR ALGEBRAS
author proved with Van Rompay and Willaert, in the mid-1990s, that there exists a quadratic
regular algebra of global dimension four on four generators with six defining relations that
has exactly one point module (up to isomorphism) and a 1-parameter family of line mods.
Some years later, in 2000, Shelton and the author proved in [29] that if a quadratic algebra
on four generators with six defining relations has a finite scheme of truncated point modules
of length three, then that scheme determines the defining relations of the algebra. One should
note that this result assumes no hypothesis of regularity nor of any other homological data.
Moreover, by [52], this result is false in general if the scheme is infinite, even if the algebra is
assumed to be regular and noetherian.
Shelton and the author also proved in [29] that if a quadratic regular algebra of global
dimension four (that satisfies a few other homological conditions) has four generators and six
defining relations and a 1-dimensional line scheme, then that scheme determines the defining
relations of the algebra.
These last two results are counter-intuitive, since they seem to be saying that if the point
scheme (respectively, line scheme) is as small as possible, then the defining relations can be
recovered from it.
However, by the start of 2001, it was still unclear whether or not any quadratic regular
algebra exists that has global dimension four, four generators, six defining relations, exactly
twenty nonisomorphic point modules and a 1-dimensional line scheme. Fortunately, this was
resolved by Shelton and Tingey in [28] in 2001 in the affirmative. Sadly, their method to
produce their example used much trial and error on a computer, which they and others were
unable to duplicate to produce more examples. This hurdle likely had a negative impact on
the development of the subject, since it is difficult to make conjectures if there is only one
known example. Hence, a quest began to find an algorithm to construct such algebras, but
it was another several years before this situation was remedied, and that is discussed in the
next section.
2. Graded Clifford Algebras, Graded Skew Clifford
Algebras and Quantum Planes
This section describes a construction of a certain type of regular algebra of arbitrary finite
global dimension; such an algebra is called a graded skew Clifford algebra as it is modelled
on the construction of a graded Clifford algebra. If the global dimension is four, then this
construction is able to produce regular algebras that have the desired properties described at
the end of the previous section. We conclude this section by revisiting the classification of
quadratic regular algebras of global dimension three, and show that almost all such algebras
may be obtained from regular graded skew Clifford algebras.
ALGEBRA & GEOMETRY IN THE SETTING OF REGULAR ALGEBRAS
9
We continue to assume that k is algebraically closed; we additionally assume char(k) 6= 2.
We write M(n, k) for the space of n × n matrices with entries in k, and Mij for the entry in
the n × n matrix M that is in row i and column j.
2.1. Graded Clifford Algebras.
Definition 2.1. [5, 18] Let M1, . . . , Mn ∈ M(n, k) denote symmetric matrices. A graded
Clifford algebra (GCA) is the k-algebra C on degree-one generators x1, . . . , xn and on degree-
two generators y1, . . . , yn with defining relations given by:
(i) (degree-2 relations) xixj + xjxi =
n
X
k=1
(Mk)ij yk for all i, j = 1, . . . , n, and
(ii) degree-3 and degree-4 relations that guarantee yk is central in C for all k = 1, . . . , n.
In general, GCAs need not be quadratic nor regular, as demonstrated by the next example.
Example 2.2. Let M1 = h 2 −1
k-algebra on degree-one generators x1, x2 with defining relations
0 i and M2 = h 0 −1
−1
2 i. The corresponding GCA is the
−1
x1x2 + x2x1 = −x2
1 − x2
2,
1x2 = x2x2
x2
1,
so this algebra is not quadratic nor regular (as (x1 + x2)2 = 0). For more details on this
algebra, the reader may consult [47, Example 2.4].
GCAs C are noetherian by [3, Lemma 8.2], since dimk(C/hy1, . . . , yni) < ∞. Moreover,
since each matrix Mk in the definition is symmetric, we may associate a quadratic form
to Mk, and thereby associate a quadric in Pn−1 to Mk for each k. This means that for each
GCA, as in Definition 2.1, there is an associated quadric system Q in Pn−1. Quadric systems
are said to be base-point free if they yield a complete intersection; that is, the intersection of
all the quadrics in the quadric system is empty. Although Example 2.2 demonstrates that a
GCA need not be quadratic nor regular, if Q is base-point free, it determines these properties
of the associated GCA as follows.
Theorem 2.3. [5, 18] The GCA C is quadratic, Auslander-regular of global dimension n
and satisfies the Cohen-Macaulay property with Hilbert series 1/(1 − t)n if and only if the
associated quadric system is base-point free; in this case, C is regular and a domain.
In spite of this result, regular GCAs of global dimension four are not candidates for generic
quadratic regular algebras of global dimension four, since, although their point schemes can
be finite ([42, 49]), the symmetry of their relations prevents their line schemes from having
dimension one ([29]). The standard argument to prove this for a quadratic regular GCA C of
10
ALGEBRA & GEOMETRY IN THE SETTING OF REGULAR ALGEBRAS
global dimension four exploits the symmetry of the defining relations of C to move the compu-
tation of §1.4 inside P(W ), where W is the 10-dimensional subspace of M(4, k) consisting of
all symmetric matrices. Hence, using the notation from §1.4, ∆ ⊂ P(W ) and the line modules
are parametrized by (Ω2 ∩ P(W )) ∩ ∆ ⊂ P(W ); thus the dimension is at least 6 + 5 − 9, so it
is at least two.
Hence, a modification of the definition of GCA is desired in such a way that enough sym-
metry is retained so as to allow an analogue of Theorem 2.3 to hold, while, at the same time,
losing some symmetry so that the line scheme might have dimension one.
2.2. Graded Skew Clifford Algebras.
In order to generalize the notion of GCA and to have a result analogous to Theorem 2.3,
we need to generalize the notions of symmetric matrix and quadric system and make use of
normalizing sequences. For any N-graded k-algebra B, a sequence {g1, . . . , gm} of homoge-
neous elements of positive degree is called normalizing if g1 is a normal element in B and, for
each k = 1, . . . , m − 1, the image of gk+1 in B/hg1, . . . , gki is a normal element.
We write k× for k \ {0}.
Definition 2.4. [9]
(a) Let µ ∈ M(n, k×) satisfy µijµji = 1 for all distinct i, j. We say a matrix M ∈ M(n, k)
is µ-symmetric if Mij = µijMji for all i, j = 1, . . . , n. We write M µ(n, k) for the subspace of
M(n, k) consisting of all µ-symmetric matrices.
(b) Fix µ as in (a) and additionally assume µii = 1 for all i. Let M1, . . . , Mn ∈ M µ(n, k). A
graded skew Clifford algebra (GSCA) associated to µ and M1, . . . , Mn is a graded k-algebra A =
A(µ, M1, . . . , Mn) on degree-one generators x1, . . . , xn and on degree-two generators y1, . . . , yn
with defining relations given by:
(i) (degree-2 relations) xixj + µijxjxi =
n
X
k=1
(Mk)ijyk for all i, j = 1, . . . , n, and
(ii) degree-3 and degree-4 relations that guarantee the existence of a normalizing sequence
{y′
1, . . . , y′
n} that spans Pn
k=1
kyk.
Clearly, symmetric matrices and skew-symmetric matrices are µ-symmetric matrices for appro-
priate µ, and GCAs are GSCAs. Moreover, by [3, Lemma 8.2], GSCAs A are noetherian since
dimk(A/hy1, . . . , yni) < ∞. Furthermore, in Definition 2.4(b)(i), for all i, j, the ji-relation
can be deduced from the ij-relation by the µ-symmetry of the Mk.
Examples 2.5.
(a) With µ as in Definition 2.4(b), skew polynomial rings on generators x1, . . . , xn with
relations xixj = −µijxjxi,
for all i 6= j, are GSCAs.
ALGEBRA & GEOMETRY IN THE SETTING OF REGULAR ALGEBRAS
11
(b) (Quantum Affine Plane) Let n = 2, and M1 = h2
0
relations of A(µ, M1, M2) have the form:
0
0i and M2 = h0
0
0
2i. The degree-2
2x2
1 = 2y1,
2x2
2 = 2y2,
x1x2 + µ12x2x1 = 0,
so that khx1, x2i/hx1x2 + µ12x2x1i −։ A(µ, M1, M2). By Theorem 2.6 below, this map is
an isomorphism (see Examples 3.2(a)).
(c) ("Jordan" Plane) Let n = 2, and M1 = h 2
µ21
relations of A(µ, M1, M2) have the form:
1
0i and M2 = h0
0
0
2i. The degree-2
2x2
1 = 2y1,
2x2
2 = 2y2,
x1x2 + µ12x2x1 = y1 = x2
1,
so that khx1, x2i/hx1x2 + µ12x2x1 − x2
1i −։ A(µ, M1, M2). By Theorem 2.6 below, this
map is an isomorphism (see Examples 3.2(b)). Depending on the choice of µ12, this family of
examples contains the Jordan plane and some quantum affine planes.
(d) The quadratic regular algebra of global dimension four found by Shelton and Tingey
in 2001, in [28], and discussed above in §1.4, that has exactly twenty nonisomorphic point
modules and a 1-dimensional line scheme is a GSCA ([9]).
One can associate a non-commutative "quadric" to each µ-symmetric matrix Mk and, in
so doing, there is also a notion of "base-point free". These ideas are discussed in §3.2 below,
and yield a generalization of Theorem 2.3 as follows.
Theorem 2.6. [9] The GSCA A is quadratic, Auslander-regular of global dimension n and
satisfies the Cohen-Macaulay property with Hilbert series 1/(1−t)n if and only if the associated
quadric system is normalizing and base-point free; in this case, A is regular and a domain and
uniquely determined, up to isomorphism, by the data µ, M1, . . . , Mn.
Theorem 2.6 allowed the production in [9] of many algebras that are candidates for generic
quadratic regular algebras of global dimension four. In particular, there exist quadratic regular
GSCAs of global dimension four on four generators with six defining relations that have exactly
twenty nonisomorphic point modules and a 1-dimensional line scheme.
It is an open problem to describe the 1-dimensional line schemes of the regular GSCAs of
global dimension four in [9] that have exactly twenty nonisomorphic point modules.
By Examples 1.7(c) and 2.5(b)(c), the regular algebras of global dimension at most two
are GSCAs, and, by §2.3, almost all quadratic regular algebras of global dimension three
are determined by GSCAs, so GSCAs promise to be very helpful in the classification of all
quadratic regular algebras of global dimension four.
12
ALGEBRA & GEOMETRY IN THE SETTING OF REGULAR ALGEBRAS
2.3. Quadratic Quantum Planes.
In the language of [1], a regular algebra of global dimension three that is generated by degree-
1 elements is sometimes called a quantum plane or quantum projective plane or a quantum P2.
The classification of the generic quantum planes is in [2, 3, 4]. In this subsection, we summarize
the results of [25], in which all quadratic quantum planes are classified by using GSCAs.
We continue to assume that k is algebraically closed, but its characteristic is arbitrary unless
specifically stated otherwise.
Let D denote a quadratic quantum plane and let X ⊂ P2 denote its point scheme. By [3,
Proposition 4.3] and [25, Lemma 2.1], there are, in total, four cases to consider:
• X contains a line, or
• X is a nodal cubic curve in P2, or
• X is a cuspidal cubic curve in P2, or
• X is a (nonsingular) elliptic curve in P2.
Theorem 2.7. [25] Suppose char(k) 6= 2. If X contains a line, then either D is a twist, by
an automorphism, of a GSCA, or D is a twist, by a twisting system, of an Ore extension of
a regular GSCA of global dimension two.
Theorem 2.8. [25]
generators x1, x2, x3 with defining relations:
If X is a nodal cubic curve, then D is isomorphic to a k-algebra on
λx1x2 = x2x1,
λx2x3 = x3x2 − x2
1,
λx3x1 = x1x3 − x2
2,
(∗)
where λ ∈ k and λ3 /∈ {0, 1}. Conversely, for any such λ, any quadratic algebra with defining
relations (∗) is a quantum plane and its point scheme is a nodal cubic curve in P2. Moreover,
if char(k) 6= 2, then D is an Ore extension of a regular GSCA of global dimension two; in
particular, if λ3 = −1, then D is a GSCA.
Theorem 2.9. [25] If char(k) = 3, then X is not a cuspidal cubic curve in P2. If char(k) 6= 3
and if X is a cuspidal cubic curve in P2, then D is isomorphic to a k-algebra on generators
x1, x2, x3 with defining relations:
x1x2 = x2x1 + x2
1,
x3x1 = x1x3 + x2
1 + 3x2
2,
x3x2 = x2x3 − 3x2
2 − 2x1x3 − 2x1x2.
(†)
Moreover, any quadratic algebra with defining relations (†) is a quantum plane; it has point
scheme given by a cuspidal cubic curve in P2 if and only if char(k) 6= 3. If char(k) 6= 2,
then any quadratic algebra with defining relations given by (†) is an Ore extension of a regular
GSCA of global dimension two.
It remains to discuss the case that X is an elliptic curve. In [2, 3], such algebras are classified
into types A, B, E, H, where some members of each type might not have an elliptic curve as
their point scheme, but a generic member does.
ALGEBRA & GEOMETRY IN THE SETTING OF REGULAR ALGEBRAS
13
Theorem 2.10. [25] Suppose that char(k) 6= 2 and that X is an elliptic curve.
(a) Quadratic quantum planes of type H are GSCAs.
(b) Quadratic quantum planes of type B are GSCAs.
(c) As in [2, 3], a quadratic quantum plane D of type A is given by a k-algebra on gener-
ators x, y, z with defining relations:
axy + byx + cz2 = 0,
ayz + bzy + cx2 = 0,
azx + bxz + cy2 = 0,
where a, b, c ∈ k×, (3abc)3 6= (a3 + b3 + c3)3, char(k) 6= 3, and either a3 6= b3, or
a3 6= c3, or b3 6= c3. In the case that a3 = b3 6= c3, D is a GSCA; whereas in the
case a3 6= b3 = c3 (respectively, a3 = c3 6= b3), D is a twist, by an automorphism, of a
GSCA.
In (c) of the last result, the case that a3 6= b3 6= c3 6= a3 is still open. Moreover, the case
when D is of type E is still open, but this case only consists of one algebra, up to isomorphism
and anti-isomorphism. However, both type A and type E have the property that the Koszul
dual of D is a quotient of a regular GSCA; so, in this sense, such algebras are weakly related
to GSCAs.
3. Complete Intersections
In this section, we define the geometric terms used in Theorem 2.6. That discussion leads
naturally into a consideration of a notion of non-commutative complete intersection that
mimics the commutative definition.
We continue to assume that the field k is algebraically closed.
3.1. Commutative Complete Intersection and Quadric Systems.
Let R denote the commutative polynomial ring on n generators of degree one. If f1, . . . , fm
are homogeneous elements of R of positive degree, then {f1, . . . , fm} is a regular sequence
in R if and only if GKdim(R/hf1, . . . , fki) = n − k ≥ 0, for all k = 1, ..., m. Geometrically,
this corresponds to the zero locus in P(R1
) of the ideal Jk = hf1, . . . , fki having dimension
n − 1 − k ≥ −1 for all k. If {f1, . . . , fm} is a regular sequence, then the zero locus of Jm
(respectively, R/Jm) is called a complete intersection (c.f., [14]).
∗
In the setting of §2.1, a quadric system Q is associated to symmetric matrices M1, . . . , Mn.
In that setting, Q corresponds to a regular sequence in R if and only if Q is a complete
intersection, that is, if and only if Q has no base points (a base point is a point that lies on
all the quadrics in Q). A non-commutative analogue of this is needed for Theorem 2.6.
14
ALGEBRA & GEOMETRY IN THE SETTING OF REGULAR ALGEBRAS
3.2. Non-Commutative Complete Intersection and Quadric Systems.
The following result uses the notion of base-point module defined in Definition 1.2.
Proposition 3.1. [9, 10] Let S denote the skew polynomial ring from §1.1, and let f1, . . . , fn
denote homogeneous elements of S of positive degree. If {f1, . . . , fn} is a normalizing sequence
in S, then the following are equivalent:
(a) {f1, . . . , fn} is a regular sequence in S,
(b) dimk(S/hf1, . . . , fni) < ∞,
(c) for each k = 1, . . . , n, we have GKdim(S/hf1, . . . , fki) = n − k,
(d) the factor ring S/hf1, . . . , fni has no right base-point modules,
(e) the factor ring S/hf1, . . . , fni has no left base-point modules.
Such a sequence {f1, . . . , fn} (respectively, S/hf1, . . . , fni) satisfying the equivalent conditions
(a)-(e) from Proposition 3.1 is called a complete intersection in [10].
In the setting of §2.2, one associates S to the GSCA by using µ. The isomorphism
M µ(n, k) → S2 defined by M 7→ (z1, . . . , zn)M(z1, . . . , zn)T associates a quadric system Q to
the µ-symmetric matrices M1, . . . , Mn; that is, Q is the span in S2 of the images of the Mk
under this map. If Q is given by a normalizing sequence in S, then it is called a normalizing
quadric system. By Proposition 3.1, if Q is normalizing, then it corresponds to a regular se-
quence in S if and only if it is a complete intersection, that is, if and only if S/hQi has no right
(respectively, left) base-point modules; this is the meaning of base-point free in Theorem 2.6.
Examples 3.2.
7→ qi = 2z2
(a) [9] We revisit the quantum affine plane from Examples 2.5(b), where n = 2. In that
case, Mi
i ∈ S2, for i = 1, 2. The sequence {q1, q2} is normalizing in S and
dim(S/hq1, q2i) < ∞. Thus, by Proposition 3.1, the corresponding quadric system is base-
point free.
(b) [9] For Examples 2.5(c), n = 2 and M1 7→ q1 = 2(z2
2. Here,
the sequence {q2, q1} is normalizing in S and dim(S/hq2, q1i) < ∞, so by Proposition 3.1, the
corresponding quadric system is base-point free.
1 + z1z2) and M2 7→ q2 = 2z2
Proposition 3.1 has recently been extended in [47] to a family of algebras that contains
the skew polynomial ring S from §1.1. In particular, an analogue of Proposition 3.1 holds
for regular GSCAs, many quantum groups, and homogenizations of finite-dimensional Lie
algebras.
Theorem 3.3. [47] Let A = L∞
i=0 Ai denote a connected, N-graded k-algebra that is generated
by A1. Suppose A is Auslander-Gorenstein of finite injective dimension and satisfies the
Cohen-Macaulay property, and that there exists a normalizing sequence {y1, . . . , yν} ⊂ A \ k
consisting of homogeneous elements such that GKdim(A/hy1, . . . , yνi) = 1. If GKdim(A) =
ALGEBRA & GEOMETRY IN THE SETTING OF REGULAR ALGEBRAS
15
n ∈ N, and if F = {f1, . . . , fn} ⊂ A \ k× is a normalizing sequence of homogeneous elements,
then the following are equivalent:
(a) F is a regular sequence in A,
(b) dimk(A/hF i) < ∞,
(c) for each k = 1, . . . , n, we have GKdim(A/hf1, . . . , fki) = n − k,
(d) the factor ring A/hF i has no right base-point modules,
(e) the factor ring A/hF i has no left base-point modules.
The reader should note that other notions of complete intersection abound in the literature,
with most emphasizing a homological approach, such as the recent work in [17].
4. Conclusion
In this section, we list some open problems and related topics. The open problems are not
listed in any particular order in regards to difficulty, and many challenge levels are included,
with some quite computational in nature, and so accessible to junior researchers.
4.1. Some Open Problems.
1. As stated at the end of §2, it is still open whether or not quadratic quantum planes of
type A with a3 6= b3 6= c3 6= a3 are directly related to GSCAs; the analogous problem is also
open for type E.
2. Is it possible to classify cubic quantum planes by using GSCAs, or by using an appropriate
analogue of a GSCA?
3. Is it possible to classify quadratic regular algebras of global dimension four by using
GSCAs? Presumably, such a classification will use both the point scheme and the line scheme.
4. Can standard results on commutative quadratic forms and quadrics be extended to non-
commutative quadratic forms and quadrics? For example, P. Veerapen and the author have
extended, in [50], the notion of rank of a (commutative) quadratic form to non-commutative
quadratic forms on n generators, where n = 2, 3; can this be done for n ≥ 4?
5. Can results concerning GCAs be carried over to GSCAs? In particular, Veerapen and
the author applied their aforementioned generalization of rank to GSCAs in a way that is
analogous to that used for the traditional notion of rank with GCAs in [49]. They proved in
[51] that various results in [49] concerning point modules over GCAs apply to point modules
over GSCAs.
6. Can standard results concerning symmetric matrices be extended or generalized to µ-
symmetric matrices?
7. Can the results in [47], mentioned above at the end of §3, on complete intersections be
extended to an even larger family of algebras than those considered in [47]?
16
ALGEBRA & GEOMETRY IN THE SETTING OF REGULAR ALGEBRAS
8. By combining results in [9] and [42], it is known that regular GSCAs of global dimension
four can have exactly N nonisomorphic point modules, where N /∈ {2, 19}; it is not yet known
if N ∈ {2, 19} is possible. In fact, by [41], N = 2 is possible if the algebra is quadratic and
regular of global dimension four but is not a GSCA, but it is not known if N = 19 is possible,
even if the algebra is not a GSCA.
9. What is the line scheme of some known quadratic regular algebras of global dimension
four? Such as those in [9, §5], double Ore extensions in [53, 54], generalized Laurent polynomial
rings in [8], etc.
10. Does the line scheme of a generic quadratic regular algebra of global dimension four
have a particular form? Perhaps a union of elliptic curves? Or, perhaps it contains at least
one elliptic curve?
11. Suppose A is as in Definition 1.1 and F is as in Theorem 3.3. Let Ik = hf1, . . . , fki for
all k ≤ n, and let [V(Ik) denote the set of isomorphism classes of right base-point modules over
A/Ik. If A is commutative, then, for each k, [V(Ik) is a scheme, and so has a dimension. In
particular, if A is the polynomial ring, then F is regular if and only if dim( [V(Ik)) = n − k − 1,
for all k ≤ n. However, if A is not commutative, is there an analogous statement and under
what hypotheses on A could it hold?
4.2. Related Topics.
Since the publication of [3], the subject has branched out in many different directions,
with the key topics being: classification of regular algebras; classification of projective sur-
faces; seeing which commutative techniques (e.g., blowing-up, blowing-down) carry over to
the non-commutative setting; and connections with differential geometry (e.g., via Poisson
geometry). Module categories and homological algebra provide a unifying umbrella over these
topics. These different directions are highlighted in the references cited in the Introduction
and throughout the text, and in the presentations from the 2013 MSRI program found in this
journal issue.
New directions continue to emerge, with one of the most recent trends being the study of
regular algebras and Hopf algebras together via the consideration of Hopf actions on regular
algebras, such as the work in [11]. However, perhaps the most recent exciting triumph of the
subject is when the universal enveloping algebra of the Witt algebra was viewed through the
geometric lens of [3] by Sierra and Walton, in [30], enabling them to solve the long-standing
problem of whether or not that algebra is noetherian.
In view of all these advances, it is now clear that the marriage of non-commutative algebra
and algebraic geometry, `a la [3], is a dynamic and evolving field of research.
ALGEBRA & GEOMETRY IN THE SETTING OF REGULAR ALGEBRAS
17
References
[1] M. Artin, Geometry of Quantum Planes, in "Azumaya Algebras, Actions and Modules,"
Eds. D. Haile and J. Osterburg, Contemp. Math. 124 (1992), 1-15.
[2] M. Artin and W. Schelter, Graded Algebras of Global Dimension 3, Adv. Math. 66 (1987),
171-216.
[3] M. Artin, J. Tate and M. Van den Bergh, Some Algebras Associated to Automorphisms of
Elliptic Curves, The Grothendieck Festschrift 1, 33-85, Eds. P. Cartier et al., Birkhauser (1990).
[4] M. Artin, J. Tate and M. Van den Bergh, Modules over Regular Algebras of Dimension 3,
Invent. Math. 106 (1991), 335-388.
[5] M. Aubry and J.-M. Lemaire, Zero Divisors in Enveloping Algebras of Graded Lie Algebras,
J. Pure & Appl. Alg. 38 (1985), 159-166.
[6] T. Cassidy, Global Dimension Four Extensions of Artin-Schelter Regular Algebras, J. Alg. 220
(1999), 225-254.
[7] T. Cassidy, Central Extensions of Stephenson's Algebras, Comm. Alg. 31 (2003), 1615-1632.
[8] T. Cassidy, P. Goetz and B. Shelton, Generalized Laurent Polynomial Rings as Quantum
Projective 3-Spaces, J. Alg. 303 (2006), 358-372.
[9] T. Cassidy and M. Vancliff, Generalizations of Graded Clifford Algebras and of Complete
Intersections, J. Lond. Math. Soc. 81 (2010), 91-112.
[10] T. Cassidy and M. Vancliff, Corrigendum to "Generalizations of Graded Clifford Algebras
and of Complete Intersections", preprint (2013).
[11] K. Chan, C. Walton, Y.-H. Wang, J. J. Zhang, Hopf Actions on Filtered Regular Algebras,
J. Alg. 397 (2014), 68-90.
[12] V. G. Drinfel'd, Quantum Groups, Proc. Int. Cong. Math., Berkeley 1 (1986), 798-820.
[13] L. D. Faddeev, N. Yu. Reshetikhin and L. A. Takhtadzhyan, Quantization of Lie Groups
and Lie Algebras, Leningrad Math. J. 1 No. 1 (1990), 193-225.
[14] D. Eisenbud, Commutative Algebra with a View Toward Algebraic Geometry, Graduate Texts
in Mathematics 150, Springer (1999).
[15] K. R. Goodearl, Semiclassical Limits of Quantized Coordinate Rings, in "Advances in Ring
Theory," Eds. D. V. Huynh et al., 165-204, Trends in Mathematics, Birkhauser Basel (2010).
[16] A. Kapustin, A. Kuznetsov and D. Orlov, Noncommutative Instantons and Twistor
Transform, Comm. Math. Phys. 220 (2001), 385-432.
[17] E. Kirkman, J. Kuzmanovich and J. J. Zhang, Noncommutative Complete Intersections,
preprint (2013).
[18] L. Le Bruyn, Central Singularities of Quantum Spaces, J. Alg. 177 No. 1 (1995), 142-153.
18
ALGEBRA & GEOMETRY IN THE SETTING OF REGULAR ALGEBRAS
[19] L. Le Bruyn and S. P. Smith, Homogenized sl(2), Proc. Amer. Math. Soc. 118 No. 3 (1993),
725-730.
[20] L. Le Bruyn and M. Van den Bergh, On Quantum Spaces of Lie Algebras, Proc. Amer.
Math. Soc. 119 No. 2 (1993), 407-414.
[21] T. Levasseur, Some Properties of Non-commutative Regular Graded Rings, Glasgow Math. J.
34 (1992), 277-300.
[22] T. Levasseur and S. P. Smith, Modules over the 4-Dimensional Sklyanin Algebra, Bull. Soc.
Math. de France 121 (1993), 35-90.
[23] T. Levasseur and J. T. Stafford, The Quantum Coordinate Ring of the Special Linear
Group, J. Pure & Appl. Alg. 86 (1993), 181-186.
[24] D.-M. Lu, J.H. Palmieri, Q.-S. Wu, J.J. Zhang, Regular Algebras of Dimension 4 and
their A∞-Ext-Algebras, Duke Math. J. 137 No. 3 (2007), 537-584.
[25] M. Nafari, M. Vancliff and J. Zhang, Classifying Quadratic Quantum P2s by using Graded
Skew Clifford Algebras, J. Alg. 346 No. 1 (2011), 152-164.
[26] D. Rogalski, An Introduction to Noncommutative Projective Algebraic Geometry, preprint
2014. ( arXiv:1403.3065 )
[27] D. Rogalski and J. J. Zhang, Regular Algebras of Dimension 4 with 3 Generators, in "New
Trends in Noncommutative Algebra," Eds. P. Ara et al, Contemp. Math. 562 (2012), 221-241.
[28] B. Shelton and C. Tingey, On Koszul Algebras and a New Construction of Artin-Schelter
Regular Algebras, J. Alg. 241 No. 2 (2001), 789-798.
[29] B. Shelton and M. Vancliff, Schemes of Line Modules I, J. Lond. Math. Soc. 65 No. 3
(2002), 575-590.
[30] S. J. Sierra and C. Walton, The Universal Enveloping Algebra of the Witt Algebra is not
Noetherian, preprint (2013).
[31] E. K. Sklyanin, Some Algebraic Structures connected to the Yang-Baxter Equation, Func.
Anal. Appl. 16 No. 4 (1982), 27-34.
[32] E. K. Sklyanin, On an Algebra generated by Quadratic Relations, Uspekhi Mat. Nauk 40 No. 2
(1985), 214.
[33] S. P. Smith and J. T. Stafford, Regularity of the Four Dimensional Sklyanin Algebra,
Compositio Math. 83 No. 3 (1992), 259-289.
[34] S. P. Smith and J. M. Staniszkis, Irreducible Representations of the 4-Dimensional Sklyanin
Algebra at Points of Infinite Order, J. Alg. 160 No. 1 (1993), 57-86.
[35] J. T. Stafford, Noncommutative Projective Geometry, Proceedings of the International Con-
gress of Mathematicians, Vol. II (Beijing, 2002), 93-103, Higher Ed. Press, Beijing, 2002.
[36] J. T. Stafford and M. Van den Bergh, Noncommutative Curves and Noncommutative
Surfaces, Bull. Amer. Math. Soc. 38 No. 2 (2001), 171-216.
ALGEBRA & GEOMETRY IN THE SETTING OF REGULAR ALGEBRAS
19
[37] D. R. Stephenson, Artin-Schelter Regular Algebras of Global Dimension Three, J. Alg. 183
No. 1 (1996), 55-73.
[38] D. R. Stephenson, Algebras associated to Elliptic Curves, Trans. Amer. Math. Soc. 349 No. 6
(1997), 2317-2340.
[39] D. R. Stephenson, Quantum Planes of Weight (1, 1, n), J. Alg. 225 (2000), 70-92.
[40] D. R. Stephenson, Corrigendum to "Quantum Planes of Weight (1, 1, n)", J. Alg. 234 (2000),
277-278.
[41] D. R. Stephenson and M. Vancliff, Some Finite Quantum P3s that are Infinite Modules
over their Centers, J. Alg. 297 No. 1 (2006), 208-215.
[42] D. R. Stephenson and M. Vancliff, Constructing Clifford Quantum P3s with Finitely Many
Points, J. Alg. 312 No. 1 (2007), 86-110.
[43] A. Sudbery, Matrix-element Bialgebras determined by Quadratic Coordinate Algebras, J. Alg.
158 (1993), 375-399.
[44] M. Vancliff, The Defining Relations of Quantum n × n Matrices, J. London Math. Soc. 52
No. 2 (1995), 255-262.
[45] M. Vancliff, Primitive and Poisson Spectra of Twists of Polynomial Rings, Algebras and
Representation Theory 2 No. 3 (1999), 269-285.
[46] M. Vancliff, Non-commutative Spaces for Graded Quantum Groups and Graded Clifford Al-
gebras, Clifford Algebras and their Applications in Mathematical Physics 1 (Ixtapa-Zihuatanejo,
1999), 303 -- 320, Progress in Physics, 18, Birkhauser Boston, Boston, MA, 2000.
[47] M. Vancliff, On the Notion of Complete Intersection outside the Setting of Skew Polynomial
Rings, Comm. Alg., to appear.
[48] M. Vancliff and K. Van Rompay, Embedding a Quantum Nonsingular Quadric in a Quan-
tum P3, J. Alg. 195 No. 1 (1997), 93-129.
[49] M. Vancliff, K. Van Rompay and L. Willaert, Some Quantum P3s with Finitely Many
Points, Comm. Alg. 26 No. 4 (1998), 1193-1208.
[50] M. Vancliff and P. P. Veerapen, Generalizing the Notion of Rank to Noncommutative Qua-
dratic Forms, in "Noncommutative Birational Geometry, Representations and Combinatorics,"
Eds. A. Berenstein and V. Retakh, Contemp. Math. 592 (2013), 241-250.
[51] M. Vancliff and P. P. Veerapen, Point Modules over Regular Graded Skew Clifford Alge-
bras, preprint (2013).
[52] K. Van Rompay, Segre Product of Artin-Schelter Regular Algebras of Dimension 2 and Em-
beddings in Quantum P3's, J. Alg. 180 (1996), 483-512.
[53] J. J. Zhang and J. Zhang, Double Ore Extensions, J. Pure & Appl. Alg. 212 No. 12 (2008),
2668-2690.
[54] J. J. Zhang and J. Zhang, Double Extension Regular Algebras of Type (14641), J. Alg. 322
No. 2 (2009), 373-409.
|
0909.0358 | 3 | 0909 | 2010-03-02T16:32:03 | Systems of Dyson-Schwinger equations | [
"math.RA"
] | We consider systems of combinatorial Dyson-Schwinger equations (briefly, SDSE) X_1=B^+_1(F_1(X_1,...,X_N))...X_N=B^+_N(F_N(X_1,...,X_N)) in the Connes-Kreimer Hopf algebra H_I of rooted trees decorated by I={1,...,N},where B^+_i is the operator of grafting on a root decorated by i, and F_1...,F_N are non-constant formal series.The unique solution X=(X_1,...,X_N) of this equation generates a graded subalgebra H_S of H_I. We characterize here all the families of formal series (F_1,...,F_N) such that H_S is a Hopf subalgebra. More precisely, we define three operations on SDSE (change of variables, dilatation and extension) and give two families of SDSE (cyclic and fundamental systems), and prove that any SDSE (S) such that H_S is Hopf is the concatenation of several fundamental or cyclic systems after the application of a change of variables, a dilatation and iterated extensions. We also describe the Hopf algebra H_S as the dual of the enveloping algebra of a Lie algebra g_S of one of the following types: 1. g_S is a Lie algebra of paths associated to a certain oriented graph. 2. g_S is an iterated extension of the Fa\`a di Bruno Lie algebra. 3. g_S is an iterated extension of an abelian Lie algebra. | math.RA | math |
Systems of Dyson-Schwinger equations
Loïc Foissy∗
Laboratoire de Mathématiques - FRE3111, Université de Reims
Moulin de la Housse - BP 1039 - 51687 REIMS Cedex 2, France
1 (F1(X1, . . . , XN )), . . ., XN = B+
ABSTRACT. We consider systems of combinatorial Dyson-Schwinger equations (briefly,
SDSE) X1 = B+
N (FN (X1, . . . , XN )) in the Connes-Kreimer
Hopf algebra HI of rooted trees decorated by I = {1, . . . , N }, where B+
is the operator of graft-
i
ing on a root decorated by i, and F1, . . . , FN are non-constant formal series. The unique solution
X = (X1, . . . , XN ) of this equation generates a graded subalgebra H(S) of HI . We characterise
here all the families of formal series (F1, . . . , FN ) such that H(S) is a Hopf subalgebra. More
precisely, we define three operations on SDSE (change of variables, dilatation and extension)
and give two families of SDSE (cyclic and fundamental systems), and prove that any SDSE (S)
such that H(S) is Hopf is the concatenation of several fundamental or cyclic systems after the
application of a change of variables, a dilatation and iterated extensions.
We also describe the Hopf algebra H(S) as the dual of the enveloping algebra of a Lie algebra
g(S) of one of the following types:
1. g(S) is a Lie algebra of paths associated to a certain oriented graph.
2. Or g(S) is an iterated extension of the Faà di Bruno Lie algebra.
3. Or g(S) is an iterated extension of an abelian Lie algebra.
KEYWORDS: Systems of combinatorial Dyson-Schwinger equations; Hopf algebras of deco-
rated rooted trees; pre-Lie algebras.
MATHEMATICS SUBJECT CLASSIFICATION. Primary 16W30. Secondary 81T15, 81T18.
Contents
1 Preliminaries
1.1 Decorated rooted trees . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
1.2 Hopf algebras of decorated rooted trees . . . . . . . . . . . . . . . . . . . . . . . .
1.3 Gradation of HD and completion . . . . . . . . . . . . . . . . . . . . . . . . . . .
1.4 Pre-Lie structure on the dual of HD . . . . . . . . . . . . . . . . . . . . . . . . .
2 Definitions and properties of SDSE
2.1 Unique solution of an SDSE . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
2.2 Graph associated to an SDSE . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
2.3 Operations on Hopf SDSE . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
2.4 Constant terms of the formal series . . . . . . . . . . . . . . . . . . . . . . . . . .
2.5 Main theorem . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
∗e-mail: [email protected]; webpage: http://loic.foissy.free.fr/pageperso/accueil.html
1
6
6
6
7
8
9
9
10
10
13
13
3 Characterisation and properties of Hopf SDSE
3.1 Subalgebras of HD generated by spans of trees
3.2 Definition of the structure coefficients
3.3 Properties of the coefficients λ(i,j)
3.4 Prelie structure on H∗
. . . . . . . . . . . . . . . . . . .
. . . . . . . . . . . . . . . . . . . . . . . .
. . . . . . . . . . . . . . . . . . . . . . . . . . .
. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
n
(S)
4 Level of a vertex
4.1 Definition of the level . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
4.2 Vertices of level 0 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
4.3 Vertices of level 1 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
4.4 Vertices of level ≥ 2 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
5 Examples of Hopf SDSE
cycles and multicycles
5.1
. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
5.2 Fundamental SDSE . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
6 Two families of Hopf SDSE
6.1 A lemma on non-self-dependent vertices
. . . . . . . . . . . . . . . . . . . . . . .
6.2 Symmetric Hopf SDSE . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
6.3 Formal series of a self-dependent vertex . . . . . . . . . . . . . . . . . . . . . . .
6.4 Hopf SDSE generated by self-dependent vertices . . . . . . . . . . . . . . . . . . .
7 The structure theorem of Hopf SDSE
7.1 Connecting vertices . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
7.2 Structure of connected Hopf SDSE . . . . . . . . . . . . . . . . . . . . . . . . . .
7.3 Connected Hopf SDSE with a multicycle . . . . . . . . . . . . . . . . . . . . . . .
. . . . . . . . . . . . . . . . . . . . . . .
7.4 Connected Hopf SDSE with finite levels
8 Lie algebra and group associated to H(S), associative case
8.1 Characterization of the associative case . . . . . . . . . . . . . . . . . . . . . . . .
8.2 An algebra associated to an oriented graph . . . . . . . . . . . . . . . . . . . . .
8.3 Group of characters . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
9 Lie algebra and group associated to H(S), non-abelian case
9.1 Modules over the Faà di Bruno Lie algebra . . . . . . . . . . . . . . . . . . . . . .
9.2 Description of the Lie algebra . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
9.3 Associated group . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
10 Lie algebra and group associated to H(S), abelian case
10.1 Modules over an abelian Lie algebra . . . . . . . . . . . . . . . . . . . . . . . . .
10.2 Description of the Lie algebra . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
10.3 Associated group . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
11 Appendix: dilatation of a pre-Lie algebra
11.1 Dilatation of a pre-Lie algebra . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
11.2 Dilatation of a Lie algebra . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
17
17
18
19
22
24
24
25
26
28
30
30
31
34
34
35
38
40
41
41
43
45
46
49
49
50
53
54
54
55
58
59
60
61
62
64
64
65
Introduction
The Connes-Kreimer Hopf algebra of rooted trees is introduced in [15] and studied in [2, 3, 6, 7,
8, 9, 14, 19]. This graded, commutative, non-cocommutative Hopf algebra is generated by the
set of rooted trees. We shall work here with a decorated version HD of this algebra, where D
is a finite, non-empty set, replacing rooted trees by rooted trees with vertices decorated by the
2
elements of D. This algebra has a family of operators (B+
d sends a
forest F to the rooted tree obtained by grafting the trees of F on a common root decorated by
d. These operators satisfy the following equation: for all x ∈ HD,
d )d∈D indexed by D, where B+
d (x) = B+
As explained in [7], this means that B+
dual to the Hochschild cohomology.
∆ ◦ B+
d (x) ⊗ 1 + (Id ⊗ B+
d is a 1-cocycle for a certain cohomology of coalgebras,
d ) ◦ ∆(x).
We are interested here in systems of combinatorial Dyson-Schwinger equations (briefly, SDSE),
that is to say, if the set of decorations is {1, . . . , N }, a system (S) of the form:
X1 = B+
1 (F1(X1, . . . , XN )),
...
XN = B+
N (FN (X1, . . . , XN )),
where F1, . . . , FN ∈ K[[h1, . . . , hN ]] are formal series in N indeterminates. These systems (in a
Feynman graph version) are used in Quantum Field Theory, as it is explained in [1, 16, 17]. They
possess a unique solution, which is a family of N formal series in rooted trees, or equivalently
elements of a completion of HD. The homogeneous components of these elements generate a
subalgebra H(S) of HD. Our problem here is to determine Hopf SDSE, that is to say SDSE (S)
such that H(S) is a Hopf subalgebra of HD. In the case of a single combinatorial Dyson-Schwinger
equation, this question has been answered in [10].
In order to answer this, we first associate an oriented graph to any SDSE, reflecting the
dependence of the different Xi's; more precisely, the vertices of G(S) are the elements of I, and
there is an edge from i to j if Fi depends on hj. We shall say that (S) is connected if G(S)
is connected. Noting that any SDSE is the disjoint union of several connected SDSE, we can
restrict our study to connected SDSE. We introduce three operations on Hopf SDSE:
• Change of variables, which replaces hi by λihi for all i ∈ I, where λi 6= 0 for all i. This
operation replaces H(S) by an isomorphic Hopf algebra and does not change G(S).
• Dilatation, which replaces each vertex of G(S) by several vertices. This operation increases
the number of vertices. For example, consider:
(S) :(cid:26) X1 = B+
X2 = B+
1 (f (X1, X2)),
2 (g(X1, X2)),
where f, g ∈ K[[h1, h2]]; then the following SDSE is a dilatation of (S):
(S′) :
X1 = B+
X2 = B+
X3 = B+
X4 = B+
X5 = B+
1 (f (X1 + X2 + X3, X4 + X5)),
2 (f (X1 + X2 + X3, X4 + X5)),
3 (f (X1 + X2 + X3, X4 + X5)),
4 (g(X1 + X2 + X3, X4 + X5)),
5 (g(X1 + X2 + X3, X4 + X5)),
• Extension, which adds a vertex 0 to G(S) with an affine formal series. This operation
increases the number of vertices by 1. For example, consider:
where f ∈ K[[h1, h2]] and a, b ∈ K; then the following SDSE is an extension of (S):
X2 = B+
(S) :(cid:26) X1 = B+
(S′) :
X0 = B+
X1 = B+
X2 = B+
3
1 (f (X1, X2)),
2 (f (X1, X2)),
0 (1 + aX1 + bX2),
1 (f (X1, X2)),
2 (f (X1, X2)),
We then introduce two families of Hopf SDSE:
• Cycles, which are SDSE such that the associated graph is an oriented graph and all the
formal series of the system are affine; see theorem 30. For example, the following system
is a 4-cycle:
The associated oriented graph is:
X1 = B+
X2 = B+
X3 = B+
X4 = B+
1 (1 + X2),
2 (1 + X3),
3 (1 + X4),
4 (1 + X1).
1
4
/ 2
3
• Fundamental SDSE, described in theorem 32. Here is an example of a fundamental SDSE:
X1 = B+
X2 = B+
X3 = B+
X4 = B+
X5 = B+
1+β1
1 (cid:18)fβ1(X1)f β2
2 (cid:18)f β1
3 (cid:18)f β1
4 (cid:18)f β1
5 (cid:18)f β1
1+β1
1+β1
1+β1
((1 + β2)h2)(1 − h3)−1(1 − h4)−1(cid:19) ,
1+β2
(X1)fβ2(h2)(1 − h3)−1(1 − h4)−1(cid:19) ,
((1 + β1)X1)f β2
1+β2
((1 + β1)X1)f β2
1+β2
((1 + β1)X1)f β2
1+β2
((1 + β2)h2)(1 − h4)−1(cid:19) ,
((1 + β2)h2)(1 − h3)−1(cid:19) ,
((1 + β2)h2)(1 − h3)−1(1 − h4)−1(cid:19) ,
where β1, β2 ∈ K − {−1} and, for all β ∈ K, fβ is the following formal series:
fβ(h) =
The associated oriented graph is:
(1 + β) · · · (1 + (k − 1)β)
k!
hk.
∞Xk=0
1
3 o
2
/ 4
'NNNNNNNNNNNNNN
wpppppppppppppp
@
^>>>>>>>
5
The main result of this paper is theorem 14, which says that any connected Hopf SDSE is ob-
tained by a dilatation and a finite number of iterated extensions of a cycle or a fundamental SDSE.
Let us now give a few explanations on the way this result is obtained. An important tool is
associated to any Hopf SDSE.
given by a family indexed by I 2 of scalar sequences (cid:16)λ(i,j)
19. Particular cases of possible sequence(cid:16)λ(i,j)
n (cid:17)n≥1
n (cid:17)n≥1
They allow to reconstruct the coefficients of the formal series of (S), as explained in proposition
are affine sequences, up to a finite number
of terms: this leads to the notion of level of a vertex.
It is shown that level decreases along
the oriented paths of G(S) (proposition 23), and this implies the following alternative if (S) is
connected: any vertex is of finite level or no vertex is of finite level. In particular, any vertex of
a fundamental SDSE is of finite level, whereas no vertex of a cycle is of finite level.
We then consider two special families of SDSE:
4
/
O
O
o
o
o
o
/
/
O
O
g
g
'
O
O
7
7
w
o
/
E
E
Y
Y
^
@
• We first assume that the graph associated to (S) does not contain any vertex related to
itself. This case includes cycles and their dilatations (called multicycles), and a special
case of fundamental SDSE called quasi-complete SDSE. We show, using graph-theoretical
considerations and the coefficients λ(i,j)
, that under an hypothesis of symmetry, they are
the only possibilities.
n
• We then assume that any vertex of (S) has an ascendant related to itself. We then prove
that (S) is fundamental.
This results are then unified in corollary 50. It says that any Hopf SDSE with a connected graph
contains a multicycle or a a fundamental SDSE (S0) and is obtained from (S0) by adding repeat-
edly a finite number of vertices. This result is precised for the multicycle case in theorem 51 and
for the fundamental case in theorem 52. The compilation of these results then proves theorem 14.
The end of the paper is devoted to the description of the Hopf algebras H(S). By the Cartier-
Quillen-Milnor-Moore theorem, they are dual of enveloping algebra U (g(S)), and it turns out
that g(S) is a pre-Lie algebra [5], that is to say it has a bilinear product ⋆ such that for all
f, g, h ∈ g(S):
(f ⋆ g) ⋆ h − f ⋆ (g ⋆ h) = (g ⋆ f ) ⋆ h − g ⋆ (f ⋆ h).
This relation implies that the antisymmetrisation of ⋆ is a Lie bracket. In our case, g(S) has a
basis (fi(k))i∈I,k≥1 and by proposition 21 its pre-Lie product is given by:
fj(l) ⋆ fi(k) = λ(i,j)
k
fi(k + l).
The product ⋆ can be associative, for example in the multicyclic case. Then, up to a change
of variables, fj(l) ⋆ fi(k) = fi(k + l) if there is an oriented path of length k from i to j in the
oriented graph associated to (S), or 0 otherwise; see proposition 57.
The fundamental case is separated into two subcases. In the non-abelian case, the Lie algebra
g(S) is described as an iterated semi-direct product of the Faà di Bruno Lie algebra by infinite
dimensional modules. Similarly, the character group of H(S) is an iterated semi-direct product
of the Faà di Bruno group of formal diffeomorphisms by modules of formal series:
Ch(H(S)) = Gm ⋊ (Gm−1 ⋊ (· · · G2 ⋊ (G1 ⋊ G0) · · · ),
where G0 is the Faà di Bruno group and G1, . . . , Gm−1 are isomorphic to direct sums of (tK[[t]], +)
as groups; see theorem 65. The second subcase is similar, replacing the Faà di Bruno Lie algebra
by an abelian Lie algebra; see theorem 72.
This text is organised as follows: the first section gives some recalls on the structure of Hopf
algebra of HD and on the pre-Lie product on g(S) = P rim(cid:16)H∗
(S)(cid:17). In the second section are
n
given the definitions of SDSE and their different operations: change of variables, dilatation and
extension. The main theorem of the text is also stated in this section. The following section
introduces the coefficients λ(i,j)
and their properties, especially their link with the pre-Lie prod-
uct of g(S). The level of a vertex is defined in the fourth section, which also contains lemmas
on vertices of level 0, 1 or ≥ 2, before that fundamental and multicyclic SDSE are introduced
in the fifth section. The next section contains preliminary results about graphs with no self-
dependent vertices or such that any vertex is the descendant of a self-dependent vertex, and
the main theorem is finally proved in the seventh section. The next three sections deals with
the description of the Lie algebra g(S) and the group Ch(cid:0)H(S)(cid:1) when g(S) is associative, in the
non-abelian, fundamental case and finally in the abelian, fundamental case. The last section
gives a functorial way to characterise pre-Lie algebra from the operation of dilatations of Hopf
SDSE.
Notations. We denote by K a commutative field of characteristic zero. All vector spaces,
algebras, coalgebras, Hopf algebras, etc. will be taken over K.
5
1 Preliminaries
1.1 Decorated rooted trees
Definition 1 [20, 21]
1. A rooted tree t is a finite graph, without loops, with a special vertex called the root of t.
The weight of t is the number of its vertices. The set of rooted trees will be denoted by T .
2. Let D be a non-empty set. A rooted tree decorated by D is a rooted tree with an application
from the set of its vertices into D. The set of rooted trees decorated by D will be denoted
by TD.
3. Let i ∈ D. The set of rooted trees decorated by D with root decorated by i will be denoted
by T (i)
D .
Examples.
1. Rooted trees with weight smaller than 5:
q ;
q; q
q
q∨q
q
q
, q
;
q∨qq q ,
q∨qq
,
q
q
q
q
q∨q
q , q
q
q
q
q
q∨qq
;
q,
✟❍q
q q
q∨q
,
q
q∨q
,
q
q∨q
q∨q
,
q
q∨q
,
q
q
q
q∨qq
q
,
q
q
q∨q
q
q
q∨q
, q
,
q
q
q
q
q
q
.
2. Rooted trees decorated by D with weight smaller than 4:
q a; a ∈ D,
q
q
b (a, b) ∈ D2;
a
b
c
q∨qq
a
c
q
= q∨q
a
b
,
q
q
q
c
b
a
, (a, b, c) ∈ D3;
b
c
d
q q
q∨q
a
d
b
= q∨qq q
a
c
b
c
q q
= q∨q
a
d
d
q q
c
= q∨q
b
a
b
d
q q
= q∨q
a
c
c
d
= q∨qq q
b
a
c
b
,
q
q
d
q
q∨q
a
d
= q∨qq
a
c
b
c
,
d
q
q∨q
b
q a
c
d
q
= q∨q
b
q a
q
q
q
q
d
c
b
a
,
, (a, b, c, d) ∈ D4.
Definition 2
1. We denote by HD the polynomial algebra generated by TD.
2. Let t1, . . . , tn be elements of TD and let d ∈ D. We denote by B+
tree obtained by grafting t1, . . . , tn on a common root decorated by d. This map B+
extended in an operator from HD to HD.
d (t1 . . . tn) the rooted
d is
For example, B+
q
d ( q
b
a
q
b
a
q c ) = q∨q
d
q
c
.
1.2 Hopf algebras of decorated rooted trees
In order to make HD a bialgebra, we now introduce the notion of cut of a tree t ∈ TD. A
non-total cut c of a tree t is a choice of edges of t. Deleting the chosen edges, the cut makes t
into a forest denoted by W c(t). The cut c is admissible if any oriented path in the tree meets
at most one cut edge. For such a cut, the tree of W c(t) which contains the root of t is denoted
by Rc(t) and the product of the other trees of W c(t) is denoted by P c(t). We also add the total
cut, which is by convention an admissible cut such that Rc(t) = 1 and P c(t) = W c(t) = t. The
set of admissible cuts of t is denoted by Adm∗(t). Note that the empty cut of t is admissible; we
put Adm(t) = Adm∗(t) − {empty cut, total cut}.
6
example. Let a, b, c, d ∈ D and let us consider the rooted tree t = q∨q
d
q
q
a
b
c
. As it as 3 edges, it
has 23 non-total cuts.
cut c
Admissible?
W c(t)
Rc(t)
P c(t)
q
a
b
q
c
q∨q
d
yes
a
b
c
q
q
q∨q
d
c
q
a
b
q
q∨q
d
yes
c
q
a
b
q
q∨q
d
yes
c
q
a
b
q
q∨q
d
yes
q
q
a
b
q
q
c
d
b
q a
q
q∨q
d
c
q
q
q
a
b
d
q c
q a q b q
q
a
b
q
q∨q
d
no
c
q
c
q
a
b
q
q∨q
d
yes
c
q
a
b
q
q∨q
d
yes
c
q
a
b
q
q∨q
d
no
c
d
q
q
a
b
q c q d
q
q a q
b
d
q c
q a q b q c q d
total
yes
a
b
c
q
q∨qq
d
a
b
q
q
q∨q
d
1
b
c
q
q
c
d
q
q
a
b
c
q
q∨q
d
q a
q
q
q
a
b
d
q c
×
×
q d
q
q
a
b
q c
q
q
b
d
q a q c
×
×
1
a
b
q
c
q∨qq
d
The coproduct of HD is defined as the unique algebra morphism from HD to HD ⊗ HD such
that for all rooted tree t ∈ TD:
∆(t) = Xc∈Adm∗(t)
P c(t) ⊗ Rc(t) = t ⊗ 1 + 1 ⊗ t + Xc∈Adm(t)
P c(t) ⊗ Rc(t).
As HD is the free associative commutative unitary algebra generated by TD, this makes sense.
This coproduct makes HD a Hopf algebra. Although it won't play any role in this text, we recall
that the antipode S is the unique algebra automorphism of HD such that for all t ∈ TD:
S(t) = − Xc cut of t
(−1)nc Wc(t),
where nc is the number of cut edges of c.
Example.
q
a
b
q
∆( q∨q
d
c
q
a
b
q
) = q∨q
d
c
⊗ 1 + 1 ⊗ q∨q
d
q
q
a
b
c
q
+ q
a ⊗ q
b
q
c + q a ⊗ q∨q
d
d
q
b
c
q
+ q c ⊗ q
q
a
b
d
q
+ q
a
b
q c ⊗ q d + q a q c ⊗ q
q
b .
d
A study of admissible cuts shows the following result:
Proposition 3 For all d ∈ D, for all x ∈ HD:
∆ ◦ B+
d (x) = B+
d (x) ⊗ 1 + (Id ⊗ B+
d ) ◦ ∆(x).
Remarks.
1. In other words, B+
d is a 1-cocycle for a certain cohomology of coalgebras, see [7].
2. If t ∈ T (i)
D , then ∆(t) − t ⊗ 1 ∈ HD ⊗ T (i)
D .
1.3 Gradation of HD and completion
We grade HD by declaring the forests with n vertices homogeneous of degree n. We denote by
HD(n) the homogeneous component of HD of degree n. Then HD is a graded bialgebra, that is
to say:
• For all i, j ∈ N, HD(i)H(j) ⊆ HD(i + j).
• For all k ∈ N, ∆(HD(k)) ⊆ Xi+j=k
HD(i) ⊗ HD(j).
7
We define, for all x ∈ HD:
n ∈ N x ∈Mk≥n
.
HD(k)
val(x) = max
dHD = Yn∈N
We then put, for all x, y ∈ HD, d(x, y) = 2−val(x−y), with the convention 2−∞ = 0. Then d is
a distance on HD. The metric space (HD, d) is not complete; its completion will be denoted by
dHD. As a vector space:
The elements of dHD will be denoted by P xn, where xn ∈ HD(n) for all n ∈ N. The product
dHD ⊗dHD to dHD, which is then an associative, commutative algebra. Similarly, the coproduct
m : HD ⊗ HD −→ HD is homogeneous of degree 0, so is continuous:
of HD can be extended as a map:
it can be extended from
HD(n).
∆ :dHD −→ HDb⊗HD = Yi,j∈N
HD(i) ⊗ HD(j).
Let f (h) =P pnhn ∈ K[[h]] be any formal series, and let X =P xn ∈dHD, such that x0 = 0.
The series of dHD of terms pnX n is Cauchy, so converges. Its limit will be denoted by f (X). In
other words, f (X) =P yn, with:
nXk=1 Xa1+···+ak=n
pkxa1 · · · xak if n ≥ 1.
y0 = p0,
1.4 Pre-Lie structure on the dual of HD
yn =
By the Cartier-Quillen-Milnor-Moore theorem [18], the graded dual H∗
algebra. Its Lie algebra P rim(H∗
D) has a basis (ft)t∈TD indexed by TD:
D of HD is an enveloping
ft :
HD −→ K
t1 . . . tn −→ (cid:26) 0 if n 6= 1,
δt,t1 if n = 1.
Recall that a pre-Lie algebra (or equivalently a Vinberg algebra or a left-symmetric algebra)
is a couple (A, ⋆), where ⋆ is a bilinear product on A such that for all x, y, z ∈ A:
(x ⋆ y) ⋆ z − x ⋆ (y ⋆ z) = (y ⋆ x) ⋆ z − y ⋆ (x ⋆ z).
Pre-Lie algebras are Lie algebras, with bracket given by [x, y] = x ⋆ y − y ⋆ x.
The Lie bracket of P rim(H∗
D) is induced by a pre-Lie product ⋆ given in the following way:
if f, g ∈ P rim(H∗
D), f ⋆ g is the unique element of P rim(H∗
D) such that for all t ∈ TD,
(f ⋆ g)(t) = (f ⊗ g) ◦ (π ⊗ π) ◦ ∆(t),
where π is the projection on V ect(T D) which vanishes on the forests which are not trees. In
other words, if t, t′ ∈ TD:
where n(t, t′; t′) is the number of admissible cuts c of t′′ such that P c(t′′) = t and Rc(t′′) = t′. It
is proved that (prim(H∗
D), ⋆) is the free pre-Lie algebra generated by the q d's, d ∈ D: see [3, 5].
Note that H∗
D is isomorphic to the Grossman-Larson Hopf algebra of rooted trees [11, 12, 13].
ft ⋆ ft′ = Xt′′∈TD
n(t, t′; t′′)ft′′,
8
2 Definitions and properties of SDSE
2.1 Unique solution of an SDSE
Definition 4 Let I be a finite, non-empty set, and let Fi ∈ K[[hj, j ∈ I]] be a non-constant
formal series for all i ∈ I. The system of Dyson-Schwinger combinatorial equations (briefly, the
SDSE) associated to (Fi)i∈I is:
∀i ∈ I, Xi = B+
i (fi(Xj, j ∈ I)),
where Xi ∈ cHI for all i ∈ I.
In order to ease the notation, we shall often assume that I = {1, . . . , N } in the proofs, with-
out loss of generality.
Notations. We assume here that I = {1, . . . , N }.
1. Let (S) be an SDSE. We shall denote, for all i ∈ I:
Fi = Xp1,··· ,pN
a(i)
(p1,··· ,pN )hp1
1 · · · hpN
N .
2. Let 1 ≤ j ≤ N . We put εj = (0, · · · , 0, 1, 0, · · · , 0) where the 1 is in position j. We shall
denote, for all i ∈ I, a(i)
j = a(i)
εj ; for all j, k ∈ I, a(i)
j,k = a(i)
εj+εk
, and so on.
Remark. We assume that there is no constant Fi. Indeed, if Fi ∈ K, then Xi is a multiple
of q i . We shall always avoid this degenerated case in all this text.
Proposition 5 Let (S) be an SDSE. Then it admits a unique solution (Xi)i∈I ∈(cid:16)cHI(cid:17)I
Proof. We assume here that I = {1, . . . , N }. If (X1, · · · , XN ) is a solution of S, then Xi is
.
a linear (infinite) span of rooted trees with a root decorated by i. We denote:
Xi = Xt∈T (i)
I
att.
These coefficients are uniquely determined by the following formulas: if
t = B+
i (cid:16)tp1,1
1,1 · · · t
p1,q1
1,q1 · · · t
pN,1
N,1 · · · t
pN,qN
N,qN (cid:17) ,
where the ti,j's are different trees, such that the root of ti,j is decorated by i for all i ∈ I,
1 ≤ j ≤ qi, then:
at = NYi=1
(pi,1 + · · · + pi,qi)!
pi,1! · · · pi,qi! ! a(i)
(p1,1+···+p1,q1 ,··· ,pN,1+···+pN,qN )ap1,1
t1,1 · · · a
So (S) has a unique solution.
pN,qN
tN,qN
.
(1)
✷
Definition 6 Let (S) be an SDSE and let X = (Xi)i∈I be its unique solution. The subal-
gebra of HI generated by the homogeneous components Xi(k)'s of the Xi's will be denoted by
H(S). If H(S) is Hopf, the system (S) will be said to be Hopf.
9
2.2 Graph associated to an SDSE
We associate a oriented graph to each SDSE in the following way:
Definition 7 Let (S) be an SDSE.
1. We construct an oriented graph G(S) associated to (S) in the following way:
• The vertices of G(S) are the elements of I.
• There is an edge from i to j if, and only if,
∂Fi
∂hj
6= 0.
2. If
6= 0, the vertex i will be said to be self-dependent.
In other words, if i is self-
∂Fi
∂hi
dependent, there is a loop from i to itself in G(S).
3. If G(S) is connected, we shall say that (S) is connected.
Remark. If (S) is not connected, then (S) is the union of SDSE (S1), · · · , (Sk) with disjoint
sets of indeterminates , so H(S) ≈ H(S1) ⊗ · · · ⊗ H(Sk). As a corollary, (S) is Hopf if, and only if,
for all j, (Sj) is Hopf.
Let (S) be an SDSE and let G(S) be the associated graph. Let i and j be two vertices of
G(S). We shall say that j is a direct descendant of i (or i is a direct ascendant of j) if there is
an oriented edge from i to j; we shall say that j is a descendant of i (or i is an ascendant of j) if
there is an oriented path from i to j. We shall write "i −→ j" for "j is a direct descendant of i".
2.3 Operations on Hopf SDSE
Proposition 8 (change of variables) Let (S) be the SDSE associated to (Fi(hj , j ∈ I))i∈I .
Let λi and µi be non-zero scalars for all i ∈ I. The system (S) is Hopf if, and only if, the SDSE
system (S′) associated to (µiFi(λjhj, j ∈ J))i∈I is Hopf.
Proof. We assume that I = {1, . . . , N }. We consider the following morphism:
φ :(cid:26)
HI −→ HI
F ∈ F −→ (µ1λ1)n1(F ) · · · (µN λN )nN (F )F,
where ni(F ) is the number of vertices of F decorated by i. Then φ is a Hopf algebra automorphism
and for all i, φ ◦ B+
i ◦ φ. Moreover, if we put Yi = 1
λi
i = µiλiB+
φ(Xi) for all i:
Yi =
=
1
λi
1
λi
φ ◦ B+
i (Fi(X1, · · · , XN ))
µiλiB+
i (Fi(φ(X1), · · · , φ(XN )))
= µiB+
i (Fi(λ1Y1, · · · , λN YN )).
So (Y1, · · · , YN ) is the solution of the system (S′). Moreover, φ sends H(S) onto H(S ′). As φ is
a Hopf algebra automorphism, H(S) is a Hopf subalgebra of HI if, and only if, H(S ′) is.
✷
Remark. A change of variables does not change the graph associated to (S).
Proposition 9 (restriction) Let (S) be the SDSE associated to (Fi(hj, j ∈ I))i∈I and let
. If (S) is
I ′ ⊆ I, non-empty. Let (S′) be the SDSE associated to (cid:16)Fi(hj, j ∈ I)hj=0, ∀j /∈I ′(cid:17)i∈I ′
Hopf, then (S′) also is.
10
Proof. We consider the epimorphism φ of Hopf algebras from HI to HI ′, obtained by sending
the forests with at least a vertex decorated by an element which is not in I ′ to zero. Then φ
sends H(S) to H(S ′). As φ is a morphism of Hopf algebras, if H(S) is a Hopf subalgebra of HI ,
H(S ′) is a Hopf subalgebra of HI ′.
✷
Remark. The restriction to a subset of vertices I ′ changes G(S) into the graph obtained by
deleting all the vertices j /∈ I ′ and all the edges related to these vertices.
Proposition 10 (dilatation) Let (S) be the system associated to (Fi)i∈I and (S′) be a
Ji, with the
system associated to a family (F ′
following property: for all i ∈ I, for all x ∈ Ii,
j)j∈J , such that there exists a partition J = [i∈I
x = FiXy∈Ij
hy, j ∈ I .
F ′
Then (S) is Hopf, if, and only if, (S′) is Hopf. We shall say that (S′) is a dilatation of (S).
Proof. We assume here that I = {1, . . . , N }.
=⇒. Let us assume that (S) is Hopf. For all i ∈ I, we can then write:
∆(Xi) =Xn≥0
P (i)
n (X1, · · · , XN ) ⊗ Xi(n),
with the convention Xi(0) = 1. Let φ : HI −→ HI ′ be the morphism of Hopf algebras such that,
for all 1 ≤ i ≤ N :
Then, immediately, for all 1 ≤ i ≤ N :
As a consequence:
Xj∈Ii
∆(X ′
j) =Xj∈IiXn≥0
∆(X ′
j) =Xn≥0
So (S′) is Hopf.
P (i)
B+
j ◦ φ.
X ′
j.
φ ◦ B+
i =Xj∈Ii
φ(Xi) =Xj∈Ii
n Xk∈I1
n Xk∈I1
P (i)
X ′
X ′
k, · · · , Xk∈IN
k, · · · , Xk∈IN
X ′
j(n).
X ′
k ⊗ X ′
k ⊗ X ′
j(n).
Conserving the terms of the form F ⊗ t, where t is a tree with root decorated by j, for all j ∈ Ii:
⇐=. By restriction, choosing an element in each Ii, if (S′) is Hopf, then (S) is Hopf.
✷
Remark. If (S′) is a dilatation of (S), then the set of vertices J of the graph G(S ′) associated
to (S′) admits a partition indexed by the vertices of G(S), and there is an edge from x ∈ Ji to
y ∈ Jj in G(S ′) if, and only if, there is an edge from i to j in G(S).
11
Example. Let f, g ∈ K[[h1, h2]]. Let us consider the following SDSE:
(S) : (cid:26) X1 = B+
X2 = B+
1 (f (X1, X2)),
2 (g(X1, X2)),
X1 = B+
X2 = B+
X3 = B+
X4 = B+
X5 = B+
1 (f (X1 + X2 + X3, X4 + X5)),
2 (f (X1 + X2 + X3, X4 + X5)),
3 (f (X1 + X2 + X3, X4 + X5)),
4 (g(X1 + X2 + X3, X4 + X5)),
5 (g(X1 + X2 + X3, X4 + X5)).
(S′) :
Then (S′) is a dilatation of (S).
Proposition 11 (extension) Let (S) be the SDSE associated to (Fi)i∈I . Let 0 /∈ I and let
(S′) be associated to (Fi)i∈I∪{0}, with:
F0 = 1 +Xi∈I
a(0)
i hi.
Then (S′) is Hopf if, and only if, the two following conditions hold:
1. (S) is Hopf.
2. For all i, j ∈ I (0) =nj ∈ I / a(0)
j
6= 0o, Fi = Fj.
If these two conditions hold, we shall say that (S′) is an extension of (S).
Proof. We assume here that I = {1, . . . , N }.
=⇒. Let us assume that (S′) is Hopf. By restriction, (S) is Hopf. Moreover:
X0 = B+
0 1 +
i Xi! = q 0 +
a(0)
NXi=1
NXi=1
a(0)
i B+
0 ◦ B+
i (fi(X1, · · · , XN )).
As H(S ′) is a graded Hopf subalgebra, the projection on H{0,··· ,N } ⊗ H{0,··· ,N }(2) gives:
So this is of the form:
a(0)
i Fi(X1, · · · , XN ) ⊗ q
q
NXi=1
0
i ∈ H(S ′)b⊗H(S ′).
P ⊗ X0(2) = P ⊗ NXi=1
a(0)
i
q
q
i! ,
0
for a certain P ∈ [H(S ′). As the q
a(0)
i Fi(X1, · · · , XN ) = a(0)
q
i P for all i, and this implies the second item.
i 's, i ∈ I, are linearly independent, we obtain that for all i, j,
0
⇐=. As (S) is Hopf, we can put for all 1 ≤ i ≤ N :
∆(Xi) = Xi ⊗ 1 +
P (i)
k ⊗ Xi(k),
+∞Xk=1
12
where P (i)
n
Fi = Fj, P (i)
is an element of the completion of H(S). By the second hypothesis, if i, j ∈ I, as
n = P (j)
n . We then denote by Pn the common value of P (i)
n for all i ∈ I. So:
∆(X0) = q 0 ⊗ 1 + 1 ⊗ q 0 +
a(0)
i ∆ ◦ B+
0 (Xi)
= X0 ⊗ 1 + 1 ⊗ X0 +
= X0 ⊗ 1 + 1 ⊗ X0 +
= X0 ⊗ 1 + 1 ⊗ X0 +
NXi=1
NXi=1
NXi=1
NXi=1
NXi=1
∞Xj=1
∞Xj=1
a(0)
i (1 + Xi) ⊗ q 0 +
a(0)
i (1 + Xi) ⊗ q 0 +
a(0)
i (1 + Xi) ⊗ q 0 +
a(0)
i (1 + Xi) ⊗ q 0 +
NXi=1
NXi=1
NXi=1
NXi=1
a(0)
i P (i)
j ⊗ B+
0 (Xi(j))
a(0)
i Pj ⊗ B+
0 (Xi(j))
Pj ⊗ B+
0
∞Xj=1
a(0)
i Xi(j)
= X0 ⊗ 1 + 1 ⊗ X0 +
Pj ⊗ X0(j + 1).
This belongs to the completion of H(S ′) ⊗ H(S ′), so (S′) is Hopf.
✷
Remarks.
1. If (S) is an extension of (S′), then G(S) is obtained from G(S ′) by adding a non-self-
dependent vertex with no ascendant.
2. If I (0) is reduced to a single element, then condition 2 is empty.
Definition 12 Let (S) a Hopf SDSE and let i ∈ I. We shall say that i is an extension vertex
if, denoting by J the set of descendants of i, the restriction of (S) to J ∪ {i} is an extension of
the restriction of (S) to J.
2.4 Constant terms of the formal series
Lemma 13 Let (S) be an Hopf SDSE. If Fi(0, · · · , 0) = 0, then Xi = 0.
Proof. If Fi(0, · · · , 0) = 0, then the homogeneous component of degree 1 of Xi is zero, so
q i /∈ H(S). Considering the terms of the form F ⊗ q i in ∆(Xi), we obtain:
Fi(Xj, j ∈ I) ⊗ q i ∈ H(S) ⊗ H(S).
As q i /∈ H(S), necessarily Fi(Xj, j ∈ I) = 0, so Xi = 0.
✷
As a consequence, if Fi(0, · · · , 0) = 0, then H(S) = H(S ′), where (S′) is the restriction of
(S) to I − {i}. Using a change of variables, we shall always suppose in the sequel that for all i,
Fi(0, · · · , 0) = 1.
2.5 Main theorem
Notations. For all β ∈ K, we put:
fβ(h) =
+∞Xk=0
(1 + β) · · · (1 + β(k − 1))
k!
hk =( (1 − βh)− 1
eh if β = 0.
β if β 6= 0,
The main aim of this text is to prove the following result:
13
Theorem 14 Let (S) be a connected SDSE. It is Hopf if and only if one of the following
assertion holds:
1. (Extended multicyclic SDSE). The set I admits a partition I = I1 ∪ · · · ∪ IN indexed by the
elements of Z/N Z, N ≥ 2, with the following conditions:
• For all i ∈ Ik:
Fi = 1 + Xj∈Ik+1
a(i)
j hj.
• If i and i′ have a common direct ascendant in G(S), then Fi = Fi′ (so i and i′ have
the same direct descendants).
2. (Extended fundamental SDSE). There exists a partition:
I =[i∈I0
with the following conditions:
Ji ∪[i∈J0
Ji ∪ K0 ∪ I1 ∪ J1 ∪ I2,
• K0, I1, J1, I2 can be empty.
• The set of indices I0 ∪ J0 is not empty.
• For all i ∈ I0 ∪ J0, Ji is not empty.
Up to a change of variables:
(a) For all i ∈ I0, there exists βi ∈ K, such that for all x ∈ Ji:
(c) For all i ∈ K0:
hy .
Fi = Yj∈I0
f βj
(b) For all i ∈ J0, for all x ∈ Ji:
Fx = fβiXy∈Ji
Fx = Yj∈I0
1+βj (1 + βj)Xy∈Jj
hy Yj∈J0−{i}
hyYj∈J0
hy Yj∈I0−{i}
1+βj (1 + βj)Xy∈Jj
1+βj (1 + βj)Xy∈Jj
hyYj∈J0
f1Xy∈Jj
f1Xy∈Jj
hy .
f1Xy∈Jj
hy .
j (cid:17)j∈I0∪J0∪K0
(d) For all i ∈ I1, there exist νi ∈ K and a family of scalars (cid:16)a(i)
hy Yj∈K0
f0(cid:16)νia(i)
hy + Xj∈K0
f 1
νia
νia(i)
j Xy∈Jj
ln1 −Xy∈Jj
hyYj∈J0
hy −Xj∈J0
νia(i)
j Xy∈Jj
ln1 −Xy∈Jj
(νi 6= 1) or (∃j ∈ I0, a(i)
j
Then, if νi 6= 0:
6= 1 + βj) or (∃j ∈ J0, a(i)
j
6= 1) or (∃j ∈ K0, a(i)
j
Fi = −Xj∈I0
a(i)
j
βj
1
νi Yj∈I0
f βj
νia
(i)
j
Fi =
If νi = 0:
(i)
j
a(i)
j
, with
6= 0).
j hj(cid:17)+1−
1
νi
.
a(i)
j hj + 1.
f βj
f βj
14
(e) For all i ∈ J1, there exists νi ∈ K − {0} and a family of scalars (cid:16)a(i)
with the three following conditions:
j (cid:17)j∈I0∪J0∪K0∪I1
,
6= 0} is not empty.
1 = {j ∈ I1 / a(i)
• I (i)
• For all j ∈ I (i)
• For all j, k ∈ I (i)
j
1 , νj = 1.
all t ∈ I0 ∪ J0 ∪ K0.
1 , Fj = Fk. In particular, we put b(i)
t = a(j)
t
for any j ∈ I (i)
1 , for
Then:
Fi =
f
1
νi Yj∈I0
Yj∈K0
b
(i)
j
f0(cid:16)b(i)
βj
−1−βj
(cid:16)b(i)
j − 1 − βj(cid:17)Xy∈Jj
j hj(cid:17) + Xj∈I (i)
a(i)
j h1 + 1 −
1
νi
1
hyYj∈J0
.
f βj
(i)
j
b
−1
(cid:16)b(i)
j − 1(cid:17)Xy∈Jj
hy
(f ) I2 = {x1, . . . , xm} and for all 1 ≤ k ≤ m, there exist a set:
I (xk) ⊆[i∈I0
Ji ∪[i∈J0
Ji ∪ K0 ∪ I1 ∪ J1 ∪ {x1, . . . , xk−1}
j (cid:17)j∈I (xk)
and a family of non-zero scalars (cid:16)a(xk)
Fxk = 1 + Xj∈I (xk)
a(xk)
j
Then:
hj.
such that for all i, j ∈ I (xk), Fi = Fj.
Here is the graph of a system of an extended multicyclic SDSE, with N = 5. The different
subset of the partition are indicated by the different colours. the multicycle corresponds to the
five boxes. An arrow between two boxes means that all vertices of the boxes are related by an
arrow.
Here is the graph of an extended fundamental SDSE. The vertices in Ji, with i ∈ I0, are
green. There are two elements in I0, one with βi = −1 (light green vertices) and one with
βi 6= −1 (dark green vertex). There are two elements in J0, corresponding to light blue and dark
blue vertices. The unique element of K0 is red; the unique element of I1 is yellow; the unique
element of J1 is orange; the dark vertices are the elements of I2. An arrow between two boxes
15
means that all vertices of the boxes are related by an arrow.
For example, the SDSE associated to the following formal series has such a graph:
F1 = fβ(h1)f1(h4 + h5)f1(h6 + h7 + h8)
F2 = F3 = (1 + h2 + h3)f β
1+β
((1 + β)h1)f1(h4 + h5)f1(h6 + h7 + h8)
F4 = F5 = f β
1+β
F6 = F7 = F8 = f β
1+β
F9 = f β
1+β
1
ν
F10 =
f β
νa
((1 + β)h1)f1(h6 + h7 + h8)
((1 + β)h1)f1(h4 + h5)
((1 + β)h1)f1(h4 + h5)f1(h6 + h7 + h8)
F11 =
(10)
1
2
6
νa
(10)
2
(cid:16)νa(10)
1 h1(cid:17) f −1
(cid:16)νa(10)
(h6 + h7 + h8)(cid:17) f0(cid:16)νa(10)
(cid:16)νa(10)
1 − 1 − β(cid:17) h1(cid:17) f −1
−1−β(cid:16)(cid:16)a(10)
−1(cid:16)(cid:16)a(10)
9 h9(cid:17) + a(11)
4 − 1(cid:17) (h4 + h5)(cid:17) f
10 h10 + 1 −
1
ν′ ,
(10)
2
β
a
,
4
νa
1
ν
(10)
4
(cid:16)νa(10)
(h4 + h5)(cid:17)
(h2 + h3)(cid:17) f 1
9 h9(cid:17) + 1 −
(cid:16)a(10)
(h2 + h3)(cid:17)
6 − 1(cid:17) (h6 + h7 + h8)(cid:17)
−1(cid:16)(cid:16)a(10)
1
(10)
6
2
a
(10)
6
f 1
νa
1
ν′ f
a
(10)
1
f
1
(10)
4
a
f0(cid:16)a(10)
F12 = F13 = 1 + a(12)
F14 = 1 + a(14)
F15 = 1 + a(15)
F16 = 1 + a(16)
F17 = 1 + a(17)
F18 = 1 + a(18)
F19 = 1 + a(19)
10 h10,
13 h13,
12 h12 + a(15)
15 h15,
2 h2,
17 h17,
17 h17,
13 h13,
where β 6= −1, ν, ν′ 6= 0, and the coefficients a(i)
j are non-zero.
16
3 Characterisation and properties of Hopf SDSE
3.1 Subalgebras of HD generated by spans of trees
Let us fix a non-empty set D.
Lemma 15 Let V be a subspace of V ect(TD) and let us consider the subalgebra A of HD
generated by V . Recall that for all d ∈ D, f q d is the following linear map:
f q d :(cid:26)
HD −→ K
t1 · · · tn −→ δt1···tn, q d .
Then A is a Hopf subalgebra if, and only if, the two following assertions are both satisfied:
1. For all d ∈ D, (f q d ⊗ Id) ◦ ∆(V ) ⊆ V + K.
2. For all d ∈ D, (Id ⊗ f q d ) ◦ ∆(V ) ⊆ A.
Proof. =⇒. If A is Hopf, then ∆(V ) ⊆ A ⊗ A. As V ⊆ V ect(TD), ∆(V ) ⊆ H ⊗ (V ect(TD) +
K). So:
∆(V ) ⊆ (A ⊗ A) ∩ (H ⊗ (V ect(TD) + K)) = A ⊗ (V ⊕ K).
This implies both assertions.
⇐=. We use here Sweedler's notations: ∆(a) = a′ ⊗ a′′ and (∆ ⊗ Id) ◦ ∆(a) = a′ ⊗ a′′ ⊗ a′′′
for all a ∈ A.
First step. Let us consider the following subspace of P rim(H∗
D):
B = {f ∈ P rim(H∗
D) / (f ⊗ Id) ◦ ∆(V ) ⊆ V + K}.
By hypothesis 1, f q d ∈ B for all d ∈ D. We recall here that ⋆ is the pre-Lie product of P rim(H∗
Let f and g ∈ B. For all v ∈ V :
D).
(f ⋆ g ⊗ Id) ◦ ∆(v) = f ◦ π(v′)g ◦ π(v′′)v′′′.
As f ∈ B, f ◦ π(v′)v′′ ∈ V + K. As g ∈ B, f ◦ π(v′)g ◦ π(v′′)v′′′ ∈ V + K. So f ⋆ g ∈ B, and B
is a sub-pre-Lie algebra of P rim(H∗
D) is generated as a pre-Lie algebra by the
f q d 's, B = P rim(H∗
D). As P rim(H∗
D).
Second step. Let us consider the following subspace of H∗
D:
B′ = {f ∈ H∗
D / (f ⊗ Id) ◦ ∆(A) ⊆ A}.
Let f ∈ P rim(H∗
D). By the first step, for all v1, · · · , vn ∈ V :
(f ⊗ Id) ◦ ∆(v1 · · · vn) = f (v′
1 · · · v′
n)v′′
1 · · · v′′
n =
v1 · · · f (v′
i)v′′
i · · · vn ∈ A,
nXi=1
so P rim(H∗
D) ⊆ B′. Let f, g ∈ B′. For all a ∈ A:
(f g ⊗ Id) ◦ ∆(a) = f (a′)g(a′′)a′′′.
As f ∈ B′, f (a′)a′′ ∈ A. As g ∈ B′, f (a′)g(a′′)a′′′ ∈ A. So B′ is a subalgebra of H∗
contains P rim(H∗
D), it is equal to H∗
D. So:
D. As it
∆(A) ⊆ HD ⊗ A + \f ∈H∗
D
Ker(f ) ⊗ HD = HD ⊗ A.
17
Third step. Let us consider the following subspace of P rim(H∗
D):
C = {f ∈ P rim(H∗
D) / (Id ⊗ f ) ◦ ∆(V ) ⊆ A}.
By the second hypothesis, f q d ∈ B for all d ∈ D. Let us take f and g ∈ C. For all v ∈ V :
(Id ⊗ (f ⋆ g)) ◦ ∆(v) = v′f ◦ π(v′′)g ◦ π(v′′′).
As g ∈ C, v′g ◦ π(v′′) ∈ A. Let us denote:
where v1, . . . , vn are elements of V . Then:
v′ ◦ π(v′′) =X v1 · · · vn,
v′f ◦ π(v′′)g ◦ π(v′′′) =X v′
By the second step, as V ⊆ V ect(TD):
1 · · · v′
nf ◦ π(v′′
1 · · · v′′
n)g ◦ π(v′′′).
∆(V ) ⊆ (HD ⊗ A) ∩ (HD ⊗ (V ect(TD) + K)) = HD ⊗ (V + K).
So:
Finally:
1 · · · v′
1 · · · v′′
n ⊗ π(v′′
n) =X nXi=1
X v′
(Id ⊗ (f ⋆ g)) ◦ ∆(v) =X nXi=1
v1 · · · v′
i · · · vn ⊗ π(v′′
i ).
v1 · · · v′
i · · · vn ⊗ f ◦ π(v′′
i ).
As f ∈ B′, this belongs to A. So f ⋆ g ∈ B′. As at the end of the first step, we conclude that
B′ = P rim(H∗
D).
Last step. As in the second step, we conclude that for all f ∈ H∗
D, (Id ⊗ f ) ◦ ∆(A) ⊆ A. So
∆(A) ⊆ A ⊗ HD, and ∆(A) ⊆ (HD ⊗ A) ∩ (A ⊗ HD) = A ⊗ A. So A is a Hopf subalgebra. ✷
3.2 Definition of the structure coefficients
Proposition 16 Let (S) be an SDSE. It is Hopf if, and only if, for all i, j ∈ I, for all n ≥ 1,
there exists a scalar λ(i,j)
n
such that for all t′ ∈ Ti(n):
Xt∈Ti(n+1)
nj(t, t′)at = λ(i,j)
n at′,
where nj(t, t′) is the number of leaves l of t decorated by j such that the cut of l gives t′.
Proof. =⇒. Let us assume that (S) is Hopf. Then H(S) is a Hopf subalgebra of HI . Let
us use lemma 15, with V = V ect(Xi(n), i ∈ I, n ≥ 1). So (f q j ⊗ Id) ◦ ∆(Xi(n + 1)) belongs
to H(S), and is a linear span of trees of degree n with a root decorated by i, so is a multiple of
Xi(n). We then denote:
(f q j ⊗ Id) ◦ ∆(Xi(n + 1)) = λ(i,j)
n Xi(n) = Xt′∈T (n)
n at′ t′.
λ(i,j)
By definition of the coproduct ∆:
(f q j ⊗ Id) ◦ ∆(Xi(n + 1)) = Xt∈T (n+1), t′∈T (n)
nj(t, t′)att′.
18
The result is proved by identifying the coefficients in the basis T (n) of these two expressions of
(f q j ⊗ Id) ◦ ∆(Xi(n + 1)).
⇐=. Let us prove that both conditions of lemma 15 are satisfied, with the same V as before.
By hypothesis, for all i, j ∈ I, for all n ≥ 2, (f q j ⊗ Id) ◦ ∆(Xi(n)) = λ(i,j)
n−1Xi(n − 1) ∈ V .
Moreover, (f q j ⊗ Id) ◦ ∆(Xi(1)) = δi,j ∈ K, so the first condition is satisfied. For the second
one:
(Id ⊗ f q j ) ◦ ∆(Xi) = (Id ⊗ f q j ) ◦ ∆(B+
i (Fi(Xj, j ∈ I))) = Fi(Xj, j ∈ I) ∈ H(S).
So H(S) is a Hopf subalgebra of HI .
✷
3.3 Properties of the coefficients λ(i,j)
n
The coefficients λ(i,j)
coefficients of the Fi's, as shown by the following result:
's are entirely determined by the a(i)
n
j 's and a(i)
j,k's, and determine the other
Lemma 17 Let us assume that (S) is Hopf, with I = {1, . . . , N }. Let us fix i ∈ I.
1. For all sequence i = i1 −→ · · · −→ in of vertices of G(S):
n = a(in)
λ(i,j)
j +
(1 + δj,ip+1)
n−1Xp=1
a(ip)
j,ip+1
a(ip)
ip+1
.
In particular, λ(i,j)
1 = a(i)
j .
2. For all p1, · · · , pN ∈ N:
a(i)
(p1,··· ,pj+1,··· ,pN ) =
1
pj + 1 λ(i,j)
p1+···+pN +1 −Xl∈I
j ! a(i)
pla(l)
(p1,··· ,pN ).
Proof. 1. Let us consider a sequence i1, · · · , in of elements of I, such that i1 = i and for all
1 ≤ p ≤ n − 1, a(ip)
ip+1
6= 0. By definition of λ(i,j)
:
n
λ(i,j)
n a
q
q...q
q
in = a
in−1
i2
i1
q
q
q...q
q
i2
i1
j + (1 + δj,in)a
in
in−1
q
n a(i1)
λ(i,j)
i2
· · · a(in−1)
in
= a(i1)
i2
· · · a(in−1)
in
j
a
inj
+
in−1
q...q∨q
n−2Xp=1
q ...q
q...q∨q
a(in)
j + (1 + δj,in)a(i1)
i2
i2
i1
i1
q
in
ip+1
ip
,
· · · a(in−1)
in,j
+
n−2Xp=1
(1 + δj,ip+1)a(i1)
i2
· · · a(ip)
j,ip+1
a(ip+1)
ip+2
· · · a(in−1)
in
,
λ(i,j)
n
= a(in)
j +
This proves the first point of the lemma.
(1 + δj,ip+1)
n−1Xp=1
a(ip)
j,ip+1
a(ip)
ip+1
.
2. Let us now fix p1, · · · , pN ∈ N. By definition, for t′ = B+
i ( q 1
p1 · · · q N
pN ):
λ(i,j)
p1+···+pN +1aB+
i ( q 1
p1 ··· q N
pN ) = (pj + 1)aB+
i ( q 1
p1 ··· q j
pj +1··· q N
pN )
+
NXl=1
aB+
i ( q 1
p1 ··· q l
pl−1··· q N
q
pN q
j ),
l
λ(i,j)
p1+···+pN +1a(i)
(p1,··· ,pN ) = (pj + 1)a(i)
(p1,··· ,pj+1,··· ,pN ) +
19
pla(i)
(p1,··· ,pN )a(l)
j .
NXl=1
This proves the second point of the lemma.
✷
Remarks.
(p1,··· ,pN ) = 0, then a(i)
1. As a consequence of the second point, if (S) is Hopf and if a(i)
(l1,··· ,lN ) = 0
if l1 ≥ p1, · · · , lN ≥ pN . In particular, as there is no constant Fi, for all i, there exists a j
such that a(i)
j
6= 0.
2. So the sequences considered in the first point of lemma 17 always exist.
3. Moreover, for all vertices i, j of G(S), i → j if and only if a(i)
j
6= 0.
4. Finally, for all i ∈ I, for all p ≥ 1, Xi(p) 6= 0.
Proposition 18 Let (S) be a Hopf SDSE.
1. Let i, j be vertices of G(S), such that j is not a descendant of i. Then for all n ≥ 1:
λ(i,j)
n = 0.
2. Let (S) be a Hopf SDSE with set of vertices I and let (S′) be a Hopf SDSE with set of
vertices J. Then (S′) is a dilatation of (S) if, and only if, J admits a partition indexed by
the elements of I and for all i, j ∈ I, for all x ∈ Ji, y ∈ Jj, for all n ≥ 1:
3. Let i ∈ I such that:
λ(i,j)
n = λ(x,y)
n
.
Fi = 1 +Xj∈I
a(i)
j hj.
Then for all direct descendant i′ of i, for all j, for all n ≥ 1:
λ(i,j)
n+1 = λ(i′,j)
n
.
As a consequence, if i′, i′′′ are two direct descendants of i, Fi′ = Fi′′.
Proof. 1. Let us consider a sequence i = i1, · · · , in of elements of I such that a(ik)
ik+1
j = 0 and a(ik)
j,ik+1
all 1 ≤ k ≤ n − 1. Then j is not a direct descendant of i1, · · · , in, so a(in)
all k. By lemma 17, λ(i,j)
n = 0.
6= 0 for
= 0 for
2. =⇒. From lemma 17-1, choosing an element xi in Ji for all i ∈ I.
⇐=. Let us consider the dilatation (S′′) of (S) corresponding to the partition of J. Then the
coefficients λ(i,j)
n
of (S′) and (S′′) are equal, so by lemma 17-2, (S′) = (S′′).
3. Let us consider a sequence i, i′ = i1, · · · , in of elements of I such that a(ik)
ik+1
6= 0 for all
1 ≤ k ≤ n − 1. By hypothesis on i, a(i)
j,i′ = 0. By lemma 17-1:
λ(i,j)
n+1 = a(in)
j + 0 +
(1 + δj,ik+1)
n−1Xk=1
a(ik)
j,ik+1
a(ik)
ik+1
= λ(i′,j)
n
.
So, if i′ and i′′ are two direct descendants of i, for all k ∈ I, for all n ≥ 1, λ(i′,k)
lemma 17-2, Fi′ = Fi′′ .
n
= λ(i′′,k)
n
. By
✷
20
Proposition 19 Let (S) be an SDSE, with I = {1, . . . , N }. It is Hopf if, and only if, the
two following conditions are satisfied:
1. There exist scalars λ(i,j)
n
satisfying, for all 1 ≤ i, j ≤ N , for all (p1, · · · , pN ) ∈ NN :
a(i)
(p1,··· ,pj+1,··· ,pN ) =
1
pj + 1 λ(i,j)
p1+···+pN +1 −Xl∈I
j ! a(i)
pla(l)
(p1,··· ,pN ).
2. For all p ≥ 1, for all i, j, d1, · · · , dp ∈ I, such that a(i)
(p1,··· ,pN ) 6= 0 where pi is the number of
dp's equal to i, for all n1, · · · , np ≥ 1:
λ(i,j)
n1+···+np+1 − a(i)
j = λ(i,j)
p+1 − a(i)
Proof. Preliminary step. Let us assume the first point and let t′ ∈ T (i)
D . We use the following
nl − a(dl)
j (cid:17) .
j +Xl∈I (cid:16)λ(dl,j)
srs .
rs.
D
t′ = B+
i Ys∈TD
pj = Xs∈T (j)
NYj=1
Ys∈TD
i q j Ys∈TD
a(i)
pj!
rs!
ars
s .
(p1,··· ,pN ) Ys∈TD
srs , t′ aB+
i ( q j Q srs )
(rs1 + 1)nj(s1, s2)a
i (cid:16) s1
B+
s2 Q srs(cid:17)
notations:
We also denote, for all j ∈ I:
Then, by (1):
Hence:
Xt∈T (i)
D
at′ =
nj(t, t′)at = njB+
+ Xs1,s2∈TD
rs2 ≥1
(pj + 1)
NYj=1
(r q j +1) Ys∈TD
pj!
rs!
a(i)
(p1,··· ,pj+1,··· ,pN )a q j Ys∈TD
ars
s
(rs1 + 1)nj(s1, s2)
rs2
rs1 + 1
at′
as1
as2
a(i)
(p1,··· ,pj+1,··· ,pN )
a(i)
(p1,··· ,pN )
p1+···+pN +1 −
pja(l)
at′ + Xs1,s2∈TD
j + Xs1,s2∈TD
rs2 >0
nj(s1, s2)rs2at′
nj(s1, s2)rs2
as1
as2
as1
as2
at′.
NXl=1
21
= (r q j +1)
= (pj + 1)
+ Xs1,s2∈TD
= λ(i,j)
=⇒. Let us assume that (S) is Hopf. We already prove the existence of the scalars λ(i,j)
n
. We
obtain from the preceding computation:
λ(i,j)
weight(t′)at′ =λ(i,j)
p1+···+pN +1 −
NXl=1
pja(l)
j + Xs2∈TD
rs2λ(d(s2),j)
weight(s2) at′,
where d(s2) is the decoration of the root of s2. Let us choose p, i, j, d1, · · · , dp, n1, · · · , np as in
the hypotheses of the proposition. Let us choose for all 1 ≤ j ≤ p a tree sj with root decorated
by dj, of weight nj, such that asj 6= 0: this always exists (for example take a convenient ladder).
Let us take t′ = B+
i (s1 · · · sp). Then at′ 6= 0 because a(i)
(p1,··· ,pN ) 6= 0, so:
λ(i,j)
n1+···+np+1 = λ(i,j)
p+1 +
pXl=1(cid:16)λ(dl,j)
nl
− a(dl)
j (cid:17) .
⇐=. Let us show the condition of proposition 16 by induction on the weight n of t′. For
n = 1, then t′ = q i . Then, by hypothesis on the a(i)
(p1,··· ,pN ), a(i)
j = λ(i,j)
1
. So:
nj(t, t′)at = q
q
j = a(i)
j = λ(i,j)
1
i
a q i .
Let us assume the result for all tree of weight < n. The preceding computation then gives:
Xt∈Ti(n+1)
nj(t, t′)at =λ(i,j)
Xt∈T (i)
D
p1+···+pN +1 −
NXl=1
pja(l)
j + Xs1,s2∈TD
rs2 >0
nj(s1, s2)rs2
as1
as2
at′.
The induction hypothesis and the condition on the coefficients λ(i,j)
to λ(i,j)
weight(t′)+1at′. So H(S) is a Hopf subalgebra of HI .
n
then give that this is equal
✷
3.4 Prelie structure on H∗
(S)
Let us consider a Hopf SDSE (S). Then H∗
g(S) = P rim(cid:16)H∗
(S) is the enveloping algebra of the Lie algebra
D) a pre-Lie product given in the following
way: for all f, g ∈ G(S), for all x ∈ H(S), f ⋆ g is the unique element of g(S) such that for all
x ∈ vect(Xi(n) / i ∈ I, n ≥ 1),
(S)(cid:17), which inherits from P rim(H∗
(f ⋆ g)(x) = (f ⊗ g) ◦ (π ⊗ π) ◦ ∆(x).
Let (fi(p))i∈I,p≥1 be the basis of g(S), dual of the basis (Xi(p))i∈I,p≥1. By homogeneity of ∆,
and as ∆(Xi(n)) is a linear span of elements − ⊗ Xi(p), 0 ≤ p ≤ n, we obtain the existence of
coefficients a(i,j)
k,l
such that, for all i, j ∈ I, k, l ≥ 1:
fj(l) ⋆ fi(k) = a(i,j)
k,l fi(k + l).
By duality, a(i,j)
k,l
the following way: for all t′ ∈ T (j)
D (l), t′′ ∈ T (i)
D (k),
is the coefficient of Xj(l) ⊗ Xi(k) in ∆(Xi(k + l)), so is uniquely determined in
Xt∈T (i)
D (k+l)
n(t′, t′′; t)at = a(i,j)
k,l at′at′′ .
22
Lemma 20 For all t′ ∈ T (j)
D (l), t′′ ∈ T (i)
D (k):
Xt∈T (i)
D (k+l)
n(t′, t′′; t)at = λ(i,j)
k
at′at′′.
Proof. By induction on k. If k = 1, then t′′ = q i , so:
Xt∈T (i)
D (k+l)
n(t′, t′′; t)at = aB+
i (t′′) = a(i)
j at′ = λ(i,j)
1
at′at′′,
as at′′ = 1. Let us assume the result at all rank ≤ k − 1. We put t′′ = B+
srs). We put
i (Ys∈TD
rs for all j ∈ I. Then:
D
pj = Xs∈T (j)
Xt∈T (i)
D (k+l)
n(t′, t′′; t)at = nt′, t′′, B+
+ Xs1,s2∈TD
rs2 ≥1
i q j Ys∈TD
srs aB+
(rs1 + 1)n(t′, s2; s1)a
i (t′ Q srs )
i (cid:16) s1
B+
s2 Q srs(cid:17)
(pj + 1)
NYj=1
(rt′+1) Ys∈TD
pj!
rs!
a(i)
(p1,··· ,pj+1,··· ,pN )at′ Ys∈TD
ars
s
(rs1 + 1)nj(s1, s2)
rs2
rs1 + 1
at′′
as1
as2
a(i)
(p1,··· ,pj+1,··· ,pN )
a(i)
(p1,··· ,pN )
at′at′′ + Xs1,s2∈TD
j + Xs1,s2∈TD
j + Xs2∈TD
rs2 >0
nj(s1, s2)rs2
as1
as2
at′′
as1
as2
at′at′′
at′at′′ ,
NXl=1
NXl=1
p1+···+pN +1 −
pja(l)
nj(s1, s2)rs2
p1+···+pN +1 −
pja(l)
rs2λ(r(s2),j)
s2
= (rt′+1)
= (pj + 1)
+ Xs1,s2∈TD
= λ(i,j)
= λ(i,j)
using the induction hypothesis on s2, denoting by r(s2) the decoration of the root of s2. By
proposition 19-2, if at′ 6= 0, then a(i)
(p1,··· ,pn) 6= 0, so:
1+P rss = λ(i,j)
λ(i,j)
t′′ = λ(i,j)
λ(i,j)
1+P rs
+Xs
rs(cid:16)λ(r(s),j)
s
p1+···+pN +1 +Xs
− a(r(s))
j
(cid:17)
−Xl
rsλ(r(s),j)
s
pla(l)
j .
So the induction hypothesis is proved at rank n.
✷
Combining this lemma with the preceding observations:
23
Proposition 21 Let (S) be a Hopf SDSE. The pre-Lie algebra g(S) = P rim(cid:16)H∗
basis (fi(k))i∈I,k≥1, and the pre-Lie product of two elements of this basis is given by:
(S)(cid:17) has a
fj(l) ⋆ fi(k) = λ(i,j)
k
fi(k + l).
4 Level of a vertex
The second item of proposition 19-2 is immediately satisfied if there exist scalars bj and a(i)
j
that λ(i,j)
level of a vertex.
for all n ≥ 1 and all i, j ∈ I. This motivates the definition of the
n = bj(n − 1) + a(i)
j
such
4.1 Definition of the level
Definition 22 Let (S) be a Hopf SDSE, and let i be a vertex of G(S). It will be said to be
of level ≤ M if for all vertex j, there exist scalar b(i)
j , a(i)
j , such that for all n > M :
n = b(i)
λ(i,j)
j (n − 1) + a(i)
j .
The vertex i will be said to be of level M if it is of level ≤ M and not of level ≤ M − 1.
Remark. In order to prove that i is of level ≤ M , it is enough to consider the j's which
n = 0 for all
are descendants of i. Indeed, if j is not a descendant of i, by proposition 18-1, λ(i,j)
n ≥ 1.
Proposition 23 Let (S) be a Hopf SDSE, i a vertex of G(S) and j a direct descendant of
G(S).
1. i has level 0 or 1 if, and only if, j as level 0.
2. Let M ≥ 2. Then i has level M if, and only if, j has level M − 1.
Moreover, if this holds, then for all k ∈ I, b(i)
k = b(j)
k .
Proof. Let i ∈ G(S) and j be a direct descendant of i. As (S) is Hopf, let us use the second
point of proposition 19, with k = 1 and d1 = j. Then for all l, for all n ≥ 1, as a(i)
j
6= 0:
λ(i,l)
n+1 = λ(i,l)
2 + λ(j,l)
n − a(j)
l
.
So for all M ≥ 1, i is of level ≤ M if, and only if, j is of level ≤ M − 1. Moreover, if this holds,
then b(i)
for all k.
k = b(j)
k
The first point is a reformulation of the preceding result for M = 1. Let us assume that
M ≥ 2. If i is of level M , then j is of level ≤ M − 1. If j is of level ≤ M − 2, then i is of level
≤ M − 1: contradiction. So j is of level M − 1. The converse is proved in the same way.
✷
Corollary 24 Let (S) be a connected Hopf SDSE. Then if one of the vertices of G(S) is of
j depend only
finite level, then all vertices of G(S) are of finite level. Moreover, the coefficients b(i)
of j. They will now be denoted by bj.
Proposition 18-1 immediately implies the following result:
Lemma 25 Let (S) be a connected Hopf SDSE and let j be a vertex of G(S) of finite level.
If there exists a vertex i in G(S) which is not a descendant of j, then bj = 0.
24
4.2 Vertices of level 0
Let (S) be a Hopf SDSE with I = {1, . . . , N }, and let us assume that i is a vertex of level 0. In
this case, the coefficients a(i)
(p1,··· ,pN ) satisfy an induction of the following form:
a(i)
(0,··· ,0) = 1,
a(i)
(p1,··· ,pj+1,··· ,pN ) =
1
pj + 1 λj +
j pl! a(i)
µ(l)
(p1,··· ,pN ).
NXl=1
In order to ease the notation, we shall write a(p1,··· ,pN ) instead of a(i)
in this section.
(p1,··· ,pN ) and F instead of Fi
Lemma 26 Under the preceding hypothesis:
1. Let us denote J = {j ∈ I / λj = 0}. There exists a partition I = I1 ∪ · · · ∪ IM ∪ J, and
scalars β1, · · · , βM , such that for all i, j ∈ I \ J = I1 ∪ · · · ∪ IM :
µ(j)
i =(cid:26) 0 if i, j do not belong to the same Il,
λiβl if i, j ∈ Il.
2. Moreover F (h1, · · · , hN ) =
MYp=1
Proof. Let us fix i 6= j. Then:
fβpXl∈Ip
λlhl.
a(p1,··· ,pi+1,··· ,pj+1,··· ,pN )
1
µ(l)
i +
pi + 1 λi + µ(j)
NXl=1
(pi + 1)(pj + 1) λi + µ(j)
(pi + 1)(pj + 1) λj + µ(i)
i pl! a(p1,··· ,pj+1,··· ,pN )
i pl! λj +
j pl! λi +
i +
j +
µ(l)
µ(l)
1
1
NXl=1
NXl=1
=
=
=
µ(l)
j pl! a(p1,··· ,pN ),
i pl! a(p1,··· ,pN ).
µ(l)
NXl=1
NXl=1
For (p1, · · · , pN ) = (0, · · · , 0), as a(0,··· ,0) = 1:
µ(j)
i λj = µ(i)
j λi.
For (p1, · · · , pN ) = εk, we obtain:
(cid:16)λi + µ(j)
i + µ(k)
i (cid:17)(cid:16)λj + µ(k)
j (cid:17) λk =(cid:16)λj + µ(i)
j + µ(k)
j (cid:17)(cid:16)λi + µ(k)
i (cid:17) λk.
µ(j)
i µ(k)
j = µ(i)
j µ(k)
i
.
So, if λk 6= 0:
(2)
(3)
If λk = 0, it is not difficult to prove inductively that a(p1,··· ,pN ) = 0 if pk > 0, so F is an element
of K[[h1, · · · , hk−1, hk+1, · · · , hN ]]. Hence, up to a restriction to I \ J, we can suppose that all
the λk's are non-zero. We then put ν(j)
for all i, j. Then (2) and (3) become: for all i, j, k,
i = µ(j)
i
λi
= ν(i)
j
ν(j)
i
i − ν(k)
j (cid:17) = 0.
ν(j)
i (cid:16)ν(k)
25
,
(4)
(5)
Let 1 ≤ i, j ≤ N . We shall say that i R j if i = j or if ν(j)
6= 0. Let us show that R is an
equivalence. By (4), it is clearly symmetric. Let us assume that i R j and j R k. If i = j or j = k
or i = k, then i R k. If i, j, k are distinct, then ν(j)
6= 0, so
j
i R k. We denote by I1, · · · , IM the equivalence classes of R .
6= 0. By (5), ν(k)
6= 0 and ν(k)
i = ν(k)
j
i
i
Let us assume that i R j, i 6= j. Then ν(j)
6= 0, so for all k, ν(k)
j = ν(k)
i
i
.
In particular,
ν(i)
j = ν(i)
i = ν(j)
i = ν(j)
j
. So, finally, there exists a family of scalars (βi)1≤i≤M , such that:
• If i, j ∈ Il, then ν(j)
i = βl, and µ(j)
i = λiβl.
• If i and j are not in the same Il, then ν(j)
i = µ(j)
i = 0.
An easy induction then proves:
a(p1,··· ,pN ) =
λp1
1 · · · λpN
N
p1! · · · pN !
This implies the assertion on F .
4.3 Vertices of level 1
(1 + βp) · · ·1 + βpXl∈Ip
MYp=1
pl − 1 .
✷
Let us now assume that i is of level 1. Then, up to a restriction to i and its direct descendants,
the coefficients a(i)
(p1,··· ,pN ) = a(p1,··· ,pN ) satisfy an induction of the form:
a(i)
(0,··· ,0) = 1,
a(i)
εj = a(i)
j ,
1
a(i)
(p1,··· ,pj+1,··· ,pN ) =
pj + 1 λj +
j pl! a(i)
µ(l)
(p1,··· ,pN ) if (p1, · · · , pN ) 6= (0, · · · , 0).
NXl=1
In order to ease the notation, we shall write a(p1,··· ,pN ) instead of a(i)
in this section.
(p1,··· ,pN ) and F instead of Fi
Lemma 27 Under the preceding hypothesis, one of the following assertions holds:
1. There exists a partition I = I1 ∪ · · · ∪ IM ∪ J, scalars β1, · · · , βM , a non-zero scalar ν such
2. There exists a partition {1, · · · , N } = I1 ∪ · · · ∪ IM ∪ J, scalars νp for 1 ≤ p ∈ M , such
that:
F (h1, · · · , hN ) =
that:
F (h1, · · · , hN ) = 1 −
alhl + 1 −
1
ν
.
alhl +Xl∈J
1
ν
1
νp
MYp=1
νalhl +Xl∈J
fβpXl∈Ip
ln1 − νpXl∈Ip
MXp=1
k (cid:17) aj.
j (cid:17) ak =(cid:16)λk + µ(j)
ak (cid:12)(cid:12)(cid:12)(cid:12)(cid:12)
λj + µ(k)
λk + µ(j)
k
= 0.
aj
j
26
(cid:16)λj + µ(k)
(cid:12)(cid:12)(cid:12)(cid:12)(cid:12)
alhl.
(6)
Proof. Let us compute aj,k in two different ways:
In other words:
Let us take J = {j / ∀k, λj + µ(k)
j = 0}. Let us consider an element j ∈ J. Then an easy
induction proves that for all (p1, · · · , pN ) such that p1 + · · · + pN ≥ 2 and pj ≥ 1, a(p1,··· ,pN ) = 0.
As a consequence:
F (h1, · · · , hN ) = F (h1, · · · , hj−1, 0, hj+1, · · · , hN ) + ajhj.
So:
F = F (hi, i /∈ J) +Xj∈J
ajhj.
We now assume that, up to a restriction, J = ∅. Let us choose an i and let us put b(p1,··· ,pN ) =
(pi + 1)a(p1,··· ,pi+1,··· ,pN ). Then, for all j ∈ I, for all (p1, · · · , pN ):
b(p1,··· ,pj+1,··· ,pN ) =
1
pj + 1 λj + µ(i)
j +
j pl! b(p1,··· ,pN ).
µ(l)
NXl=1
We deduce from lemma 26 that there exist a partition I = I1 ∪ · · · ∪ IM and scalars β1, . . . , βM ,
such that:
µ(l)
j =( 0 if j, l are not in the same Ik,
j (cid:17) βk if j, l ∈ Ik.
(cid:16)λj + µ(i)
So µ(i)
j does not depend on i such that µ(i)
j
6= 0. So there exist scalars µj such that:
µ(l)
j =(cid:26) 0 if j, l are not in the same Ik,
(λj + µj) βk if j, l ∈ Ik.
1. Let us assume that M ≥ 2. Let us choose j ∈ I1. Then for all k ∈ I2 ∪ · · · ∪ IM , (6) gives:
(cid:12)(cid:12)(cid:12)(cid:12)
aj
λj
λk ak (cid:12)(cid:12)(cid:12)(cid:12) = 0.
We denote I2 ∪ · · · ∪ Ik = {i1, · · · , iM }. We proved that the vectors (λj, λi1, · · · , λiM ) and
(aj, ai1, · · · , aiM ) are colinear. Choosing then a j ∈ I2, we obtain that there exists a scalar
ν, such that (λi)i∈I = ν(ai)i∈I . Two cases are possible.
(a) If ν 6= 0, putting a′
then the family(cid:16)a′
(p1,··· ,pN ) = νa(p1,··· ,pN ) if (p1, · · · , pN ) 6= (0, · · · , 0) and a′
(p1,··· ,pN )(cid:17) satisfies the hypothesis of lemma 26. As a consequence,
(0,··· ,0),
F (h1, · · · , hN ) satisfies the first case.
(b) If ν = 0, then we put, for all j, µj = ν′
j = ν′
k
if j and k are in the same Il: this common value is now denoted νl. It is then not
difficult to prove that:
jaj. By (6), for j and k in the same Il, ν′
F (h1, · · · , hN ) = 1 −
This is a second case.
1
νp
MXp=1
ln1 − νpXl∈Ip
alhl .
2. Let us assume that M = 1. Then (λj + µj)β1 = µ(i)
j
for all i, j ∈ I.
(a) Let us suppose that β1 6= 1. Then, for all j, k ∈ I µj = β1
1−β1
λj. So, for all j,
λj. So (6) implies that (λj)j∈I and (aj)j∈I are colinear. As in 1.(a),
λj + µj = 1
this is a first case.
1−β1
(b) Let us assume that β1 = 1. So λj = 0 for all j. As in 1.(b), this is a second case.
✷
27
4.4 Vertices of level ≥ 2
Lemma 28 Let (S) be a Hopf SDSE and let i be a vertex of G(S). We suppose that there
exists a vertex j, such that:
• j is a descendant of i.
• All oriented path from i to j are of length ≥ 3.
Then Fi = 1 + Xi−→l
a(i)
l hl.
Proof. We assume here that I = {1, . . . , N }. Let L be the minimal length of the oriented
paths from i to j. By hypothesis, L ≥ 3. Then the homogeneous component of degree L + 1 of
Xi contains trees with a leave decorated by j, and all these trees are ladders (that is to say trees
with no ramification). By proposition 16, if t′ ∈ T (i)
D (L):
λ(i,j)
L at′ = Xt∈T (i)
D (L+1)
nj(t, t′)at.
For a good-chosen ladder t′, the second member is non-zero, so λ(i,j)
is non-zero. If t′ is not a
ladder, the second member is 0, so at′ = 0. As a conclusion, Xi(L) is a linear span of ladders.
Considering its coproduct, for all p ≤ L, Xi(p) is a linear span of ladders. In particular, Xi(3) is
a linear span of ladders. But:
L
Xi(3) =Xl,m
a(i)
l a(l)
m q
q
q
m
l
i
+Xl≤m
q
a(i)
l,m q∨q
ml
,
i
so a(i)
l,m = 0 for all l, m. Hence, Fi contains only terms of degree ≤ 1.
✷
Remark. This lemma can be applied with i = j, if i is not a self-dependent vertex.
Proposition 29 Let (S) be a Hopf SDSE and let i be a vertex of G(S) of level ≥ 2. Then i
is an extension vertex.
Proof. We denote by M the level of i. By proposition 23, all the descendants of i are of
level ≤ M − 1, so i is not a descendant of itself.
Let M be the level of i and let us assume that M ≥ 3. Let j be a direct descendant of i, k
be a direct descendant of j, l be a direct descendant of k. Then j has level M − 1, k has level
M − 2, l has level M − 3. So in the graph of the restriction to {i, j, k, l} is:
i
/ j
/ k
/ l or i
/ j
/ k
/ l c
The result is then deduced from lemma 28.
Let us now assume that i is of level 2 and is not an extension vertex. Let j be a direct
descendant of i and k be a direct descendant of j. By proposition 23, j is of level 1 and k is of
level 0, so k is not a direct descendant of i. The graph of the restriction of (S) to {i, j, k} is:
i
/ j
/ k or i
/ j
/ k e
First step. Let us first prove that there exists a direct descendant j of i such that a(i)
j,j 6= 0.
Let us assume that this is not true. As i is not an extension vertex, there exist j, j′ ∈ I such
28
/
/
/
/
/
/
c
/
/
/
/
e
that a(i)
j,j ′
graph associated to the restriction to {i, j, j′, k} is:
6= 0, j 6= j′. Let k be a direct descendant of j. Considering the different levels, the
i
or
j′
j
j
<<<<<<<<
i
k
========
<<<<<<<<
Up to a change of variables, we put:
k e
========
or
j′
j
<<<<<<<<
i
========
or
j′
j
k
<<<<<<<<
i
========
j′
k e
Fi(0, · · · , 0, hj , 0, · · · , 0, hj ′, 0, · · · , 0) = 1 + hj + hj ′ + bhjhj ′ + O(h3).
Then by proposition 16, λ(i,j)
λ(i,j)
2
j ′ = a
i
+ a
a q
j ′
2
j
q
q
q
q
q∨q
i
j ′
i
q
q
a q
j = 2a
i
j
+ a
j
q
q
q
j
i
q
q∨q
i
j = 0, so λ(i,j)
2
j = b, so 0 = b: this contradicts a(i)
j,j ′ 6= 0.
= 0. On the other hand,
Second step. Let us consider a vertex j such that a(i)
we can assume that a(i)
bi = bj = 0. So, as i is of level 2, there exist scalars a, b, such that:
j = 1 and that for all direct descendant k of j, a(j)
j,j 6= 0. Up to a change of variables,
k = 1. By lemma 25,
λ(i,j)
n =
1 if n = 1,
a if n = 2,
b if n ≥ 3.
Then proposition 19-1 implies:
Fi(0, · · · , 0, hj , 0, · · · , 0) = 1 + hj +
a
2!
h2
j +
ab
6
h3
j + O(h4
j ).
By hypothesis, a 6= 0. Moreover, by proposition 16, b = λ(i,j)
3
a
k = a
j
i
q
q
q
j
k = a. So:
q
j
q
q∨q
i
Fi(0, · · · , 0, hj , 0, · · · , 0) = 1 + hj +
a
2!
h2
j +
a2
6
h3
j + O(h4
j ).
As j has level 1, we put:
n =( a(j)
λ(j,k)
k = 1 if n = 1,
c(n − 1) + d if n ≥ 2,
where c(= bk) and d are scalars. From proposition 19-1:
Fj(0, · · · , 0, hk, 0, · · · , 0) = 1 + hk +
c + d
2!
h2
k +
(c + d)(2c + d)
6
h3
k + O(h4
k).
Moreover, λ(i,k)
3
a
j
j
q∨qq
i
= a
k , so λ(i,k)
3
q
a
2 = a and λ(i,k)
3
= 2. Then λ(i,k)
3
a
k = 2a
j
i
q
q
q
k
, so
k
q
q∨q
j
q i
c + d = 2. Similarly, using
j
j
j
q
i
j
q∨q
q∨q
q q
i
j
, we obtain λ(i,k)
4 = 3. Using
3
c + d
2
= 3
(c + d)(2c + d)
6
k
.
k
q
q∨q
j
q i
, we obtain:
As c + d = 2, 2c + d = 3, so c = d = 1 and λ(j,k)
all n ≥ 1.
n = n for all n ≥ 2. As λ(j,k)
1 = 1, λ(j,k)
n = n for
29
e
e
Let now l ∈ I which is not a direct descendant of j and let k be a direct descendant of j. For
all n ≥ 1:
λ(j,l)
n = λ(j,l)
n aB+
j ( q k
n−1) = aB+
j ( q k
n−1
q
q
We proved that for any vertex l of G(S), for all n ≥ 1:
l ) = (n − 1)a(k)
l
k
.
n =( n if l is a direct descendant of j,
a(k)
l
λ(j,l)
(n − 1) if l is not a direct descendant of j,
where k is any direct descendant of j. This proves that j has level 0, so i has level 1: contradiction.
So i is an extension vertex.
✷
5 Examples of Hopf SDSE
5.1 cycles and multicycles
Notation. We denote by l(i1, · · · , in) the ladder with decorations, from the root to the leave,
i1, · · · , in. In other words:
l(i1, · · · , ip) = B+
i1
◦ · · · ◦ B+
in
q
q...q
(1) = q
in
in−1
.
i2
i1
Theorem 30 Let N ≥ 2. The SDSE associated to the following formal series is Hopf:
F1 = 1 + h2,
...
FN −1 = 1 + hN ,
FN = 1 + h1.
Proof. We identify {1, · · · , N } and Z/N Z, via the bijection i −→ i. Then, for all n ≥ 1 and
for all 1 ≤ i ≤ N , Xi(n) = l(i, · · · i + n − 1). As a consequence:
∆(Xi) = Xi ⊗ 1 + 1 ⊗ Xi +
+∞Xp=1
Xi+p ⊗ Xi(p).
So H(S) is Hopf.
✷
Note that the graph G(S) associated to such a system is an oriented cycle of length N , with
only non-self-dependent vertices.
Definition 31 Let (S) be a Hopf SDSE. It will be said to be multicyclic if, up to change of
variable, it is a dilatation of a system described in theorem 30.
The graph of a multicyclic SDSE will be called a multicycle. In other term, a N -multicycle
(N ≥ 2) is such that the set I of its vertices admits a partition I = I1 ∪ · · · ∪ IN indexed by the
elements of Z/N Z, such that the direct descendants of a vertex i in Ij are the elements of Ij+1
for all j ∈ Z/N Z. Moreover, up to a change of variables, for all i ∈ G(S):
Fi = 1 + Xi−→l
hl.
30
Here is an example of a 5-multicycle:
Note that if N = 2, G(S) is a complete bipartite graph, that is to say that the set of vertices
of G(S) admits a partition into two parts, and for all vertices i and j, there is an edge from i to
j if, and only if, i and j are not in the same part of the partition.
5.2 Fundamental SDSE
Theorem 32 Let I be a set with a partition I = I0 ∪ J0 ∪ K0 ∪ I1 ∪ J1, such that:
• I0, J0, K0, I1, J1 can be empty.
• I0 ∪ J0 is not empty.
The SDSE defined in the following way is Hopf:
1. For all i ∈ I0, there exists βi ∈ K, such that:
2. For all i ∈ J0:
3. For all i ∈ K0:
f1(hj).
f βj
1+βj
Fi = fβi(hi) Yj∈I0−{i}
Fi = Yj∈I0
Fi = Yj∈I0
((1 + βj)hj )Yj∈J0
((1 + βj)hj ) Yj∈J0−{i}
((1 + βj)hj)Yj∈J0
f βj
1+βj
f βj
1+βj
f1(hj ).
f1(hj).
4. For all i ∈ I1, there exist νi ∈ K, a family of scalars (a(i)
j )j∈I0∪J0∪K0, such that (νi 6= 1)
or (∃j ∈ I0, a(i)
j
6= 1 + βj) or (∃j ∈ J0, a(i)
j
6= 1) or (∃j ∈ K0, a(i)
j
6= 0). Then, if νi 6= 0:
Fi =
1
νi Yj∈I0
If νi = 0:
j (cid:16)νia(i)
j hj(cid:17)Yj∈J0
(i)
f βj
νia
f 1
νi a
j (cid:16)νia(i)
j hj(cid:17) Yj∈K0
(i)
f0(cid:16)νia(i)
j hj(cid:17) + 1 −
1
νi
.
Fi = −Xj∈I0
a(i)
j
βj
ln(1 − hj) −Xj∈J0
a(i)
j
ln(1 − hj) + Xj∈K0
a(i)
j hj + 1.
5. For all i ∈ J1, there exists νi ∈ K − {0}, a family of scalars (a(i)
j )j∈I0∪J0∪K0∪I1, with the
following conditions:
1 = {j ∈ I1 / a(i)
• I (i)
• For all j ∈ I (i)
j
1 , νj = 1.
6= 0} is not empty.
31
• For all j, k ∈ I (i)
t ∈ I0 ∪ J0 ∪ K0.
1 , Fj = Fk. In particular, we put b(i)
t = a(j)
t
for any j ∈ I (i)
1 , for all
Then:
Fi =
1
νi Yj∈I0
+ Xj∈I (i)
1
βj
−1−βj (cid:16)(cid:16)b(i)
j − 1 − βj(cid:17) hj(cid:17)Yj∈J0
f 1
(i)
j
b
−1(cid:16)(cid:16)b(i)
j − 1(cid:17) hj(cid:17) Yj∈K0
j hj(cid:17)
f0(cid:16)b(i)
f
b
(i)
j
a(i)
j h1 + 1 −
1
νi
.
Proof.
In order to simplify the notation, we assume that I = {1, . . . , N }. We shall use
proposition 19 with, for all i, j ∈ I:
n =( a(i)
λ(i,j)
if n = 1,
j
a(i)
j + bj(n − 1) if n ≥ 2,
the coefficients being given in the following arrays:
1. a(j)
i
:
2. a(j)
i
:
3. bj:
i \ j
∈ I0
∈ J0
∈ I0
(1 + βi) − δi,jβi
1 + βi
∈ J0
∈ K0
∈ I1
∈ J1
1
0
0
0
1 − δi,j
0
0
0
1
1 + βi
∈ K0 ∈ I1
a(j)
i
a(j)
i
a(j)
i
0
0
0
0
0
∈ J1
b(j)
i −1−βi
νj
b(j)
i −1
νj
b(j)
i
νj
a(j)
i
0
i \ j
∈ I0
∈ J0
∈ K0
∈ I1
∈ J1
∈ I0
(1 + βi) − δi,jβi
1
0
0
0
∈ J0
1 + βi
1 − δi,j
0
0
0
∈ K0
∈ I1
1 + βi νja(j)
i
νja(j)
i
νja(j)
i
0
0
1
0
0
0
∈ J1
b(j)
i − 1 − βi
b(j)
i − 1
b(j)
i
0
0
j
bj
∈ I0
1 + βj
∈ J0 ∈ K0 ∈ I1 ∈ J1
1
0
0
0
The second item of proposition 19 is immediate. Let us prove for example the first item for
i ∈ J1 and j ∈ I0. Let us fix (p1, . . . , pN ) ∈ NN − {(0, . . . , 0)}.
λ(i,j)
p1+...+pN +1 −Xl
a(l)
j pl
= b(i)
= b(i)
j − 1 − βj − (1 + βj)
(1 + βj)pl + βjpj − Xl∈I1∪J1
pl − Xl∈I0∪J0∪K0
j (cid:17) pl.
j − 1 − βj + βjpj + Xl∈I1∪J1(cid:16)1 + βj − a(l)
NXl=1
a(l)
j pl
32
If there exists l ∈ (I1 ∪ J1) − I (i)
then the result is immediate. We now suppose that pl = 0 for all l ∈ (I1 ∪ J1) − I (i)
1 , such that pl 6= 0, then a(i)
(p1,...,pj+1,...,pN ) = a(i)
(p1,...,pN ) = 0 and
1 . Then:
j pl = b(i)
a(l)
j − 1 − βj + βjpj + Xl∈I (i)
j − 1 − βj + βjpj +(cid:16)1 + βj − b(i)
j (cid:17) pl
1 (cid:16)1 + βj − a(l)
j (cid:17) Xl∈I (i)
1
pl.
= b(i)
λ(i,j)
p1+...+pN +1 −Xl
1. If Xl∈I (i)
1
pl = 0, then:
a(i)
(p1,...,pj+1,...,pN ) =(cid:16)b(i)
The first item of proposition 19 is immediate.
.
(p1,...,pN )
pj + 1
j − 1 − βjpj(cid:17) a(i)
p1+...+pN +1 −Pl a(l)
pl = 1, then a(i)
(p1,...,pj+1,...,pN ) = 0 and λ(i,j)
j pl = 0. So the first item
1
2. If Xl∈I (i)
3. If Xl∈I (i)
1
of proposition 19 holds.
pl ≥ 2, then a(i)
(p1,...,pj+1,...,pN ) = a(i)
(p1,...,pN ) = 0, so the result is immediate.
The other cases are proved in the same way, so this SDSE is Hopf.
✷
Remarks.
1. For all λ 6= 0:
(λh) =
f β
λ
λ(λ + β) · · · (λ + (k − 1)β)
k!
hk.
∞Xk=0
The second side of this formula is equal to 1 if λ = 0. So, formulas defining the SDSE of
theorem 32 are always defined.
2. The vertices of I0 ∪ J0 ∪ K0 are of level 0. A vertex i of I1 is of level 0 if νi = 1; otherwise,
it is of level 1. The vertices of J1 are of level 1.
Definition 33
1. A Hopf SDSE will be said to be fundamental if, up to a change of variables, it is the
dilatation of a system of theorem 32.
2. A fundamental Hopf SDSE (S) will be said to be abelian if for any vertex i ∈ I, bi = 0.
Remark. In other words, (S) is abelian if J0 = ∅ and if for any i ∈ I0, βi = −1. Then, for
all i ∈ K0, Fi = 1. As there is no constant Fi, we obtain K0 = ∅.
A particular case is obtained when I = J0. Then we obtain the following systems:
Theorem 34 Let I be a finite subset which is not a singleton. The SDSE associated to the
following formal series is Hopf:
Fi =Yj6=i
(1 − hj)−1, for all i ∈ I.
33
The graph associated to such an SDSE is a complete graph with only non-self-dependent
vertices, that is to say that there is an edge from i to j in G(S)if, and only if, i 6= j. In particular,
if N = 2, G(S) is 1 ←→ 2, as for the SDSE of theorem 30 with N = 2.
Definition 35 Let (S) be a Hopf SDSE. It will be said to be quasi-complete if, up to change
of variable, it is a dilatation of one of the systems described in theorem 34.
The graphs associated to quasi-complete SDSE shall be called quasi-complete. A quasi-
complete graph G has only non-self-dependent vertices; there exists a partition I = I1 ∪ · · · ∪ IM
of the set I of vertices of G(S) such that, for all x, y ∈ I, there is an edge from x to y if, and only
if, x and i are not in the same Ii. In particular, quasi complete graphs with M = 2 are complete
bipartite graphs. Moreover, if (S) is quasi-complete, up to a change of variables, for all x ∈ Ii:
Fx =Yj6=i
1 −Xy∈Ij
hy
−1
.
Here is an example of a 2-quasi-complete graph and a 3-quasi-complete graph:
Another particular case is the following: assume that I = I0 and that βx = −1 for all x ∈ I0.
Then, for all x ∈ I, Fx = 1 + hx. Note that G(S) is not connected if I ≥ 2, and this is the only
case where G(S) is not connected. The dilatation of such an SDSE will be called a non-connected
fundamental SDSE. For such an SDSE, the set of indices I admits a partition I = I1 ∪ · · · ∪ IM
(M ≥ 2) and up to a change of variables, for all 1 ≤ i ≤ M , for all x ∈ Ii:
Fx = 1 +Xy∈Ii
hy.
Remark. Note that a dilatation replacing x ∈ K0 ∪ I1 ∪ J1 by a set Jx in a system of
theorem 32 also gives a system of theorem 32. The same remark applies when the dilatation
replaces x ∈ I0, with βx = 0, by a set Jx. So we shall always assume that the dilatation giving a
fundamental SDSE from an SDSE of theorem 32 satisfies Jx = {x} for any x ∈ K0 ∪ I1 ∪ J1 and
for any x ∈ I0 such that βx = 0.
6 Two families of Hopf SDSE
We here first give characterisations of multicyclic and quasi-complete SDSE. We then consider
Hopf SDSE such that any vertex is a descendant of a self-dependent vertex. We prove that such
an SDSE is fundamental. The results of this section will be used to prove the main theorem 14.
6.1 A lemma on non-self-dependent vertices
Lemma 36 Let (S) be a Hopf SDSE and let i ∈ I such that a(i)
such that a(i)
j
6= 0, a(j)
k
6= 0 and a(i)
l
6= 0. Then a(i)
k 6= 0 or a(l)
k 6= 0.
i = 0. Let j, k and l ∈ I
Proof. Let us assume that a(i)
j λ(i,k)
Then, from proposition 16, a(i)
k = 0. As a(i)
j
2 = λ(i,k)
2
q
a q
j = a
i
q
q
6= 0, j 6= k. As a(i)
j a(j)
= a(i)
j,k = 0.
k = 0, a
k + 0; hence, λ(i,k)
2 = a(j)
k .
= a(i)
q∨q
k
k
j
j
i
q
k + a
j
i
q∨qq
i
q
34
Moreover, As a(i)
l
l a(l)
a(i)
k + 0, so λ(i,k)
2 = a(l)
k . Hence, a(l)
k = a(j)
k
6= 0.
6= 0, l 6= k. Then, by proposition 16, a(i)
l λ(i,k)
2 = λ(i,k)
2
q
a q
l = a
i
k + a
l
i
l
q
q
q
q
q∨q
i
k
=
✷
Remark. In other words, if (S) is Hopf, then, in G(S):
i
l
j
k
=⇒ i
j
or
l
/ k
i
l
<<<<<<<<
j
.
k
A special case is given by i = k:
i o
l
j
=⇒ i o
j
.
l
6.2 Symmetric Hopf SDSE
Proposition 37 Let (S) be a Hopf SDSE, such that G(S) is a N -multicycle with N ≥ 3.
Then (S) is a multicyclic SDSE.
Proof. Let I = I1 ∪ · · · ∪ IN be the partition of the set of vertices of the multicycle G(S). As
N ≥ 3, for all i ∈ I, by lemma 28 with i = j:
Fi = 1 + Xi−→j
a(i)
j hj.
Let j, j′ ∈ Im. Then any i ∈ Im−1 is a direct ascendant of j and j′. By proposition 18-3,
Fj = Fj ′. In particular, for k ∈ Im+1, a(j)
. We apply the change of variables sending hk
to 1
a(j)
k
hk if k ∈ Im+1, where j is any element of Im. Then, for any j ∈ Im:
k = a(j ′)
k
Fj = 1 + Xk∈Im+1
hk.
So (S) is multicyclic.
✷
Proposition 38 Let (S) be a Hopf SDSE, such that G(S) is M -quasi-complete graph (M ≥
2). Then (S) is a 2-multicyclic or a quasi-complete SDSE.
Proof. First, let us choose two vertices x → y in G(S). Then y → x in G(S), and by propo-
depends only
y + 0, and a(x)
y = λ(y,y)
, so λ(y,y)
x a(x)
x = a
y
sition 16, λ(y,y)
on y. So, up to a change of variables, we can suppose that all the a(x)
first study three preliminary cases.
a(y)
x = a(y)
y + a
x
y
y 's are equal to 0 or 1. We
a q
q∨q
2
2
2
x
y
y
q
q
q
q
q
First preliminary case. Let us assume that G(S) = 1 ←→ 2. We put:
F1(h2) =
∞Xi=0
aihi
2,
F2(h1) =
bihi
1,
∞Xi=0
with a1 = b1 = 1. Then λ(1,1)
2a
2
3
2
q
q
q∨q
1
= λ(1,1)
3
a
q
q
q
1 = 2a
1
2
1
1
q∨qq
2
q 1
1 , so 2a2b2 = 2a2: a2 = 0 or b2 = 1. Similarly, b2 = 0 or a2 = 1. So a2 = b2 = 0 or 1. In
= 2b2. On the other hand, λ(1,1)
3
a
2
=
2
q
q∨q
1
35
/
/
/
/
/
/
/
o
/
/
o
/
/
O
O
the first case, F1(h2) = 1 + h2 and F2(h1) = 1 + h1. In the second case, let us apply lemma 17-1
with (i1, · · · , in) = (1, 2, 1, 2, · · · ). If n = 2k is even, we obtain λ(1,2)
n = 2 + 2(k − 1) = 2k = n.
If n = 2k + 1 is odd, λ(1,2)
= n for all n ≥ 1. By proposition 19-1,
for all n ≥ 1, an+1 = an. So for all n ≥ 0, an = 1 and F1(h2) = (1 − h2)−1. Similarly,
F2(h1) = (1 − h1)−1.
= 1 + 2k = n. So λ(1,2)
n
n
Second preliminary case. Let us suppose that G(S) is the following graph (which is 3-quasi-
complete):
We put:
1 o
>>>>>>>
3
2@
F1(h2, h3) = 1 + h2 + h3 + a2h2
F2(h1, h3) = 1 + h1 + h3 + b1h2
F3(h1, h2) = 1 + h1 + h2 + c1h2
2 + a3h2
1 + b3h2
1 + c2h2
3 + a′h2h3 + O(h3),
3 + b′h1h3 + O(h3),
2 + c′h1h2 + O(h3).
By restriction, using the first preliminary case, restricting to {1, 2}, {1, 3} and {2, 3},a2 = b1,
a3 = c1 and b3 = c2 and all these elements are in {0, 1}. Moreover, by proposition 16, λ(1,2)
2 =
1
= 1 + a′. Hence,
2a
1 + a′ = 2a2. By symmetry, we obtain 1 + a′ = 2a3, so a2 = a3. Similarly, b1 = b3 and c1 = c2,
so a2 = a3 = b1 = b3 = c1 = c2 = 0 or 1.
2 = 2a2. On the other hand, λ(1,2)
, so λ(1,2)
, so λ(1,2)
2 + a
3
1
3 = a
1
a q
a q
q∨q
1
q∨q
1
2
2
2
2
2
2
3
q
q
q
q
q
q
q
If they are all equal to 0, then a′ = −1. Then λ(3,1)
1 , so λ(3,1)
3
= 1. Moreover,
3
a
2 = a
1
3
q
q
q
q
q
q
q
2
1
3
λ(3,1)
3
a
1 = a
2
3
q
q
q
1
1 , so λ(3,1)
3
q
2
q
q∨q
3
= −1: this is a contradiction, so a2 = a3 = b1 = b3 = c1 = c2 = 1,
and a′ = 1. Similarly, b′ = 1 and c′ = 1. As in the first preliminary case, using lemma 17-1, we
prove that λ(i,j)
n = n if i 6= j for all n ≥ 1, and then that F1(h2, h3) = (1 − h2)−1(1 − h3)−1.
Similarly, F2(h1, h3) = (1 − h1)−1(1 − h3)−1 and F3(h1, h2) = (1 − h1)−1(1 − h2)−1.
Third preliminary case. We now consider the 2-quasi-complete graph with three vertices
1 ←→ 2 ←→ 3. Then I1 = {1, 3} and I2 = {2}. We put:
F2(h1, h3) = 1 + h1 + h3 + a(2,0)h2
1 + a(0,2)h2
3 + a(1,1)h1h3 + O(h3).
Restricting to {1, 2}, by the first preliminary case, we obtain F1(h2) = 1 + h2 or F1(h2) =
(1 − h2)−1.
1. Let us assume that F1(h2) = 1 + h2. Then by the first case, F2(h1, 0) = 1 + h1, so
1 =
2
a(2,0) = 0. Moreover, λ(2,1)
: a(1,1) = 0. Then λ(2,3)
1 = 0, so λ(2,1)
3 = a
2
a q
a q
a q
2
1
3
q
q
q
2
2
, so λ(2,3)
2
3
= a(1,1) = 0, and λ(2,3)
2
q
a q
3 = 2a
2
3
a
1
q
q∨q
2
F2(h1, h3) = 1 + h2 + h3. Restricting to 2 ←→ 3, by the first point, F3(h2) = 1 + h2.
q
q∨q
2
: a(0,2) = 0. As a consequence,
3
2
q
q∨q
2
2. Let us assume that F1(h2) = (1 − h2)−1. Then F2(h1, 0) = (1 − h2)−1 by the first point,
so a(0,2) = 1. By the first preliminary case, this implies that F2(0, h3) = (1 − h3)−1 and
F3(h2) = (1 − h2)−1. Similarly with the first case, we prove that λ(2,i)
n = n if i = 1 or 3 for
all n ≥ 1. By proposition 19-1:
a(m+1,n) =
m + n + 1
m + 1
a(m,n),
a(m,n+1) =
m + n + 1
n + 1
a(m,n).
An easy induction proves that a(m,n) =(cid:0)m+n
m (cid:1) for all m, n, so F2(h1, h3) = (1 − h1 − h3)−1.
36
o
/
/
^
^
@
We separate the proof of the general case into two subcases.
General case, first subcase. M = 2. We put I1 = {x1, · · · , xr} and I2 = {y1, · · · , ys}. For
xi ∈ I1, we put:
Fxp = X(q1,··· ,qs)
a(xp)
(q1,··· ,qs)hq1
y1 · · · hqs
ys.
Restricting to the vertices xp and yq, by the first preliminary case, two cases are possible.
1. a(xp)
yq,yq = 0. Then, by the third preliminary case, restricting to xp, yq and yq′, for all yq, yq′,
a(xp)
yq,yq′ = 0. So:
Fxp = 1 +Xq
hyq .
2. λ(xp,yq)
n
= n for all n ≥ 1. Using proposition 19-1, we obtain:
a(xp)
(q1,··· ,qm+1,··· ,qs) =
1 + q1 + · · · + qs
qm + 1
a(xp)
(q1,··· ,qs).
An easy induction proves:
So:
a(xp)
(q1,··· ,qs) =
(q1 + · · · + qs)!
q1! · · · qs!
.
Fxp = 1 −Xq
hyq!−1
.
A similar result holds for the yq's. So, we prove that for any vertex i of G(S), one of the following
holds:
hj.
1. Fi = 1 + Xi−→j
2. Fi =1 − Xi−→j
−1
.
hj
Moreover, by the first preliminary case, if i and j are related, they satisfy both (a) or both (b).
As the graph is connected, every vertex satisfies (a) or every vertex satisfies (b).
General case, second subcase. M ≥ 3. Let us fix i ∈ G and let us denote y1, · · · , yq its direct
descendants. Restricting to the vertices i and yj, two cases are possible.
1. a(i)
yj,yj = 0. As M ≥ 3, with a good choice of yj ′, we can restrict to the second preliminary
case, and we obtain a(i)
yj ,yj = 1: contradiction. So this case is impossible.
2. λ(x,yj)
n
= n for all n ≥ 1. Using proposition 19-1, we obtain, similarly with the case M = 2,
if i ∈ Ip:
So (S) is quasi-complete.
Definition 39
Fi =Yq6=p
−1
.
1 −Xl∈hq
hl
37
✷
1. Let G be a graph. We shall say that G is symmetric if it has only non-self-dependent
vertices and if, for i 6= j, there is an edge from i to j if, and only if, there is an edge from
j to i.
2. Let (S) be an SDSE. We shall say that (S) is symmetric if G(S) is symmetric.
Theorem 40 Let (S) be a connected symmetric Hopf SDSE. Then (S) is 2-multicyclic or
quasi-complete.
Proof. By proposition 38, it is enough to prove that G(S) is a M -quasi-complete graph,
with M ≥ 2. Let us consider a maximal quasi-complete subgraph G′ of G(S). This exists, as
G(S) contains quasi-complete subgraphs (for example, two related vertices). Let us assume that
G′ 6= G(S). As G(S) is connected, there exists a vertex i ∈ G(S), related to a vertex of G′. Let us
put I ′ = I ′
M be the partition of the set of vertices of G′.
1 ∪ · · · I ′
p. Indeed, let j′ be
q, q 6= p. By lemma 36, j′ is related to i. As G(S) is symmetric,
p, it is related to any vertex of I ′
First, if i is related to a vertex j of I ′
another vertex of I ′
i is related to j′.
p and let k ∈ I ′
Let us assume that i is not related to at least two Ip's. Let us take k, l in G′, in two different
Ip's, not related to i. By the first step, j, k and l are in different Ip's, so are related. By lemma
36, k or l is related to i. As G(S) is symmetric, then i is related to k or l: contradiction. So i is
not related to at most one Ip's.
As a conclusion:
1. If i is related to every Ip's, by the first step i is related to every vertices of G′, so G′ ∪ {i}
is an M + 1-quasi-complete graph, with partition I1 ∪ · · · ∪ IM ∪ {x}: this contradicts the
maximality of G′.
2. If i is related to every Ip's but one, we can suppose up to a reindexation that i is not related
to IM . Then, by the first step, i is related to every vertices of I1 ∪ · · · ∪ IM −1. So G′ ∪ {x}
is an M -quasi-complete graph, with partition I1 ∪ · · · ∪ (IM ∪ {x}): this contradicts the
maximality of G′.
In both cases, this is a contradiction, so G(S) = G′ is quasi-complete.
✷
6.3 Formal series of a self-dependent vertex
Let (S) be a Hopf SDSE, and let us assume that i is a self-dependent vertex of G(S). Up to a
change of variables, we can suppose that a(i)
j = 0 or 1 for all j. In particular, we assume that
a(i)
i = 1.
Lemma 41 Under these hypotheses, i is of level 0 and for all j ∈ I, bj = (1 + δi,j)a(i)
i,j .
Proof. We apply lemma 17-1, with ik = i for all i. We obtain, for all n ≥ 1:
n = a(i)
λ(i,j)
j + (1 + δi,j)(n − 1)
a(i)
i,j
a(i)
i
.
So this proves the assertion.
✷
Remark. So all the descendants of i are also of level 0.
38
Lemma 42 Under the former hypotheses, there exists a partition I = I1 ∪ · · · ∪ IM ∪ J (J
eventually empty), with i ∈ I1, such that the coefficients a(k)
j
are given in the following array:
I2
I3
...
IM
J
Moreover, for all j ∈ I1:
Finally, the coefficients λ(j,k)
n
j \ k I1
I1
I2
1 − β2
1 β1 + 1
...
...
...
1
0
1
...
1
· · ·
I3
· · ·
1
1 − β3
. . .
· · ·
· · ·
· · ·
· · ·
· · ·
. . .
. . .
1
· · ·
IM
β1 + 1
1
...
1
1 − βM
0
J
∗
...
...
...
...
∗
Fj =
MYp=1
fβpXl∈Ip
hl .
are given by λ(j,k)
n = bk(n − 1) + a(j)
k
for all n ≥ 1 with:
I1
k
bk β1 + 1
I2
1
· · ·
· · ·
IM J
1
0
Proof. We can apply lemma 26 with λj = a(i)
j
and µ(l)
j = −a(l)
j + (1 + δi,j) a(i)
i,j . Then
I = I1 ∪ · · · IM ∪ J, such that −a(k)
j + (1 + δi,j) a(i)
i,j is given for all j, k by the array:
j \ k I1
β1
I1
I2
...
IM
J
0
...
0
0
I2
0
β2
. . .
· · ·
· · ·
· · ·
· · ·
. . .
. . .
0
· · ·
IM J
∗
0
...
...
...
0
...
∗
βM
0
We assume that i ∈ I1, without loss of generality. For the row j ∈ J, the result comes from the
following observation: let j, k ∈ I such that a(i)
k 6= 0, then, by proposition 19-1:
j = 0 and a(i)
As a(i)
j = 0, then a(i)
Lemma 26 also gives:
i,j = 0, so a(k)
j = 0.
a(i)
j,k =(cid:16)a(i)
j − a(k)
k = 0.
j + a(i)
i,j(cid:17) a(i)
hl .
fβpXl∈Ip
Fi =
kYp=1
So (1 + δi,j)a(i)
all j, k by the indicated array. We obtain in lemma 41 that:
i,j = β1 + 1 if j ∈ I1, 1 if j ∈ I2 ∪ · · · ∪ IM , and 0 if j ∈ J. So a(k)
j
is given by for
β1 + 1 if k ∈ I1,
1 if k ∈ I2 ∪ · · · ∪ IM ,
0 if k ∈ J.
bk =
As a conclusion, if j ∈ I1, then for all 1 ≤ k ≤ N , a(j)
k = a(i)
k and λ(j,k)
n = λ(i,k)
n
for all n ≥ 1.
✷
By proposition 19, Fi = Fj.
39
6.4 Hopf SDSE generated by self-dependent vertices
Proposition 43 Let (S′) be a Hopf SDSE, and let i be a self-dependent vertex of G(S ′).
Let (S) be the restriction of (S′) to i and all its descendants. Then (S) is fundamental, with
K0 = I1 = J1 = ∅.
Proof. We use the notations of lemma 42. Note that if i, j are in the same Ik, then
for all n ≥ 1, for all k ∈ I. So, by proposition 18-2 the Hopf SDSE formed by i
λ(i,k)
n = λ(j,k)
and its descendant is the dilatation of a system with the following coefficients λ(j,k)
n
:
n
j \ k
1
1
2
3
...
M
(β1 + 1)(n − 1) + 1
(β1 + 1)n
...
...
(β1 + 1)n
2
n
n − β2
3
· · ·
n
n
...
n
n − β3
. . .
· · ·
· · ·
· · ·
· · ·
. . .
. . .
n
M
n
n
...
n
n − βM
with i = 1. We already proved in lemma 42 that:
F1 =
MYj=1
fβj (hj).
If j 6= 1, for all (k1, · · · , kM ):
(k1+1,··· ,kM ) = (β1 + 1)
a(j)
MXl=1
kl + β1 + 1 − (β1 + 1)
kl − k1! a(j)
(k1,··· ,kM )
k1 + 1
MXl=1
= (β1 + 1 + β1k1)
a(j)
(k1,··· ,kM )
k1 + 1
,
a(j)
(k1,··· ,kj+1,··· ,kM ) = MXl=1
kl + 1 − βj −
kl + βjkj! a(j)
(k1,··· ,kM )
kj + 1
MXl=1
= (1 − βj + βjkj)
If l 6= 1 and l 6= j:
a(j)
(k1,··· ,kM )
kj + 1
.
a(j)
(k1,··· ,kl+1,··· ,kM ) = MXl=1
kl −
MXl=1
kl + βlkl! a(j)
(k1,··· ,kM )
kl + 1
= (1 + βlkl)
a(j)
(k1,··· ,kM )
kl + 1
.
So, if j 6= 1:
Fj = f β1
1+β1
((1 + β1)h1)f βj
1−βj
((1 − βj)hj ) Yk6=1,j
fβk(hk).
Let us put I ′
hj −→ 1
1−βj
0 = {j ≥ 2 / βj 6= 1} and J ′
hj for all j ∈ I ′
0:
0 = {j ≥ 2 / βj = 1}. Then, after the change of variables
f1(hj),
F1 = fβ1(h1)Yj∈I ′
0
fβj(cid:18) 1
1 − βj
Fj = f β1
1+β1
Fj = f β1
1+β1
((1 + β1)h1)f βj
1−βj
((1 + β1)h1)Yj∈I ′
0
0
hj(cid:19) Yj∈J ′
(hj) Yj∈I ′
fβj(cid:18) 1
1 − βj
0−{j}
1 − βj
fβj(cid:18) 1
hj(cid:19) Yj∈J ′
0−{j}
hj(cid:19) Yj∈J ′
0
f1(hj) if j ∈ I ′
0,
f1(hj) if j ∈ J ′
0.
40
Putting γj = βj
1−βj
for all j ∈ I0, then, as βj = γj
1+γj
and 1 − βj = 1
1+γj
:
F1 = fβ1(h1)Yj∈I ′
Fj = f β1
1+β1
0
f γj
1+γj
((1 + γj)hj)Yj∈J ′
((1 + β1)h1)fγj (hj ) Yj∈I ′
((1 + β1)h1)Yj∈I ′
f γj
1+γj
0−{j}
0
Fj = f β1
1+β1
0
f γj
1+γj
((1 + γj)hj)Yj∈J ′
0
((1 + γj)hj ) Yj∈J ′
0−{j}
f1(hj ),
f1(hj) if j ∈ I ′
0,
f1(hj ) if j ∈ J ′
0.
So this a fundamental system, with I0 = {1} ∪ I ′
0 and J0 = J ′
0.
✷
Corollary 44 Let (S) be a connected Hopf SDSE such that any vertex of G(S) is the descen-
dant of a self-dependent vertex. Then (S) is fundamental, with K0 = I1 = J1 = ∅.
Proof. Let x be a self-dependent vertex of (S). Then the system formed by x and its
the partition of the set formed by x and
and J (x)
descendants is fundamental. We then put I (x)
its descendants. We separate I (x)
into two parts:
0
0
0
I0,1 =ny ∈ I (x)
0 /βy 6= −1o , I0,2 =ny ∈ I (x)
0 /βy = −1o .
Then, after elimination of an eventual dilatation by restriction, the direct descendants of x ∈ I (x)
0,2
are x, the elements of I (x)
0,1 are the elements of I (x)
and J (x)
; the direct descendants of x ∈ J (x)
0,1 and the elements of J (x)
except x. Let us consider the following cases:
; the direct descendants of x ∈ I (x)
are the elements of I (x)
0,1 and J (x)
0,1
0
0
0
0
1. If there exists a vertex x, such that J (x)
0,1 . We then deduce that (S) is fundamental, with J0 = J (x)
0
0
6= ∅, then, as G(S) is connected, for any self-
. As a consequence, for any self-dependent vertex y,
for any self-dependent
dependent vertex y, J (y)
I (x)
0,1 = I (y)
vertex x.
0 = J (x)
0
2. If for any self-dependent vertex x, J (x)
0,2 6= ∅, then by connectivity of G(S), for any self-dependent vertex y, I (y)
0,1 = {y}, or I (y)
0,2 . Then (S) is a fundamental, with J0 = ∅.
0,2 is empty if y ∈ I (x)
0 = ∅, and if there is a self-dependent vertex x such
0,2 = I (x)
that I (x)
and I (y)
0,2
3. If for any self-dependent vertex x, J (x)
0 = ∅ = I (x)
0,2 . Then by connectivity, I = I (x)
0,1 for any
self-dependent vertex. So (S) is fundamental, with J0 = ∅.
In all cases, (S) is fundamental.
✷
7 The structure theorem of Hopf SDSE
7.1 Connecting vertices
Definition 45 Let (S) be an SDSE and let i ∈ G(S).
1. We denote by G(i)
(S) is the subgraph of G(S) formed by i and all its descendants.
2. The vertex i is a connecting vertex of G(S) if G(i)
(S) − {i} is not connected.
Lemma 46 Let (S) be a Hopf SDSE and let i ∈ G(S) be a connecting vertex. Then (i is the
descendant of a self-dependent vertex) or (i belongs to a symmetric subgraph of G(S)) or (i is not
self-dependent and relates several components of a non-connected fundamental SDSE).
41
(S) − {i} (M ≥ 2). Let xp ∈ Gp be a direct descendant of i for all p. Let x′
Proof. First step. If i is self-dependent, it is a descendant of itself and the conclusion holds.
Let us assume that i is not self-dependent. Let G1, · · · , GM be the connected components of
G(i)
p be a direct
descendant of xp. Then x′
p ∈ Gp. Choosing q 6= j and applying lemma 36, there is an edge from
i to x′
(S) − {i} is a direct descendant of
i. If i is the direct descendant of a vertex j ∈ G(i)
(S) − {i}, then i is included in the symmetric
subgraph i ←→ j of G(i)
p. Iterating this process, we deduce that any vertex of G(i)
(S), so the conclusion holds.
Second step. Let us now assume that i is not the direct descendant of any j ∈ G(i)
Let n ≥ 2, j ∈ Gp, and let i → x2 → · · · → xn in G(i)
is not related to any xl, λ(i,j)
depend on n: we put λ(i,j)
bj = 0 for all j.
n a(i)
l(i,x2,··· ,xn) = aB+
(S) − {i}.
(S), where x2, · · · , xn ∈ Gq, p 6= q. Then, as i
i ( q j l(x2,··· ,xn)), so λ(i,j)
n =
a(i)
j,x2
a(i)
x2
, and λ(i,j)
n
does not
n = λj for all j ∈ G − {i}, n ≥ 2. In other words, i has level ≤ 1, and
Third step.
In order to simplify the writing of the proof, up to a reindexation, we shall
(S) − {0} are the elements of {1, · · · , N }. By a change of
j = 1 for all 1 ≤ j ≤ N . By the second step, we can use lemma
j,k for all j, k in two different connected
for all 1 ≤ j, l ≤ N and λj = a(0)
suppose that i = 0 and the vertices of G(0)
variables, we can suppose that a(0)
27, with µ(l)
j
components of G(0)
j = −a(l)
(S) − {0}.
1. In the first case, we obtain the following values for a(k)
j
and λj:
I1
j \ k
I1 −νβ1
I2
0
I2
...
IM
J
0
...
0
0
−νβ2
. . .
· · ·
· · ·
IM
0
...
0
· · ·
J
−ν
· · ·
...
. . .
...
. . .
0 −νβM −ν
· · ·
0
0
j
λj
I1
ν
· · ·
· · ·
IM J
ν
0
As there are no vertices with no descendants, necessarily ν 6= 0 and βp 6= 0 for all p. For
the same reason, I1 ∪ · · · ∪ IM = ∅ is impossible. If J 6= ∅, then any vertex of J is related
to every vertex of I1 ∪ · · · ∪ IM , so G(0)
(S) − {0} is connected: impossible, as 0 is a connected
vertex. So J = ∅, and 0 connects several totally self-dependent subgraphs.
2. In the second case, we obtain the following values for a(k)
j
and λj:
j \ k
I1
I1 −ν1
I2
0
I2
...
IM
J
0 −ν2
...
. . .
· · ·
0
· · ·
0
· · ·
IM J
· · ·
0
0
...
...
. . .
...
. . .
0
0 −νM 0
· · ·
0
0
j
λj
I1
0
· · ·
· · ·
IM J
0
0
As there are no vertices with no descendants, J = ∅ and νl 6= 0 for all l.
42
Moreover, as bj = 1 + βj = 0 for all j ≥ 1, 0 connects several components of a non-connected
fundamental SDSE.
✷
7.2 Structure of connected Hopf SDSE
Lemma 47 Let (S) be a Hopf SDSE containing a multicycle with set of vertices I = I1 ∪
· · · ∪ IM , Then any non-self-dependent vertex of G(S) has direct descendants in at most one Ik.
Proof. Let us assume that the vertex 0 of G(S) have a direct descendant x ∈ Ik and y ∈ Il
with k 6= l. Then lemma 36 implies that any direct descendant of x is a direct descendant of 0,
so 0 has also a direct descendant in Ik+1. Similarly, 0 has a direct descendant in Il+1. Iterating
this process, 0 has direct descendants in all the Ii's. Up to a restriction, the situation is the
following:
0
1
+WWWWWWWWWWWWWWWWWWWWWWWWWWWWW
'NNNNNNNNNNNNNN
=======
/ · · ·
/ 2
/ 3
/ N
Moreover, for all 1 ≤ i ≤ k, Fi(hi+1) = 1 + hi+1, with the convention hN +1 = h1.
We first assume M ≥ 3. In order to ease the notation, we do not write the index (0) in the
. On the
, so λ(0,2)
q
= 1 + a1,2
a1
2
sequel of the proof. By proposition 16, λ(0,2)
other hand, λ(0,2)
a q
2 = 2 a2,2
a2
, so λ(0,2)
2 = 2a
0
a q
2
2
2
2
q
q
q
1 = a
0
2 + a
1
0
. Hence:
q
1
2
q
q∨q
0
q
q∨q
0
1 +
a1,2
a1
= 2
a2,2
a2
.
3 , so λ(0,2)
3
q
2
2
q
q∨q
0
= 2 a2,2
a2
. On the other hand, λ(0,2)
3
a
2 = a
1
0
q
q
q
2 , so
q
2
1
q
q∨q
0
a1,2
a1
= 2
a2,2
a2
= 1 +
a1,2
a1
.
Moreover, λ(0,2)
3
a
3 = a
2
0
q
q
q
λ(0,2)
3 = a1,2
a1
. Hence:
This is a contradiction.
Let us now prove the result for N = 2. We assume that there exists a Hopf SDSE with the
graph:
1 o
0
>>>>>>>
/ 2
and such that F1 = 1 + h2 and F2 = 1 + h1. We write:
F0 =Xi,j
a(i,j)hi
1hj
2,
with a(1,0) and a(0,1) non-zero. Then λ(0,1)
a(1,1)
λ(0,1)
2
a(0,1)
1 , so λ(0,1)
2 =
2 = a
0
+ a
a q
2
2
1
q
q
q
q
q∨q
0
2
0
q
q
a q
1 = 2a
0
1
1
q
q∨q
0
, so λ(0,1)
2 =
2a(2,0)
a(1,0)
. On the other hand,
+ 1. We obtain:
2a(2,0)
a(1,0)
=
a(1,1)
a(0,1)
+ 1.
43
'
+
/
/
/
/
f
f
o
/
Moreover, λ(0,1)
3
a
2 = a
1
0
q
q
q
so λ(0,1)
3 =
a(1,1)
a(0,1)
. So:
2 +a
q
1
1
q∨qq
0
q
q
q
q
1 , so λ(0,1)
3 =
2
1
0
2a(2,0)
a(1,0)
+1. On the other hand, λ(0,1)
3
a
1 = 2a
2
0
q
q
q
1 ,
q
1
2
q
q∨q
0
a(1,1)
a(0,1)
+ 1 =
2a(2,0)
a(1,0)
=
a(1,1)
a(0,1)
− 1.
This is a contradiction.
✷
Lemma 48 Let (S) be a Hopf SDSE, such that any vertex of G(S) has a direct ascendant.
Let i be a vertex of G(S). Then (i is a descendant of a self-dependent vertex) or (i belongs to a
multicycle of G(S)) or (i belongs to a symmetric subgraph of G(S)).
Proof. Let us first prove that i is the descendant of a vertex of a cycle of G(S). As any
vertex has a direct ascendant, it is possible to define inductively a sequence (xl)l≥0 of vertices of
G(S), such that x0 = i and xl+1 is a direct ascendant of xl for all l. As G(S) is finite, there exists
0 ≤ l < m, such that xl = xm. Then xl ← xl+1 ← · · · ← xm−1 ← xm = xl is a cycle of G(S),
and i is a descendant of any vertex of this cycle.
Let G′ = x1 → · · · → xs → x1 be a cycle such that i is a descendant of a vertex of G′, chosen
with a minimal s. As s is minimal, there are no edges from xl to xm in G(S) if m 6= l + 1, with
the convention xs+1 = x1. The situation is the following:
x1
y1
· · ·
/ xs
/ · · ·
/ yt−1
/ i
Three cases are possible:
1. If s = 1, then i is the descendant of a self-dependent vertex.
2. If s = 2, the situation is the following:
x1 o
x2
y1
/ · · ·
/ yt−1
/ i
By minimality of s, there are no self-dependent vertex in {x1, x2, y1, · · · , yt−1, i}. Applying
repeatedly lemma 36, there is an edge from y1 to x1, then from y2 to y1, · · · , then from i
to yt−1. So i belongs to a symmetric subgraph of G(S).
3. If s ≥ 3, then the subgraph formed by x1, · · · , xs is a multicycle. Let G′ be a maximal
multicycle of length s of G, such that i is a descendant of a vertex of G′. We denote by I ′
the set of vertices of G′. Let us assume that i /∈ G′. There exists x1 → y1 → · · · → yt−1 →
yt = i in G, with t ≤ 1, and x1 ∈ I ′. Up to a reindexation, we can assume that x1 ∈ I ′
1.
By lemma 36, y1 is the direct descendant of any vertex of I1 and the direct ascendant of
any vertex of I3. By lemma 47, y1 is not the direct ascendant of any vertex of I ′
if k 6= 3.
k
So I ′ ∪ {x} = I ′
1
descendant of a vertex of I ′ ∪ {i}: this contradicts the maximality of G′. So i ∈ I ′.
s gives a multicycle of length s, such that i is a
∪ {i}(cid:17) ∪ · · · ∪ I ′
∪(cid:16)I ′
2
By the preceding study of Hopf symmetric SDSE:
✷
Corollary 49 Let (S) be connected Hopf SDSE, such that any vertex of G(S) has a direct
ascendant. Then (any vertex of G(S) is the descendant of a self-dependent vertex, so (S) is
fundamental) or ((S) is quasi-complete, so (S) is fundamental) or ((S) is multicyclic).
44
/
/
/
v
v
/
/
/
o
/
/
/
/
/
Corollary 50 Let (S) be a connected Hopf SDSE. Then there exists a sequence (Gi)0≤i≤k of
subgraphs of G(S), such that:
• The system (S0) associated to the Fi's, i ∈ G0, is fundamental or is multicyclic.
• Gk = G(S).
• For all 0 ≤ i ≤ k − 1, Gi+1 is obtained from Gi by adding a non-self-dependent vertex
without any ascendant in Gi.
If G0 is fundamental, any vertex is of finite level. If G0 is multicyclic, no vertex is of finite level.
Proof. First step. Let us first prove the following (weaker) result: if (S) is a Hopf SDSE,
there exists a sequence (Gi)0≤i≤k of subgraphs of G(S), such that:
• G0 is the disjoint union of several fundamental systems or is multicyclic.
• Gk = G(S).
• For all 0 ≤ i ≤ k − 1, Gi+1 is obtained from Gi by adding a non-self-dependent vertex
without any ascendant in Gi.
Let us proceed by induction on N . If N = 1, then G(S) = G0 is formed by a single vertex which is
necessarily self-dependent, so (S) is fundamental. Let us assume the induction hypothesis at rank
≤ N − 1. If any vertex of G(S) has an ascendant, then by corollary 49, we can take G(S) = G0.
If it is not the case, let us take i being a vertex with no ascendant. The induction hypothesis
can be applied to the components of G(S) − {i}. We complete the sequence (G0, · · · , Gk) given
in this way by Gk+1 = G(S).
As a consequence, the set of descendants of any self-dependent vertex, every symmetric sub-
graph, every multicycle of G(S) is included in G0.
Second step. Let us assume that G(S) is connected. If G0 is connected, then it is fundamental
or multicyclic. If it is not, let us assume that it is not a non-connected abelian fundamental
SDSE. So one of the components H of G0 is not a fundamental abelian SDSE with I = I0.
Then for a good choice of i, the vertex added to Gi−1 to obtain Gi is a connecting vertex,
connecting a subgraph containing H and other subgraphs. By the first step, as it does not
belong to G0, this vertex is not the descendant of a self-dependent vertex and does not belong
to a symmetric subgraph. By construction, it does not connect several components of a non-
connected fundamental SDSE: this is a contradiction with lemma 46. So G0 is of the announced
form.
✷
7.3 Connected Hopf SDSE with a multicycle
Let us precise the structure of connected Hopf SDSE containing a multicycle.
Theorem 51 Let (S) be a connected Hopf SDSE containing a N -multicyclic SDSE. Then I
admits a partition I = I1 ∪ · · · ∪ IN , with the following conditions:
1. If x ∈ Ik, its direct descendants are all in Ik+1.
2. If x and x′ have a common direct ascendant, then they have the same direct descendants.
Moreover, for all x ∈ I:
If x and x′ have a common direct ascendant, then Fx = Fx′. Such an SDSE will be called an
extended multicyclic SDSE.
Fx = 1 + Xx−→y
a(x)
y hy.
45
Proof. We use the notations of corollary 50. We proceed by induction on k.
If k = 0,
(S) is a multicycle and the result is immediate. Let us assume the result at rank k − 1 and
let (S′) be the restriction of (S) to all the vertices except the last one, denoted by x. By the
induction hypothesis, the set of its vertices admits a partition I ′ = I ′
, with the required
conditions. Let us first prove that all the direct descendants of x are in the same I ′
m. Let y ∈ Ik
and z ∈ Il be two direct descendants of x, with k 6= l. Let y′ ∈ Ik+1 be a direct descendant of
y and z′ ∈ Il+1 be a direct descendant of z. Lemma 36 implies that x is a direct ascendant of
z′ and y′, as y can't be a direct ascendant of z′ and z can't be a direct ascendant of y′ because
k 6= l. So we can replace y by y′ and z by z′. Iterating the process, we can assume that y and z
are in the multicycle: this contradicts lemma 47. So the direct descendants of x are all in Im for
a good m. We then take Il = I ′
m−1 ∪ {x} and this proves the first
l
assertion on G(S).
if l 6= m − 1 and Im−1 = I ′
1 ∪· · ·∪I ′
N
We now prove the assertion on Fx. We separate the proof into two subcases. Let us first
for all i.
assume M ≥ 3. There is an oriented path x → xm → · · · → xm+M −1, with xi ∈ I ′
i
Moreover, there is no shorter oriented path from x to xm+M −1. As M ≥ 3, from lemma 28:
Fx = 1 + Xx−→y
a(x)
y hy.
Let us secondly assume that M = 2. Let 1, . . . , p be the direct descendants of x and let 0 be
a direct descendant of 1. Then as 1, . . . , p are in the same part of the partition of I ′, they are
not direct descendants of 1. Let us first restrict to {x, 1, 0}. By proposition 16, λ(x,0)
0 = 0 as
1
x
a(1)
0,0 = 0 by the induction hypothesis, λ(x,0)
= 0. Moreover, 0 = λ(x,0)
0 , so a(x)
1,1 = 0.
= a
a
a
3
3
3
1
1
q
q
q
q∨q
p,p = 0. Let us now take 1 ≤ i < j ≤ p. Then λ(x,i)
i , so a(x)
i,j = 0. As a conclusion, Fx is of the required form.
q
a q
q∨q
2
x
x
1
x
q
q
q
1
i = 0, so λ(x,i)
2 = 0
Similarly, a(x)
and 0 = λ(x,i)
2
2,2 = · · · = a(x)
q
a q
j = a
x
q
q
j
x
q
Proposition 18-3 implies that Fx = Fx′ if x and x′ have a common ascendant, and this implies
the second assertion on G(S).
✷
Remark.
In particular, the vertex added to Gi in order to obtain Gi+1 is an extension
vertex. By proposition 11, any such SDSE is Hopf.
7.4 Connected Hopf SDSE with finite levels
We now prove the following theorem:
Theorem 52 Let (S) be a connected Hopf SDSE, such that any vertex of (S) has a finite level.
Then (S) is obtained from a fundamental system by a finite number (possibly 0) of extensions.
Such an SDSE will be called an extended fundamental SDSE.
Proof. Let (S) be a connected Hopf SDSE, such that any vertex of (S) is of finite level. We
use notations of corollary 50. We shall proceed by induction on k. If k = 0, then S = S0 and
the result is obvious. Let us now assume the result at rank k − 1. By the induction hypothesis,
the system (S′) associated to Gk−1 is a dilatation of a system of theorem 32. Moreover, G is
obtained from Gk−1 by adding a vertex with all its direct descendants in Gk−1. Let us denote
by 0 this vertex. We separate the proof into three cases.
First case. Let us assume that 0 is of level 0. Then all the direct descendants of 0 are of level
0, so are in I0 ∪ J0 ∪ I1, and νx = 1 for all direct descendants of x in Ji with i ∈ I1. Moreover,
for all x ∈ I, λ(0,x)
n = bx(n − 1) + a(0)
x .
Let us take x, y ∈ I. Using proposition 19-1 into two different ways:
a(0)
x,y =(cid:16)by + a(0)
y − a(x)
y (cid:17) a(0)
x =(cid:16)bx + a(0)
x − a(y)
x (cid:17) a(0)
y .
46
So, for all x, y ∈ I:
(cid:16)by − a(x)
y (cid:17) a(0)
x =(cid:16)bx − a(y)
x (cid:17) a(0)
y .
(7)
If x and y are in the same Ii with i ∈ I0 ∪ J0, then by − a(x)
all n ≥ 1, λ(0,x)
on (S′).
y and for
. Hence, up to a restriction, we can assume that there is no dilatations
y = bx − a(y)
x
n = λ(0,y)
6= 0, so a(0)
x = a(0)
n
Let i ∈ I1. If νi 6= 1, we already know that a(0)
i = 0. Let us assume νi = 1 and let us choose
j ∈ I0 ∪ J0 ∪ K0, such that a(i)
i = 0. So
j
a(0)
i = 0 for all i ∈ I1. So the direct descendants of 0 are all in I0 ∪ J0 ∪ K0. Using proposition
19-1 with i ∈ I0 ∪ J0 ∪ K0:
6= bj. Then bi = a(j)
j (cid:17) a(0)
i = 0, so (7) gives (cid:16)bj − a(i)
i pi
bipj − a(i)
a(0)
(p1,··· ,pN )
pi + 1
a(0)
(p1,··· ,pi+1,··· ,pN ) = a(0)
= (cid:16)a(0)
F0 = Yi∈I0
a
i + bi(p1 + · · · + pN ) − Xj∈I0∪J0∪K0−{i}
i (cid:17) pi(cid:17) a(0)
i +(cid:16)bi − a(i)
i hi(cid:17)Yi∈J0
i (cid:16)a(0)
i hi(cid:17) Yi∈K0
i (cid:16)a(0)
(p1,··· ,pN )
pi + 1
f 1
(0)
a
f βi
(0)
.
So:
So (S) is a system of theorem 32, with 0 ∈ K0 ∪ I1.
f0(cid:16)a(0)
i hi(cid:17) .
Second case. Let us assume that 0 is of level 1 and is not an extension vertex. Then all the
direct descendants of 0 are of level 0, so are in I0 ∪ J0 ∪ I1, and νx = 1 for all direct descendants
of x in I1. Moreover, for all i ∈ I, λ(0,i)
First item. Let us assume that a(0)
1 = a(0)
i = 0. Then by proposition 19-1:
n = bi(n − 1) + a(0)
and λ(0,i)
if n ≥ 2.
i
i
a(0)
(p1,··· ,1,··· ,pN ) = a(0)
0 = a(0)
i + bi(p1 + · · · + pN ) −
i −Xj∈I1
a(j)
i pj a(0)
(p1,··· ,0,··· ,pN ).
NXj=1
a(j)
i pj a(0)
(p1,··· ,0,··· ,pN )
If there is a j ∈ I0 ∪ J0 ∪ K0, such that a(0)
j
If it is not the case, as 0 is not an extension vertex, there exists j, k ∈ I1, a(0)
and a(0)
k
obtain:
6= 0, then for (p1, · · · , pN ) = εj, we obtain a(0)
j,k 6= 0 (so a(0)
6= 0). Then, for (p1, · · · , pN ) = εj, (p1, · · · , pN ) = εk, and (p1, · · · , pN ) = εj + εk, we
i = 0.
6= 0
j
a(0)
i + a(j)
i = a(0)
i + a(k)
i = a(0)
i + a(j)
i + a(k)
i = 0.
So a(0)
a(j)
i a(0)
i = 0. So in all cases, a(0)
j = 0. As a conclusion, we proved:
i = 0. Moreover, for (p1, · · · , pN ) = εj for any j ∈ I1, we obtain
1. For all i ∈ I,(cid:16)a(0)
2. Let us put I (0)
i = 0(cid:17) =⇒(cid:16)a(0)
1 =ni ∈ I1 / a(0)
i = 0(cid:17).
6= 0o. Then for i ∈ I, such that a(0)
i
a(j)
i = 0.
i = 0, for all j ∈ I (0)
1 ,
Second item. Let us take i, j ∈ I. Using proposition 19-1 into two different ways:
a(0)
i,j =(cid:16)bj + a(0)
j − a(i)
j (cid:17) a(0)
i =(cid:16)bi + a(0)
i − a(j)
i (cid:17) a(0)
j
.
(8)
47
Let us take i, j ∈ I1. Then a(i)
j = a(j)
i = bi = bj = 0, so (8) gives:
So(cid:16)a(0)
i (cid:17)i∈I1
and(cid:16)a(0)
i (cid:17)i∈I1
ν ∈ K, such that for all i ∈ I1, a(0)
Then bi = a(j)
i
and bj = a(i)
a(0)
j a(0)
i = a(0)
i a(0)
j
.
are colinear. By the first item, we deduce that there exists a scalar
i = νa(0)
i
. Let us now take i, j ∈ I0 ∪ J0 ∪ K0, with i 6= j.
j , so (8) gives:
a(0)
j a(0)
i = a(0)
i a(0)
j
.
So(cid:16)a(0)
i (cid:17)i∈I0∪J0∪K0
and(cid:16)a(0)
i (cid:17)i∈I0∪J0∪K0
are colinear. By the first item, we deduce that there
exists a scalar ν′ ∈ K, such that for all i ∈ I0 ∪ J0 ∪ K0, a(0)
i ∈ I0 ∪ J0 ∪ K0 and j ∈ I1. Then bj = a(i)
words:
j = 0, so νa(0)
i =(cid:16)bi + ν′a(0)
i = ν′a(0)
j a(0)
i
i − a(j)
i (cid:17) a(0)
j
. Let us now take
. In other
∀i ∈ I0 ∪ J0 ∪ K0, ∀j ∈ I1, (ν − ν′)a(0)
i a(0)
j = (bi − a(j)
i )a(0)
j
.
(9)
Third item. Let us assume that I (0)
I0 ∪ J0 ∪ K0. Moreover, if i ∈ I0 ∪ J0 ∪ K0:
1 = ∅. Then all the direct descendants of 0 are in
a(0)
(p1,··· ,pi+1,··· ,pN ) = νa(0)
= (cid:16)νa(0)
i + bi(p1 + · · · + pN ) − Xj∈I0∪J0∪K0−{i}
i +(cid:16)bi − a(i)
i (cid:17) pi(cid:17) a(0)
(p1,··· ,pN )
pi + 1
.
bipj − a(i)
i pi
a(0)
(p1,··· ,pN )
pi + 1
It is then not difficult to show that (S) is a system of theorem 32, with 0 ∈ I1.
Fourth item. Let us assume that ν = ν′. Let j ∈ I1. If νj 6= 1, then we already know that
1 = ∅,
a(0)
j = 0. If νj = 1, then for a good choice of i, bi − a(j)
and the result is proved in the third item.
j = 0: then I (0)
6= 0 in (9), so a(0)
i
Fifth item. Let us assume that I (0)
By (9), for all i ∈ I0 ∪ J0 ∪ K0, a(j)
Fj = Fk for all j, k ∈ I (0)
I (0)
1 . Let us use proposition 19-1. For all i ∈ I0 ∪ J0 ∪ K0, if (p1, · · · , pN ) 6= (0, · · · , 0):
6= ∅. By the preceding item, ν 6= ν′. Let us take j ∈ I (0)
1 .
does not depend of j. As a consequence,
for all i ∈ I0 ∪ J0 ∪ K0, where j is any element of
i = bi − (ν − ν′)a(0)
1 . We put b(0)
i = a(j)
1
i
i
a(0)
(p1,··· ,pi+1,··· ,pN ) =ν′a(0)
For all j ∈ I (0)
1 , if (p1, · · · , pN ) 6= (0, · · · , 0):
i +(cid:16)bi − a(i)
i (cid:17) pi + (ν − ν′)a(0)
i Xj∈I (0)
1
a(0)
(p1,··· ,pN )
pi + 1
.
pj
(p1,··· ,pi+1,··· ,pN ) = νa(0)
a(0)
i
a(0)
(p1,··· ,pN )
pi + 1
.
Let us fix i ∈ I0 ∪ J0 ∪ K0 and j ∈ I (0)
1 . Then:
i + bi − a(i)
a(0)
i,i = (cid:16)ν′a(0)
i,i,j = νa(0)
a(0)
a(0)
i,j = νa(0)
a(0)
i,i,j = νa(0)
i a(0)
i a(0)
i a(0)
j (cid:16)ν′a(0)
j (cid:16)ν′a(0)
,
j
i (cid:17) a(0)
i
,
i + bi − a(i)
i (cid:17) ,
48
i + bi − a(i)
i + (ν − ν′)a(0)
i (cid:17) .
Identifying the two expressions of a(0)
all i ∈ I0 ∪ J0 ∪ K0, a(0)
is impossible. So there is an i ∈ I0 ∪ J0 ∪ K0, such that a(0)
ν′ 6= 0, and we then easily obtain that:
i,i,j, as ν 6= ν′ and a(0)
i = 0, then by the second item, for all j ∈ I (0)
j
i
i (cid:17)2
6= 0, we obtain ν(cid:16)a(0)
1 , a(j)
i = 0, then Fj = 1; this
6= 0. As a consequence, ν = 0. So
= 0. If for
F0 =
f
1
ν′ Yi∈I0
+ Xi∈I (0)
1
βi
−1−βi(cid:16)(cid:16)b(0)
i − 1 − βi(cid:17) hi(cid:17)Yi∈J0
f
1
(0)
b
i
−1(cid:16)(cid:16)b(0)
i − 1(cid:17) hi(cid:17)Yi∈I0
i hi(cid:17)
f0(cid:16)b(0)
(0)
b
i
a(0)
i hi + 1 −
1
ν′ .
So (S) is a system of theorem 32, with 0 ∈ J1.
Third case. 0 is a vertex of level ≥ 2. By proposition 29, it is an extension vertex.
✷
8 Lie algebra and group associated to H(S), associative case
Let us consider a connected Hopf SDSE (S). We now study the pre-Lie algebra g(S) of proposition
21. We separate this study into three cases:
• Associative case: the pre-Lie algebra g(S) is associative. This holds in particular if (S) is
an extended multicyclic SDSE.
• Abelian case: (S) is an extended fundamental, abelian SDSE (see definition 33).
• Non-abelian case: (S) is an extended fundamental, non-abelian SDSE.
We first treat the associative case.
8.1 Characterization of the associative case
Proposition 53 Let (S) be a Hopf SDSE. Then the pre-Lie algebra g(S) is associative if,
and only if, for all i ∈ I:
Fi = 1 + Xi−→j
a(i)
j hj.
Proof. =⇒. Let us assume that ⋆ is associative. Let i, j, k ∈ I, let us show that a(i)
j,k = 0. If
a(i)
j = 0 or a(i)
k = 0, then a(i)
j,k = 0. Let us suppose that a(i)
j
6= 0 and a(i)
k 6= 0. Then:
0 = (fk(1) ⋆ fj(1)) ⋆ fi(1) − fk(1) ⋆ (fj(1) ⋆ fi(1))
2 (cid:17) fi(3)
1 λ(i,k)
1
= λ(i,j)
= (cid:16)λ(j,k)
(cid:16)λ(j,k)
j (cid:16)a(j)
1
= a(i)
λ(i,j)
1 − λ(i,j)
1 − λ(i,k)
k − λ(i,k)
2 (cid:17) fi(3)
2 (cid:17) fi(3).
So λ(i,k)
2 = a(j)
k . Moreover, by proposition 16:
j a(j)
a(i)
k = λ(i,k)
2
q
a q
j = a
i
q
q
q
k + (1 + δj,k)a
j
i
j
q
q∨q
i
= a(i)
j a(j)
k + (1 + δj,k)a(i)
j,k.
k
So a(i)
j,k = 0. As a consequence:
Fi = 1 + Xi−→j
a(i)
j hj.
49
⇐=. Then Xi(n) is a linear span of ladders of weight n for all n ≥ 1, for all i ∈ I. As a
consequence, if x ∈ V ect(Xi(n) / i ∈ I, n ≥ 1), for all f, g ∈ g(S):
(f ⋆ g)(x) = (f ⊗ g) ◦ (π ⊗ π) ◦ ∆(x) = (f ⊗ g) ◦ ∆(x) = f (x′)g(x′′).
So if f, g, h ∈ G(S), for all x ∈ V ect(Xi(n) / i ∈ I, n ≥ 1):
((f ⋆ g) ⋆ h)(x) = f (x′)g(x′′)h(x′′′) = (f ⋆ (g ⋆ h))(x).
So (f ⋆ g) ⋆ h = f ⋆ (g ⋆ h): g(S) is an associative algebra.
✷
Corollary 54 Let (S) be a connected Hopf SDSE. Then g(S) is associative if, and only if
one of the following assertions holds:
1. (S) is an extended multicyclic SDSE.
2. (S) is an extended fundamental SDSE, with:
• For all i ∈ I0, βi = −1.
• J0, K0, I1 and J1 are empty.
If the second assertion holds, then (S) is also an extended fundamental abelian SDSE, and
another interpretation of g(S) can be given; see theorem 70.
8.2 An algebra associated to an oriented graph
Notations. Let G an oriented graph, i, j ∈ G, and n ≥ 1. We shall denote i
oriented path from i to j of length n in G.
n−→ j if there is an
Definition 55 Let G be an oriented graph, with set of vertices denoted by I. The associa-
tive, non-unitary algebra AG is generated by Pi(1), i ∈ I, and the following relations:
• If j is not a direct descendant of i in G, Pj(1)Pi(1) = 0.
• If i1 → i2 → · · · → in and i1 → i′
2 → · · · → i′
n in G, then:
Pin(1) · · · Pi2(1)Pi1 (1) = Pi′
n(1) · · · Pi′
2
(1)Pi1 (1).
Let G be an oriented graph, and let i ∈ I and n ≥ 1. For any oriented path i → i2 → · · · → in
in G, we denote Pi(n) = Pin(1) · · · Pi2(1)Pi(1).
If there is no such an oriented path, we put
Pi(n) = 0. By definition of AG (second family of relations), this does not depend of the choice
of the path.
Lemma 56 Let G be an oriented graph. Then the Pi(n)'s, i ∈ I, n ≥ 1, linearly generate
AG. Moreover, if Pi(m) and Pj(n) are non-zero, then:
Pj(n)Pi(m) =(cid:26) Pi(m + n) if i
0 if not.
m−→ j,
Proof. By the first relation, Pi(n) = Pin(1) · · · Pi2 (1)Pi(1) = 0 if (i, i1, . . . , in) is not an
oriented path in G. So the Pi(n)'s, i ∈ I, n ≥ 1, linearly generate AG.
let us fix Pi(m) = Pim(1) · · · Pi2(1)Pi(1) and Pj(n) = Pjn(1) · · · Pj2(1)Pj (1) both non-zero. If
m−→ j we can choose i2, . . . , im such that i → i2 → · · · → im → j. Then:
i
Pj(n)Pi(m) = Pjn(1) · · · Pj2(1)Pj (1)Pim (1) · · · Pi2(1)Pi(1) = Pi(m + n).
If this is not the case, then j is not a direct descendant of im, so Pj(1)Pim (1) = 0 and
Pj(n)Pi(m) = 0.
✷
50
Proposition 57 Let G be an oriented graph.
1. The following conditions are equivalent:
(a) The family (Pi(n))i∈I,n≥1 is a basis of AG.
(b) All the Pi(n) are non-zero.
(c) The graph G satisfies the following conditions:
• Any vertex of G has a direct descendant.
• If two vertices of G have a common direct ascendant, then they have the same
direct descendants.
(d) The SDSE associated to the following formal series is Hopf:
∀i ∈ I, Fi = 1 +Xi→j
hj.
2. If this holds, then AG is generated by Pi(1), i ∈ I, and the following relations:
• If j is not a direct descendant of i in G, Pj(1)Pi(1) = 0.
• If i → j and i → k in G, then Pj(1)Pi(1) = Pk(1)Pi(1).
The product of AG is given by:
Pj(n)Pi(m) =(cid:26) Pi(m + n) if i
0 if not.
m−→ j,
Moreover, if (S) is the system of condition (d), g(S) is associative and isomorphic to AG.
Proof. 1. (a) =⇒ (b) is obvious.
(b) =⇒ (c). Let us assume (b). Then for all i ∈ I, Pi(2) 6= 0, so there exists a j
such that i → j in G: any vertex of G has a direct descendant. Let us assume i → j and
i → j′ in G. Let k be a direct descendant of j. Then Pi(2) = Pj(1)Pi(i) = Pj ′(1)Pi(1) and
Pi(3) = Pk(1)Pj (1)Pi(1) = Pk(1)Pi(2) 6= 0, so Pk(1)Pi(2) = Pk(1)Pj ′(1)Pi(1) 6= 0. As a conse-
quence, Pk(1)Pj ′ (1) 6= 0 and k is a direct descendant of j′. By symmetry, the direct descendants
of j′ are also direct descendants of j: two direct descendants of a same vertex have the same
direct descendants.
(c) =⇒ (d). Then for all i ∈ I, for all n ≥ 1:
where the sum runs on all oriented paths i → i2 → · · · −→ in in G(S). So:
Xi(n) =X l(i, i2, · · · , in),
∆(Xi(n)) =X nXk=0
l(ik+1, . . . , in) ⊗ l(i, i2, · · · , ik).
If i → i2 · · · → ik → ik+1 and i → i′
and i′
So:
3 are direct descendants of i2 and i′
2 · · · → i′
k → i′
2,. . ., ik+1 and i′
k+1, the second condition on G implies that i3
k+1 are direct descendants of ik and i′
k.
∆(Xi(n)) =
nXk=0 Xi→···→ik,
k
i
−→ik+1,
ik+1→···→in
nXk=0 Xi
−→j
k
Xj(n − k) ⊗ Xi(k).
l(ik+1, . . . , in) ⊗ l(i, i2, · · · , ik) =
51
So (S) is Hopf.
(d) =⇒ (a). Then, for all i ∈ I, for all n ≥ 1:
where the sum runs on all oriented paths i → i2 → · · · −→ in in G(S). By proposition 53, g(S)
is associative. Moreover, it is quite immediate to prove that in g(S):
Xi(n) =X l(i, i2, · · · , in),
• If j is not a direct descendant of i in G, fj(1)fi(1) = 0.
• If i1 → i2 → · · · → in and i1 → i′
2 → · · · → i′
n in G, then:
fin(1) · · · fi2(1)fi1(1) = fi′
n(1) · · · fi′
2
(1)fi1 (1) = fi1(n).
So there is a morphism of algebras from AG to g(S), sending Pi(1) to fi(1). This morphism sends
Pi(n) to fi(n). As the fi(n)'s are linearly independent, so are the Pi(n)'s.
2. Let A′
G be the associative, non-unitary algebra generated by the relations of proposition
57-2. As these relation are immediatly satisfied in AG, there is a unique morphism of algebras:
Φ :(cid:26) A′
G −→ AG
Pi(1) −→ Pi(1).
Let i1 → i2 → · · · → in and i1 → i′
Pi′
of relations defining A′
(1)Pi1 (1) in A′
(1) · · · Pi′
2
k
2 → · · · → i′
n in G. Let us prove that Pik (1) · · · Pi2(1)Pi1 (1) =
G by induction on k. For k = 2, this is implied by the second family
G. Let us assume the result at rank k. Then, both in AG and A′
G:
Pik+1(1)Pik (1) · · · Pi2(1)Pi1 (1) = Pik+1(1)Pi′
k
(1) · · · Pi′
2
(1)Pi1 (1).
This is equal to Pi(k + 1) in AG, so is non-zero. As a consequence, Pik+1(1)Pi′
k → ik+1 in G. By definition of A′
i′
G, Pik+1(1)Pi′
(1) in A′
(1) = Pi′
(1)Pi′
G, so:
k
k
k+1
k
(1) 6= 0 in AG, so
Pik+1(1)Pik (1) · · · Pi2(1)Pi1 (1) = Pi′
k+1
(1)Pi′
k
(1) · · · Pi′
2
(1)Pi1 (1).
So the relations defining AG are also satisfied in A′
G, so there is a morphism of algebras:
Ψ :(cid:26) AG −→ A′
Pi(1) −→ Pi(1).
G
It is clear that Φ and Ψ are inverse isomorphisms of algebras.
✷
Corollary 58 Let (S) a Hopf SDSE. If g(S) is associative, then the graph G(S) satisfies
condition (c) of proposition 57 and g(S) is isomorphic to AG(S).
Proof. First step. Let i, j, k be vertices of G(S) and n ≥ 1 such that i
n−→ k.
Let us prove that Fj = Fk by induction on n. If n = 1, by proposition 18-3, Fj = Fk. If n ≥ 2,
then there exists vertices of G(S) such that:
n−→ j and i
i → j1 → . . . → jn−1 → j,
i → k1 → . . . → kn−1 → k.
n−1−→ j and j1
The case n = 1 implies that Fj1 = Fk1, so j1
Fj = Fk. In other words, if i n−→ j and i n−→ k, then a(j)
l = a(k)
l
for all l ∈ I.
n−1−→ k. By the induction hypothesis,
Second step. Then, for all i ∈ I, for all n ≥ 1:
Xi(n) =X a(i)
i1
· · · a(in−1)
in
l(i, i2, · · · , in),
52
Dually, putting pi(n) = a(i)
by:
n fi(n) for all 1 ≤ i ≤ N , n ≥ 1, the pre-Lie product of g(S) is given
Xj(k) ⊗ Xi(l).
where the sum runs on all oriented paths i → i2 → · · · −→ in in G(S). The first step implies
that a(i)
i1
depends only of i and n: we denote it by a(i)
. . . a(in−1)
n . Then:
in
l
k
−→j
n l(i, i2, · · · , in),
a(i)
n
l a(j)
a(i)
Xi(n) = X a(i)
∆(Xi(n)) = Xk+l=n Xi
fj(n) ⋆ fi(m) =
pj(n) ⋆ pi(m) = ( pi(m + n) if i m−→ j,
a(i)
m+n
a(i)
m a(j)
n
0 otherwise;
0 otherwise.
fi(m + n) if i
m−→ j,
Last step. It is then clear that the associative algebra g(S) is generated by the pi(1), i ∈ I, and
that these elements satisfy the relations defining AG(S). So there is an epimorphism of algebras:
Θ :(cid:26) AG(S) −→ g(S)
Pi(1) −→ pi(1).
This morphism sends Pi(n) to pi(n) for all n ≥ 1. As the pi(n)'s are a basis of AG(S), the Pi(n)'s
are linearly independent in AG(S), so the graph G(S) satisfies condition (c) of proposition 57.
Moreover, Θ is an isomorphism.
✷
8.3 Group of characters
The non-unitary, associative algebra g(S) is graded, with pi(k) homogeneous of degree k for all
1 +Xk≥1
G =
k ≥ 1. Moreover, g(S)(0) = (0). The completion dg(S) is then an associative non-unitary algebra.
We add it a unit and obtain an associative unitary algebra K ⊕dg(S). It is then not difficult to
show that the following set is a subgroup of the units of K ⊕dg(S):
xk ∀k ≥ 1, xk ∈ g(S)(k)
Proposition 59 The group of characters Ch(cid:0)H(S)(cid:1) is isomorphic to G.
in a mapbg from H(S) to K by g((1) + Ker(ε)2) = (0), where ε is the counit of H(S). Moreover,
bg ∈ dg(S). This construction implies a bijection:
Let f1, f2 ∈ Ch(cid:0)H(S)(cid:1). For all x ∈ V , we put ∆(x) = x ⊗ 1 + 1 ⊗ x + x′ ⊗ x′′. As x is a linear
Proof. We put V = V ect(Xi(k)i ∈ I, k ≥ 1). Let g ∈ V ∗. Then g can be uniquely extended
Ω :(cid:26) Ch(cid:0)H(S)(cid:1) −→ G
f −→ 1 + cfV .
span of ladders, x′ ⊗ x′′ ∈ V ⊗ V . So:
.
(f1.f2)(x) = (f1 ⊗ f2) ◦ ∆(x)
= f1(x) + f2(x) + f1(x′)f1(x′′)
= f1V (x) + f2V (x) + f1V (x′)f2V (x′′)
= df1V (x) + df2V (x) + df1V (x′)df2V (x′′)
= df1V (x) + df2V (x) +(cid:16)df1V ⋆ df2V(cid:17) (x).
53
So \(f1.f2)V = df1V + df2V + df1V ⋆ df2V . This implies that Ω is a group isomorphism.
9 Lie algebra and group associated to H(S), non-abelian case
✷
In non-abelian or abelian cases, then any vertex of G(S) is of finite level. By proposition 21, the
constant structures of the pre-Lie product satisfy:
n =( a(i)
λ(i,j)
if n = 1,
j
bj(n − 1) + a(i)
j
if n ≥ level(i) + 1,
's and bj's are scalars.
where the a(i)
j 's, a(j)
i
9.1 Modules over the Faà di Bruno Lie algebra
Let gF dB be the Faà di Bruno Lie algebra. Recall that it has a basis (e(k))k≥1, with bracket
given by:
[e(k), e(l)] = (l − k)e(k + l).
The gF dB-module V0 has a basis (f (k))k≥1, and the action of gF dB is given by:
We can then construct a semi-direct product V M
0 ⊳ gF dB, described in the following proposi-
tion:
e(k).f (l) = lf (k + l).
Proposition 60 Let M ∈ N∗. The Lie algebra V M
0 ⊳ gF dB has a basis:
and its Lie bracket given by:
(cid:16)f (i)(k)(cid:17)1≤i≤M, k≥1
∪ (e(k))k≥1,
[e(k), e(l)] = (l − k)e(k + l),
[e(k), f (i)(l)] = lf (i)(k + l),
[f (i)(k), f (j)(l)] = 0.
We now take g = V ⊕M
0
⊳ gF dB. We define a family of g-modules. Let c ∈ K and υ =
(υ1, . . . , υM ) ∈ K M . The module Wc,υ has a basis (g(k))k≥1, and the action of g is given by:
(cid:26)
e(k).g(l) = (l + c)g(k + l),
f (i)(k).g(l) = υig(k + l).
The semi-direct product is given in the following proposition:
Proposition 61 Let g be the following Lie algebra:
(cid:16)Wc1,υ(1) ⊕ . . . ⊕ WcN ,υ(N)(cid:17) ⊳(cid:0)V M
∪(cid:16)f (i)(k)(cid:17)1≤i≤M, k≥1
(cid:16)g(j)(k)(cid:17)1≤j≤N, k≥1
0 ⊳ gF dB(cid:1) .
∪ (e(k))k≥1,
It has a basis:
and its bracket is given by:
[e(k), e(l)] = (l − k)e(k + l),
[e(k), f (i)(l)] = lf (i)(k + l),
[e(k), g(i)(l)] = (l + c′
i)g(i)(k + l),
[f (i)(k), f (j)(l)] = 0,
[f (i)(k), g(j)(l)] = υ(j)
[g(i)(k), g(j)(l)] = 0.
i g(j)(k + l),
54
Let us take g as in this proposition. We define three families of modules over g:
1. Let ν = (ν1, . . . , νM ) ∈ K M . The module W ′
ν,0 has a basis (h(k))k≥1, and the action of g
2. Let ν = (ν1, . . . , νM ) ∈ K M . The module W ′
ν,1 has a basis (h(k))k≥1, and the action of g
is given by:
is given by:
e(k).g(l) = (l − 1)h(k + l),
f (i)(k).h(1) = νih(k + 1),
f (i)(k).h(l) = 0 if l ≥ 2,
g(i)(k).h(l) = 0.
e(k).h(1) = h(k + 1),
e(k).h(l) = (l − 1)h(k + l) if l ≥ 2,
f (i)(k).h(1) = νih(k + 1),
f (i)(k).h(l) = 0 if l ≥ 2,
g(i)(k).h(l) = 0.
e(k).h(l) = (l + c)h(k + l),
f (i)(k).h(l) = νih(k + l),
g(i)(k).h(1) = µih(k + 1),
g(i)(k).h(l) = 0 if l ≥ 2.
3. Let c ∈ K, ν = (ν1, . . . , νM ) ∈ K M , µ = (µ1, . . . , µN ) ∈ K N . The module W ′′
c,ν,µ has a
basis (h(k))k≥1, and the action of g is given by:
9.2 Description of the Lie algebra
Theorem 62 Let us consider a connected, fundamental non-abelian SDSE. Then g(S) has
the following form:
g(S) ≈ W ⊳(cid:16)(cid:16)Wc1,υ(1) ⊕ . . . ⊕ WcN ,υ(N)(cid:17) ⊳(cid:0)V M
0 ⊳ gF dB(cid:1)(cid:17) ,
where W is a direct sum of W ′
ν,0, W ′
ν,1 and W ′′
c,ν,µ.
Proof. First step. We first consider a Hopf SDSE (S), dilatation of a system of theorem
32, such that I = I0 ∪ J0 ∪ K0. The set J of the vertices of G(S) admits a partition J =
(Jx)x∈I0 ∪ (Jx)x∈J0 ∪ (Jx)x∈K0. We put:
A = {j ∈ J / bj 6= 0}, B = {j ∈ J / bj = 0}.
In other terms, i ∈ A if, and only if, (i ∈ Jx, with x ∈ I0 such that bx 6= −1) or (i ∈ Jx, with
x ∈ J0). As we are in the non-abelian case, A 6= ∅. Let us choose ix ∈ Jx for all x ∈ I, and
ix0 ∈ A. In order to enlighten the notations, we put i0 = ix0. We define, for all k ≥ 1:
pi0(k) =
pi(k) =
fi0(k),
(fi(k) − fi0(k)) if i ∈ Jx0 − {i0},
pix(k) =
1
bx0
pix(k) = fi(k) if x ∈ B,
fi(k) −
fi0(k) if x 6= x0 and x ∈ A,
1
bx0
1
bx0
1
bx
1
bx
pi(k) =
(fi(k) − fix(k)) if i ∈ Jx − {ix}, x 6= x0 and x ∈ A,
pi(k) = fi(k) − fix(k) if i ∈ Jx − {ix}, x ∈ B.
55
Then direct computations show that the Lie bracket of g(S) is given in the following way: for all
k, l ≥ 1,
• [pi0(k), pi0 (l)] = (l − k)pi0(k + l).
lpi(k + l) if i /∈ Jx0.
• For all i ∈ I, [pi0(k), pi(l)] =(cid:26) (l + dx0)pi(k + l) if i ∈ Jx0 − {i0},
• For all i ∈ Jx0 − {i0}, for all x 6= x0, [pix(k), pi(l)] =(cid:26) −dx0pi(k + l) if x ∈ A,
• For all x, x′ ∈ I − {x0}, i ∈ Jx′ − {ix′}, [pix(k), pi(l)] =(cid:26) 0 if x 6= x′,
• For all x, x′ ∈ I − {x0}, [pix(k), pix′ (l)] = 0.
dxpi(k + l) if x = x′.
0 if x ∈ B.
• For all x, x′ ∈ I − {x0}, i ∈ Jx − {ix}, j ∈ Jx′ − {ix′}, [pi(k), pj(l)] = 0.
We used the following notations:
if x ∈ I0, βx 6= −1,
−βx
1 + βx
1 if x ∈ I0, βx = −1,
−1 if x ∈ J0,
0 if x ∈ K0.
dx =
So the Lie algebra g(S) is isomorphic to:
W
Jx0−1
dx0 ,(−dx0 ,··· ,−dx0 ,0,··· ,0) ⊕ Mx∈I−{x0}
A basis adapted to this decomposition is:
W Ix−1
0,(0,··· ,0,dx,0,··· ,0) ⊳(cid:16)V I−1
0
⊳ gF dB(cid:17) .
(pi(k))i∈Jx0 −{i0},k≥1 ∪ [x∈I−{x0}
(pi(k))i∈Jx−{ix},k≥1 ∪ [x∈I−{x0}
(pix (k))k≥1 ∪ (pi0(k))k≥1.
Second step. We now assume that I1 6= ∅. Then the descendants of j ∈ I1 form a system of
the first step, so:
g(S) = WI1 ⊳ g(S0),
where WI1 = V ect(fj(k) / j ∈ I1, k ≥ 1} and (S0) is a restriction of (S) as in the first step. Let
us fix j ∈ I1 and let us consider the g(S0)-module Wj = V ect(fj(k) / k ≥ 1). With the notations
of the preceding step:
a(j)
i0
a(j)
i0
• [pi0(k), fj(l)] =(cid:18)l − 1 +
• [pi0(k), fj(l)] =(cid:18)l − 1 + νj
• [pix(k), fj (l)] =(cid:18) a(j)
• [pix(k), fj (l)] = νj(cid:18) a(j)
bx0(cid:19) fj(k + l) if l = 1.
bx0(cid:19) fj(k + l) if l ≥ 2.
bx0(cid:19) fj(k + l) if l = 1, x ∈ A.
bx0(cid:19) fj(k + l) if l ≥ 2, x ∈ A.
a(j)
i0
a(j)
i0
ix
bx
−
ix
bx
−
• [pix(k), fj (l)] = a(j)
ix
fj(k + l) if l = 1, x ∈ B.
• [pix(k), fj (l)] = νja(j)
ix
fj(k + l) if l ≥ 2, x ∈ B.
56
• [pi(x), fj(l)] = 0 if i is not a ix.
If νj 6= 0, we put pj(k) = fj(k) if k ≥ 2 and pj(1) = νjfj(1). Then, for all l:
• [pi0(k), pj (l)] =(cid:18)l − 1 + νj
• [pix(k), pj(l)] = νj(cid:18) a(j)
ix
bx
−
a(j)
i0
bx0(cid:19) pj(k + l).
bx0(cid:19) pj(k + l) if x ∈ A.
a(j)
i0
• [pix(k), pj(l)] = νja(j)
ix
pj(k + l) if x ∈ B.
• [pi(x), pj(l)] = 0 if i is not a ix.
So Wj is a module Wc,υ. If νj = 0 and a(j)
i0
bx0
a(j)
i0
fj(1). Then:
• [pi0(k), pj (l)] = pj(k + l) if l = 1.
6= 0, we put pj(k) = fj(k) if k ≥ 2 and pj(1) =
• [pi0(k), pj (l)] = (l − 1)pj(k + l) if l ≥ 2.
• [pix(k), fj(l)] =(cid:18) a(j)
ix
bx
−
a(j)
i0
bx0(cid:19) fj(k + l) if l = 1, x ∈ A.
• [pix(k), fj(l)] = 0 if l ≥ 2, x ∈ A.
• [pix(k), fj(l)] = a(j)
ix
fj(k + l) if l = 1, x ∈ B.
• [pix(k), fj(l)] = 0 if l ≥ 2, x ∈ B.
• [pi(x), pj(l)] = 0 if i is not a ix.
So Wj is a module W ′
ν,1. If νj = 0 and a(j)
i0
= 0, we put pj(k) = fj(k) for all k ≥ 1. Then:
• [pi0(k), pj (l)] = (l − 1)pj(k + l).
• [pix(k), fj(l)] =(cid:18) a(j)
ix
bx
−
a(j)
i0
bx0(cid:19) fj(k + l) if l = 1, x ∈ A.
• [pix(k), fj(l)] = 0 if l ≥ 2, x ∈ A.
• [pix(k), fj(l)] = a(j)
ix
fj(k + l) if l = 1, x ∈ B.
• [pix(k), fj(l)] = 0 if l ≥ 2, x ∈ B.
• [pi(x), pj(l)] = 0 if i is not a ix.
So Wj is a module W ′
ν,0.
Last step. We now consider vertices in J1. If j ∈ J1, then its descendants are vertices of the
first step and i elements of I1 such that νi = 1. As before:
g(S) = WJ1 ⊳ g(S1),
where WJ1 = V ect(fj(k) / j ∈ J1, k ≥ 1} and (S1) is a restriction of (S) as in the second step.
Let us fix j ∈ J1 and let us consider the g(S1)-module Wj = V ect(fj(k) / k ≥ 1). As νj 6= 0,
putting pj(k) = fj(k) if k ≥ 2 and pj(1) = νjfj(1):
• [pi0(k), pj (l)] =(cid:18)l − 1 + νj
a(j)
i0
bx0(cid:19) pj(k + l).
57
• [pix(k), pj (l)] = νj(cid:18) a(j)
ix
bx
−
a(j)
i0
bx0(cid:19) pj(k + l) if x ∈ A.
• [pix(k), pj (l)] = νja(j)
ix
pj(k + l) if x ∈ B.
• [pi(k), pj (l)] = νja(j)
i pj(k + l) if l = 1, i ∈ I1, with νi = 1.
• [pi(k), pj (l)] = 0 if l ≥ 2, i ∈ I1.
• [pi(x), pj (l)] = 0 if i /∈ I1 and is not a ix.
So Wj is a module W ′′
c,ν,µ.
✷
Theorem 63 Let (S) be a connected, extended, fundamental, non-abelian SDSE. Then the
Lie algebra g(S) is of the form:
gm ⊳ (gm−1 ⊳ (· · · g2 ⊳ (g1 ⊳ g0) · · · ),
where g0 is the Lie algebra associated to the restriction of (S) to the vertices which are not
extension vertices (so g0 is described in theorem 62) and, for j ≥ 1, gj is an abelian (gj−1 ⊳
(· · · g2 ⊳ (g1 ⊳ g0) · · · )-module having a basis (h(j)(k))k≥1.
Proof. The Lie algebra gj is the Lie algebra V ect(fxj (k) / k ≥ 1), where J2 = {x1, . . . , xm},
✷
with the notations of theorem 14.
9.3 Associated group
Let us now consider the character group Ch(cid:0)H(S)(cid:1) of H(S). In the preceding cases, g(S) contains
a sub-Lie algebra isomorphic to the Faà di Bruno Lie algebra, so Ch(cid:0)H(S)(cid:1) contains a subgroup
isomorphic to the Faà di Bruno subgroup:
GF dB = {x + a1x2 + a2x3 + · · · ∀i, ai ∈ K},
with the product defined by A(x).B(x) = B ◦ A(x). Moreover, each modules earlier defined on
gF dB corresponds to a module over GF dB by exponentiation:
Definition 64
1. The module V0 is isomorphic to yK[[y]] as a vector space, and the action of GF dB is given
by:
A(x).P (y) = P ◦ A(y).
2. Let G =(cid:16)V⊕M
as a vector space, and the action of G is given by:
0 (cid:17) ⋊ GF dB. Let c ∈ K, and υ = (υ1, · · · , υM ) ∈ K M . Then Wc,υ is zK[[z]]
(P1(y), · · · , PM (y), A(x)).Q(z) = exp MXi=1
υiPi(z)!(cid:18) A(z)
z (cid:19)c
Q ◦ A(z).
3. Let us consider the following semi-direct product:
G =(cid:16)W
c1,ε(1) ⊕ · · · ⊕ W
cN ,ε(N)(cid:17) ⊳(cid:16)V⊕M
0
⊳ GF dB(cid:17) .
58
(a) Let ν = (ν1, · · · , νM ) ∈ K M . Then W′
ν,0 is tK[[t]] as a vector space, and for all
X = (Q1(z), · · · , QN (z), P1(y), · · · , PM (y), A(x)) ∈ G:
X.t = 1 +
X.R(t) = (cid:18) t
νiPi(t)! t,
MXi=1
A(t)(cid:19) R ◦ A(t),
for all R(t) ∈ t2K[[t]].
(b) Let ν = (ν1, · · · , νM ) ∈ K M . Then W′
ν,1 is tK[[t]] as a vector space, and for all
X = (Q1(z), · · · , QN (z), P1(y), · · · , PM (y), A(x)) ∈ G:
X.t = 1 +
X.R(t) = (cid:18) t
MXi=1
A(t)(cid:19) R ◦ A(t),
νiPi(t)!(cid:18)t + t ln(cid:18) A(t)
t (cid:19)(cid:19) ,
for all R(t) ∈ t2K[[t]].
(c) Let c ∈ K, ν = (ν1, · · · , νM ) ∈ K M , µ = (µ1, . . . , µN ) ∈ K N . Then W′′
c,ν,µ is tK[[t]]
as a vector space, and for all X = (Q1(z), · · · , QN (z), P1(y), · · · , PM (y), A(x)) ∈ G:
X.t = (cid:18) A(t)
t (cid:19)c
X.R(t) = (cid:18) t
A(t)(cid:19)c
exp MXi=1
exp MXi=1
µiPi(t)! 1 +
MXi=1
µiPi(t)! R ◦ A(t),
µiQi(t)! A(t),
for all R(t) ∈ t2K[[t]].
Direct computations prove that they are indeed modules.
Theorem 65 Let (S) be a connected Hopf SDSE in the non-abelian, fundamental case. Then
the group Ch(cid:0)H(S)(cid:1) is of the form:
Gm ⋊ (Gm−1 ⋊ (· · · G2 ⋊ (G1 ⋊ G0) · · · ),
where G0 is a semi-direct product of the form:
G0 = W′ ⋊ (W ⋊ (V ⋊ GF dB)),
where V is a direct sum of modules V0, W a direct sum of modules Wc,υ, and W′ a direct sum
of modules W′
c,ν,µ. Moreover, for all m ≥ 1, Gm = (tK[[t]], +) as a group.
ν,1 and W′′
ν,0, W′
Proof. The group Ch(cid:0)H(S)(cid:1) is isomorphic to the group of characters of U (g)∗, where g is
described in theorem 63. This implies that this group has a structure of semi-direct product as
described in theorem 65. Let us consider the Hopf algebra H of coordinates of G0. It is a graded
Hopf algebra, and direct computations prove that its graded dual is the enveloping algebra of g0
of theorem 63. So H is isomorphic to H(S0).
✷
10 Lie algebra and group associated to H(S), abelian case
We now treat the abelian case. Recall that in this case, J0 = K0 = ∅ and, for all i ∈ I0, βi = −1.
59
10.1 Modules over an abelian Lie algebra
Let gab be an abelian Lie algebra, with basis(cid:0)e(i)(k)(cid:1)1≤i≤M,k≥1. We define a family of modules
over this Lie algebra:
Definition 66 Let υ = (υ1, · · · , υM ) ∈ K M . Then Vυ has a basis (f (k))k≥1, and the action
of gab is given by:
e(i)(k).f (l) = υif (k + l).
We can then describe the semi-direct product:
Proposition 67 Let us consider the following Lie algebra:
g = NMi=1
Vυ(i)! ⊳ gab.
It has a basis:
(e(i)(k))1≤i≤M,k≥1 ∪ (f (i)(k))1≤i≤N,k≥1,
and the Lie bracket is given by:
[e(i)(k), e(j)(l)] = 0,
[e(i)(k), f (j)(l)] = υ(j)
[f (i)(k), f (j)(l)] = 0.
i f (j)(k + l),
We now define two families of modules over such a Lie algebra.
Definition 68 Let g be a Lie algebra of proposition 67.
1. Let ν = (ν1, . . . , νM ) ∈ K M . The module Wν has a basis (g(k))k≥1, and the action of g is
2. Let ν = (ν1, . . . , νM ) ∈ K M and µ = (µ1, . . . , µN ) ∈ K N , such that for all 1 ≤ i ≤ M , for
ν,µ has a basis (g(k))k≥1, and the action
e(i)(k).g(1) = νig(k + 1),
e(i)(k).g(l) = 0 if l ≥ 2,
f (i)(k).g(l) = 0.
given by:
all 1 ≤ j ≤ N , µj(cid:16)νi − υ(j)
i (cid:17) = 0. The module W ′
Remark. The condition µj(cid:16)νi − υ(j)
of g is given by:
e(i)(k).g(l) = νig(k + l),
f (j)(k).g(1) = µjg(k + 1),
f (j)(k).g(l) = 0 if l ≥ 2.
i (cid:17) = 0 is necessary for W ′
ν,µ to be a g-module. Indeed:
[e(i)(k), f (j)(l)].g(1) = υ(j)
i µjg(k + l + 1),
e(i)(k).(cid:16)f (j)(l).g(1)(cid:17) − f (j)(l).(cid:16)e(i)(k).g(1)(cid:17) = µjνig(k + l + 1).
60
10.2 Description of the Lie algebra
We here consider a connected Hopf SDSE (S) in the abelian case.
Theorem 69 Let us consider a Hopf SDSE of abelian fundamental type, with no extension
vertices. Then g(S) has the following form:
g(S) ≈ W ⊳ ((Vυ(1) ⊕ . . . ⊕ Vυ(N) ) ⊳ gab) ,
where W is a direct sum of Wν and W ′
ν,µ.
Proof. First step. We first consider a Hopf SDSE such that:
I = [x∈I0
Jx.
For all x ∈ I0, let us fix ix ∈ Jx. We put pix(k) = fix(k) and pi(k) = fi(k)−fix(k) if i ∈ Jx −{ix}.
Then direct computations show that:
• [pix(k), pix′ (l)] = 0.
• [pix(k), pj(l)] = δx,x′pj(k + l) if j ∈ Jx′ − {ix′}.
• [pi(k), pj(l)] = 0 if i, j are not ix's.
So:
g(S) ≈Mx∈I0
V ⊕Jx−1
(0,...,0,1,0,...,0) ⊳ gab,
where gab = V ect(pix(k) / x ∈ I0, k ≥ 1).
Second step. We now assume that I1 6= ∅. Then the descendants of j ∈ I1 form a system as
in the first step, so:
g(S) = WI1 ⊳ g(S0),
where WI1 = V ect(fj(k) / j ∈ I1, k ≥ 1} and (S0) is the restriction of (S) to the regular vertices.
Let us fix j ∈ I1 and let us consider the g(S0)-module Wj = V ect(fj(k) / k ≥ 1). With the
notations of the preceding step:
• [pix(k), fj(l)] = a(j)
ix
fj(k + l) if l = 1.
• [pix(k), fj(l)] = νja(j)
ix
fj(k + l) if l ≥ 2.
• [pi(x), fj(l)] = 0 if i is not a ix.
If νj 6= 0, we put pj(k) = fj(k) if k ≥ 2 and pj(1) = νjfj(1). Then, for all l:
• [pix(k), fj(l)] = νja(j)
ix
fj(k + l).
• [pi(x), fj(l)] = 0 if i is not a ix.
So Wj is a module Vυ. If νj = 0, we put pj(k) = fj(k) for all k ≥ 1. Then:
• [pix(k), fj(l)] = a(j)
ix
fj(k + l) if l = 1.
• [pix(k), fj(l)] = 0 if l ≥ 2.
• [pi(x), fj(l)] = 0 if i is not a ix.
61
So Wj is a module Wν.
Last step. We now consider vertices in J1. If j ∈ J1, then its descendants are vertices of the
first step and vertices in I1 such that νi = 1. As before:
g(S) = WJ1 ⊳ g(S1),
where WJ1 = V ect(fj(k) / j ∈ J1, k ≥ 1} and (S1) is the restriction of (S) to the regular
vertices and the vertices of I1. Let us fix j ∈ J1 and let us consider the g(S1)-module Wj =
V ect(fj(k) / k ≥ 1). As νj 6= 0, putting pj(k) = fj(k) if k ≥ 2 and pj(1) = νjfj(1):
• [pix(k), pj (l)] = νja(j)
ix
pj(k + l).
• [pi(k), pj (l)] = 0 if i ∈ Jx − {ix}.
• [pi(k), pj (l)] = νja(j)
i pj(k + l) if l = 1 and i ∈ I1.
• [pi(k), pj (l)] = 0 if l ≥ 2 and i ∈ I1.
So Wj is a module W ′
ν,µ.
✷
Theorem 70 Let (S) be a connected Hopf SDSE in the non-abelian, fundamental case. Then
the Lie algebra g(S) is of the form:
gm ⊳ (gm−1 ⊳ (· · · g2 ⊳ (g1 ⊳ g0) · · · ),
where g0 is the Lie algebra associated to the restriction of (S) to the non-extension vertices (so
is described in theorem 69), and, for j ≥ 1, gj is an abelian (gj−1 ⊳ (· · · g2 ⊳ (g1 ⊳ g0) · · · )-module
having a basis (h(j)(k))k≥1.
Proof. Similar with the proof of theorem 62.
✷
10.3 Associated group
Let us now consider the character group Ch(cid:0)H(S)(cid:1) of H(S). In the preceding cases, g(S) contains
an abelian sub-Lie algebra gab, so Ch(cid:0)H(S)(cid:1) contains a subgroup isomorphic to the group:
, ∀1 ≤ i ≤ M, ∀k ≥ 1, a(i)
Gab =(cid:26)(cid:16)a(i)
1 x + a(i)
2 x2 + · · ·(cid:17)1≤i≤M
k ∈ K(cid:27) ,
with the product defined by (A(i)(x))i∈I .(B(i)(x))i∈I = (A(i)(x) + B(i)(x) + A(i)(x)B(i)(x))i∈I .
Note that Gab is isomorphic to the following subgroup of the following group of the units of the
ring K[[x]]M :
G1 =( 1+xf1(x)
1+xfM (x)! f1(x), . . . , fM (x) ∈ K[[x]]) .
...
The isomorphism is given by:
1 x + a(i)
(cid:16)a(i)
2 x2 + · · ·(cid:17)1≤i≤M
Gab −→ G1
−→
1+a(1)
1 x+a(1)
2 x2+...
...
1+a(M )
1
x+a(M )
2
x2+...
.
Moreover, each modules earlier defined on gab corresponds to a module over Gab by exponen-
tiation, as explained in the following definition:
Definition 71
62
1. Let υ = (υ1, . . . , υM ) ∈ K M . The module Vυ is isomorphic to yK[[y]] as a vector space,
and the action of Gab is given by:
2. Let us consider the following semi-direct product:
υiA(i)(y)! P (y).
(A(i)(x))1≤i≤M .P (y) = exp MXi=1
υ(i)! ⊳ Gab.
G = NMi=1
V
(a) Let ν = (ν1, . . . , νM ) ∈ K M . The module Wν is zK[[z]] as a vector space, and the ac-
tion of G is given in the following way: for all X = (P1(y), . . . , PN (y), A1(x), . . . , Am(x)) ∈
G,
X.z = 1 +
νiAi(z)! z,
MXi=1
X.z2R(z) = z2R(z),
for all R(z) ∈ K[[z]].
(b) Let ν = (ν1, . . . , νM ) ∈ K M and µ = (µ1, . . . , µN ) ∈ K N , such that for all 1 ≤
ν,µ is zK[[z]] as
for all X =
a vector space, and the action of G is given in the following way:
(P1(y), . . . , PN (y), A1(x), . . . , Am(x)) ∈ G,
i ≤ M , for all 1 ≤ j ≤ N , µj(cid:16)νi − υ(j)
X.z = exp MXi=1
X.z2R(z) = exp MXi=1
i (cid:17) = 0. The module W′
µiPi(z)! z,
νiAi(z)! 1 +
NXi=1
νiAi(z)! z2R(z),
for all R(z) ∈ K[[z]].
Direct computations prove that they are indeed modules. The condition µj(cid:16)νi − υ(j)
ν,µ to be a module. Indeed:
i (cid:17) = 0
is necessary for W′
Ai(x).(Pj (y).z) = (exp(νiAi(z)) + µjexp(νiAi(z))Pj (z)) z,
i Ai(y))Pj (y)Ai(x)(cid:17) .z
(Ai(x)Pj(y)).z = (cid:16)exp(υ(j)
i Ai(z))Pj (z)(cid:17) z + (exp(νiAi(z)) − 1)z
= (cid:16)1 + exp(υ(j)
= (cid:16)exp(νiAi(z)) + µjexp(υ(j)
i Ai(z))Pj (z)(cid:17) z.
Theorem 72 Let (S) be a connected Hopf SDSE in the abelian case. Then the group
Ch(cid:0)H(S)(cid:1) is of the form:
GN ⋊ (GN −1 ⋊ (· · · G2 ⋊ (G1 ⋊ G0) · · · ),
where G0 is a semi-direct product of the form:
G0 = W ⋊ (V ⋊ Gab),
where V is a direct sum of modules Vυ, and W a direct sum of modules Wν and W′
for all m ≥ 1, Gm = (tK[[t]], +) as a group.
ν,µ. Moreover,
Proof. Similar as the proof of theorem 65.
✷
63
11 Appendix: dilatation of a pre-Lie algebra
Let (S) be a Hopf SDSE with set of indices I. We choose a set J and consider the disjoint union
I ′ of several copies Ji of J indexed by I. The Lie algebra g(S) has a basis (fi(k))i∈I, k≥1 and the
Lie bracket is given by:
[fi(k), fj (l)] = λ(j,i)
l
fj(k + l) − λ(i,j)
k
fi(k + l).
Let (S′) be the dilatation of (S) with set of indices I ′. Then the Lie algebra g(S ′) has a basis
(fi(k))i∈J, k≥1 and the Lie bracket is given in the following way: for all x ∈ Ji, y ∈ Jj,
[fi(k), fj(l)] = λ(j,i)
l
fy(k + l) − λ(i,j)
k
fx(k + l).
We shall say that g(S ′) is a dilatation of g(S). We prove in this section that this construction is
equivalent to give a pre-Lie product of g(S).
11.1 Dilatation of a pre-Lie algebra
Definition 73 [4] A permutative, associative algebra is a couple (A, ·) where A is a vector
space and · is a bilinear associative (non-unitary) product on A such that for all a, b, c ∈ A:
abc = bac.
Proposition 74 Let (A, ·) be a vector space with a bilinear product. For any pre-Lie algebra
(g, ⋆), we define a product on g ⊗ A by:
(x ⊗ a) ⋆ (y ⊗ b) = (x ⋆ y) ⊗ (ab).
Then g ⊗ A is pre-Lie for any pre-Lie algebra g if, and only if, A is permutative, associative.
Proof. ⇐=. Let g be a pre-Lie algebra, and let x, y, z ∈ g, a, b, c ∈ A. Then:
((x ⊗ a) ⋆ (y ⊗ b)) ⋆ (z ⊗ c) − (x ⊗ a) ⋆ ((y ⊗ b) ⋆ (z ⊗ c))
= ((x ⋆ y) ⋆ z − x ⋆ (y ⋆ z)) ⊗ abc
= ((y ⋆ x) ⋆ z − y ⋆ (x ⋆ z)) ⊗ bac
= ((y ⊗ b) ⋆ (x ⊗ a)) ⋆ (z ⊗ c) − (y ⊗ b) ⋆ ((x ⊗ a) ⋆ (z ⊗ c)).
So g ⊗ A is pre-Lie.
=⇒. Let us assume that g ⊗ A is pre-Lie for any pre-Lie algebra g. Let us choose g as
D), with D containing three distinct elements i, j, k. Then, for any
the pre-Lie algebra P rim(H∗
a, b, c ∈ A:
((f q i ⊗ a) ⋆ (f q j ⊗ b)) ⋆ (f q k ⊗ c) − (f q i ⊗ a) ⋆ ((f q j ⊗ b) ⋆ (f q k ⊗ c))
i ⊗ (ab)c −(cid:18)f
j
k
+ f
j
q
q
q
i
q
q∨q
k
i ⊗ ((ab)c − a(bc)) − f
= f
= f
q
q
q
q
q
q
j
k
q
q
i
k
i (cid:19) ⊗ a(bc)
j
k
⊗ a(bc)
i
j
q
q∨q
k
i
j
q
q∨q
k
= ((f q j ⊗ b) ⋆ (f q i ⊗ a)) ⋆ (f q k ⊗ c) − (f q j ⊗ b) ⋆ ((f q i ⊗ a) ⋆ (f q k ⊗ c))
= f
j ⊗ ((ba)c − b(ac)) − f
⊗ b(ac).
q
So:
f
q
q
q
q
j
k
i
j
k
i ⊗ ((ab)c − a(bc)) − f
⊗ a(bc) = f
q
q
q
i
j
q∨qq
k
j ⊗ ((ba)c − b(ac)) − f
i
k
i
q
q∨q
k
⊗ b(ac).
j
q
Applying q
ciative. Applying
⊗ IdA on the two sides of this equality, we obtain (ab)c − a(bc) = 0. So A is asso-
⊗ IdA on the two sides of this equality, we obtain a(bc) = b(ac), so A is
q∨q
k
j
i
q
64
permutative, associative.
✷
Example. Let I a set, and let AI = V ect(ei)i∈I . Then A is given a permutative, associative
product: for all i, j ∈ I,
ei.ej = ej.
Let (g, ⋆) be a pre-Lie algebra. The pre-Lie product of g ⊗ A is given by:
(x ⊗ ei) ⋆ (y ⊗ ej) = x ⋆ y ⊗ ej.
The following proposition is immediate:
Proposition 75 Let (S) be a Hopf SDSE with set of indices I and (S′) be a dilatation of
(S), with set of indices J being the disjoint union of finite sets Ji indexed by i ∈ I. Let J ′ be
a set and for all i ∈ I, let φi : Ji −→ J ′ be a map. The following morphism is a morphism of
pre-Lie algebras:
(cid:26)
g(S ′) −→ g(S) ⊗ AJ ′
fx(k), x ∈ Ji −→ fi(k) ⊗ eφi(x).
It is injective (respectively surjective, bijective) if, and only if, φi is injective (respectively surjec-
tive, bijective) for all i ∈ I.
11.2 Dilatation of a Lie algebra
Let Set be the category of sets, Vect be the category of Vector spaces, and Lie the category of
Lie algebras.
Definition 76 Let V be a vector space. We define a function FV from Set to Vect in the
following way:
1. If I is a set:
FV (I) =Mi∈I
V.
The element v ∈ V in the copy of V corresponding to the index i ∈ I will be denoted by
vi.
2. If σ : I −→ J is a map:
FV (σ) :(cid:26) FV (I) −→ FV (J)
vi −→ vσ(i).
Definition 77 Let g be a Lie algebra. A dilatation of g is functor F : Set −→ Lie such
that F ({1}) = g and making the following diagram commuting:
Set
F
Lie
#HHHHHHHHH
Fg
{wwwwwwwww
Vect
where the functor from Lie to Vect is the forgetful functor.
Proposition 78 Let g be a Lie algebra. There is a bijection between the set of dilatations of
g and the set of pre-Lie product inducing the Lie bracket of g.
65
/
/
#
{
Proof. First step. Let ⋆ be a pre-Lie product inducing the Lie bracket of g. Let I be a set.
We identify v ⊗ ei ∈ g ⊗ AI and vi ∈ Fg(I). So Fg(I) is given a structure of pre-Lie algebra by:
The induced Lie bracket is given by:
vi ⋆ wj = (v ⋆ w)j .
[vi, wj] = (v ⋆ w)j − (w ⋆ v)i.
It is then easy to prove that this structure of pre-Lie algebra on Fg(I) for all I gives a dilatation
of g.
Second step. Let F be a dilatation of g. So for any set I, Fg(I) is now a Lie algebra. Moreover,
if σ : I −→ J is any map, then Fg(σ) : Fg(I) −→ Fg(J) is a Lie algebra morphism. We first
consider Fg({1, 2}). Let π2 be the projection on Fg({2}) which vanishes on Fg({1}) in Fg({1, 2}).
We define ⋆ on g in the following way: if v, w ∈ V ,
(v ⋆ w)2 = π2([v1, w2]).
Let σ : {1, 2} −→ {1, 2}, permuting 1 and 2. Then Fg(σ) permutes the two copies of g in
Fg({1, 2}), so Fg(σ) ◦ π1 = π2 ◦ Fg(σ). Moreover, Fg(σ) is a morphism of Lie algebras, so for all
v, w ∈ V :
Fg(σ) ◦ π2([w1, v2]) = π1 ◦ Fg(σ)([w1, v2]),
Fg(σ)((w ⋆ v)2) = π1([w2, v1])
(w ⋆ v)1 = −π1([v1, w2]).
So, in Fg({1, 2}):
[v1, w2] = π1([v1, w2]) + π2([v1, w2]) = (v ⋆ w)2 − (w ⋆ v)1.
Let us now consider any set I and i, j ∈ I, not necessarily distinct. Considering τ : {1, 2} −→
{i, j} sending 1 to i and 2 to j, as Fg(τ ) is a morphism of Lie algebras, for all v, w ∈ g, in Fg(I):
[vi, wj] = [Fg(τ )(v1), Fg(τ )(w2)]
= Fg(τ )([v1, w2])
= Fg(τ )((v ⋆ w)2 − (w ⋆ v)1)
= (v ⋆ w)j − (w ⋆ v)i.
In particular, if i = j = 1, in Fg({1}) = g, [v, w] = v ⋆ w − w ⋆ v: the product ⋆ induces the Lie
bracket of g(S).
Let x, y, z ∈ g. In Fg({1, 2, 3}):
0 = [x1, [y2, z3]] + [y2, [z3, x1]] + [z3, [x1, y2]]
= (x ⋆ (y ⋆ z))3 − (x ⋆ (z ⋆ y))2 − ((y ⋆ z) ⋆ x)1 + ((z ⋆ y) ⋆ x)1
+(y ⋆ (z ⋆ x))1 − (y ⋆ (x ⋆ z))3 − ((z ⋆ x) ⋆ y)2 + ((x ⋆ z) ⋆ y)2
+(z ⋆ (x ⋆ y))2 − (z ⋆ (y ⋆ x))1 − ((x ⋆ y) ⋆ z)3 + ((y ⋆ x) ⋆ z)3.
Considering the terms in the third copy of g:
(x ⋆ (y ⋆ z))3 − (y ⋆ (x ⋆ z))3 − ((x ⋆ y) ⋆ z)3 + ((y ⋆ x) ⋆ z)3 = 0.
So ⋆ is pre-Lie.
Last step. We define in the first step a correspondance sending a pre-Lie product on g to a
dilatation of g, and in the second step a correspondance sending a dilatation of g to a pre-Lie
product on g. It is clear that they are inverse one from the other.
✷
66
References
[1] Christoph Bergbauer and Dirk Kreimer, Hopf algebras in renormalization theory: locality
and Dyson-Schwinger equations from Hochschild cohomology, IRMA Lect. Math. Theor.
Phys., vol. 10, Eur. Math. Soc., Zürich, 2006, arXiv:hep-th/0506190.
[2] D. J. Broadhurst and D. Kreimer, Towards cohomology of renormalization: bigrading the
combinatorial Hopf algebra of rooted trees, Comm. Math. Phys. 215 (2000), no. 1, 217 -- 236,
arXiv:hep-th/0001202.
[3] Frédéric Chapoton, Algèbres pré-lie et algèbres de Hopf liées à la renormalisation, C. R.
Acad. Sci. Paris Sér. I Math. 332 (2001), no. 8, 681 -- 684.
[4]
, Un endofoncteur de la catégorie des opérades, Dialgebras and related operads,
Lecture Notes in Math., vol. 1763, Springer, Berlin, 2001, pp. 105 -- 110.
[5] Frédéric Chapoton and Muriel Livernet, Pre-Lie algebras and the rooted trees operad, Inter-
nat. Math. Res. Notices 8 (2001), 395 -- 408, arXiv:math/0002069.
[6] C. Chryssomalakos, H. Quevedo, M. Rosenbaum, and J. D. Vergara, Normal coordinates
and primitive elements in the Hopf algebra of renormalization, Comm. Math. Phys. 255
(2002), no. 3, 465 -- 485, arXiv:hep-th/0105259.
[7] Alain Connes and Dirk Kreimer, Hopf algebras, Renormalization and Noncommutative ge-
ometry, Comm. Math. Phys 199 (1998), no. 1, 203 -- 242, arXiv:hep-th/9808042.
[8] Héctor Figueroa and José M. Gracia-Bondia, On the antipode of Kreimer's Hopf algebra,
Modern Phys. Lett. A 16 (2001), no. 22, 1427 -- 1434, arXiv:hep-th/9912170.
[9] Loïc Foissy, Finite-dimensional comodules over the Hopf algebra of rooted trees, J. Algebra
255 (2002), no. 1, 85 -- 120, arXiv:math.QA/0105210.
[10]
, Faà di Bruno subalgebras of
the Hopf algebra of planar trees from combi-
natorial Dyson-Schwinger equations, Advances in Mathematics 218 (2008), 136 -- 162,
ArXiv:0707.1204.
[11] Robert L. Grossman and Richard G. Larson, Hopf-algebraic structure of families of trees, J.
Algebra 126 (1989), no. 1, 184 -- 210, arXiv:0711.3877.
[12]
, Hopf-algebraic structure of combinatorial objects and differential operators, Israel
J. Math. 72 (1990), no. 1-2, 109 -- 117.
[13]
, Differential algebra structures on families of trees, Adv. in Appl. Math. 35 (2005),
no. 1, 97 -- 119, arXiv:math/0409006.
[14] Michael E. Hoffman, Combinatorics of rooted trees and Hopf algebras, Trans. Amer. Math.
Soc. 355 (2003), no. 9, 3795 -- 3811.
[15] Dirk Kreimer, Combinatorics of (perturbative) Quantum Field Theory, Phys. Rep. 4 -- 6
(2002), 387 -- 424, arXiv:hep-th/0010059.
[16]
, Dyson-Schwinger equations: from Hopf algebras to number theory, Universality and
renormalization, Fields Inst. Commun., no. 50, Amer. Math. Soc., Providence, RI, 2007,
arXiv:hep-th/0609004.
[17] Dirk Kreimer and Karen Yeats, An étude in non-linear Dyson-Schwinger equations, Nuclear
Phys. B Proc. Suppl. 160 (2006), 116 -- 121, arXiv:hep-th/0605096.
67
[18] John W. Milnor and John C. Moore, On the structure of Hopf algebras, Ann. of Math. (2)
81 (1965), 211 -- 264.
[19] Florin Panaite, Relating the Connes-Kreimer and Grossman-Larson Hopf algebras built on
rooted trees, Lett. Math. Phys. 51 (2000), no. 3, 211 -- 219.
[20] Richard P. Stanley, Enumerative combinatorics. Vol. 1., Cambridge Studies in Advanced
Mathematics, no. 49, Cambridge University Press, Cambridge, 1997.
[21]
, Enumerative combinatorics. Vol. 2., Cambridge Studies in Advanced Mathematics,
no. 62, Cambridge University Press, Cambridge, 1999.
68
|
1210.3847 | 1 | 1210 | 2012-10-14T21:42:14 | Quotients of Koszul algebras and 2-d-determined algebras | [
"math.RA"
] | Vatne and Green & Marcos have independently studied the Koszul-like homological properties of graded algebras that have defining relations in degree 2 and exactly one other degree. We contrast these two approaches, answer two questions posed by Green & Marcos, and find conditions that imply the corresponding Yoneda algebras are generated in the lowest possible degrees. | math.RA | math |
QUOTIENTS OF KOSZUL ALGEBRAS AND
2-D-DETERMINED ALGEBRAS
THOMAS CASSIDY AND CHRISTOPHER PHAN
Abstract. Vatne [12] and Green & Marcos [8] have indepen-
dently studied the Koszul-like homological properties of graded
algebras that have defining relations in degree 2 and exactly one
other degree. We contrast these two approaches, answer two ques-
tions posed by Green & Marcos, and find conditions that imply the
corresponding Yoneda algebras are generated in the lowest possible
degrees.
1. Introduction
The Koszul property for graded associative algebras [11] has been
generalized in several directions, including algebras with defining re-
lations in just one degree (d-Koszul algebras [2]) and algebras with
defining relations in multiple degrees (K2 algebras [5]). d-Koszul alge-
bras share many of the nice homological properties of Koszul algebras,
but are not closed under several standard operations. The family of K2
algebras, which includes the d-Koszul algebras, has the advantage of
being closed under graded Ore extensions, regular normal extensions,
and tensor products. Interpolating between d-Koszul and K2 algebras
leads one to look for Koszul-like homological properties in algebras
with defining relations in just two degrees. This idea was considered
independently by Vatne [12] and Green & Marcos [8], who each investi-
gate graded algebras with defining relations in degree 2 and exactly one
other degree. We compare these two approaches to find sufficient con-
ditions for such an algebra to be K2, and answer two questions posed
by Green & Marcos.
Let k be a field and d an integer greater than 2. We consider graded
k-algebras of the form R = T(V )/I where V is a finite dimensional vec-
tor space, T(V ) is the free algebra generated by V and I is a homoge-
nous ideal which can be generated by elements in T(V )2 and T(V )d.
This work was partially supported by a grant from the Simons Foundation (Grant
Number 210199 to Thomas Cassidy). The authors would like to thank Andrew
Conner for his helpful suggestions.
1
2
CASSIDY AND PHAN
Properties of I determine the structure the bi-graded Yoneda algebra
E(R) := ⊕i,j Exti,j
R (k, k), where i refers to the cohomological degree
and j is the internal degree that E(R) inherits from the grading on R.
We let Ei(R) denote ⊕jEi,j(R).
Definition 1.1. R is Koszul if E(R) is generated as a k-algebra by
E1(R). R is K2 if E(R) is generated as a k-algebra by E1(R) and
E2(R). R is d-Koszul if E(R) is generated as a k-algebra by E1(R)
and E2(R), and also I is generated in degree d.
Each of these definitions requires E(R) to be generated in the lowest
possible degrees. One can determine whether or not an algebra is
Koszul or d-Koszul just by knowing the bi-degrees in the corresponding
Yoneda algebra. (Specifically, for A to be d-Koszul, we need Ei,j(A) =
0 unless j = δ(i), where δ(2m) = dm and δ(2m + 1) = dm + 1. Note
that Koszul and 2-Koszul are synonomous.)
In contrast, E(R) can
have the same bi-degrees as an algebra generated in degrees 1 and 2
even when R is not actually K2.
To explore the connections between these definitions, it is helpful
to consider the quadratic generators of I separately from the degree d
generators. Let I2 denote a linearly independent set of quadratic rela-
tions, and J a set of degree d relations, so that I is the ideal generated
by I2 and J. Note that different choices for J can produce the same
algebra R. Let A be the algebra T(V )/(cid:104)I2(cid:105) and let B = T(V )/(cid:104)J(cid:105), so
that the algebra R can be realized as either A/(cid:104)J(cid:105) or B/(cid:104)I2(cid:105).
2. Almost linear resolutions
It would be nice if the Koszul property of A and the d-Koszul prop-
erty of B would combine to imply that R is K2, but this is not neces-
sarily the case, as is shown in example 3.7 below. Indeed, any two of
the algebras A, B and R can have good homological behavior while the
third is recalcitrant. Remark 7.5 in [7] illustrates the case where R is
K2, B is 3-Koszul, and yet A is not Koszul. In the following example,
R is K2, A is Koszul, but the Yoneda algebra of B fails to be generated
in low degrees.
Example 2.1. Let V = {x, y}, I2 = {xy − yx} and J = {xyx}.
Then A is commutative and xyx is a regular element in A, hence by
[5, Corollary 9.2], R is K2. But B, as a monomial algebra, fails to be
3-Koszul by Proposition 3.8 in [2].
Clearly different hypotheses are required to get good behavior from
R. In [12], Vatne considers the case were A is Koszul, and R has an
almost linear resolution as an A-module.
QUOTIENTS OF KOSZUL ALGEBRAS AND 2-D-DETERMINED ALGEBRAS 3
A
(R, k) for all i > 0.
Definition 2.2. R has an almost linear resolution as an A-module if
Exti
A(R, k) = Exti,d−1+i
Vatne shows that if A is Koszul, d > 3 and R has an almost linear
resolution as an A-module, then E(R) has the correct bi-degrees for R
to be a K2 algebra. In fact R is K2 in this case, as the following direct
corollary of Theorem 5.15 in [7] shows.
Proposition 2.3. If A is Koszul, d ≥ 3, and R has an almost linear
resolution as an A-module, then R is K2.
In this case of monomial algebras, the almost linear condition is
relatively easy to check. This fact motivates our concluding theorem.
Proposition 2.4. Suppose that I2 and J consist of monomials and that
no element of J contains any element of I2 as a connected subword.
A(R, k) =
Then AR has an almost linear resolution if and only if Ext1
Ext1,d
A (R, k). In this case, R is K2.
A (R, k).
A(R, k) = Ext1,d
Note that A ⊗ A⊗•
A(R, k) = Ext1,d
Now, suppose Ext1
Proof. If AR has an almost linear resolution, then by definition
Ext1
A (R, k). In the remainder of this
proof, let πA : T(V ) → A be defined by πA(s) := s + (cid:104)I2(cid:105). Let M be
the set of monomials u ∈ T(V ) with πA(u) (cid:54)= 0.
+ is a projective resolution of AR. We will con-
struct a subresolution A⊗ Q• of A⊗ A•
+, which is a minimal projective
resolution. We construct Q• by choosing a monomial basis Bi for each
A (R, k), the left-ideal in
vector space Qi. Because Ext1
A generated by πA(J) is equal to the two-sided ideal generated by the
same elements. Thus we may begin by setting B1 = {πA(s) : s ∈ J}.
A(R, k) = Ext1,d
Now, suppose that Bi consists of elements of the form
πA(ui) ⊗ πA(ui−1) ⊗ ··· ⊗ πA(u1)
where each ut ∈ M. Then, we set
Bi+1 := {πA(ui+1) ⊗ πA(ui) ⊗ πA(ui−1) ⊗ ··· ⊗ πA(u1)
: ut ∈ M, πA(ui) ⊗ πA(ui−1) ⊗ ··· ⊗ πA(u1) ∈ Bi,
and πA(ui+1) is a minimal left-annihilator of πA(ui).}
Therefore, every Bi will consist of pure tensors of monomials. Fur-
thermore, since A is a quadratic monomial algebra, each minimal left-
annihilator is linear, and so
Bi ⊂ A⊗i−1
1
⊗ spank J.
Therefore, A ⊗ Q• is an almost linear resolution of AR. Since A is
monomial, it is also Koszul, and thus R is K2 by 2.3.
(cid:3)
4
CASSIDY AND PHAN
3. 2-d-determined algebras
In [8], Green and Marcos study the case where the bi-degrees of
E(R) are no greater than the bi-degrees of a K2 algebra with defining
relations in degree 2 and d. They call such an algebra 2-d-determined.
Like Vatne, Green and Marcos assume that R is a quotient of a Koszul
algebra A, but they do so via Grobner bases. They assume that I
has a reduced Grobner basis g = g2 ∪ gd, so that A = T(V )/(cid:104)g2(cid:105) and
B = T(V )/(cid:104)gd(cid:105).
At the end of [8], Green and Marcos ask three questions. In Theo-
rems 3.5 and 3.6 we provide negative answers to the first two questions
(which we have rephrased slightly)1:
(1) If C is an 2-d-determined, then is the Ext-algebra E(C) finitely
generated?
(2) If C is a 2-d-determined algebra and the Ext-algebra E(C) is
finitely generated, is C a K2 algebra (assuming that the global
dimension of C is infinite)?
We use the following construction (see [10, §III.1]):
Definition 3.1. Let A and B be graded algebras. The free product
of A and B is the algebra
A (cid:116) B :=
+ ⊗ (B+ ⊗ A+)⊗i ⊗ B⊗2
A⊗1
+ .
(cid:77)
i≥0
1,2∈{0,1}
This related result will be of importance.
Proposition 3.2 (c.f.
A and B,
[10, Proposition III.1.1]). For graded algebras
E(A (cid:116) B) (cid:39) E(A) (cid:117) E(B).
Suppose V has an ordered basis x1 < x2 < ··· < xn. Then we can
order the monomials of T(V ) by degree-lexicographical order. This
induces a filtration F on A, TorA(k, k) and E(A). (See, for example,
[10, Chapter IV] and [9].)
With this filtration F , we now have several versions of the Ext-
functor. First, there is the ungraded Ext-functor (over A-modules and
grF A-modules), which is the derived functor of the ungraded Hom-
functor. Next, since A and its associated graded algebra grF A are
both graded with respect to an internal degree, we have the N-graded
Ext-functor (over A-modules and grF A-modules), the derived functor
1In their formulation, Green and Marcos consider quotients of graph algebras.
We only consider connected-graded algebras, which suffice to answer the questions
in the negative.
QUOTIENTS OF KOSZUL ALGEBRAS AND 2-D-DETERMINED ALGEBRAS 5
of the N-graded Hom functor. Finally grF A is graded by the monoid of
monomials in T(V ), and hence we have a monomial-graded Ext-functor
(over grF A-modules).
We will use the following result:
Lemma 3.3 ([9, Theorem 1.2]). Let grF E(A) be the associated graded
algebra of ungraded algebra E(A) under the filtration F , and E(grF A)
be the graded Ext-algebra with respect to the monomial grading. Then
there is a bigraded algebra monomorphism
Λ : grF E(A) (cid:44)→ E(grF A).
In the case where each Ei,j(A) is finite-dimensional, the graded and
ungraded versions coincide (see [9, Lemma 1.4]). A slight modification
of the proof of [9, Lemma 2.11] yields:
Lemma 3.4. If Ei,j(A) is finite-dimensional for every i, j, then
dim Ei,j(A) = dim
(grF Ei(A))α
(cid:77)
α=j
where α is the length of the monomial α.
To answer to question (1), let
A :=
B :=
k(cid:104)a, b, c, d, e, f, l, m(cid:105)
(cid:104)bc − ef, ae, da − lm, cl(cid:105) ,
k(cid:104)z(cid:105)
(cid:104)z4(cid:105) ,
and
C := A (cid:116) B.
We order the monomials in
k(cid:104)a, b, c, d, e, f, l, m, z(cid:105)
by degree-lexicographical order. This creates a filtration F on A, B,
and C, as well as their corresponding Ext-algebras. Using Bergman's
diamond lemma in [3], we see that
grF A (cid:39) k(cid:104)a, b, c, d, e, f, l, m(cid:105)
(cid:104)ef, ae, lm, cl, abc, cda,(cid:105) .
Consider the structure of E(gr A). We construct a basis for a vector
space V• ⊆ (gr A)•
+ is a minimal
projective resolution of gr Ak. This is done by applying the left anni-
hilator algorithm described in [5]. Figure 1 depicts the minimal left
+ so that gr A ⊗ V• ⊆ gr A ⊗ (gr A)•
6
CASSIDY AND PHAN
cd
ab
a
b
c
d
e
f
l
m
Figure 1. A basis for V• can be constructed from the
paths ending in a first-degree monomial in this graph.
annihilation of monomials in the basis for V•. Consider all paths end-
ing with a first-degree monomial. By tensoring the vertices in all such
paths we obtain a basis for V•. For example, ab ⊗ cd ⊗ ab ⊗ c is a basis
element for V4. Other examples and applications of these graphs can
be found in [6], where the graphs are used to characterize finiteness
properties of Yoneda algebras. In Figure 1, the dotted lines represent
left-annihilation from an inessential relation (e.g. ab left-annihilates cd
because of the essential relation abc).
Likewise, we set W2i := k(z3 ⊗ z)⊗i and W2i+1 := kz ⊗ W2i, so that
B ⊗ W• ⊆ B ⊗ B⊗•
+ is a minimal projective resolution of Bk. (As with
V•, we can visualize a basis for W• using Figure 2.)
z3
z
Figure 2. A basis for W• can be constructed from the
paths ending in the first-degree monomial z in this graph.
Theorem 3.5. C is 2-4-determined, but E(C) is not finitely generated.
Proof. First, it's clear that the ideal of relations for C is generated in
degrees 2 and 4.
QUOTIENTS OF KOSZUL ALGEBRAS AND 2-D-DETERMINED ALGEBRAS 7
Now, by examining Figure 1, we see that Ei,j(grF A) = 0 unless
j ≤ 2i − 1. Thus, by Lemmas 3.3 and 3.4, we know that Ei,j(A) = 0
unless j ≤ 2i − 1. Likewise, we see that Ei,j(B) = 0 unless j ≤ δ(i)
(where δ is as defined in the paragraph after Definition 1.1). Now, since
2i − 1 ≤ δ(i) for every i, applying Proposition 3.2, we see that C is
2-4-determined.
By Proposition 3.2, to show that E(C) is not finitely generated, it
suffices to show that E(A) is not finitely generated. We shall exhibit a
projective resolution for Ak. Consider the sequence
(1)
··· Mi+1−−−→ A(−2i + 2)2 Mi−→ ··· M5−−→ A(−6)2 M4−−→ A(−4)2 M3−−→ A(−2)4
M2−−→ A(−1)8 M1−−→ A M0−−→ k → 0
where the maps are right multiplication by the matrices
M1 = (a, b, c, d, e, f, l, m)T ,
0 0 0 a
0 0
0
0 0 0 0
c 0
0 b 0 0 −e 0 0
0
−d 0 0 0 0
0 l
0
0
0
(cid:19)
,
0
(cid:18)cd
(cid:18)ab
(cid:18)cd
0
0
0
M2 =
M3 =
M2i =
M2i+1 =
0
0 0
ab 0 0
,
(cid:19)
(cid:19)
0
cd
0
ab
for i ≥ 2, and
for i ≥ 2.
Given any fixed integer N , noncommutative Grobner bases can be
used to calculate the dimensions of Ei,j(A) for j < N . We have used
the computer program Bergman [1] for this purpose.
Note that (1) satisfies im Mi ⊂ ker Mi−1. Indeed, (1) is clearly exact
up to the matrix M2. Grobner bases calculations show that E3,3(A) = 0
and E3,5(A) = 0. From above, we know that E3,j(A) = 0 for j > 5, and
so (1) is exact up to the matrix M3.
Likewise, Grobner bases calculations show that E4,4(A), E4,5(A) = 0,
and E4,7(A) = 0. From above, we know that E4,j(A) = 0 for j > 7,
and so (1) is exact up to the matrix M4.
i ≥ 3. It follows that (1) is exact.
However, im M4 = ker M3 implies that im Mi+1 = ker Mi for all
8
Therefore, we have
CASSIDY AND PHAN
(cid:40)
Ei(A) =
Ei,i(A)
Ei,2i−2(A)
if i = 0, 1,
if i ≥ 2.
Fix i ≥ 3.
If the elements of Ei,2i−2(A) were generated by lower
cohomological-degree elements of E(A), then the multiplication map
(2)
Ej,k(A) ⊗ Ei−j,2i−k−2(A) → Ei,2i−2(A)
(cid:77)
0<j<i
0<k<2i−2
would be surjective. We will show that this is not the case.
Suppose 2 ≤ j < i. Then for Ej,k(A) to be nonzero, we would need
k = 2j − 2. This implies 2i − k − 2 = 2(i − j), which is neither i − j
nor 2(i − j) − 2, and so Ei−j,2i−k−2(A) = 0.
On the other hand, suppose j = 1. Then for E1,k(A) to be nonzero,
we would need k = 1. This implies 2i − k − 2 = 2i − 3. However,
i − j = i − 1 ≥ 2, and so Ei−j,2i−k−2(A) = Ei−1,2i−3(A) = 0.
In either case, the map in (2) is zero for i ≥ 3. Hence, for i ≥ 3,
Ei(A) is not generated by lower cohomological-degree elements of E(A).
(cid:3)
Therefore, E(A) is not finitely generated.
Now we answer question (2).
Theorem 3.6. There exists a 2-d-determined algebra (with infinite
global dimension) which is not K2 even though its Yoneda algebra is
finitely generated.
Proof. Let
(cid:28) np − nq, np − nr, ps − pt, qt − qu, rs − ru, sv − sw, tw − tx,
k(cid:104)a, b, n, p, q, r, s, t, u, v, w, x, y(cid:105)
uv − ux, sv − sy, tw − ty, ux − uy, va − vb, wa − wb, xa − xb
k(cid:104)z(cid:105)
(cid:104)z4(cid:105) ,
(cid:29) ,
A :=
B :=
and
C := A (cid:116) B.
The algebra A appears in [4], and the following minimal projective
resolution for Ak is given:
0 → A(−5) → A(−3)7 → A(−2)14 → A(−1)13 → A.
Note that A is a quadratic algebra which is not Koszul, and so A is not
K2. However, as A has finite global dimension, E(A) must be finitely
QUOTIENTS OF KOSZUL ALGEBRAS AND 2-D-DETERMINED ALGEBRAS 9
generated. As shown above, the algebra B satisfies E2i(B) = E2i,4i(B),
E2i+1(B) = E2i+1,4i+1(B), and dim Ei(B) = 1. B has infinite global
dimension, but since it is K2, E(B) is finitely-generated. It follows from
Proposition 3.2, that C is 2-4-determined, has infinite global dimension,
and is not K2.
(cid:3)
There is a small omission in the statement of Theorem 20 in [8].
From their earlier proofs it is clear the authors intended to include
the hypotheses ". . . and gd is a Grobner basis for the ideal it gener-
ates." The following example shows that without this hypothesis, the
conclusions of Theorem 20 are not valid.
Example 3.7. Let V = {a, x, y, z}, ordered by z > y > x > a. Then
g = g2 ∪ g4 = {xa, az, ay} ∪ {y2z2, x2y2 + a4} forms a Grobner basis.
A =
is a quadratic monomial algebra and hence Koszul,
k(cid:104)a, x, y, z(cid:105)
has global dimension 2 and thus is 4-Koszul. But
(cid:104)g2(cid:105)
k(cid:104)a, x, y, z(cid:105)
and B =
(cid:104)gd(cid:105)
over the algebra
k(cid:104)a, x, y, z(cid:105)
(cid:104)x2y2 + a4, y2z2, xa, az, ay(cid:105) ,
R =
the module k has a minimal projective resolution of the form
0 → R(−6) → R(−7)⊕R(−6)⊕R(−5)⊕R(−3)2 → R(−4)2⊕R(−2)3 →
R(−1)4 → R → k → 0.
This means that E(R)3,6 (cid:54)= 0 and so R is neither K2 nor 2-4-determined.
We can establish sufficient conditions for R to be K2 by merging
hypotheses from Vatne with those of Green and Marcos. Our last
theorem shows that we need only slightly stronger hypotheses on the
Grobner basis to guarantee that R is K2.
Proposition 3.8. Let g2, gd and g2 ∪ gd be Grobner bases such that
g2 ∪ gd has no redundant elements (i.e. any set of defining relations for
R has at least g2 + gd elements). If either
(i) B = T(V )/(cid:104)gd(cid:105) is d-Koszul, or
(ii) Ext1
gr A(gr R, k) = Ext1,d
gr A(gr R, k),
then R is K2.
10
CASSIDY AND PHAN
Proof. Using the filtration defined by the Grobner bases we create the
monomial algebras gr A, gr B and gr R. We will show that gr R is a K2
algebra.
For case (i), assume that B is d-Koszul. Then grB is d-Koszul by
It follows from [8, Theorem 14] that gr R is 2-d-
[8, Theorem 10].
determined, and hence K2 by [8, Theorem 16].
gr A(gr R, k) then by proposition
For case (ii), if Ext1
2.4, gr R has an almost linear resolution over gr A and is therefore K2.
In either case, gr R is a monomial K2 algebra. Since g2 ∪ gd has no
redundant elements, it forms an essential Grobner basis in the sense of
[9]. Thus by [9, Theorem 1.7], R itself is K2.
(cid:3)
gr A(gr R, k) = Ext1,d
Condition (ii) is certainly not necessary. Is condition (i) necessary?
References
[1] Jorgen Backelin, Bergman. Computer software, available at http://servus.
math.su.se/bergman/.
[2] Roland Berger, Koszulity for nonquadratic algebras, J. Algebra 239 (2001),
no. 2, 705 -- 734. MR1832913 (2002d:16034)
[3] George M. Bergman, The diamond lemma for ring theory, Adv. in Math. 29
(1978), no. 2, 178 -- 218. MR506890 (81b:16001)
[4] Thomas Cassidy, Quadratic algebras with Ext algebras generated in two de-
grees, J. Pure Appl. Algebra 214 (2010), no. 7, 1011 -- 1016. MR2586982
(2011b:16099)
[5] Thomas Cassidy and Brad Shelton, Generalizing the notion of Koszul algebra,
Math. Z. 260 (2008), no. 1, 93 -- 114. MR2413345 (2009e:16047)
[6] Andrew Conner, Ellen Kirkman, James Kuzmanovich, and Frank W. Moore,
Finiteness conditions on the Yoneda algebra of a monomial algebra. Preprint.
[7] Andrew Conner and Brad Shelton, K2 factors of Koszul algebras and applica-
tions to face rings, J. Algebra 368 (2012), 251 -- 270. MR2955232
[8] Edward L. Green and E. N. Marcos, d-Koszul algebras, 2-d-determined algebras
and 2-d-Koszul algebras, J. Pure Appl. Algebra 215 (2011), no. 4, 439 -- 449.
MR2738362
[9] Christopher Phan, Generalized Koszul properties for augmented algebras, J.
Algebra 321 (2009), no. 5, 1522 -- 1537. MR2494406 (2010a:16017)
[10] Alexander Polishchuk and Leonid Positselski, Quadratic algebras, University
Lecture Series, vol. 37, American Mathematical Society, Providence, RI, 2005.
MR2177131 (2006f:16043)
[11] Stewart B. Priddy, Koszul resolutions, Trans. Amer. Math. Soc. 152 (1970),
39 -- 60. MR0265437 (42 #346)
[12] Jon Eivind Vatne, Quotients of Koszul algebras with almost linear resolution,
2012. preprint, arXiv:1103.3572.
QUOTIENTS OF KOSZUL ALGEBRAS AND 2-D-DETERMINED ALGEBRAS 11
Mathematics Department, Bucknell University, Lewisburg, PA
E-mail address: [email protected]
Department of Mathematics and Statistics, Winona State Univer-
sity, Winona, MN
E-mail address: [email protected]
|
1304.0687 | 1 | 1304 | 2013-04-02T16:49:15 | On Koszulity for operads of Conformal Field Theory | [
"math.RA"
] | We study two closely related operads: the Gelfand-Dorfman operad GD and the Conformal Lie Operad CLie. The latter is the operad governing the Lie conformal algebra structure. We prove Koszulity of the Conformal Lie operad using the Groebner bases theory for operads and an operadic analogue of the Priddy criterion. An example of deformation of an operad coming from the Hom structures is considered. In particular we study possible deformations of the Associative operad from the point of view of the confluence property. Only one deformation, the operad which governs the identity $(\alpha(ab))c=a(\alpha (bc))$ turns out to be confluent. We introduce a new Hom structure, namely Hom--Gelfand-Dorfman algebras and study their basic properties. | math.RA | math |
On Koszulity for operads of Conformal Field Theory
Natalia Iyudu, Abdenacer Makhlouf
Abstract
We study two closely related operads: the Gelfand-Dorfman operad GD and the Conformal Lie
Operad CLie. The latter is the operad governing the Lie conformal algebra structure. We prove
Koszulity of the Conformal Lie operad using the Grobner bases theory for operads and an operadic
analogue of the Priddy criterion. An example of deformation of an operad coming from the Hom
structures is considered. In particular we study possible deformations of the Associative operad from
the point of view of the confluence property. Only one deformation, the operad which governs the
identity (α(ab))c = a(α(bc)) turns out to be confluent. We introduce a new Hom structure, namely
Hom -- Gelfand-Dorfman algebras and study their basic properties.
1
Introduction
One of the key notions of the two-dimensional Conformal Field Theory is the operator
product expansion (OPE) of chiral quantum fields (see e.g. [11] ):
Φ(x)Ψ(y) = Xn∈Z
(Φ(n)Ψ)(y)
(x − y)n+1 .
This defines a binary operation Φ(n)Ψ on the space of chiral fields for each n ∈ Z. These
binary operations, along with the identity field and the unary operation ∂ = ∂/∂y form
a structure, called a vertex (or chiral) algebra, which has been axiomatized by Borcherds
[3]. The most important part of the OPE is its singular part since it allows to compute
commutators of chiral fields. This involves only the binary operations Φ(n)Ψ for each n ∈
Z+, and along with the unary operation ∂, forms a structure, called a Lie conformal algebra,
which has been axiomatized by Kac [11].
Definition. A Lie conformal algebra is a C[∂] - module R endowed with binary bilinear
(n) : R × R → R for each n ∈ Z+, which satisfy the following axioms:
over C operations
(c1) a(n)b = 0, n ≫ 0,
(c2) (∂a)(n)b = −na(n−1)b, a(n)∂b = ∂(a(n)b) + na(n−1)b,
(c3) a(n)b = −
∞
(−1)(n+j)∂(j)(b(n+j)a),
Xj=0
1
(c4) a(m)(b(n)c) − b(n)(a(m)c) =
m
Xj=0(cid:18) m
j (cid:19) (a(j)b)(m+n−j)c.
In the axiom (c3) and further on we use the divided power notation: ∂(n) = ∂n/n!.
Lie conformal algebras occur naturally in Classical Field Theory as well. Namely, con-
sider a linear local Poisson bracket
{ui(x), uj (y)} = Hij(u(y), u′(y), · · · , u(s)(y); ∂/∂y)δ(x − y),
and their derivatives, and let Hij(λ) = Ps∈Z
where H = (Hij(∂/∂y)) is a matrix differential operator of finite order, linear in the ui
λs
s! be its symbol. Then the C[∂] - module,
consisting of linear combinations of functions ui and their derivatives, is a Lie conformal
algebra, endowed with the products ui(s)uj = H s
ji, s ∈ Z+.
H s
ij
One of the simplest examples of a linear local Poisson bracket, when H has order one,
was studied by Gelfand and Dorfman [6]; it is defined by the following symbol:
Hij(λ) =Xs
cs
ijus +Xs
iju′
ds
s + λ Xs
(ds
ij + ds
ji)us! ,
ij and ds
where cs
ij are constants, satisfying certain quadratic relations. In order to explain
these relations, consider the C-span V of the ui, and denote by [, ] and ◦ the bilinear
products, with the structure constants cs
ij respectively. Then (Hij(λ)) defines a
local Poisson bracket if and only if [, ] is a Lie algebra bracket, ◦ is a left symmetric algebra,
such that the operators of right multiplications commute, and these two binary operations
satisfy the compatibility condition (5) below. Such a structure is called a Gelfand-Dorfman
algebra. The special case, when the bracket is zero, is called a Novikov algebra [2], [22].
ij and ds
We consider in this paper operads, which describe the structure of the Gelfand-Dorfman
algebras, and of the Lie conformal algebras. We denote by CLie the operad, which controls
the structure of Lie conformal algebras, and we call it the CLie (conformal Lie) operad.
In Theorem 5.2 (Section 5) we prove that CLie is Koszul, using the Grobner bases theory
(Diamond Lemma) for operads and an operadic analogue of the Priddy criterion. An
explanation of these tools will be given in Section 2.
The complete list of relations for operations ◦ and [, ], defining a left(right) Gelfand-
Dorfman algebra is the following:
(1) (a ◦ b) ◦ c − a ◦ (b ◦ c) = (b ◦ a) ◦ c − b ◦ (a ◦ c) (left-symmetric),
(2) (a ◦ b) ◦ c = (a ◦ c) ◦ b (right Novikov),
(3) [a, b] = −[b, a] ,
(4) [[a, b], c] + [[c, a], b] + [[b, c], a] = 0 ,
(5) [c ◦ a, b] − [c ◦ b, a] + [c, a] ◦ b − [c, b] ◦ a − c ◦ [a, b] = 0 .
We call an algebra satisfying these relations a right GD-algebra. The algebra with
opposite multiplication to ◦, will be called a left GD-algebra.
2
In Section 3 we touch upon the question of the deformation theory for operads. Well-
known classes of algebras, such as Associative, Lie, Jordan, Novikov, Nambu, Poisson,
Gerstenhaber, Batalin-Vilkovisky, etc. are governed by corresponding operads, they consist
of algebras over those operads.
We study several examples of deformation of operads. We emphasize that algebras
over those deformed operads are Hom -- algebras, defined in the extensive literature over
the last years [14, 15, 16, 20, 21]. This notion should be distinguished from the notion of a
deformation of an algebra over a given operad, developed and heavily used before in varying
generality [13, 1, 5]. We use here a naive version of a deformation of an operad, in order to
provide examples which could lead to the general theory. Namely, by a deformation of an
operad we mean here the following.
Consider an operad A, and introduce an extra generator h in the presentation of this
operad by generators and relations. An extra operator corresponds in the algebra represen-
tation to a unary operation α : A → A. Then we write down relations, involving h, of the
deformed operad Ah in a certain way, such that if h = id, we get the old relations. This
new operad Ah can be considered as a deformation of A.
The examples of deformations we consider here govern exactly the notion of Hom --
structures. For example, for operad Ass, defined by one generation operation m and one
relation:
-
,
consider a deformed operad Assh, defined by two generators, a binary operation m, and a
unary operation h, and a relation
-
,
The operation h denoted here by an empty dot and m by a filled one. An algebra over
the operad Assh is a Hom-associative algebra defined in [15].
We therefore study here examples of deformations of operads inspired by Hom -- structures.
We consider all operads which can be obtained from Ass by deformations, of this kind,
introducing an extra generator -- a unary operation. We classify them from the point of
view of the confluence property of the deformed operad. Application of the version of the
Grobner bases theory for nonsymmetric operads (which allows to deal with operations of
arbitrary arity), leads to the result, that only one of those deformations is an operad with
confluent presentation. This is the operad which governs the identity (α(ab))c = a(α(bc)).
In particular, this operad is Koszul.
3
In Section 4 we introduce the notion of a Hom -- Gelfand-Dorfman (Hom -- GD) algebra.
Namely we suggest to study a certain particular deformation of Gelfand-Dorfman operad.
As it was explained above the Gelfand-Dorfman structure is an enriched version of the
Novikov structure, corresponding to more general type of local Poisson bracket. The latter
have been extensively studied (see for example [22], [4] and references therein), including
the Hom versions [20]. We establish a number of basic properties of the Hom -- GD algebra,
the algebra over a certain deformation of the operad GD.
2 Grobner bases theory for operads
In this section we remind the definition of the (symmetric, shuffle, nonsymmetric) operad
and set up the basics of the Grobner bases theory for operads, analogous to the Grobner
bases theory for algebras presented by generators and relations. For more detailed intro-
duction to various aspects of operads, including those discussed here see [19] and [18].
The main ingredient for the development of the analogue of the Grobner bases theory for
algebras in the operadic setting is the notion of shuffles and their orderings, which appeared
first in this context in the paper due to Hoffbeck [10]. After introducing shuffles one can
deal with elements of the free symmetric operad (trees with marked leaves and vertices)
exactly in the same manner, as one can do with monomials in the free associative algebra.
This was written down precisely in the paper due to Dotsenko and Khoroshkin [8].
To explain main points of the Grobner bases theory for operads presented by generators
and relations, where elements of the free operad are depicted by tree-monomials, we will
need to clarify how to order tree-monomials, and what means that one tree-monomial divides
another. On the basis of this the whole Grobner bases theory works in a very much parallel
way, as in the case of algebras.
Let us start with the definition of an operad. The notion of an operad can be made very
general, depending on the category, where we prefer to work. The key point is to define the
'product', which makes a category of 'collections' into a monoidal category.
The first basic notion is a collection.
Definition. We define a nonsymmetric collection as a functor from the category of non-
empty finite ordered sets Ord (with order-preserving bijections as morphisms) to the cate-
gory of vector spaces Vect.
In this case a nonsymmetric collection is just the same as a graded vector space.
However to get a more general definition one can substitute the category of vector spaces
by graded vector spaces, or dg-vector spaces, i.e. chain complexes, which is done for objects
appearing in, say, algebraic topology.
Definition. We define a symmetric collection as a functor from the category of non-empty
finite sets Fin (with bijections as morphisms) to the category of vector spaces Vect.
In this case a symmetric collection is also called a 'S-module', which means it is a set
of right k[Sn]-modules M (n) for any n ∈ Z+.
4
Note, that it can be shown that all vector spaces P(I), for I ∈ Fin are defined by
P(n)=P({ ¯1, n}), so that P(n), n ∈ Z+ is usually what is meant by a collection.
Now we can define a natural compositions of these collections.
Definition. Let P and Q be two nonsymmetric collections. Their nonsymmetric composi-
tion is defined by the formula.
P ◦ Q(I) =Mk
P(k) ⊗ ( Mf :I→{ ¯1,k}
Q(f −1(1)) ⊗ · · · ⊗ Q(f −1(k)) )
Here the sum is taken over all monotonically non-decreasing surjections f .
Definition. Let P and Q be two nonsymmetric collections. Their shuffle composition is
defined by the formula.
P ◦sh Q(I) =Mk
P(k) ⊗ ( Mf :I→{ ¯1,k}
Q(f −1(1)) ⊗ · · · ⊗ Q(f −1(k)) )
Here the sum is taken over all shuffling surjections f , i.e. those for which f −1(i) <
f −1(j), if i < j.
Definition. Let P and Q be two symmetric collections. Their symmetric composition is
defined by the formula
P ◦ Q(I) =Mk
P(k) ⊗k[Sk] ( Mf :I→{ ¯1,k}
Here the sum is taken over all surjections f .
Q(f −1(1)) ⊗ · · · ⊗ Q(f −1(k)) )
It can be proven that defined above composition (nonsymmetric, shuffle, symmetric)
makes the set of corresponding (nonsymmetric or symmetric) collections into a monoidal
category, i.e. the pentagon and triangular diagrams are satisfied, and the proper unital map
is present.
Now any monoid in the corresponding monoidal category we call a (nonsymmetric,
shuffle, symmetric) operad.
Example Let V be a vector space, denote by EndV the operad of multilinear mappings.
It is a collection Hom(V ⊗n, V ), n ≥ 0 of all multilinear mappings, with the composition
maps, defined in a natural way.
Definition. An algebra over an operad P is a vector space V together with an operadic
morphism from P to a corresponding operad of multilinear mappings EndV .
5
The 'building blocks' for the compositions we just defined (which serves as a monoidal
product) are the following elementary compositions. We will need them later for the Grobner
bases theory for symmetric operads. The case of symmetric operads is what we mainly
will need for our results, so we deal with this version of the Grobner bases theory. The
nonsymmetric case is easier, and could be obtained as a particular case of the symmetric
one.
Definition. We say that an element a of an operad P has an arity n, if it belongs to
P(n + 1).
Definition. Let P be a symmetric operad, for elements a ∈ P(n) and b ∈ P(m) we define
1. a nonsymmetric elementary composition a ◦i b as the operation
a(x1, · · · , xi−1, b(xi, xi+1, · · · , xi+m−1), xi+m, · · · , xm+n−1)
2. a shuffle elementary composition a ◦i,σ b as the operation
a(x1, · · · , xi−1, b(xi, xσ(i+1), · · · , xσ(i+m−1)), xσ(i+m), · · · , xσ(m+n−1))
where the bijection σ : {i + 1, · · · , m + n − 1} → {i + 1, · · · , m + n − 1} is a (m − 1, n − i)-
shuffle, which means:
σ(i + 1) < · · · < σ(m + n − 1)
and
σ(i + 1) < · · · < σ(m + n − 1).
3. a symmetric elementary composition a ◦i,σ b as the operation
a(x1, · · · , xi−1, b(xi, xσ(i+1), · · · , xσ(i+m−1)), xσ(i+m), · · · , xσ(m+n−1))
where σ ia an arbitrary permutation from Sm+n−1.
Consideration of a shuffle operad instead of a symmetric operad does not lead to any
loss of information, it just corresponds to a certain choice of representatives. If we present
elements of the free symmetric operad as decorated trees in the space, then to take a shuffle
representative of that tree, means to put it into the plane in a certain, canonical way (as it
is usually done with trees). The resulting plane tree will be an element of the free shuffle
operad.
In a more rigorous way, it could be formulated as follows: for two symmetric collections
P and Q, for a forgetful functor from symmetric to shuffle categories of collections, the
following holds:
(P ◦ Q)F = P F ◦sh QF ,
6
that is, the forgetful functor is monoidal.
So from now on our main object will be the free shuffle operad. We will use the well-
known presentation of elements of the free operad by decorated trees. Our rooted tree is a
connected graph of genus 0, whose vertices are labeled (by operations). Each vertex should
have at least one input and exactly one output, corresponding to edges. The ends of edges,
which are not vertices are called leaves, except for one. There should be exactly one edge,
which has neither a vertex nor a leave at one end, but instead a distinguished output vertex,
called the root of the tree. There are obvious simplest trees: the degenerate tree, containing
no vertices, and corolla, containing only one vertex.
If we want to put a tree on the plane, we just fix an increasing (from left to right) order
for the inputs in each vertex. It is a usual procedure to order inputs of a given vertex i as
follows: we compare the minimal leaves, one can reach from the given input, if this leave is
smaller, the input is smaller. They are situated in the plane accordingly. This particular
way of putting the tree on the plane corresponds to the choice of a shuffle representative.
We will think of the free shuffle operad FP , generated by a symmetric collection P,
as a set of planar representatives of trees corresponding to basic (corolla) operations of
the collection (in each arity P(n)), and all trees obtained from them by elementary shuffle
compositions. The latter can be defined in terms of drafting and relabeling the trees. We
call those planar representatives shuffle trees tree-monomials.
Definition. We say that the degree of a tree-monomial is a number of operations (vertices)
involved in it.
Now we can define the first essential for the Grobner bases theory notion of divisibility
for the tree-monomials.
Definition. A tree-monomial w is divisible by a tree-monomial v, if the underlying tree for
w (non-labeled) contains as a (non-labeled) subtree an underlying tree of v, and decorations
are compatible in the following way: the vertices which are common for both trees w and
v have the same labels and the leaves of v are decorated according to the following rule:
first is a leave through which we can get to the smallest leave of the tree w, second -- to the
second smallest, and so on. If v has such labeling, it divides w.
Since all tree-monomials of the free operad are obtained from corollas by elementary
shuffle compositions, it is easy to see that w can be obtained form v by a sequence of
elementary shuffle compositions.
Here we can see a difference with the notion of divisibility for monomials in the free
associative algebra, which are 'one-dimensional': w is divisible by v means, there exists r, l,
such that w = lvr. Shuffle trees are 'two-dimensional', so composition could be applied not
only from the left, or from the right.
Next important step is to define an ordering on the tree-monomials, which is compatible
with shuffle compositions.
More precisely, we say, that the ordering on tree-monomials from FP is admissible, if
first, the elements of smaller arity are smaller:
7
a ∈ FP (n), b ∈ FP (m), n < m =⇒ a < b1
and if for a, a′ ∈ FP (n) and b, b′ ∈ FP (m)
then for all elementary shuffle compositions
a ≤ a′ and b ≤ b′,
a ◦i,σ b ≤ a′ ◦i,σ b′.
We give here two examples of ordering, obeying this property.
I.Path-lexicographic ordering
To explain this ordering we encode a tree by a certain sequence consisting of words in
an alphabet corresponding to operations, and one permutation of the labels on the leaves.
Let a be a tree-monomial with n leaves. Associate to a a sequence of words (u1, · · · , un),
obtained in the following way: for the leave marked by i, the word ui consists of the sequence
(of operations), marking those vertices of the tree, one has to pass going from the root to
the leave marked i. It is, of course, unique. However the information contained in these
sequences is still not enough to distinguish tree-monomials. So we add an extra bit to the
sequence: a permutation σ ∈ Sn, which lists the labels on the leaves of the tree-monomial
(according to the definition it is a planar tree, so we can list leaves from left to right). The
described map from tree-monomials to sequences of the type (u1, · · · , un, σ) is an injection.
Hence if we define how to order such sequences, it will give us an ordering on the
tree-monomials. The order on the sequences will be degree-lexicographical:
first, the longer sequence is bigger: (u1, · · · , un, σ) > (u′
if sequences have the same length, we compare sequences element-wise, using the fol-
m, σ′) if n > m;
1, · · · , u′
lowing order on words and permutations.
Words are compared degree-lexicographically (from left to right).
If all words coincide, we compare permutations, for them we use reverse lexicographical
order: suppose for two permutations σ and σ′ the first from the left position, where entries
σi and σ′
i are different is i, then if σi > σ′
i, we say that σ < σ′.
This ordering is admissible. Of course it allows slight variations, we can use the reverse
lexicographic order instead of the usual one, compare words from right to left, and so on.
The choice of a correct ordering might be very essential for the conclusions one can
deduce from the Grobner bases computations.
Let us describe here also another type of possible ordering.
II.Forest-lexicographic ordering
Let a and b be tree-monomials with leaves labeled by arbitrary finite subsets I, I ′ ⊂ N
respectively. Consider first operations at the roots of a and b, which we call A and B
1 Note that this condition could be omitted, provided the d.c.c. property holds for the ordering. In fact,
even milder restriction on the ordering is actually needed. For example, if relations are homogeneous, then
the ordering, where elements of bigger arity are smaller is also admissible
8
respectively. We have k and l trees, where k and l defined by arities of the operations A
and B, starting at A and B, so:
a = A(a1, · · · , ak); b = B(b1, · · · , bl)
Leaves of the trees a and b become disjoint unions of the sets of leaves of trees a1, · · · , ak
and b1, · · · , bl which we denote Ij ⊂ I and I ′
j ⊂ I ′ respectively. Thus,
I = I1 ⊔ · · · ⊔ Ik, I ′ = I ′
1 ⊔ · · · ⊔ I ′
l,
where, as usual, for tree-monomoials we have minI1 < · · · < minIk and minI ′
1 < · · · < minI ′
l .
We compare a and b as follows. First, we compare sets I and I ′ of leaves of those trees.
If I and I ′ are arbitrary finite subsets of natural numbers, to compare them we first order
the numbers:
I = {i1 < · · · < ik}, I ′ = {i′
1 < · · · < i′
l},
and then order them (reverse)lexicographically: suppose 1 < 2 < · · · < n, and if jth
place is the first (from the left) where ij < i′
j then I > I ′. If I < I ′ then a < b.
When I = I ′, we compare A and B: if A < B, then a < b.
In the case I = I ′ and A = B, we say that a < b if: for the smallest j, such that pairs
J and aj < bj. It is a recursive
j, bj) do not coincide, either Ij < I ′
J or Ij = I ′
(Ij, aj) 6= (I ′
definition of the ordering.
Since we have now the ordered monoid of tree-monomials with shuffle composition, as a
multiplication, compatible with the ordering, we can define reduction and proceed with the
definition of Grobner basis. Then, after defining the concept of ambiguity and s-polynomial
we can formulate a criterion that a given set of elements of the free operad forms a Grobner
basis, and the Buchberger algorithm for constructing the Grobner basis.
The following steps in the Grobner bases theory are absolutely analogous to the case of
the bases in the free associative algebras.
If the tree w is divisible by v, there is a sequence of shuffle compositions which could be
applied to v to obtain w, denote it by mw,v. It is possible to apply the same sequence of
shuffle compositions to any other tree z with the same set of inputs as in v. The result of
this operation we denote by mw,v(z).
If we have two tree-polynomials (linear combination of tree-monomials) f and g and the
highest tree-monomials of them divide one another: ¯f is divisible by ¯g, then we define a
result of a reduction of f by g, as the polynomial
rg(f ) = f −
cf
cg
m ¯f ,¯g(g).
The reduction has the property that the resulting polynomial is smaller then the initial
one, due to the admissibility of the ordering w.r.t. the composition product.
9
Definition. The Grobner basis of the operadic ideal I is the set of tree-polynomials, such
that any element of I can be reduced to zero by elements of this set.
To give a criterion, which allows to check whether the given set of tree-polynomials is a
Grobner basis, we need two more definitions.
Definition. If the monomial is divisible by two other monomials, it is called an ambiguity.
It means the monomial allows two different applications of reduction. The polynomial
which we get if we subtract results of those two different applications of reductions is called
s-polynomial corresponding to the ambiguity.
Criterion If all ambiguities formed by the set of tree-polynomials are solvable (or there
are no ambiguities), which means that corresponding s-polynomials can be reduced to zero,
then this set of tree-polynomials forms a Grobner basis of the ideal generated by them.
The Buchberger algorithm for construction of the Grobner basis, from the given set of
tree-polynomials also works exactly like in algebra case.
We will use the following Priddy's criterion of Koszulity, which in operadic setting is
explained in [19].
Theorem 2.1. If an operad P has a presentation by generators and relations with a
quadratic Grobner basis (basis of degree two), then it is Koszul.
3 Deformations of the operad Ass
In the paper [9] there is a list of 10 a priori possibilities to introduce the notion of Hom --
associative algebra. Let us have an unary linear operation α : A → A on the linear space of
an algebra A. A priori, we can consider various generalised versions of associativity, defined
w.r.t. α. The list of possible identities [9] is the following:
I1 : (ab)α(c) = α(a)bc
I2 : (aα(b))c = a(α(b)c)
I3 : (α(a)b)c = a(bα(c))
II0 : (α(ab))c = a(α(bc))
II1 : (α(a)α(b))c = a(α(b)α(c))
II2 : (α(a)b))α(c) = α(a)(bα(c))
II3 : (aα(b))α(c) = α(a)(α(b)c)
III : α((ab)c) = α(a(bc))
10
III ′ : α(ab))α(c) = α(a)(α(bc))
III ′′ : (α(a)α(b))α(c) = α(a)(α(b)α(c))
The corresponding operads, which carry information on the identities I0 -- III ′′ we
denote OI0, · · · , OIII ′′ respectively. Our goal is to classify them with respect to the conflu-
ence property of their natural presentation by generators and relations, as a nonsymmetric
operad. Those presentations are as follows. Generators of these operads are the unary
operation α, denoted by an empty dot, and the binary operation µ denoted by a filled dot:
(cid:26)
, (cid:27).
The first one corresponds to the multiplication µ and the second to the linear map α.
The relations are respectively:
I1
I2
I3
II0
II1
,
-
-
-
-
11
II2
II3
III
III ′
III ′′
-
-
-
-
-
-
Theorem 3.1. Only one Hom -- deformation of the associative operad Ass is confluent,
namely the one given by relation II0.
Proof. Applying the Grobner basis theory for nonsymmetric operads [10, 8, 19], to these
presentations, we can check that only one of them satisfies the confluence property:
12
−→
↓
↓
→
→
Using the analogue of the Priddy criterion for nonsymmertic operads we deduce that
OII0 is Koszul.
Let us give an example of the case where the confluence does not hold. Consider, say,
an ambiguity formed by relation I2.
6=
↓
−→
−→
The verification of the confluence for all other operads in the list shows that the conflu-
ence does not hold for them.
4 Hom Gelfand-Dorfman algebras
In this paragraph we define a Hom -- deformation of the Gelfand-Dorfman (GD) algebra,
which have been defined, and its origin discussed in the introduction.
We define a Hom -- GD algebra as follows. It is a quadruple (A, ◦, [, ], α), where A is a
linear space, ◦ and [, ] are bilinear operations from A × A to A, and α is a linear space
homomorphism from A to A. They should satisfy the following identities:
13
(1)∗ (a ◦ b) ◦ α(c) − α(a) ◦ (b ◦ c) = (b ◦ a) ◦ α(c) − α(b) ◦ (a ◦ c) (Hom -- left-symmetric)
(2)∗ (a ◦ b) ◦ α(c) = (a ◦ c) ◦ α(b) (Hom -- right Novikov)
(3)∗ [a, b] = −[b, a]
(4)∗ [[a, b], α(c)] + [[c, a], α(b)] + [[b, c], α(a)] = 0
(5)∗ [c ◦ a, α(b)] − [c ◦ b, α(a)] + [c, a] ◦ α(b) − [c, b] ◦ α(a) − α(c) ◦ [a, b] = 0
Since GD algebra is a Lie algebra w.r.t. the operation [, ] and a Novikov algebra w.r.t.
the operation ◦, a Hom -- GD algebra will be naturally Hom -- Lie [15] and Hom -- Novikov [20],
but with the additional compatibility condition between two operations and α, defined by
(5)∗.
First, note that for the defined in such a way Hom -- GD algebras the following natural
connection with Hom -- Novikov algebras holds.
Theorem 4.1. The algebra (A, ◦α, [◦]α) with two operations, one of which is Hom -- Novikov,
i.e. satisfies the identities (1)∗, (2)∗, and anotheris obtained from it as [a, b]α = a◦α b−b◦αa,
is a Hom -- GD algebra.
Proof. It is known that structure obtained in such a way is Hom-Novikov, and of course
from the Hom left-symmetric identity (1)∗ it follows that the associated Lie bracket [◦]α
satisfies the Hom Jacobi identity. So we need to check only that (5∗) is satisfied. Indeed,
[c ◦ a, α(b)]α − [c ◦ b, α(a)]α + [c, a]α ◦ α(b) − [c, b]α ◦ α(a) − α(c) ◦ [a, b]α =
(c ◦α a) ◦α α(b) − α(b) ◦α (c ◦α a) − (c ◦α b) ◦α α(a) + α(a) ◦α (c ◦α b)+
(c ◦α a − a ◦α c) ◦α α(b) − (c ◦α b − b ◦α c) ◦α α(a) − α(c) ◦α (a ◦α b − b ◦α a).
Clearly, the right-hand side is canceled due to (1∗) and (2∗).
We can consider a particular case of Hom -- GD algebra defined by α : A → A, which is a
homomorphism of GD algebras. Namely, we can introduce a new algebra with operations
◦α and [, ]α, where a ◦α b = α(a ◦ b) and [a, b]α = α(a ◦ b). We prove below, that this is
indeed Hom -- GD algebra.
Theorem 4.2. Let (A, ◦, [, ]) be a GD algebra and α a GD algebra morphism, then Aα =
(A, ◦α, [, ]α), where a ◦α b = α(a ◦ b) and [a, b]α = α([a, b]α), is a Hom -- GD algebra.
Proof. Since α is a GD algebra morphism, i.e. α(a◦b) = α(a)◦α(b) and α([a, b]) = [α(a), αb],
it is easy to verify, that
14
1. (a ◦α b) ◦α α(c) = α2((a ◦ b) ◦ c),
2. [[a, b]α, α(c)]α = α2([[a, b], c]),
3. [a, b]α ◦α α(c) = α2([a, b] ◦ c),
4. [a ◦α b, c]α = α2([a, b] ◦ c).
Indeed, for example,
[a, b]α ◦α α(c) = α(α([a, b]) ◦ α(c)) = α2([a, b] ◦ c),
The first equality allows to prove identities (1∗) and (2∗) for Aα, the second one could
be used to verify the Jacobi identity (3∗). The last two guarantee identity (5∗). The
anticommutativity of [, ]α is immediate.
In terms of deformations of an operad, this special case of deformations has the feature,
that set of relations of the deformed operad Ah contains as a subset relations of A, which is
of course not the case in general. In general we have A = hx F (x)i, and Ah = hx, h F (x, h)i
such that F (x, h) coincide with F (x), when h is trivial. In the case when α is an algebra
homomorphism, we have A = hx F (x), G(x, h)i, where G(x, id) = F (x).
5 Conformal Lie operad
We will study here an operad CLie, associated to Lie conformal algebra structure, and derive
its Koszulity. There is another way to describe this structure (defined in the introduction),
by taking their generating function, called the lambda bracket structure. We let for any
a, b ∈ R
[aλb] =
λn
n!
a(n)b.
∞
Xn=0
Then we get an equivalent definition of a Lie conformal algebra [11], [12]:
Definition. We define a binary operation, called λ -- bracket, on the C[∂] module R: R ×
R → C[∂, λ] ⊗C[∂] R : a ⊗ b → [aλb], as an operation satisfying the following axioms.
I.(sesquilinearity)
II. (antisymmetry)
[∂aλb] = −λ[aλb]
[aλ∂b] = (∂ + λ)[aλb]
15
III. (generalised Jacobi identity)
[bλa] = −[a−λ−∂b]
[aλ[bµc]] − [bµ[aλc]] = [[aλb]λ+µc]
The Gelfand-Dorfman algebras considered above are closely related to the Lie conformal
algebras, as it was explained in the introduction. Here we give a formal proof. Namely,
having at hands Gelfand-Dorfman algebra with two operations (◦, [, ]) we can construct the
λ bracket as follows:
Proposition 5.1. The binary operation
[aλb] = [a, b] + ∂(a ◦ b) + λ(a ◦ b + b ◦ a)
satisfies axioms of λ bracket if and only if the operations (◦, [, ]) form a Gelfand-Dorfman
algebra.
Proof. We need to check that the above λ-bracket satisfies axioms
II. (antysymmetry)
[bλa] = −[a−λ−∂b]
and
III. (generalised Jacobi identity)
[aλ[bµc]] − [bµ[aλc]] = [[aλb]λ+µc]
if and only if for all a, b the multiplication is
right symmetric:
ass(a, b, c) = ass(a, c, b) : ass(a, b, c) = (ab)c − a(bc)
left Novikov:
and left Gelfand-Dorfman:
a(bc) = b(ac)
[ac, b] − [bc, a] + b[c, a] − a[c, b] − [a, b]c = 0.
To check I, we observe that
∂(ba) + λ(ba + ab) = −∂(ab) + (λ + ∂)(ab + ba).
16
To check II, each of three terms can be rewritten in the following way:
[[aλ, b]λ+µc] = [[a, b] + ∂(ab) + λ(ab + ba)λ+µ c] =
[[a, b]λ+µ c] + [∂(ab)λ+µ c] + λ[(ab + ba)λ+µ c] =
−(λ + µ)[abλ+µ c] + λ[(ab + ba)λ+µ c] =
−µ[abλ+µ c] + λ[baλ+µ c] =
= µ([a.b]c + ∂((ab)c) + (λ + µ)((ab)c + c(ab)))+
λ([b, a]c + ∂((ba)c + (λ + µ)((ba)c + c(ba))).
Analogously we rewrite [aλ[bµc]] and [bµ[aλc]].
Then we can consider their sum as a polynomial in ∂, µ, λ, so it is zero iff 'coefficients'
of all monomials in the variables ∂, µ, λ are zero.
The fact that the coefficient of ∂ is equal to zero means that the left GD-identity is
satisfied.
The coefficient of µ is:
−[ab, c] + [b, ca] + b[a, c] + [a, c]b + [a, b]c + c[a, b] − [a, bc + cb]
According to GD identity we can substitute
−[ab, c] = −[cb, a] + c[b, a] − a[b, c] − [a, c]b
into the previous expression, which after some cancelations gives us:
[bc, a] − [ac, b] = −a[c, b] + b[c, a] + [b, a]c,
which is again GD identity.
The same will happen with the coefficient of λ.
The coefficient of ∂2 is equal to zero iff left Novikov identity is satisfied.
The coefficients of µ2, λ2, λµ, λ∂, µ∂ are zero iff the right symmetric identity holds.
17
This close connection between those two structures motivates us to look at the properties
of the operad governing conformal Lie algebras.
Let us define the operad CLie, which will actually be a family of operads CLiek,
parametrised by the positive integers. The first axiom of the Lie conformal algebra says
that big enough products are zero. Hence we define operads CLiek, k ∈ Z, such that in the
Lie conformal algebras over each of them all operations (n) for n > k are zero.
Definition. The set of generating operations for the operad CLiek is formed by the oper-
ations {a, b}n,j = ∂(j)(a(n)b), for n, j = 0, 1, 2, · · · . The set of relations on this generators is
the following:
(F 3) {a, b}n,0 = −
k
(−1)n+j{b, a}n+j,j
Xj=0
(F 4) {a, {b, c}n,0}m,0 − {b, {a, c}m,0}n,0 =
m
Xj=0(cid:18) m
j (cid:19) {{a, b}j,0, c}n+m−j,0.
Theorem 5.2. The Conformal Lie operad CLie is Koszul.
Proof. To prove Koszulity we pass to the corresponding shuffle operad CLiesh and construct
a Grobner bases there, according to Section 2. The important step will be to choose a
suitable ordering of tree-monomials. Then we will apply the Priddy criterion of Koszulity
(see Theorem2.1).
The shuffle operad is generated by elements {a, b}n,j, a < b. The presentation of those
operations by tree-monomials looks as follows:
1
2
n, j
So the relations are obtained by substitution of (F 3) to (F 4), whenever we need to
change the order of inputs.
The operad CLiesh is given by relations:
(G4) {a, {b, c}n,0}m,0 −
k
Xj=0
(−1)n+j{{a, c}m,0, b}n+j,j =
18
m
Xj=0(cid:18) m
j (cid:19) {{a, b}j,0, c}n+m−j,0
for any a < b < c, n, m = 0, 1, 2, · · · .
In the language of tree-monomials the defining relations look as follows:
1
2
3
n, 0
m, 0
−(−1)n+j
1
3
2
1
2
3
m, 0
k
Xj=0
n+j,j
−
j, 0
m
Xj=0(cid:18) m
j (cid:19)
m + n − j, 0
We will choose now an appropriate admissible ordering on the tree-monomials, which
will lead to a quadratic Grobner basis.
Let us take a path-lexicographic ordering defined in section 2, but we should decide how
to order operations, marked in our case by pairs m, j. We order those pairs lexicographically
from right to left, i.e. {(m, i)} > {(n, j)}, whenever i > j, or i = j and m > n. We suppose
symbols themselves ordered as follows: 0 < 2 < 3 < · · · < 1 − − all as numbers, but 1 is
bigger than others. In particular, the operation (m, 0) is always smaller then (n, j), j 6= 0
and (m, i) is smaller then (n, 1) if i 6= 1.
Under the above ordering, when comparing tree-monomials in the relations, we conclude
that the highest tree monomial has the shape
1
3
2
m, 0
n + 1, 1
This follows from the inequalities:
{(m, 0), (m, 0)(n, 0), (m, 0)(n, 0), (123)} < {(n + j, j)(m, 0), (n + j, j), (n + j, j)(m, 0), (132)}
{(n + m − j, 0)(j, 0), (n + m − j, 0)(j, 0), (m + n − j, 0), (123)}
< {(n + j, j)(m, 0), (n + j, j), (n + j, j)(m, 0), (132)}.
We should find out now which ambiguities this set of highest words (when m and n are
ranging over 0,1,2,..) can form. For this we need to prove the following
19
Lemma 5.3. For arbitrary n, m = 0, 1, 2, · · ·
coincide.
the operations a(n)b and ∂(a(m)b) do not
Proof. For any fixed n we construct a free C[∂] module Mn, together with operations a(n)b
on it, which satisfy axioms (c1)-(c4), but such that a(n)b 6= ∂(a(m)b) for any m = 0, 1, 2, · · · .
This means for any n we will have a concrete conformal Lie algebra Mn, where a(n)b 6=
∂(a(m)b) for any m. The statement of the lemma will follow then.
As it has been pointed out for example in the Remark(2.2) in [12], if we define operations
satisfying (c3) and (c4) on the basis of a free C[∂]-module R, they can be extended uniquely
to the whole R, according to (c2). So we construct our examples having this in mind. All
our modules Mn will be free modules over C[∂], with the basis consisting of three elements,
which we denote a, b, c. So all what we need is to define operations satisfying (c3) and (c4)
on a, b, c, such that x(n)y 6= ∂(x(m)y) for any n, m and for some elements x and y in Mn.
Let us first construct appropriate modules for n = 0, 1 for illustration, and then give a
general construction.
In M0, define a(0)b = c, b(0)a = −c and the rest of the products of elements from the
basis a, b, c are equal to zero:
a(0)c = 0, c(0)a = 0,
b(0)c = 0, c(0)b = 0, and
z(i)w = 0 for all i > 0, and z, w ∈ M0.
Surely (c3) and (c4)are satisfied in this case. We choose a(0)b and b(0)a related in such
a way, that (c3) is satisfied, and (c4) is satisfied since with such a multiplications each
product of three elements should be zero.
Now ensure that the lemma condition is satisfied for n = 0, m = 0, 1, 2, · · · . Indeed,
c = a(0)b 6= ∂(a(m)b) = ∂(0) = 0
c = a(0)b 6= ∂(a(m)b) = ∂c
for any m = 1, 2, · · · , and
for m = 0.
These hold true since c is an element in the basis of the free C[∂] module, so c 6= 0 and
c 6= ∂c.
Take now n = 1 and construct the module M1. Suppose
a(1)b = c, b(1)a = c.
Then (c3), considered as a triangular system of equations, which express operations
a(n)b via operations with bigger numbers, will give us the equation
a(0)b = b(0)a − ∂(b(1)a).
20
The possible solution is a(0)b = 0 and b(0)a = ∂c. The rest of operations on the pairs
from a, b, c we again put to be zero. The axiom (c4) is again automatically satisfied.
The conditions of the lemma hold for n = 1 and any m:
c = a(1)b 6= ∂(a(m)b)
since ∂(a(m)b) = 0 for m = 0; ∂(a(m)b) = ∂c for m = 1 and ∂(a(m)b) = 0 for m = 2, 3, · · · .
Describe now the general construction of Mn. Let Mn be a free C[∂] module with the
basis {a, b, c}, endowed with the following operations:
a(i)b = 0,
for any i > n
a(n)b = c, b(n)a = (−1)n+1c
a(n−1)b = 0, b(n−1)a = (−1)n+1∂c
a(n−2)b = 0, b(n−2)a
· · ·
a(0)b = 0, b(0)a
All products of other pairs of elements of the basis are equal to zero.
We claim that
1). a(n)b 6= ∂(a(m)b) for any m. Indeed, a(n)b = c is not equal to one of the following:
∂(a(0)b) = ∂(0) = 0
∂(a(1)b) = ∂(0) = 0
· · ·
∂(a(n−2)b) = ∂(0) = 0
∂(a(n−1)b) = ∂(0) = 0
∂(a(n)b) = ∂c
The reason why c 6= ∂c and c 6= 0 is again that c is an element of the basis of the free
C[∂] module.
2). Axioms
(c3) a(r)b = −(−1)r+j
∂(j)(b(r+j)a)
∞
Xj=0
and (c4) are satisfied in Mn for arbitrary r.
21
Indeed, for 0 ≤ r ≤ n the equality
a(r)b = −(−1)r+j
∂(j)(b(r+j)a) =
k
Xj=0
−((−1)n−1∂(n−r−1)(b(n−1)a) + (−1)n∂(n−r)(b(n)a)) = 0
holds, when b(n−1)a = (−1)n+1∂c and b(n)a = (−1)n+1c.
For r > n all terms of the axiom (c3) are zero, so it is automatically satisfied.
The axiom (c4) also holds true for such a multiplications since the product (n) of any
two elements of the basis is either zero or c. If it is zero, the product (m) with the third
element is also zero.
If it is c, due to our definitions of the products, c, multiplied by
any other element of the basis, again give zero. So any combination of products of three
elements from the basis is zero.
Remark. We can give another proof of this lemma, which rely on the construction of
current Lie algebras. Namely, take the current Lie algebra C = G⊗A, where G is non-abelian
Lie algebra. Define on C a Lie conformal structure in the following way: a(0)b = [a, b], a(n)b =
0, n = 1, 2, · · · .a, b ∈ G. Then, since according to axiom (c2) (∂(n)a)(m)b = (−1)na(m−n)b,
we can find two elements A = ∂(n)a, B = b, for which A(n)B 6= ∂A(m)B for any n, m.
Indeed, (∂(n)a)(n)b = (−1)na(0)b, ∂((∂(n)a)(m)b) = (−1)n∂(a(m−n)b), which is equal to
(−1)n∂(a(0)b), if m = nand zero otherwise.
Since we have ensured, that the operations a(n)b and ∂(a(m)b) do not coincide for arbi-
trary m and n, the highest words of the Jacobi relations can not give rise to any ambiguity.
The highest words of those relations w.r.t our ordering are of the shape {{a, c}m,0, b}n+1,1,
depicted above by the left comb labeled tree. This means this presentation by generators
and relations of the operad CLie, with respect to the chosen ordering of tree-monomials,
is formed by a combinatorially free set of quadratic (degree two) tree monomials. As a
consequence, quadratic relations form a Grobner basis.
From this we can conclude that the operad is Koszul, due to the analogues of the Priddy
criterion for operads (see Theorem 2.1 in section 2).
6 Acknowledgements
We are grateful to Victor Kac for his suggestion to look at the question of Koszulity of
an operad associated to the axioms of Conformal Field Theory, and for many valuable
discussions. We also would like to thank V. Dotsenko for a number of useful remarks. The
work is supported by the Grant 9038 of the Estonian Scientific Council.
22
References
[1] D.Balavoine, Deformations of algebras over a quadratic operad, Contemp.Math. 202
(1997), 207 -- 234.
[2] A.A.Balinskii, S.P.Novikov, Poisson brackets of hydrodynamic type, Frobenious algebras
and Lie algebras, Dokl. Acad. Nauk SSSR 283 (1985), 1036 -- 1039; English transl. Soviet
Math. Dokl. 32 (1985), 228 -- 231.
[3] R.Borcherds,
Vertex
algebras,
Kac-Moody
algebras,
and
the Monster,
Proc.Natl.Acad.Sci. USA 83 (1986), 3068 -- 3071.
[4] D. Burde, K.Dekimpe, K. Vercammen, Novikov algebras and Novikov structures on Lie
algebras, Linear Algebra Appl. 429 (2008), No. 1, 31 -- 41.
[5] A.Fialowski, G.Mukherjee, A.Naolekar, Versal deformation theory of algebras over a
quadratic operad, arXiv1202.2967v1, 2012, 1 -- 32.
[6] I.M.Gelfand, I.Ya.Dorfman, Hamiltonian operators and related algebraic structures,
Funktional Anal. i Prilozen, 13 (1976), 13 -- 29.
[7] M.Goze and E.Remm, Lie-admissible algebras and operads. J. Algebra 273 no. 1,
(2004) 129 -- 152.
[8] V.Dotsenko, A.Khoroshkin, Groebner bases for operads, Duke Math. J. 153 (2010),
Number 2, 363 -- 396.
[9] Y.Fregier, A.Gohr, On Hom type algebras, arXiv0903.3393v2 (2009), 1 -- 24.
[10] E.Hoffbeck, A Poincar-Birkhoff-Witt criterion for Koszul operads, Manuscr. Math. 131
(2010), No. 1-2, 87 -- 110.
[11] V.Kac, Vertex algebras for beginners, AMS University Lecture Series, V.10, second
edition.
[12] V. Kac, D'Andrea Structure theory of finite conformal algebras, Selecta Mathematica,
New Ser. 4 (1998), 377 -- 418.
[13] M.Kontsevich, Y.Soibelman, Deformations of algebras over operads and the Deligne
conjecture, Math.Phys.Stud. 22 (2000), 255 -- 307.
[14] D.Larsson, S.Silvestrov, Quasi-Hom-Lie algebras, Central Extensions and 2-cocycle-like
identities, J. of Algebra 288 (2005), 321 -- 344.
[15] A.Makhlouf, S.Silvestrov, Hom-algebra structures, J. Gen. Lie Theory Appl 2 (2008),
51 -- 64.
[16] A.Makhlouf, S.Silvestrov, Hom-Algebras and Hom-Coalgebras, Journal of Algebra and
its Applications 9 (2010), no. 4, 553 -- 589.
[17] Yu.I.Manin, Some remarks on Koszul algebras and quantum groups, Ann. Inst. Fourier
(Grenoble) 37 (1987), no. 4, 191 -- 205.
23
[18] M.Markl, S.Shnider and J.Stasheff, Operads in algebra, topology and physics. Math-
ematical Surveys and Monographs, 96, American Mathematical Society, Providence,
RI, 2002.
[19] J.-L.Loday, B.Vallette, Algebraic operads, Springer, 2012.
[20] D.Yau, Hom-Novikov algebras, arxiv0909.0726v2, (2011), 1 -- 18.
[21] D. Yau, Hom-algebras via PROPs, arXiv:1103.5261 [math.RA] (2011).
[22] E.Zelmanov, On a class of local translation invariant Lie algebras, Soviet.Math.Dokl
35(1987), 216 -- 218.
Natalia Iyudu
Queens's University Belfast
Department of Pure Mathematics
University road, Belfast, BT7 1NN, UK
E-mail address:
[email protected]
Abdenacer Makhlouf
Universit´e de Haute Alsace,
Laboratoire de Math´ematiques, Informatique et Applications,
4, rue des Fr`eres Lumi`ere F-68093 Mulhouse, France
E-mail address:
[email protected]
24
|
1310.1557 | 1 | 1310 | 2013-10-06T08:42:32 | Algebras whose Coxeter polynomials are products of cyclotomic polynomials | [
"math.RA",
"math.RT"
] | Let A be a finite dimensional algebra over an algebraically closed field k. Assume A is basic connected with n pairwise non-isomorphic simplemodules. We consider the Coxeter transformation ?A as the automorphism of the Grothendieck group K0(A) induced by the Auslander-Reiten translation ? in the derived category Der(modA) of the module category modA of finite dimensional left A-modules. We say that A is an algebra of cyclotomic type if the characteristic polynomial ?A of ?A is a product of cyclotomic polynomials. There are many examples of algebras of cyclotomic type in the representaton theory literature: hereditary algebras of Dynkin and extended Dynkin types, canonical algebras, some supercanonical and extended canonical algebras. Among other results, we show that: (a) algebras satisfying the fractional Calabi-Yau property have periodic Coxeter transformation and are, therefore, of cyclotomic type, and (b) algebras whose homological form hA is non-negative are of cyclotomic type. For an algebra A of cyclotomic type we describe the shape of the Auslander-Reiten components of Der(modA). 1. | math.RA | math |
ALGEBRAS WHOSE COXETER POLYNOMIALS ARE PRODUCTS
OF CYCLOTOMIC POLYNOMIALS.
JOS´E A. DE LA PE NA
Abstract. Let A be a finite dimensional algebra over an algebraically closed field
k. Assume A is basic connected with n pairwise non-isomorphic simple modules. We
consider the Coxeter transformation φA as the automorphism of the Grothendieck
group K0(A) induced by the Auslander-Reiten translation τ in the derived category
Der(modA) of the module category modA of finite dimensional left A-modules. We
say that A is an algebra of cyclotomic type if the characteristic polynomial χA of
φA is a product of cyclotomic polynomials. There are many examples of algebras
of cyclotomic type in the representaton theory literature: hereditary algebras of
Dynkin and extended Dynkin types, canonical algebras, some supercanonical and
extended canonical algebras. Among other results, we show that: (a) algebras
satisfying the fractional Calabi-Yau property have periodic Coxeter transformation
and are, therefore, of cyclotomic type, and (b) algebras whose homological form hA
is non-negative are of cyclotomic type. For an algebra A of cyclotomic type we
describe the shape of the Auslander-Reiten components of Der(modA).
1. Introduction.
Assume throughout the paper that k is an algebraically closed field. Let A be a
triangular finite dimensional k-algebra, that is, we assume that the quiver QA of A
has no oriented cycles. Hence A has finite global dimension. In this case, a theorem
of Happel [11] asserts that the bounded derived category Der(A) = Derb(modA) of
the category modA of finite dimensional (left) A-modules has Serre duality of the
form
Hom(X, Y [1]) = Hom(Y, τ X)
where τ is a self-equivalence of Der(A). In particular, Der(A) has almost-split tri-
angles and the equivalence τ serves as the Auslander-Reiten translation of Der(A).
In this setting, the Grothendieck group K0(A), formed with respect to short exact
sequences, is naturally isomorphic to the Grothendieck group of the derived cate-
gory, formed with respect to exact triangles. The Coxeter transformation φA is the
automorphism of the Grothendieck group K0(A) induced by the Auslander-Reiten
translation τ in the derived category Der(A). The characteristic polynomial χA(T )
of φA is called the Coxeter polynomial of A, and denoted simply χA. It is a monic
and self-reciprocal polynomial, i.e. χA(T ) = T nχA( 1
T ) = a0 + a1T + a2T 2 + . . . +
an−2T n−2 + an−1T n−1 + anT n ∈ Z[T ], where a0 = 1 = an. Sometimes, for the sake of
clarity, we write ai(A) for these coefficients.
1
2
JOS ´E A. DE LA PE NA
Consider the roots λ1(A), . . . , λn(A) of χA, the so called spectrum of A. There is
a number of measures associated to the absolute values λ for λ in the spectrum
Spec(φA) of A. Three important measures are the following: the spectral radius
of A is defined as ρA = max{λ : λ ∈ Spec(φA)}, the Mahler measure of φA is
M(φA) = Qn
Pi=1λi(A). In
Section 2 we show that always e(φA) ≥ n. Due to a celebrated Kronecker's theorem,
we prove the following result.
i=1 max{1,λi} and the energy function of φA is e(φA) =
n
Theorem 1. Let A be a triangular algebra. The folllowing statements are equivalent:
(1) λi(A) ≤ 1, for all 1 ≤ i ≤ n;
(2) λi(A) = 1, for all 1 ≤ i ≤ n;
(3) ρA = 1;
(4) M(φA) = 1;
(5) e(φA) = n
(6) χA(T ) factorizes as product of cyclotomic polynomials.
An algebra A satisfying these equivalent conditions is said to be of cyclotomic type.
The following finite dimensional algebras are known (or shown) to be of cyclotomic
type:
(1) hereditary algebras of finite or tame representation type;
(2) tensor products of algebras of cyclotomic type;
(3) quotient algebras An(3) of the linear quiver
1
x
/ 2
x
x
/ n
/ ···
with relations x3 = 0, for n even and many instances of n odd;
(4) all canonical algebras;
(5) (some) extended canonical algebras;
(6) (some) algebras whse module category is derived equivalent to categories of
coherent sheaves;
(7) algebras whose homological bilinear form is non-negative.
One of the purposes of this work is to show that several of the common properties of
these classes of algebras are a consequence of the cyclotomic condition, see Sections
2, 5 and 6 for details.
Recall that a triangular algebra A is said to be p
q -Calabi-Yau for integers q ≥ 1
and p ∈ Z if Sq = [p] in the derived category Der(A), where S = τ ◦ [1]. We show
that algebras satisfying the fractional Calabi-Yau property have periodic Coxeter
transformation and are, therefore, of cyclotomic type, see Sections 2 and 5.
Along this work we explore the structure of the Auslander-Reiten quiver ΓA of the
derived category of an algebra A of cyclotomic type. We introduce the class quiver
[ΓA] of A formed by the quotients [C] of components C of ΓA, obtained by identifying
/
/
/
ALGEBRAS OF CYCLOTOMIC TYPE
3
X, Y ∈ C if [X] = [Y ] in K0(Der(A)). Among other results we prove the following
theorem.
Theorem 2. Let A be a triangular algebra with n non-isomorphic simple modules
m be an irreducible decomposition of its Coxeter polynomial. Let
ΓA be the Auslander-Reiten quiver of the derived category Der(A). Then the following
holds:
and χA =Qm∈M Φe(m)
(a) if A satisfies the fractional Calabi-Yau property then φA is periodic;
(b) every component of the class quiver [ΓA] is a tube if and only if φA is periodic.
If φA is periodic then A is of cyclotomic type and
(c) the period is l.c.m.{φ(m) : m ∈ M}, where φ(−) is Euler's totient function;
(d) every component of Γ is either a tube ZA∞/(p) of finite period p where p = φ(m)
for some m ∈ M or of the form Z∆ for ∆ a Dynkin or extended Dynkin diagram or
of one of the shapes A∞, A∞
∞ or D∞. Moreover,
s
(e) if p1, . . . , ps are the periods > 1 of non-homogeneous tubes in [Γ] then
pi ≤ n.
Along the paper, we freely use specialized terminology: tame algebras, wild al-
gebras, linear growth, polynomial growth, pg-critical, and others. We refer read-
ers to the list of references where all these and related concepts are explained, see
[1, 25, 32, 35]. Results contained in this work were presented in conferences at Banff,
Canada (2010), Bielefeld, Germany and Guanajuato, M´exico (2012).
Pi=1
2. The spectrum of algebras of cyclotomic type
2.1.
If the spectrum of A lies in the unit disk, then all roots of χA(T ) lie on the unit
circle, hence A has spectral radius ρA = 1. More precisely, we recall the following
famous theorem of Kronecker.
Theorem. [18] Let f be a monic integral polynomial whose spectrum is contained in
the unit disk. Then all roots of f are roots of unity or 0. Equivalently, f factors into
cyclotomic polynomials and T m, for some integer m ≥ 0.
2.2. We recall some facts on cyclotomic polynomials.
The n-cyclotomic polynomial Φn is inductively defined by the formula
The roots of Φn are primitive φ(n)-roots of unity, where φ(−) denotes Euler's totient
function.
T n − 1 = Qdn
Φd(T ).
The Mobius function is defined as follows:
µ(n) =(0
(−1)r
if n is divisible by a square
if n = p1, . . . pr is a factorization into distinct primes.
4
JOS ´E A. DE LA PE NA
A more explicit expression for the cyclotomic polynomials is given by:
Lemma. For each n ≥ 2, we have
vµ(d)
n/d
Φn = Q1≤d<n
dn
In the lemma, we set vn = 1 + T + T 2 + . . . + T n−1. Note that vn has degree n − 1.
2.3. A path algebra k∆ is said to be of Dynkin type if the underlying graph ∆ of ∆
is one of the ADE-series, that is, of type, An, Dn, for some n ≥ 1 or Ep, for p = 6, 7, 8.
The corresponding factorization of the Coxeter polynomial χk∆ is as follows.
Dynkin
star
v-factorization
cyclotomic
Coxeter
(cid:3)
type
An
symbol
[n]
vn+1
Dn
E6
E7
E8
[2, 2, n − 2]
[2, 3, 3]
[2, 3, 4]
[2, 3, 5]
v2 (v2vn−2)
(v2vn−2)vn−1
v2v3(v3)
(v3)v4v6
v2v3(v4)
(v4)v6v9
v2v3v5
v6v10v15
v12
factorization
number
Φd
Ydn,d>1
v2(n−1) Φ2 Yd2(n−1)
d6=1,d6=n−1
n + 1
Φd 2(n − 1)
Φ3Φ12
Φ2Φ18
Φ30
12
18
30
v18
v30
In the column 'v-factorization', we have added some extra terms in the nominator
and denominator which obviously cancel.
Inspection of the table shows the following result:
Proposition. Let k be an algebraically closed field and A be a connected, hereditary k-
algebra which is representation-finite. Then the Coxeter polynomial χA(T ) determines
A up to derived equivalence.
2.4. Let A be the path algebra of a hereditary star [p1, . . . , pt] with respect to the
standard orientation, for instance
◦
◦
◦
A✂✂✂✂✂
◦
◦
◦
?⑦⑦⑦⑦
/ ◦
/ ◦.
[2, 3, 3, 4] :
◦
O
O
?
o
o
O
O
/
/
A
/
/
ALGEBRAS OF CYCLOTOMIC TYPE
5
Since the Coxeter polynomial χA does not depend on the orientation of A we will
denote it by χ[p1,...,pt]. It follows from [25, Prop. 9.1] that
(1)
χ[p1,...,pt] =
vpi (T + 1) − T
t
Yi=1
vpi ! .
vpi−1
t
Xi=1
In particular, we have an explicit formula for the sum of coefficients of χ[p1,...,pt] as
follows:
n
t
ai = χ[p1,...,pt](1) =
pi 2 −
Yi=1
(1 −
1
pi
)! .
t
Xi=1
Xi=0
This special value of χ = χ[p1,...,pt] has a specific mathematical meaning: up to the
i=1 pi this is just the orbifold-Euler characteristic of a weighted projective
line X of weight type (p1, . . . , pt). Moreover,
factor Qt
(1) χ(1) > 0 if and only if the star [p1, . . . , pt] is of Dynkin type, correspondingly
the algebra A is representation-finite.
(2) χ(1) = 0 if and only if the star [p1, . . . , pt] is of extended Dynkin type, corre-
spondingly the algebra A is of tame (domestic) type.
(3) χ(1) < 0 if and only if [p1, . . . , pt] is not Dynkin or extended Dynkin, corre-
spondingly the algebra A is of wild representation type.
To complete the picture, we consider the cyclotomic factorization of the extended
Dynkin quivers without oriented cycles. Observe that, in one case, the Coxeter poly-
nomial depends on the orientation: If p (resp. q) denotes the number of arrows in
clockwise (resp. anticlockwise) orientation (p, q ≥ 1, p + q = n + 1), that is, the quiver
has type A(p,q), the Coxeter polynomial is χ(p,q) = (T − 1)2 vpvq..
The next table displays the v-factorization of extended Dynkin quivers.
extended Dynkin type star symbol weight symbol Coxeter polynomial
Ap,q
Dn, n ≥ 4
E6
E7
E8
-
[2,2,n-2]
[3, 3, 3]
[2, 4, 4]
[2, 3, 6]
(p, q)
(2, 2, n − 2)
(2, 3, 3)
(2, 3, 4)
(2, 3, 5)
(T − 1)2vp vq
(T − 1)2v2
2vn−2
(T − 1)2v2v2
(T − 1)2v2v3v4
(T − 1)2v2v3v5
3
The above remark on representation-finite hereditary algebras extends to the tame
hereditary case. That is, the Coxeter polynomial of a connected, tame hereditary k-
algebra A (k algebraically closed) determines the algebra A up to derived equivalence.
This is no longer true for wild hereditary algebras, not even for trees.
6
JOS ´E A. DE LA PE NA
A CA.
A(X, Y ).
A + C−T
T ) = χA(T ). More-
2.5. Recall that χA is a self-reciprocal polynomial, that is, T nχA( 1
over, it was shown in [25] that χA(−1) is the square of an integer.
These conditions prevent many polynomials of being Coxeter polynomials of an
algebra. For instance, the cyclotomic polynomials Φ4, Φ6, Φ8, Φ10. Moreover, the
polynomial f (T ) = T 3 + 1, which is the Coxeter polynomial of the non simply-laced
Dynkin diagram B3, does not appear as the Coxeter polynomial of a triangular algebra
over an algebraically closed field, despite the fact that f (−1) = 0 is a square, see [25].
2.6. The homological quadratic form hA is the symmetrization of the inverse Cartan
matrix CA associated to the algebra A, that is hA(x) = xT (C−1
A )x for x ∈
K0(A). Write hu, vi = uT C−1
A v which yields a (non-symmetric) non-degenerated
bilinear form, that is, if hu, vi = 0 for a fixed u and all v then u = 0. Mooreover
hx, φA(y)iA = −hy, xiA. Therefore φA = −C−t
For any two A-modules X, Y , with classes dim X, dim Y in the Grothendieck
∞
(−1)jdimkExtj
Pj=0
Consider the quadratic form hA(x) = hx, xi. It is shown in [33] that properties of
hA characterize the representation behavior of A. Indeed, hA is positive definite (resp.
non-negative of corank 1) if and only if A is derived equivalent to the path algebra
k∆, where ∆ is a quiver with underlying graph a Dynkin diagram (resp. extended
Dynkin diagram). There are algebras A with hA non-negative of arbitrary corank.
Lemma. If u (resp. v) is an eigenvector of φA with eigenvalue λ (resp. µ) such that
λ 6= µ−1 then hu, vi = 0. Moreover, if λ 6= −1 then hA(u) = 0.
Proof. Assume C−t
A v. Then
A u = λµvtC−t
A u
group K0(A) we get hdim X, dim Y iA =
A u = −λC−1
A u and C−t
A v = −µC−1
hv, ui = vtC−t
A u = −λvtC−1
and the first claim follows. Moreover if λ2 6= 1 then hA(u) = 0. The case λ = 1 yields
hu, ui + hu, ui = 0 or equivalently, hA(u) = 0.
2.7. For an algebra A and a finite-dimensional right A-module M we call
(cid:3)
M k (cid:21)
A[M] =(cid:20) A 0
the one-point extension of A by M, see [35]. This construction provides an order to
deal inductively with triangular algebras.
For A = B[M], Happel's long exact sequence [11] relates the Hochschild cohomol-
ogy groups H i(A) and H i(B) in the following way:
0→ H 0(A)→ H 0(B)→ EndB(M)/k→ H 1(A)→ H 1(B)→ Ext1
B(M, M)→ H 2(A)→· · ·
Therefore, if M is exceptional we get H i(A) = H i(B), for i ≥ 0, and moreover, the
cohomology rings H∗(A) and H∗(B) are isomorphic. More general, if A is accessible
from B, that is, there exists a chain of algebras A1 = B, A2, . . . , As = A such that
ALGEBRAS OF CYCLOTOMIC TYPE
7
Ai+1 is a one-point extension or coextension of Ai by an exceptional Ai-module Mi,
then there is a ring isomorphism H∗(A) ∼= H∗(B). In particular, if A is accessible
(from k) then H i(A) = 0 for i > 0 and H 0(A) = k.
n
An important expression for the linear term of χA(T ) was shown in [11] as: a1 =
(−1)idimkH i(A).
Pi=0
2.8. Accessible algebras. Following [27], for an algebra B we say that A is accessible
from B if there is a sequence B = B1, B2, . . . , Bs = A of algebras such that each Bi+1
is a one-point extension (resp. coextension) of Bi for some exceptional Bi-module Mi.
As a special case, a k-algebra A is called accessible if A is accessible from k. By
construction, each accessible algebra A is connected, moreover the indecomposable
projective A-modules can be arranged to form an exceptional sequence (P1, . . . , Pn),
that is, Exti
A(Pj, Ps) = 0 for every j > s and i ≥ 0. In particular, the quiver of an
accessible algebra A has no loops or oriented cycles (we say that A is triangular) and
therefore A has finite global dimension.
If not stated otherwise we shall assume that an algebra A is defined over k and
connected, that is, its quiver is connected. Many well-known examples of algebras are
accessible: hereditary algebras of tree type, more generally tree algebras, canonical
algebras with three weights, poset algebras without crowns, representation-finite alge-
bras with vanishing first Hochschild cohomology of global dimension ≤ 2. Moreover,
extended canonical algebras with three weights and certain supercanonical algebras
also are accessible. The construction of towers of algebras A = An, An−1, . . . , A1 = k
where each algebra Ai+1 is a one-point extension or coextension of Ai by an ex-
ceptional module Mi, for i = 1, . . . , n − 1 allows to perform informative inductive
procedures. For instance, every accessible algebra A has vanishing Hochschild coho-
mology H i(A) = 0, for i > 0 and H 0(A) = k. Also, for any tree algebra A such
an inductive procedure yields that the Coxeter polynomial takes values 0 or 1 when
evaluated at −1.
2.9. The Calabi-Yau property. Let T be a triangulated category with Serre func-
tor S, that is, we have functorial isomorphisms
DHom(X, Y ) ∼= Hom(Y, SX)
for X, Y ∈ T.
Denote by [1] the suspension functor on T. If τ is the Auslander-Reiten translation
in T, then S = τ ◦ [1]. The category T is said to be p
q -Calabi-Yau for integers q ≥ 1
and p ∈ Z if Sq = [p]. If additionally q is chosen minimal, then we say that T has
Calabi-Yau dimension p
q . Notation: CY-dim T = p
q .
Example: Let E be a smooth elliptic curve. Then the category cohE of coherent
sheaves on E has Serre duality in the form DExt1(X, Y ) = Hom(Y, X), yielding τ = 1
8
JOS ´E A. DE LA PE NA
and S = [1]. Thus cohE is Calabi-Yau of dimension 1
is CY.
1. In this case we say that cohE
The following is well-known.
Lemma. Assume that the triangulated category T is p
q -Calabi-Yau. Let σ be the k-
linear automorphism of the Grothendieck group K0(T) induced by the Serre functor
S. Then the following holds:
(a) σ2q = 1;
(b) φ2q = 1, where φ = −σ is the Coxeter transformation.
Proof. Use that the suspension [1] induces on K0(T) the map −id.
2.10. We say that an algebra A is p
q -Calabi-Yau (or A satisfies the CY-property)
if Der(mod A) has the corresponding property. The remarks above show that if A
satisfies the CY-property then φA is periodic, that is, φr
A = id, for some r ≥ 1. More
generally, we have the following result.
(cid:3)
Theorem. The following conditions are equivalent for A:
(1) φA is periodic;
(2) A is of cyclotomic type and φA is diagonalizable;
m for some subset M of N and integers e(m) ≥ 1,
(3) assume that χA =Qm∈M Φe(m)
then φA has period l.c.m.{φ(m) : m ∈ M}.
Proof. (1) implies (2): Consider the Jordan block decomposition Jn1(µ1)⊕. . .⊕Jns(µs)
of φA. The r-th power of a block Jm(µ)r is the identity if and only if m = 1 and
µr = 1. Hence if φA is periodic then n1 = . . . = ns = 1 and A is of cyclotomic type.
(2) implies (1): Conversely, assume P is an invertible matrix such that P φAP −1 =
AP −1 = (P φAP −1)r = idn
s = 1, for all s = 1, . . . , n. Then P φr
diag(λ1, . . . , λn) and λr
A = P −1P = idn.
and φr
(3) implies (1) is clear.
(1) implies (3): consider φA as an operator of the n-
dimensional complex vector space V . For λ an eigenvalue of φA there is a set of
eigenvectors B(λ) of λ such that Sλ B(λ) is a basis of V . Moreover, for λ there is
some m(λ) ∈ M such that Φm(λ)(λ) = 0. In particular, λφ(m(λ)) = 1.
Set r = l.c.m.{φ(m) : m ∈ M} then for u ∈ B(λ) we have r = φ(m(λ))r′ and
A(u) = (λφ(m(λ)))r′u = u, that is, φr
φr
A = idV .
(cid:3)
3. Measures for χA
3.1. The spectral radius of a Coxeter transformation. Let A be a basic finite
dimensional k-algebra with k an algebraically closed field. We assume A = kQ/I for
a quiver Q without oriented cycles and I an ideal of the path algebra. The following
facts about the Coxeter transformation φA of A are fundamental:
(i) Let S1, . . . , Sn be a complete system of pairwise non-isomorphic simple A-
modules, P1, . . . , Pn the corresponding projective covers and I1, . . . , In the injective
ALGEBRAS OF CYCLOTOMIC TYPE
9
envelopes. Then φA is the automorphism of K0(A) defined by φA[Pi] = −[Ii], where
[X] denotes the class of a module X in K0(A).
(ii) For a hereditary algebra A = kQ, the spectral radius ρA determines the repre-
sentation type of A in the following manner:
• A is representation-finite if 1 = ρA is not a root of the Coxeter polynomial.
• A is tame if 1 = ρA ∈ Spec(φA).
• A is wild if 1 < ρA. Moreover, if A is connected, Ringel [36] shows that the
spectral radius is a simple root of χA. Then Perron-Frobenius theory yields a vector
y+ ∈ K0(A) ⊗Z R with positive coordinates such that φAy+ = ρAy+.
3.2. The energy of a matrix. Let A be an algebra and λ1, . . . , λn be the eigenvalues
of the Coxeter transformation φA = (fst)1≤s,t≤n. We may assume that λ1 ≤ λ2 ≤
. . . ≤ λn = ρA.
We recall that the singular eigenvalues of φA are the non-negative square roots of
the eigenvalues of the symmetric matrix φT
AφA, which is positive semidefinite. We
may assume that the singular eigenvalues are
The following holds,
0 ≤ σ1 ≤ σ2 ≤ . . . ≤ σn
n
Pi=1λi2 ≤ tr (φt
(a) σi ≥ λi, for all i = 1, . . . , n
σ2
Pi=1
(b)
i =
φA2 =
AφA) =
normal. We write
n
n
n
Ps,t=1fst2, equality holds if and only if φA is
Xi=1
Xs,t=1
fst2
σ2
i =
n
Theorem. The following inequatilities hold:
n ≤ e(φA) ≤ √nφA ≤ nρA
Moreover, A is of cyclotomic type if and only if any of the equalities hold.
s , λ−1
Proof. We start considering the first lower bound. Let λ be an eigenvalue of φA.
Since φA is invertible, λ 6= 0. Since the characteristic polynomial χA is real (resp.
self-reciprocal) then λ (resp. λ−1) is another root of χA. Consider the set R(s) =
{λs, λs, λ−1
s } the roots associated with λs. For λs = as + ibs 6= 0, for real numbers
as, bs, the cardinality r(s) of R(s) may be 4 (in case bs 6= 0), or 2 (in case bs = 0 and
as 6= 0, 1) or 1 (in case λs = 1).
We shall prove that µs := Pµ∈R(s) µ ≥ r(s) with equality if and only if λs has
Indeed, write λs = c(a + ib) 6= 0 in polar form, that is a, b, c are real numbers with
c > 0 and a2 + b2 = 1. In case r(s) = 2 then µs = c + c−1, which is an increasing
modulus one. The wanted bound follows.
10
JOS ´E A. DE LA PE NA
function at the interval [1,∞). Therefore with a minimal value 2 at 1. In case r(s) = 4
then
µs = (c(a + ib) + c−1(a + ib)) + (c(a − ib) + c−1(a − ib)) ≥ 4
with equality if and only if c = 1.
Next, an application of Schwartz inequality yields,
Moreover,
e(A)2 = (
n
Xi=1
λi)2 ≤ n
n
Xi=1
λi2 = nφA2
n
Finally, if A is of cyclotomic type then all λi have modulus one. Conversely,
if
e(A) = n, then µs = Pµ∈R(s) µ = r(s) and λs = 1, for all s = 1, . . . , n. If it is the
case that e(A) = √nφA then
φA2 =
λi2 ≤ nρ2
A
Xi=1
n
(
Xi=1
λi)2 = n
λi2
n
Xi=1
which implies that all λi have a common value. Since the set of eigenvalues of φA is
closed under multiplicative inverses, then this common value may only be 1. In any
case, A is of cyclotomic type.
(cid:3)
Remark: Alternatively, the first inequality in the Theorem above may be seen as
a consequence of the arithmetic mean-geometric mean inequality. Indeed, for the set
of non-negative numbers {λ1, . . . ,λn} we get
λi!
= n(det (φA))
e(A) = n
n = n.
Pi=1λi
n ≥ n n
Yi=1
1
n
n
1
3.3. Mahler measure. Given a Laurent polynomial P (z) with integer coefficients,
its Mahler measure M(P ) is defined as the geometric mean of the function P over
the real circle, that is,
M(P ) = exp(Z 1
0
log(P (e2πit))dt
For a polynomial in one variable, Mahler obtained the more elementary expression
for the measure given in the Introduction, see [29].
Kronecker's result implies that M(P (T )) = 1 precisely when P (T ) is a product
of cyclotomic polynomials and a monomial of the form T m. Rather little is known,
ALGEBRAS OF CYCLOTOMIC TYPE
11
however, about values of the Mahler measure near 1. In 1933, D. H. Lehmer found
that the polynomial
T 10 + T 9 − T 7 − T 6 − T 5 − T 4 − T 3 + T + 1
has Mahler measure µ0 = 1.176280..., and he asked if there exist any smaller values
exceeding 1. This question of determining whether 1 is a limit point for the Mahler
measure is known as Lehmers problem.
In a forthcoming publication we show that for an accessible algebra A not of cyclo-
tomic type there is a convex subcategory B of A satisfying the following properties:
(a) B is minimal not of cyclotomic type, that is, if C is any proper convex subcat-
egory of B, then C is of cyclotomic type;
(b) the Mahler measure of B is M(χB) ≥ µ0.
3.4. Proof of Theorem 1. The equivalence of (1), (2), (4) and (6) follows from
Kronecker's theorem. The remaining equivalences follow from (3.2).
(cid:3)
4. On the decomposition of the Coxeter polynomial of an algebra of cyclo-
tomic type
4.1. Let A be an algebra with n vertices whose Coxeter polynomial decomposes as
χA = Qm∈M Φm(T )e(m) for some subset M of N and integers e(m) ≥ 1. Several
aiT i and of the polynomials Φm follow, for
conditions on the coefficients of χA =
n
Pi=0
instance:
Euler's totient function;
e(m)φ(m) = n, which follows from the degree calculation, where φ(−) is
(1) Pm∈M
(2) if 1 ∈ M then e(1) is an even number, which follows from the fact that Φm for
m 6= 1, as well as χA(T ), are self-reciprocal polynomials, but Φ1(T ) = T − 1 is not
so, while Φ1(T )2 = T 2 − 2t + 1 is self-reciprocal;
(3) Pm∈M
e(m)µ(m) = −a1, which follows from the fact that the linear coeffi-
cient of Φm(T ) is a1(Φm(T )) = −µ(m). Use that a0(Φm(T )) = 1 for m 6= 1 and
a0(Φ1(T )e(1)) = 1.
Theorem. Let A be an algebra whose Coxeter polynomial factorizes as χA(T ) =
Qm∈M Φm(T )e(m) for some subset M of N and integers e(m) ≥ 1. Set ǫ(m) = 0 if
m ∈ M and ǫ(m) = 1 if m /∈ M. For any prime number p, let M(p) (resp. M′(p))
be the set of those m ∈ M of the form m = ps (resp. m = 2ps) for some 1 ≤ s. Let
f (p) = Pps=m∈M (p)
(a) χA(1) = ǫ(1) Qm∈M (p) pe(m) = ǫ(1) Qp prime pf (p)
(b) χA(−1) = ǫ(2) 2e(1)Qm∈M ′(p) pe(m) = ǫ(2) 2e(1)Qp prime pf ′(p). Moreover, if this
number is not zero then, for each prime 2 ≤ p, the number f′(p) is even.
e(m) and f′(p) = P2ps=m∈M ′(p)
e(m). Then
12
JOS ´E A. DE LA PE NA
Proof. (a) According to [34], the evaluation Φm(1) = 1 if m 6= 1 and m 6= ps, for all
prime p. Otherwise, Φ1(1) = 0 or Φps(1) = p. Then the evaluation
χA(1) = Ym∈M
Φm(1)e(m) = ǫ(1) Ym∈M (p)
pe(m)
(b) According to [34], the evaluation Φm(−1) yields Φm(−1) = −2 if m = 1,
Φm(−1) = 0 if m = 2, Φm(−1) = p if m = 2ps with p a prime number and s ≥ 1 and
Φm(−1) = 1 otherwise. Therefore
χA(−1) = Ym∈M
Φm(−1)e(m) = ǫ(2) 2e(1) Ym∈M ′(p)
pe(m)
Assume χA(−1) = r2 > 0 for some integer r, then for any prime 2 < p
P
e(m)
pe(m) = p
m∈M ′(p)
Ym∈M ′(p)
is an even power of p. For p = 2, use additionally that e(1) is an even number. The
claim follows.
(cid:3)
5. More examples
5.1. Tensor products of polynomials and algebras. Given two monic integral
polynomials f (T ) with roots a1, . . . , an and g(T ) with roots b1, . . . , bm, the tensor
product of f (T ) and g(T ) is
n
m
f (T ) ⊗ g(T ) =
Xi=1
Xj=1
(T − aibj)
By [9], the polynomial f (T ) ⊗ g(T ) is integral of degree degree(f (T ))degree(g(T )).
An important problem is to find the irreducible decomposition of f (T ) ⊗ g(T ).
We are interested in calculating Φ6(T ) ⊗ Φm(T ) for m ∈ Z. It is easy to show that
Φm(T ) ⊗ Φr(T ) = Φmr(T ), whenever m, r are relative prime.
We observe the following facts:
• Φ6(T ) ⊗ Φ2(T ) = Φ3(T )
• Φ6(T ) ⊗ Φ3(T ) = Φ1(T )2Φ2(T )2
• Φ6(T ) ⊗ Φ4(T ) = Φ12(T )
• Φ6(T ) ⊗ Φ6(T ) = Φ2(T )2Φ3(T )
• Φ6(T ) ⊗ Φ12(T ) = Φ4(T )2Φ12(T )
As it is easy to verify. For instance, we check the two last claims. Let a and b be
the roots of Φ6(T ) = T 2 − T + 1. The roots of Φ6(T ) ⊗ Φ6(T ) are a2 = a − 1 = −b,
b2 = −a and ab = 1 = ba. Let ci = exp( 2πi
12 ), for i = 1, 5, 7, 11, be the roots of
Φ12(T ) = T 4 − T 2 + 1. The roots of Φ6(T ) ⊗ Φ12(T ) are ac1, ac7 and bc5, bc11 which
are roots of Φ4(T ) and ac5 = c7, ac11 = c1, bc1 = c11, bc7 = c5.
According to [9], for m = 2r3sq, where 2 and 3 do not divide q, we have:
ALGEBRAS OF CYCLOTOMIC TYPE
13
• Φ6(T ) ⊗ Φm(T ) = Φm(T )2, in case r > 1 and s > 1, other products have more
complex expressions.
5.2. As observed by Ladkani, see [20] and also [15], at the level of algebras we have:
Proposition. Let A and B be algebras of cyclotomic type. Then A⊗ B is an algebra
of cyclotomic type with
As a particular example, the algebra R2n with 2n vertices and whose quiver is given
as
χA⊗B = χA ⊗ χB
1
1′
2
/ 2′
3
/ 3′
· · ·
/ · · ·
/ n
/ n′
with all commutative relations has Coxeter polynomial
χ2n = χAn ⊗ χA2 = vn+1 ⊗ v3
which is a product of cyclotomic polynomials, see [19].
5.3. Repetitive algebras. Let A be a triangular algebra with n vertices. Consider
the double repetitive algebra
A(2) =(cid:18)A D(A)
A (cid:19)
making use of the A − A-bimodule structure of D(A).
0
The Cartan matrix C (2)
A , its inverse and the Coxeter matrix φ(2)
C(2)
A =(cid:18)CA C t
0 CA(cid:19) , (C(2)
A )−1 =(cid:18)C −1
A
A −C −1
0
A C t
C −1
A
AC −1
A
(cid:19) and φ(2)
A of A(2) are
A =(cid:18) 0 −idn
φA (cid:19)
φ2
A
Consider an eigenvector u of φA with eigenvalue λ. Let a1 and a2 be the roots of the
equation x2 − λx + λ2 = 0. Therefore am = 1+(−1)m√−3
2
λ, for m = 1, 2. Then
−amu(cid:19) =(cid:18)
Therefore the eigenvalues of φ(2)
A (cid:18) u
φ(2)
amu
A(u) − amφA(u)(cid:19) = am(cid:18) u
−amu(cid:19)
φ2
A are of the form 1±√−3
2
λ for eigenvalues λ of φA.
Proposition. If A is an algebra of cyclotomic type then the double repetitive algebra
is also of cyclotomic type with χA(2) = Φ6 ⊗ χA.
Proof. Just observe that 1±√−3
= 1.
(cid:3)
2
/
/
/
/
/
/
/
/
/
/
/
14
JOS ´E A. DE LA PE NA
5.4. Groups of automorphisms. Let A = kQ/I be a triangular algebra given as a
quiver algebra kQ with an ideal I, equipped with a group G of automorphisms induced
by automorphisms of Q. Each g ∈ G induces a permutation matrix γ(g) ∈ GLn(Z)
which is called a symmetry. The representation γ : G → GLn(Z) is called the
canonical representation. Clearly, φA is an automorphism of γ. We shall recall well-
known results concerning the action of G on A.
Let R1, . . . , Rq be a set of representatives of the irreducible Q-representations of G,
where R1 is the trivial representation. By Maschk´e's theorem, there exists an invert-
ible rational matrix L such that the conjugate representation γL has a decomposition
q
γL =
Rr(α)
α
.
Mα=1
The dimension of Rα is denoted by dimRα, that is, Rα : G → GLdimRα
(Z). Then
we have n =
r(α)dimRα. Since φA is an automorphism of γ, by Schur's lemma,
the conjugate φL
A(:= LφAL−1) takes the block diagonal form
q
Pα=1
φ1
0
. . .
φL
A =
0
φq
where φα : Rr(α)
α → Rr(α)
α
Therefore we get a factorization
is an automorphism, that is, φα ∈ GLr(α)dimRα
(Z).
χA(T ) = χφ1(T ) . . . χφq(T )
where χφα(T ) ∈ Z[T ] is the characteristic polynomial of φα. In particular, χA(T ) has
at least {α : r(α) > 0} factors.
Let S1, . . . , Sm be a set of representatives of the irreducible C-representations of G.
Let S1 be the trivial representation. Then m is the number of conjugacy classes of G.
There is a conjugate γM of γ with a decomposition
m
γM =
Sn(β)
β
.
Mβ=1
Let χβ be the character corresponding to Sβ, that is χβ : G → C∗, g 7→ trSβ(g).
The characters 1 = χ1, . . . , χm form an orthonormal basis of the class group X(G),
with the scalar product
hχ, χ′i =
The following holds:
1
GXg∈G
χ(g)χ′(g).
ALGEBRAS OF CYCLOTOMIC TYPE
15
• For any g ∈ G, χγ(g) is the number of fixed points of g on the vertices of A.
• Burnside's lemma: the number of orbits of vertices of A is
t0(G) =
Moreover, n(1) = t0(G) = r(1).
1
GXg∈G
χγ(g).
Proposition. If A has non-trivial symmetries, then χA(T ) is not irreducible over
Z[T ]. Moreover the following holds:
(a) χA(T ) accepts a factor of degree ≤ t0(G);
(b) in case A is of cyclotomic type, there is a factor Φm(T ) of χA(T ) with m ≤
t0(G)2.
Proof. Consider e1, . . . , en the canonical basis of K0(A) = Zn. Consider the subspace
V of G-invariant vectors in K0(A) ⊗ Q. Then dimQV = t0(G).
form
(a): For a decomposition K0(A) ⊗ Q = V ⊕ V ′ the transformation φA takes the
(cid:18)φ1
0 φ2(cid:19)
∗
which yields χA(T ) = χφ1(T )χφ2(T ) with degree(χφ1(T )) = t0(G).
φ(m0) = degree(Φm0(T )) ≤ t0(G).
(b): Assume that χA(T ) = Qm∈M Φm(T )e(m) for some set M ⊂ N and numbers
e(m) ≥ 1. Then there is some m0 ∈ M satisfying
By [17], the totient function satisfies φ(m) ≥ √m, except for m = 2 and m = 6. We
distinguish two situations:
If m0 = 2 or 6 then m0 ≤ t0(G)2. Otherwise m0 ≤
(i): Assume t0(G) ≥ 3.
φ(m0)2 ≤ t0(G)2.
(ii): Assume t0(G) ≤ 2. We observe that t0(G) is the number of orbits of vertices
of Q under the action of G. Since A is triangular, there are at least two orbits,
one corresponding to sources and another corresponding to sinks in Q. Therefore
t0(G) = 2 and all sources belong to an orbit Gi0 and all sinks to an orbit Gj0.
In case G fixes i0 then Q has the following shape
i0
w♦♦♦♦♦♦♦♦♦♦♦♦♦♦♦
~⑦⑦⑦⑦⑦⑦⑦⑦
···
j2
j1
(❘❘❘❘❘❘❘❘❘❘❘❘❘❘❘❘❘
"❊❊❊❊❊❊❊❊❊
jm−1
jm
with G the group of permutations Sm for m ≥ 2. We observe that
χA = (T 2 − (m − 2)T + 1)(T + 1)m−1
which is a product of cyclotomic polynomials only if m ≤ 4. In this situation Φ1 is a
factor of χA.
w
~
"
(
16
JOS ´E A. DE LA PE NA
In case G acts freely on vertices of Q then the quotient A → B = A/G under the
action of G is a Galois quotient in the sense recalled in the next section. Then B is
the algebra of the quiver with two vertices a and b and m ≥ 2 arrows from a to b. Its
characteristic polynomial is
We shall prove in that, along with χA, the polynomial χB is cyclotomic. This is
possible only if m = 2. Therefore G is cyclic and Q has the shape
χB = T 2 − (m2 − 2)T + 1
i0
❅❅❅❅❅❅❅❅
~⑦⑦⑦⑦⑦⑦⑦⑦
/ ···
j0
i1
js
is
which has Coxeter polynomial
and accepts Φ1 as a factor.
χA = (T − 1)2(T s + T s−1 + . . . + T + 1)2
(cid:3)
5.5. Galois quotients. Let A = kQ/I be an algebra given as a quiver algebra kQ
with an ideal I, equiped with a group G of automorphisms induced by automorphisms
of Q acting without fixed vertices. Then the quotient B = A/G is an algebra with
a presentation B = kQ/I where Q = Q/G is the quotient defined by the action of
G on the quiver Q. We say that B is a Galois quotient of A and we say that A is
a Galois cover of B defined by the group G, we write B = A/G. We say that B is
simply connected if whenever B = A/G is a Galois quotient then G is trivial.
Proposition. Let B = A/G be a Galois quotient algebra of A. If A is an algebra of
cyclotomic type then B is of cyclotomic type.
Proof. Consider the Cartan matrix CA of A. The columns pi = eiCA of CA, for
i = 1, . . . , n, are the dimension vectors of the indecomposable projective A-modules
Pi = Aei. Correspondingly, the indecomposable projective B-module P s = Bes has
dimension vector ps = esCB, where s = 1, . . . , m runs among the classes of the action
of G on the vertices of Q.
Assume that A is of cyclotomic type. Let λ be an eigenvalue of φB = −CBC−t
with associated eigenvector u ∈ K0(B). We shall prove that λ is an eigenvalue of φA
which implies that λ is a root of unity.
B
Indeed, define v ∈ K0(A) a vector v(i) = u(s), where s = Gi. Then
pg(j)(i)! =
v(j)pj(i) = λ
m
λ(vC t
A)(i) = λ
n
Xj=1
u(t) Xg∈G
Xt=1
~
O
O
/
O
O
o
o
ALGEBRAS OF CYCLOTOMIC TYPE
17
m
= λ
Xt=1
u(t)pt(s) = λ(uC t
B)(s) = −(uCB)(s)
and (uCB)(s) = (vCA)(i) by the same argument. Thus λvC t
to prove.
A = −vCA as we wanted
(cid:3)
5.6. Canonical and supercanonical algebras. A canonical algebra Λ is deter-
mined by a weight sequence p = (p1, . . . , pt) of t integers pi ≥ 2 and a parameter
sequence λ = (λ3, . . . , λt) consisting of t − 2 pairwise distinct non-zero scalars from
the base field k. (We may assume λ = 1 such that for t ≤ 3 no parameters occur).
Then the algebra Λ is defined by the quiver
;✈✈✈✈✈
x1
x2 /
#❍❍❍❍❍
xt
◦
x1 /
x2 /
◦
◦
◦ xt
/ ◦
/ ◦
/ ◦
x1/
x2/
xt
x1
x2
/ · · ·◦
/ · · ·◦
/ · · ·◦ xt
x1
#❍❍❍❍❍
x2 /
/ ◦
;✈✈✈✈✈
xt
/ ◦
/ ◦
/ ◦
satisfying the t − 2 equations:
xpi
i = xp1
1 − λixp2
2 ,
i = 3, . . . , t.
For more than two weights, canonical algebras are not hereditary.
Instead, their
representation theory is determined by a hereditary category, the category cohX of
coherent sheaves on a weighted projective line X, naturally attached to Λ, see [8].
Proposition. Let Λ be a canonical algebra. Then Λ is the endomorphism ring of
a tilting object in the category cohX of coherent sheaves on the weighted projec-
tive line X. The category cohX is hereditary and satisfies Serre duality in the form
DExt1(X, Y ) = Hom(Y, τ X) for a self-equivalence τ which serves as the Auslander-
Reiten translation.
(cid:3)
Canonical algebras were introduced by Ringel [35]. They form a key class to study
important features of representation theory. In the form of tubular canonical algebras
they provide the standard examples of tame algebras of linear growth. Up to tilting
canonical algebras are characterized as the connected k-algebras with a separating
exact subcategory or a separating tubular one-parameter family (see [23] and [39]).
That is, the module category modΛ accepts a separating tubular family T = (Tλ)λ∈P1k,
where Tλ is a homogeneous tube for all λ with the exception of t tubes Tλ1, . . . , Tλt
with Tλi of rank pi (1 ≤ i ≤ t). The Coxeter polynomial of Λ is given by
t
In particular, canonical algebras are of cyclotomic type.
χΛ = (x − 1)2
vpi.
Yi=1
/
#
/
;
#
/
/
/
/
;
18
JOS ´E A. DE LA PE NA
Following [24], a supercanonical algebra is defined as follows: The double cone S
of a finite poset S is the poset obtained from S by adjoining a unique source α and
a unique sink ω, like in the pictured example
S :
◦
/ ◦
◦
<③③③
/ ◦
/ ◦
S :
α /
/ ◦
/ ◦
<③③③
"❉❉❉
◦
◦
(❘❘❘❘❘❘❘❘
<②②②
/ ◦
ω
Due to commutativity there is a unique path κS from α to ω in S. Let now S1, . . . , St
be finite posets, t ≥ 2 and λ3, . . . , λt pairwise different elements from k \ {0}. The
supercanonical algebra A = A(S1, . . . , St; λ3, . . . , λt) is obtained from the fully com-
mutative quivers S1, . . . , St by identification of the sources and sinks, respectively,
and requesting additionally the t − 2 relations κi = κ2 − λiκ1, i = 3, . . . , t, where
κi = κSi. The next figure yields an example of a supercanonical algebra with three
arms
◦
6♠♠♠♠♠♠
!❈❈❈❈❈❈❈
◦
◦
◦
/ ◦
◦
/ ◦
/ ◦
(◗◗◗◗◗◗
6♠♠♠♠♠♠
◦
6♠♠♠♠♠♠
(◗◗◗◗◗◗
(◗◗◗◗◗◗
/ ◦
=④④④④④④④
◦
where we assume that κ3 = κ2 − κ1.
In case, S1, . . . , St are linear quivers Si =
[pi] : 1 → 2 → · · · → pi−1, the algebra A(S1, . . . , St; λ1, . . . , λt) is just the canonical
algebra Λ(p1, . . . , pt; λ3, . . . , λt).
Returning to the general case, a simple calculation shows that
t
χA = (x − 1)2
χSi
Yi=1
where χSi is the Coxeter polynomial of the poset algebra Si, i = 1, . . . , t. In particular,
in case every poset algebra k[Si] is of cyclotomic type, for i = 1, . . . , t then the
supercanonical algebra A is of cyclotomic type.
5.7. One-point extensions of canonical algebras. For detailed results related
to one-point extensions by exceptional modules over canonical algebras we refer to
[25, 26].
Let Λ be a canonical algebra of weight type (p1, . . . , pt).
(a) If M is regular simple in the i-th exceptional tube Ti (of τ -period pi), then the
one-point extension Λ[M] is tilting-equivalent to the canonical algebra of weight type
(p1, . . . , pi + 1, . . . , pt) having the same parameter sequence as Λ.
(b) If M has quasi-length s in Ti (recall this means that s < pi), then Λ[M] is
derived equivalent to a supercanonical algebra in the sense of [24], where the linear
(
/
<
/
/
/
<
"
/
<
/
/
(
/
/
6
!
/
/
(
6
(
=
6
ALGEBRAS OF CYCLOTOMIC TYPE
19
arms of the canonical algebra with index different from i are kept, and the i-th linear
arm is changed to the poset K(pi, s)
1
/ 2
/ ···
/ pi − s
⋆
5❦❦❦❦❦❦❦
pi − s + 1
/ ···
/ pi − 1
We write Λ(i, s) for Λ[M] and call it a supercanonical algebra of restricted type.
hp1, . . . , pti sensibly depends on the sign of the Euler characteristic χX = 2−Pt
The structure of the bounded derived category of an extended canonical algebra
i=1(1−
1/pi) of the weighted projective line X associated to Λ. According to [26], the de-
scription of the derived category of an extended canonical algebra yields an interesting
trichotomy.
(i) Positive Euler characteristic: Let Λ be a canonical algebra of domestic represen-
tation type (p1, p2, p3), and ∆ be the Dynkin diagram [p1, p2, p3]. Then the extended
canonical algebra hp1, p2, p3i is derived equivalent to the (wild) path algebra of a
quiver Q having extended Dynkin type ∆.
(ii) Euler characteristic zero: Consider a canonical algebra with a tubular weight
sequence (p1, . . . , pt), we shall assume that 2 ≤ p1 ≤ p2 ≤ · · · ≤ pt. Then the extended
canonical algebra hp1, . . . , pti is derived canonical of type (p1, . . . , pt−1, pt + 1).
(iii) Negative Euler characteristic: Let X be a weighted projective line of negative
Euler characteristic, let A = hp1, . . . , pti be the corresponding extended canonical
algebra and R be the Z-graded singularity attached to X. Then the derived category
Der(modA) is triangle-equivalent to the triangulated category of graded singularities
DerZ
sing(R) of R, where the superscript Z refers to the grading.
5.8. Extended canonical algebras of critical type. In [26] the extended canoni-
cal algebras A with minimal weigth type (p1, p2, . . . , pt) such that Spec (χA(T )) 6⊂ S1
were classified. Recall that such an extended canonical algebra satisfies dimkH 2(A) =
t − 3. In particular, if A is accessible then t = 3. We get the following more precise
statement.
Theorem. Let A be an algebra derived equivalent to an extended canonical algebra
of weight p = (p1, p2, . . . , pt). Suppose that A is not derived equivalent to a canonical
algebra but A is of cyclotomic type then the following holds:
(a) A belongs to the Table below;
(b) suppose that B is an algebra derived equivalent to an extended canonical algebra
of weight q = (q1, q2, . . . , qt) dominated by p then B is of cyclotomic type;
(c) A is accessible if and only if t = 3;
(d) for t = 3 the algebra A is Calabi-Yau.
Statements (a), (b) and (c) are shown in [26], as explained in the next paragraph.
Statement (d) is shown in [28].
/
/
/
/
/
5
/
/
20
JOS ´E A. DE LA PE NA
Ln=0
∞
5.9. Write A = C[P ] for a wild canonical algebra C, whose module category is
derived equivalent to coh X for a weighted projective line X, and P an indecomposable
projective C-module. Then χA = PCχC, where PC is the Hilbert-Poincar´e series of
CM) and M is a rank one not
the positively graded algebra R(p, λ) =
HomC(M, τ n
∞
preprojective C-module. Equivalently, R(p, λ) =
XO), where O is the
structure sheaf on X. Recall from [21] that in case k = C, we can interpret R(p, λ) as
an algebra of entire automorphic forms associated to the action of a suitable Fuchsian
group of the first kind, acting on the upper half plane H+.
Hom(O, τ n
Ln=0
It is well-known that the k-algebra R = R(p, λ) is commutative, graded integral
Gorenstein, in particular Cohen-Macaulay, of Krull dimension two. The complexity
of the surface singularity R is described by the triangulated category
DerZ
sing(R) =
Derb(modZR)
Derb(projZR)
,
where modZR (resp. projZR) denotes the category of finitely generated (resp. finitely
generated projective) Z-graded R-modules. This category was considered by Buch-
weitz [7] and Orlov [30, 31]. It was shown in [26] that there exists a tilting object
¯T in the triangulated category DerZ
sing(R) whose endomorphism ring is isomorphic to
an extended canonical algebra C[P ].
We say that the algebra R(p, λ) (and also the weight sequence p) is formally n-
generated if
PC(T ) =
n−2
n
(1 − T ci)
Qi=1
(1 − T dj )
Qj=1
for certain natural numbers c1, . . . , cn−2 and d1, . . . , dn, all ≥ 2. The algebra R(p, λ)
(and also the weight sequence p) is formally a complete intersection if PC(T ) is a ra-
tional function f1(T )/f2(T ), where each fi(T ) is a product of cyclotomic polynomials.
In [26] it was shown that the following statements are equivalent:
(a) R(p, λ) is formally 3- or 4-generated;
(b) R(p, λ) is formally a complete intersection;
(c) A is of cyclotomic type.
Moreover, for t = 3 the algebra R(p, λ) is a graded complete intersection of the form
k[X1, . . . , Xs]/(ρ3, . . . , ρs) where s = 3 or 4 and ρ3, . . . , ρs is a homogeneous regular
sequence. For k = C the assertion also holds for t ≥ 4 for R(p, λ′) for a suitable
choice of parameters. λ′ = (λ′3, . . . , λ′t).
In case the algebra R = R(p, λ) is formally 3-generated and t = 3 then R is
always a graded complete intersection of the form k[X1, X2, X3]/(f ). In a forthcoming
ALGEBRAS OF CYCLOTOMIC TYPE
21
publication [28] we show that the singularity category DerZ
property, in particular showing statement (d) of the above theorem.
sing(R) satisfies the CY
In Table, the marks • and ⊓⊔ refer to the case k = C: those weight sequences marked
by • or ⊓⊔ correspond to algebras R(p, λ) associated to hypersurface singularities, in
those cases R(p, λ) is formally 3-generated. The marks • correspond to Arnold's
14 exceptional unimodal singularities in the theory or singularities of differentiable
maps [21]. Among those weight sequences p = (p1, p2, p3) (that is t = 3), Arnold's
singularities are exactly those rings of automorphic forms having three generators.
22
JOS ´E A. DE LA PE NA
Weight sequence
Factorization of χA
Poincar´e series
Period of φA
− − −− − −− −− − −− − −− − −− − −− − −− − −− −− − −− − −− − −−
(4, 10, 15) (30)
30
(2, 4, 5)
− − −− − −− −− − −− − −− − −− − −− − −− − −− −− − −− − −− − −−
(4, 5, 10) (20)
20
(2, 5, 5)
− − −− − −− −− − −− − −− − −− − −− − −− − −− −− − −− − −− − −−
(3, 8, 12) (24)
24
•
•
•
(2, 3, 7)
(2, 3, 8)
(2, 3, 9)
(2, 3, 10)
•
•
(2, 4, 6)
(2, 4, 7)
(2, 4, 8)
•
•
•
(2, 5, 6)
(2, 5, 7)
(2, 6, 6)
•
(3, 3, 4)
•
•
•
•
•
(3, 3, 5)
(3, 3, 6)
(3, 3, 7)
(3, 4, 4)
(3, 4, 5)
(3, 4, 6)
(3, 5, 5)
(4, 4, 4)
(4, 4, 5)
⊓⊔
(2, 2, 2, 3)
⊓⊔
⊓⊔
⊓⊔
⊓⊔
⊓⊔
⊓⊔
⊓⊔
(2, 2, 2, 4)
(2, 2, 2, 5)
(2, 2, 2, 6)
(2, 2, 3, 3)
(2, 2, 3, 4)
(2, 2, 3, 5)
(2, 2, 4, 4)
(2, 3, 3, 3)
(2, 3, 3, 4)
(3, 3, 3, 3)
(2, 2, 2, 2, 2)
(2, 2, 2, 2, 3)
(2, 2, 2, 2, 4)
(2, 2, 2, 3, 3)
(2, 2, 2, 2, 2, 2)
Φ42
Φ2 · Φ10 · Φ30
Φ3 · Φ12 · Φ24
Φ2 · Φ16 · Φ18
Φ2 · Φ6 · Φ30
Φ2
2 · Φ22
Φ2 · Φ9 · Φ18
Φ2
2 · Φ4 · Φ12 · Φ14
Φ5 · Φ20
Φ2 · Φ8 · Φ16
Φ11 · Φ12
Φ2
2 · Φ3 · Φ6 · Φ10 · Φ12
Φ3 · Φ24
Φ2 · Φ3 · Φ6 · Φ18
Φ2
3 · Φ15
Φ2 · Φ3 · Φ4 · Φ10 · Φ12
Φ2 · Φ4 · Φ16
Φ13
Φ2 · Φ3 · Φ9 · Φ10
Φ2 · Φ5 · Φ8 · Φ10
Φ2
2 · Φ2
4 · Φ6 · Φ12
Φ2 · Φ4 · Φ8 · Φ9
Φ2
2 · Φ18
Φ2
2 · Φ14
Φ2
2 · Φ3 · Φ6 · Φ12
Φ2
2 · Φ8 · Φ10
Φ2 · Φ3 · Φ4 · Φ12
Φ2
2 · Φ5 · Φ10
Φ2 · Φ7 · Φ8
Φ2
2 · Φ4 · Φ6 · Φ8
Φ2
3 · Φ9
Φ2 · Φ3 · Φ6 · Φ7
Φ2 · Φ3
Φ4
2 · Φ10
Φ3
2 · Φ4 · Φ8
2 · Φ3 · Φ2
Φ2
Φ2
2 · Φ3 · Φ5 · Φ6
Φ5
2 · Φ4 · Φ6
3 · Φ2
6
6
(6, 14, 21) (42)
(6, 8, 15) (30)
(6, 8, 9) (24)
(6, 8, 9, 10) (16, 18)
42
30
24
72
(4, 6, 11) (22)
(4, 6, 7) (18)
(4, 6, 7, 8) (12, 14)
22
18
84
(4, 5, 6) (16)
(4, 5, 6, 7) (11, 12)
(4, 5, 6, 6) (10, 12)
16
132
60
(3, 5, 9) (18)
(3, 5, 6) (15)
(3, 5, 6, 7) (10, 12)
(3, 4, 8) (16)
(3, 4, 5) (13)
(3, 4, 5, 6) (9, 10)
(3, 4, 5, 5) (8, 10)
(3, 4, 4) (12)
(3, 4, 4, 5) (8, 9)
18
15
60
16
13
90
40
12
72
(2, 4, 7) (14)
(2, 4, 5) (12)
(2, 4, 5, 6) (8, 10)
(2, 3, 6) (12)
(2, 3, 4) (10)
(2, 3, 4, 5) (7, 8)
(2, 3, 4, 4) (6, 8)
(2, 3, 3) (9)
(2, 3, 3, 4) (6, 7)
(2, 3, 3, 3) (6, 6)
(2, 2, 5) (10)
(2, 2, 3) (8)
(2, 2, 3, 4) (6, 6)
(2, 2, 3, 3) (5, 6)
(2, 2, 2, 3) (4, 6)
14
12
40
12
10
56
24
9
42
∞
10
8
∞
30
12
− − −− − −− −− − −− − −− − −− − −− − −− − −− −− − −− − −− − −−
(2, 6, 9) (18)
18
Table Weights p with ρ(φA) = 1.
ALGEBRAS OF CYCLOTOMIC TYPE
23
6. Algebras of cyclotomic type and the homological quadratic form
6.1. Let A be an algebra of cyclotomic type. As observed in [20, 25] the transfor-
mation φA is not necessarily periodic. As we remarked before, φA is periodic if and
only if A is of cyclotomic type and φA is diagonalizable.
The following result follows [16], see also [25, 37].
Theorem. If hA is non-negative, then A is of cyclotomic type. Moreover, if hA is
positive definite then φA is diagonalizable and periodic.
Proof. Let C be any invertible positive upper triangular matrix and h(x) = xt(C−1 +
C−t)x, for x ∈ K0(A), an associated quadratic form. Assume h(x) is positive definite,
we shall prove that all eigenvalues of −CC−t have modulus one.
Indeed, if C−tu = −λC−1u for an eigenvalue λ of −CC−t and a complex vector
u 6= 0, since 0 < h(u) = utC−tu = −λutC−1u then λ = 1. In particular, this shows
that hA positive definite implies A cyclotomic.
Assume that hA is non-negative. Choose ǫ > 0 such that
Cǫ = CA
∞
Xi=0
(−1)i(ǫCA)i
converges. Therefore C−1
ǫ = C−1
A (1 + ǫCA) and the real quadratic form
hǫ(x) = xtC−1
ǫ x = xtC−1
A x + ǫxtx
is positive definite.
then the same holds for the eigenvalues of hA. That is, A is of cyclotomic type.
Since limǫ→0 hǫ = hA and the eigenvalues of hǫ have modulus one, for every ǫ > 0,
Assume that hA is positive definite. Following [?] we show that φA = −CAC−t
A is
diagonalizable. Indeed, suppose, to get a contradiction, that v is an eigenvector of
φA with eigenvalue λ and 0 6= u is a vector with φAu − λu = v. We get
A u − λC−1
A u = C−1
A v
A v = λC−1
A v.
(1) −C−t
(2) −C−t
Therefore, from (2), utC−t
yields, vtC−1
contradicting the positivity of hA.
A u = −λvtC−t
A v = −λutC−1
A u. Since λ = 1, equation (1) implies hA(v) = vtC−t
A v. Taking transpose and conjugate, this
A v = 0,
Finally, φA is equivalent to a diagonal matrix D whose diagonal entries are roots
(cid:3)
of unity. Thus φA is periodic.
In certain cases the non-negativity of hA is related to the representation type of
6.2.
A, see for example [33] and the next paragraph. On the other hand, there are pairs
of algebras A and B with hA = hB and such that A is representation tame while B
is representation wild. We show a sequence of such pairs.
24
JOS ´E A. DE LA PE NA
For each number n ≥ 7, consider the quiver
/ ···
A✄✄✄
'◆◆◆◆◆◆◆
Qn : 1
/ 2
5
/ n − 1
#❋❋❋❋
5❦❦❦❦❦❦❦❦❦❦❦
n
3
4
Let An be the quotient of the path algebra kQn determined by the commutativity
relation and Bn the quotient of kQn where the long path of length n − 4 from 2 to n
vanishes. Then hAn = hBn which is a positive definite quadratic form. Observe that
An is representation finite while Bn is a wild algebra.
6.3. Consider the poset
Dn :
1
2
%❑❑❑❑❑❑
9ssssss
3
/ 4
/ ···
/ n
The supercanonical algebra A = A((1), (1), Dn; 1) is a pg-critical algebra, that is a
tame algebra which is minimal not of polynomial growth. Any pg-critical algebra is
derived equivalent to an algebra of the form A((1), (1), Dn; 1), see [23].
A semichain poset is a full convex subcategory of the poset
D(n, m) :
1
1′
B✆✆✆✆✆✆✆✆✆
✾✾✾✾✾✾✾✾✾
/ 2
2′
/ ···
A☎☎☎☎☎☎☎☎☎
✿✿✿✿✿✿✿✿✿✿
/ ···
··· m − 1
;✇✇✇✇✇✇✇✇✇✇✇✇
#●●●●●●●●●●●
m
m′
'◆◆◆◆◆◆
8qqqqqq
2m
/ ···
/ n
··· (m − 1)′
In important cases it is easy to determine the periodicity of the Coxeter transfor-
mation. Recall that a supercanonical algebra A = A(S1, . . . , St; λ3, . . . , λt) is said to
be of Dynkin class if each poset Si is of Dynkin type. In that case, let p(Si) be the
period of φSi.
Theorem. Let A = A(S1, . . . , St; λ3, . . . , λt) be a supercanonical algebra of Dynkin
class. Let p = lcm(p(S1), . . . , p(St)) and B = A/(α) be the algebra obtained by
deleting the source α.
The following are equivalent:
(a) φA is periodic;
(b) φp
(c) hA is non-negative and rad hA has rank two;
(d) hB is non-negative and rad hB has rank one.
A = 1 and p is the exact period of φA;
/
/
#
/
A
'
5
O
O
%
/
/
/
9
/
/
/
/
#
'
/
/
/
/
B
A
/
/
/
;
8
ALGEBRAS OF CYCLOTOMIC TYPE
25
7. On the structure of Auslander-Reiten components of algebras of cyclo-
tomic type
m be an irreducible
decomposition of its Coxeter polynomial.
7.1. Let A be an algebra of cyclotomic type and χA =Qm∈M Φe(m)
Consider the bounded derived category Der(A) = Derb(modA) and let Γ = ΓDer(A)
be the corresponding Auslander-Reiten quiver. Recall that for any class [X] ∈
K0(Der(A)) and X → Y → Z → T (X) an Auslander-Reiten triangle then [T (X)] =
φA([X]).
Theorem. Every component of Γ is either a tube ZA∞/(p) of finite period p where
p = φ(m) for some m ∈ M or of the form Z∆ for ∆ a Dynkin or extended Dynkin
diagram or of one of the shapes A∞, A∞
∞
Proof. By [41], every component is either a tube or of the shape Z∆ for some infinite
quiver ∆ which is locally finite and without oriented cycles.
or D∞.
Consider the case of a component C of shape Z∆ in Γ. We will show that a growth
argument restricts the possible shapes of ∆. We need some preparation. Fix a vertex
x0 in C with corresponding class X0 ∈ K0(Der(A)). For each n ∈ N the Auslander-
Reiten translation xn = T n(x0) with corresponding class Xn ∈ K0(Der(A)). Consider
the function
c(n) = Xi∈Q0
[Xn](i)
which is the restriction of an additive function on the subcategory C(x0) of C formed
by the predecessors of x0. Consider the complete section ∆ with x0 as unique sink
and c0 : ∆ → Z the vector obtained as restriction of the function c to ∆. Then
c(n) = φn
∆)n∈N is exponential,
In case ∆ is not one of the indicated graphs, the growth of (φn
∆(c0)(x0).
as it is shown in [41]. This completes the claim on the shape of ∆.
Let C be a tubular component of shape ZA∞/(p) in Γ. Let X1, X2, . . . , Xp be
objects on the mouth of C. Let V be the subspace of K0(Der(A)) ⊗ C generated by
the Xi. Then V is a φA-invariant subspace. Assume K0(Der(A)) ⊗ C = V ⊕ W then
φA takes the form
0
(cid:18)φ1
∗ φ2(cid:19)
Then χA = χφ1χφ2 and since φ1 is an irreducible matrix then χφ1 is an irreducible
factor of χA. That is χφ1 = Φm for some m ∈ M. Then p = φ(m).
(cid:3)
7.2. Let A be a triangular algebra and Γ the Auslander-Reiten quiver of its derived
category Der(A). The class quiver [Γ] of A is formed by the quotients [C] of compo-
nents C of Γ obtained by identifying X, Y ∈ C if [X] = [Y ] in K0(Der(A)). Observe
that, according to [10], [C] = C for any component of Γ with the possible exception
of [ZA∞] = ZA∞/(p) for some p.
26
JOS ´E A. DE LA PE NA
Theorem. Let A be a triangular algebra and Γ the Auslander-Reiten quiver of its
derived category Der(A). Then the following holds:
(a) if A satisfies the CY property then φA is periodic;
(b) every component of [Γ] is a tube if and only if φA is periodic.
If (b) holds then,
(c) A is of cyclotomic type, and
(d) if p1, . . . , ps are the periods > 1 of non-homogeneous tubes in [Γ] then
pi ≤ n.
s
Pi=1
Proof. Let C be a tubular component of shape ZA∞/(p) in [Γ]. As above we may
prove that there is an irreducible factor fp(T ) of χA(T ) of degree p.
Assume that every component of [Γ] is a tube. Consider a maximal set X1, . . . , Xs
in Der(A) satisfying that
Clearly,
s
Pi=1
(i) [Xi] lies on the mouth of a tube Ti in [Γ] with period pi and fpi(T ) is an
irreducible factor of χA(T ) of degree pi;
(ii) [Xj] does not lie on Ti for j 6= i;
(iii) Qs
A ([Xi]) : 1 ≤ i ≤ s, 0 ≤ m < pi} is a basis of
pi = n. In particular, {φm
i=1 fpi is a factor of χA(T ).
K0(Der(A)). Therefore φA is periodic which implies that A is of cyclotomic type.
Conversely, assume that φA is periodic. Let C be a component of shape Z∆ in Γ.
Observe that the classes [X] for X ∈ C are periodic. Hence by [10], [C] is a tube, that
is, ∆ has the shape ZA∞.
(cid:3)
References
[1] I. Assem, D. Simson and A. Skowro´nski, Elements of the Representation Theory of As-
sociative Algebras 1: Techniques of Representation Theory. London Mathematical Society
Student Texts 65, Cambridge University Press, 2006.
[2] I. Assem, D. Happel and O. Rold´an. Representation-finite trivial extension algebras. J. Pure
Appl. Algebra 33 (1984), 235–242.
[3] I. Assem and A. Skowro´nski. Tilting simply connected algebras. Comm. Algebra 22 (1994),
4611–4619.
[4] M. Barot and H. Lenzing, One-point extensions and derived equivalence, J. Algebra 264
(2003), 1–5.
[5] G. Bobi´nski, C. Geiss and A. Skowro´nski, Classification of discrete derived categories. Cent.
Eur. J. Math. 2 (2004), no. 1, 19–49.
[6] A.I. Bondal and M.M. Kapranov. Representable functors, Serre functors, and mutations.
Math. USSR, Izv. 35 (1990), 519–541.
[7] R. O. Buchweitz, Maximal Cohen-Macaulay modules and Tate-cohomology over Gorenstein
rings. Unpublished manuscript (1987) 155 pp.
[8] W. Geigle and H. Lenzing, A class of weighted projective curves arising in representation
theory of finite dimensional algebras. In Singularities, representations of algebras, and vector
bundles. Springer Lect. Notes Math. 1273 1987), 265–297.
ALGEBRAS OF CYCLOTOMIC TYPE
27
[9] S.P. Glasby, On the tensor product of polynomials over a ring. Austral. Math. Soc. 71 (2001),
307-324
[10] D. Happel, U. Preiser, and C. M. Ringel, Vinbergs characterization of Dynkin diagrams
using subadditive functions with application to DTr-periodic modules, Representation theory,
II (Proc. Second Internat. Conf., Carleton Univ., Ottawa, Ont., 1979), Lecture Notes in
Math., vol. 832, Springer, Berlin, 1980, pp. 280 - 294.
[11] D. Happel, Triangulated Categories in the Representation Theory of Finite Dimensional
Algebras. London Mathematical Society LNM 119 Cambridge University Press (1988)
[12] D. Happel, A characterization of hereditary categories with tilting object, Invent. Math. 144
(2001), 381–398.
[13] D. Happel, Hochschild cohomology of finite dimensional algebras. In S´eminaire d'Alg`ebre
Paul Dubreil et Marie-Paul Malliavin, 39`eme Ann´ee. Lecture Notes in Math. 1404,
Springer-Verlag 1989, 108–126.
[14] D. Happel, The trace of the Coxeter matrix and the Hochschild cohomology. Linear Algebra
and its Applications 258 (1997), 169-177
[15] D. Happel, The Coxeter polynomial for a one-point extension algebra. Journal of Algebra
321 (2009), 2028-2041
[16] R.B. Howlett Coxeter groups and M-matrices Bull. London Math. Soc., 14 (2) (1982), pp.
137141
[17] D.G. Kendall and R. Osborn Two simple lower bounds for Euler's function. Texas J. Sci.
17, 1965.
[18] L. Kronecker. Zwei Satze uber Gleichungen mit ganzzahligen Coefficienten. J. Crelle 1857,
Ouvres I, p. 105–108
[19] D. Kussin, H. Lenzing and H. Meltzer, Triangle singularities, ADE-chains and weighted
projective lines Preprint. Paderborn (2012).
[20] S. Ladkani. On the periodicity of Coxeter transformations and the non-negativity of their
Euler forms. Linear Algebra and its Applications Vol. 428, Issue 4, 1 (2008), 742753
[21] H. Lenzing, Wild canonical algebras and rings of automorphic forms. In Finite Dimensional
Algebras and Related Topics, V. Dlab and L.L. Scot (eds.), 191–212, Kluwer Academic
Publishers, Dordrecht (1994).
[22] H. Lenzing, Coxeter transformations associated with finite-dimensional algebras. In Compu-
tational methods for representations of groups and algebras, 287–308, Progr. Math., 173,
Birkhauser, Basel, 1999.
[23] H. Lenzing and J.A. de la Pena, Wild canonical algebras. Math. Z. 224 (1997) 403–425.
[24] H. Lenzing and J.A. de la Pena, Supercanonical algebras J. Algebra 282 (2004), 298-348.
[25] H. Lenzing and J.A. de la Pena, Spectral analysis of finite dimensional algebras and singular-
ities, inTrends in Representation Theory of Algebras and Related Topics, ed. A. Skowro´nski,
EMS Publishing House, Zurich (2008) 541–588.
[26] H. Lenzing and J.A. de la Pena, Extended canonical algebras and Fuchsian singularities,
Math. Z. 268 (2011), 143 - 167.
[27] H. Lenzing and J.A. de la Pena, Accessible algebras. J. Algebra to appear (2013).
[28] A class of nilpotent algebras. Preprint. Paderborn, Germany and M´exico (2013).
[29] K. Mahler, Lectures on trascendental numbers. Lecture Notes in Mathematics 546. Springer
Verlag, Berlin, 1976.
[30] D. Orlov, Triangulated categories of singularities and D-branes in Landau-Ginzburg models.
Proc. Steklov Inst. Math. 246 (2004) 227-248.
[31] D. Orlov, Derived categories of coherent sheaves and triangulated categories of singularities.
arXiv:math.AG/0503632v2 (2005).
28
JOS ´E A. DE LA PE NA
[32] J.A. de la Pena, Coxeter transformations and the representation theory of algebras, in Finite
Dimensional Algebras and Related Topics, V.Dlab and L.L. Scott (eds.), Kluwer Academic
Publishers, 1994, 223–253.
[33] J. A. de la Pena. The Tits form of a tame algebra. In Canadian Math. Soc. Conference
Proceedings. Vol. 19 (1996) 159–183.
[34] V. Prasolov, Polynomials. Algorithms and computation in Mathematics vol. 11. Springer
Verlag, Berlin, 2001 (second edition).
[35] C.M. Ringel, Tame algebras and integral quadratic forms. Lecture Notes in Mathematics,
1099. Springer-Verlag, Berlin, 1984.
[36] C.M. Ringel, The spectral radius of the Coxeter transformations for a generalized Cartan
matrix, Math. Ann. 300 (1994), 331–339.
[37] M. Sato, Periodic Coxeter matrices and their associated quadratic forms, Linear Algebra
Appl. 406 (2005), 99 - 108.
[38] A. Skowro´nski, Simply connected algebras in Hochschild cohomologies. In Representations
of Algebras. Canad. Math. Soc. Conf. Proc. 14, Amer. Math. Soc., Providence, RI, 1993,
431–447.
[39] A. Skowro´nski, On omnipresent tubular families of modules, CMS Conf. Proc. 18 (1996),
641–657.
[40] D. Vossieck, The algebras with discrete derived category, J. Algebra 243 (2001), 168–176.
[41] Y. Zhang, Eigenvalues of Coxeter transformations and the structure of regular components
of an Auslander-Reiten quiver Comm. Algebra 17(10):2347 - 2362 (1989)
Centro de Investigaci´on en Matem´aticas, A.C.
Guanajuato 36240 M´exico
e-mail: [email protected]
and
Instituto de Matem´aticas, UNAM. Cd. Universitaria,
M´exico 04510 D.F.
e-mail: [email protected]
|
1903.02960 | 1 | 1903 | 2019-03-06T16:36:10 | Poincare-Birkhoff-Witt theorem for pre-Lie and postLie algebras | [
"math.RA"
] | We construct the universal enveloping preassociative and postassociative algebra for a pre-Lie and a postLie algebra respectively. We show that the pairs $(\mathrm{preLie},\mathrm{preAs})$ and $(\mathrm{postLie},\mathrm{postAs})$ are Poincare-Birkhoff-Witt-pairs, for the first one it's a reproof of the result of V. Dotsenko and P. Tamaroff. | math.RA | math |
MSC (2010): 16W99
Poincar´e(cid:22)Birkhoff(cid:22)Witt theorem
for pre-Lie and postLie algebras
V. Gubarev
Abstract
We construct the universal enveloping preassociative and postassociative al-
gebra for a pre-Lie and a postLie algebra respectively. We show that the pairs
(preLie, preAs) and (postLie, postAs) are Poincar´e(cid:22)Birkhoff(cid:22)Witt-pairs, for the
first one it's a reproof of the result of V. Dotsenko and P. Tamaroff.
Keywords: Rota(cid:22)Baxter operator, Grobner(cid:22)Shirshov basis, Poincar´e(cid:22)Birk-
hoff(cid:22)Witt pair of varieties, pre-Lie algebra, postLie algebra, preassociative algebra
(dendriform algebra), postassociative algebra.
1 Introduction
In 1960s, pre-Lie algebras appeared independently in affine geometry (E. Vinberg [43];
J.-L. Koszul [30]), and ring theory (M. Gerstenhaber [17]). Arising from diverse areas,
pre-Lie algebras are known under different names like Vinberg algebras, Koszul algebras,
left- or right-symmetric algebras (LSAs or RSAs), Gerstenhaber algebras. Pre-Lie alge-
bras satisfy an identity (x1x2)x3 − x1(x2x3) = (x2x1)x3 − x2(x1x3). See [7, 36] for surveys
on pre-Lie algebras.
In 2001, J.-L. Loday [32] defined the dendriform (di)algebra (preassociative algebra)
as a vector space endowed with two bilinear operations ≻, ≺ satisfying
(x1 ≻ x2 + x1 ≺ x2) ≻ x3 = x1 ≻ (x2 ≻ x3), (x1 ≻ x2) ≺ x3 = x1 ≻ (x2 ≺ x3),
x1 ≺ (x2 ≻ x3 + x2 ≺ x3) = (x1 ≺ x2) ≺ x3.
In 1995, J.-L. Loday also defined [31] Zinbiel algebra (precommutative algebra), on
which the identity (x1x2 + x2x1)x3 = x1(x2x3) holds. Every preassociative algebra with
the identity x ≻ y = y ≺ x is a precommutative algebra (x1x2 = x1 ≻ x2) and under the
product x · y = x ≻ y − y ≺ x is a pre-Lie algebra.
In 2004, dendriform trialgebra (postassociative algebra) was introduced [35], i.e., an
algebra with bilinear operations ≺, ≻, · satisfying seven certain axioms. A space A with
two bilinear operations [, ] and · is called a post-Lie algebra (B. Vallette, 2007 [42]) if [, ]
is a Lie bracket and the next identities hold
(x · y) · z − x · (y · z) − (y · x) · z + y · (x · z) = [y, x] · z,
x · [y, z] = [x · y, z] + [y, x · z].
In last dozen years, an amount of articles devoted to post-Lie algebras in different areas
is arisen [8, 16, 38].
Let us explain the choice of terminology. Given a binary quadratic operad P, the
defining identities for pre- and post-P-algebras were found in [2]. One can define the
1
operad of pre- and post-P-algebras as P • preLie and P • postLie respectively. Here
preLie and postLie denote the operads (varieties) of pre-Lie algebras and postLie algebras
respectively, • denotes the black Manin product of operads [18]. By pre- or postalgebra
we will mean pre- or post-P-algebra for some operad P.
Before stating the main problem of the work we introduce a very useful tool to
deal with pre- and postalgebras, so called Rot(cid:22)Baxter operators, and the notion of
a Poincar´e(cid:22)Birkhoff(cid:22)Witt pair.
A linear operator R defined on an algebra A over a field k is called a Rota(cid:22)Baxter
operator (RB-operator, for short) of a weight λ ∈ k if it satisfies the relation
R(x)R(y) = R(R(x)y + xR(y) + λxy),
x, y ∈ A.
(1)
In this case, an algebra A is called Rota(cid:22)Baxter algebra (RB-algebra).
G. Baxter defined the notion of what is now called Rota(cid:22)Baxter operator on a (com-
mutative) algebra in 1960 [4], solving an analytic problem. The relation (1) with λ = 0 ap-
peared as a generalization of integration by parts formula. G.-C. Rota [40], P. Cartier [9]
and others studied different combinatorial properties of RB-operators and RB-algebras.
In 1980s, the deep connection between Lie RB-algebras and Yang(cid:22)Baxter equation was
found [5, 41]. More about Rota(cid:22)Baxter algebras see in the monograph of L. Guo [27].
In 2000, M. Aguiar [1] stated that an associative algebra with a given Rota(cid:22)Baxter
operator R of weight zero under the operations a ≻ b = R(a)b, a ≺ b = aR(b) is
a preassociative algebra. In 2002, K. Ebrahimi-Fard [14] showed that an associative RB-
algebra of nonzero weight λ under the same two products ≻, ≺ and the third operation
a · b = λab is a postassociative algebra. The analogue of the Aguiar construction for
the pair of pre-Lie algebras and Lie RB-algebras of weight zero was stated in 2000 by
M. Aguiar [1] and by I.Z. Golubchik, V.V. Sokolov [19]. In 2010 [3], this construction for
the pair of post-Lie algebras and Lie RB-algebras of nonzero weight was extended.
In 2013 [2], the construction of M. Aguiar and K. Ebrahimi-Fard was generalized for
the case of arbitrary variety.
In 2008, the notion of universal enveloping RB-algebras of pre- and postassociative
algebras was introduced [15]. In [15], it was also proved that the universal enveloping of
a free pre- or postassociative algebra is free.
In 2010, with the help of Grobner(cid:22)Shirshov bases [6], Yu. Chen and Q. Mo proved
that every preassociative algebra over a field of characteristic zero injectively embeds into
its universal enveloping RB-algebra [11].
In 2013 [25], given a variety Var, it was proved that every pre-Var-algebra (post-Var-
algebra) injectively embeds into its universal enveloping Var-RB-algebra of weight λ = 0
(λ 6= 0). Further, author constructed universal enveloping RB-algebra for a given pre- or
postalgebra in commutative [20], associative [21], and Lie [22] cases. In the associative
case it gave an answer to the question of L. Guo [27, p. 148].
The classical Poincar´e(cid:22)Birkhoff(cid:22)Witt theorem states that given a Lie algebra g with
a linear basis {xi i ∈ I}, where I is a well-ordered set, the monomials xi1 . . . xin with
i1 ≤ . . . ≤ in form a linear basis for the universal enveloping associative algebra U(g).
As a consequence, we get that the linear basis of the algebra U(g) does not depend on
the product in the Lie algebra g. Such relationships between two varieties V, W in the
2
case when there exists a functor φ : V → W associating to every algebra A ∈ V an
algebra φ(A) ∈ W by changing multiplication in A was generalized by I. Shestakov and
A.A. Mikhalev in the term Poincar´e(cid:22)Birkhoff(cid:22)Witt (PBW-) pair, see details in [37].
Now let us formulate the main problem to which the work is devoted. In advance,
preAs and postAs denote the varieties of pre- and postassociative algebras respectively.
Problem 1. a) Prove that every pre-Lie (postLie) algebra injectively embeds into its
universal enveloping preassociative (postassociative) algebra.
b) Clarify if the pairs (preAs, preLie) and (postAs, postLie) are PBW-ones.
c) Construct the universal enveloping preassociative (postassociative) algebra for
a given pre-Lie (postLie) algebra.
For pre-Lie algebras, Problem 1b and special version of Problem 1c were stated by
P. Kolesnikov in [29] in the context of Grobner(cid:22)Shirshov bases for preassociative algebras.
J.-L. Loday asked V. Dotsenko about the solution of Problem 1b around 2009 [13]. The
discussion of Problem 1 in the case of restricted pre-Lie algebras can be found in [12].
The analogues of Problem 1 for Koszul-dual objects, di- and trialgebras, were solved in
[34, 26].
Recently, in [24] and [23], the author solved Problem 1a in pre- and postalgebra cases
with the help of embedding of pre-Lie (postLie) algebras into Lie RB-algebras [25] and the
Grobner(cid:22)Shirshov bases technique developed for associative RB-algebras [28]. Actually,
the solution of Problem 1a for pre-Lie algebras can be derived from the results concerned
Hopf preassociative algebras and so called brace algebras stated in 2002 independently
by F. Chapaton [10] and M. Ronco [39]. In 2018, V. Dotsenko and P. Tamaroff by means
of the general approach arising from the category theory solved Problem 1b for pre-Lie
algebras stating that the pair of varieties (preLie, preAs) is a PBW-pair [13].
The current work is devoted to the complete solution of Problem 1 in both pre- and
postalgebra cases. Let us briefly describe the idea of the solution. For this, we need one
more embedding problem.
Let A be an associative algebra with an RB-operator R. Then the algebra A(−) is a
Lie RB-algebra under the product [x, y] = xy − yx and the same action of R. Thus, we
can state the analogue of Problem 1 for the varieties of Lie and associative RB-algebras.
Problem 2. a) Prove that every Lie RB-algebra injectively embeds into its universal
enveloping associative RB-algebra.
b) Construct the universal enveloping associative RB-algebra for a given Lie RB-
algebra.
We are not asking whether the pair (RBAs, RBLie) of the varieties of associative and
Lie RB-algebras forms a PBW-pair, since it is easy to disprove it just in 2-dimensional
case. To the moment, we are far from the solution Problem 2, and the current work as
well as [23, 24] can be also considered like a step in such direction.
The sketch of the solution of Problem 1c is following. At first, we embed a pre- or
post-Lie algebra C into its universal enveloping Lie RB-algebra L [22]. At second, with
the help of Grobner(cid:22)Shirshov bases we embed the Lie RB-algebra L into its universal
enveloping associative RB-algebra A with an RB-operator R. At third, we show that
a subalgebra U(C) generated by C in the induced pre- or postassociative algebra on the
space A is the universal enveloping pre- or postassociative algebra for C.
Actually, the same solution algorithm was used earlier by author in [23] and [24]
3
but with the following change:
in the first step it was considered injective enveloping
Lie RB-algebra from [25] instead of L, the universal enveloping one constructed in [22].
Surprisingly, a posteriori we may say that at least in the pre-Lie algebra case the injective
enveloping preassociative algebra of a given pre-Lie algebra obtained earlier in [24] is an
universal one.
At the end of Introduction, let us collect all stated in the work connections between
universal enveloping algebras of different kind in the following commutative diagram,
pre/postLie
RBLie
❏
❏
❏
❏
❏
❏
❏
❏
❏
❏
❏
❏
❏
❏
pre/postAs
$❏
RBAs
where every arrow maps the algebra from corresponding variety into its universal envelop-
ing one. An associative RB-algebra A with an RB-operator P of weight λ = 0 (λ 6= 0) is
an enveloping for a pre-Lie (postLie) algebra C in the sense that the following equalities
a ≻ b = P (a)b − bP (a),
a ≺ b = aP (b) − P (b)a (a · b = λab − λba)
(2)
hold for any a, b ∈ C.
2 Preliminaries
2.1 Some required formulas
Given a Lie algebra L, denote the product [[. . . [[y, x], x] . . .], x] ∈ Lp+1 by [y, x(p)].
Lemma [24]. Given a Lie algebra L, the equality
(l + 1)yxl =
(−1)i(cid:18)l + 1
l+1Xi=2
i (cid:19)[y, x(i−1)]xl+1−i +(cid:0)yxl + xyxl−1 + . . . + xly)
holds in the universal enveloping algebra U(L) for any x, y ∈ L and l ≥ 0.
It follows immediately from (1) that
+ λk−1 X1≤i1<i2<...<ik≤t
R(b1) . . . R(bi1) . . . R(bik ) . . . R(bt) + . . . + λt−1b1 . . . bt(cid:19) = 0,
where the signshows the omitting action of R. From (3) and (4), we derive the formula
R(R(b1)R(b2) . . . R(bl−1)blR(bl)kR(bl+k+1))
=
1
k + 1
R(b1)R(b2) . . . R(bl−1)R(bl)k+1R(bl+k+1)
4
R(b1) . . . R(bt) − R(cid:18) tXi=1
+ λ X1≤i1<i2≤t
R(b1) . . . R(bi−1) R(bi)R(bi+1) . . . R(bt)
R(b1) . . . R(bi) . . . R(bj) . . . R(bt) + . . .
(3)
(4)
/
/
/
/
$
i (cid:19)R(b1)R(b2) . . . R(bl−1)[bl, R(bl)(i−1)]R(bl)k+1−iR(bl+k+1)
R(b1) . . . R(bi−1) R(bi)R(bi+1) . . . R(bt)
R(b1) . . . R(bi) . . . R(bj) . . . R(bl+k+1) − . . .
+
1
k + 1
(−1)i(cid:18)k + 1
R(cid:18) k+1Xi=2
− X1≤i≤l−1, i=l+k+1
− λ X1≤i1<i2≤l+k+1
− λs−1 X1≤i1<...<is≤l+k+1
R(b1) . . . R(bi1) . . . R(bis) . . . R(bl+k+1)
− . . . − λl+kb1 . . . bl+k+1(cid:19),
(5)
where bl = bl+1 = . . . = bl+k and R(bl)s means (R(bl))s.
In advance we will use the formula (5) with maybe absent R(b1) and R(bl+k+1), it
means that we omit all b1 and R(b1) in the summands and all indexes ij start with two
when R(b1) is absent. We do the same if R(bl+k+1) is absent.
2.2 Embedding of pre- and postalgebras into RB-algebras
Theorem 1 [1, 2, 3, 14, 19, 33]. Let A be an RB-algebra of a variety Var and weight
λ = 0 (λ 6= 0). With respect to the operations
x ≻ y = R(x)y,
x ≺ y = xR(y)
(x · y = λxy)
(6)
A is a pre-Var-algebra (post-Var-algebra).
Denote the pre- and post-Var-algebra obtained in Theorem 1 as A(R)
λ .
Given a pre-Var-algebra hC, ≻, ≺i, universal enveloping RB-Var-algebra U of C is the
universal algebra in the class of all RB-Var-algebras of weight zero such that there exists
homomorphism from C to U (R)
. Analogously universal enveloping RB-Var-algebra of a
post-Var-algebra is defined.
0
Theorem 2 [25]. Every pre-Var-algebra (post-Var-algebra) could be embedded into
its universal enveloping RB-algebra of the variety Var and weight λ = 0 (λ 6= 0).
Let us briefly describe the idea of the proof of Theorem 2. Given a pre-Var-algebra
(post-Var-algebra) C, we define the product on the space C = C ⊕ C ′, where C ′ is a copy
of C, in such a way that C is an algebra of the variety Var. Then we define on C the
linear operator P which occurs to be a Rota(cid:22)Baxter operator of weight λ = 0 (λ 6= 0).
Finally, Theorem 2 was stated by embedding C into C (P )
by the map c → c′.
λ
2.3 Grobner(cid:22)Shirshov bases for associative RB-algebras
Let RAshXi denote the free associative algebra generated by a set X with a linear
map R in the signature. One can construct a linear basis of RAshXi (see, e.g., [15])
by induction. At first, all elements from S(X), the free semigroup generated by X, lie
in the basis. At second, if we have basic elements a1, a2, . . . , ak, k ≥ 1, then the word
w1R(a1)w2 . . . wkR(ak)wk+1 lies in the basis of RAshXi. Here w2, . . . , wk ∈ S(X) and
5
w1, wk+1 ∈ S(X) ∪ ∅, where ∅ denotes the empty word. Let us denote the basis obtained
as RS(X). Given a word u from RS(X), the number of appearances of the symbol R
in u is denoted by degR(u), the R-degree of u. We call an element from RS(X) of the
form R(w) as R-letter. By X∞ we denote the union of X and the set of all R-letters.
Given u ∈ RS(X), define deg u (degree of u) as the length of u in the alphabet X∞.
In [15], deg u was called the breadth of u.
Suppose that X is a well-ordered set with respect to <. Let us introduce by induction
the deg-lex order on S(X). At first, we compare two words u and v by the length: u < v
if u < v. At second, when u = v, u = xiu′, v = xjv′, xi, xj ∈ X, we have u < v
if either xi < xj or xi = xj, u′ < v′. We compare two words u and v from RS(X) by
R-degree: u < v if degR(u) < degR(v). If degR(u) = degR(v), we compare u and v in
deg-lex order as words in the alphabet X∞. Here we define each x from X to be less
than all R-letters and R(a) < R(b) if and only if a < b.
Let ∗ be a symbol not containing in X. By a ∗-bracketed word on X, we mean a basic
word from RAshX ∪ {∗}i with exactly one occurrence of ∗. The set of all ∗-bracketed
words on X is denoted by RS∗(X). For q ∈ RS∗(X) and u ∈ RAshXi, we define qu as
the bracketed word obtained by replacing the letter ∗ in q by u.
The order defined above is monomial, i.e., from u < v it follows that qu < qv for all
u, v ∈ RS(X) and q ∈ RS∗(X).
coefficient of ¯f in f is 1.
Given f ∈ RAshXi, by ¯f we mean the leading word in f . We call f monic if the
Definition 1 [28]. Let f, g ∈ RAshXi.
If there exist µ, ν, w ∈ RS(X) such that
w = ¯f µ = ν¯g with deg w < deg( ¯f ) + deg(¯g), then we define (f, g)w as f µ − νg and
call it the composition of intersection of f and g with respect to (µ, ν). If there exist
q ∈ RS∗(X) and w ∈ RS(X) such that w = ¯f = q¯g, then we define (f, g)q
w as f − qg
and call it the composition of inclusion of f and g with respect to q.
Definition 2 [28]. Let S be a subset of monic elements from RAshXi and w ∈ RS(X).
(1) For u, v ∈ RAshXi, we call u and v congruent modulo (S, w) and denote this by
u ≡ v mod (S, w) if u − v =P ciqisi with ci ∈ k, qi ∈ RS∗(X), si ∈ S and qisi < w.
(2) For f, g ∈ RAshXi and suitable w, µ, ν or q that give a composition of intersection
w, the composition is called trivial modulo
(f, g)w or a composition of inclusion (f, g)q
(S, w) if (f, g)w or (f, g)q
w ≡ 0 mod (S, w).
(3) The set S ⊂ RAshXi is called a Grobner(cid:22)Shirshov basis if, for all f, g ∈ S, all
w are trivial
compositions of intersection (f, g)w and all compositions of inclusion (f, g)q
modulo (S, w).
Theorem 3 [28]. Let S be a set of monic elements in RAshXi, let < be a monomial
ordering on RS(X) and let Id(S) be the R-ideal of RAshXi generated by S. If S is a
Grobner(cid:22)Shirshov basis in RAshXi, then RAshXi = kIrr(S) ⊕ Id(S) where Irr(S) =
RS(X) \ {q¯s q ∈ RS∗(X), s ∈ S} and Irr(S) is a linear basis of RAshXi/Id(S).
3 PBW-theorem for pre-Lie and postLie algebras
Let A be an associative algebra with an RB-operator R. Then the algebra A(−) is a
Lie RB-algebra under the product [x, y] = xy − yx and the same action of R.
6
Let L be a Lie RB-algebra with an RB-operator P of weight λ. Suppose that there
exists a subset X0 = {xα α ∈ Ω} in L such that X = {xα,k := P k(xα) k ∈ N, α ∈ Ω}
is a linear basis of L. Our goal is to construct the universal enveloping associative
RB-algebra of L (via Grobner(cid:22)Shirshov bases). This will lead us to the proof of the
Poincar´e(cid:22)Birkhoff(cid:22)Witt (PBW) theorem for the pairs (pre-Lie, preAs) and (postLie,
postAs).
We may assume that the set Ω is well-ordered, so we define an order < on the set X:
xα,k < xβ,l if α < β or α = β and k < l.
Consider the set S of the following elements in RAshXi:
xy − yx − [x, y], x > y, x, y ∈ X,
R(a)R(b) − R(R(a)b + aR(b) + λab),
R(R(z1)~xq1R(z2) . . . R(zs)~xqsxβ,rxk
β,r+1R(zs+1)) − ∆,
(7)
(8)
(9)
where ∆ is defined as the RHS of (5) for l = l1 + . . . + ls + s + 1 and
b1 = z1, b2 = fx1
q1, b3 = fx2
q1, . . . , bl1+1 = fxl1
bl1+...+ls−1+s = zs, bl1+...+ls−1+s+1 =fx1
q1, bl1+2 = z2, bl1+3 = fx1
ql, . . . , bl1+...+ls+s = fxls
q2, . . . ,
qs,
bl1+...+ls+s+1 = . . . = bl1+...+ls+s+k+1 = xβ,r, bl1+...+ls+s+k+2 = zs+1.
Here ~xqi = x1
qi . . . xli
qi for xj
qi ∈ X and x denotes xα,t−1 for x = xα,t ∈ X, i.e., R(x) = x.
In (8), (9), we have
s ≥ 1,
k, r ≥ 0,
a, b ∈ RS(X),
~xq1, . . . , ~xqs−1 ∈ S(X \ X0),
z2, . . . , zs ∈ RS(X) \ S(X),
~xqs ∈ S(X \ X0) ∪ ∅
and xβ,r is greater than any letter from ~xqs.
By R(z1) we denote either that z1 ∈ RS(X) \ S(X) or that R(z1) is absent, i.e.,
In particular, the values s = 1, k = 0,
R(z1) = ∅. The same holds for R(zs+1).
R(z1) = R(z2) = ~xq1 = ∅, transform (9) to the relation R(xβ,r) − xβ,r+1.
Remark 1. In (9), we use associative words ~xα ∈ S(X) instead of ordered polyno-
mials from k[X], otherwise we will have to reduce the products of such polynomials from
k[X] to the ordered ones in all possible compositions from S.
Theorem 4. The set S is a a Grobner(cid:22)Shirshov basis in RAshXi.
Proof. All compositions between two elements from (7) are trivial, as it is the
method to construct the universal enveloping associative algebra for a given Lie algebra.
Also, compositions of intersection between (8) and (8) are trivial, it is a way to get the
free associative RB-algebra. Thus, all compositions of intersection which are not at the
same time compositions of inclusion are trivial.
Let us compute a composition of inclusion between (7) and (9). Let
w = R(R(z1)~xq1R(z2) . . . R(zs)~xqsxβ,rxk
β,r+1R(zs+1))
(10)
satisfy all conditions described above. We apply the relation (7): xy − yx − [x, y] to the
subword
~xqj = ~x′
qj xy~x′′
qj ,
x > y, 1 ≤ j ≤ s.
7
Suppose that j < s. Since the image of an RB-operator is a subalgebra, [x, y] lies in the
linear combination of elements from S(X \ X0). Define w′ = wxy→yx and w′′ = wxy→[x,y],
here wα→β means the word w with the subword α replaced by β.
On the one hand, we have modulo (S, w)
w
(7)
≡ w′ + w′′
1
(9)
≡
k + 1(cid:0)R(z1) . . . R(zj)~xqj xy→yx+[x,y]R(zj+1) . . . R(zs)~xqsxk+1
β,r+1R(zs+1)
+ R(Σw′) + R(Σw′′)(cid:1),
where Σw is the expression in the RHS in the brackets under the action of R from (5).
On the other hand, modulo (S, w)
(9)
≡
w
1
k + 1(cid:0)R(z1) . . . R(zj)~xqj R(zj+1) . . . R(zs)~xqsxk+1
k + 1(cid:0)R(z1) . . . R(zj)~xqj xy→yx+[x,y]R(zj+1) . . . R(zs)~xqsxk+1
β,r+1R(zs+1) + R(Σw)(cid:1)
β,r+1R(zs+1) + R(Σw)(cid:1).
1
(7)
≡
So, the composition equals
(R(Σw) − R(Σw′) − R(Σw′′)) and we may rewrite it
briefly as
1
k + 1
1
k + 1Xs∈S
R(αs([x, y] + [x, y] + λ[x, y] − R−1([x, y]))βs)
for corresponding index set S and αs, βs ∈ RS(X). We get zero, since
[x, y] = [R(x), R(y)] = R([x, y] + [x, y] + λ[x, y])
and R has zero kernel on L.
Consider the case j = s. The triviality of the corresponding composition of inclu-
sion one can derive from the following fact. Denote as A(z1, ~xq1, z2, . . . , zs, ~xqs, zs+1) the
expression (4) for t = l1 + . . . + ls + s + 1,
b1 = z1, b2 = x1
q1
′, b3 = x2
q1
′, . . . , bl1+1 = xl1
q1
′
, bl1+2 = z2, bl1+3 = x1
q2
′, . . . ,
bl1+...+ls−1+s = zs, bl1+...+ls−1+s+1 = x1
ql
′, . . . , bl1+...+ls+s = xls
qs
′
, bl1+...+ls+s+1 = zs+1,
where z1, . . . , zs+1 and the words ~xqi = x1
one concerned xβ,r. We also have that the word ~xqs = x1
letter xβ,r, r ≥ 1, on the positions
qi . . . xli
qi satisfy the conditions for (9) except the
q1 contains the biggest
q1 . . . xls
K = {k1, k2, . . . , kp k1 < k2 < . . . < kp} ⊂ {1, 2, . . . , ls}.
So, the composition of inclusion is trivial if
A(z1, ~xq1, z2, . . . , zs, ~xqs, zs+1) ≡ 0 mod (S, w)
(11)
for w greater than all terns involved in A(. . .).
8
To prove (11), we will proceed on by induction on ls = ~xqs. For ls = 1, we are done
by (9).
Consider the equality
~xqs = ~xqs,0xp
β,r +
pXi=1
qsx2
x1
qs . . . xki−1
qs
. . . [xki
qs, wi]xp−i
β,r ,
(12)
where ~xqs,0 is obtained from ~xqs by arising all letters xβ,r with preserving order of all
remaining letters, wi is obtained from the word xki+1
qs by arising all p − i letters
xβ,r. For wi = w1
i ∈ X, the bracket [xki
. . . xls
qs, wi] in (12) means
i . . . wls−ki
, wj
i w2
qs
i
[xki
qs, wi] =
ls−kiXj=1
w1
i . . . wj−1
i
[xki
qs, wj
i ]wj+1
i
. . . wls−ki
i
.
By (9) and Lemma, we deduce that
A(z1, ~xq1, z2, . . . , zs, ~xqs, zs+1) ≡
pXi=1
ls−kiXj=1
A(z1, ~xq1, z2, . . . , zs, ~xqs(i, j), zs+1) mod (S, w),
where ~xqs(i, j) = x1
A(z1, . . . , zs, ~xqs(q, i), zs+1) ≡ 0 modulo (S, w) follows from the inductive hypothesis.
qs . . . xki−1
xp−i
β,r . The equality
i . . . wj−1
. . . wls−ki
i ]wj+1
qs, wj
. . . w1
[xki
qs
i
i
i
Consider a composition of inclusion between (8) and (9). Let w = R(w0) be defined
by (10) and b ∈ RS(X). At first, we have modulo (S, w)
R(w0)R(b)
(8)
≡ R(R(w0)b + w0R(b) + λw0b)
(9)
≡
1
k + 1
+ R(cid:0)R(z1) . . . R(zs)~xqsxβ,rxk
(9)
β,r+1R(zs+1)R(b) + λR(z1) . . . R(zs)~xqsxβ,rxk
R(cid:0)R(z1)~xq1R(z2) . . . R(zs)~xqsxk+1
≡ λR(cid:0)R(z1) . . . R(zs)~xqsxβ,rxk
k + 1(cid:0)R(cid:0)R(z1) . . . R(zs)~xqsxk+1
β,r+1R(zs+1)b + R(Σw)b(cid:1)
β,r+1R(zs+1)b(cid:1)
β,r+1R(zs+1)b + R(Σw)b(cid:1)
+ R(z1) . . . R(zs)~xqsxβ,rxk
1
+
β,r+1R(zs+1)b(cid:1)
where Σw is the expression in the RHS in the brackets under the action of R from (5).
At second, modulo (S, w)
β,r+1R(zs+1)R(b) + R(ΣwR(zs+1)→R(zs+1)R(b))(cid:1),
1
1
(8)
≡
R(w0)R(b)
(9)
≡
k + 1(cid:0)R(z1)~xq1R(z2) . . . R(zs)~xqsxk+1
β,r+1R(zs+1)R(b) + R(Σw)R(b)(cid:1)
So, the composition of inclusion multiplied by (k + 1) equals
k + 1(cid:0)R(z1) . . . R(zs)~xqsxk+1
u = R(cid:0)λ(k + 1)R(z1) . . . R(zs)~xqsxk+1
β,r+1R(zs+1)R(b) + R(R(Σw)b + ΣwR(b) + λΣwb)(cid:1).
β,r+1R(zs+1)b(cid:1)
+ R(ΣwR(zs+1)→R(zs+1)R(b)) − R(ΣwR(b) + λΣwb).
β,r+1R(zs+1)b + R(z1) . . . R(zs)~xqsxk+1
9
Depending on the last factor, Σw splits into the sum Σ′
formulas
wR(zs+1) + Σ′′
wzs+1. Applying the
λΣwb = λΣ′
wR(zs+1)b + λΣ′′
wR(zs+1)R(b) + Σ′′
wzs+1b,
wzs+1R(b),
ΣwR(b) = Σ′
ΣwR(zs+1)→R(zs+1)R(b) = Σ′
wR(zs+1)R(b) + Σ′′
w(R(zs+1)b + zs+1R(b) + λzs+1b),
we deduce
u = R(cid:0)R(z1) . . . R(zs)~xqsxk+1
+ λ(cid:0)(k + 1)R(z1) . . . R(zs)~xqsxβ,rxk
β,r+1R(zs+1)b
Writing down the sum R(Σ′′
Lemma.
β,r+1R(zs+1)b − Σ′
wR(zs+1)b(cid:1) + Σ′′
wR(zs+1)b(cid:1).
w, Σ′′
w and
wR(zs+1)b), u equals to zero by the definition of Σ′
Compute a composition of inclusion between (9) and (9). Suppose that we have w
defined by (10) and
zm = R(a1)~xt1R(a2) . . . R(an)~xtnxγ,oxm
γ,o+1R(an+1)
satisfying all conditions written above for (7).
Consider the case m ≤ s. By Σw, as earlier, we mean the expression in the RHS in
the brackets under the action of R from (5) for w. By ∆, denote the expression in the
RHS in the brackets under the action of R from (5) for zm. We also define
m = R(a1)~xt1R(a2) . . . R(an)~xtnxm+1
z′
γ,o+1R(an+1),
e∆ = ∆ − R(a1) . . . R(an)~xtnHR(an+1),
γ,o+1 +xγ,oxm
γ,o+1 + xγ,o+1xγ,oxm−1
γ,o+1+ . . . +xm
γ,o+1xγ,o,
H=
(−1)i(cid:18)m + 1
m+1Xi=2
i (cid:19)(cid:2)xγ,o, x(i−1)
γ,o+1(cid:3)xm+1−i
Σw = Σ′
sum Σ′
w + Σ′′
w and all others in Σ′′
w.
w, where we collect all summands from Σw with the factor R(zm) in the
On the one hand, modulo (S, w) we get
1
≡
(9), out
R(cid:0)R(z1)~xq1R(z2) . . . R(zs)~xqsxβ,rxk
β,r+1R(zs+1)(cid:1)
k + 1(cid:0)R(z1)~xq1R(z2) . . . ~xqm−1R(zm)~xqm . . . R(zs)~xqsxk+1
(k + 1)(m + 1)(cid:0)R(z1)~xq1R(z2) . . . ~xqm−1z′
+ R(z1)~xα1R(z2) . . . ~xqm−1R(∆)~xqm . . . R(zs)~xqsxk+1
(9), in
≡
1
β,r+1R(zs+1) + R(Σw)(cid:1)
m~xqm . . . R(zs)~xqsxk+1
β,r+1R(zs+1)
β,r+1R(zs+1) + R(Σ′
wR(zm)→z ′
m)
wR(zm)→R(∆)) + (m + 1)R(Σ′′
w).
(13)
+ R(Σ′
On the other hand,
R(cid:0)R(z1)~xq1R(z2) . . . R(zs)~xqsxβ,rxk
β,r+1R(zs+1)(cid:1)
R(cid:0)R(z1)~xq1R(z2) . . . ~xqm−1z′
m + 1
(9), in
≡
1
10
m~xqm . . . R(zs)~xqsxβ,rxk
β.r+1R(zs+1)
(9), out
≡
+ R(cid:0)R(z1)~xα1R(z2) . . . ~xqm−1R(∆)~xqm . . . R(zs)~xqsxβ,rxk
(k + 1)(m + 1)(cid:0)R(z1)~xα1R(z2) . . . ~xαm−1z′
1
m) − R(Σ′′
wzm→ e∆)
wR(zm)→z ′
+ R(z1) . . . ~xαm−1R(∆)~xαm . . . R(zs)~xαsxk+1
+ R(Σ′
Subtracting (14) from (13), we get
β,r+1R(zs+1)(cid:1)(cid:1)
m~xαm . . . R(zs)~xαsxk+1
β,r+1R(zs+1)
β,r+1R(zs+1) + R(Σwzm→∆)(cid:1) mod (S, w).
1
(k+1)(m+1) R(Σ′′
wzm→C), where
(14)
for D = H − (m + 1)xγ,oxm
C = ∆ −e∆ − (m + 1)zm = R(a1) . . . R(an)~xtr DR(an+1)
γ,o+1. Equality of D to zero follows from Lemma.
The proof in the case m = s + 1 is slightly different if only R(a1) = ∅ and n = 1. Then
we need to apply the equality A(cid:0)z1, ~xq1, z2, . . . , zs, ~xqsxβ,rxk
(S, w). Such application is correct, since all terms involved in it have less R-degree
than w.
γ,o+1, a2(cid:1) ≡ 0 modulo
β,r+1~xt1xm+1
are trivial.
It is easy to verify that the remaining compositions of inclusion between (7) and (8)
2
Corollary 1. The quotient A of RAshXi by Id(S) is the universal enveloping asso-
ciative RB-algebra for the Lie algebra L with the RB-operator R. Moreover, L injectively
embeds into A(−).
Proof. By (8), A is an associative RB-algebra. By (7) -- (9), we have that A is
enveloping of L for both: the Lie bracket [, ] and the action of R. Thus, A is an associative
enveloping of L.
Let us prove that A is the universal enveloping one. At first, A is generated by L.
At second, all elements from S are identities in the universal enveloping associative
RB-algebra URB(L).
Indeed, (7) are enveloping conditions for the product, (8) is the
RB-identity, the relations (9) as the application of (5) are direct consequences of the
RB-identity.
By Theorems 3 and 4 we get the injectivity of embedding L into A(−).
2
Let C be a pre- or post-Lie algebra with a linear basis Y . By Theorem 2, C can be
injectively embedded into the Lie algebra L with the RB-operator P of weight λ and
such subset X0 = {xα α ∈ Ω} ⊂ L that X = {xα,k := P k(xα) k ∈ N, α ∈ Ω} is a
linear basis of L. Here Ω is a well-ordered set. Then, by Corollary 1, we embed the Lie
RB-algebra L into its universal enveloping associative algebra A with the RB-operator R.
Thus, the subalgebra (in pre- or postalgebra sense) U(C) in A(R)
generated by the set Y
is an enveloping pre- or postassociative algebra of C.
λ
Now, we state the main result of the work, the analogue of the Poincar´e(cid:22)Birkhoff(cid:22)
Witt theorem for pre-Lie and postLie algebras.
Theorem 5. a) Let C be a pre- or post-Lie algebra, then U(C) is the universal
enveloping pre- or postassociative algebra of C.
b) [13] The pairs (preLie, preAs) and (postLie, postAs) are PBW-pairs.
Proof. a) Consider the postLie algebra case, the proof when C is a pre-Lie algebra
is the same. Let Y be a basis of C. It is easy to show that A is the universal enveloping
associative RB-algebra for C in the sense of the equalities (2). Indeed, given an enveloping
11
associative RB-algebra D of C, define M as the Lie RB-subalgebra of D(−) generated by
the image of C. By the universality of L, we have that M is the homomorphic image
of the Lie RB-algebra L. Thus, D as the enveloping of M is the homomorphic image of
the universal enveloping associative RB-algebra of M. The last one is the homomorphic
image of A.
Consider the universal enveloping postassociative algebra V (C) of C. Due to Theo-
rem 2, we may embed V (C) in its universal enveloping associative RB-algebra Z with
RB-operator Q. Since Z is the homomorphic image of A, V (C) as the subalgebra in Z (Q)
generated by Y is the homomorphic image of U(C).
λ
b) We get it by the construction.
Corollary 2. The pairs (preLie, RBAs) and (postLie, RBAs) are PBW-pairs.
Corollary 3. The universal enveloping associative RB-algebra of U(C) is isomorphic
2
to A.
As another corollary, we obtain the commutative diagram from Introduction.
4 Universal enveloping pre/post-associative algebra
4.1 Post-Lie case
Given a postLie algebra C with a linear basis Y , we want to construct a linear basis
of the universal enveloping postassociative algebra U(C). Define Y + = Y ∪ Y ′ ⊂ X ⊂ L
for Y ′ = {P (y) y ∈ Y }.
Given a well-ordered set Z, define Com(Z) := {w = w1 . . . wk ∈ S(Z) wj ∈ Z, w1 ≤
w2 ≤ . . . ≤ wk}.
Let us define a set E ⊂ RS(Y ) by induction. At first, Y ⊂ E. At second, the word
a = R(z1)w1R(z2) . . . R(zs)wsR(zs+1)
lies in E for s ≥ 1, w1, . . . , ws ∈ Com(Y +), z2, . . . , zs ∈ E ∩RS(Y )\Y , z1 ∈ E ∩RS(Y )\Y
or R(z1) = ∅ (the same holds for zs+1), if
1) at least one of wi contains a letter from Y ,
2) every zi is not of the following form
R(q1)u1R(q2) . . . R(qt)utyR(y)kR(qt+1)
(15)
with the same conditions written for a. Moreover, u1, . . . , ut ∈ Com(Y ′) and y ∈ Y is
greater than all letters from ut.
Theorem 6. The set {e + Id(S) e ∈ E} forms a linear basis of U(C).
Proof. At first, Theorems 3 and 4 imply that the set {e + Id(S) e ∈ E} is a
linearly independent set of elements in RAshY i.
At second, we show that U(C) ⊂ kE + Id(S). Let us prove it by induction of the
summary R-degree r of factors a, b involved in the process of generating the algebra U(C).
By the definition, Y ⊂ E. If r = 0, we get only linear combinations of elements of the
form w ∈ S(Y +) with at least one letter from Y , where we identify R(y) with P (y) for
every y ∈ Y . Because of (7), we have w ∈ E + Id(S).
12
Let us prove the inductive step for r > 0. Suppose that a, b ∈ E, more precisely,
a = R(z1)w1R(z2) . . . R(zs)wsR(zs+1),
b = R(q1)u1R(q2) . . . R(qt)utR(qt+1),
where
s, t ≥ 1, w1, . . . , ws, u1, . . . , ut ∈ S(Y +),
z2, . . . , zs, q2, . . . , qt ∈ E ∩ RS(Y ) \ Y,
by R(z1), as above, we mean that either z1 ∈ E ∩ RS(Y ) \ Y or R(z1) = ∅. The same
holds for R(zs+1), R(q1), R(qt+1).
To prove Theorem, it is enough to state that a ≻ b, a ≺ b, a · b ∈ kE + Id(S). We
have modulo Id(S)
R(z1)w1 . . . R(zs)wsu1R(q2) . . . utR(qt+1),
R(z1)w1 . . . R(zs)wsR(zs+1 ◦ q1)u1R(q2) . . . utR(qt+1), R(zs+1), R(q1) 6= ∅,
ab,
R(zs+1) = R(q1) = ∅,
otherwise,
a · b =
where zs+1 ◦ q1 = zs+1 ≻ q1 + zs+1 ≺ q1 + zs+1 · q1.
In the third case, we have an
element from E. In the first one, we need to express wsu1 ∈ U(L) via basic elements
from Com(Y +) and so a · b ∈ kE. Finally, in the second case, after rewriting zs+1 ◦ q1 as
a linear combination of elements from E + Id(S), we only need to check the condition 2
of the definition of E.
If it's required, we apply the relation (9) and we are done by
induction on r.
We have modulo Id(S)
a ≻ b =(R(a)u1R(q2) . . . utR(qt+1),
R(q1) = ∅,
R(a ◦ q1)u1R(q2) . . . utR(qt+1), R(q1) 6= ∅.
By the inductive hypothesis, a ◦ q1 ∈ kE + Id(S). We need to check the condition 2 of
the definition of E for the first R-letter of the product. As above, we apply (9).
The case of a ≺ b can be considered analogously.
2
Remark 2. We may reformulate Theorem 6 in terms avoiding Id(S) and any factor-
ization. For this, we need to define by induction the products ≻, ≺, · on the space kE.
4.2 Pre-Lie case
Given a pre-Lie algebra C with a linear basis Y , we construct a linear basis of the
universal enveloping preassociative algebra U(C).
Let us define a set E ⊂ RS(Y ) by induction. At first, Y ⊂ E. At second, the word
a = R(z1)w1R(z2) . . . R(zs)wsR(zs+1)
lies in E for s ≥ 1, w1, . . . , ws ∈ Com(Y +), z2, . . . , zs ∈ E ∩RS(Y )\Y , z1 ∈ E ∩RS(Y )\Y
or R(z1) = ∅ (the same holds for zs+1), if
1) the only wi contains a letter from Y ,
13
2) every zi is not of the following form
R(q1)u1R(q2) . . . R(qt)utyR(y)kR(qt+1)
(16)
with the same conditions written for a; u1, . . . , ut ∈ Com(Y ′) and y ∈ Y is greater than
all letters from ut.
Theorem 7. The set {e + Id(S) e ∈ E} forms a linear basis of U(C).
Proof. Analogously to the proof of Theorem 6.
Example.
2
If C is a one-dimensional pre-Lie algebra with linear basis {y}, then
{y(P (y))k} is a linear basis of U(C). Indeed, because of the condition 2, all basic elements
have no R-letters, so we have unique y and several P (y) to form the elements from E.
Moreover, in the case C = Span{y} we have the isomorphism Ugr(C) ∼= preComhyi, where
Ugr(C) means the associated graded preassociative algebra obtained from U(C) by the
filtration by the degree and preComhZi means a free precommutative algebra generated
by the set Z. If C = Span{y} has the trivial product, then U(C) ∼= preComhyi.
Remark 3. Note that given a pre-Lie algebra C, the injective enveloping preassocia-
tive algebra T constructed in [24] is isomorphic to U(C), so it is the universal enveloping
one. In the current paper, we have embedded C into its universal enveloping Lie RB-
algebra, but in [24] C was embedded into the enveloping Lie RB-algebra C arisen from
the proof of Theorem 2. A posteriori, we conclude that it is enough to embed C into the
doubling (as a vector space) Lie RB-algebra to preserve all required connections for the
construction of the universal enveloping preassociative algebra U(C).
Acknowledgements
This work was supported by the Austrian Science Foundation FWF grant P28079.
References
[1] M. Aguiar, Pre-Poisson algebras, Lett. Math. Phys. 54 (2000) 263 -- 277.
[2] C. Bai, O. Bellier, L. Guo, X. Ni, Splitting of operations, Manin products, and
Rota(cid:22)Baxter operators, Int. Math. Res. Notices 3 (2013) 485 -- 524.
[3] C. Bai, L. Guo, X. Ni, O-operators on associative algebras, associative Yang(cid:22)Baxter
equations and dendriform algebras, Conference Proceedings Quantized Algebra and
Physics (2012) 10 -- 51.
[4] G. Baxter, An analytic problem whose solution follows from a simple algebraic iden-
tity, Pacific J. Math. 10 (1960) 731 -- 742.
[5] A.A. Belavin, V.G. Drinfel'd, Solutions of the classical Yang-Baxter equation for
simple Lie algebras, Funct. Anal. Appl. 16 (3) (1982) 159 -- 180.
[6] L.A. Bokut, Yu. Chen, X. Deng, Grobner-Shirshov bases for Rota-Baxter algebras,
Sib. Math. J. 51 (6) (2010) 978 -- 988.
14
[7] D. Burde, Left-symmetric, or pre-Lie algebras in geometry and physics, Cent. Eur.
J. Math., 4 (3), 325 -- 357 (2006).
[8] D. Burde, K. Dekimpe and K. Vercammen, Affine actions on Lie groups and post-Lie
algebra structures, Linear Algebra Appl. 437 (2012), no. 5, 1250 -- 1263.
[9] P. Cartier, On the structure of free Baxter algebras, Adv. Math. 9 (1972) 253 -- 265.
[10] F. Chapoton, Un th´eor`eme de Cartier-Milnor-Moore-Quillen pour les big`ebre den-
driformes et les alg`ebre braces, J. Pure Appl. Algebra 168 (2002), 1 -- 18.
[11] Yu. Chen, Q. Mo, Embedding dendriform algebra into its universal enveloping Rota(cid:22)
Baxter algebra, Proc. Amer. Math. Soc. 139 (12) (2011) 4207 -- 4216.
[12] I. Dokas, Pre-Lie algebras in positive characteristic, J. Lie Theory 23 (4) (2013)
937 -- 952.
[13] V. Dotsenko, P. Tamaroff, Endofunctors and Poincar´e-Birkhoff-Witt theorems,
arXiv:1804.06485 [math.CT].
[14] K. Ebrahimi-Fard, Loday-type algebras and the Rota-Baxter relation, Lett. Math.
Phys. 61 (2002) 139 -- 147.
[15] K. Ebrahimi-Fard, L. Guo, Rota(cid:22)Baxter algebras and dendriform algebras, J. Pure
Appl. Algebra 212 (2) (2008) 320 -- 339.
[16] K. Ebrahimi-Fard, A. Lundervold, I. Mencattini, H. Z. Munthe-Kaas, Post-Lie Al-
gebras and Isospectral Flows, Symmetry Integr. Geom. 11 (2015), Paper 093, 16 pp.
[17] M. Gerstenhaber, The cohomology structure of an associative ring, Ann. of Math.
78 (1963) 267 -- 288.
[18] V. Ginzburg, M. Kapranov, Koszul duality for operads, Duke Math. J. 76 (1) (1994)
203 -- 272.
[19] I.Z. Golubchik, V.V. Sokolov, Generalized operator Yang-Baxter equations, inte-
grable ODEs and nonassociative algebras, J. Nonlinear Math. Phys. 7 (2) (2000)
184 -- 197.
[20] V. Gubarev, Universal enveloping commutative Rota(cid:22)Baxter algebras of pre- and
post-commutative algebras, Axioms 6 (4) (2017) 1 -- 33.
[21] V. Gubarev, Universal enveloping associative Rota(cid:22)Baxter algebras of preassocia-
tive and postassociative algebras, J. Algebra 516 (2018) 298 -- 328.
[22] V. Gubarev, Universal enveloping Lie Rota(cid:22)Baxter algebras of pre-Lie and post-Lie
algebras, Algebra and Logic (accepted), arXiv:1708.06747.
[23] V. Gubarev, Embedding of postLie algebras into postassociative algebras,
arXiv:1808.08839 [math.RA].
15
[24] V. Gubarev, Embedding of pre-Lie algebras into preassociative algebras, Algebr.
Colloq. (accepted), arXiv:1808.09822 [math.RA].
[25] V. Gubarev, P. Kolesnikov, Embedding of dendriform algebras into Rota(cid:22)Baxter
algebras, Cent. Eur. J. Math. 11 (2) (2013) 226 -- 245.
[26] V. Gubarev, P. Kolesnikov, Operads of decorated trees and their duals, Comment.
Math. Univ. Carolin. 55 (4) (2014) 421 -- 445.
[27] L. Guo, An Introduction to Rota(cid:22)Baxter Algebra, Intern. Press, Somerville, MA;
Higher education press, Beijing, 2012.
[28] L. Guo, W. Sit, and R. Zhang, Differential type operators and Grobner-Shirshov
bases, J. Symb. Comput. 52 (2013) 97 -- 123.
[29] P. Kolesnikov, Grobner(cid:22)Shirshov bases for pre-associative algebras, Commun. Al-
gebra 45 (12) (2017) 5283 -- 5296.
[30] J.-L. Koszul, Domaines born´es homog`enes et orbites de groupes de transformations
affines, Bull. Soc. Math. France, 89, 515 -- 533 (1961).
[31] J.-L. Loday, Cup-product for Leibniz cohomology and dual Leibniz algebras, Math.
Scand. 77 (2) (1995) 189 -- 196.
[32] J.-L. Loday, Dialgebras, Dialgebras and related operads, Springer-Verl., Berlin, 2001,
1 -- 61.
[33] J.-L. Loday, On the algebra of quasi-shuffles, Manuscripta Math. 123 (2007) 79 -- 93.
[34] J.-L. Loday, T. Pirashvili, Universal enveloping algebras of Leibniz algebras and
(co)homology, Math. Ann. 296 (1993) 139 -- 158.
[35] J.-L. Loday, M. Ronco, Trialgebras and families of polytopes, Comtep. Math. 346
(2004) 369 -- 398.
[36] D. Manchon, A short survey on pre-Lie algebras, Noncommutative Geometry and
Physics: Renormalisation, Motives, Index Theory, 89 -- 102 (2011).
[37] A.A. Mikhalev, I.P. Shestakov, PBW-pairs of varieties of linear algebras, Commun.
Algebra 42 (2) (2014) 667 -- 687.
[38] Yu Pan, Q. Liu, C. Bai, L. Guo, PostLie algebra structures on the Lie algebra
sl(2, C), Electron. J. Linear Al., 23, 180 -- 197 (2012).
[39] M. Ronco, Eulerian idempotents and Milnor(cid:22)Moore theorem for certain non-
cocommutative Hopf algebras, J. Algebra 254 (2002), 152 -- 172.
[40] G.-C. Rota, Baxter algebras and combinatorial identities I, Bull. Amer. Math. Soc.
75 (1969) 325 -- 329.
16
[41] M.A. Semenov-Tyan-Shanskii, What is a classical r-matrix? Funct. Anal. Appl. 17
(4) (1983) 259 -- 272.
[42] B. Vallette, Homology of generalized partition posets, J. Pure Appl. Algebra 208 (2)
(2007) 699 -- 725.
[43] E.B. Vinberg, The theory of homogeneous convex cones, Tr. Mosk. Mat. Obs., 12,
303 -- 358 (1963).
Vsevolod Gubarev
University of Vienna
Oskar-Morgenstern-Platz 1, 1090, Vienna, Austria
e-mail: [email protected]
17
|
1208.1939 | 3 | 1208 | 2013-05-29T15:05:16 | Two cores of a nonnegative matrix | [
"math.RA"
] | We prove that the sequence of eigencones (i.e., cones of nonnegative eigenvectors) of positive powers A^k of a nonnegative square matrix A is periodic both in max algebra and in nonnegative linear algebra. Using an argument of Pullman, we also show that the Minkowski sum of the eigencones of powers of A is equal to the core of A defined as the intersection of nonnegative column spans of matrix powers, also in max algebra. Based on this, we describe the set of extremal rays of the core. The spectral theory of matrix powers and the theory of matrix core is developed in max algebra and in nonnegative linear algebra simultaneously wherever possible, in order to unify and compare both versions of the same theory. | math.RA | math | TWO CORES OF A NONNEGATIVE MATRIX
PETER BUTKOVI C, HANS SCHNEIDER, SERGEI SERGEEV, AND BIT-SHUN TAM
Abstract. We prove that the sequence of eigencones (i.e., cones of nonnegative eigen-
vectors) of positive powers Ak of a nonnegative square matrix A is periodic both in max
algebra and in nonnegative linear algebra. Using an argument of Pullman, we also show
that the Minkowski sum of the eigencones of powers of A is equal to the core of A defined
as the intersection of nonnegative column spans of matrix powers, also in max algebra.
Based on this, we describe the set of extremal rays of the core.
The spectral theory of matrix powers and the theory of matrix core is developed in
max algebra and in nonnegative linear algebra simultaneously wherever possible, in order
to unify and compare both versions of the same theory.
Keywords: Max algebra, nonnegative matrix theory, Perron-Frobenius theory, matrix
power, eigenspace, core.
AMS Classification: 15A80, 15A18, 15A03,15B48
3
1
0
2
y
a
M
9
2
]
.
A
R
h
t
a
m
[
3
v
9
3
9
1
.
8
0
2
1
:
v
i
X
r
a
1. Introduction
The nonnegative reals R+ under the usual multiplication give rise to two semirings with
addition defined in two ways: first with the usual addition, and second where the role of
addition is played by maximum. Thus we consider the properties of nonnegative matrices
with entries in two semirings, the semiring of nonnegative numbers with usual addition and
multiplication called "nonnegative algebra", and the semiring called "max(-times)
algebra".
Our chief object of study is the core of a nonnegative matrix A. This concept was
introduced by Pullman in [33], and is defined as the intersection of the cones generated
by the columns of matrix powers Ak. Pullman provided a geometric approach to the
Perron-Frobenius theory of nonnegative matrices based on the properties of the core. He
This research was supported by EPSRC grant RRAH15735. Sergeı Sergeev also acknowledges the
support of RFBR-CNRS grant 11-0193106 and RFBR grant 12-01-00886. Bit-Shun Tam acknowledges
the support of National Science Council of the Republic of China (Project No. NSC 101-2115-M-032-007).
1
2
PETER BUTKOVI C, HANS SCHNEIDER, SERGEI SERGEEV, AND BIT-SHUN TAM
investigated the action of a matrix on its core showing that it is bijective and that the
extremal rays of the core can be partitioned into periodic orbits. In other words, extremal
rays of the core of A are nonnegative eigenvectors of the powers of A (associated with
positive eigenvalues).
One of the main purposes of the present paper is to extend Pullman's core to max
algebra, thereby investigating the periodic sequence of eigencones of max-algebraic matrix
powers. However, following the line of [10, 11, 24], we develop the theory in max algebra
and nonnegative algebra simultaneously, in order to emphasize common features as well
as differences, to provide general (simultaneous) proofs where this is possible. We do not
aim to obtain new results, relative to [33, 43], on the usual core of a nonnegative matrix.
However, our unifying approach leads in some cases (e.g., Theorem 6.5 (iii)) to new and
more elementary proofs than those given previously. Our motivation is closely related to
the Litvinov-Maslov correspondence principle [27], viewing the idempotent mathematics
(in particular, max algebra) as a "shadow" of the "traditional" mathematics over real and
complex fields.
To the authors' knowledge, the core of a nonnegative matrix has not received much
attention in linear algebra. However, a more detailed study has been carried out by Tam
and Schneider [43], who extended the concept of core to linear mappings preserving a
proper cone. The case when the core is a polyhedral (i.e., finitely generated) cone was
examined in detail in [43, Section 3], and the results were applied to study the case of
nonnegative matrix in [43, Section 4]. This work has found further applications in the
theory of dynamic systems acting on the path space of a stationary Bratteli diagram.
In particular, Bezuglyi et al. [4] describe and exploit a natural correspondence between
ergodic measures and extremals of the core of the incidence matrix of such a diagram.
On the other hand, there is much more literature on the related but distinct question
of the limiting sets of homogeneous and non-homogeneous Markov chains in nonnegative
algebra; see the books by Hartfiel [22] and Seneta [37] and, e.g., the works of Chi [13]
and Sierksma [42]. In max algebra, see the results on the ultimate column span of matrix
powers for irreducible matrices ([7, Theorem 8.3.11], [38]), and by Merlet [28] on the
invariant max cone of non-homogeneous matrix products.
TWO CORES OF A NONNEGATIVE MATRIX
3
The theory of the core relies on the behaviour of matrix powers. In the nonnegative
algebra, recall the works of Friedland-Schneider [17] and Rothblum-Whittle [34] (on the
role of distinguished classes which we call "spectral classes", algebraic and geometric
growth rates, and various applications). The theory of max-algebraic matrix powers is
similar. However, the max-algebraic powers have a well-defined periodic ultimate be-
haviour starting after sufficiently large time. This ultimate behaviour has been known
since the work of Cuninghame-Green [15, Theorem 27-9], Cohen et al. [14] (irreducible
case), and is described in greater generality and detail, e.g., by Akian, Gaubert and
Walsh [1], Gavalec [21], De Schutter [36], and the authors [7, 39, 40] of the present paper.
In particular, the Cyclicity Theorem of Cohen et al. [2, 7, 14, 23]) implies that extremals
of the core split into periodic orbits for any irreducible matrix (see Subsection 4.2 below)1.
Some results on the eigenvectors of max-algebraic matrix powers have been obtained
by Butkovic and Cuninghame-Green [7, 8]. The present paper also aims to extend and
complete the research initiated in that work.
This paper is organized as follows. In Section 2 we introduce the basics of irreducible
and reducible Perron-Frobenius theory in max algebra and in nonnegative linear algebra.
In Section 3 we formulate the two key results of this paper. The first key result is Main
Theorem 1 stating that the matrix core equals to the Minkowski sum of the eigencones
of matrix powers (that is, for each positive integer k, we take the sum of the eigencones
associated with Ak, and then we sum over all k). The second key result is Main Theo-
rem 2 stating that the sequence of eigencones of matrix powers is periodic and defining
the period. This section also contains a table of notations used throughout the paper.
Section 4 is devoted to the proof of Main Theorem 1, taking in "credit" the result of Main
Theorem 2 (whose proof is deferred to the end of the paper). In Section 5 we explain
the relation between spectral classes of different matrix powers, and how the eigencones
associated with general eigenvalues can be reduced to the case of the greatest eigenvalue,
see in particular Theorems 5.4 and 5.7. In Section 6 we describe extremals of the core
1In fact, many of the cited works and monographs like [2, 7, 21, 23] are written in the setting of max-
plus algebra. However, this algebra is isomorphic to the max algebra considered here, so the results
can be readily translated to the present (max-times) setting.
4
PETER BUTKOVI C, HANS SCHNEIDER, SERGEI SERGEEV, AND BIT-SHUN TAM
in both algebras extending [43, Theorem 4.7], see Theorem 6.5. Prior to this result we
formulate the Frobenius-Victory Theorems 6.1 and 6.2 giving a parallel description of
extremals of eigencones in both algebras. In Section 7, our first goal is to show that the
sequence of eigencones of matrix powers in max algebra is periodic, comparing this result
with the case of nonnegative matrix algebra, see Theorem 7.1. Then we study the inclu-
sion relation on eigencones and deduce Main Theorem 2. The key results are illustrated
by a pair of examples in Section 8.
2. Preliminaries
2.1. Nonnegative matrices and associated graphs. In this paper we are concerned
only with nonnegative eigenvalues and nonnegative eigenvectors of a nonnegative matrix.
In order to bring our terminology into line with the corresponding theory for max algebra
we use the terms eigenvalue and eigenvector in a restrictive fashion appropriate to our
semiring point of view. Thus we shall call ρ an eigenvalue of a nonnegative matrix A (only)
if there is a nonnegative eigenvector x of A for ρ. Further x will be called an eigenvector
(only) if it is nonnegative. (In the literature ρ is called a distinguished eigenvalue and x
a distinguished eigenvector of A.) For x ∈ Rn
+, the support of x, denoted by supp(x), is
the set of indices where xi > 0.
In this paper we are led to state the familiar Perron-Frobenius theorem in slightly
unusual terms: An irreducible nonnegative matrix A has a unique eigenvalue denoted
by ρ+(A), which is positive (unless A is the 1 × 1 matrix 0). Further, the eigenvector
x associated with ρ+(A) is essentially unique, that is all eigenvectors are multiples of x.
The nonnegative multiples of x constitute the cone of eigenvectors (in the above sense)
V+(A, ρ+(A)) associated with ρ+(A).
A general (reducible) matrix A ∈ Rn×n
+ may have several nonnegative eigenvalues with
associated cones of nonnegative eigenvectors (eigencones), and ρ+(A) will denote the
biggest such eigenvalue, in general. Eigenvalue ρ+(A) is also called the principal eigen-
value, and V+(A, ρ+(A)) is called the principal eigencone.
Recall that a subset V ⊆ Rn
+ is called a (convex) cone if 1) αv ∈ V for all v ∈ V
and α ∈ R+, 2) u + v ∈ V for u, v ∈ V . Note that cones in the nonnegative orthant
TWO CORES OF A NONNEGATIVE MATRIX
5
can be considered as "subspaces", with respect to the semiring of nonnegative numbers
by S ⊆ Rn
(with usual addition and multiplication). In this vein, a cone V is said to be generated
+ if each v ∈ V can be represented as a nonnegative combination v = Px∈S αxx
where only finitely many αx ∈ R+ are different from zero. When V is generated (we also
say "spanned") by S, this is denoted V = span+(S). A vector z in a cone V is called an
extremal, if z = u + v and u, v ∈ V imply z = αuu = αvv for some scalars αu and αv. Any
closed cone in Rn
+ is generated by its extremals; in particular, this holds for any finitely
generated cone.
Let us recall some basic notions related to (ir)reducibility, which we use also in max
algebra. With a matrix A = (aij) ∈ Rn×n
+ we associate a weighted (di)graph G(A) with
the set of nodes N = {1, . . . , n} and set of edges E ⊆ N × N containing a pair (i, j) if
and only if aij 6= 0; the weight of an edge (i, j) ∈ E is defined to be w(i, j) := aij. A
graph with just one node and no edges will be called trivial. A graph with at least one
node and at least one edge will be called nontrivial.
A path P in G(A) consisting2 of the edges (i0, i1), (i1, i2), . . . , (it−1, it) has length l(P ) :=
t and weight w(P ) := w(i0, i1) · w(i1, i2) · · · w(it−1, it), and is called an i − j path if i0 = i
and it = j. P is called a cycle if i0 = it. P is an elementary cycle, if, further, ik 6= il for
all k, l ∈ {1, . . . , t − 1}.
Recall that A = (aij) ∈ Rn×n
+
is irreducible if G(A) is trivial or for any i, j ∈ {1, . . . , n}
there is an i − j path. Otherwise A is reducible.
Notation A×k will stand for the usual kth power of a nonnegative matrix.
2.2. Max algebra. By max algebra we understand the set of nonnegative numbers R+
where the role of addition is played by taking maximum of two numbers: a⊕b := max(a, b),
and the multiplication is as in the usual arithmetics. This is carried over to matrices and
vectors like in the usual linear algebra so that for two matrices A = (aij) and B = (bij)
of appropriate sizes, (A ⊕ B)ij = aij ⊕ bij and (A ⊗ B)ij = Lk aikbkj. Notation A⊗k will
stand for the kth max-algebraic power.
2In our terminology, a path can visit some nodes more than once.
6
PETER BUTKOVI C, HANS SCHNEIDER, SERGEI SERGEEV, AND BIT-SHUN TAM
In max algebra, we have the following analogue of a convex cone. A set V ⊆ Rn
+ is
called a max cone if 1) αv ∈ V for all v ∈ V and α ∈ R+, 2) u ⊕ v ∈ V for u, v ∈ V .
Max cones are also known as idempotent semimodules [26, 27]. A max cone V is said
to be generated by S ⊆ Rn
+ if each v ∈ V can be represented as a max combination
v = Lx∈S αxx where only finitely many (nonnegative) αx are different from zero. When
V is generated (we also say "spanned") by S, this is denoted V = span⊕(S). When V
is generated by the columns of a matrix A, this is denoted V = span⊕(A). This cone is
closed with respect to the usual Euclidean topology [10].
A vector z in a max cone V ⊆ Rn
+ is called an extremal if z = u ⊕ v and u, v ∈ V
imply z = u or z = v. Any finitely generated max cone is generated by its extremals, see
Wagneur [45] and [10, 20] for recent extensions.
The maximum cycle geometric mean of A is defined by
(1)
λ(A) = max{w(C)1/l(C) : C is a cycle in G(A)} .
The cycles with the cycle geometric mean equal to λ(A) are called critical, and the nodes
and the edges of G(A) that belong to critical cycles are called critical. The set of critical
nodes is denoted by Nc(A), the set of critical edges by Ec(A), and these nodes and edges
give rise to the critical graph of A, denoted by C(A) = (Nc(A), Ec(A)). A maximal strongly
connected subgraph of C(A) is called a strongly connected component of C(A). Observe
that C(A), in general, consists of several nontrivial strongly connected components, and
that it never has any edges connecting different strongly connected components.
If for A ∈ Rn×n
+ we have A ⊗ x = ρx with ρ ∈ R+ and a nonzero x ∈ Rn
+, then ρ is a
max(-algebraic) eigenvalue and x is a max(-algebraic) eigenvector associated with ρ. The
set of max eigenvectors x associated with ρ, with the zero vector adjoined to it, is a max
cone. It is denoted by V⊕(A, ρ).
An irreducible A ∈ Rn×n
+
has a unique max-algebraic eigenvalue equal to λ(A) [2,
7, 15, 23].
In general A may have several max eigenvalues, and the greatest of them
equals λ(A). The greatest max eigenvalue will also be denoted by ρ⊕(A) (thus ρ⊕(A) =
λ(A)), and called the principal max eigenvalue of A. In the irreducible case, the unique
max eigenvalue ρ⊕(A) = λ(A) is also called the max(-algebraic) Perron root. When
TWO CORES OF A NONNEGATIVE MATRIX
7
max algebra and nonnegative algebra are considered simultaneously (e.g., Section 3), the
principal eigenvalue is denoted by ρ(A).
Unlike in nonnegative algebra, there is an explicit description of V⊕(A, ρ⊕(A)), see
Theorem 6.2. This description uses the Kleene star
(2)
A∗ = I ⊕ A ⊕ A⊗2 ⊕ A⊗3 ⊕ . . . .
Series (2) converges if and only if ρ⊕(A) ≤ 1, in which case A∗ = I ⊕A⊕. . .⊕A⊗(n−1) [2, 7,
23]. Note that if ρ⊕(A) 6= 0, then ρ⊕(A/ρ⊕(A)) = 1, hence (A/ρ⊕(A))∗ always converges.
The path interpretation of max-algebraic matrix powers A⊗l is that each entry a⊗l
ij
is
equal to the greatest weight of i − j path with length l. Consequently, for i 6= j, the entry
a∗
ij of A∗ is equal to the greatest weight of an i − j path (with no length restrictions).
2.3. Cyclicity and periodicity. Consider a nontrivial strongly connected graph G (that
is, a strongly connected graph with at least one node and one edge). Define its cyclicity
σ as the gcd of the lengths of all elementary cycles. It is known that for any vertices i, j
there exists a number l such that l(P ) ≡ l(mod σ) for all i − j paths P .
When the length of an i − j path is a multiple of σ (and hence we have the same for
all j − i paths), i and j are said to belong to the same cyclic class. When the length
of this path is 1 modulo σ (in other words, if l(P ) − 1 is a multiple of σ), the cyclic
class of i (resp., of j) is previous (resp., next) with respect to the class of j (resp., of i).
See [7, Chapter 8] and [5, 38, 39] for more information. Cyclic classes are also known as
components of imprimitivity [5].
The cyclicity of a trivial graph is defined to be 1, and the unique node of a trivial graph
is defined to be its only cyclic class.
We define the cyclicity of a (general) graph containing several strongly connected com-
ponents to be the lcm of the cyclicities of the components.
For a graph G = (N, E) with N = {1, . . . , n}, define the associated matrix A = (aij) ∈
{0, 1}n×n by aij = 1 ⇔ (i, j) ∈ E. This is a matrix over the Boolean semiring B := {0, 1},
where addition is the disjunction and multiplication is the conjunction operation. This
semiring is a subsemiring of max algebra, so that it is possible to consider the associated
matrix as a matrix in max algebra whose entries are either 0 or 1.
8
PETER BUTKOVI C, HANS SCHNEIDER, SERGEI SERGEEV, AND BIT-SHUN TAM
For a graph G and any k ≥ 1, define Gk as a graph that has the same vertex set as
G and (i, j) is an edge of Gk if and only if there is a path of length k on G connecting i
to j. Thus, if a Boolean matrix A is associated with G, then the Boolean matrix power
A⊗k is associated with Gk. Powers of Boolean matrices (over the Boolean semiring) are a
topic of independent interest, see Brualdi-Ryser [5], Kim [25]. We will need the following
observation.
Theorem 2.1 (cf. [5, Theorem 3.4.5]). Let G be a strongly connected graph with cyclicity
σ.
(i) Gk consists of gcd (k, σ) nontrivial strongly connected components not accessing
each other. If G is nontrivial, then so are all the components of Gk.
(ii) The node set of each component of Gk consists of σ/(gcd(k, σ)) cyclic classes of G.
Corollary 2.2. Let G be a strongly connected graph with cyclicity σ, and let k, l ≥ 1.
Then gcd(k, σ) divides gcd(l, σ) if and only if Gk and Gl are such that the node set of
every component of Gl is contained in the node set of a component of Gk.
Proof. Assume that G is nontrivial.
"If". Since the node set of each component of Gk consists of σ/gcd(k, σ) cyclic classes of
G and is the disjoint union of the node sets of certain components of Gl, and the node
set of each component of Gl consists of σ/gcd(l, σ) cyclic classes of G, it follows that the
gcd(k,σ) components of Gl.
node set of each component of Gk consists of
gcd(k,σ) /
σ
σ
gcd(l,σ) = gcd(l,σ)
Therefore, gcd(k, σ) divides gcd(l, σ).
"Only if." Observe that the node sets of the compopnents Gk and Ggcd(k,σ) (or Gl
and Ggcd(l,σ)) are the same: since gcd(k, σ) divides k, each component of Ggcd(k,σ) splits
into several components of Gk, but the total number of components is the same (as
gcd(gcd(k, σ), σ) =gcd(k, σ)), hence their node sets are the same. The claim follows since
the node set of each component of Ggcd(k,σ) splits into several components of Ggcd(l,σ). (cid:3)
Let us formally introduce the definitions related to periodicity and ultimate periodicity
of sequences (whose elements are of arbitrary nature). A sequence {Ωk}k≥1 is called
periodic if there exists an integer p such that Ωk+p is identical with Ωk for all k. The least
TWO CORES OF A NONNEGATIVE MATRIX
9
such p is called the period of {Ωk}k≥1. A sequence {Ωk}k≥1 is called ultimately periodic if
the sequence {Ωk}k≥T is periodic for some T ≥ 1. The least such T is called the periodicity
threshold of {Ωk}k≥1.
The following observation is crucial in the theory of Boolean matrix powers.
Theorem 2.3 (Boolean Cyclicity [25]). Let G be a strongly connected graph on n nodes,
with cyclicity σ.
(i) The sequence {Gk}k≥1 is ultimately periodic with the period σ. The periodicity
threshold, denoted by T (G), does not exceed (n − 1)2 + 1.
(ii) If G is nontrivial, then for k ≥ T (G) and a multiple of σ, Gk consists of σ complete
subgraphs not accessing each other.
For brevity, we will refer to T (G) as the periodicity threshold of G. We have the
following two max-algebraic extensions of Theorem 2.3.
Theorem 2.4 (Cyclicity Theorem, Cohen et al. [14]). Let A ∈ Rn×n
+
be irreducible, let σ
be the cyclicity of C(A) and ρ := ρ⊕(A). Then the sequence {(A/ρ)⊗k}k≥1 is ultimately
periodic with period σ.
Theorem 2.5 (Cyclicity of Critical Part, Nachtigall [31]). Let A ∈ Rn×n
+ , σ be the cyclicity
of C(A) and ρ := ρ⊕(A). Then the sequences {(A/ρ)⊗k
·i }k≥1, for i ∈
Nc(A), are ultimately periodic with period σ. The greatest of their periodicity thresholds,
denoted by Tc(A), does not exceed n2.
i· }k≥1 and {(A/ρ)⊗k
Theorem 2.4 is standard [2, 7, 23], and Theorem 2.5 can also be found as [7, Theorem
8.3.6]. Here Ai· (resp. A·i) denote the ith row (resp. the ith column) of A.
When the sequence {(A/ρ)⊗k}k≥1 (resp. the sequences {(A/ρ)⊗k
i· }k≥1,
{(A/ρ)⊗k
{A⊗k
·i }k≥1) are ultimately periodic, we also say that the sequence {A⊗k}k≥1 (resp.
i· }k≥1, {A⊗k
Let us conclude with a well-known number-theoretic result concerning the coin problem
·i }k≥1) is ultimately periodic with growth rate ρ.
of Frobenius, which we see as basic for both Boolean and max-algebraic cyclicity.
10
PETER BUTKOVI C, HANS SCHNEIDER, SERGEI SERGEEV, AND BIT-SHUN TAM
Lemma 2.6 (e.g.,[5, Lemma 3.4.2]). Let n1, . . . , nm be integers such that
gcd(n1, . . . , nm) = k. Then there exists a number T such that for all integers l with kl ≥ T ,
we have kl = t1n1 + . . . + tmnm for some t1, . . . , tm ≥ 0.
2.4. Diagonal similarity and visualization. For any x ∈ Rn
+, we can define X =
diag(x) as the diagonal matrix whose diagonal entries are equal to the corresponding
entries of x, and whose off-diagonal entries are zero. If x does not have zero components,
the diagonal similarity scaling A 7→ X −1AX does not change the weights of cycles and
eigenvalues (both nonnegative and max); if z is an eigenvector of X −1AX then Xz is
an eigenvector of A with the same eigenvalue. This scaling does not change the critical
graph C(A) = (Nc(A), Ec(A)). Observe that (X −1AX)⊗k = X −1A⊗kX, also showing that
the periodicity thresholds of max-algebraic matrix powers (Theorems 2.4 and 2.5) do not
change after scaling. Of course, we also have (X −1AX)×k = X −1A×kX in nonnegative
algebra. The technique of nonnegative scaling can be traced back to the works of Fiedler-
Pt´ak [16].
When working with the max-algebraic matrix powers, it is often convenient to "visual-
ize" the powers of the critical graph. Let A have λ(A) = 1. A diagonal similarity scalling
A 7→ X −1AX is called a strict visualization scaling [7, 41] if the matrix B = X −1AX has
bij ≤ 1, and moreover, bij = 1 if and only if (i, j) ∈ Ec(A)(= Ec(B)). Any matrix B
satisfying these properties is called strictly visualized.
Theorem 2.7 (Strict Visualization [7, 41]). For each A ∈ Rn×n
+ with ρ⊕(A) = 1 (that is,
λ(A) = 1), there exists a strict visualization scaling.
If A = (aij) has all entries aij ≤ 1, then we define the Boolean matrix A[1] with entries
(3)
If A has all entries aij ≤ 1 then
a[1]
ij =
1,
0,
if aij = 1,
if aij < 1.
(4)
(A⊗k)[1] = (A[1])⊗k.
TWO CORES OF A NONNEGATIVE MATRIX
11
Similarly if a vector x ∈ Rn
i = 1 if xi = 1 and x[1]
i = 0
otherwise. Obviously if A and x have all entries not exceeding 1 then (A⊗x)[1] = A[1]⊗x[1].
+ has xi ≤ 1, we define x[1] having x[1]
If A is strictly visualized, then a[1]
ij = 1 if and only if (i, j) is a critical edge of G(A). Thus
A[1] can be treated as the associated matrix of C(A) (disregarding the formal difference
in dimension). We now show that C(A⊗k) = C(A)k and that any power of a strictly
visualized matrix is strictly visualized.
Lemma 2.8 (cf. [8], [38, Prop. 3.3]). Let A ∈ Rn×n
+
and k ≥ 1.
(i) C(A)k = C(A⊗k).
(ii) If A is strictly visualized, then so is A⊗k.
Proof. Using Theorem 2.7, we can assume without loss of generality that A is strictly
visualized. Also note that both in C(A⊗k) and in C(A)k, each node has ingoing and
outgoing edges, hence for part (i) it suffices to prove that the two graphs have the same
set of edges.
Applying Theorem 2.1 (i) to every component of C(A), we obtain that C(A)k also
consists of several isolated nontrivial strongly connected graphs. In particular, each edge
of C(A)k lies on a cycle, so C(A)k contains cycles. Observe that G(A⊗k) does not have edges
with weight greater than 1, while all edges of C(A)k have weight 1, hence all cycles of C(A)k
have weight 1. As C(A)k is a subgraph of G(A⊗k), this shows that ρ⊕(A⊗k) = λ(A⊗k) = 1
and that all cycles of C(A)k are critical cycles of G(A⊗k). Since each edge of C(A)k lies
on a critical cycle, all edges of C(A)k are critical edges of G(A⊗k).
G(A⊗k) does not have edges with weight greater than 1, hence every edge of C(A⊗k) has
weight 1. Equation (4) implies that if a⊗k
ij = 1 then there is a path from i to j composed
of the edges with weight 1. Since A is strictly visualized, such edges are critical. This
shows that if a⊗k
ij = 1 and in particular if (i, j) is an edge of C(A⊗k), then (i, j) is an edge
of C(A)k. Hence A⊗k is strictly visualized, and all edges of C(A⊗k) are edges of C(A)k.
Thus C(A⊗k) and C(A)k have the same set of edges, so C(A⊗k) = C(A)k (and we also
showed that A⊗k is strictly visualized).
(cid:3)
Let T (C(A)) be the greatest periodicity threshold of the strongly connected components
of C(A). The following corollary of Lemma 2.8 will be required in Section 7.
12
PETER BUTKOVI C, HANS SCHNEIDER, SERGEI SERGEEV, AND BIT-SHUN TAM
Corollary 2.9. Let A ∈ Rn×n
+ . Then Tc(A) ≥ T (C(A)).
2.5. Frobenius normal form. Every matrix A = (aij) ∈ Rn×n
+
can be transformed by
simultaneous permutations of the rows and columns in almost linear time to a Frobenius
normal form [3, 5]
(5)
A11
0
A21 A22
...
...
0
0
...
... Aµµ
...
Ar1 Ar2
... Arr
,
where A11, ..., Arr are irreducible square submatrices of A. They correspond to the sets
of nodes N1, . . . , Nr of the strongly connected components of G(A). Note that in (5) an
edge from a node of Nµ to a node of Nν in G(A) may exist only if µ ≥ ν.
Generally, AKL denotes the submatrix of A extracted from the rows with indices in
K ⊆ {1, . . . , n} and columns with indices in L ⊆ {1, . . . , n}, and Aµν is a shorthand for
ANµNν . Accordingly, the subvector xNµ of x with indices in Nµ will be written as xµ.
If A is in the Frobenius Normal Form (5) then the reduced graph, denoted by R(A),
is the (di)graph whose nodes correspond to Nµ, for µ = 1, . . . , r, and the set of arcs is
{(µ, ν); (∃k ∈ Nµ)(∃ℓ ∈ Nν)akℓ > 0}. In max algebra and in nonnegative algebra, the
nodes of R(A) are marked by the corresponding eigenvalues (Perron roots), denoted by
µ := ρ⊕(Aµµ) (max algebra), ρ+
µ := ρ+(Aµµ) (nonnegative algebra), and by ρµ when
ρ⊕
both algebras are considered simultaneously.
By a class of A we mean a node µ of the reduced graph R(A). It will be convenient to
attribute to class µ the node set and the edge set of G(Aµµ), as well as the cyclicity and
other parameters, that is, we will say "nodes of µ", "edges of µ", "cyclicity of µ", etc.3
A class µ is trivial if Aµµ is the 1 × 1 zero matrix. Class µ accesses class ν, denoted
µ → ν, if µ = ν or if there exists a µ − ν path in R(A). A class is called initial, resp.
final, if it is not accessed by, resp. if it does not access, any other class. Node i of G(A)
accesses class ν, denoted by i → ν, if i belongs to a class µ such that µ → ν.
3The sets Nµ are also called classes, in the literature. To avoid the confusion, we do not follow this in
the present paper.
TWO CORES OF A NONNEGATIVE MATRIX
13
Note that simultaneous permutations of the rows and columns of A are equivalent to
calculating P −1AP, where P is a permutation matrix. Such transformations do not change
the eigenvalues, and the eigenvectors before and after such a transformation may only
differ by the order of their components. Hence we will assume without loss of generality
that A is in Frobenius normal form (5). Note that a permutation bringing matrix to this
form is (relatively) easy to find [5]. We will refer to the transformation A 7→ P −1AP as
permutational similarity.
2.6. Elements of the Perron-Frobenius theory. In this section we recall the spectral
theory of reducible matrices in max algebra and in nonnegative linear algebra. All results
are standard: the nonnegative part goes back to Frobenius [18], Sect. 11, and the max-
algebraic counterpart is due to Gaubert [19], Ch. IV (also see [7] for other references).
A class ν of A is called a spectral class of A associated with eigenvalue ρ 6= 0, or
sometimes (A, ρ)-spectral class for short, if
(6)
ρ⊕
ν = ρ, and µ → ν implies ρ⊕
µ ≤ ρ⊕
ν (max algebra),
ν = ρ, and µ → ν, µ 6= ν implies ρ+
ρ+
µ < ρ+
ν (nonnegative algebra).
In both algebras, note that there may be several spectral classes associated with the same
eigenvalue.
In nonnegative algebra, spectral classes are called distinguished classes [35], and there
are also semi-distinguished classes associated with distinguished generalized eigenvectors
of order two or more [44]. However, these vectors are not contained in the core4. Also,
no suitable max-algebraic analogue of generalized eigenvectors is known to us.
If all classes of A consist of just one element, then the nonnegative and max-algebraic
Perron roots are the same.
In this case, the spectral classes in nonnegative algebra
are also spectral in max algebra. However,
in general this is not so.
In particular,
for a nonnegative matrix A, a cycle of G(A) attaining the maximum cycle geometric
mean ρ⊕(A) = λ(A) need not lie in a strongly connected component corresponding to a
class with spectral radius ρ+(A). This is because, if A1, A2 are irreducible nonnegative
4For a polyhedral cone, the core of the cone-preserving map does not contain generalized eigenvectors
of order two or more [43, Corollary 4.3].
14
PETER BUTKOVI C, HANS SCHNEIDER, SERGEI SERGEEV, AND BIT-SHUN TAM
matrices such that ρ+(A1) < ρ+(A2), then we need not have ρ⊕(A1) < ρ⊕(A2). For
example, let A be the 3 × 3 matrix of all 1's, and let B(ǫ) = (3/2, ǫ, ǫ)T (3/2, ǫ, ǫ). Then
ρ+(A) = 3, ρ+(B(ǫ)) = 9/4 + 2ǫ2, so ρ+(A) > ρ+(B(ǫ)) for sufficiently small ǫ > 0, but
ρ⊕(B(ǫ)) = 9/4 > 1 = ρ⊕(A).
Denote by Λ+(A), resp. Λ⊕(A), the set of nonzero eigenvalues of A ∈ Rn×n
+
in nonneg-
ative linear algebra, resp. in max algebra. It will be denoted by Λ(A) when both algebras
are considered simultaneously, as in the following standard description.
Theorem 2.10 ([7, Th. 4.5.4], [35, Th. 3.7]). Let A ∈ Rn×n
+ . Then Λ(A) = {ρν 6=
0 : ν is spectral}.
Theorem 2.10 encodes the following two statements:
(7)
Λ⊕(A) = {ρ⊕
ν 6= 0 : ν is spectral}, Λ+(A) = {ρ+
ν 6= 0 : ν is spectral},
where the notion of spectral class is defined in two different ways by (6), in two algebras.
In both algebras, for each ρ ∈ Λ(A) define
(8)
Aρ := ρ−1
0
0 AMρMρ
0
, where
Mρ := {i : i → ν, ν is (A, ρ)-spectral} .
By "ν is (A, ρ)-spectral" we mean that ν is a spectral class of A with ρν = ρ. The next
proposition, holding both in max algebra and in nonnegative algebra, allows us to reduce
the case of arbitrary eigencone to the case of principal eigencone. Here we assume that
A is in Frobenius normal form.
Proposition 2.11 ([7, 19, 35]). For A ∈ Rn×n
+
and each ρ ∈ Λ(A), we have V (A, ρ) =
V (Aρ, 1), where 1 is the principal eigenvalue of Aρ.
For a parallel description of extremals of eigencones5 in both algebras see Section 6.1.
5In nonnegative algebra, [35, Th. 3.7] immediately describes both spectral classes and eigencones
associated with any eigenvalue. However, we prefer to split the formulation, following the exposition
of [7]. An alternative simultaneous exposition of spectral theory in both algebras can be found in [24].
TWO CORES OF A NONNEGATIVE MATRIX
15
In max algebra, using Proposition 2.11, we define the critical graph associated with
ρ ∈ Λ⊕(A) as the critical graph of Aρ. By a critical component of A we mean a strongly
connected component of the critical graph associated with some ρ ∈ Λ⊕(A).
In max
algebra, the role of spectral classes of A is rather played by these critical components,
which will be (in analogy with classes of Frobenius normal form) denoted by µ, with the
node set Nµ. See Section 5.2.
3. Notation table and key results
The following notation table shows how various objects are denoted in nonnegative
algebra, max algebra and when both algebras are considered simultaneously.
Nonnegative
Sum
Matrix power
P
A×t
Max
L
A⊗t
Both
P
At
Column span
span+(A)
span⊕(A)
span(A)
Perron root
ρ+
µ
ρ⊕
ν
ρµ
Spectrum (excl. 0)
Λ+(A)
Λ⊕(A)
Λ(A)
Eigencone
V+(A, ρ+)
V⊕(A, ρ⊕) V (A, ρ)
Sum of eigencones
Core
V Σ
+ (A)
core+(A)
V Σ
⊕ (A)
V Σ(A)
core⊕(A)
core(A)
In the case of max algebra, we also have the critical graph C(A) (with related concepts
and notation), not used in nonnegative algebra.
The core and the sum of eigencones appearing in the table have not been formally
introduced. These are the two central notions of this paper, and we now introduce them
for both algebras simultaneously.
The core of a nonnegative matrix A is defined as the intersection of the column spans
(in other words, images) of its powers:
(9)
core(A) := ∩∞
i=1 span(Ai).
16
PETER BUTKOVI C, HANS SCHNEIDER, SERGEI SERGEEV, AND BIT-SHUN TAM
The (Minkowski) sum of eigencones of a nonnegative matrix A is the cone consisting
of all sums of vectors in all V (A, ρ):
(10)
V Σ(A) := X
ρ∈Λ(A)
V (A, ρ)
If Λ(A) = ∅, which happens when ρ(A) = 0, then we assume that the sum on the right-
hand side is {0}.
The following notations can be seen as the "global" definition of cyclicity in nonnegative
algebra and in max algebra.
1. Let σρ be the the lcm of all cyclicities of spectral classes associated with ρ ∈
Λ+(A) (nonnegative algebra), or the cyclicity of critical graph associated with
ρ ∈ Λ⊕(A) (max algebra).
2. Let σΛ be the lcm of all σρ where ρ ∈ Λ(A).
The difference between the definitions of σρ in max algebra and in nonnegative algebra
comes from the corresponding versions of Perron-Frobenius theory.
In particular, let
A ∈ Rn×n
+
be an irreducible matrix. While in nonnegative algebra the eigencone associated
with the Perron root of A is always reduced to a single ray, the number of (appropriately
normalized) extremals of the eigencone of A in max algebra is equal to the number of
critical components, so that there may be up to n such extremals.
One of the key results of the present paper relates the core with the sum of eigencones.
The nonnegative part of this result can be found in Tam-Schneider [43, Th. 4.2, part (iii)].
Main Theorem 1. Let A ∈ Rn×n
+ . Then
core(A) = X
k≥1,ρ∈Λ(A)
V (Ak, ρk) = V Σ(AσΛ).
The main part of the proof is given in Section 4, for both algebras simultaneously.
However, this proof takes in "credit" some facts, which we will have to show. First of all,
we need the equality
(11)
Λ(Ak) = {ρk : ρ ∈ Λ(A)}.
This simple relation between Λ(Ak) and Λ(A), which can be seen as a special case of [24,
Theorem 3.6(ii)], will be also proved below as Corollary 5.5.
TWO CORES OF A NONNEGATIVE MATRIX
17
To complete the proof of Main Theorem 1 we also have to study the periodic sequence
of eigencones of matrix powers and their sums. On this way we obtain the following key
result, both in max and nonnegative algebra.
Main Theorem 2. Let A ∈ Rn×n
+ . Then
(i) σρ, for ρ ∈ Λ(A), is the period of the sequence {V (Ak, ρk)}k≥1, and V (Ak, ρk) ⊆
V (Aσρ, ρσρ) for all k ≥ 1;
(ii) σΛ is the period of the sequence {V Σ(Ak)}k≥1, and V Σ(Ak) ⊆ V Σ(AσΛ) for all
k ≥ 1.
Main Theorem 2 is proved in Section 7 as a corollary of Theorems 7.3 and 7.4, where
the inclusion relations between eigencones of matrix powers are studied in detail.
Theorem 6.5, which gives a detailed description of extremals of both cores, can be also
seen as a key result of this paper. However, it is too long to be formulated in advance.
4. Two cores
4.1. Basics. In this section we investigate the core of a nonnegative matrix defined by (9).
In the main argument, we consider the cases of max algebra and nonnegative algebra
simultaneously.
One of the most elementary and useful properties of the intersection in (9) is that,
actually,
(12)
span(A) ⊇ span(A2) ⊇ span(A3) ⊇ . . .
Generalizing an argument of Pullman [33] we will prove that
(13)
core(A) = X
k≥1
V Σ(Ak) = X
k≥1,ρ∈Λ(A)
V (Ak, ρk)
also in max algebra.
Note that the following inclusion is almost immediate.
Lemma 4.1. Pk≥1 V Σ(Ak) ⊆ core(A).
Proof. x ∈ V (Ak, ρ) implies that Akx = ρx and hence x = ρ−tAktx for all t ≥ 1 (using
the invertibility of multiplication). Hence x ∈ Tt≥1 span Akt = Tt≥1 span(At).
(cid:3)
18
PETER BUTKOVI C, HANS SCHNEIDER, SERGEI SERGEEV, AND BIT-SHUN TAM
So it remains to show the opposite inclusion
(14)
core(A) ⊆ X
k≥1
V Σ(Ak).
Let us first treat the trivial case ρ(A) = 0, i.e., Λ(A) = ∅. There are only trivial classes
in the Frobenius normal form, and G(A) is acyclic. This implies Ak = 0 for some k ≥ 1.
In this case core(A) = {0}, the sum on the right-hand side is {0} by convention, so (13)
is the trivial "draw" {0} = {0}.
4.2. Max algebra: cases of ultimate periodicity. In max algebra, unlike the non-
negative algebra, there are wide classes of matrices for which (14) and (13) follow almost
immediately. We list some of them below.
S1 : Irreducible matrices.
S2 : Ultimately periodic matrices. This is when the sequence {A⊗k}k≥1 is ultimately peri-
odic with a growth rate ρ (in other words, when the sequence {(A/ρ)⊗k}k≥1 is ultimately
periodic). As shown by Moln´arov´a-Pribis [30], this happens if and only if the Perron roots
of all nontrivial classes of A equal ρ⊕(A) = ρ.
+ the orbit {A⊗k ⊗ x}k≥1 hits an eigenvector of
S3 : Robust matrices. For any vector x ∈ Rn
A, meaning that A⊗T ⊗ x is an eigenvector of A for some T . This implies that the whole
remaining part {A⊗k ⊗ x}k≥T of the orbit (the "tail" of the orbit) consists of multiples of
that eigenvector A⊗T ⊗ x. The notion of robustness was introduced and studied in [9].
+ the orbit {A⊗k ⊗ x}k≥1 hits an eigen-
S4 : Orbit periodic matrices: For any vector x ∈ Rn
vector of A⊗σx for some σx, implying that the remaining "tail" of the orbit {(A⊗k ⊗ x}k≥1
is periodic with some growth rate. See [40, Section 7] for characterization.
S5 : Column periodic matrices. This is when for all i we have (A⊗(t+σi))·i = ρσi
i A⊗t
·i
for all
large enough t and some ρi and σi.
Observe that S1 ⊆ S2 ⊆ S4 ⊆ S5 and S3 ⊆ S4.
Indeed, S1 ⊆ S2 is the Cyclicity
Theorem 2.4. For the inclusion S2 ⊆ S4 observe that, if A is ultimately periodic then
A⊗(t+σ) = ρσA⊗t and hence A⊗(t+σ) ⊗ x = ρσA⊗t ⊗ x holds for all x ∈ Rn
+ and all big
enough t. Observe that S3 is a special case of S4, which is a special case of S5 since the
columns of matrix powers can be considered as orbits of the unit vectors.
TWO CORES OF A NONNEGATIVE MATRIX
19
To see that (14) holds in all these cases, note that in the column periodic case all
·i}t≥1 end up with periodically repeating eigenvectors of A⊗σi or
column sequences {At
the zero vector, which implies that span⊕(A⊗t) ⊆ Lk≥1 V Σ
⊕ (A⊗k) ⊆ core⊕(A) and hence
span⊕(A⊗t) = core⊕(A) for all large enough t. Thus, finite stabilization of the core occurs
in all these classes. A necessary and sufficient condition for this finite stabilization is
described in [12].
4.3. Core: a general argument. The original argument of Pullman [33, Section 2] used
the separation of a point from a closed convex cone by an open homogeneous halfspace
(that contains the cone and does not contain the point).
In the case of max algebra, Nitica and Singer [32] showed that at each point x ∈ Rn
+ there
are at most n maximal max-cones not containing this point. These conical semispaces,
used to separate x from any max cone not containing x, turn out to be open. Hence they
can be used in the max version of Pullman's argument.
However, for the sake of a simultaneous proof we will exploit the following analytic
argument instead of separation. By B(x, ǫ) we denote the intersection of the open ball
centered at x ∈ Rn
+ of radius ǫ with Rn
+. In the remaining part of Section 4 we consider
both algebras simultaneously.
Lemma 4.2. Let x1, . . . , xm ∈ Rn
+ be nonzero and let z /∈ span(x1, . . . , xm). Then there
exists ǫ > 0 such that z /∈ span(B(x1, ǫ), . . . , B(xm, ǫ)).
Proof. By contradiction assume that for each ǫ there exist points yi(ǫ) ∈ B(xi, ǫ) and
nonnegative scalars µi(ǫ) such that
(15)
z =
m
X
i=1
µi(ǫ)yi(ǫ).
Since yi(ǫ) → xi as ǫ → 0 and xi are nonzero, we can assume that yi(ǫ) are bounded
from below by nonzero vectors vi, and then z ≥ Pm
i=1 µi(ǫ)vi for all ǫ, implying that
µi(ǫ) are uniformly bounded from above. By compactness we can assume that µi(ǫ)
converge to some µi ∈ R+, and then (15) implies by continuity that z = Pm
i=1 µixi, a
(cid:3)
contradiction.
20
PETER BUTKOVI C, HANS SCHNEIDER, SERGEI SERGEEV, AND BIT-SHUN TAM
Theorem 4.3 ([33, Theorem 2.1]). Assume that {Kl} for l ≥ 1, is a sequence of cones in
Rn
+ such that Kl+1 ⊆ Kl for all l, and each of them generated by no more than k nonzero
l=1Kl is also generated by no more than k vectors.
vectors. Then the intersection K = ∩∞
Proof. Let Kl = span(yl1, . . . , ylk) (where some of the vectors yl1, . . . , ylk may be repeated
when Kl is generated by less than k nonzero vectors), and consider the sequences of
normalized vectors {yli/yli}l≥1 for i = 1, . . . , k, where u := max ui (or any other
norm). As the set {u : u = 1} is compact, we can find a subsequence {lt}t≥1 such that
for i = 1, . . . , k, the sequence {ylti/ylti}t≥1 converges to a finite vector ui, which is
nonzero since ui = 1. We will assume that ylti = 1 for all i and t.
We now show that u1, . . . , uk ∈ K. Consider any i = 1, . . . , k. For each s, ylti ∈ Ks
for all sufficiently large t. As {ylti}t≥1 converges to ui and Ks is closed, we have ui ∈ Ks.
Since this is true for each s, we have ui ∈ ∩∞
s=1Ks = K.
Thus u1, . . . , uk ∈ K, and span(u1, . . . , uk) ⊆ K. We claim that also K ⊆ span(u1, . . . , uk).
Assume to the contrary that there is z ∈ K that is not in span(u1, . . . , uk). Then by
Lemma 4.2 there exists ǫ > 0 such that z /∈ span(B(u1, ǫ), . . . , B(uk, ǫ)). Since the se-
quence {ylti}t≥1 converges to ui, we have ylti ∈ B(ui, ǫ) for t large enough, and
span(ylt1, . . . , yltk) ⊆ span(B(u1, ǫ), . . . , B(uk, ǫ))
But z belongs to Klt = span(ylt1, . . . , yltk) since it belongs to the intersection of all these
cones, a contradiction.
(cid:3)
Theorem 4.3 applies to the sequence {span(At)}t≥1, so core(A) is generated by no more
than n vectors.
Proposition 4.4 ([33, Lemma 2.3]). The mapping induced by A on its core is a surjection.
Proof. First note that A does induce a mapping on its core. If z ∈ core(A) then for each t
there exists xt such that z = Atxt. Hence Az = At+1xt, so Az ∈ ∩t≥2 span At = core(A).
Next, let m be such that Am has the greatest number of zero columns (we assume
that A is not nilpotent; recall that a zero column in Ak remains zero in all subsequent
powers).
If z = Atxt for t ≥ m + 1, we also can represent it as Am+1ut, where ut :=
At−m−1xt. The components of ut corresponding to the nonzero columns of Am+1 are
TWO CORES OF A NONNEGATIVE MATRIX
21
bounded since Am+1ut = z. So we can assume that the sequence of subvectors of ut
with these components converges. Then the sequence yt := Amut also converges, since
the indices of nonzero columns of Am coincide with those of Am+1, which are the indices
of the converging subvectors of ut. Let y be the limit of yt. Since ys = As−1xs are in
span(At) for all s > t, and since span(At) are closed, we obtain y ∈ span(At) for all t.
Thus we found y ∈ core(A) satisfying Ay = z.
(cid:3)
Theorem 4.3 and Proposition 4.4 show that the core is generated by finitely many
vectors in Rn
+ and that the mapping induced by A on its core is "onto".
Now we use the fact that a finitely generated cone in the nonnegative orthant (and
more generally, closed cone) is generated by its extremals both in nonnegative algebra
and in max algebra, see [10, 45].
Proposition 4.5 ([33, Theorem 2.2]). The mapping induced by A on the extremal gen-
erators of its core is a permutation (i.e., a bijection).
Proof. Let core(A) = span(u1, . . . , uk) where u1, . . . , uk are extremals of the core. Suppose
that xj is a preimage of uj in the core, that is, Axj = uj for some xj ∈ core(A), j =
1, . . . , k. Then xj = Pk
i=1 αiui for some nonnegative coefficients α1, . . . , αk, and uj =
Pk
i=1 αiAui. Since uj is extremal, it follows that uj is proportional to Aui for some i.
Thus for each j ∈ {1, . . . , k} there exists an i ∈ {1, . . . , k} such that Aui is a positive
multiple of uj. But since for each i ∈ {1, . . . , k} there is at most one j such that Aui
is a positive multiple of uj, it follows that A induces a bijection on the set of extremal
generators of its core.
(cid:3)
We are now ready to prove (13) and Main Theorem 1 taking the periodicity of the
eigencone sequence (Main Theorem 2) in "credit".
Proof of Main Theorem 1. Proposition 4.5 implies that all extremals of core(A) are eigen-
vectors of Aq, where q denotes the order of the permutation induced by A on the extremals
of core(A). Hence core(A) is a subcone of the sum of all eigencones of all powers of
A, which is the inclusion relation (14). Combining this with the reverse inclusion of
Lemma 4.1 we obtain that core(A) is precisely the sum of all eigencones of all powers of
22
PETER BUTKOVI C, HANS SCHNEIDER, SERGEI SERGEEV, AND BIT-SHUN TAM
A, and using (11) (proved in Section 5 below), we obtain the first part of the equality
of Main Theorem 1. The last part of the equality of Main Theorem 1 now follows from
the periodicity of eigencones formulated in Main Theorem 2, or more precisely, from the
weaker result of Theorem 7.1 proved in Section 7.
(cid:3)
5. Spectral classes and critical components of matrix powers
This section is rather of technical importance. It shows that the union of node sets of
all spectral classes is invariant under matrix powering, and that access relations between
spectral classes in all matrix powers are essentially the same. Further, the case of an
arbitrary eigenvalue can be reduced to the case of the principal eigenvalue for all powers
simultaneously (in both algebras). At the end of the section we consider the critical
components of max-algebraic powers.
5.1. Classes and access relations. As in Section 4, the arguments are presented in
both algebras simultaneously. This is due to the fact that the edge sets of G(A⊗k) and
G(A×k) are the same for any k and that the definitions of spectral classes in both algebras
are alike. Results of this section can be traced back, for the case of nonnegative algebra,
to the classical work of Frobenius [18], see remarks on the very first page of [18] concerning
the powers of an irreducible nonnegative matrix6.
The reader is also referred to the monographs of Minc [29], Berman-Plemmons [3], Brua-
ldi-Ryser [5], and we will often cite the work of Tam-Schneider [43, Section 4] containing
all of our results in this section, in nonnegative algebra.
6Frobenius defines (what we could call) the cyclicity or index of imprimitivity k of an irreducible S
as the number of eigenvalues that lie on the spectral circle. He then remarks "If A is primitive, then
every power of A is again primitive and a certain power and all subsequent powers are positive". This is
followed by "If A is imprimitive, then Am consists of d irreducible parts where d is the greatest common
divisor of m and k. Further, Am is completely reducible. The characteristic functions of the components
differ only in the powers of the variable" (which provides a converse to the preceding assertion). And
then "The matrix Ak is the lowest power of A whose components are all primitive". The three quotations
cover Lemma 5.1 in the case of nonnegative algebra.
TWO CORES OF A NONNEGATIVE MATRIX
23
Lemma 5.1 (cf.
[3, Ch. 5, Ex. 6.9], [43, Lemma 4.5] ). Let A be irreducible with the
unique eigenvalue ρ, let G(A) have cyclicity σ and k be a positive integer.
(i) Ak is permutationally similar to a direct sum of gcd(k, σ) irreducible blocks with
eigenvalues ρk, and Ak does not have eigenvalues other than ρk.
(ii) If k is a multiple of σ, then the sets of indices in these blocks coincide with the
cyclic classes of G(A).
(iii) If supp(x) is a cyclic class of G(A), then supp(Ax) is the previous cyclic class.
Proof. (i): Assuming without loss of generality ρ = 1, let X = diag(x) for a positive
eigenvector x ∈ V (A, ρ) and consider B := X −1AX which is stochastic (nonnegative
algebra), or max-stochastic, i.e., such that Ln
j=1 bij = 1 holds for all i (max algebra).
By Theorem 2.1, Bk is permutationally similar to a direct sum of gcd(k, σ) irreducible
isolated blocks. These blocks are stochastic (or max-stochastic), hence they all have an
eigenvector (1, . . . , 1) associated with the unique eigenvalue 1. If x ∈ V (Ak, ρ) for some
ρ, then its subvectors corresponding to the irreducible blocks of Ak are also eigenvectors
of those blocks, or zero vectors. Hence ρ = 1, which is the only eigenvalue of Ak.
(ii): By Theorem 2.1, G(A) splits into gcd(k, σ) = σ components, and each of them
contains exactly one cyclic class of G(A).
(iii): Use the definition of cyclic classes and that each node has an ingoing edge.
(cid:3)
Lemma 5.2. Both in max algebra and in nonnegative linear algebra, the trivial classes
of Ak are the same for all k.
Proof. In both algebras, an index belongs to a class with nonzero Perron root if and only
if the associated graph contains a cycle with a nonzero weight traversing the node with
that index. This property is invariant under taking matrix powers, hence the claim. (cid:3)
In both algebras, each class µ of A with cyclicity σ corresponds to an irreducible
submatrix Aµµ. It is easy to see that (Ak)µµ = (Aµµ)k. Applying Lemma 5.1 to Aµµ we
see that µ gives rise to gcd(k, σ) classes in Ak, which are said to be derived from their
common ancestor µ. If µ is trivial, then it gives rise to a unique trivial derived class of
Ak, and if µ is non-trivial then all the derived classes are nontrivial as well. The classes
24
PETER BUTKOVI C, HANS SCHNEIDER, SERGEI SERGEEV, AND BIT-SHUN TAM
of Ak and Al derived from the common ancestor will be called related. Note that this is
an equivalence relation on the set of classes of all powers of A. Evidently, a class of Ak is
derived from a class of A if and only if its index set is contained in the index set of the
latter class. It is also clear that each class of Ak has an ancestor in A.
We now observe that access relations in matrix powers are "essentially the same". This
has identical proof in max algebra and nonnegative algebra.
Lemma 5.3. Let A ∈ Rn×n
+ . For all k, l ≥ 1 and ρ > 0, if an index i ∈ {1, . . . , n}
accesses (resp. is accessed by) a class with Perron root ρk in Ak then i accesses (resp. is
accessed by) a related class with Perron root ρl in Al.
Proof. We deduce from Lemma 5.1 and Lemma 5.2 that the index set of each class of Ak
with Perron root ρk is contained in the ancestor class of A with Perron root ρ. Then, i
accessing (resp. being accessed by) a class in Ak implies i accessing (resp. being accessed
by) its ancestor in A. Since ρ > 0, this ancestor class is nontrivial, so the access path can
be extended to have a length divisible by l, by means of a path contained in the ancestor
class. By Lemma 5.1, the ancestor decomposes in Al into several classes with the common
Perron root ρl, and i accesses (resp. is accessed by) one of them.
(cid:3)
Theorem 5.4 ([43, Corollary 4.6]). Let A ∈ Rn×n
+ .
(i) If a class µ is spectral in A, then so are the classes derived from it in Ak. Con-
versely, each spectral class of Ak is derived from a spectral class of A.
(ii) For each class µ of A with cyclicity σ, there are gcd(k, σ) classes of Ak derived
from it. If k is a multiple of σ then the index sets of derived classes are the cyclic
classes of µ.
Proof. (i): We will prove the following equivalent statement: For each pair µ, ν where µ
is a class in A and ν is a class derived from µ in Ak, we have that µ is non-spectral if and
only if ν is non-spectral.
Observe that by Lemma 5.2, the Perron root of µ is 0 if and only if the Perron root of
ν is 0. In this case, both µ and ν are non-spectral (by definition). Further, let ρ > 0 be
the Perron root of µ. Then, by Lemma 5.1, the Perron root of ν is ρk. Let i be an index
in ν. It also belongs to µ.
TWO CORES OF A NONNEGATIVE MATRIX
25
If µ is non-spectral, then i is accessed in A by a class with Perron root ρ′ such that
ρ′ > ρ in max algebra, resp. ρ′ ≥ ρ in nonnegative algebra. By Lemma 5.3, there is a
class of Ak, which accesses i in Ak and has Perron root (ρ′)k. Since we have (ρ′)k > ρk in
max algebra or resp. (ρ′)k ≥ ρk in nonnegative algebra, we obtain that ν, being the class
to which i belongs in Ak, is also non-spectral.
Conversely, if ν is non-spectral, then i is accessed in Ak by a class θ with Perron root
equal to ρk for some ρ, and such that ρk > ρk in max algebra, resp. ρk ≥ ρk in nonnegative
algebra. The ancestor of θ in A accesses7 i in A and has Perron root ρ. Since we have
ρ > ρ in max algebra or resp. ρ ≥ ρ in nonnegative algebra, we obtain that µ, being the
class to which i belongs in A, is also non-spectral. Part (i) is proved.
(ii): This part follows directly from Lemma 5.1 parts (i) and (ii).
(cid:3)
Corollary 5.5. Let A ∈ Rn×n
+
and k ≥ 1. Then Λ(Ak) = {ρk : ρ ∈ Λ(A)}.
Proof. By Theorem 2.10, the nonzero eigenvalues of A (resp. Ak) are precisely the Perron
roots of the spectral classes of A (resp. Ak). By Theorem 5.4(i), if a class of A is spectral,
then so is any class derived from it in Ak. This impliesthat Λ(Ak) ⊆ {ρk : ρ ∈ Λ(A)}. The
converse inclusion follows from the converse part of Theorem 5.4(i).
(cid:3)
Let us note yet another corollary of Theorem 5.4. For A ∈ Rn×n
+
and ρ ≥ 0, let
N(A, ρ) be the union of index sets of all classes of A with Perron root ρ, and N s(A, ρ)
be the union of index sets of all spectral classes of A with Perron root ρ. Obviously,
N s(A, ρ) ⊆ N(A, ρ), and both sets (as defined for arbitrary ρ ≥ 0) are possibly empty.
Corollary 5.6. Let A ∈ Rn×n
+ , ρ ∈ R+ and k ≥ 1. Then
(i) N(Ak, ρk) = N(A, ρ),
(ii) N s(Ak, ρk) = N s(A, ρ).
Proof. (i): This part follows from Lemmas 5.1 and 5.2.
(ii):
Inclusion N s(A, ρ) ⊆
N s(Ak, ρk) follows from the direct part of Theorem 5.4(i), and inclusion
N s(Ak, ρk) ⊆ N s(A, ρ) follows from the converse part of Theorem 5.4(i).
(cid:3)
7This can be observed immediately, or obtained by applying Lemma 5.3.
26
PETER BUTKOVI C, HANS SCHNEIDER, SERGEI SERGEEV, AND BIT-SHUN TAM
For the eigencones of A ∈ Rn×n
+ , the case of an arbitrary ρ ∈ Λ(A) can be reduced to the
case of the principal eigenvalue: V (A, ρ) = V (Aρ, 1) (Proposition 2.11). Now we extend
this reduction to the case of V (Ak, ρk), for any k ≥ 1. As in the case of Propositon 2.11,
we assume that A is in Frobenius normal form.
Theorem 5.7. Let k ≥ 1 and ρ ∈ Λ(A).
(i) The set of all indices having access to the spectral classes of Ak with the eigenvalue
ρk equals Mρ, for each k.
(ii) (Ak)MρMρ = ρk(Aρ)k
MρMρ.
(iii) V (Ak, ρk) = V ((Aρ)k, 1).
Proof. (i): Apply Corollary 5.6 part (ii) and Lemma 5.3. (ii): Use that Mρ is initial in
G(A). (iii): By Proposition 2.11 we have (assuming that Ak is in Frobenius normal form)
that V (Ak, ρk) = V ((Aρk )k, 1) where, instead of (8),
(16)
Ak
ρk := ρ−k
0
0 Ak
0
M k
ρ M k
ρ
, and
M k
ρ := {i : i → ν, ν is (Ak, ρk)-spectral}
By part (i) M k
ρ = Mρ, hence Ak
ρk = (Aρ)k and the claim follows.
(cid:3)
5.2. Critical components. In max algebra, when A is assumed to be strictly visualized,
each component µ of C(A) with cyclicity σ corresponds to an irreducible submatrix A[1]
µµ
(as in the case of classes, Aµµ is a shorthand for AN µN µ). Using the strict visualization
and Lemma 2.8 we see that (A⊗k)[1]
µµ we see
that µ gives rise to gcd(k, σ) critical components in A⊗k. As in the case of classes, these
µµ = (A[1]
µµ)⊗k. Applying Lemma 5.1(i) to A[1]
components are said to be derived from their common ancestor µ.
Evidently a component of C(A⊗k) is derived from a component of C(A) if and only if
its index set is contained in the index set of the latter component. Following this line we
now formulate an analogue of Theorem 5.4 (and some other results).
Theorem 5.8 (cf. [7, Theorem 8.2.6], [8, Theorem 2.3]). Let A ∈ Rn×n
+ .
TWO CORES OF A NONNEGATIVE MATRIX
27
(i) For each component µ of C(A) with cyclicity σ, there are gcd(k, σ) components of
C(A⊗k) derived from it. Conversely, each component of C(A⊗k) is derived from
a component of C(A).
If k is a multiple of σ, then index sets in the derived
components are the cyclic classes of µ.
(ii) The sets of critical indices of A⊗k for k = 1, 2, . . . are identical.
(iii) If A is strictly visualized, xi ≤ 1 for all i and supp(x[1]) is a cyclic class of µ, then
supp((A ⊗ x)[1]) is the previous cyclic class of µ.
Proof. (i),(ii): Both statements are based on the fact that C(A⊗k) = (C(A))k, shown in
Lemma 2.8. To obtain (i), also apply Theorem 2.1 to a component µ of C(A). (iii): Use
(A ⊗ x)[1] = A[1] ⊗ x[1], the definition of cyclic classes and the fact that each node in µ has
an ingoing edge.
(cid:3)
6. Describing extremals
The aim of this section is to describe the extremals of the core, in both algebras. To this
end, we first give a parallel description of extremals of eigencones (the Frobenius-Victory
theorems).
6.1. Extremals of the eigencones. We now describe the principal eigencones in non-
negative linear algebra and then in max algebra. By means of Proposition 2.11, this
description can be obviously extended to the general case. As in Section 2.6, both de-
scriptions are essentially known: see [7, 18, 19, 35].
We emphasize that the vectors x(µ) and x(µ) appearing below are full-size.
Theorem 6.1 (Frobenius-Victory[35, Th. 3.7]). Let A ∈ Rn×n
+
have ρ+(A) = 1.
(i) Each spectral class µ with ρ+
µ = 1 corresponds to an eigenvector x(µ), whose support
consists of all indices in the classes that have access to µ, and all vectors x of
V+(A, 1) with supp x = supp x(µ) are multiples of x(µ).
(ii) V+(A, 1) is generated by x(µ) of (i), for µ ranging over all spectral classes with
ρ+
µ = 1.
(iii) x(µ) of (i) are extremals of V+(A, 1). (Moreover, x(µ) are linearly independent.)
28
PETER BUTKOVI C, HANS SCHNEIDER, SERGEI SERGEEV, AND BIT-SHUN TAM
Note that the extremality and the usual linear independence of x(µ) (involving linear
combinations with possibly negative coefficients) can be deduced from the description
of supports in part (i), and from the fact that in nonnegative algebra, spectral classes
associated with the same ρ do not access each other. This linear independence also means
that V+(A, 1) is a simplicial cone. See also [35, Th. 4.1].
Theorem 6.2 ([7, Th. 4.3.5], [41, Th. 2.8]). Let A ∈ Rn×n
+
have ρ⊕(A) = 1.
(i) Each component µ of C(A) corresponds to an eigenvector x(µ) defined as one of the
columns A∗
·i with i ∈ Nµ, all columns with i ∈ Nµ being multiples of each other.
(i') Each component µ of C(A) is contained in a (spectral) class µ with ρ⊕
µ = 1, and
the support of each x(µ) of (i) consists of all indices in the classes that have access
to µ.
(ii) V⊕(A, 1) is generated by x(µ) of (i), for µ ranging over all components of C(A).
(iii) x(µ) of (i) are extremals in V⊕(A, 1). (Moreover, x(µ) are strongly linearly inde-
pendent in the sense of [6].)
To verify (i'), not explicitly stated in the references, use (i) and the path interpretation
of A∗.
Vectors x(µ) of Theorem 6.2 are also called the fundamental eigenvectors of A, in max
algebra. Applying a strict visualization scaling (Theorem 2.7) allows us to get further
details on these fundamental eigenvectors.
Proposition 6.3 ([41, Prop. 4.1]). Let A ∈ Rn×n
+
be strictly visualized (in particular,
ρ⊕(A) = 1). Then
(i) For each component µ of C(A), x(µ) of Theorem 6.2 can be canonically chosen as
A∗
·i for any i ∈ Nµ, all columns with i ∈ Nµ being equal to each other.
(ii) x(µ)
i ≤ 1 for all i. Moreover, supp(x(µ)[1]) = Nµ.
6.2. Extremals of the core. Let us start with the following observation in both algebras.
Proposition 6.4. For each k ≥ 1, the set of extremals of V Σ(Ak) is the union of the sets
of extremals of V (Ak, ρk) for ρ ∈ Λ(A).
TWO CORES OF A NONNEGATIVE MATRIX
29
Proof. Due to the fact that Λ(Ak) = {ρk : ρ ∈ Λ(A)}, we can assume without loss of
generality that k = 1.
1. As V Σ(A) is the sum of V (A, ρ) for ρ ∈ Λ(A), it is generated by the extremals of
V (A, ρ) for ρ ∈ Λ(A). Hence each extremal of V Σ(A) is an extremal of V (A, ρ) for some
ρ ∈ Λ(A).
2. Let x ∈ V (A, ρµ), for some spectral class µ, be extremal. Assume without loss
of generality that ρµ = 1, and by contradiction that there exist vectors yκ, all of them
extremals of V Σ(A), such that x = Pκ yκ. By above, all vectors yκ are eigenvectors
of A. If there is yκ associated with an eigenvalue ρν > 1, then applying At we obtain
x = (ρν)tyκ + . . ., which is impossible at large enough t. So ρν ≤ 1. With this in
mind, if there is yκ associated with ρν < 1, then 1) in nonnegative algebra we obtain
Ax > APκ yκ, a contradiction; 2) in max algebra, all nonzero entries of A ⊗ yκ go below
the corresponding entries of x meaning that yκ is redundant. Thus we are left only with
yκ associated with ρ⊕
ν = 1, which is a contradiction: an extremal x ∈ V⊕(A, 1) appears as
a "sum" of other extremals of V⊕(A, 1) not proportional to x.
(cid:3)
A vector x ∈ Rn
+ is called normalized if max xi = 1. Recall the notation σρ introduced
in Section 3.
Theorem 6.5 (cf. [43, Theorem 4.7]). Let A ∈ Rn×n
+ .
(i) The set of extremals of core(A) is the union of the sets of extremals of V (Aσρ, ρσρ)
for all ρ ∈ Λ(A).
(ii) In nonnegative algebra, each spectral class µ with cyclicity σµ corresponds to a
set of distinct σµ normalized extremals of core+(A), such that there exists an index
in their support that belongs to µ, and each index in their support has access to µ.
In max algebra, each critical component µ with cyclicity σµ associated with some
ρ ∈ Λ⊕(A) corresponds to a set of distinct σµ normalized extremals x of core⊕(A),
which are (normalized) columns of (Aσρ
ρ )∗ with indices in Nµ.
(iii) Each set of extremals described in (ii) forms a simple cycle under the action of A.
(iv) There are no normalized extremals other than those described in (ii). In non-
negative algebra, the total number of normalized extremals equals the sum of
30
PETER BUTKOVI C, HANS SCHNEIDER, SERGEI SERGEEV, AND BIT-SHUN TAM
cyclicities of all spectral classes of A.
In max algebra, the total number of
normalized extremals equals the sum of cyclicities of all critical components of A.
Proof. (i) follows from Main Theorem 1 and Proposition 6.4.
For the proof of (ii) and (iii) we can fix ρ = ρµ ∈ Λ(A), assume A = Aρ (using
Theorem 5.7) and σ := σρ. In max algebra, we also assume that A is strictly visualized.
(ii) In nonnegative algebra, observe that by Theorem 5.4, each spectral class µ
of A gives rise to σµ spectral classes in A×σ, whose node sets are cyclic classes of µ
(note that σµ divides σ). According to Frobenius-Victory Theorem 6.1, these classes give
rise to normalized extremals of V+(A×σ, 1), and the conditions on support follow from
Theorem 6.1 and Lemma 5.3.
(iii): Let x be an extremal described above. Then supp(x) ∩ Nµ is a cyclic class of µ
and supp(Ax) ∩ Nµ is the previous cyclic class of µ, by Lemma 5.1 part (iii). It can be
checked that all indices in supp(Ax) also have access to µ. By Proposition 4.5, Ax is an
extremal of core+(A), and hence an extremal of V+(A×σ, 1). Theorem 6.1 identifies Ax
with the extremal associated with the previous cyclic class of µ.
Vectors x, Ax, . . . , A×σµ−1x are distinct since the intersections of their supports with
Nµ are disjoint, so they are exactly the set of extremals associated with µ. Note that
A×σµx = x, as supp(A×σµx) ∩ Nµ = supp(x) ∩ Nµ, and both vectors are extremals of
V+(A×σ, 1).
(ii): In max algebra, observe that by Theorem 5.8(i) each component µ of C(A) gives
rise to σµ components of C(A⊗σ), whose node sets are the cyclic classes of µ (note that
σµ divides σ). These components correspond to σµ columns of (A⊗σ)∗ with indices in
different cyclic classes of µ, which are by Theorem 5.8(i) the node sets of components
of C(A⊗σ). By Theorem 6.2 these columns of (A⊗σ)∗ are extremals of V⊕(A⊗σ, 1), and
Proposition 6.3(ii) implies that they are normalized.
(iii): Let x be an extremal described above. By Proposition 6.3 and Theorem 5.8(i)
supp(x[1]) is a cyclic class of µ, and by Theorem 5.8(iii) supp((A ⊗ x)[1]) is the previous
cyclic class of µ. By Proposition 4.5, A ⊗ x is an extremal of core⊕(A), and hence an
extremal of V⊕(A⊗σ, 1). Proposition 6.3 identifies A⊗x with the extremal associated with
the previous cyclic class of µ.
TWO CORES OF A NONNEGATIVE MATRIX
31
Vectors x, A⊗x, . . . , A⊗σ µ−1x are distinct since their booleanizations x[1], (A⊗x)[1], . . . , (A⊗σ µ−1⊗
x)[1] are distinct, so they are exactly the set of extremals associated with µ. Note that
A⊗σ µ ⊗ x = x, as (A⊗σ µ ⊗ x)[1] = x[1] and both vectors are extremals of V⊕(A⊗σ, 1).
(iv): In both algebras, the converse part of Theorem 5.4 (i) shows that there are
no spectral classes of Aσ other than the ones derived from the spectral classes of A. In
nonnegative algebra, this shows that there are no extremals other than described in
(ii). In max algebra, on top of that, the converse part of Theorem 5.8 (i) shows that
there are no components of C(A⊗σ
ρ ) other than the ones derived from the components
C(Aρ), for ρ ∈ Λ⊕(A), hence there are no extremals other than described in (ii). In both
algebras, it remains to count the extremals described in (ii).
(cid:3)
7. Sequence of eigencones
The main aim of this section is to investigate the periodicity of eigencones and to prove
Main Theorem 2. Unlike in Section 4, the proof of periodicity will be different for the
cases of max algebra and nonnegative algebra. The periods of eigencone sequences in
max algebra and in nonnegative linear algebra are also in general different, for the same
nonnegative matrix (see Section 8 for an example). To this end, recall the definitions of
σρ and σΛ given in Section 3, which will be used below.
7.1. Periodicity of the sequence. We first observe that in both algebras
(17)
k divides l ⇒ V (Ak, ρk) ⊆ V (Al, ρl) ∀ρ ∈ Λ(A),
k divides l ⇒ V Σ(Ak) ⊆ V Σ(Al).
We now prove that the sequence of eigencones is periodic.
Theorem 7.1. Let A ∈ Rn×n
+
and ρ ∈ Λ(A).
(i) V (Ak, ρk) = V (Ak+σρ, ρk+σρ), and V (Ak, ρk) ⊆ V (Aσρ, ρσρ) for all k ≥ 1.
(ii) V Σ(Ak) = V Σ(Ak+σΛ) and V Σ(Ak) ⊆ V Σ(AσΛ) for all k ≥ 1.
Proof. We will give two separate proofs of part (i), for the case of max algebra and the
case of nonnegative algebra. Part (ii) follows from part (i).
32
PETER BUTKOVI C, HANS SCHNEIDER, SERGEI SERGEEV, AND BIT-SHUN TAM
In both algebras, we can assume without loss of generality that ρ = 1, and using
Theorem 5.7, that this is the greatest eigenvalue of A.
In max algebra, by Theorem 2.5, columns of A⊗r with indices in Nc(A) are periodic
for r ≥ Tc(A). Recall that by Corollary 2.9, Tc(A) is not less than T (C(A)), which
is the greatest periodicity threshold of the strongly connected components of C(A). By
Theorem 2.3 part (ii), (C(A))tσ consists of complete graphs for tσ ≥ T (C(A)), in particular,
it contains loops (i, i) for all i ∈ Nc(A). Hence
a⊗(tσ)
ii
= 1 ∀i ∈ Nc(A), t ≥ ⌈
Tc(A)
σ
⌉,
and
a⊗(l+tσ)
ki
≥ a⊗l
ki a⊗(tσ)
ii
= a⊗l
ki
∀i ∈ Nc(A), ∀k, l, ∀t ≥ ⌈
Tc(A)
σ
⌉,
or, in terms of columns of matrix powers,
A⊗(l+tσ)
·i
≥ A⊗l
·i
∀i ∈ Nc(A), ∀l, ∀t ≥ ⌈
Tc(A)
σ
⌉.
Multiplying this inequality repeatedly by A⊗l we obtain A⊗(kl+tσ)
and A⊗(k(l+tσ))
for all k ≥ 1. Hence we obtain
≥ A⊗(kl)
·i
·i
·i
≥ A⊗(kl)
·i
for all k ≥ 1,
(18)
(A⊗(l+tσ))∗
·i ≥ (A⊗l)∗
·i ∀i ∈ Nc(A), ∀l, ∀t ≥ ⌈
Tc(A)
σ
⌉.
On the other hand, using the ultimate periodicity of critical columns we have
(A⊗(l+tσ))∗
·i = M{A⊗s
·i
: s ≡ kl(modσ), k ≥ 1, s ≥ Tc(A)}
for all l and all tσ ≥ Tc(A), while generally
(A⊗l)∗
·i ≥ M{A⊗s
·i
: s ≡ kl(modσ), k ≥ 1, s ≥ Tc(A)} ∀l,
implying the reverse inequality with respect to (18). It follows that
(19)
(A⊗(l+tσ))∗
·i = (A⊗l)∗
·i ∀i ∈ Nc(A), ∀l, ∀t ≥ ⌈
Tc(A)
σ
⌉,
therefore (A⊗(l+σ))∗
·i for all critical indices i and all
l. Since V (A⊗l, 1) is generated by the critical columns of (A⊗l)∗, and the critical indices of
·i = (A⊗(l+tσ+σ))∗
·i = (A⊗(l+tσ))∗
·i = (A⊗l)∗
A⊗l are Nc(A) by Theorem 5.8(ii), the periodicity V⊕(A⊗l, ρl) = V⊕(A⊗(l+σ), ρl+σ) follows.
TWO CORES OF A NONNEGATIVE MATRIX
33
Using this and (17) we obtain V⊕(A⊗l, ρl) ⊆ V⊕(A⊗(lσ), ρlσ) = V⊕(A⊗σ, ρσ) for each l and
ρ ∈ Λ⊕(A).
In nonnegative algebra, also assume that all final classes (and hence only them)
have Perron root ρ = 1. Final classes of A×l are derived from the final classes of A; they
(and no other classes) have Perron root ρl. By Theorem 5.4(i) and Corollary 2.2, for any
t ≥ 0, the spectral classes of A×l and A×(l+tσ) with Perron root 1 have the same sets of
nodes, which we denote by N1, . . . , Nm (assuming that their number is m ≥ 1).
By the Frobenius-Victory Theorem 6.1, the cone V+(A×l, 1) is generated by m extremals
x(1), . . . , x(m) with the support condition of Theorem 6.1(i) from which we infer that
the subvectors x(µ)
µ
Nν ) are zero for all µ 6= ν
from 1 to m, since the different spectral classes by (6) do not access each other, in
Nµ) are positive, while x(µ)
(i.e., x(µ)
(i.e., x(µ)
ν
the nonnegative linear algebra. Analogously the cone V+(A×(l+tσ), 1) is generated by m
eigenvectors y(1), . . . , y(m) such that the subvectors y(µ)
µ are positive, while y(µ)
ν = 0 for all
µ 6= ν from 1 to m.
Assume first that l = σ. As V+(A×σ, 1) ⊆ V+(A×(tσ), 1), each x(µ) is a nonnegative
linear combination of y(1), . . . , y(m), and this implies x(µ) = y(µ) for all µ = 1, . . . , m.
Hence V+(A×(tσ), 1) = V+(A×σ, 1) for all t ≥ 0.
We also obtain V+(A×l, 1) ⊆ V+(A×(σl), 1) = V+(A×σ, 1) for all l. Thus V+(A×l, 1) ⊆
V+(A×(tσ), 1), and therefore V+(A×l, 1) ⊆ V+(A×(l+tσ), 1). Now if V+(A×l, 1), resp. V+(A×(l+tσ), 1)
are generated by x(1), . . . , x(m), resp. y(1), . . . , y(m) described above and each x(µ) is a
nonnegative linear combination of y(1), . . . , y(m), this again implies x(µ) = y(µ) for all
µ = 1, . . . , m, and V+(A×(l+tσ), 1) = V+(A×l, 1) for all t ≥ 0 and all l.
Using this and (17) we obtain V+(A×l, ρl) ⊆ V+(A×(lσ), ρlσ) = V+(A×σ, ρσ) for each l
and ρ ∈ Λ+(A).
(cid:3)
7.2. Inclusion and divisibility. We now show that the inclusion relations between the
eigencones of different powers of a matrix, in both algebras, strictly follow divisibility
of exponents of matrix powers with respect to σρ and σΛ. We start with a corollary of
Theorem 7.1.
Lemma 7.2. Let k, l ≥ 1 and ρ ∈ Λ(A).
34
PETER BUTKOVI C, HANS SCHNEIDER, SERGEI SERGEEV, AND BIT-SHUN TAM
(i) V (Ak, ρk) = V (Agcd(σρ,k), ρgcd(σρ,k)) and V Σ(Ak) = V Σ(Agcd(σΛ,k)).
(ii) gcd(k, σρ) = gcd(l, σρ) implies V (Ak, ρk) = V (Al, ρl), and gcd(k, σΛ) = gcd(l, σΛ)
implies V Σ(Ak) = V Σ(Al).
Proof. (i): Let σ := σρ, and s :=gcd(k, σ).
If s = σ then k is a multiple of σ and
V (Ak, ρk) = V (As, ρs) by Theorems 7.1(i). Otherwise, since s divides k, we have V (As, ρs) ⊆
V (Ak, ρk).
In view of the periodicity (Theorem 7.1(i)), it suffices to find t such that
V (Ak, ρk) ⊆ V (As+tσ, ρs+tσ). For this, observe that s + tσ is a multiple of s = gcd(k, σ).
By Lemma 2.6 (the Frobenius coin problem), for big enough t it can be expressed as
t1k + t2σ where t1, t2 ≥ 0. Moreover t1 6= 0, for otherwise we have s = σ. Then we obtain
V (Ak, ρk) ⊆ V (At1k, ρt1k) = V (At1k+t2σ, ρt1k+t2σ)
= V (As+tσ, ρs+tσ) = V (As, ρs),
and the first part of the claim follows. The second part is obtained similarly, using
Theorem 7.1(ii) instead of Theorems 7.1(i).
(ii) follows from (i).
(cid:3)
Theorem 7.3. Let A ∈ Rn×n
+
and σ be either the cyclicity of a spectral class of A (non-
negative algebra) or the cyclicity of a critical component of A (max algebra).The
following are equivalent for all positive k, l:
(i) gcd(k, σ) divides gcd(l, σ) for all cyclicities σ;
(ii) gcd(k, σρ) divides gcd(l, σρ) for all ρ ∈ Λ(A);
(iii) gcd(k, σΛ) divides gcd(l, σΛ);
(iv) V (Ak, ρk) ⊆ V (Al, ρl) for all ρ ∈ Λ(A) and
(v) V Σ(Ak) ⊆ V Σ(Al).
Proof. (i)⇒(ii)⇒(iii) follow from elementary number theory.
(ii)⇒ (iv) and (iii)⇒(v)
follow from (17) and Lemma 7.2 part (i) (which is essentially based on Theorem 7.1).
(iv)⇒(v) is trivial. It only remains to show that (v)⇒ (i).
(v)⇒ (i): In both algebras, take an extremal x ∈ V (Ak, ρk). As V Σ(Ak) ⊆ V Σ(Al),
this vector can be represented as x = Pi yi, where yi are extremals of V Σ(Al). Each yi
is an extremal of V (Al, ρl) for some ρ ∈ Λ(A) (as we will see, only the extremals with
TWO CORES OF A NONNEGATIVE MATRIX
35
ρ = ρ are important). By Frobenius-Victory Theorems 6.1 and 6.2 and Theorem 5.4(i),
there is a unique spectral class µ of A to which all indices in supp(x) have access. Since
supp(yi) ⊆ supp(x), we are restricted to the submatrix AJJ where J is the set of all
indices accessing µ in A. In other words, we can assume without loss of generality that
µ is the only final class in A, hence ρ is the greatest eigenvalue, and ρ = 1. Note that
supp(x) ∩ Nµ 6= ∅.
In nonnegative algebra, restricting the equality x = Pi yi to Nµ we obtain
(20)
supp(xµ) = [
i
supp(yi
µ).
If supp(yi
µ) is non-empty, then yi is associated with a spectral class of A×l whose nodes
are in Nµ. Theorem 6.1(i) implies that supp(yi
µ) consists of all indices in a class of A×l
µµ.
As x can be any extremal eigenvector of A×k with supp x ∩ Nµ 6= ∅, (20) shows that
each class of A×k
µµ (corresponding to
yi). By Corollary 2.2 this is only possible when gcd(k, σ) divides gcd(l, σ), where σ is the
µµ (corresponding to x) splits into several classes of A×l
cyclicity of the spectral class µ.
In max algebra, since ρ = 1, assume without loss of generality that A is strictly
visualized. In this case A and x have all coordinates not exceeding 1. Recall that x[1]
is the Boolean vector defined by x[1]
i = 1 ⇔ xi = 1. Vector x corresponds to a unique
critical component µ of C(A) with the node set Nµ. Then instead of (20) we obtain
(21)
x[1] = M
i
yi[1] ⇒ supp(x[1]
µ ) = [
i
supp(yi[1]
µ ),
µ ). If supp(yi[1]
µ ) is non-empty then also supp(yi
where supp(x[1]) = supp(x[1]
supp(yi[1]) = supp(yi[1]
µ ) by Proposition 6.3(ii) and Theorem 5.8(i), and hence also
Nµ) is non-empty so
that yi is associated with the eigenvalue 1. As yi is extremal, Proposition 6.3(ii) implies
that supp(yi[1]
µµ)⊗l. As x can be any extremal
eigenvector of A⊗k with supp(x[1]) ∩ Nµ 6= ∅, (21) shows that each class of (A[1]
µµ)⊗k splits
into several classes of (A[1]
µµ)⊗l. By Corollary 2.2 this is only possible when gcd(k, σ) divides
(cid:3)
µ ) consists of all indices in a class of (A[1]
gcd(l, σ), where σ is the cyclicity of the critical component µ.
Let us also formulate the following version restricted to some ρ ∈ Λ(A).
36
PETER BUTKOVI C, HANS SCHNEIDER, SERGEI SERGEEV, AND BIT-SHUN TAM
Theorem 7.4. Let A ∈ Rn×n
+ , and let σ be either the cyclicity of a spectral class (non-
negative algebra) or the cyclicity of a critical component (max algebra) associated
with some ρ ∈ Λ(A). The following are equivalent for all positive k, l:
(i) gcd(k, σ) divides gcd(l, σ) for all cyclicities σ;
(ii) gcd(k, σρ) divides gcd(l, σρ);
(iii) V (Ak, ρk) ⊆ V (Al, ρl).
Proof. (i)⇒(ii) follows from the elementary number theory, and (ii)⇒(iii) follows from (17)
and Lemma 7.2(i). The proof of (iii)⇒(i) follows the lines of the proof of Theorem 7.3
(v)⇒ (i), with a slight simplification that ρ = ρ and further, x and all yi in x = Pi yi are
associated with the same eigenvalue.
(cid:3)
We are now ready to deduce Main Theorem 2
Proof of Main Theorem 2. We prove the first part. The inclusion V (Ak, ρk) ⊆ V (Aσ, ρσ)
was proved in Theorem 7.1 (i), and we are left to show that σρ is the least such p that
V (Ak+p, ρk+p) = V (Ak, ρk) for all k ≥ 1. But taking k = σρ and using Theorem 7.4
(ii)⇔(iii), we obtain gcd(σρ + p, σρ) = σρ, implying that σρ divides σρ + p, so σρ divides
p. Since Theorem 7.1 (i) also shows that V (Ak+σρ, ρk+σρ) = V (Ak, ρk) for all k ≥ 1, the
result follows.
The second part can be proved similarly, using Theorem 7.1(ii) and Theorem 7.3
(iii)⇔(v).
(cid:3)
8. Examples
We consider two examples of reducible nonnegative matrices, examining their core in
max algebra and in nonnegative linear algebra.
Example 1. Take
0.1206
0.5895 0.2904
0
0
1
0
0.8797 0.4253
0
(22)
A =
0.2262 0.6171 0.3439
1
0.3127
.
0.3846 0.2653 0.5841 0.2607
1
0.5830
1
0.1078 0.5944 0.1788
TWO CORES OF A NONNEGATIVE MATRIX
37
A has two classes with node sets {1} and {2, 3, 4, 5}. Both in max algebra and in nonnega-
tive linear algebra, the only spectral class arises from M := {2, 3, 4, 5}. The max-algebraic
Perron root of this class is ρ⊕(A) = 1, and the critical graph consists of just one cycle
2 → 3 → 4 → 5 → 2.
The eigencones V⊕(A, 1), V⊕(A⊗2, 1), V⊕(A⊗3, 1) and V⊕(A⊗4, 1) are generated by the
last four columns of the Kleene stars A∗, (A⊗2)∗, (A⊗3)∗, (A⊗4)∗. Namely,
V⊕(A, 1) = V⊕(A⊗3, 1) = span⊕{(0 1 1 1 1)},
V⊕(A⊗2, 1) = span⊕{(0, 1, 0.8797, 1, 0.8797), (0, 0.8797, 1, 0.8797, 1)},
V⊕(A⊗4, 1) = span⊕{(0, 1, 0.6807, 0.7738, 0.8797), (0, 0.8797, 1, 0.6807, 0.7738),
(0, 0.7738, 0.8797, 1, 0.6807), (0, 0.6807, 0.7738, 0.8797, 1)}
By Main Theorem 1, core⊕(A) is equal to V⊕(A⊗4, 1). Computing the max-algebraic
powers of A we see that the sequence of submatrices A⊗t
M M becomes periodic after t = 10,
with period 4. In particular,
α
0
0
0.4511 0.7738 0.6807
0
1
0
0.8797
(23)
A⊗10 =
0.5128 0.8797 0.7738 0.6807
1
0.5830
1
0.8797 0.7738 0.6807
0.5895 0.6807
1
0.8797 0.7738
,
where 0 < α < 0.0001. Observe that the last four columns are precisely the ones that
generate V⊕(A⊗4, 1). Moreover, if α was 0 then the first column would be the following
max-combination of the last four columns:
a⊗10
41 A⊗10
·2 ⊕ a⊗10
51 A⊗10
·3 ⊕ a⊗10
21 A⊗10
·4 ⊕ a⊗10
31 A⊗10
·5
.
On the one hand, the first column of A⊗t cannot be a max-combination of the last four
columns for any t > 0 since a⊗t
11 > 0. On the other hand, a⊗t
11 → 0 as t → ∞ ensuring that
the first column belongs to the core "in the limit".
Figure 1 gives a symbolic illustration of what is going on in this example.
In nonnegative algebra, the block AM M with M = {2, 3, 4, 5} is also the only spectral
block . Its Perron root is approximately ρ+(A) = 2.2101, and the corresponding eigencone
38
PETER BUTKOVI C, HANS SCHNEIDER, SERGEI SERGEEV, AND BIT-SHUN TAM
Figure 1. The spans of matrix powers (upper curve) and the periodic
sequence of their eigencones (lower graph) in Example 1 (max algebra).
is
V+(A, ρ+(A)) = span+{(0, 0.5750, 0.5107, 0.4593, 0.4445)}.
Taking the usual powers of (A/ρ+(A)) we see that
(cid:0)A/ρ+(A)(cid:1)×12
=
,
α
0.2457 0.2752 0.2711 0.3453 0.2693
0
0
0
0
0.2182 0.2444 0.2408 0.3067 0.2392
0.1963 0.2198 0.2165 0.2759 0.2151
0.1899 0.2127 0.2096 0.2670 0.2082
where 0 < α < 0.0001, and that the first four digits of all entries in all higher powers
are the same. It can be verified that the submatrix (A/ρ+(A))×12
M M is, approximately, the
outer product of the Perron eigenvector with itself, while the first column is also almost
proportional to it.
TWO CORES OF A NONNEGATIVE MATRIX
39
Example 2. Take
(24)
A =
0
1
1
0
0
0
0
0
0.6718 0.2240 0.5805 0.1868
0.6951 0.6678 0.4753 0.3735
.
This matrix has two classes µ and ν with index sets {1, 2} and {3, 4}, and both classes
are spectral, in both algebras. In max algebra ρ⊕
µ = 1 and ρ⊕
ν = a33 < 1. The eigencones
of matrix powers associated with ρ⊕
µ = 1 are
V⊕(A, 1) = span⊕{(1, 1, 0.6718, 0.6951)},
V⊕(A⊗2, 1) = span⊕{(1, 0, 0.3900, 0.6678), (0, 1, 0.6718, 0.6951)},
and the eigencone associated with ρ⊕
ν is generated by the third column of the matrix:
V⊕(A, ρ⊕
ν ) = span⊕{(0, 0, 0.5805, 0.4753)}.
By Main Theorem 1, core⊕(A) is equal to the (max-algebraic) sum of V⊕(A⊗2, 1) and
V⊕(A, ρ⊕
ν ). To this end, observe that already in the second max-algebraic power
(25)
A⊗2 =
1
0
0
1
0
0
0
0
0.3900 0.6718 0.3370 0.1084
0.6678 0.6951 0.2759 0.1395
the first two columns are the generators of V⊕(A⊗2, 1). However, the last column is still
not proportional to the third one which shows that span⊕(A⊗2) 6= core⊕(A). However, it
can be checked that this happens in span⊕(A⊗4), with the first two columns still equal to
the generators of V⊕(A⊗2, 1), which shows that span⊕(A⊗4) is the sum of above mentioned
max cones, and hence span⊕(A⊗4) = span⊕(A⊗5) = . . . = core⊕(A). Hence we see that
A is column periodic (S5) and the core finitely stabilizes. See Figure 2 for a symbolic
illustration.
40
PETER BUTKOVI C, HANS SCHNEIDER, SERGEI SERGEEV, AND BIT-SHUN TAM
Figure 2. The spans of matrix powers (upper graph) and the periodic
sequence of their eigencones (lower graph) in Example 2 (max algebra)
In nonnegative algebra, ρ+
µ = 1 and ρ+
ν = 0.7924. Computing the eigenvectors of A
and A×2 yields
V+(A, 1) = span+{(0.1326, 0.1326, 0.6218, 0.7604)},
V+(A×2, 1) = span+{(0.2646, 0, 0.5815, 0.7693), (0, 0.2566, 0.6391, 0.7251)},
and
V+(A, ρ+
ν ) = span+{(0, 0, 0.6612, 0.7502)}.
Here core+(A) is equal to the ordinary (Minkowski) sum of V+(A×2, 1) and V+(A, ρ+
ν ).
To this end, it can be observed that, within the first 4 digits, the first two columns
of A×t become approximately periodic after t = 50, and the columns of powers of the
normalized submatrix Aνν/ρ+
ν approximately stabilize after t = 7. Of course, there is no
finite stabilization of the core in this case. However, the structure of the nonnegative core
is similar to the max-algebraic counterpart described above.
TWO CORES OF A NONNEGATIVE MATRIX
41
Acknowledgement
The authors are grateful to the anonymous referee for careful reading and a great
number of useful suggestions which helped to improve the presentation and to eliminate
some flaws and errors.
References
[1] M. Akian, S. Gaubert, and C. Walsh. Discrete max-plus spectral theory. In G.L. Litvinov and V.P.
Maslov, editors, Idempotent Mathematics and Mathematical Physics, volume 377 of Contemporary
Mathematics, pages 53 -- 77. AMS, 2005.
[2] F. L. Baccelli, G. Cohen, G. J. Olsder, and J. P. Quadrat. Synchronization and Linearity: an Algebra
for Discrete Event Systems. Wiley, 1992.
[3] A. Berman and R.J. Plemmons. Nonnegative Matrices in the Mathematical Sciences. Society for
Industrial and Applied Mathematics, Philadelphia, 1994.
[4] S. Bezuglyi, J. Kwiatkowski, K. Medynets, and B. Solomyak. Invariant measures on stationary
Bratteli diagrams. Ergodic Theory Dynam. Systems, 30(4):973-1007, 2010.
[5] R.A. Brualdi and H.J. Ryser. Combinatorial Matrix Theory. Cambridge Univ. Press, 1991.
[6] P. Butkovic. Max-algebra: the linear algebra of combinatorics? Linear Alg. Appl., 367:313 -- 335,
2003.
[7] P. Butkovic. Max-linear Systems: Theory and Algorithms. Springer, 2010.
[8] P. Butkovic and R. A. Cuninghame-Green. On matrix powers in max algebra. Linear Alg. Appl.,
421(2-3):370 -- 381, 2007.
[9] P. Butkovic, R.A. Cuninghame-Green, and S. Gaubert. Reducible spectral theory with applica-
tions to the robustness of matrices in max-algebra. SIAM J. on Matrix Analysis and Applications,
31(3):1412 -- 1431, 2009.
[10] P. Butkovic, H. Schneider, and S. Sergeev. Generators, extremals and bases of max cones. Linear
Alg. Appl., 421:394 -- 406, 2007.
[11] P. Butkovic, H. Schneider, and S. Sergeev. Z-matrix equations in max algebra, nonnegative lin-
ear algebra and other semirings. Linear and Multilinear Alg., 60(10):1191-1210, 2012. E-print
arXiv:1110.4564.
[12] P. Butkovic, H. Schneider, and S. Sergeev. On the max-algebraic core of a nonnegtaive matrix. In
preparation, 2012.
[13] M.-H. Chi. The long-run behavior of Markov chains. Linear Alg. Appl., 244:111 -- 121, 1996.
[14] G. Cohen, D. Dubois, J.P. Quadrat, and M. Viot. Analyse du comportement p´eriodique de syst`emes
de production par la th´eorie des dioıdes. INRIA, Rapport de Recherche no. 191, F´evrier 1983.
42
PETER BUTKOVI C, HANS SCHNEIDER, SERGEI SERGEEV, AND BIT-SHUN TAM
[15] R. A. Cuninghame-Green. Minimax Algebra, volume 166 of Lecture Notes in Economics and Math-
ematical Systems. Springer, Berlin, 1979.
[16] M. Fiedler and V. Pt´ak. Diagonally dominant matrices. Czechoslovak Math. J., 17(3):420 -- 433, 1967.
[17] S. Friedland and H. Schneider. The growth of powers of a nonnegative matrix. SIAM J. Alg. Discr.
Meth., 1(2):185 -- 200, 1980.
[18] G. F. Frobenius. Uber Matrizen aus nicht negativen Elementen. Sitzungsber. Kon. Preuss. Akad.
Wiss., 1912. In: Ges. Abh., Springer, vol. 3, 1968, pp. 546-557.
[19] S. Gaubert. Th´eorie des syst`emes lin´eaires dans les dioıdes. PhD thesis, Ecole des Mines de Paris,
1992.
[20] S. Gaubert and R. D. Katz. The Minkowski theorem for max-plus convex sets. Linear Alg. Appl.,
421(2-3):356 -- 369, 2007. E-print arXiv:math.GM/0605078.
[21] M. Gavalec. Periodicity in Extremal Algebras. Gaudeamus, Hradec Kr´alov´e, 2004.
[22] D. J. Hartfiel. Nonhomogeneous matrix products. World Scientific, Singapore, 2002.
[23] B. Heidergott, G.-J. Olsder, and J. van der Woude. Max-plus at Work. Princeton Univ. Press, 2005.
[24] R.D. Katz, H. Schneider, and S. Sergeev. On commuting matrices in max algebra and in classical
nonnegative algebra. Linear Alg. Appl., 436(2):276-292, 2012.
[25] K.H. Kim. Boolean Matrix Theory and Applications. Marcel Dekker, New York, 1982.
[26] V. N. Kolokoltsov and V. P. Maslov. Idempotent analysis and its applications. Kluwer Academic
Pub., 1997.
[27] G.L. Litvinov and V.P. Maslov. The correspondence principle for idempotent calculus and some
computer applications. In J. Gunawardena, editor, Idempotency, pages 420 -- 443. Cambridge Univ.
Press, 1998. E-print arXiv:math/0101021.
[28] G. Merlet. Semigroup of matrices acting on the max-plus projective space. Linear Alg. Appl.,
432(8):1923 -- 1935, 2010.
[29] H. Minc. Nonnegative matrices. Wiley, 1988.
[30] M. Moln´arov´a and J. Pribis. Matrix period in max-algebra. Discrete Appl. Math., 103:167 -- 175, 2000.
[31] K. Nachtigall. Powers of matrices over an extremal algebra with applications to periodic graphs.
Mathematical Methods of Operations Research, 46:87 -- 102, 1997.
[32] V. Nitica and I. Singer. Max-plus convex sets and max-plus semispaces. I. Optimization, 56:171 -- 205,
2007.
[33] N.J. Pullman. A geometric approach to the theory of nonnegative matrices. Linear Alg. Appl., 4:297 --
312, 1971.
[34] U.G. Rothblum and P. Whittle. Growth optimality for branching Markov decision chains. Mathe-
matics of operations research, 7(4):582 -- 601, 1982.
TWO CORES OF A NONNEGATIVE MATRIX
43
[35] H. Schneider. The influence of the marked reduced graph of a nonnegative matrix on the Jordan
form and on related properties: A survey. Linear Alg. Appl., 84:161 -- 189, 1986.
[36] B. De Schutter. On the ultimate behavior of the sequence of consecutive powers of a matrix in the
max-plus algebra. Linear Alg. Appl., 307:103 -- 117, 2000.
[37] E. Seneta. Non-negative matrices and Markov chains. Springer, 1981.
[38] S. Sergeev. Max algebraic powers of irreducible matrices in the periodic regime: An application of
cyclic classes. Linear Alg. Appl., 431(6):1325 -- 1339, 2009.
[39] S. Sergeev. Max-algebraic attraction cones of nonnegative irreducible matrices. Linear Alg. Appl.,
435(7):1736 -- 1757, 2011.
[40] S. Sergeev and H. Schneider. CSR expansions of matrix powers in max algebra. Transactions of the
AMS, 364:5969-5994, 2012.
[41] S. Sergeev, H. Schneider, and P. Butkovic. On visualization scaling, subeigenvectors and Kleene stars
in max algebra. Linear Alg. Appl., 431(12):2395 -- 2406, 2009.
[42] G. Sierksma. Limiting polytopes and periodic Markov chains. Linear and Multilinear Algebra, 46:281 --
298, 1999.
[43] B.-S. Tam and H. Schneider. On the core of a cone-preserving map. Transactions of the AMS,
343(2):479 -- 524, 1994.
[44] B.-S. Tam and H. Schneider. On the invariant faces associated with a cone-preserving map. Trans-
actions of the AMS, 353:209-245, 2000.
[45] E. Wagneur. Moduloıds and pseudomodules. 1. Dimension theory. Discrete Math., 98:57 -- 73, 1991.
Peter Butkovic, University of Birmingham, School of Mathematics, Watson Building,
Edgbaston B15 2TT, UK
E-mail address: [email protected]
Hans Schneider, University of Wisconsin-Madison, Department of Mathematics, 480
Lincoln Drive, Madison WI 53706-1313, USA
E-mail address: [email protected]
Sergeı Sergeev, University of Birmingham, School of Mathematics, Watson Building,
Edgbaston B15 2TT, UK
E-mail address: [email protected]
Bit-Shun Tam, Tamkang University, Department of Mathematics, Tamsui, Taiwan 25137,
R.O.C.
E-mail address: [email protected]
|
1102.2629 | 2 | 1102 | 2011-06-29T03:25:33 | Outer restricted derivations of nilpotent restricted Lie algebras | [
"math.RA"
] | In this paper we prove that every finite-dimensional nilpotent restricted Lie algebra over a field of prime characteristic has an outer restricted derivation whose square is zero unless the restricted Lie algebra is a torus or it is one-dimensional or it is isomorphic to the three-dimensional Heisenberg algebra in characteristic two as an ordinary Lie algebra. This result is the restricted analogue of a result of T\^og\^o on the existence of nilpotent outer derivations of ordinary nilpotent Lie algebras in arbitrary characteristic and the Lie-theoretic analogue of a classical group-theoretic result of Gasch\"utz on the existence of $p$-power automorphisms of $p$-groups. As a consequence we obtain that every finite-dimensional non-toral nilpotent restricted Lie algebra has an outer restricted derivation. | math.RA | math |
OUTER RESTRICTED DERIVATIONS OF NILPOTENT
RESTRICTED LIE ALGEBRAS
J ORG FELDVOSS, SALVATORE SICILIANO, AND THOMAS WEIGEL
Abstract. In this paper we prove that every finite-dimensional nilpotent re-
stricted Lie algebra over a field of prime characteristic has an outer restricted
derivation whose square is zero unless the restricted Lie algebra is a torus
or it is one-dimensional or it is isomorphic to the three-dimensional Heisen-
berg algebra in characteristic two as an ordinary Lie algebra. This result is
the restricted analogue of a result of Togo on the existence of nilpotent outer
derivations of ordinary nilpotent Lie algebras in arbitrary characteristic and
the Lie-theoretic analogue of a classical group-theoretic result of Gaschutz on
the existence of p-power automorphisms of p-groups. As a consequence we
obtain that every finite-dimensional non-toral nilpotent restricted Lie algebra
has an outer restricted derivation.
1. Introduction
In 1966 W. Gaschutz proved the following celebrated result:
Theorem. (W. Gaschutz [3]) Every finite p-group of order > p has an outer au-
tomorphism of p-power order.
Since every finite nilpotent group is a direct product of its Sylow p-subgroups,
the outer automorphism group of a finite nilpotent group is a direct product of
the outer automorphism groups of its Sylow p-subgroups. Therefore it is a direct
consequence of Gaschutz' theorem that every finite nilpotent group of order greater
than 2 has an outer automorphism. This answers a question raised by E. Schenkman
and F. Haimo in the affirmative (see [9]).
Since groups and Lie algebras often have structural properties in common, it
seems rather natural to ask whether an analogue of Gaschutz' theorem holds in the
setting of ordinary or restricted Lie algebras. In the case of ordinary Lie algebras
a stronger version of such an analogue is already known and was established by
S. Togo around the same time as Gaschutz proved his result.
Theorem. (S. Togo [11, Corollary 1]) Every nilpotent Lie algebra of finite dimen-
sion > 1 over an arbitrary field has an outer derivation whose square is zero.
In fact, Togo's result is more general (see [11, Theorem 1]) and is a refinement
of a theorem of E. Schenkman that establishes the existence of outer derivations for
non-zero finite-dimensional nilpotent Lie algebras (see [7, Theorem 4]). Much later
the first author proved a restricted analogue of Schenkman's result for p-unipotent
2000 Mathematics Subject Classification. 17B05, 17B30, 17B40, 17B50, 17B55, 17B56.
Key words and phrases. Restricted Lie algebra, nilpotent Lie algebra, p-unipotent restricted
Lie algebra, torus, Heisenberg algebra, restricted derivation, outer restricted derivation, nilpotent
outer restricted derivation, restricted cohomology, p-supplement, abelian p-ideal, maximal abelian
p-ideal, maximal p-ideal, free module.
1
2
J ORG FELDVOSS, SALVATORE SICILIANO, AND THOMAS WEIGEL
restricted Lie algebras (see [2, Corollary 5.2]).
(Here we follow [1, Section I.4,
Exercise 23, p. 97] by calling a restricted Lie algebra (L, [p]) p-unipotent if for every
x ∈ L there exists some positive integer n such that x[p]n
= 0.) In this paper we
prove that every finite-dimensional nilpotent restricted Lie algebra L over a field
F of characteristic p > 0 has an outer restricted derivation whose square is zero
unless: (1) L is a torus, or (2) dimF L = 1, or (3) p = 2 and L is isomorphic to the
three-dimensional Heisenberg algebra h1(F) over F as an ordinary Lie algebra (see
Theorem 1). Indeed, in these three cases every nilpotent restricted derivation is
inner. As a consequence we also obtain a generalization of [2, Corollary 5.2] to non-
toral nilpotent restricted Lie algebras (see Theorem 2) which is the full analogue of
Schenkman's result for nilpotent restricted Lie algebras.
In the following we briefly recall some of the notation that will be used in this
paper. A derivation D of a restricted Lie algebra (L, [p]) is called restricted if
D(x[p]) = (adL x)p−1(D(x)) for every x ∈ L (see [6, Section 4, (15), p. 21]) and the
set of all restricted derivations of L is denoted by Derp(L). Observe that Derp(L) is
a restricted Lie algebra (see [6, Theorem 4]) and that every inner derivation of L is
restricted. In this paper we will use frequently without further explanation that the
vector space of outer restricted derivations Derp(L)/ ad(L) of L is isomorphic to the
first adjoint restricted cohomology space H 1
∗ (L, L) of L in the sense of Hochschild
(see [4, Theorem 2.1]). More generally, we will need for any restricted L-module
M the vector space
Z 1
∗ (L, M ) := { D ∈ Der(L, M ) ∀ x ∈ L : D(x[p]) = (x)p−1
M (D(x)) }
of restricted 1-cocycles of L with values in M (see again [4, Theorem 2.1]) and the
vector space
B1
∗(L, M ) := { D ∈ HomF(L, M ) ∃ m ∈ M ∀ x ∈ L : D(x) = x · m }
of restricted 1-coboundaries of L with values in M . Then
H 1
∗ (L, M ) := Z 1
∗ (L, M )/B1
∗(L, M )
denotes the first restricted cohomology space of L with values in M . For any L-
module M we denote by M L := { m ∈ M ∀ x ∈ L : x · m = 0 } the space of
L-invariants of M and by SocL(M ) the largest semisimple L-submodule of M .
Moreover, u(L) denotes the restricted universal enveloping algebra of L (see [6,
Section 2], [8, Section V.7, Theorem 12], or [10, Section 2.5]). For a subset S of
L we denote by hSip the p-subalgebra of L generated by S and by CentL(S) the
centralizer of S in L. Finally, Z(L) := CentL(L) is the center of L and L′ is
the derived subalgebra of L. For more notation and well-known results from the
structure theory of ordinary and restricted Lie algebras we refer the reader to the
first two chapters in [10].
2. Preliminaries
2.1. Restricted derivations and cohomology. Let L be a restricted Lie algebra
over a field of characteristic p > 0 and let I be a p-ideal of L. Then the centralizer
CentL(I) = LI of I in L is a p-ideal of L. By virtue of the five-term exact sequence
associated to the Hochschild-Serre spectral sequence the canonical mapping
ι : H 1
∗ (L/I, CentL(I)) −→ H 1
∗ (L, L)
NILPOTENT OUTER RESTRICTED DERIVATIONS
3
is injective. Hence one has the following sufficient condition for the existence of
outer restricted derivations:
Lemma 1. Let L be a restricted Lie algebra over a field of characteristic p > 0, let
I be a p-ideal of L, and let D be a restricted derivation of L satisfying I ⊆ Ker(D)
as well as Im(D) ⊆ CentL(I), but Im(D) 6⊆ [L, CentL(I)]. Then D is not inner.
Proof. Let D ∈ Z 1
hypothesis, Im( D) 6⊆ [L, CentL(I)] and thus D 6∈ B1
D + B1
∗(L/I, CentL(I)) be the linear transformation induced by D. By
∗(L/I, CentL(I)). Hence
∗) 6= 0, so that D is not inner. (cid:3)
∗ 6= 0, and therefore D + ad(L) = ι( D + B1
∗ := B1
By using the injectivity of ι again, one deduces the following criterion for the
existence of outer restricted derivations whose square is zero.
Lemma 2. Let L be a restricted Lie algebra over a field of characteristic p > 0
that contains p-ideals I and J with the following properties:
(i) J ⊆ I ∩ CentL(I) = Z(I) and
(ii) the canonical mapping γ : H 1
∗ (L/I, J) → H 1
∗ (L/I, CentL(I)) is non-trivial.
Then there exists an outer restricted derivation of L whose square is zero.
Proof. Let D ∈ Z 1
Z 1
∗(L, L) be the linear transformation induced by D. Then it follows from
∗ (L/I, J) be such that γ( D + B1
∗(L/I, J)) 6= 0 and let D ∈
that D2 = 0. Moreover,
Im(D) ⊆ J ⊆ I ⊆ Ker(D)
D + ad(L) = ι(γ( D + B1
∗(L/I, J))) 6= 0 ,
and thus D is an outer restricted derivation of L with D2 = 0.
(cid:3)
2.2. Abelian p-ideals. In the proof of our main result we will need the following
criterion for the existence of certain p-supplements of abelian p-ideals (for the first
part of the proof see also [4, Lemma 3.1]).
Proposition 1. Let L be a restricted Lie algebra over a field of characteristic p > 0
and let A be an abelian p-ideal of L containing Z(L). If H 2
∗ (L/A, A) = 0, then there
exists a p-subalgebra H of L such that L = A + H and A ∩ H = Z(L).
Proof. Let BA be a vector space basis of A being contained in a vector space basis
B of L. According to [8, Section V.7, Theorem 11] or [10, Theorem 2.2.3], the
function ψ : B → L defined by
ψ(x) := (cid:26) 0
x[p]
if x ∈ BA
if x ∈ B \ BA
can be extended uniquely to a p-mapping [p]′
Lie algebra. By construction,
: L → L making (L, [p]′) a restricted
(1)
x[p] − x[p]′
∈ Z(L)
for every x ∈ L .
The ideal A -- with trivial p-mapping -- is also a p-ideal of (L, [p]′), and the identity
mapping induces an isomorphism between (L/A, [p]) and (L/A, [p]′). It follows from
H 2
∗ (L/A, A) = 0 that there exists a p-subalgebra C of (L, [p]′) such that L = A ⊕ C
(see [4, Theorem 3.3]). Then by (1), H := Z(L) ⊕ C is a p-subalgebra of (L, [p])
with the desired properties.
(cid:3)
4
J ORG FELDVOSS, SALVATORE SICILIANO, AND THOMAS WEIGEL
Remark. A similar proof shows that Proposition 1 holds more generally for abelian
p-ideals A of L that do not necessarily contain the center of L if one replaces
everywhere Z(L) by the image of A under the p-mapping of (L, [p]).
The next result shows that maximal abelian p-ideals of finite-dimensional non-
abelian nilpotent restricted Lie algebras are self-centralizing and contain the center
properly.
Proposition 2. Let L be a finite-dimensional non-abelian nilpotent restricted Lie
algebra over a field of characteristic p > 0 and let A be a maximal abelian p-ideal
of L. Then Z(L) ( A = CentL(A).
Proof. Suppose by contradiction that C := CentL(A) ) A. As C/A is a non-zero
ideal of the nilpotent Lie algebra L/A, there is an element x in L that does not
belong to A such that x + A ∈ Z(L/A) ∩ C/A. Put X := A + hxip. Then X is an
abelian p-ideal of L properly containing A which by the maximality of A implies
that L = X is abelian contradicting the hypothesis that L is non-abelian.
It follows from A = CentL(A) that A is a faithful L/A-module under the induced
adjoint action. Suppose now that Z(L) = A. Then Z(L) is a faithful and trivial
L/ Z(L)-module under the induced adjoint action which implies that L = Z(L) is
abelian, which again is a contradiction.
(cid:3)
Let Tp(L) denote the set of all semisimple elements of a finite-dimensional nilpo-
tent restricted Lie algebra L. Since L is nilpotent, we have that Tp(L) ⊆ Z(L) and
thus Tp(L) is the unique maximal torus of L. Furthermore, Tp(L) is a p-ideal of
L and it follows from [10, Theorem 2.3.4] that L/I is p-unipotent for every p-ideal
I of L that contains Tp(L). These well-known results will be used in the following
proofs without further explanation.
Proposition 3. Let L be a finite-dimensional non-abelian nilpotent restricted Lie
algebra over a field F of characteristic p > 0 and let A be a maximal abelian p-ideal
of L. If furthermore every restricted derivation D of L with D2 = 0 is inner, then A
is a free u(L/A)-module of rank r := dimF Z(L). In particular, dimF L = d + r · pd,
where d := dimF L/A.
Proof. Suppose that H 1
∗ (L/A, A) 6= 0. Then by Lemma 2 (applied to I := J := A)
in conjunction with Proposition 2, there exists an outer restricted derivation of L
whose square is zero, a contradiction. Hence H 1
∗ (L/A, A) = 0.
Another application of Proposition 2 yields Tp(L) ⊆ A. Consequently, L/A is
p-unipotent and it follows from [2, Proposition 5.1] that A is a free u(L/A)-module.
Moreover, by virtue of the Engel-Jacobson theorem (see [10, Corollary 1.3.2(1)]),
the trivial L/A-module F is the only irreducible restricted u(L/A)-module up to
isomorphism. Thus
SocL/A(A) = AL/A = Z(L) ,
and A is a free u(L/A)-module of rank r := dimF Z(L). In particular, dimF A = r·pd
and the dimension formula for L follows.
(cid:3)
Note that dimF A = dimF Z(L) · pdimF L/A shows again that A contains Z(L)
properly if L is not abelian.
NILPOTENT OUTER RESTRICTED DERIVATIONS
5
2.3. Maximal p-ideals of nilpotent restricted Lie algebras. For the conve-
nience of the reader we include a proof of the following result.
Lemma 3. Let L be a finite-dimensional non-toral nilpotent restricted Lie algebra
over a field of characteristic p > 0. Then every maximal p-ideal of L containing
the unique maximal torus of L has codimension one in L.
Proof. Let I be a maximal p-ideal of L that contains the unique maximal torus
Tp(L) of L. Then Z(L/I) 6= 0 and because L/I is p-unipotent, there exists a non-
zero central element z of L/I such that z[p] = 0. Then Z := Fz is a one-dimensional
p-ideal of L/I and the inclusion-preserving bijection between p-ideals of L/I and
p-ideals of L that contain I in conjunction with the maximality of I yields that
L/I = Z is one-dimensional.
(cid:3)
3. Main results
The main goal of this paper is to establish the existence of nilpotent outer re-
stricted derivations of finite-dimensional nilpotent restricted Lie algebras.
It is
well-known that the restricted cohomology of tori vanishes (see [4] and the main
result of [5] or [2, Corollary 3.6]). Hence tori have no outer restricted derivations.
For the convenience of the reader we include the following straightforward proof of
the latter statement.
Proposition 4. Every restricted derivation of a torus over any field of character-
istic p > 0 vanishes.
Proof. Suppose that L is a torus and let D be any restricted derivation of L.
Then we have D(x[p]) = (adL x)p−1(D(x)) = 0 for every x ∈ L, and inductively,
D(x[p]n
) = 0 for every x ∈ L and every positive integer n. Since x ∈ hx[p]ip, we
conclude that D(x) = 0 for every x ∈ L.
(cid:3)
If L is one-dimensional, then either L is a torus or L = Fe with e[p] = 0. In the
second case the (outer) restricted derivations of L coincide with the vector space
endomorphisms EndF(L) = F · idL of L and therefore no non-zero (outer) restricted
derivation is nilpotent.
Indeed there is only one more finite-dimensional nilpotent restricted Lie algebra
(up to isomorphism of ordinary Lie algebras) that has no nilpotent outer restricted
derivations, namely the three-dimensional restricted Heisenberg algebra over a field
of characteristic two. Since the derived subalgebra of any Heisenberg algebra is
central, it follows from [10, Example 2, p. 72] that Heisenberg algebras are re-
strictable. For the proof of the next result one only needs that the image of every
p-mapping is central so that the result does not depend on the particular choice of
the p-mapping.
Proposition 5. Every nilpotent restricted derivation of a three-dimensional re-
stricted Heisenberg algebra over a field of characteristic two is inner.
Proof. Let F be any field of characteristic 2 and let L be the three-dimensional
Heisenberg algebra h1(F) = Fx + Fy + Fz defined by [x, y] = z ∈ Z(L). Since L′ is
central, we have that
(2)
L[2] ⊆ Z(L) = Fz .
6
J ORG FELDVOSS, SALVATORE SICILIANO, AND THOMAS WEIGEL
Suppose that D is any nilpotent outer restricted derivation of L. It follows from
[x, D(z)] = [y, D(z)] = 0
that z is an eigenvector of D. This forces D(z) = 0, as D is nilpotent. Moreover,
by (2) we obtain that
and then
(adL x)(D(x)) = D(x[2]) = 0 ,
(adL y)(D(y)) = D(y[2]) = 0 ,
D(x) = αx + λz ,
D(y) = βy + µz
for suitable α, β, λ, µ ∈ F. Furthermore, D([x, y]) = 0 yields α = β.
Now, observe that D2 = αD and therefore Dn = αn−1D for every positive
integer n. As D is nilpotent, we conclude that α = 0. Thus one has
D = adL(λy + µx) ,
so that D is inner, as claimed.
(cid:3)
Now we are ready to prove our first main result:
Theorem 1. Let L be a nilpotent restricted Lie algebra of finite dimension > 1
over a field F of characteristic p > 0. Then L has an outer restricted derivation D
with D2 = 0 unless L is a torus or p = 2 and L ∼= h1(F) as ordinary Lie algebras.
Proof. Assume that L is neither a torus nor one-dimensional. In the following T
will denote the unique maximal torus Tp(L) of L and by assumption L/T 6= 0. We
proceed by a case-by-case analysis.
Case 1: There exists a maximal p-ideal I of L containing T such that Z(L) * I.
According to Lemma 3, I has codimension 1 in L and therefore L = Fx ⊕ I for
any x ∈ Z(L)\I. Let 0 6= z ∈ Z(I) and consider the linear transformation D of L
defined by setting D(x) := z and D(y) := 0 for every y ∈ I. Then D is a derivation
of L (see [1, Section I.4, Exercise 8(a), p. 92]) with D2 = 0. Moreover, the inner
derivation adL a vanishes on x for every a ∈ L and thus D cannot be inner. Finally,
as L/I is p-unipotent and I has codimension 1 in L, it follows that L[p] ⊆ I. This
implies that D(l[p]) = 0 = (adL l)p−1(D(l)) for every l ∈ L, and so D is restricted.
Note that Case 1 covers already the abelian case. So for the rest of the proof we
may assume that L is not abelian.
Case 2: Z(L) ⊆ I for every maximal p-ideal I of L containing T .
Suppose that CentL(I) 6⊆ I for some maximal p-ideal I of L containing T . Then
there exists x ∈ CentL(I)\I such that L = Fx ⊕ I; in particular, x ∈ Z(L) ⊆ I, a
contradiction. Hence we may assume from now on that CentL(I) = Z(I) for every
maximal p-ideal I of L containing T .
Case 2.1: There exists a maximal p-ideal I of L containing T such that CentL(I)
= Z(L).
Let x ∈ L\I be such that L = Fx ⊕ I. Let 0 6= z ∈ Z(L) and consider the vector
space endomorphism D of L given by D(x) := z and D(y) := 0 for every y ∈ I.
NILPOTENT OUTER RESTRICTED DERIVATIONS
7
Then as above, D is a restricted derivation of L with D2 = 0. By construction,
Ker(D) = I and Im(D) = Fz. Since by hypothesis [L, CentL(I)] = 0, it follows
from Lemma 1 that D is not inner.
Case 2.2: Z(L) ( CentL(I) = Z(I) for every maximal p-ideal I of L containing
T .
Suppose that every restricted derivation D of L satisfying D2 = 0 is inner. Let A be
a maximal abelian p-ideal of L. Since L is not abelian, it follows from Proposition
2 that A contains Z(L) properly. According to Proposition 3, A is a free u(L/A)-
module of rank r := dimF Z(L) and dimF L = d + r · pd, where d := dimF L/A.
Since A is a free u(L/A)-module, [2, Proposition 5.1] implies that H 2
∗ (L/A, A) =
0. Hence by Proposition 1, there exists a p-subalgebra H of L such that L = A + H
and A ∩ H = Z(L). In particular, H contains T . As A contains Z(L) properly,
one has H 6= L. Choose now a proper p-subalgebra J of L with H ⊆ J of maximal
dimension. Then J is a maximal p-subalgebra of L. Since J is properly contained
in L, the nilpotency of L implies that J is properly contained in the p-subalgebra
NL(J) := { x ∈ L [x, J] ⊆ J }. Hence the maximality of J yields NL(J) =
L. Consequently, J is a maximal p-ideal of L containing H and in particular T .
According to Lemma 3, J is of codimension 1 in L. Since L = A + J, one has
dimF A/A ∩ J = dimF A + J/J = dimF L/J = 1 ,
so that A ∩ J has codimension 1 in A.
Let B := CentL(J) = Z(J). By hypothesis, there exists an element e ∈ B\ Z(L).
We obtain from B ⊆ CentL(H) and L = A + H that B ∩ A = Z(L) which implies
that e 6∈ A. Since B is an abelian p-ideal of L, e[p] ∈ Z(L) ⊆ A. Hence it follows
from [10, Lemma 2.1.2] that Fe ⊕ A is a p-subalgebra of L.
Let E := Fe ⊕ A/A be the one-dimensional p-unipotent p-subalgebra generated
by the image of e in L/A. Note that A ∩ J ⊆ AE and dimF A ∩ J = r · pd − 1. By
construction, A is a free u(E)-module of rank r · pd−1. Consequently,
r · pd − 1 = dimF A ∩ J ≤ dimF AE = dimF SocE(A) = r · pd−1 .
But this implies pd = 2 and r = 1. Hence F has characteristic 2 and dimF L = 3.
Since the Heisenberg algebra is the only non-abelian nilpotent three-dimensional
Lie algebra (up to isomorphism), this finishes the proof.
(cid:3)
Theorem 1 in conjunction with the introductory remarks of this section yields a
characterization of finite-dimensional nilpotent restricted Lie algebras in terms of
the existence of nilpotent outer restricted derivations.
Corollary 1. For a finite-dimensional nilpotent restricted Lie algebra (L, [p]) over
a field F of characteristic p > 0 the following statements are equivalent:
(i) L is a torus or dimF L = 1 and L[p] = 0 or p = 2 and L ∼= h1(F) as ordinary
Lie algebras.
(ii) L has no nilpotent outer restricted derivation.
(iii) L has no outer restricted derivation D with D2 = 0.
Proof. Clearly, (ii) implies (iii) and it follows from Theorem 1 that (iii) implies (i).
It remains to prove the implication (i)=⇒(ii) which is an immediate consequence
of Proposition 4, the paragraph thereafter, and Proposition 5.
(cid:3)
8
J ORG FELDVOSS, SALVATORE SICILIANO, AND THOMAS WEIGEL
If we exclude the one-dimensional case and assume that the characteristic of the
ground field is greater than two, then Theorem 1 can be used to characterize the tori
among nilpotent restricted Lie algebras in terms of the non-existence of restricted
derivations with various nilpotency properties.
Corollary 2. For a nilpotent restricted Lie algebra L of finite dimension > 1 over
a field of characteristic p > 2 the following statements are equivalent:
(i) L is a torus.
(ii) L has no non-zero restricted derivation.
(iii) L has no nilpotent outer restricted derivation.
(iv) L has no outer restricted derivation D with D2 = 0.
Proof. Obviously, the implications (ii)=⇒(iii) and (iii)=⇒(iv) hold. The impli-
cations (iv)=⇒(i) and (i)=⇒(ii) are immediate consequences of Theorem 1 and
Proposition 4, respectively.
(cid:3)
Finally, we obtain from Theorem 1 the following generalization of [2, Corollary
5.2]:
Theorem 2. Every finite-dimensional non-toral nilpotent restricted Lie algebra
over a field of characteristic p > 0 has an outer restricted derivation.
Proof. According to Theorem 1 and the first paragraph after Proposition 4, it
is enough to show the assertion for the three-dimensional restricted Heisenberg
algebra L := h1(F) over a field F of characteristic 2. Denote the one-dimensional
center of L by Z. If L is 2-unipotent, then the claim follows from [2, Corollary 5.2].
Suppose now that L is not 2-unipotent. Then the center Z of L is a one-dimensional
torus. According to [2, Proposition 3.9] in conjunction with [2, Corollary 3.6], we
obtain that H 1
∗ (L/Z, L). Since L′ is central, we have L[2] ⊆ Z, and
consequently, L/Z is 2-unipotent. Suppose now that L has no outer restricted
derivation and therefore H 1
∗ (L, L) = 0. Then an application of [2,
Proposition 5.1] yields that L is a free u(L/Z)-module and it follows that
∗ (L/Z, L) ∼= H 1
∗ (L, L) ∼= H 1
3 = dimF L ≥ dimF u(L/Z) = 4 ,
a contradiction.
(cid:3)
Remark. It can also be seen directly that any three-dimensional restricted Heisen-
berg algebra h1(F) over a field F of characteristic 2 has an outer restricted deriva-
tion. Namely, let D be a linear transformation of h1(F) vanishing on its center
Z(h1(F)) and inducing the identity transformation on h1(F)/ Z(h1(F)). Then it is
straightforward to see that D is indeed an outer restricted derivation of h1(F).
Neither Theorem 1 nor Theorem 2 does generalize to non-toral solvable restricted
Lie algebras with non-zero center as the direct product of a two-dimensional non-
abelian restricted Lie algebra and a one-dimensional torus over any field of charac-
teristic p > 0 shows. In this case it is not difficult to see that there are no outer
restricted derivations.
[1] N. Bourbaki: Lie Groups and Lie Algebras: Chapters 1 -- 3 (2nd printing), Springer-Verlag,
Berlin/Heidelberg/New York/London/Paris/Tokyo, 1989.
References
NILPOTENT OUTER RESTRICTED DERIVATIONS
9
[2] J. Feldvoss: On the cohomology of restricted Lie algebras, Comm. Algebra 19 (1991), no. 10,
2865 -- 2906.
[3] W. Gaschutz: Nichtabelsche p-Gruppen besitzen aussere p-Automorphismen, J. Algebra 4
(1966), no. 1, 1 -- 2.
[4] G. Hochschild: Cohomology of restricted Lie algebras, Amer. J. Math. 76 (1954), no. 3,
555 -- 580.
[5] G. Hochschild: Representations of restricted Lie algebras of characteristic p, Proc. Amer.
Math. Soc. 5 (1954), no. 4, 603 -- 605.
[6] N. Jacobson: Restricted Lie algebras of characteristic p, Trans. Amer. Math. Soc. 50 (1941),
no. 1, 15 -- 25.
[7] N. Jacobson: A note on automorphisms and derivations of Lie algebras, Proc. Amer. Math.
Soc. 6 (1955), no. 2, 281 -- 283.
[8] N. Jacobson: Lie Algebras, Dover Publications, Inc., New York, 1979 (unabridged and cor-
rected republication of the original edition from 1962).
[9] E. Schenkman: The existence of outer automorphisms of some nilpotent groups of class 2,
Proc. Amer. Math. Soc. 6 (1955), no. 1, 6 -- 11; Erratum, Proc. Amer. Math. Soc. 7 (1956),
no. 6, 1160.
[10] H. Strade and R. Farnsteiner: Modular Lie Algebras and Their Representations, Mono-
graphs and Textbooks in Pure and Applied Mathematics, vol. 116, Marcel Dekker, Inc., New
York/Basel, 1988.
[11] S. Togo: Outer derivations of Lie algebras, Trans. Amer. Math. Soc. 128 (1967), no. 2,
264 -- 276.
Department of Mathematics and Statistics, University of South Alabama, Mobile,
AL 36688 -- 0002, USA
E-mail address: [email protected]
Dipartimento di Matematica "E. de Giorgi", Universit`a del Salento, Via Provinciale
Lecce-Arnesano, I-73100 Lecce, Italy
E-mail address: [email protected]
Dipartimento di Matematica e Applicazioni, Universit`a di Milano-Bicocca, Via R.
Cozzi 53, I-20125 Milano, Italy
E-mail address: [email protected]
|
1612.07407 | 4 | 1612 | 2017-05-10T00:54:10 | Attaching topological spaces to a module (I): Sobriety and spatial frames of submodules | [
"math.RA"
] | In this paper we study some frames associated to an $R$-module $M$. We define semiprimitive submodules and we prove that they form an spatial frame canonically isomorphic to the topology of $Max(M)$. We characterize the soberness of $Max(M)$ in terms of the point space of that frame. Beside of this, we study the regularity of an spatial frame associated to $M$ given by annihilator conditions. | math.RA | math |
ATTACHING TOPOLOGICAL SPACES TO A MODULE (I):
SOBRIETY AND SPATIALITY
MAURICIO MEDINA B ´ARCENAS
LORENA MORALES CALLEJAS
MARTHA LIZBETH SHAID SANDOVAL MIRANDA
´ANGEL ZALD´IVAR CORICHI
ABSTRACT. In this paper we study some frames associated to an R-module
M . We define semiprimitive submodules and we prove that they form an spatial
frame canonically isomorphic to the topology of Max(M ). We characterize the
soberness of Max(M ) in terms of the point space of that frame. Beside of this,
we study the regularity of an spatial frame associated to M given by annihilator
conditions.
1. INTRODUCTION
For the study of a module categories over a unitary ring one technique is through
the examination of complete lattices of submodules, such as in [Alb14b], [Alb14a],
[Simc] and [MSZ16]. Following this idea, we give a module-theoretical counter-
part of the ring-theoretical examination developed by H. Simmons in [Sim89] and
[Sima]. We observe some point-free topology aspects of certain frames that we
construct trough the manuscript and that consideration eventually leads to the con-
struction of some spatial frames, that is, they behave like a topology ([Joh86],
[Sim01] and [PP12]).
In order to obtain these frames and prove their spatiality, we make use of results
and techiniques arised from lattice theory as in [Ros90], [MSZ16] and [Simb], and
the usual algebraic techniques developed in categories, in particular those for the
category σ[M ] (see 2.2) studied in [Wis91], [BJKN80], [Bea02]. Besides these
tools, we will introduce the point-free topology perspective applied to frames pro-
vided by modules as in [MSZ16].
In [MSZ16], it was studied the prime spectrum of a module as well as a gener-
alization of quantales, namely quasi-quantales making use of the techniques men-
tioned above. One of the main results in that article is that the frame of semiprime
modules is an spatial frame like in the classical case of commutative ring theory,
[MSZ16, Proposition 4.27]. In this manuscript we continue this study for the case
of the space of (fully invariant) maximal submodules of a given module. It should
be noted that many of the results will use the theory developed in [MSZ16].
In this article we study primitive submodules of a module M , which are just
annihilators of simple modules in σ[M ], and so will get the frame of semiprimitive
submodules which turns out to be spatial. Moreover this one is isomorphic to the
frame of open sets of the space of maximal submodules of M with the hull-kernel
topology ([Sim01], [Ros90]). An application of the above result is that for a pm-
module (3.19), the space of maximal submodules concides with that of its primitive
1
2
MEDINA, MORALES, SANDOVAL, ZALD´IVAR
submodules and the point space of the frame of semiprimitive submodules, see
Proposition 3.20. As in the case of commutative rings, we study the sobriety of
the space of prime submodules and the space of maximal submodules of a given
module. For instance, we get necessary and sufficient conditions for the sobriety
on the space of maximal submodules.
Since the idea of annihilators of submodules is present in the study of these
frames, following the idea of [Sima] we introduce the frame Ψ(M ) for a module
M which is given in terms of annihilators ( see 5.1) this frame turns to be spatial
and we give sufficient conditions for it to be regular.
The interest of the sudy of these aspects comes from the neccesity of triyng to
give a module-theoretic counterpart of the ring-theoretic aspect of sheaves repre-
sentations of [BSvdB84], [BVDB83] and [BV+83] and also the localic point of
view of these representations given in [Sim89] and [Sim85].
The organization of the paper is as follows: Section 1 is this introduction, Sec-
tion 2 provides the necessary (perhaps not in full detail) material that is needed
for the reading of the next sections. Section 3 is concerned with the first spatial
frame associated to a module, the frame of semiprimitve submodules, we exami-
nate its point space and we see when it coincides with the primitive submodules.
In Section 4 it is proved the sobriety of the space Spec(M ) (the space of prime
submodules) and we give necessary and sufficient conditions for Max(M ) to be
a sober space. In section 5 we introduce a set of submodules Ψ(M ) defined by
annihilator conditions. We prove that this set is an spatial frame and we explore its
regularity.
2. PRELIMINARIES.
2.1. Idiomatic-quantale preliminaries. First we give the idiomatic preliminaries
that are necessary for our analysis.
Definition 2.1. An idiom (A, ≤,W, ∧, ¯1, 0) is a upper-continuous and modular lat-
tice, that is, A is a complete lattice that satisfies the following distributive laws:
(IDL)
holds for all a ∈ A and X ⊆ A directed; and
a ∧ (_ X) = _{a ∧ x x ∈ X}
(ML)
a ≤ b ⇒ (a ∨ c) ∧ b = a ∨ (c ∧ b)
for all a, b, c ∈ A.
A good account of the many uses of these lattices can be found in [Sim14b] and
[Sim14a].
Examples of idioms are given by submodules of a module over a ring. First
some conventions, in the whole text R will be an associative ring with identity, not
necessarily commutative. The word ideal will mean two-sided ideal, unless explic-
itly stated the side (left or right ideal). All modules are unital and left modules.
Let M be an R-module, a submodule N of M is denoted by N ≤ M , whereas
we write N < M when N is a proper submodule of M . Recall that N ≤ M is
called fully invariant submodule, denoted by N ≤f i M , if for every endomorphism
f ∈ EndR(M ), it follows that f (N ) ⊆ N. Denote by
Λ(M ) = {N N ≤ M }
ATTACHING TOPOLOGICAL SPACES TO A MODULE
3
Λf i(M ) = {N N ≤f i M }
It is straightforward to see that Λ(M ) and Λf i(M ) are idioms, in particular the
left (right) ideals of the ring R, Λ(R) constitutes an idiom.
Another important example come from topology, consider any topological space
S and denote by O(S) its topology, then it is know that O(S) is a complete lattice
with satisfies the distributivity laws of definition 2.1, in fact it satisfies the arbi-
trary distributivity law not just for directed subsets. This kind of lattices are called
frames.
Definition 2.2. A complete lattice (A, ≤,W, ∧, ¯1, 0) is a frame, if A satisfies
(FDL)
for all a ∈ A and X ⊆ A any subset.
a ∧ (_ X) = _{a ∧ x x ∈ X}
Frames are certain kind of idioms:
Proposition 2.3 ([Sim14b] Lemma 1.6). Let A be an idiom, A is a frame if and
only if A is a distributive idiom, that is,
or equivalently
a ∨ (b ∧ c) = (a ∨ b) ∧ (a ∨ c)
a ∧ (b ∨ c) = (a ∧ b) ∨ (a ∧ c)
holds for every a, b, c ∈ A.
Thus idioms are a generalization of frames. There exists other structures that
are generalizations of frames.
Definition 2.4. A complete lattice A is a quantale if A has a binary associative
operation · : A × A → A such that
(LQ)
(RQ)
l(cid:16)_ X(cid:17) = _ {lx x ∈ X} , and
(cid:16)_ X(cid:17) r = _ {xr x ∈ X}
hold for all l, r ∈ A and X ⊆ A.
See [Ros90] for theory of quantales. The canonical example of a quantale is
provided by Λ(R) with R a ring, with the usual product of ideals and as we will
see for a module M (with some extra properties) we can give to Λ(M ) a structure
of quantale. By idiomatic-quantale we will mean an idiom that is also a quantale.
One of the main tools for the study of all of these structures are adjoint situa-
tions.
Definition 2.5. A morphism ofW-semilattices, f : A → B is a monotone function
that preserves arbitrary suprema, that is, f [W X] = W f [X] for all X ⊆ A.
We are going to use next fact repetitively. If the partial ordered set A is a W-
semilattice we have the following important fact.
4
MEDINA, MORALES, SANDOVAL, ZALD´IVAR
Proposition 2.6. Given any morphism of W-semilattices, f : A → B there exits
f∗ : B → A such that
f ∗(a) ≤ b ⇔ a ≤ f∗(b)
that is, f and f∗ form an adjunction
A
f∗
f ∗
B
This is a particular case of the General Adjoint functor theorem, a proof of this
can be found in any standard book of category theory for instance [Lei14, Theorem
6.3.10].
The following is easy to check.
Proposition 2.7 (Lemma 3.3, [Sim14b]). From this situation, we can consider the
two compositions
f∗f ∗ : A → A and f ∗f∗ : B → B.
Then:
(1) f∗f ∗ is an inflator, that is, is a monotone and a ≤ f∗f ∗(a) for all a ∈ A.
(2) f ∗f∗ is a deflator, that is, is a monotone and f ∗f∗(b) ≤ b for all b ∈ B.
(3) f∗f ∗ and f ∗f∗ are idempotent.
Now if we are dealing with idioms, and the morphism f : A → B also satisfies
f (a ∧ b) = f (a) ∧ f (b) then from Proposition 2.7 we have:
Proposition 2.8 (Lemma 3.12, [Sim14b]). The closure operators f∗f ∗ and f ∗f∗
satisfies:
and
f∗f ∗(a ∧ b) = f∗f ∗(a) ∧ f∗f ∗(b)
f ∗f∗(a ∧ b) = f ∗f∗(a) ∧ f ∗f∗(b)
Next we will review the above properties in the context of idioms and quantales.
A good account of all these facts and their proofs are in [Sim14b] and [Ros90].
Definition 2.9. Let A be an idiom. A nucleus on A is a function j : A → A such
that:
(1) j is an inflator.
(2) j is idempotent.
(3) j is a prenucleus, that is, j(a ∧ b) = j(a) ∧ j(b).
There also exists the quantale counterpart of this notions:
Definition 2.10. A quantic nucleus k over a quantale A is an idempotent inflator
such that k(a)k(b) ≤ k(ab) for all a, b ∈ A.
As we are dealing with idiomatic-quantales, we require that our nuclei control
the product of the quantale and the ∧ in some nice way, thus we have the following
improvement,
Definition 2.11. A multiplicative nucleus, k : A → A is an idempotent inflator
such that k(a) ∧ k(b) = k(ab).
5
5
u
u
ATTACHING TOPOLOGICAL SPACES TO A MODULE
5
There are several uses of these operators in literature for example they provide
a relative version of A, that is:
Proposition 2.12. Let A be an idiomatic-quantale, let j be a nucleus (or a multi-
plicative nucleus or a quantic nucleus) then consider the set
Aj = {a ∈ A j(a) = a}
the set of all fixed points of j. Then
(1) If j is a nucleus then Aj is an idiom.
(2) If j is a quantic nucleus then Aj is a quantale.
(3) If j is a multiplicative nucleus and the following inequality holds
for all a, b ∈ A then Aj is an idiomatic-quantale.
j(a)j(a) ≤ j(a) ∧ j(b)
The deatails about these facts can be found in [Sim14b] and in a more general
context in [MSZ16, Propositions 3.9, 3.10] where the notion of quasi-quantale is
introduced,
Definition 2.13. Let A be a W-semilattice. We say that A is a quasi-quantale if it
has an associative product A × A → A such that for all directed subsets X, Y ⊆ A
and a ∈ A:
We say A is a left (resp. right, resp. bilateral) quasi-quantale if there exists e ∈ A
such that e(a) = a (resp. (a)e = a, resp. e(a) = a = (a)e) for all a ∈ A.
Definition 2.14. Let A be a quasi-quantale and B a subW-semilattice. We say that
B is a subquasi-quantale of A if
for all directed subsets X, Y ⊆ B and a ∈ B.
In that investigation the authors use this kind of operators to give a radical the-
ory for modules and they observed that this theory is related with some spatial
properties of these structures, let us recall some of that material.
As we saw any topological space S determines a frame, its topology O(S),
this defines a functor from the category of topological spaces into the category of
frames O( ) : T op → F rm. There exists a functor in the other direction:
Definition 2.15. Let A be a frame. An element p ∈ A is a point or a ∧-irreducible
if p 6= 1 and a ∧ b ≤ p ⇒ a ≤ p or b ≤ p.
Denote by pt(A) the set of all points of A.
We can topologize this set as follows, for each a ∈ A define:
p ∈ UA(a) ⇔ a (cid:2) p
and
and
(_ X)a = _{xa x ∈ X},
a(_ Y ) = _{ay y ∈ Y }.
(_ X)a = _{xa x ∈ X}
a(_ Y ) = _{ay y ∈ Y },
6
MEDINA, MORALES, SANDOVAL, ZALD´IVAR
for p ∈ pt(A). The collection O pt(A) = {UA(a) a ∈ A} constitutes a topology
for pt(A). We have a frame morphism
UA : A → O pt(A)
that determines a nucleus on A by Proposition 2.6, this nucleus or the adjoint situ-
ation is called the hull-kernel adjunction. With this, the frame A is spatial if UA is
an injective morphism (hence an isomorphism).
With some work one can prove that this defines a functor pt( ) : F rm → T op
in such way that the pair
T op
O( )
pt( )
, F rm
form a adjoinction. For details see [Joh86], [Sim06] and [PP12].
For a (quasi-)quantale there exists a similar way to produce a space.
Definition 2.16. Let A be a quasi-quantale and B a subquasi-quantale of A. An
element 1 6= p ∈ A is a prime element relative to B if whenever ab ≤ p with
a, b ∈ B then a ≤ p or b ≤ p.
Denote SpecB(A) := {p p is prime inA relative to B}. We can topologize it
imitating the process for pt(A), and obtain the Zariski-like topology for SpecB(A).
Theorem 2.17. Let A be a quasiquantale, B a sub-quasiquantal of A. Then, the
following conditions hold.
(1) SpecB(A) is a topological space where the closed sets are given by V (b) =
{p ∈ SpecB(A) b ≤ p} with b ∈ B. Duality, the open sets are give by
U (b) = {p ∈ SpecB(A) b 6≤ p}. [MSZ16, Proposition 3.18]
SpecB(A) p ∈ V (b)}. [MSZ16, Proposition 3.20]
(2) For every b ∈ B, µ(b) is the greatest element in B such that µ(b) ≤ V{p ∈
(3) The W −semilattice morphism U : B → O(SpecB(A)) has a right adjoint
U∗ : O(SpecB(A)) → B given by U∗(W ) := W{b ∈ B U (b) ⊆ W } and
the composition µ = U∗ ◦ U : B → B is a multiplicative nucleus. [MSZ16,
Proposition 3.21]
(4) Bµ := {x ∈ B µ(x) = x} is upper continuous. [MSZ16, Corollary 3.22]
(5) If in addition, B is a quasi-quantale such that (W X)a = W{xa x ∈ X}, for
every X ⊆ A and a ∈ A, then Bµ is a frame. [MSZ16, Corollary 3.11]
The following statements are important to prove Theorem 5.6, they are hard to
find in the literature thus for convenience of the reader we provide their proofs.
Our prototype of idiomatic-quantale is Λ(R). Note that this lattice is also com-
pactly generated:
Definition 2.18. Let A be an idiomatic-quantale:
(1) An element c ∈ A is compact
for each X directed subset of A. This is equivalent to consider any subset X
c ≤ _ X ⇒ (∃x ∈ X)[c ≤ x]
of A such that c ≤ W X then c ≤ W F for a F ⊆ X finite.
(2) A is compactly generated if for each a ∈ A we have a = W C for some set C
of compact elements.
,
k
k
ATTACHING TOPOLOGICAL SPACES TO A MODULE
7
Now consider a sublattice Λ of A and a compact element m ∈ A. Set
X (m) = {x ∈ Λ m 6≤ x}.
Observe that this subset depends on Λ, and if m = 0 then X (m) = ∅.
Proposition 2.19. Let A be an idiomatic-quantale, Λ a sublattice of A and m ∈ A
a compact element. Then,
(1) The family X (m) is closed under directed unions.
(2) Each member of X (m) is less than or equal to a maximal member of X (m).
Proof. (1) Let D be a directed subset of X (m). Suppose that m ≤ W D. Since
m is a compact element, we have m ≤ W F for some finite family F ⊆ D. Thus
there exists d ∈ D such that m ≤ W F ≤ d. This gives d /∈ X (m) which is a
contradiction.
(2) This is an application of (1) and Zorn's Lemma.
(cid:3)
We know that if A is an idiom then A is a frame provided it is a distributive
lattice (Proposition 2.3), there exists many non-distributive idioms, but we can
subtract the distributive part of a idiom:
Definition 2.20. Let A be an idiom. An element a ∈ A is distributive if it satisfies
the following equivalent conditions:
(1) (∀x, y ∈ A[a ∨ (x ∧ y) = (a ∨ x) ∧ (a ∨ y)]).
(2) (∀x, y ∈ A[a ∧ (x ∨ y) = (a ∧ x) ∨ (a ∧ y)]).
Denote by F (A) the set of all distributive elements of A, it is easy to see that
F (A) is a sub-idiom of A which is distributive and thus it is a frame.
Proposition 2.21. Let A be an idiomatic-quantale, Λ a sub-idiom of F (A) and
m ∈ A a compact element. Then each maximal member of X (m) is a point of Λ.
Proof. Let p be a maximal element of X (m). Since m 6≤ p then p 6= 1. Now,
suppose a, b ∈ Λ are elements such that a 6≤ p and b 6≤ p. We shall show that
a ∧ b 6≤ p. We have that p < a ∨ p and p < b ∨ p since a 6≤ p and b 6≤ p. Thus
a ∨ p /∈ X (m) and b ∨ p /∈ X (m) by the maximality of p. This gives
m ≤ (a ∨ p) ∧ (b ∨ p) = (a ∧ b) ∨ p
where the right hand holds since Λ is distributive. Finally, since m 6≤ p, then
a ∧ b 6≤ p, as required.
(cid:3)
Let Λ as above. Denote by pt(Λ) the set of points of Λ then it is not empty. Let
S ⊆ pt(Λ). For each a ∈ A we define the following subset of S,
p ∈ d(a) ⇔ a (cid:2) p.
Therefore we can topologize S with the family O(S) = {d(a) a ∈ A}, thus we
have an idiom epimorphism
d : Λ → O(S)
with this we can prove the following which has an important consequence and il-
lustrate the spatial behaviour of distributive lattices of compactly generated idioms.
8
MEDINA, MORALES, SANDOVAL, ZALD´IVAR
Theorem 2.22. Let A be an idiomatic-quantale compactly generated and Λ a sub-
idiom of F (A). Then the associated indexing morphism
is injective. Hence Λ is spatial.
d : Λ → O pt(Λ)
Proof. We show that d(a) ⊆ d(b) implies a ≤ b for a, b ∈ Λ. Suppose a 6≤ b
for some a, b ∈ Λ. Since A is compactly generated there exists m ∈ A such that
m ≤ a and m 6≤ b. Then b ∈ X (m) and hence, by Proposition 2.19 we have
b ≤ p fore some maximal element p in X (m). Note that p /∈ d(b) and a 6≤ p since
m 6≤ p. Thus p /∈ d(b) and p ∈ d(a). This gives d(a) 6⊆ d(b).
(cid:3)
Remark 2.23. In [Sima, Section 2] a more general situation is given.
2.2. Module theoretic preliminaries. As it was mentioned in the Introduction,
we want to translate some notions of rings to the module context. In order to do
this, we will work in the category σ[M ] where M is an R-module. The category
σ[M ] is the full subcategory of R -Mod consisting of all modules that can be em-
bedded in a M -generated module. This category is more general than R -Mod in
the sense that if M = R then σ[M ] = R -Mod. It can be seen that σ[M ] is a cate-
gory of Grothendieck [Wis91]. Many ring-theoretic aspects have been translated to
modules in this category, see for example [CR12b], [CR12a], [CR14], [CMRZ16],
[Wis91], [Wis96], etc.
In [BJKN80] is defined a product of modules as follows:
Definition 2.24. Let M and K be R-modules. Let N ≤ M . The product of N
with K is defined as:
NM K = X{f (N ) f ∈ Hom(M, K)}
This product generalizes the usual product of an ideal and an R-module. For
properties of this product see [CR12b, Proposition 1.3]. In particular we have a
product of submodules of a given module.
Given a submodule N of a module M , we will denote the least fully invariant
submodule of M containing N by N . This submodule can be described as
N = NM M.
Since we have a product, it is natural to ask for an annihilator. Next definition
was given in [Bea02]:
Definition 2.25. Let M and K be R-modules. The annihilator of K in M is
defined as:
AnnM (K) = T{Ker(f ) f ∈ Hom(M, K)}
This annihilator is a fully invariant submodule of M and it is the greatest sub-
module of M such that AnnM (K)M K = 0.
Now we present two lemmas that will be needed in what follows.
Lemma 2.26. Let M be projective in σ[M ] and {Ni i ∈ I} a family of modules
in σ[M ]. Then T AnnM (Ni) = AnnM (P Ni).
ATTACHING TOPOLOGICAL SPACES TO A MODULE
9
Now, let f : M → N be a non zero morphism and consider the canonical epi-
Proof. Denote N = PI Ni. Since Ni ≤ N then AnnM (N ) ≤ TI AnnM (Ni).
morphism ρ : LI Ni → N . So, since M is projective in σ[M ] there exists
g : M → LI Ni such that ρg = f .
M
①
①
①
g
①
f
/ N
{①
LI Ni ρ
are the canonical projections, hence g(x) = 0. Thus f (x) = ρ(g(x)) = 0. This
(cid:3)
Let x ∈ TI AnnM (Ni), then πi(g(x)) = 0 for all i ∈ I, where πi : LI Ni → Ni
implies TI AnnM (Ni) ≤ AnnM (N ).
If N is a submodule of M , then
Lemma 2.27. Let M be projective in σ[M ].
AnnM (N ) = AnnM (N ).
Proof. Let N be a submodule of M . We always have AnnM (N ) ≤ AnnM (N ).
On the other hand, using the associativity of the product in M , we have
AnnM (N )M N = AnnM (N )M NM M = 0.
(cid:3)
Other concepts related to the product that come up are primeness and semiprime-
ness.
Definition 2.28 (Definition 13 [RRR+05]). Let N ≤ M be a proper fully invariant
submodule. N is prime in M if whenever LM K ≤ N with L, K ≤f i M then
K ≤ N or L ≤ N . We say that M is a prime module if 0 is prime in M .
Definition 2.29 ([RRR+09]). Let N ≤ M be a proper fully invariant submodule.
N is semiprime in M if whenever LM L ≤ N with L ≤f i M then L ≤ N . We say
that M is a semiprime module if 0 is semiprime in M .
The product of submodules is neither associative nor distributes sums from the
right, in general. If we assume that M is projective in σ[M ] then the product is
associative and distributive over sums [Bea02, Proposition 5.6] and [CMR, Lemma
1.1]. Moreover if N, L ≤f i M then NM L ≤f i M . See [MSZ16, Remark 4.2].
Now, we give some properties of prime and semiprime submodules:
Proposition 2.30. Let M be projective in σ[M ] and P a fully invariant submodule
of M . The following conditions are equivalent:
(1) P is prime in M .
(2) For any submodules K, L of M such that KM L ≤ P , then K ≤ P or L ≤ P .
(3) For any submodules K, L of M containing P and such that KM L ≤ P , then
K = P or L = P .
(4) M/P is a prime module.
Proof. It follows from [CR12b, Proposition 1.11], [Bea02, Proposition 5.5] and
[RRR+05, Proposition 18].
(cid:3)
Proposition 2.31. Let M be projective in σ[M ] and N a fully invariant submodule
of M . The following conditions are equivalent:
{
/
10
MEDINA, MORALES, SANDOVAL, ZALD´IVAR
(1) N is semiprime in M .
(2) For any submodule K of M , KM K ≤ N implies K ≤ N .
(3) For any submodule K ≤ M containing N such that KM K ≤ N , then K =
N .
(4) M/N is a semiprime module.
(5) If m ∈ M is such that RmM Rm ≤ N , then m ∈ N .
(6) N is an intersection of prime submodules.
Proof. (1) ⇔ (2) ⇔ (3) ⇔ (4) It is analogous to Proposition 2.30.
(1) ⇔ 5 ⇔ (6) See [CMRZ16, Proposition 1.11].
(cid:3)
3. THE FRAME OF SEMIPRIMITIVE SUBMODULES.
In this section, we define primitive submodules and semiprimitive submodules.
We wil prove that the set of all semiprimitive submodules and M is a spatial frame,
furthermore , we will see that this frame is isomorphic to the frame of open set of
the topological space Max(M ). Since we have this frame, we also ask us how is
given its point space, and then we give sufficient conditions for its point space to
be precisely the primitive submodules.
Definition 3.1. Let M be an R-module and N < M . It is said that N is a primitive
submodule if N = AnnM (S) for some S ∈ σ[M ] a simple module. We say that
M is a primitive module if 0 is a primitive submodule.
Denote Prt(M ) = {P < M P is primitive }.
Remark 3.2. If 0 6= M then there are simple modules in σ[M ]. Consider Rm ≤
M with 0 6= m, since Rm is cyclic then it has maximal submodules, say M ≤
Rm, then Rm/M ∈ σ[M ] is simple.
Proposition 3.3. Let M be projective in σ[M ]. If P < M is a primitive submodule
then P is prime in M .
Proof. Let N, L be fully invariant submodules of M such that NM L ≤ P . Since
P is a primitive submodule of M there exists S ∈ σ[M ] simple module such that
P = AnnM (S), then (NM L)M S = 0. Furthermore, NM (LM S) = 0 because M
is projective in σ[M ]. Since LM S ≤ S, we have that LM S = 0 or LM S = S. If
LM S = 0 then L ≤ P . On the other hand, if LM S = S then 0 = NM (LM S) =
NM S. Thus N ≤ P .
(cid:3)
Remark 3.4. It is possible that M has no primitive submodules. Although, if M
has a maximal submodule M, then P = AnnM (M/M) is a primitive submodule.
Definition 3.5. Let M be an R-module.
proper submodule is contained in a maximal submodule.
It is said that M is coatomic if every
Example 3.6. The following are examples of coatomic modules:
(a) Finitely generated modules,
(b) semisimple modules,
(c) semiperfect modules,
(d) multiplication modules over a commutative ring,
(e) modules over a left perfect ring,
ATTACHING TOPOLOGICAL SPACES TO A MODULE
11
We will denote by Max(M ) and Maxf i(M ) the maximal elements in the lattices
Λ(M ) and Λf i(M ) respectively.
Lemma 3.7. Let M be coatomic and projective in σ[M ]. If N ∈ Maxf i(M ) then
N is a primitive submodule.
Proof. Since M is coatomic, there exists a maximal submodule M such that N ≤
M. Since M is projective in σ[M ] and N is fully invariant, then NM (M/M) = 0
by [CR12b, Proposition 1.3 and Proposition 1.8]. Thus N ≤ AnnM (M/M), but
N ∈ Maxf i(M ), therefore N = AnnM (M/M).
(cid:3)
Remark 3.8. The converse of the last Lemma is not true in general. For instance,
consider the following examples:
(1) The ring R constructed by Bergman and described in [CH80, pp. 27]. It can
be seen that R is a primitive ring and has a unique non zero two-sided ideal U .
Therefore, 0 is a primitive ideal but it is not a maximal ideal.
(2) Let K be a field and V a K−vector space with dimK(V ) = ℵ0 and let
R := EndK (V ). Notice that V is an R−module, AnnR(V ) = 0 is not a
maximal ideal in R and I = {f ∈ R dimK (Im(f )) is finite} is the only
maximal ideal of R. Let f : V → V such that dimK (Im(f )) = 1. Thus, Rf
is a projective simple R−module and AnnR(Rf ) = 0 is not maximal. See
[Lam03, Ex. 3.15, Ex. 4.8].
Now, we are interested in a particular class of submodules defined by the primi-
tive submodules. We will prove that this class can be seen as the fixed points of a
suitable operator on Λf i(M ).
Definition 3.9. A submodule N ≤ M is called a semiprimitive submodule if N is
an intersection of primitive submodules.
Let M be an R-module and Max(M ) = {M < M M is maximal }. If M is
projective in σ[M ], then Max(M ) is a topological space with open sets {m(N )
N ≤f i M } where m(N ) = {M ∈ Max(M ) N (cid:2) M}. See [MSZ16].
If M is coatomic, then Max(M ) 6= ∅. Hence, we have an adjunction
Λf i(M )
m
m∗
. O(Max(M ))
This adjuntion defines a multiplicative nucleus τ := m∗ ◦ m on Λf i(M ) (see
[MSZ16, Theorem 3.21]).
Remark 3.10. Given N ≤f i M , by [MSZ16, Proposition 3.20] τ (N ) is the largest
fully invariant submodule contained in T{M ∈ Max(M )N ≤ M}.
Consider Λf i(M )µ = {N ∈ Λf i(M ) τ (N ) = N }.
Proposition 3.11. Let M be an R-module, then Prt(M ) ⊆ Λf i(M )µ.
Proof. If P ∈ Prt(M ), then P = AnnM (S) = T{Ker(f ) f ∈ Hom(M, S)},
with S a simple module. Thus, we can take only the nonzero morphisms in
Hom(M, S), hence P is an intersection of maximal submodules, therefore
τ (P ) ≤ \{M ∈ Max(M ) P ≤ M} ≤ P.
Since always P ≤ τ (P ), then P = τ (P ). Hence, Prt(M ) ⊆ Λf i(M )µ.
(cid:3)
.
m
m
12
MEDINA, MORALES, SANDOVAL, ZALD´IVAR
Proposition 3.12. Let M be projective in σ[M ] and K be a proper fully invariant
submodule of M . Then
Proof. Let M ∈ Max(M ) such that K ≤ M. Since AnnM (M/M) ≤ M, then
\{AnnM (M/M) K ≤ M ∈ Max(M )} = \{M ∈ Max(M ) K ≤ M}.
\{AnnM (M/M) K ≤ M ∈ Max(M )} ≤ \{M ∈ Max(M ) K ≤ M}.
Now, let f ∈ Hom(M, M/M) be nonzero and A = ker(f ). Notice that A
is a maximal submodule of M . Since M
A ) = 0. Therefore
A
K ≤ AnnM (M/A) ≤ A. This implies that AnnM (M/M) is an intersection of
maximal submodules containing K. Thus
M , then KM ( M
∼= M
\{M ∈ Max(M ) K ≤ M} ≤ \{AnnM (MM) K ≤ M ∈ Max(M )}.
(cid:3)
Theorem 3.13. Let M be projective in σ[M ]. Then
Λf i(M )µ = {N ∈ Λf i(M ) N is a semiprimitive submodule } ∪ {M }
is an spatial frame. Moreover,
Λf i(M )µ ∼= O(Max(M ))
Proof. Let N be a semiprimitive submodule, then there exist {Pi}i∈I ⊆ Λf i(M )µ
M(cid:1) K ≤ M ∈ Max(M )}, for every K ∈ SP m(M ).
and K = T{AnnM (cid:0) M
such that N = \{Pi i ∈ I}. Since every primitive submodule is an intersection
N ≤ τ (N ) ≤ \{M ∈ Max(M ) N ≤ M} ≤ \I
of maximal submodules containing N , then
{Pi Pi ∈ Prt(M )} = N.
Thus τ (N ) = N . Clearly, M ∈ SP m(M ).
Now let K be a proper submodule of M such that K = τ (K). Since K is fully
M ) = 0, and K ≤ AnnM (M/M) ≤ M for every maximal M
invariant, KM ( M
containing K. By Proposition 3.12 we have that
M ) is a primitive submodule and is fully invariant. Since K =
K ≤ \{AnnM (M/M) K ≤ M∈ Max(M )} = \{M∈ Max(M ) K ≤ M}.
Each AnnM ( M
τ (K) is the largest fully invariant submodule contained in T{M ∈ Max(M )
K ≤ M}, then K = T{AnnM (cid:0) M
struction Λf i(M )µ ∼= O(Max(M )).
By [MSZ16, Corolary 3.11], it follows that Λf i(M )µ is a frame and by con-
M(cid:1) K ≤ M ∈ Max(M )}.
Remark 3.14. Inasmuch as Theorem 3.13, for the remainder of the text, we will
denote by SP m(M ) the spatial frame of all semiprimitive submodules Λf i(M )µ
for a projective module M in σ[M ].
Proposition 3.15. Let M be projective in σ[M ]. Then
Prt(M ) ⊆ pt(SP m(M )) ⊆ Spec(M ).
(cid:3)
ATTACHING TOPOLOGICAL SPACES TO A MODULE
13
Proof. Since KM L ≤ K ∩ L for all K, L ∈ Λf i(M ), every prime submodule is
∧-irreducible. Then Prt(M ) ⊆ pt(SP m(M )).
Let Q ∈ pt(SP m(M )) and N, L ∈ Λf i(M ) such that NM L ≤ Q. We can
apply the nucleus τ and we get
τ (N ) ∩ τ (L) = τ (N ∩ L) = τ (NM L) ≤ τ (Q) = Q
Since Q is a point in SP m(M ) then N ≤ τ (N ) ≤ Q or L ≤ τ (L) ≤ Q.
(cid:3)
After last Proposition is natural to ask when Prt(M ) = pt(SP m(M )).
In
general this equality does not hold, as the following example shows:
Example 3.16. Consider Z the ring of integers. Then Λ(Z) = Λf i(Z). Since Z is
commutative Prt(Z) = {pZ p is prime } = Max(Z).
Now, note that 0 = T{pZ p is prime } ∈ SP m(Z). Since Z is uniform then
0 is ∩−irreducible, that is, 0 ∈ pt(SP m(Z)) but it is not a primitive ideal. Thus
pt(SP m(Z)) 6= Prt(Z).
Denote by M -Simp a complete set of representatives of isomorphism classes of
simple modules in σ[M ]. Next propositions give sufficient conditions on a module
M in order to get the equality pt(SP m(M )) = Prt(M ).
Proposition 3.17. Let M be projective in σ[M ] such that M -Simp is finite. Then
pt(SP m(M )) = Prt(M ).
Proof. Let K ∈ pt(SP m(M )). By Theorem 3.13,
K = \ {AnnM (M/M) K ≤ M ∈ Max(M )} .
By hypothesis M -Simp is finite then this intersection is finite. Every AnnM (M/M)
is in SP m(M ) and K is a ∧-irreducible then, there exists M ∈ Max(M ) such
(cid:3)
that K = AnnM (cid:0) M
M(cid:1). Thus K is primitive.
Example 3.18. Examples of rings satisfying the last proposition are:
(a) Commutative artinian rings.
(b) R = (cid:18)Q 0
R R(cid:19)
(c) Rings of upper triangular matrices with coefficients in a field.
Now, recall that a ring R is called pm−ring if every prime ideal is contained
in a unique maximal ideal. In the study of Spec(R) and M ax(R) for a commu-
tative ring, pm−rings had taken an important role, for instance see [Sun92]. So,
as a generalization of the above concept, we introduce the following definition for
modules
Definition 3.19. Let M be an R-module. It is said M is a pm-module if every
prime submodule is contained in a unique maximal submodule.
Proposition 3.20. Let M be projective in σ[M ]. If M is a pm-module then
pt(SP m(M )) = Prt(M ) = Max(M ).
Proof. We have that a primitive submodule P is an intersection of maximal sub-
modules, so if M is a pm-module then P has to be maximal. On the other hand, if
14
MEDINA, MORALES, SANDOVAL, ZALD´IVAR
M ∈ Max(M ) the AnnM (M/M) is primitive submodule and it is contained in
M. Thus M = AnnM (M/M). Hence we have that Max(M ) = Prt(M ).
By Proposition 3.15 pt(SP m(M )) ⊆ Spec(M ). Now, let Q ∈ pt(SP m(M )),
then Q is semiprimitive, that is, Q is an intersection of primitive submodules, but
the primitive submodules are the maximal submodules. Since Q is prime in M ,
by hypothesis Q is contained in a unique maximal submodule thus Q is maximal.
Therefore pt(SP m(M )) ⊆ Max(M ) = Prt(M ) and by Proposition 3.15 we have
that Prt(M ) ⊆ pt(SP m(M )).
(cid:3)
Now we are going to show that the converse of Proposition 3.17 is not true in
general.
Example 3.21. (a) Let X be a discrete space. Let C(X) be the (commutative)
ring of all continuous functions from X to R with the pointwise operations.
By [GJ13, Theorem 2.11] the ring C(X) is a pm-module (over itself). Thus
by Proposition 3.20 Max(C(X)) = Prt(C(X)) = pt(SP m(C(X))). The
maximal ideals of C(X) can be described as: M ≤ C(X) is a maximal ideal
if and only if M = hfxi for some x ∈ X, where fx is the function defined as
fx(y) = 0 if x = y
fx(y) = 1 if x 6= y
Therefore, the ring C(X) has as many maximal ideals as points in X. Thus, if
X in not finite, C(X)-Mod has infinitely many non isomorphic simple mod-
ules and Prt(C(X)) = pt(SP m(C(X))).
(b) Consider the boolean ring R := Zℵ0
2
and note that M ∈ M ax(R) if and
only if M is prime. Also, their factors are not isomorphic R−modules and
R − simp = ℵ0.
4. SOBRIETY
In this section we will study the sobriety of the spaces Spec(M ) and Max(M ),
where M is an R-module projective in σ[M ]. We will see that in this case, Spec(M )
is sober. On the other hand, in general Max(M ) is not sober and then we will give
sufficient and necessary conditions for being that.
For properties of sober spaces and theory related to this topic we refer to the
reader to [Sim06], [Joh86] and [PP12].
Given S a topological space and X any subset of S, we will denote X − and X ′
closure of X and the complement of X in S, respectively.
Definition 4.1. A nonempty subset X of a topological space S is closed irreducible
if it is closed, and satisfies
(X ∩ U ) and (X ∩ V ) are not empty ⇒ X ∩ (U ∩ V ) is not empty
for all U, V ∈ O(S).
Definition 4.2. A topological space S is sober if each closed irreducible subset
X has a unique generic point, that is, there exists a unique point s ∈ X such that
X = s−.
ATTACHING TOPOLOGICAL SPACES TO A MODULE
15
The full subcategory of sober spaces is reflective in the category of topological
spaces, therefore for every space S we can associate a sober space sob(S) and a
continuous morphism ζ : S → sob(S), this assignation is called the soberfication
of S (see [Sim06] ).
The space sob(S) consists of all closed irreducible subsets of S and its topology
is given by the family of sets:
♮U = {X ∈ sob(S) X ∩ U 6= ∅}
with U ∈ O(S); and the continuous map is given by ζ(s) = s−.
As examples of sober spaces we have the following two results.
Theorem 4.3. For each frame A the point space pt(A) is sober.
Proof. See [Sim06, Theorem 3.6].
(cid:3)
Remark 4.4. Using this Theorem, it can be seen that for every topological space
S, the assignation (−)′ : pt(O(S)) → sob(S), such that U 7→ U ′, is a homeomor-
phism (see [Sim06]).
Proposition 4.5. Let M be projective in σ[M ]. Then Spec(M ) is sober.
Proof. Spec(M ) is a topological space with open sets {U (N ) N ∈ Λf i(M )}
where U (N ) = {P ∈ Spec(M ) N (cid:2) P }. So the hull-kernel adjuntion gives a
multiplicative nucleus µ on Λf i(M ) defined as
Hence by [MSZ16, Proposition 4.24] we have that the set of fixed points of µ is
µ(N ) = \{P N ≤ P }.
SP (M ) = {M } ∪ {N N is semiprime in M }.
By [MSZ16, Proposition 4.29] Spec(M ) = pt(SP (M )) ∼= pt(O(Spec(M )).
Hence, by Remark 4.4
sob(Spec(M )) = {U (P )′ P is prime in M } = {P − P ∈ Spec(M )}.
Thus, by [Sim06, Lemma 1.10], we conclude that Spec(M ) is sober.
(cid:3)
An immediately consequence of the above is:
Corollary 4.6. For every ring R, the spectrum Spec(R) is a sober space.
Proposition 4.7. Let D be a principal ideal domain. Then sob(Max(D)) ∼=
Spec(D) as topological spaces. That is, the soberfication of Max(D) is Spec(D).
Proof. Inasmuch as D is a PID, every non zero prime ideal is maximal and the
primitive ideals are the maximal ideals, so Prt(D) = Max(D).
Since D is a domain, pt(SP m(D)) = {0} ∪ {M < D M is maximal }.
Under the isomorphism SP m(D) ∼= O(Max(D)) we have that
pt(O(Max(D))) = {∅} ∪ {U (M) M is maximal }.
Thus, using the fact that if M is maximal then it is closed, we have that
sob(Max(D)) = {Max(D)} ∪ {M− M is maximal}
= {Max(D)} ∪ {{M} M is maximal}.
16
MEDINA, MORALES, SANDOVAL, ZALD´IVAR
Notice that if U ∈ O(Max(D)) then
♮U = {X ∈ sob(Max(D)) X ∩ U 6= ∅} =
{Max(D)} ∪ {{M} ∈ sob(Max(D)) M ∈ U }.
Because of the open sets of Max(D) are U (I) = {M I * M} for some I ≤ D,
it follows that ♮(U (I)) = {Max(D)} ∪ U (I).
Applying the universal property of sob(Max(D)) ([Sim06, Theorem 3.9]), we
get the following commutative diagram
Max(D)
Spec(D)
7♦
ζ
♦
♦
♦
∃!ϕ#
♦
♦
sob(Max(D))
where ϕ#(M) = M and ϕ#(Max(D)) = 0. Recall that a non trivial open set in
Spec(D) has the form:
{P ∈ Spec(D) I (cid:2) P } = {0} ∪ {M ∈ Max(D) I (cid:2) M}.
(cid:3)
From Proposition 4.7, we can see that the space Max(M ) is not sober in general.
Next Theorem gives necessary and sufficient conditions for Max(M ) to be sober.
Lemma 4.8. Let M be projective in σ[M ]. Then M− = m(AnnM (M/M))′ for
any M ∈ Max(M ).
Proof. It is clear M ∈ m(AnnM (M/M))′. On the other hand, let m(K)′ a
closed set in Max(M ) containing M.Then K ⊆ M. Since M is projective in
σ[M ], KM M/M = 0. Hence K ⊆ AnnM (M/M). Thus m(AnnM (M/M))′ ⊆
m(K)′.
(cid:3)
Definition 4.9. Let M be an R-module. It is said M is quasi-duo if Max(M ) =
Maxf i(M ). A ring R is called left quasi-duo if as left R-module is quasi-duo.
Examples of left quasi-duo rings are the upper triangular matrices of n by n with
coefficients in a field (See [Yu95, Proposition 2.1]).
Lemma 4.10. Let M be projective in σ[M ] such that Max(M ) 6= ∅. The following
conditions are equivalent:
(1) M is a quasi-duo module.
(2) Max(M ) is T1
(3) Max(M ) is T0
Proof. (1)⇒(2) Let M ∈ Max(M ). Since M is fully invariant, M = m(M)′.
(2)⇒(3) It is clear.
(3)⇒(1) Let M ∈ Max(M ). Let 0 6= f : M → M/M. Since M/ ker(f ) ∼=
M/M then M− = m(AnnM (M/M))′ = m(AnnM (M/ ker(f )))′ = ker(f )−.
But Max(M ) is T0, thus ker(f ) = M. This implies that AnnM (M/M) = M.
(cid:3)
Theorem 4.11. Let M be projective in σ[M ]. The following conditions hold:
(1) If Max(M ) is sober then pt(SP m(M )) = Prt(M ).
/
/
7
ATTACHING TOPOLOGICAL SPACES TO A MODULE
17
(2) If M is a quasi-duo module and pt(SP m(M )) = Prt(M ) then Max(M ) is
sober.
Proof. Firstly, by Remark 4.4 it follows that
sob(Max(M )) = {m(K)′ K ∈ pt(SP m(M ))}.
(1) Assume that Max(M ) is sober and let m(K)′ ∈ sob(Max(M )). Then by
Lemma 4.8, m(K)′ = M− = m(AnnM (M/M))′ for some M ∈ Max(M ).
Therefore m(K) = m(AnnM (M/M)).
Since m : SP m(M ) → O(Max(M )) is a bijection it follows that K =
AnnM (M/M), that is K is a primitive submodule.
(2) By Lemma 4.10 Max(M ) is T0, hence it is enough to show that every
closed irreducible is a point closure. Assume that K = AnnM (S) with S a
simple module in σ[M ]. Let 0 6= f : M → S. Then ker(f ) ∈ Max(M ) and
K = AnnM (M/ ker(f )). Thus m(K)′ = ker(f )−.
(cid:3)
Corollary 4.12. If R is a left quasi-duo ring such that R-Simp is finite then Max(R)
is sober. In particular, Max(R) is sober for every commutative artinian ring R.
Proof. It follows by Proposition 3.17.
Corollary 4.13. If R is a pm-ring then Max(R) is a sober space.
Proof. If follows by Proposition 3.20.
(cid:3)
(cid:3)
5. THE FRAME Ψ(M )
In this section, we introduce Ψ(M ) which is a frame given by condition on
annihilators. In fact, Ψ(M ) turns out to be a spatial frame. Firstly, we give an
explicit definition of this frame and after that we get it as the set of fixed point of
an operator in Λf i(M ). Also, we research about the case when Ψ(M ) = Λf i(M )
and we characterize the modules with this property.
Also on the definition of Ψ(M ) arise the regular core for an idiomatic-quantale.
So, taking this in account, we give sufficient condition to get that the regular core
of Λf i(M ) concides with Ψ(M ).
Definition 5.1. For a module M , set
Ψ(M ) := {N ∈ Λf i(M ) ∀n ∈ N, [N + AnnM (Rn) = M ]}.
Definition 5.2. We say that a module M is self-progenerator in σ[M ] if M is
projective in σ[M ] and generates all its submodules.
Proposition 5.3. Let M be self-progenerator in σ[M ]. Then for every N ∈ Ψ(M )
we have that N 2 = N.
Proof. First, for each n ∈ N we have that
Rn=MM Rn = [N +AnnM (Rn)]M Rn = NM Rn+AnnM (Rn)M Rn = NM Rn.
Then, by [CMR, Lemma 2.1],
N 2 = NM N = NM P Rn = P(NM Rn) = P Rn = N.
(cid:3)
If N, L ∈ Ψ(M ) then
Proposition 5.4. Let M be self-progenerator in σ[M ].
N ∩L ∈ Ψ(M ). Moreover, if N ∈ Ψ(M ) and K ∈ Λf i(M ) then K ∩N = NM K.
18
MEDINA, MORALES, SANDOVAL, ZALD´IVAR
Proof. Let N, L ∈ Ψ(M ) and n ∈ N ∩ L, then
M = MM M = [N + AnnM (Rn)]M [L + AnnM (Rn)]
NM L + NM AnnM (Rn) + AnnM (Rn)2 + AnnM (Rn)M L
≤ NM L + AnnM (Rn) ≤ N ∩ L + AnnM (Rn).
Therefore N ∩ L ∈ Ψ(M ).
Now let N ∈ Ψ(M ) and K ∈ Λf i(M ). By Proposition 5.3, if n ∈ K ∩ N then
(cid:3)
Rn = NM Rn ≤ NM K ≤ N ∩ K, thus N ∩ K = NM K.
Proposition 5.5. Let M be projective in σ[M ]. If {Nα α ∈ I} ⊆ Ψ(M ) then
P{Nα α ∈ I} ∈ Ψ(M )
Proof. Let N = PI Nα and x ∈ N . Then x = nα1 + · · · nαk with nαi ∈
Nαi. By hypothesis Nαi + AnnM (Rnαi) = M for all 1 ≤ i ≤ k. Hence N +
AnnM (Rnαi) = M . Therefore, taking the product of M k-times:
M = MM · · · M M = (N + AnnM (Rnα1))M · · · M (N + AnnM (Rnαk ))
= N k + · · · + (AnnM (Rnα1)M · · · M AnnM (Rnαk ))
≤ N + (AnnM (Rnα1)M · · · M AnnM (Rnαk ))
≤ N + (AnnM (Rnα1)\ · · ·\ AnnM (Rnαk )).
On the other hand, since Rx ≤ Rnα1 + · · · + Rnαk it follows that
AnnM (Rnα1 + · · · + Rnαk ) ≤ AnnM (Rx).
Thus, by Lemma 2.26 M ≤ N + (AnnM (Rnα1)T · · ·T AnnM (Rnαk )) ≤ N +
AnnM (Rx).
(cid:3)
From Propositions 5.4 and 5.5 we have the following:
Theorem 5.6. Let M be self-progenerator in σ[M ]. Then Ψ(M ) is a spatial frame.
Proof. From Proposition 5.4, Ψ(M ) is a distributive lattice. Therefore by Lemma
2.3 it is a frame. Thus Ψ(M ) is an spatial frame by Theorem 2.22.
(cid:3)
We can produce Ψ(M ) from another perspective. We can recover Ψ(M ) as the
fixed point of a suitable operator in Λf i(M ). In this case such operator is not an
inflator but a deflator.
Definition 5.7. For each N ∈ Λf i(M ), define
Ler(N ) = {m ∈ M N + AnnM (Rm) = M }.
Proposition 5.8. Let M be projective in σ[M ]. For each N ∈ Λf i(M ), Ler(N )
is a submodule of M .
Proof. First, let l, k ∈ Ler(N ). Then N + AnnM (Rl) = M = N + AnnM (Rk)
and thus M = MM M = (N + AnnM (Rl))M (N + AnnM (Rk)). Using the fact
that this product is distributive (because M is projective), it follows that
M = NM N + AnnM (Rl)M N + NM AnnM (Rk) + AnnM (Rl)M AnnM (Rk).
Hence M ⊆ N + AnnM (R(l + k)). Therefore, l + k ∈ Ler(N ).
ATTACHING TOPOLOGICAL SPACES TO A MODULE
19
Now, let r ∈ R and n ∈ Ler(N ). Since n ∈ Ler(N ), we have that N +
AnnM (Rn) = M. Inasmuch as R(rn) ⊆ Rn it follows that AnnM (Rn) ⊆
AnnM (R(rn)). Hence, M = N + AnnM (Rn) ⊆ N + AnnM (R(rn)), Thus
M = N + AnnM (R(rn)), that is, rn ∈ Ler(N ).
(cid:3)
Lemma 5.9. Let M be self-progenerator in σ[M ] and N ∈ Λf i(M ). Then
Ler(N ) ≤ N .
Proof. Let N ∈ Λf i(M ) and m ∈ Ler(N ). Then M = N + AnnM (Rm).
Therefore, Rm = MM Rm = (N + AnnM (Rm))M Rm = NM Rm ≤ N.
(cid:3)
Proposition 5.10. Let M be projective in σ[M ] and N1, ..., Nn ∈ Λf i(M ). Then
n
Ler(
\i=1
Ni) =
n
\i=1
Ler(Ni).
Proof. Let x ∈ Tn
i=1 Ler(Ni), hence M = Ni + AnnM (Rx) for all 1 ≤ i ≤ n.
M = MM M = (Ni + AnnM (Rx))M (Nj + AnnM (Rx))
≤ NiM Nj + AnnM (Rx) ≤ Ni ∩ Nj + AnnM (Rx)
Thus, x ∈ Ler(Ni ∩ Nj). Therefore Tn
We always have, Ler(Tn
i=1 Ni) ≤ Tn
i=1 Ler(Ni) ≤ Ler(Tn
i=1 Ler(Ni).
Proposition 5.11. Let M be self-progenerator in σ[M ]. For each N ∈ Λf i(M ),
Ler(N ) = N if and only if N ∈ Ψ(M ).
i=1 Ni).
(cid:3)
Proof. It follows from Definiton 5.1, Definiton 5.7 and Lemma 5.9
(cid:3)
Now we are going to characterize the extreme case when Ψ(M ) = Λf i(M ).
Definition 5.12. [Goo91, page 89] Let R be a ring. It is said that R is biregular if
every cyclic two sided ideal is generated by a central idempotent.
Proposition 5.13. A ring R is biregular if and only if R = RaR + AnnR(RaR)
for all a ∈ R.
Proof. Let a ∈ R and suppose R = RaR + AnnR(RaR). Then 1 = sar + t with
r, s ∈ R and t ∈ AnnR(RaR). This implies that
sar = (sar)sar + (sar)t = (sar)2
so, sar and t are idempotents. Now, let k ∈ R. We have that
0 = (sar)kt = (1 − t)kt = kt − tkt
hence kt = tkt. In the same way tk = tkt, thus t is central. This implies that sar
is central. Notice that a = asar, hence RaR = RsarR. Thus R is biregular.
Reciprocally, if a ∈ R there exists a central idempotent e ∈ R such that
RaR = ReR and R = ReR + R(1 − e)R. Since e(1 − e) = 0 then R(1 − e) ≤
AnnR(RaR). Thus R = RaR + AnnR(RaR).
(cid:3)
In view of the above result, we introduce the concept of birregular module as
follows.
20
MEDINA, MORALES, SANDOVAL, ZALD´IVAR
Definition 5.14. Let M be an R-module. We say that M is birregular if
M = Rm + AnnM (Rm),
for every m ∈ M.
Proposition 5.15. Let M be projective in σ[M ]. M is a birregular module if and
only if Ψ(M ) = Λf i(M ).
Proof. First, suppose that M is a birregular module. Let N ∈ Λf i(M ) and n ∈ N.
By Lemma 2.27
M = Rn + AnnM (Rn) = Rn + AnnM (Rn) ⊆ N + AnnM (Rn).
So N ∈ Ψ(M ) and thereby Ψ(M ) = Λf i(M ).
Conversely, assume that Ψ(M ) = Λf i(M ). Then, for each m ∈ M we have
that Rm ∈ Ψ(M ) = Λf i(M ), consequently by Lemma 2.27, M = Rm +
AnnM (Rm). Hence, M is birregular.
(cid:3)
In what follows, we want to study a property of regularity in the frame Ψ(M ).
In order to do that, we recall some definitions and facts concerning to the regu-
larity property. The background of this topics can see for example in [Joh86] and
[Sim89].
Definition 5.16. Let Ω be a frame and x, a ∈ Ω.
It is said x is rather below
a, denoted by x 0 a, if a ∨ ¬x = 1. Denote Ω¬ the set of all a ∈ Ω with
a = W{x x 0 a}.
Remark 5.17. It is well known that Ω¬ is a subframe of Ω.
Definition 5.18. It is said that a frame Ω is regular if Ω¬ = Ω.
In general for a frame the subframe Ω¬ not need to be regular there is an en-
hancement to this situation. We can associate to any idiomatic-quantale A, a reg-
ular frame Areg called the regular core of A [Sim89]. The construction of this
regular frame is as follows:
Denote by Ar the set of fixed points of the operator r : A → A, given by r(a) =
W{x ∈ A a ∨ xr = 1} where xr = W{y yx = 0}. Now, Ar has his own r
deflator hence we can consider Ar(2) := (Ar)r. Inductively, it is defined:
Ar(0) := A Ar(α+1) := Ar(α)r Ar(λ) := \{Ar(α) α < λ}
for each non-limit ordinal α and limit ordinal λ respectively. This chain is decreas-
ing, therefore by a cardinality argument it eventually stabilizes in some ordinal.
Let us denote the least of those ordinals by ∞ and Areg := Ar(∞).
In [Sim89, Theorem 3.4] it is proved that Areg is a regular frame and every
regular subframe of A is contained in it.
Proposition 5.19. Let M be projective in σ[M ] and denote A = Λf i(M ). Then
the deflator r : A → A is Ler and Ar = Ψ(M ).
Proof. Denote A = Λf i(M ) and let N ∈ A. So, r(N ) = P {K∈A N +K r=M }
where K r = P{L ∈ A LM K = 0} = AnnM (K). Hence,
r(N ) = X{K ∈ A N + AnnM (K) = M }.
ATTACHING TOPOLOGICAL SPACES TO A MODULE
21
By definition, Ler(N ) = {m ∈ M N + AnnM (Rm) = M }. Then
Ler(N ) ⊆ X{K ∈ A N + AnnM (K) = M }.
On then other hand, if K ∈ A is such that M = N + AnnM (K) then for every
k ∈ K, N + AnnM (Rk) = M . Hence K ⊆ Ler(N ). Thus r(N ) = Ler(N ).
This implies that Ar = Ψ(M ) by Proposition 5.11.
(cid:3)
In what follows, we want to study the case when Ψ(M ) is the regular core of
Λf i(M ). At this moment, we just have sufficient conditions for this to happen.
Theorem 5.20. Let M be projective in σ[M ]. If Ler(AnnM (K)) = AnnM (K)
for all K ∈ Λf i(M ) then Ψ(M ) = (Λf i(M ))reg.
Proof. Denote A = Λf i(M ). Let us show that (Ar)r = Ar. Consider r : Ar →
Ar. For N ∈ Ar r(N ) = P{K ∈ Ar N + K r = Ler(M )} where K r =
P{L ∈ Ar LM K = 0.}
Notice that Ler(M ) = M . By hypothesis Ler(AnnM (K)) = AnnM (K) for
all K ∈ A. Hence AnnM (K) ∈ Ar for all K ∈ Ar. Thus
r(N ) = X{K ∈ Ar N + AnnM (K) = M }.
Let B ∈ A such that N + AnnM (B) = M . By Proposition 5.4,
BM AnnM (B) ⊆ B ∩ AnnM (B) = AnnM (B)M B = 0.
Hence, B ⊆ AnnM (AnnM (B)). So,
M = N + AnnM (B) ⊆ N + AnnM (AnnM (AnnM (B))).
By hypothesis, AnnM (AnnM (B)) ∈ Ar then,
N = X{B ∈ A N + AnnM (B) = M }
⊆ X{AnnM (AnnM (B)) ∈ Ar N + AnnM (AnnM (AnnM (B))) = M }
⊆ r(N ).
Thus, r(N ) = N and (Ar)r = Ar.
(cid:3)
Lemma 5.21. Let M be projective in σ[M ] and semiprime , if
with N ∈ Λf i(M ) then L ∈ Λf i(M )
M = N ⊕ L
Proof. Since NM L ⊆ N ∩ L = 0, then NM L = 0. Now M is semiprime then
NM L = 0 = LM N , thus LM M = LM N ⊕ L2 = L2 ≤ L then L = LM M . (cid:3)
Corollary 5.22. Let M be projective in σ[M ] and semiprime. If AnnM (N ) is a
direct summand of M for all N ∈ Λf i(M ) then Ψ(M ) = (Λf i(M ))reg.
Proof. Let N ∈ Λf i(M ) then M = AnnM (N ) ⊕ L by Lemma 5.21 we know that
L ∈ Λf i(M ) thus L ⊆ AnnM (AnnM (N )) then
AnnM (AnnM (N )) = AnnM (AnnM (N )) ∩ M =
AnnM (AnnM (N )) ∩ (AnnM (N ) ⊕ L) = L.
22
MEDINA, MORALES, SANDOVAL, ZALD´IVAR
Thus M = AnnM (N )⊕AnnM (AnnM (N )). Then AnnM (N ) ≤ Ler(AnnM (N )).
Thus AnnM (N ) = Ler(AnnM (N )). By Theorem 5.20, Ψ(M ) = (Λf i(M ))reg.
(cid:3)
Remark 5.23. In [RR04, Definition 3.2] are defined quasi-Baer modules. Given
N ∈ Λf i(M ), HomR(M, N ) is a two-sided ideal of EndR(M ). Then by [RR04,
Remark 3.3] in every quasi-Baer module M , AnnM (N ) is a direct summand for
all N ∈ Λf i(M ).
6. ACKNOWLEDGEMENTS
This investigation is the result of a two-semester research seminar about the
notes [Sima], the authors are thankful to Professor Harold Simmons who kindly
provide to us that set of notes and many other interesting subjects around this.
The authors are very thankful to the referee for his/her comments on this paper
that helped to improve it.
This work was supported by the grant UNAM-DGAPA-PAPIIT IN10057.
REFERENCES
[Alb14a]
Toma Albu, The osofsky-smith theorem for modular lattices, and applications (ii), Com-
munications in Algebra 42 (2014), no. 6, 2663 -- 2683.
[Alb14b]
, Topics in lattice theory with applications to rings, modules, and categories,
[Bea02]
Lecture Notes, XXIII Brazilian Algebra Meeting, Maring´a, Paran´a, Brasil, 2014.
J. Beachy, M-injective modules and prime M-ideals, Communications in Algebra 30
(2002), no. 10, 4649 -- 4676.
[BJKN80] L. Bican, P. Jambor, T. Kepka, and P. Nemec, Prime and coprime modules, Fundamenta
Matematicae 107 (1980), 33 -- 44.
[BSvdB84] Francis Borceux, Harold Simmons, and Gilberte van den Bossche, A sheaf representa-
tion for modules with applications to gelfand rings, Proceedings of the London Mathe-
matical Society 3 (1984), no. 2, 230 -- 246.
Francis Borceux, Gilberte VanDenBossche, et al., Algebra in a localic topos with appli-
cations to ring-theory, Lecture Notes in Mathematics 1038 (1983), R3.
[BV+83]
[BVDB83] Francis Borceux and Gilberte Van Den Bossche, Recovering a frame from its sheaves of
[CH80]
[CMR]
algebras, Journal of Pure and Applied Algebra 28 (1983), no. 2, 141 -- 154.
Arthur W Chatters and Charudatta R Hajarnavis, Rings with chain conditions, vol. 44,
Pitman Advanced Pub. Program, 1980.
Jaime Castro, Mauricio Medina, and Jos´e R´ıos, On the structure of goldie modules,
arXiv:1601.03444 [math.RA].
[CMRZ16] Jaime Castro, Mauricio Medina, Jos´e R´ıos, and Angel Zald´ıvar, On semiprime Goldie
[CR12a]
[CR12b]
[CR14]
[GJ13]
[Goo91]
[Joh86]
[Lam03]
[Lei14]
modules, Communications in Algebra 44 (2016), no. 11, 4749 -- 4768.
Jaime Castro and Jos´e R´ıos, FBN modules, Communications in Algebra 40 (2012),
no. 12, 4604 -- 4616.
, Prime submodules and local gabriel correspondence in σ[M ], Communica-
tions in Algebra 40 (2012), no. 1, 213 -- 232.
, Krull dimension and classical krull dimension of modules, Communications in
Algebra 42 (2014), no. 7, 3183 -- 3204.
Leonard Gillman and Meyer Jerison, Rings of continuous functions, Springer Science &
Business Media, 2013.
Kenneth R Goodearl, Von neumann regular rings, Krieger Pub Co, 1991.
Peter T Johnstone, Stone spaces.
T.Y. Lam, Exercises in classical ring theory, Springer-Verlag, 2003.
Tom Leinster, Basic category theory, vol. 143, Cambridge University Press, 2014.
ATTACHING TOPOLOGICAL SPACES TO A MODULE
23
[PP12]
[Ros90]
[MSZ16] Mauricio Medina, Lizbeth Sandoval, and Angel Zald´ıvar, A generalization of quantales
with applications to modules and rings, Journal of pure and applied algbra 220 (2016),
no. 5, 1837 -- 1857.
J Picado and A Pultra, Frames and locales, Topology without points (2012).
Kimmo I Rosenthal, Quantales and their applications, vol. 234, Longman Scientific and
Technical, 1990.
S Tariq Rizvi and Cosmin S Roman, Baer and quasi-baer modules, Communications in
Algebra 32 (2004), no. 1, 103 -- 123.
[RR04]
[RRR+05] Francisco Raggi, Jos´e R´ıos, Hugo Rinc´on, Rogelio Fern´andez-Alonso, and Carlos Sig-
noret, Prime and irreducible preradicals, Journal of Algebra and its Applications 4
(2005), no. 04, 451 -- 466.
[RRR+09]
[Sima]
[Simb]
[Simc]
[Sim85]
, Semiprime preradicals, Communications in Algebra 37 (2009), no. 8, 2811 --
2822.
H Simmons, Some methods of attaching a topological space to a ring.
, Two sided multiplicative lattices and ring radicals, preprint.
Harold Simmons, A lattice theoretic analysis of a result due to hopkins and levitzki.
, Sheaf representations of strongly harmonic rings, Proceedings of the Royal
Society of Edinburgh: Section A Mathematics 99 (1985), no. 3-4, 249 -- 268.
[Sim89]
, Compact representations -- the lattice theory of compact ringed spaces, Journal
[Sim01]
[Sim06]
of Algebra 126 (1989), no. 2, 493 -- 531.
, The point-free approach to sheafification, 2001.
H Simmons, The point space of a frame, http://www.cs.man.ac.uk/hsimmons/FRAMES/C-POINTSPACE.pdf,
2006.
[Sim14a]
,
Cantor-bendixson,
socle,
and
atomicity,
http://www.cs.man.ac.uk/hsimmons/00-IDSandMODS/002-Atom.pdf,
2 2014.
[Sim14b]
, An introduction to idioms, http://www.cs.man.ac.uk/hsimmons/00-IDSandMODS/001-Idioms.pdf,
[Sun92]
[Wis91]
2014.
Shu-Hao Sun, Rings in which every prime ideal is contained in a unique maximal right
ideal, Journal of Pure and Applied Algebra 78 (1992), no. 2, 183 -- 194.
Robert Wisbauer, Foundations of module and ring theory, vol. 3, Reading: Gordon and
Breach, 1991.
[Wis96]
, Modules and algebras: Bimodule structure on group actions and algebras,
[Yu95]
vol. 81, CRC Press, 1996.
Hua-Ping Yu, On quasi-duo rings, Glasgow Mathematical Journal 37 (1995), no. 1,
21 -- 31.
†.
MAURICIO MEDINA B ´ARCENAS
Instituto de Matem´aticas, Universidad Nacional Aut´onoma de M´exico, ´area de Investigaci´on Cient´ıfica,
Circuito Exterior, C.U., 04510, M´exico, D.F., M´exico.
∗, LORENAMORALES CALLEJAS
⋆, ANGEL ZALD´IVAR CORICHI
MARTHA LIZBETH SHAID SANDOVAL MIRANDA
Facultad de Ciencias, Universidad Nacional Aut´onoma de M´exico.
Circuito Exterior, Ciudad Universitaria, 04510, M´exico, D.F., M´exico.
e-mails: ∗ [email protected]
(‡).
⋆ [email protected]
† [email protected]
‡ [email protected]
|
1106.4332 | 2 | 1106 | 2011-07-12T19:49:23 | Basic polynomial invariants, fundamental representations and the Chern class map | [
"math.RA",
"math.AG",
"math.RT"
] | Consider a crystallographic root system together with its Weyl group $W$ acting on the weight lattice $M$. Let $Z[M]^W$ and $S^*(M)^W$ be the $W$-invariant subrings of the integral group ring $Z[M]$ and the symmetric algebra $S^*(M)$ respectively. A celebrated theorem of Chevalley says that $Z[M]^W$ is a polynomial ring over $Z$ in classes of fundamental representations $w_1,...,w_n$ and $S^*(M)^{W}$ over rational numbers is a polynomial ring in basic polynomial invariants $q_1,...,q_n$, where $n$ is the rank. In the present paper we establish and investigate the relationship between $w_i$'s and $q_i$'s over the integers. As an application we provide an annihilator of the torsion part of the 3rd and the 4th quotients of the Grothendieck gamma-filtration on the variety of Borel subgroups of the associated linear algebraic group. | math.RA | math |
BASIC POLYNOMIAL INVARIANTS, FUNDAMENTAL
REPRESENTATIONS AND THE CHERN CLASS MAP
S. BAEK, E. NEHER, AND K. ZAINOULLINE
Introduction
Consider a crystallographic root system Φ together with its Weyl group W acting
on the weight lattice Λ of Φ. Let Z[Λ]W and S∗(Λ)W be the W -invariant subrings
of the integral group ring Z[Λ] and the symmetric algebra S∗(Λ). A celebrated
theorem of Chevalley says that Z[Λ]W is a polynomial ring over Z in classes of
fundamental representations ρ1, . . . , ρn and S∗(Λ)W ⊗ Q is a polynomial ring over
Q in basic polynomial invariants q1, . . . , qn, where n = rank(Φ).
In the present paper we establish and investigate the relationship between ρi's
and qi's. To do this we introduce an equivariant analogue of the Chern class map φi
m and S∗(Λ)/I j
that provides an isomorphism between the truncated rings Z[Λ]/I j
a
modulo powers of the respective augmentation ideals. This allows us to express
basic polynomial invariants in terms of fundamental representations and vice versa,
hence relating the geometry of the variety of Borel subgroups X with representation
theory of the respective Lie algebra g.
A multiple of φi restricted to the respective cohomology (K0 and CH∗) of X
gives the classical Chern class map ci : K0(X) → CH i(X). This geomeric interpre-
tation provides a powerful tool to compute the annihilators of the torsion of the
Grothendieck γ-filtration on K0 of twisted forms of X as well as a tool to estimate
the torsion part of its Chow groups in small codimensions.
The paper is organized as follows. In the first section we introduce the I-adic
filtrations on Z[Λ] and S∗(Λ) together with an isomorphism φi on their truncations.
Then we study the subrings of invariants and introduce the key notion of an expo-
nent τi of a W -action on a free abelian group Λ. Roughly speaking, the integers τi
measure how far is the ring S∗(Λ)W (with integer coefficients) from being a poly-
nomial ring in qi's. In section 5 we compute all the exponents up to degree 4 and
show that they all coincide with the Dynkin index of the Lie algebra g. In section
6 we apply the obtained results to estimate the torsion in Grothendieck γ-filtration
of some twisted flag varieties. In section 7 we compute the second exponent τ2 for
a non-crystallographic group H2.
Acknowledgments. The first author has been partially supported from the NSERC
grants of the other two authors and from the Fields Institute. The second author
gratefully acknowledges support through NSERC Discovery grant 8836-20121. The
last author has been supported by the NSERC Discovery grant 385795-2010, Ac-
celerator Supplement 396100-2010 and an Early Researcher Award (Ontario).
2000 Mathematics Subject Classification. Primary 13A50; Secondary 14L24.
Key words and phrases. Dynkin index, polynomial invariant, fundamental representation, in-
variant theory, Chern class map, finite reflection group.
1
2
S. BAEK, E. NEHER, AND K. ZAINOULLINE
1. Two filtrations
Consider the two covariant functors S∗(−) and Z[−] from the category of abelian
groups to the category of commutative rings
S∗(−) : Λ 7→ S∗(Λ) and Z[−] : Λ 7→ Z[Λ]
given by taking the symmetric algebra of an abelian group Λ and the integral group
ring of Λ respectively. The ith graded component Si(Λ) is additively generated by
monomials λ1λ2 . . . λi with λj ∈ Λ and the ring Z[Λ] is additively generated by
exponents eλ, λ ∈ Λ.
The trivial group homomorphism induces the ring homomorphisms
ǫa : S∗(Λ) → Z and ǫm : Z[Λ] → Z
called the augmentation maps. By definition ǫa sends every element of positive
degree to 0 and ǫm sends every eλ to 1. Let Ia and Im denote the kernels of ǫa and
ǫm respectively. Observe that Ia = S>0(Λ) consists of elements of positive degree
and Im is generated by differences (1 − e−λ), λ ∈ Λ. Consider the respective I-adic
filtrations:
and let
S∗(Λ) = I 0
a ⊇ Ia ⊇ I 2
gr∗a(Λ) = Mi≥0
a ⊇ . . . and Z[Λ] = I 0
m ⊇ Im ⊇ I 2
m ⊇ . . .
a/I i+1
I i
a
and gr∗m(Λ) = Mi≥0
m/I i+1
I i
m
denote the associated graded rings. Observe that gr∗a(Λ) = S∗(Λ).
1.1. Example. If Λ ≃ Z, then the ring S∗(Λ) can be identified with the poly-
nomial ring in one variable Z[ω], where ω is a generator of Λ and the ring Z[Λ]
can be identified with the Laurent polynomial ring Z[x, x−1] where x = eω. The
augmentations ǫa and ǫm are given by
ǫa : ω 7→ 0 and ǫm : x 7→ 1.
We have Ia = (ω) and Im is additively generated by differences (1 − xn), n ∈ Z.
Note that the rings Z[ω] and Z[x, x−1] are not isomorphic, however they become
isomorphic after the truncation. Namely for every i ≥ 0 there is ring isomorphism
φi : Z[x, x−1]/I i+1
m ≃→ Z[ω]/I i+1
a
defined by φi : x 7→ (1 − ω)−1 = 1 + ω + . . . + ωi with the inverse defined by
φ−1
i
: ω 7→ 1 − x−1. It is useful to keep the following picture in mind
Z[x, x−1]
Z[ω]
O
O
O
O
O
O
O
O
O
Z[x, x−1]/I i+1
m
φi
≃
O
'O
/ Z[ω]/I i+1
a
observing that the inverse φ−1
can't.
i
can be lifted to the map Z[ω] → Z[x, x−1] but φi
The example can be generalized as follows:
'
o
o
/
POLYNOMIAL INVARIANTS AND FUNDAMENTAL REPRESENTATIONS
3
1.2. Lemma. [GaZ, 2.1] Assume that Λ is a free abelian group of finite rank n. The
rings Z[Λ] and S∗(Λ) become isomorphic after truncation. Namely, if {ω1, . . . , ωn}
is a Z-basis of Λ, then for every i ≥ 0 there is a ring isomorphism
defined by φi(1) = 1 and
φi : Z[Λ]/I i+1
m ≃→ S∗(Λ)/I i+1
a
with the inverse defined by φ−1
i
φi(ePn
n
j=1 aj ωj ) =
Yj=1
(ωj) = 1 − e−ωj .
(1 − ωj)−aj
Note that the map φi preserves the I-adic filtrations.
m) ⊆ I j
a for every 0 ≤ j ≤ i. Moreover, we have the following
φi(I j
Indeed, by definition
m/I i+1
1.3. Lemma. The isomorphism φi restricted to the subsequent quotients I i
doesn't depend on the choice of a basis of Λ. Hence, there is an induced canonical
isomorphism of graded rings
m
φ∗ = ⊕i≥0φi : gr∗m(Λ) ≃−→ gr∗a(Λ) = S∗(Λ).
Proof. Indeed, in this case we can define the inverse φ−1
: I i
i
a/I i+1
a → I i
m/I i+1
m by
φ−1
i
(λ1λ2 . . . λi) = (1 − e−λ1 )(1 − e−λ2 ) . . . (1 − e−λi).
It is well-defined since (1 − e−λ−λ′
) = (1 − e−λ) + (1 − e−λ′
Consider the composite of the map φi with the projections
) modulo I 2
m.
(cid:3)
φ(i) : Z[Λ] → Z[Λ]/I i+1
m
φi−→ S∗(Λ)/I i+1
a → Si(Λ).
The map φ(i), and therefore φi, can be computed on generators eλ, λ ∈ Λ as follows:
Let f (z) = Qj(1 − ωjz)−aj , where λ = Pj ajωj. Then
(cid:12)(cid:12)(cid:12)z=0
To compute the derivatives of f (z) we observe that f′(z) = f (z)g(z), where g(z) =
(1−ωj z)i+1 . Hence, starting with g0 = 1 we
Pj ajωj(1 − ωjz)−1 and di g(z)
obtain the following recursive formulas
d zi = Pj
φ(i)(ePj aj ωj ) =
dif (z)
i! aj ωi+1
1
i!
dzi
j
di f (z)
d zi = f (z) · gi(z), where gi(z) = g(z)gi−1(z) + g′i−1(z).
1.4. Example. For small values of i we obtain
i! · φ(i)(eλ) =
i
1 λ
2 λ2 + λ(2)
3 λ3 + 3λ(2)λ + 2λ(3)
4 λ4 + 6λ(4) + 6λ(2)λ2 + 8λ(3)λ + 3λ(2)2
where given a presentation λ = Pn
{ω1, ω2, . . . .ωn}, the character λ(m), m ≥ 1 is defined by
j=1 aj,λωj, aj,λ ∈ Z in terms of the basis
n
λ(m) =
aj,λωm
j .
Xj=1
4
S. BAEK, E. NEHER, AND K. ZAINOULLINE
2. Invariants and exponents
Let W be a finite group which acts on a free abelian group Λ of finite rank by
Z-linear automorphisms. Consider the induced action of W on Z[Λ] and S∗(Λ).
Observe that it is compatible with the I-adic filtrations, i.e. W (I i
m and
W (I i
m) ⊆ I i
a) ⊆ I i
Note that the isomorphisms φi and φ−1
a for every i ≥ 0.
i
ever, by Lemma 1.3 their restrictions to the subsequent quotients I i
a/I i+1
I i
a = Si(Λ) are W -equivariant and we have
are not necessarily W -equivariant. How-
m and
m/I i+1
(I i
m/I i+1
m )W ≃ (I i
a/I i+1
a
)W .
Let I W
m denote the ideal of Z[Λ] generated by W -invariant elements from the
augmentation ideal Im, i.e., by elements from Z[Λ]W ∩ Im. Similarly, let I W
a denote
the ideal of S∗(Λ) generated by W -invariant elements from Ia, i.e., by elements
from S∗(Λ)W ∩ Ia.
For each χ ∈ Λ let ρ(χ) = Pλ∈W (χ) eλ denote the sum over all elements of the
W -orbit of χ. Every element in I W
m can be written as a finite linear combination
with integer coefficients of the elements ρ(χ) = ρ(χ) − ǫm(ρ(χ)), χ ∈ Λ. Therefore,
the ideal I W
m is generated by the elements ρ(χ), i.e.,
I W
m = hρ(χ) χ ∈ Λi.
The image of I W
m by means of the composite
Z[Λ] → Z[Λ]/I i+1
m
φi−→ S∗(Λ)/I i+1
a
.
is an ideal in S∗(Λ)/I i+1
the image of I W
by φ(j)(ρ(χ)), where 1 ≤ j ≤ i, χ ∈ Λ, i.e.
generated by the elements φi(ρ(χ)), χ ∈ Λ. Therefore,
m in Si(Λ) is the ith homogeneous component of the ideal generated
a
φ(i)(I W
m ) = hf · φ(j)(ρ(χ)) 1 ≤ j ≤ i, f ∈ Si−j(Λ), χ ∈ ΛiZ .
We are ready now to introduce the central notion of the present paper:
2.1. Definition. We say that an action of W on Λ has finite exponent in degree i
if there exists a non-zero integer Ni such that
where (I W
the i-th exponent of the W -action and will be denoted by τi.
a ∩ Si(Λ). In this case the g.c.d. of all such Nis will be called
a )(i) = I W
Ni · (I W
a )(i) ⊆ φ(i)(I W
m ),
Observe that if φ(i)(I W
the exponent of φ(i)(I W
and τi τi+1 for every i ≥ 0.
m ) is a subgroup of finite index in (I W
a )(i), then τi is simply
a )(i). Note also that by the very definition τ0 = 1
m ) in (I W
3. Essential actions
In the present section we study W -actions that have no W -invariant linear forms,
i.e. we assume that ΛW = 0. In the theory of reflection groups such actions are
called essential (see [B, V, §3.7] or [H]). Note that this immediately implies that
τ1 = 1.
3.1. Lemma. For every χ ∈ Λ and m ∈ N+ we have Pλ∈W (χ) λ(m) = 0.
POLYNOMIAL INVARIANTS AND FUNDAMENTAL REPRESENTATIONS
5
Proof. Let ω1, ω2, . . . .ωn be a Z-basis of Λ. For m ∈ N+ we have
Xj=1(cid:0) Xλ∈W (χ)
λ(m) = Xλ∈W (χ)(cid:0)
Xλ∈W (χ)
j (cid:1) =
aj,λωm
Xj=1
n
n
j .
aj,λ(cid:1)ωm
In particular, for m = 1 we obtain
n
λ =
Xλ∈W (χ)
Xj=1(cid:0) Xλ∈W (χ)
aj,λ(cid:1)ωi.
Since ΛW = 0, we have Pλ∈W (χ) λ = 0. Since ωj, 1 ≤ j ≤ n are Z-free, we have
Pλ∈W (χ) aj,λ = 0 for all 1 ≤ j ≤ n.
3.2. Corollary. For every χ ∈ Λ we have
φ(2)(ρ(χ)) = 1
λ2.
(cid:3)
2 Xλ∈W (χ)
In particular, the quadratic form φ(2)(ρ(χ)) is W -invariant, i.e.
φ(2)(ρ(χ)) ∈ S2(Λ)W .
Proof. By the formula for φ(2) in Example 1.4 and by Lemma 3.1 we obtain that
φ(2)(cid:0) Xλ∈W (χ)
eλ(cid:1) = 1
2 Xλ∈W (χ)
(λ2 + λ(2)) = 1
2 Xλ∈W (χ)
λ2. (cid:3)
3.3. Corollary. If S2(Λ)W = hqi for some q, then φ(2)(I W
index in (I W
a )(2).
m ) is a subgroup of finite
Proof. The image of the ideal I W
ΛW = 0, φ(1)(ρ(χ)) = Pλ∈W (χ) λ = 0 and by Corollary 3.2, φ(2)(I W
only by the W -invariant quadratic forms φ(2)(ρ(χ)). For every χ ∈ Λ let
m is generated by φ(1)(ρ(χ)) and φ(2)(ρ(χ)). Since
m ) is generated
(1)
Then the subgroup φ(2)(I W
a )(2) of exponent
φ(2)(ρ(χ)) = Nχ · q, Nχ ∈ N.
m ) is a subgroup of (I W
Nχ. (cid:3)
τ2 = gcd
χ∈Λ
We now investigate the invariants of degree 3 and 4.
3.4. Lemma. For every χ ∈ Λ we have
φ(3)(ρ(χ)) = 1
(λ3 + 3λ(2)λ).
6 Xλ∈W (χ)
Proof. By the formula for φ(3) in Example 1.4 and by Lemma 3.1 we obtain that
φ(3)(ρ(χ)) = 1
6 Xλ∈W (χ)
(λ3 + 3λ(2)λ + 2λ(3)) = 1
6 Xλ∈W (χ)
(λ3 + 3λ(2)λ). (cid:3)
3.5. Lemma. For every χ ∈ Λ we have
φ(4)(ρ(χ)) = 1
[λ4 + 6λ(2)λ2 + 8λ(3)λ + 3λ(2)2].
24 Xλ∈W (χ)
Proof. It follows from Example 1.4 and Lemma 3.1.
(cid:3)
6
S. BAEK, E. NEHER, AND K. ZAINOULLINE
4. The Dynkin index
In the present section we show that the action of the Weyl group W of a crys-
tallographic root system Φ on the weight lattice Λ has finite exponent in degree 2
which coincides with the Dynkin index of the respective Lie algebra.
Let W be the Weyl group of a crystallographic root system Φ and let Λ be its
weight lattice as defined in [H, §2.9]. Let {ω1, . . . , ωn} be a basis of Λ consisting of
fundamental weights (here n is the rank of Φ).
The Weyl group W acts on λ ∈ Λ by means of simple reflections
sj(λ) = λ − hα∨j , λi · αj,
j = 1 . . . n
where α∨j
hα∨j , ωii = δij , where δij is the Kronecker symbol.
is the j-th simple coroot and h−,−i is the usual pairing. Note that
The subring of invariants Z[Λ]W is the representation ring of the respective Lie
algebra g. By a theorem of Chevalley it is the polynomial ring in fundamental
representations ρ(ωj) ∈ Z[Λ]W , i.e.
Z[Λ]W ≃ Z[ρ(ω1), . . . , ρ(ωn)].
Observe that the dimension of the fundamental representation ρ(ωj) equals to the
number of elements in the orbit that is ǫm(ρ(ωj)).
Therefore, the ideal I W
m is generated by the elements ρ(ωj), j = 1 . . . n and
m ) is the i-th homogeneous component of the ideal generated by
its image φ(i)(I W
φ(j)(ρ(ωl)), 1 ≤ j ≤ i, l = 1 . . . n.
4.1. Lemma. We have ΛW = 0 and hence also
m ) = 0.
φ(1)(Z[Λ]W ) = φ(1)(I W
Proof. Let η ∈ ΛW . Since η = sαj (η) = η −hη, α∨
for all simple roots αj which implies that η = 0.
4.2. Lemma. We have S2(Λ)W = hqi.
Proof. By [GN, Prop. 4] there exists an integer valued W -invariant quadratic form
on Λ which has value 1 on short coroots. As the group S2(Λ)W is identical to the
group of all integral W -invariant quadratic forms on T∗ ⊗ R, the result follows. (cid:3)
4.3. Corollary. The image φ(2)(I W
j iαj we have hη, α∨
j i = 2(αj ,η)
(αj ,αj ) = 0
(cid:3)
m ) is a subgroup of (I W
a )(2) of finite index.
Proof. This follows from Corollary 3.3 and Lemma 4.1.
(cid:3)
We recall briefly the notion of indices of representations introduced by Dynkin
[D, §2] (See also [BR]).
Let f : g → g′ be a morphism between Lie algebras. Then there exists a unique
number jf ∈ C, called the Dynkin index of f , satisfying
(f (x), f (y)) = jf (x, y),
for all x, y ∈ g, where ( -- , -- ) is the Killing form on g and g′ normalized such that
(α, α) = 2 for any long root α. In particular, if f : g → sl(V ) is a linear represen-
tation, jf is a positive integer, called the Dynkin index of the linear representation
f , defined by
tr(f (x), f (y)) = jf (x, y).
POLYNOMIAL INVARIANTS AND FUNDAMENTAL REPRESENTATIONS
7
The Dynkin index of g is defined to be the greatest common divisor of all the
Dynkin indices of all linear representations of g. By [D, (2.24) and (2.25)], the
Dynkin index of g is the greatest common divisor of the Dynkin index of its fun-
damental representations. Moreover, all the Dynkin indices of the fundamental
representations were calculated in [D, Table 5].
Using the sl2-representation theory, the Dynkin index of a linear representation
f : g → sl(V ) can be described as follows. Let α be a long root. For the formal
character ch(V ) = Pλ nλeλ, one has (see [LS, Lemma 2.4] or [KNR, 5.1 and
Lemma 5.2])
jf =
1
2 Xλ
hλ, α∨i2.
(2)
4.4. Theorem. The integers N (ωj) for the j-th fundamental weight as defined in
(1) coincide with the Dynkin index of the fundamental representation with highest
weight ωj. In particular, the second exponent τ2 coincides with the Dynkin index of
g.
Proof. To find the precise value of τ2 we use the explicit formula for φ(2), that is
φ(2)(ρ(χ)) = 1
2 Xλ∈W (χ)
λ2.
We know that τ2 is the greatest common divisor of the integers Nj = Nωj using
the notation of the proof of Corollary 3.3, where ωj is the j-th fundamental weight
of g. As the Dynkin index is the greatest common divisor of the Dynkin indices of
the fundamental representations ωj, it suffices to show that Nj coincides with the
Dynkin index of the representation Vj corresponding to ωj. We can view φ(2)(ρ(χ))
for χ = ωj as a function on the lattice hZ = SpanZ{α∨ α ∈ Φ long}. Since Vj
has character ch(Vj ) = Pλ∈W (ωj ) eλ, by (2) the Dynkin index of the representation
Vj is 1
2 Pλ∈W (ωj )hλ, α∨i2, where α is any long root in Φ. Thus, φ(2)(ρ(ωj)) is the
constant function with value Nj.
(cid:3)
We note that a different proof of Theorem 4.4 was given in [GaZ, §2].
5. Exponents of degrees 3 and 4
In the present section we show that τ2 = τ3 = τ4 for all crystallographic root
systems
Let S = {λ1, . . . , λr} be a finite set of weights. We denote by −S the set of
opposite weights {−λ1, . . . ,−λr}, by S+ the set of sums {λi + λj}i<j, by S− the
set of differences {λi − λj}i<j and by S± the disjoint union S+ ∐ S−. By definition
we have S+ = S− = (cid:0)r
2(cid:1).
Using the fact that (λ + λ′)(m) = λ(m) + λ′(m) for every λ, λ′ ∈ Λ and m ≥ 0
we obtain the following lemma which will be extensively used in the computations
5.1. Lemma. (i) For every integer m1, m2, x, y ≥ 0 and a finite subset S ⊂ Λ we
have
Xλ∈S∐−S
λ(m1)xλ(m2)y = (1 + (−1)x+y)Xλ∈S
In particular, Pλ∈S∐−S λ(2)λ2 = 0.
λ(m1)xλ(m2)y.
8
S. BAEK, E. NEHER, AND K. ZAINOULLINE
(ii) For every subset S ⊂ Λ with S = r and for every m1, m2 ≥ 0 we have
Xλ∈S+
λ(m1)λ(m2) = (r − 1)Xλ∈S
λ(m1)λ(m2) = (r − 1)Xλ∈S
Xλ∈S−
λ(m1)λ(m2) +Xi6=j
λ(m1)λ(m2) −Xi6=j
λi(m1)λj (m2) and
λi(m1)λj (m2).
In particular, this implies that Pλ∈S± λ(m1)λ(m2) = 2(r − 1)Pλ∈S λ(m1)λ(m2).
An-case. Let Φ be of type An for n ≥ 3. We denote the canonical basis of Rn+1
by ei with 1 ≤ i ≤ n + 1. According to [H, §3.5 and §3.12] the basic polynomial
invariants of the W -action on Λ (algebraically independent homogeneous generators
of S∗(Λ)W as a Q-algebra) are given by the symmetric power sums
qi := ei
1 + ··· + ei
n+1,
2 ≤ i ≤ n + 1.
Let si denote the ith elementary symmetric function in e1, . . . , en+1. Using the
classical identities
q1 = s1,
qi = s1qi−1 − s2qi−2 + . . . + (−1)isi−1q1 + (−1)i+1i · si,
and the fact that s1 = 0, we obtain that
1 < i < n + 1
generate (with integral coefficients) the ideal I W
q2/2 = −s2, q3/3 = s3, and q4/2 = s2
2 − 2s4.
a up to degree 4.
The fundamental weights of Φ can be expressed as follows
ω1 = e1, ω2 = e1 + e2,
. . . , ωn−1 = e1 + . . . + en−1, ωn = −en+1,
where e1 + e2 + . . . + en+1 = 0. The orbits of ω1, ω1 + ωn, ωn and ω2, ωn−1 under
the action of the Weyl group W = Sn+1 are given by
W (ω1) = {e1, . . . , en+1} = −W (ωn), W (ω1 + ωn) = {ei − ej}i6=j and
W (ω2) = {ei + ej}i<j = −W (ωn−1).
Therefore, W (ω1 + ωn) = S− ∐ −S− and W (ω2) = S+, where S = W (ω1).
Applying Lemma 3.5 and Lemma 5.1 we obtain that
φ(4)(ρ(ω1) + ρ(ωn)) = 1
12 Xλ∈S
24 Xλ∈S±∐−S±
φ(4)(ρ(ω1 + ωn) + ρ(ω2) + ρ(ωn−1)) = 1
6 Xλ∈S
24 Xλ∈S±∐−S±
λ4 + n
= 1
Then the difference
(8λ(3)λ + 3λ(2)2).
(λ4 + 8λ(3)λ + 3λ(2)2) and
(λ4 + 8λ(3)λ + 3λ(2)2) =
φ(4)(ρ(ω1 + ωn) + ρ(ω2) + ρ(ωn−1)) − 2n · φ(4)(ρ(ω1) + ρ(ωn)) =
= 1
24 Xλ∈S±∐−S±
λ4 − n
6 Xλ∈S
λ4 =
(3)
is a symmetric function in e1, . . . , en+1 and, therefore, it can be always written as
a polynomial in qis. Indeed, since
Xλ∈S±∐−S±
λ4 = 2Xi<j
((ei + ej)4 + (ei − ej)4) = 4nXλ∈S
λ4 + 24Xi<j
i e2
e2
j ,
POLYNOMIAL INVARIANTS AND FUNDAMENTAL REPRESENTATIONS
9
the difference (3) equals
= Xi<j
e2
i e2
j = (q2
2 − q4)/2.
5.2. Lemma. For a root system of type An, n ≥ 2, we have τ2 = τ3 = τ4 = 1.
Proof. It is enough to show that the generators q2/2, q3/3 and q4/2 are in the ideal
generated by the image of φ(i), i ≤ 4.
By Corollary 3.2 we have φ(2)(ρ(ω1)) = 1
2 Pλ∈S λ2 = q2/2. By Lemma 3.4 we
have q3/3 = φ(3)(ρ(ω1))− φ(3)(ρ(ωn)) (see also [GaZ, §1C]). If Φ is of type A2, then
s4 = 0 and, hence, q4 = q2
2/2. If Φ is of type An, n ≥ 3, then by (3) the generator
q4/2 belongs to the ideal generated by the images of φ(2) and φ(4).
(cid:3)
5.3. Lemma. For any crystallographic root system Φ the third exponent τ3 of the
W -action coincides with τ2 (the Dynkin index).
Proof. If Φ is of type An, this follows from Lemma 5.2; for the other types there are
no basic polynomial invariants of degree 3 [H, §3.7 Table 1]. Therefore, τ3 = τ2. (cid:3)
Bn, Cn and Dn cases. Let Φ be of type Bn or Cn for n ≥ 2 or of type Dn for
n ≥ 4. We denote the canonical basis of Rn by ei with 1 ≤ i ≤ n. By [H, §3.5
and §3.12] the basic polynomial invariants of the W -action on Λ are given by even
power sums
The first two fundamental weights of Φ are given by ω1 = e1, ω2 = e1 + e2 and
q2i := e2i
1 + ··· + e2i
n ,
1 ≤ i ≤ n.
their W -orbits are
W (ω1) = {±e1, . . . ,±en} and W (ω2) = {±ei ± ej}i<j.
Hence W (ω1) = S ∐ −S and W (ω2) = S± ∐ −S±, where S = {e1, . . . , en}.
Applying Lemma 3.5 and Lemma 5.1 we obtain that
φ(4)(ρ(ω1)) = 1
λ4 + 1
(8λ(3)λ + 3λ(2)2) and
12 Xλ∈S
12 Xλ∈S
24 Xλ∈S±∐−S±
φ(4)(ρ(ω2)) = 1
λ4 + n−1
6 Xλ∈S
(8λ(3)λ + 3λ(2)2).
Then similar to the An-case we obtain
φ(4)(ρ(ω2)) − 2(n − 1)φ(4)(ρ(ω1)) = (q2
2 − q4)/2,
(4)
1 + . . . + ei
n.
where qi = ei
5.4. Lemma. For a root system of type Bn or Cn, n ≥ 2 or Dn, n ≥ 4 we have
τ4 = τ2.
Proof. It is enough to show that q4/2 is in the ideal generated by the image of φ(2)
and φ(4).
By Corollary 3.2 we have φ(2)(ρ(ω1)) = Pλ∈S λ2 = q2. Therefore, by (4)
q4/2 = (q2/2) · φ(2)(ρ(ω1)) − φ(4)(ρ(ω2)) + 2(n − 1)φ(4)(ρ(ω1))
and the proof is finished.
(cid:3)
5.5. Theorem. For any crystallographic root system Φ we have τ2 = τ3 = τ4.
10
S. BAEK, E. NEHER, AND K. ZAINOULLINE
Proof. The equality τ2 = τ3 is proven in Lemma 5.3. If Φ is of type An, τ4 = 1 fol-
lows from Lemma 5.2. If Φ is of type Bn, Cn or Dn, τ4 = τ2 follows from Lemma 5.4.
For all other types τ4 = τ2 since there are no basic polynomial invariants of degree
3 and 4 (see [H, §3.7 Table 1]).
(cid:3)
6. Torsion in the Grothendieck γ-filtration
The goal of the present section is to provide geometric interpretation (see (6))
of the map φi and the exponents τi.
Let G be a simple simply-connected Chevalley group over a field k. We fix a
maximal split torus T of G and a Borel subgroup B ⊃ T . Let Λ be the group of
characters of T . Since G is simply-connected, Λ coincides with the weight lattice
of G.
Let X denote the variety of Borel subgroups of G (conjugate to B). Consider
the Chow ring CH∗(X) of algebraic cycles modulo rational equivalence and the
Grothendieck ring K0(X). Following [De74, §1] to every character λ ∈ Λ we may
associate the line bundle L(λ) over X. It induces the ring homomorphisms (called
the characteristic maps)
ca : S∗(Λ) → CH∗(X) and cm : Z[Λ] ։ K0(X)
by sending λ 7→ c1(L(λ)) and eλ 7→ [L(λ)] respectively. Note that the map ca is an
isomorphism in codimension one, hence, giving
ca : S1(Λ) = Λ ≃→ P ic(X) = CH 1(X)
a and I W
and the map cm is surjective. Let W be the Weyl group and let I W
m denote
the respective W -invariant ideals. Then according to [De73, §4 Cor.2,§9] and [CPZ,
§6]
(5)
and ker ca is generated by elements of S∗(Λ) such that their multiples are in I W
a .
Consider the Grothendieck γ-filtration on K0(X) (see [GaZ, §1]). Its ith term is
ker cm = I W
m
an ideal generated by products
γi(X) := h(1 − [L∨1 ])(1 − [L∨2 ]) · . . . · (1 − [L∨i ])i,
where L1,L2, . . . ,Li are line bundles over X. Consider the ith subsequent quo-
tient γi(X)/γi+1(X). The usual Chern class ci induces a group homomorphism
ci : γi(X)/γi+1(X) → CH i(X).
6.1. Proposition. For every i ≥ 0 there is a commutative diagram of group homo-
morphisms
(−1)i−1(i−1)!·φi
m/I i+1
I i
m
cm
Si(Λ)
ca
γi(X)/γi+1(X)
ci
/ CHi(X)
(6)
Proof. Indeed, the γ-filtration on K0(X) is the image of the Im-adic filtration on
Z[Λ], i.e. γi(X) = cm(I i
m) for every i ≥ 0. The Proposition then follows from the
identity
ci(cid:16)(1 − [L∨1 ])(1 − [L∨2 ]) . . . (1 − [L∨i ])(cid:17) = (−1)i−1(i − 1)! · c1(L1)c1(L2) . . . c1(Li),
/
/
/
POLYNOMIAL INVARIANTS AND FUNDAMENTAL REPRESENTATIONS
11
where L1,L2, . . . ,Li are line bundles over X and L∨i denotes the dual of Li.
6.2. Remark. Note that Z[Λ] can be identitfied with the T -equivariant K0 of a
point pt = Spec k and S∗(Λ) with the T -equivariant CH of a point (see [GiZ]).
The maps ca and cm then can be identified with the pull-backs KT (pt) → KT (G)
and CHT (pt) → CHT (G) induced by the structure map G → pt.
In view of these identifications the map φi can be viewed as an equivariant
(cid:3)
analogue of the Chern class map ci.
Consider the diagram (6) with Q-coefficients. In this case the Chern class map
ci will become an isomorphism (by the Riemann-Roch theorem), the characteristic
map ca will turn into a surjection and the map (−1)i−1(i − 1)! · φi will be an
isomorphism as well. In view of (5) we obtain an isomorphism
φ(i) ⊗ Q : I W
m ∩ I i
m/I W
m ∩ I i+1
m ⊗ Q −→ (I W
a )(i) ⊗ Q
on the kernels of cm and ca. By the very definition of the exponents τi this implies
that
6.3. Corollary. The action of the Weyl group of a crystallograhic root system has
finite exponent τi for every i.
We are now ready to prove the main result of this section
6.4. Theorem. The integer τi · (i− 1)! annihilates the torsion of the ith subsequent
quotient γi(X)/γi+1(X) of the γ-filtration on K0(X) for i = 3, 4.
6.5. Remark. Note that by [SGA, Expos´e XIV, 4.5] for groups of types An and
Cn the quotients γi(X)/γi+1(X) have no torsion.
Proof. Assume that α is a torsion element in γi(X)/γi+1(X). Then ci(α) = 0 since
m ⊆ Z[Λ]/I i+1
CH i(G/B) has no torsion. Let α be a preimage of α via cm in I i
m .
a )(i)
By the same analysis as in [GaZ, §1B, §1C] one can show that ker(ca)(i) = (I W
for i ≤ 4. By (6) we obtain that
m/I i+1
By definition of the index τi we have
(i − 1)! φi(α) ∈ (I W
a )(i)
Applying φ−1
i
τi · (i − 1)! φi(α) = φi(β), where β ∈ I W
to the both sides we obtain
m /I i+1
m ∩ I W
m .
τi · (i − 1)! · α = β ∈ I W
m /I i+1
m ∩ I W
m
Applying cm to the both sides and observing that I W
τi · (i − 1)! · α = 0.
m = ker cm we obtain that
(cid:3)
Let ξX be a twisted form of the variety X by means of a cocycle ξ ∈ Z 1(k, G).
By [P, Thm. 2.2.(2)] the restriction map K0(ξX) → K0(X) (here we identify
K0(X) with the K0(X ×k ¯k) over the algebraic closure ¯k) is an isomorphism. Since
the characteristic classes commute with restrictions, this induces an isomorphism
between the γ-filtrations, i.e. γi(ξX) ≃ γi(X) for every i ≥ 0, and between the
respective quotients
γi(ξX)/γi+1(ξX) ≃ γi(X)/γi+1(X)
In view of this fact Theorem 6.4 imply that
for every i ≥ 0.
12
S. BAEK, E. NEHER, AND K. ZAINOULLINE
6.6. Corollary. Let G be a split simple simply connected group of type Bn (n ≥ 3)
or Dn (n ≥ 4). Then for every ξ ∈ Z 1(k, G) the torsion in γ4(ξX)/γ5(ξX) is
annihilated by 12.
Consider the topological filtration on K0(Y ) given by the ideals
τ i(Y ) := h[OV ] V ֒→ Y, codimV Y ≥ ii.
It is known (see [GaZ, §2]) that γi(Y ) ⊆ τ i(Y ) for every i ≥ 0.
6.7. Corollary. In the notation of Corollary 6.6 assume in addition that the induced
map
is surjective. Then the 2-torsion of CH4(ξX) is annihilated by 8.
γ4(ξX)/γ5(ξX) → τ 4(ξX)/τ 5(ξX)
Proof. By the Riemann-Roch theorem [F, Ex.15.3.6], the composition
CH4(ξX) ։ τ 4(ξX)/τ 5(ξX) c4→ CH4(ξX)
is the multiplication by (−1)4−1(4 − 1)! = −6, where the first map is surjective.
Hence, the torsion subgroup of CH4(ξX) is annihilated by 72 and so the result
follows.
(cid:3)
7. 'The Dynkin index' in the H2 case
Note that the notion of an exponent τi can be defined over a unique factorisation
domain in the same way. As an example we compute the second exponent τ2 for the
action of the Weyl group of of the non-crystallographic root system H2 over the base
ring Z[ 1+√5
], hence, giving rise to an interesting question about its geometric/Lie
algebra interpretation.
2
7.1. Theorem. For the non-crystallographic root system H2 := I2(5), the second
exponent τ2 is √5.
Proof. We follow the notations in [CMP]. In the root system H2, the Weyl group
W is the dihedral group of order 10 and M is the Z[τ ]-lattice generated by two
simple roots α1 and α2, where τ = (1 + √5)/2. Observe that Z[τ ] is an Euclidean
domain.
The dual basis {ω1, ω2} is defined by
3−τ (2α1 + τ α2)
3−τ (τ α1 + 2α2)
(cid:26) ω1 = 1
ω2 = 1
or (cid:26) α1 = 2ω1 − τ ω2
α2 = −τ ω1 + 2ω2
One computes the orbits of ω1 and ω2 as follows:
W (ω1) = {ω1,−ω2,−ω1 + τ ω2,−τ ω1 + ω2, τ ω1 − τ ω2},
W (ω2) = −W (ω1).
As the action of W on M is essential, by Corollary 3.2, we have
φ(2)(ρ(ω2)) = φ(2)(ρ(ω1)) =
1
2
(ω2
1 + ω2
= (1 + τ 2)ω2
2 + (ω1 − τ ω2)2 + (τ ω1 − ω2)2 + (τ ω1 − τ ω2)2)
1 + (1 + τ 2)ω2
(7)
2 − (2τ + τ 2)ω1ω2.
Since φ(2)(ρ(ω2)) is W -invariant by Corollary 3.2, we have
τ2 = gcd(1 + τ 2, 2τ + τ 2) = gcd(2 + τ, 2τ − 1).
POLYNOMIAL INVARIANTS AND FUNDAMENTAL REPRESENTATIONS
13
But 2τ − 1 = √5 is a prime in Z[τ ], and we have 2 + τ = (2τ − 1)τ proving that
τ2 = √5.
(cid:3)
References
[B]
Bourbaki, N. Lie groups and Lie algebras: Chapters 4-6, Springer-Verlag, Berlin, (2002).
[BR] Braden, H. W. Integral pairings and Dynkin indices, J. London Math. Soc. (2) 43 (1991),
313 -- 323.
[CPZ] Calm´es, B., Petrov, V., Zainoulline, K. Invariants, torsion indices and oriented cohomology
of complete flags, Preprint arXiv 2010, 36pp.
[CMP] Chen, L., Moody, R. V., Patera, J. Non-crystallographic root systems, Quasicrystals and
discrete geometry, Fields Inst. Monogr. Vol 10, Amer. Math. Soc., Providence, RI, (1998),
135 -- 178.
[De74] Demazure, M. D´esingularisation des vari´et´es de Schubert g´en´eralis´ees, Ann. Sci. ´Ecole
Norm. Sup. (4) 7 (1974), 53 -- 88.
[De73] Demazure, M. Invariants sym´etriques entiers des groupes de Weyl et torsion, Invent.
[D]
[F]
[LS]
Math. 21 (1973), 287 -- 301.
Dynkin, E. B. Semisimple subalgebras of semisimple Lie algebras, Amer. Math. Soc.
Transl., Ser. II, 6 (1957), 111 -- 244.
Fulton, W. Intersection Theory, Springer-Verlag, Berlin (1984).
Laszlo, Y., Sorger, C. The line bundles on the moduli of parabolic G-bundles over curves
and their sections, Ann. Ec. Norm. Sup. (4) 30 (1997), 499 -- 525.
[GaZ] Garibaldi, S., Zainoulline, K. The gamma-filtration and the Rost invariant, Preprint arXiv
2010, 19pp.
[GiZ] Gille, S., Zainoulline, K. Equivariant theories and invariants of torsors. Preprint arXiv
2011, 23pp.
[GN] Gross, B. H., Nebe, G. Globally maximal arithmetic groups, J. Algebra 272 (2004), no. 2,
[H]
625 -- 642.
Humphreys, J. Reflection groups and Coxeter groups. Cambridge studies in Advanced
Math. 29, Cambridge Univ. Press (1990).
[KNR] Kumar, S., Narasimhan, M. S., Ramanathan, A. Infinite Grassmannians and moduli
[P]
spaces of G-bundles, Math. Ann. 300 (1994), 41 -- 75.
Panin, I. A. On the algebraic K-theory of twisted flag varieties, K-Theory 8 (1994), no. 6,
541 -- 585.
[SGA] SGA 6, Th´eorie des intersections et Th´eor`eme de Riemann-Roch, Lecture Notes in Math.
225, Springer-Verlag. 1971.
Sanghoon Baek, Department of Mathematics and Statistics, University of Ottawa
E-mail address: [email protected]
Erhard Neher, Department of Mathematics and Statistics, University of Ottawa
E-mail address: [email protected]
Kirill Zainoulline, Department of Mathematics and Statistics, University of Ot-
tawa, 585 King Edward, Ottawa ON K1N 6N5, Canada
E-mail address: [email protected]
|
1502.01744 | 4 | 1502 | 2015-09-02T16:07:35 | Exotic Elliptic Algebras | [
"math.RA",
"math.QA"
] | This paper examines a general method for producing twists of a comodule algebra by tensoring it with a torsor then taking co-invariants. We examine the properties that pass from the original algebra to the twisted algebra and vice versa. We then examine the special case where the algebra is a 4-dimensional Sklyanin algebra viewed as a comodule algebra over the Hopf algebra of functions on the non-cyclic group of order 4 with the torsor being the 2x2 matrix algebra. The twisted algebra is an "exotic elliptic algebra". We show that the twisted algebra has many of the good properties that the Sklyanin algebra has, and that it has some new properties that make it quite unusual by comparison. | math.RA | math |
EXOTIC ELLIPTIC ALGEBRAS
ALEX CHIRVASITU AND S. PAUL SMITH
Abstract. The 4-dimensional Sklyanin algebras, over C, A(E, τ ), are constructed from an elliptic
curve E and a translation automorphism τ of E. The Klein vierergruppe Γ acts as graded algebra
automorphisms of A(E, τ ). There is also an action of Γ as automorphisms of the matrix algebra
M2(C) making it isomorphic to the regular representation. The main object of study in this paper
is the algebra eA := (cid:0)A(E, τ ) ⊗ M2(C)(cid:1)Γ. Like A(E, τ ), eA is noetherian, generated by 4 elements
modulo six quadratic relations, Koszul, Artin-Schelter regular of global dimension 4, has the same
Hilbert series as the polynomial ring on 4 variables, satisfies the χ condition, and so on. These
results are special cases of general results proved for a triple (A, T, H) consisting of a Hopf algebra
H, a (often graded) H-comodule algebra A, and an H-torsor T . Those general results involve
transferring properties between A, A ⊗ T , and (A ⊗ T )coH. We then investigate eA from the point of
view of non-commutative projective geometry. We examine its point modules, line modules, and a
certain quotient eB := eA/(Θ, Θ′) where Θ and Θ′ are homogeneous central elements of degree two.
In doing this we show that eA differs from A in interesting ways. For example, the point modules
for A are parametrized by E and 4 more points whereas eA has exactly 20 point modules. Although
eB is not a twisted homogeneous coordinate ring in the sense of Artin and Van den Bergh a certain
quotient of the category of graded eB-modules is equivalent to the category of quasi-coherent sheaves
on the curve E/E[2] where E[2] is the 2-torsion subgroup. We construct line modules for eA that
are parametrized by the disjoint union (E/hξ1i) ⊔ (E/hξ2i) ⊔ (E/hξ3i) of the quotients of E by its
three subgroups of order 2.
Contents
1.
Introduction
2. Preliminaries
3. Torsors, twisting, and descent
4. Homological properties under twisting
5. "Exotic" elliptic algebras
7. Γ acts on E as translation by the 2-torsion subgroup
6. Generators and relations for fQ4
8. Properties of eB
9. Point modules for eQ
11. Line Modules for eQ
Appendix A. Equivariant structures
References
10. Secant lines to E and line modules for Q
1
4
5
10
13
14
16
19
24
30
31
39
44
1.1. The 3- and 4-dimensional Sklyanin algebras are among the most interesting algebras that have
appeared in non-commutative algebraic geometry. Such an algebra determines and is determined
1. Introduction
2010 Mathematics Subject Classification. 16E65, 16S38, 16T05, 16W50.
Key words and phrases. Sklyanin algebras, comodule algebras, torsors, descent.
1
2
ALEX CHIRVASITU AND S. PAUL SMITH
by an elliptic curve, E, a translation automorphism, τ , of E, and an invertible OE-module L of
degree 3, and 4, respectively. The representation theory of the Sklyanin algebra A(E, τ,L) and,
what is almost the same thing, the geometric features of the non-commutative projective space
Projnc(cid:0)A(E, τ,L)(cid:1), is governed by the geometry of E and τ when E is embedded as a cubic or
quartic curve in P(cid:0)H 0(E,L)∗(cid:1). We refer the reader to [1] and [29] for overviews of the 3- and 4-
dimensional Sklyanin algebras. The n in "n-dimensional" refers to the Gelfand-Kirillov dimension
of A(E, τ ), or its global dimension, or the dimension of A(E, τ,L)1 which is equal to H 0(E,L).
Odesskii and Feigin have defined generalizations of the 4-dimensional Sklyanin algebras in [22],
[23], and [11]. The algebras they construct depend on a pair (E, τ ), as before, but now a higher
degree line bundle is used to construct A(E, τ,L). In particular, when deg(L) = n2, n ≥ 2, Odesskii
and Feigin construct an algebra that they denote by Qn2(E, τ ).
Following an idea of Odesskii in [21], described in §1.5 below, we construct for every such pair
(E, τ ) and integer n ≥ 2 a connected graded algebra eQ = eQn2(E, τ ) by a kind of Galois descent
procedure applied to Qn2(E, τ ). We show that the algebras obtained in this manner inherit many
of the good properties enjoyed by Qn2(E, τ ). For example, they are Artin-Schelter regular.
1.2. This paper examines the case n = 2 and shows that the algebras eQ exhibit a range of novel
features. They are still governed very strongly by the geometry of E and τ . For this reason we call
them "elliptic algebras", the name Odesskii and Feigin adopted for their algebras, and we append
the adjective "exotic" to indicate that they are somewhat novel when compared to the familiar
4-dimensional Sklyanin algebras and other 4-dimensional Artin-Schelter regular algebras.
we consider the noetherian property, that of being finite as a module over its center, and numerous
homological properties that play an important role in non-commutative algebraic geometry. When
In §4 we assume that dimk(H) < ∞, and (usually) A is a connected graded H-comodule algebra.
sional Hopf algebra over a field k and A an H-comodule algebra. One might also require A to be a
graded algebra and that every homogeneous component be a subcomodule. Let T be an H-torsor
(see §3.1) and define the algebra A′ := A⊗ T . If A is graded one places T in degree zero to make A′
1.3. The procedure we use to construct the algebras eQ is quite general. Let H be a finite dimen-
a graded algebra. Let eA denote the subalgebra of A′ consisting of the H-coinvariant elements. In
§3 and §4 we show how various properties pass back and forth between A, A′, and eA. For example,
H is commutative, which is the case in the definition of eQ, A′ is an H-comodule algebra.
We show A is Koszul (m-Koszul) if and only if eA is. We show A is Artin-Schelter regular of
dimension d if and only if eA is. We show eA satisfies the χ condition, introduced in [6], if A does.
1.4. The construction A eA, and our results about properties shared by A and eA, should be
useful in other situations. It would be sensible to examine the effect of this construction on 2- and 3-
dimensional Artin-Schelter regular algebras now that J.J. Zhang and his co-authors have determined
(many/all?) the finite dimensional Hopf algebras for which such algebras can be comodule algebras.
Even the case when A is a polynomial ring, or an enveloping algebra, deserves investigation.
It was shown in [31] that Γ =
1.5. Let Q = A(E, τ,L) be a 4-dimensional Sklyanin algebra.
(Z/2)× (Z/2) acts as graded algebra automorphisms of Q when k = C. The action there is induced
by the translation action of the 2-torsion subgroup, E[2], on E. Here, working over an arbitrary
algebraically closed field k of characteristic 6= 2, we define an action of Γ as graded k-algebra
automorphisms of Q and show that this "corresponds" to the translation action of E[2] on E.
In the language of §1.3, we take H to be the Hopf algebra of k-valued functions on Γ and T
to be M2(k), the ring of 2 × 2 matrices, with an appropriate H-comodule algebra structure. We
properties Q has. It is a noetherian domain, has global dimension 4, has the same Hilbert series as
the polynomial ring on 4 indeterminates, is Artin-Schelter regular, satisfies the χ condition, etc.
then have eQ = (Q ⊗ T )coH = (Q ⊗ T )Γ. The results in §3 and §4 show that eQ has "all" the good
EXOTIC ELLIPTIC ALGEBRAS
3
1.6. Among the most important results about Sklyanin algebras are classifications of their point
and line modules. The point modules of a 3-dimensional Sklyanin algebra are naturally parametrized
by E or, more informatively, by a natural copy of E embedded as a smooth cubic curve in
P2 = P(Q∗1). The point modules for a 4-dimensional Sklyanin are parametrized by a natural copy of
E as a smooth quartic curve in P3 = P(Q∗1) and 4 additional points, those being the vertices of the
4 singular quadrics that contain the copy of E. The line modules are, in both cases, parametrized
by the secant lines to E, the lines in P(Q∗1) that meet E with multiplicity ≥ 2.
The results for eQ are very different. For example, eQ has only 20 point modules.
In a note
circulated in 1988 [10], Van den Bergh showed that a generic 4-dimensional AS-regular algebra
(with some other properties) has exactly 20 point modules. Since then, there have been a number
of examples showing that particular algebras, rather than the ephemeral "generic algebras", have
exactly 20 point modules. We believe that ours are the first such examples that turn up "in vivo",
so to speak.
1.8.
1.7. Van den Bergh and Tate [38] showed that the Odesskii-Feigin algebras Qn2 are noetherian,
Koszul, Artin-Schelter regular algebras of dimension n2 with Hilbert series (1 − t)−n2. It follows
from the relations for Qn2 that Γ = (Z/n) × (Z/n), realized as the n-torsion subgroup E[n] ⊂ E,
acts as graded algebra automorphisms of Qn2. It is an easy matter to see that the ring of n × n
matrices Mn(C) is an H-torsor where H is the Hopf algebra of k-valued functions on Γ. In §5 we
show that for all n ≥ 2, gQn2 =(cid:0)Qn2 ⊗ Mn(k)(cid:1)Γ has "the same" properties as Qn2.
In §6 we begin a detailed examination of the algebra eQ in §1.5. We give explicit generators
and relations for eQ. It has 4 generators and 6 quadratic relations (Proposition 6.1). Since Γ =
(Z/2) × (Z/2) acts on Q1 it acts as automorphisms of P(Q1)∗ = P3. This P3 contains a natural
copy of E embedded as a quartic curve and Γ restricts to an action as automorphisms of E.
In §7 we show that this action is the same as the translation action of the 2-torsion subgroup E[2].
Each γ ∈ Γ acts as an auto-equivalence M γ∗M of the graded-module category Gr(Q). Because
Γ acts as E[2] does, if Mp, p ∈ E, is the point module corresponding to p ∈ E, then γ∗Mp ∼= Mp+ω
for a suitable ω ∈ E[2]. There is a similar result for line modules: γ∗Mp,q ∼= Mp+ω,q+ω.
1.9. By [30], there is a regular sequence in Q consisting of two homogeneous central elements of
degree 2, Ω and Ω′ say, such that Q/(Ω, Ω′) is a twisted homogeneous coordinate ring, B(E, τ,L),
in the sense of Artin and Van den Bergh [5]. The main result in [5] tells us that the quotient
category QGr(B(E, τ,L)) is equivalent to Qcoh(E), the category of quasi-coherent sheaves on E.
1.10.
The algebra eQ also has a regular sequence consisting of two homogeneous central elements of
degree 2, Θ and Θ′ say. Although eB := eQ/(Θ, Θ′) is not a twisted homogeneous coordinate ring,
Theorem 8.1 proves that QGr(eB) is equivalent to Qcoh(E/E[2]).1 Nevertheless, eB has no point
modules. The points on E/E[2] correspond to fat point modules of multiplicity 2 over eB. Another
new feature is that eB is not a domain although B is. Nevertheless, eB is a prime ring.
In §9 we prove that eQ has exactly 20 point modules. These modules correspond to 20 points
in P3 = P(eQ∗1) that we determine explicitly. The "meaning" of these 20 points eludes us. Let P
of order 2. Shelton and Vancliff [27] have shown that the data (P, θ) determines eQ in the sense that
that eQ is isomorphic to T (Q1)/(R), the tensor algebra on Q1 modulo the ideal generated by R.
denote that set of 20 points. The degree shift functor M M (1) induces a permutation θ : P → P
the subspace R ⊆ Q1 ⊗ Q1 of bihomogeneous forms vanishing on the graph of θ has the property
1Although E/E[2] is isomorphic to E it is "better" to think of QGr( eB) as equivalent to Qcoh(E/E[2]).
4
ALEX CHIRVASITU AND S. PAUL SMITH
In §11, we exhibit three families of line modules for eQ parametrized by (cid:0)E/hξi(cid:1) ⊔(cid:0)E/hξ′i(cid:1) ⊔
(cid:0)E/hξ′′i(cid:1) where {ξ, ξ′, ξ′′} is the set of 2-torsion points on E. These are not all the line modules for
eQ.
In §§8 and 11, we examine Γ-equivariant objects in Gr(Q) and other categories of interest.
1.11.
So as not to interrupt the flow of the paper we collect some basic facts about group actions on
categories and equivariant objects in an Appendix. The material there is known in one form or
another, and in various degrees of generality but we have not found a suitable reference. The reader
might find the appendix useful in filling in the details of some of the proofs in §10.
1.12.
In late January 2015, after proving most of the results in this paper, we found an announce-
ment on the web of a seminar talk by Andrew Davies at the University of Manchester in January
2014 that appeared to contain some of the results we prove here. On 1/20/2015, we found a copy of
his Ph.D. thesis ([8], [9]) which has substantial overlap with this paper. Davies also proves several
and Stafford [2]. Nevertheless, most of what we do is more general, and most of our arguments
differ from his. For example, when we deal with the 4-dimensional Sklyanin algebras we make no
assumption on the order of τ , we do not restrict our base field to the complex numbers, and we
things we don't. For example, he describes eB (when τ has infinite order) in the manner of Artin
describe some of the line modules for eQ. Also, the results in §3 and §4 for arbitrary H and T are
proved by Davies only in the case H is the ring of k-valued functions on a finite abelian group.
Acknowledgement. We are very grateful to Kenneth Chan for numerous useful conversations
while working on this paper and in particular for providing some of the insight on Azumaya algebras
and related topics necessary in Section 8. We thank Pablo Zadunaisky and Michaela Vancliff for
pointing out errors in an earlier version of this paper.
2. Preliminaries
In Sections 2 to 4, we work over an arbitrary field k. Once we begin discussing the 4-dimensional
Sklyanin algebras k will be an algebraically closed field of characteristic 6= 2.
2.1. We will use what is now standard terminology and notation for graded rings and non-
commutative projective algebraic geometry. There are several sources for unexplained terminology:
the Artin-Tate-Van den Bergh papers ([3], [4]) that started the subject of non-commutative projec-
tive algebraic geometry; Stafford and Van den Bergh's survey [34]; papers by Stafford and Smith [30]
and Levasseur and Smith [16] on 4-dimensional Sklyanin algebras; the survey [29] on 4-dimensional
Sklyanin algebras; Artin and Van den Bergh's paper on twisted homogeneous coordinate rings [5];
Artin and Zhang's on non-commutative projective schemes [6].
Suppose A is an N-graded k-algebra such that dimk(Ai) < ∞ for all i. The category of Z-graded
left A-modules with degree-preserving A-module homomorphisms is denoted by Gr(A). The full
subcategory of Gr(A) consisting of modules that are the sum of their finite dimensional submodules
is denoted by Fdim(A). This is a Serre subcategory so we can form the quotient category
QGr(A) :=
Gr(A)
Fdim(A)
.
In fact, Fdim(A) is a localizing subcategory so the quotient functor π∗ : Gr(A) → QGr(A) has a
right adjoint π∗. The functor π∗ is exact. By definition, QGr(A) has the same objects as Gr(A).
Since π∗π∗ is isomorphic to the identity functor we may view objects in QGr(A) as objects in Gr(A).
2.2. We write Vect for the category of vector spaces over k.
EXOTIC ELLIPTIC ALGEBRAS
5
2.3. Throughout this paper, H is a Hopf algebra over k with bijective antipode. We write HM
for the category of left H-comodules and MH for the category of right H-comodules. Furthermore
A denotes a right H-comodule-algebra, i.e., an algebra object in MH .
Let Υ be an abelian group. We call A an Υ-graded H-comodule algebra or an Υ-graded algebra
in MH if it is an H-comodule algebra such that each homogeneous component, Ai, is an H-
subcomodule. For example, if V is a right H-comodule and R ⊆ V ⊗ V an H-subcomodule, then
the tensor algebra, T V , and its quotient T V /(R), are Z-graded algebras in MH.
We write Mod(R) for the the category of left modules over a ring R. We write AMH for the
category of A-modules internal to the category of H-comodules, i.e., vector spaces V equipped with
an A-module structure and an H-comodule structure such that A⊗ V → V is an H-comodule map.
If A is an Υ-graded algebra in MH we write Gr(A)MH for the category of Υ-graded A-modules
internal to MH , i.e. each homogeneous component Mi is an H-comodule. Similar conventions
apply to right A-modules, with the algebra subscripts appearing on the right in that case.
In this section we prove some general results on the inheritance of various properties for certain
3. Torsors, twisting, and descent
rings of (co)invariants, relating various good properties of A to those of the algebra eA defined in
In §§3.1-3.3, the only assumption on H is that it is a Hopf algebra with bijective
(3-4) below.
antipode. In §3.4 we add the hypothesis that H is commutative.
3.1. Torsors. A left H-torsor (or just torsor for short) is a left H-comodule-algebra T such that
(1) T ∼= H in HM,
(2) the ring of coinvariants, coHT , is k, and
(3) the linear map
(3-1)
T ⊗ T
ρ ⊗ id
H ⊗ T ⊗ T
id ⊗ m
H ⊗ T
is bijective where ρ : T → H ⊗ T is the comodule structure and m : T ⊗ T → T is
multiplication.
Throughout Section 3, T denotes a left H-torsor.
3.1.1. A comodule algebra for which the composition in (3-1) is an isomorphism is sometimes
called a left H-Galois object (see e.g. [7, Defn. 1.1]). Loc. cit. and the references therein are good
sources for background on torsors. Left H-torsors classify exact monoidal functors MH → Vect,
the functor corresponding to T being
(3-2)
M 7→ M (cid:3)HT := {x ∈ M ⊗ T (ρM ⊗ id)(x) = (id ⊗ ρ)(x)},
where ρM : M → M ⊗ H and ρ : T → H ⊗ T are the comodule structure maps. The vector space
M (cid:3)HT is called the cotensor product of M and T .
3.1.2. Left versus right torsors. Since the antipode, s : H → H, is an algebra anti-isomorphism,
the categories HM and MH are equivalent:
if ρ : X → H ⊗ X is a left H-comodule, then X
becomes a right H-comodule with respect to the structure map
(3-3)
X
ρ
/ H ⊗ X
s⊗id
/ H ⊗ X
τ
/ X ⊗ H
where the right-most map is τ (h ⊗ x) = x ⊗ h.
/
/
/
6
ALEX CHIRVASITU AND S. PAUL SMITH
3.1.3. Left versus right comodule algebras. The operation (3-3) does not turn a left H-comodule
algebra into a right H-comodule algebra. However, if X is a left H-comodule algebra and X op
denotes X with the opposite multiplication, then X op becomes a right H-comodule algebra with
respect to the structure map (3-3). To see this, first denote the composition in (3-3) by ρ◦ and,
when x ∈ X, write x◦ for x viewed as an element in X op. Thus, if x, y ∈ X, then x◦y◦ = (yx)◦.
Therefore if x, y ∈ X and ρ(x) = x−1 ⊗ x0, then ρ◦(x◦) = x◦0 ⊗ s(x−1) so
ρ◦(x◦y◦) = ρ◦(cid:0)(yx)◦(cid:1) = τ (s ⊗ id)ρ(yx) = τ (s ⊗ id)(y−1x−1 ⊗ y0x0) = y0x0 ⊗ s(x−1)s(y−1)
which is equal to(cid:0)x◦0 ⊗ s(x−1)(cid:1)(cid:0)y◦0 ⊗ s(y−1)(cid:1) = ρ◦(x◦)ρ◦(y◦).
Since T is a left H-torsor, T op with the structure map ρ◦ : T op → T op ⊗ H is a right H-torsor.
3.1.4. The monoidal functore• : M 7→ fM . By [39, Lemma 1.4], the functor M 7→ M (cid:3)HT in §3.1.1
is a monoidal functor. We denote it bye• : M 7→ fM . It is naturally equivalent to M 7→ (M ⊗ T )coH.
3.1.5. Sincee• is a monoidal functor, it sends algebras in MH to algebras in Vect, and hence for
In the expression M (cid:3)HT we treat T as a left H-comodule. In the expression (M ⊗ T )coH we
treat T as a right H-comodule using the new structure map in (3-3). The algebra structure on T
in not used in constructing either M (cid:3)HT or (M ⊗ T )coH.
A ∈ MH as in §2.3,
(3-4)
has a natural algebra structure. We treat T as a right H-comodule in the expression (A ⊗ T )coH .
Although T has two algebra structures, its original one and the opposite one, neither makes
A⊗ T into an H-comodule algebra unless additional hypotheses are made (see §3.4). Nevertheless,
eA is a subalgebra of A ⊗ T (T having its initial algebra structure, not the opposite one). In §3.4
below we specialize to commutative H, in which case A ⊗ T is a comodule algebra.
e• lifts to a functor AMH → Mod(eA), and similarly when everything in sight is Υ-graded for
some abelian group Υ. We denote all of these functors by the same symbol, relying on context to
differentiate between them.
In the definition of a torsor, the condition that T ∼= H in HM makes the Galois object cleft;
3.1.6.
this condition follows automatically from (3-1) when H is finite-dimensional, which is the case we
are really interested in here. This is (part of) [7, Thm. 1.9], which cites [15] for a proof.
eA := (A ⊗ T )coH
Cleft objects have an alternative characterization by means of Hopf cocycles. Recall (e.g.
[7,
Example 1.3]) that the latter are linear maps σ : H ⊗ H → k satisfying certain conditions that we
will not spell out here and which are reminiscent of those from group cohomology.
By [7, Theorem 1.8], every left torsor in the sense of §3.1 can be obtained from such a gadget σ
by twisting H: T can be identified with H as a vector space, but has a new multiplication defined
by
s ◦ t = s1t1σ(s2 ⊗ t2)
for all s, t ∈ H.
Here, s 7→ s1 ⊗ s2 is the comultiplication in H and juxtaposition on the right hand side means
with the modified multiplication
multiplication in H. Similarly, the algebra eA can be identified with the vector space A endowed
a ◦ b = a0b0σ(a1 ⊗ b1)
for all a, b ∈ A,
where a 7→ a0 ⊗ a1 is the H-comodule structure.
characteristic of k this construction specializes in the following way.
When H is the function algebra of an abelian group Γ whose order is not divisible by the
H can be identified with the group algebra kbΓ of the character group of Γ, i.e. A is bΓ-graded.
A Hopf cocycle H ⊗ H → k then turns out to be the same thing as (the linear extension of)
EXOTIC ELLIPTIC ALGEBRAS
7
a normalized group 2-cocycle µ : bΓ ×bΓ → k× in the usual sense. Now, denoting by Aα the α-
homogeneous component of A with respect to thebΓ-grading, the twisted algebra eA can be identified
with the vector space A together with the new multiplication
a ◦ b = µ(α, β)ab
for all α, β ∈bΓ, a ∈ Aα, b ∈ Aβ.
3.2. Generalities. We prove some auxiliary general results of use below.
Lemma 3.1. The categories MH
T op and Vect are equivalent via the mutually quasi-inverse functors
(3-5)
Vect
• ⊗ T op
•coH
MH
T op
Proof. By [25, Thm. I] applied to the comodule algebra T op ∈ MH the assertion follows from the
torsor condition (3-1) if T op is injective as an H-comodule. It is because T ∼= H as a left comodule
and every coalgebra is self-injective in the same way that every algebra is self-projective.
(cid:4)
Proposition 3.2. There is an isomorphism
HomH
HomH(M, N ⊗ T ) ∼= Hom(fM , eN ),
A (M, N ⊗ T ) ∼= Hom eA(fM , eN )
(3-6)
functorial in M, N ∈ MH. Moreover, it restricts to a functorial isomorphism
(3-7)
for M, N ∈ AMH.
Proof. By the adjunction between scalar extension • ⊗ T op : MH → MH
T op and scalar restriction
(i.e. simply forgetting the T op-action) the left hand side of (3-6) is naturally isomorphic to the
space HomH
T op(M ⊗ T op, N ⊗ T op), where T op acts on just the T op tensorands. In turn, this is
naturally isomorphic to the right hand side of (3-6) by Lemma 3.1.
To verify the second assertion note that the left hand side of (3-7) can be realized as an equalizer
(3-8)
HomH
A (M, N ⊗ T )
HomH (M, N ⊗ T )
HomH (A ⊗ M, N ⊗ T )
f7→⊲◦(idA⊗f )
where the upper and lower ⊲ symbols denote the action A⊗ M → M and A⊗ N → N respectively.
Applying the natural isomorphism from the first part of the proposition to the two parallel arrows
f7→f◦⊲
in (3-8), and keeping in mind the fact thate• is a monoidal functor, we get the arrows
f7→f◦⊲
Hom(fM , eN )
f7→⊲◦(idA⊗f )
Hom(eA ⊗fM , eN ).
Their equalizer is precisely the right hand side of (3-7).
(cid:4)
3.2.1. There is a graded version of Proposition 3.2 with virtually the same proof (M and N are
graded comodules, etc.).
The following simple observation turns out to be rather important.
Lemma 3.3. Suppose H is finite-dimensional. The functorse• : AMH → Mod(eA) and
e• : Gr(A)MH → Gr(eA) send projective objects to projective objects.
8
ALEX CHIRVASITU AND S. PAUL SMITH
Proof. Let A♯H∗ denote the smash product. The category AMH can be identified with Mod(A♯H∗).
Under this identification, projectives are direct summands of direct sums of copies of A♯H∗. It
3.3. The noetherian property and GK-dimension.
Proposition 3.4. Let Υ be an abelian group and A an Υ-graded H-comodule algebra. Then
As an A-module A♯H∗ is simply A ⊗ H∗ with the A-action on the left tensorand. As an H-
therefore suffices to show that the image of A♯H∗ undere• is projective over eA.
comodule A♯H∗ is the tensor product A ⊗ H∗, with H coacting on H∗ regularly. Since e• is a
monoidal functor, it sends A♯H∗ ∈ AMH to eA ⊗ fH∗ with the obvious action of eA. This is a direct
sum of copies of eA in Mod(eA) and hence projective.
dimk(Ai) = dimk(eAi) for all i ∈ Υ.
Proof. We are assuming T ∼= H in MH so W ⊗ T ∼= W ⊗ H in MH for all W ∈ MH. As in the
proof of Proposition 3.9, the map W ⊗ H → W ⊗ H, w ⊗ h 7→ w0 ⊗ w1h, is an isomorphism from
W ⊗ H with the diagonal H-coaction to W ⊗ H with the regular H-coaction on the right-hand
tensorand. As a consequence, there is a vector space isomorphism W ∼= (W ⊗ T )coH. Now apply
this fact with W equal to each homogeneous component of A.
Lemma 3.5. [14, Lem. 6.1] Let A be an N-graded k-algebra such that dimk(Ai) < ∞ for all i, and
M a finitely generated graded A-module. Then
(cid:4)
(cid:4)
GKdim(M ) = 1 + lim sup logn(dimk(Mn)).
Proof. Suppose A is left noetherian.
right-handed version of Proposition 3.2.)
Proposition 3.6. If A is a Z-graded comodule algebra such that dimk(Ai) < ∞ for all i, then A
and eA have the same Gelfand-Kirillov dimension.
Lemma 3.7. The functor forget : Gr(A)MH → Gr(A) preserves projectivity, as does the analogous
functor for ungraded modules.
Proof. This follows from the fact that forget is left adjoint to an exact functor, namely • ⊗ H :
Gr(A) → Gr(A)MH . The same proof works in the ungraded case.
Proposition 3.8. Suppose H is finite-dimensional. If A is left or right noetherian then so is eA.
Let S be an arbitrary set. The goal is to show that for any eA-module map f : eA⊗S → eA the
→ eA stabilize as S′ ⊆ S ranges over ever larger finite subsets.
images of the restrictions fS ′ : eA⊕S ′
By Proposition 3.2, f can be identified with some A-module H-comodule map ϕ : A⊕S → A⊗ T .
for finite subsets
By naturality, this identification is compatible with taking restrictions ϕS ′ to A⊕S ′
S′ ⊆ S (in the sense that fS ′ gets identified with ϕS ′).
From the proof of Proposition 3.2 we see that the image of fS ′ consists of the H-coinvariants of
the T op-submodule of A ⊗ T generated by the image of ϕS ′. Hence, it suffices to show that the
images of ϕS ′ stabilize as S′ increases. This, however, is a consequence of the noetherianness of A
and the fact that T is finite-dimensional (so that the A-module A ⊗ T is finitely generated).
(cid:4)
3.4. The case when H is commutative, and the algebra A′. In this section we assume that
H is commutative, i.e., the ring of regular functions on an affine group scheme (not necessarily
reductive or reduced).
(The right noetherian case has a similar proof using the
(cid:4)
Because H is commutative, if V and W are right H-comodules, the map V ⊗ W → W ⊗ V ,
v ⊗ w 7→ w ⊗ v, is an isomorphism of right H-comodules. It follows from this that if T is made into
a right H-comodule via the procedure in §3.1.2, then
(3-9)
A′
:= A ⊗ T
EXOTIC ELLIPTIC ALGEBRAS
9
becomes a right H-comodule algebra with its usual tensor product algebra structure. We emphasize
that the T factor in A ⊗ T has its original multiplication and is made into a right H-comodule
algebra by the procedure in §3.1.2 and not by giving T the opposite multiplication.
As mentioned in §3.1.5, eA is a subalgebra of A′. The following result therefore makes sense.
Proposition 3.9. The categories A′MH and Mod(eA) are equivalent via the mutually quasi-inverse
functors
A′⊗ eA•
(3-10)
Mod(eA)
•coH
A′MH
Furthermore, the extension eA → A′ is faithfully flat on the right and on the left.
Proof. It will be convenient to phrase the proof in terms of left comodules. Note that since H
is commutative its antipode is an automorphism and therefore the equivalence between MH and
HM described in §3.1.2 is a monoidal equivalence. In this manner, we think of A and A′ as left
comodule algebras for the duration of the proof, and show that the two functors above implement
an equivalence between Mod(eA) and H
A′M. We will also freely interchange the order of tensorands,
By [25, Thm. I], both assertions follow if A′ is injective as an H-comodule and the map
as permitted by the commutativity of H.
(3-11)
A′ ⊗ A′
ρ ⊗ id
H ⊗ A′ ⊗ A′
id ⊗ m
H ⊗ A′
analogous to (3-1) is onto, where ρ : A′ → H ⊗ A′ is the left comodule structure mentioned at the
beginning of the proof and m is multiplication.
The H-comodule T ∼= H is injective in HM (every coalgebra is self-injective, in the same way
that every algebra is self-projective). Now, for any left H-comodule M , the map
H ⊗ M → H ⊗ M,
h ⊗ m 7→ hm−1 ⊗ m0
is an isomorphism from M ⊗ H ∼= H ⊗ M with the tensor product comodule structure to M ⊗ H
with the comodule structure coming from the right hand tensorand alone. In other words M ⊗ H is
isomorphic in HM to a direct sum of dimk(M ) copies of H and in particular is injective. Applying
this to M = A, it follows that A′ = A ⊗ T ∼= A ⊗ H is injective in HM.
To check the surjectivity of (3-11) note that since (3-1) is an isomorphism so is the composition
T ⊗ A′ = T ⊗ T ⊗ A → H ⊗ T ⊗ A′ = H ⊗ T ⊗ T ⊗ A → H ⊗ T ⊗ A = H ⊗ A′,
i.e. the restriction of (3-11) to T ⊗ A′ ⊆ A′ ⊗ A′ already surjects onto H ⊗ A′.
Lemma 3.10. Keeping the notation above, if N ∈ A′MH is finitely generated over A′, then N coH
is finitely generated over eA.
Proof. Finite generation can be characterized in category-theoretic terms as follows. Let I be a
filtered small category in the sense of [17, Section IX.1]: Every two objects i, i′ fit inside a diagram
(cid:4)
i
i′
k
10
ALEX CHIRVASITU AND S. PAUL SMITH
and every solid left hand wedge as in the picture below can be completed to a commutative diagram
by a dotted right hand wedge
j
i
i′
k
For any functor F : I → Mod(A′) we have a canonical map
(3-12)
lim
HomA′(N, F (i)) → HomA′(N, lim
−→i
F (i)).
−→i∈I
We leave it to the reader to check that N is finitely generated if and only if for every filtered I and
every functor F such that every arrow F (i → i′) is an embedding the map (3-12) is an isomorphism.
Also, the hom spaces on the two sides of the arrow are H-comodules, and the isomorphism respects
these comodule structures.
Let F : I → A′MH be a functor from a filtered small category such that all F (i → i′) are
monomorphisms. Since by Proposition 3.9 the equivalence A′MH ≡ Mod(eA) is effected by the
functor (•)coH which preserves filtered colimits, the analogue of (3-12) over AcoH is obtained by
applying this functor to (3-12). Since (3-12) is an isomorphism, so is its image under (•)coH.
(cid:4)
There are analogous graded versions of Lemma 3.10 and Proposition 3.9.
4. Homological properties under twisting
We keep the notation and conventions from the previous section, under the assumption that H
is finite dimensional. We do not assume H is commutative until Theorem 4.12.
4.1. Let A be a (usually connected) graded k-algebra. For M, N ∈ Gr(A) we define the graded
vector space
Hom(M, N ) := Md∈Z
Hom(M, N (d)),
where N (d) is the degree shift of N by d and Hom here is understood from context to be the space of
degree-preserving A-module maps. Just like ordinary Hom, Hom has derived functors Exti taking
values in the category of graded vector spaces. We denote the degree-j component of Exti(M, N )
by Exti(M, N )j, as usual.
If A is noetherian and M is finitely generated then Ext(M,−) and Ext(M,−) agree or, more
precisely, Ext(M,−) is the vector space obtained by forgetting the grading on Ext(M,−). This is
not the case in general though.
4.2. Let A be a connected graded k-algebra in MH. If we make the smash product A♯H∗ into a
Z-graded k-algebra by placing H∗ in degree 0, then Gr(A)MH is equivalent to Gr(A♯H∗). Therefore
every M ∈ Gr(A)MH has a resolution by projective objects in Gr(A)MH. Let (P∗, d) be such a
projective resolution; it is also a projective resolution in Gr(A) by Lemma 3.7. If N ∈ Gr(A)MH ,
then the homology of HomA(P∗, N ) is in MH. Thus, if M, N ∈ Gr(A)MH, then every Exti
A(M, N )j
is in MH :
Lemma 4.1. Let A be a connected graded H-comodule algebra and M, N ∈ Gr(A)MH. Then the
components Exti
A(M, N )j acquire H-comodule structures natural in M, N ∈ Gr(A)MH.
Similarly, if M, N ∈ AMH, then Exti
The following result will be used repeatedly.
A(M, N ) ∈ MH , naturally in M and N .
EXOTIC ELLIPTIC ALGEBRAS
11
Theorem 4.2. Let A be a connected graded H-comodule algebra and M, N ∈ Gr(A)MH. There is
a natural isomorphism of bigraded vector spaces
Ext∗A(M, N ) q ∼= Ext∗eA(cid:0)fM , eN(cid:1) q.
(4-1)
Proof. Let (P∗, d) be a projective resolution of AM in Gr(A)MH (and hence also in Gr(A) by
Lemma 3.7). Then (eP∗,ed) is a projective resolution offM in Gr( eA)M by Lemma 3.3, and Ext∗eA
(fM , eN ) q
is the cohomology of the complex Hom eA(eP∗, eN ). By Proposition 3.2 (or rather its graded version;
see §3.2.1), this is the same as the cohomology of the complex
(4-2)
A (P∗, N ⊗ T ) ∼= HomA(P∗, N ⊗ T )coH ∼= (HomA(P∗, N ) ⊗ T )coH,
HomH
where the second isomorphism uses the finite-dimensionality of T .
The right-most complex is the image of HomA(P∗, N ) (regarded as a complex of Z-graded H-
left-hand side of (4-1) into its right-hand side.
comodules) under the functore• to graded vector spaces. Since this functor is exact, it turns the
cohomology of HomA(P∗, N ), i.e., Ext∗A(M, N ) q, into that of (4-2). In other words, e• turns the
Finally,e• is isomorphic to the forgetful functor MH → Vect as a linear functor (though not as
a monoidal functor) because T ∼= H as a comodule; the conclusion follows.
There is a version of Theorem 4.2 for ungraded modules M, N ∈ AMH ; the same proof, with
(cid:4)
the obvious modifications, works.
Corollary 4.3. Let A be a connected graded H-comodule algebra.
T V /(eR) where eR and R are isomorphic as graded vector spaces.
Proof. This follows by applying Theorem 4.2 to M = N = k from the fact that there are isomor-
phisms Ext1
(cid:4)
A(k, k) ∼= R∗ of bigraded vector spaces.
A(k, k) ∼= V ∗ and Ext2
If A ∼= T V /(R), then A ∼=
4.3. The Koszul property.
Definition 4.4. Let m be an integer ≥ 2. A connected graded algebra A is m-Koszul if A ∼= T V /(R)
with deg(V ) = 1, R ⊆ V ⊗m, and Exti
Corollary 4.5. Let m be an integer ≥ 2. A connected graded H-comodule algebra A is m-Koszul
if and only if eA is.
Proof. This follows immediately from Corollary 4.3 and Theorem 4.2 applied to M = N = k. (cid:4)
A(k, k) is concentrated in just one degree for all i.
(cid:7)
4.4. Artin-Schelter regularity. We begin by recalling the relevant notions.
Definition 4.6. A connected graded k-algebra A is Artin-Schelter Gorenstein (AS-Gorenstein for
short) of dimension d if the left and right injective dimensions of A as a graded A-module equal d
and
(4-3)
for some integer ℓ.
Exti
A(k, A) = Exti
A◦(k, A) ∼= δid k(ℓ).
If A is AS-Gorenstein we say it is Artin-Schelter regular (AS-regular for short) of dimension d if
(cid:7)
in addition gldim(A) = d < ∞.
Artin and Schelter's original definition of regularity included a restriction on the growth of
dimk(Ai) but in some situations it is sensible to avoid that restriction. We will show that if A is
AS-regular of dimension d then so is eA. Since dimk(Ai) = dimk(eAi) for all i (Proposition 3.6), if A
is AS-regular with the growth restriction so is eA.
Proposition 4.7. For all noetherian connected graded algebras A ∈ MH, gldim(eA) = gldim(A).
12
ALEX CHIRVASITU AND S. PAUL SMITH
Proof. This follows immediately from Proposition 3.8, Theorem 4.2 and the fact that for noetherian
connected graded algebras the homological dimension can be computed as the supremum of those
i for which Exti(k, k) is non-zero.
(cid:4)
Theorem 4.8. If a noetherian connected graded algebra A ∈ MH is AS-regular of dimension d so
is eA.
Proof. By Proposition 4.7, gldim(eA) = d. Theorem 4.2 and its right handed version applied to
M = k and N = A show that (4-3) holds (or does not hold) simultaneously for A and eA.
Corollary 4.9. If A is a noetherian twisted Calabi-Yau algebra, so is eA.
Proof. By [24, Lem. 1.2], an algebra is twisted Calabi-Yau if and only if it is AS-regular.
(cid:4)
(cid:4)
We can drop the noetherian hypothesis from Theorem 4.8 and Corollary 4.9 if we assume that
H is cosemisimple, i.e. its category of comodules is semisimple.
4.5. Condition χ. In this subsection we prove that the finiteness condition χ introduced in [6] is
preserved under twisting. Throughout, A will be an N-graded algebra.
Definition 4.10. [6, Defn. 3.7] We say that A has property χ if for all non-negative integers i, d
and all finitely-generated graded A-modules N there is an integer n0 such that Exti
A(A/A≥n, N )≥d
is finitely generated over A for all n ≥ n0. (The left A-module structure on Ext comes from the
right A-action on A/A≥n.)
(cid:7)
The χ condition is crucial in proving Serre-type results on finiteness of cohomology for non-
d holds.
d hold by descending induction on i. We now do this.
Proof. If the finite generation condition from 4.10 holds for all N for a fixed choice of i and d we
say that condition χi
dimension. This latter condition means that all sufficiently high Exti vanish, so that we can prove
that all χi
commutative projective schemes (see e.g. [6, Thm. 7.4]).
Theorem 4.11. If the noetherian connected graded algebra A ∈ MH of finite global dimension has
property χ then so does eA.
By Propositions 3.4, 3.8 and 4.7, eA is also noetherian connected graded and of finite global
d holds for all d and all j > i. Fix N ∈ Gr(eA) and d as
in 4.10. Because eA is noetherian, N is the cokernel in a short exact sequence
tion hypothesis we conclude that it suffices to prove that the graded eA-module Exti
(eA/eA≥n, eA⊕S)≥d is the image of
is finitely generated for sufficiently large n.
Just as in the proof of Theorem 4.2, Exti
eA
of finitely generated graded modules. Applying the resulting long exact Ext sequence and the induc-
0 → K → eA⊕S → N → 0
(eA/eA≥n, eA⊕S)≥d
Fix i and suppose we have proved that χj
eA
Un = Exti
A(A/A≥n, A⊕S)≥d ∈ Gr(A)MH
Un is also an H-comodule, it is finitely generated over A♯H∗ and hence is a quotient of some finite
under the functore•. By hypothesis, Un is finitely generated over A for sufficiently large n. Since
direct sum of copies of A♯H∗ in Gr(A)MH. Applyinge• we obtain
(eA/eA≥n, eA⊕S )≥d
as a quotient of a finite direct sum of copies of ^A♯H∗ ∼= eA ⊗ fH∗ ∈ Gr(eA).
When H is commutative the noetherian and global dimension hypotheses are not needed.
fUn = Exti
(cid:4)
eA
EXOTIC ELLIPTIC ALGEBRAS
13
Theorem 4.12. If H is commutative and the graded algebra A ∈ MH satisfies condition χ then
so does eA.
Proof. Let N be a finitely generated graded eA-module and i, d fixed integers. Because A has
property χ, there is some n0 for which the finiteness condition in 4.10 holds for the graded A-
module N′ = A′⊗ eA N (the A-module structure is obtained by restricting scalars from A′ = A⊗ T op
to A). We will show that n0 satisfies the requirements of 4.10 for N .
Apply the graded analogue of Proposition 3.9 to identify Gr(eA) with Gr(A′)MH . Arguing as in
(eA/eA≥n, N )≥d that we are interested in
the proof of Theorem 4.2 we see that the eA-module Exti
≥n, N′)≥d ∼= Exti
A(A/A≥n, N′)≥d.
To conclude, apply Lemma 3.10 (substituting (4-4) for N in that result).
is precisely the space of H-coinvariants in
(4-4)
Exti
A′(A′/A′
(cid:4)
eA
We now apply the above results to Sklyanin algebras.
5. "Exotic" elliptic algebras
5.1. Fix an integer n ≥ 3. Let k = C. Fix a primitive (n2)th root of unity ε ∈ k.
Let Q = Qn2,1(E, τ ) be the Sklyanin algebra defined in [22].
By [22, §1, Remark 2], the finite Heisenberg group of order n6, Hn2, acts as automorphisms of
Q. There is a basis xi, 1 ≤ i ≤ n2, for the degree-1 component of Q on which the generators of the
Heisenberg group act as xi 7→ xi+1 and xi 7→ εixi where the indices are labelled modulo n2. The
nth powers of the two generators generate a subgroup Γ ⊆ Hn2 that is isomorphic to (Z/n)2. The
generators of Γ act by xi 7→ xi+n and xi 7→ ζ ixi where ζ = εn.
Let H = k(Γ) denote the algebra of k-valued functions on Γ and let Mn(k) denote the n × n
matrix algebra. We make Γ act on Mn(k) by having its generators act as conjugation by
0
0
...
1
0
and
1 0 ···
0 ζ
···
...
...
. . .
0 0 ···
0
0
...
ζ n−1
.
0 1
0 0
...
...
0 0
1 0
0
1
. . .
0
0
···
···
. . .
. . .
···
By duality, the action of Γ as automorphisms of Mn(k) gives Mn(k) the structure of an H-
comodule algebra.
Lemma 5.1. The above action makes Mn(k) into a left H-torsor in the sense of §3.1.
Proof. Every character of Γ appears with multiplicity one in Mn(k). In particular, Mn(k)coH =
Mn(k)Γ = k.
A k-algebra on which Γ acts as automorphisms is the same thing as a k-algebra with a grading
by the character group of Γ. Every homogeneous component of T = Mn(k) is the k-span of an
invertible matrix. Hence, if χ and χ′ are characters of Γ, then TχTχ′ = Tχχ′. In other words, T is
a strongly graded algebra. A result of Ulbrich shows that for every group Υ the Υ-graded algebras
that are Galois as comodules over the group algebra kΥ are exactly the strongly graded ones [19,
Thm. 8.1.7]. Let Υ be the character group of Γ. Using the natural isomorphism, Pontryagin
duality, kΥ ∼= k(Γ) = H, so T is a left H-torsor.
(cid:4)
Let eQ = (Q ⊗ Mn(k))coH.
Proposition 5.2. The algebra eQ is AS-regular of dimension n2, Koszul, and noetherian, and has
Hilbert series (1 − t)−n2.
14
ALEX CHIRVASITU AND S. PAUL SMITH
Proof. By [38, Thm. 1.1, Cor. 1.3], all the hypotheses of Propositions 3.4 and 3.8, Corollary 4.5,
and Theorem 4.8 are satisfied.
(cid:4)
Lemma 5.1 and Proposition 5.2 hold when n = 2 and k is any algebraically closed field of
characteristic 6= 2. See Section 6.
Let k be an algebraically closed field whose characteristic is not 2.
We now specialize the discussion from Section 5 to n = 2, considering the action of the group
6. Generators and relations for fQ4
Γ = Z/2 × Z/2 on Q = Qn2 = Q4.
6.1. Let α1, α2, α3 ∈ k be such that α1 + α2 + α3 + α1α2α3 = 0 and {α1, α2, α3} ∩ {0,±1} = ∅.
Often we write α = α1, β = α2, and γ = α3.
We fix a, b, c, i ∈ k such that a2 = α, b2 = β, c2 = γ, and i2 = −1.
When k = C and E = C/Λ, α, β, and γ, are the values at τ of certain elliptic functions with
period lattice Λ [28, §2], [30, §2.10]. Thus, when k = C we can take
c = i
b = i
a =
,
,
θ11(τ )θ01(τ )
θ10(τ )θ11(τ )
,
θ11(τ )θ10(τ )
θ11(τ )θ01(τ )
θ11(τ )θ00(τ )
θ01(τ )θ10(τ )
where θ11, θ00, θ01, θ10 are Jacobi's four theta functions as defined at [42, p.71].
6.2. Let Q = k[x0, x1, x2, x3] be the quotient of the free algebra khx0, x1, x2, x3i by the six relations
(6-1)
x0xi − xix0 = αi(xjxk + xkxj),
x0xi + xix0 = xjxk − xkxj,
where (i, j, k) runs over the cyclic permutations of (1, 2, 3).
6.3. The earlier results will be applied to the Hopf algebra H of k-valued functions on
and its action as k-algebra automorphisms of Q given by
Γ = {1, γ1, γ2, γ3 = γ1γ2} ∼= Z2 × Z2
x2
x0
γ1 x0
γ2 x0 −x1
γ3 x0 −x1 −x2
x1
x3
x1 −x2 −x3
x2 −x3
x3
Table 1. The action of Γ as automorphisms of Q
The irreducible characters of Γ are labelled χ0, χ1, χ2, χ3 in such a way that γ(xj) = χj(γ)xj for
all γ ∈ Γ and j = 0, 1, 2, 3.
6.4. A quaternionic basis for M2(k) and the conjugation action of Γ on M2(k). Define
0 1(cid:19) ,
q0 =(cid:18)1 0
q1 =(cid:18)i
0 −i(cid:19) ,
0
i 0(cid:19) ,
q2 =(cid:18)0 i
0 (cid:19) .
q3 =(cid:18)0 −1
1
3 = −1 and, if (i, j, k) is a cyclic permutation of (1, 2, 3), qiqj = qk and qiqj +qjqi =
(6-2)
Then q2
0.
1 = q2
2 = q2
Define an action of Γ as automorphisms of M2(k) by γj(a) := qjaq−1
, i.e., g(qj) = χj(g)qj .
j
As before, eQ = (Q ⊗ M2(k))Γ. If γ ∈ Γ, then γ(xiqj) = χi(γ)χj (γ)xiqj so
y2 := x2q2,
y3 := x3q3,
y1 := x1q1,
y0 := x0,
are Γ-invariant elements of Q ⊗ M2(k).
EXOTIC ELLIPTIC ALGEBRAS
15
and
y0yi + yiy0 = yjyk + ykyj,
y0yi − yiy0 = αi(yjyk − ykyj)
subject to 6 degree-two relations. Since y0, y1, y2, y3 are Γ-invariant elements of degree one, they
(6-3)
were (i, j, k) is a cyclic permutation of (1, 2, 3). The function yj 7→ −yj, j = 0, 1, 2, 3, extends to
Proposition 6.1. The algebra eQ is generated by y0, y1, y2, y3 modulo the relations
an algebra anti-automorphism of eQ.
Proof. Because eQ is Koszul with Hilbert series (1 − t)−4, it is generated by 4 degree-one elements
generate eQ. It follows from the quadratic relations for Q4 that (x0xi−xix0)qi = αi(xjxk +xkxj)qjqk
Since eQ is a regular noetherian algebra of global dimension and GK-dimension 4, it is a domain
Proposition 6.2. There is an action of Γ as graded k-algebra automorphisms of eQ given by
and (x0xi + xix0)qi = (xjxk − xkxj)qjqk. Rewriting these relations in terms of y0, y1, y2, y3 gives
the relations in (6-3).
by [4, Thm.3.9].
(cid:4)
y2
y0
γ1 y0
γ2 y0 −y1
γ3 y0 −y1 −y2
y1
y3
y1 −y2 −y3
y2 −y3
y3
Table 2. The action of Γ as automorphisms of eQ
Using the conjugation action of Γ as automorphisms of M2(k), this gives an action of Γ as auto-
morphisms of eQ ⊗ M2(k). The invariant subalgebra (eQ ⊗ M2(k))Γ is generated by
z1 := y1q1,
z2 := y2q2,
z3 := y3q3
z0 := y0,
and is isomorphic to Q via zj 7→ xj.
(yjyk + ykyj)qjqk. Rewriting these relations in terms of z0, z1, z2, z3 gives the relations z0zi − ziz0 =
αi(zjzk + zkzj) and z0zi + ziz0 = zjzk − zkzj.
Proof. A calculation shows that the action of Γ respects the relations (6-3). Because (eQ⊗ M2(k))Γ
is Koszul with Hilbert series (1− t)−4, it is generated by 4 degree-one elements subject to 6 degree-
two relations. The elements z0, z1, z2, z3 are Γ-invariant so generate (eQ ⊗ M2(k))Γ. It follows from
the quadratic relations for eQ that (y0yi − yiy0)qi = αi(yjyk − ykyj)qjqk and (y0yi + yiy0)qi =
6.5. Central elements in eQ. In [28, Thm.2], Sklyanin proved that
1 +(cid:18) 1 + α1
1 − α2(cid:19)x2
2 +(cid:18) 1 − α1
1 + α3(cid:19)x2
Ω := −x2
belong to the center of Q when k = C. By the Principle of Permanence of Algebraic Identities, Ω
and Ω′ are central for all k.
1, x2
3 are fixed by the action of Γ. Since y2
j for j = 1, 2, 3, the
The elements x2
Ω′ := x2
0 + x2
2 + x2
3
1 + x2
0, x2
2, x2
(6-4)
and
(cid:4)
3
elements
1 + y2
0 + y2
Θ := y2
1 − α2(cid:19)y2
1 +(cid:18)1 + α1
belong to the center of eQ. We note that Θ = −Ω and Θ′ = −Ω′.
Θ′ := y2
2 + y2
3
and
j = −x2
2 +(cid:18) 1 − α1
1 + α3(cid:19)y2
3
16
ALEX CHIRVASITU AND S. PAUL SMITH
7. Γ acts on E as translation by the 2-torsion subgroup
If we use x0, x1, x2, x3 as an ordered set of coordinate functions on Q∗1, then the action of Γ
7.1.
on Q∗1 induced by its action on Q1 is given by the formulas
(7-1)
γ1(δ0, δ1, δ2, δ3) = (δ0, δ1,−δ2,−δ3)
γ2(δ0, δ1, δ2, δ3) = (δ0,−δ1, δ2,−δ3)
γ3(δ0, δ1, δ2, δ3) = (δ0,−δ1,−δ2, δ3).
We will write P3 for P(Q∗1), the projective space of lines in Q∗1. The action of Γ on Q∗1 induces
an action of Γ as automorphisms of P3 given by the formulas in (7-1).
The relations for Q, which are elements of Q1 ⊗ Q1, are bi-homogeneous forms on P3 × P3. We
mult
write R = ker(Q1 ⊗ Q1
−→ Q2) and define the subscheme
V := {(u, v) r(u, v) = 0 for all r ∈ R} ⊆ P3 × P3.
Let pri : P3 × P3 → P3, i = 1, 2, be the projections of V onto the left and right copies of P3.
Proposition 7.1. [30, Props. 2.4, 2.5] With the above notation,
pr1(V ) = pr2(V ) = E ∪ (cid:8)(1, 0, 0, 0), (1, 0, 0, 0), (1, 0, 0, 0), (1, 0, 0, 0)(cid:9)
where E is the intersection of the quadrics
(1 − γ)x2
Furthermore, E is an elliptic curve.
x2
0 + x2
1 + x2
2 + x2
2 + (1 + α)x2
3 = 0,
3 = 0.
1 + (1 + αγ)x2
The reader will notice that we use the same notation for elements in Q as for elements in the
3 is an element in S(Q1), i.e.,
symmetric algebra S(Q1). Thus, in Proposition 7.1, x2
a degree-two form on P3, whereas in (6-4), −x2
It is clear that Γ fixes the points in {(1, 0, 0, 0), (1, 0, 0, 0), (1, 0, 0, 0), (1, 0, 0, 0)}. It is also clear
that E is stable under the action of Γ (indeed, that must be so because R is Γ-stable). The map
Γ → Aut(E) is injective so we will identity Γ with a subgroup of Aut(E). Once we have fixed a
group law ⊕ on E we will identify E with the subgroup of Aut(E) consisting of the translation
automorphisms, i.e., E → Aut(E) sends a point v ∈ E to the automorphism u 7→ u ⊕ v.
3 denotes an element in Q.
Once we have defined the group (E,⊕) we will write o for its identity element and
1 + x2
0 + x2
2 + x2
0 + x2
1 + x2
2 + x2
E[2]
:= {v ∈ E v ⊕ v = o}.
The next main result, Theorem 7.6, shows we can define ⊕ such that Γ = E[2] as subgroups of
Aut(E). We will then identify Γ with E[2]. In anticipation of that result we define an involution
⊖ : E → E and a distinguished point o ∈ E by
(7-2)
and
where
⊖ (w, x, y, z) := (−w, x, y, z)
o := (cid:0)0,√ν − 1,p1 − µ, √µ − ν(cid:1)
and
ν :=
µ :=
1 − γ
1 + α
1 + γ
1 − β
and √ν − 1, √1 − µ, and √µ − ν are some fixed square roots.2 The restrictions on the values of α,
β, γ, imply that {1, µ, ν} = 3. We use this fact in the proof of Lemma 7.5.
2The choice of square root doesn't matter -- as one takes the different square roots one obtains 4 different candidates
for o. But, as we will see, with the choice of ⊕ we eventually make, those 4 points are the points in E[2]. The situation
is analogous to that of a smooth plane cubic: there are nine inflection points and if one chooses the group law so that
one of those points is the identity, then the inflection points are the points in E[3], the 3-torsion subgroup.
EXOTIC ELLIPTIC ALGEBRAS
17
Lemma 7.2. E ∩ {x0 = 0} = np ∈ E(cid:12)(cid:12)(cid:12) p = ⊖po = n(cid:0)0,±√ν − 1,±√1 − µ,±√µ − ν(cid:1)o.
Proof. It follows from the definition of ⊖ that E ∩ {x0 = 0} = {p ∈ E p = ⊖p}. Computing
E ∩ {x0 = 0} reduces to computing the intersection of the plane conics x2
3 = 0 and
3 = 0. The conics meet at four points, namely(cid:0)±√ν − 1, ±√1 − µ, ±√µ − ν(cid:1) ∈ P2.
µx2
The result follows.
Lemma 7.3. There is a degree-two morphism π : E → P1 such that π(p) = π(⊖p) for all p ∈ E,
In particular, the ramification locus of π is
i.e., the fibers of π are the sets {p,⊖p}, p ∈ E.
{p ∈ E p = ⊖p} = {o, ξ1, ξ2, ξ3} where
1 + νx2
2 + x2
1 + x2
2 + x2
(cid:4)
o := (cid:0)0,√ν − 1,p1 − µ, √µ − ν(cid:1)
ξ1 := γ1(o) = (cid:0)0, √ν − 1, −p1 − µ, √µ − ν(cid:1)
ξ2 := γ2(o) = (cid:0)0, −√ν − 1,p1 − µ, √µ − ν(cid:1)
ξ3 := γ3(o) = (cid:0)0, −√ν − 1, −p1 − µ, √µ − ν(cid:1).
Proof. The conic C, given by µx2
by π(w, x, y, z) = (x, y, z). The result is now obvious.
Proposition 7.4. Let E′ ⊆ P2 be the curve y2z = x(x − z)(x − λz) where
1 + νx2
2 + x2
3 = 0, is smooth so isomorphic to P1. Define π : E → C
(cid:4)
λ :=
ν − µν
ν − µ
=
1
1 + α! α + γ
γ 1 + γ
1 − β!,
and consider the group (E′,⊕) in which (0, 1, 0) is the identity and three points of E′ sum to zero
if and only if they are collinear.
(1) There is an isomorphism of varieties g : E → E′ such that
g(o) = ∞ = (0, 1, 0),
g(ξ1) = (0, 0, 1),
g(ξ2) = (1, 0, 1),
g(ξ3) = (λ, 0, 1).
(2) If (E,⊕) is the unique group law such that g : (E,⊕) → (E′,⊕) is an isomorphism of
(3) p ⊕ (⊖p) = o for all p ∈ E, and
(4) 4 points on E are coplanar if and only if their sum is zero.
groups, then E[2] = {p p = ⊖p} = {o, ξ1, ξ2, ξ3}, and
Proof. (1) Let π : E → C = {µx2
in Lemma 7.3 and f : C → P1 the isomorphism
1 + νx2
2 + x2
3 = 0} be the morphism π(x0, x1, x2, x3) = (x1, x2, x3)
f (x1, x2, x3) = (√−νx2 + √µx1, x3) = (x3, √−νx2 − √µx1)
with inverse
f−1(s, t) =(cid:16) 1√µ (s2 − t2),
1√−ν (s2 + t2), 2st(cid:17).
Let h = f ◦ π : E → P1. The ramification locus of π, and hence of h, is obviously {p ∈ E p = ⊖p}.
Let E′ be the plane cubic y2z = x(x− z)(x− λz) and h′ : E′ → P1 the morphism h′(x, y, z) = (x, z).
Consider the following diagram:
(7-3)
E
g
π
E′
C
h′
f
P1
The following result is implicit in [13, Ch.4, §4]: If E and E′ are elliptic curves and h : E → P1 and
h′ : E′ → P1 are degree 2 morphisms having the same branch points, then there is an isomorphism
of varieties g : E → E′ such that h′g = h.
18
ALEX CHIRVASITU AND S. PAUL SMITH
The four branch points for h are
(cid:16) ± √µν − ν ± √µν − µ, √µ − ν(cid:17) =(cid:16)√µ − ν, ±√µν − ν ∓ √µν − µ(cid:17).
The cross-ratios of these four points are(cid:8)λ, 1
λ , 1 − λ,
1
λ (cid:9) where
1−λ , λ
λ−1 , λ−1
γ 1 + γ
1 + α! α + γ
1 − β!.
1
λ :=
ν − µν
ν − µ
=
The four branch points for h′ : E′ → P1 have the same cross-ratios so E ∼= E′. In particular,
there is an isomorphism of varieties g : E → E′ such that
g(o) = ∞ = (0, 1, 0),
g(ξ1) = (0, 0, 1),
g(ξ2) = (1, 0, 1),
g(ξ3) = (λ, 0, 1).
(2) Let ⊕ be the unique group law on E such that g(p ⊕ p′) = g(p) ⊕ g(p′) for all p, p′ ∈ E.
Then g is an isomorphism of algebraic groups. Since E′[2] = {0, 1, 0), (0, 0, 1), (1, 0, 1), (λ, 0, 1)},
E[2] = {o, ξ1, ξ2, ξ3} = {p ∈ E p = ⊖p}.
(3) Since g : E → E′ is a group isomorphism it suffices to show that g(p)⊕ g(⊖p) = o. The fibers
of h consist of points that sum to zero so it suffices to show that h(g(p)) = h(g(⊖p)). However,
hg = f π and π(p) = π(⊖p) so hg(p) = hg(⊖p).
It is easy to show that if D and D′ are divisors of the same degree, then D ∼ D′ if and only if
Φ(D) = Φ(D′). The points {o, ξ1, ξ2, ξ3} are coplanar. Four points q0, . . . , q3 ∈ E are coplanar if
and only if (o) + (ξ1) + (ξ2) + (ξ3) ∼ (q0) + (q1) + (q2) + (q3). Since o⊕ ξ1⊕ ξ2⊕ ξ3 = o, q0, . . . , q3 ∈ E
are coplanar if and only if q0 ⊕ q1 ⊕ q2 ⊕ q3 = o.
Lemma 7.5. There are exactly four singular quadrics that contain E, namely
(4) Let Φ : Div(E) → E be the map Φ(cid:0)(q1) + . . . + (qm)− (r1) . . .− (rn)(cid:1) := q1⊕ . . .⊕ qm⊖ r1⊖ rn.
(cid:4)
Q0 = {µx2
Q1 = {µx2
Q2 = {νx2
Q3 = {x2
3 = 0},
2 + (µ − 1)x2
1 + (ν − 1)x2
1 + (1 − ν)x2
Let p ∈ E. For each i, the line through ⊖p and γi(p) lies on Qi.
Proof. Since the equation defining each Qi is a linear combination of the equations in Proposition 7.1,
Qi contains E. Each Qi has a unique singular point, namely ei where
2 + x2
1 + νx2
0 + (µ − ν)x2
0 + (ν − µ)x2
0 + (1 − µ)x2
3 = 0},
3 = 0},
2 = 0}.
e0 := (1, 0, 0, 0),
e1 := (0, 1, 0, 0),
e2 := (0, 0, 1, 0),
e3 := (0, 0, 0, 1).
Thus Qi is a union of lines and every line on Qi passes through ei.
Let f1, f2 be quadratic forms such that E = {f1 = f2 = 0}. A quadric contains E if and only if
it is the zero locus of λ1f1 + λ2f2 for some (λ1, λ2) ∈ P1; conversely, for all (λ1, λ2) ∈ P1 the zero
locus of λ1f1 + λ2f2 is a quadric that contains E. Since {1, µ, ν} = 3, there are exactly 4 singular
quadrics in the pencil of quadrics that contain E; these are the quadrics Qi (see [16, Prop. 3.4]).
Let p = (w, x, y, z) ∈ E. Let L be line through ⊖p and e0. Thus L = {(t− sw, sx, sy, sz) (s, t) ∈
P1}. The line L lies on Q0 and meets E when
(t − sw)2 + (sx)2 + (sy)2 + (sz)2 = µ(sx)2 + ν(sy)2 + (sz)2 = 0.
The second expression is zero for all s. The first expression is zero if and only if t2 − 2stw = 0; one
solution to this is t = 0 and it corresponds to the point ⊖p ∈ L ∩ E. The other solution occurs
when t − 2sw = 0 and corresponds to the point (w, x, y, z) = p. In other words, if w 6= 0, then the
line through p and ⊖p lies on Q0.
EXOTIC ELLIPTIC ALGEBRAS
19
The line through ⊖p and e1 is {(−sw, sx + t, sy, sz) (s, t) ∈ P1}. It lies on Q1 and meets E
when
(−sw)2 + (sx + t)2 + (sy)2 + (sz)2 = µ(−sw)2 + ν(sy)2 + (sz)2 = 0.
The second expression is zero for all s and the first is zero if and only if t2 + 2stx = 0. The solution
t = 0 to this equation corresponds to the point ⊖p ∈ L ∩ E. The other solution occurs when
t + 2sx = 0 and gives the point (−w,−x, y, z) = γ1(p). Another way of saying this is that if x 6= 0,
then the line through (−w, x, y, z) and (w, x,−y,−z) lies on Q1; i.e., the line through ⊖p and γ1(p)
lies on Q1.
Similar calculations show that the line through ⊖p and γi(p) lies on Qi for i = 2, 3.
The statement of Lemma 7.5 doesn't quite make sense if ⊖p = γi(p). It should be changed to
say that the line through ei and ⊖(p) meets E again at γi(p), i.e., the line is tangent to E.
Theorem 7.6. There is a group law ⊕ on E such that each element in Γ acts as translation by a
point in E[2].
(cid:4)
Proof. Let γi be the automorphism in Table 1 and let ξi be the point in Lemma 7.3. We will show
that γi is translation by ξi, i.e., ξi = γi(o).
Let p and q be arbitrary points of E. The line through ⊖p and γi(p) lies on Qi. So does the line
through ⊖q and γi(q). Because these lines are on Qi they meet at ei. The lines therefore span a
plane, i.e., ⊖p, γi(p), ⊖q, and γi(q), are coplanar. Therefore (⊖p)⊕ γi(p)⊕ (⊖q)⊕ γi(q) = o. Taking
q = o and rearranging the equation gives p = γi(p) ⊕ γi(o) or, γi(p) = p ⊖ γi(o) = p ⊕ γi(o).
7.2. Twisting a Q-module by γi. Let γ ∈ Γ and M a graded left Q-module. We define γ∗M to
be the graded Q-module which is equal to M as a graded vector space and has the new Q-action
(cid:4)
r qγm := γ−1(r)m
for r ∈ Q and m ∈ γ∗M = M . We make γ∗ into an auto-equivalence of Gr(Q) in the obvious way
and we note that these auto-equivalences have the property that γ∗δ∗ = (γδ)∗.
Proposition 7.7. Let p, q ∈ E and let Mp and Mp,q be the associated point and line modules. Then
γ∗i Mp ∼= Mp+ξi and γ∗i Mp,q ∼= Mp+ξi,q+ξi.
Proof. Let r ∈ Q1 and p ∈ P3 = P(Q∗1). The action of γi on Q1 and Q∗1 is such that γi(r)(p) =
r(γ−1
(p)) = r(γi(p)). Thus, r(p) = 0 if and only if γi(r) vanishes at γi(p). Since Mp = Q/Qp⊥
where p⊥ is the subspace of Q1 vanishing at p, γ∗i Mp = Q/Q(p + ξi)⊥. A similar argument works
for line modules.
(cid:4)
i
By [30, §3.9], Q/(Ω, Ω′) is isomorphic to the twisted homogeneous coordinate ring B(E, τ,L).
Since Ω and Ω′ are fixed by Γ, there is an induced action of Γ on Q/(Ω, Ω′).
8. Properties of eB
The main result in this subsection is
The quotient eQ/(Ω, Ω′) is isomorphic to eB := (B(E, τ,L) ⊗ M2(k))Γ.
8.1. The category QGr(eB). Let B = B(E, τ,L), B′ = B ⊗ M2(k), eB = (B′)Γ, and B = M2(OE).
Theorem 8.1. There is an equivalence of categories QGr(eB) ≡ Qcoh(E/E[2]).
Corollary 8.2. The set of isomorphism classes of simple QGr(eB)-objects is in natural bijection
with E/E[2].
20
ALEX CHIRVASITU AND S. PAUL SMITH
The plan is to work our way through the chain of equivalences
(8-1)
The notation needs some unpacking.
QGr(eB) ≡ QGr(B′)Γ ≡ Qcoh(B)Γ ≡ Qcoh(BΓ) ≡ Qcoh(E/E[2]).
First, Γ acts on the categories QGr(B′) and Qcoh(B). Such an action comprises an auto-
equivalence γ∗ of the respective category for each γ ∈ Γ together with natural isomorphisms
tγ,δ : γ∗ ◦ δ∗ ∼= (γδ)∗ for γ, δ ∈ Γ such that
(8-2)
γ∗ ◦ δ∗ ◦ ε∗
(γδ)∗ ◦ ε∗
γ∗ ◦ (δε)∗
/ (γδε)∗
commutes for all γ, δ, ε ∈ Γ.
The action of Γ as automorphisms of B′ induces an action of Γ on Gr(B′) as described in §7.2.
Since the subcategory Fdim(B′) is stable under each γ∗, the Γ-action passes to the quotient category
QGr(B′). The action on Qcoh(B) comes from translation on E by E[2] together with twisting via
the Γ-action on the M2(k) tensorand in B = OE ⊗ M2(k).
If Γ acts on a category C we can then form the category of Γ-equivariant objects CΓ. The objects
of CΓ are objects c ∈ C equipped with isomorphisms ϕγ : c → γ∗c for γ ∈ Γ such that
(8-3)
ϕγ
γ∗c
c
ϕγδ
γ ∗(ϕδ )
γ∗(δ∗c)
tγ,δ
(γδ)∗c
commutes and the morphisms are those in C that preserve all the structure. Explicitly, if (ϕγ)γ∈Γ
and (ϕ′γ)γ∈Γ are equivariant structures on objects c and c′, respectively, a morphism f : (ϕγ)γ∈Γ →
(ϕ′γ)γ∈Γ is a morphism f : c → c′ in C such that α∗(f )ϕγ = ϕ′γf for all γ ∈ Γ. This elucidates the
notation CΓ in (8-1) for C = QGr(B′) or Qcoh(B).
Finally, BΓ denotes the sheaf of algebras on E/E[2] obtained by descent from B. To make sense
of this, let ρ : E → E/E[2] be the ´etale quotient morphism. Now recall
Proposition 8.3. [20, Prop. 2, p.70] The functors
are mutually inverse equivalences between Qcoh(E/E[2]) and Qcoh(E)Γ.
G ρ∗G
and
F (ρ∗F)Γ
The equivalences in Proposition 8.3 are monoidal, because ρ∗ is, so they identify Γ-equivariant
sheaves of algebras on E with sheaves of algebras on E/E[2]. Keeping this in mind, BΓ is simply
shorthand for the sheaf of algebras on E/E[2] corresponding to B ∈ Qcoh(E)Γ, i.e. (ρ∗B)Γ.
Proof of Theorem 8.1. We go through the equivalences in (8-1) one by one, moving rightward.
First equivalence. The graded version of Proposition 3.9 (applied to B′ coacted upon by the
function algebra of Γ) provides the equivalence Gr(eB) and Gr(B)Γ. The equivalence restricts to the
subcategories Fdim(eB) and Fdim(B′)Γ so descends to the quotient categories QGr.
Second equivalence. By [5, Thm. 3.12], QGr(B) ≡ Qcoh(OE). Since B = OE ⊗ M2(k), Morita
equivalence lifts this to
(8-4)
Now note that Γ acts on the geometric data (E, τ,L) that gives rise to B = B(E, τ,L) in the sense
that it acts on E, commutes with τ , and there is an Γ-equivariant structure on L. Moreover, it acts
QGr(B′) ≡ Qcoh(B).
/
/
/
EXOTIC ELLIPTIC ALGEBRAS
21
in the same way on the M2(k) tensorands in B′ = B⊗M2(k) and B = OE⊗M2(k). This observation
together with the precise description of the equivalence QGr(B) ≡ Qcoh(E) from [5, Thm. 3.12]
shows that (8-4) intertwines the Γ-actions on the two categories. This implies the desired result
that it lifts to an equivalence between the respective categories of Γ-equivariant objects.
Third equivalence. This also follows from Proposition 8.3. As observed before that equivalence
is monoidal, and it identifies B ∈ Qcoh(E)Γ with BΓ ∈ Qcoh(E/E[2]). The monoidality then ensures
that it implements an equivalence between the categories of modules over B and BΓ internal to
Qcoh(E)Γ and Qcoh(E/E[2]) respectively.
Fourth equivalence. Because ρ : E → E/E[2] is ´etale and ρ∗(BΓ) ∼= M2(OE), BΓ is a sheaf
of Azumaya algebras on E/E[2]. The fourth equivalence now follows from Morita equivalence and
the fact that BΓ is Azumaya and hence (because we are working over an algebraically closed field)
of the form End(V) for some vector bundle V on E/E[2].
(cid:4)
We can actually find an explicit vector bundle V on E/E[2] such that BΓ ∼= End(V).
(cid:4)
Proposition 8.4. Let V be the unique non-split extension 0 −→ OE/E[2] −→ V −→ OE/E[2] −→ 0.
There is an isomorphism of OE/E[2]-algebras BΓ ∼= End(V).
Proof. We already know that BΓ is trivial Azumaya, hence BΓ ∼= End(V) for some rank 2 vector
bundle V. By Atiyah's classification of vector bundles on elliptic curves, either V is decomposable,
or isomorphic to V ⊗ L for some L ∈ Pic(E/E[2]). If V is decomposable, the OE/E[2]-module BΓ
contains two copies of OE/E[2] as direct summands, whence dim H 0(BΓ) > 2. Since dim H 0(BΓ) =
dim H 0(B)Γ = 1, we must have BΓ ∼= End(V ⊗ L) ∼= End(V).
8.2. E/E[2] is a closed subvariety of Projnc(eQ). The title of this subsection is made precise in
the following way. In [40, §3.4], a subcategory B of an abelian category D is said to be closed if it
is closed under subquotients and the inclusion functor i∗ : B → D is fully faithful and has a left
adjoint i∗ and a right adjoint i!. In [32, Thm. 1.2], which corrects an error in [33], it is shown that
if J is a two-sided ideal in an N-graded k-algebra A, then the inclusion functor Gr(A/J) → Gr(A)
induces a fully faithful functor i∗ : QGr(A/J) → QGr(A) whose essential image is closed in the sense
of [40, §3.4]. In particular, since eB is a quotient of eQ, this result in conjunction with Theorem 8.1
shows that the essential image of the composition Qcoh(E/E[2]) → QGr(eB) → QGr(eQ) is closed in
8.3. Fat point modules for eB. Let p ∈ E. Let p⊥ ⊂ Q1 be the subspace of Q1 vanishing at p.
natural way. Then Mp ⊗ k2 is a left Q ⊗ M2(k)-module, and hence a left eQ-module.
Since (Ω, Ω′) annihilates Mp, Mp ⊗ k2 is a eB-module.
Lemma 8.5. If p ∈ E, then at most one of {x0, x1, x2, x3} vanishes at p.
Proof. Suppose xr(p) = xs(p) = 0 and r 6= s. Let t ∈ {0, 1, 2, 3}−{r, s}. There are non-zero scalars
λ, µ, ν such that λx2
3 vanishes
on E so it would follow that xj(p) = 0 for all j. That is absurd.
(cid:4)
We call Mp := Q/Qp⊥ the point module associated to p. We view k2 as a left M2(k)-module in the
t vanishes on E so xt(p) = 0 also. But x2
the sense of [40, §3.4].
r + µx2
0 + x2
1 + x2
2 + x2
s + νx2
Proposition 8.6. Let p ∈ E. If m ⊗ v is a non-zero element in (Mp ⊗ k2)n, then eQ(m ⊗ v) ⊇
(Mp ⊗ k2)≥n+1. In particular, every quotient of Mp ⊗ k2 by a non-zero graded eQ-submodule has
finite dimension; i.e., Mp ⊗ k2 is 1-critical.
Proof. Let N be a non-zero graded eQ-submodule of Mp ⊗ k2. Let en ⊗ v be a non-zero element in
N where {en} is a basis for the degree-n component of Mp and v ∈ k2 − {0}.
Every non-zero matrix in (kq0 + kq2) ∪ (kq0 + kq3) ∪ (kq1 + kq2) ∪ (kq1 + kq3) has rank 2 so
(kq0 + kq2)v = (kq0 + kq3)v = (kq1 + kq2)v = (kq1 + kq3)v = k2.
22
ALEX CHIRVASITU AND S. PAUL SMITH
for the degree-(n + 1) component of Mp such that xjen = λjen+1 for j = 0, . . . , 3.
If p + nτ = (λ0, λ1, λ2, λ3) with respect to the coordinates x0, . . . , x3, then there is a basis {en+1}
By Lemma 8.5, at least one element in {x0, x1} and at least one element in {x2, x3} does not
vanish at p + nτ . Suppose, for the sake of argument, that x1(p + nτ ) 6= 0 and x2(p + nτ ) 6= 0. Then
x1en and x2en are non-zero. It follows that (kx1 ⊗ q1 + kx2 ⊗ q2) · (en ⊗ v) = en+1 ⊗ k2. Thus,
eQ1(en ⊗ v) = en+1 ⊗ k2. The same sort of argument can be used in the other cases (for example,
if x0(p + nτ ) and x2(p + nτ ) are non-zero) to show that eQ1(en ⊗ v) is always equal to en+1 ⊗ k2.
It now follows by induction on n that eQ(en ⊗ v) ⊇ (Mp)≥n+1 ⊗ k2. The result follows.
Corollary 8.7. Every simple object in QGr(eB) is isomorphic to π∗(Mp ⊗ k2) for some p ∈ E.
The previous result is the reason that Mp ⊗ k2 is called a fat point module for eQ: "point"
HomQGr( eQ)(eQ, π∗(Mp ⊗ k2)) = 2, not 1.
Proposition 8.8. If ω ∈ E[2] and p ∈ E, then there is an isomorphism of eQ-modules
because in algebraic geometry simple objects in Qcoh(X) correspond to closed points, "fat" because
(cid:4)
Mp ⊗ k2 ∼= Mp+ω ⊗ k2.
Proof. Write E[2] = {o, ξ1, ξ2, ξ3}. If ω = o the identity map is an isomorphism. Fix i ∈ {1, 2, 3}.
Let {en n ≥ 0} be a homogeneous basis for Mp with deg(en) = n. For each n, let ξnj ∈ k,
j = 0, 1, 2, 3, be the unique scalars such that
xjen = ξnjen+1.
Thus, (ξn0, ξn1, ξn2, ξn3) = p + nτ . Let ξ′n0 = ξn0, ξ′ni = ξni, and ξ′nj = −ξnj when j ∈ {1, 2, 3}−{i}.
Therefore p + nτ + ξi = (ξ′n0, ξ′n1, ξ′n2, ξ′n3). Let {fn n ≥ 0} be the unique homogeneous basis for
Mp+ξi with deg(fn) = n such that xjfn = ξ′njfn+1 for j = 0, 1, 2, 3.
Define ϕi : Mp ⊗ k2 −→ Mp+ξi ⊗ k2 by ϕi(en ⊗ v) := fn ⊗ qiv. It follows that
ϕi(cid:0)yj · (en ⊗ v)(cid:1) = ϕi(xjen ⊗ qjv) = ϕi(ξjen+1 ⊗ qjv) = ξjfn+1 ⊗ qiqjv
yj · ϕi(en ⊗ v) = yj · (fn ⊗ qiv) = xjfn+1 ⊗ qjqiv) = ξ′jfn+1 ⊗ qjqiv.
and
For all j, ξjfn+1 ⊗ qiqjv = ξ′jfn+1 ⊗ qjqiv because
• if j ∈ {0, i}, then ξj = ξ′j and qiqj = qjqi;
• if j ∈ {1, 2, 3} − {i}, then ξj = −ξ′j and qiqj = −qjqi.
Therefore ϕi(cid:0)yj · (en ⊗ v)(cid:1) = yj · ϕi(en ⊗ v) for j = 0, 1, 2, 3. This proves that ϕi is a homomorphism
of graded eQ-modules. It is obviously bijective so the proof is complete.
8.4. eB is a prime ring. Davies [9, Cor. 5.3.21] proved that eB is a prime ring when τ has infinite
order [9, Hypothesis 5.0.2]. We use a different method to prove the result without any restriction
on τ .
(cid:4)
Proposition 8.9. Let I1 and I2 be graded ideals in an N-graded left and right noetherian k-
algebra A. Suppose there is a projective scheme X and an equivalence of categories Φ : QGr(A) →
Qcoh(X). By [32], there are functors α1∗ and α2∗, and closed subschemes Z1, Z2 ⊆ X such that
the essential image of Φαi∗ is equal to Qcoh(Zi), and there is a commutative diagram
EXOTIC ELLIPTIC ALGEBRAS
23
(8-5)
Gr(A/I1)
f1∗
π1
Gr(A)
π
f2∗
Gr(A/I2)
π2
QGr(A/I1) α1∗
/ QGr(A)
a2∗
QGr(A/I2)
Φ
Qcoh(X).
in which fi∗ : Gr(A/Ii) → Gr(A), i = 1, 2, are the natural inclusion functors, and π1, π2, and π
denote the quotient functors. If I1 ∩ I2 = 0 and X is reduced and irreducible, then Z1 ∪ Z2 = X.
Proof. Let Ox be the skyscraper sheaf at a closed point x ∈ X and M an A-module such that
ΦπM ∼= Ox and π∗(M/N ) = 0 for all non-zero N ⊆ M .
If I2M = 0, then Ox ∼= Φi2∗π2M so
x ∈ Z2. On the other hand, suppose I2M 6= 0. Then π(M/I2M ) = 0 so π(I2M ) ∼= Ox. Since
I1I2M = 0, π(I2M ) = πf1∗(I2M ) = i1∗π1M which implies that i1∗π1M ∼= Ox. Hence x ∈ Z1.
Thus, every closed point of X belongs to Z1 ∪ Z2. The proposition now follows from the fact
that X is reduced and irreducible.
(cid:4)
Theorem 8.10. Let A be a connected, N-graded, left and right noetherian k-algebra Suppose there
If A is
is a projective scheme X and an equivalence of categories Φ : QGr(A) → Qcoh(X).
semiprime and X is reduced and irreducible, then A is a prime ring.
Proof. Suppose the result is false. Then there are non-zero elements x and y such that xAy = 0.
If xm and yn are the top-degree components of x and y, then xmAyn = 0. Let I1 = AxmA and
I2 = AynA. Then I1 and I2 are graded ideals such that I1I2 = 0. Since (I1 ∩ I2)2 ⊆ I1I2, the fact
that A is semiprime implies I1 ∩ I2 = 0. Hence Z1 ∪ Z2 = X. But X is irreducible so either Z1 = X
or Z2 = X.
Without loss of generality suppose that Z1 = X. Then the functor i1∗ : QGr(A/I1) → QGr(A) is
an equivalence. In particular, there is a module M ∈ Gr(A/I1) such that πA ∼= i1∗π1M = πf1∗M .
Hence, if ω is the right adjoint to π constructed by Gabriel, ωπA ∼= ωπf1∗M . By Step 2 in the
proof of [32, Thm. 1.2], ωπf1∗M ∼= f1∗ω′π′M where ω′ is right adjoint to π′. It follows that I1
annihilates ωπA.
There is an exact sequence 0 → T → A → ωπA where T is the largest finite dimensional
submodule of A. Since A0 = k, T ⊆ A≥1. It follows that T n = 0 for n ≫ 0. But A is semiprime so
T = 0. Therefore I1 annihilates A. Hence I1 = 0.
Corollary 8.11. eB is a prime ring.
Cor. 1.5(1)] shows that (B ⊗ M2(k))Γ, which is eB, is a semiprime ring. Therefore Theorems 8.1
and 8.10 imply that eB is a prime ring.
show that A does not contain a non-zero left ideal of finite dimension. For eB, one can prove that
without appealing to the fact that eB is connected. Since eB = eQ/(Θ, Θ′) where Θ, Θ′ is a regular
sequence on eQ of length 2, the projective dimension of eB as a left eQ-module is 2. Hence, by [16,
Prop. 2.1(e)], eB does not contain a non-zero left ideal of finite dimension.
Proof. As observed in [8, Cor. 5.1.8], because B is a domain B ⊗ M2(k) is a prime ring, so [18,
8.4.2. The twisted homogeneous coordinate ring of a reduced and irreducible variety, in particular
B(E, τ,L), is a domain.
8.4.1. Remark. The hypothesis in Theorem 8.10 that the algebra A is connected was needed to
(cid:4)
(cid:4)
/
/
o
o
/
o
o
24
ALEX CHIRVASITU AND S. PAUL SMITH
Proposition 8.12. eB is not a domain. In particular, in eB, 0 = y2
(y0 − y1 − y2 − y3)2 = (y0 − y1 + y2 + y3)2 = (y0 + y1 − y2 + y3)2 = (y0 + y1 + y2 − y3)2.
Proof. This is a straightforward calculation: (y0 − y1 − y2 − y3)2 equals
0 + y2
2 + y2
1 + y2
3 =
y2
0 + y2
1 + y2
2 + y2
3 −
(y0yi + yiy0 − yjyk − ykyj)
3Xi=1
where (i, j, k) is a cyclic permutation of 1, 2, 3. But y0yi + yiy0 = yjyk + ykyj when (i, j, k) is a
cyclic permutation of 1, 2, 3 and y2
3 = −Ω which is zero in eB. Similar calculations
show that the squares of the other 3 elements are zero in eB; alternatively, one can use the fact that
Γ acts as automorphisms of eB and these four elements in eB1 form a Γ-orbit.
0 + y2
1 + y2
2 + y2
(cid:4)
A point module for a connected graded algebra A is a graded left A-module M such that M =
AM0 and dimk(Mi) = 1 for all i ≥ 0. The importance of point modules is that they are simple
objects in QGr(A).
9. Point modules for eQ
9.1. Suppose M is a point module for eQ. Its degree-zero component, M0, is annihilated by a 3-
dimensional subspace of eQ1. That 3-dimensional subspace determines and is determined by a point
in P3, its vanishing locus. We will show that the only points in P3 that arise in this way are those
in Table 4 where the coordinates are written with respect to the coordinate system (y0, y1, y2, y3).
We write P for this set of points.
Recall that a, b, c, i are fixed square roots of α, β, γ,−1.
P∞
(1, 0, 0, 0)
(0, 1, 0, 0)
(0, 0, 1, 0)
(0, 0, 0, 1)
P0
P1
P2
P3
(1, 1, 1, 1)
(1, 1,−1,−1)
(1,−1, 1,−1)
(1,−1,−1, 1)
(bc,−i,−ib,−c)
(ac,−a,−i,−ic)
(bc,−i, ib, c)
(ac,−a, i, ic)
(bc, i,−ib, c)
(ac, a,−i, ic)
(bc, i, ib,−c)
(ac, a, i,−ic)
Table 3. The points in P.
(ab,−ia,−b,−i)
(ab,−ia, b, i)
(ab, ia,−b, i)
(ab, ia, b,−i)
Γ
γ1
γ2
γ3
The points in P∞ are fixed by Γ and every other Pi is a Γ-orbit. If u is the topmost point in
one of the columns Pi, i = 0, 1, 2, 3, the other points in that column are γ1(u), γ2(u), and γ3(u),
in that order.
We define a permutation θ of P with the property θ2 = idP by
(9-1)
θ(u) :=(u
γi(u)
if u ∈ P∞ ∪ P0
if u ∈ Pi, i = 1, 2, 3.
9.2. The point scheme, P. Let V denote the linear span of y0, y1, y2, y3. The defining relations for
Let
eQ belong to V ⊗2. Non-zero elements in V ⊗2 are forms of bi-degree (1, 1) on P(V ∗)×P(V ∗) = P3×P3.
P : = the subscheme of P3 × P3 where the quadratic relations for eQ vanish.
We will show that P is a reduced scheme consisting of 20 points.
Lemma 9.1. If (u, v) ∈ P, then (v, u) ∈ P.
EXOTIC ELLIPTIC ALGEBRAS
25
Proof. As remarked in Proposition 6.1, there is an anti-automorphism of eQ given by yi 7→ −yi
for i = 0, 1, 2, 3. Thus, if r = P µijyi ⊗ yj is a quadratic relation for eQ so is r′ = P µijyj ⊗ yi.
Obviously, r vanishes at (u, v) ∈ P3× P3 if and only if r′ vanishes at (v, u). The lemma now follows
from the fact that P is the zero locus of the set of quadratic relations for eQ.
9.2.1. From point modules to points in P. Suppose M is a point module for eQ. Let e0, e1, . . . be a
basis for M with deg(en) = n. Define λnj ∈ k by the requirement that yjen = λnjen+1. Because
M is a point module, for each n, some λnj is non-zero. The point pn := (λn0, λn1, λn2, λn3) ∈ P3
does not depend on the basis {en}n≥0. Since yj(pn) = λnj, the pn's belong to P(V ∗).
Because M is a eQ-module, each quadratic relation r ∈ V ⊗2 has the property that r · en = 0 for
all n. Thus, r viewed as a (1,1) form on P3 × P3 vanishes at (pn+1, pn). Hence (pn+1, pn) ∈ P.
9.3. The point modules Mu, u ∈ P.
Proposition 9.2. Let u ∈ P. Let θ be the function defined at (9-1) and for each n ≥ 0 write
θn(u) = (λn0, λn1, λn2, λn3) where the coordinates are written with respect to (y0, y1, y2, y3). There
is a point module, Mu, with homogeneous basis e0, e1, . . ., deg(en) = n, and action
(cid:4)
(9-2)
yjen := λnjen+1.
These 20 point modules are pair-wise non-isomorphic.
Proof. It is clear that Mu is generated by e0 so it suffices to show that (9-2) really does define a
left eQ-module. To do this we must show that every relation for eQ annihilates every en. In other
words, we must show that every quadratic relation for eQ, when viewed as a form of bi-degree (1, 1)
on P3 × P3, vanishes at(cid:0)(λn+1,0, λn+1,1, λn+1,2, λn+1,3), (λn0, λn1, λn2, λn3)(cid:1) ∈ P3 × P3 for all n ≥ 0.;
i.e., it suffices to show that these forms vanish at (θ(v), v) for all v ∈ P. Since θ2 = 1, this is
equivalent to showing they vanish at (v, θ(v)) for all v ∈ P.
The relations for eQ are the entries in the matrix M1y where
M1 =
−y1
y0
αy3 −αy2
βy1
−y2 −βy3
y0
y0
−y3
γy2 −γy1
−y2
y0
y1
−y3
−y3
y2
y0
−y1
y0
−y2 −y1
y3
We must therefore show that M1(v)θ(v)T = 0 for all v ∈ P. This is a routine calculation. We give
one example to illustrate the process.
Let v = (δ0, δ1, δ2, δ3) ∈ P1. Then θ(v) = γ1(v) = (δ0, δ1,−δ2,−δ3) so
M1(v)θ(v)T =
−δ1
δ0
αδ3 −αδ2
βδ1
−δ2 −βδ3
δ0
δ0
γδ2 −γδ1
−δ3
δ1
δ0
−δ3
−δ2
−δ3
−δ1
δ0
δ2
δ0
−δ1
δ3
−δ2
It is easy to check that this 6 × 1 matrix is 0 for all v ∈ P1.
The annihilator of e0 in eQ1 is the subspace that vanishes at u. Hence if u and v are different
points of P, Mu 6∼= Mv.
Theorem 9.3. The 20 point modules Mu, u ∈ P, in Proposition 9.2 are all the eQ-point modules.
(cid:4)
y0
y1
y2
y3
.
and y =
= 2
δ0
δ1
−δ2
−δ3
−δ0δ2 − βδ3δ1
−δ0δ3 + γδ1δ2
δ0δ1 + δ2δ3
0
0
0
.
26
ALEX CHIRVASITU AND S. PAUL SMITH
Proof. Let M be a point module for eQ. Let {en n ≥ 0} be a homogeneous basis for M with
deg(en) = n. Let pn, n ≥ 0, be the points in P3 determined by the procedure described in §9.2.1.
Then (pn+1, pn) ∈ P for all n ≥ 0. By Lemma 9.1, (pn, pn+1) ∈ P. Thus, to prove the Theorem it
suffices to show that P =(cid:8)(cid:0)u, θ(u)(cid:1)(cid:12)(cid:12) u ∈ P(cid:9). This is what we do in Theorem 9.4 below.
(cid:4)
Theorem 9.4. Let P ⊆ P3 × P3 be the subscheme defined in §9.2. Then
P = {(u, v) ∈ P3 × P3 M1(u)v = 0} = (cid:8)(cid:0)u, θ(u)(cid:1)(cid:12)(cid:12) u ∈ P(cid:9).
In particular, P is the graph of the automorphism θ of P.
Proof. Let pr1, pr2 : P → P3 denote the projections onto the first and second factors of P3 × P3.
We will show that pr1(P) = P. Let u ∈ pr1(P). There is a point v ∈ P3 such that (u, v) ∈ P, i.e.,
such that M1(u)v = 0. This implies that rank(M1(u)) ≤ 3. Thus the 4 × 4 minors of M1 vanish at
u. We used SAGE [35] to compute these minors. After removing a common factor of 2, they are
− bγy0y3
1 − αγy0y1y2
2 + βγy2
1y2y3 + aγy3
0 + βγy2
1 + αγy2
2y3 − αβy0y1y2
2 + αβy2
3),
3 + αβy2y3
3 − y3
0y1 + y2
0y2y3
2 + βγy3
1y3 + αγy1y2
2y3 − αβy0y2y2
3 + αβy1y3
3 − y3
0y2 + y2
0y1y3
0 + βγy2
1 + αγy2
1y3 − αγy0y2
1 + αγy2
0 + βγy2
3),
2 + αβy2
2y3 + αβy1y2y2
2 + αβy2
1y2
3 − αβy0y3
3 + y2
0y1y2 − y3
0y3
3),
0y2
3 + αy2
1 + y2
3 − y2
3 + y3
2y3 − βy0y1y2
2y2
0y2
2,
0y1 − y0y1y2
2
− βγy0y2
βγy3
= (y2y3 − y0y1)(y2
1y2 − αγy0y3
= (y1y3 − y0y2)(y2
1y2 + αγy1y3
= (y1y2 − y0y3)(y2
1y2
1y2y3 + αβy2y3
= (y0y1 − αy2y3)(y2
3 + αβy2
2y2
2 − βγy0y2
− αβy2
− αβy2
2 + βy2
0y2y3 + αy3
2 − βy2
3),
0y1y3 − βy3
2 + αy2
3),
0y2y3 + αy2y3
2 − αy2
2 + y2
3),
0y2
1 − αy2
0y2
0y2
3 − βy2
1 − αy2
3 + βy0y3
1 − y2
0 + βy2
3 − αy0y3
2 + βy2
1 − αy2
0 − y2
1 − γy0y1y2
1 − γy2
0 + γy2
1 + γy2
1y2
2 − αy2
1y2 − αy0y2
0y1y2 + γy3
0 − y2
1 − αy2
2 + αy2
3),
3 + αy2
− αβy1y2
2y3 + αβy1y3
αγy2
= (y0y2 + βy1y3)(y2
1y2y3 − αγy3
2y3 + γy0y3
= (y0y1 + αy2y3)(−y2
1y2
2 − αγy2
0y2
αγy2
3 − γy2
2y2
2 − αγy1y2y2
3 − γy2
αγy1y3
= (y0y3 − γy1y2)(y2
1y3 + αy0y2y2
3 + y3
0y2 − y0y2
1y2
3 − y3
0y1 + y0y1y2
3,
2y2
0y2
0y2
3 + y2
1 − y2
3,
0y3 − y0y2
3 + y3
2y3 + αy0y3
1y3
EXOTIC ELLIPTIC ALGEBRAS
27
1 − y2
0y2y3 + y2
1y2y3
0y1y3 − βy1y3
3 − y3
0y2 + y0y2y2
3
1y3 + βy0y3
3 − y3
0y3 + y0y2
2y3
αy0y1y2
− βγy3
− βγy3
0y1 + y0y3
3 − y3
2 − αy2
3),
2 + βy2
2 + y2
3),
2 − βy0y2
2y3 − αy0y1y2
2 + αy3
3 − αy2y3
= (y0y1 + y2y3)(−y2
1 + αy2
0 + y2
1y3 + βγy1y2
2y3 + γy0y2
1y2 − γy0y3
= (y0y2 − βy1y3)(−y2
1 − γy2
0 + γy2
1y2 + βγy1y2y2
3 − γy2
0y1y2 + γy1y3
1 + y2
= (y0y3 + γy1y2)(−y2
0 − βy2
2 + βy2
3),
− βγy2
1y2
2 + βγy2
1y2
3 − γy2
0y2
2 + γy2
1y2
2 − βy2
0y2
3 + βy2
3 − y3
3 + βy1y3
1y3 + βy0y2y2
1y−βy3
− βy0y2
0 − βy2
= (y0y2 + y1y3)(−y2
1 + y2
2 + βy2
2y3 − y2
1y3 − γy0y2
0 + γy2
1 − γy2
= (y0y3 + y1y2)(−x2
0y1y2 − y3
2 + y2
1y2 − γy1y3
0y2 + y0y3
2 + γy0y2
3),
3).
γy3
1y2
2 − y2
3 − y2
0y2
2 + y2
0y1y3 + y1y2
0y2
3,
2y3
0y3 + y1y2y2
3 + y0y3
3
Some reorganization and changes of sign show that the linear span of the above 15 polynomials is
the same as the linear span of the following 15 polynomials:
1 + αγy2
1 + αγy2
1 + αγy2
2 + αβy2
3)
2 + αβy2
3)
2 + αβy2
3)
0 + βγy2
(y2y3 − y0y1)(y2
0 + βγy2
(y1y3 − y0y2)(y2
0 + βγy2
(y1y2 − y0y3)(y2
0 − y2
(y0y1 + y2y3)(y2
0 − y2
(y0y2 + βy1y3)(y2
0 − y2
(y0y3 − γy1y2)(y2
0 + βy2
(y0y1 − αy2y3)(y2
0 + βy2
(y0y2 + y1y3)(y2
0 + βy2
(y0y3 + γy1y2)(y2
0 − γy2
(y0y1 + αy2y3)(y2
0 − γy2
(y0y2 − βy1y3)(y2
0 − γy2
(y0y3 + y1y2)(y2
3 − αβy2
2y2
αβy2
3 + βy2
1y2
3 + γy2
1y2
2 − βγy2
1y2
βγy2
αγy2
1y2
2 − αγy2
2y2
3 + αy2
2 + αy2
1 − αy2
3)
2 + αy2
1 − αy2
3)
2 + αy2
1 − αy2
3)
2 − βy2
1 − y2
3)
2 − βy2
1 − y2
3)
2 − βy2
1 − y2
3)
2 − y2
1 + γy2
3)
2 − y2
1 + γy2
3)
2 − y2
1 + γy2
3)
1 − βy2
1y2
3 + αy2
0y2
2 + βy2
1y2
2 − γy2
0y2
2y2
3 − γy2
0y2
1 + γy2
1 − y2
2 − αy2
2y2
3 + y2
0y2
0y2
0y2
2,
0y2
2 − y2
0y2
3 + y2
1y2
3 − βy2
0y2
3,
1y2
2 − αy2
0y2
3 + y2
0y2
1 − y2
0y2
3.
The proof of Proposition 9.2 showed that M1(u)θ(u)T = 0 for all u ∈ P so these 15 polynomials
vanish at the points in P. One can also check this directly by evaluating these quartic polynomials
at u ∈ P. For example, it is obvious that yiyj vanishes on P∞ if i 6= j from which it immediately
follows that all 15 polynomials vanish on P∞. As another example, y2y3 − y0y1, y1y3 − y0y2, and
y1y2 − y0y3, vanish on P0, whence the first 3 of the 15 polynomials vanish on P0; the other twelve
polynomials belong to the ideal (y2
0 − y2
3) so they too vanish on P0. As a final
2 + αβy2
example, consider P2. The first three quartics vanish on P2 because y2
3
does. The second three quartics vanish on P2 because y2
0 − y2
3 does. The third three
quartics vanish on P2 because the ideal (y0y1 − αy2y3, y0y2 + y1y3, y0y3 + γy1y2) does. The fourth
three quartics vanish on P2 because y2
3 does. A calculation shows the last three
quartics vanish on P2.
1 − αy2
0 − γy2
1 + αγy2
0 + βγy2
0 − y2
0 − y2
2 − y2
2 + αy2
1 + γy2
2, y2
1, y2
28
ALEX CHIRVASITU AND S. PAUL SMITH
Suppose these 15 quartics vanish at a point u ∈ P3. To complete the proof we will show that
The determinant
u ∈ P.
det
1 βγ αγ αβ
1 −1 −α
α
β −1 −β
1
γ −1
1 −γ
= −(1 + αβ + βγ + γα)2
is non-zero: the hypothesis that α+β +γ +αβγ = 0 implies 1+αβ +βγ +γα = (1+α)(1+β)(1+γ)
which is non-zero because we are assuming that {α, β, γ} ∩ {0,±1} = ∅. Because the determinant
is non-zero the polynomials
1 + αγy2
2 + αβy2
3,
0 + βγy2
y2
0 − y2
y2
0 + βy2
y2
0 − γy2
y2
1 − αy2
1 − y2
1 + γy2
2 + αy2
3,
2 − βy2
3,
2 − y2
3,
(9-3)
(9-4)
(9-5)
(9-6)
(9-7)
(9-8)
are linearly independent. Their linear span is therefore the same as that of {y2
3}. Hence
the common zero locus of the polynomials in (9-3) is empty and at most three of them vanish at u.
1, y2
2, y2
0, y2
We now do some case-by-case analysis to show that u belongs to some Pi.
2 + αβy2
P∞ ∪ P0. Suppose u is not in the zero locus of y2
1 + αγy2
0 + βγy2
3. Then
y0y1 − y2y3 = y0y2 − y1y3 = y0y3 − y1y2 = 0
at u. If one of the coordinate functions y0, y1, y2, y3 vanishes at u, then three of do so
If none of y0, y1, y2, y3 vanishes at u, then it follows from (9-4) that
u ∈ {(1, 0, 0, 0), (0, 1, 0, 0), (0, 0, 1, 0), (0, 0, 0, 1)} = P∞.
u ∈ {(1, 1, 1, 1), (1, 1,−1, −1), (1, −1, 1,−1), (1,−1, −1, 1)} = P0.
P1. Suppose u is not in the zero locus of y2
0 − y2
1 − αy2
2 + αy2
3 and not in P∞ ∪ P0. Then
y0y1 + y2y3 = y0y2 + βy1y3 = y0y3 − γy1y2 = 0
If one of y0, y1, y2, y3 vanishes at u, then three of them do so u ∈ P∞. This is not the
at u.
case so none of y0, y1, y2, y3 vanishes at u. Without loss of generality we can, and do, assume that
0(y1y2y3) = βγ(y1y2y3)2. Therefore bc = y1y2y3. It
u = (bc, y1, y2, y3). It follows from (9-6) that y3
2 = −βy2
also follows from (9-6) that βγy2
3. Some case-by-case analysis shows that
1 = γy2
u ∈ {(bc,−i, ib, c), (bc,−i,−ib,−c), (bc, i, ib, −c), (bc, i,−ib, c)} = P1.
P2. Suppose u is not in the zero locus of y2
0 + βy2
1 − y2
2 − βy2
3 and not in P∞ ∪ P0. Then
y0y1 − αy2y3 = y0y2 + y1y3 = y0y3 + γy1y2 = 0
at u. As in the previous paragraph, y0y1y2y3 does not vanish at u. Without loss of generality we
can, and do, assume that u = (ac, y1, y2, y3). The same sort of analysis as that in the previous
paragraph shows that
u ∈ {(ac, a,−i, ic), (ac, a, i,−ic), (ac,−a, −i,−ic), (ac,−a, i, ic) = P2.
P3. Suppose u is not in the zero locus of y2
0 − γy2
1 + γy2
2 − y2
3 and not in P∞ ∪ P0. Then
at u. Proceeding as before, we eventually see that
y0y1 + αy2y3 = y0y2 − βy1y3 = y0y3 + y1y2 = 0
u ∈ {(ab, ia, b,−i), (ab, ia,−b, i), (ab,−ia, b, i), (ab,−ia, −b,−i)} = P3.
This completes the proof that pr1(P) ⊂ P. Thus pr2(P) = P.
EXOTIC ELLIPTIC ALGEBRAS
29
By Lemma 9.1, pr2(P) = P also. Since pr2(P) does not contain a line, the rank of M1(u) is 3 for
all u ∈ pr1(P). Let u ∈ P. Since M1(u)θ(u)T = 0, θ(u)T is the only v ∈ P3 such that M1(u)vT = 0.
Hence (u, θ(u)) is the only point in pr−1
1 (u). It follows that P = {(u, θ(u)) u ∈ P}.
(cid:4)
0 + y2
1 + y2
2 + y2
3 does not annihilate any point modules
Proposition 9.5. The central element Θ = y2
for eQ. Consequently, eB has no point modules.
Proof. Let u ∈ P.
To describe the action of Θ on Mu we must fix a basis for Mu. We pick a basis for Mu that
is compatible with the entries in Table 4. To do this it is helpful, for a moment, to think of the
entries in Table 4 as points in k4. Suppose u = (δ0, δ1, δ2, δ3). Let e0 be any non-zero element in
(Mu)0. Let e1 be the unique element in (Mu)1 such that yie0 = δie1 for i = 0, 1, 2, 3. Likewise,
if (δ′0, δ′1, δ′2, δ′3) is the entry in Table 4 for θ(u), there is a unique element e2 ∈ (Mu)2 such that
yie1 = δ′ie2 for i = 0, 1, 2, 3.
If u ∈ P∞, then Θe0 = e2. If u ∈ P0, then Θe0 = 4e2.
Let u = (bc,−i,−ib,−c) ∈ P1. Then θ(u) = (bc,−i, ib, c). Therefore
Θe0 = (y2
0 + y2
1 + y2
2 + y2
3)e0
= (bcy0 − iy1 − iby2 − cy3)e1
= (cid:0)(bc)2 − 1 + b2 − c2(cid:1)e2
= (β − 1)(γ + 1)e2
Likewise, if u = (bc, i,−ib, c) ∈ P1, then θ(u) = (bc, i, ib,−c) and a similar calculation shows that
Θe0 = (β − 1)(γ + 1)e2. Thus, Θe0 = (β − 1)(γ + 1)e2 for all u ∈ P1.
Similar calculations show that Θe0 = (α + 1)(γ − 1)e2 for all u ∈ P2. Finally, if u ∈ P3, then
Θe0 = (α − 1)(β + 1)e2.
(cid:4)
9.4. Not only do the relations for eQ determine P, but P determines the defining relations for
eQ: the quadratic relations for eQ are precisely the elements of V ⊗2 that vanish at P. This is a
consequence of the following remarkable result.
Theorem 9.6 (Shelton-Vancliff). [27] Let V be a 4-dimensional vector space and R ⊆ V ⊗2 a 6-
dimensional subspace. Let T V denote the tensor algebra on V and let P ⊂ P(V ∗) × P(V ∗) be the
scheme-theoretic zero locus of R. If dim(P) = 0, then
R = {f ∈ V ⊗2 fP = 0}.
9.5. There has been some interest in Artin-Schelter regular algebras with Hilbert series (1 − t)−4
that have only finitely many point modules [41], [26], [36], [37]. The interest arises because this
phenomenon does not occur for Artin-Schelter regular algebras with Hilbert series (1 − t)−3; the
point modules for the latter algebras are parametrized either by a cubic divisor in P2 or by P2. In
1988, M. Van den Bergh circulated a short note showing that a generic 4-dimensional Artin-Schelter
regular algebra with Hilbert series (1 − t)−4 has exactly 20 point modules [10]. Van den Bergh's
example is a generic Clifford algebra. In particular, it is a finite module over its center.
Davies [8, §5.1] shows, when the translation automorphism has infinite order, that eQ is not
Proposition 9.7. The point modules Mu for u ∈ P∞ ∪ P0 are quotient rings of eQ.
isomorphic to any of the previously found examples of 4-dimensional regular algebras having 20
point modules.
(λ0, λ1, λ2, λ3) ∈ P∞ ∪ P0, then
If u =
Mu ∼=
(λjyi − λiyj 0 ≤ i, j ≤ 3) ∼= k[t].
eQ
30
ALEX CHIRVASITU AND S. PAUL SMITH
scheme with 20 points.
Proposition 9.8. The scheme-theoretic zero locus in P3 × P3 of the relations for eQ is a reduced
Proof. (Van den Bergh [10].) We have already seen that the relations for eQ vanish at 20 points in
P3 × P3. Let X denote the image of the Segre embedding P3 × P3 → P15. If we view P15 as the
space of 4× 4 matrices, then X is the space of rank-one matrices. By [12, §18.15], for example, the
degree of X is(cid:0)6
3(cid:1) = 20. The 6 defining relations for eQ are linear combinations of terms xixj which,
under the Segre embedding, become linear combinations of the coordinate functions xij. Hence
the vanishing locus of the relations in P15 is the vanishing locus of 6 linear forms, hence a linear
subspace, L say, of dimension 9. Hence, by B´ezout's Theorem, if the scheme-theoretic intersection
L ∩ X is finite it has degree 20. But, L ∩ X consists of 20 different points so it is reduced.
(cid:4)
10. Secant lines to E and line modules for Q
The relevance of this section will become apparent in §11 when we construct some line modules
for eQ that are parametrized by certain lines in P(Q∗1). To make the word "parametrized" precise
we will show that the parametrizing space is a closed subvariety of the Grassmannian of lines in
P(Q∗1).
10.1. Secant lines. The second symmetric power of E is the quotient variety S2E := (E × E)/Z2
where Z2 acts by (p, q) 7→ (q, p). We think of the points in S2E as effective divisors of degree 2 on
E and write (p) + (q) for the image of (p, q) ∈ E × E.
Because the quartic curve E ⊂ P(Q∗1) = P3 has no trisecants, there is a well-defined morphism
E × E → G(1, 3) that sends (p, q) ∈ E × E to pq, the line in P(Q∗1) = P3 whose scheme-theoretic
intersection with E is (p)+(q). By the universal property of the quotient (E×E)/Z2 this morphism
factors through a morphism γ : S2E → G(1, 3). The image of γ is a closed subscheme of G(1, 3)
called the variety of secant lines to E. See [12, Ex. 8.3], for example.
Proposition 10.1. The map γ : S2E → G(1, 3) defined by γ(cid:0)(p)+(q)(cid:1) := pq is a closed immersion.
Proof. The morphism γ is injective because E has no trisecants, so it suffices to argue that the
image of the morphism is smooth. This follows from the standard description of the singular points
of a secant variety: a line in the image of γ is singular if and only if it is a trisecant (see e.g. the
discussion on page 312 of [12] regarding Exercise 16.11 in that book).
(cid:4)
10.2. The line modules Mp,q. A line module for Q, or eQ, is a cyclic graded module whose Hilbert
series is (1 − t)−2.
Theorem 10.2. [16, Thm. 4.5] The function that sends (p) + (q) ∈ S2E to Q/Qx + Qx′ where
pq = {x = x′ = 0} is a bijection from S2E to the set of isomorphism classes of line modules for Q.
10.3.
If (p) + (q) ∈ S2E and pq = {x = x′ = 0} we write Mp,q := Q/Qx + Qx′.
In §11 we will show that if y = y′ = 0 is a line in P(eQ∗1) = P(eQ∗1) that meets E at
(p) + (p + ξ) for some p ∈ E and ξ ∈ E[2] − {o}, then eQ/eQy + eQy′ is a line modules for eQ.
Such lines will be parametrized by the subscheme of G(1, 3) that is the image of the composition
E/hξi → S2E → G(1, 3).
Lemma 10.3. The morphism β : E/hξi → S2E defined by β(p + hξi) = (p) + (p + ξ) is a closed
immersion.
Proof. It is clear that β is injective as a set map on the closed points of E/hξi, so it suffices to
prove that its derivative is one-to-one on each tangent space, or equivalently that the composition
of β with the ´etale map π : E → E/hξi has this same property.
The composition βπ is
EXOTIC ELLIPTIC ALGEBRAS
31
where the left hand arrow sends p to (p, p + ξ) and the right hand arrow is the quotient morphism.
Since the latter is ´etale off the diagonal ∆ ⊂ E × E and the former is a closed immersion into
E × E \ ∆ the conclusion follows.
(cid:4)
E → E × E → S2E,
11. Line Modules for eQ
In this section we exhibit three families of line modules for eQ parametrized by the disjoint
11.1.
union of the three elliptic curves E/hξi as ξ ranges over the three 2-torsion points of E. The
isomorphism classes of the line modules parametrized by E/hξi are in natural bijection with the
lines p, p + ξ, p ∈ E; the union of these lines is an elliptic scroll in P(S∗1).
These are not all the line modules for eQ.
labelled in such a way that γ∗i Mp,q ∼= Mp+ξi,q+ξi. Thus, γ∗(cid:0)Mp,q ⊕ Mr,s(cid:1) ∼= Mp,q ⊕ Mr,s for all γ ∈ Γ
11.2. By Proposition 7.7, the elements in Γ = {γ0, γ1, γ2, γ3} and E[2] = {ξ0, ξ1, ξ2, ξ3} may be
if and only if {p, q, r, s} is an E[2]-coset.
11.3. Recall that Q′ = Q ⊗ M2(k). The next result follows from Proposition 3.9.
Proposition 11.1. The function M 7→ M Γ is a bijection from isomorphism classes of Γ-equivariant
Q′-modules with Hilbert series 4(1 − t)−2 to isomorphism classes of eQ-line modules.
By Morita equivalence, a Γ-equivariant Q′-module M with Hilbert series 4(1− t)−2 is isomorphic
to N ⊗ k2 for some Q-module N with Hilbert series 2(1 − t)−2 (a "fat line" of multiplicity two
over Q). Moreover, by the remark in §11.2, the equivariance ensures/requires that the isomorphism
class of M is invariant under translation by the 2-torsion subgroup.
The main ingredient in constructing eQ-lines will be Q-modules with Hilbert series 2(1 − t)−2.
The obvious such modules are those of the form Mp,q⊕ Mr,s where the invariance condition requires
{p, q, r, s} to be an E[2]-coset. Theorem 11.6 will provide the examples announced in §11.1.
Lemma 11.2. Let x, y ∈ E/E[2] and let ξ and ω be 2-torsion points. Define
(11-1)
Mx,ξ := (cid:0)Mp,p+ξ ⊕ Mp+ξ ′,p+ξ ′′(cid:1) ⊗ k2
where p is any point in E such that x = p + E[2].
(1) The Q′-module Mx,ξ does not depend on the choice of p.
(2) Mx,ξ ∼= My,ω if and only if (x, ξ) = (y, ω).
(3) The map Φ : k× × k× → AutQ′(cid:0)Mx,ξ(cid:1), Φ(λ, λ′)(m, m′) := (λm, λ′m′), is an isomorphism.
Proof. Let E[2] = {o, ξ, ξ′, ξ′′}.
(1) Suppose x is also the image of q ∈ E. Since ξ′ + ξ′′ = ξ,
(cid:8){q, q + ξ},{q + ξ′, q + ξ′′}(cid:9) = (cid:8){p, p + ξ},{p + ξ′, p + ξ′′}(cid:9).
Therefore Mp,p+ξ ⊕ Mp+ξ ′,p+ξ ′′ = Mq,q+ξ ⊕ Mq+ξ ′,q+ξ ′′. Hence Mx,ξ does not depend on the choice
of p. In particular, if (x, ξ) = (y, ω), then Mx,ξ = My,ω.
(2) Suppose that the Q′-modules Mx,ξ and My,ω are isomorphic. Let q ∈ E be be such that
y = q + E[2]. By Morita equivalence, there is an isomorphism of Q-modules
Mp,p+ξ ⊕ Mp+ξ ′,p+ξ ′′ ∼= Mq,q+ω ⊕ Mq+ω′,q+ω′′
where E[2] = {o, ω, ω′, ω′′}. Since isomorphism classes of line modules for Q are in natural bijection
with effective divisors of degree 2 on E,
(cid:8){q, q + ω},{q + ω′, q + ω′′}(cid:9) = (cid:8){p, p + ξ},{p + ξ′, p + ξ′′}(cid:9).
32
ALEX CHIRVASITU AND S. PAUL SMITH
It follows immediately from this equality that q + E[2] = p + E[2], i.e., x = y. Since ω can be
(3) Every line module for Q is cyclic so its graded automorphism group is isomorphic to k×, each
recovered from(cid:8){q, q + ω},{q + ω′, q + ω′′}(cid:9) as the difference between the elements in {q, q + ω}
and also as the difference between the elements in {q + ω′, q + ω′′}, it follows that ω = ξ.
λ ∈ k× acting on the line module by scalar multiplication.
By Morita equivalence, AutQ′(Mx,ξ) = AutQ(Mp,p+ξ ⊕ Mp+ξ ′,p+ξ ′′) ∼= k× × k× where the iso-
morphism is because Mp,p+ξ 6∼= Mp+ξ ′,p+ξ ′′. An automorphism (λ, λ′) ∈ (k×)2 acts on Mx,ξ =
(Mp,p+ξ ⊗ k2) ⊕ (Mp+ξ ′,p+ξ ′′ ⊗ k2) as multiplication by λ on the first summand and multiplication
by λ′ on the second summand.
Lemma 11.3. Let E[2] = {o, ξ, ξ′, ξ′′}. Let x ∈ E/E[2] and write M = Mx,ξ.
(1) If γ ∈ Γ, then γ∗M ∼= M as Q′-modules.
(2) If γ ∈ Γ and a ∈ AutQ′(M ), then there is a unique element γ ⊲ a ∈ AutQ′(M ) such that
(cid:4)
ϕγ
M
γ⊲a
γ∗M
γ ∗(a)
γ∗M
commutes for all isomorphisms ϕγ : M → γ∗M .
M
ϕγ
(3) The map (γ, a) 7→ γ ⊲ a defines a left action of Γ on AutQ′(M ).
(4) If we identify k× × k× with AutQ′(cid:0)Mx,ξ(cid:1) via the isomorphism Φ in Lemma 11.2, then the
Γ-action on AutQ′(M ) is
ξ ⊲ (λ, λ′) = (λ, λ′)
and
ξ′ ⊲ (λ, λ′) = ξ′′ ⊲ (λ, λ′) = (λ′, λ)
for all (λ, λ′) ∈ k× × k×.
Proof. Let p ∈ E be such that x = p + E[2]. Thus M =(cid:0)Mp,p+ξ ⊕ Mp+ξ ′,p+ξ ′′(cid:1) ⊗ k2.
(1) This follows from the remark in §11.2.
(2) Choose an isomorphism ϕγ : M → γ∗M . Define γ ⊲ a := ϕ−1
γ γ∗(a)ψγ = ϕ−1
γ γ∗(a)ϕγ. Certainly the diagram
commutes. If ψγ : M → γ∗M is another isomorphism, then ψγ is a multiple of ϕγ by an element
in AutQ′(γ∗M ). But AutQ′(γ∗M ) is abelian so ψ−1
γ γ∗(a)ϕγ .
(3) This is standard. See, for example, Lemma A.1.
(4) By Proposition 7.7, ξ∗Mp,p+ξ ∼= Mp,p+ξ and ξ∗Mp+ξ ′,p+ξ ′′ ∼= Mp+ξ ′,p+ξ ′′ so ϕξ preserves the
summands Mp,p+ξ ⊗ k2 and Mp+ξ ′,p+ξ ′′ ⊗ k2. Therefore ξ acts on (k×)2 trivially. On the other
hand, (ξ′)∗Mp,p+ξ ∼= (ξ′′)∗Mp,p+ξ ∼= Mp+ξ ′,p+ξ ′′ so ξ′ and ξ′′ act on (k×)2 by switching the two
components.
(cid:4)
A Γ-equivariant structure on a Q′-module M is the same thing as a left Q′-module M endowed
with a left action Γ × M → M , (γ, m) 7→ mγ, such that (xm)γ = γ(x)mγ for all x ∈ Q′, m ∈ M ,
and γ ∈ Γ. We adopt this point of view several times in the rest of this section.
Recall that the action of Γ as automorphisms of Q′ is defined in terms of the actions of Γ as
automorphisms of Q and M2(k) (see §6.4).
Lemma 11.4. Let N be a graded left Q-module that is generated by N0. The function that sends
a Γ-equivariant structure {ϕγ : N ⊗ k2 −→ γ∗(N ⊗ k2) γ ∈ Γ} on the Q′-module N ⊗ k2 to the
is injective.
Proof. Certainly, if the maps {ϕγ : N ⊗ k2 −→ γ∗(N ⊗ k2) γ ∈ Γ} give N ⊗ k2 the structure of
a Γ-equivariant Q′-module, then their restrictions to the degree zero components give N0 ⊗ k2 the
structure of a Γ-equivariant M2(k)-module.
Γ-equivariant structure {ϕγ(cid:12)(cid:12)N0⊗k2 : N0⊗ k2 −→ γ∗(N0 ⊗ k2) γ ∈ Γ} on the M2(k)-module N0⊗ k2
EXOTIC ELLIPTIC ALGEBRAS
33
Since Q′ is generated as an algebra by Q′0 and Q′1, the formula (xm)γ = γ(x)mγ implies that
the action of γ on Nn+1 ⊗ k2 is completely determined by the action of γ on Nn ⊗ k2. Thus, if two
Γ-equivariant structures on N ⊗ k2 agree on N0 ⊗ k2, then they agree on N .
11.3.1. Warning. The result in Lemma 11.4 does not extend to a result saying that two equivari-
ant structures on N are isomorphic if and only if their restrictions to N0 ⊗ k2 are isomorphic.
Proposition 11.5 says that all Γ-equivariant structures on N0 ⊗ k2 are isomorphic to each other.
The group Γ acts as k-algebra automorphisms of M2(k), We fixed a basis for k2 such that ω ∈ Γ
acts on M2(k) as conjugation by the quaternionic basis element qω defined in §6.4. We use that
basis in the next result.
Proposition 11.5. Fix ζ, η, ξ ∈ Γ such that qζ, qη, qξ is a cyclic permutation of q1, q2, q3.
(1) Let φω : M2(k) → M2(k), ω ∈ Γ, be the linear isomorphisms that take the following values
(cid:4)
on the basis 1, qζ, qη, qξ for M2(k):
q
φ0(q)
φζ(q)
φη(q)
φξ(q)
qη
qη
1
1
1
1 −qζ
1 −qζ −qη
qζ
qξ
qζ
qξ
qζ −qη −qξ
qη −qξ
qξ
Table 4. Action of Γ on M2(k)
The action of Γ on M2(k) given by the maps φω, together with the action of M2(k) on M2(k)
by left multiplication, gives M2(k) the structure of a Γ-equivariant left M2(k)-module.
(2) Every Γ-equivariant M2(k)-module is isomorphic to a direct sum of copies of the Γ-equivariant
M2(k)-module in (2).
(3) Let V be a finite dimensional Γ-equivariant M2(k)-module. As a Γ-module, V is isomorphic
to a direct sum of copies of the regular representation. If ω ∈ {ζ, η, ξ}, then the (+1)- and
(−1)-eigenspaces for the action of ω on V have dimension 1
2 dimk(V ).
Proof. (1) Whenever a group Γ acts as automorphisms of a ring R, R viewed as left R-module via
multiplication is a Γ-equivariant R-module with respect to the action of Γ as automorphisms of R.
The value of φω(qω′) in the table is qωqω′q−1
ω so, by the previous sentence, this action of Γ makes
M2(k) a Γ-equivariant M2(k)-module.
(2) By Lemma 3.1, there is an equivalence from the category of Γ-equivariant M2(k)-modules
to the category of vector spaces, the functor implementing the equivalence being M M Γ. Since
M2(k)Γ ∼= k, the result follows.
Alternatively, a Γ-equivariant left M2(k)-module is the same thing as a left module over the skew
group ring M2(k) ⋊ Γ which has dimension16; the Γ-equivariant M2(k)-module in (2) is irreducible
of dimension 4 so we conclude that M2(k) ⋊ Γ ∼= M4(k). The result follows.
(3) follows from (2) because M2(k) is isomorphic as a Γ-module to the regular representation. (cid:4)
Theorem 11.6. Let E[2] = {o, ξ, ξ′, ξ′′}. Let M be the Q′-module (Mp,p+ξ ⊕ Mp+ξ ′,p+ξ ′+ξ) ⊗ k2.
(1) There are exactly two Γ-equivariant structures on M up to isomorphism.
(2) The group H 1(Γ, AutQ′(M )) acts simply transitively on this two-element set.
(3) Up to isomorphism one equivariant structure is obtained from the other by interchanging
the (+1)- and (−1)-eigenspaces for the action of ξ on M and simultaneously interchanging
the (+1)- and (−1)-eigenspaces for the action of ξ + ξ′ on M , and leaving the (+1)- and
(−1)-eigenspaces for the action of ξ′ unchanged.
34
ALEX CHIRVASITU AND S. PAUL SMITH
Proof. If x = p + E[2], then M is the module Mx,ξ in Lemmas 11.2 and 11.3.
Step 1: Existence of an equivariant structure. Let ϕγ : M → γ∗M , γ ∈ Γ, be arbitrary
Q′-module isomorphisms. An arbitrary choice of such isomorphisms need not give an equivariant
structure on M ; i.e., there is no reason the diagrams (8-3) should commute. The failure of (8-3) to
commute is measured by the elements
(11-2)
aγ,δ := ϕ−1
γδ ◦ tγ,δ ◦ γ∗(ϕδ) ◦ ϕγ,
γ, δ ∈ Γ
in AutQ′(M ) where tγ,δ is as in (8-3) and the right-hand side of (11-2) is the clockwise composition
of the automorphisms in (8-3).
A tedious calculation (see Lemma A.2) shows that the function (γ, δ) 7→ aγ,δ is a 2-cocycle for
Γ valued in the Γ-module AutQ′(M ) ∼= (k×)2 defined in Lemma 11.3. Let ξ′ ∈ Γ − hξi. Since
Γ = hξi × hξ′i it follows from the Hochschild-Serre spectral sequence
(11-3)
Ea,b
2 = H a(hξi, H b(hξ′i, (k×)2)) ⇒ H a+b(Γ, (k×)2)
γ now form an equivariant structure on M .
and the cohomology of Z/2 that H 2(Γ, (k×)2) is trivial. Hence the obstruction cocycle (aδ,γ) is
cohomologous to zero. Thus (aδ,γ) is the coboundary of some function Γ → AutQ′(M ), γ 7→ aγ;
the isomorphisms ϕγa−1
Step 2: Classification of equivariant structures. By Step 1, there is at least one Γ-
equivariant structure on M . Suppose the maps ϕγ : M → γ∗M , γ ∈ Γ, provide such an equivariant
structure.
Let (ψγ)γ∈Γ be another equivariant structure on M . Running through the compatibility condi-
tions comprising equivariance, the maps aγ = (ϕγ)−1ψγ can be seen to form a 1-cocycle of Γ valued
in the Γ-module AutQ′(M ) ∼= (k×)2. We similarly leave it to the reader to check that cocycles (aγ)
and (a′γ) give rise to isomorphic equivariant structures
ψγ = ϕγaγ and ψ′γ = ϕγa′γ
if and only if they are cohomologous. In other words, the set of isomorphism classes of equivariant
structures on M is acted upon simply and transitively by H 1(Γ, (k×)2). Using the Hochschild-Serre
spectral sequence once more we get H 1(Γ, (k×)2) ∼= Z/2 (see the proof of (3) below).
This completes the proof of (1) and (2).
(3) The Hochschild-Serre spectral sequence yields an isomorphism
(11-4)
H 1(Γ, Aut(M )) ∼= H 1(cid:0)hξi, H 0(hξ′i, (k×)2)(cid:1) ⊕ H 0(cid:0)hξi, H 1(hξ′i, (k×)2)(cid:1).
Since ξ′ interchanges the two copies of k×, the H 1 term in the second summand vanishes so we are
left with a natural isomorphism
H 1(Γ, Aut(M )) ∼= H 1(hξi, k×) ∼= HomZ(hξi, k×),
where this time k× is the diagonal subgroup of AutQ′(M ).
The function f : Γ → AutQ′(M ) defined by f (ξ) = f (ξ′ + ξ) = (−1,−1) and f (o) = f (ξ′) = (1, 1)
is a 1-cocycle whose class [f ] in H 1(Γ, Aut(M )) is non-trivial.
If the Q′-module isomorphisms
{φγ : M → γ∗M γ ∈ Γ} give M a Γ-equivariant structure, then the Γ-equivariant structure on M
associated to the result of [f ] acting on the given equivariant structure is given by the isomorphisms
{φγ ◦ f (γ) : M → γ∗M γ ∈ Γ}. Recall that γ∗M is M as a graded vector space. The (+1)-
eigenspace for the action of ξ on M with equivariant structure {φγ}γ∈Γ is {m ∈ M φξ(m) =
m} which is the (−1)-eigenspace for φξ ◦ f (ξ). Likewise, the (−1)-eigenspace for the action of
φξ+ξ ′ ◦ f (ξ + ξ′) is the (+1)-eigenspaces for the action of φξ+ξ ′. On the other hand, the eigenspaces
for ξ′ are the same for both equivariant structures on Mx,ξ.
(cid:4)
EXOTIC ELLIPTIC ALGEBRAS
35
11.3.2. There is a lack of symmetry in part (3) of Theorem 11.6: the eigenspaces for ξ + ξ′ are
switched but those for ξ′ are not. The explanation is that the equivariant structure obtained by
interchanging the eigenspaces for ξ′ but not ξ + ξ′ (but still exchanging the eigenspaces for ξ) is
isomorphic to that obtained by switching the eigenspaces for ξ + ξ′ but not those for ξ′.
11.3.3. The proof of Theorem 11.6 illustrates a familiar pattern in obstruction theory. The class
of structures we are interested in, isomorphism classes of equivariant structures in this case, is a
pseudotorsor over a cohomology group. Whether or not it is empty is controlled by an obstruction
living in a cohomology group, H 2 for us, as in Step 1 of the proof, and when this obstruction
vanishes the cohomology group of one degree lower, H 1 in our case, acts on the class of structures
simply transitively.
11.4. An explicit equivariant structure on Mx,ξ. Let {ξ1, ξ2, ξ3} denote both the 2-torsion
points on E and the corresponding elements in Γ, labelled so that the action of Γ as automorphisms
of M2(k) is such that each ξj acts as conjugation by the element qj in (6-2).
Let p ∈ E and let x = p + hξ1i ∈ E/hξ1i. Let M = Mx,ξ1 = (Mp,p+ξ1 ⊕ Mp+ξ2,p+ξ3) ⊗ k2. Fix a
basis e for the degree-zero component of Mp,p+ξ1 and a basis e′ for the degree-zero component of
Mp+ξ2,p+ξ3.
If u =(cid:0) 0
1(cid:1) and v =(cid:0) 1
0(cid:1), then
q1u = −iu,
q1v = iv,
q2u = iv,
q2v = iu,
q3u = −v,
q3v = u.
Lemma 11.7. Let β0x0 + β1x1 + β2x2 + β3x3 be a linear form that vanishes at p and p + ξ1. Then
(1) the line through p and p + ξ1 is β0x0 + β1x1 = β2x2 + β3x3 = 0,
(2) the line through p + ξ2 and p + ξ3 is β0x0 − β1x1 = β2x2 − β3x3 = 0,
(3) β0y0 + iβ1y1 and iβ2y2 + β3y3 annihilate e ⊗ u + e′ ⊗ v and are linearly independent, and
(4) β0y0 − iβ1y1 and iβ2y2 − β3y3 annihilate e ⊗ v + e′ ⊗ u and are linearly independent.
Proof. By Lemma 8.5, at least three of the coordinate functions x0.x1, x2, x3 are non-zero at p.
Thus (β0, β1) 6= (0, 0) and (β2, β3) 6= (0, 0). Therefore the equations in (2) and (3) really do define
lines in P(Q∗1). It also follows that β0y0 + iβ1y1 and iβ2y2 + β3y3 are linearly independent.
(1) Translation by ξ1 leaves the set {p, p + ξ1} stable so ξ1(β0x0 + β1x1 + β2x2 + β3x3) also
vanishes at p and p + ξ1. Since ξ1(β0x0 + β1x1 + β2x2 + β3x3) = β0x0 + β1x1 − β2x2 − β3x3, (1)
follows.
(2) Since translation by ξ2 sends {p, p+ξ1} to {p+ξ2, p+ξ3}, ξ2(β0x0 +β1x1) and ξ2(β2x2 +β3x3)
(3) Since
vanish at p + ξ2 and p + ξ3. Thus (2) is true.
y0 · (e ⊗ u + e′ ⊗ v) = (x0 ⊗ q0) · (e ⊗ u + e′ ⊗ v) = x0e ⊗ u + x0e′ ⊗ v,
y1 · (e ⊗ u + e′ ⊗ v) = (x1 ⊗ q1) · (e ⊗ u + e′ ⊗ v) = −ix1e ⊗ u + ix1e′ ⊗ v,
y2 · (e ⊗ u + e′ ⊗ v) = (x2 ⊗ q2) · (e ⊗ u + e′ ⊗ v) = ix2e ⊗ v + ix2e′ ⊗ u, and
y3 · (e ⊗ u + e′ ⊗ v) = (x3 ⊗ q3) · (e ⊗ u + e′ ⊗ v) = −x3e ⊗ v + x3e′ ⊗ u,
(β0y0 + iβ1y1 − iβ2y2 − β3y3) · (e ⊗ u + e′ ⊗ v) equals
(β0x0 + β1x1)e ⊗ u + (β2x2 + β3x3)e ⊗ v + (β2x2 − β3x3)e′ ⊗ u + (β0x0 − β1x1)e′ ⊗ v.
Since e ∈ (Mp,p+ξ1)0 it follows from (1) that (β0x0 + β1x1)e = (β2x2 + β3x3)e = 0. Since e′ ∈
(Mp+ξ2,p+ξ3)0 it follows from (2) that (β0x0 − β1x1)e′ = (β2x2 − β3x3)e′ = 0. Therefore (3) is true.
The proof of (4) is similar.
(cid:4)
36
ALEX CHIRVASITU AND S. PAUL SMITH
φ1
φ2
e′ ⊗ u e′ ⊗ v
e ⊗ u
e ⊗ u −e ⊗ v −e′ ⊗ u e′ ⊗ v
e′ ⊗ v
e ⊗ u
Table 5. Equivariant structure on M0
e ⊗ v
e′ ⊗ u
e ⊗ v
Let φ0 be the identity map on M0 and let φ1, φ2 ∈ GL(M0) be the linear automorphisms which
act on the basis {e ⊗ u, e ⊗ v, e′ ⊗ u, e′ ⊗ v} as in Table 5.
Let φ3 = φ1φ2.
The following observation is elementary.
Lemma 11.8. Let a be an element in a ring R such that a2 = 1. There is a group homomorphism
Z/2 → Aut(R) given by sending the non-identity element to the automorphism b 7→ aba−1. Let
M be a left R-module and define the group homomorphism Z/2 → AutZ(M ) by sending the non-
identity element to the automorphism m 7→ am. This action of Z/2 makes M a Z/2-equivariant
R-module.
Theorem 11.9. Let each ξi act on M0 as the linear map φi in Table 5.
(1) This action of Γ on M0 extends to an action of Γ on M that makes M a Γ-equivariant
Q′-module.
Proof. (1) We will use Lemma 11.8 to show that M0 is a Γ-equivariant M2(k)-module.
(3) If β0x0 + β1x1 = β2x2 + β3x3 = 0 is the line in P(Q∗1) that passes through p and p + ξ1, then
(2) The eQ-line module M Γ is generated by e ⊗ u + e′ ⊗ v.
the line in P(eQ∗1) corresponding to M Γ is β0y0 + iβ1y1 = iβ2y2 + β3y3 = 0.
First, consider the action of ξ1 by φ1 on e ⊗ k2. With respect to the ordered basis {e ⊗ u, e ⊗ v},
ξ1 acts on e ⊗ k2 as multiplication by 1 ⊗(cid:0)1
0 −1(cid:1). The action of ξ1 on M2(k) is b 7→ q1bq−1
1 .
0 −1(cid:1), Lemma 11.8 tells us that e ⊗ k2 is a
Since conjugation by q1 is the same as conjugation by(cid:0)1
Now consider the action of ξ1 by φ1 on e′⊗ k2. With respect to the ordered basis {e⊗ u, e⊗ v}, ξ1
0 1(cid:1). Since conjugation by q1 is the same as conjugation
acts on e′ ⊗ k2 as multiplication by 1 ⊗(cid:0)−1 0
0 1(cid:1), Lemma 11.8 tells us that e′ ⊗ k2 is a hξ1i-equivariant M2(k)-module.
by(cid:0)−1 0
Thus, M0 is hξ1i-equivariant M2(k)-module. A similar argument shows that M0 is a hξji-
equivariant M2(k)-module for the other j's. Since {φ0, φ1, φ2, φ3} is a subgroup of GL(M0) iso-
morphic to Γ, these Z/2-equivariant structures fit together to make M0 = (e ⊗ k2) ⊕ (e′ ⊗ k2) a
Γ-equivariant M2(k)-module.
hξ1i-equivariant M2(k)-module.
0
0
To extend the equivariant structure to all of M , simply define automorphisms φi of M by
φi(am) = ξi(a)φi(m),
∀a ∈ Q′, m ∈ M0.
That this action is well-defined boils down to checking that whenever a ∈ Q′ annihilates m ∈ M0,
ξi(a) annihilates φi(m). For this it suffices to assume that m is an eigenvector of φi (since M0
breaks up as a direct sum of Γ-eigenspaces), and hence to prove that
∀a ∈ Q′, m ∈ M0.
am = 0 ⇒ ξi(a)m = 0,
The conclusion follows from the fact that all twists ξ∗i M are isomorphic to M as Q′-modules
(because we already know there are equivariant structures on M ).
is in M Γ
0 so it generates the eQ-line module M Γ.
(2) By Proposition 11.1, M Γ is a line module for eQ. One sees from Table 1 that e ⊗ u + e′ ⊗ v
(3) The correspondence between line modules for eQ and lines in P(eQ∗1) is given by sending a line
module eQ/eQy + eQy′ to the line y = y′ = 0. Thus, (3) follows from Lemma 11.7(3).
(cid:4)
11.5. 3 elliptic curves parametrizing some line modules. Let G(1, 3) be the Grassmannian
EXOTIC ELLIPTIC ALGEBRAS
37
class of L is determined by the isomorphism class of L/L≥2. Thus, there is a well-defined map
{isomorphism classes of line modules for eQ} −→ G(1, 3).
G(1, 3) ←→ {isomorphism classes of cyclic graded eQ-modules with Hilbert series 1 + 2t}
of lines in P(eQ∗1). There is a bijection
given by the function sending a line y = y′ = 0 to the module eQ/eQy + eQy′ + eQ≥2 and its inverse
which sends a cyclic graded eQ-module N with Hilbert series 1 + 2t to the vanishing locus of the
subspace of eQ1 that annihilates N0.
Let L be a line module for eQ. The Hilbert series for L/L≥2 is 1 + 2t so L determines a point in
G(1, 3). Since L ∼= eQ/eQy +eQy′ for some linearly independent elements y, y′ ∈ eQ1, the isomorphism
Proposition 11.10. Let g : P(Q∗1) → P(eQ∗1) be the isomorphism induced by the linear isomorphism
eQ1 → Q1,
y1 7→ −ix1,
The function f : E/hξ1i → G(1, 3) defined by
is a closed immersion and f(cid:0)E/hξ1i(cid:1) parametrizes the isomorphism classes of Γ-equivariant Q′-
modules of the form Mx,ξ1, x ∈ E/E[2]. If x = p+E[2], then the lines f(cid:0)p+hξ1i(cid:1) and f(cid:0)p+ξ2+hξ1i(cid:1)
correspond to the two non-isomorphic equivariant structures on Mx,ξ1.
Proof. The map that sends a point p ∈ E to the line through p and p + ξ1 is a morphism from E to
the Grassmanian of lines in P(Q∗1). Composing that map with g gives a morphism h : E → G(1, 3).
Since h(p) = h(p + ξ1), h factors as a composition
(11-5)
f(cid:0)p + hξ1i(cid:1) := g(the line in P(Q∗1) that passes through p and p + ξ1)
y0 7→ x0,
y2 7→ −ix2,
y3 7→ x3.
E −→ E/hξ1i −→ G(1, 3)
where the first map is the quotient map and the second is f . By the universal property of the quo-
tient map, f is a morphism. In fact, f is the composition γβ of the two maps from Proposition 10.1
and Lemma 10.3 and hence is a closed immersion.
The line in P(Q∗1) through p and p + ξ1 is of the form β0x0 + β1x1 = β2x2 + β3x3 = 0. Therefore
corresponds to the Γ-equivariant structure on M = Mx,ξ1 with the equivariant structure described
in Theorem 11.9.
(cid:4)
f(cid:0)p + hξ1i(cid:1) is the line g(β0x0 + β1x1) = g(β2x2 + β3x3) = 0, i.e., the line iβ0y0 − β1y1 = β2y2 −
iβ3y3 = 0. Thus, f (p + hξ1i) is the line in P(eQ∗1) that corresponds to the eQ-line module, M Γ, that
In particular, by
Proposition 11.10 there are morphisms E/hξ1i → G(1, 3), E/hξ2i → G(1, 3), and E/hξ3i → G(1, 3).
It is easy to see that these morphisms are injective but we have not yet shown that the images are
smooth. It is clear that the images of these morphisms are disjoint from one another.
There are versions of all the results in §11.4 with ξ2 and ξ3 in place of ξ1.
Theorem 11.11. The set of Γ-equivariant Q′-modules in Theorem 11.6 is parametrized by
where {ξ, ξ′, ξ′′} is the set of 2-torsion points on E.
(cid:0)E/hξi(cid:1) ⊔ (cid:0)E/hξ′i(cid:1) ⊔ (cid:0)E/hξ′′i(cid:1)
In fact, we can say more about these three components of the scheme of line modules. We will
say that a closed subscheme of a projective space PN is spatial if its inclusion factors through some
linear P3 ⊂ PN but not through a linear P2 ⊂ PN .
Proposition 11.12. For each 2-torsion point ξ the elliptic curve E/hξi ⊂ G(1, 3) ⊂ P5 is spatial
of degree four.
38
ALEX CHIRVASITU AND S. PAUL SMITH
Proof. That E/hξi is contained in a P3 ⊂ P5 follows from its construction in Proposition 11.10.
Indeed, suppose in order to fix notation that ξ = ξ1 and denote E = E/hξi.
If the Plucker
coordinates of the line
are the minors Mij, 0 ≤ i < j ≤ 3 of the matrix
3Xj=0
3Xj=0
λ′0 λ′1 λ′2 λ′3(cid:19)
M =(cid:18)λ0 λ1 λ2 λ3
λjyj =
λ′jyj = 0
supported on columns i and j, then the two coordinates M01 and M23 vanish on E by part (3) of
Theorem 11.9.
The fact that E is not contained in a P2 will follow once we prove that the degree of the embedding
into P5 is four, as claimed in the statement.
To check the degree assertion we will intersect E with a hyperplane section of G(1, 3) ⊂ P5,
judiciously chosen so that it is not tangent to E and the number of intersection points is clearly
four.
For every line ℓ in P3 the collection of all lines in G(1, 3) intersecting ℓ is a hyperplane section
Hℓ of G(1, 3) ⊂ P5. Let ℓ = pq be a secant line of E. The points in E ∩ Hℓ are the classes modulo
hξi of those u ∈ E for which the secant line u(u + ξ) intersects ℓ.
If
q 6= p + ξ
and 3p + q + ξ 6= 0, p + 3q + ξ 6= 0
(11-6)
then there are exactly four such classes modulo hξi, namely those of p, q, u and u + ξ′, where
u + (u + ξ) + p + q = 0 and E[2] − {0} ∋ ξ′ 6= ξ.
It remains to check that p, q ∈ E can be chosen so that Hℓ is not tangent to E at any of the four
points where they intersect, in addition to satisfying (11-6).
Identify, as usual, the tangent space to G(1, 3) at some line m (simultaneously regarded as a
2-plane in the 4-dimensional vector space V ) with the space of linear maps m → V /m. Generally,
we will conflate linear subspaces of V and their projectivized versions.
For any u ∈ E, the tangent line to E ⊂ G(1, 3) at u(u + ξ) can be identified with the space of
linear maps u(u + ξ) → V /u(u + ξ) that send the lines u and u + ξ in V to the 2-planes TuE and
Tu+ξE in V respectively modulo u(u + ξ).
On the other hand, reverting to the notation introduced above for u ∈ E so that 2u+ξ +p+q = 0,
the tangent space at u(u + ξ) ∈ G(1, 3) to Hℓ consists of those linear maps u(u + ξ) → V /u(u + ξ)
that send the intersection s = pq ∩ u(u + ξ) to pq modulo u(u + ξ) (see e.g. [12, Example 16.6]).
Since the line s ⊂ V is in the span of u and u + ξ, we would be certain that the tangent space in
the previous paragraph does not contain the tangent line described two paragraphs up if we knew
that the tangents to E at u and u + ξ are coplanar. This is indeed the case if 4u = 0, so simply
take u ∈ E[4] and afterwards select p and q so that (11-6) holds.
(cid:4)
11.5.1. There is another perspective on the Γ-equivariant Q′-modules parametrized by E/hξi. The
family of Q′-modules Mx,ξ is parametrized by x ∈ E/E[2]. The quotient of the fundamental groups,
π1(E/E[2])/π1(E/hξi), which is naturally isomorphic to E[2]/hξi, acts freely and transitively on
each fiber of the natural map E/hξi → E/E[2].
If we identify the fiber over x with the set of
isomorphism classes of equivariant structures on Mx,ξ, then H 1(Γ, Aut(Mx,ξ)) also acts on the fiber
over x. As the paragraph explains, these actions of E/hξi and H 1(Γ, Aut(Mx,ξ)) on the fibers are
compatible in a natural way.
The Weil pairing h·,·i : E[2] × E[2] → µ2 = {±1} ⊆ k× is a non-degenerate skew-symmetric
bilinear form on E[2] viewed as a 2-dimensional vector space over F2. Since hξ, ξi = 1, there is an
EXOTIC ELLIPTIC ALGEBRAS
39
induced non-degenerate bilinear map hξi × E[2]/hξi → µ2 or, what is essentially the same thing, a
group isomorphism
E[2]/hξi −→ HomZ(hξi, µ2) = HomZ(hξi, k×) ∼= H 1(Γ, Aut(Mx,ξ))
where the right-most isomorphism was established in the proof of Theorem 11.6(3).
irreducible component of the scheme parametrizing the line modules has dimension ≥ 1 [27, Cor.2.6]
and that every point module is a quotient of a line module [27, Prop.3.1]. We will investigate this
11.6. Under quite general conditions, which eQ satisfies, Shelton and Vancliff prove that every
relationship in a subsequent paper. We also show there that the line modules for eQ described above
are not all the line modules.
Appendix A. Equivariant structures
A.1. Groups acting on categories. An action of a group Γ on a category C consists of data
{α∗, tα,β α, β ∈ Γ} where each α∗ : C → C is an auto-equivalence and each tα,β : α∗β∗ → (αβ)∗ is
a natural isomorphism such that the diagrams
α∗ ◦ β∗ ◦ γ∗
tα,β·γ ∗
(αβ)∗ ◦ γ∗
α∗·tβ,γ
tαβ,γ
α∗ ◦ (βγ)∗
tα,βγ
/ (αβγ)∗
commute for all α, β, γ ∈ Γ.
Lemma A.1. Let x ∈ Ob(C) and φ = {φα : x → α∗x α ∈ Γ} a set of isomorphisms. If Aut(x) is
abelian, then there is an action of Γ on Aut(x) given by the formula
Γ × Aut(x) → Aut(x)
(α, f ) 7→ α · f := φ−1
α α∗(f )φα.
This action does not depend on the choice of the φα's.
Proof. Because tα,β : α∗ ◦ β∗ → (αβ)∗ is a natural transformation, the diagram
α∗(β∗x)
(tα,β )x
(αβ)∗x
α∗β ∗(f )
(αβ)∗(f )
α∗(β∗x)
/ (αβ)∗x
(tα,β )x
commutes for all f ∈ Aut(x) and all α, β ∈ Γ. In other words,
(A-1)
(αβ)∗(f ) = (tα,β)x ◦ α∗β∗(f ) ◦ (tα,β)−1
x .
Since Aut(α∗β∗x) is abelian, (tα,β)−1
x φαβφ−1
α α∗(φβ)−1 commutes with α∗β∗(f ). This fact can be
expressed as
which we re-write as
(A-2)
α α∗(φ−1
φ−1
β ) ◦ α∗β∗(f ) ◦ α∗(φβ)φα = φ−1
α α∗(cid:16)φ−1
β β∗(f )φβ(cid:17)φα = φ−1
φ−1
αβ (tα,β)x ◦ α∗β∗(f ) ◦ (tα,β)−1
x φαβ
αβ (tα,β)x ◦ α∗β∗(f ) ◦ (tα,β)−1
x φαβ.
The left-hand side of (A-2) is φ−1
(A-2) is equal to
α α∗(β · f )φα = α · (β · f ) and, by (A-1), the right-hand side of
φ−1
αβ (αβ)∗(f )φαβ
/
/
/
/
/
/
40
ALEX CHIRVASITU AND S. PAUL SMITH
which equals (αβ) · f . Thus α · (β · f ) = (αβ) · f .
To see that the action does not depend on the choice of the φα's suppose that {φ′α : x →
α∗x α ∈ Γ} is another collection of isomorphisms. There are automorphisms ψα ∈ Aut(α∗x) such
that φ′α = ψαφα. The action of Γ on Aut(x) associated to the φ′α, α ∈ Γ, is
α α∗(f )ψαφα;
(α, f ) 7→ (φ′α)−1α∗(f )φ′α = φ−1
α ψ−1
but ψ−1
equation is equal to α · f .
α α∗(f )ψα = α∗(f ) because Aut(α∗x) is abelian, so the right-hand side of the displayed
(cid:4)
A.2. Equivariant objects. Suppose Γ acts on C. A Γ-equivariant structure on an object x ∈ C is
a set of isomorphisms {φα : x → α∗x α ∈ Γ} such that the diagrams
(A-3)
x
φαβ
(αβ)∗x
φα
α∗x
α∗(φβ )
α∗(β∗x)
(tα,β )x
commute for all α, β, γ ∈ Γ.
An arbitrary set of isomorphisms φα : x → α∗x, α ∈ Γ, will not usually give an equivariant
structure on x. Their failure to do so, i.e., the failure of (A-3) to commute, is measured by the
automorphisms
(A-4)
aα,β := φ−1
αβ ◦ (tα,β)x ◦ α∗(φβ) ◦ φα
of x.
Lemma A.2. Let x ∈ Ob(C) and and let {φα : x → α∗x α ∈ Γ} be a set of isomorphisms. If
Aut(x) is abelian, then the function
is a 2-cocycle.
a : Γ × Γ → Aut(x),
(α, β) 7→ aα,β,
Proof. We must show that aαβ,γ ◦ aα,β = aα,βγ ◦ (α · aβ,γ) for all α, β, γ ∈ Γ.
First, aαβ,γ ◦ aα,β equals
φ−1
αβγ ◦ (tαβ,γ)x ◦ (αβ)∗(φγ) ◦ φαβ ◦ φ−1
= φ−1
αβγ ◦ (tαβ,γ)x ◦ (αβ)∗(φγ) ◦ (tα,β)x ◦ α∗(φβ) ◦ φα
= φ−1
αβγ ◦ (tαβ,γ)x ◦ (tα,β)γ ∗x ◦ α∗β∗(φγ) ◦ α∗(φβ) ◦ φα
αβ ◦ (tα,β)x ◦ α∗(φβ) ◦ φα
where the last equality follows from the commutative diagram
α∗β∗x
(tα,β )x
(αβ)∗x
α∗β ∗(φγ )
(αβ)∗(φγ )
α∗β∗(γ∗x)
/ (αβ)∗(γ∗x)
(tα,β )γ∗ x
which exists by virtue of the fact that tα,β is a natural transformation (applied to the isomorphism
φγ : x → γ∗x).
/
/
o
o
/
/
/
EXOTIC ELLIPTIC ALGEBRAS
41
On the other hand, aα,βγ ◦ (α · aβ,γ) equals
φ−1
αβγ ◦ (tα,βγ)x ◦ α∗(φβγ) ◦ φα ◦ φ−1
= φ−1
αβγ ◦ (tα,βγ)x ◦ α∗((tβ,γ)x) ◦ α∗β∗(φγ) ◦ α∗(φβ) ◦ φα
= φ−1
αβγ ◦ (tα,βγ)x ◦ (α∗ · tβ,γ)x ◦ α∗β∗(φγ) ◦ α∗(φβ) ◦ φα
= φ−1
αβγ ◦ (tαβ,γ)x ◦ (tα,β)γ ∗x ◦ α∗β∗(φγ) ◦ α∗(φβ) ◦ φα
α ◦ α∗(cid:0)φ−1
βγ ◦ (tβ,γ)x ◦ β∗(φγ) ◦ φβ(cid:1) ◦ φα
Thus, aα,βγ ◦ (α · aβ,γ) = aαβ,γ ◦ aα,β.
Proposition A.3. Let x ∈ Ob(C) and suppose Aut(x) is abelian. If the 2-cocycle (α, β) 7→ aα,β
defined in (A-4) is the coboundary of the function f : Γ → Aut(x), α 7→ aα, then the isomorphisms
{φαa−1
Proof. The hypothesis says that
α : x → α∗x α ∈ Γ} form an equivariant structure on x.
(cid:4)
αβ ◦ (tα,β)x ◦ α∗(φβ) ◦ φα = (df )(α, β) = (α · aβ) ◦ a−1
φ−1
αβ ◦ aα
for all α, β ∈ Γ. Since Aut(x) is abelian, we can rewrite this as
φ−1
αβ ◦ (tα,β)x ◦ α∗(φβ) ◦ φα = a−1
= a−1
αβ ◦ aα ◦ (α · aβ)
αβ ◦ aα ◦ φ−1
α α∗(aβ)φα
whence (tα,β)x ◦ α∗(φβ) = φαβa−1
αβ ◦ aαφ−1
α ◦ α∗(aβ). In other words, the diagram
x
φαβa−1
αβ
φαa−1
α
α∗x
α∗(φβ a−1
β )
(αβ)∗x
α∗(β∗x)
(tα,β )x
commutes; i.e., the maps {φαa−1
A.3. Classification of equivariant structures. In order to classify equivariant structures we
must first say what it means for two equivariant structures to be the "same".
α : x → α∗x α ∈ Γ} form an equivariant structure on x.
(cid:4)
Suppose that Γ acts on C. The objects in the category CΓ of Γ-equivariant objects in C are pairs
(x, φ) consisting of an object x in C and a set of isomorphisms φ = {φα : x → α∗x α ∈ Γ} that
give x the structure of a Γ-equivariant object. A morphism f : (x, φ) → (y, ψ) in CΓ is a morphism
f : x → y in C such that the diagram
f
x
y
φα
ψα
α∗x
α∗(f )
/ α∗y
We will classify equivariant structures on an x ∈ Ob(C) up to isomorphism in the special case
commutes for all α ∈ Γ.
when Aut(x) is abelian.
Lemma A.4. Let x ∈ Ob(C). Suppose that {φα : x → α∗x α ∈ Γ} and {ψα : x → α∗x α ∈ Γ} are
equivariant structures on x. If Aut(x) is abelian, then the function f : Γ → Aut(x), f (α) := ψ−1
α φα,
is a 1-cocycle.
/
/
o
o
/
/
/
42
ALEX CHIRVASITU AND S. PAUL SMITH
Proof. By definition,
(A-5)
Because the φ's and ψ's define equivariant structures,
β φβ) ◦(cid:0)ψ−1
(df )(α, β) = (α · ψ−1
αβ φαβ = (cid:16)tα,βα∗(ψβ)ψα(cid:17)−1
αβ φαβ(cid:1)−1 ◦ ψ−1
◦(cid:16)tα,βα∗(φβ)φα(cid:17)
α φα.
ψ−1
= ψ−1
α α∗(ψ−1
β φβ)φα
Therefore
(df )(α, β) = φ−1
= idx.
α α∗(ψ−1
β φβ)φα ◦(cid:0)ψ−1
α α∗(ψ−1
β φβ)φα(cid:1)−1 ◦ ψ−1
α φα
Thus, f is a 1-cocycle as claimed.
(cid:4)
Let x ∈ Ob(x). We write Φ(x) for the set of equivariant structures on x and Φ(x)Isom for the set
of isomorphism classes of equivariant structures on x. If φ = {φα : x → α∗x α ∈ Γ} ∈ Φ(x) we
write [φ] for the isomorphism class of φ; i.e., φ 7→ [φ] denotes the obvious function Φ(x) → Φ(x)Isom.
Proposition A.5. Let x ∈ Ob(C) and suppose Aut(x) is abelian. If φ = {φα : x → α∗x α ∈ Γ}
is an equivariant structure on x and f : Γ → Aut(x), α 7→ fα, a 1-cocycle, then
(f · φ) := {φαfα : x → α∗x α ∈ Γ}
is an equivariant structure on x that depends only on the class of f in H 1(Γ, Aut(x)). This gives an
action of H 1(Γ, Aut(x)) on Φ(x)Isom. Furthermore, if Φ(x) 6= ∅, then H 1(Γ, Aut(x)) acts simply
transitively on Φ(x)Isom.
Proof. Let [f ] ∈ H 1(Γ, Aut(x)) where f is a 1-cocycle. Let φ = {φα} ∈ Φ(x). Because f is a
1-cocycle, (α · fβ)f−1
αβ fα = idx. Because Aut(x) is abelian this equality can be rewritten as
Since the φα's form an equivariant structure on x,
fαβ = (α · fβ)fα = φ−1
α α∗(fβ)φαfα.
φαβ = (tα,β)xα∗(φβ)φα
for all α, β ∈ Γ. Therefore
φαβfαβ =(cid:16)(tα,β)xα∗(φβ)φα(cid:17) ◦(cid:16)φ−1
α α∗(fβ)φαfα(cid:17) = (tα,β)xα∗(φβfβ)φαfα.
In other words, the diagram
x
φαβfαβ
φαfα
α∗x
α∗(φβ fβ)
(αβ)∗x
α∗(β∗x)
(tα,β )x
commutes; i.e., the maps {φαfα : x → α∗x α ∈ Γ} form an equivariant structure on x.
We now show that the isomorphism class of (x, f · φ) depends only on the cohomology class of
f . Let f, f′ : Γ → Aut(x) be 1-cocycles. They are cohomologous if and only if f′f−1 = dg for some
g ∈ C 0(Γ, Aut(x)) = Aut(x), i.e., if and only if there is g ∈ Aut(x) such that
f′αf−1
α = (dg)(α) = (α · g)g−1
/
/
o
o
EXOTIC ELLIPTIC ALGEBRAS
43
for all α ∈ Γ. On the other hand, (x, f · φ) ∼= (x, f′ · φ) if and only if there is an isomorphism
g : x → x such that the diagram
g
x
x
φαfα
φαf ′
α
α∗x
α∗(g)
/ α∗x
commutes for all α ∈ Γ; i.e., if and only if α∗(g)φαfα = φαf′αg or, equivalently, φ−1
α α∗(g)φαfα = f′αg
α α∗(g)φαg−1 =
for all α ∈ Γ. Since Aut(x) is abelian, this is equivalent to the condition that φ−1
α . This completes the proof that (x, f · φ) ∼= (x, f′ · φ) if
f′αf−1
and only if [f ] = [f′]. Thus, once we have show that ([f ], [φ]) 7→ [f · φ], really is an action, as we
do in the next paragraph, we will have shown that H 1(Γ, Aut(x)) acts on Φ(x)Isom and all isotropy
groups are trivial.
α for all α ∈ Γ, i.e., (α · g)g−1 = f′αf−1
We now check that ([f ], φ) 7→ (f·φ) is an action of H 1(Γ, Aut(x)) on Φ(x). Let f, f′ : Γ → Aut(x)
be 1-cocycles. Then f · (f′ · φ) = {φαf′αfα α ∈ Γ}. Since fα and f′α are elements in the abelian
group Aut(x), f′αfα = fαf′α, from which it follows that f · (f′ · φ) = (f f′) · φ.
It remains to show that H 1(Γ, Aut(x)) acts transitively on Φ(x)Isom is transitive. Let φ, φ′ ∈ Φ(x).
We will show there is a 1-cocycle f such that φ′ ∼= f · φ. By Lemma A.6 below, the function
f : Γ → Aut(x) defined by f (α) := φ−1
α φ′α is a 1-cocycle. But (f ·φ)α = φαfα = φ′α so φ′ = f ·φ. (cid:4)
Lemma A.6. Let x ∈ Ob(C) and suppose that Aut(x) is abelian If φ, ψ ∈ Φ(x), then φ−1ψ :=
{φ−1
Proof. We must show that d(φ−1ψ)(α, β) is the identity for all α, β ∈ Γ. This is the case because
α ψα α ∈ Γ} is a 1-cocycle for Γ with values in Aut(x).
d(φ−1ψ)(α, β) = α · (φ−1
= α · (φ−1
= φ−1
= φ−1
= φ−1
α α∗(φ−1
α α∗(φ−1
α α∗(φ−1
β ψβ) ◦ (φ−1
β ψβ) ◦ φ−1
αβ ψαβ)−1 ◦ φ−1
α ψα ◦ (φ−1
α ψα
αβ ψαβ)−1
α ψα ◦ ψ−1
αβ φαβ
α α∗(ψβ)−1(tα,β)−1
β ψβ)φα ◦ φ−1
β ψβ)ψα ◦ ψ−1
β ψβ)α∗(ψβ)−1α∗(φβ)φα
x ◦ (tα,β)xα∗(φβ)φα
which is certainly equal to idx.
(cid:4)
A.4. Equivariant modules. Let Γ act as k-algebra automorphisms of a k-algebra R. If α ∈ Γ and
M is a left R-module we define α∗M to be M as a k-vector space with a new action of R, namely
x ·α m := α−1(x)m. If f : M → N is an R-module homomorphism we define α∗(f ) : α∗M → α∗N
to be the function f , now viewed as a homomorphism from α∗M to α∗N . In this way, α∗ becomes
an auto-equivalence of the category of left R-modules, Mod(R). Since α∗β∗ = (αβ)∗ this gives an
action of Γ on Mod(R).
α (xm) = α(x)φ−1
Suppose M is a Γ-equivariant left R-module via the isomorphisms φα : M → α∗M , α ∈ Γ.
Since α∗M = M , each φα is a k-linear map φα : M → M and it has the property that φα(xm) =
x·αφα(m) = α−1(x)φα(m) or, equivalently, φ−1
α (m), for all x ∈ R and m ∈ M . If we
write mα := φ−1
α (m), then we obtain a left action of Γ on M with the property that (xm)α = α(x)mα
for all x ∈ R, α ∈ Γ, and m ∈ M .
Conversely, if M is a left R-module with a left action of Γ on M such that (xm)α = α(x)mα for
all x ∈ R, α ∈ Γ, and m ∈ M , then the maps φα : M → α∗M defined by φα(m) = mα−1 gives M
the structure of a Γ-equivariant R-module.
Thus, a Γ-equivariant R-module is an R-module, M say, together with an action of Γ via a group
homomorphism Γ → AutZ(M ), α 7→ (m 7→ mα), such that (xm)α = α(x)mα for all α ∈ Γ and
m ∈ M .
/
/
/
44
ALEX CHIRVASITU AND S. PAUL SMITH
References
[1] M. Artin. Geometry of quantum planes. In Azumaya algebras, actions, and modules (Bloomington, IN, 1990),
volume 124 of Contemp. Math., pages 1 -- 15. Amer. Math. Soc., Providence, RI, 1992.
[2] M. Artin and J. T. Stafford. Semiprime graded algebras of dimension two. J. Algebra, 227(1):68 -- 123, 2000.
[3] M. Artin, J. Tate, and M. Van den Bergh. Some algebras associated to automorphisms of elliptic curves. In The
Grothendieck Festschrift, Vol. I, volume 86 of Progr. Math., pages 33 -- 85. Birkhauser Boston, Boston, MA, 1990.
[4] M. Artin, J. Tate, and M. Van den Bergh. Modules over regular algebras of dimension 3. Invent. Math.,
106(2):335 -- 388, 1991.
[5] M. Artin and M. Van den Bergh. Twisted homogeneous coordinate rings. J. Algebra, 133(2):249 -- 271, 1990.
[6] M. Artin and J. J. Zhang. Noncommutative projective schemes. Adv. Math., 109(2):228 -- 287, 1994.
[7] J. Bichon. Hopf-Galois objects and cogroupoids. ArXiv e-prints, June 2010.
[8] A. P. Davies. Cocyle Twists
of Algebras. PhD thesis,
2014.
University
of Manchester,
https://www.escholar.manchester.ac.uk/uk-ac-man-scw:229719. Retrieved 01-20-2014.
[9] A. P. Davies. Cocycle Twists of Algebras. ArXiv e-prints, February 2015.
[10] M. Van den Bergh. An example with 20 point modules. circulated privately, 1988.
[11] B. Feigin and A. Odesskii. A family of elliptic algebras. Internat. Math. Res. Notices, (11):531 -- 539, 1997.
[12] J. Harris. Algebraic geometry, volume 133 of Graduate Texts in Mathematics. Springer-Verlag, New York, 1992.
A first course.
[13] R. Hartshorne. Algebraic geometry. Springer-Verlag, New York-Heidelberg, 1977. Graduate Texts in Mathemat-
ics, No. 52.
[14] G. Krause and T. H. Lenagan. Growth of algebras and Gelfand-Kirillov dimension, volume 22 of Graduate Studies
in Mathematics. American Mathematical Society, Providence, RI, revised edition, 2000.
[15] H. F. Kreimer and P. M. Cook, II. Galois theories and normal bases. J. Algebra, 43(1):115 -- 121, 1976.
[16] T. Levasseur and S. P. Smith. Modules over the 4-dimensional Sklyanin algebra. Bull. Soc. Math. France,
121(1):35 -- 90, 1993.
[17] S. Mac Lane. Categories for the working mathematician, volume 5 of Graduate Texts in Mathematics. Springer-
Verlag, New York, second edition, 1998.
[18] S. Montgomery. Fixed rings of finite automorphism groups of associative rings, volume 818 of Lecture Notes in
Mathematics. Springer, Berlin, 1980.
[19] S. Montgomery. Hopf algebras and their actions on rings, volume 82 of CBMS Regional Conference Series
in Mathematics. Published for the Conference Board of the Mathematical Sciences, Washington, DC; by the
American Mathematical Society, Providence, RI, 1993.
[20] D. Mumford. Abelian varieties. Tata Institute of Fundamental Research Studies in Mathematics, No. 5. Published
for the Tata Institute of Fundamental Research, Bombay; Oxford University Press, London, 1970.
[21] A. V. Odesskii. Introduction to the theory of elliptic algebras. http://data.imf.au.dk/conferences/FMOA05/ellect.pdf.
Retrieved 09-18-2014.
[22] A. V. Odesskii and B. L. Feigin. Sklyanin's elliptic algebras. Funktsional. Anal. i Prilozhen., 23(3):45 -- 54, 96,
1989.
[23] A. V. Odesskii and B. L. Feigin. Constructions of elliptic Sklyanin algebras and of quantum R-matrices. Funkt-
sional. Anal. i Prilozhen., 27(1):37 -- 45, 1993.
[24] M. Reyes, D. Rogalski, and J. J. Zhang. Skew Calabi-Yau algebras and homological identities. Adv. Math.,
264:308 -- 354, 2014.
[25] H.-J. Schneider. Principal homogeneous spaces for arbitrary Hopf algebras. Israel J. Math., 72(1-2):167 -- 195,
1990. Hopf algebras.
[26] B. Shelton and M. Vancliff. Some quantum P3s with one point. Comm. Algebra, 27(3):1429 -- 1443, 1999.
[27] B. Shelton and M. Vancliff. Schemes of line modules. I. J. London Math. Soc. (2), 65(3):575 -- 590, 2002.
[28] E. K. Sklyanin. Some algebraic structures connected with the Yang-Baxter equation. Funktsional. Anal. i
Prilozhen., 16(4):27 -- 34, 96, 1982.
[29] S. P. Smith. The four-dimensional Sklyanin algebras. In Proceedings of Conference on Algebraic Geometry and
Ring Theory in honor of Michael Artin, Part I (Antwerp, 1992), volume 8, pages 65 -- 80, 1994.
[30] S. P. Smith and J. T. Stafford. Regularity of the four-dimensional Sklyanin algebra. Compositio Math., 83(3):259 --
289, 1992.
[31] S. P. Smith and J. M. Staniszkis. Irreducible representations of the 4-dimensional Sklyanin algebra at points of
infinite order. J. Algebra, 160(1):57 -- 86, 1993.
[32] S.P. Smith. Corrigendum to "Maps between non-commutative spaces" [Trans. Amer. Math. Soc., 356(7) (2004)
2927-2944]. ArXiv e-prints, 1507.02497.
[33] S.P. Smith. Maps between non-commutative spaces. Trans. Amer. Math. Soc., 356(7):2927 -- 2944 (electronic),
2004.
EXOTIC ELLIPTIC ALGEBRAS
45
[34] J. T. Stafford and M. Van den Bergh. Noncommutative curves and noncommutative surfaces. Bull. Amer. Math.
Soc. (N.S.), 38(2):171 -- 216, 2001.
[35] W. A. Stein et al. Sage Mathematics Software (Version 6.4.1). The Sage Development Team, 2015.
http://www.sagemath.org.
[36] D. R. Stephenson and M. Vancliff. Some finite quantum P3s that are infinite modules over their centers. J.
Algebra, 297(1):208 -- 215, 2006.
[37] D. R. Stephenson and M. Vancliff. Constructing Clifford quantum P3s with finitely many points. J. Algebra,
312(1):86 -- 110, 2007.
[38] J. Tate and M. Van den Bergh. Homological properties of Sklyanin algebras. Invent. Math., 124(1-3):619 -- 647,
1996.
[39] K.-H. Ulbrich. Galois extensions as functors of comodules. Manuscripta Math., 59(4):391 -- 397, 1987.
[40] M. Van den Bergh. Blowing up of non-commutative smooth surfaces. Mem. Amer. Math. Soc., 154(734):x+140,
2001.
[41] M. Vancliff, K. Van Rompay, and L. Willaert. Some quantum P3s with finitely many points. Comm. Algebra,
26(4):1193 -- 1208, 1998.
[42] Heinrich Weber. Lehrbuch der algebra. Band III. Chelsea Publishing Co., New York, 1908.
Department of Mathematics, Box 354350, University of Washington, Seattle, WA 98195,USA.
E-mail address: [email protected], [email protected]
|
0906.4396 | 2 | 0906 | 2010-05-28T09:28:46 | Algebras Defined by Monic Gr\"obner Bases over Rings | [
"math.RA"
] | Let $K\langle X\rangle =K\langle X_1,...,X_n\rangle$ be the free algebra of $n$ generators over a field $K$, and let $R\langle X\rangle =R\langle X_1,...,X_n\rangle$ be the free algebra of $n$ generators over an arbitrary commutative ring $R$. In this semi-expository paper, it is clarified that any monic Gr\"obner basis in $K\langle X\rangle$ may give rise to a monic Gr\"obner basis of the same type in $R\langle X\rangle$, and vice versa. This fact turns out that many important $R$-algebras have defining relations which form a monic Gr\"obner basis, and consequently, such $R$-algebras may be studied via a nice PBW structure theory as that developed for quotient algebras of $K\langle X\rangle$ in ([LWZ], [Li2, 3]). | math.RA | math | Algebras Defined by Monic Grobner Bases over Rings ∗
Huishi Li
Department of Applied Mathematics
College of Information Science and Technology
Hainan University
Haikou 570228, China
0
1
0
2
y
a
M
8
2
]
.
A
R
h
t
a
m
[
2
v
6
9
3
4
.
6
0
9
0
:
v
i
X
r
a
Abstract. Let KhXi = KhX1, ..., Xni be the free algebra of n generators over
a field K, and let RhXi = RhX1, ..., Xni be the free algebra of n generators over
an arbitrary commutative ring R. In this semi-expository paper, it is clarified
that any monic Grobner basis in KhXi may give rise to a monic Grobner
basis of the same type in RhXi, and vice versa. This fact turns out that many
important R-algebras have defining relations which form a monic Grobner basis,
and consequently, such R-algebras may be studied via a nice PBW structure
theory as that developed for quotient algebras of KhXi in ([LWZ], [Li2, 3]).
0. Introduction
In the structure theory and the representation theory of associative algebras over a ground field
K, it is well known that numerous popularly studied algebras have defining relations which
form a Grobner basis in the classical sense (e.g., [Mor], [Gr]), and such algebras can be studied
in a computational way via their Grobner defining relations (e.g., see [An], [CU], [GI-L], [Gr],
[Li2, 3], [Uf1, 2]); also we know that algebras defined by the relations of the same type over a
commutative ring R are equally important, for instance, the algebras over rings considered in
[Yam], [Ber], [CE], and [LVO2]. So, naturally we expect that certain algebras over rings could
be studied by means of Grobner basis theory as in loc. cit., and thus we hope that the following
statement would hold true:
• Let KhXi = KhX1, ..., Xni be the free algebra of n generators over a field K, and let
RhXi = RhX1, ..., Xni be the free algebra of n generators over an arbitrary commutative
ring R. By a Grobner basis for an ideal in a free algebra we mean the one as defined in
[Mor] and [Gr]. If, with respect to some monomial ordering ≺ on KhXi, a subset G ⊂ KhXi
is a Grobner basis for the ideal hGi in KhXi, then, taking a counterpart of G in RhXi (if
∗Project supported by the National Natural Science Foundation of China (10571038, 10971044).
2000 Mathematics Subject Classification: 16W70; 16Z05.
Key words and phrases: Monic Grobner basis, Termination theorem, PBW R-basis, PBW isomorphism.
1
it exists, again denoted by G) and using the same monomial ordering ≺ on RhXi, G is a
Grobner basis for the ideal hGi in RhXi.
To see at what level the above statement may hold true, it is necessary to see whether a version of
the classical termination theorem ([Mor], [Gr]) works well for verifying Grobner bases in RhXi.
Let KhXi = KhX1, ..., Xni be the free associative K-algebra of n generators over a field
K, and let B be the standard K-basis of KhXi consisting of monomials (words in alphabet
X = {X1, ..., Xn}, including empty word which is identified with the multiplicative identity
element 1 of KhXi). Given a monomial ordering ≺ on B (i.e. a well-ordering ≺ on B satisfying:
u ≺ v implies wus ≺ wvs for all w, u, v, s ∈ B), and f, g ∈ KhXi − {0}, if there are monomials
u, v ∈ B such that
(1) LM(f )u = vLM(g), and
(2) LM(f )6 v and LM(g) 6 u,
then the element
o(f, u; v, g) =
(f · u) −
(v · g)
1
LC(f )
1
LC(g)
is referred to as an overlap element of f and g, where, with respect to ≺, LM( ) denotes
the function taking the leading monomial and LC( ) denotes the function taking the leading
coefficient on elements of KhXi respectively. Over the ground field K, the termination theorem
in the sense of ([Mor], [Gr]), which is known an algorithmic version of Bergman's diamond lemma
[Ber1], states that
• if G is an LM-reduced subset of KhXi (i.e., LM(gi)6 LM(gj ) for gi, gj ∈ G with i 6= j), then
G is a Grobner basis for the ideal I = hGi if and only if for each pair gi, gj ∈ G, including
G
gi = gj, every overlap element o(gi, u; v, gj) of gi and gj has the property o(gi, u; v, gj )
= 0,
that is, o(gi, u; v, gj ) is reduced to 0 by division by G;
and it follows that there is a noncommutative analogue of the Buchberger algorithm for con-
structing a (possibly infinite) Grobner basis starting with a given finite subset in KhXi. Note
that the algorithmic feasibility of the above criterion lies in the fact that
(a) for each pair (gi, gj) there are only finitely many associated overlap elements, and
(b) there is no trouble with taking the inverse of a nonzero coefficient when the division algo-
rithm is performed, for, K is a field.
However, if the field K is replaced by a commutative ring R, and if G ⊂ RhXi = RhX1, ..., Xni
is taken such that the leading coefficient LC(g) of some g ∈ G is not invertible, then, even if R
is an arithmetic ring (e.g. the ring Z of integers) as recently considered by [Gol], there seems
no implementable termination theorem (as we mentioned above) for G. Nevertheless, we have
the following observations:
(1) If G is a Grobner basis for an ideal I in the free algebra KhXi over a field K, then we may
always assume that all elements of G are monic, i.e., LC(g) = 1 for every g ∈ G.
(2) If S is a subset consisting of monic elements in the free algebra RhXi over a ring R, i.e.,
LC(f ) = 1 for every f ∈ S, then, with respect to any monomial ordering ≺ on RhXi, a
2
division algorithm by S can be implemented in RhXi exactly as in KhXi.
Recalling the proof of the classical termination theorem ([Mor], [Gr]), the above observations
provide us with sufficient reason to have an implementable termination theorem (as mentioned
above) for verifying whether certain given monic elements of RhXi form a Grobner basis in the
classical sense, so that our foregoing expectation may come true. To present the details, we
organize this paper as follows. In Section 1, after giving a quick introduction of the notion of a
monic Grobner basis in RhXi, we examine carefully that a version of the termination theorem
in the sense of ([Mor], [Gr]) holds true for verifying LM-reduced monic Grobner bases in RhXi,
just for convincing ourselves and also for the reader's convenience from the viewpoint of "to
see is to believe". This enables us to clarify that every monic Grobner basis in the free algebra
KhXi over a field K may give rise to a monic Grobner basis of the same type in the free algebra
RhXi, and vice versa.
In Section 2, by strengthening and generalizing a result of [Li2], we
show how PBW R-bases and monic Grobner bases of certain type can determine each other.
In the final Section 3, we show that the working principle via PBW isomorphism developed in
[LWZ] and [Li3] can be generalized to study quotient algebras of RhXi, so that many global
structural properties of R-algebras defined by monic Grobner bases may be determined through
their N-leading homogeneous algebras and BR-leading homogeneous algebras.
Unless otherwise stated, rings considered in this paper are associative rings with multiplica-
tive identity 1, ideals are meant two-sided ideals, and modules are unitary left modules. For
a subset U of a ring S, we write hU i (or ShU iS if necessary) for the ideal generated by U .
Moreover, we use N, respectively Z, to denote the set of nonnegative integers, respectively the
set of integers.
1. Monic Grobner Bases over K vs Monic Grobner Bases over R
Let R be an arbitrary commutative ring, RhXi = RhX1, ..., Xni the free R-algebra of n gen-
erators, and BR the standard R-basis of RhXi consisting of monomials (words in alphabet
X = {X1, ..., Xn}, including empty word which is identified with the multiplicative identity
element 1 of RhXi). Unless otherwise stated, monomials in BR are denoted by lower case letters
u, v, w, s, t, · · ·. In this section, after introducing the notion of a monic Grobner basis in RhXi
and examining carefully that a version of the termination theorem in the sense of ([Mor], [Gr])
holds true for verifying LM-reduced monic Grobner bases in RhXi, we clarify that every monic
Grobner basis in the free algebra KhX1, ..., Xni over a field K has a counterpart in the free
algebra RhXi, and vice versa.
First note that all monomial orderings used for free algebras over a field can be well defined
on the standard R-basis BR of RhXi. In particular, by an N-graded monomial ordering on BR,
denoted ≺gr, we mean a monomial ordering on BR which is defined subject to a well-ordering
3
≺ on BR, that is, for u, v ∈ BR, u ≺gr v if either degu < degv or degu = degv but u ≺ v,
where deg( ) denotes the degree function on elements of RhXi with respect to a fixed weight
N-gradation of RhXi (i.e. each Xi is assigned a positive degree ni, 1 ≤ i ≤ n). For instance, the
usual N-graded (reverse) lexicographic ordering is a popularly used N-graded monomial ordering.
i=1 λiwi ∈ RhXi, where λi ∈ R − {0}
and wi ∈ BR, such that w1 ≺ w2 ≺ · · · ≺ ws, then the leading monomial of f is defined as
LM(f ) = ws and the leading coefficient of f is defined as LC(f ) = λs. For a subset H ⊂ RhXi,
we write LM(H) = {LM(f ) f ∈ H} for the set of leading monomials of S. We say that a
subset G ⊂ RhXi is monic if LC(g) = 1 for all g ∈ G. Moreover, for u, v ∈ BR, as usual we say
that v divides u, denoted vu, if u = wvs for some w, s ∈ BR.
If ≺ is a monomial ordering on BR and f = Ps
With notation and all definitions as above, it is easy to see that a division algorithm by a
monic subset G is valid in RhXi with respect to any fixed monomial ordering ≺ on BR. More
precisely, let f ∈ RhXi. Noticing LC(g) = 1 for all g ∈ G, if LM(g)LM(f ) for some g ∈ G, then
f can be written as f = LC(f )ugv +f1 with u, v ∈ BR, f1 ∈ RhXi satisfying LM(f1) ≺ LM(f );
if LM(g)6 LM(f ) for all g ∈ G, then f = f1 + LC(f )LM(f ) with f1 = f − LC(f )LM(f )
satisfying LM(f1) ≺ LM(f ). Next, consider the divisibility of LM(f1) by LM(g) with g ∈ G,
and so forth. Since ≺ is a well-ordering, after a finite number of successive division by elements
in G in this way, we see that f can be written as
f = Pi,j λijuijgjvij + rf , where λij ∈ R, uij, vij ∈ BR, gj ∈ G,
and rf =Pp λpwp with λp ∈ R, wp ∈ BR,
satisfying LM(uijgjvij) (cid:22) LM(f ) whenever λij 6= 0,
LM(rf ) (cid:22) LM(f ) and LM(g) 6 wp for every g ∈ G whenever λp 6= 0.
If, in the presentation of f above, rf = 0, then we say that f is reduced to 0 by division by G,
and we write f
= 0 for this property.
G
The validity of such a division algorithm by G leads to the following definition.
1.1. Definition Let ≺ be a fixed monomial ordering on BR, and I an ideal of RhXi. A monic
Grobner basis of I is a subset G ⊂ I satisfying:
(1) G is monic; and
(2) f ∈ I and f 6= 0 implies LM(g)LM(f ) for some g ∈ G.
By the division algorithm presented above, it is clear that a monic Grobner basis of I is
first of all a generating set of the ideal I, and moreover, a monic Grobner basis of I can be
characterized as follows.
1.2. Proposition Let ≺ be a fixed monomial ordering on BR, and I an ideal of RhXi. For a
monic subset G ⊂ I, the following statements are equivalent:
(i) G is a monic Grobner basis of I;
4
(ii) Each nonzero f ∈ I has a Grobner representation:
f = Pi,j λijuijgjvij, where λij ∈ R, uij, vij ∈ BR, gj ∈ G,
satisfying LM(uijgjvij) (cid:22) LM(f ) whenever λij 6= 0,
G
or equivalently, f
(iii) hLM(G)i = hLM(I)i.
= 0;
(cid:3)
Let ≺ be a monomial ordering on the standard R-basis BR of RhXi, and let G be a monic
subset of RhXi. We call an element f ∈ RhXi a normal element (mod G) if f =Pj µjvj with
µj ∈ R, vj ∈ BR, and f has the property that LM(g) 6 vj for every g ∈ G and all µj 6= 0. The
set of normal monomials in BR (mod G) is denoted by N (G), i.e.,
N (G) = {u ∈ BR LM(g)6 u, g ∈ G}.
Thus, an element f ∈ RhXi is normal (mod G) if and only if f ∈Pu∈N (G) Ru.
1.3. Proposition Let G be a monic Grobner basis of the ideal I = hGi in RhXi with respect
to some monomial ordering ≺ on BR. Then each nonzero f ∈ RhXi has a finite presentation
λijsijgiwij + rf , λij ∈ R, sij, wij ∈ BR, gi ∈ G,
f =Xi,j
where LM(sijgiwij) (cid:22) LM(f ) whenever λij 6= 0, and either rf = 0 or rf is a unique normal
element (mod G). Hence, f ∈ I if and only if rf = 0, solving the "membership problem" for I.
Proof By the division algorithm by G, f can be written as f =Pi,j λijsijgiwij + rf where either
rf = 0 or rf is normal. Suppose after division by G we also have f =Pt,j λtjstjgtwtj +r, where r
is normal (mod G). Then r − rf ∈ I and hence there is some g ∈ G such that LM(g)LM(r − rf ).
But by the definition of a normal element this is possible only if r = rf .
(cid:3)
The foregoing discussion enables us to obtain further characterization of a monic Grobner
basis G, which, in turn, gives rise to the fundamental decomposition theorem of RhXi (viewed
as an R-module) by the ideal I = hGi, and gives rise to a free R-basis for the R-algebra RhXi/I.
1.4. Theorem Let I = hGi be an ideal of RhXi generated by a monic subset G. With notation
as above, the following statements are equivalent.
(i) G is a monic Grobner basis of I.
(ii) The R-module RhXi has the decomposition
RhXi = I ⊕ Xu∈N (G)
Ru = hLM(I)i ⊕ Xu∈N (G)
Ru.
(iii) The canonical image N (G) of N (G) in RhXi/hLM(I)i and RhXi/I forms a free R-basis for
RhXi/hLM(I)i and RhXi/I respectively.
5
(cid:3)
With notation and every definition introduced so far, we proceed now to show that a version
of the termination theorem in the sense of ([Mor], [Gr]) holds true for verifying an LM-reduced
monic Grobner basis in RhXi (see the definition below).
Given a monomial ordering ≺ on BR, we say that a subset G ⊂ RhXi is LM-reduced if
LM(gi) 6 LM(gj ) for all gi, gj ∈ G with gi 6= gj.
If a subset G ⊂ RhXi is both LM-reduced and monic, then we call G an LM-reduced monic
subset. Thus we have the notion of an LM-reduced monic Grobner basis.
Let I be an ideal of RhXi.
If G is a monic Grobner basis of I and g1, g2 ∈ G such that
g1 6= g2 but LM(g1)LM(g2), then clearly g2 can be removed from G and the remained subset
G − {g2} is again a monic Grobner basis for I. Hence, in order to have a better criterion for
monic Grobner basis we need only to consider the subset which is both LM-reduced and monic.
Let ≺ be a monomial ordering on BR. For two monic elements f, g ∈ RhXi − {0}, including
f = g, if there are monomials u, v ∈ BR such that
(1) LM(f )u = vLM(g), and
(2) LM(f ) 6 v and LM(g) 6 u,
then the element
o(f, u; v, g) = f · u − v · g
is called an overlap element of f and g. From the definition we are clear about the fact that
LM((o(f, u; v, g)) ≺ LM(f u) = LM(vg),
and moreover, there are only finitely many overlap elements for each pair (f, g) of elements in
RhXi.
With the preparation made above, below we mention a termination theorem for checking LM-
reduced monic Grobner bases in RhXi, and, also we present a careful step-by-step verification
of its correctness for the reason that we are working on an arbitrary ring instead of a field after
all, though the process is only a light modification of the argument given in [Gr].
1.5. Theorem (Termination theorem) Let ≺ be a fixed monomial ordering on BR.
If G is
an LM-reduced monic subset of RhXi, then G is an LM-reduced monic Grobner basis for the
ideal I = hGi if and only if for each pair gi, gj ∈ G, including gi = gj, every overlap element
= 0, that is, by division by G, every
o(gi, u; v, gj ) of gi, gj has the property o(gi, u; v, gj)
o(gi, u; v, gj ) is reduced to zero.
G
Proof Since o(gi, u; v, gj ) ∈ I, the necessity follows from Proposition 1.2.
6
Under the assumption on overlap elements we prove the sufficiency by showing that if f ∈ I
then LM(g)LM(f ) for some g ∈ G. Suppose the contrary that LM(g)6 LM(f ) for any g ∈ G.
Then we proceed to derive a contradiction.
Since I = hGi, f may be presented as a finite sum
(1)
f =Xi,j
λijvijgiwij, where λij ∈ R, vij, wij ∈ BR, and gi ∈ G.
Let u be the largest monomial occurring on the right hand side of (1). Then noticing that each
gi is monic, u occurs as some vijLM(gi)wij. Since LM(gi)6 LM(f ) for the gi occurring in (1),
it follows that LM(f ) ≺ u and u must occur at least twice on the right hand side of (1) for a
cancelation, that is, we may have
(2)
u = vijLM(gi)wij = vkℓLM(gk)wkℓ.
Among all such presentations of f we can choose one such that
(3) u has the fewest occurrences on the right hand side of (1) and u is as small as possible.
To go further, let us simplify notation by writing v = vij, g = gi, w = wij, v′ = vkℓ, g′ = gk, and
w′ = wkℓ. Thus, the above (2) is turned into the form
(4)
u = vLM(g)w = v′LM(g′)w′.
Moreover, as usual we use l(s) to denote the length of a monomial s ∈ BR. Below we show,
through a case by case study of the above (4), that the choice of f satisfying the above (3) is
impossible.
Case A: l(v) < l(v′).
Case A.1: l(w) < l(w′).
This implies that LM(g) contains LM(g′) as a subword, and hence, LM(g′)LM(g), contradict-
ing the hypothesis on G.
Case A.2: l(w) ≥ l(w′). Then we have to deal with two possibilities.
Case A.2.1: l(v′) ≥ l(vLM(g)).
This implies that there is no overlap element of LM(g) and LM(g′) in u. By the assumption
on lengths, it follows that there is a segment w′′ of v′ such that v′ = vLM(g)w′′ and w =
w′′LM(g′)w′, i.e.,
u = vLM(g)w′′LM(g′)w′.
Rewriting g as g = LM(g) +P λiui, g′ = LM(g′) +P µiu′
vgw = vgw′′g′w′ − vgw′′ (g′ − LM(g′)) w′
i, then
iw′
= vLM(g)w′′g′w′ +Pλivuiw′′g′w′ −Pµivgw′′u′
= v′g′w′ +Pλivuiw′′g′w′ −Pµivgw′′u′
iw′.
Thus, in writing vgw this way, the number of occurrences of u in the chosen presentation of f
satisfying the above (3) can be reduced, a contradiction.
7
Case A.2.2: l(v′) < l(vLM(g)).
This implies that there is an overlap element of LM(g) and LM(g′) in u, that is, there is a
segment r of w and a segment s of v′ such that vs = v′, rw′ = w and LM(g)r = sLM(g′).
Hence,
(5)
u = vLM(g)rw′ = vsLM(g′)w′ and o(g, r; s, g′) = gr − sg′.
Furthermore, it follows from gr = o(g, r; s, g′) + sg′ that
(6)
vgw = vgrw′ = v · o(g, r; s, g′) · w′ + v′g′w′.
By the assumption, o(g, r; s, g′) is reduced to 0 under the division by G, namely,
(7)
satisfying
(8)
o(g, r; s, g′) =Xk,j
ckjvkjgkwkj, vkj, wkj ∈ BR, gk ∈ G, ckj ∈ R,
if ckj 6= 0 then vkjLM(gk)wkj ≺ LM(g · r) = LM(s · g′).
Combining the above (5) -- (8), once again we see that the number of occurrences of u in the
chosen presentation of f satisfying the foregoing (3) can be reduced, a contradiction.
Case B: l(v) = l(v′).
This implies LM(g)LM(g′) or LM(g′)LM(g), which contradicts the assumption that G is
LM-reduced.
Case C: l(v) > l(v′).
This is similar to Case 1.
(cid:3)
Remark (i) Let G be an LM-reduced monic subset of RhXi and I = hGi. Since for each pair
gi, gj ∈ G, including gi = gj, there are only finitely many associated overlap elements, by
Theorem 1.5 we can check effectively whether G is a Grobner basis of I or not, when G is a finite
subset.
(ii) Obviously, if G ⊂ RhXi is an LM-reduced subset with the property that each g ∈ G has the
leading coefficient LC(g) which is invertible in R, then Theorem 1.5 is also valid for G.
(iii) It is obvious as well that Theorem 1.5 does not necessarily induce an analogue of the
Buchberger algorithm as in the classical case.
(iv) It is not difficult to see that all discussion we presented so far is valid for getting monic
Grobner bases in a commutative polynomial ring R[x1, ..., xn] over an arbitrary commutative
ring R where overlap elements are replaced by S-polynomials.
Note that, throughout the proof of Theorem 1.5, nothing involves the invertibility of an
element in the ring R; moreover, division algorithm by monic elements in a free algebra (over a
field or over a commutative ring) never touches on the invertibility of a coefficient. Therefore, we
8
are now clear about the relation between monic Grobner bases over a field and monic Grobner
bases over a commutative ring.
1.6. Proposition Let KhXi = KhX1, ..., Xni be the free algebra of n generators over a field K,
and let RhXi = RhX1, ..., Xni be the free algebra of n generators over an arbitrary commutative
ring R. With notation as before, fixing the same monomial ordering ≺ on both KhXi and RhXi,
the following statements hold.
(i) If a monic subset G ⊂ KhXi is a Grobner basis for the ideal hGi in KhXi, then, taking a
counterpart of G in RhXi (if it exists), again denoted by G, G is a monic Grobner basis for the
ideal hGi in RhXi.
(ii) If a monic subset G ⊂ RhXi is a Grobner basis for the ideal hGi in RhXi, then, taking a
counterpart of G in KhXi (if it exists), again denoted by G, G is a Grobner basis for the ideal
hGi in KhXi.
(cid:3)
From the literature we know that numerous well-known K-algebras over a field K, such as
the n-th Weyl algebra over K, the enveloping algebra of a K-Lie algebra, a K-exterior algebra,
a K-Clifford algebra, a down-up K-algebra, etc., all have defining relations that form an LM-
reduced monic Grobner basis in a free K-algebra. Hence, by Proposition 1.6, if the field K
is replaced by a commutative ring R, then all of these R-algebras have defining relations that
form an LM-reduced monic Grobner basis in a free R-algebra. Below we give another example
illustrating Theorem 1.5 and Proposition 1.6.
Example 1. Let R be a commutative ring. Consider in RhXi = RhX1, ..., Xni the subset
G = Ω ∪ R consisting of
Ω ⊆ {gi = X p
i
R = {gji = XjXi − λjiXiXj 1 ≤ i < j ≤ n} with λji ∈ R,
1 ≤ i ≤ n} with p ≥ 2 a fixed integer,
that is, λji may be zero.
In the case that R = K is a field, it was verified in ([Li4], Example 4) that, under the N-graded
lexicographic ordering ≺gr such that X1 ≺gr X2 ≺gr · · · ≺gr Xn, G forms an LM-reduced monic
Grobner basis for the ideal I = hGi in KhXi. Hence, by Proposition 1.6, G is an LM-reduced
monic Grobner basis for the ideal I = hGi in RhXi. Furthermore, the division by LM(G) yields
1 X α2
2
· · · X αn
n
αi ∈ N and 0 ≤ αs ≤ p − 1 if X p
It follows from Theorem 1.4 that both the algebras RhXi/I and RhXi/hLM(I)i have the free
R-basis
α1
1 X
α2
2 · · · X
αi ∈ N and 0 ≤ αs ≤ p − 1 if X p
where each X i is the canonical image of Xi in RhXi/I and RhXi/hLM(I)i respectively.
N (G) =nX α1
N (G) =nX
(cid:12)(cid:12)(cid:12)
n (cid:12)(cid:12)(cid:12)
αn
9
s ∈ Ωo .
s ∈ Ωo ,
Here let us point out that this example contains two families of special R-algebras, that is,
in the case where Ω = ∅, the R-algebra RhXi/I is similar to the coordinate ring of a quantum
affine n-space over a field (such a quantum coordinate ring over a field is defined with all the
λji 6= 0); and in the case where Ω = {gi = X 2
i 1 ≤ i ≤ n}, the algebra RhXi/I is similar to the
quantum grassmannian (or quantum exterior) algebra over a field in the sense of [Man] (such a
quantum grassmannian algebra over a field is defined with all the λji 6= 0).
We finish this section by a useful corollary of Theorem 1.5.
1.7. Corollary Let R be a commutative ring and R′ a subring of R with the same identity
If we consider the free R-algebra RhXi = RhX1, ..., Xni and the free R′-algebra
element 1.
R′hXi = R′hX1, ..., Xni, then the following two statements are equivalent for a subset G ⊂ R′hXi:
(i) G is an LM-reduced monic Grobner basis for the ideal I = hGi in R′hXi with respect to some
monomial ordering ≺ on the standard R′-basis BR′ of R′hXi;
(ii) G is an LM-reduced monic Grobner basis for the ideal J = hGi in RhXi with respect to
the monomial ordering ≺ on the standard R-basis BR of RhXi, where ≺ is the same monomial
ordering used in (i).
Proof Let G ⊂ R′hXi be an LM-reduced monic subset. Noticing that BR = BR′, each pair
gi, gj ∈ G has the same set of overlap elements in both R′hXi and RhXi. Moreover, notic-
ing that performing the division of an overlap element o(gi, u; v, gj ) by G in both R′hXi and
G
RhXi uses only coefficients from R′. It follows that o(gi, u; v, gj )
= 0 in R′hXi if and only
G
= 0 in RhXi. Hence the equivalence of (i) and (ii) is proved by the termina-
if o(gi, u; v, gj )
tion theorem for LM-reduced monic Grobner bases in R′hXi and the termination theorem for
LM-reduced monic Grobner bases in RhXi, respectively.
(cid:3)
2. PBW R-bases vs Specific Monic Grobner Bases
1 aα2
2 · · · aαn
Let R be a commutative ring and A = R[a1, ..., an] a finitely generated R-algebra with generators
a1, ..., an. If the set B = {aα1
n αj ∈ N} forms a free R-basis of A, that is, A is, as an
R-module, free with the basis B, then, in honor of the classical PBW (Poincar´e-Birkhoff-Witt)
theorem for enveloping algebras of Lie algebras over a ground field K, the set B is usually
referred to as a PBW R-basis of A. Presenting A as a quotient algebra of the free R-algebra
RhXi = RhX1, ..., Xni, i.e., A = RhXi/I with I an ideal of RhXi, the aim of this section
is to show, under a mild condition, that A has a PBW R-basis is equivalent to that I has a
specific monic Grobner basis. This result enables us to obtain PBW R-bases by means of monic
Grobner bases on one hand; and on the other hand, since it is well known that in practice there
are different ways to find a PBW basis of a given algebra provided it exists (e.g., [Ros], [Yam],
[Rin], [Ber]), this result also enables us to obtain monic Grobner bases via already known PBW
R-bases.
10
Throughout this section, we let R be a commutative ring, RhXi = RhX1, ..., Xni the free
R-algebra of n generators, and BR the standard R-basis of RhXi. All notations and notions
concerning monic Grobner bases in RhXi are maintained as before.
Let I be an ideal of RhXi such that the R-algebra A = RhXi/I has the PBW R-basis
α1
1 X
α2
2 · · · X
αn
contains necessarily a subset G consisting of n(n−1)
n αi ∈ N(cid:9), where each X i is the canonical image of Xi in A. Then I
elements of the form:
B = (cid:8)X
2
gji = XjXi −Xα
λαwα, where 1 ≤ i < j ≤ n, λα ∈ R, wα = X α1
1 X α2
2
· · · X αn
n .
In light of Theorem 1.4 and the observation made above, below we give the main result of
this section which, indeed, strengthens and generalizes ([Li2], CH.III, Theorem 1.5).
2
2.1. Theorem Let I be an ideal of RhXi, A = RhXi/I. Suppose that I contains a monic
subset of n(n−1)
elements G = {gji 1 ≤ i < j ≤ n} such that, with respect to some monomial
ordering ≺ on the standard R-basis BR of RhXi, LM(gji) = XjXi for 1 ≤ i < j ≤ n. The
following two statements are equivalent.
(i) The R-algebra A has the PBW R-basis B = {X
the canonical image of Xi in A.
(ii) Any monic subset G of I containing G is a monic Grobner basis for I with respect to ≺.
αn
n αj ∈ N} where each X i is
α2
2 · · · X
α1
1 X
Proof (i) ⇒ (ii) Let G be a monic subset of I containing G, and let
N (G) = {u ∈ BR LM(g) 6 u, g ∈ G}
be the set of normal monomials in BR (mod G). If f ∈ I and f 6= 0, then, after implementing
the division of f by G (with respect to the given monomial ordering ≺) we have
f = Pi,j λijuijgivij + rf , where λij ∈ R, uij, vij ∈ BR, gi ∈ G,
satisfying LM(wijgivij) (cid:22) LM(f ) whenever λij 6= 0,
and rf =Pp λpwp with λp ∈ R and wp ∈ N (G).
1 X α2
2
· · · X αn
n
It follows that N (G) ⊆
Note that gji ∈ G ⊆ G and LM(gji) = XjXi by the assumption.
{X α1
αj ∈ N}. Thus, since B is a free R-basis of A, rf = Pp λpwp = f −
Pi,j uijgivij ∈ I implies λp = 0 for all p. Consequently rf = 0. This shows that every nonzero
element of I has a Grobner representation by the elements of G. Hence G is a monic Grobner
basis for I by Proposition 1.2.
(ii) ⇒ (i) By (ii), the subset G itself is a monic Grobner basis of I with respect to ≺. Let
N (G) be the set of normal monomials in BR (mod G). Noticing that LM(gji) = XjXi for every
gji ∈ G, it follows that N (G) = {X α1
1 X α2
αj ∈ N}, and thereby the algebra A has the
2
desired PBW R-basis B by Theorem 1.4.
(cid:3)
· · · X αn
n
11
We illustrate Theorem 2.1 by several examples. The first four examples given below serve
to obtain monic Grobner bases by means of already known PBW R-bases which are obtained
in the literature without using the theory of Grobner basis.
Example 1. (This is a special case of Example 3 given later.) Let g = R[V ] be the R-Lie algebra
defined by the free R-module V = ⊕n
jixℓ,
1 ≤ i < j ≤ n, λℓ
ji ∈ R. By the classical PBW theorem, the universal enveloping algebra U (g)
of g has the PBW R-basis B = {xα1
n αj ∈ N}. If, with respect to the natural N-
gradation of RhXi = RhX1, ..., Xni, we use an N-graded monomial ordering ≺gr on the standard
R-basis BR of RhXi such that X1 ≺gr X2 ≺gr · · · ≺gr Xn (i.e., degXi = 1, 1 ≤ i ≤ n), then the
set of defining relations
i=1Rxi and the bracket product [xj, xi] = Pn
ℓ=1 λℓ
1 xα2
2 · · · xαn
G =( gji = XjXi − XiXj −
n
Xℓ=1
1 ≤ i < j ≤ n)
λℓ
jiXℓ (cid:12)(cid:12)(cid:12)(cid:12)(cid:12)
of U (g) satisfies LM(gji) = XjXi for 1 ≤ i < j ≤ n. Hence, by Theorem 2.1, G is a monic
Grobner basis for the ideal I = hGi in RhXi.
Example 2. Let U +
q (AN ) be the (+)-part of the Drinfeld-Jimbo quantum group of type AN
over a commutative ring R, where q ∈ R is invertible and q8 6= 1. This example shows that
the defining relations (Jimbo relations) of U +
q (AN ) over R form a monic Grobner basis in a free
R-algebra. By Proposition 1.6, we reach this property over a field K.
In [Ros] and [Yam] it was proved that, over a field K, U +
q (AN ) has a PBW K-basis with
respect to the defining relations (Jimbo relations) of U +
q (AN ); later in [BM] such a PBW basis
was recaptured by showing that the Jimbo relations form a Grobner basis ([BM] Theorem 4.1),
where the proof was sketched to check that all compositions (overlaps) of Jimbo relations reduce
to zero on the base argument of [Yam]. Recently, a very detailed elementary verification of
the fact that all compositions (overlaps) of Jimbo relations reduce to zero and hence the Jimbo
relations form a Grobner basis (namely Theorem 4.1 of [BM]) was carried out by [CSS]. Now,
by using Theorem 2.1 we will see that it is indeed very easy to conclude: the Jimbo relations
form a Grobner basis.
Recall that the Jimbo relations (as described in [Yam]) are given by
xmnxij − q−2xijxmn,
xmnxij − xijxmn,
xmnxij − xijxmn + (q2 − q−2)xinxmj,
xmnxij − q2xijxmn + qxin,
((i, j), (m, n)) ∈ C1 ∪ C3,
((i, j), (m, n)) ∈ C2 ∪ C6,
((i, j), (m, n)) ∈ C4,
((i, j), (m, n)) ∈ C1 ∪ C3,
where with ΛN = {(i, j) ∈ N × N 1 ≤ i < j ≤ N + 1},
C1 = {((i, j), (m, n)) i = m < j < n}, C2 = {((i, j), (m, n)) i < m < n < j},
C3 = {((i, j), (m, n)) i < m < j = n}, C4 = {((i, j), (m, n)) i < m < j < n},
C5 = {((i, j), (m, n)) i < j = m < n}, C6 = {((i, j), (m, n)) i < j < m < n}.
12
By [Yam], for q8 6= 1, U +
q (AN ) has the PBW basis consisting of elements
xi1j1xi2j2 · · · xik jk with (i1, j1) ≤ (i2, j2) ≤ · · · ≤ (ik, jk), k ≥ 0,
where (iℓ, jℓ) ∈ ΛN and < is the lexicographic ordering on ΛN . If we use the N-graded monomial
ordering ≺gr (on the standard K-basis B of the corresponding free algebra) subject to
xij ≺gr xmn ⇐⇒ (i, j) < (m, n),
then it is clear that for each pair ((i, j), (m, n)) ∈ Ci with (i, j) < (m, n), the leading monomial
of the corresponding relation is of the form xmnxij as required by Theorem 2.1.
Example 3. With RhXi = RhX1, ..., Xni, where R is an arbitrary commutative ring, recall
from [Ber] that a q-algebra A = RhXi/hGi over R is defined by the set G of quadric relations
gji = XjXi − qjiXiXj − {Xj, Xi}, 1 ≤ i < j ≤ n, where qji ∈ R − {0},
and {Xj, Xi} =Xαkℓ
satisfying if αkl
ji XkXℓ +XαhXh + cji, αkℓ
ji 6= 0, then i < k ≤ ℓ < j, and k − i = j − ℓ.
ji , αh, cji ∈ R,
Define two R-submodules of the free R-module RhXi:
1 ≤ i < j ≤ no ,
E1 = R-Spanngji (cid:12)(cid:12)(cid:12)
E2 = R-SpannXigji, gjiXi, Xjgji, gjiXj (cid:12)(cid:12)(cid:12)
If, for 1 ≤ i < j < k ≤ n, every Jacobi sum
1 ≤ i < j ≤ no .
J(Xk, Xj, Xi) = {Xk, Xj}Xi − λkiλjiXi{Xk, Xj}−
−λji{Xk, Xi}Xj + λkjXj{Xk, Xi}+
+λkjλki{Xj , Xi}Xk − Xk{Xj, Xi}
is contained in E1 + E2, then A is called a q-enveloping algebra. Clearly, enveloping algebras
of R-Lie algebras are special q-enveloping algebras with q = 1. In [Ber], a q-PBW theorem for
q-enveloping algebras over a commutative ring was obtained along the line similar to the classical
argument on enveloping algebras of Lie algebras as given in [Jac], that is, if A is a q-enveloping
R-algebra then A has the PBW R-basis B = {X
αn
n αj ∈ N}.
α2
2 · · · X
α1
1 X
Now, if we use the N-graded monomial ordering X1 ≺gr X2 ≺gr · · · ≺gr Xn on BR with
respect to the natural N-gradation of RhXi (i.e., degXi = 1, 1 ≤ i ≤ n), then G satisfies
LM(gji) = XjXi for all 1 ≤ i < j ≤ n. Hence, by Theorem 2.1, the set G of the defining
relations of a q-enveloping R-algebra is a monic Grobner basis for the ideal I = hGi in RhXi.
In particular, all quantum algebras over R = C[[h]] which are q-enveloping algebras appeared in
[Ber] are defined by monic Grobner bases.
Remark It is necessary to point out that if R = K is a field, then the fact that the set of
defining relations G of a q-enveloping K-algebra A forms a Grobner basis of the ideal I = hGi
13
was proved in ([Li2], CH.III) directly by using the termination theorem through the division
algorithm. Here our last example provides the general result for all q-enveloping algebras over
an arbitrary commutative ring.
Example 4. This example generalizes the previous three examples but uses an ad hoc monomial
ordering. As an application we show that, over a commutative ring R, the PBW generators of the
quantum algebra U +
q (AN ).
With RhXi = RhX1, ..., Xni, consider the R-algebra A = RhXi/hGi defined by the subset G
q (AN ) derived in [Rin] provides another Grobner defining set for U +
consisting of n(n−1)
2
elements
gji = XjXi − qjiXiXj −Pα λαX α1
i1
where qji, λα, λji ∈ R, αk ∈ N, i < i1 ≤ i2 ≤ · · · ≤ is < j.
X α2
i2
· · · X αs
is
+ λji, 1 ≤ i < j ≤ n,
α1
1 X
α2
2 · · · X
It is well-known that numerous iterated skew polynomial algebras over R are defined subject to
αn
such relations, and consequently they have the PBW R-basis B = {X
n αj ∈ N}.
Under the assumption that A has the PBW R-basis as described we aim to show that G is a monic
Grobner basis of hGi. In view of Theorem 2.1, it is sufficient to introduce a monomial ordering
on BR so that LM(gji) = XjXi for all 1 ≤ i < j ≤ n. To this end, let R[t] = R[t1, ..., tn] be the
commutative polynomial R-algebra of n variables. Consider the canonical algebra epimorphism
π: RhXi → R[t] with π(Xi) = ti. If we fix the lexicographic ordering X1 <lex X2 <lex · · · <lex
Xn on BR of RhXi (note that <lex is not a monomial ordering on BR) and fix an arbitrarily
chosen monomial ordering ≺ on the standard R-basis BR = {tα1
αj ∈ N} of R[t],
respectively, then, as in [EPS], a monomial ordering ≺et on BR, which is called the lexicographic
extension of the given monomial ordering ≺ on BR, may be obtained as follows: for u, v ∈ BR,
2 · · · tαn
1 tα2
n
u ≺et v if
π(u) ≺ π(v),
or
π(u) = π(v) and u <lex v in BR.
In particular, with respect to the monomial ordering ≺et obtained by using the lexicographic
ordering tn ≺lex tn−1 ≺lex · · · ≺lex t1 on BR, we see that LM(gji) = XjXi for all 1 ≤ i < j ≤ n,
as required by Theorem 2.1.
In [Rin] it was proved that U +
q (AN ) has m = N (N +1)
2
generators x1, ..., xm satisfying the
relations:
xjxi = qvji xixj − rji, 1 ≤ i < j ≤ m, where vji = (wt(xi), wt(xj )), and
i+1 xαi+2
rji is a linear combination of monomials of the form xαi+1
i+2 · · · xαj−1
j−1 ,
and that U +
above relations. Thus U +
G = {gji = xjxi − qvij xixjrji 1 ≤ i < j ≤ m} forms a monic Grobner defining set of U +
with respect to the monomial ordering ≺et as described before.
q (AN ) is an iterated skew polynomial algebra generated by x1, ..., xm subject to the
m αj ∈ N}, and consequently
q (AN )
q (AN ) has the PBW basis {xα1
2 · · · xαm
1 xα2
14
Remark If, in the defining relations given in the last example, the condition i < i1 ≤ i2 ≤ · · · ≤
is < j is replaced by 1 ≤ i1 ≤ i2 ≤ · · · ≤ is ≤ i − 1, then a similar result holds.
The next three examples provide monic Grobner bases which are not necessarily the type as
described in previous Examples 3 -- 4, but they all give rise to PBW R-bases.
Example 5. Let R be a commutative ring, and let I be the ideal of the free R-algebra RhXi =
RhX1, X2i generated by the single element
g21 = X2X1 − qX1X2 − αX2 − f (X1),
where q, α ∈ R, and f (X1) is a polynomial in the variable X1. Assigning to X1 the degree 1,
then in either of the following two cases:
(a) degf (X1) ≤ 2, and X2 is assigned the degree 1;
(b) degf (X1) = n ≥ 3, and X2 is assigned the degree n,
G = {g21} forms an LM-reduced monic Grobner basis for I. For, in both cases we may use the
N-graded lexicographic ordering X1 ≺gr X2 with respect to the natural N-gradation of KhXi,
respectively the weight N-gradation of RhXi with weight {1, n}, such that LM(g21) = X2X1,
and then we see that the only overlap element of G is o(g21, 1; 1, g21) = 0. Thus, by Theorem
2.1 in both cases the algebra A = RhXi/I has the PBW R-basis B = {X
β
2 α, β ∈ N}.
α
1 X
Example 6. Let R be a commutative ring, and let RhXi = RhX1, X2, X3i be the free R-
algebra generated by X = {X1, X2, X3}. This example provides a family of algebras similar to
the enveloping algebra U (sl(2, R)) of the R-Lie algebra sl(2, R), that is, we consider the algebra
A = KhXi/hGi with G consisting of
g31 = X3X1 − λX1X3 + γX3,
g12 = X1X2 − λX2X1 + γX2,
g32 = X3X2 − ωX2X3 + f (X1),
where λ, γ, ω ∈ R, and f (X1) is a polynomial in the variable X1. It is clear that A = U (sl(2, R))
in case λ = ω = 1, γ = 2 and f (X1) = −X1.
Suppose f (X1) has degree n ≥ 1. Then we can always equip RhXi with a weight N-gradation
by assigning to X1, X2 and X3 the positive degree n1, n2, n3 respectively (for instance, (1, 1, 1)
if degf (X1) = n ≤ 2; (1, n, n) if degf (X1) = n > 2), such that LM(G) = {X3X1, X1X2, X3X2}
with respect to the N-graded monomial ordering X2 ≺gr X1 ≺gr X3 on BR. In the case that
R = K is a field, it was verified in ([Li4], Example 7) that G is a Grobner basis for the ideal hGi
in KhXi with respect to the same ≺gr. Hence, by Proposition 1.6, G is a Grobner basis for the
ideal hGi in RhXi. It follows from Theorem 2.1 that the algebra A = RhXi/hGi has the PBW
R-basis B = {X2
α3 αj ∈ N}.
α2X1
α1X3
Let us point out that in the case that f (X1) has degree ≤ 2, i.e., f (X1) is of the form
f (X1) = aX 2
1 + bX1 + c with a, b, c ∈ R,
15
if degX1 = degX2 = degX3 = 1 is used, the algebra A provides R-versions of some popularly
studied algebras over a field K, for instance,
(a) let ζ ∈ R be invertible, and put λ = ζ 4, ω = ζ 2, γ = −(1 + ζ 2), a = 0 = c, and b = −ζ,
then A is just the R-version of the Woronowicz's deformation of U (sl(2, K)) introduced in the
noncommutative differential calculus;
(b) if λγwb 6= 0 and c = 0, then A is just the R-version of Le Bruyn's conformal sl(2, K)
enveloping algebra [LB] which provides a special family of Witten's deformation of U (sl(2, K))
in quantum group theory.
Example 7. Let G be the subset of the free R-algebra RhXi = KhX1, X2, X3i consisting of
g21 = X2X1 − X1X2,
g31 = X3X1 − λX1X3 − µX2X3 − γX2,
g32 = X3X2 − X2X3.
λ, µ, γ ∈ R,
Then, under the N-graded lexicographic ordering X1 ≺gr X2 ≺gr X3 with respect to the natural
N-gradation of RhXi, LM(gji) = XjXi, 1 ≤ i < j ≤ 3, and the only nontrivial overlap element
of G is S321 = o(g32, X1; X3, g21) = −X2X3X1 + X3X1X2. One checks easily that S321
= 0.
By Theorem 2.1, G is an LM-reduced monic Grobner basis for the ideal hGi. Hence, by Theorem
2.1 the algebra A = RhXi/hGi has the PBW R-basis B = {X1
α3 αj ∈ N}.
G
α1X2
α2 X3
3. PBW Isomorphisms and Applications
In this section we show that the working principle via PBW isomorphism developed in [LWZ] and
[Li3] can be generalized to study algebras defined by monic Grobner bases over a commutative
ring R. All notions and notations used in previous sections are maintained.
Let R be an arbitrary commutative ring, RhXi = RhX1, ..., Xni the free R-algebra of n
generators, and BR the standard free R-basis of RhXi. Consider a weight N-gradation of RhXi
subject to deg(Xi) = ni > 0, 1 ≤ i ≤ n, that is, RhXi = ⊕p∈NRhXip with RhXip = R-
span{w ∈ B deg(w) = p}. For an element f ∈ RhXi, say f = F0 +F1 +· · ·+Fp with Fi ∈ RhXii
and Fp 6= 0, let LHN(F ) denote the N-leading homogeneous element of f , i.e., LHN(f ) = Fp.
Then every ideal I of RhXi is associated to an N-graded ideal hLHN(I)i generated by the set of N-
leading homogeneous elements LHN(I) = {LHN(f ) f ∈ I}. Adopting the notion and notation
as in [Li3], we call the N-graded algebra AN
LH = RhXi/hLHN(I)i the N-leading homogeneous
algebra of the algebra A = RhXi/I. On the other hand, noticing that RhXi is also a BR-graded
algebra by the multiplicative monoid BR, i.e., RhXi = ⊕w∈BRRhXiw with RhXiw = Rw, if ≺
i=1 λiwi ∈ RhXi with w1 ≺ w2 ≺ · · · ≺ wn, then
the term λnwn is called the BR-leading homogeneous element of f and is denoted by LHBR(f ).
Thus each ideal I of RhXi is associated to a BR-graded ideal hLHBR(I)i generated by the set of
BR-leading homogeneous elements LHBR(I) = {LHBR(f ) f ∈ I}, and similarly, the BR-graded
is a monomial ordering on BR and if f = Pn
16
algebra ABR
LH = RhXi/hLHBR(I)i is referred to as the BR-leading homogeneous algebra of the
algebra A = RhXi/I. Furthermore, consider the N-grading filtration F NRhXi of RhXi defined
by
and the BR-grading filtration F BRRhXi of RhXi defined by
F N
p RhXi = ⊕i≤pRhXii,
p, i ∈ N,
F BR
w RhXi = ⊕u(cid:22)wRhXiu, w, u ∈ BR.
If I is an ideal of RhXi, then the algebra A = RhXi/I has the N-filtration F NA induced by
F NRhXi, i.e.,
respectively the BR-filtration F BRA induced by F BRRhXi, i.e.,
F N
p A = (F N
p RhXi + I)/I,
p ∈ N,
F BR
w A = (F BR
w RhXi + I)/I, w ∈ BR.
Note that if each Xi has degree 1, 1 ≤ i ≤ n, then the filtration F NA is just the commonly
used natural N-filtration. Let GN(A) = ⊕p∈NGN(A)p with GN(A)p = F N
p−1A be the asso-
ciated N-graded algebra of A determined by F NA, respectively GBR(A) = ⊕w∈BRGBR(A)w with
GBR(A)w = F BR
≺wA the associated BR-graded algebra of A determined by F BRA, where
F BR
u A. We have the following analogue of ([Li3], Theorem 1.1). Since the proof
of this result is similar to that given in loc. cit., we omit it here.
≺wA = ∪u≺wF BR
w A/F BR
p A/F N
3.1. Theorem With notation as above, there are graded R-algebra isomorphisms:
AN
LH = RhXi/hLHN(I)i ∼= GN(A), ABR
LH = RhXi/hLHBR(I)i ∼= GBR(A).
(cid:3)
Since we are using an arbitrary commutative ring R (instead of a field) as the coefficient
ring, the next lemma makes the key bridge for us to generalize the working principle of [LWZ]
and [Li3] to quotient algebras of RhXi defined by monic Grobner bases.
3.2. Lemma Let RhXi be equipped with the fixed weight N-gradation as before, and I an ideal
of RhXi. Put J = hLHN(I)i. The following two statements hold.
(i) If h is a nonzero homogeneous element of RhXi, then h ∈ J if and only if h ∈ LHN(I). Hence
LHN(J) = LHN(I).
(ii) Let ≺gr be an N-graded monomial ordering on BR with respect to the fixed weight N-
gradation of RhXi. Then LHBR(J) = LHBR(I) and LM(J) = LM(I).
(iii) Let ≺gr be an N-graded monomial ordering on BR with respect to the fixed weight N-
gradation of RhXi. If G is a monic Grobner basis of I, then
hLHBR(J)i = hLHBR(I)i = hLM(G)i = hLM(I)i = hLM(J)i.
17
Proof (i) Let h be a nonzero homogeneous element in RhXi. If h ∈ J, then
HijLHN(fi)Tij, where Hij, Tij are homogeneous elements and fi ∈ I.
h =Xi,j
If we write fi = LHN(fi) + f ′
i, where deg(f ′
i) < deg(fi), then f =Pi,j HijfiTij ∈ I and
Hijf ′
iTij.
f =Xij
HijLHN(fi)Tij +Xi,j
Hijf ′
iTij = h +Xi,j
It follows immediately that h = LHN(f ) ∈ LH(I). This shows that LHN(J) ⊆ LHN(I) and
hence the equality holds.
(ii) Note that ≺gr is an N-graded monomial ordering on BR, every element of BR is an N-
homogeneous element, and thus for f ∈ RhXi we have
(∗)
LHBR(f ) = LHBR(LHN(f )) and LM(f ) = LM(LHN(f ))
It follows from (i) and the above formula (∗) that
LHBR(J) = LHBR(LHN(J)) = LHBR(LHN(I)) = LHBR(I),
LM(J) = LM(LHBR(J)) = LM(LHBR(I)) = LM(I).
(iii) Let f ∈ RhXi be a monic element with respect to the fixed monomial ordering ≺gr, say
f = w +P λiwi with w, wi ∈ BR, λi ∈ R and LM(f ) = w. Then it is clear that
LM(f ) = w = LHBR(f ).
(∗∗)
So, if G is a monic Grobner basis of I with respect to ≺gr, then the above formula (∗∗) implies
LM(G) = LHBR(G) ⊂ LHBR(I). Hence, by (ii) and Proposition 1.2 we obtain the desired
equalities:
hLHBR(J)i = hLHBR(I)i = hLM(G)i = hLM(I)i = hLM(J)i.
(cid:3)
Next, we show that an analogue of ([LWZ], Theorem 2.3.2 (i) ⇔ (iii)) holds true for monic
Grobner bases in RhXi.
3.3. Theorem Let I be an ideal of RhXi. With notation as above, if ≺gr is an N-graded
monomial ordering on BR with respect to a fixed weight N-gradation of RhXi, the following two
statements are equivalent for a subset G ⊂ I:
(i) G is a monic Grobner basis of I ;
(ii) LHN(G) = {LHN(g) g ∈ G} is a monic Grobner basis for the N-graded ideal hLHN(I)i.
Proof Since we are using the N-graded monomial ordering ≺gr on BR, by Lemm 3.2 or its proof,
a subset G of RhXi is monic if and only if LHN(G) is monic, and we have
hLM(I)i = hLM(G)i if and only if hLM(hLHN(I)i)i = hLM(LHN(G))i.
18
It follows from Proposition 1.2 that G is a monic Grobner basis for the ideal I if and only if
LHN(G) is a monic Grobner basis for the N-graded ideal hLHN(I)i, proving the equivalence of
(i) and (ii).
(cid:3)
Remark Also we point out that an analogue of ([LWZ], Theorem 2.3.2 (i) ⇔ (ii)) works well
for monic Grobner bases when the homogenization of I in RhXi[t] is considered.
Combining the previous 3.1 -- 3.3, we get immediately the result presenting the associated
N-graded algebra, respectively the associated BR-graded algebra via a monic Grobner basis.
3.4. Theorem Let RhXi be equipped with a fixed weight N-gradation as before, and I an ideal
of RhXi. If G is a monic Grobner basis of I with respect to an N-graded monomial ordering ≺gr
on BR, then we have the graded algebra isomorphisms
LH = RhXi/hLHN(I)i = RhXi/hLHN(G)i ∼= GN(A),
AN
LH = RhXi/hLHBR(I)i = RhXi/hLM(G)i ∼= GBR(A),
ABR
LH)BR
(AN
LH = RhXi/hLHBRhLHN(I)i)i = RhXi/hLM(G)i ∼= GBR(AN
LH).
(cid:3)
As in [Li3] we call the graded algebra isomorphisms presented in the last theorem the N-
PBW isomorphism (for the first one) and BR-PBW isomorphism (for the last two), determined
by the given monic Grobner basis G respectively.
Focusing on the first isomorphism of Theorem 3.4, typical examples can be given by using
the Grobner defining relations of Weyl algebras and enveloping algebras of Lie algebras, or more
generally, the Grobner defining relations of q-enveloping algebras determined in Example 3 of the
last section, over a commutative ring. Here we specify several other examples. In all examples
given below, R is an arbitrary commutative ring.
Example 1. Let X = {Xi}i∈J and C = RhXi/hGi the Clifford algebra over R, where G consists
of
i − qi,
gi = X 2
gkℓ = XkXℓ + XℓXk − qkℓ, k, ℓ ∈ J, k > ℓ, qkℓ ∈ R.
i ∈ J, qi ∈ R,
Note that if all the qi = 0, qk,ℓ = 0, we get the defining relations of an R-exterior algebra. It is
well known that if R = K is a field, then, under the N-graded lexicographic ordering ≺gr such
that degXi = 1, i ∈ J, and
Xℓ ≺gr Xk,
ℓ, k ∈ J, ℓ < k,
G forms a Grobner basis for the ideal hGi in KhXi (e.g., see CH.II of [Li2]). It follows from
Proposition 1.6 that G is a Grobner basis for the ideal hGi in RhXi. By Theorem 3.4, with respect
19
to the natural N-filtration F NC of C, the associated N-graded algebra GN(C) ∼= RhXi/hLHN(G)i
of C is nothing but an exterior algebra E over R.
Example 2. Let A = RhX1, X2i/hGi be a down-up R-algebra in the sense of [Ben], where G
consists of
g1 = X 2
g2 = X1X 2
1 X2 − αX1X2X1 − βX2X 2
2 − αX2X1X2 − βX 2
1 − γX1,
2 X1 − γX2,
α, β ∈ R.
It is well known that if R = K is a field, then, under the N-graded lexicographic ordering ≺gr
such that degX1 = degX2 = 1 and X2 ≺gr X1, G forms a Grobner basis for the ideal hGi in KhXi
(e.g., see CH.II of [Li2]). It follows from Proposition 1.6 that G is a Grobner basis for the ideal
hGi in RhXi. By Theorem 3.4, with respect to the natural N-filtration F NA of A, the associated
N-graded algebra GN(A) ∼= RhX1, X2i/hLHN(G)i of A is a down-up algebra over R with the
set of defining relations LHN(G) = {LHN(g1) = X 2
1 , LHN(g2) =
X1X 2
2 X1}; in particular, one sees that if α = 2 and β = −1, then GN(A)
is nothing but the universal enveloping algebra of the (−)-part (or (+)-part) of the Kac-Moody
1 X2 − αX1X2X1 − βX2X 2
2 − αX2X1X2 − βX 2
R-Lie algebra associated to the Cartan matrix 2 −1
2 !.
−1
Example 3. Let A = RhX1, X2i/hg21i be the R-algebra as given in (Section 2, Example 5).
Then by Theorem 3.4, with respect to both the natural N-filtration and the weight N-filtration
induced by the weight N-grading filtration of RhX1, X2i, A has the associated N-graded algebra
GN(A) ∼= RhX1, X2i/hX2X1 − qX1X2i, which, in the case that q is invertible, is the coordinate
ring of the quantum plane over R.
Example 4. Let A = RhX1, X2, X3i/hGi be the R-algebra as given in (Section 2, Example 6).
In the case that f (X1) has degree ≤ 2, i.e., f (X1) is of the form
f (X1) = aX 2
1 + bX1 + c with a, b, c ∈ R,
then by Theorem 3.4, with respect to the natural N-filtration F NA, A has the associated N-
graded algebra GN(A) ∼= RhX1, X2, X3i/hLHN(G)i with
LHN(G) = {X3X1 − λX1X3, X1X2 − λX2X1, X3X2 − ωX2X3 + aX 2
1 };
while in the case that f (X1) has degree n ≥ 3, if the weight (1, n, n) is used, then by Theorem
3.4, with respect to the weight N-filtration F NA induced by the weight N-grading filtration of
RhX1, X2i, A has the associated N-graded algebra GN(A) ∼= RhX1, X2, X3i/hLHN(G)i with
LHN(G) = {X3X1 − λX1X3, X1X2 − λX2X1, X3X2 − ωX2X3}
By referring to the well-known filtered-graded comparison principle for algebras with an N-
filtration ([MR], [Li1], [LVO1], [Li3]), we now summarize, without proof, several applications of
20
Theorem 3.3 and Theorem 3.4. Let R be an arbitrary commutative ring. For the convenience,
in what follows we let the free R-algebra RhXi = RhX1, ..., Xni be equipped with a fixed weight
N-gradation, I = hGi an ideal of RhXi generated by a monic Grobner basis G with respect to
an N-graded monomial ordering ≺gr on the standard R-basis BR of RhXi, and A = RhXi/I.
Then the following diagram may indicate how all results to be given will work:
A = RhXi/hGi
✒
lifting
GBR(A)
∧
∼= BR−PBW
■❅
❅
❅ lifitng
❅
❅
❅
GN(A)
∼=←−
N−PBW
RhXi
hLHN(G)i
= AN
LH
∧
lifting
ABR
LH =
RhXi
hLM(G)i
∼=
BR−PBW
> GBR(AN
LH)
3.5. Theorem Under the respective canonical algebra epimorphism, the set N (G) of normal
monomials in BR (mod G), projects to a free R-basis for the algebras A = RhXi/I, AN
LH =
RhXi/hLHN(I)i , and ABR
LH = RhXi/hLM(I)i respectively, and thereby to a free R-basis for
GN(A), GBR(A), and GBR(AN
(cid:3)
LH(A)), respectively.
LH is BR-graded left Artinian, that is, ABR
LH = RhXi/hLM(G)i in mind, the following statements hold.
LH is a (semi-)prime ring, then AN
LH is a (semi-)prime ring (hence GN(A) is a (semi-)prime
LH is BR-graded left Noetherian, that is, every BR-graded left ideal of G(A) is finitely
LH is left Noetherian (hence GN(A) is left Noetherian), and A is left Noetherian.
LH satisfies the descending chain condition for
LH is left Artinian (hence GN(A) is left Artinian), and A is left
3.6. Theorem Bearing ABR
(i) If ABR
ring), and A is a (semi-)prime ring.
(ii) If ABR
generated, then AN
(iii) If ABR
BR-graded left ideals, then AN
Artinian.
(iv) If ABR
ideal, then AN
R-algebra.
(v) If the Krull dimension (K.dim in the sense of Gabriel and Rentschler, e.g. see [MR] for the
definition) of ABR
LH (hence of GN(A)) and A is
defined and K.dimA ≤ K.dim GN(A) = K.dimAN
(vi) If ABR
GN(A) is semisimple (simple) Artinian), and A is semisimple (simple) Artinian.
(vii) Let gl.dim abbreviate the phrase "global homological dimension". We have gl.dimA ≤
gl.dimGN(A) = gl.dimAN
LH is well-defined, then the Krull dimension of AN
LH ≤ K.dimABR
LH.
LH is a BR-graded simple R-algebra, that is, ABR
LH does not have nontrivial BR-graded
LH is a simple R-algebra (hence GN(A) is a simple R-algebra), and A is a simple
LH is semisimple (simple) Artinian, then AN
LH is semisimple (simple) Artinian (hence
LH ≤ gl.dimABR
LH.
21
LH is left hereditary, then AN
LH ≤ gl.wdimABR
LH.
LH is left hereditary (hence GN(A) is left hereditary), and
(viii) If ABR
A is left hereditary.
(ix) Let gl.wdim abbreviate the phrase "global week homological dimension". We have
gl.wdimA ≤ gl.wdimGN(A) = gl.wdimAN
(x) If ABR
is a Von Neuman regular ring), and A is a Von Neuman regular ring.
LH is Von Neuman regular ring (hence GN(A)
(cid:3)
LH is a Von Neuman regular ring, then AN
LH = RhXi/hLHN(G)i in mind, if the role of ABR
3.7. Theorem Bearing AN
then the analogues of Theorem 4.6 (i) -- (x) hold true. Moreover, we have:
(i) If AN
(ii) If AN
definition), then A is a Noetherian domain and maximal order in its quotient ring.
(iii) If AN
Auslander regular ring.
LH is a domain, then A is a domain.
LH is a Noetherian domain and maximal order in its quotient ring (see e.g. [MR] for the
LH is an Auslander regular ring (see e.g. [Li1], [LVO] for the definition), then A is an
(cid:3)
LH is replaced by AN
LH,
References
[An] D. J. Anick, On the homology of associative algebras, Trans. Amer. Math. Soc.,
2(296)(1986), 641 -- 659.
[Ben] G. Benkart, Down-up algebras and Witten's deformations of the universal enveloping
algebra of sl2, Contemp. Math., 224(1999), 29 -- 45.
[Ber] R. Berger, The quantum Poincar´e-Birkhoff-Witt theorem, Comm. Math. Physics,
143(1992), 215 -- 234.
[Ber1] G. Bergman, The diamond lemma for ring theory, Adv. Math., 29(1978), 178 -- 218.
[BM] L. A. Bokut and P. Malcolmson, Grobner-Shirshov basis for quantum enveloping algebras,
Israel Journal of Mathematics, 96(1996), 97 -- 113.
[CE] B.L. Cox and T.J. Enright, Representations of quantum groups defined over commutative
rings, Comm. Alg, 23(6)(1995), 2215 - 2254.
[CSS] Y. Chen, H. Shao and K. P. Shum, Rosso-Yamane Theorem on PBW basis of Uq(AN ),
CUBO, A Mathematical Journal, 10(3)(2008), 171 -- 194. arXiv:0804.0954v1 [math.RA]
[CU] S. Cojocaru and V. Ufnarofski, BERGMAN under MS-DOS and Anick's resolution, Dis-
crete Mathematics and Theoretical Computer Science, 1(1997), 139 -- 147.
[EPS] D. Eisenbud, I. Peeva and B. Sturmfels, Non-commutative Grobner bases for commutative
algebras, Proc. Amer. Math. Soc., 126(1998), 687 -- 691.
[GI-L] T. Gateva-Ivanova and V. Latyshev, On recognizable properties of associative algebras, J.
Symbolic Computation, 6(1988), 371 -- 388.
[Gol] E. S. Golod, On noncommutative Grobner bases over rings, Journal of Mathematical
Science, 140(2)(2007), 239 -- 242.
[Gr] E. L. Green, Noncommutative Grobner bases and projective resolutions, in: Proceedings
22
of the Euroconference Computational Methods for Representations of Groups and Alge-
bras, Essen, 1997, (Michler, Schneider, eds), Progress in Mathematics, Vol. 173, Basel,
Birkhauser Verlag, 1999, 29 -- 60.
[Jac] N. Jacobson, Lie Algebras, New York, Wiley, 1962.
[LB] L. Le Bruyn, Conformal sl2 enveloping algebras, Comm. Alg., 23(1995), 1325 -- 1362.
[Li1] H. Li, Noncommutative Zariskian Filtered Rings, PhD Thesis, Antwerp University, 1989.
[Li2] H. Li, Noncommutative Grobner Bases and Filtered-Graded Transfer, LNM, 1795,
Springer-Verlag, 2002.
[Li3] H. Li, Γ-leading homogeneous algebras and Grobner bases, in: Advanced Lectures in Math-
ematics, Vol.8, International Press, Boston, 2009, 155 -- 200.
[Li4] H. Li, Looking for Grobner basis theory for (almost) skew 2-nomial algebras, Journal of
Symbolic Computation, 45(2010), in press, online 10 may 2010, doi:10.1016/j.jsc.2010.05.002.
(also see arXiv:math.RA/0808.1477)
[LVO1] H. Li and F. Van Oystaeyen, Zariskian Filtrations, K-Monograph in Math., Vol.2, Kluwer
Academic Publishers, 1996.
[LVO2] H. Li and F. Van Oystaeyen, Reductions and global dimension of quantized algebras over
a regular commutative domain, Comm. Alg., 26(4)(1998), 1117 -- 1124.
[LWZ] H. Li, Y. Wu and J. Zhang, Two applications of noncommutative Grobner bases, Ann.
Univ. Ferrara - Sez. VII - Sc. Mat., XLV(1999), 1 -- 24.
[Man] Yu.I. Manin, Quantum Groups and Noncommutative Geometry, Les Publ. du Centre de
R´echerches Math., Universite de Montreal, 1988.
[MR] J.C. McConnell and J.C. Robson, Noncommutative Noetherian Rings, John Wiley & Sons,
1987.
[Mor] T. Mora, An introduction to commutative and noncommutative Grobner bases, Theoretic
Computer Science, 134(1994), 131 -- 173.
[Rin] C. M. Ringel, PBW-bases of quantum groups, J. Reine Angew. Math. 470 (1996), 51 -- 88.
[Ros] M. Rosso, An analogue of the Poincare-Birkhoff-Witt theorem and the universal R-matrix
of Uq(sl(N + 1)), Comm. Math. Phys., 124(2)(1989), 307 -- 318.
[Uf1] V. Ufnarovski, A growth criterion for graphs and algebras defined by words, Mat. Zametki,
31(1982), 465 -- 472 (in Russian); English translation: Math. Notes, 37(1982), 238 -- 241.
[Uf2] V. Ufnarovski, On the use of graphs for computing a basis, growth and Hilbert series of
associative algebras, (in Russian 1989), Math. USSR Sbornik, 11(180)(1989), 417-428.
[Yam] I. Yamane, A Poincare-Birkhoff-Witt theorem for quantized universal enveloping algebras
of type AN , Publ., RIMS. Kyoto Univ., 25(3)(1989), 503 -- 520.
23
|
1708.03032 | 1 | 1708 | 2017-08-09T23:26:46 | Group gradings on the Jordan algebra of upper triangular matrices | [
"math.RA"
] | Let G be an arbitrary group and let K be a field of characteristic different from 2. We classify the G-gradings on the Jordan algebra of upper triangular matrices of order n over K. It turns out that there are, up to a graded isomorphism, two families of gradings: the elementary gradings (analogous to the ones in the associative case), and the so called mirror type (MT) gradings. Moreover we prove that the G-gradings on this algebra are uniquely determined, up to a graded isomorphism, by the graded identities they satisfy. | math.RA | math |
Group gradings on the Jordan algebra of upper
triangular matrices
Plamen Emilov Koshlukov∗
Felipe Yukihide Yasumura†
Department of Mathematics, State University of Campinas
651 Sergio Buarque de Holanda
13083-859 Campinas, SP, Brazil
e-mail addresses: plamen, [email protected]
Abstract
Let G be an arbitrary group and let K be a field of characteristic
different from 2. We classify the G-gradings on the Jordan algebra UJn
of upper triangular matrices of order n over K. It turns out that there
are, up to a graded isomorphism, two families of gradings: the elementary
gradings (analogous to the ones in the associative case), and the so called
mirror type (MT) gradings. Moreover we prove that the G-gradings on
UJn are uniquely determined, up to a graded isomorphism, by the graded
identities they satisfy.
Keywords: Jordan algebras, Group gradings, Upper triangular matrices.
MSC: 17C05; 17C99; 16W50; 17C50, 16R99
Introduction
Graded algebras were first introduced and studied in Commutative algebra in
order to generalize properties of polynomial rings. The study of the gradings on
associative algebras was initiated by Wall [12]. He described the finite dimen-
sional graded simple algebras with grading group Z2, the cyclic group of order 2.
Much later, around 1985, Kemer developed the structure theory of the T-ideals
(ideals of identities) in the free associative algebra, see for instance [8]. One of
the principal ingredients of that theory is the study of Z2-graded algebras and
their graded identities. The theory developed by Kemer has since immensely
influenced the research in PI theory, and motivated further study of gradings
and on graded polynomial identities in associative algebras. Motivated in part
∗Partially supported by FAPESP grant No.
2014/09310-5, and by CNPq grant No.
304632/2015-5
†Supported by PhD grant 2013/22802-1 from FAPESP
1
by Kemer's results, gradings on algebras became an object of extensive study.
We cite here the recent monograph [7] and the bibliography therein for further
and more detailed information concerning gradings on algebras. We recall some
of the cornerstone results in the area that will be used in the exposition below.
In order to do it in a precise way we have to introduce now part of the notions
we will need.
Let G be a group, K a field, and A an algebra (that is a vector space over K
equipped with a bilinear multiplication). We do not require the multiplication
in A to be either commutative or associative. The algebra A is G-graded if
A=⊕g∈GAg where the subspaces Ag ⊆ A satisfy AgAh ⊆ Agh, g, h ∈ G. A vector
subspace V of A is homogeneous, or graded, if V = ⊕g∈GV ∩ Ag. In the same
way one defines graded subalgebras and ideals of A.
The gradings on the (associative) matrix algebras of order n were completely
described by Bahturin and Zaicev, see for example [1]. Later on this description
was extended, in the case of abelian groups, to simple associative algebras with
a minimal one sided ideal (over an algebraically closed field), see [7, pp. 27, 28].
Similar results were obtained for simple Lie algebras, see for example [7]; simple
Jordan algebras, see [2] and its bibliography.
We note that relatively little is known about the classification of the grad-
ings on important algebras that are not simple. The Grassmann (or exterior)
algebra appears naturally in various branches of Mathematics and Physics but
the gradings on this algebra are known under restrictions on the gradings [6].
described in [9]. Note that the latter algebra is a Jordan algebra of a symmetric
bilinear form though the form is degenerate.
The gradings on the Jordan algebra of the 2× 2 upper triangular matrices were
The elementary gradings on the associative algebra U Tn(K) of the upper
triangular matrices of order n over a field K were described in [5]. Also, in [11],
the authors proved that every grading on U Tn is, up to a graded isomorphism,
elementary. Hence one has a complete classification of the gradings on the
associative algebra of upper triangular matrices.
UJ2 given in [9].
In this paper we study the gradings on the Jordan algebra UJn of the upper
triangular matrices of order n over an infinite field. We describe completely all
these gradings. Our principal result states that every group grading on UJn is
either elementary or is of the so-called mirror type (MT for short).
For the particular case n = 2, we recall the description of the gradings on
Theorem 1 ([9]). Denote 1 = e11 + e22, a = e11 − e22, b = e12 ∈ UJ2. Let
UJ2 = J0 + J1 be Z2-graded. Then the grading is isomorphic to one of the
0. The trivial grading: J0 = UJ2;
1. The associative grading: J0 = 1K ⊕ bK, J1 = aK;
2. The scalar grading: J0 = 1K, J1 = aK ⊕ bK;
3. The classical grading: J0 = 1K ⊕ aK, J1 = bK.
following gradings:
2
The above theorem will be a particular case of our results. We recall that
in [9] the authors also described the graded polynomial identities satisfied by
each of the possible gradings, including the trivial one. Here we are not going
to do that. It seems to us that the description of the graded identities satisfied
by UJn is a very complicated problem, and even in the simplest cases it turned
out to be far from our reach.
n
the vector space of the upper triangular
1 Notations
Here, we will denote UJn = U T (+)
matrices of size n equipped with the Jordan product a○ b= ab+ ba. We denote
the associator by (a, b, c)=(a○ b)○ c− a○(b○ c). We shall omit the parentheses
in left normed products, that is a○ b○ c stands for (a○ b)○ c, and similarly for
We denote by eij the matrix units having entry 1 at position (i, j), and 0
elsewhere. Given a matrix x ∈ UJn we denote by (x)(i,j) the (i, j)-th entry of
x. Also for i, m∈ N we define
products of more than 3 elements.
ei∶m = ei,i+m,
(x)(i∶m) =(x)(i,i+m),
e−i∶m = en−i−m+1,n−i+1,
(x)(−i∶m) =(x)(n−i−m+1,n−i+1).
Let G be a group with neutral element 1. Unless otherwise stated we use
the multiplicative notation, even if the group is abelian. Of course we write
additively the groups (Z,+), and (Zn,+).
First we classify the elementary and MT gradings (to be defined below), and
then prove that every grading on UJn is, up to a graded isomorphism, either
elementary or MT.
2 Elementary Gradings
Let G be any group. We call a G-grading on UJn elementary if all matrix units
eij are homogeneous in the grading.
Lemma 2. Let UJn be equipped with an elementary G-grading. Then
(i) deg eii = 1, i= 1, . . . , n.
(ii) The sequence η = (deg e12, deg e23, . . . , deg en−1,n) defines completely the
grading.
(iii) The support of the grading is commutative.
Proof. The statements of the lemma and their proofs are standard facts, we give
these proofs for the sake of completeness.
(i) Since eii ○ eii = 2eii we have (deg eii)2 = deg eii hence deg eii = 1.
(ii) It follows from eij = ei,i+1○ ei+1,i+2○⋯○ ej−1,j.
3
(iii) Let t1 = deg e12, t2 = deg e23, . . . , tn−1 = deg en−1,n. By (ii), it suffices to
prove that titj = tjti for all i, j ∈{1, 2, . . . , n− 1}. But if i< j then
ei,i+1○(ej,j+1○(ei+1,i+2 ○⋯○ ej−1,j))= ej,j+1 ○(ei,i+1○ ei+1,i+2○⋯○ ej−1,j).
Thus titjti+1⋯tj−1 = tjtiti+1⋯tj−1 and titj = tjti.
Notation. We denote by (UJn, η) the elementary grading defined by η ∈
Gn−1. This grading is defined by putting deg ei,i+1 = gi, for each i, where
η =(g1, g2, . . . , gn−1). We denote by rev η =(gn−1, gn−2, . . . , g1).
Lemma 3. Let η ∈ Gn−1. The map ϕ∶(UJn, η) → (UJn, rev η) given by eij ↦
From here on in this section, we assume that G is abelian.
en−j+1,n−i+1 is an isomorphism of G-graded algebras.
Proof. The proof is a direct and easy verification.
We recall notations and results of [4, 13]. Denote by Sm the set of permu-
tations of m elements and let
Tm ={σ ∈ Sm σ(1)>⋯> σ(t)= 1, σ(t+ 1)<⋯< σ(m)}.
Using same argument as [13, Lemma 3 (ii)], one can prove
Lemma 4. Let r1, . . . , rm be strictly upper triangular matrix units such that
the associative product r1⋯rm ≠ 0, and let σ ∈ Sm. Then rσ−1(1)○⋯○ rσ−1(m) ≠ 0
if and only if σ ∈ Tm.
In analogy with [5] we define
Definition 5. Let G be a group and let (UJn, η) be an elementary G-grading.
Let µ=(a1, . . . , am)∈ Gm be any sequence.
1. (See [5]) The sequence µ is associative η-good if there exist strictly upper
triangular matrix units r1, . . . , rm ∈ U Tn such that r1⋯rm ≠ 0 and deg ri =
ai for every i= 1, . . . , m. Otherwise µ is associative η-bad sequence.
matrix units r1, . . . , rm such that r1○⋯○ rm ≠ 0 and deg ri = ai, for every
i= 1, . . . , m. Otherwise µ is Jordan η-bad sequence.
2. The sequence µ is Jordan η-good if there exist strictly upper triangular
Definition 6. If µ=(a1, a2,⋯, am)∈ Gm we define
○⋯○ f (am)
○ f (a2)
3h ),
3h−2, x(1)
3h−1, x(1)
x(a)
h ,
fµ = f (a1)
h = (x(1)
where
f (a)
m
1
2
if a= 1,
if a≠ 1
.
The following lemma is proved exactly in the same way as Proposition 2.2
of [5].
4
for (UJn, η).
Lemma 7. A sequence µ is Jordan η-bad if and only if fµ is a G-graded identity
If S is any set and s = (s1, . . . , sm) ∈ Sm is any sequence of symbols, we
define the left action of Sm on Sm by
σs =(sσ−1(1), . . . , sσ−1(m)), σ ∈ Sm.
Lemma 7, we obtain
The unique non-zero associative product of n−1 strictly upper triangular matrix
units of U Tn is e12e23⋯en−1,n (see [5]), so combining this fact, Lemma 4, and
Lemma 8. A sequence µ ∈ Gn−1 is Jordan η-good for (UJn, η) if and only if
µ= ση for some σ ∈ Tn−1.
The following lemma was proved in [13].
Lemma 9 ([13]). Let s, s′ ∈ Sm be any sequences where S is a set. Then s = s′
or s = rev s′ if and only if for every σ, τ ′ ∈ Tn−1 we can find σ′, τ ∈ Tn−1 such
that σs = σ′s′ and τ s = τ ′s′.
Combining Lemmas 8 and 9, we obtain
Corollary 10. Let η, η′ ∈ Gn−1 with η ≠ η′ and η ≠ rev η′. Then (UJn, η) ~≃
(UJn, η′).
Proof. By Lemma 9, there exists σ ∈ Tm such that ση ≠ σ′η′ for each σ′ ∈ Tm,
but Jordan η′-bad sequence, hence fση is not a graded identity for (UJn, η), but
it is a graded identity for (UJn, η′). In particular, (UJn, η) ~≃(UJn, η′).
interchanging η and η′ if necessary. By Lemma 8, ση is Jordan η-good sequence
In this way we have a classification of the elementary gradings on UJn:
Theorem 11. The support of an elementary G-grading on UJn is commutative.
Let G be an abelian group and define the equivalence relation on Gn−1 as
follows. Let µ1 and µ2 = (a1, a2, . . . , an−1) ∈ Gn−1, then µ1 ∼ µ2 whenever
µ1 = µ2 or µ1 =(an−1,⋯, a2, a1).
Then there is 1 -- 1 correspondence between Gn−1~ ∼ and the class of non-
isomorphic elementary G-gradings on UJn.
3 MT Gradings
Notation. If i, m∈ N we denote Y +
Remark. In the above notation, if n− m is odd then Y +
i∶m = 0 for i=(n− m+ 1)~2.
i∶m = ei∶m + e−i∶m, and Y −
Y −
i∶m = ei∶m − e−i∶m.
i∶m = 2ei∶m = 2e−i∶m, and
Definition 12. A G-grading on UJn is of mirror pattern type (or just MT) if
all Y +
i∶m, Y −
Lemma 13. One has Y +
i∶m are homogeneous and deg Y +
i+1∶1 ○⋯○ Y +
i∶1○ Y +
i∶m.
i∶m ≠ deg Y −
i+m−1∶1 = λY +
i∶m for some λ= 2p, p ∈ Z.
5
If Y +
Proof. Induction on m. When m= 1 the statement is trivial, so assume m > 1.
i+m∶1 =
λ′(ei∶m+1+ e−i∶m+1).
i∶m = λ(ei∶m + e−i∶m) then (λY +
i+m−1∶1 = λY +
i+1∶1 ○ ⋯ ○ Y +
i∶m) ○ Y +
i∶1 ○ Y +
Lemma 14. Let a G-grading on UJn be MT, then
(i) deg Y +
element of order 2.
(ii) Let q = ऄ n−1
the element t= deg Y −
i∶0 = 1 for every i, and deg Y −
1∶0 = deg Y −
2 अ, then the sequence η = (deg Y +
1∶1, deg Y +
1∶0 completely define the grading.
2∶0 = ⋯ = deg Y −
⌊ n
2 ⌋∶0 = t is an
q∶1) and
2∶1, . . . , deg Y +
(iii) The support of the grading is commutative.
Moreover, if the elements Y ±
geneous, then the grading is necessarily MT.
i∶m, for each i and for m = 0 and m = 1, are homo-
1∶0○Y +
i+1∶0○Y ±
i∶0)2 = 2Y +
1∶1 = Y −
1∶1, (Y ±
i∶1 = Y −
i∶0○Y ±
i∶0 and Y −
i∶1
Proof. (i) The equalities Y −
yield the proof.
(ii) It follows from Lemma 13.
(iii) According to (ii), the elements deg Y −
i∶1, for all i, generate
the support of the group. Using Lemma 13 and the same idea as of Lemma
2.(iii), we prove the statement.
1∶0 and deg Y +
then the decomposition of A into symmetric and skew-symmetric elements with
If, moreover, A is endowed
We assume from now on in this section G abelian. We denote by (UJn, t, η)
the MT-grading defined by t∈ G and the sequence η ∈ Gq.
It is well known that, if we have an associative algebra with involution(A,∗),
respect to ∗ gives rise to a Z2-graded algebra.
with an H-grading and ∗ is a graded involution (that is, deg a∗ = deg a, for
all homogeneous a ∈ A), then the decomposition cited yields an H × Z2-graded
The upper triangular matrices possess a natural involution, given by ψ ∶
ei∶m ∈ U Tn ↦ e−i∶m ∈ U Tn. For an elementary grading η on U Tn, ψ will be a
graded involution if and only if η = rev η. It is easy to see that the obtained
Jordan algebra.
grading by the involution is an MT-grading.
The following example, in its present form, was suggested by the Referee.
Z4, and deg Y −
Example. Let G = Z4 and take the MT-grading on UJ4 given by deg Y +
i∶1 = 1 ∈
i∶0 = 2 ∈ Z4, for every i. Since Z4 is an indecomposable group, it
cannot be written in the form Z2 × H. Therefore there exist MT-gradings that
cannot be given by the involution.
Below we classify all MT gradings. We recall that, according to [3], ev-
ery automorphism of UJn is given either by an automorphism of U Tn, or by
an automorphism of U Tn followed by the involution ei∶m ↦ e−i∶m. In particu-
lar, the ideal J of all strictly upper triangular matrices is invariant under all
automorphisms of UJn.
6
2 ⌉ and t1, t2 ∈ G are elements of order
Lemma 15. Let η, η′ ∈ Gq where q =⌈ n−1
2. If t1 ≠ t2 then (UJn, t1, η) ~≃(UJn, t2, η′).
Proof. If ψ∶(UJn, t1, η)→(UJn, t2, η′) is a graded isomorphism then ¯ψ∶ UJn~J →
UJn~J will be a graded isomorphism which is impossible when t1 ≠ t2.
Lemma 16. Let t ∈ G be an element of order 2 and let η = (g1, . . . , gq), η′ =
(g′
2 ⌉. Assume that one of the following holds:
2 ⌋ such that gi ~≡ g′
i (mod ⟨t⟩), or
q)∈ Gq where q =⌈ n−1
• there is an i, 1≤ i≤⌊ n−1
• n is even and gq ≠ g′
1, . . . , g′
q.
Then (UJn, t, η) ~≃(UJn, t, η′).
Proof. Let ϕ∶ G → G0 = G~⟨t⟩ be the canonical projection. The induced G0-
grading on UJn by ϕ and by(UJn, t, η) coincides with the elementary G0-grading
(UJn, η0) where
η0 = (ϕ(g1), ϕ(g2), . . . , ϕ(gq), ϕ(gq), ϕ(gq−1), . . . , ϕ(g1)), if n is odd,
(ϕ(g1), ϕ(g2), . . . , ϕ(gq−1), ϕ(gq), ϕ(gq−1), . . . , ϕ(g1)), if n is even.
A G-graded isomorphism ψ∶(UJn, t, η) → (UJn, t, η′) induces a G0-graded iso-
morphism (UJn, η0) → (UJn, η′
0 = rev η′
Now, assume n even and gq ≠ g′
i∶m (i, m)~∈{(q, 1),(q, 0)}}
0) if and only if η0 = η′
q. Let T = Span{Y ±
Theorem 11. This proves the first condition.
(note that T is invariant under all automorphisms of UJn). Then T is a graded
2 ⌉. Assume that
2 ⌋,
i (mod ⟨t⟩), for i= 1, 2, . . . , p where p =⌊ n−1
ideal, and UJn~T ≃ UJ2. But (UJn~T, t, η) ~≃(UJn~T, t, η′) if gq ≠ g′
Lemma 17. Let t ∈ G be of order 2, η = (g1, . . . , gq), η′ = (g1, . . . , g′
where q =⌈ n−1
i) gi ≡ g′
ii) if n is even then gq = g′
Then (UJn, t, η)≃(UJn, t, η′).
Proof. For every i = 1, 2, . . . , p, let ǫi = 1 if gi = g′
i and ǫi = −1 if gi ≠ g′
ǫ = ǫ1ǫ2⋯ǫp and A = diag(ǫ, ǫ, . . . , ǫ, ǫp−1ǫp−2⋯ǫ1, ǫp−2⋯ǫ1, . . . , ǫ2ǫ1, ǫ1, 1). The
map (UJn, t, η)→(UJn, t, η′) given by x↦ AxA−1 is a graded isomorphism.
q) ∈ Gq
0 (since η′
i. Let
0), by
q.
q.
The following theorem classifies the MT gradings on UJn.
Theorem 18. Every MT grading has commutative support. If G is abelian,
then there is 1 -- 1 correspondence between the non-isomorphic MT gradings on
UJn and the set M where
1. if n is odd, M ={(t, η) t∈ G, o(t)= 2, η ∈(G~⟨t⟩) n−1
2 },
2 × G}.
2. if n is even, M ={(t, η) t∈ G, o(t)= 2, η ∈(G~⟨t⟩) n−2
7
4 General gradings on UJn
Let G be any group and fix a G-grading on UJn. The ideal J of all strictly
upper triangular matrices is homogeneous since J = (UJn, UJn, UJn). Also the
element e1n is always homogeneous since Span{e1n}= J n−1.
As a consequence B = AnnUJn{e1n} = {x ∈ UJn (x)(1,1) + (x)(n,n) = 0}
is homogeneous, and B2 = B ○ B = {x ∈ UJn (x)(1,1) = (x)(n,n)} is as
It follows C = B ∩ B2 = {x ∈ UJn (x)(1,1) = (x)(n,n) = 0} is homo-
1 = {x ∈ UJn (x)(i,j) =
geneous. Let U1 = AnnUJn(C~J) and let T1 = U ○n
0, for i≠ 1 or (i, j)≠(i, n)}, the n-th power of U1. It is easy to see that T1 is
an ideal (moreover, a graded ideal). A similar trick in the associative case can
be found in the proof of Lemma 2 of [10].
well.
Lemma 19. There exists a homogeneous element e2 ∈ T1 such that (deg e2)2 = 1
and e2 ≡ e11− enn (mod T1∩ J).
Proof. Note first that A= T1~T1∩ J is an associative graded algebra whose unit
is ¯e1 = ¯e11 + ¯enn. Hence ¯e1 is graded and deg ¯e1 = 1. Moreover, we can choose
the homogeneous element x ∈ T1 and we can assume that ¯x and ¯e1 are linearly
independent in A. If deg ¯x = 1 then we are done. Otherwise deg ¯x ≠ deg(¯x○ ¯x)
which implies ¯x○ ¯x is a multiple of ¯e1, and this proves the lemma.
Lemma 20. Up to a graded isomorphism, e1 = e11+ enn and e2 = e11− enn are
homogeneous and deg e1 =(deg e2)2 = 1.
Proof. Let e2 be as in the previous lemma, and let e1 = 1
(a) e1 ≡ e11+ enn (mod T1∩ J).
(b) (e1)(1,i) =(e2)(1,i) and (e1)(i,n) =−(e2)(i,n), for i= 2, 3, . . . , n− 1.
(c) (e1)(1,n) =∑n−1
As a consequence of the above properties, the associative product x= e1(e1−1)=
(a) (x)(1,i) = (e1)(1,1)(e1 − 1)(1,i) +(e1)(1,i)(e1− 1)(i,i) = 0, for every i = 1, 2,
(b) (x)(i,n) = 0, for i= 2, 3, . . . , n− 1.
i=2 (e2)(1,i)(e2)(i,n).
2 e2○ e2. Note that
. . . , n− 1.
0. Indeed,
(c) Using the above relations one obtains
(x)(1,n) = nQi=1(e1)(1,i)(e1− 1)(i,n)
=(e1)(1,1)(e1− 1)(1,n)
´¹¹¹¹¹¹¹¹¹¹¹¹¹¹¹¹¹¹¹¹¹¹¹¹¹¹¹¹¹¹¹¹¹¹¹¹¹¹¹¹¹¹¹¹¹¹¹¹¹¹¹¹¹¹¹¹¹¹¹¹¸¹¹¹¹¹¹¹¹¹¹¹¹¹¹¹¹¹¹¹¹¹¹¹¹¹¹¹¹¹¹¹¹¹¹¹¹¹¹¹¹¹¹¹¹¹¹¹¹¹¹¹¹¹¹¹¹¹¹¹¹¶
(e1)(1,n)
+ n−1Qi=2(e1)(1,i)(e1− 1)(i,n)
´¹¹¹¹¹¹¹¹¹¹¹¹¹¹¹¹¹¹¹¹¹¹¹¹¹¹¹¸¹¹¹¹¹¹¹¹¹¹¹¹¹¹¹¹¹¹¹¹¹¹¹¹¹¹¶
(e1)(i,n)
= 0.
(d) All remaining entries are evidently zero.
8
then ψ induces a new G-grading on UJn, isomorphic to the original one, such
These equalities show that the minimal polynomial of e1 is z(z− 1), hence e1 is
diagonalizable. If ψ∶ UJn → UJn is the conjugation such that ψ(e1)= e11+ enn,
that e1 = e11+ enn is homogeneous of degree 1.
Consider again the element e2 from Lemma 19. Let r2 = e2○ e1− e2. Then
r2 = e11− enn+ αe1n for some α ∈ F , and moreover, r2 is diagonalizable. Since
inner automorphism ψ′ such that ψ′(e1)= e11+ enn and ψ′(r2)= e11− enn. This
If (A,○) is a Jordan algebra, not necessarily with unit and x∈ A, we denote
(1− x)○ A = {y − x○ y y ∈ A}. Suppose the element e1 = e11 + enn ∈ UJn is
e1 and r2 commute, they are simultaneously diagonalizable, and we can find an
homogeneous. Then the following set is also homogeneous:
concludes the lemma.
∆=(1− e1)○(1− 1
2
e1)○ UJn = UJn−2.
Thus we write UJn = T1 ⊕ ∆. Note that every inner automorphism (conju-
gation) of ∆ by a matrix M can be extended to an inner automorphism of UJn
by the matrix
M ′ =⎛⎜⎝
⎞⎟⎠ .
0
1
0
0 M 0
0
1
0
Therefore we can repeat the argument above for ∆. Thus we suppose that, up
we can also assume the existence of the elements e1 and e2 as in Lemma 20.
to a graded isomorphism, the elements u1 = e22+ en−1,n−1 and u2 = e22− en−1,n−1
are homogeneous and deg u1 =(deg u2)2 = 1. Since M ′ei(M ′)−1 = ei for i= 1, 2,
Take a homogeneous element z1 ∈ J n−2 = Span{e1,n−1, e2n, e1n} such that
(z)(1,n−1) ≠ 0. We can change z to z ○ u1, if necessary, in order to obtain
(z)(1,n) = 0. In this case u2○ z =−e2○ z, hence we have
Lemma 21. In the notation introduced above, deg e2 = deg u2.
If deg e2 = 1 then e11 and enn are homogeneous of degree 1.
Lemma 22. If deg e2 = 1 then, up to a graded isomorphism, the grading is
Proof. If n= 2 then the elements e11, e22, e12 are (up to a graded isomorphism)
homogeneous hence the grading is elementary. If n = 3 we consider the decom-
position UJ3 = T1 ⊕ ∆. Since dim ∆ = 1 it is easy to prove that the grading is
Thus we assume n> 3. We decompose UJn = T1⊕ ∆. In the notation intro-
duced above, by Lemma 21 we have deg u2 = deg e2 = 1. We use an induction
e22 and en−1,n−1 are homogeneous. If z ∈ J is homogeneous with (z)(1,2) = 1
then e12 = (z ○ e11) ○ e22 is homogeneous.
In the same way we obtain that
en−1,n is homogeneous. This implies that the elements e12, e23, . . . , en−1,n are
homogeneous. Therefore the grading is elementary.
to conclude that the grading on ∆ is elementary. In particular the elements
again elementary.
elementary.
9
assume by induction that up to a graded isomorphism, ∆ is equipped with an
MT grading.
Lemma 23. If deg e2 ≠ 1 then, up to a graded isomorphism, the grading is MT.
Proof. First we assume n > 3. Decompose UJn = T1 ⊕ ∆, by Lemma 21, we
Let z′′ ∈ J ∩ T1 be a homogeneous element with (z′′)(1,2) = 1, and let z′ =
z′′ ○ u1. The only non-zero entries of z′ can be (1, 2), (n − 1, n), (1, n − 1),
(2, n), and z′ is homogeneous with (z′)(1,2) = 1. If z = 1
2(z′ ○ u2 + z′ ○ e2) then
z is homogeneous and z = e12 + aen−1,n for some a. Since deg e2 ≠ 1 we have
deg z ≠ deg(z○ e2), hence a≠ 0. Let A= diag(1, 1, . . . , 1, a). Then ψ∶ UJn → UJn
defined by x ↦ AxA−1 is an isomorphism. The induced grading is such that
ψ(z) = e12 + en−1,n, ψ(z ○ e2) = e12 − en−1,n and ψ(Y ±
i∶1, for i = 2, 3, . . . ,
⌈ n−1
2 ⌉. As the latter are homogeneous elements the induced grading is MT.
Now, as in the previous Lemma one proves that whenever deg e2 ≠ 1 and
n = 2 or 3, the grading is MT. When n = 3 and deg e2 ≠ 1, we can find a
homogeneous element of type z = e12 + ae23, where a ∈ F is non-zero. Thus we
can conjugate with the diagonal matrix diag(1, 1, a) as in the general case in
i∶1) = Y ±
order to obtain a MT-grading.
Theorem 24. Every G-grading on UJn has commutative support and, up to a
graded isomorphism, the grading is either elementary or MT.
5 On the graded identities
1
1
1
○ x(t1)
A1, but it is one for A2.
○⋯○ x(t1)
)○n = x(t1)
We have seen that non-isomorphic elementary gradings satisfy different graded
identities. Let G be a group and assume A1 = (UJn, t1, η) is an MT grading,
and either A2 = (UJn, t2, η′) with t1 ≠ t2 or A2 = (UJn, η′′) is an elementary
grading. Then f =(x(t1)
Now let A1 =(UJn, t, η) and A2 =(UJn, t, η′), and assume A1 ~≃ A2. We shall
for it. If ϕ∶ G → G0 = G~⟨t⟩ is the canonical projection we denote the elementary
gradings induced on A1 and on A2 by ϕ as ¯A1 = (UJn, η0) and ¯A2 = (UJn, η′
0).
m ) such that f is a
(a) η0 ≠ η′
0, hence we can find a polynomial f(x(¯g1)
use the notation of the proof of Lemma 16, and we will give an alternative proof
graded identity for ¯A1, but not for ¯A2 (interchanging ¯A1 and ¯A2 if necessary).
This means that
Then we have two possibilities:
is not a graded identity for
, . . . , x(¯gm)
1
1
m )= f(x(g1)
1
+ x(g1t)
1
, . . . , x(gm)
m )
m + x(gmt)
0. In this case, necessarily n is even and, up to a graded isomorphism,
2 , where Ji =(Ai, Ai, Ai), i= 1 and i= 2. Let f = z(1)
q) with gq ≠ g′
3j−1, x(gi)
q. Note that Supp J1~J 2
1 ≠
q ○
2 ○⋯○ z(q)
3j ). Then f is a graded identity
1 ○ z(2)
1
1
, x(g1t)
, . . . , x(gm)
m , x(gmt)
is a graded identity for A1 but not for A2.
g(x(g1)
(b) η0 = η′
η = (g1, . . . , gq) and η′ = (g1, . . . , gq−1, g′
Supp J2~J 2
q−1 ○⋯○ z(n−1)
3j−2, x(1)
z(q+1)
for A2, but not for A1.
=(x(1)
, where z(j)
1
i
10
In this way we have the final result.
Theorem 25. Let A1 and A2 be two G-gradings on UJn. Then A1 ≃ A2 as
graded algebras if and only if TG(A1)= TG(A2).
Acknowledgements
We are thankful to the Referee whose numerous remarks and comments helped
us improve the exposition, and correct some arguments in the main proofs of
the paper. The Example before Lemma 3 was also a Referee's suggestion which
we accepted with gratitude.
References
[1] Yu. A. Bahturin, M. V. Zaicev, Graded algebras and graded identities,
Polynomial identities and combinatorial methods (Pantelleria, 2001), Lect.
Notes Pure Appl. Math., vol. 235, Dekker, NY, 2003, 101 -- 139.
[2] Yu. A. Bahturin, I. P. Shestakov, M. V. Zaicev, Gradings on simple Jordan
and Lie algebras, J. Algebra 283(2), (2005), 849 -- 868.
[3] K. I. Beidar, M. Bresar, M. A. Chebotar, Jordan isomorphisms of triangular
matrix algebras over a connected commutative ring, Linear Algebra Appl.
312(1) (2000), 197 -- 201.
[4] D. Blessenohl, H. Laue, Algebraic combinatorics related to the free Lie al-
gebra. S´eminaire Lotharingien de Combinatoire (Thurnau, 1992), Pr´epubl.
Inst. Rech. Math. Av 1993 (1992): 33.
[5] O. M. Di Vincenzo, P. Koshlukov, A. Valenti, Gradings on the algebra of
upper triangular matrices and their graded identities, J. Algebra 275(2),
(2004), 550 -- 566.
[6] O. M. Di Vincenzo, V. R. T. Silva, On Z2-graded polynomial identities of
the Grassmann algebra, Linear Algebra Appl. 431, (2009), 56 -- 72.
[7] A. Elduque, M. Kochetov, Gradings on simple Lie algebras, Math. Surveys
Monographs, vol. 189, Amer. Math. Soc., AARMC, Providence, RI, and
Halifax, Nova Scotia, 2013.
[8] A. Kemer, Ideals of Identities of Associative Algebras, in: Transl. Math.
Monogr., Vol. 87, Amer. Math. Soc., Providence, RI, 1991.
[9] P. Koshlukov, F. Martino, Polynomial identities for the Jordan algebra
of upper triangular matrices of order 2, J. Pure Appl. Algebra 216(11),
(2012), 2524 -- 2532.
[10] A. Valenti, M. V. Zaicev, Abelian gradings on upper triangular matrices,
Archiv Math. 80(1) (2003), 12 -- 17.
11
[11] A. Valenti, M. V. Zaicev, Group gradings on upper triangular matrices,
Archiv Math. 89(1) (2007), 33 -- 40.
[12] C. T. C. Wall, Graded Brauer groups, J. Reine Angew. Math. 213
(1963/1964), 187 -- 199.
[13] F. Yasumura, E. A. Hitomi, On the combinatorics of commutators of Lie
algebras, arXiv:1609.09407.
12
|
1108.3548 | 1 | 1108 | 2011-08-17T18:46:13 | Periodic derivations and prederivations of Lie algebras | [
"math.RA"
] | We consider finite-dimensional complex Lie algebras admitting a periodic derivation, i.e., a nonsingular derivation which has finite multiplicative order. We show that such Lie algebras are at most two-step nilpotent and give several characterizations, such as the existence of gradings by sixth roots of unity, or the existence of a nonsingular derivation whose inverse is again a derivation. We also obtain results on the existence of periodic prederivations. In this context we study a generalization of Engel-4-Lie algebras. | math.RA | math |
PERIODIC DERIVATIONS AND PREDERIVATIONS OF LIE
ALGEBRAS
DIETRICH BURDE AND WOLFGANG ALEXANDER MOENS
Abstract. We consider finite-dimensional complex Lie algebras admitting a peri-
odic derivation, i.e., a nonsingular derivation which has finite multiplicative order.
We show that such Lie algebras are at most two-step nilpotent and give several
characterizations, such as the existence of gradings by sixth roots of unity, or the
existence of a nonsingular derivation whose inverse is again a derivation. We also
obtain results on the existence of periodic prederivations. In this context we study
a generalization of Engel-4-Lie algebras.
1. Introduction
Let g be a Lie algebra over a field k. A derivation of g is called nonsingular,
if it is injective as a linear map. Lie algebras admitting nonsingular derivations
have been studied in many different contexts. First, they play an important role
in the existence question of left-invariant affine structures on Lie groups. Here
nonsingular derivations arise as a special case of invertible 1-cocylces for the Lie
algebra cohomology with coefficients in g-modules M with dim(M) = dim(g). If
D is a nonsingular derivation, then the formula x · y = D−1([x, D(y)]) defines a
left-symmetric structure on g. For a survey on this topic see [4] and the references
therein. An important structure result for Lie algebras g of characteristic zero with
a nonsingular derivation has been given by Jacobson [6].
It says that such Lie
algebras must be nilpotent. For Lie algebras in prime characteristic the situation is
more complicated. In that case there exist non-nilpotent Lie algebras, even simple
ones, which admit nonsingular derivations [2]. This is very interesting for the coclass
theory of pro-p groups and Lie algebras, see [10] and the references given therein. In
this context also the orders of nonsingular derivations have been studied. This leads
naturally to a subclass of nonsingular derivations, given by periodic derivations,
where the derivation has finite multiplicative order. Again it is interesting to obtain
structure results on Lie algebras admitting periodic derivations. Whereas this has
been studied intensively in prime characteristic, there seems to be only one result
for the characteristic zero case, which is proved [7]. It says that such Lie algebras
are abelian, if the order of the periodic derivation is not a multiple of six. There is
nothing said in [7] on Lie algebras admitting a periodic derivation of order which
Date: October 1, 2018.
2000 Mathematics Subject Classification. Primary 17B10, 17B25.
The authors were supported by the FWF, Projekt P21683.
1
2
D. BURDE AND W. MOENS
is a multiple of six. The aim of this paper is to close this gap by characterizing
such Lie algebras. We first prove that such Lie algebras are abelian or two-step
nilpotent. Then we show that the existence of a periodic derivation is equivalent to
the existence of a so called hexagonal grading by sixth roots of unity. This again
is equivalent to the existence of a nonsingular derivation whose inverse is again a
derivation. As it turns out, not all two-step nilpotent Lie algebras admit a periodic
derivation. This leads us to consider further invariants of two-step nilpotent Lie
algebras admitting periodic derivations.
In the last section, we study the much more difficult case of periodic prederivations.
A Lie algebra admitting a periodic prederivation is again nilpotent. However, it is
not clear how to find a good bound on the nilpotency class. In many cases we can
prove that such Lie algebras are at most 4-step nilpotent. This involves Engel-4-Lie
algebras and so called pre-Engel-4 Lie algebras. On the other hand, however, we
also construct a 5-step nilpotent Lie algebra admitting a periodic prederivation.
2. Periodic derivations
Let g be a Lie algebra. We always assume that g is finite-dimensional and complex,
if not mentioned otherwise. Denote by Der(g) the Lie algebra of derivations of g.
Periodic derivations have been defined in [7].
Definition 2.1. Let g be a Lie algebra. A derivation D of g is called periodic, if
there is an integer m ≥ 1 such that Dm = id.
If g has a periodic derivation D with m = 1, then we have [x, y] = D([x, y]) =
[D(x), y] + [x, D(y)] = 2[x, y], so that g is abelian. Conversely, an abelian Lie
algebra has periodic derivations of any possible order m ≥ 1. Indeed, just define
D = ζm id ∈ End(g) = Der(g), where ζm is a primitive m-th root of unity.
Our aim is to characterize complex Lie algebras admitting a periodic derivation. We
need an elementary lemma related to roots of unity.
Lemma 2.2. Let α, β be complex numbers with α = β = α + β = 1. Then
β = ωα with a primitive third root of unity ω.
Proof. The points 0, α and α + β are the vertices of an equilateral triangle. Hence
β = ωα. To see this algebraically, let γ = −(α + β). Then α + β + γ = 0 and
α = β = γ = 1. We may write β = eiϕα and γ = eiψα. Substituting in
α + β + γ = 0 one obtains 1 + eiϕ + eiψ = 0. Equating real and imaginary parts
yields ψ = −ϕ and cos(ϕ) = −1/2. We may take ϕ to be positive, so that ϕ = 2π/3
and ψ = −2π/3. Then ω = eiϕ = e2πi/3 and β = ωα.
(cid:3)
Corollary 2.3. Let α, β, γ ∈ C. Then at least one of the numbers α, β, γ, α + γ, β +
γ, α + β + γ is not an m-th root of unity.
Proof. Assume that all of these numbers are m-th roots of unity. Then by lemma
2.2 we have α = ωr(β + γ) = ωt and β = ωsγ for some r, s, t = ±1. This implies
PERIODIC DERIVATIONS
3
ωr(1 + ωs) = ωt. Raising this to the third power yields −1 = (1 + ωs)3 = 1, a
contradiction.
(cid:3)
Corollary 2.4. Let α1, . . . , α4 be m-th roots of unity. Then at least one of the
numbers αi + αj for i 6= j is not an m-th root of unity.
Proof. Assume that all of these numbers are m-th roots of unity. Then by lemma 2.2
we have α2 = ωrα1, α3 = ωsα1 and α4 = ωtα1 for some r, s, t ∈ {1, −1}. Now two
of these exponents must coincide, say r = s. This yields 1 = ωr + ωs = 2ωr = 2,
a contradiction.
(cid:3)
The only result on periodic derivations in characteristic zero we could find is the
following proposition of [7].
Proposition 2.5. Let g be a nonabelian Lie algebra of characteristic zero, which
admits a periodic derivation of order m. Then m is a multiple of six.
Proof. There is a proof given in [7] using a binomial formula and determinants,
which works for arbitrary characteristic. For the complex numbers there is a shorter
argument available. Let D be a periodic derivation of order m. Since any complex
endomorphism of finite order is semisimple, D is semisimple. Since g is nonabelian
there exist eigenvectors u and v with eigenvalues α and β such that [u, v] is a nonzero
eigenvector with eigenvalue α + β.
Indeed, D([u, v]) = [D(u), v] + [u, D(v)] =
(α + β)[u, v]. This means αm = βm = (α + β)m = 1, so that β = αω with a primitive
third root of unity ω, by lemma 2.2. Then we have α + β = α(1 + ω). Raising this
to the m-th power we obtain (1 + ω)m = 1, with 1 + ω being a primitive sixth root
of unity, so that 6 m.
(cid:3)
The result can be reformulated as a structure result for g. A Lie algebra admitting
a periodic derivation of order m coprime to six is abelian. Jacobson has proved [6],
that a Lie algebra of characteristic zero admitting a nonsingular derivation must be
nilpotent. Since a periodic derivation is nonsingular as a linear transformation, we
obtain the following structure result.
Proposition 2.6. Let g be a Lie algebra admitting a periodic derivation. Then g is
nilpotent.
In characteristic p > 0 this need not be true. There are non-nilpotent Lie algebras,
even simple Lie algebras, which have periodic derivations, see [10]. For complex Lie
algebras we obtain a stronger result than the above proposition.
Proposition 2.7. Let g be a Lie algebra admitting a periodic derivation. Then g is
at most two-step nilpotent.
Proof. Let D be a periodic derivation of g. Hence it is semisimple. Assume that
[g, [g, g]] 6= 0. Then there exist eigenvectors x, y, z of D with eigenvalues α, β, γ such
that [x, [y, z]] 6= 0. Hence [y, z] 6= 0. By the Jacobi identity we may assume that
also [y, [x, z]] 6= 0, hence [x, z] 6= 0. We conclude that α, β, γ, α + γ, β + γ, α + β + γ
are all m-th roots of unity. This contradicts corollary 2.3.
(cid:3)
4
D. BURDE AND W. MOENS
Proposition 2.8. Let g be a Lie algebra admitting a periodic derivation D. Then
the inverse D−1 is again a derivation of g.
Proof. We know that D is semisimple. Let e1, . . . , en be a basis of eigenvectors
with eigenvalues α1, . . . , αn. As in the proof of proposition 2.5, we see that for two
noncommuting eigenvectors ei and ej with eigenvalues αi and αj we have αi = αjω.
Then D−1([ei, ej]) = [D−1(ei), ej] + [ei, D−1(ej)] follows from
i + α−1
α−1
j = α−1
j (1 + ω−1) = α−1
j (1 + ω)−1 = (αi + αj)−1.
If a decomposable Lie algebra g admits a periodic derivation of order a multiple of
six, say 12, then we may not obtain a periodic derivation of order six just by rescaling.
Indeed, for g = C2 the linear map D = diag(ζ, ζ 2), with ζ being a primitive 12-th
root of unity, is a periodic derivation of order 12. However, no multiple λD has
order six: 1 = (λζ 2)6 = (λζ)6 yields 1 = ζ 6 = −1, which is a contradiction.
Definition 2.9. Denote by N(c, g) the free-nilpotent Lie algebra of nilpotency class
c and g generators.
(cid:3)
Let c = 2. The Lie algebra N(2, 1) is abelian, and N(2, 2) is the Heisenberg Lie
algebra, with [x1, x2] = x3. It admits periodic derivations, e.g., D = diag(1, ω, 1 + ω)
or
D =
.
0 −1 0
1
0
1
0
1
0
The latter derivation is integral, based on an element of order six in SL2(Z). This
example generalizes to all Heisenberg Lie algebras hm of dimension 2m + 1, with Lie
brackets [xi, yi] = z for 1 ≤ i ≤ m. Indeed, D = diag(1, . . . , 1, ω, . . . , ω, 1 + ω) is a
periodic derivation of hm.
The Lie algebra N(2, 3), with basis x1, . . . , x6 and brackets
[x1, x2] = x4, [x1, x3] = x5, [x2, x3] = x6
is our prototype of a Lie algebra admitting a periodic derivation.
Example 2.10. The Lie algebra N(2, 3) has a periodic derivation of order six, given
by
D = diag(1, ω, ω2, 1 + ω, 1 + ω2, ω + ω2).
It is easy to see that D−1 = diag(1, ω2, ω, 1 + ω2, 1 + ω, ω + ω2). This is again a
derivation.
In low dimensions every two-step nilpotent Lie algebra admits a periodic deriva-
tion. This can be seen by explicit calculations, using a list of complex two-step
nilpotent Lie algebras. It also follows more easily from the results of section 4.
Proposition 2.11. Let g be a two-step nilpotent Lie algebra of dimension n ≤ 6.
Then g admits a periodic derivation.
PERIODIC DERIVATIONS
5
Corollary 2.12. Let g be a two-step nilpotent Lie algebra with g ≤ 3 generators,
i.e. with first Betti number b1 ≤ 3. Then g admits a periodic derivation.
Proof. Such a Lie algebra is a quotient of N(2, 3). In the latter case it would be at
most 6-dimensional.
(cid:3)
In general, not every two-step nilpotent Lie algebra admits a periodic derivation.
Already in dimension 7 there are counter examples. Consider the Lie algebra g =
N(2, 4)/I5 in theorem 7.15 of [5], with Lie brackets
[x1, x2] = x5, [x1, x3] = x6, [x2, x3] = x7, [x3, x4] = −x5.
Example 2.13. The two-step nilpotent Lie algebra g = N(2, 4)/I5 of dimension 7
has no periodic derivation.
Indeed, it is easy to see that a derivation of g has eigenvalues of the form
These cannot be all m-th roots of unity.
α, β, γ, β − α, α + γ, β − γ, α + β − γ.
In contrast to this, all two-step nilpotent Lie algebras admit a periodic automorphism
of any possible order m ≥ 1. See [6] for the discussion on periodic automorphisms.
More examples of higher dimension without periodic derivations are given by the
following result.
Proposition 2.14. The free-nilpotent Lie algebra N(2, g) admits a periodic deriva-
tion if and only if g ≤ 3.
Proof. The case g ≤ 3 has been treated above. Assume that g ≥ 4 and that
f = N(2, g) admits a periodic derivation D. Then D is semisimple and we can
find eigenvectors x1, . . . , xg which span a subspace complementary to [f, f]. The
commutator is spanned by the linearly independent eigenvectors [xi, xj] for 1 ≤ i <
j ≤ g. If the xi have eigenvalue λi, then the [xi, xj] have eigenvalue λi +λj, for i 6= j.
We obtain m-rooths of unity λ1, λ2, λ3, λ4 and λi + λj for i 6= j. This contradicts
corollary 2.4.
(cid:3)
3. Gradings
We introduce the following grading, motivated by the periodic derivation
D = diag(α, β, γ, α + β, α + γ, β + γ) = diag(1, ω, ω2, 1 + ω, 1 + ω2, ω + ω2)
of example 2.10.
Definition 3.1. A Lie algebra g is called hexagonally graded, if it admits a vector
space decomposition
g = gα ⊕ gβ ⊕ gγ ⊕ gα+β ⊕ gα+γ ⊕ gβ+γ
with all indices being distinct complex numbers, satisfying
(1) [gi, gj] ⊆ gi+j for all i, j ∈ {α, β, γ, α + β, α + γ, β + γ}
6
D. BURDE AND W. MOENS
(2) [gi, g] = 0 for all i ∈ {α + β, α + γ, β + γ}.
Remark 3.2. Note that a hexagonally graded Lie algebra g satisfies [g, [g, g]] = 0.
The term hexagonally graded can be illustrated as follows. Let ω = e2πi/3. With
α = 1, β = ω and γ = ω2 we have
(α, β, γ, α + β, α + γ, β + γ) = (1, ω, ω2, 1 + ω, 1 + ω2, ω + ω2),
which consists just of all the sixth roots of unity.
ω
1 + ω
PSfrag replacements
ω + ω2
1
ω2
1 + ω2
Remark 3.3. We can rewrite the definition in a shorter way. A Lie algebra g is
hexagonally graded, if it admits an additive grading by sixth roots of unity, g =
⊕ζ 6=1gζ, such that the subspace ⊕ζ 3=1gζ corresponding to the third roots of unity
is central. In this formulation, the third roots form a subgroup. Note that in the
above picture we should rotate the roots to achieve this.
Lemma 3.4. Let g be hexagonally graded. Then g admits a periodic dervation of
order six.
Proof. Let g = ⊕ζ 6=1gζ be a hexagonal grading. We may assume that g is non-
abelian. Define a linear map D : g → g by its restrictions to the subspaces gζ, i.e.,
by D(x) = ζx for x ∈ gζ. This is a periodic derivation of g of order six.
(cid:3)
Definition 3.5. A Lie algebra g is called triangularly graded, if it admits an additive
grading by the non-zero complex numbers, g = Lα∈C× gα with [gα, gβ] ⊆ gα+β, such
that [gα, gβ] 6= 0 implies β = ωα with a primitive third root of unity ω.
If β = ωα, then α, β, −(α+β) lie on a circle and form the vertices of an equilateral
triangle.
Lemma 3.6. Let g be a nonabelian triangularly graded Lie algebra. Then g is
two-step nilpotent.
Proof. The proof goes exactly like the one of proposition 2.7. Suppose [g, [g, g]] 6= 0.
Then there exist nonzero scalars α, β and γ such that [gα, [gβ, gγ]] 6= 0. This implies
[gβ, gγ] 6= 0. By the Jacobi identity we may assume that [gα, gγ] is nonzero. Now
the conditions on α, β, γ contradict again corollary 2.3.
(cid:3)
It is clear that a hexagonally graded Lie algebra is also triangularly graded. But
also the converse is true.
PERIODIC DERIVATIONS
7
Proposition 3.7. A Lie algebra g is hexagonally graded if and only if it is triangu-
larly graded.
Proof. Let g = ⊕gα be triangularly graded. Then [g, g] = ⊕αWα for linear subspaces
Wα of gα. Write gα = Vα ⊕ Wα with a complementary vector space Vα and define
V = ⊕αVα. Then g = V ⊕ [g, g] = Lα(Vα ⊕ Wα). Define an equivalence relation on
C× by
α ∼ β ⇔ α3 = β 3.
For α ∼ β we have [Vα, Vβ] ⊆ [gα, gβ], which maybe nonzero except for α = β where
[gα, gα] ⊆ g2α = 0. For α 6∼ β we have [Vα, Vβ] ⊆ [gα, gβ] = 0.
Now consider for each α ∈ C×, with β = ωα, γ = ω2α the linear subspaces
g(α) = Vα ⊕ Vβ ⊕ Vγ ⊕ [Vα, Vβ] ⊕ [Vα, Vγ] ⊕ [Vβ, Vγ]
of g. We have g(α) = g(α′) for α ∼ α′ and [g(α), g(α′)] = 0 for α 6∼ α′, because g is
two-step nilpotent and [Vα, Vα′] = 0 in this case. We want to show that g is a direct
sum of ideals g(α), i.e.,
g(α).
g = Mα/∼
to see that all g(α) are Lie subalgebras. Then consider the sum Pα/∼
This would immediately imply that g itself is hexagonally graded. First it is easy
g(α) of
commuting subalgebras.
It is a Lie subalgebra of g and contains V . Hence it
coincides with g, since V generates g as a Lie algebra. Then we have [g, g(α)] ⊆
[g(α), g(α)] ⊆ g(α), so that g(α) is a Lie ideal of g. It remains to show that the
sum is direct. Recall that g = Lα(Vα ⊕ Wα). Suppose that a subspace Vθ occurs
in the decomposition of two ideals g(α) and g(α′). Then θ = ωiα = ωjα′ for some
i, j ∈ {0, 1, 2}. Hence α3 = (ωiα)3 = θ3 = (ωjα′)3 = (α′)3, so that α ∼ α′ and
g(α) = g(α′). Thus each Vθ occurs at most once in the above sum. A similar
argument shows that each Wθ appears at most once in the sum. Hence the sum is
direct.
(cid:3)
Corollary 3.8. Let g be a Lie algebra admitting a periodic derivation. Then g is
hexagonally graded.
Proof. Let D be a periodic derivation of g. Then D is semisimple and we may
consider the eigenspace decomposition g = ⊕αgα of g with respect to D. If [gα, gβ] 6=
0, then α, β and α + β are eigenvalues of D, hence m-th roots of unity. By lemma
2.2 we have β = ωα. Hence g is triangularly graded and the claim follows from
proposition 3.7.
(cid:3)
Corollary 3.9. Let g be a Lie algebra admitting a nonsingular derivation whose
inverse is again a derivation. Then g is hexagonally graded.
Proof. If there is such a derivation D, then we may assume that D and D−1 are
semisimple, by replacing them with the corresponding semisimple parts which are
again nonsingular derivations. Consider again the eigenspace decomposition g =
8
D. BURDE AND W. MOENS
⊕αgα of g with respect to D. If [gα, gβ] 6= 0, then α−1, β−1 are eigenvalues of D−1,
and [gα, gβ] this is an eigenvector of D−1 with eigenvalue α−1 + β−1 = (α + β)−1.
The latter equation again implies that β = ωα. Hence g is triangularly graded and
the claim follows from proposition 3.7.
(cid:3)
Now we can summarize our characterizations as follows.
Theorem 3.10. For a complex Lie algebra the following statements are equivalent.
(1) g admits a periodic derivation.
(2) g is hexagonally graded.
(3) g admits a nonsingular derivation whose inverse is again a derivation.
Proof. First, (1) implies (2) by corollary 3.8. Then (2) implies (3) by lemma 3.4 and
proposition 2.8. Finally (3) implies (1) by corollary 3.9 and lemma 3.4.
(cid:3)
Another consequence is the following characterization.
Corollary 3.11. A Lie algebra admits a periodic derivation if and only if it admits
a periodic derivation of order six.
Proof. We may assume that the Lie algebra g is nonabelian. Suppose that g admits
a periodic derivation. Its order is a multiple of six. Then g is hexagonally graded,
so that lemma 3.4 yields a periodic derivation of order six.
(cid:3)
Corollary 3.12. If g admits a periodic derivation of order six, then it also admits
a periodic derivation of order 6k for all k ≥ 1.
Proof. If D is a periodic derivation of order six, then D = diag(α1, . . . , αn) with αi
being sixth roots of unity. By multiplying D with a suitable sixth root of unity,
we may assume that one of the αi equals 1, say αj = 1. If we then multiply by a
primitive 6k-th root of unity, for each k ≥ 1, then we obtain a diagonal derivation
of order exactly 6k, since the j-th entry has order exactly 6k in C×.
(cid:3)
4. Quotients of N(2, g) by homogeneous ideals
Every two-step nilpotent Lie algebra g is a quotient of f = N(2, g) by some ideal
I contained in [f, f]. We always assume that g is nonabelian, so that g ≥ 2. The
dimension r = dim(I) ≥ 0 is an invariant of g, the number of relations. We have
dim(f) = g +(cid:18)g
2(cid:19), dim(g) = g +(cid:18)g
2(cid:19) − r.
Denote by X a minimal generating subset of f, with g elements.
Definition 4.1. An ideal J of f = N(2, g) is said to partition f homogeneously, if
there exists a partition X = X ∪ Y ∪ Z of X into three subsets X, Y, Z such that
J = JX + JY + JZ + JX,Y + JX,Z + JY,Z,
where JX = h[x, x′]x, x′ ∈ Xi and JX,Y is a linear subspace of h[x, y]x ∈ X, y ∈ Y i,
and so on.
PERIODIC DERIVATIONS
9
As an example, consider f with g = 2m generators, and partition
X = X ∪ Y ∪ Z = {x1, . . . , xm} ∪ {y1, . . . , ym} ∪ ∅.
Define
JX = h[xi, xj] 1 ≤ i, j ≤ mi,
JY = h[yi, yj] 1 ≤ i, j ≤ mi,
JX,Y = h[xi, yi] − [xj, yj] 1 ≤ i, j ≤ mi,
and JZ = JX,Z = JY,Z = 0. Then J = JX + JY + JX,Y partitions f homogeneously,
and the quotient f/J is a two-step nilpotent Lie algebra with 2m generators and
1-dimensional commutator, which is obviously the Heisenberg Lie algebra hm of
dimension 2m + 1.
We obtain another characterization of Lie algebras admitting a periodic derivation
as follows.
Proposition 4.2. A Lie algebra is hexagonally graded if and only if it is a quotient
of N(2, g) by a homogeneously partitioning ideal.
Proof. If g = N(2, g)/J is such a quotient, then it is easy to check that
g = gα ⊕ gβ ⊕ gγ ⊕ gα+β ⊕ gα+γ ⊕ gβ+γ
= hXi ⊕ hY i ⊕ hZi ⊕ h[X, Y ]i ⊕ h[X, Z]i ⊕ h[Y, Z]i mod J
yields a well-defined hexagonal grading.
Conversely, assume that g is hexagonally graded. Let X = {x1, . . . , xr} be a basis
for g(α), Y = {y1, . . . , xs} a basis for g(β) and Z = {z1, . . . , zt} a basis for g(γ). We
may assume that
[g, g] = span{[xi, yj], [xi, zj], [yi, zj]},
so that X = X ∪ Y ∪ Z is a minimal generating set for g as a Lie algebra. Denote by
f the free-nilpotent Lie algebra of class 2 on the generators X1, . . . , Xr, Y1, . . . , Ys,
Z1, . . . , Zt. There exists a Lie algebra epimorhism π : f → g mapping Xi to xi, Yi to
yi and Zi to zi. Thus g is isomorphic to f/J with J = ker(π). It suffices to show
that J partitions f homogeneously. First observe that J ⊆ [f, f]. So every R ∈ J is
of the form
R = Rx + Ry + Rz + Rxy + Rxz + Ryz,
where Rx ∈ span([xi, xj]), Rxy ∈ span([xi, yj]), etc. For Xi and Xj in X, we obtain
π([Xi, Xj]) = [π(Xi), π(Xj)] = [xi, xj] = 0 so that [Xi, Xj] ∈ J for all Xi, Xj.
Similarly, [Yi, Yj], [Zi, Zj] ∈ J. So we already know that Rx, Ry, Rz ∈ J.
It now
suffices to show that Rxy, Rxz and Ryz are also contained in J. Note that
0 = π(R) = π(Rxy) + π(Rxz) + π(Ryz).
Since the terms in this sum belong to the linearly independent subspaces
span([xi, yj]) = g(α + β), span([xi, zj]) = g(α + γ), span([yi, zj]) = g(β + γ)
10
D. BURDE AND W. MOENS
respectively, they must vanish. We conclude that Rx, Ry, Rz, Rxy, Rxz and Ryz are
contained in J.
(cid:3)
Theorem 3.10 implies the following result.
Corollary 4.3. A Lie algebra admits a periodic derivation if and only if it is a
quotient of N(2, g) by a homogeneously partitioning ideal.
Proposition 4.4. Let g = f/I of dimension n with g generators and dim(I) = r.
Assume that g admits a periodic derivation. Then we have the following estimates:
g ≤ n ≤
g(g − 3)
6
≤ r ≤
g2
3
+ g,
g(g − 1)
2
.
Proof. By theorem 3.10, g is hexagonally graded, so that we may assume that I
partitions f homogeneously. Let X, Y and Z as in the definition 4.1. Then g =
X + Y + Z is a partition of g. Since I contains the span of [x, x′], [y, y′] and
[z, z′] for x, x′ ∈ X; y, y′ ∈ Y ; z, z′ ∈ Z, its dimension r is at least (cid:0)X2 (cid:1) + (cid:0)Y 2 (cid:1) +
(cid:0)Z2 (cid:1) ≥ g(g − 3)/6. On the other hand we always have r ≤ dim([f, f]) = g(g − 1)/2.
r = g +(cid:0)g
This shows the second estimate. The first one follows form this by substituting
(cid:3)
We can use this result to classify Lie algebras g = f/I with small r, admitting a
√24r+9+3
periodic derivation. The second estimate is equivalent to
.
For a given small r it restricts the possible values for g very much. If g is nonabelian,
then we have
2(cid:1) − n.
√8r+1+1
≤ g ≤
√8r+1+1
2
< g.
2
2
Proposition 4.5. Let g = N(2, g)/I be a two-step nilpotent Lie algebra with small
r = dim(I), namely 0 ≤ r ≤ 2. Then g admits a periodic derivation if and only if it
is isomorphic to a Lie algebra in the table below.
g
N(2, 2)
N(2, 3)
r g
0 2
0 3
1 3
1 4
2 3 N(2, 3)/h[x1, x2], [x1, x3]i
2 4 N(2, 4)/h[x1, x2], [x3, x4]i
N(2, 4)/h[x2, x4], [x3, x4]i
2 5 N(2, 5)/h[x1, x2], [x3, x4]i
N(2, 3)/h[x2, x3]i
N(2, 4)/h[x1, x2]i
dim(g)
3
6
5
9
4
8
8
13
Proof. All two-step nilpotent Lie algebras with 0 ≤ r ≤ 2 and 2 ≤ g ≤ 5 have
been classified in [5]. An easy calculation then shows which of them admit a peri-
odic derivation. The result also follows by considering the possible homogeneously
partitioning ideals and classifiying the resulting quotients.
(cid:3)
PERIODIC DERIVATIONS
11
Remark 4.6. For r = 3 the result becomes much more complicated. Then the
above estimate gives 4 ≤ g ≤ 6. The case r = 3, g = 4 is again classified in [5],
Theorem 7.15. There are six two-step nilpotent Lie algebras N(2, 4)/Ik, 1 ≤ k ≤ 6
of dimension 7. The only one not admitting a periodic derivation is N(2, 4)/I5, see
example 2.13. This yields another interesting fact. Admitting a periodic derivation
is not a degeneration invariant.
A Lie algebra g is said to degenerate to a Lie algebra h, if h lies in the orbit closure
of g under the action of the general linear group acting by base changes. In fact,
N(2, 4)/I6 degenerates to N(2, 4)/I5, but N(2, 4)/I6 admits a periodic derivation,
whereas N(2, 4)/I5 does not.
It is also possible to determine the Lie algebras with small commutator subalgebra,
admitting a periodic derivation.
Proposition 4.7. Let g be a two-step nilpotent Lie algebra with dim([g, g]) ≤ 2.
Then g admits a periodic derivation.
Proof. It is well-known that a nilpotent Lie algebra with 1-dimensional commutator
subalgebra is isomorphic to the direct sum of the Heisenberg Lie algebra hm and
some abelian Lie algebra Ck. Since both summands admit a periodic derivation, so
does g. For the case that dim([g, g]) = 2 we can refer to theorem 1 in [1]. It says
that g admits a hexagonal grading, and hence also a periodic derivation.
(cid:3)
Remark 4.8. As example 2.13 with N(2, 4)/I5 shows, a two-step nilpotent Lie algebra
g with 3-dimensional commutator subalgebra need not admit a periodic derivation.
5. Periodic prederivations
Lie algebra prederivations are a generalization of Lie algebra derivations. They
have been studied in connection with Lie algebra degenerations, Lie triple systems
and bi-invariant semi-Riemannian metrics on Lie groups, see [3] and the references
given therein.
Definition 5.1. A linear map P : g → g is called a prederivation of g if
P ([x, [y, z]]) = [P (x), [y, z]] + [x, [P (y), z]] + [x, [y, P (z)]]
for every x, y, z ∈ g. It is called periodic, if P m = id for some m ≥ 1.
If g has a periodic prederivation P with m = 1, then g is at most two-step
nilpotent. Conversely, a Lie algebra of nilpotency class at most two admits periodic
prederivations of any any possible order m ≥ 1. Jacobsons result of [6] generalizes to
prederivations, and even to k-Leibniz derivations, see [9]. This implies the following
result.
Proposition 5.2. Let g be a Lie algebra admitting a periodic prederivation. Then
g is nilpotent.
12
D. BURDE AND W. MOENS
Consider two first examples of nilpotency class 4, namely the filiform nilpotent Lie
algebras g1 and g2 of dimension 5. The Lie brackets of g1 are given by [x1, xi] = xi+1
for i = 2, 3, 4. For g2 there is the additional bracket [x2, x3] = x5.
Example 5.3. The Lie algebra g1 admits periodic prederivations of any possible
even order m ≥ 2, whereas g2 admits no periodic prederivations.
Indeed, P = diag(α, −α, −α, α, α) with αm = (−α)m = 1 is a periodic pred-
erivation of g1. On the other hand, a prederivation of g2 has the eigenvalues
α, β, 2β − α, 2α − β, α + 2β. These cannot be all m-th roots of unity.
A Lie algebra admits periodic prederivations of odd order only in a trivial way. To
show this, we need another lemma on roots of unity.
Lemma 5.4. There exists m-th roots of unity α, β, γ such that α + β + γ is again
an m-th root of unity if and only if m is even.
Proof. If m is even then we may take α = β = −γ = 1. Conversely, suppose that
αm = βm = γm = δm = 1 with δ = −(α + β + γ) and α + β + γ + δ = 0. We obtain
a rhombus, so that two sides are vectors of opposite direction, say β = −α. Then
1 = βm = (−α)m = (−1)m and hence m is even.
(cid:3)
Proposition 5.5. A Lie algebra g admits a periodic prederivation of odd order if
and only if g it is nilpotent of class at most two.
Proof. Assume that [g, [g, g]] 6= 0 and P m = id for some odd m ≥ 1. Then P is
semisimple, and there exist eigenvectors x, y, z with eigenvalues α, β, γ such that
[x, [y, z]] 6= 0. In particular, [x, [y, z]] is an eigenvector with eigenvalue α + β + γ.
Then α, β, γ and α + β + γ are m-th roots of unity. This contradicts lemma 5.4 since
m is odd. It follows that [g, [g, g]] = 0.
(cid:3)
Remark 5.6. Let g be a Lie algebra admitting a periodic prederivation P . Then
it is not difficult to show that the inverse P −1 is again a prederivation of g. On
the other hand, if g admits a nonsingular prederivation P , such that P −1 is again a
prederivation, we do not know whether or not this implies the existence of a periodic
prederivation. Possibly theorem 3.10 can be extended to prederivations.
We introduce the following class of Lie algebras.
Definition 5.7. A Lie algebra g is called an pre-Engel-m Lie algebra, if the subspace
Em(g) = span{x ∈ g ad(x)m = 0} has full dimension, i.e., if dim(Em(g)) = dim(g).
In other words, g is a pre-Engel-m Lie algebra, if it admits an ad-nilpotent basis
of degree m. A special class if given by Engel-m Lie algebras. These are Lie algebras
g, where ad(x)m = 0 holds for all x ∈ g. They have received a lot of attention in the
literature, in particular concerning the possible nilpotency class of such algebras.
For m ≥ 5 this is a difficult question. For m = 4 the answer is, that an Engel-4 Lie
algebra has nilpotency class at most 7, and there are such examples. For details see
[11] and the references given therein.
PERIODIC DERIVATIONS
13
Remark 5.8. Note that a simple Lie algebra may have an ad-nilpotent basis, e.g.,
g = sl2(C) with e1 = ( 0 1
ourself to nilpotent Lie algebras.
−1 −1(cid:1). However, here we restrict
0 0 ), e2 = ( 0 0
1 0 ) and e3 = (cid:0) 1
1
We are in particular interested in nilpotent pre-Engel-4 Lie algebras, because they
are important for the existence of periodic prederivations. Clearly, any Lie algebra
of nilpotency class c ≤ 4 is a pre-Engel-4 Lie algebra. Thus it is more interesting to
consider pre-Engel-4 Lie algebras of nilpotency class c ≥ 5. For the filiform nilpotent
case and the free-nilpotent case it is easy to obtain a classification.
Proposition 5.9. A filiform Lie algebra g of nilpotency class c ≥ 5 is a pre-Engel-4
Lie algebra if and only if g is 6-dimensional and 2-step solvable.
Vergne. In particular we have ad(e1)n−3 6= 0 and ad(e1)n−2 = 0. Let x = Pn
Proof. We have dim(g) ≥ 6 because of c ≥ 5. Let g be a filiform nilpotent Lie
algebra of dimension n ≥ 7 and consider an adpated basis e1, . . . , en, in the sense of
i=1 λiei
and assume that ad(x)4 = 0. Then it is easy to see that we always obtain λ1 = 0.
Here we need n ≥ 7. This implies that E4(g) ⊆ span{e2, . . . , en}, so that g cannot
be a pre-Engel-4 Lie algebra. Hence we are left with the case dim(g) = 6 and c = 5,
where we have 5 different algebras. Two of them are 3-step solvable and three of
them are 2-step solvable. The result follows by an easy computation.
(cid:3)
Proposition 5.10. The free-nilpotent Lie algebra N(c, g) is a pre-Engel-4 Lie al-
gebra if and only if c ≤ 4.
Proof. We may assume that g ≥ 2. Suppose that c ≥ 5 and denote the generators by
x1, . . . , xg. Let e1, . . . , en be any basis of f = N(c, g). It contains a subset which is a
basis for f/[f, f]. We may assume that this subset is given by {e1, . . . , eg}. Define a Lie
algebra automorphism α : f → f by α(xi) = ei for all 1 ≤ i ≤ g (see [5]). Therefore
the identity ad(ei)t(ej) = 0 for i < j is equivalent to the identity ad(xi)t(xj) = 0
for i < j, which holds if and only if t ≥ c. It follows that ad(ei)4(ej) 6= 0 for all
1 ≤ i < j ≤ g. Hence N(c, g) is not a pre-Engel-4 Lie algebra for c ≥ 5.
(cid:3)
Pre-Engel-4 Lie algebras are related to the existence of periodic prederivations as
follows.
Proposition 5.11. Let g be a nilpotent Lie algebra which is not a pre-Engel-4 Lie
algebra. Then g does not admit a periodic derivation.
Proof. Assume that g admits a periodic prederivation P . Then P is semisimple, and
g admits a basis of eigenvectors x1, . . . , xn with corresponding eigenvalues α1, . . . , αn
of absolute value one. By assumption every basis x1, . . . , xn contains an element xi
with ad(xi)4 6= 0. Hence y = [xi, [xi, [xi, [xi, xj]]]] is nonzero for two eigenvectors xi
and xj, and y is again an eigenvector with P (y) = (4αi + αj)y. Since P is periodic,
4αi + αj = 1. This is a contradiction to αi = αj = 1.
(cid:3)
Corollary 5.12. Let g be a filiform Lie algebra of nilpotency class c ≥ 5. Then g
does not admit a periodic prederivation.
14
D. BURDE AND W. MOENS
Proof. We have dim(g) ≥ 6 because of c ≥ 5. All such filiform Lie algebra, except
for three algebras of dimension 6 are not pre-Engel-4 Lie algebras. Hence they do
not admit a periodic prederivation by proposition 5.11. For the remaining three
algebras the claim can be verified directly.
(cid:3)
Corollary 5.13. The free-nilpotent Lie algebra N(c, g) with c ≥ 5 does not admit
a periodic prederivation.
At this point there is the interesting question whether the above result can be
generalized to all nilpotent Lie algebras of class c ≥ 5. In low dimensions the answer
is positive. Indeed, such algebras only exist in dimensions n ≥ 6, and for n = 6 they
are filiform. The first interesting dimension then is seven.
Proposition 5.14. Let g be a nilpotent Lie algebra of dimension 7 and nilpotency
class c ≥ 5. Then g does not admit a periodic prederivation.
Proof. If c = 6, then g is filiform and the claim follows from corollary 5.12. For c = 5
there is a list of indecomposible algebras, consisting of 30 single algebras and one
family g(λ) of algebras (see for example [8]). The claim is clear for those algebras
which are not pre-Engel-4 Lie algebras. There remain 14 pre-Engel-4 Lie algebras to
be checked, where we have computed all prederivations. The result is that none of
these algebras admits a periodic prederivation. For decomposible algebras we have
g = f ⊕ C, where f is filiform of dimension 6. Again a direct computation gives the
result.
(cid:3)
In general, however, the result of corollary 5.12 cannot be generalized. We present
a counter example in dimension 8: take the free-nilpotent Lie algebra N(2, 5) with
2 generators x1, x2 and nilpotency class c = 5, and consider the quotient g given by
the following Lie brackets,
x3 = [x1, x2],
x4 = [x1, [x1, x2]] = [x1, x3],
x5 = [x2, [x1, x2]] = [x2, x3],
x6 = [x2, [x1, [x1, x2]]] = [x2, x4],
x6 = [x1, [x2, [x1, x2]]] = [x1, x5],
x7 = [x2, [x2, [x1, x2]]] = [x2, x5],
x8 = [x1, [x2, [x2, [x1, x2]]]] = [x1, x7],
x8 = [x2, [x1, [x2, [x1, x2]]]] = [x2, x6].
Note that this Lie algebra is a pre-Engel-4 Lie algebra, with ad(xi)4 = 0 for
1 ≤ i ≤ 8. However, it is not an Engel-4 Lie algebra because of ad(x1 + x2)4 6= 0.
PERIODIC DERIVATIONS
15
Example 5.15. The above 5-step nilpotent Lie algebra admits periodic prederiva-
tions of every even order m ≥ 2.
Indeed, a direct computation shows that for all α, β, γ ∈ C,
P = diag(α, β, γ, 2α + β, α + 2β, α + β + γ, 2β + γ, 2α + 3β)
is a prederivation of g. For β = −α, γ = α we obtain
P = diag(α, −α, α, α, −α, α, −α, −α).
With α being a primitive m-th root of unity satisfying (−α)m = 1, this prederivation
is periodic of order m.
There is another class of Lie algebras not admitting periodic prederivations.
Definition 5.16. Let g be a Lie algebra. We say that g satisfies property F , if every
basis contains a triple (x1, x2, x3) of basis elements such that for all i 6= j in {1, 2, 3}
holds: [xi, [xi, xj]] 6= 0 or [xj, [xj, xi]] 6= 0.
For a given Lie algebra, it seems difficult to check this for every basis. Therefore
it is more convenient to rewrite the definition as follows. A Lie algebra g does not
satisfy property F if it has a basis x1, . . . , xn such that for all triples (xi, xj, xk) there
exists a subset {xℓ, xm} ⊆ {xi, xj, xk} such that
[xℓ, [xℓ, xm]] = [xm, [xm, xℓ]] = 0.
Example 5.17. The Lie algebra of example 5.15 does not have property F .
To see this, we need to find a basis y1, . . . , y8 such that for each of the pos-
sible (cid:0)8
3(cid:1) = 56 triples (yi, yj, yk) there is a pair (yℓ, ym) such that ad(yℓ)2(ym) =
ad(ym)2(yℓ) = 0. It turns out that we may choose the original basis x1, . . . , x8 of
g together with the pairs (xℓ, xm) for (ℓ, m) = (1, 2), (1, 3), (2, 3), (1, 4), (2, 4), (3, 4),
(5, 6), (5, 7), (5, 8), (6, 7), (6, 8), (7, 8).
Proposition 5.18. The free-nilpotent Lie algebra N(c, g) has property F if and only
if either c ≥ 3, g ≥ 3 or c ≥ 4, g = 2.
Proof. Suppose first that c ≥ 3 and g ≥ 3. Denote the generators of f = N(c, g) by
x1, . . . , xg. Let e1, . . . , en be any basis of f. It contains a subset, say {e1, . . . , eg},
which is a basis for f/[f, f]. We obtain a Lie algebra automorphism α : f → f by
setting α(xi) = ei for all 1 ≤ i ≤ g. We may choose three different vectors from
{e1, . . . , eg}, say e1, e2, e3. Since [xi, [xi, xj]] 6= 0 for all i 6= j in {1, 2, 3}, we obtain
[ei, [ei, ej]] 6= 0 for all i 6= j in {1, 2, 3}. This shows that f has property F .
Now suppose that c ≥ 4 and g = 2. Denote the generators by x1, x2 and let
e1, . . . , enx be again any basis of f, such that e1 and e2 are a basis of f/[f, f]. We may
choose e3 such that e1, e2, e3 is a basis of f/[f, [f, f]]. As before, define an automor-
phism by α(ei) = xi for i = 1, 2. Then we may write
α(e3) = λ1x1 + λ2x2 + λ3[x1, x2] + v
16
D. BURDE AND W. MOENS
for some λi ∈ C with λ3 6= 0 and v ∈ [f, [f, f]]. A short computation shows that the
three terms [e1, [e1, e2]], [e1, [e1, e3]], [e2, [e2, e3]] are nonzero, by applying α and using
[xi, [xi, xj]] 6= 0 for all i 6= j. Hence f has property F .
(cid:3)
The relation to periodic prederivations is as follows.
Proposition 5.19. Let g be a Lie algebra having property F . Then g does not admit
a periodic prederivation.
Proof. Suppose that g admits a periodic prederivation P . Then P is semisimple,
and there is a basis of eigenvectors e1, . . . , en with P (ei) = αiei and αi = 1 for all
i = 1, . . . , n. Since g has property F , there exists a triple (e1, e2, e3) such that for all
i 6= j in {1, 2, 3} either [ei, [ei, ej]] or [ej, [ej, ei]] (or both) are nonzero eigenvalues,
i.e., either 2αi + αj = 1 or 2αj + αi = 1. In other words, for all i 6= j we have
αi = −αj. It follows α1 = α2 = α3 = 0, which is a contradiction.
(cid:3)
Corollary 5.20. The free-nilpotent Lie algebra N(c, g) admits a periodic prederiva-
tion if and only if c ≤ 2, or if c = 3, g = 2.
Proof. If not c ≤ 2 or c = 3, g = 2, then N(c, g) does not admit a periodic
prederivation by propositions 5.18 and 5.19. Conversely, if c ≤ 2, then any en-
domorphism is a prederivation. Hence there exists a periodic prederivation.
If
c = 3, g = 2, then we have the Lie algebra N(3, 2) with basis x1, . . . , x5 and brack-
ets [x1, x2] = x3, [x1, x3] = x4, [x2, x3] = x5. A periodic prederivation is given by
P = diag(α, −α, 1, α, −α).
(cid:3)
References
[1] C. Bartolone, A. Di Bartolo, G. Falcone: Nilpotent Lie algebras with 2-dimensional commu-
tator ideals. Lin. Alg. Appl. 434 (2011), 650 -- 656.
[2] G. Benkart, A. I. Kostrikin, M. I. Kuznetsov: Finite-dimensional simple Lie algebras with a
nonsingular derivation. Journal of Algebra 171 (1995), 894 -- 916.
[3] D. Burde: Lie Algebra Prederivations and strongly nilpotent Lie Algebras. Comm. in Algebra
30 (2002), no. 7, 3157 -- 3175.
[4] D. Burde: Left-symmetric algebras, or pre-Lie algebras in geometry and physics. Central Eu-
ropean Journal of Mathematics 4 (2006), no. 3, 323 -- 357.
[5] M. A. Gauger: On the classification of metabelian Lie algebras. Trans. Amer. Math. Soc. 179
(1973), 293-329.
[6] N. Jacobson: A note on automorphisms and derivations of Lie algebras. Proc. Amer. Math.
Soc. 6 (1955), 281 -- 283.
[7] A. I. Kostrikin, M. I. Kuznetsov: Two remarks on Lie algebras with a nonsingular derivation.
Proceedings of the Steklov Institute of Mathematics 208 (1995), 166-171.
[8] L. Magnin: Adjoint and Trivial Cohomology Tables for Indecomposable Nilpotent Lie Algebras
of Dimension ≤ 7 over C. Online e-book (2007), 1 -- 810.
[9] W. A. Moens: A characterization of nilpotent Lie algebras by invertible Leibniz-derivations.
Preprint (2011).
[10] A. Shalev: The orders of nonsingular derivations. J. Austral. Math. Soc. 67 (1999), 254 -- 260.
[11] G. Traustason: Engel Lie-algebras. Quart. J. Math. Oxford Ser. (2) 44 (1993), no. 175, 355-
384.
PERIODIC DERIVATIONS
17
Fakultat fur Mathematik, Universitat Wien, Nordbergstrasse 15, 1090 Wien,
Austria
E-mail address: [email protected]
Fakultat fur Mathematik, Universitat Wien, Nordbergstrasse 15, 1090 Wien,
Austria
E-mail address: [email protected]
|
1911.10526 | 1 | 1911 | 2019-11-24T13:43:21 | Inclusion among commutators of elementary subgroups | [
"math.RA",
"math.GR"
] | In the present paper we continue the study of the elementary commutator subgroups $[E(n,A),E(n,B)]$, where $A$ and $B$ are two-sided ideals of an associative ring $R$, $n\ge 3$. First, we refine and expand a number of the auxiliary results, both classical ones, due to Bass, Stein, Mason, Stothers, Tits, Vaserstein, van der Kallen, Stepanov, as also some of the intermediate results in our joint works with Hazrat, and our own recent papers [40,41]. The gimmick of the present paper is an explicit triple congruence for elementary commutators $[t_{ij}(ab),t_{ji}(c)]$, where $a,b,c$ belong to three ideals $A,B,C$ of $R$. In particular, it provides a sharper counterpart of the three subgroups lemma at the level of ideals. We derive some further striking corollaries thereof, such as a complete description of generic lattice of commutator subgroups $[E(n,I^r),E(n,I^s)]$, new inclusions among multiple elementary commutator subgroups, etc. | math.RA | math |
INCLUSIONS AMONG COMMUTATORS OF
ELEMENTARY SUBGROUPS
NIKOLAI VAVILOV AND ZUHONG ZHANG
Abstract. In the present paper we continue the study of the elementary com-
mutator subgroups [E(n, A), E(n, B)], where A and B are two-sided ideals of an
associative ring R, n ≥ 3. First, we refine and expand a number of the auxiliary
results, both classical ones, due to Bass, Stein, Mason, Stothers, Tits, Vaserstein,
van der Kallen, Stepanov, as also some of the intermediate results in our joint works
with Hazrat, and our own recent papers [40, 41]. The gimmick of the present paper
is an explicit triple congruence for elementary commutators [tij(ab), tji(c)], where
a, b, c belong to three ideals A, B, C of R. In particular, it provides a sharper coun-
terpart of the three subgroups lemma at the level of ideals. We derive some further
striking corollaries thereof, such as a complete description of generic lattice of com-
mutator subgroups [E(n, I r), E(n, I s)], new inclusions among multiple elementary
commutator subgroups, etc.
1. Introduction
Let GL(n, R) be the general linear group of degree n ≥ 3 over an associative ring
R with 1. For an ideal A E R we denote by GL(n, R, A) the principal congruence sub-
group of level A and by E(n, R, A) the corresponding relative elementary subgroup.
The study of commutator subgroups
[GL(n, R, A), GL(n, R, B)],
[GL(n, R, A), E(n, R, B)],
[E(n, R, A), E(n, R, B)]
It goes back to the
and other related birelative groups has a venerable history.
beginnings of algebraic K-theory in the works of Hyman Bass [4, 5, 6], and was then
continued, at the stable level, by Alec Mason, Wilson Stothers [24, 23], and many
others.
The next breakthrough, for rings satisfying commutativity conditions, came with
the works by Andrei Suslin, Leonid Vaserstein, Zenon Borewicz and the first author,
Anthony Bak, Alexei Stepanov, and many others [32, 33, 7, 2, 31]. However, these
papers mostly addressed only the case where one of the ideals A or B was the ring R
itself. These results depended on new powerful localisation methods introduced by
Daniel Quillen and Suslin in connection with Serre's problem, and their off-springs,
and also on remarkable geometric methods, see [3, 13] for a systematic description of
that stage.
Key words and phrases. general linear group, congruence subgroups, elementary subgroups, stan-
dard commutator formulae.
This publication is supported by Russian Science Foundation grant 17-11-01261.
1
2
NIKOLAI VAVILOV AND ZUHONG ZHANG
The next stage started with our joint papers with Roozbeh Hazrat and Alexei
Stepanov [36, 19, 37], where we addressed the general case, first for GL(n, R) over
commutative rings, then over quasi-finite rings. Our works [14, 15, 20, 17, 30] address
generalisations to other groups, such as Chevalley groups and Bak's unitary groups.
These results are systematically described in our surveys and conference papers [10,
11, 12, 18].
One of the pivotal results of our theory, initially established in a somewhat weaker
form by Hazrat and the second author in [20], and then stated in a more precise form
in our joint papers [16, 18], is a description of generators of [E(n, R, A), E(n, R, B)]
as a subgroup. In those papers it was proven over quasi-finite rings, and involved
three types of generators, the Stein -- Tits -- Vaserstein generators tji(c)tij(ab)tji(−c)
and tji(c)tij(ba)tji(−c), the elementary commutators [tij(a)tji(b)], where in both cases
i 6= j, a ∈ A, b ∈ B, c ∈ R, and also a third type of generators.
In 2018 we observed that the generators of the third type were redundant, see,
[34, 39]. Then in 2019 we noticed that even the elementary commutators should be
only taken for one root, and finally that this generation result holds over arbitrary
associative rings, see [40], Theorem 1 (which we reproduce below as Lemma 3). This
proof is then repeated in a slightly more transparent form in [41], § 5, as part of the
proof of the elementary multiple commutator formula
In turn, our proof of that result hinges on the following result, which is a stronger,
and more precise version of [40], Lemma 5. In this form, it is proven as [41], Lemma
11. Here we denote by A ◦ B = AB + BA the symmetrised product of two-sided
ideals A and B. For commutative rings, A ◦ B = AB = BA is the usual product of
ideals A and B.
Theorem A. Let R be an associative ring with 1, n ≥ 3, and let A, B be two-sided
ideals of R. Then for any 1 ≤ i 6= j ≤ n, any 1 ≤ k 6= l ≤ n, and all a ∈ A, b ∈ B,
c ∈ R, one has
yij(ac, b) ≡ ykl(a, cb) (mod E(n, R, A ◦ B)) .
During the talks "commutators of relative and unrelative elementary subgroups"
presenting our papers [40] -- [43], both at the algebraic group seminar at Chebyshev Lab
on October 22 (see http://chebyshev.spbu.ru/en/schedule/?week=1571605200), and
the at the Conference "Homological algebra, ring theory and Hochschild cohomology"
at EIMI on October 29 (see http://www.pdmi.ras.ru/EIMI/2019/CR/index.html)
the first author was invariably writing a ∈ A, b ∈ B, c ∈ C. During the second
talk, Pavel Kolesnikov, who was keeping notes, asked, what was that C? After a
little thinking, we decided that C should be C E R here, rather than R itself. Of
course, modulo a smaller subgroup E(n, R, ABC) the elementary commutator in the
left hand side will be congruent to a product of two other elementary commutators,
rather than to a single such commutator, as in the absolute case.
Theorem 1. Let R be an associative ring with 1, n ≥ 3, and let A, B, C be two-sided
ideals of R. Then for any three distinct indices i, j, h such that 1 ≤ i, j, h ≤ n, and
all a ∈ A, b ∈ B, c ∈ C, one has
yij(ab, c)yjh(ca, b)yhi(bc, a) ≡ 1 (mod E(n, R, ABC + BCA + CAB)) .
INCLUSIONS AMONG COMMUTATORS OF ELEMENTARY SUBGROUPS
3
Observe, that the ideal defining the congruence module is precisely
ABC + BCA + CAB = A ◦ BC + B ◦ CA + C ◦ AB.
This can be easily established in the same style as in our recent papers [40] -- [43].
Morally, it is essentially the same computation by Mennicke [25] in the form given to
it by Bass -- Milnor -- Serre in [6], Theorem 5.4. Subsequently, it was used in virtually
each and every paper on bounded generation. Alternatively, one could use the Hall --
Witt identity. However, everywhere before C = R so that the last factor becomes
trivial.
Theorem 1, and its immediate consequence Theorem 6, the three ideals lemma
[E(n, AB), E(n, C)] ≤ [E(n, BC), E(n, A)] · [E(n, CA), E(n, B)],
are the high points of the present paper, the rest are either preparations to their proof,
or corollaries of the above trirelative congruence. However, a number of intermediate
results generalise classical results and are decidedly interesting in themselves. Let us
list other principal results of the present paper.
• Theorem 2: generators of partially relativised elementary subgroup E(n, B, A),
where A, B E R, generalising a classical result by Stein -- Tits -- Vaserstein.
• Theorem 3: reduced set of generators of E(n, R, A), in terms of the unipotent
radicals of two opposite parabolic subgroups, generalising results by van der Kallen
and Stepanov.
• Theorem 4: the three ideals lemma for partially relativised subgroups E(n, A, BC).
• Theorem 5: a stable version of the standard commutator formula,
[GL(n − 1, R, A), E(n, R, B)] = [E(n, A), E(n, B)],
for arbitrary associative rings.
• Propositions 2 and 3: new inclusions for multiple elementary commutators.
• Theorem 7: a complete description of the generic lattice of inclusions among
[E(n, I r), E(n, I s)], for powers of one ideal I E R.
• Proposition 5: inclusion [E(n, A + B), E(n, A ∩ B)] ≤ [E(n, A), E(n, B)].
• Proposition 6: an explicit example, where E(n, R, A∩ B) is strictly smaller than
E(n, R, A) ∩ E(n, R, B).
The paper is organised as follows. In § 2 and § 3 we review some notation and briefly
recall the requisite facts on elementary subgroups in GL(n, R). The next six sections
are of a technical nature, they develop technical tools for the rest of the present paper.
Namely, in § 4 we consider partially relativised elementary subgroups E(n, B, A) and
prove Theorem 2, which gives their sets of generators. In § 5 we recall some basic facts
on intersections of parabolic subgroups with congruence subgroups, after which in § 6
we establish Theorem 3, which is a further strengthening of results by van der Kallen
and Stepanov, generation of E(n, R, A) in terms of unipotent radicals of two opposite
parabolic subgroups. In § 7 we prove a toy version of our main results, the three
ideals lemma for partially relativised elementary groups E(n, C, AB), Theorem 4. In
§ 8 we establish Theorem 5, which is a stable version of the standard commutator
4
NIKOLAI VAVILOV AND ZUHONG ZHANG
formula, valid for all associative rings. In § 9 we discuss the important special case,
behaviour of elementary commutators modulo relative elementary subgroups. The
core of the present paper is § 10, where we prove Theorem 1 and using that derive an
inclusion among birelative commutators, Theorem 6. This is kind of a three ideals
lemma, to be used in all subsequent results. The balance of this paper is dedicated
to applications. In § 11 we consider an application to the only outstanding case in
our multiple elementary commutator paper [41], quadruple commutators in GL(3, R),
and obtain some new inclusions among multiple elementary commutator subgroups.
In § 12 we obtain definitive results for the crucial case of the powers of one ideal
and prove Theorem 7. These results will be instrumental in the sequel of the present
paper dedicated to the case of Dedekind rings. In § 13 we compare the commutator
of two elementary subgroups of levels A and B with the commutator of elementary
subgroups of levels A ∩ B and A + B.
In § 14 we construct a counter-example
concerning intersections of relative elementary subgroups. Finally, in § 15 we make
some further related observations, and state some unsolved problems.
Initially, we planned to include in this paper also explicit computations over Dede-
kind rings. But then we realised that the topic is so extensive that it would be more
appropriate to publish those results separately.
2. Notation
2.1. Commutators. Let G be a group. A subgroup H ≤ G generated by a subset
X ⊆ G will be denoted by H = hXi. For two elements x, y ∈ G we denote by
xy = xyx−1 and yx = x−1yx the left and right conjugates of y by x, respectively.
Further, we denote by
[x, y] = xyx−1y−1 = xy · y−1 = x · yx−1
the left-normed commutator of x and y. Our multiple commutators are also left-
normed. Thus, by default, [x, y, z] denotes [[x, y], z], we will use different notation for
other arrangement of brackets. Throughout the present paper we repeatedly use the
customary commutator identities, such as their multiplicativity with respect to the
factors:
and a number of other similar identities, such as
[x, yz] = [x, y] · y[x, z],
[xy, z] = x[y, z] · [x, z],
[x, y]−1 = [y, x],
z[x, y] = [zx, zy],
[x−1, y] = [y, x]x,
[x, y−1] = [y, x]y,
usually without any specific reference.
Iterating multiplicativity we see that the
commutator [x1 . . . xm, y] is the product of conjugates of the commutators [xi, y],
i = 1, . . . , m. Obviously, a similar claim holds also for [x, y1 . . . ym].
Further, for two subgroups F, H ≤ G one denotes by [F, H] their mutual commu-
tator subgroup, spanned by all commutators [f, h], where f ∈ F , h ∈ H. Clearly,
[F, H] = [H, F ], and if F, H E G are normal in G, then [F, H] E G is also normal.
Similarly, for F, H, K ≤ G three subgroups of G their triple commutator [F, H, K]
is spanned by [f, h, k], where f ∈ F , h ∈ H and k ∈ K. We will use the following
version of the three subgroups lemma.
INCLUSIONS AMONG COMMUTATORS OF ELEMENTARY SUBGROUPS
5
Lemma 1. If F, H, K ≤ G be three subgroups of G. Assume that two of the subgroups
[F, H, K], [H, K, F ], [K, F, H] are normal in G. Then the third of them is also normal
and
[F, H, K] ≤ [H, K, F ] · [K, F, H].
Often times, elementary textbooks needlessly assume that the subgroups F, H, K
themselves are normal in G. This depends, of course, on the exact form of the Hall --
Witt identity one is using. In the correct form, the only conjugations occur outside
of the commutators, one such form is
[x, y−1, z−1]x · [z, x−1, y−1]z · [y, z−1, x−1]y = 1.
2.2. General linear group. Let R be an associative ring with 1, R∗ be the multi-
plicative group of the ring R. For two natural numbers m, n we denote by M(m, n, R)
the additive group of m × n-matrices with entries in R. By M(n, R) = M(n, n, R)
we denote the full matrix ring of degree n over R.
Let G = GL(n, R) = M(n, R)∗ be the general linear group of degree n over R. In
the sequel for a matrix g ∈ G we denote by gij its matrix entry in the position (i, j),
so that g = (gij), 1 ≤ i, j ≤ n. The inverse of g will be denoted by g−1 = (g′
ij),
1 ≤ i, j ≤ n.
As usual we denote by e the identity matrix of degree n and by eij a standard
matrix unit, i. e., the matrix that has 1 in the position (i, j) and zeros elsewhere. An
elementary transvection tij(ξ) is a matrix of the form tij(c) = e + ceij, 1 ≤ i 6= j ≤ n,
c ∈ R.
Further, let A be a two-sided of R. We consider the corresponding reduction
homomorphism
ρA : GL(n, R) −→ GL(n, R/A),
(gij) 7→ (gij + A).
Now, the principal congruence subgroup GL(n, R, A) of level A is the kernel ρA,
For a commutative ring R we denote by SL(n, R) the corresponding general linear
group. All other subgroups are interpreted similarly. Thus, for instance, the principal
congruence subgroup SL(n, R, A) is defined as SL(n, R, A) = GL(n, R, A)∩ SL(n, R).
3. Generation of relative elementary subgroups
The unrelative elementary subgroup E(n, A) of level A in GL(n, R) is generated by
all elementary matrices of level A. In other words,
E(n, A) = heij(a), 1 ≤ i 6= j ≤ n, a ∈ Ai.
In general E(n, A) has little chances to be normal in GL(n, R). The relative elemen-
tary subgroup E(n, R, A) of level A is defined as the normal closure of E(n, A) in the
absolute elementary subgroup E(n, R):
E(n, R, A) = heij(a), 1 ≤ i 6= j ≤ n, a ∈ AiE(n,R).
The following lemma on generation of relative elementary subgroups E(n, R, A) is
a classical result discovered in various contexts by Stein, Tits and Vaserstein, see, for
6
NIKOLAI VAVILOV AND ZUHONG ZHANG
instance, [33] (or [18], Lemma 3, for a complete elementary proof). It is stated in
terms of the Stein -- Tits -- Vaserstein generators):
zij(a, c) = tij(c)tji(a)tij(−c),
1 ≤ i 6= j ≤ n,
a ∈ A,
c ∈ R.
Lemma 2. Let R be an associative ring with 1, n ≥ 3, and let A be a two-sided ideal
of R. Then as a subgroup E(n, R, A) is generated by zij(a, c), for all 1 ≤ i 6= j ≤ n,
a ∈ A, c ∈ R.
The following result is a generalisation of Lemma 2 to mutual commutator sub-
groups [E(n, R, A), E(n, R, B)] of relative elementary subgroups. a further type of
generators occur, the elementary commutators:
yij(a, b) = [tij(a), tji(b)],
1 ≤ i 6= j ≤ n,
a ∈ A,
b ∈ B.
In slightly less precise forms, Theorem A was discovered by Roozbeh Hazrat and
the second author, see [20], Lemma 12 and then in our joint paper with Hazrat [18],
Theorem 3A. The strong form reproduced above was only established in our paper
[40], Theorem 1, as an aftermath of our papers [34, 39].
Lemma 3. Let R be any associative ring with 1, let n ≥ 3, and let A, B be two-
sided ideals of R. Then the mixed commutator subgroup [E(n, R, A), E(n, R, B)] is
generated as a group by the elements of the form
• zij(ab, c) = tij(c)tji(ab)tij(−c) and zij(ba, c) = tij(c)tji(ba)tij(−c),
• yij(a, b) = [tij(a), tji(b)],
where 1 ≤ i 6= j ≤ n, a ∈ A, b ∈ B, c ∈ R. Moreover, for the second type of
generators, it suffices to fix one pair of indices (i, j).
Since all generators listed in Lemma 3 belong already to the commutator subgroup
of unrelative elementary subgroups, we get the following corollary, [40], Theorem 2.
Lemma 4. Let R be any associative ring with 1, let n ≥ 3, and let A, B be two-sided
ideals of R. Then one has
(cid:2)E(n, R, A), E(n, R, B)(cid:3) = (cid:2)E(n, R, A), E(n, B)(cid:3) = (cid:2)E(n, A), E(n, B)(cid:3).
Let us state also some subsidiary results we use in our proofs. The following level
computation is standard, see, for instance, [36, 37, 18], and references there.
Lemma 5. R be an associative ring with 1, n ≥ 3, and let A and B be two-sided
ideals of R. Then
E(n, R, A ◦ B) ≤ (cid:2)E(n, A), E(n, B)(cid:3) ≤ (cid:2)E(n, R, A), E(n, R, B)(cid:3) ≤ GL(n, R, A ◦ B).
However, when applying this lemma to multiple commutators, one should bear in
mind that the symmetrised product is not associative. Thus, when writing something
like A◦ B ◦ C, we have to specify the order in which products are formed. Of course,
for commutative rings this dependence on the original bracketing disappears.
For quasi-finite rings the following result is [37], Theorem 5 and [18], Theorem 2A,
but for arbitrary associative rings it was only established in [41], Theorem 2.
INCLUSIONS AMONG COMMUTATORS OF ELEMENTARY SUBGROUPS
7
Lemma 6. Let R be any associative ring with 1, let n ≥ 3, and let A and B be
two-sided ideals of R. If A and B are comaximal, A + B = R, then
[E(n, A), E(n, B)] = E(n, R, A ◦ B).
4. Partially relativised elementary subgroups
Actually, the recent work by Alexei Stepanov [28, 29, 30, 1] makes apparent that in
many contexts it is very useful to consider partially relativised subgroups. One such
context is relative localisation, as introduced in the papers by Roozbeh Hazrat and
the second author [19, 20], then expanded and developed in a series of our joint papers
with Hazrat, and reconsidered by Stepanov, see [14, 15, 17, 10, 11, 12, 18, 29, 30].
Namely, for two ideals A, B E R we denote by E(n, B, A) the smallest subgroup
containing E(n, A) and normalised by E(n, B):
E(n, B, A) = E(n, A)E(n,B).
In particular, when B = R we get the usual relative group E(n, R, A), as defined
above. Clearly, if B ≤ C, then E(n, B, A) ≤ E(n, C, A). It follows that
E(n, A) = E(n, 0, A) ≤ E(n, B, A) ≤ E(n, R, A)
On the other hand,
(cid:2)E(n, A), E(n, B)(cid:3) ≤ E(n, B, A) ∩ E(n, A, B).
Thus, Lemma 5 implies the following inclusion, which is a broad generalisation of [1],
Lemma 4.1, in the linear case.
Proposition 1. Let R be any associative ring with 1, let n ≥ 3, and let A, B be
two-sided ideals of R. Then one has
E(n, R, A ◦ B) ≤ E(n, B, A) ∩ E(n, A, B).
Now, we start working towards a partially relativised generalisation of Lemma 2.
Lemma 7. Let R be any associative ring with 1, let n ≥ 2, and let A, B be two-sided
ideals of R. Then one has
(cid:10)E(n, A), E(n, B)(cid:11) = E(n, A + B).
Proof. Clearly, the left hand side is contained in the right hand side. On the other
hand E(n, A + B) is generated by the elementary transvections tij(a + b), where
1 ≤ i 6= j ≤ n, a ∈ A, b ∈ B. But every tij(a+b) = tij(a)tij(b) ∈ E(n, A)E(n, B). (cid:3)
In particular, even E(n, A)E(n, B) = E(n, A + B), if the left hand side is a sub-
Indeed, the above lemma implies that in the
group. But this is easy to remedy.
definition of partially relativised subgroups one can assume that A ≤ B.
Corollary 1. Let R be any associative ring with 1, let n ≥ 2, and let A, B be two-
sided ideals of R. Then E(n, B, A) = E(n, A + B, A).
But since E(n, B, A) is normalised by both E(n, A) and E(n, B), it is normal in
E(n, A + B).
8
NIKOLAI VAVILOV AND ZUHONG ZHANG
Corollary 2. Let R be any associative ring with 1, let n ≥ 2, and let A, B be two-
sided ideals of R. Then
E(n, B, A)E(n, B) = E(n, A + B).
Passing to the normal closures in E(n, R) we get the familiar equality, see, in
particular, [36], Lemma 1.
Corollary 3. Let R be any associative ring with 1, let n ≥ 2, and let A, B be two-
sided ideals of R. Then
E(n, R, A)E(n, R, B) = E(n, R, A + B).
The following result is a generalisation of a classical result on generation of relative
elementary subgroups E(n, R, A), discovered in various contexts by Stein, Tits and
Vaserstein, see, for instance, [33]. It is stated in terms of the Stein -- Tits -- Vaserstein
generators):
zij(a, c) = tij(c)tji(a)tij(−c),
1 ≤ i 6= j ≤ n,
a ∈ A,
c ∈ R.
Essentially, its proof follows the proofs of Lemma 2 (as reproduced in [38], The-
orem 1 or [18], Lemma 3, for instance). But of course a posteriori we can take
advantage of the simplifications that result from Lemma 3.
Theorem 2. Let R be an associative ring with identity 1, n ≥ 3, and let A and B be
two-sided ideals of R. Then the partially relativised elementary subgroup E(n, B, A)
is generated the elements zij(a, b), for all 1 ≤ i 6= j ≤ n, a ∈ A, b ∈ B.
Proof. Clearly, E(n, B, A) is generated by yx with x ∈ E(n, A) and y ∈ E(n, B).
Since yx = [y, x] · x we get [y, x] ∈ [E(n, A), E(n, B)]. By definition, the factor x
is a product of elementary matrices tij(a) with i 6= j and a ∈ A. By Lemma 3 the
commutator subgroup [E(n, A), E(n, B)] is generated by E(n, R, A◦B) together with
the elementary commutators
yij(a, b) = [tij(a), tji(b)] = tij(a) · tji(b)tij(−a) = zij(a, 0)zij(a, b),
Thus,
where 1 ≤ i 6= j ≤ n, a ∈ A and b ∈ B.
it only remains to show that the generators zij(ab, c) = tji(c)tij(ab) and
zij(ba, c) = tji(c)tij(ba) of the relative elementary group E(n, R, A ◦ B) are products
of generators listed in the statement of the lemma. Here, as above, 1 ≤ i 6= j ≤ n,
a ∈ A, b ∈ B and c ∈ C.
We choose a h 6= i, j, then
tji(c)tij(ab) = tji(c)[tih(a), thj(b)] = [tji(c)tih(a), tji(c)thj(b)] =
(cid:2)[tji(c), tih(a)]tih(a), [tji(c), thj(b)]thj(b)(cid:3) = (cid:2)tjh(ca)tih(a), thi(−bc)thj(b)(cid:3) =
tjh(ca)[tih(a), thi(−bc)thj(b)] · [tjh(ca), thi(−bc)thj(b)].
To finish the proof, we consider the two above commutators separately
v = [tjh(ca), thi(−bc)thj(b)]
u = [tih(a), thi(−bc)thj(b)],
INCLUSIONS AMONG COMMUTATORS OF ELEMENTARY SUBGROUPS
9
The following computation shows that u is a products of generators listed in the
statement of the lemma.
u = [tih(a), thi(−bc)tkj(b)] = [tih(a), thi(−bc)] · thi(−bc)[tih(a), thj(b)] =
[tih(a), thi(−bc)] · thi(−bc)tij(ab) = [tih(a), thi(−bc)] · [thi(−bc), tij(ab)]tij(ab) =
[tih(a), thi(−bc)] · thj(−bcab)tij (ab) = tih(a) · thi(−bc)tih(−a)thj(−bcab)tij(ab),
A similar computation shows that the same holds also for v, which finishes the proof.
(cid:3)
Lemma 8. Let R be an associative ring with identity 1, and let A, B, C and D be
its two-sided ideals. Then we have the following commutator formula for partially
relativised elementary subgroups
[E(n, B, A), E(n, D, C)] = [E(n, A), E(n, C)].
Proof. Combining the inclusions among partially relativised subgroups with Lemma
4, we get
[E(n, A), E(n, C)] = [E(n, 0, A), E(n, 0, C]] ≤ [E(n, B, A), E(n, D, C)] ≤
[E(n, R, A), E(n, R, C)] = [E(n, A), E(n, C)].
(cid:3)
5. Parabolic subgroups
In this section and the next one we collect some results on generation of E(n, R, A)
by the elements in unipotent radicals, or their conjugates. We start with the absolute
case.
5.1. Standard parabolic subgroups. Denote by Rn the free right R-module con-
sisting of columns of height n with components from R. Similarly, nR denotes the free
left R-module, consisting of rows of length n with components from R. The module
nR is dual to Rn, with the pairing of nR and Rn defined by the multiplication of a row
by a column, nR × Rn −→ R, (v, u) 7→ vu ∈ R. The standard based of Rn and nR
will be denoted by e1, . . . , en and f1, . . . , fn, respectively. Recall that ei is the column
of height n, whose i-th component equals 1, while all other components are zeroes.
Similarly, fi is the row of length n, whose i-th component equals 1, while all other
components are zeroes. The base f1, . . . , fn is dual to e1, . . . , en, with respect to the
above pairing. The group G = GL(n, R) acts on Rn on the left, by multiplication of
a column u ∈ Rn by a matrix g ∈ G, (g, u) 7→ gu. By the same token, the group G
acts on nR on the right by multiplication: (v, g) 7→ vg for v ∈ nR, g ∈ G.
Denote by Pm the m-th standard maximal parabolic subgroup in G = GL(n, R).
From a geometric viewpoint the subgroup Pi, m = 1, . . . , n − 1, is precisely the
stabiliser of the submodule Vm in V , generated by e1, . . . , em. In matrix form Pm can
be thought of as the group of upper block triangular matrices
Pm = (cid:26)(cid:18)x y
0 z(cid:19) x ∈ GL(m, R), y ∈ M(m, n − m, R), z ∈ GL(n − m, R)(cid:27) .
10
NIKOLAI VAVILOV AND ZUHONG ZHANG
Simultaneously we consider the opposite maximal parabolic subgroup P −
m
P −
m = (cid:26)(cid:18)x 0
w z(cid:19) x ∈ GL(m, R), w ∈ M(n − m, m, R), z ∈ GL(n − m, R)(cid:27) .
These subgroups admit Levi decompositions Pm = Lm ⋌ Um and P −
m = Lm ⋌ U −
m
with common Levi subgroup
Lm = (cid:26)(cid:18)x y
0 z(cid:19) x ∈ GL(i, R), z ∈ GL(n − m, R)(cid:27) ,
and opposite unipotent radicals
Um = (cid:26)(cid:18)e y
0 e(cid:19) y ∈ M(m, n − m, R)(cid:27) , U −
m = (cid:26)(cid:18) e 0
w e(cid:19) w ∈ M(n − m, m, R)(cid:27) .
In particular, Lm, and thus all of its subgroups, normalise both Um and U −
m.
5.2. Generation by two opposite unipotent radicals. The following result as-
serts that E(n, R) is generated by the unipotent radicals of two standard parabolic
subgroups. It is obvious from the Chevalley commutator formula, and well known.
Actually, in more general settings this is the definition of elementary subgroups, see
the paper by Victor Petrov and Anastasia Stavrova [26] and references there1.
Lemma 9. Let R be an associative ring with 1, n ≥ 3, and 1 ≤ m ≤ n − 1. Then
E(n, R) = hUm, U −
mi.
What is important, is that both generators here are normalised by the Levi sub-
group Lm and all of its subgroups.
As usual, for m < n we consider the stability embedding
GL(m, R) −→ GL(n, R),
g 7→ (cid:18)g 0
0 e(cid:19) .
This embedding is compatible with elementary subgroups, congruence subgroups, rel-
ative elementary subgroups, etc. When we consider GL(m, R, A), etc., as a subgroup
of GL(n, R) we always mean its image under this embedding.
5.3. Unipotent radicals of level A. Now, let A E R be an ideal of R. We denote
by Um(A) and U −
n (A) the intersections of Um and U −
m with GL(n, R, A):
Um(A) = (cid:26)(cid:18)e y
m(A) = (cid:26)(cid:18) e 0
0 e(cid:19) y ∈ M(m, n − m, A)(cid:27) ,
w e(cid:19) w ∈ M(n − m, m, A)(cid:27) .
U −
The following lemma is a direct corollary of the Levi decomposition for Pm and its
opposite P −
m.
1Of course, with this advanced approach one has to prove that this definition is correct, in other
words that various parabolic subgroups lead to the same elementary subgroup. This is precisely
what is accomplished in [26].
INCLUSIONS AMONG COMMUTATORS OF ELEMENTARY SUBGROUPS
11
Lemma 10. Let R be an associative ring with 1, n ≥ 3, 1 ≤ m ≤ n− 1, and let A, C
be two-sided ideals of R. Then one has
[GL(m, R, A), Um(C)] ≤ Um(AC),
m(C)] ≤ U −
[GL(m, R, A), U −
m(CA).
0 e(cid:19) ,(cid:18)e y
Proof. Let g ∈ GL(m, R, A), y ∈ M(m, n − m, C), and w ∈ M(n − m, m, C). Then,
clearly,
(cid:20)(cid:18)g 0
w(g−1 − e) e(cid:19) ,
where y ≡ 0 (mod C) , w ≡ 0 (mod C) and g ≡ g−1 ≡ e (mod A) . Then all outer-
diagonal entries of the matrices in the right hand sides are congruent to 0 modulo
AC, as claimed.
(cid:3)
0 e(cid:19)(cid:21) = (cid:18)e (g − e)y
0 e(cid:19) ,(cid:18) e 0
w e(cid:19)(cid:21) = (cid:18)
0
e
(cid:19) ,
(cid:20)(cid:18)g 0
e
0
6. Limiting the set of generators for E(n, R, A)
Let us state a result by Wilberd van der Kallen, [21], Lemma 2.2. Morally, it is a
trickier and mightier version of Lemma 1, with a smaller set of generators.
Lemma 11. Let A E R be an ideal of an associative ring, n ≥ 3. Then as a subgroup
E(n, R, A) is generated by E(n, A) and zin(a, d), for all 1 ≤ i ≤ n − 1, a ∈ A, d ∈ R
Generalisations of this result to unipotent radicals of parabolics in arbitrary Cheval-
ley groups were obtained by Alexei Stepanov in [28, 29, 30], (of course, in these papers
R was assumed commutative).
Below we extract the rationale behind these results by van der Kallen and Stepanov
and prove a still stronger version of their results, for the case of GL(n, R). Formally,
it is not necessary for the rest of the paper, Lemma 11 would suffice. Yet, it is so
natural in itself, that it will be certainly prove useful for future applications. Also,
it is a midget version of our main result, Theorem 1, with one ideal instead of three,
and with yij(ab, c) replaced by zij(a, d). Actually, in the next section we breed it up
to a life-size toy version of our main results. Eventually, we believe it should be taken
as the correct definition of relative elementary groups in more general settings, viz.,
for isotropic reductive groups, see [26, 27].
First, we reproduce the proof of Lemma 11 for GL(3, R).
Lemma 12. Let AER be an ideal of an associative ring, n ≥ 3, and let i, j, h be three
pair-wise distinct indices. Then if a subgroup E(n, A) ≤ H ≤ GL(n, R) contains
• either zih(a, d) and zjh(a, d),
• or zhi(a, d) and zhj(a, d),
for all a ∈ A, d ∈ R, then it also contains zij(a, d) and zji(a, d), for all such a and d.
Proof. We prove the desired inclusions for the first case, the second one is treated
similarly. Indeed,
zij(a, d) = tji(d)tij(a) = tji(d)[tih(a), thj(1)] = [tih(a)tjh(da), thj(1)thi(−d)].
Expanding the commutator with respect to the second factor, we see that
zij(a, d) = [tih(a)tjh(da), thj(1)] · thj (1)[tih(a)tjh(da), thi(−d)].
12
NIKOLAI VAVILOV AND ZUHONG ZHANG
Now, expanding both commutators with respect to the first factor, we see that
zij(a, d) = tjh(da)[tih(a), thj(1)] · [tjh(da), thj(1)]·
thj (1)tih(a)[tjh(da), thi(−d)] · thj (1)[tih(a), thi(−d)].
Consider the four factors on the right hand side individually
• tjh(da)[tih(a), thj(1)] = tij(a)tih(−ada),
• [tjh(da), thj(1)] = tjh(da) · zjh(−da, 1),
• thj (1)tih(a)[tjh(da), thi(−d)] = tji(−dad)thi(−dad) · zjh(dada, 1),
• thj (1)[tih(a), thi(−d)] = tih(a)tij(−a)· thj (1)zih(−a,−d), but by Theorem A the last
factor only differs from zih(−a,−d) ∈ GL(2, R, A) by a factor from E(n, A).
We see that all factors only involve matrices from E(n, A) and elementary conju-
(cid:3)
gates of the form zih(b, c) and zjh(b, c), for some b ∈ A, c ∈ R, as claimed.
Theorem 3. Let R be an associative ring with identity 1, n ≥ 3, and let A be a two-
sided ideal of R. Fix an m, 1 ≤ m ≤ n − 1. Then the relative elementary subgroup
E(n, R, A) is generated by
U −
m(A)
and
uvu−1, where
v ∈ Um(A), u ∈ U −
m.
Proof. Let H be the group generated by the above elements. First, observe that
E(n, A) ≤ H. Indeed, the following case analysis shows that H contains all generators
of E(n, A):
tij(a) = zij(a, 0) belongs to H by assumption.
• When i ≤ m and j ≥ m + 1, or when i ≥ m + 1 and j ≤ m, the generator
• When i, j ≤ m take any h ≥ m + 1. Then tij(a) = tih(a) · thj (1)tih(−a) ∈ H.
• When i, j ≥ m + 1 take any h ≤ m. Then tij(a) = tih(1)thj(a) · thj(−a) ∈ H.
Now, we are done by repeated application of Lemma 12. Indeed, zij(a, d) ∈ H by
assumption when i ≤ m, j ≥ m + 1.
• When i, j ≤ m take any h ≥ m + 1. Then zih(a, d), zjh(a, d) ∈ H and thus
zij(a, d), zji(a, d) ∈ H by the first item of Lemma 12.
• When i, j ≥ m + 1 take any h ≤ m. Then zhi(a, d), zhj(a, d) ∈ H and thus
zij(a, d), zji(a, d) ∈ H by the second item of Lemma 12.
Finally, when i ≥ m + 1 and j ≤ m, one has to distinguish two cases.
• If m ≥ 2, one can choose h ≤ m, h 6= j. Then zhj(a, d) ∈ H by assumption,
whereas zhi(a, d) ∈ H by the first item above. Thus, zij(a, d) ∈ H by the second item
of Lemma 12.
• If m ≤ n−2, one can choose h ≥ m+1, h 6= j. Then zih(a, d) ∈ H by assumption,
whereas zjh(a, d) ∈ H by the second item above. Thus, zij(a, d) ∈ H by the first item
of Lemma 12.
It remains only to refer to Lemma 2 -- or Theorem 2, for that matter.
(cid:3)
INCLUSIONS AMONG COMMUTATORS OF ELEMENTARY SUBGROUPS
13
Corollary. Let R be an associative ring with identity 1, n ≥ 3, and let A be a two-
sided ideal of R. Fix an m, 1 ≤ m ≤ n − 1. Then the relative elementary subgroup
E(n, R, A) is generated by the group E(n, A) and the elements zij(a, d), for all i 6= j,
1 ≤ i ≤ m, m + 1 ≤ j ≤ n. a ∈ A, d ∈ R.
7. Three ideals lemma for E(n, C, AB)
It is natural to ask, whether Theorem 2 admits a similar stronger version. Un-
fortunately, the above proof of Theorem 3 does not generalise immediately to the
partially relativised case.
Problem 1. Let R be an associative ring with identity 1, n ≥ 3, and let A, B be a
two-sided ideals of R. Fix an m, 1 ≤ m ≤ n−1. Is the partially relativised elementary
subgroup E(n, B, A) generated by
U −
m(A)
and
uvu−1, where
v ∈ Um(A), u ∈ U −
m(B)?
Instead, we prove the following refinement of Lemma 12, which gives a full scale
generalisation of Proposition 1, and a toy version of our Theorems 1 and 6.
Theorem 4. Let R be an associative ring with identity 1, n ≥ 3, and let A, B, C be
a two-sided ideals of R. Then E(n, C, AB) is contained in any of the following three
spans:
(cid:10)E(n, BC, A), E(n, B, CA)(cid:11),
(cid:10)E(n, A, BC), E(n, CA, B)(cid:11),
(cid:10)E(n, BC, A), E(n, CA, B)(cid:11).
Proof. By Theorem 2 the group E(n, C, AB) is generated by the elementary commu-
tators zij(ab, c), where a ∈ A, b ∈ B, c ∈ C. Now we imitate the proof of Lemma 12,
but now monitor the levels of the occurring parameters of the zrs's in the right hand
side, rather than their positions. As in Lemma 12 we take and h 6= i, j and rewrite
the generator zij(c, ab) as a commutator:
zij(ab, c) = tji(c)tij(ab) = tji(c)[tih(a), thj(b)] = [tih(a)tjh(ca), thj(b)thi(−bc)].
Expanding the commutator with respect to the second factor, we see that
zij(ab, c) = [tih(a)tjh(ca), thj(b)] · thj (b)[tih(a)tjh(ca), thi(−bc)].
Now, expanding both commutators with respect to the first factor, we see that
zij(ab, c) = tjh(ca)[tih(a), thj(b)] · [tjh(ca), thj(b)]·
thj (b)tih(a)[tjh(ca), thi(−bc)] · thj (b)[tih(a), thi(−bc)].
Consider the four factors on the right hand side individually. Observe that by Lemma
4 the commutator subgroup [E(n, A), E(n, B)] and other such double commutators
are normal in E(n, R), so that we can ignore all occurring elementary conjugations.
Amazingly, the only problematic factor is the first one!
• Clearly, tjh(ca)[tih(a), thj(b)] = tij(ab)tih(−abca) belongs to
E(n, AB) ≤ E(n, A) ∩ E(n, B) ≤ E(n, BC, A) ∩ E(n, CA, B).
14
NIKOLAI VAVILOV AND ZUHONG ZHANG
• Further, [tjh(ca), thj(b)] belongs to
[E(n, CA), E(n, B)] ≤ E(n, CA, B) ∩ E(n, B, CA).
• Next, thj (b)tih(a)[tjh(ca), thi(−bc)] belongs to
[E(n, CA), E(n, BC)] ≤ E(n, CA, BC) ∩ E(n, BC, CA) ≤
E(n, A, BC) ∩ E(n, B, CA) ∩ E(n, BC, A) ∩ E(n, CA, B).
• Finally, thj (b)[tih(a), thi(−bc)] belongs to
[E(n, A), E(n, BC)] ≤ E(n, A, BC) ∩ E(n, BC, A).
We see that the third factor belongs to all four subgroups, and can be discarded,
whereas the other three factors are contained in two of the subgroups E(n, BC, A),
Inspecting the cases listed in the
E(n, B, CA), E(n, A, BC), E(n, CA, B), each.
statement, we see that all of them contain all three factors.
(cid:3)
The other three possible combinations of the subgroups E(n, BC, A), E(n, B, CA),
E(n, A, BC), E(n, CA, B), do not seem to work in general. Thus, for instance,
(cid:10)E(n, A, BC), E(n, B, CA)(cid:11) does not contain the first factor, and so on.
8. Stable version of the standard commutator formula
We start with a slightly more general form of [40], Lemma 3 and [41]. Lemma 9.
Essentially, it is a classical corollary of surjective stability for K1, but again we need
a birelative version.
The following lemma is what stays behind [40], Lemma 3, and [41], Lemma 9. Our
argument here is both much more general, and much easier, since it avoids all explicit
computations.
Lemma 13. Let R be an associative ring with 1, n ≥ 3, and let A be a two-sided
ideal of R. Then for any g ∈ GL(n − 1, R, A) and any x ∈ E(n, R) one has
xg ≡ g (mod E(n, R, A)) .
Proof. By Lemma 9 any x ∈ E(n, R) can be expressed as a product x = y1 . . . ym,
where yi alternatively belong to Un−1 or U −
n−1. Consider a shorts such product. We
argue by induction on m.
Let x = yz, where y ∈ E(n, R) is shorter than x, whereas z ∈ Un−1 or z ∈
U −
n−1. By Lemma 10 [g, z] ∈ Un−1(A) in the first case, and [g, z] ∈ U −
n−1(A) in
the first case. Since Un−1(A), U −
n−1(A) ≤ E(n, A) ≤ E(n, R, A), this means that
zg ≡ g (mod E(n, R, A)) . This means that xg ≡ yg (mod E(n, R, A)) . But yg ≡
g (mod E(n, R, A)) by induction hypothesis.
(cid:3)
Of course, one would love to have a similar birelative lemma, asserting that for any
g ∈ GL(n − 1, R, A) and any x ∈ E(n, R, C) one has xg ≡ g (mod E(n, R, A)) . This
would give plenty of leverage, to establish very strong results, including Theorem 1,
with minimum direct calculations.
Unfortunately, it seems that such a lemma does not hold. What we can see easily, is
only the weaker congruence xg ≡ g (mod [E(n, A), E(n, C)]) . The following result is
INCLUSIONS AMONG COMMUTATORS OF ELEMENTARY SUBGROUPS
15
a version of the standard commutator formula that survives for arbitrary associative
rings. Various forms of this result are known for decades, since the groundbreaking
paper by Hyman Bass [4], and the refinements by Alec Mason and Wilson Stothers
[24], see our exposition in [18]. However, the proofs proceeded as follows. First, one
established a more sophisticated double relative version of Whitehead lemma, and
then invoked deep results, such as Bass -- Vaserstein injective stability for K1. Our
proof below is entirely elementary, works for all associative rings, and only uses the
sharp generation results obtained in the previous sections.
Theorem 5. Let R be an associative ring with 1, n ≥ 3, and let A and C be two-sided
ideals of R. Then
[GL(n − 1, R, A), E(n, R, C)] = [E(n, A), E(n, C)].
Proof. Indeed, by Theorem 3 the group E(n, R, C) is generated by w ∈ U −
n−1(C) and
by uvu−1, where v ∈ Un−1(C), u ∈ U −
n−1. Take an arbitrary g ∈ GL(n − 1, R, A).
Then [g, w] ∈ E(n, CA) by Lemma 10. On the other hand, for the other type of
generators one has
[g, uvu−1] = [g, u] · u[g, v] · uv[g, u−1] = [g, u] · u[g, v] · u[v, [g, u−1]] · u[g, u−1].
Now, by Lemma 11 one has [g, v] ∈ E(n, AC), so that u[g, v] ∈ E(n, R, AC). Sim-
ilarly, [g, u−1] ∈ E(n, A), so that [v, [g, u−1]] ∈ [E(n, A), E(n, C)].
It follows from
in fact), that [E(n, A), E(n, C)] is normal in
Lemma 4 (but was known before,
E(n, R). Thus, u[v, [g, u−1]] ∈ [E(n, A), E(n, C)].
By Lemma 5, one has E(n, R, AC) ≤ [E(n, A), E(n, C)] so that both central factors
belong to [E(n, A), E(n, C)]. On the other hand, u[g, u−1] = [g, u]−1. Again invoking
the fact that [E(n, A), E(n, C)] is normal in E(n, R) we see that the commutator
[g, uvu−1] belongs to [E(n, A), E(n, C)].
Since the elements uvu−1 generate E(n, R, C) and are themselves elementary, the
left hand side of the equality in the statement of the theorem is contained in the right
hand side. The other inclusion is obvious.
(cid:3)
Of course, when surjective stability holds for K1(n − 1, R, A), one has
GL(n, R, A) = GL(n − 1, R, A)E(n, R, A),
so that Theorem 4 implies the usual standard commutator formula
[GL(n, R, A), E(n, R, C)] = [E(n, A), E(n, C)].
Otherwise, we use Theorem 4 in the following form.
Corollary. Let R be an associative ring with 1, n ≥ 3, and let A and C be two-sided
ideals of R. Then for any g ∈ GL(n − 1, R, A) and any x ∈ E(n, R, C) one has
xg ≡ g (mod [E(n, A), E(n, C)]) .
16
NIKOLAI VAVILOV AND ZUHONG ZHANG
9. Elementary commutators modulo E(n, R, A ◦ B)
In the present section we collect special cases of the previous results concerning
the behaviour of elementary commutators modulo the level.
Since the elementary commutator yij(a, b), where 1 ≤ i 6= j ≤ n, a ∈ A, b ∈ B, has
level A ◦ B, we get the following result, which is [40], Lemma 3, and [41], Lemma 9.
Lemma 14. Let R be an associative ring with 1, n ≥ 3, and let A, B be two-sided
ideals of R. Then for any 1 ≤ i 6= j ≤ n, a ∈ A, b ∈ B, and any x ∈ E(n, R) one
has
xyij(a, b) ≡ yij(a, b) (mod E(n, R, A ◦ B)) .
It is well known that the absolute elementary group E(n, R) contains all permuta-
tion matrices, maybe after correcting the sign of one entry. Thus, already this lemma
implies that elementary commutators yij(a, b) and yhk(a, b) are congruent modulo
E(n, R, A ◦ B). Of course, we still need Theorem A, since we need to move around
not only the indices, but also the parameters.
The following result is [41], Lemmas 10 and 12.
Lemma 15. Let R be an associative ring with 1, n ≥ 3, and let A, B be two-sided
ideals of R. Then for any 1 ≤ i 6= j ≤ n, a, a1, a2 ∈ A, b, b1, b2 ∈ B one has
yij(a1 + a2, b) ≡ yij(a1, b) · yij(a1, b) (mod E(n, R, A ◦ B)) ,
yij(a, b1 + b2) ≡ yij(a, b1) · yij(a, b2) (mod E(n, R, A ◦ B)) ,
yij(a, b)−1 ≡ yij(−a, b) ≡ yij(a,−b) (mod E(n, R, A ◦ B)) ,
yij(ab1, b2) ≡ yij(a1, a2b) ≡ e (mod E(n, R, A ◦ B)) ,
yij(a1a2, b) ≡ yij(a, b1b2) ≡ e (mod E(n, R, A ◦ B)) .
Together with Theorem A this lemma asserts that modulo E(n, R, A ◦ B) the
elementary commutators yij(a, b) do not depend on the choice of a pair (i, j), i 6= j,
and can be considered as symbols
σ : A/A(A + B) ⊗R B/B(A + B) −→ [E(n, A), E(n, B)]/E(n, R, A ◦ B),
(a + A(A + B)) ⊗ (b + B(A + B)) 7→ y12(a, b) (mod E(n, R, A ◦ B)) .
Let us reiterate [40], Problem 1, and [41], Problem 2.
(1)
(2)
Problem 2. Give a presentation of
by generators and relations. Does this presentation depend on n ≥ 3?
(cid:2)E(n, A), E(n, B)(cid:3)/E(n, R, A ◦ B)
The following lemma is classically known and obvious. It follows from the fact that
in a Dedekind ring R for any two ideals A and B there exists an ideal C ∼= A such
that C + B = R.
Lemma 16. Let R be a Dedekind ring, A and B be ideals of R. Then
A/A(A + B) ∼= B/B(A + B) ∼= R/(A + B).
INCLUSIONS AMONG COMMUTATORS OF ELEMENTARY SUBGROUPS
17
Thus, in this case the above symbols σ can be considered as symbols
σ : R/(A + B) ⊗R R/(A + B) −→ [E(n, A), E(n, B)]/E(n, R, A ◦ B),
closely related to the usual Mennicke symbols. We intend to address Problem 2 for
Dedekind rings in a subsequent paper.
10. Proof of Theorem 1
Now, we are all set to prove the technical heart of the present paper, Theorem 1.
Proof. We take any h 6= i, j and rewrite the elementary commutator yij(ab, c) =
(cid:2)tij(ab), tji(c)(cid:3) as
yij(ab, c) = tij(ab) · tji(c)tij(−ab) = tij(ab) · tji(c)(cid:2)tih(a), thj(−b)(cid:3).
Expanding the conjugation by tji(b), we see that
yij(ab, c) = tij(ab) ·(cid:2)tji(c)tih(a), tji(c)thj(−b)(cid:3) = tij(ab) · [tjh(ca)tih(a), thj(−b)thi(bc)(cid:3).
Expanding the commutator in the right hand side, using multiplicativity of the com-
mutator w.r.t. the second argument, we get
yij(ab, c) = tij(ab) ·(cid:2)tjh(ca)tih(a), thj(−b)(cid:3) · thj (−b)(cid:2)tjh(ca)tih(a), thi(bc)(cid:3).
Expanding the first commutator in the right hand side, and using multiplicativity of
the commutator w.r.t. the first argument, we get
(cid:2)tjh(ca)tih(a), thj(−b)(cid:3) = tjh(ca)(cid:2)tih(a), thj(−b)(cid:3) ·(cid:2)tjh(ca), thj(−b)(cid:3) =
tij(−ab) · tih(abca) · yjh(ca,−b)
Now, the first factor cancels with tij(ab), the second factor belongs to E(n, ABC), and
can be discarded, so that the first commutator is congruent modulo E(n, R, ABC)
to yjh(ca,−b). By Lemma 14 one has
yjh(ca,−b) ≡ yjh(ca, b)−1 (mod E(n, R, CA ◦ B)) .
Next, we look at the second commutator in the right hand side of the formula for
yij(ab, c), and using multiplicativity of the commutator w.r.t. the first argument, we
get
thj (−b)(cid:2)tjh(ca)tih(a), thi(bc)(cid:3) = thj (−b)tjh(ca)(cid:2)tih(a), thi(bc)(cid:3) · thj (−b)(cid:2)tjh(ca), thi(bc)(cid:3) =
thj (−b)tjh(ca)yih(a, bc) · thj (−b)tji(cabc).
Now, the second factor belongs to E(n, ABC), and stays there after an elementary
conjugation, so it can be discarded. The first factor is congruent to yih(a, bc) modulo
E(n, R, A ◦ BC) by Lemma 14. Again by Lemma 14 one has
yih(a, bc) ≡ yhi(bc,−a) ≡ yhi(bc, a)−1 (mod E(n, R, A ◦ BC)) .
Summarising the above, we see that
yij(ab, c)yjh(ca, b)yhi(bc, a) ≡ 1 (mod E(n, R, ABC + BCA + CAB)) ,
18
NIKOLAI VAVILOV AND ZUHONG ZHANG
as claimed. Observe that since all factors are central in E(n, R) modulo the normal
subgroup E(n, R, ABC + BCA + CAB), which equals
E(n, R, A ◦ BC) · E(n, R, B ◦ CA) · E(n, R, C ◦ AB),
their order does not matter.
(cid:3)
By the last remark in the proof of Theorem 1, the levels of all three commutators in
the next result are contained in the normal subgroup E(n, R, ABC + BCA + CAB).
Thus, Theorem 1 immediately implies the following result that can be interpreted as
a three ideals lemma.
Theorem 6. Let R be an associative ring with 1, n ≥ 3, and let A, B, C be two-sided
ideals of R. Then
[E(n, AB), E(n, C)] ≤ [E(n, BC), E(n, A)] · [E(n, CA), E(n, B)].
Modulo Lemma 14 on triple commutator subgroups, it becomes a special case of
the three subgroups lemma, at least in the commutative case. However, the proof
of Lemma 14 itself crucially depends on a version of Theorem A, Theorem 1, or a
similar calculation.
11. Applications to multiple commutators
Our proof of elementary multiple commutator formulas in [41] is an easy induction
that proceeds from the following two special cases, triple commutators, and quadruple
commutators. The following results are [41], Lemma 7 and Lemma 8, respectively.
Lemma 17. Let R be an associative ring with 1, n ≥ 3, and let A, B, C be two-sided
ideals of R. Then
(cid:2)(cid:2)E(n, A), E(n, B)(cid:3), E(n, C)(cid:3) = (cid:2)E(n, A ◦ B), E(n, C)(cid:3).
Lemma 18. Let R be an associative ring with 1, n ≥ 4, and let A, B, C, D be two-
sided ideals of R. Then
(cid:2)(cid:2)E(n, A), E(n, B)(cid:3),(cid:2)E(n, C), E(n, D)(cid:3)(cid:3) = (cid:2)E(n, A ◦ B), E(n, C ◦ D)(cid:3).
These results were first proven for quasi-finite rings by Roozbeh Hazrat and the
second author, under assumption n ≥ 3, see [20]. However, in that paper the proof
was based on (a weaker version of) Theorem A and the usual (commutative!) local-
isation, so that there is no chance to make it work over arbitrary associative rings.
Here, for quadruple commutators we assume that n ≥ 4. The reason was that in [41]
the proof of Lemma proceeds as follows. By Theorem A and Lemma 3 one only has
to prove that
[yij(a, b), yhk(c, d)] ∈ (cid:2)E(n, A ◦ B), E(n, C ◦ D)(cid:3),
for 1 ≤ i 6= j ≤ n, 1 ≤ h 6= k ≤ n, a ∈ A, b ∈ B, c ∈ C, d ∈ D. By Lemma 13
conjugations by elements x ∈ E(n, R) do not matter, since they amount to extra
factors from the above triple commutators, which are already accounted for. Now,
for n ≥ 4 this finishes the proof, since in this case modulo E(n, R, C◦D) we can move
yhk(c, d) to a position, where it commutes with yij(a, b), by Lemma 13 or Theorem A.
INCLUSIONS AMONG COMMUTATORS OF ELEMENTARY SUBGROUPS
19
Problem 3. Prove that Lemma 18 holds also for n = 3, or construct a counter-
example.
Actually,
it seems to us that either way it will be non-trivial. To prove the
lemma one will have to verify that the commutator [yij(a, b), yih(c, d)] of two in-
terlaced elementary commutators belongs where it should, and that's a non-trivial
calculation. On the other hand, since Lemma 18 holds for quasi-finite rings, none
of the usual counter-examples will work, so that one will have to construct a truly
non-commutative counter-example. But to imitate Gerasimov's universal counter-
example would be an extremely troublesome business.
Observe that in fact the three subgroups lemma and Lemma 17 imply the following
poor man's version of our Theorem 1. In the commutative case it is essentially a slight
generalisation of a result by Himanee Apte and Alexei Stepanov [1], Lemma 3.4.
Proposition 2. Let R be an associative ring with identity 1, n ≥ 3, and let A, B
and C be its two-sided ideals. Then
[E(n, A ◦ B), E(n, C)] ≤ [E(n, A ◦ C), E(n, B)] · [E(n, A), E(n, B ◦ C)].
Proof. By the triple commutator formula of elementary subgroups
By the three subgroups lemma
[E(n, A ◦ B), E(n, C)] = (cid:2)[E(n, A), E(n, B)], E(n, C)(cid:3).
(cid:2)[E(n, A), E(n, B)], E(n, C)(cid:3) ≤
Now, applying the triple commutator formula of elementary subgroups to the factors
in the right hand side, we get
(cid:2)[E(n, A), E(n, C)], E(n, B)(cid:3) ·(cid:2)E(n, A), [E(n, B), E(n, C)](cid:3).
(cid:2)[E(n, A), E(n, C)], E(n, B)(cid:3) = (cid:2)E(n, A ◦ C), E(n, B)(cid:3),
(cid:2)E(n, A), [E(n, B), E(n, C)](cid:3) = (cid:2)E(n, A), E(n, B ◦ C)](cid:3).
(cid:3)
By analogy, we can do the same applying the three subgroup lemma to a quadruple
commutator, and then combine it with Lemma 15 again. Of course, this might be
interesting only for the case n = 3.
Proposition 3. Let R be an associative ring with 1, n ≥ 3, and let A, B, C, D be
two-sided ideals of R. Then
(cid:2)[E(n, A), E(n, B)], [E(n, C), E(n, D)](cid:3) ≤
Proof. Indeed, by the three subgroups lemma one has
[E(n, (A ◦ B) ◦ C), E(n, D)] · [E(n, (A ◦ B) ◦ D), E(n, C)].
(cid:2)[E(n, A), E(n, B)], [E(n, C), E(n, D)](cid:3) ≤
(cid:2)(cid:2)[E(n, A), E(n, B)], E(n, C)(cid:3), E(n, D)(cid:3) ·(cid:2)(cid:2)[E(n, A), E(n, B)], E(n, D)(cid:3), E(n, D)(cid:3).
It remains to twice apply Lemma 15 to each factor.
(cid:3)
20
NIKOLAI VAVILOV AND ZUHONG ZHANG
However, it is not feasible to prove Lemma 16 using calculations at the level of
subgroups. Rather, one should go to the level of individual elements. We have a
blueprint how to do that by first applying the Hall -- Witt identity, and then the
identity in Theorem 1 to both resulting factors. However, the calculations seem to
be formidable, and as of today we have not succeeded.
12. Inclusions among commutators for powers of one ideal
Let us state the following exciting special case of Theorem 6.
Proposition 4. Let I be an ideal of an associative ring R. Then
[E(n, I r+s), E(n, I t)] ≤ [E(n, I r), E(n, I s+t)] · [E(n, I s), E(n, I r+t)].
In the case of an even exponent this gives an inclusion among two such commuta-
tors.
Corollary. Let I be an ideal of an associative ring R. Then
[E(n, I 2r), E(n, I t)] ≤ [E(n, I r), E(n, I r+t)].
Iterated application of this proposition allows to establish all inclusions among the
commutator subgroups [E(n, I r), E(n, I s)] of a given level. The following remark-
ably easy argument was suggested to the authors by Fedor Petrov. In the following
theorem we call inclusions that result from Proposition 4, generic.
It is not only
interesting in itself, but also very relevant to obtain definitive results for Dedekind
rings. These inclusions hold for arbitrary associative rings. Of course, for a specific
ring some of them may become equalities.
Theorem 7. Let I be an ideal of an associative ring R, m ≥ 1. Then the generic
lattice of elementary commutator subgroups
H(r) = [E(n, I r), E(n, I m−r)] ≤ E(n, R, I m),
0 ≤ r ≤ m,
of level I m is isomorphic to the lattice of divisors of m. In other words, generically,
[E(n, I r), E(n, I m−r)] ≤ [E(n, I s), E(n, I m−s)] ⇐⇒ gcd(s, m) gcd(r, m).
Proof. Let us understand r in the definition of H(r) modulo m. Then, clearly, one
has H(r) = H(m − r) = H(−r) and H(r + s) ≤ H(r)H(s). Indeed, for r, s ≤ m/2
this is precisely Proposition 4. When r > m/2 or/and s > m/2, we replace one or
both of them by m − r or/and m − s, and then apply Proposition 4.
Indeed, by
induction H(kr) ≤ H((k − 1)r)H(r) ≤ H(r). This establishes the second, and thus
also the first claim of the Theorem.
In particular, this means that H(kr) ≤ H(r) for all k ∈ Z/mZ.
(cid:3)
• In particular, this theorem implies that
H(r) = H(s) ⇐⇒ gcd(r, m) = gcd(s, m)
and that
H(r) ≤ H(s1) . . . H(sl) ⇐⇒ gcd(gcd(s1, m), . . . , gcd(sl, m)) gcd(r, m).
INCLUSIONS AMONG COMMUTATORS OF ELEMENTARY SUBGROUPS
21
r + s = p, are equal.
• At level I p, where p is a prime, all non-trivial double commutators [E(n, I r), E(n, I s)],
• At level I 4 one has
E(n, R, I 4) ≤ [E(n, I 2), E(n, I 2)] ≤ [E(n, I 3), E(n, I)].
The second claim of [24], Theorem 5.4 asserts that the second inclusion may be strict!
Actually, it is strict already in the simplest example, where R = Z[i] is the ring of
Gaussian integers, and I = p = (1 + i)Z[i] is the prime divisor of 2,
[E(n, Z[i], p
2), E(n, Z[i], p
2)] < [E(n, Z[i], p
3), E(n, Z[i], p)],
of index 2. In other words,
whereas
[E(n, Z[i], p
2), E(n, Z[i], p
2)] = E(n, Z[i], p
4),
[E(n, Z[i], p
3), E(n, Z[i], p)] = SL(n, Z[i], p
4).
• At level I 6 one has
E(n, R, I 6) ≤ [E(n, I 3), E(n, I 3)], [E(n, I 4), E(n, I 2)] ≤ [E(n, I 5), E(n, I)],
and there are no obvious inclusions between [E(n, I 3), E(n, I 3)] and [E(n, I 4), E(n, I 2)].
However, the third claim of [24], Theorem 5.4 asserts in the above example of Gauss-
ian integers one has
[E(n, Z[i], p
4), E(n, Z[i], p
2)] = E(n, Z[i], p
6),
whereas
[E(n, Z[i], p
3), E(n, Z[i], p
3)] = [E(n, Z[i], p
5), E(n, Z[i], p)],
is strictly larger, being a proper intermediate subgroup between E(n, Z[i], p6) and
SL(n, Z[i], p6), both indices being equal to 2.
• At level I 30, any of the three subgroups
[E(n, I 6), E(n, I 10)],
[E(n, I 6), E(n, I 15)],
[E(n, I 10), E(n, I 15)]
is contained in the product of the other two.
13. When [E(n, A + B), E(n, A ∩ B)] = [E(n, A), E(n, B)]?
For the ideals themselves, one has an obvious inclusion
(A + B) ◦ (A ∩ B) = (A + B)(A ∩ B) + (A ∩ B)(A + B) ≤ AB + BA = A ◦ B.
Only very rarely this inclusion is always an equality. In fact, it is classically known
that among commutative integral domains (A + B)(A∩ B) = AB characterises Prfer
domains.
A Noetherian Prufer domain is a Dedekind domain, so any Noetherian domain
that is not Dedekind provides a counterexample to the equality. Let, for instance,
R = K[x, y], A = xR, B = yR. Then A+B = xR+yR, whereas A∩B = AB = xyR,
since R is factorial and x and y are coprime. Thus,
(A + B)(A ∩ B) = x2yR + xy2R < AB.
22
NIKOLAI VAVILOV AND ZUHONG ZHANG
The following inclusion can be verified in the style of the original proof of Lemma
7, see [37]. But it is also an immediate corollary of Lemmas 2 -- 6.
Proposition 5. For any two ideals A, B E R, n ≥ 3, one has
[E(n, A + B), E(n, A ∩ B)] ≤ [E(n, A), E(n, B)]
Proof. Lemma 3 and the formula at the beginning of this section show that the level
of the left hand side is contained in the level of the right hand side,
Thus, it only remains to prove that the elementary commutators yij(a+b, c), where
E(cid:0)n, R, (A + B) ◦ (A ∩ B)(cid:1) ≤ E(n, R, A ◦ B).
By Lemma 5, one has
a ∈ A, b ∈ B, c ∈ A ∩ B, in the left hand side belong to the right hand side.
yij(a + b, c) ≡ yij(a, c) · yij(b, c) (mod E(cid:0)n, R, (A + B) ◦ (A ∩ B)(cid:1)) .
Thus, this congruence holds also modulo the larger subgroup E(n, R, A ◦ B).
On the other hand, Lemma 4 implies that
Combining the above congruences, we see that
yij(b, c) ≡ yij(c,−b) (mod E(n, R, A ◦ B)) .
yij(a + b, c) ≡ yij(a, c) · yij(c,−b) (mod E(n, R, A ◦ B)) ,
where both commutators in the right hand side belong to [E(n, A), E(n, B)], which
proves the desired inclusion.
(cid:3)
Of course, when A and B are comaximal, by Lemma 7 one has
[E(n, A + B), E(n, A ∩ B)] = [E(n, A), E(n, B)].
Indeed, in this case A ∩ B = AB so that both sides are equal to E(n, R, AB). This
is also true in the opposite case, when A = B, as in all counter-examples listed in
[41]. But, as we've seen, in general it may break already as regards the levels of these
subgroups, since the level of the left hand side may be strictly smaller, than the level
of the right hand side. Thus, the question remains
Problem 4. When
[E(n, A + B), E(n, A ∩ B)] = [E(n, A), E(n, B)]?
In a subsequent paper we will show that this equality, and in fact much more
general statements, hold for Dedekind rings.
14. Intersections of elementary subgroups
Let A and B be two ideals of a commutative ring R, n ≥ 3. Clearly,
[E(n, A), E(n, B)] ≤ E(n, R, A) ∩ E(n, R, B)].
There is no obvious counter-example to the following stronger claim.
Problem 5. Is it true that for all ideals A, B E R, n ≥ 3, one has
[E(n, A), E(n, B)] ≤ E(n, R, A ∩ B)].
INCLUSIONS AMONG COMMUTATORS OF ELEMENTARY SUBGROUPS
23
This is obviously true when [E(n, A), E(n, B)] = E(n, R, AB). But in all examples
where [E(n, A), E(n, B)] > E(n, R, AB) we are aware of, one still has
[E(n, A), E(n, B)] ≤ E(n, R, A ∩ B).
In these examples usually A = B, when the above inclusion is obvious.
Dually to the Corollary 3 of Lemma 3 one has
GL(n, R, A) ∩ GL(n, R, B) = GL(n, R, A ∩ B),
this equality is classically known, and obvious. The same holds also for congruence
subgroups in SL(n, R).
However, a similar statement for sums of ideals is obviously false for GL(n, R, A),
already in the case R = Z. In other words, in general GL(n, R, A) GL(n, R, B) is
strictly smaller than GL(n, R, A + B), even for comaximal A and B. The trivial
reason is that R may have more invertible elements than just those that can be
expressed as products of invertible elements congruent to 1 modulo A or modulo B.
In fact, even for the easier case of SL(n, R) the equality
SL(n, R, A) SL(n, R, B) = SL(n, R, A + B)
only holds under some very strong assumptions, such as one of the factor-rings R/A
or R/B being semi-local2, see [5], Corollary 9.3, p. 267, or [24], Theorem 2.2.
Also the corresponding property for intersections of elementary subgroups fails in
general.
Proposition 6. For two ideals A, B E R the group E(n, R, A ∩ B) can be strictly
smaller than E(n, R, A) ∩ E(n, R, B).
Proof. Here is the smallest such example. Let R be the ring of integers of the
imaginary quadratic field Q(cid:0)√−7(cid:1). Then R = Z[ζ], where ζ =
R∗ = µ(R) = {±1}.
Now, set p1 = ζR, and p2 = p1 = ζR. Since ζ + ζ = 1, the ideals p1 and p2 are
coprime (= comaximal, in this case). And since ζ · ζ = 2, one has 2 = p1p2 so that
the prime 2 completely decomposes in R.
Recall the formula of Bass -- Milnor -- Serre, [6] for the exponent of the p-part of the
1 + i√7
and
2
order of SK1(R, I):
vp(cid:0) SK1(R, I)(cid:1) = min
pp (cid:20)vp(I)
vp(p) −
1
p − 1(cid:21)[0,vp(µ(R))]
.
Here the minimum is taken over all prime divisors of p in R, while [x][0,m] is the closest
integer in the interval [0, m] to the integer part [x] of x.
Now, set A = p2
1, B = p2
2. The ideals A and B are still comaximal, A + B = R. In
particular, A ∩ B = AB.
2This is automatically the case, for instance, for non-zero ideals in Noetherian integral domains
of dimension 1. Say, for Dedekind rings.
24
NIKOLAI VAVILOV AND ZUHONG ZHANG
Now, this formula implies that SK1(R, A) = SK1(R, B) = 1, in other words,
E(n, R, A) = SL(n, R, A),
E(n, R, B) = SL(n, R, B),
and thus
E(n, R, A) ∩ E(n, R, B) = SL(n, R, A) ∩ SL(n, R, B) = SL(n, R, AB).
On the other hand, SK1(R, A ∩ B) = SK1(R, AB) = {±1}, so that the subgroup
E(n, R, A ∩ B) = E(n, R, AB) has index 2 in E(n, R, A) ∩ E(n, R, B).
(cid:3)
Of course, since A and B are comaximal, by Lemma 7 one has
[E(n, A), E(n, B)] = E(n, R, AB) = E(n, R, A ∩ B),
so that we do not get a counter-example to Problem 3. By the same token, there are
no such counter-examples for imaginary quadratic rings. On the other hand, Lemma
9 the last equality always holds for Dedekind rings of arithmetic type with infinite
multiplicative group, so that in this case there are no counter-examples to Problem
3 either.
15. Final remarks
It would be natural to generalise results of the present paper to more general
contexts.
Problem 6. Generalise Theorem 1 and other results of the present paper to Chevalley
groups.
We do not see any difficulties in treating the simply laced case. However, for doubly
laced systems and for type G2 one might get longer and fancier formulas, than those
in Theorem 1.
Problem 7. Generalise Theorems 1 and other results of the present paper to Bak's
unitary groups.
It was a great experience to collaborate in this field with Roozbeh Hazrat and
Alexei Stepanov over the last decades. Also, we are very grateful to Pavel Kolesnikov
for his questions during our talk, and to Fedor Petrov for suggesting the above proof
of Theorem 6.
References
[1] H. Apte, A. Stepanov, Local-global principle for congruence subgroups of Chevalley groups,
Cent. Eur. J. Math. 12 (2014), no. 6, 801 -- 812.
[2] A. Bak, Non-abelian K-theory: The nilpotent class of K1 and general stability, K -- Theory 4
(1991), 363 -- 397.
[3] A. Bak, N. Vavilov, Structure of hyperbolic unitary groups. I. Elementary subgroups, Algebra
Colloq. 7 (2000), no. 2, 159 -- 196.
[4] H. Bass, K-theory and stable algebra, Inst. Hautes ´Etudes Sci. Publ. Math. (1964), no. 22, 5 -- 60.
[5] H. Bass, Algebraic K-theory. Benjamin, New York, 1968.
[6] H. Bass, J. Milnor, J.-P. Serre, Solution of the congruence subgroup problem for SLn (n ≥ 3)
and Sp2n (n ≥ 2), Inst. Hautes ´Etudes Sci. Publ. Math. 33 (1967) 59 -- 133.
INCLUSIONS AMONG COMMUTATORS OF ELEMENTARY SUBGROUPS
25
[7] Z. I. Borewicz, N. A. Vavilov, The distribution of subgroups in the general linear group over a
commutative ring, Proc. Steklov Inst. Math. 3 (1985), 27 -- 46.
[8] S. C. Geller, C. A. Weibel, Subroups of elementary and Steinberg groups of congruence level I 2,
J. Pure Appl. Algebra 35 (1985), 123 -- 132.
[9] V. N. Gerasimov, Group of units of a free product of rings, Math. U.S.S.R. Sb., 134 (1989),
no. 1, 42 -- 65.
[10] R. Hazrat, A. Stepanov, N. Vavilov, Zuhong Zhang, The yoga of commutators, J. Math. Sci.
179 (2011), no. 6, 662 -- 678.
[11] R. Hazrat, A. Stepanov, N. Vavilov, Zuhong Zhang, Commutator width in Chevalley groups,
Note di Matematica 33 (2013), no. 1, 139 -- 170.
[12] R. Hazrat, A. Stepanov, N. Vavilov, Zuhong Zhang, The yoga of commutators, further applica-
tions, J. Math. Sci. 200 (2014), no. 6, 742 -- 768.
[13] R. Hazrat, N. Vavilov, Bak's work on K-theory of rings (with an appendix by Max Karoubi )
J. K-Theory 4 (2009), no. 1, 1 -- 65.
[14] R. Hazrat, N. Vavilov, Z. Zhang, Relative unitary commutator calculus and applications, J.
Algebra 343 (2011) 107 -- 137.
[15] R. Hazrat, N. Vavilov, Z. Zhang, Relative commutator calculus in Chevalley groups, J. Algebra
385 (2013), 262 -- 293.
[16] R. Hazrat, N. Vavilov, Zuhong Zhang Generation of relative commutator subgroups in Chevalley
groups, Proc. Edinburgh Math. Soc., 59, (2016), 393 -- 410.
[17] R. Hazrat, N. Vavilov, Zuhong Zhang, Multiple commutator formulas for unitary groups. Israel
J. Math., 219 (2017), 287 -- 330.
[18] R. Hazrat, N. Vavilov, Zuhong Zhang, The commutators of classical groups, J. Math. Sci., 222
(2017), no. 4, 466 -- 515.
[19] R. Hazrat, Zuhong Zhang, Generalized commutator formula, Commun. Algebra, 39 (2011),
no. 4, 1441 -- 1454.
[20] R. Hazrat, Zuhong Zhang, Multiple commutator formula, Israel J. Math., 195 (2013), 481 -- 505.
[21] W. van der Kallen, A group structure on certain orbit sets of unimodular rows, J. Algebra 82
(1983), 363 -- 397.
[22] A. Lavrenov, S. Sinchuk, A Horrocks-type theorem for even orthogonal K2, arXiv:1909.02637
v1 [math.GR] 5 Sep 2019, pp. 1 -- 23.
[23] A. W. Mason, On subgroups of GL(n, A) which are generated by commutators. II, J. reine
angew. Math., 322 (1981), 118 -- 135.
[24] A. W. Mason, W. W. Stothers, On subgroups of GL(n, A) which are generated by commutators,
Invent. Math., 23 (1974), 327 -- 346.
[25] J. L. Mennicke, Finite factor groups of the unimodular group, Ann. Math., 81 (1965), 31 -- 37.
[26] V. Petrov, A. Stavrova, Elementary subgroups in isotropic reductive groups, St.Petersburg
Math. J. 20 (2009), no. 4, 625 -- 644.
[27] A. Stavrova, A. Stepanov, Normal structure of isotropic reductive groups over rings, arXiv:1801.
08748v1 [math.GR] 26 Jan 2018, 1 -- 20.
[28] A. Stepanov, Elementary calculus in Chevalley groups over rings, J. Prime Res. Math., 9 (2013),
79 -- 95.
[29] A. V. Stepanov, Non-abelian K-theory for Chevalley groups over rings, J. Math. Sci., 209
(2015), no. 4, 645 -- 656.
[30] A. Stepanov, Structure of Chevalley groups over rings via universal localization, J. Algebra,
450 (2016), 522 -- 548.
[31] A. Stepanov, N. Vavilov, Decomposition of transvections: a theme with variations, K-Theory,
19 (2000), no. 2, 109 -- 153.
[32] A. A. Suslin, The structure of the special linear group over polynomial rings, Math. USSR Izv.,
11 (1977), no. 2, 235 -- 253.
26
NIKOLAI VAVILOV AND ZUHONG ZHANG
[33] L. N. Vaserstein, On the normal subgroups of the GLn of a ring, Algebraic K-Theory, Evanston
1980, Lecture Notes in Math., vol. 854, Springer, Berlin et al., 1981, pp. 454 -- 465.
[34] N. Vavilov, Unrelativised standard commutator formula, Zapiski Nauchnyh Seminarov POMI.
470 (2018), 38 -- 49.
[35] N. Vavilov, Commutators of congruence subgroups in the arithmetic case, Zapiski Nauchnyh
Seminarov POMI. 479 (2019), 5 -- 22.
[36] N. A. Vavilov, A. V. Stepanov, Standard commutator formula, Vestnik St. Petersburg State
Univ., Ser. 1, 41 (2008), no. 1, 5 -- 8.
[37] N. A. Vavilov, A. V. Stepanov, Standard commutator formulae, revisited, Vestnik St. Petersburg
State Univ., Ser.1, 43 (2010), no. 1, 12 -- 17.
[38] N. A. Vavilov, A. V. Stepanov, Linear groups over general rings I. Generalities, J. Math. Sci.,
188 (2013), no. 5, 490 -- 550.
[39] N. Vavilov, Z. Zhang, Generation of relative commutator subgroups in Chevalley groups. II,
Proc. Edinburgh Math. Soc., (2019), 1 -- 16.
[40] N. Vavilov, Z. Zhang, Commutators of relative and unrelative elementary groups, revisited, J.
Math. Sci. 481 (2019), 1 -- 14.
[41] N. Vavilov, Z. Zhang, Multiple commutators of elementary subgroups: end of the line, Israel J.
Math., (2019), 1 -- 14.
[42] N. Vavilov, Z. Zhang, Commutators of relative and unrelative elementary subgroups in Chevalley
groups, , (2019), 1 -- 19.
[43] N. Vavilov, Z. Zhang, Unrelativised commutator formulas for unitary groups, (2019), 1 -- 19.
[44] C. Weibel, K2, K3 and nilpotent ideals, J. Pure Appl. Algebra 18 (1980), 333 -- 345.
[45] V. Wendt On homotopy invariance for homology of rank two groups, J. Pure Appl. Algebra,
216 (2012), no. 10, 2291 -- 2301.
[46] Hong You, On subgroups of Chevalley groups which are generated by commutators, J. Northeast
Normal Univ., (1992), no. 2, 9 -- 13.
St. Petersburg State University
E-mail address: [email protected]
Department of Mathematics, Beijing Institute of Technology, Beijing, China
E-mail address: [email protected]
|
1308.6596 | 2 | 1308 | 2015-12-01T10:04:13 | Weitzenboeck derivations of free metabelian associative algebras | [
"math.RA",
"math.AC"
] | By the classical theorem of Weitzenboeck the algebra of constants (i.e., the kernel) of a nonzero locally nilpotent linear derivation of the polynomial algebra K[X] in d variables over a field K of characteristic 0 is finitely generated. As a noncommutative generalization one considers the algebra of constants of a locally nilpotent linear derivation of a d-generated relatively free algebra F(V) in a variety V of unitary associative algebras over K. It is known that the algebra of constants of F(V) is finitely generated if and only if V satisfies a polynomial identity which does not hold for the algebra of 2 x 2 upper triangular matrices. Hence the free metabelian associative algebra F(M) is a crucial object to study. We show that the vector space of the constants in the commutator ideal F'(M) is a finitely generated module of the algebra of constants of the polynomial algebra K[U,V] in 2d variables, where the derivation acts on U and V in the same way as on X. For small d, we calculate the Hilbert series of the constants in F'(M) and find the generators of the related module. This gives also an (infinite) set of generators of the algebra of constants in F(M). | math.RA | math | WEITZENB OCK DERIVATIONS
OF FREE METABELIAN ASSOCIATIVE ALGEBRAS
RUMEN DANGOVSKI, VESSELIN DRENSKY, S¸EHMUS FINDIK
5
1
0
2
c
e
D
1
]
.
A
R
h
t
a
m
[
2
v
6
9
5
6
.
8
0
3
1
:
v
i
X
r
a
Abstract. By the classical theorem of Weitzenbock the algebra of con-
stants K[Xd]δ of a nonzero locally nilpotent linear derivation δ of the
polynomial algebra K[Xd] = K[x1, . . . , xd] in several variables over a
field K of characteristic 0 is finitely generated. As a noncommutative
generalization one considers the algebra of constants Fd(V)δ of a lo-
cally nilpotent linear derivation δ of a finitely generated relatively free
algebra Fd(V) in a variety V of unitary associative algebras over K. It
is known that Fd(V)δ is finitely generated if and only if V satisfies a
polynomial identity which does not hold for the algebra U2(K) of 2 × 2
upper triangular matrices. Hence the free metabelian associative alge-
bra Fd = Fd(M) = Fd(N2A) = Fd(var(U2(K))) is a crucial object to
d)δ in the com-
study. We show that the vector space of the constants (F ′
d is a finitely generated K[Ud, Vd]δ-module, where δ acts
mutator ideal F ′
on Ud and Vd in the same way as on Xd. For small d, we calculate the
d)δ and find the generators of the K[Ud, Vd]δ-module
Hilbert series of (F ′
d)δ. This gives also an (infinite) set of generators of the algebra F δ
(F ′
d .
Free metabelian associative algebras; algebras of constants; Weitzenbock
derivations.
[2010]16R10; 16S15; 16W20; 16W25; 13N15; 13A50.
1. Introduction
Recall that a derivation of an algebra R over a field K is a linear operator
δ : R → R such that δ(uv) = δ(u)v + uδ(v) for all u, v ∈ R. In this paper
we fix a field K of characteristic 0, an integer d ≥ 2 and a set of variables
Xd = {x1, . . . , xd}. For the polynomial algebra K[Xd] = K[x1, . . . , xd] in
d variables, every mapping δ : Xd → K[Xd] can be extended to a deriva-
tion of K[Xd] which we shall denote by the same symbol δ. We assume
that δ is a Weitzenbock derivation, i.e., acts as a nonzero nilpotent linear
operator of the vector space KXd with basis Xd. Up to a change of the
basis of KXd, the derivation δ is determined by its Jordan normal form
J(δ) with Jordan cells J1, . . . , Js with zero diagonals. Hence the essen-
tially different Weitzenbock derivations are in a one-to-one correspondence
with the partition (p1 + 1, . . . , ps + 1) of d, where p1 ≥ · · · ≥ ps ≥ 0,
(p1 + 1) + · · · + (ps + 1) = d, and the correspondence is given in terms of the
size (pi + 1) × (pi + 1) of the Jordan cells Ji of J(δ), i = 1, . . . , s. We shall
denote the derivation corresponding to this partition by δ(p1, . . . , ps). The
1
2
DANGOVSKI, DRENSKY, FINDIK
algebra of constants of δ
K[Xd]δ = ker δ = {u ∈ K[Xd] δ(u) = 0}
can be studied also with methods of classical invariant theory because it
coincides with the algebra of invariants K[Xd]U T2(K) of the unitriangular
group U T2(K) ∼= {exp(αδ) α ∈ K}. The classical theorem of Weitzenbock
[11] states that for any Weitzenbock derivation δ the algebra of constants
K[Xd]δ is finitely generated, see the book by Nowicki [9] for more details.
The polynomial algebra K[Xd] is free in the class of all commutative al-
gebras. As a noncommutative generalization one considers the relatively
free algebra Fd(V) in a variety V of unitary associative algebras over K,
see e.g., [3] for a background on varieties of algebras. As in the polynomial
case, Fd(V) is freely generated by the set Xd and every map Xd → Fd(V)
can be extended to a derivation of Fd(V). Again, we shall call the deriva-
tions δ which act as nilpotent linear operators of the vector space KXd
Weitzenbock derivations and shall denote them by δ(p1, . . . , ps). In the the-
ory of varieties of unitary associative algebras there is a dichotomy. Either
all polynomial identities of the variety V follow from the metabelian identity
[x1, x2][x3, x4] = 0 (which is equivalent to the condition that V contains the
algebra U2(K) of 2 × 2 upper triangular matrices), or V satisfies an Engel
] = x2adnx1 = 0. Drensky and Gupta
polynomial identity [x2, x1, . . . , x1
[5] studied Weitzenbock derivations δ acting on Fd(V).
In particular, if
U2(K) ∈ V, then the algebra of constants Fd(V)δ is not finitely generated.
If U2(K) does not belong to V, a result of Drensky [4] gives that the algebra
Fd(V)δ is finitely generated. Hence the free metabelian associative algebra
Fd = Fd(M) = Fd(N2A) = Fd(var(U2(K)))
is a crucial object in the study of Weitzenbock derivations.
The present paper may be considered as a continuation of our recent
d)δ of the
paper [2] where we have studied the algebra of constants (Ld/L′′
Weitzenbock derivation δ acting on the free metabelian Lie algebra Ld/L′′
d,
where Ld is the free d-generated Lie algebra. As in the Lie case we show that
the algebra of constants F δ
d when δ acts on the free metabelian associative
algebra Fd = Fd(var(U2(K))) is very close to finitely generated. In the Lie
case the commutator ideal L′
d has a natural structure of a K[Xd]-module.
d)δ is a finitely
In [2] we have seen that the vector space of constants (L′
generated K[Xd]δ-module. In the associative case the commutator ideal F ′
d
d)δ is a
is a K[Xd]-bimodule. We prove that the vector space of constants (F ′
finitely generated K[Ud, Vd]δ-module, where δ acts on the variables Ud and
Vd in the same way as on the variables Xd. Then, using methods of [1] we
give an algorithm how to calculate the Hilbert series of F δ
d and calculate it
for small d. If the Jordan form of δ contains a 1 × 1 cell, we may assume
that δ acts as a nilpotent linear operator on KXd−1 and δ(xd) = 0. In this
situation we express the Hilbert (or Poincar´e) series of F δ
d and the generators
d/L′′
d/L′′
n times
{z
}
WEITZENB OCK DERIVATIONS OF FREE ALGEBRAS
3
of the K[Ud, Vd]δ-module (F ′
the generators of the K[Ud−1, Vd−1]δ-module (F ′
generators of the K[Ud, Vd]δ-module (F ′
which gives also an explicit (infinite) set of generators of the algebra F δ
d .
d−1 and
d−1)δ. Finally, we find the
d)δ for d ≤ 6 (and δ 6= δ(5), δ(3, 1)),
d)δ in terms of the Hilbert series of F δ
2. Finite generation
Let Ad = KhXdi be the free unitary associative algebra of rank d with
d be its commutator ideal,
free generating set Xd = {x1, . . . , xd}, and let A′
i.e., the ideal generated by all commutators
[f, g] = f adg = f g − gf,
f, g ∈ Ad.
The metabelian variety M consists of all associative algebras satisfying the
polynomial identity [x1, x2][x3, x4] = 0. The free metabelian algebra Fd =
d)2. Clearly,
Fd(M) of rank d is isomorphic to the factor-algebra Ad/(A′
K[Xd] ∼= Ad/A′
d. We assume that all Lie commutators are left
d
normed and
∼= Fd/F ′
[x1, . . . , xn−1, xn] = [[x1, , . . . , xn−1], xn] = [x1, . . . , xn−1]adxn.
It is well known, see [3], that the commutator ideal F ′
consisting of all
1 · · · xad
xa1
The metabelian identity implies the identity
d [xj1, xj2, xj3, . . . , xjn],
ai ≥ 0, 1 ≤ ji ≤ d, j1 > j2 ≤ j3 ≤ · · · ≤ jn.
d of Fd has a basis
xiρ(1) · · · xiρ(m)[xj1, xj2, xjσ(3), . . . , xjσ(n)] = xi1 · · · xim[xj1, xj2, xj3, . . . , xjn],
where ρ and σ are arbitrary permutations of 1, . . . , m and 3, . . . , n, respec-
tively. This identity allows to define an action of the polynomial algebra
K[Ud, Vd] on F ′
d, where Ud = {u1, . . . , ud} and Vd = {v1, . . . , vd} are two sets
of commuting variables: If f ∈ F ′
d, then
f ui = xif,
f vi = f adxi,
i = 1, . . . , d.
In this way, the vector subspace F ′
bimodule).
d of Fd is a K[Ud, Vd]-module (or a K[Xd]-
Now we need an embedding of Fd into a wreath product. The construction
is a partial case of the construction of Lewin [7] used in [5] and is similar to
the construction of Shmel'kin [10] in the case of free metabelian Lie algebras
(as used in [2]). Let Yd = {y1, . . . , yd} be a set of commuting variables and
let
Md =
aiK[Ud, V ′
d] = AdK[Ud, V ′
d]
d
Mi=1
be the free K[Ud, V ′
d]-module of rank d freely generated by Ad = {a1, . . . , ad},
where V ′
d}. We equip Md with trivial multiplication Md · Md =
0 and with a structure of a free K[Yd]-bimodule with the action of K[Yd]
defined by
1, . . . , v′
d = {v′
yjai = aiuj,
aiyj = aiv′
j,
i, j = 1, . . . , d.
4
DANGOVSKI, DRENSKY, FINDIK
Then the algebra Wd = K[Yd] ⋌ Md satisfies the metabelian identity and
hence belongs to M. The following proposition is a partial case of [7].
Proposition 2.1. The mapping ı : xj → yj + aj , j = 1, . . . , d, defines an
embedding ı of Fd into Wd.
If
then
w =Xi>jXk,l
βijklX k
d [xi, xj]adlXd,
βijkl ∈ K, X k
d = xk1
1 · · · xkd
d ,
adlXd = (adl1x1) · · · (adldxd),
βijkl(aivj − ajvi)U k
d V l
d , U k
d = uk1
1 · · · ukd
d ,
d = vl1
V l
1 · · · vkl
d ,
ı(w) =Xi>jXk,l
and Vd = {v1, . . . , vd}, vi = v′
i−ui, i = 1, . . . , d. We may replace the variables
V ′
d with Vd. In this way Md becomes a free K[Ud, Vd]-module. To simplify
the notation we shall omit ı and shall consider Fd as a subalgebra of Wd.
Since the action of K[Ud, Vd] on F ′
d agrees with its action on Md, we shall
also think that F ′
d is a K[Ud, Vd]-submodule of Md. If δ is a Weitzenbock
derivation of Fd, such that
δ(xj ) =
αijxi, αij ∈ K,
i, j = 1, . . . , d,
we define an action of δ on Wd assuming that
δ(aj ) =
αijai,
δ(yj) =
d
Xi=1
d
Xi=1
d
Xi=1
αijyi,
d
Xi=1
d
Xi=1
δ(uj ) =
αijui,
δ(vj ) =
αijvi,
j = 1, . . . , d.
Obviously, if w ∈ Fd, then ı(δ(w)) = δ(ı(w)). It is clear that the vector space
d of the constants of δ in the K[Ud, Vd]-module Md is a K[Ud, Vd]δ-module.
M δ
The following lemma is a partial case of [4, Proposition 3].
Lemma 2.2. The vector space M δ
d is a finitely generated K[Ud, Vd]δ-module.
The next theorem is a direct consequence of the lemma.
Theorem 2.3. Let δ be a Weitzenbock derivation of the free metabelian
d)δ of the constants
associative algebra Fd = Fd(M). Then the vector space (F ′
d of Fd is a finitely generated K[Ud, Vd]δ-
of δ in the commutator ideal F ′
module.
Proof. By Lemma 2.2 the K[Ud, Vd]δ-module M δ
d is finitely generated. By
the theorem of Weitzenbock [11] the algebra K[Ud, Vd]δ is also finitely gen-
d)δ, are also
erated. Hence all K[Ud, Vd]δ-submodules of M δ
finitely generated.
(cid:3)
d , including (F ′
WEITZENB OCK DERIVATIONS OF FREE ALGEBRAS
5
3. Hilbert series
The free metabelian associative algebra Fd = Fd(M) is a graded vector
is the homogeneous component of degree n of Fd, then the
space. If F (n)
Hilbert (or Poincar´e) series of Fd is the formal power series
d
H(Fd, z) =Xn≥0
dim F (n)
d zn.
The algebra Fd is also multigraded, with a Zd-grading which counts the
is the
degree of each variable xj in the monomials in Fd.
multihomogeneous component of degree (n1, . . . , nd), then the corresponding
Hilbert series of Fd is given by
If F (n1,...,nd)
d
H(Fd, z1, . . . , zd) = Xnj ≥0
dim F (n1,...,nd)
d
1 · · · znd
zn1
d .
If δ is a Weitzenbock derivation of Fd, then the algebra of constants F δ
also graded and its Hilbert series is
d is
H(F δ
d , z) =Xn≥0
dim(F δ
d )(n)zn.
The Hilbert series of Fd(M) is well known. We shall give the proof for
self-containedness.
Lemma 3.1. The Hilbert series of the free metabelian associative algebra
Fd is
H(Fd, z1, . . . , zd) =
d
Yj=1
1
1 − zj
2 + (z1 + · · · + zd − 1)
d
Yj=1
1
1 − zj
.
Proof. The Hilbert series of a unitary relatively free algebra Fd(V) can be
expressed in terms of the Hilbert series of the subalgebra Bd(V) of the so
called proper (or commutator) polynomials, see, e.g. [3]:
H(Fd(V), z1, . . . , zd) = H(K[Xd], z1, . . . , zd)H(Bd(V), z1, . . . , zd)
=
d
Yi=1
1
1 − zi
H(Bd(V), z1, . . . , zd).
For the metabelian variety M the algebra of proper polynomials is spanned
by the base field K and the commutator ideal L′
d of the free metabelian
Lie algebra. The Hilbert series of L′
d/L′′
d/L′′
d is
H(L′
d/L′′
d, z1, . . . , zd) = 1 + (z1 + · · · + zd − 1)
1
1 − zi
.
d
Yi=1
Hence
H(Fd(M), z1, . . . , zd) =
1
1 − zi
d
Yi=1
(1 + H(L′
d/L′′
d, z1, . . . , zd))
6
DANGOVSKI, DRENSKY, FINDIK
which gives immediately the result.
The Hilbert series of Fd(M) can be calculated directly using that Fd(M)
has a vector space basis consisting of
X a
d [xj1, xj2, xj3, . . . , xjn], X a
d , X a
d = xa1
1 · · · xad
d , j1 > j2 ≤ j3 ≤ · · · ≤ jn, n ≥ 2.
The products X a
d contribute to the factor
1
1 − zi
. The expressions [xj1, xj2, xj3, . . . , xjn]
for n ≥ 1 and without the restriction j1 > j2 behave as the pairs (xj1, xj2xj3 · · · xjn) ∈
Xd × [Xd], where [Xd] is the semigroup of all monomials in K[Xd]. Hence
they give the series
d
Yi=1
(z1 + · · · + zd)
1
1 − zi
.
d
Yi=1
1
1 − zi
d
Yi=1
Now we have to subtract the series
− 1 which corresponds to the
expressions [xj1, xj2, xj3, . . . , xjn] for n ≥ 1 and j1 ≤ j2 (and behave as the
monomials xj1xj2xj3 · · · xjn ∈ [Xd]).
(cid:3)
Starting from the Hilbert series of the algebra Fd = Fd(M) and the sizes
of the Jordan cells of the Weitzenbock derivation δ, there are many ways to
compute the Hilbert series of the algebra of constants F δ
d , see the comments
in [1] and [2]. We shall sketch the way used in [2] in the Lie case.
It is
an improvement (given in [1]) of the method of Elliott [6] and its further
development by McMahon [8], the so called partition analysis or Ω-calculus.
Let δ = δ(p1, . . . , ps) be a Weitzenbock derivation of Fd. As we al-
ready mentioned, the algebra of constants F δ
d coincides with the algebra
of U T2(K)-invariants F U T2(K)
, where U T2(K) ∼= {exp(αδ) α ∈ K}. The
idea is to extend the action of U T2(K) on Fd to an action of the general
linear group GL2(K). Then Fd is a direct sum of irreducible GL2(K)-
modules. Each irreducible component contains a one-dimensional subspace
of U T2(K)-invariants and the algebra F δ
is spanned by these
subspaces. In order to compute the Hilbert series of F δ
d it is sufficient to
find the number of the irreducible GL2(K)-components in each homoge-
neous subspace F (n)
of Fd. The Hilbert series H(Fd, z1, . . . , zd) plays the
role of the character of GLd(K) in the following sense. If g belongs to the
diagonal subgroup Dd(K) of GLd(K) and has eigenvalues ζ1, . . . , ζd, then
d = F U T2(K)
d
d
d
H(Fd, ζ1z, . . . , ζdz) =Xn≥0
tr
F
(n)
d
(g)zn.
The irreducible polynomial GL2(K)-modules are indexed by partitions λ =
(λ1, λ2), λ1 ≥ λ2. If W = W (λ) is an irreducible component of Fd, then
it is a direct sum of one-dimensional subspaces W α = W (α1,α2) such that
α1 + α2 = λ1 + λ2 and λ1 ≥ α1, α2 ≥ λ2. If g ∈ D2(K) has eigenvalues ξ1, ξ2
and w ∈ W α, then g(w) = ξα1
2 w. The unique (up to a multiplicative
1 ξα2
WEITZENB OCK DERIVATIONS OF FREE ALGEBRAS
7
constant) nonzero element w ∈ W λ spans the U T2(K)-invariant subspace of
W . In our case GL2(K) acts polynomially on the vector space KXd and the
vector space KXd has a basis with the property that the diagonal subgroup
D2(K) of GL2(K) acts as a group of d × d diagonal matrices. Then the
GL2(K)-character of Fd is a formal power series
χFd(g, z) =Xn≥0
g =(cid:18)ξ1
0
0
χ
F
(n)
d
(g)zn =Xn≥0
tr
F
(n)
d
(g)zn,
ξ2(cid:19) = ξ1e11 + ξ2e22 ∈ D2(K) ⊂ GL2(K).
s
[i=1
The characters of the irreducible GL2(K)-modules W (λ) are equal to the
Schur functions Sλ(t1, t2) and for g = ξ1e11 + ξ2e22 ∈ D2(K) ⊂ GL2(K)
χFd(g, z) =Xn≥0 X(λ1,λ2)
m(λ1, λ2, n)S(λ1,λ2)(ξ1, ξ2)zn.
Here the nonnegative integer m(λ1, λ2, n) is the multiplicity of W (λ1, λ2)
in F (n)
d can be
solved in two steps. First we have to find the Hilbert series
. Then the problem of computing of the Hilbert series of F δ
d
HGL2(Fd, t1, t2, z) =Xn≥0 X(λ1,λ2)
m(λ1, λ2, n)S(λ1,λ2)(t1, t2)zn
which corresponds to the GL2(K)-action on Fd. Then
H(F δ
d , z) =Xn≥0 X(λ1,λ2)
m(λ1, λ2, n)zn
and we have to use HGL2(Fd, t1, t2, z) and to evaluate the series H(F δ
d , z).
The first part of the problem is solved in the following way. We assume
that Xd is a Jordan basis for the action of δ on the vector space KXd. If
Yi = {xj, xj+1, . . . , xj+pi} is the part of the basis Xd corresponding to the
i-th Jordan cell of δ, we have an action of the linear automorphism exp(δ) on
the vector space KYi which is the same as its action on the vector space of the
binary forms of degree pi. We extend this action to the action of GL2(K). In
this way KXd is a direct sum of the GL2(K)-modules of the binary forms of
degree pi, i = 1, . . . , s. Then we extend this action of GL2(K) diagonally on
the whole Fd. The basis Xd =
Yi consists of eigenvectors of the diagonal
subgroup D2(K) of GL2(K). If g = ξ1e11 + ξ2e22 ∈ D2(K), then
g(xj+k) = ξpi−k
1
ξk
2 xj+k,
k = 0, 1, . . . , pi.
This defines a bigrading on Fd assuming that the bidegree of xj+k is (pi −
k, k). We express the Hilbert series of Fd as a bigraded vector space. For
this purpose we replace in the Hilbert series H(Fd, z1, . . . , zd) the variables
8
DANGOVSKI, DRENSKY, FINDIK
1 z, tpi−1
zj, zj+1, . . . , zj+pi−1, zj+pi corresponding to each set Yi = {xj, xj+1, . . . , xj+pi−1, xj+pi}
by tpi
HGL2(Fd, t1, t2, z) = H(Fd, tp1
2 z, respectively, and obtain the Hilbert series
t2z, . . . , tps
2 z)
t2z, . . . , t1tpi−1
1 z, tps−1
1
2
1
z, tpi
1 z, tp1−1
2 z, . . . , tps
m(λ1, λ2, n)S(λ1,λ2)(t1, t2)zn.
t2z, . . . , tp1
1
=Xn≥0 X(λ1,λ2)
The variable z gives the total degree and t1, t2 count the bidegree induced by
the action of the diagonal subgroup of GL2(K). The coefficient of tn1
2 zn
in HGL2(Fd, t1, t2, z) is equal to the dimension of the elements of Fd which
are linear combinations of products of length n in the variables Xd and are
of bidegree (n1, n2). Then the bigraded Hilbert series of the algebra F δ
1 tn2
d is
HGL2(F δ
d , t1, t2, z) =Xn≥0 X(λ1,λ2)
d , z) =Xn≥0
dim(F δ
and the Hilbert series of F δ
d as a Z-graded vector space is
H(F δ
d )(n)zn = HGL2(F δ
d , 1, 1, z).
m(λ1, λ2, n)tλ1
1 tλ2
2 zn
In this way the second part of the problem reduces to the computing of
the series HGL2(F δ
d , t1, t2, z). The Hilbert series HGL2(Fd, t1, t2, z) is a sym-
metric function with respect to the variables t1, t2.
If we already have a
function f (t1, t2, z) it is much easier to check (even by hand) whether it is
equal to the Hilbert series HGL2(F δ
d , t1, t2, z). It is known that f (t1, t2, z) =
HGL2(F δ
d , t1, t2, z) if and only if
HGL2(Fd, t1, t2, z) =
1
t1 − t2
(t1f (t1, t2, z) − t2f (t2, t1, z)).
But in our situation, given HGL2(Fd, t1, t2, z) only, we want to compute
HGL2(F δ
d , t1, t2, z). For this purpose we use the Elliott-McMahon method.
The Hilbert series HGL2(Fd, t1, t2, z) is a so called nice rational symmetric
function in t1, t2, i.e., its denominator is a product of binomials of the form
1 − tm1
2 zm. The Schur function S(λ1,λ2)(t1, t2) has the expression
1 tm2
S(λ1,λ2)(t1, t2) = (t1t2)λ2 tλ1−λ2+1
1
2
t1 − t2
− tλ1−λ2+1
.
If
then
HGL2(Fd, t1, t2, z) =Xn≥0 X(λ1,λ2)
(t1 − t2)HGL2(Fd, t1, t2, z) =Xn≥0 X(λ1,λ2)
m(λ1, λ2, n)S(λ1,λ2)(t1, t2)zn =Xn≥0
hn(t1, t2)zn,
m(λ1, λ2, n)(tλ1+1
1
tλ2
2 − tλ2
1 tλ1+1
2
)zn
=Xn≥0 Xn1,n2
an1,n2tn1
1 tn2
2 zn.
WEITZENB OCK DERIVATIONS OF FREE ALGEBRAS
9
Hence we have to calculate "half" of the series (t1 − t2)HGL2(Fd, t1, t2, z),
i.e., to evaluate the component with n1 > n2. Then
HGL2(F δ
d , t1, t2, z) =
=Xn≥0 X(λ1,λ2)
1
t1Xn≥0 Xn1>n2
an1,n2tn1
1 tn2
2 zn
m(λ1, λ2, n)tλ1
1 tλ2
2 zn.
The idea of Elliott and McMahon is to replace t1 and t2 with t1u and t2/u,
respectively, and to obtain the Laurent series
(cid:18)t1u −
=
t2
t2
u
, z) =
u(cid:19) HGL2(Fd, t1u,
Xk=−∞
fk(t1, t2, z)uk =Xk<0
+∞
The Hilbert series we need is
+∞
Xk=−∞
Xn≥0 Xn1−n2=k)
fk(t1, t2, z)uk +Xk>0
an1,n2tn1
1 tn2
2 zn
uk
fk(t1, t2, z)uk.
HGL2(F δ
d , t1, t2, z) =
For a nice rational function
fk(t1, t2, z).
1
t1Xk>0
f (t1, t2, z) =Xn≥0 Xn1,n2
an1,n2tn1
1 tn2
2 zn = p(t1, t2, z)Y
1
1 − tm1
1 tm2
2 zm ,
p(t1, t2, z) ∈ K[t1, t2, z], we can use two ways to calculate the component
with n1 > n2. In both cases we replace t1 and t2, respectively, with t1u and
t2/u in f (t1, t2, z). The first method uses the approach of Elliott [6]. The
denominator of f (t1u, t2/u, z) is a product of three kinds of binomials:
1 − tm1
1 tm2
2 zmuk,
1 tm2
1 − tn1
1 tn2
2 znu−l,
1 tn2
2 zmuk and 1−tn1
2 znu−l appear, we use the identity
k, l > 0,
1 − tr1
1 tr2
2 zr.
If both kinds 1−tm1
1
=
1
1 − v1v2(cid:18) 1
1 − v1
+
1
1 − v2
− 1(cid:19)
(1 − v1)(1 − v2)
2 zmuk and v2 = tm1
1 tm2
1 tm2
1 tm2
2 ustk−l)(1−tm1
1 tm2
2 zmu−l and present f (t1u, t2/u, z) as a
2 zmuk)(1 −
2 znu−l) in the denominator is replaced by one of the three kinds of
for v1 = tm1
sum of nice rational functions in which the product (1 − tm1
tn1
1 tn2
products
(1−ts1
1 ts2
1 ts2
Continuing this process, we present f (t1u, t2/u, z) as a sum of a Laurent
polynomial in u and nice rational functions which do not contain both
factors 1 − tm1
2 znu−l. Then the contribution to
2 zmuk and 1 − tn1
2 zmuk), (1−ts1
Pk>0 fk(t1, t2, z)uk is due to the monomials of positive degree with respect
to u in the Laurent polynomial, the nice rational functions with denomina-
tors not depending on factors of the form 1 − tn1
2 znu−l and the constant
1 tn2
2 ustk−l)(1−tn1
1 tn2
2 znu−l), 1−ts1
1 tm2
1 tn2
1 ts2
2 ustk−l.
10
DANGOVSKI, DRENSKY, FINDIK
1 tn2
terms of the nice functions depending on 1 − tn1
2 znu−l. In this way we
also prove that the Hilbert series HGL2(K)(F δ
d , t1, t2, z) is a nice rational
function. We refer to [1] for details.
In our computations we have used
another method. Considering f (t1u, t2/u, z) as a rational function in the
variable u, we present it as a sum of a polynomial p0(t1, t2, z, u) in u with
rational in t1, t2, z coefficients and of partial fractions of the form
p1(t1, t2, z)
uk
,
p2(t1, t2, z)
(1 − tn1
2 znuk)r ,
1 tn2
p3(t1, t2, z)
(1 − tn1
2 znu−k)r ,
1 tn2
where pi(t1, t2, z) are rational functions. Then HGL2(K)(F δ
sum of p0(t1, t2, z, u), the fractions p2(t1, t2, z)/(1 − tn1
functions p2(t1, t2, z), after replacing u with 1.
1 tn2
d , t1, t2, z) is a
2 znuk)r and the
Now we shall give the Hilbert series of the subalgebras of constants of
Weitzenbock derivations of free metabelian associative algebras with small
number of generators.
In some of the cases we give both Hilbert series,
as graded and bigraded vector spaces, because we shall use the results in
the last section of our paper. We do not give results for derivations with
a one-dimensional Jordan cell because we shall handle them in the next
section. We shall work out in detail the case d = 3 and δ = δ(2) only. The
computations in the other cases are similar.
Example 3.2. Let d = 3 and δ = δ(2), i.e., the matrix of δ (acting on KX3)
consists of one Jordan 3 × 3 block. The Hilbert series of the commutator
ideal F ′
3 of F3 with respect to the canonical GL3(K)-action is
H(F ′
3, z1, z2, z3) =
1
(1 − z1)(1 − z2)(1 − z3)
×(cid:18)1 + (z1 + z2 + z3 − 1)
1
(1 − z1)(1 − z2)(1 − z3)(cid:19) .
With respect to the action of D2(K) the elements x1, x2, x3 of the basis of
KX3 are homogeneous of degree (2, 0), (1, 1), (0, 2), respectively. Hence we
replace the variables z1, z2, z3 with t2
2z, respectively. In this way
the Hilbert series of the GL2(K)-module F ′
1z, t1t2z, t2
3 is
HGL2(K)(F ′
3, t1, t2, z) =
1
(1 − t2
1z)(1 − t1t2z)(1 − t2
2z)
×(cid:18)1 + ((t2
1 + t1t2 + t2
2)z − 1)
Now we consider the function
1
(1 − t2
1z)(1 − t1t2z)(1 − t2
2z)(cid:19) .
f (t1, t2, z) = (t1 − t2)HGL2(K)(F ′
3, t1, t2, z),
replace there t1 with t1u and t2 by t2/u and express f (t1u, t2/u, z) as a sum
of partial fractions with respect to u:
t2
1t2zu
(1 − t2
1zu2)2(1 − t1t2z)2(1 − t2
1t2
2z2)
+
(t2
1t2
2z2 − t1t2z − 1)t2
1t2zu
(1 − t2
1zu2)(1 − t1t2z)2(1 + t1t2z)(1 − t2
1t2
2z2)
WEITZENB OCK DERIVATIONS OF FREE ALGEBRAS
11
+
1t4
t3
2z3u
(u2 − t2
2z)(1 − t1t2z)2(1 + t1t2z)(1 − t2
1t2
2z2)
−
t1t4
2z2u
(u2 − t2
2z)2(1 − t1t2z)2(1 − t2
1t2
2z2)
.
The expansion as a Laurent series in u
u
u2 − t2
2z
=
1
u(1 − t2
2z/u)
(t2
2z)k−1
uk
=Xk≥1
contains only negative powers of u. Hence we have to remove the third
summand in f (t1u, t2/u, z), and similarly for the fourth summand. Hence
the Hilbert series HGL2(K)(F ′
3, t1, t2, z) is obtained replacing u by 1 in the
sum
t2
1t2zu
+
(t2
1t2
2z2 − t1t2z − 1)t2
1t2zu
1zu2)(1 − t1t2z)2(1 + t1t2z)(1 − t2
1t2
2z2)
.
1t2
1zu2)2(1 − t1t2z)2(1 − t2
(1 − t2
After simple computations we obtain
1t2z2(1 + (t1 + t2)t2z − t2
t3
(1 − t2
3)δ, t1, t2, z) =
2z2)
1z)2(1 − t1t2z)3(1 + t1t2z)2 ,
1t2
HGL2(K)((F ′
(1 − t2
2z2)
H((F ′
3)δ, z) = HGL2(K)(F ′
3, 1, 1, z) =
z2(1 + 2z − z2)
(1 − z)3(1 − z2)2 ,
Similarly, starting from the Hilbert series of K[X3]
H(K[X3], z1, z2, z3) =
1
(1 − z1)(1 − z2)(1 − z3)
we obtain
HGL2(K[X3]δ, t1, t2, z) =
1
,
1t2
2z2)
1z)(1 − t2
(1 − t2
1
,
H(K[X3]δ, z) =
(1 − z)(1 − z2)
3 , z) = H(K[X3]δ, z) + H((F ′
3)δ, z).
H(F δ
When δ acts on the linear component of K[U3, V3], its matrix consists of
two Jordan 3 × 3 blocks. The Hilbert series of K[U3, V3] with respect to the
canonical GL6(K)-action is
H(K[U3, V3], z1, . . . , z6) =
1
1 − zi
.
6
Yi=1
With respect to the action of D2(K) the elements {u1, v1}, {u2, v2}, {u3, v3}
are homogeneous of degree (2, 0), (1, 1), (0, 2), respectively. Hence we replace
the variables z1, z4 with t2
2z. Again,
decomposing (t1u − t2/u)HGL2(K)(K[U3, V3], t1u, t2/u, z) as a sum of partial
fractions with respect to u and taking only the fractions which contribute
with positive degrees in the expansion as a Laurent series in u, we obtain
1z, z2, z5 with t1t2z, and z3, z6 with t2
HGL2(K[U3, V3]δ, t1, t2, z) =
1t2z2
1 + t3
1z)2(1 − t2
2z2)3 ,
1t2
(1 − t2
12
DANGOVSKI, DRENSKY, FINDIK
H(K[U3, V3]δ, z) =
1 + z2
(1 − z)2(1 − z2)3 .
Once found, it is easy to check that HGL2(K)(F δ
series of F δ
3 , t1, t2, z) is really the Hilbert
3 as a bigraded vector space. It is sufficient to verify the equality
HGL2(F3, t1, t2, z) =
1
t1 − t2
(t1HGL2(F δ
3 , t1, t2, z) − t2HGL2(F δ
3 , t2, t1, z))
which can be seen by direct computations.
Example 3.3. Let δ = δ(p1, . . . , ps) be the Weitzenbock derivation acting
on KXd with Jordan cells of size p1 + 1, . . . , ps + 1. Let δ act on the vector
spaces KUd and KVd in the same way as on KXd. Then the Hilbert series
of the algebras of constants F δ
d = 2, δ = δ(1):
d and K[Ud, Vd]δ are:
HGL2(F δ
2 , t1, t2, z) =
1
1 − t1z
+
t1t2z2
(1 − t1z)2(1 − t1t2z2)
;
H(F δ
2 , z) =
1
1 − z
+
z2
(1 − z)2(1 − z2)
;
HGL2(K[U2, V2]δ, t1, t2, z) =
1
(1 − t1z)2(1 − t1t2z2)
;
H(K[U2, V2]δ, z) =
1
(1 − z)2(1 − z2)
;
d = 3, δ = δ(2):
HGL2(F δ
3 , t1, t2, z) =
1
(1 − t2
1z)(1 − t2
1t2
2z2)
+
1t2(1 + t1t2z)(1 + t1t2z + t2
t3
1t2
1z)2(1 − t2
(1 − t2
2z − t2
2z2)3
1t2
2z2)z2
;
H(F δ
3 , z) =
1
(1 − z)(1 − z2)
+
z2(1 + z)(1 + 2z − z2)
(1 − z)2(1 − z2)3
;
HGL2(K[U3, V3]δ, t1, t2, z) =
1 + t3
1t2z2
1z)2(1 − t2
2z2)3 ;
1t2
(1 − t2
H(K[U3, V3]δ, z) =
1 + z2
(1 − z)2(1 − z2)3 ;
d = 4, δ = δ(3):
H(F δ
4 , z) =
1 − z + z2
(1 − z)2(1 − z4)
+
z2p(z)
(1 − z)4(1 − z2)2(1 − z4)3 ,
p(z) = 2+z+3z2+4z3−6z4−13z5+13z6−14z7+2z8+9z9−5z10+4z11+2z12;
H(K[U4, V4]δ, z) =
1 − 2z + 4z2 − 3z4 − 3z8 + 4z10 − 2z11 + z12
(1 − z)4(1 − z2)2(1 − z4)3
;
WEITZENB OCK DERIVATIONS OF FREE ALGEBRAS
13
d = 4, δ = δ(1, 1):
H(F δ
4 , z) =
1
(1 − z)2(1 − z2)
+
z2(4 + 2z + z2 − 22z3 + 9z4 + 10z5 − 3z6 − 2z7 + z8)
(1 − z)4(1 − z2)5
,
H(K[U4, V4]δ, z) =
1 + z2 − 4z3 + z4 + z6
(1 − z)4(1 − z2)5
;
d = 5, δ = δ(4):
H(F δ
5 , z) =
1 − z + z2
(1 − z)2(1 − z2)(1 − z3)
+
z2q(z)
(1 − z)5(1 − z2)3(1 − z3)3 ,
q(z) = 2+4z+5z2−6z3−15z4+11z5−10z6+3z7+5z8+2z9+z10−5z11+4z12−z13;
H(K[U5, V5]δ, z) =
d = 5, δ = δ(2, 1):
1 − 3z + 6z2 − 7z3 + 3z4 + 2z5 + z6 − 9z7 + 8z8 − 3z9 + z10
(1 − z)5(1 − z2)3(1 − z3)3
;
H(F δ
5 , z) =
1 + z2
(1 − z)2(1 − z2)(1 − z3)
+
z2r(z)
(1 − z)5(1 − z2)3(1 − z3)3 ,
r(z) = 4 + 8z + 8z2 − 9z3 − 20z4 − 5z5 − 2z6 + 9z7 + 7z8 − 2z11 + 3z12 − z13;
H(K[U5, V5]δ, z) =
1 + 5z2 + z3 − 4z4 − 3z5 − 3z6 − 4z7 + z8 + 5z9 + z11
(1 − z)5(1 − z2)3(1 − z3)3
.
As in Example 3.2 we can check these results using the equality
HGL2(Fd, t1, t2, z) =
1
t1 − t2
(t1HGL2(F δ
d , t1, t2, z) − t2HGL2(F δ
d , t2, t1, z)).
4. Derivations with one-dimensional Jordan cell
In this section we study the Weitzenbock derivations δ of the form δ =
δ(p1, . . . , ps−1, 0). In other words, the Jordan normal form J(δ) of the linear
operator δ acting on KXd has a 1×1 cell. Without loss of generality we may
assume that δ acts on the vector spaces KXd, KUd, and KVd as a nilpotent
linear operator such that the vector subspaces KXd−1, KUd−1, and KVd−1
are δ-invariant, and δ(xd) = δ(ud) = δ(vd) = 0. Now let ω(K[Vd−1]) denote
the augmentation ideal of K[Vd−1], i.e., the ideal consisting all polynomials
without constant term, and let K[Ud−1, Vd−1]ω denote the tensor product of
the K-algebras K[Ud−1] and ω(K[Vd−1]), i.e.,
K[Ud−1, Vd−1]ω = K[Ud−1] ⊗K ω(K[Vd−1]).
This algebra can be considered as an ideal of the algebra K[Ud−1, Vd−1], and
as a K[Ud−1, Vd−1]-module generated by Vd−1. By [4, Proposition 3] the
algebra of constants K[Ud−1, Vd−1]δ
ω of K[Ud−1, Vd−1]ω is a finitely generated
K[Ud−1, Vd−1]δ-module.
14
DANGOVSKI, DRENSKY, FINDIK
Lemma 4.1. The Hilbert series of (F ′
related by
d)δ, (F ′
d−1)δ and K[Ud−1, Vd−1]δ
ω are
HGL2((F ′
d)δ, t1, t2, z) =
d−1)δ, t1, t2, z)
1
(1 − z)2(cid:16)HGL2((F ′
ω, t1, t2, z)(cid:17) ,
+zHGL2(K[Ud−1, Vd−1]δ
H((F ′
d)δ, z) =
1
(1 − z)2 (H((F ′
d−1)δ, z) + zH(K[Ud−1, Vd−1]δ
ω, z)).
Proof. If δ acts on KUd−1 and KVd−1 as δ = δ(p1, . . . , ps−1), then it acts on
KUd and KVd as δ(p1, . . . , ps−1, 0). By Lemma 3.1 the Hilbert series of the
commutator ideal of Fd is
H(F ′
d, z1, . . . , zd) =
1
1 − zj
d
Yj=1
H(L′
d/L′′
d, z1, . . . , zd),
d/L′′
where L′
d is the commutator ideal of the free metabelian Lie algebra.
Following the procedure described in Section 3 we replace its variables zj
with tqj
2 z, where the nonnegative integers qj, rj depend on the size of the
corresponding Jordan cell and the position of the variable xj in the Jordan
basis of KXd. In particular, we have to replace the variable zd with z. Hence
1 trj
HGL2(F ′
d, t1, t2, z) =
1
1 − z
d−1
Yj=1
1
1 trj
1 − tqj
2 z
We know from [2, Lemma 4.1] that
HGL2(L′
d/L′′
d, t1, t2, z).
HGL2(L′
d/L′′
d, t1, t2, z) =
d−1/L′′
d−1, t1, t2, z)
1
1 − z(cid:0)HGL2(L′
+zHGL2(ω(K[Vd−1]), t1, t2, z)) .
Thus
=
HGL2(F ′
d, t1, t2, z) =
1
(1 − z)2(cid:0)HGL2(F ′
d−1, t1, t2, z)
+z(HGL2(K[Ud−1], t1, t2, z)HGL2(ω(K[Vd−1]), t1, t2, z)))
1
d−1, t1, t2, z) + zHGL2(K[Ud−1, Vd−1]ω, t1, t2, z)).
(1 − z)2 (HGL2(F ′
Using the fact that the formal power series f (t1, t2, z) is the Hilbert series
of the δ-constants of the Z-graded GL2(K)-module W if and only if
HGL2(W, t1, t2, z) =
1
t1 − t2
(t1f (t1, t2, z) − t2f (t2, t1, z)),
we obtain that the equality
HGL2(F ′
d) =
(1 − z)2 (HGL2(F ′
1
d−1) + zHGL2(K[Ud−1, Vd−1]ω))
WEITZENB OCK DERIVATIONS OF FREE ALGEBRAS
15
is equivalent to the desired equality
HGL2((F ′
d)δ, t1, t2, z) =
d−1)δ, t1, t2, z)
+zHGL2(K[Ud−1, Vd−1]δ
1
(1 − z)2(cid:16)HGL2((F ′
ω, t1, t2, z)(cid:17) .
Clearly this implies that
1
H((F ′
d)δ, z) =
(1 − z)2 (H((F ′
d−1)δ, z) + zH(K[Ud−1, Vd−1]δ
ω, z)).
Recall that F ′
d is a K[Ud, Vd]-module. For every monomial
uj1 · · · ujm ∈ K[Ud−1], vj1 · · · vjn ∈ ω(K[Vd−1]), m, n ≥ 1,
we define a K-linear map π : K[Ud−1, Vd−1]ω → F ′
d by
(cid:3)
π(vj1 · · · vjn) =
[xd, xjk ]vj1 · · · vjk−1vjk+1 · · · vjn,
n
Xk=1
π(ui1 · · · uimvj1 · · · vjn) = π(vj1 · · · vjn)ui1 · · · uim,
and
π(ui1 · · · uim) = 0.
Lemma 4.2. (i) ([2]) The map π satisfies the equality
π(v1v2) = π(v1)v2 + π(v2)v1,
v1, v2 ∈ ω(K[Vd−1]).
(ii) The derivation δ and the map π commute.
Proof. The case (i) was already proved in [2]. It is sufficient to prove the case
(ii) when w = uv ∈ K[Ud−1, Vd−1]ω, and u ∈ K[Ud−1] and v ∈ ω(K[Vd−1])
are monomials. The map π sends v to the commutator ideal L′
d of the free
metabelian Lie algebra L′
d. In [2] we have seen that δ(π(v)) = π(δ(v)).
Now
d/L′′
d/L′′
δ(π(w)) = δ(π(uv)) = δ(π(v)u) = δ(π(v))u+π(v)δ(u) = π(δ(v))u+π(v)δ(u)
= π(δ(v)u) + π(vδ(u)) = π(δ(v)u + vδ(u)) = π(δ(vu)) = π(δ(w))
and this establishes (ii).
(cid:3)
The next theorem and its corollary are the main results of the section.
Theorem 4.3. Let the Weitzenbock derivation δ acting on the vector space
KXd have a Jordan form with a 1 × 1 cell. Let δ act as a nilpotent lin-
ear operator on KXd−1 and δ(xd) = 0, with a similar action on KUd and
KVd. Let {di i ∈ I}, {gj ∈ K[Ud−1, Vd−1]ω j ∈ J} be, respectively, ho-
d−1)δ and of
mogeneous vector space bases of the K[Ud−1, Vd−1]δ-module (F ′
d)δ has a
K[Ud−1, Vd−1]δ
basis
ω with respect to both Z- and Z2-gradings. Then (F ′
{dium
d vn
d , π(gj )um
d vn
d i ∈ I, j ∈ J, m, n ≥ 0}.
16
DANGOVSKI, DRENSKY, FINDIK
d−1)δ and K[Ud−1, Vd−1]δ
Proof. The Hilbert series of (F ′
tively, to the generating functions of their bases. Hence
2 zmi ,
HGL2((F ′
tqi
1 tri
d−1)δ, t1, t2, z) =Xi∈I
ω are equal, respec-
where di is of bidegree (qi, ri) and of total degree mi. Since ud and vd are
of bidegree (0, 0) and of total degree 1, the generating function of the set
D = {dium
d i ∈ I, n ≥ 0} is
d vn
G(D, t1, t2, z) = Xm,n≥0Xi∈I
tqi
1 tri
2 zmizm+n =
1
(1 − z)2 HGL2((F ′
d−1)δ, t1, t2, z).
The map π sends the monomials of K[Ud−1, Vd−1]δ
ω to linear combinations of
commutators with an extra variable xd in the beginning of each commutator.
Hence, if the Hilbert series of K[Ud−1, Vd−1]δ
ω is
HGL2(K[Ud−1, Vd−1]δ
ω, t1, t2, z) =Xj∈J
tkj
1 tlj
2 znj ,
where the bidegree of gj is (kj, lj) and its total degree is nj, then the gener-
ating function of the set E = {π(gj)um′
d j ∈ J, m′, n′ ≥ 0} is
d vn′
2 znj+1zm+n =
tkj
1 tlj
z
(1 − z)2 HGL2(K[Ud−1, Vd−1]δ
ω, t1, t2, z).
G(E, t1, t2, z) = Xm,n≥0Xj∈J
Hence, by Lemma 4.1
HGL2((F ′
d)δ, t1, t2, z) = G(D, t1, t2, z) + G(E, t1, t2, z).
d)δ, we shall conclude that
Since both sets D and E are contained in (F ′
d)δ if we show that the elements of D ∪ E are linearly
D ∪ E is a basis of (F ′
independent. For this purpose it is more convenient to work in the wreath
product Wd.
The elements di belong to F ′
d−1 ⊂ Wd−1 and hence are of the form
d−1
Xk=1
di =
Therefore
akwki(Ud−1, Vd−1), wki(Ud−1, Vd−1) ∈ K[Ud−1, Vd−1]ω.
dium
d vn
d =
akwki(Ud−1, Vd−1)um
d vn
d .
d−1
Xk=1
On the other hand, the elements π(gj)um
d vn
d are of the form
π(gj)um′
d vn′
d =
d−1
Xk=1
[xd, xk]hkj(Ud−1, Vd−1)um′
d vn′
d
=
d−1
Xk=1
(advk − akvd)hkj(Ud−1, Vd−1)um′
d vn′
d ,
WEITZENB OCK DERIVATIONS OF FREE ALGEBRAS
17
where hkj(Ud−1, Vd−1) ∈ K[Ud−1, Vd−1]. Let
w =X ξimndium
d vn
d +X ηjm′n′π(gj)um′
d vn′
d = 0
d vn
d , π(gj)um′
for some ξimn, ηjm′n′ ∈ K. Clearly, we may assume that the elements
dium
d are homogeneous with respect to each of the groups
of variables Ud−1, Ad−1 ∪ Vd−1, {ud}, and {ad, vd}. It follows from the defi-
nition of π that
d vn′
π(vj1 · · · vjq ) =
[xd, xjk ]vj1 · · · vjk−1vjk+1 · · · vjq
q
Xk=1
=
(advjk − ajk vd)vj1 · · · vjk−1vjk+1 · · · vjq
q
Xk=1
= qadvj1 · · · vjq −
ajk vj1 · · · vjk−1vjk+1 · · · vjq vd.
q
Xk=1
Hence, if gj(Ud−1, Vd−1) is homogeneous of degree q with respect to Vd−1,
then
π(gj)um′
d vn′
d = qadgj(Ud−1, Vd−1) −
akhkj(Ud−1, Vd−1)vd! um′
d vn′
d .
d−1
Xk=1
In the linear dependence between the elements dium
replace these elements with their expressions in Wd and obtain
d vn
d and π(gj)um′
d vn′
d we
d−1
w =X ξimn
Xk=1
+X ηjm′n′ qadgj(Ud−1, Vd−1) −
d−1
Xk=1
akwki(Ud−1, Vd−1)um
d vn
d
akhkj(Ud−1, Vd−1)vd! um′
d vn′
d = 0.
Hence m′ = m, n′ = n − 1, and, canceling um
d and vn−1
d
, we obtain
w = qadX ηj,m,n−1gj(Ud−1, Vd−1) −X ηj,m,n−1
d−1
Xk=1
akhkj(Ud−1, Vd−1)vd
+X ξimn
d−1
Xk=1
akwki(Ud−1, Vd−1)vd = 0.
Since the elements gj are linearly independent in K[Ud−1, Vd−1]δ
ω, comparing
the coefficient of ad in w we conclude that ηj,m,n−1 = 0. Then, using that
d−1)δ, we derive that ξimn = 0.
the elements di are linearly independent in (F ′
Hence the set D ∪ E is a basis of (F ′
(cid:3)
d)δ.
18
DANGOVSKI, DRENSKY, FINDIK
Every polynomial f (Ud−1, Vd−1) ∈ K[Ud−1, Vd−1]δ can be written in the
form
f (Ud−1, Vd−1) = f ′(Ud−1) + f ′′(Ud−1, Vd−1),
where f ′(Ud−1) does not depend on Vd−1 and every monomial of f ′′(Ud−1, Vd−1)
contains a variable from Vd−1. Since δ(f ′(Ud−1)) and δ(f ′′(Ud−1, Vd−1)) pre-
serve these properties, both f ′(Ud−1) and f ′′(Ud−1, Vd−1) belong to K[Ud−1, Vd−1]δ.
Hence we may fix a system of generators of the algebra K[Ud−1, Vd−1]δ
{e1(Ud−1), . . . , ek(Ud−1), f1(Ud−1, Vd−1), . . . , fl(Ud−1, Vd−1)},
where ei(Ud−1) ∈ K[Ud−1]δ and all monomials of fj(Ud−1, Vd−1) depend
ω. Every element
on Vd−1,
1 · · · f bl
1 · · · eak
of K[Ud−1, Vd−1]δ
l ,
where b1 + · · · + bl > 0. Hence the set
i.e., fj(Ud−1, Vd−1) belongs to K[Ud−1, Vd−1]δ
ω is a linear combination of products ea1
k f b1
{f1(Ud−1, Vd−1), . . . , fl(Ud−1, Vd−1)}
generates the K[Ud−1, Vd−1]δ-module K[Ud−1, Vd−1]δ
ω. We also fix a set
d−1)δ. Without
{c1, . . . , cm} of generators of the K[Ud−1, Vd−1]δ-module (F ′
loss of generality we may assume that the polynomials in these systems are
homogeneous. Our purpose is to find a generating set of the K[Ud, Vd]δ-
module (F ′
d)δ.
Corollary 4.4. Let Xd be a Jordan basis of the Weitzenbock derivation δ
and let δ have a 1 × 1 Jordan cell corresponding to xd. Let δ act in the
same way on KUd and KVd. Let {c1, . . . , cm} be a homogeneous generat-
ing set of the K[Ud−1, Vd−1]δ-module (F ′
d−1)δ. Let {e1, . . . , ek, f1, . . . , fl} be
a generating set of the algebra K[Ud−1, Vd−1]δ, where e1, . . . , ek do not de-
pend on Vd−1 and every monomial of f1, . . . , fl depends on Vd−1. Then the
K[Ud, Vd]δ-module (F ′
d)δ is generated by the set
{c1, . . . , cm} ∪ {π(f1), . . . , π(fl)}.
k f b1
1 · · · f bl
1 · · · eak
Proof. Clearly, the K[Ud−1, Vd−1]δ-module (F ′
ments ciea1
di from the basis of the vector space (F ′
we obtain also all elements dium
ω is spanned by the products ea1
K[Ud−1, Vd−1]δ
the property b1 + · · · + bl > 0. By Lemma 4.2
d−1)δ is spanned by the ele-
l . In particular, in this way we obtain all elements
d−1)δ. Since ud, vd ∈ K[Ud, Vd]δ,
d . Similarly, the K[Ud−1, Vd−1]δ-module
l which satisfy
1 · · · eak
k f b1
1 · · · f bl
d vn
π(ea1
1 · · · eak
k f b1
1 · · · f bl
l ) =
l
Xj=1
π(fj)ea1
1 · · · eak
k f b1
1 · · · f bj −1
· · · f bl
l .
j
Since K[Ud, Vd]δ = (K[Ud−1, Vd−1]δ)[ud, vd], the elements π(gj )um
d belong
to the K[Ud, Vd]δ-module generated by π(f1), . . . , π(fl). Hence {c1, . . . , cm}∪
{π(f1), . . . , π(fl)} generate the K[Ud, Vd]δ-module (F ′
(cid:3)
d vn
d)δ.
WEITZENB OCK DERIVATIONS OF FREE ALGEBRAS
19
5. Generating sets for small number of generators
In this section we shall find the generators of the K[Ud, Vd]δ-module (F ′
d)δ
for d ≤ 3.
Example 5.1. Let d = 2, δ = δ(1), and let δ(x1) = 0, δ(x2) = x1. It is
well known, see e.g., [9], that K[U2, V2]δ is generated by the algebraically
independent polynomials u1, v1 and u1v2 − u2v1. Hence
HGL2(K[U2, V2]δ, t1, t2, z) =
1
(1 − t1z)2(1 − t1t2z2)
,
as indicated in Example 3.3. The same example gives that
HGL2((F ′
2)δ, t1, t2, z) =
t1t2z2
(1 − t1z)2(1 − t1t2z2)
.
2)δ and is of bidegree
It is easy to see that the element [x2, x1] is belong to (F ′
2)δ generated by
(1, 1). Since the Hilbert series of K[U2, V2]δ-submodule of (F ′
2)δ, we conclude
[x2, x1] is equal to the Hilbert series of the whole module (F ′
2)δ is a free cyclic K[U2, V2]δ-module generated by [x2, x1]. As a
that (F ′
1 (u1v2−u2v1)n,
vector space (F2)δ is spanned by the elements xk
k, l, m, n ≥ 0. Recall that the action of u1v2 − u2v1 on the commutator ideal
F ′
1, [x2, x1]ul
1vm
2 is given by
w(u1v2 − u2v1) = x1[w, x2] − x2[w, x1], w ∈ F ′
2.
Knowing the basis of F δ
ated by the infinite set
2 it is easy to derive that the algebra (F2)δ is gener-
{x1, [x1, x2](u1v2 − u2v1)n n ≥ 0}.
Example 5.2. Let d = 3 and let the Jordan normal form of δ have two cells,
of size 2 × 2 and 1 × 1, respectively. Hence δ = δ(1, 0) in our notation and
we may apply Corollary 4.4. By Example 5.1 for d = 2 and δ = δ(1), the
algebra K[U2, V2]δ is generated by u1, v1 and u1v2 − u2v1. The K[U2, V2]δ-
2)δ is generated by the single element [x2, x1]. In the notation of
module (F ′
Corollary 4.4,
e1 = u1,
f1 = v1,
Hence the K[U3, V3]δ-module (F ′
[x3, x1], and
f2 = u1v2 − u2v1,
3)δ is generated by c1 = [x2, x1], π(f1) =
c1 = [x2, x1].
π(f2) = π(u1v2 − u2v1) = π(v2)u1 − π(v1)u2 = x1[x3, x2] − x2[x3, x1].
These three elements satisfy the relation
c1u1v3 − π(f1)(u1v2 − u2v1) + π(f2)v1 = 0,
i.e.,
x1[[x2, x1], x3]−(x1[[x3, x1], x2]−x2[[x3, x1], x1])+[x1[x3, x2]−x2[x3, x1], x1] = 0.
20
DANGOVSKI, DRENSKY, FINDIK
This result agrees with the Hilbert series
HGL2((F ′
3)δ(1,0), t1, t2, z) =
1t2z4
t1z2 + t1t2z2 + t1t2z3 − t2
(1 − z)2(1 − t1z)2(1 − t1t2z2)
of K[U3, V3]δ-module (F ′
inator correspond to the three generators c1, π(f1), π(f2), and −t2
responds to the relation.)
3)δ. (The summands t1z2, t1t2z2, t1t2z3 in the nom-
1t2z4 cor-
Example 5.3. Let d = 3, δ = δ(2), and let δ(x1) = 0, δ(x2) = x1, δ(x3) =
x2. The generators of K[U3, V3]δ are given in [9]. In our notation they are
f1 = u1,
f2 = v1,
f3 = u2
2 − 2u1u3,
f4 = v2
2 − 2v1v3,
f5 = u1v3 − u2v2 + u3v1,
f6 = u1v2 − u2v1.
There are two more generators in [9],
f7 = 2u2
1v3 − 2u1u2v2 + u2
2v1,
f8 = u1v2
2 − 2v1u2v2 + 2u3v2
1,
but they can be expressed by the other ones:
f7 = f2f3 + 2f1f5,
f8 = f1f4 + 2f2f5.
The generators f1, f2, f3, f4, f5, f6 of K[U3, V3]δ satisfy the defining relation
We easily conclude from the Hilbert series
f 2
6 = f 2
1 f4 + f 2
2 f3 + 2f1f2f5.
HGL2(K[U3, V3]δ, t1, t2, z) =
1t2z2
1 + t3
1z)2(1 − t2
1t2
2z2)3
(1 − t2
that K[U3, V3]δ is a free K[f1, f2, f3, f4, f5]-module with generators 1 and
f6, and that the algebra K[U3, V3]δ has the presentation
1 f4 + f 2
K[U3, V3]δ ∼= K[f1, f2, f3, f4, f5, f6 f 2
2 f3 + 2f1f2f5].
6 = f 2
In particular, as a vector space K[U3, V3]δ has a basis
{f q1
4 f q5
3 f q4
By Example 3.3 the Hilbert series of (F ′
5 , f q1
1 f q2
2 f q3
4 f q5
1 f q2
2 f q3
3 f q4
5 f6 q1, q2, q3, q4, q5 ≥ 0}.
3)δ is
1t2
= t3
2z − t2
HGL2((F ′
1z)2(1 − t2
1t2z2(1 + 2t2
(1 + t1t2z + t2
2z2)(t3
1t2
3)δ, t1, t2, z) =
1t2z2 + t4
2z2)3
2z3(2t2
1 + t1t2) + · · ·
3)δ has a generator c1 of degree
This suggests that the K[U3, V3]δ-module (F ′
2 and bidegree (3, 1). It together with c1f1 and c1f2 give the contribution
1t2z2(1 + 2t2
t3
1z). We also expect three generators c2, c3 and c4 of degree 3
and bidegree (4, 2), (4, 2) and (3, 3), respectively. By easy calculations we
have found the explicit form of c1, c2, c3, c4:
(1 − t2
1t2
1z) + t2
2z3)
1t2
.
c1 = [x2, x1],
c2 = [x3, x1]v1 − [x2, x1]v2 = [x3, x1, x1] − [x2, x1, x2],
c3 = [x3, x1]u1 − [x2, x1]u2 = x1[x3, x1] − x2[x2, x1],
c4 = [x3, x2]u1 − [x3, x1]u2 + [x2, x1]u3 = x1[x3, x2] − x2[x3, x1] + x3[x2, x1].
WEITZENB OCK DERIVATIONS OF FREE ALGEBRAS
21
For example, c4 is a linear combination of all elements of degree 3 and
bidegree (3, 3) of the form: [x3, x2]u1, [x3, x1]u2, and [x2, x1]u3:
c4 = γ1x1[x3, x2] + γ2x2[x3, x1] + γ3x3[x2, x1],
γ1, γ2, γ3 ∈ K,
and the condition δ(c4) = 0 gives
0 = γ1x1[x3, x1] + γ2(x1[x3, x1] + x2[x2, x1]) + γ3x2[x2, x1]
= (γ1 + γ2)x1[x3, x1] + (γ2 + γ3)x2[x2, x1].
Hence
and, up to a multiplicative constant, the only solution is
γ1 + γ2 = γ2 + γ3 = 0
γ1 = 1,
γ2 = −1,
γ3 = 1.
The Hilbert series of the free K[U3, V3]δ-module generated by four elements
of bidegree (3, 1), (4, 2), (4, 2), and (3, 3) is
HGL2(t1, t2, z) =
Hence
t3
1t2z2(1 + (2t1 + t2)t2z)(1 + t3
1t2z2)
(1 − t2
1z)2(1 − t2
1t2
2z2)3
.
HGL2(t1, t2, z) − HGL2((F ′
3)δ, t1, t2, z) = (t2
1 − t2
2)t4
1t2
2z4 + · · ·
which suggests that there is a relation of bidegree (6, 2) and a generator
of bidegree (4, 4). Continuing in the same way, we have found one more
generator
c5 = [x3, x1]u3v1−[x3, x1]u1v3+[x3, x2]u1v2−[x3, x2]u2v1−[x2, x1]u3v2+[x2, x1]u2v3
of bidegree (4, 4) and the relations
R1(6, 2) : c1f6 = c3f2 − c2f1,
R2(7, 3) : c2f6 = c4f 2
2 − c1(f1f4 + f2f5),
R3(7, 3) : c3f6 = c4f1f2 + c1(f1f5 + f2f3),
R4(6, 4) : c4f6 = c2f3 + c3f5 + c5f1,
R5(6, 4) : c5f2 = c2f5 + c3f4,
R6(7, 5) : c5f6 = c1(f3f4 − f 2
5 ) + c4(f1f4 + f2f5),
The above relations show that cjf6, j = 1, . . . , 5, and c5f2 can be replaced
with a linear combination of other generators. Hence the K[U3, V3]δ-module
generated by c1, . . . , c5 is spanned by
E = {cjf mj
1 f nj
3 f qj
2 f pj
∪{c5f m
4 f rj
1 f p
5 mj, nj, pj, qj, rj ≥ 0, j = 1, 2, 3, 4}
3 f q
5 m, p, q, r ≥ 0}.
4 f r
It is easy to check that the generating function of the set E is equal to
3)δ. Hence, if we show that the elements of E are
the Hilbert series of (F ′
22
DANGOVSKI, DRENSKY, FINDIK
linearly independent, we shall conclude that the K[U3, V3]δ-module (F ′
generated by c1, . . . , c5. Let
3)δ is
cjsj = 0,
5
Xj=1
where sj are polynomials in f1, f2, f3, f4, f5, j = 1, . . . , 5, and s5 does not
depend on f2 = v1. We shall show that this implies that sj = 0, j = 1, . . . , 5.
The bidegrees of the polynomials f1, f2, f3, f4, f5 consist of pairs of even
numbers which implies that we can rewrite the equation above as
c1s1 + c4s4 = 0,
c2s2 + c3s3 + c5s5 = 0,
since the only bidegrees consisting of odd numbers are the bidegrees of c1
and c4, which are (3, 1) and (3, 3), respectively. We shall work in the abelian
wreath product W3 and shall denote by wi the coordinate of ai of w ∈ W3.
First we consider the equation c1s1 + c4s4 = 0. The three coordinates wi of
w = c1s1 + c4s4 = a1w1 + a2w2 + a3w3 = 0
define a linear homogeneous system
wi = 0,
i = 1, 2, 3,
with unknowns s1 and s4. Since w3 = (u1v2 − u2v1)s4 = 0, then s4 = 0 and
thus s1 = 0. Now let us consider
w = c2s2 + c3s3 + c5s5 = 0.
We may assume that not all monomials of s2, s3, s5 depend on v2. We
substitute v2 = 0 in the wreath product. Then f1, f2, f3, f4, f5 become
¯f1 = u1,
¯f2 = v1,
¯f3 = u2
2 − 2u1u3,
¯f4 = −2v1v3,
¯f5 = u1v3 + u3v1
and the generators c2, c3 and c5 become
¯c2 = (−a1v3 + a3v1)v1,
¯c3 = −a1u1v3 − a2u2v1 + a3u1v1,
¯c5 = a1(u1v3 − u3v1)v3 + a2u2v1v3 − a3(u1v3 − u3v1)v1
in ¯W3. Also, at least one of the polynomials ¯s2, ¯s3, ¯s5 is different from 0. The
equality ¯w = ¯c2¯s2 + ¯c3¯s3 + ¯c5¯s5 = 0 gives the equalities of the coordinates
¯w1 = −v3(v1¯s2 + u1¯s3 − (u1v3 − u3v1)¯s5) = 0,
¯w2 = −u2v1(¯s3 − 2v3¯s5) = 0,
v1(v1¯s2 + u1¯s3 − (u1v3 − u3v1)¯s5) = 0.
We consider these three equalities as a homogeneous system with unknowns
¯s2, ¯s3, ¯s5. Its only solution is
¯s2 =
(u1v3 + u3v1)¯s5
v1
,
¯s3 = 2v3¯s5.
Since we work with polynomials ¯s2, ¯s3, ¯s5, we conclude that v1 divides ¯s5
which contradicts with the assumption that s5 does not depend on v1. Hence
WEITZENB OCK DERIVATIONS OF FREE ALGEBRAS
23
¯s2 = ¯s3 = ¯s5 = 0. Since ¯f1, ¯f2, ¯f3, ¯f4, and ¯f5 are algebraically independent
in K[U3, v1, v3], we obtain that s2 = s3 = s5 = 0. This completes the proof
that the K[U3, V3]δ-module (F ′
3)δ is generated by c1, . . . , c5.
3 . Obviously, x1 ∈ F δ
2 − 2x1x3 which generate F δ
Now we want to find generators of the algebra F δ
3 modulo the ideal (F ′
3 . We start with x1
and x2
3)δ. We want to lift
them to elements in F δ
3 . It is easy to check that the
element x2
3 and acts by multiplication on F ′
3
in the same way as x2
2 − (x1x3 + x3x1) is the lifting of
x2
2 − 2x1x3 which we are searching for. Then we shall take a subset of E
which, together with x1 and x2
3, then
wf1 = x1w, wf3 = (x2 − (x1x3 + x3x1))w and we can remove the elements
of E
2 − (x1x3 + x3x1) belongs to F δ
2 − 2x1x3. Hence x2
2 − 2x1x3, generates F δ
3 .
If w ∈ F ′
cjf mj
1 f nj
2 f pj
3 f qj
4 f rj
5 ,
j = 1, 2, 3, 4,
c5f m
1 f p
3 f q
4 f r
5
which contain f1 and f3. Hence we may assume that mj, pj, m, p = 0. Also,
wf2 = wx1 −x1w and we may consider only those elements of E with nj = 0.
Similarly,
wf4 = w(x2
2 − (x1x3 + x3x1)) − (x2
2 − (x1x3 + x3x1))w + 2wf5
and we assume that qj = q = 0. Hence, as a vector space
(F ′
3)δ =
K[x1, x2
2 − 2x1x3](cjK[f5])K[x1, x2
2 − 2x1x3].
5
Xj=1
As a consequence we obtain that the algebra F δ
{x1, x2
2 − (x1x3 + x3x1), cjf pj
5
3 is generated by
j = 1, . . . , 5, pj ≥ 0}.
Acknowledgements
The research of the first named author was a part of his project in the
frames of the High School Student Institute at the Institute of Mathematics
and Informatics of the Bulgarian Academy of Sciences. The research of the
second named author was partially supported by Grant Ukraine 01/0007
of the Bulgarian Science Fund for Bilateral Scientific Cooperation between
Bulgaria and Ukraine. The research of the third named author was partially
supported by the Council of Higher Education (Y OK) in Turkey during his
visit as a post-doctoral fellow at the Institute of Mathematics and Infor-
matics of the Bulgarian Academy of Sciences. He is very thankful to the
Institute for the creative atmosphere and the warm hospitality for the period
when this project was carried out. For the revised version of the paper, the
research of the second and the third named authors was partially supported
by Grant I02/18 of the Bulgarian National Science Fund.
References
[1] F. Benanti, S. Boumova, V. Drensky, G.K. Genov, P. Koev, Computing with rational
symmetric functions and applications to invariant theory and PI-algebras, Serdica
Math. J. 38 (2012), 137 -- 188.
24
DANGOVSKI, DRENSKY, FINDIK
[2] R. Dangovski, V. Drensky, S¸. Fındık, Weitzenbock derivations of free metabelian Lie
algebras, Linear Algebra and its Applications 439 (2013), No. 10, 3279 -- 3296.
[3] V. Drensky, Free Algebras and PI-Algebras (Springer-Verlag, Singapore, 1999).
[4] V. Drensky, Invariants of unipotent transformations acting on noetherian relatively
free algebras, Serdica Math. J. 30 (2004), 395 -- 404.
[5] V. Drensky, C.K. Gupta, Constants of Weitzenbock derivations and invariants of
unipotent transformations acting on relatively free algebras, J. Algebra 292 (2005),
393 -- 428.
[6] E.B. Elliott, On linear homogeneous diophantine equations, Quart. J. Pure Appl.
Math. 34 (1903), 348 -- 377.
[7] J. Lewin, A matrix representation for associative algebras . I, Trans. Amer. Math.
Soc. 188 (1974), 29 -- 308.
[8] P.A. MacMahon, Combinatory Analysis, vols. 1 and 2 (Cambridge Univ. Press. 1915,
1916). Reprinted in one volume: (Chelsea, New York, 1960).
[9] A. Nowicki, Polynomial Derivations and Their Rings of Constants (Uniwer-
sytet Mikolaja Kopernika, Torun, 1994). www-users.mat.umk.pl/anow/ps-dvi/pol-
der.pdf.
[10] A.L. Shmel'kin, Wreath products of Lie algebras and their application in the theory
of groups (Russian), Trudy Moskov. Mat. Obshch. 29 (1973), 247 -- 260. Translation:
Trans. Moscow Math. Soc. 29 (1973), 239-252.
[11] R. Weitzenbock, Uber die Invarianten von linearen Gruppen, Acta Math. 58 (1932),
231 -- 293.
Massachusetts Institute of Technology,, 428 Memorial Drive, Cambridge,
MA 02139, U.S.A.
E-mail address: [email protected]
Institute of Mathematics and Informatics,, Bulgarian Academy of Sciences,
1113 Sofia, Bulgaria
E-mail address: [email protected]
Department of Mathematics,, C¸ ukurova University, 01330 Balcalı, Adana,
Turkey
E-mail address: [email protected]
|
1711.03216 | 1 | 1711 | 2017-11-09T00:19:16 | Truncated quantum Drinfeld Hecke algebras and Hochschild cohomology | [
"math.RA",
"math.QA"
] | We consider deformations of quantum exterior algebras extended by finite groups. Among these deformations are a class of algebras which we call truncated quantum Drinfeld Hecke algebras in view of their relation to classical Drinfeld Hecke algebras. We give the necessary and sufficient conditions for which these algebras occur, using Bergman's Diamond Lemma. We compute the relevant Hochschild cohomology to make explicit the connection between Hochschild cohomology and truncated quantum Drinfeld Hecke algebras. To demonstrate the variance of the allowed algebras, we compute both classical type examples and demonstrate an example that does not arise as a factor algebra of a quantum Drinfeld Hecke algebra. | math.RA | math |
TRUNCATED QUANTUM DRINFELD HECKE ALGEBRAS AND
HOCHSCHILD COHOMOLOGY
L. GRIMLEY AND C. UHL
Abstract. We consider deformations of quantum exterior algebras extended by finite groups.
Among these deformations are a class of algebras which we call truncated quantum Drinfeld
Hecke algebras in view of their relation to classical Drinfeld Hecke algebras. We give the
necessary and sufficient conditions for which these algebras occur, using Bergman's Diamond
Lemma. We compute the relevant Hochschild cohomology to make explicit the connection
between Hochschild cohomology and truncated quantum Drinfeld Hecke algebras. To demon-
strate the variance of the allowed algebras, we compute both classical type examples and
demonstrate an example that does not arise as a factor algebra of a quantum Drinfeld Hecke
algebra.
1. Introduction
Drinfeld Hecke algebras occur naturally as deformations of the skew group algebra S(V ) ⋊ G,
the (semi-direct product) algebra formed by a finite group G acting on the polynomial algebra
S(V ) over a vector space V . Drinfeld Hecke algebras appear in diverse areas of mathematics,
including representation theory and orbifold theory. For the development of (quantum) Drinfeld
orbifold algebras and their relation to Hochschild cohomology see [5], [14], and [17]. In [15], She-
pler and Witherspoon used Hochschild cohomology to characterize Drinfeld Hecke algebras. The
same authors further developed deformations of skew group algebras, of which Drinfeld Hecke
algebras arise as a special case, in [16]. Allowing for the introduction of possible noncommutativ-
ity, Levandovskyy and Shepler defined a generalization of Drinfeld Hecke algebras in [7]. These
algebras were analogously defined as a deformation of the skew group algebra Sq(V ) ⋊ G, where
Sq(V ) is the quantum polynomial algebra over V . The class of algebras Hq,κ, a factor algebra of
the tensor algebra of V extended by G, which satisfy the Poincar´e-Birkhoff-Witt (PBW) prop-
erty, they name quantum Drinfeld Hecke algebras. In their paper, Levandovskyy and Shepler
used noncommutative Grobner bases theory to enumerate the necessary and sufficient conditions
for Hq,κ to be a quantum Drinfeld Hecke algebra. Naidu and Witherspoon used this criteria to
classify quantum Drinfeld Hecke algebras in terms of Hochschild cohomology in [10], generating
examples of quantum Drinfeld Hecke algebras coming from actions of complex reflection groups.
In this article, we replace Sq(V ) with the quantum exterior algebra, Λq(V ), and explore
deformations of Λq(V ) ⋊ G. One such class of deformations arises as a factor algebra of the
Hq,κ of [7]. Mirroring the language of [7], we call these algebras for which the PBW property is
satisfied truncated quantum Drinfeld Hecke algebras. In Theorem 2.8, we record the necessary
and sufficient conditions for an algebra to be a truncated quantum Drinfeld Hecke algebra,
modifying the technique of [7] and using Bergman's [2] Diamond Lemma. Our conditions are
nearly identical to those of [7, Theorem 7.6] but with an analogous version of condition (i) and
the addition of conditions (v) and (vi) to respect the truncation on Λq(V ). Because of condition
(i), our characterization allows for truncated quantum Drinfeld Hecke algebras that do not arise
as a factor algebra of a quantum Drinfeld Hecke algebra as our choice of group actions is more
flexible. We include one such example at the end of Section 5. Section 3 establishes the conditions
1
2
L. GRIMLEY AND C. UHL
on Hochschild cohomology given by truncated quantum Drinfeld Hecke algebras.
In Section
4, we compute the relevant elements in Hochschild cohomology, culminating in establishing
the cohomological elements which produce truncated quantum Drinfeld Hecke algebras. With
this result, we further develop the connection between Hochschild cohomology and truncated
quantum Drinfeld Hecke algebras, allowing for another perspective for computations. Finally, in
addition to the example mentioned previously, Section 5 contains examples of truncated quantum
Drinfeld Hecke algebras arising from diagonal group actions, whose Hochschild cohomology was
computed separately by the first author in [6].
Let K be a field of characteristic not 2 and, unless otherwise noted, ⊗ = ⊗K.
2. Truncated Quantum Drinfeld Hecke Algebras
Let q = {qij ∈ K∗} be a set of nonzero elements of a field K with char(K) 6= 2 and let V be a
K-vector space with basis {v1, v2, ..., vn}. Let G ⊂ GL(V ) be a finite group. Restrict to K with
char(K) ∤ G. We denote g ∈ G acting on vj by gvj.
Definition 2.1. Let A be a K-algebra and assume a finite group G acts on A. The skew group
algebra (also called cross product algebra) A ⋊ G is A⊗ KG as a vector space with multiplication
(ag)(bh) = a(gb)gh for all a, b ∈ A and g, h ∈ G.
We suppress the ⊗ in A ⋊ G whenever the context is clear.
Let t be an indeterminate and κ : V × V → KG be a bilinear function. Define an associative
K-algebra, our main algebra of study,
Hq,κ,t := T (V ) ⋊ G[t]/(vjvi − qij vivj − κ(vi, vj)t, v2
i ).
We may also denote κ(vi, vj) =Pg∈G κg(vi, vj )g, dividing κ into its g-components. One should
Hq,κ,t of this paper to Hq,κ,t of [10], in which no truncation conditions occur, and
compare
Hq,κ of [7], in which t = 1 and truncation conditions are not included.
Assigning elements in KG and t degree 0 and vi degree 1,
Hq,κ,t is a filtered algebra.
Definition 2.2. We call
a Poincar´e-Birkhoff-Witt (PBW) basis over K[t].
Hq,κ,t a truncated quantum Drinfeld Hecke algebra over K[t] if it has
In the special case that t = 1, we call the algebras
Hq,κ,1 for which there exists a PBW basis
a truncated quantum Drinfeld Hecke algebra over K. The t = 1 case will be the main interest
of this paper. The inclusion of the indeterminate is necessary for comparing to conditions on
deformations discussed in Sections 3 and 4.
The goal of this Section is to determine precisely the conditions on κ and q for which Hq,κ,1
is a truncated quantum Drinfeld Hecke algebra over K. First we will need to develop some
i ∈ K. As in
Levandovskyy and Shepler [7, Definition 3.1], we make use of the quantum minor determinant.
notation and definitions. For each g ∈ G, let gvj = Pn
Definition 2.3. The quantum (i, j, k, l)-minor determinant of a group element g ∈ G is given
by detijkl(g) := gi
i vi for some scalars gj
i=1 gj
kgj
l gj
l − qij gi
k.
Let
Λq(V ) := Khv1, v2, ..., vnvjvi = qijvivj and v2
i = 0 for i 6= j ∈ {1, 2, ..., n}i
be the quantum exterior algebra on V .
Lemma 2.4 ([10], Lemma 4.2). G acts as an automorphism on Λq(V ) if and only if for all
g ∈ G,
TRUNCATED QUANTUM DRINFELD HECKE ALGEBRAS
3
detsrji(g) = −qijdetsrij(g) for all 1 ≤ s, r, i, j ≤ n and s 6= r, i 6= j,
i gr
gr
j (1 + qij) = 0 for all i < j.
We include a proof here for completeness.
Proof. The proof is a generalization of [7, Lemma 3.2]. Consider
g(vr)g(vs) − qsr
g(vs)g(vr) =Xj,i
detsrij(g)vj vi
where s 6= r. We can rewrite the right hand side as
Xj≤i
detsrij (g)vjvi +Xj>i
detsrij(g)vjvi =Xj<i
=Xj<i
detsrij(g)vivj +Xj
detsrij (g)vjvi + qijXj>i
(detsrij(g) + qjidetsrji(g))vj vi +Xj
2 ··· vαn
1 vα2
detsrjj (g)vjvj.
detsrjj (g)vjvj
Since vj vj = 0 in Λq(V ) and Λq(V ) has a K-basis given by {vα1
right hand side vanishes exactly when
n αi ∈ {0, 1}}, the
detsrji(g) = −qijdetsrij(g) for all 1 ≤ s, r, i, j ≤ n and i 6= j, s 6= r.
get the second condition of the Lemma.
Similarly, we consider g(vrvr) − g(vr)g(vr) = 0 −Pi,j gr
Definition 2.5 ([7], Definition 3.4). The parameter κ is called a quantum 2-form if, for all i
and j, κ(vi, vi) = 0 and κ(vj, vi) = −q−1
Proposition 2.6. If
j vivj. Simplifying this expression, we
(cid:3)
Hq,κ,1 is a truncated quantum Drinfeld Hecke algebra, then
ij κ(vi, vj).
i gr
(i) κ is a quantum 2-form,
(ii) qij = q−1
for i 6= j, and
ji
(iii) G acts on Λq(V ) by automorphisms.
Proof. Suppose Hq,κ,1 is a truncated quantum Drinfeld Hecke algebra. Then Hq,κ,1 has a PBW
basis. Since v2
i = 0 = qiiv2
i + κ(vi, vi), we have that κ(vi, vi) = 0. When i 6= j, we have
vjvi = qijvivj + κ(vi, vj)
= qij(qjivjvi + κ(vj, vi)) + κ(vi, vj)
= qijqjivjvi + qij κ(vj, vi) + κ(vi, vj).
Thus we have that qij 6= 0 for all i 6= j, as well as qij = q−1
for all i 6= j. Furthermore
ij κ(vi, vj). The two conclusions on the parameters is exactly conditions (i) and
ji
κ(vj , vi) = −q−1
(ii).
Now, for all h ∈ G and r 6= s, we have
0 = h(vsvr)h−1 − h(qrsvrvs +Xg∈G
hvs −Xg∈G
hvr − qrs
= hvs
hvr
κg(vr, vs)g)h−1
κg(vr, vs)hgh−1
=Xi,j
=Xi,j
(hs
i hr
i hs
j vivj − qrshr
detrsji(h)vivj −Xg∈G
j vivj ) −Xg∈G
κh−1gh(vr, vs)g
κg(vr, vs)hgh−1
4
L. GRIMLEY AND C. UHL
=Xi<j
(detrsji(h) + qijdetrsij (h))vivj +Xi
detrsii(h)v2
i
+Xg∈G
(Xi<j
detrsij (h)κg(vi, vj) − κh−1gh(vr, vs))g.
Since each term needs to be zero, we recover that detrsij(g) = −qji detrsji(g) for r 6= s and
j ) = 0 for all
(cid:3)
i 6= j. Similarly we consider 0 = h(vrvr)h−1 − h(0)h−1 to recover (1 + qij)(hr
i < j. Therefore, by Lemma 2.4, G acts on Λq(V ) by automorphisms.
i hr
This result agrees with [7, Proposition 3.5] except that conditions (ii) and (iii) are modified by
the added truncation. Notice, in particular, that we get no conditions on the diagonal terms qii.
Had we considered truncation at larger powers or no truncation on vi, we would have required
qii = 1.
With all of this structure, we wish to determine the necessary and sufficient criteria for which
Hq,κ,1 will be a truncated quantum Drinfeld Hecke algebra. Recall that this means determining
the conditions for which Hq,κ,1 is a PBW algebra. According to Bergman's Diamond Lemma [2],
it is enough to check a minimal set of words in an algebra to confirm that the algebra is PBW.
While Bergman says that the main results of his paper are trivial, but far from clear explicitly, it
is the crux of our argument. For completeness, we include the statement of Bergman's Diamond
Lemma here.
Theorem 2.7 ([2], Theorem 1.2). Let S be a reduction system for a free associative algebra
KhXi, and ≤ a semigroup partial ordering on hXi, compatible with S, and having descending
chain condition. All ambiguities of S are resolvable relative to ≤ if and only if all elements of
khXi are reduction-unique under S.
In this setting X = {v1,··· , vn} ∪ G and S are the relations on Hq,κ,1. This result implies
i = 0, v2
we must check the following ambiguities: vkvjvi for k > j > i, vjv2
j vi
for j > i and v2
i = 0 and g ∈ G, gvjvi for j > i and g ∈ G as these are the
monomials with more than one reduction using the relations. Bergman's Diamond Lemma says
it is enough to check that the above monomials are reduction-unique.
i for j > i and v2
j = 0, gv2
i for v2
Theorem 2.8. The factor algebra
only if
Hq,κ,1 is a truncated quantum Drinfeld Hecke algebra if and
(i) G acts on Λq(V ) by automorphisms and qij = q−1
ji
(ii) κ is a quantum 2-form,
(iii) For all h in G and 1 ≤ i < j < k ≤ n,
for i 6= j,
hvj)κh(vi, vk) + (hvi − qij qikvi)κh(vj , vk),
detrsij(h)κg(vi, vj),
hvk − vk)κh(vi, vj) + (qjkvj − qij
0 = (qikqjk
(iv) For all g, h in G and all 1 ≤ r < s ≤ n,
κh−1gh(vr, vs) =Xi<j
(v) For all h in G and 1 ≤ i < j ≤ n,
0 = qij viκh(vi, vj) + hviκh(vi, vj ),
and
0 = qij
(vi) For all g, h ∈ G and i < j
hvjκh(vi, vj) + vjκh(vi, vj), and
(gr
i gr
j )κh(vi, vj) = 0.
Xi<j
TRUNCATED QUANTUM DRINFELD HECKE ALGEBRAS
5
Hq,κ,1 is a truncated quantum Drinfeld Hecke algebra. Then by
Proof. Suppose factor algebra
Hq,κ,1 is an associative algebra,
Proposition 2.6 we have that (i) and (ii) are satisfied. Since
we know that for all 1 ≤ i < j < k ≤ n, 0 = vk(vj vi) − (vkvj)vi. We reorder the vi, vj, vk in
ascending order to recover condition (iii).
0 =vk(vj vi) − (vkvj)vi
=qij vkvivj + vkκ(vi, vj) − qjkvjvkvi − κ(vj, vk)vi
=qij qikvivkvj + qijκ(vi, vk)vj + vkκ(vi, vj) − qjkqikvj vivk − qjkvjκ(vi, vk) − κ(vj, vk)vi
=qij qikvi(qjkvjvk + κ(vj, vk)) + qij κ(vi, vk)vj + vkκ(vi, vj)
− qjkqik(qijvivj + κ(vi, vj))vk − qjkvjκ(vi, vk) − κ(vj, vk)vi
=qij qikqjkvivjvk + qij qikviκ(vj , vk) +Xg∈G
−Xg∈G
gvkκg(vi, vj)g − qjkvjκ(vi, vk) −Xg∈G
qjkqik
qij
gviκg(vj , vk)g.
gvjκg(vi, vk)g + vkκ(vi, vj) − qjkqikqij vivjvk
By rearranging terms, we have that for all g ∈ G, 0 = (qijqikvi − gvi)κg(vj, vk) + (qij
qjkvj)κg(vi, vk) + (vk − qjkqik
is the result of considering 0 = h(vsvr)h−1 − h(qrsvrvs + κ(vr, vs))h−1. If we consider
gvj −
gvk)κg(vi, vj) which is equivalent to condition (iii). Condition (iv)
i ) − (vj vi)vi
0 = vj(v2
= −qijvivjvi − κ(vi, vj)vi
= −qijvi(qijvivj + κ(vi, vj)) − κ(vi, vj)vi
= −qijviκ(vi, vj) − κ(vi, vj)vi,
then for all g ∈ G and all i < j, we need (−gvi − qij vi)κg(vi, vj) = 0. This is the first part of
condition (v). Similarly, 0 = (v2
j )vi − vj(vj vi) gives us the second part of condition (v).
Finally, we consider
0 = g(v2
r ) − (gvr)vr
gr
i vigvr
=Xi
= (Xi
= (Xi<j
= (Xi<j
=Xi<j
gr
gr
i gr
gr
j vj)g
i vi)(Xj
j vivj +Xj<i
j vivj +Xi<j
gr
i gr
gr
i gr
j vivj)g
qij gr
j gr
i vivj + gr
j gr
i κ(vik, vj))g
(gr
i gr
j + qijgr
j gr
i )vivjg + gr
j gr
i κ(vi, vj)g
which gives condition (vi).
We have checked all of the monomials required by Bergman's Diamond Lemma to see if they
are reduction-unique. Therefore the factor algebra Hq,κ,1 is a truncated quantum Drinfeld Hecke
algebra if and only if conditions (i)-(vi) are met.
(cid:3)
We get comparable conditions on Hq,κ,1 being a truncated quantum Drinfeld Hecke algebra
to the non-truncated case [7, Theorem 7.6]. As it compares to the conditions in [7], condition
6
L. GRIMLEY AND C. UHL
(i) is analogous, conditions (ii)-(iv) are identical, and the added conditions (v)-(vi) correspond
to the added truncation.
We end this Section by providing some insight into the conditions for which κ will lead to a
truncated quantum Drinfeld Hecke algebra for a fixed set of quantum scalars. The parameter κ
defines a linear transformation κ : V ⊗ V → KG. The set of all parameters Hom K(V ⊗ V, KG)
is a K-vector space. We are interested in the subset for which κ defines a truncated quantum
Drinfeld Hecke algebra.
Definition 2.9. A parameter κ is admissible if it defines a truncated quantum Drinfeld Hecke
algebra
Hq,κ,t.
That is, κ is admissible if
Hq,κ,t satisfies the conditions in Theorem 2.8. See [4] for the
definition in the setting of affine Hecke algebras. We call the set
PG = {κ ∈ Hom K(V ⊗ V, KG) κ is admissible} ⊂ Hom K(V ⊗ V, KG)
the parameter space. As PG is closed under addition and scalar multiplication, PG is a subspace
of Hom K(V ⊗V, KG). We use the dimension of PG over K to characterize the truncated quantum
Drinfeld Hecke algebras that arise from specific finite groups with fixed system q of quantum
parameters.
Proposition 2.10. In a truncated quantum Drinfeld Hecke algebra, only group elements that
act diagonally on the vector space can support the parameter space.
j = −qji, and gk
k = qkiqkj for all k 6= i, j.
Hq,κ,t is a truncated quantum Drinfeld Hecke algebra and κg(vi, vj) 6= 0. Then
Proof. Suppose
condition (iii) and condition (v) from Theorem 2.8 imply that g is a diagonal action where
i = −qij, gj
gi
(cid:3)
Corollary 2.11. The dimension of the parameter space of a truncated quantum Drinfeld Hecke
algebra is bounded by (cid:0)n
2(cid:1) where n is the dimension of the vector space.
Example 5.3 gives an example where the bound is met.
3. Truncated quantum Drinfeld Hecke algebras as deformations
Truncated quantum Drinfeld Hecke algebras are deformations of the algebra Λq(V ) ⋊ G. In
this Section, we give the conditions for which the algebraic description of truncated quantum
Drinfeld Hecke algebras coincide with the notion of a formal deformation.
Definition 3.1. Let A be an algebra over K. A deformation of A over K[t] is an associative
K[t] algebra with a deformed multiplication determined by bilinear maps µi : A ⊗ A → A where
a ∗ b = ab + µ1(a ⊗ b)t + µ2(a ⊗ b)t2 + ... + µj(a ⊗ b)tk
for all a, b ∈ A and ab is multiplication in A.
Naidu and Witherspoon showed that quantum Drinfeld Hecke algebras over C[t] are defor-
mations of Sq′(V ) over C[t] with deg µi = −2i for all i > 0 in [10, Theorem 2.2] (see Section 5.3
for the definition of Sq′(V )). Using a generalization of their proof, which was a rephrasing of
the proof of [21, Theorem 3.2], we recover a similar result here but with the slight modification
that the deformed multiplication must respect the truncation on Λq(V ).
Theorem 3.2. The truncated quantum Drinfeld Hecke algebras over K[t] are the deformations
of Λq(V ) ⋊ G over K[t] with deg µi = −2i and µi(vj, vj) = 0 for all j ∈ {1, 2, ..., n} and i > 0.
TRUNCATED QUANTUM DRINFELD HECKE ALGEBRAS
7
Proof. Assume
Section 4], the associated graded algebra of
show that
Hq,κ,t is a truncated quantum Drinfeld Hecke algebra over K[t]. Then, by [13,
Hq,κ,t is isomorphic to Λq(V ) ⋊ G[t]. We want to
Hq,κ,t is a formal deformation of Λq(V ) ⋊ G which meets the conditions on µi.
Note that Λq(V ) ⋊ G has a K-basis given by {vα1
n gαi ∈ {0, 1}}. Thus, because
Hq,κ,t is a truncated quantum Drinfeld Hecke algebra, all elements in Hq,κ,t are a K[t]-linear
combination of terms of the form vα1
2 ··· vαn
n g for αi ∈ {0, 1} and g ∈ G. Consider two basis
n h ∈ Hq,κ,t. Denote their product in Hq,κ,t by v ∗ w.
elements v = vα1
Thus v ∗ w = vα1
1 vα2
1 ··· vβn
n )gh. We obtain
n g, w = vβ1
1 ··· vβn
1 ··· vαn
1 ··· vαn
2 ··· vαn
1 vα2
g(vβ1
n
v ∗ w = vw + µ1(v ⊗ w)t + µ2(v ⊗ w)t2 + ··· + µk(v ⊗ w)tk
1 vα2
2 ··· vαn
using defining relations on Hq,κ,t to reorder the vi terms in ascending order. That is, vw is a
linear combination of terms of the form vα1
n gh with αi ∈ {0, 1}, µ1(v ⊗ w) is the sum
of all terms coming from reordering a single adjacent pair vivj , µ2(v ⊗ w) is the sum of all terms
coming from reordering a two adjacent pairs, and similar for higher t powers.
Because the degree drops by two with every application of the relation, the sum is finite (k
is the number of flips of adjacent terms to put the vj's in ascending order) and deg µi = −2i
for all i > 0. Considering the special case where v = vk and w = vk gives µi(vk, vk) = 0 for
all k ∈ {1, 2, ..., n} and i > 0. The operation ∗ is K-linear, making µi linear. Finally, as ∗ is
associative,
On the other hand, assume D is a deformation of Λq(V ) ⋊ G over K[t] with the degree of µi
equal to −2i and µi(vk, vk) = 0 for all k ∈ {1, 2, ..., n} and i > 0. We want to show that D is
Hq,κ,t. Because D is a deformation, D ∼= Λq(V ) ⋊ G[t] as a vector space
isomorphic to some
over K[t]. Consider the map φ : T (V ) ⋊ G[t] → D that sends
φ(vi) = vi and φ(g) = g.
Hq,κ,t is a deformation of Λq(V ) ⋊ G over C[t].
We may extend φ, in the first component, to a unique algebra homomorphism from T (V ) to D.
By the degree condition, we have µi(KG, KG) = 0 = µi(V, KG) = µi(KG, V ). Therefore
φ((vi1 ··· vil g)(vj1 ··· vjm h)) = vi1 ∗ ··· ∗ vil ∗ g ∗ vj1 ∗ ··· ∗ vjm ∗ h
= vi1 ∗ ··· ∗ vil ∗ gvj1 ∗ ··· ∗ gvjm ∗ gh
= φ(vi1 ··· vil
gvj1 ··· gvjm gh)
and the map can be extended to K[t]-algebra homomorphism. The first isomorphism theorem,
with this φ, gives us our desired isomorphism between D and Hq,κ,t.
We will show that φ is surjective by strong induction on degree. By construction, φ(g) = g
and φ(vig) = vi ∗ g = vig + µ1(vi, g)t = vig. Now, let vi1 ··· vim g be an arbitrary monomial
of D. By the induction hypothesis, vi2 ··· vim g ∈ Im(φ). Thus φ(X) = vi2 ··· vim g for some
X ∈ T (V ) ⋊ G[t]. Then
φ(vi1 X) = vi1 ∗ φ(X)
= vi1 ∗ (vi2 ··· vim g)
= vi1 vi2 ··· vim g + µ1(vi1 , vi2 ··· vim g)t + µ2(vi1 , vi2 ··· vim g)t2 + ···
Because deg(µi) = 2i, we know deg(µj (vi1 , vi2 ··· vim g)tj) < m + 1 for all j in the finite sum
above. That is, by the induction hypothesis, µj(vi1 , vi2 ··· vim g)tj ∈ Im φ for all j. Subtracting
the appropriate pre-images, we have vi1 vi2 ··· vim g is in the image of φ.
i ∈ Ker φ because µk(vi ⊗ vi) = 0 for all
k > 0 and v2
Now we determine the kernel of φ. It is clear that v2
i = 0. Now assume i 6= j, then
φ(vivj) = vi ∗ vj = vivj + µ1(vi, vj)t and φ(vj vi) = vj ∗ vi = vjvi + µ1(vj, vi)t.
8
L. GRIMLEY AND C. UHL
Note we do not get higher µi terms because of the degree condition. Also because of the degree
conditions, we know µ1(vi, vj) ∈ KG and µ1(vj , vi) ∈ KG. Using the above expression we get
0 = φ(vj vi − qij vivj) = vjvi + µ1(vj, vi)t − qij vivj − qij µ1(vi, vj)t
= vjvi − qijvivj + t(µ1(vj , vi) − qij µ1(vi, vj))
= 0 + (µ1(vj, vi) − qijµ1(vi, vj))t.
Because φ(g) = g for all g ∈ G,
(3.3)
vj vi − qij vivj − µ1(vj , vi)t + qij µ1(vi, vj)t ∈ Ker φ.
Let I[t] be the ideal generated by the terms of (3.3) and v2
i . It is clear that I[t] ⊆ Ker φ. Note
φ induces a surjection T (V ) ⋊ G[t]/I[t] → D. Because the dimension of D in each degree is less
or equal to the corresponding dimension of T (V ) ⋊ G[t]/I[t], I[t] is, in fact, equal Ker φ.
Therefore, by the first isomorphism theorem, φ induces an isomorphism of algebras D ∼=
Hq,κ,t. Moreover, the
T (V ) ⋊ G[t]/I[t] where the right hand side is also isomorphic to an
associated graded algebra of T (V )⋊G[t]/I[t] is isomorphic to Λq(V )⋊G[t], making D a truncated
quantum Drinfeld Hecke algebra.
(cid:3)
In order for the deformed algebra of A to be associative, µ1 must also be a Hochschild 2-
cocycle. That is, for all a, b, c ∈ A,
aµ1(b ⊗ c) − µ1(ab ⊗ c) + µ1(a ⊗ bc) − µ1(a ⊗ b)c = 0.
As in [10], the proof reveals a relation between the functions κg of Section 2 and Hochschild
2-cocycles µ1 of this Section and the next. Namely,
(3.4)
κg(vj, vi)g = µ1(vi ⊗ vj − qjivj ⊗ vi)
Xg∈G
for i 6= j and κg(vi, vi) = 0. This relation on κ also motivates the definition of ǫβ in the
construction of our projective resolution in the next Section. As deformations of an algebra are
so intimately tied to the Hochschild cohomology of that algebra, we next look at the Hochschild
cohomology of Λq(V ) ⋊ G in order to understand the structure of truncated quantum Drinfeld
Hecke algebras from this perspective.
4. Hochschild cohomology
We use this Section to develop all the homological information we will need.
Definition 4.1. Let A be an algebra over a field K and let M be an A-bimodule. The Hochschild
cohomology of A with coefficients in M is
HH∗(A, M ) := Ext∗
Ae (A, M )
where Ae = A ⊗ Aop and A-bimodules are (left) Ae-modules by left and right action in the
corresponding tensor factor.
When M = A, we call HH∗(A, A) simply the Hochschild cohomology of A and the notation
is shortened to HH∗(A). The standard choice of projective resolution, on which Hochschild
cohomology was originally defined, is the bar resolution.
Definition 4.2. Let A be an algebra over a field K. The bar resolution of A is
B(A) : ...
δ3−→ A ⊗ A ⊗ A ⊗ A
δ2−→ A ⊗ A ⊗ A
δ1−→ A ⊗ A
δ0−→ A → 0
where δm(a0 ⊗ a1 ⊗ ... ⊗ am+1) =Pm
i=0(−1)ia0 ⊗ ... ⊗ aiai+1 ⊗ ... ⊗ am+1.
TRUNCATED QUANTUM DRINFELD HECKE ALGEBRAS
9
In the case of skew group algebras, Hochschild cohomology has a particular form given by the
isomorphism
HH∗(A ⋊ G) ∼= HH∗(A, A ⋊ G)G
if G is a finite group acting on a K-algebra A and char K ∤ G. See [19] for a proof of this result in
the more general setting of Hopf Galois extensions and [8] for the case of Hochschild homology.
Recall from the outset we restricted G to groups such that char K ∤ G. This is one of the settings
in which we explicitly need this assumption. The above isomorphism allows us to compute
HH∗(Λq(V ) ⋊ G) by instead finding the G-invariant subspace of HH∗(Λq(V ), Λq(V ) ⋊ G) =
Lg∈G HH∗(Λq(V ), Λq(V )g). Thus we would like to construct a small resolution of Λq(V ) to
compute Hochschild cohomology. One such small resolution of Λq(V ) was given in [6] which we
describe in the next subsection.
4.1. Projective resolution of Λq(V ). The bar resolution can be computationally cumbersome
so we define a sub-resolution on which we will compute cohomology. Let us start by defining a
generalization of the generators f n
i of [3]. We begin with a more intuitive description. Define
ǫi1,i2,...,in = Xα∈S(i1,...,in)
(−q)α ⊗ vα ⊗ 1
where S(i1, ..., in) is the set of multi-set permutations on a set with i1 1's, i2 2's, ..., and in n's,
vα is the (i1 + i2 + ... + in)-fold tensor product in which the orders of v1, v2, ..., vn are given
by the multi-set permutation α, and (−q)α is the product of the negative quantum coefficients
that occur when permuting the vj's from ascending order to the arrangement given by α. For
example,
ǫ1,0,0,...0 = 1 ⊗ v1 ⊗ 1,
ǫ1,1,0,...,0 = 1 ⊗ v1 ⊗ v2 ⊗ 1 − q21 ⊗ v2 ⊗ v1 ⊗ 1,
ǫ2,1,0,...,0 = 1 ⊗ v1 ⊗ v1 ⊗ v2 ⊗ 1 − q21 ⊗ v1 ⊗ v2 ⊗ v1 ⊗ 1 + q2
21 ⊗ v2 ⊗ v1 ⊗ v1 ⊗ 1.
To more precisely define ǫβ, let's introduce some notation. Let β ∈ Nn and α ∈ {0, 1}n.
Define ǫβ = ǫβ1,β2,...,βn using the multi-index notation, define fβ similarly, and define vα =
vαa
1 vα2
2 ··· vαn
n . Denote β = β1 + β2 + ··· + βn and finally, denote by [j] = (0, ..., 0, 1, 0, ..., 0)
the standard basis n-tuple with one 1 in the jth coordinate.
Let f(0,0,...,0) = 1, f[j] = vj for all j ∈ {1, 2, ..., n}. If βj < 0 for some j ∈ {1, 2, ..., n}, then
setfβ = 0. For arbitrary β ∈ Nn,
fβ =
(−qkj )βk fβ−[j] ⊗ vj.
n
Xj=1Yk>j
Intuitively, each term in the sum of fβ is the consequence of moving an arbitrary vj to the end
of the tensor product and multiplying by the negative of the quantum scalars that would have
occurred in Λq(V ) from this movement. The terms ǫβ are then precisely 1 ⊗ fβ ⊗ 1. Note, with
this new notation, we can update the relation (3.4) to be κ(vj , vi) = µ1(ǫ[i]+[j]) for all i 6= j
further motivating its construction.
Consider the complex, built upon these ǫβ terms,
C : ...
dm+1−−−→ Mβ=m
n
where
Λq(V )ǫβΛq(V )
dm−−→ Mβ=m−1
Λq(V )ǫβΛq(V )
dm−1−−−→ ...
dm(ǫβ) =
Xj=1
(−1)Pl<j βl(Yl<j
qβl
jl vjǫβ−[j] + (−1)βjYl>j
qβl
lj ǫβ−[j]vj).
10
L. GRIMLEY AND C. UHL
It is simple to check that d2 = 0. By [6, Lemma 3.4], (C, d) is a subcomplex of the bar resolution
(B(Λq(V )), δ).
Proposition 4.3. (C, d) is a graded projective resolution of Λq(V ) as a Λq(V )e-module.
Proof. In [6, Section 3], the first author constructed a graded projective (Λq(V ))e-module res-
olution of Λq(V ) given by the total complex of the twisted tensor product of n copies of the
graded projective K[vi]/(v2
i )e-module resolution of K[vi]/(v2
i )
di
Di : ...
3−→ (K[vi]/(v2
2−→ (K[vi]/(v2
i ))eh2i
m(1 ⊗ 1) = vi ⊗ 1 + (−1)m1 ⊗ vi. See [6] for the details of
where µ is multiplication, and di
this construction. For the purposes of this article, we note that the argument hinged upon
an isomorphism between graded modules in the total complex of the twisted tensor product
1−→ (K[vi]/(v2
−→ K[vi]/(v2
i ))eh1i
i ) → 0
i ))e µ
di
di
(Tot(D1 ⊗t1 D2 ⊗t2 ···⊗tn−1 Dn)m and Lβ=m Λq(V )ehβi for β ∈ Zn given by [1, Lemma 4.3].
The former is a graded projective resolution of Λq(V ) as a Λq(V )e-module by [1, Lemma 4.5]
and the latter induced complex we claim is isomorphic to (C, d). The chain map φ : Cm →
Lβ=m Λq(V )ehβi given by sending ǫβ to the copy of 1 ⊗ 1 with homological degree β, induces
the graded isomorphism we need between (C, d) and the latter complex.
(cid:3)
Definition 4.4. A projective resolution is G-compatible if G acts on each algebra in the resolu-
tion and the action commutes with the differentials.
The bar resolution, B(Λq(V )), is G-compatible for all group actions on Λq(V ). It turns out
that this is all we need for the subresolution C to be G-compatible as well.
Lemma 4.5. The group action on V extends to a G-compatible action on C if and only if
(1 + qij )(gr
i gs
j − qsrgs
i gr
i gr
j = 0
(1 + qij )gr
j ) = 0 and
for all r 6= s and i 6= j.
Comparing to Lemma 2.4, C is G-compatible precisely when G acts by automorphisms on
Λq(V ). To close this section, let's record how elements of G must act on the generators of the
resolution, ǫβ. Lemma 4.5 implies
gǫ[r]+[s] =Xl≤k
(gr
l gs
l gr
k − qsrgs
k)ǫ[l]+[k] and gǫ2[r] =Xl≤k
gr
l gr
kǫ[l]+[k]
for all g ∈ G and r < s.
4.2. Relevant 2-cocycles. To compute HH∗(Λq(V ), Λq(V )g), we must apply the functor
Hom Λq(V )e (−, Λq(V )g) to our projective resolution from the previous section.
After passing through the isomorphism Hom (Λq(V ))e(Λq(V )ǫβΛq(V ), Λq(V )g) ∼= Hom K(ǫβ, Λq(V )g),
on which homomorphisms are completely determined by the image of ǫβ, we get the complex
β be the dual Λq(V )e-module homomorphism, sending ǫβ to 1 and all other terms to 0.
Let ǫ∗
Hom (Λq(V ))e(C, Λq(V )g) : ... dm−1
−−−→ Mβ=m−1
Λq(V )gǫ∗
β
dm
−−→ Mβ=m
Λq(V )gǫ∗
β
dm+1
−−−→ ...
where
n
dm((vαg)ǫ∗
β) =
Xj=1
(−1)Pl<j βl(Yl<j
qβl
jl (vjvαg) − (−1)βjYl>j
qβl
lj (vα(gvj)g))ǫ∗
β+[j]
TRUNCATED QUANTUM DRINFELD HECKE ALGEBRAS
11
n
n
=
Xj=1
(−1)Pl<j βl(Yl<j
qβl−αl
jl
(vα+[j]g) − (−1)βjYl>j
qβl
lj
Xi=1
gj
i Yi<l
qαl
il (vα+[i]g))ǫ∗
β+[j].
For the purposes of recovering the structure of truncated quantum Drinfeld Hecke algebras,
we are interested in those 2-cocycles that meet the degree requirement of Theorem 3.2. That
is, we want to find G-invariant elements η ∈ Hom (Λq(V ))e(C2, Λq(V )g) such that d3(η) = 0 and
deg(η) = −2. All elements in C2 have homological and polynomial degree 2 therefore the image
of η must be in Kg. We will call such maps constant maps.
Let η be an arbitrary constant 2-cocycle. Then η is of the form
η = Xg∈G X1≤r≤s≤n
(κg
rsg)ǫ∗
[r]+[s]
for some scalars κg
(κg
rsg)ǫ∗
rs ∈ K. Note ǫ∗
[r]+[s] is the homomorphism sending
[r]+[s] is order independent but κg
rs is only defined for r ≤ s.
1 ⊗ vr ⊗ vs ⊗ 1 − qsr ⊗ vs ⊗ vr ⊗ 1 7→ κg
rsg.
rs}r<s to all r 6= s by defining κg
We may extend {κg
sr to be g-component of the image of
−qsrǫ[r]+[s] = 1 ⊗ vs ⊗ vr ⊗ 1 − qrs ⊗ vs ⊗ vr ⊗ 1 under η. With this compatible extension,
sr = −qrsκg
κg
rs.
The necessary and sufficient conditions for which the coefficient of ǫ[i]+[j]+[k] is 0 in d3(η) for
any i ≤ j ≤ k in HH(Λq(V ), Λq(V )g) are
(4.6)
gvj) + κg
κg
jk(vi−qjiqki(gvi)) − κg
ik(qjivj − qkj
kj vk−gvk) + κg
jk(vj + qkj
ij (qkiqkj vk − gvk) = 0,
gvj ) = 0,
(4.7)
(4.8)
(4.9)
κg
jj(q2
κg
jj (vi−q2
ji
ij (qjivj + gvj) = 0, and
gvi) − κg
κg
kk(vk − gvk) = 0.
What remains is to determine the G-invariant subspace, placing additional constraints on our
2-cocycles. Let h ∈ G and r < s, then
(κg
(κg
(κg
ij g)ǫ∗
ij g)ǫ∗
(hη)(ǫ[r]+[s]) =hXg∈GXi<j
=hXg∈GXi<j
=hXg∈GXi<j
=Xg∈GXi<j
(hη)(ǫ2[r]) =hXg∈GXi<j
=hXg∈GXi<j
=Xg∈GXi<j
κg
ij[(h−1)r
and
[i]+[j](h−1
ǫ[r]+[s])h−1
[i]+[j](Xl≤k
[(h−1)r
l (h−1)s
k − qsr(h−1)s
l (h−1)r
k]ǫ[k]+[l])h−1
ij g)[(h−1)r
i (h−1)s
j − qsr(h−1)s
i (h−1)r
j ]h−1
i (h−1)s
j − qsr(h−1)s
i (h−1)r
j ]hgh−1
ǫ2[r])h−1
(h−1)r
l (h−1)r
kǫ[k]+[l])h−1
(κg
ijg)ǫ∗
[i]+[j](h−1
(κg
ijg)ǫ∗
[i]+[j](Xl≤k
κg
ij(h−1)r
i (h−1)r
j hgh−1.
12
L. GRIMLEY AND C. UHL
In order for hη = η, we must have, for all h, g ∈ G and r < s,
j ] = κhgh−1
(4.10)
κg
ij [(h−1)r
i (h−1)s
i (h−1)r
rs
j − qsr(h−1)s
Xi<j
(h−1)r
i (h−1)r
j κg
ij = 0.
and Xi<j
Now that we have the necessary information for determining Hochschild 2-cocycles of
Λq(V ) ⋊ G, we can characterize those that are truncated quantum Drinfeld Hecke algebras.
Theorem 4.11. If the G action on V extends to an action on Λq(V ), then each constant
Hochschild 2-cocycle of Λq(V )⋊G that sends ǫ2[i] 7→ 0 for all i ∈ {1, 2, ..., n} produces a truncated
quantum Drinfeld Hecke algebra.
Proof. Let
be a constant 2-cocycle.
η = Xg∈G X1≤r≤s≤n
(κg
rsg)ǫ∗
[r]+[s]
We compare the conditions of 2-cocycles from this section to the conditions on truncated
quantum Drinfeld Hecke algebras given in Theorem 2.8. To compare the two sections, we set
κg
ij = κg(vj, vi), as suggested by the proof of Theorem 3.2. Theorem 2.8 condition (i) holds as a
result of Lemma 4.5. The requirement that cocycles send ǫ2[i] 7→ 0 for all i ∈ {1, 2, ..., n}, forces
κg
ii = 0 ∀i ∈ {1, 2, ..., n} by definition. Therefore κ is a quantum 2-form and condition (ii) is
met.
ij = κg(vj , vi), to compare to Theorem 2.8, we should first substitute κg
Because we set κg
ji in all cases containing such terms for i 6= j. We describe all other manipulations
with −qjiκg
necessary to match the conditions in Theorem 2.8 precisely below.
ij
As a result of κg
ii being 0 for all i ∈ {1, 2, ..., n}, equations 4.7 and 4.8 become
κg
jk(vj + qkj
gvj) = 0 and κg
ij(qjivj + gvj) = 0
respectively for i < j < k. Multiply the first equation by −q2
to match the first and second conditions of (v).
jk and the second equation by −q2
If we multiply equation (4.6) by qij qikqjk, then (4.6) becomes condition (iii). Condition (iv)
is the first condition in equation (4.10), after rewriting the scalar using Lemma 4.5. The first
condition of (vi) is met by the second condition in equation (4.10), after using the relation
−qijgr
(cid:3)
j from Lemma 4.5 to rewrite the scalar.
j = gr
i gr
i gr
ij
5. Basic examples
We conclude this paper by providing examples that show the range of the class of truncated
quantum Drinfeld Hecke algebras.
5.1. Diagonal actions. We begin by comparing the PBW and Hochschild cohomological con-
ditions for group actions that are strictly diagonal.
In this case, Hochschild cohomology is
known.
Theorem 5.1 ([6], Theorem 3.3). Assume G acts diagonally on Λq(V ) by gvi = gi
i ∈ K∗, then HHm(Λq(V ), Λq(V ) ⋊ G) is isomorphic to
gi
ivi for some
Mg∈G Mβ∈Nn
β=m Mα∈{0,1}n
β−α∈Cg
SpanK{(vαg)ǫ∗
β}
where Cg = {γ ∈ (Z≥−1)n∀ i, γi = −1 or (−1)γiQj6=i qγj
is isomorphic to the G-invariant subspace of HHm(Λq(V ), Λq(V ) ⋊ G).
ij = gi
i}. Moreover, HHm(Λq(V ) ⋊ G)
TRUNCATED QUANTUM DRINFELD HECKE ALGEBRAS
13
The curious reader should compare this theorem to the results of Naidu, Shroff, and Wither-
spoon [9, Theorem 4.1] in the non-truncated setting.
ij = gi
With this result in mind, we can systematically determine the n-tuples γ = β − α satisfying
γi = −1 or (−1)γiQj6=i qγj
i for all i ∈ {1, 2, ..., n} such that β = 2. We can simplify this
expression further to isolate those which result in a truncated quantum Drinfeld Hecke algebra.
Corollary 5.2. The subspace of HHm(Λq(V ), Λq(V )⋊G) consisting of constant 2-cocycles which
send vi ⊗ vi 7→ 0 is isomorphic to
SpanK{(g)ǫ∗
[i]+[j]} Mi s.t.
∀j,k6=i
qij qik=gi
i
Mg∈G Mi s.t.
∀j6=i
−qij =gi
i
SpanK{(g)ǫ∗
[j]+[k]}.
Using Corollary 5.2, we can check fewer relations to generate truncated quantum Drin-
In the examples below, we include the full description of Hochschild 2-
feld Hecke algebras.
cohomology, given by Theorem 5.1, for a more complete description of possible deformations.
Recall that the translation between Hochschild cohomology and truncated quantum Drinfeld
Hecke algebras is given by Pg∈G κg(vj , vi)g = µ1(ǫ[i]+[j]) for µ1 a 2-cocycle and i < j.
We start with an example of a truncated quantum Drinfeld Hecke algebra where the bound
on the dimension of the parameter space is met. See Section 2 for details on the parameter
space.
Hq,κ,1 is a truncated quantum Drinfeld Hecke algebra where
0 −1(cid:17) ,(cid:16) 1 0 0
0 0 1(cid:17)o = {g, I}. Let q12 = −1 and q13 = q23 =
Hq,κ,1 is generated by
0
0 −1 0
0
0 1 0
Example 5.3. Let G = n(cid:16) −1 0
1. Then
v1, v2, v3 and h ∈ G with relations
v2v1 = −v1v2 + m1I,
v3v1 = v1v3 + m2g,
v3v2 = v2v3 + m3g,
v2
i = 0 for i = 1, 2, 3
where m1, m2, m3 ∈ K and the dimension of the parameter space is 3 =(cid:0)3
to the G-invariant subspace of 2-cocycles for this example which is
0,1,1}.
SpanK{(I)ǫ∗
1,1,0, (g)ǫ∗
1,0,1, (g)ǫ∗
2(cid:1). Compare the above
We include a non-truncated example of a quantum Drinfeld Hecke algebra, given in [20], to
contrast with the truncated example immediately following it.
Example 5.4. Let g =(cid:16) −ω2 0 0
3 and
V = C3. Then Hq,κ is a quantum Drinfeld Hecke algebra where Hq,κ is generated by v1, v2, v3
and h ∈ G with relations
0 1(cid:17), G = hgi and q12 = −1, q23 = ω = q31 where ω = e
0 −ω 0
0
2πi
v2v1 = −v1v2 + m0I + m1g + m2g2 + m3g3 + m4g4 + m5g5,
v3v2 = ωv2v3,
v3v1 = ω2v1v3,
and mi ∈ C. Here we have that κh(v1, v2) is arbitrary for all h ∈ G and that dim(PG) = 6 = G.
2 6= −q21 as would be required in the truncated
Note that κg(v1, v2) 6= 0, but g1
1 6= −q12 and g2
version.
14
L. GRIMLEY AND C. UHL
Example 5.5. As in the above example, let g = (cid:16) −ω2 0 0
q12 = −1, q23 = ω = q31 where ω = e
is a truncated quantum Drinfeld Hecke algebra where
with relations
0 1(cid:17), G = hgi and quantum scalars
Hq,κ,1
Hq,κ,1 is generated by v1, v2, v3 and h ∈ G
3 and V = C3. In the truncated setting, we have
0 −ω 0
0
2πi
v2v1 = −v1v2 + m0I,
v3v2 = ωv2v3,
v3v1 = ω2v1v3,
v2
i = 0 for i = 1, 2, 3
where m0 ∈ C. The dimension of the parameter space is 1. Compare this to the G-invariant
subspace of 2-cocycles for this example which is
0,0,2, (g2)ǫ∗
0,0,2, (v1v2g3)ǫ∗
0,0,2, (v2g3)ǫ∗
2,0,0,
SpanK{(v1v2g2)ǫ∗
(v1g3)ǫ∗
(v1v3I)ǫ∗
0,2,0, (v1v2g4)ǫ∗
1,0,1, (v2v3I)ǫ∗
0,0,2, (v1v2I)ǫ∗
0,1,1, (I)ǫ∗
0,0,2, (v1v2I)ǫ∗
1,1,0, and (v1v2g5)ǫ∗
1,1,0,
0,0,2}.
Of these, the only constant generators are (I)ǫ∗
dition that cocycles must send v1 ⊗ vi → 0 excludes the latter cocycle.
1,1,0 and (g2)ǫ∗
0,0,2. However, including the con-
Note the Hochschild cohomology in this case gave rise to other deformations that are not
truncated quantum Drinfeld Hecke algebras.
Comparing this example to the previous, we see that in the non-truncated setting there is
some extra freedom in the dimension of the parameter space. Specifically, the dimension of the
parameter space in the non-truncated setting is not bound by (cid:0)n
setting there is more freedom on the groups that we can consider as shown in Example 5.10.
2(cid:1). However, in the truncated
We conclude the diagonal action section with an example with a nontrivial parameter on a
non-identity group element.
Example 5.6. Let g = (cid:16) ω 0 0
ω = e
generated by v1, v2, v3 and h ∈ G with relations
3 and V = C3.
0 ω2 0
2πi
0 0 1(cid:17), G = hgi, and quantum scalars q12 = −ω = q23 = q31 where
Hq,κ,1 is a truncated quantum Drinfeld Hecke algebra where
Hq,κ,1 is
v2v1 = −ωv1v2 + m0g,
v3v1 = −ω2v1v3,
v3v2 = −ωv2v3,
v2
i = 0 for i = 1, 2, 3
where m0 ∈ C. The dimension of the parameter space is 1. Note that when attempting the
homological computation, we get the system of equations for g-acting
(−1)γ1 (−ω)γ2(−ω2)γ3 = ω
(−1)γ2(−ω2)γ1(−ω)γ3 = ω2
(−1)γ3(−ω)γ1(−ω2)γ2 = 1
where γ is the difference in homological and polynomial degree. This gives us the eligible repre-
sentatives are
(v1v2g)ǫ∗
(v1v2I)ǫ∗
0,0,2, (g)ǫ∗
1,1,0, (v1v3I)ǫ∗
1,1,0, (g)ǫ∗
0,0,2, (v1g2)ǫ∗
1,0,1, and (v2v3I)ǫ∗
0,2,0
0,1,1.
TRUNCATED QUANTUM DRINFELD HECKE ALGEBRAS
15
The constant terms are (g)ǫ∗
again agrees with the conditions above.
1,1,0, (g)ǫ∗
0,0,2 and the only one for which vi⊗vi 7→ 0 is (g)ǫ∗
1,1,0 which
5.2. Complex reflection groups. We consider the infinite family of complex reflection groups,
G(r, p, n) described by G.C. Shephard and J.A. Todd [12] to compare with the results of Naidu
and Witherspoon [10] who consider them in the non-truncated setting. The group G(r, p, n) is
the finite group of n × n monomial matrices, whose nonzero entries are rth roots of unity and
the product of nonzero entries is a r
Example 5.7 (Symmetric Group). Consider S3 acting on V = C3 and q12 = q13 = q23 = −1.
Then Hq,κ,1 is a truncated quantum Drinfeld Hecke algebra generated by v1, v2, v3, and h ∈ S3
with relations
p root of unity for a fixed r, p, n ∈ Z where p r.
v2v1 = −v1v2 + mI,
v3v1 = −v1v3 + mI,
v3v2 = −v2v3 + mI,
v2
i = 0 for i = 1, 2, 3
where m ∈ C and I is the identity of the group, is a truncated quantum Drinfeld Hecke algebra.
The dimension of the parameter space is 1.
Theorem 5.8. Let q = {−1}. For the infinite family of complex reflection groups G(r, p, n), the
only nontrivial truncated quantum Drinfeld Hecke algebras are G(1, 1, n) = Sn whose parameter
space has dimension 1 and G(2, 2, 2) whose parameter space also has dimension 1.
Proof. Proposition 2.10 focuses our attention to group elements with diagonal action. When
r > 1 and n 6= 2, it is easy to see that condition (iv) of Theorem 2.8 forces κ ≡ 0. For G(2, 2, 2),
the identity of the group supports the parameter space.
Theorem 5.9. Let q = {1}. For the infinite family of complex reflection groups G(r, p, n) there
are no nontrivial truncated quantum Drinfeld Hecke algebras.
(cid:3)
Proof. For n > 2, the only group elements that could support the parameter space would be
elements that act diagonally on the vector space, sending two vectors to their negatives and
the rest to themselves. The presence of the permutation (1 2 3) ∈ G(r, p, n) forces κ ≡ 0.
(Conjugating the group element by (1 2 3) results in another group element with diagonal action
that commutes with original element and that in addition to condition (iv) from Theorem 2.8
forces the dimension of the parameter space to be zero.) For n = 2, either there is not an
element that satisfies Corollary 2.11 or there is, but condition (iv) is violated by an element that
commutes with that element.
(cid:3)
5.3. A non-truncated truncated quantum Drinfeld Hecke algebra. As shown in Section
3, truncated quantum Drinfeld Hecke algebras only need the group to act on Λq(V ) by auto-
morphisms and not Sq′(V ) as is needed in the non-truncated case. To compare actions on the
two algebras, we must now restrict q such that qii = 1 for all i ∈ {1, 2, ..., n}. We define
Sq′(V ) = Khv1, v2, ..., vnvjvi = qij vivj for i, j ∈ {1, 2, ..., n}i.
Note Sq′(V ) is the quantum polynomial algebra S−q(V ), associated with the quantum scalar
set −q and consistent with our definition of Λq(V ).
We now provide an example of a group acting on Λq(V ) by automorphism, which does not
act on Sq′ (V ) by automorphisms. Thus, by considering the truncated case, we actually have
some extra freedom that we do not have in the non-truncated case. This also gives an example
of a nonmonomial group.
16
L. GRIMLEY AND C. UHL
Example 5.10. Let g = √1−η3
η2
0
all i 6= j where η = e
quantum Drinfeld Hecke algebra where
2πi
5 and V = C3 in the truncated setting, we have
η
−√1−η3 0
0
0
1!, G = hgi, and quantum scalars qij = −1, for
Hq,κ,1 is a truncated
Hq,κ,1 is generated by v1, v2, v3 and h ∈ G with relations
v2v1 = −v1v2,
v3v1 = −v1v3 + mI,
v3v2 = −v2v3 + η3(1 −p1 − η3)mI,
v2
i = 0 for i = 1, 2, 3
where m ∈ C and I is the identity of the group. The dimension of the parameter space is 1.
Note that G does not act on Sq′ (V ) by automorphisms as qij = −1 for all i 6= j and G is not
a monomial matrix group as required by [7, Theorem 11.6]. However, by changing the quantum
scalars to q12 = 1 and q23 = −1 = q31, this group would act on Sq′(V ) by automorphisms and
also produces a nontrivial quantum Drinfeld Hecke algebra in that setting (see [20]).
References
[1] P. A. Bergh and S. Oppermann, Cohomology of twisted tensor products, J. Algebra 320 (2008), 3327 -- 3338.
[2] G. Bergman, The Diamond Lemma for Ring Theory, Advances in Mathematics 29 (1978), 178 -- 218.
[3] R.O. Buchweitz, E. L. Green, D. Madsen, and Ø. Solberg, Finite Hochschild cohomology without finite
global dimension, Math. Research Letters 12 (2005), 805 -- 816.
[4] V. G. Drinfel'd, Degenerate affine Hecke algebras and Yangians, Funktsional. Anal. i Prilozhen, (1986).
[5] B. Foster-Greenwood and C. Kriloff, Drinfeld orbifold algebras for symmetric groups J. Algebra 491 (2017),
573 -- 610.
[6] L. Grimley, Hochschild cohomology
of
group extensions of
quantum complete
intersections,
arXiv:1606.01727 (2016).
[7] V. Levandovskyy and A. Shepler, Quantum Drinfeld Hecke algebras, Canad. J. Math. 66 (2014), no. 4,
874 -- 901.
[8] M. Lorenz, On the homology of graded algebras, Comm. Algebra 20 (1992), 489 -- 507.
[9] D. Naidu, P. Shroff, S. Witherspoon, Hochschild cohomology of group extensions of quantum symmetric
algebras, Proc. Amer. Math. Soc. 139 (2011) 1553 -- 1567
[10] D. Naidu and S. Witherspoon, Hochschild cohomology and quantum Drinfeld Hecke algebras,
arXiv:1111.5243 (2014), to appear in Selecta Math.
[11] J. Shakalli, Deformations of quantum symmetric algebras extended by groups, arXiv:1204.2042.
[12] G. C. Shephard and J. A. Todd, Finite unitary reflection groups Canadian Journal of Mathematics
[13] A. Shepler and S. Witherspoon, Poincare-Birkhoff-Witt Theorems, MSRI Proceedings Commutative Alge-
bra and Noncommutative Algebraic Geometry, Vol. 1, ed. D. Eisenbud, S. B. Iyengar, A. K. Singh, J. T.
Stafford, and M. Van den Bergh, Cambridge Univ. Press (2015).
[14] A. Shepler and S. Witherspoon, Drinfeld orbifold algebras, Pacific J. Math. 259 (2012), no. 1, 161 -- 193.
[15] A. Shepler and S. Witherspoon, Hochschild cohomology and graded Hecke algebras, Trans. Amer. Math.
Soc. 360 (2003), no. 8, 3975 -- 4005.
[16] A. Shepler and S. Witherspoon, Group actions on algebras and the graded Lie structure of Hochschild
cohomology, J. Algebra 351 (2012), 350 -- 381.
[17] P. Shroff, Quantum Drinfeld orbifold algebras, Comm. in Algebra 43 (2015), 1563 -- 1570.
[18] P. Shroff and S. Witherspoon, PBW deformations of quantum symmetric algebras and their group exten-
sions, to appear in J. Algebra Appl., arXiv:1410.8212.
[19] D. S¸tefan, Hochschild cohomology on Hopf Galois extensions, J. Pure Appl. Algebra 103 (1995), 221 -- 233.
[20] C. Uhl, Quantum Drinfeld Hecke Algebras, dissertation; Denton, Texas, (August 2016).
[21] S. Witherspoon, Twisted graded Hecke algebras, J. Algebra 317 (2007), 30 -- 42.
TRUNCATED QUANTUM DRINFELD HECKE ALGEBRAS
17
Lauren Grimley, Mathematics Department, Spring Hill College, Mobile, AL 36608, USA
E-mail address: [email protected]
Christine Uhl, Department of Mathematics, St. Bonaventure University, St Bonaventure, NY
14778
E-mail address: [email protected]
|
1608.06687 | 1 | 1608 | 2016-08-24T02:02:39 | Pure resolutions of unbounded complexes of modules | [
"math.RA",
"math.AC"
] | Let $R$ be a commutative ring. We show that pure injective resolutions and pure projective resolutions can be constructed for unbounded complexes of $R$-modules. We use these to obtain a closed symmetric monoidal structure on the unbounded pure derived category. | math.RA | math | Pure resolutions of unbounded complexes of modules
Abhishek Banerjee
g
u
A
4
2
]
.
A
R
h
t
a
m
[
1
v
7
8
6
6
0
.
8
0
6
1
:
v
i
X
r
a
Dept. of Mathematics, Indian Institute of Science, Bangalore-560012, India.
Email: [email protected]
Let R be a commutative ring. We show that pure injective resolutions and pure projective
resolutions can be constructed for unbounded complexes of R-modules. We use these to obtain
a closed symmetric monoidal structure on the unbounded pure derived category.
Abstract
MSC (2010) Subject Classification: 13D09, 16E35, 18D10.
Keywords: Pure resolutions, pure projectives, pure injectives, pure acyclic.
1
Introduction
Given a ring R, there are two well known exact structures (in the sense of Quillen [16]) on the
category of R-modules: the usual exact structure and the pure exact structure (see [22]). The
usual derived category D(R), which is constructed by inverting quasi-isomorphisms in the homotopy
category of R-modules, has been studied extensively in homological algebra. Additionally, there
has been recent interest by several authors (see, for instance, [3], [5], [8], [15], [18], [22]) in studying
the pure exact structure on the category of R-modules. This raises the question of which of the
properties of the usual derived category D(R) can also be extended to the "pure derived category"
Dpur(R), which is obtained by inverting pure quasi-isomorphisms in the homotopy category of R-
modules. When R is a commutative ring, the purpose of this paper is to construct pure injective
and pure projective resolutions of complexes of R-modules. We then use these resolutions to exhibit
a closed symmetric monoidal structure on the category Dpur(R).
It is important to note that in this article, we work with unbounded complexes of R-modules
and consider the unbounded pure derived category Dpur(R). The question behind this paper was
motivated naturally by reading the recent work of Zheng and Huang [22], where the authors study
pure resolutions of bounded complexes. Further, the authors in [22] have also shown that any
bounded above (resp. bounded below) complex of R-modules admits a pure injective (resp. pure
projective) resolution. However, the general question of pure resolutions for arbitrary unbounded
1
complexes is still left open in [22]. Since our methods require us to have a closed symmetric monoidal
structure on the category of R-modules, we will limit ourselves to commutative rings. It should
be mentioned here that even in the case of the classical derived category D(R), the construction
of projective and injective resolutions of arbitrary unbounded complexes presents some difficulties.
In [19], Spaltenstein showed the existence of projective and injective resolutions for unbounded
complexes, which allowed him to remove the boundedness conditions for the existence of certain
derived functors of functors such as Hom and tensor product. The derived Hom and derived
tensor product functors on the unbounded derived category were also constructed by Bokstedt and
Neeman in [2], where the authors used the method of homotopy colimits. For more on resolutions
of unbounded complexes, the reader may also see the work of Alonso Tarr´ıo et al. [1] and Serp´e [18].
Our methods in this paper are a combination of the classical method of constructing resolutions
of bounded complexes (see, for instance, [12]) along with the techniques of Spaltenstein [19] for
treating arbitrary unbounded complexes.
We now describe the structure of the paper in detail. In Section 2, we briefly recollect the notions of
pure acyclic complexes, pure projective modules, pure injective modules, pure quasi-isomorphisms
and pure resolutions that we will need in the paper.
In Section 3, we show how to construct
pure injective resolutions for unbounded complexes of R-modules. Thereafter, pure projective
resolutions of unbounded complexes are constructed in Section 4. It should be noted that due to
the fact that tensoring preserves cokernels but not kernels, we will need to somewhat adjust our
methods in Section 4, i.e., our arguments for pure projective resolutions are not exactly the dual
of our arguments for pure injective resolutions. Finally, in Section 5, we use pure projective and
pure injective resolutions to obtain a "pure derived Hom functor":
P Hom•
R( ,
) : Dpur(R)op × Dpur(R) −→ Dpur(R)
(1.1)
Then, in the case of a commutative ring R, (1.1) removes the boundedness condition for the
existence of the pure derived Hom functor in [22, § 4]. We conclude by showing that we have
natural isomorphisms:
P Hom•
R((A• ⊗R B•)•, C •) ∼= P Hom•
R(A•, P Hom•
R(B•, C •))
HomDpur(R)((A• ⊗R B•)•, C •) ∼= HomDpur (A•, P Hom•
R(B•, C •))
(1.2)
for any complexes A•, B• and C • of R-modules, thus giving the unbounded pure derived category
Dpur(R) the structure of a closed symmetric monoidal category.
Acknowledgements: The author is grateful for the hospitality of the Academy of Mathematics
and Systems Science at the Chinese Academy of Sciences in Beijing, where some of this paper was
written.
2 Pure acyclic complexes
In this section, we will briefly recall some facts on pure acyclic complexes, pure projectives, pure
injectives and pure resolutions that we will use in the rest of the paper. Throughout, we let R be
2
a commutative ring with identity. We denote by R − M od the category of R-modules. We will
denote by C(R) the category of cochain complexes:
(M •, d•
M ) :
. . . −−−−→ M i−1
di−1
M−−−−→ M i
di
M−−−−→ M i+1 −−−−→ . . .
(2.1)
of R-modules and by K(R) the homotopy category of R − M od. By abuse of notation, we will
usually denote an object (M •, d•
M ) ∈ C(R) simply by M •. For any n ∈ Z, we let M [n]• denote
the shifted cochain complex given by M [n]i := M i+n with differential di
M . The
unbounded derived category of R-modules is denoted by D(R). A cochain complex M • is said to
be bounded above (resp. bounded below) if M i vanishes for all i sufficiently large (resp.
for all
i sufficiently small). Given complexes (M •, d•
N ) ∈ C(R), we have an internal Hom
R(M •, N •), d•) ∈ C(R) given by:
object (Hom•
R(M •, N •) := Qj∈Z
HomR(M j, N i+j) ∀ i ∈ Z
M [n] = (−1)ndi+n
M ) and (N •, d•
Homi
di(f ) := dN ◦ f − (−1)if ◦ dM ∀ f ∈ Homi
R(M •, N •)
We now recall the following definitions.
Definition 2.1. (see [21]) (a) A monomorphism f : M ′ −→ M in R − M od is said to be pure
if the induced morphism N ⊗R f : N ⊗R M ′ −→ N ⊗R M is a monomorphism for each module
N ∈ R − M od.
(b) A short exact sequence :
0 −→ M ′ −→ M −→ M ′′ −→ 0
in R − M od is said to be pure acyclic if the induced sequence
0 −→ N ⊗R M ′ −→ N ⊗R M −→ N ⊗R M ′′ −→ 0
(2.2)
(2.3)
is acyclic for each module N ∈ R − M od. In such a situation, the morphism M −→ M ′′ is said to
be a pure epimorphism.
(c) (see [7]) More generally, a complex M • ∈ C(R) is said to be pure acyclic if the induced sequence
N ⊗R M • is acyclic for each module N ∈ R−M od. This is equivalent to the complex Hom•
R(F, M •)
being acyclic for each finitely presented R-module F .
(d) An acyclic complex (M •, d•
0 −→ Im(dn−1
M ) ∈ C(R) is said to be pure acyclic at some given n ∈ Z if
M ) = Ker(dn
M ) −→ M n −→ Coker(dn−1
M ) −→ 0
(2.4)
is a pure short exact sequence in the sense of (b). The complex (M •, d•
in the sense of (c) if and only if it is pure acyclic at each n ∈ Z.
M ) ∈ C(R) is pure acyclic
Given a morphism f • : M • −→ N •, its mapping cone C •
M [1]• ⊕ N • with differential d•
Cf
given by
f is taken to be the complex C •
f :=
di
Cf = (cid:18)−di+1
M
f i+1
0
di
N(cid:19) : M i+1 ⊕ N i −→ M i+2 ⊕ N i+1
(2.5)
3
We notice that the canonical projections pi : M i⊕N i−1 −→ M i determine a morphism of complexes
from C •
f to M [1]•.
Definition 2.2. (see [22, Definition 2.7]) A morphism f • : M • −→ N • in C(R) is a pure quasi-
isomorphism if its cone C •
f is pure acyclic.
Equivalently, f • is a pure quasi-isomorphism if M ′ ⊗R f • : M ′ ⊗R M • −→ M ′ ⊗R N • is a quasi-
isomorphism for each module M ′ ∈ R − M od.
Definition 2.3. (see [21]) A module P ∈ R − M od is pure projective if the functor Hom•
)
carries pure acyclic complexes to pure acyclic complexes. Similarly, a module I ∈ R − M od is pure
injective if the functor Hom•
R( , I) preserves pure acyclic complexes.
R(P,
The category of pure projectives in R−M od will be denoted by PP and the category of pure injectives
in R − M od by PI.
We mention here (see [22, Remark 2.6]) that a complex M • ∈ C(R) is pure acyclic if and only if
Hom•
R(M •, I)
being acyclic for any pure injective I ∈ PI.
R(P, M •) is acyclic for any pure projective P ∈ PP. This is also equivalent to Hom•
On the other hand, (see [22, Remark 2.8]) a morphism f • : M • −→ N • of complexes is a pure quasi-
isomorphism if and only if Hom•
R(f •, I) :
R(M •, I) ) is a quasi-isomorphism for each P ∈ PP (resp. for each I ∈ PI).
Hom•
R(P, N •) (resp. Hom•
R(N •, I) −→ Hom•
R(P, f •) : Hom•
R(P, M •) −→ Hom•
Definition 2.4. (see [22]) (a) Let M • ∈ C(R). A morphism f • : P • −→ M • is said to be a pure
projective resolution of M • if it satisfies the following conditions:
(i) P • is a complex of pure projective modules and f • is a pure quasi-isomorphism.
(ii) The functor Hom•
) preserves pure acyclic complexes.
R(P •,
(b) Dually, a morphism g• : M • −→ I • is said to be a pure injective resolution of M • if it satisfies:
(i) I • is a complex of pure injective modules and g• is a pure quasi-isomorphism.
(ii) The functor Hom•
R( , I •) preserves pure acyclic complexes.
Given an R-module M , it is known (see [21, Corollary 6]) that there exists a pure injective module
I ∈ PI and a pure monomorphism M ֒→ I. In fact, this holds more generally in any locally finitely
presented additive category (see Herzog [9]). The aim of this paper is to show that any unbounded
complex of R-modules admits a "pure projective resolution" and a "pure injective resolution". In
[22], Zheng and Huang have already shown that any bounded above (resp. bounded below) complex
in C(R) admits a pure injective resolution (resp. a pure projective resolution) . The proof of Zheng
and Huang in [22] uses the techniques of homotopy (co)limits due to Bokstedt and Neeman [2] (see
also [14]). However, our proof will use a combination of the classical technique for constructing
resolutions along with the methods of Spaltenstein [19] for treating arbitrary unbounded complexes.
4
3 Pure injective resolutions of unbounded complexes
In this section, we will need one more concept, that of a "K-pure injective complex", which will be
analogous to the classical notion of a K-injective complex (see, for instance, [19, § 1]). Although
this notion already appears implicitly in Definition 2.4, we propose the following definition, which
does not seem to have appeared explicitly before in the literature.
Definition 3.1. We will say that a cochain complex M • ∈ C(R) is K- pure injective if the functor
Hom•
R( , M •) : C(R) −→ C(R) takes pure acyclic complexes to pure acyclic complexes.
Proposition 3.2. Let M • ∈ C(R). Then, the following statements are equivalent:
(a) M • is a K-pure injective complex.
(b) For any pure acyclic complex A• ∈ C(R), we have HomK(R)(A•, M •) = 0.
Proof. (a) ⇒ (b): For any two complexes B•, C • ∈ C(R), it is well known (see, for instance, [19,
§0.4]) that
H k(Hom•
R(B•, C •)) = HomK(R)(B•, C[k]•)
∀ k ∈ Z
(3.1)
Let A• ∈ Ch(R) be pure acyclic. Since M • is K-pure injective, it follows that Hom•
pure acyclic and, in particular, acyclic. Then, HomK(R)(A•, M •) = H 0(Hom•
(b) ⇒ (a): Shifting the indices in (3.1), we obtain:
R(A•, M •)) = 0.
R(A•, M •) is
H k(Hom•
R(B•, C •)) = HomK(R)(B[−k]•, C •)
∀ k ∈ Z
(3.2)
for any two complexes B•, C • ∈ C(R). Now let M • ∈ C(R) be such that HomK(R)(A•, M •) = 0 for
every pure acyclic A• ∈ C(R). We have to show that Hom•
R(A•, M •) is pure acyclic, or equivalently
that Hom•
R(F ⊗R A•, M •) is acyclic for every finitely presented R-
module F . If A• is pure acyclic, it is immediate from Definition 2.1 that (F ⊗R A•)[−k]• is also
pure acyclic for any k ∈ Z. It now follows from (3.2) that:
R(A•, M •)) ∼= Hom•
R(F, Hom•
H k(Hom•
R(F ⊗R A•, M •)) = HomK(R)((F ⊗R A•)[−k]•, M •) = 0
∀ k ∈ Z
(3.3)
This proves the result.
Let Kpac(R) denote the full subcategory of K(R) consisting of complexes that are also pure acyclic.
The cone of a morphism of pure acyclic complexes is still pure acyclic and hence Kpac(R) is a
triangulated subcategory. Then, since Kpac(R) is closed under direct summands, it follows from
Rickard's criterion [17, Proposition 1.3] (see also [13]) that Kpac(R) is a thick subcategory. We
then consider the "pure derived category" Dpur(R) as in [22, §3] given by the Verdier quotient:
Dpur(R) := K(R)/Kpac(R)
(3.4)
5
Consider a "right roof" in K(R) given by a pair of morphisms (f •, u•) of the form:
A• f •
−→ C • u•
⇐= B•
(3.5)
where u• is a pure quasi-isomorphism. Then, the morphisms in Dpur(R) from A• to B• can be
given by equivalence classes of right roofs as in (3.5) (see [6] for details). We will now characterize
K-pure injective complexes in terms of the pure derived category Dpur(R).
Proposition 3.3. Let M • ∈ C(R). Then, the following statements are equivalent:
(a) M • is a K-pure injective complex.
(b) For any complex A• ∈ C(R), we have HomK(R)(A•, M •) ∼= HomDpur(R)(A•, M •).
Proof. (b) ⇒ (a): Let A• ∈ C(R) be pure acyclic. Then, A• = 0 in the pure derived category
Dpur(R). Hence, HomK(R)(A•, M •) ∼= HomDpur(R)(A•, M •) = 0 and it follows from Proposition
3.2 that M • is K-pure injective.
(a) ⇒ (b): Suppose that M • is a K-pure injective complex. We first show that if u• : M • −→ N •
is a pure quasi-isomorphism, it must have a left inverse up to homotopy, i.e., we must have some
v• : N • −→ M • such that v• ◦ u• ∼ 1.
u denotes the cone of u•, applying the functor
HomK(R)( , M •) to the distinguished triangle M • u•
−→ N • −→ C •
u gives an induced exact sequence:
If C •
HomK(R)(C •
u, M •) → HomK(R)(N •, M •) → HomK(R)(M •, M •) → HomK(R)(Cu[−1]•, M •)
Since u• is a pure quasi-isomorphism, its cone C •
u is pure acyclic and M • being K-pure injec-
tive, we get HomK(R)(C •
u, M •) = HomK(R)(Cu[−1]•, M •) = 0. This gives us an isomorphism
∼=−→ HomK(R)(M •, M •). Choosing 1 ∈ HomK(R)(M •, M •),
HomK(R)(u•, M •) : HomK(R)(N •, M •)
it follows that there exists v• : N • −→ M • such that v• ◦ u• ∼ 1. Now, given a morphism
(f •, u•) ∈ HomDpur(R)(A•, M •) having the form of a right roof
A• f •
−→ N • u•
⇐= M •
(3.6)
we associate (f •, u•) ∈ HomDpur(R)(A•, M •) to v• ◦ f • ∈ HomK(R)(A•, M •). Conversely, any
morphism g• ∈ HomK(R)(A•, M •) is associated to the roof (g•, 1) ∈ HomDpur(R)(A•, M •). Since
the right roofs (v• ◦ f •, 1), (f •, u•) are equivalent in HomDpur(R)(A•, M •), it is clear that these two
associations are inverse to each other. Hence, we have HomK(R)(A•, M •) ∼= HomDpur(R)(A•, M •).
Proposition 3.4. Let (I •, d•
I • is K-pure injective.
I ) ∈ C(R) be a bounded below complex of pure injective modules. Then,
6
Proof. For the sake of definiteness, we suppose that I j = 0 for every j < 0 and I 0 6= 0. Using
Proposition 3.2, it suffices to show that HomK(R)(A•, I •) = 0 for any pure acyclic (A•, d•
A) ∈ C(R).
We will show that any morphism f : (A•, d•
I ) in C(R) is null-homotopic. For some
given integer K, we suppose that we have constructed maps tk : Ak −→ I k−1 for all integers k ≤ K
such that tk ◦ dk−1
◦ tk−1 = f k−1. This is already true for K = 0. We now note that:
A) −→ (I •, d•
◦(tk ◦dk−1
A ) = (f k ◦dk−1
A −dk−1
I
◦f k−1)+dk−1
I
◦dk−2
I
◦tk−1 = 0
I
I
(f k −dk−1
◦tk)◦dk−1
A + dk−2
A = f k ◦dk−1
Hence, the morphism (f k − dk−1
Im(dk
is pure injective, the morphism Im(dk
satisfies (f k − dk−1
◦ tk) = tk+1 ◦ dk
A −dk−1
I
A). From Definition 2.1, it follows that Im(dk
I
◦ tk) : Ak −→ I k factors through Ak/Im(dk−1
A) ∼=
A) ֒→ Ak+1 is a pure monomorphism. Since I k
A) −→ I k extends to a morphism tk+1 : Ak+1 −→ I k that
A ) = Ak/Ker(dk
I
A. This proves the result.
As mentioned in Section 2, it is well known (see [21, Corollary 6]) that any R-module M admits a
pure monomorphism M ֒→ I into a pure injective module I. We will now show that any bounded
below complex of R-modules admits a pure injective resolution.
Proposition 3.5. Let M • ∈ C(R) be a cochain complex of R-modules that is bounded below. Then,
there exists a pure quasi-isomorphism u• : M • −→ I • such that I • is a bounded below complex of
pure injective modules.
Proof. For the sake of definiteness, we suppose that the complex (M •, d•) satisfies M j = 0 for
each j < 0 but M 0 6= 0. We put I j = 0 for each j < 0. We choose a pure monomorphism
u0 : M 0 ֒→ I 0 with I 0 pure injective. Then, for every i < 1, we have already constructed pure
injective modules I i, morphisms ui : M i −→ I i and differentials ei−1 : I i−1 −→ I i such that we
have induced isomorphisms:
H i−1(M • ⊗R N )
∼
−→ Ker(ei−1 ⊗R N )/Im(ei−2 ⊗R N )
∀ N ∈ R − M od
(3.7)
and monomorphisms:
Coker(di−1 ⊗R N ) ֒→ Coker(ei−1 ⊗R N )
∀ N ∈ R − M od
(3.8)
We suppose that we have already done this for all i ∈ Z less than some given integer k ≥ 1. We
will now show that we can choose a pure injective I k, a morphism uk : M k −→ I k and a differential
ek−1 : I k−1 −→ I k such that:
H k−1(M • ⊗R N ) ∼−→ Ker(ek−1 ⊗R N )/Im(ek−2 ⊗R N )
∀ N ∈ R − M od
Coker(dk−1 ⊗R N ) ֒→ Coker(ek−1 ⊗R N )
∀ N ∈ R − M od
(3.9)
For this we consider the colimit C k defined by the following pushout square:
M k−1
dk−1
−−−−→ M k
Coker(ek−2) −−−−→ C k
y
y
7
(3.10)
and choose a pure monomorphism C k ֒→ I k with I k pure injective. Then, for any N ∈ R − M od,
C k ⊗R N −→ I k ⊗R N is a monomorphism and the following square is still a pushout:
M k−1 ⊗R N
dk−1⊗RN
−−−−−−→ M k ⊗R N
Coker(ek−2) ⊗R N = Coker(ek−2 ⊗R N ) −−−−→ C k ⊗R N
y
y
(3.11)
We now define ek−1 : I k−1 −→ I k to be the composition I k−1 −→ Coker(ek−2) −→ C k ֒→ I k
and uk : M k −→ I k to be the composition M k −→ C k −→ I k. Applying the dual of [12, Lemma
68] to the pushout square (3.11) along with the monomorphism C k ⊗R N ֒→ I k ⊗R N gives us a
monomorphism Coker(dk−1 ⊗R N ) ֒→ Coker(ek−1 ⊗R N ).
We now put dj
N := ej ⊗R N for any integer j and any R-module N . Since the
morphisms M k−1 ⊗R N −→ M k ⊗R N and M k−1 ⊗R N −→ Coker(ek−2 ⊗R N ) both factor through
the epimorphism M k−1 ⊗R N −→ (M k−1 ⊗R N )/Im(dk−2
N ), we can simply replace M k−1 ⊗R N by
(M k−1 ⊗R N )/Im(dk−2
N ) in (3.11) and still obtain a pushout square. Now if we let CN be the colimit
of the system Coker(ek−2
N ), we obtain the following
commutative diagram:
N ) ←− (M k−1 ⊗R N )/Im(dk−2
N := dj ⊗R N and ej
N ) −→ Im(dk−1
(M k−1 ⊗R N )/Im(dk−2
N ) = Coker(dk−2
N ) −−−−→ Im(dk−1
N ) −−−−→ M k ⊗R N
y
Coker(ek−2
N )
y
−−−−→ CN
y
−−−−→ C k ⊗R N
(3.12)
N ) gives an epimorphism Coker(ek−2
where all the squares are pushouts. The pushout of the epimorphism (M k−1 ⊗R N )/Im(dk−2
Im(dk−1
abelian category, it follows that the pushout of the monomorphism Im(dk−1
monomorphism CN ֒→ C k ⊗R N . Accordingly, the morphism ek−1
be factored as the epimorphism:
N ) −→
N ) ։ CN . On the other hand, since R − M od is an
N ) ֒→ M k ⊗R N gives a
N : I k−1 ⊗R N −→ I k ⊗R N can
I k−1 ⊗R N ։ (I k−1 ⊗R N )/Im(ek−2
N ) = Coker(ek−2
N ) ։ CN
followed by the monomorphism:
CN ֒→ C k ⊗R N ֒→ I k ⊗R N
(3.13)
(3.14)
From (3.13) and (3.14) it follows that CN ∼= (I k−1 ⊗R N )/Ker(ek−1
we have a monomorphism Coker(dk−2
N ). By assumption,
N ). This gives us the following pushout square:
N ) ֒→ Coker(ek−2
N ) = Im(ek−1
(M k−1 ⊗R N )/Im(dk−2
N ) = Coker(dk−2
N )
epic
−−−−→ Im(dk−1
N ) = (M k−1 ⊗R N )/Ker(dk−1
N )
monicy
(I k−1 ⊗R N )/Im(ek−2
N ) = Coker(ek−2
N )
y
(3.15)
−−−−→ Im(ek−1
N ) = (I k−1 ⊗R N )/Ker(ek−1
N )
epic
8
The fact that (3.15) is a pushout square and that Coker(dk−2
N ) is a monomor-
phism shows that we have an isomorphism of the kernels of the two horizontal morphisms. It is
also clear that the kernels of these horizontal morphisms are H k−1(M • ⊗R N ) and Ker(ek−1 ⊗R
N )/Im(ek−2 ⊗R N ) respectively. Thus, we can construct inductively a pure quasi-isomorphism
M • −→ I • where I • is a bounded below complex of pure injectives.
N ) −→ Coker(ek−2
1 : M •
Proposition 3.6. Let f • : M •
Let u•
pure injectives. Then, there exists a bounded below complex I •
isomorphism u•
1 be a pure quasi-isomorphism from M •
1 be a morphism of bounded below complexes in C(R).
1 of
2 of pure injectives, a pure quasi-
1 to a bounded below complex I •
1 fitting into a commutative diagram:
2 and a morphism g• : I •
2 −→ M •
1 −→ I •
2 −→ I •
2 : M •
2 −→ I •
M •
2
f •y
M •
1
u•
2−−−−→ I •
2
g•y
u•
1−−−−→ I •
1
(3.16)
Proof. It is clear that the mapping cone C •
1 is a bounded
below complex in C(R). Applying Proposition 3.5, we choose a pure quasi-isomorphism v• :
u1f −→ I • to a bounded below complex of pure injectives. Thereafter, we consider the composition
C •
h. We now have the following commutative diagram
v•
−→ I • and its mapping cone C •
u1f of the composition u•
1 ◦ f • : M •
1 −→ C •
2 −→ I •
u1f
h• : I •
in K(R):
u•
1◦f •
M •
2
u•
2
Ch[−1]• g•
I •
1
1
/ I •
1
C •
u1f
v•
/ I •
h•
/ M2[1]•
u2[1]•
/ C •
h
(3.17)
Since the horizontal rows in (3.17) are distinguished triangles, the triangulated structure on K(R)
2 −→ Ch[−1]• making the diagram commute. It is clear
implies that we have a morphism u•
that Ch[−1]• is a bounded below complex of pure injectives and we set I •
2 := Ch[−1]•. Now, for
any module N ∈ R − M od, we have an induced commutative diagram:
2 : M •
N ⊗R M •
2
N ⊗R(u•
1◦f •)
N ⊗R I •
1
N ⊗R C •
u1f
/ (N ⊗R M •
2 )[1]
(3.18)
N ⊗Ru•
2
1
N ⊗Rv•
(N ⊗Ru•
2)[1]
N ⊗R I •
2 = N ⊗R Ch[−1]• N ⊗Rg•
/ N ⊗R I •
1 N ⊗Rh•
/ N ⊗R I •
/ (N ⊗R I •
2 )[1]
Since v• is a pure quasi-isomorphism, N ⊗R v• is a quasi-isomorphism. Since the mapping cone
commutes with the functor N ⊗R , the horizontal rows in (3.18) are still distinguished triangles.
u1f −→ N ⊗R I • are quasi-isomorphisms
Now, since 1 : N ⊗R I •
1 and N ⊗R v• : N ⊗R C •
1 −→ N ⊗R I •
9
/
/
/
/
/
/
/
/
/
/
/
/
/
/
/
/
(and hence isomorphisms in the derived category D(R)), the third morphism N ⊗Ru•
2 −→ I •
N ⊗R I •
isomorphism that fits into the commutative square (3.16).
2 is also a quasi-isomorphism (see [20, Tag 014A]). Hence, u•
2 : M •
2 : N ⊗RM •
2 −→
2 is a pure quasi-
We will now proceed to construct pure injective resolutions for arbitrary, unbounded complexes of
R-modules. We will first need the notion of a "special inverse system".
Definition 3.7. (see [19, Definition 2.1]) Let T be a class of complexes in C(R) that is closed
under isomorphisms.
(a) A T -special inverse system of complexes is an inverse system {I •
satisfying the following conditions for each n ∈ Z:
n}n∈Z of complexes in C(R)
(1) The cochain map I •
(2) The kernel K •
(3) The short exact sequence of complexes:
n := Ker(I •
n −→ I •
n −→ I •
n−1 is surjective.
n−1) lies in the class T .
0 −→ K •
n
i•
n−→ I •
n
p•
n−→ I •
n−1 −→ 0
(3.19)
is "semi-split", i.e., it is split in each degree.
(b) The class T ⊆ C(R) is said to be closed under special inverse limits if the inverse limit of every
T -special inverse system in C(R) is contained in T .
By slight abuse of notation, we will refer to {I •
for all n < 0 makes {I •
n}n∈Z into a T -special inverse system in the sense of Definition 3.7 above.
n}n≥0 as a T -special inverse system if setting I •
n = 0
Proposition 3.8. (a) Let C ⊆ C(R) be a class of complexes. Let T (C) denote the class of complexes
M • ∈ C(R) such that Hom•
R(A•, M •) is pure acyclic for each A• ∈ C. Then, T (C) is closed under
special inverse limits.
(b) The class of all K-pure injective complexes is closed under special inverse limits.
Proof. (a) We begin by setting:
T ′ := {B• ∈ C(R) B• = F ⊗R A• for some finitely presented R-module F and some A• ∈ C}
R(F ⊗R A•, M •) ∼= Hom•
Now since Hom•
R(A•, M •)), a complex M • ∈ T (C) if and only
if Hom•
R(B•, M •) is acyclic for each B• ∈ T ′. It now follows from [19, Corollary 2.5] that T (C) is
closed under special inverse limits. The result of (b) follows directly from (a) by taking C to be the
class of all pure acyclic complexes in C(R).
R(F, Hom•
10
Given a complex (M •, d•) ∈ C(R), we recall that for any n ∈ Z, its truncation τ ≥nM • is given by
setting
(τ ≥nM )i =
M i
Coker(dn−1)
0
if i > n
if i = n
if i < n
(3.20)
Then, it is clear that H i(τ ≥nM •) = H i(M •) for all i ≥ n and H i(τ ≥nM •) = 0 otherwise. Further,
the canonical morphisms τ ≥n−1M • −→ τ ≥nM • can be used to express M • as an inverse limit
M • = lim
←−
n≥0
τ ≥−nM •.
Proposition 3.9. For any complex M • ∈ C(R) there exists a special inverse system {I •
K-pure injective complexes and a morphism {fn : τ ≥−nM • −→ I •
the following conditions:
n}n≥0 of
n}n≥0 of inverse systems satisfying
(a) Each I •
n is a bounded below complex of pure injectives.
(b) Each fn is a pure quasi-isomorphism.
Proof. Using Proposition 3.5, we choose a pure quasi-isomorphism f0 : τ ≥0M • −→ I •
0 with I •
0
a bounded below complex of pure injectives. For some n ≥ 1, we assume that we have already
chosen pure quasi-isomorphisms fj : τ ≥−jM • −→ I •
j for all 0 ≤ j ≤ n − 1 satisfying the required
conditions. We now set:
I • := I •
n−1
N • := τ ≥−nM •
f : N • = τ ≥−nM • −→ τ ≥−n+1M •
fn−1−−−−→ I •
n−1 = I •
(3.21)
f −→ J • with J • a bounded below complex of pure injective modules. Since C •
We let C •
f denote the cone of f . Again using Proposition 3.5, we choose a pure quasi-isomorphism
f = N [1]• ⊕ I •
g : C •
−→ J •
as a Z-graded module, g induces morphisms g′ : N [1]• −→ C •
f
of graded modules and g′′ is actually a morphism of complexes. We rewrite g′ : N [1]• −→ J • as a
morphism g′ : N • −→ J[−1]• and consider the following morphism for each i ∈ Z:
−→ J • and g′′ : I • −→ C •
f
g
g
hi : N i −→ C−g′′[−1]i = I i ⊕ J[−1]i
n 7→ (f (n), g′(n)) = (f (n), g(n, 0))
(3.22)
−g′′ is the cone of −g′′. We claim that h• = {hi}i∈Z is a morphism of complexes. For this,
where C •
we note that for some i ∈ Z and for any n ∈ N i we have:
h ◦ dN (n) = (f ◦ dN (n), g(dN (n), 0))
dC−g′′ [−1] ◦ h(n) = (dI ◦ f (n), g′′ ◦ f (n) − dJ ◦ g(n, 0))
= (dI ◦ f (n), g′′ ◦ f (n) − g ◦ dCf (n, 0))
= (dI ◦ f (n), g′′ ◦ f (n) − g(−dN (n), f (n)))
= (f ◦ dN (n), g(dN (n), 0))
(3.23)
where dN is the differential on N •. As a graded module, it is immediate that the cone C •
satisfies:
h of h•
C •
h = N [1]• ⊕ C−g′′[−1]• = N [1]• ⊕ I • ⊕ J[−1]• = (N [2]• ⊕ I[1]• ⊕ J •)[−1] = C−g[−1]•
(3.24)
11
To show that (3.24) is an isomorphism of complexes, for some given i ∈ Z, we choose any x ∈ N i+1,
y ∈ I i and z ∈ J i−1. Then, we see that:
dCh(x, y, z)
= (−dN (x), h(x) + dC−g′′ [−1](y, z))
= (−dN (x), f (x) + dI (y), g(x, 0) + g′′(y) − dJ (z))
= (−dN (x), f (x) + dI (y), g(x, y) − dJ (z))
dC−g [−1](x, y, z) = −dC−g (x, y, z)
(3.25)
= −(−dCf (x, y), −g(x, y) + dJ (z))
= (dCf (x, y), g(x, y) − dJ (z))
= (−dN (x), f (x) + dI (y), g(x, y) − dJ (z))
∼= C−g[−1]• of complexes. Since g is a pure quasi-
Accordingly, we have an isomorphism C •
h
In other words, h• : N • = τ ≥−nM • −→
isomorphism, it now follows that C •
n := C−g′′[−1]• is
C−g′′[−1]• is a pure quasi-isomorphism. From the definitions, it is clear that I •
a bounded below complex of pure injectives. Finally, the definition in (3.22) makes it clear that
{I •
n}n≥0 is a special inverse system of K-pure injective complexes. This proves the result.
h is pure acyclic.
Lemma 3.10. Let {g•
of complexes of R-modules such that each g•
we can choose a positive integer N (i) such that the morphisms
n }n≥0 be a morphism of inverse systems {X •
n }n≥0
n is a quasi-isomorphism. Suppose that for each i ≥ 0,
n}n≥0 and {Y •
n : X •
n −→ Y •
τ ≥−iX •
n+1 −→ τ ≥−iX •
n
τ ≥−iY •
n+1 −→ τ ≥−iY •
n
(3.26)
are epimorphisms in each degree for each n ≥ N (i). Then, the induced morphism on the inverse
limits g• : X • := lim
←−
n≥0
n −→ Y • := lim
←−
n≥0
n is a quasi-isomorphism.
X •
Y •
It is clear from the definition of the truncations in (3.20)
n −→ Y •
n −→
n of complexes. Further, the transition maps of the inverse systems {τ ≥−iX •
n}n≥N (i) and
n }n≥N (i) are all surjections. It now follows from [11, Corollary 3.11] that the induced mor-
Proof. First we fix some i ≥ 0.
that the quasi-isomorphisms g•
τ ≥−iY •
{τ ≥−iY •
phism
n induce quasi-isomorphisms τ ≥−ig•
n : τ ≥−iX •
n : X •
lim
←−
n≥0
τ ≥−iX •
n = lim
←−
n≥N (i)
τ ≥−iX •
n −→ lim
←−
n≥N (i)
τ ≥−iY •
n = lim
←−
n≥0
τ ≥−iY •
n
is a quasi-isomorphism. From the definition of truncations in (3.20) it is also clear that:
H j(lim
←−
n≥0
τ ≥−iX •
n) = H j(lim
←−
n≥0
X •
n)
H j(lim
←−
n≥0
τ ≥−iY •
n ) = H j(lim
←−
n≥0
Y •
n )
(3.27)
(3.28)
for all integers j ≥ −i + 2. Choosing i to be arbitrarily large, we now see that H j(X •)
for all j ∈ Z, i.e., g• : X • −→ Y • is a quasi-isomorphism.
∼
−→ H j(Y •)
12
We now make one more observation. Let {I (j)}j≥0 be a family of pure injective modules and consider
the product I := Qj≥0 I (j). If F is a finitely presented R-module and A• ∈ C(R) is pure acyclic,
R(F ⊗R A•, I (j)) is
we observe that Hom•
acyclic. It follows that the product I is also pure injective. We now prove the main result of this
section.
R(F ⊗R A•, I) = Qj≥0 Hom•
R(A•, I)) = Hom•
R(F, Hom•
Proposition 3.11. Let R be a commutative ring and let M • ∈ C(R) be a cochain complex of
R-modules. Then, M • has a pure injective resolution.
Proof. We consider the special inverse system {I •
quasi-isomorphisms {fn : τ ≥−nM • −→ I •
projective R-module P . Then, we have quasi-isomorphisms:
n}n≥0 of K-pure injective complexes and the pure
n}n≥0 given by Proposition 3.9. We choose any pure
Hom•
R(P, fn) : Hom•
R(P, τ ≥−nM •) −→ Hom•
R(P, I •
n)
(3.29)
We first consider the inverse system {Hom•
that for n ≥ i + 2, the canonical morphisms
R(P, τ ≥−nM •)}n≥0. Choose any i ≥ 0. Then, it is clear
τ ≥−iHom•
R(P, τ ≥−n−1M •) −→ τ ≥−iHom•
R(P, τ ≥−nM •)
(3.30)
are all identity. Further, from the proof of Proposition 3.9, we know that each I •
n −→ J •
as I •
Hom•
) commutes with mapping cones (see, for instance, [4, (A.2.1.2)]), we get
xn is the mapping cone of a morphism x•
n+1 = Cxn[−1]•, where C •
n : I •
n+1 can be expressed
n. Since the functor
R(P,
Hom•
R(P, I •
n+1) = Cone(Hom•
R(P, x•
n))[−1]•
(3.31)
Accordingly, the morphisms Hom•
n) are all
surjective for n ≥ 0. It follows easily that for any i ≥ 0, the induced morphism on the truncations:
n+1) = Cone(Hom•
n))[−1]• −→ Hom•
R(P, x•
R(P, I •
R(P, I •
τ ≥−iHom•
R(P, I •
n+1) −→ τ ≥−iHom•
R(P, I •
n)
(3.32)
is surjective for each n ≥ 0. We now set I • := lim
←−
n≥0
n and consider the induced morphism f • :
I •
M • = lim
←−
n≥0
τ ≥−nM • −→ I •. From (3.30) and (3.32) and applying Lemma 3.10, it follows that we
have a quasi-isomorphism on inverse limits:
Hom•
R(P, M •) = lim
←−
n≥0
Hom•
R(P, τ ≥−nM •)
Hom•
R(P,f •)y
R(P, I •
Hom•
n) = Hom•
R(P, I •)
lim
←−
n≥0
(3.33)
for any pure projective R-module P . From (3.33), we conclude that f • : M • −→ I • is a pure quasi-
isomorphism. Further, we know from Proposition 3.9 that {I •
n}n≥0 is a special inverse system of
13
K-pure injective complexes. It now follows from Proposition 3.8(b) that the limit I • of the special
inverse system {I •
n}n≥0 is still K-pure injective. It remains to show that I • is a complex of pure
injectives. For this, we notice that for any j ∈ Z, I j is the limit of the following inverse system:
... −→ I j
n+1 = I j
n ⊕ J j−1
n
pn+1−−−−→ I j
n = I j
n−1 ⊕ J j−1
n−1
pn−−−−→ ... −→ I j
0
(3.34)
where each pn is the canonical projection onto the direct summand. Then, I j can be expressed as
the direct product I j = I j
n−1 of pure injectives. Hence, I j is pure injective.
0 ⊕Qn≥1 J j−1
4 Pure projective resolutions of unbounded complexes
In this section, we will construct pure projective resolutions for arbitrary complexes of R-modules.
As in the previous section, our methods are an adaptation of the classical method for constructing
projective resolutions of bounded above complexes (see, for example, [12]) along with the techniques
of Spaltenstein [19] for treating unbounded complexes. Unfortunately, the proofs in this section
are not always the dual of the arguments in Section 3. However, we will try to be as concise as
possible by pointing out all those arguments that are dual to the case of pure injective resolutions.
Definition 4.1. We will say that a cochain complex M • ∈ C(R) is K-pure projective if the functor
Hom•
) : C(R) −→ C(R) carries pure acyclic complexes to pure acyclic complexes.
R(M •,
We make the following observation: given a pure acyclic complex A• ∈ C(R) and any finitely
presented R-module F , we consider the complex Hom•
R(F, A•). Now if F ′ is any other finitely
presented R-module, the tensor product F ′ ⊗R F is still finitely presented. Hence, Hom•
R(F ′ ⊗R
F, A•) is acyclic. Therefore, Hom•
R(F ′ ⊗R F, A•) is acyclic for any
finitely presented R-module F ′ and we conclude that Hom•
R(F, A•)) ∼= Hom•
R(F, A•) is actually pure acyclic.
R(F ′, Hom•
Proposition 4.2. For a complex M • ∈ C(R), the following statements are equivalent:
(a) M • is K-pure projective.
(b) For any pure acyclic complex A• ∈ C(R), we have HomK(R)(M •, A•) = 0.
(c) For any complex A• ∈ C(R), we have HomK(R)(M •, A•) ∼= HomDpur(R)(M •, A•).
Proof. (a) ⇒ (b) : This is dual to the corresponding argument in the proof of Proposition 3.2.
(b) ⇒ (a): Let A• ∈ C(R) be pure acyclic. We have to show that Hom•
From the observation above, we know that for any finitely presented R-module F , Hom•
(and hence Hom•
R(F, A•)[k] for any k ∈ Z) is pure acyclic. Then, we get:
R(M •, A•) is pure acyclic.
R(F, A•)
H k(Hom•
R(F, Hom•
R(M •, A•))) = H k(Hom•
R(M •, Hom•
= HomK(R)(M •, Hom•
R(F, A•)))
R(F, A•)[k]) = 0
(4.1)
14
R(M •, A•) is pure acyclic.
for any k ∈ Z. Hence, Hom•
(c) ⇒ (b): This is clear because any pure acyclic A• is 0 in Dpur(R).
(a) ⇒ (c): The proof is dual to the corresponding argument in the proof of Proposition 3.3 if we
consider the morphisms in the derived category Dpur(R) as "left roofs" in place of the "right roofs"
appearing in (3.6).
Proposition 4.3. Let P • ∈ C(R) be a bounded above complex of pure projective modules. Then,
P • is K-pure projective.
Proof. Following Proposition 4.2, it suffices to show that HomK(R)(P •, A•) = 0 for any pure acyclic
A• ∈ C(R). The rest of the argument is now dual to that in the proof of Proposition 3.4.
It is known (see [21, Proposition 1]) that for any R-module M , there exists a pure epimorphism P ։
M with P a pure projective module. We are now ready to show that any bounded above complex
of R-modules admits a pure projective resolution. Now, a key step in the proof of Proposition
3.5, which gives the corresponding result for pure injective resolutions, is the fact that the functor
⊗R N preserves cokernels for any N ∈ R − M od. Since this no longer holds for kernels, we must
modify our approach somewhat to obtain pure projective resolutions, i.e., we cannot simply dualize
the proof of Proposition 3.5 here.
Proposition 4.4. (a) Let M • ∈ C(R) be a cochain complex that is bounded above. Then, there
exists a pure quasi-isomorphism v• : P • −→ M • such that P • is a bounded above complex of pure
projective modules.
2 −→ M •
(b) Let f • : M •
a pure quasi-isomorphism to M •
exists a bounded above complex P •
and a morphism g : P •
2 −→ P •
1 be a morphism of bounded above complexes in C(R). Let v•
2 : P •
2 be
2 of pure projectives. Then, there
1 −→ M •
1
2 −→ M •
1 : P •
2 from a bounded above complex P •
1 of pure projectives, a pure quasi-isomorphism v•
1 fitting into a commutative diagram:
P •
2
g•y
P •
1
v•
2−−−−→ M •
2
f •y
v•
1−−−−→ M •
1
(4.2)
Proof. (a) For the sake of definiteness, we suppose that the complex (M •, d•) satisfies M 0 6= 0
and M i = 0 for each i > 0. We set P i = 0 for each i > 0 and choose a pure epimorphism
v0 : P 0 −→ M 0 with P 0 pure projective. Then, for every i ≥ 0, we have already obtained pure
projectives P i, morphisms vi : P i −→ M i along with differentials ei : P i −→ P i+1 such that we
have induced isomorphisms:
Ker(HomR(Q, ei+1))/Im(HomR(Q, ei))
∼
−→ H i+1(Hom•
R(Q, M •))
∀ Q ∈ PP
(4.3)
15
and epimorphisms:
Ker(HomR(Q, ei)) ։ Ker(HomR(Q, di))
∀ Q ∈ PP
(4.4)
We suppose that we have already done this for all integers i greater than some given integer k ≥ −1.
We will show that there exists a pure projective P k, a morphism vk : P k −→ M k and a differential
ek : P k −→ P k+1 such that:
Ker(HomR(Q, ek+1))/Im(HomR(Q, ek)) ∼−→ H k+1(Hom•
Ker(HomR(Q, ek)) ։ Ker(HomR(Q, dk))
R(Q, M •))
∀ Q ∈ PP
We now consider the object Lk defined by the following pullback square:
Lk −−−−→ Ker(ek+1)
y
M k
y
dk
−−−−→ M k+1
∀ Q ∈ PP
(4.5)
(4.6)
and choose a pure epimorphism P k −→ Lk with P k pure projective. Then, for any Q ∈ PP,
the induced morphism HomR(Q, P k) −→ HomR(Q, Lk) is still an epimorphism and we obtain a
pullback square
HomR(Q, Lk)
−−−−→ HomR(Q, Ker(ek+1)) = Ker(HomR(Q, ek+1))
y
HomR(Q, M k)
HomR(Q,dk)
−−−−−−−−→
y
HomR(Q, M k+1)
(4.7)
We set ek : P k −→ P k+1 to be the composition P k −→ Lk −→ Ker(ek+1) −→ P k+1 and vk :
P k −→ M k to be the composition P k −→ Lk −→ M k. Now applying to the pullback square (4.7)
arguments that are dual to those applied to the pushout square (3.11) in the proof of Proposition 3.5,
we obtain the required results in (4.5). The induced morphisms Hom•
R(Q, P •) −→
Hom•
R(Q, M •) being quasi-isomorphisms for each pure projective Q ∈ PP, it follows that v• :
P • −→ M • is a pure quasi-isomorphism.
R(Q, v•) : Hom•
(b) As in part (a), we use the functors HomR(Q,
) with Q ∈ PP in place of the functors ⊗R N
) also preserves
with N ∈ R − M od appearing in the proof of Proposition 3.6. Since HomR(Q,
mapping cones, we can now apply arguments dual to those in the proof of Proposition 3.6 to prove
this result.
In order to proceed to pure projective resolutions of unbounded complexes, we will need to consider
"special direct systems".
Definition 4.5. (see [19, Definition 2.6] ) Let T be a class of complexes in C(R) that is closed
under isomorphisms.
16
(a) A T -special direct system of complexes is a direct system {P •
fying the following conditions for each n ∈ Z:
n }n∈Z of complexes in C(R) satis-
(1) The cochain map P •
n is injective.
n−1 −→ P •
(2) The cokernel C •
(3) The short exact sequence of complexes:
n−1 −→ P •
n := Coker(P •
n ) lies in the class T .
0 −→ P •
n−1
i•
n−→ P •
n
p•
n−→ C •
n −→ 0
(4.8)
is "semi-split", i.e., it is split in each degree.
(b) The class T ⊆ C(R) is said to be closed under special direct limits if the direct limit of every
T -special direct system in C(R) is contained in T .
By slight abuse of notation, we will refer to {P •
for all n < 0 makes {P •
n }n≥0 as a special direct system if setting P •
n }n∈Z into a special direct system in the sense of Definition 4.5 above.
n = 0
Proposition 4.6. (a) Let C ⊆ C(R) be a class of complexes. Let T (C) denote the class of complexes
M • ∈ C(R) such that Hom•
R(M •, A•) is pure acyclic for each A• ∈ C. Then, T (C) is closed under
special direct limits.
(b) The class of all K-pure projective complexes is closed under special direct limits.
Proof. (a) We set:
T ′ := {B• ∈ C(R) B• = Hom•
R(F, A•) for some finitely presented R-module F & some A• ∈ C}
R(M •, A•)) ∼= Hom•
R(F, Hom•
Now since Hom•
R(F, A•)), it follows that a complex M • ∈
T (C) if and only if Hom•
R(M •, B•) is acyclic for each B• ∈ T ′. From [19, Corollary 2.8], we now
see that T (C) is closed under special direct limits. Finally, the result of (b) follows from (a) by
letting C be the class of all pure acyclic complexes in C(R).
R(M •, Hom•
For a complex (M •, d•) ∈ C(R) and any given integer n, we now recall that its truncation τ≤nM •
is given by:
(τ≤nM )i =
0
Ker(dn)
M i
if i > n
if i = n
if i < n
It is clear that the complex M • may be expressed as the direct limit M • = lim
−→
n≥0
(4.9)
τ≤nM •.
17
Proposition 4.7. For any complex M • ∈ C(R) there exists a special direct system {P •
K-pure projective complexes and a morphism {fn : P •
the following conditions:
n }n≥0 of
n −→ τ≤nM •}n≥0 of direct systems satisfying
(a) Each P •
n is a bounded above complex of pure projectives.
(b) Each fn is a pure quasi-isomorphism.
Proof. The proof of this is dual to that of Proposition 3.9.
Lemma 4.8. Let {P (j)}j≥0 be a family of pure projective modules. Then, the direct sum Lj≥0
P (j)
is pure projective.
P (j) and consider a pure acyclic complex A•. We have to check that
Proof. We set P := Lj≥0
Hom•
R(P, A•) is pure acyclic. For this, we choose a finitely presented R-module F and see that:
Hom•
R(F, Hom•
R(P, A•)) = Hom•
R(F,Yj≥0
Hom•
R(P (j), A•)) = Yj≥0
Hom•
R(F, Hom•
R(P (j), A•))
Since each P (j) is pure projective, each of the complexes Hom•
each Hom•
be acyclic, we conclude that Hom•
F . This proves the result.
R(P (j), A•) is pure acyclic and hence
R(P (j), A•)) is acyclic. Since the product of acyclic complexes in R−M od must
R(P, A•)) is acyclic for any finitely presented R-module
R(F, Hom•
R(F, Hom•
Proposition 4.9. Let R be a commutative ring and let M • ∈ C(R) be a cochain complex of
R-modules. Then, M • has a pure projective resolution.
Proof. We consider the special direct system {P •
the pure quasi-isomorphisms {fn : P •
P • := lim
−→
n≥0
n }n≥0 of K-pure projective complexes along with
n −→ τ≤nM •}n≥0 from the proof of Proposition 4.7. We set
n . It is clear that pure quasi-isomorphisms commute with direct limits. This gives us
P •
an induced pure quasi-isomorphism:
f • : P • = lim
−→
n≥0
P •
n −→ lim
−→
n≥0
τ≤nM • = M •
(4.10)
Using Proposition 4.6(b), we see that the direct limit P • is K-pure projective. It remains to show
that P • is a complex of pure projective modules. From the construction of each P •
n in Proposition
4.7 which is dual to the construction in Proposition 3.9, it follows that each term in the direct limit
P • is actually a direct sum of a family of pure projective modules. It now follows from Lemma 4.8
that each term in P • is pure projective.
18
5 Closed monoidal structure on the pure derived category
In this section, we will use the pure projective and pure injective resolutions developed so far to
give a closed monoidal structure on the pure derived category Dpur(R). Now, if u• : M • −→ N •
is a pure quasi-isomorphism and A is any R-module, it is immediate from the definitions that
the induced morphism A ⊗R u• : A ⊗R M • −→ A ⊗R N • is a pure quasi-isomorphism. In order
to produce a tensor structure on Dpur(R), we will need to extend this fact to tensor products of
(possibly unbounded) complexes of R-modules.
Given cochain complexes (M •, d•
A) ∈ C(R), we recall that their tensor product (M • ⊗R
A•)• is the total complex associated to the double complex (M • ⊗R A•)ij := M i ⊗R Aj with
differentials dij
A : (M • ⊗R
A•)ij −→ (M • ⊗R A•)i,j+1.
M ⊗R Aj : (M • ⊗R A•)ij −→ (M • ⊗R A•)i+1,j and dij
2 := M i ⊗R dj
M ), (A•, d•
1 := di
Lemma 5.1. Let u• : M • −→ N • be a pure quasi-isomorphism. Then, for any given integer n ∈ Z,
the induced morphism on the truncations τ≤nu• : τ≤nM • −→ τ≤nN • is a pure quasi-isomorphism.
Proof. We will show that Hom•
R(P, τ≤nN •) is a quasi-
isomorphism for any pure projective P ∈ PP. Since u• is a pure quasi-isomorphism, we already
know that Hom•
R(P, N •) is a quasi-isomorphism. This induces
a quasi-isomorphism on the truncations:
R(P, τ≤nM •) −→ Hom•
R(P, M •) −→ Hom•
R(P, τ≤nu•) : Hom•
R(P, u•) : Hom•
τ≤nHom•
R(P, u•) : τ≤nHom•
R(P, M •) −→ τ≤nHom•
R(P, N •)
(5.1)
Further, since the functor HomR(P,
the truncations satisfy:
) preserves kernels, it is clear from the definition in (4.9) that
Hom•
R(P, τ≤nM •) = τ≤nHom•
R(P, M •)
Hom•
R(P, τ≤nN •) = τ≤nHom•
R(P, N •)
(5.2)
for any integer n ∈ Z. Combining (5.1) and (5.2), the result follows.
Proposition 5.2. Let u• : M • −→ N • be a pure quasi-isomorphism of complexes. Then, for any
cochain complex A• ∈ C(R), the induced morphism A• ⊗R u• : (A• ⊗R M •)• −→ (A• ⊗R N •)• is a
pure quasi-isomorphism.
Proof. We choose any R-module B and set C • := B ⊗R A•. Further, for any integer m ∈ Z, we set
C •
m := τ≤mC •. Now for any m, n ∈ Z, we claim that the induced morphism:
(C •
m ⊗R τ≤nM •)• −→ (C •
m ⊗R τ≤nN •)•
(5.3)
is a quasi-isomorphism. From Lemma 5.1, we know that τ≤nu• : τ≤nM • −→ τ≤nN • is a pure
quasi-isomorphism. It follows that for any fixed i ∈ Z, the morphism
C i
m ⊗R τ≤nu• : C i
m ⊗R (τ≤nM •) −→ C i
m ⊗R (τ≤nN •)
(5.4)
19
is a quasi-isomorphism. Since C •
m, τ≤nM • and τ≤nN • are all bounded above complexes, it now
follows from a standard spectral sequence argument (see, for instance, [20, Tag 0132]) that the
quasi-isomorphisms in (5.4) induce a quasi-isomorphism of the total complexes in (5.3). Taking
direct limits of quasi-isomorphisms over all m, n ∈ Z, we obtain a quasi-isomorphism:
B ⊗R (A• ⊗R M •)• = (C • ⊗R M •)• −→ (C • ⊗R N •)• = B ⊗R (A• ⊗R N •)•
(5.5)
for any R-module B. This proves the result.
From Proposition 5.2, it is clear that the tensor product ⊗ : C(R) × C(R) −→ C(R) descends to a
tensor product:
⊗ : Dpur(R) × Dpur(R) −→ Dpur(R)
(5.6)
on the pure derived category of R-modules.
Proposition 5.3. Let u• : M • −→ N • be a pure quasi-isomorphism of complexes. Then, the
following hold:
(a) Let P • be a K-pure projective complex. Then, Hom•
is a pure quasi-isomorphism.
(b) Let I • be a K-pure injective complex. Then, Hom•
is a pure quasi-isomorphism.
R(P •, M •) −→ Hom•
R(N •, I •) −→ Hom•
R(P •, u•) : Hom•
R(u•, I •) : Hom•
R(M •, I •)
R(P •, N •)
Proof. (a) Let C •
u denote the pure acyclic complex that is the mapping cone of the pure quasi-
isomorphism u• : M • −→ N •. For any complex B• ∈ C(R), we know that the functor Hom•
)
on C(R) commutes with mapping cones (see, for instance, [4, (A.2.1.2)]). In particular, we see that:
R(B•,
Cone(Hom•
R(P •, u•)) = Hom•
R(P •, C •
u)
(5.7)
Since P • is a K-pure projective complex, it follows from the definition that Hom•
acyclic. Combining this with (5.7), we conclude that Cone(Hom•
hence Hom•
R(P •, u•) is a pure quasi-isomorphism.
u) is pure
R(P •, u•)) is pure acyclic and
R(P •, C •
R( , B•) on C(R) preserves map-
(b) For any complex B• ∈ C(R), the contravariant functor Hom•
ping cones up to a shift (see, for instance, [10, (1.5.3)]). Since the shift of a pure acylic complex is
still pure acyclic, the result of part (b) now follows by applying an argument dual to that in part
(a).
Let M •, N • ∈ C(R) be two arbitrary complexes of R-modules. From the results of Section 4,
M −→ M • giving a pure projective resolution of M •.
we may choose a pure quasi-isomorphism P •
Similarly, using the results of Section 3, we may choose a pure quasi-isomorphism N • −→ I •
N giving
a pure injective resolution of N •. We can now define a "pure derived Hom functor":
P Hom•
R( ,
P Hom•
) : Dpur(R)op × Dpur(R) −→ Dpur(R)
R(M •, N •) := Hom•
M , I •
R(P •
N )
(5.8)
20
The functor in (5.8) is well defined due to the pure quasi-isomorphisms
Hom•
R(P •
M , N •) −→ Hom•
R(P •
M , I •
N ) ←− Hom•
R(M •, I •
N )
(5.9)
that both follow from Proposition 5.3.
We conclude by showing that we have obtained a closed symmetric monoidal structure on the pure
derived category Dpur(R).
Proposition 5.4. For any A•, B•, C • ∈ C(R), we have natural isomorphisms:
P Hom•
R((A• ⊗R B•)•, C •) ∼= P Hom•
R(A•, P Hom•
R(B•, C •))
HomDpur(R)((A• ⊗R B•)•, C •) ∼= HomDpur (A•, P Hom•
R(B•, C •))
(5.10)
Proof. We choose pure projective resolutions P •
pure injective resolution of C •. From Proposition 5.2, it follows that (P •
isomorphic to (A• ⊗R B•)•. From the definitions, it is also clear that (P •
projective complex each of the terms of which is pure projective. Hence, (P •
projective resolution of (A• ⊗R B•)•. It now follows that:
A and P •
A ⊗R P •
C be a
B)• is pure quasi-
B)• is a K-pure
B)• is a pure
A ⊗R P •
A ⊗R P •
B of A• and B• respectively. Let I •
P Hom•
R((A• ⊗R B•)•, C •) = Hom•
∼= Hom•
= Hom•
= P Hom•
R((P •
R(P •
R(P •
A ⊗R P •
A, Hom•
A, P Hom•
B)•, I •
C)
R(P •
B, I •
R(B•, C •))
C ))
R(A•, P Hom•
R(B•, C •))
To prove the second isomorphism in (5.10), we proceed as follows: it is already clear that:
HomDpur(R)((A• ⊗R B•)•, C •) ∼= HomDpur(R)((P •
A ⊗R P •
B)•, I •
C )
Since I •
C is K-pure injective, it follows from Proposition 3.3 that:
HomDpur(R)((P •
A ⊗R P •
B)•, I •
C ) ∼= HomK(R)((P •
A ⊗R P •
B)•, I •
C )
The closed monoidal structure on K(R) now gives:
HomK(R)((P •
A ⊗R P •
B)•, I •
C ) ∼= HomK(R)(P •
A, Hom•
R(P •
B, I •
C))
(5.11)
(5.12)
(5.13)
(5.14)
We can put P Hom•
from Proposition 4.2 that:
R(B•, C •) = Hom•
R(P •
B, I •
C). Further, since P •
A is K-pure projective, it follows
HomK(R)(P •
A, Hom•
R(P •
B, I •
C )) ∼= HomDpur(R)(P •
A, P Hom•
R(B•, C •))
(5.15)
Since P •
A
phism in (5.10).
∼= A• in Dpur(R), the sequence of isomorphisms (5.12)-(5.15) proves the second isomor-
21
References
[1] L. Alonso Tarr´ıo, A. Jerem´ıas L´opez, M. J. Souto Salorio, Localization in categories of complexes and
unbounded resolutions, Canad. J. Math. 52 (2) (2000) 225 -- 247.
[2] M. Bokstedt, A. Neeman, Homotopy limits in triangulated categories, Compositio Math. 86 (1993), no.
2, 209 -- 234.
[3] J. D. Christensen, M. Hovey, Quillen model structures for relative homological algebra, Math. Proc.
Cambridge Philos. Soc. 133 (2002), no. 2, 261 -- 293.
[4] L. W. Christensen, Gorenstein dimensions, Lecture Notes in Mathematics, 1747, Springer-Verlag,
Berlin, 2000.
[5] S. Estrada, J. Gillespie, S. Odaba¸si, Pure exact structures and the pure derived category of a scheme,
arXiv:1408.2846.
[6] P. Gabriel, M. Zisman, Calculus of fractions and homotopy theory, Ergebnisse der Mathematik und ihrer
Grenzgebiete, Band 35 Springer-Verlag New York, Inc., New York 1967.
[7] J. R. Garc´ıa-Rozas, Covers and envelopes in the category of complexes of modules, Research Notes in
Mathematics, 407, Chapman & Hall/CRC, Boca Raton, FL, 1999.
[8] J. Gillespie, The derived category with respect to a generator, Annali di Matematica Pura ed Applicata,
Vol. 195, No. 2, 371 -- 402 (2016).
[9] I. Herzog, Pure-injective envelopes, J. Algebra Appl., 2 (2003), no. 4, 397 -- 402.
[10] J. Lipman, M. Hashimoto, Foundations of Grothendieck duality for diagrams of schemes, Lecture Notes
in Mathematics, 1960. Springer-Verlag, Berlin, 2009.
[11] V. A. Lunts, M. O. Schnurer, New enhancements of derived categories of coherent sheaves and applica-
tions, J. Algebra, 446 (2016), 203 -- 274.
[12] D. Murfet, Derived Categories Part I (2006), available online at : therisingsea.org.
[13] A. Neeman, The derived category of an exact category, J. Algebra, 135 (1990) 388 -- 394.
[14] A. Neeman, Triangulated categories, Annals of Mathematics Studies, 148, Princeton University Press,
Princeton, NJ, 2001.
[15] A. Neeman, The homotopy category of flat modules, and Grothendieck duality, Invent. Math. 174
(2008) 255 -- 308.
[16] D. Quillen, Higher Algebraic K-Theory I, Lecture Notes in Math., vol. 341, Springer, Berlin, 1973.
[17] J. Rickard, Derived categories and stable equivalence, J. Pure Appl. Algebra, 61 (1989), no. 3, 303 -- 317.
[18] C. Serp´e, Resolution of unbounded complexes in Grothendieck categories, J. Pure Appl. Algebra, 177
(2003), no. 1, 103 -- 112.
[19] N. Spaltenstein, Resolutions of unbounded complexes, Compositio Math. 65 (1988), no. 2, 121 -- 154.
[20] The Stacks Project, available at http://stacks.math.columbia.edu.
[21] R. B. Warfield Jr., Purity and algebraic compactness for modules, Pacific J. Math., 28 (1969) 699 -- 719.
[22] Y. Zheng, Z. Huang, On pure derived categories, J. Algebra, 454, (2016), 252 -- 272.
22
|
1911.05015 | 1 | 1911 | 2019-11-12T17:19:25 | local derivations on Witt algebras | [
"math.RA"
] | In this paper, we prove that every local derivation on Witt algebras $W_n, W_n^+$ or $W_n^{++} $ is a derivation for any $n\in\mathbb{N}$. As a consequence, we obtain that every local derivation on a centerless generalized Virasoro algebra of higher rank is a derivation. | math.RA | math |
LOCAL DERIVATIONS ON WITT ALGEBRAS
YANG CHEN, KAIMING ZHAO, AND YUEQIANG ZHAO
n or W ++
Abstract. In this paper, we prove that every local derivation on Witt algebras
is a derivation for any n ∈ N. As a consequence we obtain that
Wn, W +
every local derivation on a centerless generalized Virasoro algebra of higher rank is a
derivation.
n
Keywords: Lie algebra, Witt algebra, generalized Virasoro algebra, derivation, local
derivation.
AMS Subject Classification: 17B05, 17B40, 17B66.
1. Introduction
Let A a Banach (or associative) algebra, X be an A-bimodule. A linear mapping
∆ : A → X is said to be a local derivation if for every x in A there exists a derivation
Dx : A → X, depending on x, satisfying ∆(x) = Dx(x). When X is taken to be
A, such a local derivation is called a local derivation on A. The concept of local
derivation for Banach (or associative) algebras was introduced by Kadison [11], Larson
and Sourour [12] in 1990. Since then there have been a lot of studies on local derivations
on various algebras. See for example the recent papers [1 -- 3, 13] and the references
therein. Local derivations on various algebras are some kind of local properties for the
algebras, which turn out to be very interesting. Kadison actually proved in [11] that
each continuous local derivation of a von Neumann algebra M into a dual Banach M-
bimodule is a derivation, he established the existence of local derivations on the algebra
C(x) of rational functions which are not derivations and showed that any local derivation
of the polynomial ring C[x1, · · · , xn] is a derivation. Recently, several papers have
been devoted to studying local derivations for Lie (super)algebras. In [3], Ayupov and
Kudaybergenov proved that every local derivation on a finite dimensional semisimple Lie
algebra over an algebraically closed field of characteristic 0 is automatically a derivation,
and gave examples of nilpotent Lie algebra (so-call filiform Lie algebras) which admit
local derivations which are not derivations. In [15] and [4] Ayupov and Yusupov studied
2-local derivations on univariate Witt algebras. In [16], we study 2-local derivations on
multivariate Witt algebras. In the present paper we study local derivations on Witt
algebras.
Let n ∈ N. The Witt algebra Wn of vector fields on an n-dimensional torus is the
derivation Lie algebra of the Laurent polynomial algebra An = C[t±1
n ]. Witt
algebras were one of the four classes of Cartan type Lie algebras originally introduced
in 1909 by Cartan [7] when he studied infinite dimensional simple Lie algebras. Over
the last two decades, the representation theory of Witt algebras was extensively studied
by many mathematicians and physicists; see for example [5, 6, 10]. Very recently, Billig
and Futorny obtained the classification for all simple Harish-Chandra Wn-modules in
their remarkable paper [5].
2 , · · · , t±1
1 , t±1
The present paper is arranged as follows. In Section 2 we recall some known results
and establish some related properties concerning Witt algebras. In Section 3 we prove
1
2
YANG CHEN, KAIMING ZHAO, AND YUEQIANG ZHAO
that every local derivation on Witt algebras Wn is a derivation. As a consequence
we obtain that every local derivation on a centerless generalized Virasoro algebra of
higher rank is a derivation. The methods used in [3] for finite dimensional semisimple
Lie algebras no longer work for Witt algebras since Witt algebras have very different
algebraic structure with finite dimensional semisimple Lie algebras. We have to establish
new methods in the proofs of this section. Finally, in Section 4 we show that the above
methods and conclusions are applicable for Witt algebras W +
n and W ++
n .
Throughout this paper, we denote by Z, N, Z+ and C the sets of all integers, positive
integers, non-negative integers and complex numbers respectively. All algebras are over
C.
2. The Witt algebras
In this section we recall definitions, symbols and some known results for later use in
this paper.
A derivation on a Lie algebra L is a linear map D : L → L which satisfies the Leibniz
law
D([x, y]) = [D(x), y] + [x, D(y)], ∀x, y ∈ L.
The set of all derivations of L is a Lie algebra and is denoted by Der(L). Clearly
derivations on L are local derivations, but the converse may not be true in general. For
any a ∈ L, the map
ad(a) : L → L, ad(a)x = [a, x], ∀x ∈ L
is a derivation and derivations of this form are called inner derivations. The set of all
inner derivations of L, denoted by Inn(L), is a Lie ideal of Der(L).
For n ∈ N, let An = C[t±1
n ] be the Laurent polynomial algebra and
Wn = Der(An) be the Witt algebra of vector fields on an n-dimensional torus. Thus
Wn has a natural structure of a left A-module, which is free of rank n. Denote d1 =
t1
, which form a basis of this A-module:
, . . . , dn = tn
1 , t±1
2 , · · · , t±1
∂
∂t1
∂
∂tn
Wn =
n
Mi=1
Andi.
Denote tα = tα1
basis of Zn. Then we can write the Lie bracket in Wn as follows:
n for α = (α1, . . . , αn) ∈ Zn and let {ǫ1, . . . , ǫn} be the standard
1 · · · tαn
[tαdi, tβdj] = βitα+βdj − αjtα+βdi, i, j = 1, . . . , n; α, β ∈ Zn.
The subspace h spanned by d1, . . . , dn is the Cartan subalgebra of Wn. We may write
any nonzero element in Wn as Pα∈S tαdα, where S is the finite subset consisting of all
α ∈ Zn with dα ∈ h \ {0}. For dα = c1d1 + · · · + cndn ∈ h and β = (β1, β2, · · · , βn) ∈ Zn,
define
Then we get the following formula
(dα, β) = c1β1 + · · · + cnβn.
[tαdα, tβdβ] = tα+β((dα, β)dβ − (dβ, α)dα), ∀α, β ∈ Zn, dβdα ∈ h.
For convenience when we write
n
X = Xα∈Zn
Xi=1
cα,itαdi ∈ Wn,
(2.1)
LOCAL DERIVATIONS ON WITT ALGEBRAS
3
where cα,i ∈ C, the coefficient cα,i will be denoted by (X)tαdi. We make the convention
that when X ∈ Wn is written as in (2.1) we always assume that the sum is finite, i.e.,
there are only finitely many cα,i nonzero.
Definition 2.1. We call a vector µ ∈ Cn generic if µ · α 6= 0 for all α ∈ Zn \ {0} ,
where µ · α is the standard inner product on Cn.
For a generic vector µ = (µ1, µ2, · · · , µn) ∈ Cn let dµ = µ1d1 + · · · + µndn. Then we
have the Lie subalgebra of Wn:
Wn(µ) = Andµ,
which is called (centerless) generalized Virasoro algebra of rank n, see [14].
From Proposition 4.1 and Theorem 4.3 in [8] we know that any derivation on Wn is
inner. Then for the Witt algebra Wn, the above definition of the local derivation can
be reformulated as follows. A linear map ∆ on Wn is a local derivation on Wn if for
every elements x ∈ Wn there exists an element ax ∈ Wn such that ∆(x) = [ax, x].
3. Local derivations on Wn
In this section we shall mainly prove the following result concerning local derivations
on Wn for n ∈ N.
Theorem 3.1. Every local derivation on the Witt algebra Wn is a derivation.
Since the proof of this theorem is long, we first setup six lemmas as preparations.
Let µ = (µ1, µ2, · · · , µn) be a fixed generic vector in Cn and dµ = µ1d1 + · · · + µndn.
For a given α ∈ Zn, we define an equivalence relation α∼ on Zn for β, γ ∈ Zn:
β α∼ γ if and only if γ − β = kα, for some k ∈ Z.
Let [γ] := {β ∈ Zn γ α∼ β} denote the equivalence class containing γ. The set of
equivalence classes of Zn defined by α is denoted by Zn/α.
Let ∆ be a local derivation on Wn with ∆(dµ) = 0. For tαdµ with α 6= 0, since ∆ is
β ∈ Wn, where d′′
β ∈ h, such that
(3.1)
(3.2)
(3.3)
a local derivation there is an element a = Pβ∈Zn tβd′′
Xk=pγ
∆(tαdµ) = [a, tαdµ] = X[γ]∈F
qγ
tγ+kαdγ+kα,
where F is a finite subset of Zn/α and pγ ≤ qγ ∈ Z. It is clear that
dα = (d′′
0, α)dµ.
For tαdµ + xdµ where x ∈ C∗, since ∆ is a local derivation there is an element
q′
γ
X[γ]∈F
Xk=p′
γ
tγ+kαd′
γ+kα ∈ Wn,
where d′
γ+kα ∈ h and p′
γ ≤ q′
γ ∈ Z, such that
∆(tαdµ) = ∆(tαdµ + xdµ) = [X[γ]∈F
q′
γ
Xk=p′
γ
tγ+kαd′
γ+kα, tαdµ + xdµ]
= X[γ]∈F
q′
γ+1
Xk=p′
γ
tγ+kα((d′
γ+(k−1)α, α)dµ − (dµ, γ + (k − 1)α)d′
γ+(k−1)α
− x(dµ, γ + kα)d′
γ+kα),
4
YANG CHEN, KAIMING ZHAO, AND YUEQIANG ZHAO
where we have assigned d′
(3.1) and (3.3).
γ −1)α = d′
γ+(p′
γ+(q′
γ +1)α = 0. Note that we have the same F in
Lemma 3.2. Let ∆ be a local derivation on Wn such that ∆(dµ) = 0. Then F = {[0]}
in (3.1) and (3.3).
γ α 6= 0 and d′
Proof. We have assumed that α 6= 0. Suppose that dγ+pγ α 6= 0 and dγ+qγ α 6= 0,
d′
γ α 6= 0 for some [γ] 6= [0]. Comparing the right hand sides of (3.1)
γ+p′
and (3.3) we see that pγ = p′
γ + 1, from (3.1) and (3.3) we
deduce that
γ + 1. If qγ < q′
γ and qγ ≤ q′
γ+q′
(d′
γ α, α)dµ − (dµ, γ + q′
γα)d′
γ+q′
γ α = 0.
γ+q′
Since (dµ, γ + q′
γα) 6= 0, we see that d′
γ α = cdµ for some c ∈ C∗, and furthermore
γ+q′
γ − 1)α)dµ = 0.
Then γ + (q′
γα)cdµ = c(dµ, −γ − (q′
(cdµ, α)dµ − (dµ, γ + q′
γ − 1)α = 0, i.e., [γ] = [0], a contradiction. Thus qγ = q′
Comparing (3.1) and (3.3) we deduce that (at least two equations)
− x(dµ, γ + pγα)d′
(d′
γ+pγ α, α)dµ − (dµ, γ + pγα)d′
· · · · · · · · · · · · ;
(d′
γ+pγ α − x(dµ, γ + (pγ + 1)α)d′
γ+(qγ −2)α, α)dµ − (dµ, γ + (qγ − 2)α)d′
γ+pγ α = dγ+pγ α;
γ+(qγ−2)α − x(dµ, γ + (qγ − 1)α)d′
γ+(qγ−1)α
γ + 1, and pγ < qγ.
γ+(pγ+1)α = dγ+(pγ +1)α;
= dγ+(qγ −1)α;
(d′
γ+(qγ −1)α, α)dµ − (dµ, γ + (qγ − 1)α)d′
γ+(qγ−1)α = dγ+qγ α.
Since (dµ, γ + kα) 6= 0 for k ∈ Z, eliminating d′
substitution we see that
γ+pγ α, . . . , d′
(3.4)
γ+(qγ −1)α in this order by
(3.5)
where ∗ ∈ h are independent of x. We always find some x ∈ C∗ not satisfying (3.5),
which is a contradiction. The lemma follows.
(cid:3)
dγ+qγ α + ∗x−1 + · · · + ∗x−qγ +pγ = 0,
Now (3.1) and (3.3) become
q
Xk=p
tkαdkα =
q′+1
Xk=p′
tkα((d′
(k−1)α, α)dµ − (dµ, (k − 1)α)d′
(k−1)α − x(dµ, kα)d′
kα),
(3.6)
where p = p0, q = q0, p′ = p′
may assume that dpα 6= 0 and dqα 6= 0, d′
and p′ = p if p′ 6= 0. Our destination is to prove that p = q = 1.
0 and we have assigned d′
p′α 6= 0 and d′
0, q′ = q′
(p′−1)α = d′
(q′+1)α = 0. We
q′α 6= 0. Clearly p′ ≤ p ≤ q ≤ q′+1,
Lemma 3.3. Let ∆ be a local derivation on Wn such that ∆(dµ) = 0. Then p′ ≥ 0 and
p ≥ 1 in (3.6).
Proof. To the contrary we assume that p′ < 0. Then p′ = p. If further q′ ≥ −1, from
(3.6) we obtain a set of (at least two) equations
pα = dpα;
pα, α)dµ − (dµ, pα)d′
− x(dµ, pα)d′
(d′
· · · · · · ;
(d′
−α, α)dµ − (dµ, −α)d′
−α = d0.
pα − x(dµ, (p + 1)α)d′
(p+1)α = d(p+1)α;
(3.7)
LOCAL DERIVATIONS ON WITT ALGEBRAS
5
If d0 6= 0, using the same arguments as for (3.4), the equations (3.7) makes contradic-
tions. So we consider the case that d0 = 0. From the last equation in (3.7) we see that
d′
−α = 0. We continue upwards in (3.7) in this manner to some step. We get
(l−1)α = 0.
d0 = d−α = · · · = dlα = 0, d(l−1)α 6= 0, and d′
−α = · · · = d′
If p + 1 < l(≤ 0), then (3.7) becomes
pα = dpα;
pα, α)dµ − (dµ, pα)d′
− x(dµ, pα)d′
(d′
· · · · · · ;
(d′
(l−2)α, α)dµ − (dµ, (l − 2)α)d′
(l−2)α = d(l−1)α.
pα − x(dµ, (p + 1)α)d′
(p+1)α = d(p+1)α;
(3.8)
Using the same arguments as for (3.4), the equations (3.8) makes contradictions. We
need only to consider the case that p + 1 = l, i.e., l − 1 = p. In this case we have that
0 = d′
p′α 6= 0, again a contradiction. Therefore q′ < −1.
(l−1)α = d′
If q < q′ + 1, from (3.6) we see that
(d′
q′α, α)dµ − (dµ, q′α)d′
q′α = 0.
Since (dµ, q′α) 6= 0, we see that d′
q′α = cdµ for some c ∈ C∗, and furthermore
(cdµ, α)dµ − (dµ, q′α)cdµ = c(dµ, −(q′ − 1)α)dµ = 0.
Then (q′ − 1)α = 0, i.e., q′ = 1, a contradiction. So q = q′ + 1 and p < q. We obtain a
set of (at least two) equations from (3.6)
pα = dpα;
pα, α)dµ − (dµ, pα)d′
− x(dµ, pα)d′
(d′
· · · · · · ;
(d′
(q−1)α, α)dµ − (dµ, (q − 1)α)d′
(q−1)α = dqα.
pα − x(dµ, (p + 1)α)d′
(p+1)α = d(p+1)α;
(3.9)
Using the same arguments as for (3.4), the equations (3.9) makes contradictions. Hence
p′ ≥ 0.
If p′ ≥ 1, then p = p′ ≥ 1. If p′ = 0, then d0 = (d′
−α, α)dµ − (dµ, −α)d′
−α = 0 by (3.6).
(cid:3)
So p ≥ 1 also.
Lemma 3.4. Let ∆ be a local derivation on Wn such that ∆(dµ) = 0. Then
∆(tαdµ) ∈ Ctαdµ, ∀α ∈ Zn.
Proof. From Lemma 3.3, we know that q ≥ p ≥ 1. We need only to prove that q = 1 in
(3.6). Otherwise we assume that q > 1, and then q′ > 0.
Case 1: q′ > 1.
In this case we can show that q = q′ + 1 as in the above arguments. If p′ ≥ 1, we see
that p = p′ and p < q. From (3.6) we obtain a set of (at least two) equations
pα = dpα;
pα, α)dµ − (dµ, pα)d′
− x(dµ, pα)d′
(d′
· · · · · · ;
(d′
(q−1)α, α)dµ − (dµ, (q − 1)α)d′
(q−1)α = dqα.
pα − x(dµ, (p + 1)α)d′
(p+1)α = d(p+1)α;
(3.10)
Using the same arguments as for (3.4), the equation (3.10) makes contradictions. So
p′ = 0. Now we have
p′ = 0, p ≥ 1, q = q′ + 1 > 2.
6
YANG CHEN, KAIMING ZHAO, AND YUEQIANG ZHAO
By (3.6) and (3.2) we have
(d′
0, α)dµ − x(dµ, α)d′
α = dα = (d′′
0, α)dµ;
(d′
α, α)dµ − (dµ, α)d′
α − x(dµ, 2α)d′
2α = d2α.
(3.11)
(3.12)
Equation (3.11) implies that d′
to −x(dµ, 2α)d′
α = cdµ, c ∈ C. Then Equation (3.12) can be simplified
2α = d2α. Again from (3.6) we obtain a set of (at least two) equations
− x(dµ, 2α)d′
2α = d2α
(d′
2α, α)dµ − (dµ, 2α)d′
· · · · · ·
(d′
(q−1)α, α)dµ − (dµ, (q − 1)α)d′
(q−1)α = dqα.
2α − x(dµ, 3α)d′
3α = d3α
(3.13)
Using the same arguments again, (3.13) makes contradictions. So q′ = 1. Now we have
Case 2: q′ = 1.
We need only consider the case that q = 2. In this case we still have Equations (3.11)
α = cdµ, c ∈ C. Then Equation
2α = 0. Equation (3.11) implies that d′
and (3.12) with d′
(3.12) implies that d2α = 0 which is a contradiction.
Therefore p = q = 1 and ∆(tαdµ) = tαdα = (d′′
0, α)tαdµ by (3.2). The lemma
(cid:3)
follows.
Lemma 3.5. Let ∆ be a local derivation on Wn such that ∆(dµ) = ∆(tidµ) = 0 for a
given 1 ≤ i ≤ n. Then ∆(tm
i dµ) = 0 for any m ∈ Z.
Proof. We may assume that m 6= 0, 1. By Lemmas 3.4, there is c ∈ C and
qγ
X[γ]∈F
Xk=pγ
tγ+k(m−1)ǫidγ+k(m−1)ǫi ∈ Wn,
where F is a finite subset of Zn/(m − 1)ǫi and pγ ≤ qγ ∈ Z, dγ+k(m−1)ǫi ∈ h, such that
ctm
i dµ =∆(tm
i dµ) = ∆(tm
i dµ + tidµ) = [X[γ]∈F
Xk=pγ
dk(m−1)ǫi + X[0]6=[γ]∈F
qγ
tk(m−1)
i
tk(m−1)
i
dk(m−1)ǫi, tm
i dµ + tidµ]
=[
=[
q0
Xk=p0
Xk=p0
q0
qγ
Xk=pγ
tγ+k(m−1)ǫidγ+k(m−1)ǫi, tm
i dµ + tidµ]
tγ+k(m−1)ǫidγ+k(m−1)ǫi, tm
i dµ + tidµ]
=
q0+1
Xk=p0
tk(m−1)+1
i
((d(k−1)(m−1)ǫi, mǫi)dµ − (dµ, (k − 1)(m − 1)ǫi)d(k−1)(m−1)ǫi)
+ ((dk(m−1)ǫi, ǫi)dµ − (dµ, k(m − 1)ǫi)dk(m−1)ǫi).
(3.14)
where we have assigned d(p0−1)(m−1)ǫi = d(q0+1)(m−1)ǫi = 0, and we have used the fact
that
qγ
[ X[0]6=[γ]∈F
Xk=pγ
tγ+k(m−1)ǫidγ+k(m−1)ǫi, tm
i dµ + tidµ] = 0
LOCAL DERIVATIONS ON WITT ALGEBRAS
7
since it cannot contain elements from tm
i h or eliminate any term in
q0
[
Xk=p0
tk(m−1)
i
dk(m−1)ǫi, tm
i dµ + tidµ].
We may assume that dp0(m−1)ǫi 6= 0 and dq0(m−1)ǫi 6= 0. Our destination is to prove that
c = 0, To the contrary we assume that c 6= 0. Then
p0 ≤ 1, q0 ≥ 0, and p0 ≤ q0.
Claim 1. p0 = 0 or 1.
Suppose that p0 < 0, by (3.14) we deduce that
(dp0(m−1)ǫi, ǫi)dµ − (dµ, p0(m − 1)ǫi)dp0(m−1)ǫi = 0.
Since (dµ, p0(m − 1)ǫi) 6= 0, we have dp0(m−1)ǫi = c′dµ for some c′ ∈ C∗ and furthermore
(dp0(m−1)ǫi, ǫi)dµ − (dµ, p0(m − 1)ǫi)dp0(m−1)ǫi
=(c′dµ, ǫi)dµ − (dµ, p0(m − 1)ǫi)c′dµ
=c′(dµ, (1 − p0(m − 1))ǫi)dµ = 0
.
It follows that 1 − p0(m − 1) = 0, and thus p0 = −1, m = 0, a contradiction. Hence
p0 = 0 or 1.
Claim 2. q0 = 0.
Suppose that q0 > 0, by (3.14) we deduce that
(dq0(m−1)ǫi, mǫi)dµ − (dµ, q0(m − 1)ǫi)dq0(m−1)ǫi = 0.
Since (dµ, q0(m − 1)ǫi) 6= 0, we have dq0(m−1)ǫi = c′dµ for some c′ ∈ C∗ and furthermore
(dq(m−1)ǫi, mǫi)dµ − (dµ, q(m − 1)ǫi)dq(m−1)ǫi
=(c′dµ, mǫi)dµ − (dµ, q0(m − 1)ǫi)c′dµ
=c′(dµ, (m − q0(m − 1))ǫi)dµ = 0
.
It follows that m − q0(m − 1) = 0, and thus m = 2 and q0 = 2. Now from (3.14) we
deduce that
(d0, ǫi)dµ = 0;
(d0, 2ǫi)dµ + ((dǫi, ǫi)dµ − (dµ, ǫi)dǫi) = cdµ;
((dǫi, 2ǫi)dµ − (dµ, ǫi)dǫi) + ((d2ǫi, ǫi)dµ − (dµ, 2ǫi)d2ǫi) = 0;
(d2ǫi, 2ǫi)dµ − (dµ, 2ǫi)d2ǫi = 0.
(3.15)
(3.16)
(3.17)
(3.18)
We see that dǫi = c′′dµ, c′′ ∈ C by (3.18) and (3.17). Substituting into (3.16) it follows
that c = 0, a contradiction. Claim 2 follows.
From Claims 1 and 2 we have p0 = q0 = 0. By (3.14) we have (d0, ǫi)dµ = 0 and still
(cid:3)
cdµ = (d0, mǫi)dµ = 0. The proof is completed.
Lemma 3.6. Let ∆ be a local derivation on Wn such that ∆(dµ) = ∆(tidµ) = 0 for all
1 ≤ i ≤ n. Then ∆(tαdµ) = 0 for all α ∈ Zn.
Proof. From Lemma 3.5, we have
∆(tmǫidµ) = 0, ∀1 ≤ i ≤ n, m ∈ Z.
We may assume that n > 1 and α ∈ Zn \ ∪1≤i≤nZǫi. Take an fixed integer m > αi for
any 1 ≤ i ≤ n. Define
I = {i : αi ≥ 0} and I ′ = {i : αi < 0}.
8
YANG CHEN, KAIMING ZHAO, AND YUEQIANG ZHAO
Then by Lemma 3.4 and Lemma 3.5 there is c ∈ C and an element x = Pβ∈Zn tβdβ ∈ Wn
such that
ctαdµ = ∆(tαdµ) = ∆(tαdµ +Xi∈I
tm
i dµ +Xi∈I ′
t−m
i dµ)
tβdβ, tαdµ +Xi∈I
= [Xβ∈Zn
i dµ] = 0 for some 1 ≤ i ≤ n, where γ ∈ Zn \ {0}, dγ ∈ h \ {0}
i dµ +Xi∈I ′
t−m
i dµ].
tm
(3.19)
Claim 1. If [tγdγ, tk
and k ∈ Z, then tγdγ = citk
i dµ for some ci ∈ C∗.
Since
[tγdγ, tk
i dµ] = tγ+kǫi((dγ, kǫi)dµ − (dµ, γ)dγ) = 0.
and (dµ, γ) 6= 0, we have dγ = cidµ for some ci ∈ C∗ and furthermore
(dγ, kǫi)dµ − (dµ, γ)dγ = (cidµ, kǫi)dµ − (dµ, γ)cidµ = ci(dµ, kǫi − γ)dµ = 0.
It implies that γ = kǫi. Claim 1 follows.
Take a nonzero term tγdγ in the expression of x with maximal degree with respect to
ti for some i ∈ I.
Claim 2. If γi ≥ 0 and γ 6= 0, then tγdγ = citm
The term [tγdγ, tm
i dµ] is of maximal degree with respect to ti in the expression of
i dµ for some ci ∈ C∗.
[Xβ∈Zn
tβdβ, tαdµ +Xi∈I
tm
i dµ +Xi∈I ′
t−m
i dµ].
Since γ + mǫi 6= α, it follows that [tγdγ, tm
ci ∈ C∗ by Claim 1.
i dµ] = 0. We see that tγdγ = citm
i dµ for some
i dµ for some ci ∈ C∗ if γi ≤ 0 and γ 6= 0. Moreover, citm
Similarly, a nonzero term tγdγ in the expression of x with minimal degree with respect
i dµ (resp.
i dµ) is the only possible term of non-negative maximal degree (resp. non-positive
If there is such a
i dµ, i ∈ I without loss of generality. Now, to delete the term
to ti for i ∈ I ′ is cit−m
cit−m
minimal degree) with respect to ti, i ∈ I (resp.
term, we consider citm
[citm
i ∈ I ′) in x.
i dµ, tαdµ] = ci(dµ, α − mǫi)tα+mǫidµ 6= 0 in
tβdβ, tαdµ +Xi∈I
[Xβ∈Zn
tm
i dµ +Xi∈I ′
t−m
i dµ],
there must be citαdµ in Pβ∈Zn tβdβ. To delete the term [citαdµ, tm
(resp. [citαdµ, t−m
j dµ], j ∈ I ′) in
j dµ] 6= 0, i 6= j ∈ I
[Xβ∈Zn
tβdβ, tαdµ +Xi∈I
j dµ, i 6= j ∈ I (resp. cit−m
tm
i dµ +Xi∈I ′
t−m
i dµ],
there must be citm
j dµ, j ∈ I ′). Thus ci = c1 for all 1 ≤ i ≤ n.
Let x′ = x − c1(tαdµ +Pi∈I tm
i dµ +Pi∈I ′ t−m
i dµ). Then
ctαdµ = [x′, tαdµ +Xi∈I
tm
i dµ +Xi∈I ′
t−m
i dµ].
Note that, if tγdγ is a nonzero term in the expression of x′, then γ = 0 or
(cid:26)γi < 0
γi > 0
if i ∈ I,
if i ∈ I ′.
Claim 3. In the expression of x′, the term t0d0 = 0.
(3.20)
(3.21)
LOCAL DERIVATIONS ON WITT ALGEBRAS
9
If d0 6= 0, considering the highest (resp.
lowest) degree term with respect to ti for
i ∈ I (resp. i ∈ I ′) in (3.20), we deduce that (d0, ǫi) = 0 for all 1 ≤ i ≤ n. Thus d0 = 0.
Claim 3 follows.
Suppose that there exists a nonzero term tγdγ in the expression of x′ with maximal
i dµ] is of
degree with respect to some ti for i ∈ I. Then γ 6= 0, and the term [tγdγ, tm
maximal degree with respect to ti in the expression of
[x′, tαdµ +Xi∈I
tm
i dµ +Xi∈I ′
t−m
i dµ].
It is only possible that 0 6= [tγdγ, tm
i dµ] ∈ tαh by Claim 1. Then γ + mǫi = α. We have
0 > γj = αj ≥ 0, i 6= j ∈ I and 0 < γj = αj < 0, i 6= j ∈ I ′ by (3.21), a contradiction.
So x′ = 0. Therefore ∆(tαdµ) = 0.
(cid:3)
Lemma 3.7. Let ∆ be a local derivation on Wn such that ∆(dµ) = 0. Then ∆ h= 0.
Proof. This is trivial for n = 1. Next we assume that n > 1. For a given i ∈ {1, . . . , n},
there is an element Pα∈Zn tαd(i)
α ∈ Wn, where d(i)
α ∈ h such that
∆(di) = [Xα∈Zn
tαd(i)
α , di] = − Xα∈Zn
αitαd(i)
α .
So we may suppose that
∆(di) = Xα∈Zn\{0} X1≤j≤n
c(i)
α,jtαdj,
where c(i)
cα,j = c(k)
α,j ∈ C. For a given α ∈ Zn \ {0}, we have αk 6= 0 for some 1 ≤ k ≤ n. Let
α,j/αk. For i 6= k, let d = αkdi − αidk. We have
α,j − αic(k)
α,j.
(∆(d))tαdj = αkc(i)
On the other hand, there is an element Pα∈Zn P1≤j≤n bα,jtαdj ∈ Wn where bα,j ∈ C
such that
(∆(d))tαdj = ([Xα∈Zn X1≤j≤n
α,j = αi(c(k)
Then c(i)
α,j/αk) = αicα,j, yielding that
bα,jtαdj, αkdi − αidk])tαdj = −bα,jαkαi + bα,jαiαk = 0.
0 = (∆(dµ))tαdj =
n
Xi=1
µic(i)
α,j = cα,j
n
Xi=1
µiαi.
We deduce that cα,j = 0. So
c(i)
α,j = 0, ∀1 ≤ i, j ≤ n, α ∈ Zn \ {0},
that is, ∆(di) = 0 for 1 ≤ i ≤ n. Therefore ∆ h= 0.
(cid:3)
Now we are in the position to prove Theorem 3.1.
Proof of Theorem 3.1. Let ∆ be a local derivation on Wn. We fix an arbitrary generic
µ ∈ Cn. There is an element a ∈ Wn such that ∆(dµ) = [a, dµ]. Set ∆1 = ∆ − ad(a).
Then ∆1 is a local derivation such that ∆1(dµ) = 0. From Lemma 3.7, we know that
∆1(h) = 0. By Lemma 3.4, there are ci ∈ C such that
∆1(tidµ) = citidµ, ∀1 ≤ i ≤ n.
10
YANG CHEN, KAIMING ZHAO, AND YUEQIANG ZHAO
Set ∆2 = ∆1 −Pn
i=1 ciad(di). Then ∆2 is a local derivation such that
∆2(h) = 0, and ∆2(tidµ) = 0, ∀1 ≤ i ≤ n.
By Lemma 3.6, for any generic µ ∈ Cn we have
∆2(tαdµ) = 0, ∀α ∈ Zn.
For any generic λ ∈ Cn that is not a multiple of µ since ∆2(dλ) = 0, from Lemma 3.4
there is cα ∈ C such that ∆2(tαdλ) = cαtαdλ. There is Pβ∈Zn tβdβ, where dβ ∈ h, such
that
cαtαdλ =∆2(tαdα) = ∆2(tαdµ + tαdλ) = [Xβ∈Zn
tβdβ, tαdµ + tαdλ]
tα+β((dβ, α)(dµ + dλ) − (dµ + dλ, β)dβ).
= Xβ∈Zn
We see that cαtαdλ = (d0, α)(dµ + dλ), yielding that cα = 0. Thus for any generic vector
λ,
∆2(tαdλ) = 0, ∀α ∈ Zn.
ad(a) +Pn
i=1 ciad(di) is a derivation. The proof is completed.
Since the set {dλ : λ is generic} can span h we must have ∆2 = 0. Hence ∆ =
✷
By Theorem 3.4 in [9] any derivation on the generalized Virasoro algebra Wn(µ) can
be seen as the restriction of a inner derivation on Wn. All the proofs in this section with
minor modifications are valid for the generalized Virasoro algebra Wn(µ). Therefore we
obtain the following consequence.
Corollary 3.8. Let n ∈ N, and let µ ∈ Cn be generic. Then any local derivation on
the generalized Virasoro algebra Wn(µ) is a derivation.
4. Local derivations on W +
n
and W ++
n
For n ∈ N, we have the Witt algebra W +
n = Der(C[t1, t2, · · · , tn]) which is a subal-
gebra of Wn. We use h to denote the Cartan subalgebra of Wn which is also a Cartan
subalgebra (not unique) of W +
n . We know that
W +
n = Xα∈Zn
+
tαh +
n
Xi=1
Ct−1
i di.
Furthermore W +
n has a subalgebra
W ++
n = Xα∈Zn
+
tαh.
It is well-known that W +
n is a simple Lie algebra, but W ++
n
is not.
From Proposition 4.1 and Theorem 4.3 in [8] we know that any derivation on W +
n is
inner . Using same arguments as the proof of Proposition 3.3 in [9] we can show that
any derivation on W ++
is inner.
n
Hence, the proofs and conclusions with slight modifications in Section 3 are applicable
n . It is routine to verify this. We omit the details and directly state the
n and W ++
to W +
following theorem.
Theorem 4.1. Every local derivation on Witt algebras W +
n or W ++
n
is a derivation.
Acknowledgements. This research is partially supported by NSFC (11871190) and
NSERC (311907-2015).
LOCAL DERIVATIONS ON WITT ALGEBRAS
11
References
[1] Sh. A. Ayupov, K. K. Kudaybergenov, Local derivations on measurable operators and commuta-
tivity, Eur. J. Math. 2 (2016), no. 4, 1023-1030. 1
[2] Sh. A. Ayupov, K. Kudaybergenov, A. Peralta, A survey on local and 2-local derivations on C∗-
and von Neumann algebras. Topics in functional analysis and algebra, 73-126, Contemp. Math.,
672, Amer. Math. Soc., Providence, RI, 2016. 1
[3] Sh. A. Ayupov, K. K. Kudaybergenov, Local derivations on finite-dimensional Lie algebras, Linear
Algebra Appl. 493 (2016), 381-398. 1, 2
[4] Sh. A. Ayupov, B. B. Yusupov, 2-Local derivations on infinite-dimensional Lie algebras,
arXiv:1901.04261. 1
[5] Y. Billig, V. Futorny, Classification of simple Wn-modules with finite-dimensional weight spaces,
J. Reine Angew. Math. 720 (2016), 199-216. 1
[6] Y. Billig, A. Molev, R. Zhang, Differential equations in vertex algebras and simple modules for
the Lie algebra of vector fields on a torus, Adv. Math., 218 (2008), no.6, 1972-2004. 1
[7] E. Cartan, Les groupes de transformations continus, infinis, simples. (French) Ann. Sci. École
Norm. Sup. (3) 26 (1909), 93-161. 1
[8] D. Z. Djokovic, K. Zhao, Generalized Cartan type W Lie algebras in characteristic zero, J. Algebra,
195 (1997), 170-210. 3, 10
[9] D. Z. Djokovic, K. Zhao, Derivations, isomorphisms and second cohomology of generalized Witt
algebra, Trans. Amer. Math. Soc., (2) 350 (1998), 643-664. 10
[10] X. Guo, G. Liu, R. Lu, K. Zhao, Simple Witt modules that are finitely generated over the Cartan
subalgebra, to appear in Moscow Mathematical Journal. 1
[11] R.V. Kadison, Local derivations, J. Algebra, 130 (1990), 494-509. 1
[12] D.R. Larson and A.R. Sourour, Local derivations and local automorphisms of B(X), Proc. Sym-
pos. Pure Math., 51, Part 2, Providence, Rhode Island 1990, 187-194. 1
[13] D. Liu, J. Zhang, Local Lie derivations of factor von Neumann algebras, Linear Algebra Appl.
519 (2017), 208-218. 1
[14] J. Patera, H. Zassenhaus, The higher rank Virasoro algebras, Comm. Math. Phys. 136 (1991),
1-14. 3
[15] B. B. Yusupov, 2-local derivations on Witt algebras, Uzbek Math. J., 125 (2018), 160-166. 1
[16] Y. Zhao, Y. Chen, K. Zhao, 2-Local derivations on Witt algebras, arXiv:1909.06242.
1
Mathematics Postdoctoral Research Center, Hebei Normal University, Shiji-
azhuang 050016, Hebei, China
E-mail address: [email protected]
Department of Mathematics, Wilfrid Laurier University, Waterloo, ON, Canada
N2L 3C5, and College of Mathematics and Information Science, Hebei Normal Univer-
sity, Shijiazhuang 050016, Hebei, China
E-mail address: [email protected]
School of Mathematical Sciences, Hebei Normal University, Shijiazhuang 050016,
Hebei, China
E-mail address: [email protected]
|
1612.05040 | 3 | 1612 | 2018-03-27T10:01:47 | Reachability of eigenspaces for interval circulant matrices in max-algebra | [
"math.RA"
] | A nonnegative matrix A is said to be strongly robust if its max-algebraic eigencone is universally reachable, i.e., if the orbit of any initial vector ends up with a max-algebraic eigenvector of A. Consider the case when the initial vector is restricted to an interval and A can be any matrix from a given interval of nonnegative circulant matrices. The main aim of this paper is to classify and characterize the six types of interval robustness in this situation. This naturally leads us also to study the max-algebraic spectral theory of circulant matrices and the relation of inclusion between attraction cones of circulant matrices in max-algebra. | math.RA | math |
Reachability of eigenspaces for interval circulant
matrices in max-algebra
J´an Plavkaa, Sergeı Sergeevb,1,∗
aDepartment of Mathematics and Theoretical Informatics, Technical University,
bUniversity of Birmingham, School of Mathematics, Edgbaston B15 2TT, UK
B. Nemcovej 32, 04200 Kosice, Slovakia
Abstract
A nonnegative matrix A is said to be strongly robust if its max-algebraic
eigencone is universally reachable, i.e., if the orbit of any initial vector ends
up with a max-algebraic eigenvector of A. Consider the case when the initial
vector is restricted to an interval and A can be any matrix from a given
interval of nonnegative circulant matrices. The main aim of this paper is to
classify and characterize the six types of interval robustness in this situation.
This naturally leads us also to study the max-algebraic spectral theory of
circulant matrices and the relation of inclusion between attraction cones of
circulant matrices in max-algebra.
Keywords: Max-algebra, circulant matrices, interval analysis, reachability.
AMS classification: 15A18, 15A80, 65G40, 93C55
1. Introduction
Max-algebra has applications in such fields as discrete event systems and
scheduling theory (among others) [2, 4, 11], and plays a crucial role in the
study of discrete event systems in connection with optimization problems
such as scheduling or project management in which the objective function
depends on the maximum and times operations (or equivalently maximum
and plus via a logarithmic transform). Notice that the main principle of
∗Corresponding author.
Email addresses: [email protected] (J´an Plavka), [email protected] (Sergeı
Sergeev)
1Supported by EPSRC grant EP/P019676/1
Preprint submitted to Elsevier
August 24, 2018
discrete events systems consisting of n entities is that the entities work in-
teractively, i.e., a given entity must wait before proceeding to its next job
until certain others have completed their current jobs. The steady states
of such systems correspond to the max-algebraic eigenvectors of the matri-
ces that describe them, therefore the investigation of reachability of the set
of eigenvectors from a given state by a given system is important for such
applications. Matrices for which the steady states of the corresponding sys-
tems are reached with any nontrivial starting vector are called robust, see [4]
Section 8.6.
In practice, matrix entry values are not exact numbers and usually are
contained within intervals, and therefore interval arithmetic is an efficient
way to represent matrices in a guaranteed way on a computer. A max-
algebraic (tropical) version of interval analysis was developed, e.g., in [12],
which emphasized the polynomiality of some algorithms of max-algebraic
interval analysis. That polynomiality was in striking contrast with NP-
hardness of relevant algorithms previously known in usual interval analysis.
Independently, [7] developed a theory of some max-algebraic linear systems
with interval coefficients and optimization problems over such systems.
When developing interval extensions of linear algebra problems a whole
range of solvability problems routinely arises, by considering all possible com-
binations of quantifiers (as in Definition 2.8 of the present paper). In classical
linear algebra this leads to the notions of united solutions, controllable solu-
tions and tolerable solutions [17, 20]. In max-algebra we similarly have, e.g.,
four types of interval extensions of the max-algebraic spectral problem [8] or
two types of interval extensions of robustness studied in [15].
Similarly to [15], the present paper also considers max-algebraic interval
extensions of robustness and reachability problems. However, we focus on
matrices of a certain special type: circulants.
In usual algebra, circulant
matrices have a number of geometric applications [6]. A more recent appli-
cation of circulants can be found in [21]. There, an algebraic construction
based on circulant matrices allows for designing LDPC codes with efficient
encoder implementation, in contrast to designing LDPC codes based on ran-
dom construction techniques which make it difficult to store and assess a
large parity-check matrix or to analyze the performance of the code.
In
max-algebra, circulant matrices appear to describe the periodic regime of se-
quences of matrix powers [4, 18]. It is also easy to see that circulant matrices
of a given dimension form a commutative semigroup, both in max-algebra
and in usual linear algebra.
2
When considering matrices of special type, it is natural to require that
the set of matrices that is an interval extension of such a matrix can contain
matrices of that type only. This is a basic idea behind the notion of interval
circulant matrix defined here. The main aim of the present paper is thus
to classify and characterize the six types of interval robustness for circulant
matrices in max-algebra. However, obtaining such a characterization is not
possible without a deeper study of properties of circulant matrices in max-
algebra, which is itself of some theoretical interest.
We now outline the organization of the paper and the results obtained
there. Section 2 is devoted to some basic notions of max-algebra and its
connections to the theory of digraphs and max-algebraic convexity. In par-
ticular, we revisit the max-algebraic spectral theory here, focusing on the
eigencone and the attraction cone associated with an arbitrary eigenvalue,
the cyclicity of critical graphs and the ultimate periodicity of max-algebraic
matrix powers and orbits.
Section 3 presents some known as well as some new results on the spec-
tral theory and attraction cones of circulant matrices. In particular, Propo-
sition 3.7 describes the critical node sets of circulant matrices and presents
several formulae for the cyclicity of the critical graph of a circulant matrix.
This result combines together some facts that have been previously obtained
or stated in [14, 15, 22]. The main new result of this section is Theorem 3.10,
which deals with a particular problem of inclusion of the attraction cones of
circulant matrices A and B satisfying A ≤ B and having the same maxi-
mum cycle mean. It appears that inclusion attr(A) ⊆ attr(B) holds for such
circulant matrices. Note that it does not hold for general matrices, as Exam-
ple 2.24 demonstrates. Section 3 also contains several motivating examples.
The proofs of Proposition 3.7 and Theorem 3.10 are deferred to Section 5.
Based on the result about inclusion of attraction cones of Theorem 3.10,
Section 4 characterizes various types of interval robustness which are de-
scribed in Definition 2.8. Some of them can be verified in polynomial time, see
Theorems 4.7, 4.9, 4.15. Other types of robustness reduce to max-algebraic
two-sided systems of equations and inequalities for which efficient algorithms
exist but the problem of constructing a polynomial algorithm remains open.
See Theorems 4.11, 4.13, 4.14.
Subsection 5.1 presents a proof of Proposition 3.7. The proof uses the
fact that any circulant matrix is strictly visualized in the sense of [19] and
relies in part on the results of [9, 10].
Subsection 5.2 presents a proof of Theorem 3.10. In particular, the proof
3
draws upon the role of cyclic classes in the max-linear systems of equations
describing attraction cones, as presented in [4] Chapter 8 and [18].
2. Preliminaries
2.1. Main definitions and problem statements
By max-algebra we mean the set of nonnegative numbers R+ equipped
with the usual multiplication a · b and the idempotent addition a ⊕ b :=
max(a, b). These arithmetical operations are then routinely extended to ma-
trices and vectors: in particular, (A ⊗ B)i,k = Lj Ai,j · Bj,k and (A ⊕ B)i,j =
Ai,j ⊕ Bi,j for any two nonnegative matrices A, B of appropriate sizes. We
will also consider the max-algebraic powers of matrices Ak := A ⊗ . . . ⊗ A
.
}
In what follows, we will be interested in the orbits of vectors under the
{z
k
action of matrices, that is, the sets
O(A, x) = {x, A ⊗ x, A2 ⊗ x, . . .},
(1)
and especially in the case when the orbit of a vector hits an eigenvector of A.
Let us now give formal definitions related to the max-algebraic eigenproblem.
Definition 2.1 (Eigenvalues and Eigenvectors). A value λ ∈ R+ is called
a (max-algebraic) eigenvalue of A ∈ Rn×n
The greatest eigenvalue of A will be denoted by λ(A).
if A⊗x = λx for some x ∈ Rn
+\{0}.
+
A vector x ∈ Rn
+\{0} satisfying A ⊗ x = λx is called a (max-algebraic)
eigenvector associated with A.
The eigencone of A associated with eigenvalue λ is defined as the set
containing all eigenvectors of A with associated eigenvalue λ as well as the
zero vector:
V (A, λ) = {x ∈ Rn
+ : A ⊗ x = λ ⊗ x}.
One of the key notions of the paper is that of attraction cone: the set
which comprises all vectors whose orbit hits a given eigencone.
Definition 2.2 (Attraction cones). The attraction cone of A ∈ Rn×n
sociated with eigenvalue λ is the set
+
as-
attr(A, λ) = {x ∈ Rn
+ : O(A, x) ∩ V (A, λ) 6= ∅}.
We also denote attr(A) = attr(A, λ(A)).
4
Any eigencone or any attraction cone is a max cone, in the sense of the
following definition.
Definition 2.3 (Max cones). A set V ⊆ Rn
+ is called a max cone if for
all x ∈ V , y ∈ V any max-linear combination αx ⊕ βy (where α, β ∈ R+)
belongs to V .
We will use the following notational shortcuts.
Definition 2.4 (Index Sets N and N0). We denote
N = {1, . . . , n}, N0 = {0, . . . , n − 1}.
In this paper we deal with the following special class of matrices in max-
algebra.
Definition 2.5 (Circulant Matrices). A matrix A ∈ Rn×n
is called circu-
lant, if it has entries Ai,j = at for i, j ∈ N, t ∈ N0 such that t ≡ (j−i)(mod n)
and a0, a1, . . . , an−1 ∈ R+. Equivalently, A is a circulant matrix if it is of the
form
+
A =
a0
a1 a2
an−1 a0 a1
...
...
...
a2 a3
a1
. . . an−2 an−1
. . . an−3 an−2
...
a0
...
. . . an−1
.
for some a0, a1, . . . , an−1 ∈ R+. Such a circulant matrix we will also denote
by Z(a0, . . . , an−1).
Circulant matrices will be the main topic of Section 3 and Section 5,
where we will study their spectral theory and attraction cones.
The final part of this paper is devoted to intervals and interval circulant
matrices.
Definition 2.6 (Intervals). A set X ⊆ Rn
the form
+ is called an interval if it is of
X = ×n
i=1
X i,
for X i nonempty subsets of R+ taking any of the following four forms:
[xi, xi], (xi, xi), (xi, xi], [xi, xi),
for xi, xi ∈ R+.
5
Definition 2.7 (Interval Circulant Matrices). By Z C(a0, . . . , an−1) we
denote the set of all circulant matrices A such that Ai,j ∈ at for i, j ∈ N
and t ∈ N0 such that t ≡ (j − i)(mod n), where a0, . . . , an−1 are intervals
independently taking any of the four forms listed in Definition 2.6.
A set of circulant matrices that is of the form Z C(a0, . . . , an−1) for inter-
vals a0, . . . , an−1 is called an interval circulant matrix.
+, see [4] Section 8.6.
+
In the literature on max-algebra, A ∈ Rn×n
is called robust if attr(A) =
Rn
In this paper we consider various extensions of
this notion to interval circulant matrices. These extensions are listed in the
following definition.
Definition 2.8 (Interval Robustness). Let X ⊆ Rn
+ be an interval and
Z C(a0, . . . , an−1) be an interval circulant matrix. Then Z C(a0, . . . , an−1) is
called
possibly X−robust if (∃A ∈ Z C(a0, . . . , an−1))(∀x ∈ X)[ x ∈ attr(A) ] ,
universally X−robust if (∀A ∈ Z C(a0, . . . , an−1))(∀x ∈ X)[ x ∈ attr(A) ] ,
tolerance X−robust if (∀A ∈ Z C(a0, . . . , an−1))(∃x ∈ X)[ x ∈ attr(A) ] ,
weakly tolerance X−robust if (∃A ∈ Z C(a0, . . . , an−1))(∃x ∈ X)[x ∈ attr(A)]
and X is called
possibly Z C−robust if (∃x ∈ X)(∀A ∈ Z C(a0, . . . , an−1))[ x ∈ attr(A) ] ,
tolerance Z C−robust if (∀x ∈ X)(∃A ∈ Z C(a0, . . . , an−1))[ x ∈ attr(A) ] .
In particular, the aim of Section 4 will be to derive an efficient character-
ization of these types of interval robustness.
2.2. Associated graphs, critical graphs and periodicity
Let us start with the following basic definition. For relevant definitions
see also, e.g., [4] Section 1.5.
Definition 2.9 (Digraphs, Walks, Cycles and Connectivity). Let G be
a digraph with set of nodes N and set of edges E. A walk on G is a se-
quence W = (i0, i1, . . . , il) with i0, i1, . . . , il ∈ N where each pair (is−1, is) for
6
s ∈ {1, . . . , l} is an edge. If i0 = i and il = j then W is said to be connecting
i to j, and l is called the length of W .
G is called strongly connected if for each i, j ∈ N with i 6= j there exists
a walk on G connecting i to j.
For A ∈ Rn×n
+ , the weighted digraph G(A) associated with A is the digraph
with set of nodes N = {1, . . . , n} and set of edges E = {(i, j) : Ai,j 6= 0},
where Ai,j is the weight of an edge (i, j).
If G = G(A) then the weight of W = (i0, i1, . . . , il) is defined by Ai0,i1 ·
Ai1,i2 · . . . · Ail−1,il. This walk is called a cycle if il = i0, with the cycle
(geometric) mean defined by (Ai0,i1 · Ai1,i2 · . . . · Ail−1,i0)1/l.
Let us also give a separate definition of the maximum cycle mean.
Definition 2.10 (Maximum cycle (geometric) mean). The maximum
cycle (geometric) mean of any A ∈ Rn×n
or of G(A) is
+
n
max
k=1
max
1≤i1,...,ik≤n
(Ai1,i2 · Ai2,i3 . . . Aik,i1)1/k.
(2)
The striking importance of this concept in max-algebra is due to the following
fact.
Proposition 2.11 (e.g., [4], Corollary 4.5.6). For any A ∈ Rn×n
greatest max-algebraic eigenvalue (λ(A)) is equal to (2).
+ , its
The concept of irreducible matrix is common for max-algebra and non-
negative linear algebra, and it is most conveniently defined via the associated
digraph.
Definition 2.12 (Irreducible, Reducible and Completely Reducible).
A is called irreducible if G(A) is strongly connected, and reducible otherwise.
Digraph G is called completely reducible if it consists of several strongly
connected subgraphs called components such that there are no walks con-
necting a node from one component to a node of another component. A is
called completely reducible if so is G(A).
Note that any irreducible matrix is completely reducible. Observe also
the following criterion of complete reducibility.
Proposition 2.13. A digraph G = (N, E) is completely reducible if and only
if every edge of E lies in a cycle of G.
7
Proof. "If": Suppose that G contains two maximal strongly connected sub-
graphs G1 and G2 and that there is a walk connecting one subgraph to the
other. Without loss of generality we can assume that the walk does not con-
tain nodes from any other subgraphs, so that it contains an edge (i, j) with
i ∈ G1 and j ∈ G2. As this edge is on a cycle, there is also a walk from j
to i. However, this implies that G1 and G2 both belong to a larger strongly
connected subgraph of G thus contradicting their maximality. Thus the "if"
part is proved.
"Only if": If G is completely reducible then each edge (i, j) belongs to a
strongly connected subgraph of G, and it belongs to a cycle since there exists
a walk connecting j back to i.
(cid:3)
The following subdigraph of G(A) is crucial for the max-algebraic spectral
theory and it is an example of completely reducible digraph.
Definition 2.14 (Critical Digraphs). The critical digraph of A, denoted
by Gc(A), consists of all nodes and edges of the cycles of G(A) at which the
maximum cycle mean of A (2) is attained. These cycles are called critical
cycles. The nodes of Gc(A) are called critical nodes and their set is denoted
by Nc(A), and the edges of Gc(A) are called critical edges and their set is
denoted by Ec(A).
Corollary 2.15. Any critical graph is completely reducible.
Proof. By Definition 2.14, every edge of Gc(A) belongs to a cycle of Gc(A).
The claim now follows from Proposition 2.13.
(cid:3)
The concept of the digraph's cyclicity is crucial for the study of attraction
cones (Definition 2.2) and the ultimate periodicity of {At}t≥1 (to be defined
soon).
Definition 2.16 (Cyclicity). For a strongly connected digraph, its cyclic-
ity is defined as the g.c.d. of the lengths of all cycles of that digraph.
Cyclicity of a completely reducible digraph is defined as the l.c.m. of the
cyclicities of its components.
Cyclicity of a digraph G is denoted by σ(G).
We now discuss the ultimate periodicity of max-algebraic matrix powers.
8
Definition 2.17 (Ultimate Periodicity). Let {αk}k≥1 be a sequence of
some elements. If there exists T such that αt+σ = αt for all t ≥ T and some
σ (i.e., αt+σ and αt are identical), then {αk}k≥1 is called ultimately periodic.
The least T and the least σ for which the above property holds are called
the transient and the ultimate period of {αk}k≥1 respectively.
Proposition 2.18 ([5]). Let A ∈ Rn×n
be an irreducible matrix with λ(A) 6=
0.Then {(A/λ(A))t}t≥1 is ultimately periodic and σ(Gc(A)) is the ultimate
period of that sequence.
+
In this paper we also need the following trivial extension of Proposi-
tion 2.18 and its consequence for orbits of vectors.
Corollary 2.19. Let A ∈ Rn×n
be a completely reducible matrix with λ(A) 6=
0, such that the maximum cycle mean of each component of G(A) is the same
(and equal to λ(A)).Then {(A/λ(A))t}t≥1 is ultimately periodic and σ(Gc(A))
is the ultimate period of that sequence.
+
Corollary 2.20. Under the conditions of Proposition 2.18 or Corollary 2.19,
{(A/λ(A))t ⊗ x}t≥1 is ultimately periodic for any x ∈ Rn
+.
Let us now introduce some notation related to the ultimate periodicity.
Definition 2.21. Let A ∈ Rn×n
have λ(A) 6= 0. If {(A/λ(A))t}t≥1 is ulti-
mately periodic then denote by T (A) the transient and by per(A) the ulti-
mate period of that sequence.
+
Thus per(A) = σ(Gc(A)) for any A satisfying the condition of Proposi-
tion 2.18 or Corollary 2.19.
The ultimate period of {(A/λ(A))t ⊗ x}t≥1 does not necessarily equal the
cyclicity of Gc(A), and the attraction cone associated with λ(A) consists of
the vectors for which the ultimate period of {(A/λ(A))t ⊗ x}t≥1 is equal to
1. More precisely, we have the following.
Proposition 2.22. Let A ∈ Rn×n
be a completely reducible matrix with
λ(A) 6= 0 such that the maximum cycle mean of each component of G(A)
is the same (and equal to λ(A)). Then
+
attr(A) = {x ∈ Rn
+ : λ(A)At ⊗ x = At+1 ⊗ x}, where t ≥ T (A).
9
Proof. By definition x ∈ attr(A) if and only if As+1 ⊗ x = λ(A)As ⊗ x for
some s, hence λ(A)At ⊗ x = At+1 ⊗ x for some t ≥ T (A) is sufficient for
x ∈ attr(A). For the necessity observe that As+1 ⊗ x = λ(A)As ⊗ x im-
plies As′+1 ⊗ x = λ(A)As′ ⊗ x for some s′ ≥ max(s, T (A)) and such that
(A/λ(A))s′ = (A/λ(A))t, and hence At+1 ⊗ x = λ(A)At ⊗ x.
(cid:3)
Corollary 2.23. Under the conditions of Proposition 2.22 attr(A) is a closed
max-cone.
Proof. Under these conditions attr(A) is the solution set of the system
λ(A)At ⊗ x = At+1 ⊗ x. This solution set is a max-cone since it is closed
under taking max-linear combinations (see Definition 2.3) and it is a closed
set since all arithmetic operations of max-algebra are continuous.
(cid:3)
Let us finally consider the attraction cones of the following two matrices
satisfying the conditions of Proposition 2.22.
Example 2.24. Take
A =
1
0.2 0
0.5
1
0.5 0.2 0
0.2 0.2 0.2 0
1
0
0
0
, B =
1
0.2 0
0.5
1
0.5 0.3 0
0.4 0.4 0.4 0
1
0
0
0
The ultimate periods of {A, A2, A3, . . .} and {B, B2, B3, . . .} equal 2. In the
first case, the periodicity starts from A2 (i.e., we have A2 = A4), and in the
second case it starts from B3 (i.e., we have B3 = B5). The attraction cones
are
attr(A) = {x : A3 ⊗ x = A4 ⊗ x},
attr(B) = {x : B3 ⊗ x = B4 ⊗ x},
where
A3 =
B3 =
1
0.5
0.2
0.2
0.5
0
1
0
0.2 0.2 0.04 0
0
1
0
0
1
0.5
0.2
0.3
0
0.5
0
1
0.4 0.4 0.12 0
0
1
0
0
, A4 =
, B4 =
10
0.2
0.2
0.5
1
1
0
0.5
0
0.2 0.2 0.04 0
0
1
0
0
0.3
0.2
0.5
1
0
1
0.5
0
0.4 0.4 0.12 0
0
1
0
0
,
.
We further see that in both cases, the systems defining these attraction cones
reduce to just one equation:
attr(A) = {x : 0.5x1 ⊕ x2 ⊕ 0.2x3 = x1 ⊕ 0.5x2 ⊕ 0.2x3},
attr(B) = {x : 0.5x1 ⊕ x2 ⊕ 0.2x3 = x1 ⊕ 0.5x2 ⊕ 0.3x3},
Observe that x = [1 1 5 1] belongs to attr(A) but not to attr(B), and
x = [0.5 1 10
(cid:3)
3 1] belongs to attr(B) but not to attr(A).
Example 2.24 also shows that Theorem 3.10, the main result of the next
section which claims that attr(A) ⊆ attr(B) for any circulant A, B with
A ≤ B and λ(A) = λ(B), is not true for general completely reducible (or
irreducible) matrices.
3. Circulant matrices: critical graph and attraction cones
Let us start with the following statement, which is well known in usual
linear algebra. See, e.g., [6] Theorem 3.1.1. A proof of it in max-algebra,
which works equally well in the usual linear algebra case, is given below for
the reader's convenience.
Proposition 3.1. Let A, B ∈ Rn×n
be circulant matrices. Then A ⊗ B
is also circulant. In particular, any max-algebraic power of A (or B) is a
circulant.
+
Proof. Observe that A is a circulant matrix if and only if we can represent
A = a0I ⊕ a1P ⊕ . . . ⊕ an−1P n−1, where
P =
0
0
...
0
1
1
0
. . .
0
0
0
1
. . .
. . .
. . .
. . . 0
. . . 0
...
1
0
. . .
0
0
.
In this case A = Z(a0, a1, . . . , an−1). Computing A ⊗ B amounts to multi-
plying a0I ⊕ a1P ⊕ . . . ⊕ an−1P n−1 by b0I ⊕ b1P ⊕ . . . ⊕ bn−1P n−1, assuming
11
that A = Z(a0, a1, . . . , an−1) and B = Z(b0, b1, . . . , bn−1). This multiplica-
tion results in an expression of the form c0I ⊕ c1P ⊕ . . . ⊕ cn−1P n−1, thus also
a circulant.
Writing At as At−1 ⊗A for every t ≥ 2, we also show that At is a circulant
(cid:3)
by a simple inductive argument.
The following observation will play a key role in proving many properties
of circulants.
Lemma 3.2. Let A ∈ Rn×n
be a nonzero circulant matrix and let Ai,j =
µ 6= 0 for some i, j ∈ N. Then (i, j) belongs to a cycle (i1, . . . , in, i1) with
it −it−1 ≡ (j −i)(mod n) for all t ∈ {2, . . . , n}, and i1 −in ≡ (j −i)(mod n).
The weight of each edge in (i1, . . . , in, i1) equals Ai,j = µ.
+
Proof. Consider an infinite sequence {iℓ}ℓ≥1 where i1 = i, i2 = j, iℓ+1 − iℓ ≡
(j − i)(mod n) for all ℓ ≥ 1 and iℓ ∈ N for all ℓ ≥ 1. By Definition 2.5,
Aiℓ,iℓ+1 = Ai,j for all ℓ ≥ 1. However, we also have that in+1 = i1 since
in+1 − i1 ≡ n · (j − i)(mod n) = 0(mod n). Hence the claim follows.
(cid:3)
Proposition 3.3. Let A = Z(a0, . . . , an−1). Then A has a unique max-
algebraic eigenvalue equal to
λ(A) =
n−1
max
k=0
ak.
(3)
If A 6= 0 then λ(A) 6= 0 and all nodes in N are critical.
Proof. If A 6= 0 then max(a0, . . . , an−1) > 0. In this case, let i and j be
such that Ai,j = µ > 0. By Lemma 3.2 (i, j) belongs to a cycle (i1, . . . , in, i1)
where the weights of all edges are equal to µ. It follows that the cycle mean
of that cycle is also µ. Thus, the maximal cycle mean is equal to the maximal
weight of edges, which shows (3). Taking k such that ak = λ(A), for each
i ∈ N we have j with k ≡ (j − i)(mod n) such that Ai,j = λ(A), hence each
i ∈ N is on a critical cycle. Since all nodes G(A) are critical, A has a unique
eigenvalue equal to λ(A) as it follows, e.g., from [4] Corollary 4.5.8.
If A = 0 then max(a0, . . . , an−1) = 0 = λ(A).
(cid:3)
Note that equation (3) was obtained already in [16], Theorem 2.1. How-
ever, we preferred to give a partially self-contained proof of this equation for
the reader's convenience.
12
Corollary 3.4. Let A ∈ Rn×n
only if A = 0.
+
be a circulant matrix. Then λ(A) = 0 if and
Proof. Obviously, λ(A) = 0 if A = 0. The "only if" part is equivalent to the
implication (A 6= 0) ⇒ (λ(A) 6= 0) stated in Proposition 3.3.
(cid:3)
We now formulate the following immediate corollary of Proposition 2.13.
Corollary 3.5. Any circulant matrix A is completely reducible.
Proof. If A = 0 then G(A) has no edges and is completely reducible. Oth-
erwise, by Lemma 3.2 any edge of G(A) belongs to a cycle, and the claim
follows from Proposition 2.13.
(cid:3)
Proposition 3.6. For any nonzero circulant matrix A ∈ Rn×n
sequence {(A/λ(A))t}t≥1 is ultimately periodic, and T (A) ≤ (n − 1)2 + 1.
the matrix
+
Proof. For the first part of the claim observe that any circulant matrix is
completely reducible by Corollary 3.5, and that by Proposition 3.3 λ(A) is
the maximum cycle mean of any maximal strongly connected component of
G(A).
Since λ(A) = 0 implies A = 0 by Corollary 3.4, we can assume λ(A) = 1
without loss of generality. Since all nodes of G(A) are critical, the transient
of periodicity of {A, A2, A3, . . .} is the same as the greatest transient of peri-
odicity of any sequence of rows of these powers {Ai•, A2
i•, . . .} where i is
critical. However, these transients are bounded by (n − 1)2 + 1 by [13] Main
Theorem 1.
(cid:3)
i•, A3
The following proposition gives more information on the critical graph
and cyclicity of circulant matrices.
Proposition 3.7. Let A = Z(a0, . . . , an−1) 6= 0 and let p1, . . . , ps ∈ {1, . . . , n−
1} be the nonzero indices for which ap1 = . . . = aps = λ(A) (if such indices
exist) and such that p1 > p2 > . . . > ps. Then
(i) Gc(A) consists of m = gcd(n, p1, . . . , ps) isomorphic strongly connected
components. Node set of the ith component, for i ∈ {1, . . . , m}, is
{i, i + m, . . . , i + (n/m − 1)m}.
13
(ii) per(A), equal to the cyclicity of each of these components, is 1 if a0 =
λ(A) and
per(A) = gcd(
n
gcd(n, p1)
,
p1 − p2
gcd(p1, p2)
,
p1 − p3
gcd(p1, p3)
, . . . ,
p1 − ps
gcd(p1, ps)
)
= gcd(
= gcd(
n
gcd(n, p1)
n
gcd(n, p1)
,
,
if a0 6= λ(A).
p1 − p2
gcd(p1, p2)
p1 − p2
gcd(n, p1, p2)
,
p2 − p3
gcd(p2, p3)
, . . . ,
ps−1 − ps
gcd(ps−1, ps)
)
,
p1 − p3
gcd(n, p1, p2, p3)
. . . ,
p1 − ps
gcd(n, p1, . . . , ps)
(4)
)
Parts of this statement can be found in [15] Theorem 4.1 and Lemma 4.1.
Essentially, part (i) was proved in [14] Lemma 4.2 and Lemma 4.3, although
in the max-min algebra setting. The number gcd(n, p1, . . . , ps) also appeared
in [22] Theorem 4 as the "eigenspace dimension". The result of part (ii)
relies on [9] (Theorems 3.1 and 3.3) where the cyclicity of threshold circulant
graphs (see Definition 5.3) was studied. We will give a complete proof of (i)
and a reduction of (ii) to the results of [9] in Subsection 5.1, for the reader's
convenience.
Let us now describe the attraction cone of a circulant matrix as a solution
set of a max-algebraic two-sided system of equations.
Proposition 3.8. Let A ∈ Rn×n
+
be a circulant matrix. Then
attr(A) = {x : λ(A)An2
⊗ x = An2+1 ⊗ x}.
Proof. By Corollary 3.4, λ(A) = 0 if and only if A = 0, in which case
+ and An2 = An2+1 = 0, and the claim holds trivially. Oth-
attr(A) = Rn
erwise, by Corollary 3.5 A is completely reducible and by Proposition 3.3
the maximal cycle mean of each component of G(A) is the same. The claim
then follows since A satisfies the conditions of Proposition 2.22 and since
n2 ≥ T (A) by Proposition 3.6.
(cid:3)
Let us examine the attraction cone of a 4 × 4 circulant matrix.
14
Example 3.9. Consider
A =
0 0 1 t
t 0 0 1
1 t 0 0
0 1 t 0
,
where t : 0 < t < 1. This is a circulant matrix, λ(A) = 1, and Gc(A) consists
of two disjoint cycles: (1 3) and (2 4). The cyclicity of Gc(A) is thus equal
to 2 and so is the ultimate period of the max-algebraic matrix powers of A.
Taking the max-algebraic powers of A we obtain
A2 =
A2k−1 =
1
0
t2
t
t2
t
1
t3
t
1
0
t2
t3
t2
t
1
t2
t
1
0
1
t3
t2
t
0
t2
t
1
t
1
t3
t2
, A2k =
1
t3
t2
t
t
1
t3
t2
t2
t
1
t3
t3
t2
t
1
∀k ≥ 2,
∀k ≥ 2.
In particular, the periodicity transient is T (A) = 3. By Proposition 3.8 we
have attr(A) = {x : A16 ⊗ x = A17 ⊗ x}, implying that the attraction cone is
precisely the set of vectors x = (x1 x2 x3 x4) that satisfy
x1 ⊕ tx2 ⊕ t2x3 ⊕ t3x4 = t2x1 ⊕ t3x2 ⊕ x3 ⊕ tx4
tx1 ⊕ t2x2 ⊕ t3x3 ⊕ x4 = t3x1 ⊕ x2 ⊕ tx3 ⊕ t2x4.
(5)
System (5) can be further reduced using the cancellation rule
a ⊕ b = ta ⊕ c ⇔ a ⊕ b = c,
where t < 1 and a, b, c are arbitrary. Repeatedly applying this rule we obtain
the system
x1 ⊕ tx2 = x3 ⊕ tx4
tx1 ⊕ x4 = x2 ⊕ tx3,
(6)
equivalent to (5).
Now observe that x = [t 1 t2 1] satisfies this system of equations and
belongs to the attraction cone. In particular, the ultimate period of {Atx}t≥1
is 1, however, A ⊗ x 6= x which shows that attr(A) is not the same as the
(max-algebraic) eigencone of A in this case.
15
The following theorem is one of the main results of the paper. Its proof
is postponed to Subsection 5.2.
Theorem 3.10. Let A, B ∈ Rn×n
λ(B) and A ≤ B. Then attr(A) ⊆ attr(B).
+
be two circulant matrices such that λ(A) =
Let us give two examples demonstrating this theorem. In the first example
we have two 0-1 matrices, and in the second one we consider the matrix of
Example 3.9 with two different values of t.
Example 3.11. Let us first consider a pair of 0-1 matrices:
A =
0 1 0 0 0 0
0 0 1 0 0 0
0 0 0 1 0 0
0 0 0 0 1 0
0 0 0 0 0 1
1 0 0 0 0 0
, B =
0 1 0 1 0 0
0 0 1 0 1 0
0 0 0 1 0 1
1 0 0 0 1 0
0 1 0 0 0 1
1 0 1 0 0 0
.
Observe that the sequence {At}t≥1 is periodic from the very beginning. The
system A36 ⊗ x = A37 ⊗ x, being the same as A ⊗ x = A2 ⊗ x, reduces to
x1 = x2 = x3 = x4 = x5 = x6.
The sequence {Bt}t≥1 becomes periodic from T (B) = 2. More precisely, we
have
B2k+1 =
0 1 0 1 0 1
1 0 1 0 1 0
0 1 0 1 0 1
1 0 1 0 1 0
0 1 0 1 0 1
1 0 1 0 1 0
, B2k =
1 0 1 0 1 0
0 1 0 1 0 1
1 0 1 0 1 0
0 1 0 1 0 1
1 0 1 0 1 0
0 1 0 1 0 1
,
k ≥ 1,
and the system B36 ⊗ x = B37 ⊗ x reduces to x1 ⊕ x3 ⊕ x5 = x2 ⊕ x4 ⊕ x6,
thus attr(A) ⊆ attr(B).
(cid:3)
Example 3.12. Take
A =
0
0
t1
0
1 t1
0
1 t1
1
0
0
0
1 t1
0
0
0
t2
0
1 t2
0
1 t2
1
0
0
0
1 t2
0
, B =
16
with 0 < t1 < t2 < 1. Then attr(A) is the set of all x satisfying (6) with
t = t1, which is
x1 ⊕ t1x2 = x3 ⊕ t1x4
t1x1 ⊕ x4 = x2 ⊕ t1x3,
and attr(B) is the set of all x satisfying
x1 ⊕ t2x2 = x3 ⊕ t2x4
t2x1 ⊕ x4 = x2 ⊕ t2x3,
(7)
(8)
We next show that attr(A) ⊆ attr(B) in this example, by considering various
special cases.
Suppose first that we have t1x2 = t1x4 ≥ x1 ⊕ x3 in the first equation
of (7). This implies x2 = x4 ≥ t2(x1 ⊕ x3) ≥ t1(x1 ⊕ x3) and t2x2 = t2x4 ≥
(x1 ⊕ x3). This shows that in this case x belongs to both attr(A, 1) and
attr(B, 1). The case when t1x1 = t1x3 ≥ x2 ⊕ x4 in the second equation
of (7) is treated similarly.
Suppose now that x ∈ attr(A) and t1x2 = x3 ≥ x1 ⊕ t1x4. As we cannot
have t1x1 = x2 and x4 = t1x3 in the second equation of (7), assume that x2 =
x4 ≥ t1(x1 ⊕ x3). But this implies t1x2 = t1x4, and as t1x2 is the maximum
in the first equation, this returns us to the case which we considered first,
where x ∈ attr(B). We also note three other similar cases that are treated
in the same way.
The remaining case when x ∈ attr(A), x1 = x3 ≥ t1(x2 ⊕ x4) and
(cid:3)
x2 = x4 ≥ t1(x1 ⊕ x3) is impossible when t1 < 1.
4. Interval robustness of circulant matrices
In this section we characterize the six types of interval robustness of
Definition 2.8 for interval circulant matrix Z C(a0, . . . , an−1) and interval
X = ×n
X i where X i and ai are intervals independently taking one of the
following four forms:
i=1
and
[xi, xi], (xi, xi), (xi, xi], [xi, xi)
[aj, aj], (aj, aj), (aj, aj], [aj, aj)
for xi, xi ∈ R+ and i ∈ N, and aj, aj ∈ R+ and j ∈ N0, respectively.
17
4.1. Universal and possible X−robustness
Let us introduce the following notation.
Definition 4.1 (Matrices A(k) and vectors x(k)). For a given index k ∈
N0 denote
A(k) = Z(a0, a1, . . . , ak−1, ak, ak+1, . . . , an−1),
and
x(k) = (x1, x2, . . . , xk−1, xk, xk+1, . . . , xn)
The following lemma explains the use of vectors x(k).
Lemma 4.2. Let X ⊆ Rn
attr(A) if and only if x(i) ∈ attr(A) for each i ∈ N.
+ be an interval and let A ∈ Rn×n
+ . Then X ⊆
Proof. Observe first that since the cone attr(A) is a closed set by Corol-
lary 2.23, the inclusion X ⊆ attr(A) is equivalent to cl(X) ⊆ attr(A), where
cl is a Euclidean closure. Since x(i) ∈ cl(X) for all i ∈ N (as vertices of the
box cl(X)), it follows that the condition is necessary. Let us show that this
condition is also sufficient. For this we will show that
x =
n
M
k=1
xk
xk
x(k).
(9)
Indeed, observe that when k 6= i we have that xk/xk ≤ 1 implies (xk/xk)x(k)
xi, and when k = i we obtain (xi/xi)x(i)
i ≤
i = xi. Since xi ≤ xi, we obtain that
n
M
k=1
xk
xk
i = (xi/xi)x(i)
x(k)
i = xi,
for all i, so (9) holds. Thus x can be expressed as a max-linear combination of
x(k) for k ∈ N and x ∈ attr(A) since attr(A) is a max-cone (Definition 2.3). (cid:3)
Definition 4.3 (Matrix A). For a = max
k∈N0
ak define
where
A = Z(a0, a1, . . . , an−1),
ai = min{a, ai}, for each i ∈ N0.
18
Let us characterize the cases when A = 0 and when A ∈ Z C(a0, . . . , an−1).
Proposition 4.4. Let Z C(a0, . . . , an−1) be given. Then
(i) A = 0 ⇔ a = 0 ⇔ A = 0 ⇔ λ(A) = 0.
(ii) If ∀i : ai = [ai, ai], then A ∈ Z C(a0, . . . , an−1).
(iii) A ∈ Z C(a0, . . . , an−1) ⇔ ∀i : (a ≥ ai ⇒ ai ∈ ai)&(a ≤ ai ⇒ a ∈ ai).
Proof. (i): Let us show that A = 0 ⇔ a = 0. By Definition 4.3 it is immedi-
ate that a = 0 implies A = 0. Next, assume that A = 0. Then ai = 0 for all
i, which implies ai = 0 for all i, hence a = 0. The equivalence a = 0 ⇔ A = 0
is obvious, and A = 0 ⇔ λ(A) = 0 follows from Corollary 3.4.
(ii) and (iii): Straightforward.
(cid:3)
Matrices A and A(k) for k = 0, . . . , n − 1 have the following useful prop-
erties.
Lemma 4.5. If A 6= 0, then (∀A ∈ Z C(a0, . . . , an−1))[(A/λ(A)) ≤ ( A/λ( A)].
Proof. Observe that A 6= 0 implies that A = 0 does not belong to the interval
matrix Z C(a0, . . . , an−1). Recalling that ai = min(ai, a) for all i we see that
ai ≤ a for all i and that ak = a for k such that ak = a. Hence λ( A) = a by
Proposition 3.3. Showing (A/λ(A)) ≤ ( A/λ( A) means showing
ai/ max
k
ak ≤ min(ai, a)/a ∀i.
To prove (10) we observe that it follows from the inequality
ai · a ≤ max
j
aj · min(ai, a) ∀i,
which is
when min(ai, a) = a, and
ai · a ≤ max
j
aj · a
ai · max
i
ai ≤ ai · max
j
aj
(10)
(11)
(12)
(13)
when min(ai, a) = ai. Both (12) and (13) are obvious. This shows (11) and
hence (10) and (A/λ(A)) ≤ ( A/λ( A)).
(cid:3)
19
Lemma 4.6. For any nonzero A ∈ Z C(a0, . . . , an−1) there exists A(k) 6= 0
for some k ∈ N0 such that [(A(k)/λ(A(k))) ≤ (A/λ(A))].
Proof. Let A = Z(a0, . . . , an−1) and let k be such that ak = maxj∈N aj. Con-
sider A(k). Since ak ≥ ak > 0 but the rest of the components defining A(k)
are ai ≤ ai for i 6= k, we have λ(A(k)) = ak and (A(k)/λ(A(k))) ≤ (A/λ(A)). (cid:3)
We now characterize possibly X-robust and universally X-robust interval
circulant matrices.
+ be an interval, and let Z C(a0, . . . , an−1) ⊆
be an interval circulant matrix containing A. Then Z C(a0, . . . , an−1)
Theorem 4.7. Let X ⊆ Rn
Rn×n
is possibly X-robust if and only if we have x(i) ∈ attr( A) for all i ∈ N.
+
Proof. We need to show that there exists A ∈ Z C(a0, . . . , an−1) such that
X ⊆ attr(A) if and only if x(i) ∈ attr( A) for all i ∈ N.
If A = 0 then
attr( A) = Rn
+ and the claim is obvious. Next we suppose that A 6= 0
which implies λ( A) 6= 0 by Corollary 3.4. By Proposition 4.4 part (i),
Z C(a0, . . . , an−1) contains only nonzero matrices in this case.
"If": By Lemma 4.2, the condition implies that X ⊆ attr( A). The claim
then follows since A ∈ Z C(a0, . . . , an−1).
"Only if": Let A ∈ Z C(a0, . . . , an−1) be such that X ⊆ attr(A).
By Lemma 4.5 we have (A/λ(A)) ≤ ( A/λ( A), and Theorem 3.10 yields
that x ∈ attr A. As x ∈ attr A for all x ∈ X, the claim then follows from
Lemma 4.2.
(cid:3)
Corollary 4.8. Let x ∈ Rn
be an interval
circulant matrix containing A. Then (∃A ∈ Z C(a0, . . . , an−1))[x ∈ attr(A)]
if and only if x ∈ attr( A).
+ and let Z C(a0, . . . , an−1) ⊆ Rn×n
+
Proof. Take X = {x} then the possible X-robustness means existence of
A ∈ Z C(a0, . . . , an−1) such that x ∈ attr(A) and x(i) = x for all i ∈ N. The
claim then follows from Theorem 4.7.
(cid:3)
+ be an interval, and let Z C(a0, . . . , an−1) ⊆
be an interval circulant matrix. Then Z C(a0, . . . , an−1) is universally
Theorem 4.9. Let X ⊆ Rn
Rn×n
X-robust if and only if x(j) ∈ attr(A(i)) for all i ∈ N0 and j ∈ N.
+
20
Proof. We need to show that X ⊆ attr(A) for all A ∈ Z C(a0, . . . , an−1) if
and only if x(j) ∈ attr(A(i)) for all i ∈ N0 and j ∈ N.
"If": Let x(j) ∈ attr(A(i)) hold for all i ∈ N0 and j ∈ N. Take A ∈
Z C(a0, . . . , an−1).
+. Otherwise, by
Lemma 4.6 there exists k ∈ N0 such that A(k) 6= 0 and (A(k)/λ(A(k))) ≤
(A/λ(A)). Applying Theorem 3.10 to (A(k)/λ(A(k))) and (A/λ(A)) we obtain
x(j) ∈ attr(A) for all nonzero x(j), hence X ⊆ attr(A).
If A = 0 then x(j) ∈ attr(A) = Rn
"Only if": Take a sequence {As}s≥1 ⊆ Z C(a0, . . . , an−1) such that
lims→∞ As = A(k), and take any x ∈ X. Since x ∈ attr(As) for all s, by
Proposition 3.8 we have λ(As)An2
s ⊗ x for all s, and by the conti-
nuity of the arithmetic operations of max-algebra we obtain λ(A(k))(A(k))n2 ⊗
x = (A(k))n2+1 ⊗ x. As x ∈ attr(A(k)) for all x ∈ X, the claim then follows
from Lemma 4.2.
(cid:3)
s ⊗ x = An2+1
Corollary 4.10. Let x ∈ Rn
be an inter-
val circulant matrix. . Then (∀A ∈ Z C(a0, . . . , an−1)) [x ∈ attr(A) ] if and
only if x ∈ attr(A(k)) for each k ∈ N0.
+, and let Z C(a0, . . . , an−1) ⊆ Rn×n
+
Proof. Take X = {x} then the universal X-robustness means that
x ∈ attr(A) for all A ∈ Z C(a0, . . . , an−1). The claim then follows from
Theorem 4.9.
(cid:3)
4.2. Tolerance and weak tolerance X−robustness
Theorem 4.11. Let X ⊆ Rn
Rn×n
X−robust if and only if (∀k ∈ N0)[(attr(A(k)) ∩ X) 6= ∅].
+ be a closed interval, and let Z C(a0, . . . , an−1) ⊆
be an interval circulant matrix. Then Z C(a0, . . . , an−1) is tolerance
+
Proof. "If": Take A ∈ Z C(a0, . . . , an−1). If A = 0 then attr(A) = Rn
+, hence
attr(A) ∩ X 6= ∅. Otherwise, for each i ∈ N0 take y(i) ∈ (X ∩ attr(A(i))], By
Lemma 4.6 there exists k ∈ N0 with (A(k)/λ(A(k))) ≤ (A/λ(A)). Applying
Theorem 3.10 to (A(k)/λ(A(k))) and (A/λ(A)) we obtain y(k) ∈ attr(A), hence
the implication.
"Only if": For any k ∈ N0 take a sequence {As}s≥1 ⊆ Z C(a0, . . . , an−1)
such that lims→∞ As = A(k). For each of these matrices there exists xs ∈ X
such that xs ∈ attr(As). Then by Proposition 3.8 we have λ(As)An2
s ⊗ xs =
An2+1
s ⊗xs for all s. Since X is compact, we can assume that lims→∞ xs exists
21
and denote it by y(k). Then we obtain that by the continuity of arithmetic
operations of max-algebra λ(A(k))(A(k))n2 ⊗ y(k) = (A(k))n2+1 ⊗ y(k). Hence
y(k) ∈ attr(A(k)).
(cid:3)
Corollary 4.12. Under the conditions of Theorem 4.11, Z C(a0, . . . , an−1)
is tolerance X−robust if and only if all systems
λ(A(k))(A(k))n2
⊗ y = (A(k))n2+1 ⊗ y,
y ∈ X,
(14)
with k ∈ N0 such that A(k) 6= 0 are solvable.
We now characterize the weak tolerance robust matrices.
+
+ be an interval and let Z C(a0, . . . , an−1) ⊆
Theorem 4.13. Let X ⊆ Rn
be an interval circulant matrix containing A. Then Z C(a0, . . . , an−1)
Rn×n
is weakly tolerance X−robust if and only if λ( A)( A)n2 ⊗ x = ( A)n2+1 ⊗ x is
solvable with x ∈ X.
Proof. By Corollary 4.8, x ∈ X and A ∈ Z C(a0, . . . , an−1) such that
x ∈ attr(A) exist if and only if x ∈ attr( A) for some x ∈ X. This, by
Proposition 3.8, is equivalent to λ( A)( A)n2 ⊗ x = ( A)n2+1 ⊗ x being solvable
with x ∈ X.
(cid:3)
4.3. Possible and tolerance Z C(a0, . . . , an−1)−robustness
We now characterize the remaining two types of robustness.
+
+ be an interval, and let Z C(a0, . . . , an−1) ⊆
be an interval circulant matrix. Then X is possibly Z C(a0, . . . , an−1)−robust
Theorem 4.14. Let X ⊆ Rn
Rn×n
if and only if there exists x ∈ X that satisfies λ(A(i))(A(i))n2⊗x = (A(i))n2+1⊗
x for all i ∈ N0 such that A(i) 6= 0.
Proof. By Corollary 4.10, x ∈ X belongs to attr(A) for all A ∈ Z C(a0 . . . , an−1)
if and only if it belongs to attr(A(i)) for all i ∈ N0 with A(i) 6= 0. By Propo-
sition 3.8 this is equivalent to x satisfying λ(A(i))(A(i))n2 ⊗ x = (A(i))n2+1 ⊗ x
for all such i.
(cid:3)
22
+
+ be an interval, and let Z C(a0, . . . , an−1) ⊆
Theorem 4.15. Let X ⊆ Rn
be an interval circulant matrix containing A. Then interval vector
Rn×n
X is tolerance Z C(a0, . . . , an−1)−robust if and only if Z C(a0, . . . , an−1) is
possibly X−robust.
Proof. Suppose that X is tolerance Z C(a0, . . . , an−1)−robust, then we have
the following
(∀x ∈ X)(∃A ∈ Z C(a0, . . . , an−1))[ x ∈ attr(A) ]
Cor.4.8
⇐⇒ (∀x ∈ X)[ x ∈ attr( A) ]
⇒ (∃A ∈ Z C(a0, . . . , an−1))(∀x ∈ X)[ x ∈ attr(A) ],
and hence we have that Z C(a0, . . . , an−1) is possibly X−robust.
The converse implication is trivial.
(cid:3)
4.4. Computational complexity
We close the section with a couple of remarks on the computational com-
plexity of the different types of interval robustness.
Remark 4.16. By Theorems 4.7 and 4.15 the verification of whether
(i) Z C(a0, . . . , an−1) is possibly X-robust,
(ii) Z C(a0, . . . , an−1) is universally X-robust,
(iii) X is tolerance Z C(a0, . . . , an−1)-robust
reduces, under some assumptions, to the verification whether some vectors
satisfy some two-sided max-linear systems with n2 and n2 + 1 powers of some
matrices. Hence these types of robustness are of polynomial complexity.
Remark 4.17. By Corollary 4.12, Theorem 4.14 and Theorem 4.15, verify-
ing whether
(i) Z C(a0, . . . , an−1) is tolerance X-robust,
(ii) Z C(a0, . . . , an−1) is weakly tolerance X-robust,
(iii) X is possibly Z C(a0, . . . , an−1)-robust
23
reduces, under some assumptions, to verifying the non-emptyness of solution
set of some system of max-affine inequalities, where some of the inequalities
(among those defining X) can be strict. This problem was generally shown
to be polynomially equivalent to solving a mean-payoff game [1], for which
efficient pseudopolynomial algorithms exist, but existence of a polynomial
algorithm has been a long-standing open question.
5. Proofs of Proposition 3.7 and Theorem 3.10
5.1. Cyclicity of circulants: Proof of Proposition 3.7
Let us start with the following elementary but useful statement.
Lemma 5.1. Let p1, . . . , ps, n ∈ N (the set of natural numbers). Then the
equation
p1x1 + . . . + psxs ≡ m(mod n)
(15)
has a solution (x1, . . . , xs) ∈ (N ∪ {0})s if and only if m is a multiple of
gcd(p1, . . . , ps, n).
Proof. "Only if": Observe that p1x1 + . . . + psxs and n are always multiples
of gcd(p1, . . . , ps, n), and if (15) holds then so is m as well.
"If": The claim is well known for s = 1 (elementary number theory).
The same fact also implies existence of xs ∈ N ∪ {0} such that
psxs ≡ m(mod gcd(n, p1, . . . , ps−1)).
(16)
We now prove the claim by induction assuming that it holds for s−1. Observe
that (16) implies that there exists also k ∈ N ∪ {0} such that
psxs + k gcd(n, p1, . . . , ps−1) ≡ m(mod n).
(17)
But by induction there exist x1 ∈ N ∪ {0}, . . . , xs−1 ∈ N ∪ {0} such that
p1x1 + . . . + ps−1xs−1 ≡ k gcd(n, p1, . . . , ps−1)(mod n).
Combining (17) and (18) we get the claim.
(18)
(cid:3)
Let us now introduce the following definition that appeared in [19] (see
also [4]).
24
Definition 5.2 (Visualized Matrices). A nonzero A ∈ Rn×n
+
is called
(i) visualized if Ai,j ≤ λ(A) for all i, j, and
(ii) strictly visualized if it is visualized and Ai,j = λ(A) if and only if (i, j) ∈
Gc(A).
By (3) we have that λ(A) = max(a0, a1, . . . , an−1) for A = Z(a0, . . . , an−1),
Ai,j for any circulant A. That is, any circulant
implying that λ(A) =
n
max
i,j=1
matrix is visualized. We will now argue that it is also strictly visualized.
Definition 5.3 (Threshold Digraphs). Let A ∈ Rn×n
and h ∈ R+. De-
fine the threshold digraph of A with respect to h as the subgraph of G(A)
containing all edges (i, j) with Ai,j ≥ h, and all nodes that are beginning
and end nodes of those edges. Denote this threshold graph by G(A, h).
+
Proposition 5.4. Let A ∈ Rn×n
strictly visualised, and Gc(A) = G(A, λ(A)).
+
be a nonzero circulant matrix. Then it is
Proof: By (3) no entry of A exceeds λ(A). Hence A is visualized. Also recall
that λ(A) > 0 by Corollary 3.4.
If Ai,j < λ(A) then the mean weight of any cycle with edge (i, j) is strictly
less than λ(A), so (i, j) is not critical. In other words, (i, j) being critical
implies Ai,j = λ(A).
It remains to show that if Ai,j = λ(A), which is equivalent to (i, j) being
an edge of G(A, λ(A)), then (i, j) is critical. In this case by Lemma 3.2 (i, j)
lies in a cycle with all edge weights equal to λ(A). The weights of all edges
in this cycle are equal to λ(A), hence the mean weight of this cycle is λ(A),
i.e., it is a critical cycle and (i, j) is critical. This completes the proof.
(cid:3)
Proof of Proposition 3.7. First observe that Proposition 5.4 implies that
Gc(A) = G(A, λ(A)) and hence the set of critical edges of a circulant matrix
A is given by
Ec(A) = {(i, j) : i = j if a0 = λ(A) or j − i ≡ pk(mod n), k ∈ {1, . . . , s}}
(19)
where p1, . . . , ps are such that ap1 = . . . = aps = λ(A) (and p1 > p2 > . . . >
ps).
25
We now consider the component of Gc(A) which contains node i, for i
from the set {1, . . . , gcd(n, p1, . . . , ps)}.
Let us argue that the node set of this component is given by
{k ∈ N : k ≡ i + l1p1 + . . . + lsps(mod n), l1, . . . , ls ∈ N ∪ {0}},
(20)
Indeed, by (19) edges (i, j) where j ≡ (l+pt)(mod n)) for some t ∈ {1, . . . , s}
are the only edges that issue from i and are critical. Using this observation,
the claim follows by simple induction.
Using Lemma 5.1 we now observe that (20) is the same as
{i + k gcd(n, p1, . . . , ps)) : k ∈ {0, . . . , (n/ gcd(n, p1, . . . , ps)) − 1}.
This set does not intersect with the node set of any component containing a
different node in {1, . . . , gcd(n, p1, . . . , ps)}, and this yields gcd(n, p1, . . . , ps)
strongly connected components of Gc(A). Isomorphism between two compo-
nents containing i1 ∈ {1, . . . , gcd(n, p1, . . . , ps)} and i2 ∈ {1, . . . , gcd(n, p1, . . . , ps)}
is induced by the following mapping on their set of nodes:
i1 + k gcd(n, p1, . . . , ps)) 7→ i2 + k gcd(n, p1, . . . , ps)).
This completes the proof of part (i) of Proposition 3.7.
If a0 = λ(A) then Gc(A) contains all loops of the form (i, i) for 1 ≤ i ≤ n,
and the cyclicity of every component of Gc(A) is 1 since it contains a loop.
When a0 < λ(A), we can use the result of [9] Theorem 3.3 part (i) since
this result describes the cyclicity of any component of the threshold digraph
G(A, λ(A)) (see [9] Theorem 3.1.), and since Gc(A) = G(A, λ(A)) by Propo-
sition 5.4. According to this result, that cyclicity is equal to any of the three
expressions given in (4). This completes the proof of part (ii).
(cid:3)
5.2. Inclusion of attraction cones: Proof of Theorem 3.10
Before considering the problem of our interest, let us recall the notion of
cyclic classes which will be necessary for some proofs.
Definition 5.5 (Cyclic Classes). Let G = (N, E) be a strongly connected
graph with cyclicity σ(G), and let i, j ∈ N. Nodes i, j are said to belong to
the same cyclic class if the lengths of some (and hence all) walks connecting
i to j are a multiple of σ(G).
26
2
1
3
6
4
5
2
1
3
6
4
5
Figure 1: Cyclic classes of two graphs of Example 5.6 (shown in different shades).
The cyclic class of i will be denoted by [i]. We also write [i] →1 [j] if
the lengths of some (and hence all) walks connecting a member of [i] to a
member of [j] have length congruent to 1 modulo σ(G).
By cyclic classes of a completely reducible digraph we mean cyclic classes
of its (strongly connected) components.
Example 5.6. Consider two associated graphs of 0-1 matrices of Exam-
ple 3.11 shown in Figure 1. On the left, the graph consists just of one cycle
of length 6, hence its cyclicity is 6 and the cyclic classes are {1}, {2}, {3},
{4}, {5} and {6}. On the right, the cyclicity of the graph is 2 and the cyclic
classes are {1, 3, 5} and {2, 4, 6}.
(cid:3)
Cyclic classes are also called components of imprimitivity. We refer the
reader to [3] Lemma 3.4.1 for a proof that belonging to the same cyclic class
is a well-defined equivalence relation.
Lemma 5.7. Let G be a strongly connected digraph.
(i) Let σ(G) > 1 and let i0, i1, . . . , ik be a walk on G. Then [il−1] →1 [il]
for each l ∈ {1, . . . , k}.
(ii) Let C be a cycle of G. Then C contains a member of each cyclic class
of G.
Proof. (i): Each edge is a walk of length 1. Therefore [il−1] →1 [il] for each
l ∈ {1, . . . , k}.
(ii): Let i be a node which is not in C. Let us show that C contains a node
27
in the cyclic class of i. Since G is strongly connected, there exists a walk
connecting i to a node j of C. If the length of this walk is a multiple of σ(G)
then j ∈ [i]. Otherwise, we concatenate this walk with a walk from j to some
node k ∈ C whose edges belong to C and such that the length of resulting
walk is a multiple of σ(G). Then k ∈ [i] and the claim is proved.
(cid:3)
We now derive a convenient form of a system defining the attraction cone
for circulant matrices, based on the results of [18]. Here At
i• denotes the ith
row of At. We also write i ∼A j when i and j belong to the same component
of Gc(A).
Proposition 5.8. Let A ∈ Rn×n
+
be a nonzero circulant matrix. Then
x ∈ attr(A) ⇔ An2
i• ⊗ x = An2
j• ⊗ x ∀i, j ∈ N s.t. [i] →1 [j]
and
x ∈ attr(A) ⇔ An2
i• ⊗ x = An2
j• ⊗ x ∀i, j ∈ N s.t. i ∼A j.
Proof. By Proposition 3.8
attr(A) = {x : An2
⊗ x = An2+1 ⊗ x}
(21)
(22)
(23)
j• = An2+1
i•
Since A is a circulant matrix, by Proposition 5.4 it is visualized, and then
by [18] Proposition 2.8 we also have An2
for any i, j ∈ Nc(A) such
that [i] →1 [j]. This shows (21). To show (22) recall that if a component of
Gc(A) has more than one cyclic class then for every two nodes i, j of the com-
ponent there is a walk i0 = i, i1, i2, . . . , ik = j on Gc(A) where [il−1] →1 [il]
for each l ∈ {1, . . . , k} by Lemma 5.7 part (i). Hence An2
j• ⊗x holds
for all nodes i, j in that component. If a component has only one cyclic class
then [18] Proposition 2.8 implies that all rows with indices in that component
are equal to each other, so the equations An2
j• ⊗ x hold trivially for
all pairs of nodes from that component.
(cid:3)
i• ⊗ x = An2
i• ⊗x = An2
It can be seen that we wrote out system (21) for all examples of Section 3.
In the case of Example 3.11, for which G(A) = Gc(A) and the cyclic classes
are shown on Figure 1, system (21) reduces to x1 = x2 = x3 = x4 = x5 = x6
for A and to x1 ⊕ x3 ⊕ x5 = x2 ⊕ x4 ⊕ x6 for B.
We will also need the following observations.
28
Lemma 5.9. Let A, B ∈ Rn×n
and A ≤ B. Then Gc(A) ⊆ Gc(B).
+
be two matrices such that λ(A) = λ(B) 6= 0
Proof. Since A ≤ B the mean weight of each cycle in B is not less than the
mean weight of the same cycle in A. If that cycle is critical in A then its
mean weight λ(A) cannot increase in B since λ(A) = λ(B). Hence it equals
λ(B), i.e., the cycle belongs to Gc(B).
(cid:3)
Lemma 5.10. Let A, B ∈ Rn×n
λ(B) 6= 0, A ≤ B. Then
+
be two circulant matrices with λ(A) =
x ∈ attr(B) ⇔ Bn2
i• ⊗ x = Bn2
j• ⊗ x ∀i, j ∈ N, s.t. i ∼A j.
Proof. By (22),
x ∈ attr(B) ⇔ Bn2
i• ⊗ x = Bn2
j• ⊗ x ∀i, j ∈ N s.t. i ∼B j.
(24)
(25)
We also have Gc(A) ⊆ Gc(B) by Lemma 5.9 and hence each x ∈ attr(B)
satisfies the system in (24).
Suppose now that x satisfies the system in(24). We will show that x also
satisfies
Bn2
i• ⊗ x = Bn2
j• ⊗ x ∀i, j ∈ N s.t. [i] →1 [j]
(26)
so that x ∈ attr(B) by Proposition 5.8. Since Gc(A) ⊆ Gc(B), each compo-
nent α of Gc(A) belongs to a component β of Gc(B), and each component of
Gc(B) contains a component of Gc(A) because Nc(A) = Nc(B) = N. Hence
it amounts to show that if x satisfies the subsystem of equations in (24) cor-
responding to a component α of Gc(A) then it also satisfies the subsystem
of equations in (26) corresponding to the component β of Gc(B) such that
α ⊆ β. But by Lemma 5.7 part (ii) each cyclic class of β has a member in
any cycle of β and hence in any cycle of α (because α ⊆ β). This shows that
for each i, j with [i] →1 [j] in Gc(B) there exist k ∈ [i] and l ∈ [j] on a cycle
of α and then Bn2
i• and
Bn2
(cid:3)
l• ⊗ x holds by (24). However, Bn2
j• by [18] Proposition 2.8. Hence the claim follows.
k• ⊗ x = Bn2
k• = Bn2
l• = Bn2
Let us now introduce Kleene stars, as they will also be useful in the proof
of Theorem 3.10.
29
Definition 5.11 (Kleene Stars). Let A ∈ Rn×n
+
have λ(A) ≤ 1. Then
A∗ = I ⊕ A ⊕ A2 ⊕ . . . ⊕ An−1
is called the Kleene star of A.
Proposition 5.12 (e.g., [4], Corollary 1.6.16). Let A ∈ Rn×n
A if and only if one of the following equivalent conditions hold:
+ . Then A∗ =
(i) A2 = A and Ai,i = 1 for all i ∈ N;
(ii) Ai,i = 1 and Ai,jAj,k ≤ Ai,k for all i, j, k ∈ N.
More specifically, we will make use of the following.
Lemma 5.13. Let A 6= 0 be a circulant matrix. Then (A/λ(A))n2 is a
Kleene star.
Proof. Note that λ(A) 6= 0 by Corollary 3.4. By Proposition 5.12 it suffices to
show that (A/λ(A))n2 is an idempotent matrix and that ((A/λ(A))n2)i,i = 1
for all i. For the idempotency, observe that by Proposition 3.7 part (ii) per(A)
divides n2, and that T (A) ≤ n2 by Proposition 3.6. Hence (A/λ(A))2n2 =
(A/λ(A))n2.
For the remaining part of the claim, assume λ(A) = 1 and recall that for
any t ≥ 1 and any i, j ∈ N, entry (At)i,j is equal to the greatest weight
of a walk of length t connecting i to j (e.g.,[4], Example 1.2.3). Take
i ∈ {1, . . . , n} and observe that G(A) contains a critical cycle of length n
going through i. The weights of all entries of that cycle equal to 1. Taking
n copies of this cycle we obtain a cycle in G(A) of weight 1 and length n2.
The claim ((A/λ(A))n2)i,i = 1 follows since the weights of all entries and
(therefore) of all walks are bounded by 1.
(cid:3)
We are now ready to prove the main result of Section 3.
The proof will also make use of the following notation.
Definition 5.14. Denote by k[mod n], respectively by k[mod′ n], the only
number in N0 = {0, . . . , n − 1}, respectively in N = {1, . . . , n}, which is
congruent to k modulo n.
30
Proof of Theorem 3.10. The case λ(A) = λ(B) = 0 is trivial since in
that case A = B = 0 by Corollary 3.4 and hence attr(A) = attr(B) = Rn
+.
Otherwise, as attr(A/λ(A)) = attr(A) and attr(B/λ(B)) = attr(B) (which
follows, e.g., from Proposition 2.22), we can assume without loss of generality
λ(A) = λ(B) = 1 and consider matrices C = An2 and D = Bn2. By
Proposition 3.1 C and D are circulants, hence C = Z(c0, . . . , cn−1) and D =
Z(d0, . . . , dn−1) for some c0, . . . , cn−1 and d0, . . . , dn−1. Using A ⊕ B = B and
the expansion for (A ⊕ B)n2 we obtain An2 ≤ (A ⊕ B)n2 = Bn2, thus C ≤ D.
By Lemma 5.13 both of them are also Kleene stars. By Proposition 5.12 we
have D1,(α+γ)[mod′ n] ≥ D1,α · Dα,(α+γ)[mod′ n] and hence
d(α+γ−1)[mod n] ≥ dα−1 · dγ
(27)
for any α ∈ {1, . . . , n} and γ ∈ {0, . . . , n − 1}. In what follows we are going
to show that the assumption that attr(A) ⊆ attr(B) does not hold leads to
a contradiction with (27) for some α and γ.
By Lemma 5.9 we have Gc(A) ⊆ Gc(B). By Proposition 3.7 part (i), Gc(A)
consists of l components whose node sets are of the form
{k, k + l, k + 2l, . . . , k + (n/l − 1)l} for k ∈ {1, . . . , l},
(28)
where l is a divisor of n. Each of these node sets belongs to some component
of Gc(B).
By Proposition 5.8 x ∈ attr(A) if and only if
Ck• ⊗ x = Ck+l• ⊗ x = . . . = Ck+(n/l−1)l• ⊗ x for k ∈ {1, . . . , l},
(29)
and by Lemma 5.10 x ∈ attr(B) if and only if
Dk• ⊗ x = Dk+l• ⊗ x = . . . = Dk+(n/l−1)l• ⊗ x for k ∈ {1, . . . , l},
(30)
We will refer to (29) or (30) for fixed k as to a chain of equations.
Suppose by contradiction that x ∈ attr(A) but x /∈ attr(B). The latter
means that there exist k and s such that Dk• ⊗ x > Dk+ls• ⊗ x for some
integers k and s. Assume without loss of generality that k = 1 then
D1• ⊗ x = d0x1 ⊕ d1x2 ⊕ . . . ⊕ dn−1xn.
Let dα−1 ·xα be one of the terms where the maximum in the above expres-
sion is attained. In D1+ls• ⊗ x we find a term dα−1 · xβ where α ≡ β(mod l),
and we have the inequality dα−1 · xα > dα−1 · xβ and hence xα > xβ.
31
Observe that c0 = d0 = 1 since C and D are Kleene stars. Since
α ≡ β(mod l) there exists a chain of equations among those of (29), which
contains both c0xα = xα and c0xβ = xβ. The corresponding chain of equa-
tions holds (since x ∈ attr(A)), but xα > xβ and therefore in the expres-
sion containing c0xβ there is a term cγx(β+γ)[mod′ n] (for some γ) such that
cγx(β+γ)[mod′ n] ≥ xα > 0, and hence
dγx(β+γ)[mod′ n] ≥ xα.
(31)
Going back to the terms in the inequality D1,•x > D1+ls,•x and know-
ing that the maximum in D1,•x is attained at dα−1xα and D1+ls,•x con-
tains a term of the form dα−1xβ, we see that D1+ls,•x also contains the term
d(α+γ−1)[mod n]x(β+γ)[mod′ n] and that
dα−1xα > d(α+γ−1)[mod n]x(β+γ)[mod′ n].
(32)
Multiplying (31) by dα−1, combining with (32) and canceling x(β+γ)[mod′ n] >
0 we have
dα−1dγ > d(α+γ−1)[mod n],
which contradicts with the Kleene star property (27). The proof is com-
plete.
(cid:3)
6. Acknowledgement
We are grateful to the referees for their careful reading as well as numerous
questions and helpful comments on the initial version of the paper, and we
also would like to thank Dr. Michelle Delcourt for some useful suggestions
regarding Abstract and Introduction.
References
[1] X. Allamigeon, A. Legay, U. Fahrenberg, R. Katz, S. Gaubert. Tropical
Fourier-Motzkin elimination, with an application to real-time verifica-
tion. Internat. J. of Algebra and Computation 24:5 (2014) 569-607.
[2] F.L. Baccelli, G. Cohen, G.J. Olsder and J.P. Quadrat.
Syn-
chronization and Linearity. Wiley and Sons, 1992. Available online:
https://www.rocq.inria.fr/metalau/cohen/documents/BCOQ-book.pdf
32
[3] R.A. Brualdi and H.J. Ryser. Combinatorial Matrix Theory. Cambridge
Univ. Press, 1991.
[4] P. Butkovic, Max-linear Systems: Theory and Algorithms. Springer,
London, 2010.
[5] G. Cohen, D. Dubois, J.P. Quadrat, M. Viot. Analyse du comportement
p´eriodique de syst`emes de production par la th´eorie des dioıdes. INRIA,
Rapport de Recherche No. 191, F´evrier, 1983.
[6] P.J. Davis. Circulant Matrices. Wiley, 1979
[7] M. Fiedler, J. Nedoma, J. Ram´ık, J. Rohn, K. Zimmermann. Linear
Optimization Problems with Inexact Data. Springer, Berlin, 2006.
[8] M. Gavalec, J. Plavka, D. Ponce Tolerance types of interval eigenvectors
in max-plus algebra. Information Science 367-368 (2016) 14 -- 27.
[9] M. Gavalec. Periods of special fuzzy matrices. Tatra Mt. Math. Publ.
16 (1999) 47-60.
[10] M. Gavalec. Periodicity in Extremal Algebra. Gaudeamus, Hradec
Kr´alov´e 2004.
[11] B. Heidergott, G.-J. Olsder, and J. van der Woude. Max-plus at Work.
Princeton Univ. Press, 2005.
[12] G.L. Litvinov, A.N. Sobolevskiı. Idempotent interval analysis and opti-
mization problems. Reliable Computing 7 (2001) 353 -- 377.
[13] G. Merlet, T. Nowak, H. Schneider and S. Sergeev. Generalizations of
bounds on the index of convergence to weighted digraphs. Discr. Appl.
Math. 178 (2014) 121 -- 134.
[14] M. Moln´arov´a, H. Myskov´a, J. Plavka. The robustness of interval fuzzy
matrices. Linear Algebra and Its Applications 438 (2013) 3350 -- 3364.
[15] H. Myskov´a, J. Plavka. The robustness of interval matrices in max-plus
algebra. Linear Algebra and Its Applications 445 (2014) 85 -- 102.
[16] J. Plavka. On eigenproblem for circulant matrices in max-algebra. Op-
timization 50:5-6 (2001) 477-483.
33
[17] J. Rohn. Solvability of systems of linear interval equations, SIAM J. on
Matrix Anal. Appl. 25:1 (2003) 237 -- 245.
[18] S. Sergeev. Max-algebraic cones of nonnegative irreducible matrices.
Linear Algebra and its Applications 435 (2011) 1736-1757.
[19] S. Sergeev, H. Schneider and P. Butkovic. On visualization scaling,
subeigenvectors and Kleene stars in max algebra. Linear Algebra and
Its Applications 431 (2009) 2395 -- 2406.
[20] S. P. Shary. A new technique in systems analysis under interval uncer-
tainty and ambiguity. Reliable Computing 8 (2002) 321 -- 418.
[21] R. M. Tanner at al. LDPC Block and Convolutional Codes Based on
Circulant Matrices. IEEE Trans. on Inform. Theory 50:12 (2004) 2966 --
2984.
[22] H. Tom´askov´a. Eigenproblem for circulant matrices in max-plus alge-
bra. In: Proceedings of the 29th Conference on Mathematical Methods,
Computational Techniques, Intelligent Systems (MAMECTIS-29), 2010.
34
|
1904.00033 | 2 | 1904 | 2019-10-21T15:40:40 | On graded Brown--McCoy radicals of graded rings | [
"math.RA"
] | We investigate the graded Brown--McCoy and the classical Brown--McCoy radical of a graded ring, which is the direct sum of a family of its additive subgroups indexed by a nonempty set, under the assumption that the product of homogeneous elements is again homogeneous. There are two kinds of the graded Brown--McCoy radical, the graded Brown--McCoy and the large graded Brown--McCoy radical of a graded ring. Several characterizations of the graded Brown--McCoy radical are given, and it is proved that the large graded Brown--McCoy radical of a graded ring is the largest homogeneous ideal contained in the classical Brown--McCoy radical of that ring. | math.RA | math |
On graded Brown -- McCoy radicals of graded rings
Emil Ili´c-Georgijevi´c
Abstract
We investigate the graded Brown -- McCoy and the classical Brown --
McCoy radical of a graded ring, which is the direct sum of a family of
its additive subgroups indexed by a nonempty set, under the assump-
tion that the product of homogeneous elements is again homogeneous.
There are two kinds of the graded Brown -- McCoy radical, the graded
Brown -- McCoy and the large graded Brown -- McCoy radical of a graded
ring. Several characterizations of the graded Brown -- McCoy radical are
given, and it is proved that the large graded Brown -- McCoy radical of a
graded ring is the largest homogeneous ideal contained in the classical
Brown -- McCoy radical of that ring.
1
Introduction
Graded radicals and radicals of group-graded rings have been investigated
by many authors. Particularly, the graded Brown -- McCoy and the Brown --
McCoy radical (classical) of group-graded rings have been studied recently,
as well as in the recent decades, regarding many important open problems.
For example, in [20], results related to the Brown -- McCoy radical of a ring
graded by the additive group of integers are obtained in order to examine
open problems on the Brown -- McCoy radical of a polynomial ring (see also
[21]). These are, on the other hand, related to the famous Kothe's Conjec-
ture (consult also references of [20]). In [8], G-systems are introduced, where
G is a group, and their Brown -- McCoy radicals are investigated, while in [4],
the graded Brown -- McCoy and the Brown -- McCoy radical of G-graded rings
are studied. For research related to some problems on monomial rings, see
[12]. There is, however, a more general notion of a graded ring and the aim
of this paper is to study the graded Brown -- McCoy and the classical Brown --
McCoy radical of such rings. Results dealing with the Jacobson radical of
these rings can be found in [16]; see also [14] and [15].
Definition 1.1 ([16, 15]). Let R be a ring, and S a partial groupoid, that
is, a set with a partial binary operation. Also, let {Rs}s∈S be a family
of additive subgroups of R. We say that R = Ls∈S Rs is S-graded and
R induces S (or R is an S-graded ring inducing S) if the following two
conditions hold:
i) RsRt ⊆ Rst whenever st is defined;
ii) RsRt 6= 0 implies that the product st is defined.
2010 Mathematics Subject Classification 16W50, 16N80
Key words and phrases. Graded rings and modules, Brown -- McCoy radical
1
An S-graded ring inducing S from above is also called a homogeneous sum
[16, 15]. However, we will call it simply a graded ring in the sequel. If R =
Ls∈S Rs is a graded ring, the set A = Ss∈S Rs is called the homogeneous
part of R, and elements of A are called homogeneous elements of R. The
previous definition applies to both associative and nonassociative rings, but
all rings in this paper are assumed to be associative.
Clearly, every usual group (semigroup, groupoid) graded ring is graded in
the above sense. However, there are many other important examples of
such rings. For instance, let A, B be rings and V, W be an (A, B)-bimodule
and a (B, A)-bimodule, respectively, and let the quadruple (A, V, W, B) be
a Morita context, that is, the set R = (cid:18) A V
W B (cid:19) of matrices forms a
ring under matrix addition and multiplication (see, for instance, [7]). Then
R = L(i,j)∈{1,2}2 R(i,j) is a graded ring in the above sense if R(i,j) denotes
the set of matrices of R whose all entries are zero except the (i, j) entry
which does not have to be zero. The following are also examples of graded
rings in the above sense: a semidirect sum of rings (see [15]), particular case
of which is the Dorroh extension of a ring, a ring which is the direct sum of
its left ideals, a path algebra (see [15]).
A notion of a graded ring, equivalent to that in Definition 1.1, was studied
in [6, 10, 18] from a different point of view. Namely, let R be a graded
ring and A its homogeneous part. Consider A with induced partial addition
and induced multiplication from R. Then A is called an anneid [6, 10, 18].
Origins of this approach can be found in [17]. Anneid A has a partial
addition since the sum of nonzero elements x, y ∈ A does not have to be a
homogeneous element of R. However, if x + y belongs to A, elements x and
y are called addable and we write x#y [6, 10, 18]. Multiplication is defined
everywhere, according to the very definition of a graded ring. As we will
see in the next section, studying graded rings is equivalent to studying their
corresponding anneids [6, 10, 18]. We will therefore, in this paper, conduct
research on anneids, all of which are assumed to be associative.
The graded Jacobson radical of an anneid is thoroughly investigated in [9,
10]. Similar studies on 2-graded rings can be found in [22]. Here, we use
results of [9, 10] in order to introduce and investigate the graded Brown --
McCoy radical of an anneid.
Inspired by [9, 10], we give two notions of
this radical, the graded Brown -- McCoy and the large graded Brown -- McCoy
radical of an anneid. Several characterizations of the graded Brown -- McCoy
radical of an anneid are given, and as the main result we prove that the
large graded Brown -- McCoy radical of an anneid is the homogeneous part
of the largest homogeneous ideal contained in the classical Brown -- McCoy
radical of the corresponding graded ring (Theorem 3.24).
Obtained results can be easily translated to the language of graded rings,
and represent generalizations of results which hold for usual group-graded
rings [8, 4].
2
2 Preliminaries
Here, we give some notions and properties related to graded rings described
above. Everything stated in this section can be found in more detail in [18]
and references therein (particularly [6, 10]) or in [19, 23].
Theorem 2.1 ([6, 10, 18]). Anneid A can be characterized by the following
axioms:
a1) A is a groupoid with respect to multiplication and possesses an element
0 for which we have a0 = 0 = 0a for all a ∈ A;
a2) 0 is addable with every element of A, and the relation of addibility is
reflexive and almost transitive: (∀a, b, c ∈ A) a#b∧b#c∧b 6= 0 ⇒ a#c;
a3) for all 0 6= a ∈ A, the set A(a) of all elements from A, which are
addable with a, is a commutative group with respect to addition induced
by that of A and such a group is called the addibility group of a;
a4) for all a, b, c ∈ A, a#b implies ca#cb, ac#bc, c(a + b) = ca + cb,
(a + b)c = ac + bc.
Clearly, 0 is the neutral element of all addibility groups. Also, two distinct
addibility groups have only one common element, and it is 0. Let ∆∗ denote
the set of addibility groups of A and let ∆ = ∆∗ ∪ 0. If 0 6= δ ∈ ∆, then let
A(δ) denote the addibility group of A that defines δ. Otherwise, that is, if
δ = 0, A(δ) = 0. Then, if ξ, η ∈ ∆, we put
ξη = (cid:26) 0,
ζ,
if A(ξ)A(η) = 0;
if A(ξ)A(η) ⊆ A(ζ)
(see, for instance, [6]). This multiplication on ∆ is justified since it is known
that if x, y, a, b ∈ A, then x#y and a#b imply xa#yb (see, for instance,
[18]). Now, to each anneid A, we may associate a graded ring A, called the
linearization of A, in the following manner. We will follow the exposition
from [10] (see also [6]). Define A to be the direct sum of A(δ), where δ
runs through ∆∗. The multiplication of A is obtained by extending linearly
the multiplication of A. The set ∆∗ plays the role of S from Definition 1.1.
Product of two elements of ∆∗ is defined if and only if their product is
distinct from 0 in ∆, in which case these products coincide. Then A =
Lδ∈∆∗ A(δ) = Lδ∈∆ A(δ) is a graded ring whose homogeneous part is A. If
0 6= a ∈ A, then the degree of a [6, 10, 18], denoted by δ(a), is the element
of ∆ such that a ∈ A(δ(a)). It is assumed to be equal to 0 if a = 0. If
a 6= 0, then of course, A(δ(a)) = A(a). Hence δ(ab) = δ(a)δ(b) if ab 6= 0, and
δ(a)(δ(b)δ(c)) = (δ(a)δ(b))δ(c) if abc 6= 0 [6, 10, 18]. If δ is an idempotent
element of ∆∗, that is, if δδ = δ, then clearly, A(δ) is a ring with respect to
operations induced from A.
3
Definition 2.2 ([6, 10, 18]). An anneid A is said to be regular if each of the
conditions ac#bc or ca#cb implies a#b for ac 6= 0, bc 6= 0, ca 6= 0, cb 6= 0,
a, b, c ∈ A.
In the language of graded rings, regularity is equivalent to the notion of
cancellativity [16, 15].
A nonempty subset I of an anneid A is called a right ideal of A [6, 10, 18]
if: i) for all a, b ∈ I, for which a#b, we have a − b ∈ I; ii) for all a ∈ I and
x ∈ A, we have ax ∈ I. A left ideal of an anneid A is defined similarly, and a
subset of A which is a left and a right ideal is called an ideal or a two-sided
ideal. If a nonempty subset I of A satisfies the first condition only, then I
is called a subanneid of A.
If A and A′ are two anneids, then the mapping f : A → A′ is called a
homomorphism [6, 10, 18] if for all a, b ∈ A :
i) a#b ⇒ f (a)#f (b) ∧ f (a + b) = f (a) + f (b);
ii) f (ab) = f (a)f (b);
iii) f (a)#f (b) ∧ f (a), f (b) 6= 0 ⇒ a#b.
Let I be an ideal (left, right, two-sided) of an anneid A and let us define
a ∼ b if and only if either both a and b belong to I or a#b and a − b ∈ I. It
is easy to verify that ∼ is an equivalence relation on A. We also say that a
and b are congruent modulo I and write a ∼ b mod I. Denote the set A/ ∼
by A/I. Clearly, [a]∼ = a + I = {a + x x ∈ I ∧ a#x}. Now, let I be a
two-sided ideal, and let f : A → A/I be the mapping defined by f (a) = [a]∼.
Put [a]∼[b]∼ := f (ab), a, b ∈ A. We say that [a]∼ and [b]∼ are addable and
write [a]∼#[b]∼ if a#b. If [a]∼#[b]∼, we put [a]∼ + [b]∼ := f (a + b). Then
A/I becomes an anneid, called the factor anneid [6, 10, 18] (see also [13]).
Also, f is a homomorphism of anneids and if A is regular, A/I is regular
too. Isomorphism theorems for anneids are analogous to those for ordinary
rings.
3 The graded Brown -- McCoy radical
Let A be an anneid. We will fix some notation. If I is a right ideal of A, then
I denotes the largest ideal of A contained in I. The class of simple regular
anneids with unity, that is, of regular anneids with unity whose only ideals
are 0 and the anneid itself, will be denoted by M. Also, if a is an element
of A, hai denotes the ideal of A generated by a, that is, the set comprised of
i, where
i ∈ A, and where all the summands are mutually
all elements of the form na + Pfinite axi + Pfinite yia + Pfinite x′
iay′
n is an integer, xi, yi, x′
addable [6, 10, 18].
Following [1], it makes sense to define the graded Brown -- McCoy radical
i, y′
4
G(A) of an anneid A to be the intersection of all maximal right ideals I of
A such that A/ I ∈ M.
Remark 3.1. The linearization G(A) = Lδ∈∆∗ G(A) ∩ A(δ) of G(A) is a
homogeneous (graded) ideal of a graded ring A. This graded ideal will be
referred to as a graded Brown -- McCoy radical of a graded ring. We could
have simply called G(A) the Brown -- McCoy radical of an anneid A, just as
it was done in [9, 10] in the case of the graded Jacobson radical of an anneid,
which was called just radical in accordance with [11]. We added the adjective
'graded' in order to avoid confusion with the classical Brown -- McCoy radical
of a ring while discussing relations between these notions.
A right ideal I of an anneid A is called a modular right ideal if there exists
e ∈ A such that ea ∼ a mod I for all a ∈ A [9]. If I is a proper modular
right ideal, then the degree of e is an idempotent element of ∆∗ [9, 10]. This
notion served in a description of the graded Jacobson radical of a regular
anneid in [9, 10] by analogy with classical rings. In order to do the same
with the graded Brown -- McCoy radical of a regular anneid, we introduce the
following natural notion.
Definition 3.2. An ideal I of an anneid A is called modular if there exists
an element e ∈ A such that for all a ∈ A, ea ∼ a mod I and ae ∼ a mod I.
Element e is called a unity modulo I. We also say that an ideal I is modular
with respect to e.
Remark 3.3. If I is a proper modular ideal of an anneid A with respect
to e, then δ(e), the degree of e, is an idempotent element of ∆∗, that is,
δ(e)δ(e) = δ(e). We proceed as in [10]. Indeed, since e is a unity modulo I,
for all a ∈ A, either ea, a ∈ I or ea#a and ea − a ∈ I, and, either ae, a ∈ I
or ae#a and ae − a ∈ I. Since I is proper, e /∈ I. Also, e2#e and e2 − e ∈ I.
Since e /∈ I, we have that e2 6= 0. This and the fact that e2 and e belong to
the same addibility group, that is, e2#e, imply the claim.
The proof of the following lemma is included for the sake of completeness.
Lemma 3.4. An ideal I of an anneid A is a maximal modular ideal if and
only if A/I is a simple anneid with unity.
Proof. Let I be a maximal modular ideal and e a unity modulo I. If a + I
is an arbitrary element of A/I, then, since e is a unity modulo I, ea + I =
a + I = ae + I. Hence (e + I)(a + I) = (a + I)(e + I) = a + I, which proves
that e + I is a unity of A/I. Clearly, A/I is a simple anneid, since I is a
maximal ideal.
Conversely, if A/I is a simple anneid with unity e + I, then, for all a ∈ A, we
have (e + I)(a + I) = (a + I)(e + I) = a + I, that is, ea + I = a + I = ae + I.
Hence ea ∼ a mod I and ae ∼ a mod I. Therefore I is a modular ideal.
Also, since A/I is simple, I is a maximal ideal.
5
Corollary 3.5. If A is a regular anneid, then G(A) coincides with the
intersection of all maximal right ideals I of A such that the ideals I are
modular.
Proof. If I is a maximal right ideal of A such that I is modular, then A/ I
is a simple anneid with unity. Moreover, A/ I is a regular anneid since A is
regular. Conversely, if I is a maximal right ideal of A such that A/ I is a
regular simple anneid with unity, then I is a modular ideal of A.
Remark 3.6. If I is a maximal right ideal such that I is modular with respect
to e, then e is also a unity modulo I, that is, ea ∼ a mod I and ae ∼ a
mod I for all a ∈ A.
It is sometimes more pleasable to describe radicals of rings by their modules.
It is shown in [2] that any special radical of an associative ring may be defined
by some class of modules. For analogous results on special radicals of group-
graded rings, see [3]. Here, we aim to describe the graded Brown -- McCoy
radical of an anneid A by the means of right A-moduloids, and so in the
next paragraph we recall the notion of a moduloid.
If a group is the direct sum of a family of its subgroups, indexed by a
nonempty set, then such a group is called a graded group [18]. Let M =
Ld∈D M d be a commutative graded group and A = Lδ∈∆ Aδ a graded ring,
where D and ∆ are nonempty sets. Moreover, suppose that M is a right
A-module. A right A-module M is called graded [6, 10, 18] if for all d ∈ D
and δ ∈ ∆ there exists t ∈ D such that M dAδ ⊆ M t. A right A-moduloid
[6, 10, 18] M is just the homogeneous part of M , that is, M = Sd∈D M d, with
induced partial addition from M and induced outer operation M × A → M
from M × A → M , where A is the corresponding anneid of A. Similarly
to the case of anneids, to an A-moduloid M one can associate a graded
A-module M , called the linearization of an A-moduloid M, where A is the
linearization of an anneid A (see, for instance, [19]).
A right A-moduloid M is said to be regular [6, 10, 18] if xa#xb, xa, xb 6= 0,
implies a#b, where x ∈ M, a, b ∈ A.
If M and M ′ are two right A-moduloids, then the mapping f : M → M ′
is called a homomorphism [6, 10, 18] if for all x, y ∈ M, a ∈ A : i) x#y ⇒
f (x)#f (y) ∧ f (x + y) = f (x) + f (y); ii) f (xa) = f (x)a; iii) f (x)#f (y) ∧
f (x), f (y) 6= 0 ⇒ x#y.
Submoduloids are defined as usual (see [6]). A right A-moduloid M is said
to be irreducible [10] if M A 6= 0 and only submoduloids of M are 0 and
M. If x is an element of a right A-moduloid, then (0 : x) denotes the set
{a ∈ A xa = 0}, which is a right ideal of an anneid A [10]. Correspondingly,
the annihilator of a right A-moduloid M, that is, the set {a ∈ A M a = 0},
will be denoted by (0 : M )A or just by (0 : M ) (see [10]). If A is an anneid
and I a right ideal of A, then A/I is a right A-moduloid with respect to
(a+I)+(b+I) = (a+b)+I if a#b, and (a+I)c = ac+I for a+I, b+I ∈ A/I,
6
c ∈ A (see [10]). Also, f : A → A/I, defined by f (a) = a + I for a ∈ A, is a
homomorphism of right A-moduloids [10].
The notion of a simplicity is commonly used to mean that an algebraic
structure in hand has only trivial substructures. Inspired by the notion from
[3] for usual group-graded modules, we give simplicity a different meaning
in the following definition.
Definition 3.7. An irreducible right A-moduloid M over an anneid A is
called simple if M A 6= 0 and if for all ideals I of A for which M I 6= 0, there
exists b ∈ I = Lδ∈∆∗ A(δ) ∩ I such that xb = x for all x ∈ M.
We will soon see, after proving two lemmas, that the graded Brown -- McCoy
radical of a regular anneid A coincides with the intersection of annihilators
of all regular simple right A-moduloids. Therefore, inspired by the notion of
the large graded Jacobson radical of an anneid from [9, 10], it makes sense
to introduce the following notion.
Definition 3.8. The intersection of annihilators of all simple right A-modul-
oids, which are not necessarily regular, is called the large graded Brown --
McCoy radical of an anneid A and is denoted by Gl(A).
Lemma 3.9. Let M be a regular simple right A-moduloid. Then, if I is an
ideal of A such that M I 6= 0, then there exists b ∈ I such that xb = x for all
x ∈ M, and, moreover, the degree of b is an idempotent element of ∆∗.
Proof. Let M be a regular simple right A-moduloid, and I an ideal of A
such that M I 6= 0. By definition, there exists b ∈ I such that xb = x for all
x ∈ M. Then we have b = b1 +· · ·+bn for some natural number n, and where
bi ∈ I, i ∈ {1, . . . , n}. For an arbitrary x ∈ M, x = xb = x(b1 + · · · + bn) =
xb1 + · · · + xbn. This means that xbi are mutually addable, and since M is
regular, bi are mutually addable. Therefore b = b1 + · · · + bn ∈ I. Let us
now prove that the degree of b is an idempotent element of ∆∗. Indeed, since
xb = x for all x ∈ M, we have xb2 = xb 6= 0, for x 6= 0. Now, the regularity
of M implies b2#b, and the claim follows.
Lemma 3.10. If M is a regular simple right A-moduloid, then M ∼= A/I,
where I is a maximal right ideal of A such that I is a modular ideal of A.
Conversely, if I is a maximal right ideal of A such that I is a modular ideal
of A, then A/I is a simple right A-moduloid.
Proof. Let M be a regular simple right A-moduloid. Then, since M A 6= 0,
and since M is regular, there exist 0 6= x ∈ M and a ∈ A such that xa 6= 0.
Therefore M hai 6= 0. Since M is simple and regular, Lemma 3.9 implies
that there exists a′ ∈ hai such that ma′ = m for all m ∈ M. Let x 6= 0 be
an element of M. Since M is irreducible, we know from [10] that M = xA,
by analogy with the case of classical modules [11]. On the other hand,
7
from [10] we know that xA ∼= A/(0 : x) by the first isomorphism theorem
for moduloids (for the theorem, see [6, 10]). Therefore, according to [10],
(0 : x) is a maximal modular right ideal of A, that is, there exists e ∈ A
such that ea ∼ a mod (0 : x) for all a ∈ A. Denote (0 : x) by I. It suffices
now to prove that I is a modular ideal of A. Since I is a modular right
ideal of A, we have that I = {a ∈ A Aa ⊆ I}, by analogy with classical
rings (see, for instance, [5]). We already know that there exists a′ ∈ A
such that ma′ = m for every m ∈ M, and particularly, xa′ = x. Also, for
every b ∈ A, and every c ∈ A, we have x(cba′) = (xcb)a′ = xcb. Then, if
x(cba′) = xcb 6= 0, the regularity of M implies cba′#cb and ba′#b. Hence
cba′ − cb ∈ I and ba′ − b ∈ I. If x(cba′) = xcb = 0, then, of course, both cba′
and cb belong to I, meaning that ba′, b ∈ I. Also, for every b ∈ A, and every
c ∈ A, x(ca′b) = (xca′)b = xcb, which again, by the regularity of M, implies
ca′b − cb ∈ I if x(ca′b) = xcb 6= 0, and ca′b, cb ∈ I if x(ca′b) = xcb = 0.
Therefore I is a modular ideal of A.
Conversely, if I is a maximal right ideal of A such that I is modular, it is
clear that M = A/I is an irreducible right A-moduloid. Let us recall, if I
is a modular ideal with respect to e, then e is a unity modulo I. If K is an
ideal of A such that M K 6= 0, then M K = M. Hence there exists f ∈ K
such that e + I = f + I, and so for all a + I ∈ M, af + I = a + I. Therefore
M is a simple right A-moduloid.
Corollary 3.11. The graded Brown -- McCoy radical of a regular anneid A
coincides with the intersection of annihilators of all regular simple right A-
moduloids.
Remark 3.12. If A is a regular anneid, notice that we have Gl(A) ⊆ G(A).
Let us now deal with the elementwise characterization of the graded Brown --
McCoy radical of a regular anneid, inspired by a similar study on the graded
Jacobson radical of a regular anneid from [9, 10].
The following lemma follows the proof of the analogous lemma for modular
right ideals of regular anneids from [10].
Lemma 3.13. If I is a proper modular ideal of a regular anneid A, then all
unities modulo I have the same degree. Moreover, this degree is an idempo-
tent element of ∆∗.
Proof. Let f : A → A/I be the canonical mapping, and e, e′ be two unities
modulo I. Then for all a ∈ A, f (a) = f (ea) = f (e′a) = f (ae) = f (ae′).
Hence in particular, ¯0 6= f (ea)#f (e′a) 6= ¯0 if a /∈ I. Since f is a homo-
morphism, it follows that ea#e′a. Now, the regularity of A implies e#e′.
Therefore, according to Remark 3.3, e and e′ have the same degree, which
is an idempotent element of ∆∗.
Definition 3.14. The degree of all unities modulo a proper modular ideal
I of a regular anneid is called the degree of I.
8
It is stated in [9] and proved in [10] that there exists a one-to-one correspon-
dence between the maximal modular right ideals of a regular anneid A of
degree ǫ and the maximal modular right ideals of A(ǫ). Here, we prove the
same for maximal modular ideals if we additionally assume that ∆ is such
that the product of nonidempotent elements cannot be a nonzero idempo-
tent. We assume the same in Theorem 3.17, Theorem 3.18, Corollary 3.19
and Theorem 3.20.
Theorem 3.15. Let A be a regular anneid and ǫ an idempotent element of
∆∗. If I is a maximal right ideal of A such that I is a modular ideal of A of
degree ǫ, then Iǫ = I ∩ A(ǫ) is a maximal right ideal of A(ǫ) such that Iǫ is
a modular ideal of A(ǫ). If I is a maximal right ideal of A(ǫ) such that I is
modular, then K = {x ∈ A xA ∩ A(ǫ) ⊆ I} is a maximal right ideal of A
such that K is a modular ideal of A of degree ǫ.
Proof. Let I be a maximal right ideal of A(ǫ) such that I is a modular ideal
of A(ǫ) with respect to e. Since A is regular, according to [9, 10] we have
that K = {x ∈ A xA ∩ A(ǫ) ⊆ I} is a maximal modular right ideal of
A with respect to e, and K ∩ A(ǫ) = I. It is clear that K ∩ A(ǫ) ⊆ I. Let
x ∈ I, and let a ∈ A. If b ∈ A is such that 0 6= axb ∈ A(ǫ), then, by our
assumption on ∆, we have that both a and b belong to A(ǫ). Since x ∈ I, it
follows that axb ∈ I. Therefore I ⊆ K ∩ A(ǫ). Thus K ∩ A(ǫ) = I, and so
K 6= 0. It follows that K is a maximal ideal of A. Let a ∈ A. If a ∈ K, then
ea, ae ∈ K. If a /∈ K, then there exist x, y ∈ A such that xay ∈ A(ǫ) but
xay /∈ I. In particular, xay 6= 0. Since A is regular, our assumption on ∆
implies that δ(x) = δ(y) = δ(a) = ǫ. By assumption, I is a modular ideal of
A(ǫ) with respect to e. Therefore ea#a, ae#a, and ea − a, ae − a ∈ I ⊆ K.
Since K is a maximal ideal of A, it follows that A/ K is a regular simple
anneid with unity e + K. Therefore K is modular.
Conversely, if I is a maximal right ideal of A such that I is modular, then
clearly, Iǫ = I ∩ A(ǫ) is a modular right ideal of A(ǫ) such that Iǫ is modular.
Also, according to [9, 10], Iǫ is maximal.
Definition 3.16. An element x of an anneid A is said to be G-regular if
there exists no proper ideal I of A such that x is a unity modulo I. An
anneid (ideal) is called G-regular if every of its elements is G-regular.
The following theorem is an analogue of the characterization of quasi-regular
elements of a regular anneid from [9, 10] with the same proving technique.
Theorem 3.17. Let A be a regular anneid. An element x ∈ A is G-regular
if and only if one of the following two conditions is satisfied:
i) δ(x) is not an idempotent element of ∆∗;
ii) ǫ = δ(x) is an idempotent element of ∆∗ and x is a G-regular element
of the ring A(ǫ).
9
Proof. According to Remark 3.3, every unity modulo a proper modular ideal
has the degree which is an idempotent element of ∆∗. Hence, if x ∈ A satisfies
i), then x is G-regular. If x ∈ A is such that ǫ = δ(x) is an idempotent
element of ∆∗, we claim that x is G-regular if and only if x is G-regular in
A(ǫ). It suffices to prove that x ∈ A is a unity modulo a maximal modular
ideal of A if and only if it is a unity modulo a maximal modular ideal of
A(ǫ). However, this follows from Theorem 3.15.
Theorem 3.18. Let A be a regular anneid. Then G(A) = A if and only if
A is a G-regular anneid.
Proof. Let G(A) = A. According to Remark 3.6, no element of A can be
a unity modulo a proper ideal, whence every element of A is a G-regular
element.
Conversely, suppose that every element of a regular anneid A is G-regular.
Take any right ideal I of A such that I 6= A. Suppose that A/ I has a unity
a + I, a /∈ I. Then a2 + I = a + I. This and the fact that a /∈ I imply
0 6= a2#a. Hence ǫ = δ(a) is an idempotent element of ∆∗. According to
the previous theorem, a is a G-regular element of A(ǫ), that is, a belongs to
G(a) := {ax − x +Pn
i=1(yiazi − yizi) x, yi, zi ∈ A(ǫ), n a natural number}
(see, for instance, [7]). However, for any element ax − x +Pn
i=1(yiazi − yizi)
i=1((yi + I)(a + I)(zi +
from G(a), we have (a + I)(x + I) − (x + I) + Pn
I) − (yi + I)(zi + I)) = 0 + I, since a + I is by assumption a unity of A/ I.
i=1(yiazi − yizi) ∈ I. Therefore G(a) ⊆ I, and particularly,
Hence ax − x +Pn
a ∈ I, a contradiction. We proved that there is no proper right ideal I of A
such that A/ I is a simple regular anneid with unity, and so A is a graded
Brown -- McCoy radical anneid, that is, G(A) = A.
Corollary 3.19. Let A be a regular anneid. Then G(A) is a G-regular ideal
that contains all G-regular ideals of A.
If A is an anneid and if ǫ is an idempotent element of ∆∗, then we will
denote the classical Brown -- McCoy radical of the ring A(ǫ) by G(A(ǫ)). The
proof of the following theorem follows Halberstadt's proof of the equality
J(A(ǫ)) = J(A)∩ A(ǫ), where J(A(ǫ)) denotes the classical Jacobson radical
of the ring A(ǫ), and J(A) the graded Jacobson radical of a regular anneid
A [10]. We include it for completeness.
Theorem 3.20. If A is a regular anneid, then for every idempotent element
ǫ of ∆∗, we have G(A(ǫ)) = G(A) ∩ A(ǫ).
Proof. Ideal G(A) ∩ A(ǫ) is a G-regular ideal of A(ǫ), according to Theorem
3.17 and the previous corollary. Hence, according to [7, Corollary 4.8.3],
G(A) ∩ A(ǫ) is contained in G(A(ǫ)).
Conversely, let x ∈ G(A(ǫ)) and suppose that x /∈ G(A). Then, according
to Theorem 3.15 and Corollary 3.5, there exists y ∈ A such that ǫ′ = δ(xy)
10
is an idempotent element of ∆∗, but xy /∈ G(A(ǫ′)). Clearly, ǫ 6= ǫ′. Since
ǫδ(y) = ǫ′ and A is a regular anneid, ǫ, ǫ′ and δ(y) are mutually distinct.
Notice that in any ring R, the set I = {r ∈ R RrR = 0} is an ideal of R
and I 3 = 0, which implies that I is contained in the Jacobson radical of R,
and hence, I is contained in the Brown -- McCoy radical of R. In our case,
xy /∈ G(A(ǫ′)), and so there exists z ∈ A(ǫ′) such that zxy 6= 0. This implies
ǫ′(ǫδ(y)) = (ǫ′ǫ)δ(y), and so (ǫ′ǫ)δ(y) = ǫ′2 = ǫ′ = ǫδ(y) 6= 0. Therefore,
since A is regular, ǫ′ǫ = ǫ = ǫ2 6= 0. Again, by the regularity of A, we have
ǫ′ = ǫ, a contradiction.
Remark 3.21. One cannot discard the assumption made on ∆. Namely, ac-
cording to Example in [8], page 352, there exists a simple C2-graded ring
R without unity such that Re is not a Brown -- McCoy radical ring. Since
a simple graded ring is graded simple, it follows that A = Re ∪ Rg is a
graded Brown -- McCoy radical anneid. Therefore, G(Re) $ G(A)∩Re. (Here,
A(e) = Re.)
However, the first statement of Theorem 3.15 is true in general, and implies
the following result.
Theorem 3.22. Let A be a regular anneid and let ǫ be a nonzero idempotent
element of ∆. Then G(A(ǫ)) ⊆ G(A) ∩ A(ǫ).
Proof. Let I be a maximal right ideal of A such that I is a modular ideal of
A of degree ǫ. Also, let e be a unity modulo I. Then I is a maximal modular
right ideal of A with respect to e. Therefore for every x ∈ A we either have
that ex, x ∈ I or ex#x and ex − x ∈ I. Let x ∈ A(φ), where φ is a nonzero
idempotent element of ∆ distinct from ǫ. Then, since A is regular, we must
have that x ∈ I. Hence I ∩ A(φ) = A(φ). On the other hand, according to
the first statement of Theorem 3.15, we have that Iǫ = I ∩ A(ǫ) is a maximal
right ideal of A(ǫ) such that Iǫ is a modular ideal of A(ǫ) with respect to e.
Therefore, G(A) ∩ A(ǫ) = T I∈M I ∩ A(ǫ) equals the intersection of maximal
right ideals I of A, such that I are modular ideals of A of degree ǫ, with
A(ǫ). Hence G(A(ǫ)) ⊆ G(A) ∩ A(ǫ).
In case R = Lδ∈∆ Rδ is a strongly graded ring (see [15]) with unity,
where ∆ is a finite group with identity ǫ, it is known from [12] that G(Rǫ) =
G(R) ∩ Rǫ, where G denotes the classical Brown -- McCoy radical of a ring.
In the proof of this result, the fact that unity element of R belongs to Rǫ is
essential.
According to [10], if A is a regular anneid, and A its linearization, then
A is a graded ring with unity 1 if and only if the following hold:
i) For every nonzero idempotent element ǫ ∈ ∆, the ring A(ǫ) is with
unity 1ǫ;
11
ii) For every x ∈ A there exist idempotent elements ǫ, φ ∈ ∆∗ such that
1ǫx = x = x1φ;
iii) ∆∗ contains only a finite number of idempotent elements.
Moreover, 1 = Pǫ 1ǫ. Therefore, if A is a regular anneid with unity 1, then
∆∗ contains exactly one idempotent element ǫ and 1 ∈ A(ǫ).
If A is a regular anneid, by J(A) we denote the graded Jacobson radical
[9] of A.
Theorem 3.23. Let A be a regular anneid with unity such that A is strongly
graded. Also, let us assume that J(A) ∩ A(δ) is properly contained in A(δ)
for every δ ∈ ∆∗. If ǫ is the idempotent element of ∆∗, then G(A(ǫ)) =
G(A) ∩ A(ǫ).
Proof. Since A(δ) * J(A) for every δ ∈ ∆∗, it follows by [9, 10] that for
every δ ∈ ∆∗ there exists a unique δ−1 ∈ ∆∗ such that δδ−1 = ǫ. Since
ǫδ = δǫ = δ for every δ ∈ ∆, and since A is regular, it follows easily that
δδ−1 = δ−1δ = ǫ. Now, as in the case of group-graded rings, if I is an ideal
of A, then Iǫ = I ∩ A(ǫ) is an ideal of A(ǫ) and A(δ)IǫA(δ−1) = Iǫ for
every δ ∈ ∆∗. We refer to such ideals of A(ǫ) as ∆∗-invariant. Also, it can
be proved that I is an ideal of A if and only if I = AIǫ = IǫA. If I is an
ideal of A(ǫ), then the mapping I 7→ IA defines a one-to-one correspondence
between the maximal ∆∗-invariant ideals of A(ǫ) and the maximal ideals of
A. Since A is with unity, if I is a maximal ideal of A(ǫ), then A(δ)IA(δ−1)
is a maximal ideal of A(ǫ) for every δ ∈ ∆∗, and Tδ∈∆∗ A(δ)IA(δ−1) is a
maximal ∆∗-invariant ideal of A(ǫ), just as in the proof of Proposition 2
in [12]. It follows that G(A(ǫ)) equals the intersection of all maximal ∆∗-
invariant ideals of A(ǫ). Therefore G(A(ǫ)) = G(A) ∩ A(ǫ).
Next we prove that the homogeneous part of the largest homogeneous ideal
contained in the classical Brown -- McCoy radical of a graded ring coincides
with the large graded Brown -- McCoy radical of the corresponding anneid.
For a similar result on the Jacobson radical, see [9, 10].
Theorem 3.24. Let A be an anneid, A its linearization, and G(A) the
classical Brown -- McCoy radical of the ring A. Then Gl(A) = G(A) ∩ A.
Proof. Let M be a simple right A-module. Then M may be viewed as a
simple right A-moduloid. Indeed, M is obviously a right A-moduloid and
M A 6= 0. Also, let I be an ideal of A such that M I 6= 0. Suppose that for all
a ∈ I there exists y ∈ M such that ya 6= y. On the other hand, M I 6= 0 and
so there exists b ∈ I such that xb = x for every x ∈ M . In particular, yb = y,
a contradiction. Hence M is a simple right A-moduloid. Also, it is obvious
that (0 : M )A ⊆ (0 : M )A, that is, the annihilator of a right A-moduloid M
is contained in the annihilator of a right A-module M . Since the classical
12
Brown -- McCoy radical of a ring equals the intersection of annihilators of all
simple right modules over that ring [2], we have Gl(A) ⊆ G(A). Therefore
Gl(A) ⊆ G(A) ∩ A.
Conversely, let a ∈ G(A) ∩ A and let M be an arbitrary simple right A-
moduloid. We claim that a ∈ (0 : M ). Suppose a /∈ (0 : M ). Then there
exists x ∈ M such that xa 6= 0. This implies M hai 6= 0. Since M is simple,
there exists b ∈ hai such that mb = m for all m ∈ M. Particularly, xb = x.
For all ¯a ∈ A, we have x¯a = xb¯a, which implies ¯a − b¯a ∈ (0 : x). Since
b ∈ hai, and a ∈ G(A), we have b ∈ G(A). Therefore there exist w, yi zi ∈ A
i=1 yi(bzi − zi) (see,
for instance, [7, Theorem 4.8.2]). However, as we have seen, for all ¯a ∈ A,
we have ¯a − b¯a ∈ (0 : x), that is, b¯a − ¯a ∈ (0 : x). Particularly, bw − w,
bzi − zi ∈ (0 : x), and therefore b ∈ (0 : x), which implies 0 = xb = x, a
contradiction. Hence a ∈ (0 : M ), and so G(A) ∩ A ⊆ Gl(A). Therefore
Gl(A) = G(A) ∩ A.
and a natural number n such that b = bw − w + Pn
Remark 3.25. Let A be a regular anneid. Since Gl(A) ⊆ G(A), the previ-
ous theorem implies that Lδ∈∆∗ G(A) ∩ A(δ) ⊆ G(A), that is, the largest
homogeneous ideal contained in the classical Brown -- McCoy radical of the
ring A is contained in the graded Brown -- McCoy radical of A (contrast with
[4, Proposition 3.5]).
If R = Ls∈S Rs is a group-graded ring, and if e is the neutral element of S,
then we know from [8] that the Brown -- McCoy radical of Re is contained in
the Brown -- McCoy radical of R. Here we have the following result.
Theorem 3.26. Let A be a regular anneid, A its linearization, and ǫ an
idempotent element of ∆∗. If G(A) = Gl(A), then the classical Brown --
McCoy radical G(A(ǫ)) of A(ǫ) is contained in the classical Brown -- McCoy
radical G(A) of A.
Proof. Since by assumption, G(A) = Gl(A), and since by Theorem 3.22,
G(A(ǫ)) ⊆ G(A) ∩ A(ǫ), we have G(A(ǫ)) ⊆ Gl(A). On the other hand,
Theorem 3.24 tells us that Gl(A) = G(A) ∩ A, and so G(A(ǫ)) ⊆ G(A).
References
[1] A. V. Andrunakievich, V. A. Andrunakievich, One-sided ideals and
radicals of rings, Algebra and Logic 20 (6) (1981), 319 -- 333.
[2] V. A. Andrunakievich, Yu. M. Ryabuhin, Special modules and special
radicals, Dokl. AN SSSR 147 (6) (1962), 1274 -- 1277.
[3] I. N. Balaba, Special radicals of graded rings, Bul. Acad. S¸tiint¸e Mold.
Mat. 44 (1) (2004), 26 -- 33.
13
[4] M. Beattie, P. Stewart, Graded radicals of graded rings, Acta Math.
Hung. 58 (3-4) (1991), 261 -- 272.
[5] B. Brown, An extenstion of the Jacobson radical, Proc. Amer. Math.
Soc. 2 (1951), 114 -- 117.
[6] M. Chadeyras, Essai d'une th´eorie noetherienne pour les anneaux com-
mutatifs, dont la graduation est aussi g´en´erale que possible, M´emoire
de la Soci´et´e Math´ematique de France 22 (1970), 3 -- 143.
[7] B. J. Gardner, R. Wiegandt, Radical Theory of Rings, Pure and Applied
Mathematics 261, Marcel Dekker 2004.
[8] P. Grzeszczuk, On G-systems and G-graded rings, Proc. Amer. Math.
Soc. 95 (3) (1985), 348 -- 352.
[9] E. Halberstadt, Le radical d'un anneide r´egulier, C. R. Acad. Sci.,
Paris, S´er. A, Paris 270 (1970), 361 -- 363.
[10] E. Halberstadt, Th´eorie artinienne homog`ene des anneaux gradu´es `a
grades non commutatifs r´eguliers, PhD Thesis, University of Pierre and
Marie Currie (Paris VI), 1971.
[11] N. Jacobson, Structure of rings, Colloquium Publications, 37, American
Mathematical Society, 1964.
[12] E. Jespers, E. R. Puczy lowski, The Jacobson and Brown -- McCoy rad-
icals of rings graded by free groups, Comm. Algebra 19 (2) (1991),
551 -- 558.
[13] A. V. Kelarev, Combinatorial properties and homomorphisms of semi-
groups, Internat. J. Algebra Comput. 4 (3) (1994), 443 -- 450.
[14] A. V. Kelarev, On groupoid graded rings, J. Algebra 178 (1995), 391 --
399.
[15] A. V. Kelarev, Ring constructions and applications, Series in Algebra,
Vol. 9, World Scientific, 2002.
[16] A. V. Kelarev, A. Plant, Bergman's lemma for graded rings, Comm.
Algebra 23 (12) (1995), 4613 -- 4624.
[17] M. Krasner, Une g´en´eralisation de la notion de corps-corpo´ıde. Un
corpo´ıde remarquable de la th´eorie des corps valu´es, C. R. Acad. Sci.,
Paris 219 (1944), 345 -- 347.
[18] M. Krasner, Anneaux gradu´es g´en´eraux, Publications math´ematiques
et informatiques de Rennes, Colloque d'alg`ebre, Fascicule S3, (1980),
209 -- 308.
14
[19] M. Krasner, M. Vukovi´c, Structures paragradu´ees (groupes, anneaux,
modules), Queen's Papers in Pure and Applied Mathematics, No. 77,
Queen's University, Kingston, Ontario, Canada 1987.
[20] P.-H. Lee, E. R. Puczy lowski, On prime ideals and radicals of poly-
nomial rings and graded rings, J. Pure Appl. Algebra 218 (2) (2014),
323 -- 332.
[21] A. Smoktunowicz, A note on nil and Jacobson radicals in graded rings,
J. Algebra Appl. 13 (4) (2014), 1350121, 8.
[22] A. Suli´nski, J. F. Watters, On the Jacobson radical of associative 2-
graded rings, Acta Math. Hung. 43 (1-2) (1984), 13 -- 16.
[23] M. Vukovi´c, Structures gradu´ees et paragradu´ees, Prepublication de
l'Institut Fourier, Universit´e de Grenoble I, 536 (2001), pp. 1 -- 40.
Emil Ili´c-Georgijevi´c
University of Sarajevo
Faculty of Civil Engineering
Patriotske lige 30, 71000 Sarajevo
Bosnia and Herzegovina
e-mail: [email protected]
15
|
1807.05227 | 2 | 1807 | 2019-10-01T10:01:12 | The special linear group for nonassociative rings | [
"math.RA"
] | We extend to arbitrary rings a definition of the octonion special linear group due to Baez. At the infinitesimal level we get a Lie ring, which we describe over some large classes of rings, including all associative rings and all algebras over a field. As a corollary we compute all the groups Baez defined. | math.RA | math |
THE SPECIAL LINEAR GROUP FOR NONASSOCIATIVE RINGS
HARRY PETYT
Abstract. We extend to arbitrary rings a definition of the octonion special linear
group due to Baez. At the infinitesimal level we get a Lie ring, which we describe over
some large classes of rings, including all associative rings and all algebras over a field.
As a corollary we compute all the groups Baez defined.
1. Introduction
The special linear groups SL2(R) and SL2(C) are, respectively, the double covers of
SO0(2, 1) and SO0(3, 1) -- the isometry groups of the hyperbolic plane and hyperbolic
3-space. The pattern continues with the quaternions H, as shown by Kugo and Townsend
in [KT83], and Sudbery deals with the final normed real division algebra, the octonions
O, in [Sud84]. Unfortunately the way Sudbery defines the special linear group over O
only makes sense in dimensions two and three.
In his celebrated survey [Bae02], Baez suggests a unified definition of SLm(O) for all
m, and shows that it agrees with Sudbery's definition when m = 2. He does not discuss
the case m > 2, and it seems that until now no further investigation has been made.
Motivated by this, in Section 2 we reformulate Baez's definition of the special linear
group and algebra in a natural way that lends itself to computation, and note that it
naturally extends to arbitrary nonassociative rings (in the present paper, we do not in
general assume rings to be associative). We then determine the corresponding special
linear ring (we do not necessarily get an algebra structure) for all associative rings. In
Section 3 we cover the two dimensional case for unital real composition algebras.
In
Section 4 we characterise SLm, with m > 2, for a large class of algebras that includes O.
This allows us to compute Baez's groups. In doing so, we find that in three dimensions
his definition disagrees with Sudbery's, which gives a real form of the exceptional Lie
group E6.
An alternative definition for SL2(O) has been proposed by Hitchin [Hit18]. This defi-
nition is motivated by a dimension argument, and does not give a Lie group.
I would like to thank Dmitriy Rumynin for introducing me to the problem and for his
helpful comments and suggestions.
2. Preliminaries
2.1. A composition algebra is a not necessarily unital or associative algebra C over a
field F, together with a nondegenerate quadratic form · 2 that is multiplicative in the
sense that zw2 = z2w2. Such algebras come with an anti-involution, which we call
conjugation and denote by a bar, e.g. ¯z. They are also necessarily alternative. That is,
the associator [·, ·, ·] : C 3 → C, given by [z, w, u] = (zw)u − z(wu), is alternating.
1
THE SPECIAL LINEAR GROUP FOR NONASSOCIATIVE RINGS
2
If the characteristic of F is not two, then all unital composition algebras can be obtained
from F by the Cayley -- Dickson construction (a description of which can be found in
[Sch95]), and have famously been classified by Jacobson [Jac58]. We are mainly interested
in real composition algebras, and we state his classification in this case.
Theorem 2.1 (Jacobson). The unital real composition algebras are exactly:
(i) R
(ii) C
(iv) H
(vi) O
(iii) R2, with quadratic form (a, b)2 = ab
(v) M2(R), the 2 × 2 matrices over R, with · 2 = det
(vii) The split octonions O′
We write 1, e1, . . . , ed−1 for an orthonormal basis of a unital composition algebra of
i = ±1 for all i, and eiej = −ej ei whenever i 6= j. Letting
dimension d. We then have e2
Lz : C → C denote the left multiplication map w 7→ zw, alternativity of C gives us
(1)
LeiLei = Le2
i
= ±1
Moreover, whenever ei 6= ej we have
0 = [ei, ej , z] + [ej , ei, z] = Leiej+ej ei(z) − (LeiLej + Lej Lei)(z)
and hence
(2)
LeiLej = −Lej Lei
Let Rp,q denote Rp+q with the standard quadratic form of signature (p, q). The algebras
in the left-hand column of Theorem 2.1 have signature (d, 0), and those in the right-hand
column have signature ( d
2 , d
2 ).
2.2. For a ring R we write Mm(R) to mean the space of m × m matrices with entries
in R, and Eij for an element of the standard basis. The trace of a matrix x is written
tr x, and left multiplication maps are again denoted Lx. The definition of the octonion
special linear group and algebra given by Baez is as follows [Bae02, p.177].
Definition 2.2 (Baez). The octonion special linear algebra slm(O) is the Lie algebra
generated under commutators by the set {Lx : x ∈ Mm(O), tr x = 0}. The octonion
special linear group SLm(O) is the Lie group generated by exponentiating slm(O).
This definition is not well suited to computation, and we prefer to use the following,
which is easily seen to agree with Definition 2.2 in the case R = O.
Definition 2.3. For R a not necessarily associative or unital ring, slm(R) is the ring
generated by (cid:8)LaEij : a ∈ R, i 6= j(cid:9) under commutators. Similarly, SLm(R) is the group
generated by (cid:8)LI+aEij : a ∈ R, i 6= j(cid:9) under composition.
Straight from the definition we can obtain a nice description of the special linear
algebra of an associative ring.
Theorem 2.4. Let R be an associative (not necessarily unital) ring. Then there is an
isomorphism slm(R) ∼= {x ∈ Mm(R) : tr x ∈ [R, R]}
THE SPECIAL LINEAR GROUP FOR NONASSOCIATIVE RINGS
3
Proof. Since R is associative, we have LaLb = Lab for all a, b ∈ R, so we can identify
LaEij with the matrix aEij and consider slm(R) ⊂ Mm(R). Now slm(R) contains all
matrices with all diagonal entries zero, as these form the linear span of the generators.
Furthermore, the commutator of two generators is [aEij, bEkl] = δjkabEil − δilbaEkj,
where δ denotes the Kronecker delta. If δjk and δil are not both 1 then we get either
zero or a generator. If both are 1 then we get abEii − baEjj. Clearly this has trace lying
in [R, R], and by varying a and b we can get the whole of [R, R]. Then varying i and
j gives the right hand side of the result. Note that the commutator of such a diagonal
matrix with a generator is traceless, so all further commutators have trace in [R, R], and
we are done.
(cid:3)
Theorem 2.4 shows that Definition 2.3 gives a true generalisation of the usual special
linear algebra, for if R is a field then [R, R] = 0, and, moreover, if R = H then [H, H] =
{z ∈ H : Re(z) = 0}, which gives the standard definition of slm(H) [Har90, p.52].
3. The Two Dimensional Case
Let C be a unital real composition algebra. For x = (xij) ∈ Mm(C), the hermitian
conjugate of x is x∗ = (xji). If x∗ = x then x is said to be hermitian, and the set of such
matrices is denoted hm(C). Note that all diagonal entries of a hermitian matrix lie in
R. We restrict our attention to the case m = 2, where alternativity of C ensures that
the determinant map x 7→ x11x22 − x12x12 is a well defined quadratic form on h2(C) (see
[Bae02, p.176]).
If C has dimension d and signature (p, q), then writing z = z0 + z1e1 + · · · + zd−1ed−1
for an element of C, we have that h2(C) is isometric to Rq+1,p+1 via the map
(cid:18) r
¯z
z
s (cid:19) 7−→ (cid:18) r + s
2
,
r − s
2
, z0, z1, . . . , zd−1(cid:19)
We now define a representation of SL2(C) on h2(C). Let y = LI+aEij be a generator
of SL2(C), and for x ∈ h2(C) set y · x = (I + aEij)x(I + ¯aEji). This product is well
defined because C is alternative, and we extend to SL2(C) in the obvious way.
Lemma 3.1. The action of SL2(C) on h2(C) is by isometries. That is, if x ∈ h2(C) and
y ∈ SL2(C) then det(y · x) = det x.
Proof. It suffices to show that this holds for generators of SL2(C). The two cases are
similar, so we just do y = LI+aE21. Let x = (cid:18) r
¯z
z
s (cid:19), recalling that r, s ∈ R. Since
w ¯w = ¯ww for all w ∈ C we obtain
det(y · x) = det(cid:18)
r
ra + ¯z
r¯a + z
ra¯a + ¯z¯a + az + s (cid:19) = rs − ¯zz = det x
(cid:3)
It follows that there is a homomorphism of connected Lie groups
ψ : SL2(C) −→ SO0(q + 1, p + 1)
In order to analyse ψ, we describe a basis of sl2(C), but first a remark.
Remark 3.2. If F = C then the same argument as the one above gives a homomorphism
SL2(C) → SO(d + 2, C).
THE SPECIAL LINEAR GROUP FOR NONASSOCIATIVE RINGS
4
Lemma 3.3. sl2(C) is based by the set
nLE12, LE21, [LE12 , LE21], αi = LeiE12 , βi = LeiE21 , γi = [LE12 , βi], εij = [αi, βj ] : i < jo
In particular, dim SL2(C) = 3 + 3(d − 1) + (d−1)(d−2)
= (d+1)(d+2)
.
2
2
Proof. It follows from identities (1) and (2) that the set in question bases the subspace
spanned by products of length at most two, so it suffices to show that this is the whole
of sl2(C). Products of length three are spanned by generators and elements δ and δT ,
where
δ = (cid:18) 0 LeiLej Lek + Lek Lej Lei
0
0
(cid:19)
There are three cases for δ, depending on the choice of ei, ej , and ek.
Case 1: ej is equal to either ei or ek. Then δ is a generator by identity (1).
Case 2: ei = ek 6= ej. Using both identities (1) and (2) we see that δ is a generator.
Case 3: ei, ej , ek are distinct. Then δ = 0 by identity (2).
Thus products of length three are spanned by generators, which completes the proof. (cid:3)
Lemma 3.4. ker dψ = 0
Proof. The action of SL2(C) on h2(C) induces an action of sl2(C):
if x ∈ h2(C) and
y is a generator of sl2(C) then y · x = yx + xy∗. By definition, any element of ker dψ
acts trivially, so we calculate the action of the basis of Lemma 3.3 on an arbitrary
i=1 ziei ∈ C, and below the λi and κij
i=1 λiei.
z
¯z
s (cid:19) ∈ h2(C). Here r, s ∈ R, z = z0 +Pd−1
x = (cid:18) r
are real. In several places we find it convenient to write w = λ0 +Pd−1
(a) : (cid:16)λ0LE12 +
(b) : (cid:16)λ0LE21 +
λiαi(cid:17) · x = LwE12 · x = (cid:18) 2 Re(w¯z) sw
0 (cid:19)
λiβi(cid:17) · x = LwE21 · x = (cid:18) 0
rw 2 Re(wz) (cid:19)
Xi=1
Xi=1
r ¯w
s ¯w
d−1
d−1
(c) : (cid:16)λ0[LE12 , LE21] +
d−1
Xi=1
λiγi(cid:17) · x = [LE12 , LwE21] · x
= LE12 ·(cid:18) 0
= 2(cid:18)
−sλ0 (cid:19)
¯z ¯w − w¯z
wz − z ¯w
rλ0
r ¯w
rw 2 Re(wz) (cid:19) − LwE21 ·(cid:18) 2 Re(¯z)
s
s
0 (cid:19)
THE SPECIAL LINEAR GROUP FOR NONASSOCIATIVE RINGS
5
(d) : (cid:16) X0<i<j<d
= Xi<j
= Xi<j
κij εij(cid:17) · x = Xi<j
κij(cid:18)(cid:18) 2 Re(eirej) 2 Re(ej z)ei
2κij (cid:18)
2 Re(ej z) ¯ei
j zj ei + e2
i ziej
−e2
0
0
e2
j zj ei − e2
i ziej
0
(cid:19)
κij (cid:0)LeiE12 · (Lej E21 · x) − Lej E21 · (LeiE12 · x)(cid:1)
(cid:19) −(cid:18)
0
2 Re(ei ¯z)ej
2 Re(ei ¯z) ¯ej
2 Re(ej sei) (cid:19)(cid:19)
Together, these describe the action of every element of sl2(C) on h2(C). Now assume
that y acts trivially. Writing y in the basis of Lemma 3.3 and using the four equations
above (with respective sets of coefficients λi, µi, νi, and κij), consider y · x.
and z ∈ C. Taking z = 0, r = 1 gives ν0 = 0, and then cycling z through the ei gives
The upper-left entry is 2 Re(cid:0)(λ0+Pd−1
i=1 λiei)¯z(cid:1)+2rν0, which must be zero for all r ∈ R
λi = 0. Similarly, considering the lower right entry gives 2 Re(cid:0)(µ0 +P µiei)z(cid:1)−2sν0 = 0.
We already have ν0 = 0, and cycling z through the ei gives µi = 0. Finally, considering
the top right entry, we are left with
d−1
Xi=1
2(νieiz − zνi ¯ei) + X0<i<j<d
2κij(e2
j zjei − e2
i ziej) = 0
Taking z = 1 gives Pd−1
1
z = e1, z = e2, . . . we find that all κ1j , κ2j , . . . are zero. Thus ker dψ = 0.
4νiei = 0, so all νi are zero. Then successively considering
(cid:3)
Knowing that dψ has full rank is enough to prove the main theorem of this section.
Theorem 3.5. If C is a unital real composition algebra of dimension d and signature
(p, q) then SL2(C) ∼= Spin(p + 1, q + 1).
Proof. Because Spin(p + 1, q + 1) ∼= Spin(q + 1, p + 1) it suffices to show that ψ is
onto and has two-point kernel. By Lemma 3.3 we have dim SL2(C) = (d+1)(d+2)
=
dim SO0(q + 1, p + 1), so by Lemma 3.4 dψ is onto. Hence ψ is onto. Indeed,
2
ψ(SL2(C)) = exp(dψ(sl2(C))) = exp(so(q + 1, p + 1)) = SO0(q + 1, p + 1)
It remains to show that ψ has two-point kernel. Consider the real matrices
a = (cid:18) 1 −1
1 (cid:19) ,
0
b = (cid:18) 1 0
1 1 (cid:19) ,
c = (cid:18) 1 −2
1 (cid:19)
0
and the linear map ι = LaLbLcLbLa ∈ SL2(C), which acts as −I on C 2. Because
a, b, c ∈ M2(R), the expression for ι · x associates, so ι · x = (−I)x(−I) = x, and
ι ∈ ker ψ. Hence ker ψ consists of at least two elements.
Claim: If C is associative then ker ψ = {1, ι}.
Proof: As in the proof of Theorem 2.4 we can consider elements of SL2(C) to be matrices.
Then ι = −I. Also, if y = (yij) ∈ SL2(C) acts trivially on h2(C) we have
(cid:18) r 0
0 s (cid:19) = y ·(cid:18) r 0
0 s (cid:19) = (cid:18) ry11y11 + sy12y12
ry21y11 + sy22y12
ry11y21 + sy12y22
ry21y21 + sy22y22 (cid:19)
THE SPECIAL LINEAR GROUP FOR NONASSOCIATIVE RINGS
6
Taking r = 1, s = 0 in this gives y11y11 = 1 and y21 = 0. Similarly, taking r = 0, s = 1
gives y22y22 = 1 and y12 = 0. Now we have
¯z 0 (cid:19) = (cid:18) y11
(cid:18) 0 z
0
0
y22 (cid:19) ·(cid:18) 0 z
¯z 0 (cid:19) = (cid:18)
0
y11zy22
y22z y11
0
(cid:19)
Taking z = 1 gives y11y22 = 1. But y11y11 = 1, so y22 = y11. From this we get
z = y11zy11, so zy11 = y11z for all z ∈ C. Hence y11 = y22 = ±1, and thus y ∈ {1, ι}. ♦
For the case C = O, note that since π1(SO(9, 1)) = Z2 [Hal03, pp.335, 343], any
proper cover is a double cover.
This just leaves the case C = O′. The complexification C ⊗ O′ is isomorphic to the
bioctonions C ⊗ O, a unital complex composition algebra. Thus SL2(O′) is a real form of
SL2(C ⊗ O), which covers SO(10, C) by Remark 3.2. Since π1(SO(10, C)) = Z2 [Hal03,
p.343], the covering is 2:1, with kernel {1, ι}. We thus have a commutative diagram
SL2(O′)
ψ
SO0(5, 5)
complexify
complexify
SL2(C ⊗ O)
2:1
SO(10, C)
which completes the proof in the final case, C = O′.
(cid:3)
4. The General Case
Let R be a commutative associative unital ring, and A a finite dimensional R -- algebra
that is free as an R -- module, with basis {e1, . . . , en}. For example, A could be any finite
dimensional algebra over a field.
Write MA for the left multiplication algebra of A. That is, MA is the R -- algebra
generated by {Lrei : r ∈ R}. Since A is free over R we have MA ⊂ EndR(Rn) = Mn(R).
Theorem 4.1. Under the above assumptions, if m ≥ 3 then slm(A) ∼= (cid:8)x ∈ Mm(MA) :
tr x ∈ [MA, MA](cid:9). In particular slm(A) ⊂ slmn(R).
Proof. The nonzero products of two generators are:
[LaEij , LbEji] = LLaLbEii − LLbLaEjj
[LaEij , LbEjk ] = LLaLbEik
Since m ≥ 3, it follows by taking successive products that slm(A) is the span of the set
nLαEij
: α ∈ MA, i 6= jo[nLαβEii − LβαEjj
: α, β ∈ MA, i 6= jo
Clearly all such matrices have trace in [MA, MA], and varying α and β gives the whole
of [MA, MA]. Varying i and j then gives the result.
(cid:3)
This reduces the problem of determining slm(A) to that of finding MA. In the case
of the R-algebra O, the group generated by left multiplications by units is SO(8) [CS03,
p.92]. This R-spans the full matrix algebra M8(R), so MO = M8(R). We can thus
calculate Baez's groups:
Corollary 4.2. If m ≥ 3 then SLm(O) ∼= SL8m(R).
THE SPECIAL LINEAR GROUP FOR NONASSOCIATIVE RINGS
Proof. Theorem 4.1 gives slm(O) ∼= sl8m(R). Exponentiating gives the result.
7
(cid:3)
Together with Theorem 3.5 this describes all the groups Baez defined. In fact, the
same argument works for O′, and we similarly obtain SLm(O′) ∼= SL8m(R) for m ≥ 3.
Remark 4.3. Corollary 4.2 shows that Baez's definition of SL3(O) disagrees with Sud-
bery's, which gives a real form of E6. For m = 2, 3, Sudbery defines slm(O) to be the
left multiplications by traceless elements of hm(O), together with its derivations. The
isomorphism with e6 is due to Chevalley and Schafer [CS50]. It is natural to ask what
happens to this construction when n is greater than 3. In this case, it is shown in [Pet19,
Thm. 3.3] that the derivation algebra of hm(O) is g2 ⊕ som, but it is not clear how this
interacts with the multiplication operators. In particular, when m > 3 the commutator
of two such operators may fail to be a derivation.
References
[Bae02] John C. Baez. The octonions. Bull. Amer. Math. Soc. (N.S.), 39(2):145 -- 205, 2002.
[CS50] Claude Chevalley and R. D. Schafer. The exceptional simple Lie algebras F4 and E6. Proc. Nat.
Acad. Sci. U.S.A., 36:137 -- 141, 1950.
[CS03] John H. Conway and Derek A. Smith. On quaternions and octonions: their geometry, arithmetic,
and symmetry. A K Peters, Ltd., Natick, MA, 2003.
[Hal03] Brian C. Hall. Lie groups, Lie algebras, and representations: an elementary introduction, volume
222 of Graduate Texts in Mathematics. Springer-Verlag, New York, 2003.
[Har90] F. Reese Harvey. Spinors and calibrations, volume 9 of Perspectives in Mathematics. Academic
Press, Inc., Boston, MA, 1990.
[Hit18] Nigel Hitchin. SL(2) over the octonions. Math. Proc. R. Ir. Acad., 118A(1):21 -- 38, 2018.
[Jac58] N. Jacobson. Composition algebras and their automorphisms. Rend. Circ. Mat. Palermo (2),
7:55 -- 80, 1958.
[KT83] Taichiro Kugo and Paul Townsend. Supersymmetry and the division algebras. Nuclear Phys. B,
221(2):357 -- 380, 1983.
[Pet19] Harry Petyt. Derivations of octonion matrix algebras. Comm. Algebra, 47(10):4216 -- 4223, 2019.
[Sch95] Richard D. Schafer. An introduction to nonassociative algebras. Dover Publications, Inc., New
York, 1995. Corrected reprint of the 1966 original.
[Sud84] A. Sudbery. Division algebras, (pseudo)orthogonal groups and spinors. J. Phys. A, 17(5):939 --
955, 1984.
|
1104.1341 | 1 | 1104 | 2011-04-07T14:29:58 | The higher rank numerical range of matrix polynomials | [
"math.RA"
] | The notion of the higher rank numerical range $\Lambda_{k}(L(\lambda))$ for matrix polynomials $L(\lambda)=A_{m}\lambda^{m}+...+A_{1}\lambda+A_{0}$ is introduced here and some fundamental geometrical properties are investigated. Further, the sharp points of $\Lambda_{k}(L(\lambda))$ are defined and their relation to the numerical range $w(L(\lambda))$ is presented. A connection of $\Lambda_{k}(L(\lambda))$ with the vector-valued higher rank numerical range $\Lambda_{k}(A_{0},..., A_{m})$ is also discussed. | math.RA | math | The higher rank numerical range
of matrix polynomials
Aik. Aretaki and J. Maroulas∗
September 10, 2018
1
1
0
2
r
p
A
7
]
.
A
R
h
t
a
m
[
1
v
1
4
3
1
.
4
0
1
1
:
v
i
X
r
a
Abstract
The notion of the higher rank numerical range Λk(L(λ)) for matrix
polynomials L(λ) = Amλm + . . . + A1λ + A0 is introduced here and
some fundamental geometrical properties are investigated. Further, the
sharp points of Λk(L(λ)) are defined and their relation to the numerical
range w(L(λ)) is presented. A connection of Λk(L(λ)) with the vector-
valued higher rank numerical range Λk(A0, . . . , Am) is also discussed.
Key words: higher rank numerical range, matrix polynomials, quantum error
correction.
AMS Subject Classifications: 15A60, 15A90, 81P68.
1
Introduction
Let Mn(C) be the algebra of matrices A = [aij]n
and
i,j=1 with entries aij ∈ C
L(λ) = Amλm + Am−1λm−1 + . . . + A1λ + A0
Λk(L(λ)) = {λ ∈ C : P L(λ)P = 0n f or some P ∈ Pk} ,
be a matrix polynomial with Ai ∈ Mn(C) and Am 6= 0. For a positive
integer k ≥ 1, we define the higher rank numerical range of L(λ) as
(1.1)
where Pk is the set of all orthogonal projections P of Cn onto any k-
dimensional subspace K of Cn. Equivalently,
(1.2)
Λk(L(λ)) = {λ ∈ C : Q∗L(λ)Q = 0k f or some Q ∈ Mn,k with Q∗Q = Ik} ,
since P = QQ∗, with Q ∈ Mn,k(C) and Q∗Q = Ik. In case k = 1, the set
reduces to the well known numerical range w(L(λ)) of L(λ) [14]
(1.3)
Λ1(L(λ)) ≡ w(L(λ)) = {λ ∈ C : x∗L(λ)x = 0 f or some x ∈ Cn,kxk = 1} .
∗Department of Mathematics, National Technical University of Athens, Zografou
Campus, Athens 15780, Greece. E-mail address: [email protected].
1
The (1.1) (or (1.2)) is an interesting generalization of numerical range, since
matrix polynomials play a significant role in several problems of computa-
tional chemistry and structural molecular biology [9]. They consist algebraic
tools to computing all conformations of ring molecules and they model va-
rious problems in terms of polynomial equations.
If L(λ) = Iλ − A, then
Λk(Iλ − A) = {λ ∈ C : P AP = λP f or some P ∈ Pk}
(1.4)
= {λ ∈ C : Q∗AQ = λIk , Q∗Q = Ik, Q ∈ Mn,k(C)} ,
namely, it coincides with the higher rank numerical range of a matrix A ∈
Mn. The concept of higher rank numerical range has been studied exten-
sively by Choi et al in [4, 5, 6, 7] and later by other researchers in [18, 20, 23].
We should note that for k = 1, Λk(Iλ − A) yields the classical numerical
range of a matrix A, i.e.
(1.5)
F (A) = {x∗Ax : x ∈ Cn,kxk = 1} ,
whose basic properties can be found in [10, 11, 12].
A multi-dimensional higher rank numerical range is the joint higher rank
numerical range [19]
(1.6)
Λk(A) =(cid:8)(µ0, µ1, . . . , µm) ∈ Cm+1 : ∃ P ∈ Pk such that
P AiP = µiP, i = 0, . . . , m} ,
where A = (A0, A1, . . . , Am) is an (m + 1)-tuple of matrices Ai ∈ Mn(C)
for i = 0, . . . , m. Apparently, for k = 1, Λ1(A) is identified with the joint
numerical range, denoted by
(1.7)
w(A) = {(x∗A0x, . . . , x∗Amx) : x ∈ Cn,kxk = 1} .
In the context of quantum information theory, Λk(A) is closely related to
a quantum error correcting code, since the latter exists as long as the joint
higher rank numerical range associated with the error operators of a noisy
quantum channel is a non empty set.
In section 2, we present some familiar properties of Λk(L(λ)) and we
provide a description of the set through intersections of numerical ranges of
all compressions of the matrix polynomial L(λ) to (n − k + 1)-dimensional
subspaces. This study originates from an analogous expression for matrices,
presented and proved in [1]. It also motivates us to investigate the geometry
of Λk(L(λ)) proving conditions for its boundedness and elaborating a basic
property on the number of its connected components.
In section 3, a connection of the boundary points of Λk(L(λ)) with re-
spect to the boundary points of w(L(λ)) is considered. Particularly, intro-
ducing the notion of sharp points for Λk(L(λ)), we show that a sharp point
2
of w(Aλ − B) with algebraic multiplicity k with respect to the spectrum
σ(Aλ − B) is also a sharp point of Λj(Aλ − B), for j = 2, . . . , k.
In se-
ction 4, a relationship between Λk(L(λ)) and Λk(CL(λ)) is presented, where
CL(λ) is the companion polynomial of L(λ). Also, we treat a sufficient con-
dition for boundary points of w(A) to be boundary points of Λk(A), where
A = (A0, . . . , Am) and evenly, we investigate an interplay of Λk(L(λ)) and
Λk(A).
2 Geometrical Properties
In the beginning of this section, we present some basic properties as in [20]
for the higher rank numerical range of a matrix polynomial L(λ).
Proposition 1. Let L(λ) =Pm
0, then
j=0 Ajλj be a matrix polynomial, where Am 6=
(a) Λk(L(λ)) is closed in C.
(b) For any α ∈ C, Λk(L(λ + α)) = Λk(L(λ)) − α.
j=0 Am−jλj then Λk(Q(λ)) \ {0} =(cid:8)µ−1 : µ ∈ Λk(L(λ))(cid:9).
(c) If Q(λ) =Pm
(d) If Ai, i = 0, . . . , m have a common totally isotropic subspace S =
span{x1, . . . , xk} with orthonormal vectors xj ∈ Cn, j = 1, . . . , k,
i.e. x∗l Aixj = 0 for any l, j = 1, . . . , k and i = 0, . . . , m, then
Λk(L(λ)) = C.
ing are equivalent:
j=0 Ajλj be a matrix polynomial, the follow-
Proposition 2. Let L(λ) =Pm
(i) µ ∈ Λk(L(λ))
(ii) there exists M ∈ Mn,k(C) with rankM = k such that M∗L(µ)M = 0k
(iii) there exists an L(µ)-orthogonal k-dimensional subspace K of Cn
(iv) there exist {ui}k
i=1 orthonormal vectors such that u∗j L(µ)ui = 0 for
every i, j = 1, . . . , k
(v) there exists a k-dimensional subspace K of Cn such that v∗L(µ)v = 0
for every v ∈ K
(vi) there exists a unitary matrix U ∈ Mn(C) such that
L2(µ) L3(µ)(cid:21) ,
U∗L(µ)U =(cid:20) 0k
L1(µ)
where L1(λ), L2(λ) and L3(λ) are suitable matrix polynomials.
3
j=1 Ajλj, then
Proof. The arguments (i)-(vi) are equivalent, since µ ∈ Λk(L(λ)) is equiva-
lent to 0 ∈ Λk(L(µ)). Further, we refer to the Proposition 1.1 in [4].
Proposition 3. Let L(λ) =Pm
Proof. For any j ∈ {2, . . . , k}, let µ0 ∈ Λj(L(λ)). Then 0 ∈ Λj(L(µ0)) ⊆
Λj−1(L(µ0)) and consequently, by 0 ∈ Λj−1(L(µ0)), we conclude that µ0 ∈
Λj−1(L(λ)).
Corollary 4. Let L(λ) be an n× n matrix polynomial. Then for any k ≤ n
Λk(L(λ)) ⊆ Λk−1(L(λ)) ⊆ . . . ⊆ Λ1(L(λ)).
k
) = w(L(λ)),
Λk(L(λ) ⊕ . . . ⊕ L(λ)
}
{z
i.e. Λk(⊕kL(λ)) is a non-empty set.
Proof. Due to Proposition 3, µ0 ∈ Λk(⊕kL(λ)) ⊆ w(⊕kL(λ)). Hence 0 ∈
F (⊕kL(µ0)) = F (L(µ0)), equivalently µ0 ∈ w(L(λ)) and then we obtain
Λk(⊕kL(λ)) ⊆ w(L(λ)).
In addition, µ0 ∈ w(L(λ)) ⇒ 0 ∈ F (L(µ0)) =
∩kF (L(µ0)) ⊆ Λk(⊕kL(µ0)), according to a relation in [7]. Thus w(L(λ)) ⊆
Λk(⊕kL(λ)) and the proof is established.
The following result sketches the higher rank numerical range of a square
matrix through numerical ranges.
Theorem 5. Let A ∈ Mn(C). Then
Λk(A) =\M
F (M∗AM ),
where M is any n × (n − k + 1) isometry.
The preceding expression of Λk(A) indicates the "convexity of Λk(A)"
in another way, since the Toeplitz-Hausdorff theorem ensures that each
F (M∗AM ) is convex. For k = n, clearly Λn(A) = Tx∈Cn,kxk=1 F (x∗Ax)
and should be Λn(A) 6= ∅ precisely when A is scalar.
discs as in [2, 3], i.e.
By Theorem 5, we may also describe Λk(A) as intersections of circular
Λk(A) =\M
\γ∈C
D (γ,kM∗AM − γIn−k+1k2)
.
Since Λk(Iλ − A) is identified with the higher rank numerical range of a
matrix A ∈ Mn(C), Theorem 5 paves also the way for a characterization of
Λk(L(λ)), demonstrated in the next proposition.
4
Proposition 6. Suppose L(λ) =Pm
Λk(L(λ)) =\M
w(M∗L(λ)M ) =[N
j=1 Ajλj, then
where M ∈ Mn,n−k+1(C), N ∈ Mn,k(C) are isometries.
Proof. Obviously, by Theorem 5
Λk(N∗L(λ)N ),
Evenly, considering the equation Λk(A) =SN Λk(N∗AN ) [1], we have
µ0 ∈ Λk(L(λ)) ⇔ 0 ∈ Λk(L(µ0)) ⇔
F (M∗L(µ0)M ) ⇔ µ0 ∈\M
µ0 ∈ Λk(L(λ)) ⇔ 0 ∈ Λk(L(µ0)) ⇔
Λk(N∗L(µ0)N ) ⇔ µ0 ∈[N
0 ∈\M
0 ∈[N
w(M∗L(λ)M ).
Λk(N∗L(λ)N ).
We should note that Proposition 6 provides us an estimation of the
boundary of Λk(L(λ)) through the numerical approximation of the nume-
rical range w(L(λ)). Although the higher rank numerical range Λk(Iλ − A)
is always connected and convex [20, 23], Λk(L(λ)) need not satisfy these
properties, as we will see in the next example.
Example 1. Let
L(λ) = 3I5λ3 +" 1
2
3
4
5
0 −1 −2 −3 −4
2i 3i 4i 5i
i
−2 1
2
0.3 0
2
0
1
0
0 # λ2 +" 1 2 0 0 0
2 3 4 0 0
0 4 5 6 0
0 0 6 7 8
2 4 5 #
0 0 0 7 8# λ +" 4 −i 1 0 −2
i 2i −6i 1 0
4 2 0
0 1
−i 3i
0 2 4
3 1
3
2
1
0
−1
−2
−3
−4
i
s
x
A
y
r
a
n
g
a
m
i
I
−5
−3
−2
−1
0
Real Axis
1
2
3
The intersection of the numerical ranges w(M∗L(λ)M ) by 400 randomly
chosen 5×4 isometries M , approximates the set Λ2(L(λ)) and it is illustrated
by the areas of "white" holes inside the figure. Note that all figure constitutes
the numerical range w(L(λ)).
Investigating the non emptyness of Λk(L(λ)), it is noticed that the ne-
cessary and sufficient condition n ≥ 3k − 2 for Λk(A) 6= ∅ of A ∈ Mn [18]
fails in general for matrix polynomials, as shown in the next two results.
The first proposition refers to the emptyness of the set Λk(Aλ + B), where
A, B are n × n complex hermitian matrices.
5
Proposition 7. Let the n × n selfadjoint pencil L(λ) = Aλ + B such that
w(M∗(Aλ + B)M ) 6= C for any n × (n − k + 1) isometry M . If A is a
positive semidefinite matrix where the algebraic multiplicity of the eigenvalue
µA = 0 is greater than k − 1 and B is positive (or negative) definite, then
Λk(Aλ + B) = ∅, for any k = 2, 3, . . . , n.
Proof. Suppose B is a positive definite matrix. Due to the condition of
the multiplicity of µA = 0, the matrices M∗AM and M∗BM are positive
semidefinite and positive definite, respectively, for any n×(n−k+1) isometry
M . Moreover, w(M∗(Aλ+B)M ) 6= C and w(M∗(Aλ+B)M ) = (−∞,− 1
],
[21, Th.9], where νM is the maximum eigenvalue of (M∗BM )−1M∗AM .
Then, by Proposition 6, we obtain
νM
Λk(Aλ + B) =\M
w(M∗(Aλ + B)M ) =\M
(−∞,−
1
νM
] = Rc = ∅.
Similarly, if B is a negative definite matrix.
Moreover, in the next proposition Λk(L(λ)) appears to be non empty,
with L(λ) of special form.
Proposition 8. Let L(λ) = (λ − λ0)mAm be an n × n matrix polyno-
mial, where Am 6= 0 and 0 /∈ Λk(Am). Then Λk(L(λ)) is a singleton, i.e.
Λk(L(λ)) = {λ0}, λ0 ∈ C.
Proof. Since 0 /∈ Λk(Am), by Theorem 5, there exists an n × (n − k +
1) isometry M0 such that 0 /∈ F (M∗0 AmM0) and evenly w(M∗0 L(λ)M0) =
w((λ − λ0)mM∗0 AmM0) = {λ0}. Due to the special form of M∗L(λ)M ,
λ0 ∈ w(M∗L(λ)M ) for all n × (n − k + 1) isometries M , whereupon by
Proposition 6, we have
Λk(L(λ)) =\M
w(M∗L(λ)M ) = w(M∗0 L(λ)M0) = {λ0} .
In order to obtain Λk(L(λ)) 6= ∅ for any matrix polynomial L(λ) =
Pm
l=0 Alλl with Am 6= 0, we are led to the common roots of the k2 > 1
scalar polynomials bij(λ, Q) = q∗i L(λ)qj, i, j = 1, . . . , k for some isometries
qk(cid:3) ∈ Mn,k. Adapting the notion of the Sylvester matrix Rs
Q =(cid:2)q1
appeared in [15] and the discussion therein to the polynomials
. . .
bij(λ, Q) = q∗i Amqjλm + . . . + q∗i Alqjλl + . . . + q∗i A0qj
ij (Q)
ij (Q)λm + . . . + b(l)
ij (Q)λl + . . . + b(0)
= b(m)
(2.1)
for all i, j = 1, . . . , k and for some n × k isometry Q = (cid:2)q1
have a condition for the polynomials bij(λ, Q) to share polynomial common
qk(cid:3), we
. . .
6
factors. Denote by σ ≤ m to be the largest degree of the k2 polynomials
bij(λ, Q) and let, as in (2.1)
(2.2)
bi1,j1(λ, Q) = b(σ)
i1,j1(Q)λσ + . . . + b(l)
i1,j1(Q)λl + . . . + b(0)
i1,j1(Q),
If τ ≤ σ is the largest degree of the
for some indices i1, j1 ∈ {1, . . . , k}.
remaining polynomials, then the generalized Sylvester matrix is
Rs(Q) =
Rk2(Q)
R1(Q)
(2.3)
...
,
where R1(Q) is the stripped τ × (σ + τ ) matrix
···
b(σ)
i1,j1(Q)
b(σ−1)
i1,j1 (Q)
b(σ−1)
i1,j1 (Q)
b(σ)
i1,j1(Q)
. . .
0
b(σ)
i1,j1(Q)
b(0)
i1,j1(Q)
0
. . .
···
b(σ−1)
i1,j1 (Q)
. . .
···
b(0)
i1,j1(Q)
R1(Q) =
Rp(Q) =
and for p = 2, . . . , k2, Rp(Q) are the following σ × (σ + τ ) matrices
b(0)
ip,jp
b(τ )
ip,jp
(Q)
0
(Q)
·
·
b(τ )
ip,jp
(Q)
b(τ )
ip,jp
·
(Q)
·
·
·
·
·
b(0)
ip,jp
(Q)
0
with ip, jp ∈ {1, . . . , k} and ip 6= i1, jp 6= j1. Hence, the degree δ(Q) 6= 0
of the greatest common divisor of bij(λ, Q) (i, j = 1, . . . , k) for some n × k
isometry Q satisfies the relation
(2.4)
rankRs(Q) = τ + σ − δ(Q) ≤ 2m − δ(Q)
and clearly, Λk(L(λ)) 6= ∅ if and only if there exists an n × k isometry Q
such that rankRs(Q) < 2m.
Following, we investigate the boundedness of Λk(L(λ)) and we state the
next helpful lemma.
Lemma 9. Let A ∈ Mn(C) with A 6= 0. For the function f : Mn,k(C) →
Mk(C) defined by f (Q) = Q∗AQ, we have int(kerf ) = ∅.
Proof. For k = 1, let int(ker f ) 6= ∅ and a vector x0 ∈ Cn ∩ int(ker f ). Then
there exists an open ball B(x0, ε) ⊂ ker f with ε > 0. For any y ∈ Cn with
y ∈ B(0, ε) and real t < 1, clearly ty ∈ B(0, tε) ⊂ B(0, ε) and x0 + ty ∈
B(x0, ε). Hence,
f (x0 + y) = f (x0 + ty) = 0
7
q02
. . .
(2.5)
and consequently we have (t2 − t)y∗Ay = 0 for any y ∈ B(0, ε). Therefore,
A = 0, which is a contradiction.
For k > 1, suppose Q0 ∈ Mn,k(C) ∩ int(ker f ) and let the open ball
qk(cid:3) ∈ B(Q0, ε) and
B(Q0, ε) ⊂ ker f . If an n × k matrix Q =(cid:2)q1
denote Q0 =(cid:2)q01
for i = 1, . . . , k, where ei ∈ Cn is the i-th vector of the standard basis of Cn
and k·k2 is the spectral norm. Hence, by Q∗AQ = Q∗0AQ0 = 0 we obtain
f (qi) = q∗i Aqi = 0 and f (q0i) = q∗0iAq0i = 0 (i = 1, . . . , k) and by (2.5) we
conclude qi ∈ B(q0i, ε), i.e. B(q0i, ε) ⊂ ker f . This contradicts the emptyness
of int(ker f ) in the vector case.
kqi − q0ik2 = k(Q − Q0)eik2 ≤ kQ − Q0k2 < ε
q0k(cid:3), then
q2
. . .
Proposition 10. Let L(λ) = Amλm + Am−1λm−1 + . . . + A1λ + A0 be an
n× n matrix polynomial, where Am 6= 0. If 0 /∈ Λk(Am), then Λk(L(λ)) 6= ∅
is bounded.
Conversely, assume that rankRs(Q) < 2m, where Rs(Q) is the Sylvester
matrix in (2.3) of k2 scalar polynomials, elements of matrix Q∗L(λ)Q, for all
isometries Q ∈ Mn,k such that Q∗AmQ = zIk (z ∈ C\{0}). If Λk(Am) 6=
{0} and Λk(L(λ)) is bounded, then 0 /∈ Λk(Am).
Proof. Initially, we should remark that we investigate the boundedness of
Λk(L(λ)) taking into account the condition (2.4), so that it is not empty and
all the sets Λ1(L(λ)) ⊇ . . . ⊇ Λk−1(L(λ)) are not bounded. If 0 /∈ Λk(Am),
then by Theorem 5 there exists an n× (n− k + 1) isometry M0 such that 0 /∈
F (M∗0 AmM0). Hence, w(M∗0 L(λ)M0) is bounded [14] and by Proposition 6,
as Λk(L(λ)) ⊆ w(M∗0 L(λ)M0), we conclude that Λk(L(λ)) is bounded.
For the converse, suppose that 0 ∈ Λk(Am) 6= {0} and Λk(L(λ)) is
bounded. We may find a sequence {zν} ⊆ Λk(Am) such that limν→∞ zν = 0
and consequently, a sequence of n× k isometries {Qν} such that Q∗νAmQν =
zνIk → 0k. Due to the compactness of the group of n × k isometries, there
is a subsequence {Qρ} of {Qν} such that limρ→∞ Qρ = Q0, with Q0 ∈ Mn,k
be an isometry. Hence, by continuity, limρ→∞ Q∗ρAmQρ = Q∗0AmQ0 = 0k
and by Lemma 9, should be Q∗ρAmQρ = zρIk 6= 0. Note that in (2.3),
the Sylvester matrix Rs(Qρ) has dimensions k2m × 2m, since in (2.2), σ =
τ = m and due to rankRs(Qρ) < 2m, the equation Q∗ρL(λ)Qρ = 0k always
guarantees roots.
wise Λk(L(λ)) ≡ C) and evenly, (cid:13)(cid:13)Q∗ρAjQρ(cid:13)(cid:13) ≥ ε for some fixed ε > 0 and
Moreover, there exists an index j 6= m such that Q∗0AjQ0 6= 0k (other-
sufficiently large ρ. Hence, the (m − j)th elementary symmetric function
± 1
Q∗ρAjQρ of the roots of the matrix polynomial Q∗ρL(λ)Qρ [8, Th.4.2] is
not bounded, concluding that Λk(L(λ)) is not bounded. This contradicts
the assumption and the proof is complete.
zρ
8
Obviously, if L(λ) is a monic matrix polynomial, then Λk(L(λ)) is always
bounded. Following, we present an illustrative example of Proposition 10.
Example 2.
I. Let the matrix polynomial
L(λ) =
0
i
i
0
0
0
1
0
0
0
2
2
−i 0 −2 8
5
λ2 +
i 2 i 3
3 0 0 0
0 4 5 0
i 0 i 0
1 2 3 4
2 3 4 5
3 4 5 6
5 6 7 8
λ +
.
i
s
x
A
y
r
a
n
g
a
m
i
I
4
3
2
1
0
−1
−2
−3
−4
−5
−5
−4
−3
−2
−1
0
1
2
3
4
5
Real Axis
i
s
x
A
y
r
a
n
g
a
m
i
I
2
1.5
1
0.5
0
−0.5
−1
−1
−0.5
0
0.5
1
Real Axis
1.5
2
2.5
The uncovered area in the left picture approximates the set Λ2(L(λ)), which
is bounded, although Λ1(L(λ)) = C. The boundary of Λ2(A2) of the leading
coefficient A2 is illustrated on the right and we observe that 0 /∈ Λ2(A2).
II. For the converse, let the 4 × 4 matrix polynomial (m = 1)
0 0 0
0 1 0
0 1 0
0 0 0
0
0
0
= A1λ + A0.
λ +
3 0 0 0
0 0 0 0
0 0 0 0
0 0 0 4
L(λ) =
0 0
0
0 2
0
0 0 −1 0
0
0 0
Firstly, we observe that 0 ∈ Λ2(A1) = [0, 3]. On the other hand, Λ2(L(λ))
is equal to the bounded set {0}.
In fact, if we take the 4 × 3 isometries
M1 =(cid:20) 1 0 0
0 0 1(cid:21) and M2 =(cid:20) 1 0 0
0 0 1(cid:21), then M∗1 A1M1 and M∗1 A0M1 are both posi-
tive semidefinite matrices and consequently, [21, Th.9], w(M∗1 L(λ)M1) =
(−∞, 0]. Similarly, M∗2 A1M2, M∗2 A0M2 are positive and negative semi-
definite, respectively, which verifies w(M∗2 L(λ)M2) = [0,∞). Clearly,
Λ2(L(λ)) ⊆ w(M∗1 L(λ)M1) ∩ w(M∗2 L(λ)M2) = {0} and 0 ∈ Λ2(L(λ)) 6= ∅,
i.e. Λ2(L(λ)) = {0}.
In addition, for the isometry Q =
we have Q∗A1Q = I2 and
in (2.3) the Sylvester matrix Rs(Q) =
has rankRs(Q) = 2, not
less than 2, as it is required.
III. Consider the 4 × 4 matrix polynomial L(λ) = I2 ⊗ (Bλ + I2), with
0
1/√3
−√6/4 1/√3
√6/4 1/√3
1/2
0 −3√2/4
0 −3√2/4
0
3/8
1/3
1
1
9
B = [ 1 1
0 0 ]. Then Λ2(I2 ⊗ B) 6= {0} and additionally, 0 ∈ Λ2(I2 ⊗ B). In
this case, for any 4 × 2 isometry Q such that Q∗(I2 ⊗ B)Q = zI2 6= 02,
the Sylvester matrix in (2.3) is Rs(Q) =" 1 1/z
0 0 # with rankRs(Q) = 1 < 2.
Since, 0 ∈ F (A2), then w(L(λ)) as well as Λ2(L(λ)⊕ L(λ)) (Corollary 4) are
unbounded. It was expected by the converse of Proposition 10.
1 1/z
0 0
Further, we study the connectedness of Λk(L(λ)), attempting to specify
a bound for the number of its connected components.
Proposition 11. Let L(λ) = Amλm + . . . + A1λ + A0 be an n × n matrix
polynomial, with Am 6= 0 and let Λk(L(λ)) 6= ∅ have ρ connected components.
Moreover, rankRs(Q) < 2m, where Rs(Q) is the Sylvester matrix in (2.3)
of k2 polynomials (elements of Q∗L(λ)Q), for any n × k isometry Q such
that Q∗AmQ = γIk with γ ∈ Λk(Am) \ {0}.
If Λk(Am) \ {0} is connected, then ρ ≤ l ≤ m, where l is the minimum
number of distinct roots of the equation Q∗L(λ)Q = 0 for any n×k isometry
Q such that Q∗AmQ = γIk, with γ ∈ Λk(Am) \ {0}.
Otherwise, if Λk(Am) \ {0} = C1 ∪ C2, C1 ∩ C2 = ∅ and Ci, i = 1, 2 are
connected, then ρ ≤ l1 +l2 ≤ 2m, where li is the minimum number of distinct
roots of Q∗L(λ)Q = 0 for any n × k isometry Q that corresponds to points
γ ∈ Ci, for i = 1, 2.
Proof. Let C1 be a connected component of Λk(Am)\{0} and the n×k isome-
q1k(cid:3) correspond to Q∗0AmQ0 =
tries Q0 = (cid:2)q01
γ0Ik and Q∗1AmQ1 = γ1Ik, with γ0, γ1 ∈ C1. Evenly, we consider that
Q∗0L(λ)Q0 = 0k, Q∗1L(λ)Q1 = 0k and in particular, Q0 has the property
that provides the minimum number of distinct roots. We shall prove that
there exists a continuous function of isometries Q(t) : [0, 1] → Mn,k(C), with
Q(0) = Q0, Q(1) = Q1U for some unitary matrix U such that corresponds
to a continuous path γ(t) ∈ C1 joining γ0 to γ1.
origin, consider the continuous function
In case γ0 6= γ1 and the line segment joining γ0, γ1 does not contain the
q0k(cid:3), Q1 = (cid:2)q11
. . .
. . .
(2.6)
where U = diag(eiθ1 , . . . , eiθk ), with θj ∈ [0, 2π], j = 1, . . . , k and
Q(t) = (p1 − t2Q0 + tQ1U )C(t, U ),
t ∈ [0, 1],
C(t, U ) = diag(c−1
1 (t, θ1), . . . , c−1
k (t, θk)) ∈ Mk,
where cj(t, θj) = k√1 − t2q0j + teiθj q1jk2, j = 1, . . . , k. Clearly, Q(0) = Q0,
Q(1) = Q1U and Q∗(t)Q(t) = Ik, since the subspaces Kj = span{q0j, q1j}
are pairwise orthogonal for all j = 1, . . . , k [11, p.318]. Hence, after some
manipulations we obtain
Q∗(t)AmQ(t) = C(t, U )hγ(t)Ik + tp1 − t2(Q∗0AmQ1U + U∗Q∗1AmQ0)i C(t, U ),
10
where γ(t) = γ0 + t2(γ1 − γ0) for t ∈ [0, 1]. Moreover, according to the
conditions (i)-(iii) in the proof of Theorem 2.2 in [14], we may have a suitable
unitary matrix U0 = diag(eiθ01 , . . . , eiθ0k ) such that the matrix function
g(U ) = Q∗0AmQ1U + U∗Q∗1AmQ0
satisfies one of the following conditions:
(i) g(U0) = 0k,
(ii) g(U0) = ξ(γ1 − γ0)Ik for some real ξ 6= 0.
Then, Q∗(t)AmQ(t) = hγ0 + (t2 + ξt√1 − t2)(γ1 − γ0)i C 2(t, U0) 6= 0k and
for all j = 1, . . . , k the line segments hj(t) = γ0+(t2+ξt√1−t2)(γ1−γ0)
6= 0 join
the points γ0, γ1 without these necessarily be endpoints. Apparently, due to
the convexity of Λk(Am), we have that the isometries Q(t) generate the line
segment γ(t) ∈ C1.
In case the origin belongs to the line segment [γ0, γ1], (γ0 6= γ1), we may
consider another γ2 ∈ C1 such that γ2 6= γ0, γ1 and [γ0, γ2] ∪ [γ2, γ1] ⊆ C1.
This is true because of the convexity of Λk(Am) and the fact that the points
γ0, γ1 belong to the same connected component.
c2
j (t,θ0j )
Finally, if γ0 = γ1 and Am is a scalar matrix, then instead of (2.6)
consider the continuous function of n × k isometries
Q(t) = (p1 − t2Q0 + tQ1)C(t, Ik),
Otherwise, if Am is not scalar, we refer to (2.6).
t ∈ [0, 1].
Thus, we have constructed a continuous function of n × k isometries
Q(t) such that Q∗(t)AmQ(t) = γ(t)Ik 6= 0k, t ∈ [0, 1] and this asserts
that the Sylvester matrix Rs(Q(t)) ∈ Mk2m,2m for t ∈ [0, 1], since σ =
τ = m in (2.2). Hence, by the assumption rankRs(Q(t)) < 2m for all
t ∈ [0, 1], we have that the equation Q∗(t)L(λ(t))Q(t) = 0 has roots, let
λ1(t), . . . , λr(t) (r ≤ m). Due to the continuity of Q(t), the roots λj(t) :
[0, 1] → Λk(L(λ)) are continuous paths in Λk(L(λ)), connecting the roots
of equations Q∗0L(λ)Q0 = 0k and Q∗1L(λ)Q1 = 0k and thus the proof is
completed.
Example 3.
Let the 4 × 4 quadratic matrix polynomial
L(λ) =(cid:2) 2i 0
0 −2i(cid:3) ⊗ I2.
Obviously, Λ2(L(λ)) = {0} ∪ Λ2(Dλ + 4I4) and 0 /∈ Λ2(Dλ + 4I4) 6= ∅,
that is {0} is an isolated point. We also note that µ0 ∈ Λ2(Dλ + 4I4)
i/2i ⊗ I2) \ {0}, therefore Λ2(L(λ)) has three
if and only if µ−1
0 −2i(cid:3) ⊗ I2λ2 + 4I4λ = λ(Dλ + 4I4), with D =(cid:2) 2i 0
0 ∈ Λ2(h −i/2 0
0
11
connected components, two on the imaginary axis, the sets (−∞,−2], [2,∞)
and {0}. Moreover, for the 8 × 4 Sylvester matrix Rs(Q) in (2.3) we have
0 λ0 0 4# < 4 for all isometries Q ∈ M4,2 such that
rankRs(Q) = rank" λ0 0 4 0
Q∗DQ = λ0I2 6= 02. Also Λ2(D) \ {0} has two connected components and
Proposition 11 is confirmed.
0 λ0 0 4
λ0 0 4 0
3 Sharp points
In this section, following [16], we define the notion of sharp points. Parti-
cularly, z0 ∈ ∂Λk(L(λ)) is called to be a sharp point if for a connected
component Λ(s)
k (L(λ)) of Λk(L(λ)) there exist a disc S(z0, ε), with ε > 0
and two angles θ1 < θ2, with θ1, θ2 ∈ [0, 2π), such that
Re (eiθz0) = maxnRe z : e−iθz ∈ Λ(s)
k (L(λ)) ∩ S(z0, ε)o ∀ θ ∈ (θ1, θ2).
The following proposition presents a condition for a boundary point of
w(L(λ)) to be a boundary point of Λk(L(λ)), as well. We should remark
that the term 'multiplicity' as mentioned below is referred to the algebraic
multiplicity of an eigenvalue.
Proposition 12. Let the n × n matrix polynomial L(λ). If γ ∈ σ(L(λ)) ∩
∂w(L(λ)) with multiplicity k, then for j = 2, . . . , k
γ ∈ ∂Λj(L(λ)).
Proof. Clearly, by the assumption, γ is seminormal eigenvalue of the matrix
polynomial L(λ) of multiplicity k [13, Th.6]. That is, there exists a unitary
matrix U such that
U∗L(γ)U = 0k ⊕ R(γ),
where R(λ) is an (n − k) × (n − k) matrix polynomial and γ /∈ intw(R(λ)).
Hence, by Propositions 2(vi) and 3, it is implied that γ ∈ Λj(L(λ)) ⊆
Λj−1(L(λ)) for j = 2, . . . , k and due to γ /∈ intw(L(λ)) (≡ intΛ1(L(λ)), we
obtain γ ∈ ∂Λj(L(λ)), for j = 2, . . . , k.
For the pencil Iλ − A, we obtain the following corollary.
Corollary 13. Let A ∈ Mn(C). If γ ∈ ∂F (A) is eigenvalue of A of multi-
plicity k, then
γ ∈ ∂Λj(A),
j = 2, . . . , k.
The converse of Corollary 13 and consequently of Proposition 12 is not
true, as it is illustrated in the next example.
12
Example 4.
Let A = diag(3 + 4i, 4 − i,−3 − 2i,−3,−3 + 3i). The outer polygon of the
figure is F (A), whereas the inner shaded polygon is Λ2(A), which is the in-
tersection of all(cid:18)5
4(cid:19) convex combinations of the eigenvalues λj1, λj2, λj3, λj4
of A, with 1 ≤ j1 ≤ . . . ≤ j4 ≤ 5. Notice that λ0 = −3 ∈ ∂F (A) ∩ ∂Λ2(A),
but it is a simple eigenvalue of matrix A. In addition, Λ3(A) = ∅.
i
s
x
A
y
r
a
n
g
a
m
i
I
4
3
2
1
0
−1
−2
−3
−4
−4
−3
−2
−1
0
1
2
3
4
Real Axis
In view of the definition of sharp points, for a pencil Aλ − B, we have
the next proposition.
Proposition 14. Let the pencil L(λ) = Aλ− B ∈ Mn(C) and z0 be a sharp
point of w(Aλ− B) of multiplicity k with respect to the spectrum σ(Aλ− B),
then z0 is also a sharp point of Λj(Aλ − B), for j = 2, . . . , k.
Proof. Since the sharp point z0 of w(Aλ − B) is also an eigenvalue of the
pencil Aλ − B [17, Th.1.3], with multiplicity k by hypothesis, we deduce by
Proposition 12 that z0 ∈ ∂Λj(Aλ − B), for j = 2, . . . , k. It only suffices to
prove that for any disc S(z0, ε) with ε > 0, z0 satisfies the equality
Re(eiθz0) = maxnRe z : e−iθz ∈ Λj(Aλ − B) ∩ S(z0, ε)o
or equivalently, due to Proposition 6
Re(eiθz0) = max(Re z : z ∈\M (cid:16)w(eiθM∗(Aλ − B)M ) ∩ S(eiθz0, ε)(cid:17))
The inclusion relation w(M∗(Aλ− B)M ) ⊆ w(Aλ− B) for any n× (n−
for every angle θ ∈ (θ1, θ2) with 0 ≤ θ1 < θ2 < 2π.
j + 1) isometry M , j = 2, . . . , k verifies the inequality
(3.1)
max
TM (w(eiθM ∗(Aλ−B)M )∩S(eiθ z0,ε))
for any disc S(eiθz0, ε) and every θ ∈ (θ1, θ2).
Re z ≤
w(eiθ(Aλ−B))∩S(eiθ z0,ε)
max
Re z = Re(eiθz0)
13
Moreover, ker (Az0 − B) ∩ Im (M M∗) 6= ∅, since dim ker (Az0 − B) +
dim Im (M M∗) = k+n−j +1 ≥ n+1. Therefore, for an eigenvector x0 ∈ Cn
of Aλ − B corresponding to z0 there exists a vector y0 ∈ Cn such that x0 =
M M∗y0. Obviously, M∗y0 ∈ Cn−j+1 is an eigenvector of M∗(Aλ − B)M
corresponding to z0, yielding z0 ∈ σ(M∗(Aλ − B)M ) ⊆ w(M∗(Aλ − B)M )
for any n × (n − j + 1) isometry M.
Thus, z0 ∈TM w(M∗(Aλ − B)M ), i.e. Re z0 ∈ Re(TM w(M∗(Aλ − B)M )),
whereupon we confirm the inequality
(3.2)
Re (eiθz0) ≤
max
TM (w(eiθM ∗(Aλ−B)M )∩S(eiθz0,ε))
Re z
for any disc S(eiθz0, ε) and every θ ∈ (θ1, θ2). Therefore, by (3.1) and (3.2)
Re (eiθz0) = max(Re z : z ∈\M (cid:16)w(eiθM∗(Aλ − B)M ) ∩ S(eiθz0, ε)(cid:17))
for any disc S(eiθz0, ε) and every θ ∈ (θ1, θ2), establishing the assertion.
Because of the previous results, we obtain an interesting corollary con-
cerning the sharp points of the higher rank numerical range of a matrix
A ∈ Mn(C).
Corollary 15. Let A ∈ Mn(C) and z0 ∈ ∂F (A) be a sharp point of F (A)
of multiplicity k with respect to σ(A), then z0 is also a sharp point of Λj(A),
for j = 2, . . . , k.
Analogous statement to Proposition 14 for the "sharp points" of Λj(L(λ))
we may confirm taking into consideration Theorem 1.4 in [17].
4 Connection between Λk(L(λ)) and Λk(A)
mn companion pencil
Let the matrix polynomial L(λ) =Pm
i=0 Aiλi and the corresponding mn ×
0
0
...
In
Am−1
well known as linearization of L(λ), since there exist suitable matrix poly-
nomials E(λ), F (λ) with constant nonzero determinants such that
In
0
...
0
0
0
0
...
0
A0
0
0
...
0
Am
0
0
. . .
···
,
···
···
. . .
CL(λ) =
···
···
. . .
In
0
0
In
. . .
···
0
In
λ −
(cid:20)L(λ)
0
0
In(m−1)(cid:21) = E(λ)CL(λ)F (λ).
Next, we generalize a corresponding relation in [16] between the higher
rank numerical ranges of L(λ) and CL(λ).
14
Proposition 16.
Λk(L(λ)) ∪ {0} ⊆ Λk(CL(λ)).
Proof. By Proposition 6 and the relationship w(L(λ)) ∪ {0} ⊆ w(CL(λ)) in
[16], we have
(4.1) Λk(L(λ)) ∪ {0} = \M
w(M∗L(λ)M )! ∪ {0} ⊆\M
w(CM ∗LM (λ)),
where M ∈ Mn,n−k+1(C), with M∗M = In−k+1 and CM ∗LM (λ) is the
linearization of the matrix polynomial M∗L(λ)M . Since,
CM ∗LM (λ) = (Im ⊗ M )∗
···
= (Im ⊗ M )∗CL(λ)(Im ⊗ M ),
0
λIn −In
. . .
λIn −In
0
...
0
A0
···
···
. . .
0
0
...
−In
Amλ + Am−1
(Im ⊗ M )
Imn−k+1, we have
considering the isometry Q =(cid:2)Im ⊗ M V(cid:3) ∈ Mmn,mn−k+1(C), with Q∗Q =
\M
w(CM ∗LM (λ)) = \M
⊆ \Q
w((Im ⊗ M )∗CL(λ)(Im ⊗ M ))
w(Q∗CL(λ)Q) ⊆\X
w(X∗CL(λ)X) = Λk(CL(λ)),
(4.2)
where X ∈ Mmn,mn−k+1(C) with X∗X = Imn−k+1. Thus by (4.1) and (4.2)
the proof is completed.
Furthermore, Λk(L(λ)) appears to be associated with the joint higher
rank numerical range Λk(A) of an (m + 1)-tuple of n × n matrices A =
(A0, A1, . . . , Am). In fact,
Λk(L(λ)) = {λ ∈ C : P AmP λm + . . . + P A1P λ + P A0P = 0n , P ∈ Pk}
⊇ {λ ∈ C : (µmλm + . . . + µ1λ + µ0)P = 0n , (µ0, µ1, . . . , µm) ∈ Λk(A)}
= {λ ∈ C : µmλm + . . . + µ1λ + µ0 = 0 , (µ0, µ1, . . . , µm) ∈ Λk(A)}
= {λ ∈ C : h(1, λ, . . . , λm), ui = 0 , u = (µ0, µ1, . . . , µm) ∈ Λk(A)} .
The above inclusion justifies that Q∗AjQ may not be scalar matrices for
j = 0, . . . , m and for all isometries Q ∈ Mn,k(C).
sition 12.
The notion of the joint spectrum in [13], leads to an extension of Propo-
15
Proposition 17. Let A = (A0, . . . , Am) be an (m + 1)-tuple of n × n ma-
If (µ0, . . . , µm) ∈ ∂w(A) is a normal joint eigenvalue of A with
trices.
geometric multiplicity k, then
(µ0, . . . , µm) ∈ ∂Λj(A),
j = 2, . . . , k.
Proof. Since (µ0, . . . , µm) is a normal joint eigenvalue with geometric mul-
tiplicity k [13], there exists a unitary matrix U ∈ Mn(C) such that
(U∗A0U, . . . , U∗AmU ) = (µ0Ik ⊕ B0, . . . , µmIk ⊕ Bm),
where (B0, . . . , Bm) is an (m + 1)-tuple of (n − k) × (n − k) matrices and
(µ0, . . . , µm) /∈ σ(B0, . . . , Bm). Thus, (µ0, . . . , µm) ∈ Λk(A). Since the point
(µ0, . . . , µm) ∈ ∂w(A) and Λj(A) ⊆ Λj−1(A) for every j = 2, . . . , k [19], we
establish (µ0, . . . , µm) ∈ ∂Λj(A) for all j = 2, . . . , k.
Finally, we obtain the following result relative to that in [22].
i=0 Aiλi. Then
Λk(L(λ)) ⊇ {λ ∈ C : h(1, λ, . . . , λm), ui = 0 , u ∈ coΛk(A)} ,
Proposition 18. Let the matrix polynomial L(λ) =Pm
where A = (A0, A1, . . . , Am) is the (m + 1)-tuple of n × n matrices Ai.
Proof. Let Ω = {λ ∈ C : h(1, λ, . . . , λm), ui = 0 , u ∈ coΛk(A)}. To prove
the inclusion, suppose λ0 ∈ Ω, that is
(4.3)
for some u = (u0, u1, . . . , um) ∈ coΛk(A). We have Λk(A) ⊆ Cm+1 ≡ R2m+2
and by Caratheodory's theorem in Convex Analysis, there are at most 2m+3
elements of Λk(A) such that
h(1, λ0, . . . , λm
0 ), ui = 0
ρ
ρ
Xj=1
Xj=1
µjuj : uj ∈ Λk(A), µj ≥ 0 ,
µj = 1, with ρ ≤ 2m + 3
coΛk(A) =
Hence, for u = (u0, u1, . . . , um) ∈ coΛk(A) there are suitable µj ≥ 0,
Pρ
j µj = 1, ρ ≤ 2m + 3 such that
u = µ1u1 + . . . + µρuρ = µ1
+ . . . + µρ
uρ0
...
uρm
u10
...
u1m
(4.4)
.
,
and (4.4), we obtain:
where uj =(cid:2)uj0
h(1, λ0, . . . , λm
. . . ujm(cid:3)T ∈ Λk(A), j = 1, . . . , ρ and by equations (4.3)
0 (cid:3)
uρ0
...
+ . . . + µρ(cid:2)1 . . . λm
uρm
0 (cid:3)
0 ), ui = µ1(cid:2)1 . . . λm
u10
...
u1m
16
i.e.
(4.5)
µ1p1(λ0) + . . . + µρpρ(λ0) = 0,
where pj(λ) = ujmλm + . . . + uj1λ + uj0 for j = 1, . . . , ρ. Evenly, by uj =
(uj0, . . . , ujm) ∈ Λk(A), there exist rank-k orthogonal projections Pj, j =
1, . . . , ρ such that PjAiPj = ujiPj, i = 0, . . . , m and consequently
pj(λ0)Pj = uj0Pj + uj1Pjλ0 + . . . + ujmPjλm
0
= PjA0Pj + PjA1Pjλ0 + . . . + PjAmPjλm
0
= PjL(λ0)Pj,
which means that pj(λ0) ∈ Λk(L(λ0)). Due to the convexity of the higher
rank numerical range of the matrix L(λ0) and the equation (4.5), 0 ∈
Λk(L(λ0)), equivalently λ0 ∈ Λk(L(λ)).
References
[1] Aik. Aretaki and J. Maroulas, Presentation at the 10th Workshop on "Numeri-
cal Ranges and Numerical Radii", Krakow, Poland, 2010.
[2] F.F. Bonsall and J. Duncan, Numerical Ranges of Operators on Normed Spaces
and of Elements of Normed Algebras, London Mathematical Society Lecture
Note Series, Cambridge University Press, New York, 1971.
[3] F.F. Bonsall and J. Duncan, Numerical Ranges II, London Mathematical So-
ciety Lecture Notes Series, Cambridge University Press, New York, 1973.
[4] M.D. Choi, M. Giesinger, J.A. Holbrook and D.W. Kribs, Geometry of higher-
rank numerical ranges, Linear and Multilinear Algebra, 56(1), 53-64, 2008.
[5] M.D. Choi, J.A. Holbrook, D.W. Kribs and K. Zyczkowski, Higher-
preprint,
of unitary and normal matrices,
rank numerical
http://arxiv.org/quant-ph/0608244.
ranges
[6] M.D. Choi, D.W. Kribs and K. Zyczkowski, Quantum error correcting codes
from the compression formalism, Reports on Mathematical Physics, 58, 77-86,
2006.
[7] M.D. Choi, D.W. Kribs and K. Zyczkowski, Higher-rank numerical ranges
and compression problems, Linear Algebra and its Applications, 418, 828-839,
2006.
[8] J.E. Dennis, Jr., J.F. Traub, R.P. Weber, The Algebraic Theory of Matrix
Polynomials, SIAM Journal on Numerical Analysis, 13, 831-845, 1976.
[9] I.Z. Emiris, E.D. Fritzilas and D. Manocha, Algebraic Algorithms for struc-
ture Determination in Biological chemistry, International Journal of Quantum
Chemistry, 106, 190-210, 2006.
[10] K.E. Gustafson and D.K.M. Rao, Numerical Range. The Field of Values of
Linear Operators and Matrices, Springer-Verlag, New York, 1997.
17
[11] P.R. Halmos, A Hilbert Space Problem Book, 2nd Ed., Springer-Verlag, New
York, 1982.
[12] R.A. Horn and C.R. Johnson, Topics in Matrix Analysis, Cambridge University
Press, Cambridge, 1991.
[13] P. Lancaster and P. Psarrakos, Normal and seminormal eigenvalues of matrix
functions, Integral Equations and Operator Theory, 41, 331-342, 2001.
[14] C.K. Li, L. Rodman, Numerical range of matrix polynomials, SIAM J. Matrix
Analysis and Applications, 15, 1256-1265, 1994.
[15] J. Maroulas and D. Dascalopoulos, Applications of the Generalized Sylvester
Matrix, Applied Math. and Computations, 8, 121-135, 1981.
[16] J. Maroulas and P. Psarrakos, Geometrical Properties of numerical range of
matrix polynomials, Computers Math. Applic., 31, 41-47, 1996.
[17] J. Maroulas and P. Psarrakos, The boundary of the numerical range of matrix
polynomials, Linear Algebra and its Applicatrions, 267, 101-111, 1997.
[18] C.K. Li, Y.T. Poon and N.S. Sze, Condition for the higher rank numerical
range to be non-empty, Linear and Multilinear Algebra, 57(4), 365-368, 2009.
[19] C.K Li and Y.T. Poon, Quantum error correction and generalized numerical
ranges, preprint, http://arxiv.org/0812.4772v1 [math.FA], 2008.
[20] C.K. Li and N.S. Sze, Canonical forms, higher rank numerical ranges, totally
isotropic subspaces, and matrix equations, Proceedings of the American Math-
ematical Society, 136, 3013-3023, 2008.
[21] P.J. Psarrakos, Numerical range of linear pencils, Linear Algebra and its Ap-
plicatrions, 317, 127-141, 2000.
[22] P.J. Psarrakos and M.L. Tsatsomeros, On the relation between the numerical
range and the joint numerical range of matrix polynomials, Electronic Journal
of Linear Algebra, 6, 20-30, 2000.
[23] H.J. Woerdeman, The higher rank numerical range is convex, Linear and Mul-
tilinear Algebra, 56(1), 65-67, 2007.
18
|
1507.08361 | 1 | 1507 | 2015-07-30T02:18:26 | The characteristic polynomial of an algebra and representations | [
"math.RA"
] | In this short note, we give a new sufficient condition for a linear map from a product of copies of a field to endomorphisms of a finite dimensional vector space over the same field to be an algebra homomorphism. We expect that this result can be applied to study representations of higher-degree Clifford algebras and finite extensions of commutative rings. | math.RA | math |
The characteristic polynomial of an algebra and
representations
Rajesh S. Kulkarni ∗, Yusuf Mustopa †, and Ian Shipman ‡
January 17, 2021
Suppose that k is a field and let A be a finite dimensional, associative, unital k-algebra. Often one
is interested in studying the finite-dimensional representations of A. Of course, a finite dimensional
representation of A is simply a finite dimensional k-vector space M and a k-algebra homomorphism
A → Endk(M ).
In this article we will not consider representations of algebras, but rather how
to determine if a k-linear map φ : A → Endk(M ) is actually a homomorphism. We restrict our
attention to the case where A is a product of field extensions of k.
If φ : A → Endk(M ) is a
representation then certainly, if a ∈ A satisfies am = 1 then φ(a)m = id as well. Our first Theorem
is a remarkable converse to this elementary observation.
Theorem A. Suppose that A = kd and φ : A → Endk(M ) is a k-linear map. Let n > 2 be a
natural number and assume that k has n primitive nth roots of unity. If φ(1A) = id and for each
a ∈ A such that an = 1A we have φ(a)n = id, then φ is an algebra homomorphism.
Consider the regular representation µL : A → Endk(A) of A on itself by left multiplication. For
a ∈ A, let χa(t) and χa(t) be the characteristic and minimal polynomials of µL(a), respectively.
We note that χa(a) = χa(a) = 0 in A. Therefore if M is a finite dimensional left A module with
structure map φ : A → Endk(M ) then χa(φ(a)) = χa(φ(a)) = 0 in Endk(M ). The notion of
assigning a characteristic polynomial to each element of an algebra and considering representations
which are compatible with this assignment has appeared in [Pro87]. This idea has been applied
to some problems in noncommutative geometry as well [LB03]. However, as far as we know the
following related notion is new.
Definition 1. Suppose that φ : A → Endk(M ) is a k-linear map, where M is a finite dimensional
k-vector space. We say that φ is a characteristic morphism if χa(φ(a)) = 0 for all a ∈ A. We say
that φ is minimal-characteristic if, moreover, χa(φ(a)) = 0 for all a ∈ A.
It is natural to ask whether or not the notions of characteristic morphism and minimal characteristic
morphism are weaker than the notion of algebra morphism. Let us address minimal-characteristic
morphisms first.
Corollary. Assume that A = kd and that k has a full set of dth roots of unity. Then a minimal-
characteristic morphism φ : A → Endk(M ) is an algebra morphism.
Proof. First note that χ1(t) = t − 1. Hence φ(1) = id. Furthermore if a ∈ A satisfies ad = 1 then
χa(t) divides td − 1. Therefore, φ(a)d = id. Hence, if d > 2 then Theorem A implies that φ is an
algebra morphism. We leave the cases d = 1, 2 for the reader.
(cid:4)
∗Michigan State University, East Lansing, Michigan. [email protected]
†Tufts University, Medford, Massachusetts. [email protected]
‡University of Utah, Salt Lake City, Utah. [email protected]
1
Example 2. Let a, b ∈ k be such that a + b 6= 0. Then the map φ : k×2 → Mat2(k) given by
φ(e1) =(cid:18)1 a
0(cid:19) , φ(e2) =(cid:18)0
0 1(cid:19)
0
b
is a characteristic morphism that is not a representation.
Characteristic morphisms form a category in a natural way. Any linear map φ : A → Endk(M )
endows M with the structure of a T(A) module, where T(A) denotes the tensor algebra on A. We
can view the characteristic polynomial of elements of A as a homogeneous form χ(t) ∈ Sym•(A∨)[t]
of degree d, monic in t. Pappacena [Pap00] associates to such a form an algebra
where if χa(t) =Pd
i=0 ci(a)ti then
C(A) =
T(A)
hχa(a) : a ∈ Ai
,
χa(a) :=
d
Xi=0
ci(a)a⊗i ∈ T(A).
Clearly, the action map φ : A → Endk(M ) of a T(A)-module M is a characteristic morphism if and
only if the action of T(A) factors through C(A). We declare the category of characteristic morphisms
to be the category of finite-dimensional C(A)-modules. So we have a notion of irreducible charac-
teristic morphism. The characteristic morphism constructed in Example 2 is not irreducible, being
an extension of two irreducible characteristic morphisms. However, every irreducible characteristic
morphism k×2 → Endk(M ) is actually an algebra morphism. The following example shows this is
not always the case.
Example 3. The linear map φ : k×3 → Mat3(k) defined by
e1 7→
1
2
1
1
1 −1
0
0
0 −1
1
,
e3 7→
1 −1
1
−1
1
1
1
2
0
0
0
e2 7→
1
2
0 −1
1
0
1
1
−1 0
1
,
is an irreducible characteristic morphism, but not an algebra morphism; while e2
checked that φ(e1)2 6= e1.
1 = e1, it can be
Given a linear map φ : A → Endk(M ), let Tφ ∈ A∨ ⊗ Endk(M ) be the element that corresponds
to φ under the isomorphism Homk(A, Endk(M )) ∼= A∨ ⊗ Endk(M ). We view Tφ as an element of
Sym•(A∨) ⊗ Endk(M ). The equation χa(φ(a)) = 0 for all a ∈ A holds if and only if χA(Tφ) = 0 in
Sym•(A∨)⊗ Endk(M ). We can just as easily view Tφ as an element of T(A∨)⊗ Endk(M ). Moreover,
we can lift χ from Sym•(A∨)[t] to T(A∨) ∗ k[t] by the naıve symmetrization map Sym•(A∨)[t] →
T(A∨) ∗ k[t].
Theorem B. Assume that char(k) is either 0 or greater than d. Let A = k×d and φ : A → Endk(M )
a k-linear map. The map φ factors through a homomorphism A → Endk(M ) if and only if χ(T ) = 0
in T(A∨) ⊗k B.
Acknowledgments. We would like to thank Ted Chinburg and Zongzhu Lin for interesting con-
versations. I.S. was partially supported during the preparation of this paper by National Science
Foundation award DMS-1204733. R. K. was partially supported by the National Science Foundation
awards DMS-1004306 and DMS-1305377.
2
Proofs
We now turn to the proofs of the results in the introduction. The proof of the Theorem A depends
on an arithmetic Lemma.
Lemma 4. Let ζ ∈ k be a primitive nth root of unity. Suppose that a, b, c, d ∈ Z satisfy b, d 6= 0
mod n and
ζ a − 1
ζ b − 1
=
ζ c − 1
ζ d − 1
.
Then either:
1. a ≡ b mod n and c ≡ d mod n, or
2. a ≡ c mod n and b ≡ d mod n.
Proof. After possibly passing to a finite extension we may assume that k admits an automorphism
sending ζ to ζ −1. Thus we have
ζ −a − 1
ζ −b − 1
=
ζ −c − 1
ζ −d − 1
,
which we rewrite
ζ −a
ζ −b ·
1 − ζ a
1 − ζ b =
ζ −c
ζ −d ·
1 − ζ c
1 − ζ d .
Using our assumption we find that ζ b−a = ζ d−c. Thus b − a ≡ d − c mod n. Let e = b − a ≡ d − c
mod n. Then we have
ζ b−e − 1
ζ b − 1
=
ζ d−e − 1
ζ d − 1
which implies that
Finally we see that
ζ b−e + ζ d = ζ d−e + ζ b.
ζ d − ζ b = (ζ d − ζ b)ζ −e
Therefore either e ≡ 0 mod d so that (1) holds, or d ≡ b so that (2) holds.
(cid:4)
Proof of Theorem A. Whether or not φ is an algebra homomorphism is stable under passage to the
algebraic closure of k. So we may assume that k is algebraically closed. Let e1, . . . , ed ∈ A be a
complete set of orthogonal idempotents. Put αi = φ(ei) and note that by hypothesis α1 + · · · + αd =
id. Fix a primitive nth root of unity ξ. Then x = 1 + (ξ − 1)ei satisfies xn = 1. Therefore φ(x)d = id.
This implies that φ(x) is diagonalizable and each eigenvalue is an nth root of unity. Now, since φ is
linear,
αi =
φ(x) − id
ξ − 1
and hence αi is diagonalizable as well. Let λ be an eigenvalue of αi. Then for some a we have
Now for any b, φ(1 + (ξb − 1)ei)d = id. So we see that
λ =
ξa − 1
ξ − 1
.
1 + λ(ξb − 1)
3
must be a root of unity for every b. However, if
1 + λ(ξb − 1) = ξc
then Lemma 4 implies that either a ≡ 1 mod n, λ = 0, or b ≡ 1 mod n. Now, b is under our
control and since n ≥ 3 we can choose b 6= 0, 1 mod n, excluding the third case. If a ≡ 1 mod n
then λ = 1 and otherwise λ = 0. Thus αi is semisimple with eigenvalues equal to zero or one. So
α2
i = αi.
Let i 6= j and consider
ya = id +(ξa − 1)(αi + αj)
Clearly, yn
a = id and thus ya is semisimple. We compute
(ya − id)2 = (ξa − 1)2(αiαj + αjαi) + (ξa − 1)(ya − id)
and deduce that
(ξa − 1)−2(ya − id)(ya − ξa) = (αiαj + αjαi).
(1)
Assume that b 6= 0 mod n. Observe that ya − id = ξa−1
simultaneously diagonalizable. Suppose that ξc is an eigenvalue of yb unequal to 1. Then
ξb−1 (yb − id) and therefore, ya and yb are
ξa − 1
ξb − 1
(ξc − 1) + 1 = ξe
is an eigenvalue of ya. Since n ≥ 3 we can assume that a 6= b, 0 mod n. Then Lemma 4 implies
that e ≡ a mod n and b ≡ c mod n. We see that the only eigenvalues of yb are 1 and ξb.
Because yb is semisimple, this means that the right side of (1) vanishes. So αiαj = −αjαi for all i, j.
Suppose that αi(m) = m. Then αj(αi(m)) = αj(m) = −αi(αj(m)). Since −1 is not an eigenvalue
of αi we see that αj(m) = 0. Now let m ∈ M and write m = m0 + m1 where αi(m0) = 0 and
αi(m1) = m1. Then
αi(αj(m)) = αi(αj (m0)) = −αj(αi(m0)) = 0.
Thus we see that in fact αiαj = 0. So α1, . . . , αd satisfy the defining relations of k×d and φ is
actually an algebra homomorphism.
(cid:4)
We now turn to the proof of Theorem B. The key idea is to use the fact that the single equation
χ(Tφ) = 0 over the tensor algebra encodes many relations for the matrices defining φ. It is convenient
to consider αi = φ(ei), where ei is the standard basis of idempotents in k×d. Furthermore we write
χd for the characteristic polynomial of k×d viewed as an element of khx1, . . . , xd, ti (where x1, . . . , xd
is the dual basis to e1, . . . , ed).
Lemma 5. Suppose that k is a field with char(k) > d. Let α1, . . . , αd ∈ Mn(k) and put T =
x1α1 + . . . + xdαd. If T satisfies χd then
1. for some i = 1, . . . , d, αi has a 1-eigenvector, and
2. if m ∈ kn satisfies αim = m then αjm = 0 for all j 6= i.
Proof. (1.) Let S = k[x1, . . . , xd] as an A = khx1, . . . , xdi module in the obvious way. Then the
image of χd in k[x1, . . . , xd, t] is p(t) = n!(t − x1) · · · (t − xd), where now the order of the terms
does not matter. Hence T satisfies (T − x1) · · · (T − xd) = 0 in Mn(S). So we can view Sn as an
R = k[x1, . . . , xd, t]/(p(t))-module M . For each i consider the quotient Si := R/(t − xi), which is
isomorphic to S under the natural map S → Si. Define Mi = M ⊗R Si. Since the map S → S1 ×· · ·×
4
Sd is an isomorphism after inverting a =Qi6=j(xi − xj ) and a is a nonzerodivisor on M , the natural
map M → M1 ⊕· · ·⊕Md is injective. Hence there is some i such that Mi has positive rank. Consider
¯M := M/(x1, . . . , xi−1, xi − 1, xi+1, . . . , xd)M and ¯Mi := Mi/(x1, . . . , xi−1, xi − 1, xi+1, . . . , xd)Mi.
Now since Mi (is finitely generated and) has positive rank ¯Mi 6= 0. Observe that since M = Sd,
the natural map kd → ¯M is an isomorphism. Moreover the action of t on ¯M is identified with the
action of αi. Now, ¯Mi = ¯M /(t − xi) ¯M = ¯M /(αi − 1) ¯M 6= 0. Hence αi − 1 is not invertible, αi − 1
has nonzero kernel, and αi has a 1-eigenvector.
(2.) Let us compute χ(x1, . . . , xd, T ). We denote by δi
j the Kronecker function. We have
d
d
xiαi − xσ(d))
(
xiαi − xσ(1)) · · · (
Xi=1
Xi=1
χd(x1, . . . , xd, T ) = Xσ∈Sd
= Xσ∈Sd
Xi=1
Yj=1(cid:0)
σ(j))(cid:1)
xi1 · · · xim(cid:0) Xσ∈Sd
= X1≤i1,...,id≤d
xi(αi − δi
d
d
(αi1 − δi1
σ(1)) · · · (αid − δid
σ(d))(cid:1).
In the second line the term order matters so the product is taken in the natural order j = 1, 2, . . . , d.
Now suppose that χd(x1, . . . , xd, T ) = 0. Then for all 1 ≤ i1, . . . , id ≤ d we have
(αi1 − δi1
σ(1)) · · · (αid − δid
σ(d)) = 0.
Xσ∈Sd
For each j 6= i, we consider the noncommutative monomial xixjxd−2
i
and its equation (2),
(αi − δi
σ(1))(αj − δj
σ(2))(αi − δi
σ(3)) · · · (αi − δi
σ(d)) = 0.
Xσ∈Sd
(2)
(3)
Note that since αi(m) = m, we calculate
(αi − δi
σ(3)) · · · (αi − δi
σ(d))m =(m i /∈ {σ(3), . . . , σ(d)},
i ∈ {σ(3), . . . , σ(d)}.
0
Therefore applying (3) to m and simplifying we get
Xσ∈Sd,σ(1)=i
(αi − 1)(αj − δj
σ(2))m + Xσ∈Sd,σ(2)=i
αiαjm = (d − 1)!(cid:0)(αi − 1)(αj − δjσ(2))m + αiαjm(cid:1)
= (d − 1)!(cid:0)(αi − 1)αjm + αiαjm(cid:1)
= (d − 1)!(2αi − 1)αjm
= 0,
where passing from the first line to the second we use the fact that (αi − 1)δj
σ(2)m = 0.
Now, consider the special case of (2) corresponding to xd
i :
(αi − δi
σ(1)) · · · (αi − δi
σ(d)) =
Xσ∈Sd
d
Xj=1 Xσ∈Sd,σ(j)=i
αj−1
i
(αi − 1)αd−j−1
i
= d!αd−1
i
(αi − 1) = 0.
Since αd−1
i
(αi−1) = 0 it follows that 2αi−1 is invertible. However, (2αi−1)αjm = 0 so αjm = 0. (cid:4)
5
Proof of Theorem B. (⇐) We proceed by induction on dim(M ) and fix an identification M ∼= kn.
Suppose n = 1. Then by Lemma 5, there is some i and some m ∈ k1 such that αi(m) = m.
Moreover, αjm = 0 for all j 6= 0. Since m spans k1, the αi satisfy the neccessary relations for φ to
factor through an algebra morphism.
Now given α1, . . . , αd ∈ Mn(k), Lemma 5 implies that we can find an element m ∈ kd such that
km ⊂ kd is stable under the action of α1, . . . , αd. Let Mn(k, m) ⊂ Mn(k) be the algebra of operators
that preserve km. Then there is a surjective algebra homomorphism Mn(k, m) ։ Mn−1(k). Since
α1, . . . , αd ∈ Mn(k, m) we find that T ∈ Mn(k, m) ⊗k khx1, . . . , xdi. So if α′
d ∈ Mn−1(k)
are the images of α1, . . . , αd then T ′ = x1α′
d satisfies χd. By induction we see that
j = 0 for i 6= j. In particular, there is a codimension 1 subspace of kn−1 preserved
(α′
d. Its inverse image in kn (we identify kn−1 with kn/km) is then a codimension one
by α′
subspace V ′ ⊂ kd which is invariant under α1, . . . , αd. Again by induction, α2
i − αi and αiαj(i 6= j)
annihilate V ′. There is some i such that αi acts by the identity on kn/V ′. Since αd−1
(αi − 1) = 0,
the geometric multiplicity of 1 as an eigenvalue of αi is equal to its algebraic multiplicity. So there
is a 1-eigenvector m ∈ kn whose image in kn/V ′ is nonzero. Again Lemma 5 implies that αjm = 0
for j 6= i. Hence the relations α2
i − αi and αiαj annihilate a basis for kn and hence annihilate kn.
1 + . . . + xdα′
1, . . . , α′
1, . . . , α′
i)2 = α′
i and α′
iα′
i
(⇒) Suppose that φ is an algebra map. Then we have α2
i = αi for all i and αj αi = 0 if i 6= j.
Decompose kn = V1 ⊕ . . . ⊕ Vd where Vi = αi(kn). Then T preserves Vi ⊗ khx1, . . . , xdi for each
i=1 Endk(Vi) ⊗ khx1, . . . , xdi ⊂ Mn(khx1, . . . , xdi). Since
(T − xi) vanishes identically on Vi ⊗ khx1, . . . , xdi we see that for each σ ∈ Sd and each i the image
of (T − xσ(1)) · · · (T − xσ(d)) vanishes in Endk(Vi) ⊗ khx1, . . . , xdi and hence in Mn(khx1, . . . , xdi).
Since all of the terms of χd(T ) vanish in Mn(khx1, . . . , xdi), so does χd(T ).
(cid:4)
i. So we can view T as an element of Qd
Questions
There are many natural questions that surround the notion of characteristic morphism. We point
out a few of them.
Question 1. What are the irreducible characteristic morphisms for A = k×d? Are there infinitely
many for d ≥ 3?
Replacing a commutative semisimple algebra with a semisimple algebra, Theorem A fails to hold.
Indeed, the map φ : Matd(k) → Matd(k) defined by φ(M ) = M T is not a homomorphism, but does
satisfy the hypotheses of Theorem A. Moreover, φ is a characteristic morphism.
Question 2. Is there a characterization of when a linear map φ : Matd(k) → Matr(k) is a homo-
morphism along the lines of Theorem A?
Let V is a finite dimensional vector space and F (t) ∈ Sym•(V ∨)[t] be monic and homogeneous.
Given v ∈ V we can consider the image Fv(t) of F (t) under the homomorphism Sym•(V ∨)[t] → k[t]
induced by v : V ∨ → k. The main theorem of [CK15] implies that there always exists a linear
map φ : V → Matr(k) for some r such that Fv(φ(v)) = 0 for all v ∈ V . There is a natural
non-commutative generalization of this problem.
Question 3. For which monic, homogenous elements F (t) of T(V ∨) ∗ k[t], does there exist an
element φ∨ ∈ V ∨ ⊗ Matr(V ) for some r such that F (φ∨) = 0 in T(V ∨) ⊗ Matr(k)?
If F (t) is the symmetrization of the characteristic polynomial of an algebra structure on V then we
have an affirmative answer. However, if F (t) = t2 − u ⊗ v where u, v are linearly independent, then
there is no such element.
6
References
[CK15] Adam Chapman and Jung-Miao Kuo. On the generalized Clifford algebra of a monic
polynomial. Linear Algebra Appl., 471:184 -- 202, 2015.
[LB03] L. Le Bruyn. Three talks on noncommutative geometry @n.
arXiv:math/0312221
[math.AG], 2003.
[Pap00] Christopher J. Pappacena. Matrix pencils and a generalized Clifford algebra. Linear Algebra
Appl., 313(1-3):1 -- 20, 2000.
[Pro87] Claudio Procesi. A formal inverse to the Cayley-Hamilton theorem. J. Algebra, 107(1):63 --
74, 1987.
7
|
1309.7325 | 2 | 1309 | 2015-07-03T18:08:04 | A rational construction of Lie algebras of type E_7 | [
"math.RA",
"math.AG"
] | We give an explicit construction of Lie algebras of type $E_7$ out of a Lie algebra of type $D_6$ with some restrictions. Up to odd degree extensions, every Lie algebra of type $E_7$ arises this way. For Lie algebras that admit a $56$-dimensional representation we provide a more symmetric construction based on an observation of Manivel; the input is seven quaternion algebras subject to some relations. | math.RA | math |
A rational construction of Lie algebras
of type E7
Victor Petrov∗
January 25, 2020
Abstract
We give an explicit construction of Lie algebras of type E7 out of
a Lie algebra of type D6 with some restrictions. Up to odd degree
extensions, every Lie algebra of type E7 arises this way. For Lie al-
gebras that admit a 56-dimensional representation we provide a more
symmetric construction based on an observation of Manivel; the input
is seven quaternion algebras subject to some relations.
1
Introduction
In [16] Jacques Tits wrote the following: "It might be worthwile trying to
develop a similar theory for strongly inner groups of type E7. For instance,
can one give a general construction of such groups showing that there exist
anisotropic strongly inner K-groups of type E7 as soon as there exist cen-
tral division associative 16-dimensional K-algebras of order 4 in Br K whose
reduced norm is not surjective?"
The goal of the present paper is to give such (and much more general)
construction. We deal with Lie algebras; of course, the corresponding group
is just the automorphism group of its Lie algebra. By rational constructions
we mean those not appealing to the Galois descent, that is involving only
terms defined over the base field.
Let us recall several milestones in the theory. Freudenthal in [6] gave
an elegant explicit construction of the split Lie algebra of type E7. On the
∗This research is supported by Russian Science Foundation grant 14-21-00035
1
language of maximal Lie subalgebras it is a particular case of A7-construction.
Another approach was proposed by Brown in [4] (see also [7] for a recent
exposition); this is an E6-construction. It gives only isotropic Lie algebras.
In full generality A7-construction was described by Allison and Faulkner in
[1] as a particular case of a Cayley-Dickson doubling; generically it produces
anisotropic Lie algebras of type E7. Another construction with this property
was discovered by Tits in [16]; in our terms it is an A3 +A3 +A1-construction.
On the other hand, some Lie algebras of type E7 can be obtained via the
Freudenthal magic square, see [14] (or [8] for a particular case).
Our strategy is to define a Lie triple system structure on the (64-dimen-
sional over F ) simple module of the even Clifford algebra of a central simple
algebra of degree 12 with an orthogonal involution under some restrictions.
Then the embedding Lie algebra is of type E7. Our construction is of type
D6 + A1.
For Lie algebras that admit a 56-dimensional representation we provide
another construction based on an observation of Manivel [11] and a result of
Qu´eguiner -- Mathieu and Tignol [12]; it is of type 7A1.
The author is grateful to Ivan Panin, Anastasia Stavrova and Alexander
Luzgarev for discussing earlier attempts to this work, to Nikolai Vavilov for
pointing out Manivel's paper [11] and to Anne Qu´eguiner -- Mathieu for her
hospitality and fruitful discussions.
2 Lie triple systems and quaternionic gifts
Let F be a field of characteristic not 2. Recall that a Lie triple system is a
vector space W over F together with a trilinear map
W × W × W → W
(u, v, w) 7→ [u, v, w] = D(u, v)w
satisfying the following axioms:
D(u, u) = 0
D(u, v)w + D(v, w)u + D(w, u)v = 0
D(u, v)[x, y, z] = [D(u, v)x, y, z] + [x, D(u, v)y, z] + [x, y, D(u, v)z].
A derivation is a linear map D : W → W such that
D[x, y, z] = [Dx, y, z] + [x, Dy, z] + [x, y, Dz].
2
The vector space of all derivations form a Lie algebra Der (W ) under the
usual commutator map.
The vector space Der (W ) ⊕ W under the map
[D + u, E + v] = [D, E] + D(u, v) + Dv − Eu
form a Z /2-graded Lie algebra called the embedding Lie algebra of W . Con-
versely, degree 1 component of any Z /2-graded Lie algebra is a Lie triple
system under the triple commutator map.
Consider a Z-graded Lie algebra
L = L−2 ⊕ L−1 ⊕ L0 ⊕ L1 ⊕ L2
with one-dimensional components L−2 = F f , L2 = F e, such that each Li is
an eigenspace of the map [[e, f ], ·] with the eigenvalue i. Then e, f and [e, f ]
form an sl2-triple, and the maps [e, ·] and [f, ·] are mutualy inverse isompor-
phisms of L−1 and L1. Moreover, maps [x, ·] with x from he, f, [e, f ]i = sl2
defines a structure of left M2(F )-module on L1 ⊕ L−1, that by inspection
coincides with the usual structure on F 2 ⊗ L1 (after identification of L−1 and
L1 mentioned above).
Now L defines two kind of structures: one is a Lie triple structure on
L1 ⊕ L−1, and the other is a ternary system considered by Faulkner in [5]
on L1 (roughly speaking, it is an asymmetric version of a Freudenthal triple
system). Namely, define maps h·, ·i and h·, ·, ·i by formulas
[u, v] = hu, vie
hu, v, wi = [[[f, u], v], w].
Note that h·, ·i allows to identify the dual L∗
1 with L1, and so the map
L1 ⊗ L1 → End (L1)
corresponding to h·, ·, ·i produces a linear map
namely
π : End (L1) → End (L1),
π(h·, uiv) = hu, v, wi.
By the Morita equivalence, we can consider π as a map
End M2(F )(F 2 ⊗ L1) → End M2(F )(F 2 ⊗ L1).
3
Also, the same equivalence gives rise to a Hermitian (with respect to the
canonical symplectic involution on M2(F )) form
φ(cid:18)(cid:18)u1
u2(cid:19) ,(cid:18)v1
hu2, v2i −hu2, v1i(cid:19) .
v2(cid:19)(cid:19) = (cid:18)hu1, v2i −hu1, v1i
Now we want to relate the two structures on V1 ⊕ V−1 ≃ F 2 ⊗ V1. Direct
calculation shows that
D(u, v) =
1
2(cid:0)π(φ(·, u)v − φ(·, v)u) + φ(v, u) − φ(u, v)(cid:1).
(1)
This description admits a Galois descent. Namely, let Q be a quaternion
algebra over F , W be a left Q-module equipped with a Hermitian (with
respect to the canonical involution on Q) form φ and a linear map
π : End Q(W ) → End Q(W ).
Assume that φ and π become maps as above over a splitting field of Q.
In terms of [8] this means that End Q(W ) together with π and the sym-
plectic involution adjoint to φ form a gift (an abbreviation for a generalized
Freudenthal (or Faulkner) triple); one can state the conditions on π and φ as
a list of axioms not appealing to the descent (Garibaldi assumes that W is
of dimension 28 over Q, but this really doesn't matter, at least under some
additional restrictions on the characteristic of F ). Then equation (1) defines
on W a structure of a Lie triple system, hence the embedded Lie algebra
Der (W ) ⊕ W .
3 D6 + A1-construction
We say that a map of functors A → B from fields to sets is surjective at 2 if
for any field F and b ∈ B(F ) there exists an odd degree separable extension
E/F and a ∈ A(E) such that the images of a and b in B(E) coincide.
We enumerate simple roots as in [2]. Erasing vertex 1 from the extended
Dynkin diagram of E7 we see that the simply connected split group Esc
7
contains a subgroup of type D6 + A1, namely (Spin12 × SL2)/µ2. Its image
in the adjoint group Ead
is (HSpin12 × SL2)/µ2, which we denote by H for
7
brevity.
Theorem 1. The map H1(F, H) → H1(F, Ead
7 ) is surjective at 2.
4
Proof. Note that W (D6 + A1) and W (E7) has the same Sylow 2-subgroup.
Then the result follows by repeating the argument from the proof of Propo-
sition 14.7, Step 1 in [9] (this argument is a kind of folklore).
The long exact sequence
H1(F, µ2) → H1(F, H) → H1(PGO+
12 × PGL2) → H2(F, µ2)
shows that the orbits of H1(F, H) under the action of H1(F, µ2) are the isome-
try classes of central simple algebras of degree 12 with orthogonal involutions
(A, σ) and fixed isomorphism Cent(C0(A, σ)) ≃ F ×F , with [C+
0 (A, σ)] = [Q]
in Br(F ) for some quaternion algebra Q, where C0 stands for the Clifford al-
gebra and C+
0 for its first component (see [10, § 8] for definitions). Now
C+
0 (A, σ) ≃ End Q(W )
for a 16-dimensional space W over Q, and the canonical involution on C+
0 (A, σ)
induces a Hermitian form φ on W up to a scalar factor. It is not hard to see
that H1(F, H) parametrizes all the mentioned data together with φ (and not
only its similarity class), and H1(F, µ2) multiplies φ by the respective con-
stant. Over a splitting field of Q the 32-dimensional half-spin representation
carries a structure of Faulkner ternary system, so we are in the situation of
Section 2. The resulting embedding 66 + 3 + 64 = 133-dimensional Lie alge-
bra Der (W ) ⊕ W is the twist of the split Lie algebra of type E7 obtained by
a cocycle representing the image in H1(F, Ead
7 ). Theorem 1 shows that any
Lie algebra of type E7 over F arises this way up to an odd degree extension.
7 ) is
Recall that the class of Tits algebra of a cocycle class from H1(F, Ead
its image under the connecting map of the long exact sequence
H1(F, Esc
7 ) → H1(F, Ead
7 ) → H2(F, µ2).
The sequence fits in the following diagram:
H1(F, (Spin12 × SL2)/µ2)
H1(F, H)
H2(F, µ2)
H1(F, Esc
7 )
/ H1(F, Ead
7 )
/ H2(F, µ2).
The fundamental relation for groups of type D6 (see [10, 9.14]) shows that
7 ) corresponding
the class of the Tits algebra in Br(F ) of the class in H1(F, Ead
to Der (W ) ⊕ W is [A] + [Q].
5
/
/
/
/
/
/
4
7A1-construction
Consider a split simply connected group Esc
7 over a field F of characteris-
tic not 2 and the connected maximal rank subgroup H ′ of type 7A1 there
obtained by Borel -- de Siebenthal procedure.
Theorem 2. The map H1(F, H ′) → H1(F, Esc
7 ) is surjective at 2.
Proof. By Theorem 1 it suffices to show that the map
H1(F, H ′) → H1(F, H)
is surjective at 2.
Let ξ be an element in H1(F, H); denote by Q the quaternion algebra cor-
responding to the image of ξ in H1(F, PGL2). Twisting by Q, the cocycle class
corresponding to ξ comes from an element ξ′ ∈ H1(F, Spin(M6(Q), hyp)),
where hyp stands for the involution adjoint to the hyperbolic anti-Hermitian
form with respect to the canonical involution on Q (see [10] for references).
Denote by (A, σ) the algebra with involution (M6(Q), hyp) twisted by the
image of ξ′ in H1(F, PGO(M6(Q), hyp)). By [12, Cor. 3.3] we have
(A, σ) ∈ (Q1,¯) ⊗ (H1,¯) ⊞ (Q2,¯) ⊗ (H2,¯) ⊞ (Q3,¯) ⊗ (H3,¯)
(2)
for some quaternion algebras Qi, Hi over F such that [Q] = [Qi] + [Hi] in
Br(F ) and [H1] + [H2] + [H3] = 0 in Br(F ). Denote by H ′′ the maximal rank
subgroup in Spin(M6(Q), hyp) isogeneous to Spin(M2(Q), hyp)×3. Since H ′′
contains the center of Spin(M6(Q), hyp), it follows that ξ′ comes from ξ′′ ∈
H1(F, H ′′). Twisting back by Q we finally obtain an element ξ′′′ ∈ H1(F, H ′)
whose image is ξ.
We provide now a rational interpretation of the cohomological construc-
tion used in the proof. This is a twisted version of Manivel's construction
from [11, Theorem 4].
First note that the relations between classes [Q], [Qi] and [Hi] can be
summarized as follows. Put the quaternions at the point of the Fano plane as
the picture below shows; we will denote the quaternion algebra corresponding
to a vertex v by Qv. Now for any three collinear points u, v, w we have the
relation
[Qu] + [Qv] + [Qw] = 0 ∈ Br(F ).
6
Q1
✡✡✡✡✡✡✡✡✡✡
✹✹✹✹✹✹✹✹✹✹
H2
■■■■■
✍✍✍✍✍✍✍✍✍✍✍✍
qqqqqqqqqqqqqq
H3
✵✵✵✵✵✵✵✵✵✵✵✵
✉✉✉✉✉
▼▼▼▼▼▼▼▼▼▼▼▼▼▼
Q2
Q
H1
Q3
As in [11], we use a bijection between the lines and the quadruples of
points that are not incident to a line. If we have two lines α and β, then the
corresponding quadruples have two points in common, say, α = (xyuv) and
β = (xyzw), and α + β = (uvzw) is the third line such that α, β and α + β
are concurrent.
Further, for any line α = (xyuv) we have
[Qx] + [Qy] + [Qu] + [Qv] = 0 ∈ Br(F ),
so
Qx ⊗ Qy ⊗ Qu ⊗ Qv ≃ End (Vα)
for a 16-dimensional F -vector space Vα.
Note that the group H = Qv SL1(Qv) acts on the left-hand side by algebra
homomorphisms, so we obtain a projective representation H → PGL(Vα)
that lifts to a usual representation H on Vα.
Now in the Lie algebra ξe7 twisted by the image of ξ ∈ H1(F, Ead
7 ) there
is a Lie subalgebra h = "v sl1(Qv), and as h-module g decomposes as follows
(see [11, Theorem 4] in the split case, the general case readily follows by
descent):
ξe7 = h ⊕Mα
Vα.
(3)
We want to reconstruct the Lie bracket. From the O-grading we see
that for any α = (xyuv) the submodule h ⊕ Vα is actually a Lie subalgebra
isomorphic to
spin((Qx,¯) ⊗ (Qy,¯) ⊞ (Qu,¯) ⊗ (Qv,¯)) × "z6=x,y,u,v sl1(Qz)
7
(see [11, § 2] in the split case and use descent in the general case).
Further,
[Vα, Vβ] ≤ Vα+β for α 6= β.
Explicitly, if α = (xyuv) and β = (xyzw), we have a H-equivariant algebra
isomorphism
End (Vα) ⊗ End (Vβ) ≃ End (Qx) ⊗ End (Qy) ⊗ End (Vα+β).
This gives rise to a H-equivariant map
Vα ⊗ Vβ
/ Qx ⊗ Qy ⊗ Vα+β
TrdQx ⊗ TrdQy ⊗ id
/ Vα+β
defined up to a constant.
Now we compute the Rost invariant r(ξ) of a cocycle ξ from H1(F, Esc
7 )
that can be obtained this way (see [9] or [10, § 31] for definitions). As in the
proof of Theorem 2 we find Q, ξ′ ∈ H1(F, Spin(M6(Q), hyp)) and (A, σ). The
embedding D6 → E7 has the Rost multiplier 1, hence r(ξ) = r(ξ′). In terms
of [12, § 2.5] we have
r(ξ) = e3(A, σ) mod F × ∪ [Q].
(4)
5 Tits construction
Now we reproduce a construction from [16] in our terms. Let D be an algebra
of degree 4 and µ be a constant from F ×. By the exceptional isomorphism
A3 = D3 the group PGL1(D) defines a 3-dimensional anti-Hermitian form h
over Q up to a constant, where [Q] = 2[D] in Br(F ). Consider the algebra
M6(Q) with the orthogonal involution σ adjoint to the 6-dimensional form
h1, −µi ⊗ h. One of the component of C0(M6(Q), σ) is trivial in Br(F ) and
the other is Brauer-equivalent to Q. Choose some φ on W = Q16. The class
of the Tits algebra of the corresponding cocycle class in H1(F, Ead
7 ) is trivial,
so the cocycle class comes from some ξ ∈ H1(F, Esc
7 ).
Theorem 3. For D and µ as above, there is a cocycle from H1(F, Esc
7 ) whose
Rost invariant is (µ) ∪ [D]. In particular, if this element cannot be written
as a sum of two symbols from H3(F, Z /2) with a common slot, then there is
a strongly inner anisotropic group of type E7 over F .
8
/
/
Proof. It follows from [12, Corollary 2.17] that
r(ξ) = (µ) ∪ [D] + (λ) ∪ [Q]
[10, p. 441]), so there is
for some λ. Changing φ to λφ adds (λ) ∪ [Q] (cf.
a cocycle class from H1(F, Esc
7 ) whose Rost invariant is (µ) ∪ [D]. The last
claim follows from the easy computation of the Rost invariant of cocycles
corresponding to isotropic groups of type E7 (cf.
[9, Appendix A, Proposi-
tion]).
References
[1] B.N. Allison, J.R. Faulkner, A Cayley-Dickson Process for a Class of
Structurable Algebras, Trans. Amer. Math. Soc. 283 (1984), 185 -- 210.
[2] N. Bourbaki, Groupes et alg`ebres de Lie. Chapitres 4 -- 6, Hermann, Paris,
1968.
[3] I. Dejaiffe, Somme orthogonal d'algebres a involution et algebra de Clif-
ford, Comm. in Algebra 26 (1998), 1589 -- 1612.
[4] R.B. Brown, A minimal representation for the Lie algebra E7, Ill. J. Math.
12 (1968), 190 -- 200.
[5] J.R. Faulkner, A construction of Lie algebras from a class of ternary
algebras, Trans. Amer. Math. Soc. 155 (1971), 397 -- 408.
[6] H. Freudenthal, Sur le groupe exceptionnel E7, Proc. Nederl. Akad.
Wetensch. Ser. A 56 (1953), 81 -- 89.
[7] S. Garibaldi, Structurable algebras and groups of type E6 and E7, J. Al-
gebra 236 (2001), 651 -- 691.
[8] S. Garibaldi, Groups of type E7 over arbitrary fields, Comm. in Algebra,
29 (2001), 2689 -- 2710.
[9] S. Garibaldi, Cohomological invariants: Exceptional groups and Spin
groups, Memoirs of the AMS 937, 2009.
[10] M.-A. Knus, A. Merkurjev, M. Rost, J.-P. Tignol, The book of involu-
tions, AMS Colloquium Publ. 44, 1998.
9
[11] L. Manivel, Configuration of lines and models of Lie algebras, J. Algebra
304 (2006), 457 -- 486.
[12] A. Qu´eguiner -- Mathieu, J.-P. Tignol, The Arason invariant of orthogonal
involutions of degree 12 and 8, and quaternionic subgroups of the Brauer
group, preprint, available at http://arxiv.org/abs/1406.7705
[13] T. Springer, Linear algebraic groups, 2nd ed., Birkhauser, Boston, 1998.
[14] J. Tits, Alg`ebres alternatives, alg`ebres de Jordan et alg`ebres de Lie ex-
ceptionelles. I. Construction, Indag. Math. 28 (1966), 223 -- 237.
[15] J. Tits, Classification of algebraic semisimple groups, Algebraic groups
and discontinuous subgroups, Proc. Sympos. Pure Math. 9, Amer. Math.
Soc., Providence RI, 1966, 33 -- 62.
[16] J. Tits, Strongly inner anisotropic forms of simple algebraic groups, J.
of Algebra 131 (1990), 648 -- 677.
Victor Petrov
PDMI RAS, Nab. Fontanki, 27, Saint Peterburg, 191023 Russia
Chebyshev Laboratory, St. Petersburg State University, 14th
Line, 29b, Saint Petersburg, 199178 Russia
[email protected]
10
|
1709.08582 | 1 | 1709 | 2017-09-25T16:35:15 | The Second Cohomology Group of Elementary Quadratic Lie Superalgebras and Classifying a Subclass of 8-dimensional Solvable Quadratic Lie Superalgebras | [
"math.RA"
] | By definition, a quadratic Lie superalgebra is a Lie superalgebra endowed with a non-degenerate supersymmetric bilinear form which satisfies the even and invariant properties. In this paper we calculate all of the second cohomology group of elementary quadratic Lie superalgebras which have been classified in \cite{DU14} by applying the super-Poisson bracket on the super exterior algebra. Besides, we give the classification of 8-dimensional solvable quadratic Lie superalgebras having 6-dimensional indecomposable even part. The method is based on the double extension and classification results of adjoint orbits of the Lie algebra $\mathfrak{s}\mathfrak{p}(2)$. | math.RA | math |
The Second Cohomology Group of Elementary
Quadratic Lie Superalgebras and Classifying a
Subclass of 8-dimensional Solvable Quadratic
Lie Superalgebras
Cao Tran Tu Hai∗, Duong Minh Thanh† , Le Anh Vu∗∗
∗ Le Quy Don High School, Ninh Thuan, Vietnam
Email: [email protected]
†Ho Chi Minh city University of Education, Vietnam
Email: [email protected]
∗∗University of Economics and Law, VNU-HCMC, Vietnam
Email: [email protected]
Abstract
By definition, a quadratic Lie superalgebra is a Lie superalgebra en-
dowed with a non-degenerate supersymmetric bilinear form which satis-
fies the even and invariant properties. In this paper we calculate all of
the second cohomology group of elementary quadratic Lie superalgebras
which have been classified in [7] by applying the super-Poisson bracket
on the super exterior algebra. Besides, we give the classification of 8-
dimensional solvable quadratic Lie superalgebras having 6-dimensional
indecomposable even part. The method is based on the double extension
and classification results of adjoint orbits of the Lie algebra sp(2).
Keywords: cohomology, quadratic Lie superalgebras, super-exterior algebra, double
extension, adjoint orbits.
MSC (2010): Primary 17B, Secondary 17B56, 17B60.
Introduction
As far as we know, the Killing form of a Lie superalgebra is supersymmetric, invariant and
even. In some special cases, it also satisfies the non-degeneracy. Those lead to study of Lie
1
2
superalgebras endowed with a supersymmetric, invariant, even and non-degenerate bilinear
form. Such Lie superalgebras are called quadratic Lie superalgebras.
Consider the constructive aspect, a non-trivial quadratic Lie algebra could be considered
as a double extension (a combination of central extension and semi-direct product) of a
quadratic Lie algebra of smaller dimension (see [10]). Moreover, every solvable quadratic
Lie algebra is also isometrically isomorphic to either a T*-extension of a certain Lie algebra
with its dual space or an ideal of codimension one of a T*-extension (see [2]). Both of these
conceptions were generalized by S. Bajo, H. Benamor, S. Benayadi and M. Bordemann for
quadratic Lie superalgebras in [3] and [4]. In what follows, it is of interest to research algebras
endowed with an invariant and non-degenerate bilinear form as well as their applications.
Another concerned problem is to describe the cohomology of Lie superalgebras, which
is an important tool in mathematrics and theoretical physics. A classical example of a
constant such that D. B. Fuchs and D. A. Leites in [8] calculated the cohomology groups
of the classical Lie superalgebras with trivial coefficients, Y. C. Su and R. B. Zhang in [15]
computed explicitly the first and second cohomology groups of the classical Lie superalgebras
slmn and osp22n with coefficients in the finite-dimensional irreducible modules and the Kac
modules, W. Bai and W. Liu in [1] described the cohomology groups of Heisenberg Lie
superalgebras.
We shall now describe a situation in which g is a quadratic Lie superalgebra, then the
algorithm of describing the second cohomology groups with coefficients in C of g is relevant
to describing the super-Poisson bracket of the super-exterior algebra of g. In a consequence
of this result, we can list all of their one-dimensional double extensions. This provides much
information for the classification of quadratic Lie superalgebras. Our goal in this article
is to calculate the second cohomology group of all elementary quadratic Lie superalgebras
classified in [7] and give a classification of 8-dimensional solvable quadratic Lie superalgebras
having 6-dimensional indecomposable even part. The classification is based on the double
extension and the classification of adjoint orbits of the Lie algebra sp(2).
The paper will be organized as follows: The first section is devoted to recall some basic
concepts and results of Lie superalgebras, cohomology of Lie superalgebras and quadratic
Lie superalgebras. The second section gives the second cohomology group of all elementary
quadratic Lie superalgebras classified in [7] by using the super Z × Z2− Poisson bracket in
the super-exterior algebra. The last section gives the classification of 8-dimensional solvable
quadratic Lie superalgebras having 6-dimensional indecomposable even part by applying the
results of a double extension of quadratic Lie superalgebras in [3], [4] and the classification
of adjoint orbits of the Lie algebra sp(2).
All vector spaces considered in throughout the paper are finite-dimensional complex
vector spaces.
1 Cohomology of Lie Superalgebras and
Quadratic Lie Superalgebras
In this section, we recall some preliminary concepts and basic results which will be used later.
For details we refer the reader to the paper [8] of D.B.Fuchs, D.A.Leites and the paper [13]
of G. Pinczon, R. Ushirobira.
1.1 Lie Superalgebras and Cohomology
Definition 1.1.1. A Lie superalgebra g is a Z2−graded vector space g = g0 ⊕ g1 endowed
with a Lie super bracket [.,.] that satisfies the following conditions:
(i) The Lie super bracket [.,.] is bilinear and [gx, gy] ⊂ gx+y (grading);
(ii) [X, Y ] = −(−1)xy[Y, X] (skew-supersymmetry);
3
(iii) (−1)zx [[X, Y ], Z] + (−1)xy [[Y, Z], X] + (−1)yz [[Z, X], Y ] = 0 (super Jacobi identity)
for all x, y, z ∈ Z2, X ∈ gx, Y ∈ gy, Z ∈ gz .
Definition 1.1.2. Let g = g0 ⊕ g1 be a Lie superalgebra. Denote by Alt(g0, C) the algebra of
alternating multilinear forms on g0 and by Sym(g1, C) the algebra of symmetric multilinear
forms on g1. We define a Z × Z2-gradation on Alt(g0, C) and on Sym(g1, C) by
and
Alt(i,0)(g0, C) = Alti(g0, C), Alt(i,1)(g0, C) = {0}
Sym(i,i)(g1, C) = Symi(g1, C), Sym(i,j)(g1, C) = {0}
where i, j ∈ Z; i, j ∈ Z2 are respectively the residue classes modulo 2 of i, j and i 6= j. The
super-exterior algebra of g is C(g, C) = Alt(g0, C) ⊗ Sym(g1, C) endowed with the super-
exterior product on C(g, C) defined by
(Ω ⊗ F ) ∧(cid:0)Ω′ ⊗ F ′(cid:1) = (−1)f ω′ (cid:0)Ω ∧ Ω′(cid:1) ⊗ F F ′,
for all Ω ∈ Alt(g0, C), Ω′ ∈ Altω′
(g0, C), F ∈ Symf (g1, C), F ′ ∈ Sym(g1, C).
Remark that C(g, C) is a Z×Z2−graded algebra. More precisely, in terms of Z−gradation,
one has
C n(g, C) =
and in terms of Z2−gradation,
n
⊕
m=0 (cid:0)Altm(g0, C) ⊗ Symn−m(g1, C)(cid:1) , C0(g, C) = C,
C0(g, C) = Alt(g0, C) ⊗(cid:18) ⊕
and C1(g, C) = Alt(g0, C) ⊗(cid:18) ⊕
Sym2j (g1, C)(cid:19)
Symm2j+1(g1, C)(cid:19) .
j≥0
j≥0
Definition 1.1.3. Denote by End(C(g, C)) the space of endomorphisms on C(g, C). A
homogeneous endomorphism D ∈ End(C(g, C)) of degree (n, d) is called a superderivation of
C(g, C) if it satisfies the following condition:
D(A ∧ A′) = D(A) ∧ A′ + (−1)na+dbA ∧ D(A), ∀A ∈ C(a,b)(g, C), ∀A′ ∈ C(g, C).
Denote by Dern
d (C(g, C)) the space of superderivations of degree (n, d) of C(g, C). Then
we have a Z × Z2−gradation of the space of superderivations of C(g, C):
Der(C(g, C)) =
⊕
(n,d)∈Z×Z2
Dern
d (C(g, C)).
Example 1.1.4. Let X ∈ gx be a homogeneous element in g of degree x and define the
endomorphism iX of C(g, C) by
iX (A) (X1, . . . , Xa−1) = (−1)xbA (X, X1, . . . , Xa−1)
for all A ∈ C(a,b)(g, C); X1, . . . , Xa−1 ∈ g. Then
iX (A ∧ A′) = iX (A) ∧ A′ + (−1)−a+xbA ∧ iX (A)
for all A ∈ C(a,b)(g, C), A′ ∈ C(g, C). It means that iX is a superderivation of degree (−1, x).
Given k ≥ 0, the differential operator δk : C k(g, C) → C k+1(g, C) is a superderivation of
degree (cid:0)1, 0(cid:1) defined by
δkω (X0, . . . , Xk)
= Xr<s
(−1)s+xs(xr+1+...+xs−1)ω(cid:16)X0, . . . , Xr−1, [Xr, Xs] , Xr+1, . . . , cXs, . . . , Xk(cid:17) ,
4
for all ω ∈ C k(g, C), X0 ∈ gx0 , . . . , Xk ∈ gxk , where the sign cXs indicates that the element
Xs is omitted. It is easy to check that δ2 = δ ◦ δ = 0. By convention, δ0 = 0.
An element ω ∈ C k(g, C) is called a k-cocycle if δkω = 0 or a k-coboundary if there
exists ϕ ∈ C k−1(g, C) such that ω = δk−1ϕ.
We denote by Z k(g, C) the set of all k-cocycles and by Bk(g, C) the set of all k-coboundaries.
That is Z k(g, C) = Kerδk and Bk(g, C) = Imδk−1. Clearly, Bk(g, C) ⊂ Z k(g, C). The quo-
tient space Z k(g, C)/Bk(g, C) is denoted by H k(g, C) and called the k -cohomology groups of
g with trivial coefficients.
Definition 1.1.5. The dimension of the k-cohomology group H k(g, C) is called the k-th
Betti number of g and denoted by bk(g).
Example 1.1.6. (see [1]) Let the Heisenberg Lie superalgebra
h2n+1,m = g0 ⊕ g1 = C{Z, X1, . . . , Xn, Xn+1, . . . , X2n} ⊕ C{Y1, . . . , Ym}
with non-zero super brackets
[Xi, Xn+i] = Z, [Yj, Yj ] = Z, ∀i = 1, n, j = 1, m.
It is easy to compute that δX ∗
i = δY ∗
j = 0, for all i = 1, 2n, j = 1, m and
δZ ∗ =
nXi=1
X ∗
n+i ∧ X ∗
i −
1
2
mXj=1
j Y ∗
Y ∗
j .
For the second cohomology group, we have δ (Z ∗ ∧ ω) = δZ ∗ ∧ ω − Z ∗ ∧ δω = 0 if and only
if ω = 0. Then
Z 2(h2n+1,m, C) =(cid:8)X ∗
i ∧ X ∗
j , X ∗
i ⊗ Y ∗
k , Y ∗
k Y ∗
l
and dim Z 2(h2n+1,m, C) =(cid:18)2n
2 (cid:19) + 2n.m + m +(cid:18)m
: i, j = 1, 2n, i 6= j, k, l = 1, m(cid:9)
2(cid:19) = 2n2 − n + 2nm +
2
m2 + m
.
Accordingly, b2(h2n+1,m) = 2n2 − n + 2nm +
m2 + m
2
− 1.
1.2 Quadratic Lie Superalgebras
Definition 1.2.1. Let g = g0 ⊕ g1 be a Lie superalgebra. Assume that B is a bilinear form
defined on g such that it satisfies the following properties:
(i) B(X, Y ) = (−1)xyB(Y, X), ∀X ∈ gx, Y ∈ gy (supersymmetric);
(ii) B ([X, Y ] , Z) = B (X, [Y, Z]) for all X, Y, Z ∈ g (invariant);
(iii) B(X, Y ) = 0, ∀Y ∈ g implies X = 0 (non-degenerate).
The pair (g, B) is called a quadratic Lie superalgebra if B is even, that is
B(X, Y ) = 0; ∀X ∈ g0, Y ∈ g1.
In this case, it is easy to check that(cid:16)g0, B(cid:12)(cid:12)(cid:12)g0
is a g0−module endowed with a symplectic structure.
× g0(cid:17) is a quadratic Lie algebra and(cid:16)g1, B(cid:12)(cid:12)(cid:12)g1
× g1(cid:17)
Let (g, B) , (g′, B′) be two quadratic Lie superalgebras. We say (g, B) and (g′, B′) isomet-
rically isomorphic (or i-isomorphic, for short) if there exists a Lie superalgebra isomorphism
A from g onto g′ satisfying
B′ (A(X), A(Y )) = B (X, Y ) , ∀X, Y ∈ g.
Then A is called an i-isomorphism. We write (g, B)
i
∼= (g′, B′).
5
Definition 1.2.2. Let (g, B) be a quadratic Lie superalgebra and ℑ be a graded ideal of g.
(i) ℑ is called non-degenerate if the restriction of B to ℑ × ℑ is non-degenerate. Otherwise,
we say ℑ degenerate.
(ii) (g, B) is called irreducible if g does not have any non-degenerate graded ideal excepting
{0} and ℑ.
(iii) A non-degenerate ideal ℑ is called irreducible if ℑ does not have any non-degenerate
graded ideal excepting {0} and ℑ.
(iv) Ideal ℑ is called totally isotropic if B (ℑ, ℑ) = {0}.
The following proposition reduces the study of quadratic Lie superalgebras to non-
degenerate graded ideals.
Proposition 1.2.3 (see [4]). Let (g, B) be a quadratic Lie superalgebra and ℑ be a graded
ideal of g. Then ℑ⊥ is also a graded ideal of g. In addition, if ℑ is non-degenerate then so
is ℑ⊥, (cid:2)ℑ, ℑ⊥(cid:3) = {0} and ℑ ∩ ℑ⊥ = {0}. In this case, we denote g = ℑ
2 The Second Cohomology Group of
⊥
⊕ ℑ⊥.
(cid:3)
Elementary Quadratic Lie Superalgebras
In this section, we will compute the second cohomology group of all elementary quadratic
Lie superalgebras classified in [7]. Firstly, we recall the concept of the super Z × Z2− Poisson
bracket on the super-exterior algebra of a quadratic Lie superalgebra, which is used to give
a new way of description of cohomology.
2.1 The Super Z × Z2−Poisson Bracket on The Super-
exterior Algebra
Let g = g0 ⊕ g1 be a Z2−graded vector space equipped with a non-degenerate even supersym-
metric bilinear form B. In this case, g1 is a symplectic vector space. Hence, the dimension
of g1 must be even and g is aslo called a quadratic Z2−graded vector space. Now we recall
the definition of the Poisson bracket on Sym(g1) and the super-Poisson bracket on Alt(g0, C)
which are used later.
Let {X1, . . . , Xn, Y1, . . . , Yn} be a Darboux basis of g1, i.e. we have
B(Xi, Xj) = B(Yi, Yj ) = 0, B(Xi, Yj) = δij ,
for all i, j = 1, n. Let {p1, . . . , pn, q1, . . . , qn} be its dual basis. Then the algebra Sym(g1, C)
regarded as the polynomial algebra C[p1, . . . , pn, q1, . . . , qn] is equipped with the Poisson
bracket as follows:
{F, G} =
(cid:18) ∂F
∂pi
nXi=1
∂G
∂qi
−
∂G
∂pi(cid:19), ∀F, G ∈ Sym(g1, C).
∂F
∂qi
For any X ∈ g0, let iX be the derivation of Alt(g0, C) defined by
ιX (Ω)(Z1, . . . , Zk) = Ω(X, Z1, . . . , Zk), ∀Ω ∈ Altk+1(g0, C), X, Z1, . . . , Zk ∈ g0, k ≥ 0
and ιX (1) = 0. Let {Z1, . . . , Zm} be a fixed orthonormal basis of g0. The super-Poisson
bracket on Alt(g0, C) is defined by (see [13]):
(cid:8)Ω, Ω′(cid:9) = (−1)k+1
mXj=1
ιZj (Ω) ∧ ιZj (Ω′), ∀Ω ∈ Altk(g0, C), Ω′ ∈ Alt(g0, C).
6
Next, the super Z × Z2−Poisson bracket on C(g, C) is given by:
(cid:8)Ω ⊗ F, Ω′ ⊗ G(cid:9) = (−1)f ω′ (cid:0)(cid:8)Ω, Ω′(cid:9) ⊗ F G + Ω ∧ Ω′ ⊗ {F, G}(cid:1) ;
for any Ω ∈ Alt(g0, C), Ω′ ∈ Altω′
Proposition 2.1.1 (see [7], [13]). The algebra C(g, C) is a graded Lie algebra with the super
Z × Z2−Poisson bracket. More precisely, for all A ∈ C(a,b)(g, C), A′ ∈ C(a′,b′)(g, C) and
A′′ ∈ C(g, C), one has
(g0, C), F ∈ Symf (g1, C), G ∈ Sym(g1, C).
(i) {A′, A} = −(−1)aa′+bb′
(ii) {{A, A′}, A′′} = {A, {A′, A′′}} − (−1)aa′+bb′
Furthermore, {A, A′ ∧ A′′} = {A, A′} ∧ A′′ + (−1)aa′+bb′
{A, A′},
{A′, {A, A′′}}.
A′ ∧ {A, A′′}.
(cid:3)
Now, we choose an arbitrary basis nX 1
basis. Let nY 1
, . . . , Y m
0 o be the basis of g0 defined by αi = B(Y i
, . . . , Xm
0 o of g0. Denote by {α1, . . . , αm} its dual
, (cid:5)). That means
0
0
0
Set nX 1
1
, . . . , X n
1
, Y 1
1
, . . . , Y n
bracket on C(g, C) is also given by
(cid:8)A, A′(cid:9) = (−1)ω+f +1
B(Y i
0 , X j
0
) = δij ; ∀i, j = 1, . . . , n.
1 o be a Darboux basis of g1. Then the super Z × Z2−Poisson
mXi,j=1
0 , Y j
(A) ∧ ι
B(Y i
).ιX i
(A′)
X
0
0
j
0
+ (−1)ω
(cid:18)ιX k
1
nXk=1
(A) ∧ ιY k
1
(A′) − ιY k
1
(A) ∧ ιX k
1
(A′)(cid:19),
for all A ∈ Altω(g0, C) ⊗ Symf (g1, C), A′ ∈ C(g, C) (see [7]).
Remark 2.1.2. In [7], the authors introduced a useful tool. That was the 3-form I defined
on any quadratic Lie superalgebra (g, B) as follows
I (X, Y, Z) = B([X, Y ], Z), ∀X, Y, Z ∈ g.
This 3-form is called the 3-form associated to g. It is easy to prove that I is the homogeneous
element of degree (3, 0) in the Z × Z2−graded algebra C(g, C) = Alt(g0, C) ⊗ Sym(g1, C),
{I, I} = 0 and δ = − {I, .} (see [7], Proposition 1.11). Using this proposition, the cohomology
group H k(g, C) can be computed through the super Z × Z2−Poisson bracket.
2.2 Elementary Quadratic Lie Superalgebras
The main result of Section 2 is the description of the second cohomology group of elementary
quadratic Lie superalgebras which have classified in [7]. There are exactly three superalgebras
as follows.
2.2.1. gs
4,1 = (cid:0)CX0 ⊕ CY0(cid:1) ⊕(cid:0)CX1 ⊕ CY1(cid:1), where g0 = span{X0, Y0}, g1 = span{X1, Y1}.
The bilinear form B is defined by B(cid:0)X0, Y0(cid:1) = 1, B(cid:0)X1, Y1(cid:1) = 1, the others are zero
and the Lie super bracket is given by
(cid:2)Y1, Y1(cid:3) = −2X0,(cid:2)Y0, Y1(cid:3) = −2X1.
2.2.2. gs
4,2 = (cid:0)CX0 ⊕ CY0(cid:1) ⊕(cid:0)CX1 ⊕ CY1(cid:1), where g0 = span{X0, Y0}, g1 = span{X1, Y1}.
The bilinear form B is defined by B(cid:0)X0, Y0(cid:1) = 1, B(cid:0)X1, Y1(cid:1) = 1, the others are zero
and the Lie super bracket is given by
(cid:2)X1, Y1(cid:3) = X0,(cid:2)Y0, X1(cid:3) = X1,(cid:2)Y0, Y1(cid:3) = −Y1.
7
2.2.3. gs
6 = (cid:0)CX0 ⊕ CY0(cid:1) ⊕ (cid:0)CX1 ⊕ CY1 ⊕ CZ1 ⊕ CT1(cid:1), where g0 = span{X0, Y0}, g1 =
span{X1, Y1, Z1, T1}. The bilinear form B is defined by B(cid:0)X0, Y0(cid:1) = 1, B(cid:0)X1, Z1(cid:1) =
1, B(cid:0)Y1, T1(cid:1) = 1, the others are zero and the Lie super bracket is given by
(cid:2)Z1, T1(cid:3) = −X0,(cid:2)Y0, Z1(cid:3) = −Y1,(cid:2)Y0, T1(cid:3) = −X1.
2.3 The Second Cohomology Group of Elementary
Quadratic Lie Superalgebras
Now we will introduce the first result of the paper. Namely, we will describe the second coho-
mology group of the elementary quadratic Lie superalgebras which have listed in Subsection
2.2.
Theorem 2.3.1. With notations being as above in Subsection 2.2, the second cohomology
group of the elementary quadratic Lie superalgebras are described as follows
(i) H 2(gs
4,1, C) = spannhY ∗
0
⊗ X ∗
1i ,hX ∗
1
the dual basic of {X0, Y0, X1, Y1}.
Y ∗
1
− 2X ∗
0
∧ Y ∗
0 io where {X ∗
0
, Y ∗
0
, X ∗
1
, Y ∗
1
} is
(ii) H 2(gs
4,2, C) = {0} .
(iii) H 2(gs
6, C) = span(cid:26)hY ∗
0
⊗ X ∗
1i ,hY ∗
0
⊗ Y ∗
1 i ,(cid:20)(cid:16)Z ∗
hX ∗
1
1(cid:17)2(cid:21) ,(cid:20)(cid:16)T ∗
Z ∗
1
− X ∗
0
∧ Y ∗
1(cid:17)2(cid:21) ,
0 i ,hY ∗
1
T ∗
1
− X ∗
0
∧ Y ∗
0 io
where {X ∗
0
, Y ∗
0
, X ∗
1
, Y ∗
1
, Z ∗
1
, T ∗
1
} is the dual basic of {X0, Y0, X1, Y1, Z1, T1}.
The Proof of Theorem 2.3.1
(i) Firstly, we consider gs
g1 = span{X1, Y1}.
4,1 = (cid:0)CX0 ⊕ CY0(cid:1) ⊕(cid:0)CX1 ⊕ CY1(cid:1), where g0 = span{X0, Y0},
⊗(cid:16)Y ∗
1 (cid:17)2
. By a
According to the paper [7], the associated 3-form of gs
straightforward computation, we obtain
4,1 is I = Y ∗
0
0
nI, X ∗
nI, X ∗
nI, X ∗
(cid:26)I,(cid:16)X ∗
0
1 o = 0,
0
0
0
0
0
0
Y ∗
1
∧ Y ∗
⊗ Y ∗
⊗ Y ∗
⊗ X ∗
⊗ X ∗
⊗ Y ∗
1
⊗ X ∗
1
+ 2X ∗
0
, nI, Y ∗
1o = 2Y ∗
1(cid:16)Y ∗
1 (cid:17)2
1o = X ∗
1o = 0, nI, Y ∗
1 (cid:17)2(cid:27) = 0, nI, X ∗
1 (cid:17)2
0o =(cid:16)Y ∗
0 o = 0, nI, X ∗
, nI, Y ∗
⊗(cid:16)Y ∗
, nI, X ∗
0 o = Y ∗
1 (cid:17)2
1 o = 0,
1 o = −(cid:16)Y ∗
1 (cid:17)3
, nI, Y ∗
, (cid:26)I,(cid:16)Y ∗
1(cid:17)2(cid:27) = 4Y ∗
1 o = 2Y ∗
1 (cid:27)
Imδ1 = span(cid:26)(cid:16)Y ∗
1 (cid:17)2
0 ⊗ Y ∗
1 ,(cid:16)Y ∗
1 (cid:17)2
1i ,hX ∗
4,1, C) = Kerδ2/Imδ1 = spannhY ∗
1 Y ∗
Kerδ2 = span(cid:26)Y ∗
0 ⊗ X ∗
1 , Y ∗
0 ⊗ Y ∗
0 ⊗ X ∗
H 2(gs
, X ∗
1 Y ∗
1 − 2X ∗
, Y ∗
Y ∗
1
Then we get
and
Therefore
∧ Y ∗
0
⊗ Y ∗
1
,
⊗(cid:16)Y ∗
1 (cid:17)2
.
0
0 (cid:27) .
0 ∧ Y ∗
1 − 2X ∗
0 ∧ Y ∗
0 io .
8
(ii) Next, we consider gs
g1 = span{X1, Y1}.
From [7], we obtain the associated 3-form I = Y ∗
0
as above, we have
4,2 =(cid:0)CX0 ⊕ CY0(cid:1) ⊕(cid:0)CX1 ⊕ CY1(cid:1), where g0 = span{X0, Y0} and
⊗ X ∗
1
Y ∗
1
. By a similar computation
Kerδ2 = Imδ1 = spannX ∗
1 Y ∗
1 , Y ∗
0 ⊗ X ∗
1 , Y ∗
1 o .
0 ⊗ Y ∗
Therefore we get H 2(gs
4,2, C) = {0}.
(iii) Finally, we consider gs
span{X0, Y0}, g1 = span{X1, Y1, Z1, T1}.
By a similar computation, we have
6 = (cid:0)CX0 ⊕ CY0(cid:1) ⊕ (cid:0)CX1 ⊕ CY1 ⊕ CZ1 ⊕ CT1(cid:1), where g0 =
I = Y ∗
0 ⊗ Z ∗
1 ; Imδ1 = spannZ ∗
1 T ∗
1 T ∗
1 , Y ∗
0 ⊗ T ∗
1 , Y ∗
Kerδ2 = span(cid:26)Y ∗
0 ⊗ X ∗
1 , Y ∗
0 ⊗ Y ∗
1 , Y ∗
0 ⊗ Z ∗
1 , Y ∗
0 ⊗ T ∗
1 T ∗
Z ∗
1 , X ∗
1 Z ∗
1 − X ∗
1o ;
0 ⊗ Z ∗
,(cid:16)T ∗
1(cid:17)2
1 − X ∗
1 T ∗
,
1 ,(cid:16)Z ∗
1(cid:17)2
0 , Y ∗
0 ∧ Y ∗
0 o .
0 ∧ Y ∗
Therefore we get
H 2(gs
6, C) = Kerδ2/Imδ1 = span(cid:26)hY ∗
1(cid:17)2(cid:21) ,hX ∗
(cid:20)(cid:16)T ∗
0 ⊗ X ∗
0 ⊗ Y ∗
1i ,hY ∗
1 − X ∗
1 Z ∗
0 ∧ Y ∗
1(cid:17)2(cid:21) ,
1 i ,(cid:20)(cid:16)Z ∗
0 i ,hY ∗
1 T ∗
1 − X ∗
0 ∧ Y ∗
0 i(cid:27) .
The proof is complete.
(cid:3)
3 Classification of 8-dimensional Solvable
Quadratic Lie Superalgebras Having
6-dimensional Indecomposable Even Part
The main result of the paper is a classification of all solvable quadratic Lie superalgebras of
dimension 8 having indecomposable even part of dimension 6. In order to get this classifica-
tion, it is necessary to observe consequence of adjoint orbits of the Lie algebra sp(2) and the
double extension.
3.1 Adjoint Orbits of Symplectic Lie Algebra sp(2)
Every non-zero element (cid:18)a
c −a(cid:19) in the Lie algebra sp(2) is either nilpotent or semisimple.
b
It means that we can choose a basis of C2 such that the matrix representation of an element
in sp(2) with respect to this basis is diagonal or strictly upper triangliar.
Lemma 3.1.1. Let A, B ∈ sp(2) such that [A, B] = 0. Then A and B are linearly dependent.
Proof. The Lie algebra sp(2) has a basis {H, X, Y } with
[H, X] = 2X, [H, Y ] = −2Y, [X, Y ] = H.
Set
A = aH + bX + cY, B = a′H + b′X + c′Y.
From [A, B] = 0, it implies
(bc′ − b′c)H + 2(ab′ − a′b)X − 2(ac′ − a′c)Y = 0.
That means A and B are linearly dependent.
As an immediate consequence of Lemma 3.1.1, we have the following lemma.
9
(cid:3)
Lemma 3.1.2. Let A, B, C ∈ sp(2). If [A, B] = C and [A, C] = [B, C] = 0 then C = 0. (cid:3)
Lemma 3.1.3. Let A, B ∈ sp(2) such that B 6= 0 and [A, B] = B. Then A is the semisimple
element with eigenvalues
1
2
, −
1
2
and B is nilpotent.
Proof. We can choose a basis such that A is
(cid:18)0
0
1
0(cid:19) or (cid:18)λ
0 −λ(cid:19) .
0
If A =(cid:18)0
0
1
0(cid:19) = X, let B = aH + bX + cY . Since [A, B] = B one has
−2aX + cH = aH + bX + cY.
So B = 0 and this is a contradiction. Thus, A =(cid:18)λ
0 −λ(cid:19) = λH . Set B = aH + bX + cY .
0
By [A, B] = B, we have
This implies
Then
2λbX − 2cλY = aH + bX + cY.
a = 0, 2λb = b, −2cλ = c.
a = 0, λ =
1
2
, c = 0 or a = 0, λ = −
1
2
, b = 0.
The proof is complete.
(cid:3)
3.2 Double extension of quadratic Lie superalgebras
In this subsection, we recall the notion of a double extension of quadratic Lie superalgebras
which is introduced in [4] and some its properties because it will be used later. For details
we refer the reader to the paper [9] of V. G. Kac, the paper [4] of I. Bajo, S. Benayadi, M.
Bordemann and the paper [3] of I. Bajo, S. Benayadi. Firstly, we recall the definition of
homogeneous superderivations.
Definition 3.2.1 (see [9]). Let g be a Lie superalgebra. An endomorphism D ∈ Homα(g, g)
(where α ∈ Z2) is called a homogeneous superderivation of degree α of g if
D[X, Y ] = [DX, Y ] + (−1)αx[X, DY ]; ∀x ∈ Z2, ∀X ∈ gx, ∀Y ∈ g.
Denote by (Der(g))α ⊂ Homα(g, g) the space of all homogeneous superderivations of
degree α. Assuming that Der(g) = (Der(g))0 ⊕ (Der(g))1, it is easily seen that Der(g) is
Lie subsuperalgebra of Lie superalgebra Hom(g, g) and we call it the Lie superalgebra of
superderivations of g.
Definition 3.2.2 (see [4]). Let (g, B) be a quadratic Lie superalgebra. A homogeneous
superderivation D of degree α of g is called skew-supersymmetric if
B(D(X), Y ) = −(−1)αxB(X, D(Y )); ∀x ∈ Z2, ∀X ∈ gx, ∀Y ∈ g.
10
It is proved that the vector subspace of Der(g) generated by the set of all homogeneous
skew-supersymmetric superderivations of (g, B) is a Lie subsuperalgebra of Der(g) and it is
denoted by Dera(g, B). In the remainder of this subsection we give the notion of a double
extension of quadratic Lie superalgebras.
Proposition 3.2.3 (see [4], Theorem 2.4). Let (g, B) be a quadratic Lie superalgebra and h
be a Lie superalgebra. Suppose that ψ : h → Dera (g, B) is a morphism of Lie superalgebras.
Define a bilinear map ϕ from g × g to h∗ by
ϕ(X, Y )(Z) = (−1)(x+y)zB(ψ(Z)(X), Y ); ∀x, y, z ∈ Z2, ∀X ∈ gx, Y ∈ gy, Z ∈ hz.
Let πh by the coadjoint representation of h. Then the Z2−graded vector space g = h ⊕ g ⊕ h∗
becomes a Lie superalgebra with the bracket defined by
[Z + X + f, W + Y + g] = [Z, W ]h + [X, Y ]g + ψ(Z)(Y ) − (−1)xyψ(W )(X)
+ π(Z)(g) − (−1)xy π(W )(f ) + ϕ(X, Y )
for all Z + X + f ∈ gx, W + Y + g ∈ gy; x, y, z ∈ Z2. Furthermore, let γ be an even
supersymmetric invariant bilinear form on h. Then g becomes a quadratic Lie superalgebra
with the bilinear form defined by
B(Z + X + f, W + Y + g) = B(X, Y ) + γ(Z, W ) + f (W ) + (−1)xyg(Z).
(cid:3)
The quadratic Lie superalgebra g in Proposition 3.2.3 is called a double extension of
(g, B) by h (by means of ψ).
In particular, if h = Ce, h∗ = Cf then D = ψ(e) is an even skew-symmetric superderiva-
tion on g and g = Ce ⊕ g ⊕ Cf . In this case, g is called a 1-dimensional double extension of
g by means of D. The Lie super bracket on g is defined as follows
[X, Y ]g = [X, Y ]g + (D(X), Y )f, [e, X] = D(X), [f, g] = 0; ∀X, Y ∈ g.
The invariant bilinear form B on g is defined by
B(e, f ) = 1, B(e, g) = B(f, g) = 0, B(X, Y ) = B(X, Y ), ∀X, Y ∈ g.
The following proposition presents the crucial role of one-dimensional double extensions
for the construction of quadratic Lie superalgebras.
Proposition 3.2.4 (see [3]). Let (g, B) be an irreducible quadratic Lie superalgebra of di-
mension n such that n > 1. If z(g) ∩ g0 6= {0} then g is a double extension of a quadratic
Lie superalgebra of dimension n − 2 by the one-dimensional Lie algebra.
(cid:3)
3.3 Classification of 8-dimensional Solvable Quadratic Lie
Superalgebras having 6-dimensional Indecomposable
Even Part
The main result of this section is the classification of 8-dimensional solvable quadratic Lie
algebras having 6-dimensional indecomposable even part. Before starting the main theorem,
we first recall the classification of the 6-dimensional quadratic Lie algebras in [12].
Proposition 3.3.1 (see [12]). Let (g, B) be a solvable quadratic Lie algebra of dimension
6. Assume g indecomposable . Then there exists a basis {Z1, Z2, Z3, X1, X2, X3} of g such
that the bilinear form B is defined by B(Xi, Zj) = δi,j; 1 ≤ i, j ≤ 3, the other are zero and
g is i-isomorphic to each of Lie algebras as follows:
(1) g6,1 : [X1, X2] = Z3, [X2, X3] = Z1, [X3, X1] = Z2.
11
(2) g6,2(λ) : [X3, Z1] = Z1, [X3, Z2] = λZ2, [X3, X1] = −X1, [X3, X2] = −λX2, [Z1, X1] =
Z3, [Z2, X2] = λZ3 where λ ∈ C and λ 6= 0. In this case g6,2(λ1) and g6,2(λ2) is i-
isomorphic if and only if λ2 = ±λ1 or λ2 = λ−1
1 .
(3) g6,3 : [X3, Z1] = Z1, [X3, Z2] = Z1 + Z2, [X3, X1] = −X1 − X2, [X3, X2] = −X2,
[Z1, X1] = [Z2, X1] = [Z2, X2] = Z3.
Now, we consider an 8-dimensional solvable quadratic Lie superalgebras(cid:0)g = g0 ⊕ g1, B(cid:1)
having 6-dimensional indecomposable even part. Because g is solvable, so is g0. Since
the even part g0 is indecomposable, g0 is isometrically isomorphic to each of quadratic Lie
superalgebras g6,1, g6,2(λ), g6,3 which have been listed in Proposition 3.3.1.
(cid:3)
Assume that g is decomposable. Consider an arbitrary proper non-degenerate ideal
a non-zero and non-degenerate ideal of g0. Because g0 is indecomposable, then j0 = g0 or
j = j0 ⊕ j1 of g. Then j⊥ is also a proper non-degenerate ideal of g. Hence, j0 or (cid:0)j⊥(cid:1)0 is
(cid:0)j⊥(cid:1)0 = g0. Therefore, without loss of generality, we may assume that j = g0 ⊕ j1. Then we
get
j1 = {0}, (cid:0)j⊥(cid:1)0 = {0} and (cid:0)j⊥(cid:1)1 = g1.
That means that g = g0 ⊕ g1 where g0 is an 6-dimensional indecomposable quadratic Lie
algebra, g1 is a symplectic vector space of dimension 2 and [g0, g1] = {0}.
Next, we consider g as an indecomposable quadratic Lie superalgebra with the even and
odd parts are given as follows
g0 = span{Xi, Zi}, B (Xi, Xj) = B (Zi, Zj ) = 0, B (Xi, Zj ) = δij ; 1 ≤ i, j ≤ 3;
g1 = span(cid:8)Y1, T1(cid:9) , B(cid:0)Y1, T1(cid:1) = 1.
By Proposition 3.3.1, there are three cases for g0. Now, we will consider these cases one
by one.
(1) g0 = g6,1
In this case, the Lie brackets are defined by
[X1, X2] = Z3, [X2, X3] = Z1, [X3, X1] = Z2.
Remark that
ad(X)g1
∈ sp(cid:16)g1, Bg1×g1(cid:17) ∼= sp (2) , ∀X ∈ g0.
By Lemma 3.1.2, we get
ad(Z1)g1
= ad(Z2)g1
= ad(Z3)g1
= 0.
By Lemma 3.1.1, any pair of elements in the set {ad(X1)g1
is linearly dependent. Because g is indecomposable, one of ad(X1)g1
ad(X3)g1
is non-zero. On the other hand, by the classification of Sp(cid:0)g1(cid:1)-orbits
of the Lie algebra sp(cid:0)g1(cid:1), we can choose a Darboux basic (cid:8)Y1, T1(cid:9) such that the
either are concurrently
representation matrices of ad(X1)g1
diagonal or strictly upper triangliar. Therefore, we need consider two different cases.
}
, ad(X3)g1
and
, ad(X2)g1
, ad(X2)g1
, ad(X3)g1
, ad(X2)g1
=(cid:18)µ
0 −µ(cid:19) , ad(X3)g1
0
=(cid:18)ν
0 −ν(cid:19),
0
=(cid:18)λ
0 −λ(cid:19) , ad(X2)g1
0
1a) ad(X1)g1
where λ, µ, ν ∈ C such that they are not simultaneously zero. In this case, we have
(cid:2)X1, Y1(cid:3) = λY1, (cid:2)X1, T1(cid:3) = −λT1, (cid:2)X2, Y1(cid:3) = µY1,
12
(cid:2)X2, T1(cid:3) = −µT1, (cid:2)X3, Y1(cid:3) = νY1, (cid:2)X3, T1(cid:3) = −νT1.
Because B is invariant, non-degenerate and even, we obtain
[Y1, Y1] = 0, [T1, T1] = 0, [Y1, T1] = λZ1 + µZ2 + νZ3.
We denote this Lie superalgebra by gs
not simultaneously zero.
1
1b) ad(X3)g1
where λ, µ, ∈ C. In this case, we have
0(cid:19) , ad(X1)g1
=(cid:18)0
0
8,2,1(λ, µ, ν) where λ, µ, ν ∈ C such that they are
=(cid:18)0
0
λ
0(cid:19) , ad(X2)g1
=(cid:18)0 µ
0(cid:19),
0
(cid:2)X3, T1(cid:3) = Y1, (cid:2)X1, T1(cid:3) = λY1, (cid:2)X2, T1(cid:3) = µY1.
Because B is invariant, non-degenerate and even, we get [Y1, g1] = 0. In a similar way
of argument to the one used in the case (1a), we have
and the others on g1 are zero. We denote this Lie superalgebra by gs
λ, µ, ∈ C.
8,2,2(λ, µ) where
[T1, T1] = λZ1 + µZ2 + Z3,
(2) g0 = g6,2(λ)
The Lie bracket is given by
[X3, Z1] = Z1, [X3, Z2] = λZ2, [X3, X1] = −X1,
[X3, X2] = −λX2, [Z1, X1] = Z3, [Z2, X2] = λZ3
for λ 6= 0. By Lemma 3.1.2, ad (Z3)(cid:12)(cid:12)(cid:12)g1
V = spannad(Xi)(cid:12)(cid:12)(cid:12)g1
Then ad(X3)(cid:12)(cid:12)(cid:12)g1
2a) Assume that dim V = 1
We get
= 0. Let
, ad(Zj )(cid:12)(cid:12)(cid:12)g1
: i = 1, 2, 3; j = 1, 2o .
6= 0 since g is indecomposable. By Lemma 3.1.1, we get dim V = 1, 2.
(ii) ad(X3) =
, we denote this Lie superalgebra by
and
ad(X1)(cid:12)(cid:12)(cid:12)g1
= 0, ad(X2)(cid:12)(cid:12)(cid:12)g1
= 0, ad(Z1)(cid:12)(cid:12)(cid:12)g1
= 0 , ad(Z2)(cid:12)(cid:12)(cid:12)g1
= 0
[g1, g1] ⊂ [g0, g0]⊥ = Z(g0) = CZ3.
According to Proposition 3.2.3, g is a one-dimensional double extension of q0 ⊕ g1
(where q0 = span {Z1, Z2, X1, X2}) by means of ad(X3) ∈ o(q0)⊕ sp(g1). By applying
the classification of Sp(cid:0)g1(cid:1) −orbits of sp(cid:0)g1(cid:1), we have the following cases of ad(X3).
(i) ad(X3) =
, we denote this Lie superalgebra by
1
0
0
0
0
0
0
0
0
0
1
0
0
0
0
−λ
0
0
0
0
0
λ
0 −1
0
0
0
0
0
0
0
0
0
0
0
0
gs
8,2,3(λ) where λ ∈ C and λ 6= 0 .
0
0
0
0
0
0
0
0
0
µ
0 −µ
0
0
λ
0
0 −1
0
0
0
0
0
0
0
0
0
−λ
0
0
1
0
0
0
0
0
gs
8,2,4(λ, µ) where λ, µ, ∈ C and λ 6= 0, µ 6= 0.
13
2b) Assume that dim V = 2
,
Moreover,
6= 0. We have
ad(X1)(cid:12)(cid:12)(cid:12)g1
, ad(Z1)(cid:12)(cid:12)(cid:12)g1
=had(X3)(cid:12)(cid:12)(cid:12)g1
Without loss of generality, we assume that ad(Z1)(cid:12)(cid:12)(cid:12)g1
= xad(Z1)(cid:12)(cid:12)(cid:12)g1
for some x ∈ C since ad(X1)(cid:12)(cid:12)(cid:12)g1
−ad(X1)(cid:12)(cid:12)(cid:12)g1
, ad(X1)(cid:12)(cid:12)(cid:12)g1i = xad(Z1)(cid:12)(cid:12)(cid:12)g1
clearly force ad(X1)(cid:12)(cid:12)(cid:12)g1
infers that ad(X3)(cid:12)(cid:12)(cid:12)g1
, ad(Z1)(cid:12)(cid:12)(cid:12)g1i = ad(Z1)(cid:12)(cid:12)(cid:12)g1
had(X3)(cid:12)(cid:12)(cid:12)g1
are linearly dependent.
= 0.
1
2
is nilpotent
(see Lemma 3.1.3). Consequently, we can choose a Darboux basis {Y1, T1} of g1 such
that
is semisimple with eigenvalues
, −
Assume that
ad(X3)(cid:12)(cid:12)(cid:12)g1
ad(X2)(cid:12)(cid:12)(cid:12)g1
0
2
0 − 1
=(cid:18) 1
= αad(Z1)(cid:12)(cid:12)(cid:12)g1
2(cid:19) , ad(Z1)(cid:12)(cid:12)(cid:12)g1
, ad(Z2)(cid:12)(cid:12)(cid:12)g1
1
1
2
and ad(Z1)(cid:12)(cid:12)(cid:12)g1
=(cid:18)0
0(cid:19) .
= βad(Z1)(cid:12)(cid:12)(cid:12)g1
0
for some α, β ∈ C. Then
had(X3)(cid:12)(cid:12)(cid:12)g1
implies that λ = 1 or ad(Z2)(cid:12)(cid:12)(cid:12)g1
had(X3)(cid:12)(cid:12)(cid:12)g1
yields that λ = −1 or ad(X2)(cid:12)(cid:12)(cid:12)g1
= 0.
, ad(Z2)(cid:12)(cid:12)(cid:12)g1i = λad(Z2)(cid:12)(cid:12)(cid:12)g1
, ad(X2)(cid:12)(cid:12)(cid:12)g1i = −λad(X2)(cid:12)(cid:12)(cid:12)g1
= 0.
Note that when λ = −1, it can be returned to the case λ = 1 by replacing X2 by Z2
and Z2 by X2.
Therefore, we have two following cases.
(i) If ad(Z2)(cid:12)(cid:12)(cid:12)g1
= ad(X2)(cid:12)(cid:12)(cid:12)g1
= 0, then
[Y1, T1] =
1
2
Z3, [T1, T1] = X1.
Denote this Lie superalgebra by gs
8,2,5(λ) where λ ∈ C and λ 6= 0.
(ii) If λ = 1, then
ad(X2)(cid:12)(cid:12)(cid:12)g1
= 0, ad(Z2)(cid:12)(cid:12)(cid:12)g1
=(cid:18)0 µ
0(cid:19) , µ 6= 0.
0
By the invariance of the bilinear form B, we obtain
[Y1, T1] =
1
2
Z3, [T1, T1] = X1 + µX2.
Denote this Lie superalgebra by gs
8,2,6(λ, µ) where λ, µ, ∈ C and λ 6= 0, µ 6= 0. .
14
(3) g0 = g6,3
The non-zero Lie brackets:
[X3, Z1] = Z1, [X3, Z2] = Z1 + Z2, [X3, X1] = −X1 − X2,
[X3, X2] = −X2, [Z1, X1] = [Z2, X1] = [Z2, X2] = Z3.
Denote by
, ad(X3)(cid:12)(cid:12)(cid:12)g1
V = spannad(X1)(cid:12)(cid:12)(cid:12)g1
, ad(X2)(cid:12)(cid:12)(cid:12)g1
Similarly to Case (2), we get ad(Z3)(cid:12)(cid:12)(cid:12)g1
sp(g1). Applying the classification of Sp(cid:0)g1(cid:1) −orbits of sp(cid:0)g1(cid:1), we have the following
3a) Assume that dim V = 1.
By the argument analogous to that used in Case (2), g is a one-dimensional double
extension of q0 ⊕ g1 (where q0 = span {Z1, Z2, X1, X2}) by means of ad(X3) ∈ o(q0) ⊕
, ad(Z2)(cid:12)(cid:12)(cid:12)g1o .
, ad(Z1)(cid:12)(cid:12)(cid:12)g1
= 0 and dim V = 1, 2.
cases of ad(X3).
(i) ad(X3) =
, denote this Lie superalgebra by gs
8,2,7.
1
0
0
0
0
0
0
0
1
0
1
0
0 −1
0
0 −1 −1
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
1
0
(ii) ad(X3) =
0
0
1
0
1
0
0 −1
0
0 −1 −1
0
0
0
0
where 0 6= λ ∈ C.
1
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
λ
0
0 −λ
3b) Assume that dim V = 2.
or ad(X1)(cid:12)(cid:12)(cid:12)g1
Firstly, we see that ad(Z2)(cid:12)(cid:12)(cid:12)g1
in assuming ad(Z2)(cid:12)(cid:12)(cid:12)g1
=(cid:18) 1
ad(X3)(cid:12)(cid:12)(cid:12)g1
ad(Z1)(cid:12)(cid:12)(cid:12)g1
= ad(X1)(cid:12)(cid:12)(cid:12)g1
By the invariance of B, we have
0
2
0 − 1
[X3, Y1] =
1
Y1, [X3, T1] = −
2
Denote this Lie superalgera by gs
8,2,9.
1
2
, denote this Lie superalgebra by gs
8,2,8(λ)
is non-zero. There is no loss of generality
6= 0 . By Lemma 3.1.3, we obtain
1
0(cid:19) ,
=(cid:18)0
2(cid:19) , ad(Z2)(cid:12)(cid:12)(cid:12)g1
= ad(X2)(cid:12)(cid:12)(cid:12)g1
0
= 0.
T1, [Z2, T1] = Y1, [Y1, T1] =
1
2
Z3, [T1, T1] = X2.
Now, combining the above calculations, we obtain the main theorem of the paper which gives
the classification of 8-dimensional solvable quadratic Lie superalgebras having 6-dimensional
indecomposable even part.
Theorem 3.3.2. Let g be a solvable quadratic Lie superalgebra of dimension 8 such that
its even part is indecomposable and 6-dimensional.
(i) If g is decomposable then g is isomorphic to g = g0
⊥
⊕g1 where g0 is isometrically
isomorphic to g6,1, g6,2(λ), g6,3 which are given in Proposition 3.3.1; g1 is a symplectic
vector space of dimension 2 and [g0, g1] = {0}.
(ii) If g is indecomposable then g is i-isomorphic to each of the following quadratic Lie
8,2,6(λ, µ),
(cid:3)
8,2,1(λ, µ, ν), gs
8,2,2(λ, µ), gs
8,2,4(λ, µ), gs
8,2,5(λ), gs
8,2,3(λ), gs
superalgebras gs
gs
8,2,7 , gs
8,2,8(λ) and gs
8,2,9.
15
References
[1] W. Bai, W. Liu, Cohomology of Heisenberg Lie Superalgebras, J. of Math. Physics 58
(2017) 021701.
[2] M. Bordemann, Nondegenerate invariant bilinear forms on nonassociative algebras,
Acta. Math. Uni. Comenianac LXVI (2) (1997) 151-201.
[3] I. Bajo, S. Benayadi, Lie algebras admitting a unique quadratic structure, Commun. in
Algebra 25 (9) (1997) 27952805.
[4] I. Bajo, S. Benayadi, M. Bordemann, Generalized double extension and descriptions of
quadratic Lie superalgebras, arXiv:0712.0228v1 (2007).
[5] T. T. H. Cao, M. T. Duong, The Betti numbers and the vector space of skew-symmetric
derivations of solvable quadratic Lie algebras with dimension ≤ 7, J. of Science, Ho Chi
Minh city University of Education, No 5 (70) (2015) 86-96 (in Vietnamese).
[6] M. T. Duong, The cohomology group H 2(g, C) of the elementary quadratic Lie algebras,
J. of Science, Ho Chi Minh city University of Education, No 47 (81) (2013) 25 36.
[7] M. T. Duong, R. Ushirobira, Singular quadratic Lie superalgebras, J. of Algebra 407
(2014) 372412.
[8] D.B.Fuchs, D.A.Leites, Cohomology of Lie superalgebras, Dokl.Bolg. Akad. Nauk 37
(10) (1984) 1294-1296.
[9] V. G. Kac, A sketch of Lie superalgebra theory, Commun. Math. Physics, 53 (1977)
3164.
[10] A. Medina, P. Revoy, Algbres de Lie et produit scalaire invariant, Ann. Sci. c. Norm.
Sup., 4me sr. 18 (1985) 553-561.
[11] I.A. Musson, G. Pinczon, R. Ushirobira, Hochschild cohomology and deformations of
CliffordWeyl algebras, SIGMA 5 27 (2009).
[12] T. D. Pham, M. T. Duong, A. V. Le, Solvable quadratic Lie algebras in low dimensions,
East-West J. of Math. 14 (2) (2012) 208-218.
[13] G. Pinczon, R. Ushirobira, New Applications of Graded Lie Algebras to Lie Algebras,
Generalized Lie Algebras, and Cohomology, J. Lie Theory, 17 (2007) 633-667.
[14] M.Scheunert, R.B.Zhang, Cohomology of Lie superalgebras and their generalizations,
J.Math. Phy. 39 (9) (1998) 5024-5061.
[15] Y.C.Su, R.B.Zhang, Cohomology of Lie superalgebras slmn and osp22n , Proc. London
Math. Soc. 94 (3) (2007) 91-136.
|
1712.00318 | 1 | 1712 | 2017-12-01T13:49:55 | On filiform Lie algebras. Geometric and algebraic studies | [
"math.RA"
] | A finite dimensional filiform K-Lie algebra is a nilpotent Lie algebra g whose nil index is maximal, that is equal to dim g -1. We describe necessary and sufficient conditions for a filiform algebra over an algebraically closed field of characteristic 0 to admit a contact linear form (in odd dimension) or a symplectic structure (in even dimension). If we fix a Vergne's basis, the set of filiform n-dimensional Lie algebras is a closed Zariski subset of an affine space generated by the structure constants associated with this fixed basis. Then this subset is an algebraic variety and we describe in small dimensions the algebraic components. | math.RA | math |
ON FILIFORM LIE ALGEBRAS. GEOMETRIC AND ALGEBRAIC
STUDIES
ELISABETH REMM
Abstract. A finite dimensional filiform K-Lie algebra is a nilpotent Lie algebra g whose nil
index is maximal, that is equal to dim g− 1. We describe necessary and sufficient conditions
for a filiform algebra over an algebraically closed field of characteristic 0 to admit a contact
linear form (in odd dimension) or a symplectic structure (in even dimension). If we fix a
Vergne's basis, the set of filiform n-dimensional Lie algebras is a closed Zariski subset of an
affine space generated by the structure constants associated with this fixed basis. Then this
subset is an algebraic variety and we describe in small dimensions the algebraic components.
Conventions. All Lie algebras considered in this paper will be defined over an alge-
braically closed fixed field K of characteristic 0.
Introduction. The problem of classification up to isomorphism is a substantial problem
in the study of finite dimensional Lie algebras, even over an algebraically closed field K of
characteristic 0. This problem has a solution if we consider in the simple or semisimple
algebras, that is, non abelian with no non-trivial ideals or with no non-zero abelian ideals.
Indeed, in 1884, Elie Cartan gave the classification of the complex and real simple finite
dimensional Lie algebras. His work is based on the works of Killing. He shows that this
classification is reduced to 4 classes and 5 exceptional Lie algebras. The Levi decomposition,
which was a conjecture of Killing and Cartan and was proved by Eugenio Elia Levi (1905),
states that any finite-dimensional real or complex Lie algebra is the semidirect product of a
solvable ideal and a semisimple subalgebra. From this result, the problem of classification
is reduced to the classification of solvable Lie algebras and to the problem of representation
of semisimple Lie algebras. Thus we are led to classify the solvable Lie algebras. But there
are only few results on this topic. We know this classification up to the dimensions 5 or 6.
Without effective solutions to these problems, since the structure of a solvable Lie algebra
is determined by its nilradical, that is the maximal nilpotent ideal, we are interested by the
classification of nilpotent Lie algebras. It is one of the firstly aims of this work. Nowadays,
only the classifications of complex or real nilpotent Lie algebras of dimension lower or equal
to 7 was established. Even this partial result was presented after numerous tries by various
authors and in often different approaches. It is surely unrealistic to hope for classifications in
bigger dimensions. Indeed, in dimension 7, we find non isomorphic families of one parameter
of nilpotent Lie algebras but also a very large number (more than 100) of not parametrized
and not isomorphic Lie algebras. In bigger dimension, this number of Lie algebras has to give
the dizziness and will become very difficult to verify (it is enough to see the different tries of
classifications in dimension 7 and the story of the dimension 7 to realize it already). Moreover,
when we have a given Lie algebra, it is often difficult to find this algebra in the official list of
the classification because it is not the same invariants which are used, and finding the change
THIS WORK HAS BEEN PRESENTED TO THE 13-TH INTERNATIONAL WORKSHOP ON
DIFFERENTIAL GEOMETRY AND ITS APPLICATIONS, ROMANIA, 2017
1
2
ELISABETH REMM
of basis is a little bit tedious. Thus to pursue the work of classification up an isomorphism
in bigger dimension seems utopian. Some works showed that particular families of nilpotent
algebras were parametrized by tensor spaces. This means that it equivalent to classify Lie
algebras and arbitrary bilinear maps. Therefore, it does not seem wise to study particular
properties of Lie algebras by starting on existing classifications. For example, the properties
that we consider in this paper are the existence of contact forms or of symplectic forms,
and also topological and algebraic properties based on the deformation theory and on the
rigidity property. Thus the approach is more original.
It was introduced in a previous
paper concerning the study of the k-step nilpotent Lie algebras. We globally study some
reduced families which are invariant by isomorphism and which are closed, that is defined
by a finite polynomial system. For this families, we can define an adapted cohomology and
then introduce a notion of local rigidity, that is we consider only deformations of elements
of a given family which stay in this family. In [13], we have for the first time studied some
geometrical properties of k-step nilpotent Lie algebras by considering family adapted to the
characteristic sequence of a nilpotent Lie algebra (see e.g.
[19] to a presentation of this
invariant).
We consider in this work filiform Lie algebras, that is nilpotent Lie algebra whose nil
index is maximal, that is n − 1 if n is the dimension of the Lie algebra. Of course, for
this family we know the classification up to the dimension 7. In [10], classifications are also
given for the dimensions 8 and 9 for this particular family of nilpotent Lie algebras. But
there was no general consensus on these classifications. Thus we begin our work with these
dimensions. We are interested in topological properties, as the rigidity, which we study by
developing this notion of restricted cohomology. In particular we come back on a result of
[3] concerning the local rigidity. We describe also in dimension 8 the family (which is neither
open nor closed) of symplectic Lie algebras, that is Lie algebras provided with a bilinear
symplectic form. In dimension 9, we show that there are no rigid filiform Lie algebras and
we determine all the contact Lie algebras. We show also that any symplectic 8-filiform
Lie algebra is obtained as a quotient of a contact filiform Lie algebra by its center and
we find again the first given description of symplectic Lie algebra. We do the same thing
for the dimensions 10 and 11, giving the description of the contact 11-dimensional filiform
Lie algebras and consequently the description of the symplectic 10-dimensional filiform Lie
algebras. We remark also, that in these dimensions, all none of the Lie algebras are rigid. For
the general dimension, we determine the family of contact (2p + 1)-dimensional filiform Lie
algebras. We expose also a model for this geometrical property, that is a family of contact
Lie algebras such that any filiform contact Lie algebra is a deformation of an algebra of this
model. Let us recall also that the reduction of the polynomial Jacobi system, that is the
system of polynomial equations given by the Jacobi conditions, is the fundamental problem.
We don't have many tools to find the generators of the ideal generated by these equations in
i,j, 1 ≤ i < j ≤ n, 1 ≤ k ≤ n]. In general this ideal I is not equal to √I and the
the ring K[C k
associated affine scheme is not reduced. We described for the family model (parametrized
by (p − 1) parameters) a process of reduction and we give the associated reduced system.
To end this study, we determine from the family of (2p + 1)-dimensional contact filiform
Lie algebra the family of symplectic (2p)-dimensional filiform Lie algebras and we propose
a notion of deformation of symplectic Lie algebra based on deformations of the contact Lie
algebra which is a the one dimensional central extension . Let us note also, that in [5], we
have studied filiform Lie algebras admitting a G-grading, where G is an abelian group.
[X0, Xi] = Xi+1, 1 ≤ i ≤ n − 1, [X1, Xn−1] = 0,
[Xi, Xj] = Xk≥i+j
i,jXk,
C k
ON FILIFORM LIE ALGEBRAS
3
1. Generalities on Filiform Lie algebras
1.1. The Vergne's basis. Let g be a n-dimensional Lie algebra over the field K. The
ascending central series {Cig} of g is defined by
for i > 0 and the descending central series {Ci
C0g = {0},
Cig = {X ∈ g / [X, g] ⊂ Ci−1g}
g} of g is defined by
g = g,
C0
Ci
g = [g,Ci−1
g],
i > 0.
Definition 1. The n-dimensional Lie algebra g is called filiform if we have dim Cig = i for
0 ≤ i ≤ n − 2.
A filiform Lie algebra is nilpotent and we have
Cig = Cn−i−1
g, 0 ≤ i ≤ n − 1.
Proposition 2. Let g be a (n + 1)-dimensional filiform Lie algebra. There exists a basis
{X0, X1,· · · , Xn} called the Vergne basis of g such that
Another characterization of a filiform Lie algebra is given by its characteristic sequence.
In fact if X is a vector of the nilpotent Lie algebra g, the characteristic sequence c(X) of the
adjoint operator adX is the decreasing sequence of the dimensions of the Jordan blocks of
the nilpotent operator adX. The characteristic sequence c(g) of g is the following sequence
max{c(X), X ∈ g − C1(g)}, the maximum corresponding to the lexicographic order. Any
vector X whose characteristic sequence c(X) of adX is equal to c(g) is called characteristic
vector of the nilpotent Lie algebra g (so according to the lexicographic ordering, we have
c(Y ) ≤ c(X) for any Y ∈ g if X is a characteristic vector). The n-dimensional nilpotent
Lie algebra g is filiform if and only if c(g) = n − 1. If {X0, X1,· · · , Xn} is a Vergne basis
of a filiform Lie algebra g, the characteristic sequence c(X0) of the adjoint operator adX0 is
equal to (n, 1). An interesting example of (n + 1)-dimensional filiform Lie algebra is the Lie
algebra Ln+1 often called the model filiform Lie algebra ([11]) whose Lie bracket µ0 is
(1)
(cid:26) µ0(X0, Xi) = Xi+1, 1 ≤ i ≤ n − 1,
µ0(Xi, Xj) = 0, 1 ≤ i < j ≤ n.
1.2. Geometric structure on filiform Lie algebras.
1.2.1. Contact and symplectic structures. Let g be a (2p)-dimensional K-Lie algebra. A
symplectic form on g is a closed 2-form θ, that is satisfying
θ([X, Y ], Z) + θ([Y, Z], X) + θ([Z, X], Y ) = 0
for any X, Y, Z ∈ g and which is also nondegenerate that is
θp = θ ∧ · · · ∧ θ 6= 0.
A Lie algebra provided with a symplectic form θ is called a symplectic Lie algebra and
denoted by the pair (g, θ). There exist filiform Lie algebras without symplectic structures.
For example, in [6], on find the list of filiform Lie algebras up the dimension 6. When
4
ELISABETH REMM
∗ the dual space of g such that
the symplectic form is exact, that is if there exists ω in g
θ = dω where dω is the bilinear form defined by dω(X, Y ) = −ω([X, Y ]) for any X, Y ∈ g,
the symplectic Lie algebra is called frobeniusian. But, from [14], there are no frobeniusian
nilpotent Lie algebras.
Let g be a n = 2p + 1-dimensional K-Lie algebra. A contact form on g is non zero linear
form ω on g satisfying
ω ∧ (dω)p 6= 0
where dω is the bilinear form defined by dω(X, Y ) = −ω([X, Y ]) for any X, Y ∈ g and
(dω)p = dω ∧ · · · ∧ dω) p-times. In case of nilpotent Lie algebras, there is an obstruction to
the existence of contact form ([14]), the center of g can be of dimension 1. But the center
of any filiform Lie algebra is always of dimension 1, then this necessary conditions is always
satisfied. Let us note that this does'nt imply that any filiform odd dimensional Lie algebras
admits a contact form.
Proposition 3. Let (g, θ) be a 2p-dimensional filiform symplectic Lie algebra. Then the one
dimensional central extension gθ = g ⊕ KZ whose bracket is given by
(cid:26) [X, Y ]gθ = [X, Y ] + θ(X, Y )Z, ∀X, Y ∈ g
[X, Z]gθ = 0, ∀X ∈ g
is a 2p + 1-dimensional filiform contact Lie algebra.
Proof. Let {X0, X1,· · · , X2p−1} be a Vergne basis of g and let be {ω0,· · · , ω2p−1} is the dual
basis. If θ is a symplectic form on g then
dθ(X0, Xi, X2p−1) = 0 = θ(Xi+1, X2p−1),
i = 1,· · · , 2p − 1.
This implies that θ = ω2p−1 ∧ (a0ω0 + a1ω1) + θ1.
Conversely, if g is a (2p + 1)-dimensional contact nilpotent Lie algebra, then its center
Z(g) is one dimensional ([14]) and the factor algebra g/Z(g) is a symplectic 2p-dimensional
nilpotent Lie algebra. If g is filiform, then g/Z(g) is also filiform. In [16], one proves that
g admits a contact form iff the linear form ω2p is a contact form where {ω0,· · · , ω2p} is the
dual basis of the Vergne basis of g. We deduce
Proposition 4. A (2p)-dimensional Lie algebra (g, θ) is symplectic iff in the central exten-
sion gθ, the linear form ω2p is a contact form, where {ω0,· · · , ω2p} is the dual basis of the
Vergne basis of gθ.
1.2.2. Complex structures.
Definition 5. A complex structure on an (2p)-dimensional R-Lie algebra g is a linear en-
domorphism J of g such that:
(1) J 2 = −Id;
(2) [JX, JY ] − [X, Y ] − J[JX, Y ] − J[X, JY ] = 0,
Let gC = g ⊗R C denote the complexification of g, and
∀X, Y ∈ g.
σ : gC → gC
the corresponding conjugation. The second condition is equivalent to the splitting
where g
1,0 and g
0,1 are complex Lie subalgebras of gC and g
0,1 = σ(g
1,0).
gC = g
0,1
1,0 ⊕ g
ON FILIFORM LIE ALGEBRAS
5
Proposition 6. [15] There are no filiform Lie algebra admitting a complex structure.
1.2.3. Affine structures.
Definition 7. An affine structure on a Lie algebra g is a K-bilinear multiplication, denoted
X · Y which is left-symmetric, that is
X · (Y · Z) − (X · Y ) · Z = Y · (X · Z) − (Y · X) · Z
for all X, Y, Z ∈ g.and satisfies
where [X, Y ] denotes the Lie bracket of g.
[X, Y ] = X · Y − Y · X
The problem, which concerns also the linear representations of Lie algebra ([7]), is no
completely solved even for filiform Lie algebras. We know that, as soon as the dimension is
greater or equal to 10, there exist filiform Lie algebras without affine structure. However,
let us recall this classical result:
Proposition 8. Any symplectic Lie algebra is affine.
Proof. Let (g, θ) be a symplectic Lie algebra. We consider the product XY given by XY =
f (X)Y where f : g → End(g) is defined implicitly by θ(f (X)(Y ), Z) = −θ(Y, [X, Z]) for
any X, Y, Z ∈ g. Since θ is symplectic, this product XY is well defined and provides g with
an affine structure.
In the case of contact Lie algebras, see for example ([18]).
2. Filiform Lie algebras of dimension 8
2.1. Topological description. The classification of filiform Lie algebras over K of dimen-
sion less than or equal to 7 is well known ([11]). The aim of this section is to come back to
the classification proposed in [3] and corrected some mistakes of this last paper.
Let g be a 8-dimensional filiform Lie algebra.
{X0,· · · , X7} a Vergne basis, then the Jacobi identity implies
µ(X0, µ(Xi, Xj)) = µ(Xi, Xj+1) + µ(Xi+1, Xj).
If we denote by µ its Lie bracket and
These identities imply that the structure constants C k
i,j for k < 7 are linear combinations
of C 7
ij. We deduce
Proposition 9. [3] Any 8-dimensional filiform Lie algebra over K is given in a Vergne basis
by
µ(X0, Xi) = Xi+1, 1 ≤ i ≤ 6,
µ(X2, X5) = a1X7,
µ(X1, X5) = a1X6 + a2X7,
µ(X3, X4) = −a1X7,
µ(X2, X4) = a4X7,
µ(X1, X4) = a1X5 + (a2 + a4)X6 + a5X7,
µ(X2, X3) = a4X6 + a6X7,
µ(X1, X3) = a1X4 + (a2 + 2a4)X5 + (a5 + a6)X6 + a7X7,
µ(X1, X2) = a1X3 + (a2 + 2a4)X4 + (a5 + a6)X5 + a7X6 + a8X7,
(2)
with a1(5a4 + 2a2) = 0.
Consequence: Let Lie8 be the algebraic variety over K of 83-uple (C k
and 0 ≤ k ≤ 7 satisfying
C k
i,j = −C k
7
Xl=0
i,jC s
l,i = 0, ∀s = 0,· · · , 7.
l,k + C l
C l
j,i
j,kC s
l,i + C l
k,iC s
6
ELISABETH REMM
i,j) with 0 ≤ i < j ≤ 7
A 8-dimensional Lie algebra with Lie bracket µ is identified with a point of Lie8 consid-
ering the structural constants of µ in a given the basis. An action of the algebraic group
GL(8, K) on Lie8 corresponds to the changes of basis and the orbit of a Lie algebra is its
isomorphism class. Let N il8 be the algebraic subvariety of Lie8 whose elements correpond
the 8-dimensional nilpotent Lie algebra and F il8 be the set of 8-dimensional filiform Lie al-
gebras. It is a Zariski open set of N il8 and from Proposition 9 it is the orbit of the subvariety
F il8 of N il8 whose elements are the Lie algebras defined in (2). We deduce that the study
of the open set F il8 can be deduced directly from the study of F il8.
The set F il8 is an algebraic variety embedded in K8 and parametrized by the structural
constants a1, a2, a4, a5, a6, a7, a8. It is the union of two irreducible connected algebraic com-
ponents
ij) of N il8 (resp. F il8) where the C k
(1) F il8(1) defined by a1 = 0 which is a 6-dimensional plane,
(2) F il8(2) defined by 5a4 + 2a2 = 0 which is also a 6-dimensional plane.
We deduce that F il8 is the union of two irreducible algebraic components, F il8(1) and
F il8(2) which are respectively the orbits of F il8(1) and F il8(2).
2.2. Deformations. In the following, we identify the Lie bracket of a 8-dimensional nilpo-
tent (resp. filiform) Lie algebra with the point (C k
i,j are
its structural constants related to the given Vergne basis {X0,· · · , X7}.
Definition 10. Let µ0 be a Lie bracket belonging to F il8. A deformation µ in F il8 of µ0 is
a formal deformation in the Gerstenhaber sense such that µ ∈ F il8 ⊗ K[[t]].
From [9], any deformation in F il8 is isomorphic to a linear deformation µ0 + tψ where
ψ is a bilinear skew-symmetric application which is a 2-cocycle for the Chevalley-Eilenberg
cohomology of µ0 and satisfying also the Jacobi identity. Moreover since µ0 + tψ is filiform,
ψ is a nilpotent Lie bracket.
The determination of deformations in F il8 of µ0 ∈ F il8 reduces to the study of bilinear
skew-symmetric applications ψ such that µ0 + tψ is in F il8.
Lemma 11. Let µ0 be in F il8. Then µ0 + tψ is a linear deformation in F il8 of µ0 if and
only if ψ is given by
(3)
ψ(X2, X5) = u1X7,
ψ(X1, X5) = u1X6 + u2X7,
ψ(X3, X4) = −u1X7,
ψ(X2, X4) = u4X7,
ψ(X1, X4) = u1X5 + (u2 + u4)X6 + u5X7,
ψ(X2, X3) = u4X6 + u6X7,
ψ(X1, X3) = u1X4 + (u2 + 2u4)X5 + (u5 + u6)X6 + u7X7,
ψ(X1, X2) = u1X3 + (u2 + 2u4)X4 + (u5 + u6)X5 + u7X6 + u8X7,
ON FILIFORM LIE ALGEBRAS
7
with u1(5a4 + 2a2) + a1(5u4 + 2u2) + tu1(5u4 + 2u2) = 0 where (a1, a2, a4, a5, a6, a7, a8) are
the parameters of µ0.
2.3. Study of the component F il8(1). Any Lie algebra in F il8(1) is isomorphic to a Lie
algebra of F il8(1) whose Lie bracket is defined by
(4)
µ0(X0, Xi) = Xi+1, 1 ≤ i ≤ 6,
µ0(X1, X5) = a2X7,
µ0(X2, X4) = a4X7,
µ0(X1, X4) = (a2 + a4)X6 + a5X7,
µ0(X2, X3) = a4X6 + a6X7,
µ0(X1, X3) = (a2 + 2a4)X5 + (a5 + a6)X6 + a7X7,
µ0(X1, X2) = (a2 + 2a4)X4 + (a5 + a6)X5 + a7X6 + a8X7.
We have seen that F il8(1) is an algebraic subvariety of N il8.
In [13], we have define
for each set k-N iln of k-step nilpotent n-dimensional Lie algebra a cochain complex whose
associated cohomology parametrizes the "internal" deformations, that is deformations of k-
step nilpotent Lie algebras which are also k-step nilpotent. When k is maximal, that is for the
filiform case, this cohomology is the Vergne cohomology because any nilpotent deformation of
a filiform Lie algebra is always filiform. We consider here the same approach, considering the
cohomology adapted to the internal deformations in F il8(1) that is deformations of elements
of F il8(1) which remain in this variety. Since we are only concerned by the second space
of this cohomology, we shall describe it. Let µ be in F il8(1). We denote by Z 2
CR(µ, µ) the
space of 2-cochains, that is bilinear skew-symmetric maps ψ on K8 with values on K8 which
are defined by
(5)
ψ(X1, X5) = u2X7,
ψ(X2, X4) = u4X7,
ψ(X1, X4) = (u2 + u4)X6 + u5X7,
ψ(X2, X3) = u4X6 + u6X7,
ψ(X1, X3) = (u2 + 2u4)X5 + (u5 + u6)X6 + u7X7,
ψ(X1, X2) = (u2 + 2u4)X4 + (u5 + u6)X5 + u7X6 + u8X7.
If ∂µ is the coboundary operator of the Chevalley-Eilenberg complex of µ, then ∂µ(ψ) = 0
and any cochain is closed. Let B2
CR(µ, µ) the space of ∂µ(f ) for f ∈ End(K8) such that
i=1 βiXi. We have
f (X0) = P7
(6)
i=0 αiXi and f (X1) = P7
δf (X1, X5) = v2X7,
δf (X2, X4) = v4X7,
δf (X1, X4) = (v2 + v4)X6 + v5X7,
δf (X2, X3) = v4X6 + v6X7,
δf (X1, X3) = (v2 + 2v4)X5 + (v5 + v6)X6 + v7X7,
δf (X1, X2) = (v2 + 2v4)X4 + (v5 + v6)X5 + v7X6 + v8X7,
ELISABETH REMM
8
with
(7)
2 − 5a2a4 − 5a2
4),
v2 = a2(β1 − 2α0),
v4 = a4(β1 − 2α0),
v5 = a5(β1 − 3α0) + α1(−2a2
v6 = a6(β1 − 3α0) + α1(−3a2a4 − 3a2
4),
v7 = a7(β1 − 4α0) − 2a4β3 − α1(a5 + a6)(5a2 + 9a4),
v8 = a8(β1 − 5α0) − 3a7α1(2a2 + 3a4) − 2a6β3 − 3a4β4
+3α3a4(a2 + 2a4) − α1(a5 + a6)(3a5 + 2a6).
CR(µ, µ) and the quotient space
Then B2
CR(µ, µ) is a linear subspace of Z 2
H 2
CR(µ, µ) = Z 2
CR(µ, µ)/B2
CR(µ, µ)
parametrizes the deformations in F il8(1). We deduce, using the classical theory of Nijenhuis-
Richardson, that a Lie algebra g with bracket µ is such that H 2
CR(µ, µ) = 0 is rigid in F il8(1),
that is, its orbit in F il8 is open. But, since F il8(1) is isomorphic to a linear 6-dimensional
space, its associated affine scheme is naturally reduced and the converse is also true: if g is
rigid, then dim H 2
CR(µ, µ) = 0. But any cocycle ψ with u2u4 6= 0 can not be cohomologous
with a cocycle where u2 = u4 = 0. We deduce that dim H 2
CR(µ, µ) ≥ 1. We deduce
Proposition 12. No Lie algebra belonging to F il8(1) is rigid in F il8 and also in N il8 and
in Lie8 where N il8 (respectively Lie8) is the algebraic variety of 8-dimensional nilpotent Lie
algebras (respectively the algebraic variety of 8-dimensional Lie algebras).
Let us determine the Lie algebras of this component which satisfy dim H 2
CR(µ, µ) = 1. Let
g be such a Lie algebra. From the previous remark, any cocycle ψ ∈ Z 2
CR(µ, µ) must be
cohomologous to a cocycle with u5 = u6 = u7 = u8 = 0 and a2 or a4 not zero. Suppose
a4 6= 0. The coefficients β3 and β4 can always be chosen such that u7−v7 = 0 and u8−v8 = 0.
(1) If a2 + a4 6= 0 and a5 6= 0 we can choose α1, β1 − 3α0 and β1 − 2α0 such that u6 − v6 =
a4 6= −5 ± i√15
(here K = C) we can choose α1, β1 − 3α0 and β1 − 2α0 such that u6 − v6 = u5 − v5 =
u2 − v2 = 0. The corresponding Lie algebra satisfies dim H 2
(3) a2 + a4 = 0 and a6 6= 0 we can choose α1, β1 − 3α0 and β1 − 2α0 such that u6 − v6 =
u5 − v5 = u2 − v2 = 0. The corresponding Lie algebra satisfies dim H 2
u5 − v5 = u2 − v2 = 0. The corresponding Lie algebra satisfies dim H 2
a2
(2) If a2 + a4 6= 0, a5 = 0, a6 6= 0 and 2a2
4 6= 0 that is
2 + 5a2a4 + 5a2
CR(µ, µ) = 1.
CR(µ, µ) = 1.
4
CR(µ, µ) = 1.
In all other cases dim H 2
CR(µ, µ) > 1.
To simplify denote by µ(a2, a4, a5, a6, a7, a8) a Lie bracket of a Lie algebra belonging to
F il8(1). The previous computations shows that the Lie algebras µ(α, 1, 0, 1, 0, 0) with α such
that 2α2+5α+5 6= 0 and µ(α, 1, 1, 0, 0, 0) with α 6= −1 satisfy dim H 2
CR(µ, µ) = 1. From ([3]),
these two Lie algebras are isomorphic. So consider the one-parameter family T 1
α constituted
of µ(α, 1, 1, 0, 0, 0) with α 6= −1. Any deformation in F il8(1) of an algebra of this family
belongs to this family. Since F il8(1) is a 6-dimensional plane, a reduced algebraic variety,
we deduce that the closure of T 1
α is F il8(1).
ON FILIFORM LIE ALGEBRAS
9
Proposition 13. The family T 1
α
µ(X0, Xi) = Xi+1, 1 ≤ i ≤ 6,
µ(X1, X5) = αX7,
µ(X2, X4) = X7,
µ(X1, X4) = (α + 1)X6 + X7,
µ(X2, X3) = X6,
µ(X1, X3) = (α + 2)X5 + X6,
µ(X1, X2) = (α + 2)X4 + X5.
is rigid in F il8(1) and F il8(1) = O(T 1
α ).
Remarks
(1) This generalized notion of rigidity which concerns a one-parameter family of Lie
algebras has already been defined in([9].
(2) To compare these results with [3], we give the complex classification, up to isomor-
phism, of elements of F il8. Recall that such results would be utopic to establish for
greater dimensions.
Proposition 14. Let us write (a2, a4, a5, a6, a7, a8) a Lie algebra of F il8(1). Then any Lie
algebra of F il8(1) is isomorphic to one of the following:
(λ, 1,−1, 1, 0, 0)
(0, 0, λ, 1, 1, 0)
(1, 0, 0, 0, 0, 1)
(0, 0, 0, 0, 1, 0)
(λ, 1, 0, 0, 0, 0)
(0, 0, λ, 1, 0, 0)
(1, 0, 0, 0, 0, 0)
(0, 0, 0, 0, 0, 1)
(−2, 1, 1, 0, 0, 0)
(λ, 0, 0, 0, 1, 1)
(0, 0, 1, 0, 1, 0)
(0, 0, 0, 0, 0, 0).
(1, 0,−1, 1, λ, 0)
(1, 0, 0, 0, 1, 0)
(0, 0, 1, 0, 0, 0)
2.4. Study of the component F il8(2). We consider the Lie algebras of (2) with a4 = − 2
5 a2.
Any Lie algebra of F il8(2) is isomorphic to a Lie algebra of F il8(2) with Lie bracket defined
by:
(8)
(9)
µ(X0, Xi) = Xi+1, 1 ≤ i ≤ 6,
µ(X2, X5) = a1X7,
µ(X1, X5) = a1X6 + a2X7,
µ(X3, X4) = −a1X7,
µ(X2, X4) = −
a2X7,
3
µ(X1, X4) = a1X5 +
5
2
5
µ(X2, X3) = −
µ(X1, X3) = a1X4 +
2
5
µ(X1, X2) = a1X3 +
a2X6 + a5X7,
a2X6 + a6X7,
1
5
1
5
a2X5 + (a5 + a6)X6 + a7X7,
a2X4 + (a5 + a6)X5 + a7X6 + a8X7,
Then F il8(2) is a 6-dimensional plane parametrized by (a1, a2, a5, a6, a7, a8). We consider
similarly to the previous section the linear deformations in F il8(2) of the Lie brackets be-
longing to F il8(2). We denote always by H 2
CR(µ, µ) the space which parametrizes these de-
formations. The space of 2-cocycles Z 2
CR(µ, µ) is constituted of the skew-symmetric bilinear
10
ELISABETH REMM
applications ψ given by
(10)
u2X6 + u5X7,
ψ(X2, X5) = u1X7,
ψ(X1, X5) = u1X6 + u2X7,
ψ(X3, X4) = −u1X7,
ψ(X2, X4) = −
u2X7,
3
ψ(X1, X4) = u1X5 +
5
2
5
ψ(X2, X3) = −
ψ(X1, X3) = u1X4 +
2
5
u2X6 + u6X7,
ψ(X1, X2) = u1X3 +
1
5
1
5
u2X5 + (u5 + u6)X6 + u7X7,
u2X4 + (u5 + u6)X5 + u7X6 + u8X7.
Let f be a linear endomorphism of g such that δf ∈ Z 2
f (X1) = P7
i=1 βiXi then
CR(µ, µ). If f (X0) = P7
i=0 αiXi and
(11)
with
(12)
δf (X2, X5) = v1X7,
δf (X1, X5) = v1X6 + v2X7,
δf (X3, X4) = −v1X7,
δf (X2, X4) = −
v2X7,
3
δf (X1, X4) = v1X5 +
5
2
5
δf (X2, X3) = −
δf (X1, X3) = v1X4 +
2
5
v2X6 + v6X7,
δf (X1, X2) = v1X3 +
v2X6 + v5X7,
1
5
1
5
v2X5 + (v5 + v6)X6 + v7X7,
v2X4 + (v5 + v6)X5 + v7X6 + v8X7,
v1 = a1(β1 − α0 − α1a1),
v2 = a2(β1 − 2α0 − 3α1a1),
v5 = a5(β1 − 3α0 − 5a1α1) − α1(2a1a6 + 4
2) + 2α3a2
5a2
1 − 2β3a1,
v6 = a6(β1 − 3α0 − 2a1α1) + α1(a1a5 + 18
25 a2
2) − 2α3a2
1 + 2β3a1,
v7 = a7(β1 − 4α0 − 5a1α1) − 7
5α1a2(a5 + a6) − 4
5 β3a2,
v8 = a8(β1 − 5α0 − 5a1α1) − α1( 12
5 a2a7 + (a5 + a6)(3a5 + 2a6))+
2) − 4
25 a2
5 α4a1a2 + 2α5a2
CR(µ, µ) ≥ 3.
α3(2a1(a5 + 2a6) − 6
1 − 2β3a6 + 6
5 α3a1a2 + 4
(1) If a1 = a2 = 0 then dim H 2
(2) If a1 = 0 and a2 6= 0, then dim H 2
(3) If a2 = 0, a1 6= 0 then dim H 2
(4) Assume now a2 6= 0, a1 6= 0, We will develop this case because it is the part of [3]
which presents a mistake. Let us compute the kernel of the linear system {vi = 0}.
Since a1a2 6= 0 then v1 = v2 = 0 is equivalent to
CR(µ, µ) ≥ 2.
CR(µ, µ) ≥ 1.
5β4a2 − 2β5a1,
β1 = −α1a1, α0 = −2α1a1.
ON FILIFORM LIE ALGEBRAS
11
We can also choose β5 to obtain v8 = 0. Then the system is reduced to
v5 = −α1(2a6a1 + 4
5a2
v6 = (3a6a1 + a5a1 + 18
v7 = (2a7a1 − 7
The matrix of this system is
2) + 2α3a2
25a2
2)α1 − 2α3a2
1 − 2β3a1,
1 + 2β3a1,
5α3a1a2 + 4
5 β3a2,
5a2(a5 + a6))α1 − 4
−2a6a1 − 4
M =
5a2
2
3a6a1 + a5a1 + 18
2a7a1 − 7
Since this matrix is singular, then dim KerM ≥ 1 and dim H 2
CR(µ, µ) ≥ 1. Since
the affine scheme associated with this component is reduced, the Lie algebras with
a1a2 6= 0 are not rigid. We can now study the conditions to have dim H 2
CR(µ, µ) = 1,
this is equivalent to rank(M) = 2 that is
2a2
−2a1
1
−2a2
2a1
5 a1a2 + 4
5a2
5a2(a5 + a6) − 4
25 a2
2
1
or
a1a2(cid:18)18
5
a1(a5 + a6) −
a6(cid:19) +
26
5
a5 +
2
25
a2
2 6= 0
72
125
a3
2 − 4a2
1a7 6= 0.
In particular, we can take a1 = a2 = 1, a5 = −a6 = t. For each value of t, the
dimension of H2,r(g, g) of the corresponding Lie algebra is equal to 1. We deduce
Proposition 15. None of the Lie algebras of F il8(2) is rigid in F il8. This component is
the closure of the one dimensional rigid family T 2
t (8) of Lie algebras isomorphic to
(13)
µ(X0, Xi) = Xi+1, 1 ≤ i ≤ 6,
µ(X2, X5) = X7,
µ(X1, X5) = X6 + X7,
µ(X3, X4) = −X7,
µ(X2, X4) = −
X7,
3
µ(X1, X4) = X5 +
5
2
5
X6 + tX7,
µ(X2, X3) = −
µ(X1, X3) = X4 +
2
5
X6 − tX7,
µ(X1, X2) = X3 +
1
5
1
5
X5,
X4.
Remark. The problem of classification of 8-dimensional filiform Lie algebras has been
already solved. We can find this classification in [9]. A lot of the results previously obtained
of course are direct consequence of this classification. But to obtain a general result of
the classification problem is certainly utopian. This implies to develop another way. We
began a new approach in [13] by considering subfamilies of k-step nilpotent Lie algebras and
defining an adapted cohomology of deformations. The previous calculus are made in this
way. Nevertheless, since this classification is known and since we have given this classification
for the algebras of the first component, it would be surprising not to give it for the second
component.
12
ELISABETH REMM
Proposition 16. Let g be a filiform Lie algebra belonging to F il8(2). Then any Lie algebra
of F il8(2) is isomorphic to one of the following corresponding to
(a1, a2, a5, a6, a7, a8) ∈ {(1, 0, 0, 0, 0, 0), (1, 0, 0, 0, 1, 0), (1, 0, 1, 0, λ, 0), (1, 1, λ,−2, 0, 0)}.
2.5. Symplectic structures. We determine all filiform 8-dimensional symplectic Lie al-
gebras. A direct approach consists to write the conditions related to the existence of a
symplectic form for Lie algebras belonging to each component. Let g ∈ F il8 and θ be a
closed 2-form on g. Let {X0,· · · , X7} the Vergne basis (2). Computing dθ(X0, Xi, X7) = 0,
we obtain that θ(Xi, X7) = 0 for i = 2,· · · , 6. As consequence, θ(Xi, Xj) = 0 as soon as
i + j ≥ 9 and dθ(Xi, Xj, Xk) = 0 is always satisfies for i + j + k ≥ 9. We call the weight
of the equation dθ(Xi, Xj, Xk) = 0 the integer p = i + j + k and we solve this equation for
p = 8, 7,· · · , 3.
(1) p = 8.
(cid:26) 0 = (3a1 + 2a3)θ(X3, X5) − a1θ(X1, X7) + a1θ(X2, X6),
0 = (2a1 + a3)θ(X3, X5) − a3θ(X1, X7).
(2) p = 7.
Then, we have to consider the matrix
0 = θ(X2, X6) + θ(X1, X7),
0 = θ(X3, X5) − a1θ(X0, X7) + θ(X2, X6),
0 = a3θ(X0, X7) − θ(X3, X5),
0 = (3a1 + 2a3)θ(X3, X4) − (a1 + a3)θ(X1, X6) − a4θ(X1, X7)
+(2a1 + a3)θ(X2, X5) + (a2 + a4)θ(X2, X6).
M =
0 −a1 a1 3a1 + 2a3
0 −a3
2a1 + a3
0
1
−a1
0
a3
0
0
1
1
0
0
1
−1
=
0 −a1 a1
a1
0
0
a1
0
1
0
−a1
1
1
−a1
0 −1
a1
1
0
0
If θ is symplectic, one of the scalar θ(X0, X7) or θ(X1, X7) is non zero which is equivalent
to say that the rank of M is less than 4. But rankM = 4 if and only if a1 6= 0 so any
8-dimensional filiform symplectic Lie algebra g belongs to F il8(1). Moreover, the symplectic
form satisfies θ(Xi, Xj) = 0 for i + j ≥ 8. If we compute the relations of weight 6 we obtain
0 = θ(X2, X5) + θ(X1, X6) − a2θ(X0, X7),
0 = θ(X3, X4) + θ(X2, X5) − a4θ(X0, X7),
0 = (a2 + 2a4)θ(X3, X4) − (a2 + 2a4)θ(X2, X5) + a4θ(X1, X6).
But θ is non degenerate if and only if θ(0, 7)θ(1, 6)θ(2, 5)θ(3, 4) 6= 0. This implies
(1) 2a2 + 5a4 6= 0, that is g ∈ F il8(1) and g /∈ F il8(2) and a4(a2 + 2a4)(2a2 − a4) 6= 0
(2) or a2 = a4 = 0.
Proposition 17. An 8-dimensional filiform Lie algebra is symplectic if and only if it is
isomorphic to a Lie algebra g ∈ F il8(1)− F il8(2) and a4(a2 + a4)(2a2 − a4)(a2 + 2a4) 6= 0 or
g ∈ F il8(1) ∩ F il8(2) and a2 = a4 = 0.
ON FILIFORM LIE ALGEBRAS
13
We can note that any Lie algebra of the rigid family T 1
of α which are −2, 1
2,− 5
2.
α is symplectic, except for three values
2.6. Determination of symplectic 8-dimensional filiform Lie algebras using contact
9-dimensional filiform Lie algebras. If (g, θ) is a 2p-dimensional symplectic Lie algebra,
then the Lie algebra gθ the one dimensional central extension
Recall that the Lie bracket mu1of gθ is given by
gθ = g ⊕θ KZ.
(cid:26) µ1(X, Y ) = θ(X, Y )Z + µ(X, Y )
µ1(X, Z) = 0
for any X, Y ∈ g and gθ is a contact (2p + 1)-dimensional Lie algebra and Z generates the
center. From Proposition 4, g is a factor algebra g1/Z(g1) of a filiform contact algebra and
the linear form ω2p is a contact form. Then we can determine all the symplectic filiform
algebra in dimension 8 starting from the contact filiform 9-dimensional Lie algebras. This
study is the aim of the last section, but we can already use these results, all the proofs are
given in the following section.
Proposition 18. Any 9-dimensional filiform Lie algebra provided with a contact form is
isomorphic to a Lie algebra of the following family:
µ(X0, Xi) = Xi+1, 1 ≤ i ≤ 7,
µ(X1, X6) = a2X8,
µ(X2, X5) = a4X8,
µ(X1, X5) = (a2 + a4)X7 + a5X8,
µ(X3, X4) = a6X8
µ(X2, X4) = (a4 + a6)X7 + a7X8,
µ(X1, X4) = (a2 + 2a4 + a6)X6 + (a5 + a7)X7 + a8X8,
µ(X2, X3) = (a4 + a6)X6 + a7X7 + a9X8,
µ(X1, X3) = (a2 + 3a4 + 2a6)X5 + (a5 + 2a7)X6 + (a8 + a9)X7 + a10X8,
µ(X1, X2) = (a2 + 3a4 + 2a6)X4 + (a5 + 2a7)X5 + (a8 + a9)X6 + a10X7 + a11X8,
4 + 2a2
with −3a2
Since the center is generated by X8, we deduce
6 + 2a2a6 + a4a6 = 0 and a2a4a6 6= 0.
Proposition 19. Any symplectic 8-dimensional filiform Lie algebra is isomorphic to a Lie
algebra of the following family
with −3a2
4 + 2a2
µ(X0, Xi) = Xi+1, 1 ≤ i ≤ 6,
µ(X1, X5) = (a2 + a4)X7,
µ(X2, X4) = (a4 + a6)X7,
µ(X1, X4) = (a2 + 2a4 + a6)X6 + (a5 + a7)X7,
µ(X2, X3) = (a4 + a6)X6 + a7X7,
µ(X1, X3) = (a2 + 3a4 + 2a6)X5 + (a5 + 2a7)X6 + (a8 + a9)X7,
µ(X1, X2) = (a2 + 3a4 + 2a6)X4 + (a5 + 2a7)X5 + (a8 + a9)X6 + a10X7
6 + 2a2a6 + a4a6 = 0 and a2a4a6 6= 0. This is equivalent to say
14
ELISABETH REMM
We come back to the notations (2) and we put
b2 = a2 + a4, b4 = a4 + a6, b5 = a5 + a7, b6 = a7, b7 = a8 + a9, b8 = a10.
Then the conditions −3a2
4 + 2a2
6 + 2a2a6 + a4a6 = 0 and a2a4a6 6= 0 implies that
b2 = b4 = 0, or b4(b2 + b4)(2b2 − b4)(b2 + 2b4) 6= 0.
Any symplectic 8-dimensional filiform Lie algebra is isomorphic to a Lie algebra of the
following family
µ(X0, Xi) = Xi+1, 1 ≤ i ≤ 6,
µ(X1, X5) = b2X7,
µ(X2, X4) = b4X7,
µ(X1, X4) = (b2 + b4)X6 + b5X7,
µ(X2, X3) = b4X6 + b6X7,
µ(X1, X3) = (b2 + 2b4)X5 + (b5 + b6)X6 + b7X7,
µ(X1, X2) = (b2 + 2b4)X4 + (b5 + b6)X5 + b7X6 + b8X7
with b2 = b4 = 0 or b4(b2 + b4)(b2 − b4)(b2 + 2b4) 6= 0. We find again all the conditions of
Proposition 17.
This last way to determine the symplectic structures permits also the introduce a notion
of symplectic deformation. Recall that a deformation of a symplectic Lie algebra can be
non symplectic. The simplest example is given by the even dimensional abelian Lie algebra.
This algebra is symplectic and any Lie algebra is isomorphic to a deformation of this abelian
algebra and it is clear that exists non symplectic Lie algebras as soon as the dimension is
strictly greater than 2. Likewise if a symplectic Lie algebra g1 s a deformation of a Lie
algebra g0, this last is not necessarily symplectic. Then the classical notion of deformation
is not well adapted to the notion of symplectic structures. But it is not the case for the
contact structures. Any deformation of a contact Lie algebra is always a contact Lie algebra
(see also [14]). This remark leads to introduce a restricted notion of deformation that we
shall call symplectic deformation:
9
Definition 20. Let g0 and g1 be two symplectic 8-dimensional filiform Lie algebras and let g
0
9
9
and g
i ), i = 1, 2 where π
1 be 9-dimensional contact filiform Lie algebras such that gi = π(g
is the canonical projection π : g → g/Z(g). We shall call that g1 is a symplectic deformation
9
9
of g0 if g
1 is a (classical) deformation of g
0.
2.7. Affine structures. From [7], we know that any 8-filiform Lie algebra admits an affine
structure. To prove this, we construct affine structure of adjoint type, that is, if Li denote
the linear map Li(X) = XiX then L0 = adX0. Since LI for i ≥ 2 is given by Li = [L0, Li−1],
such affine structure is completely determinate by L1. For exemple if we consider the rigid
family T 2
t (8) in F il8(2), we consider for L1 the linear map whose matrix in the Vergne's
basis is
ON FILIFORM LIE ALGEBRAS
15
α1
0
0
− 1
5
0
0
0
0
0
0
0 2(70α6−25t−42)
0
0
375
2α3
α4
0
0
0
0
0
0
α2
0
0
0
0
0
70α6−25t−42
375
α3
− 2
25 (5α6 − 3)
α5
0
0
0
0
0
0
1
5
0
0
0
0
0
0
0
0
0
0
0
0
0
0
25 (5α6 − 3) α6 − 1
2 0
0
0
0
0
0
0
0
t
2 − 3
α1
0
0
3
5
0
0
0
0
0
0
0 2(−210α6+125t−42)
0 2(−210α6+125t−42)
0
375
α4
375
0
0
0
0
2
5
0
−210α6+125t−42
375
α3
α2
0
0
0
0
− 2
α5
5
− 14
25 (5α6 + 1)
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
− 3
0
0
25(5α6 + 1) α6 − 1
2 0
0
0
0
0
0
0
0
5
t
2 − 21
or
Note that all these affine structure are complete, that is the linear map RY : X → X · Y
is nilpotent for any Y , or equivalently the trace of RY is zero. Note also that the linear
representation of g:
given by ρ(X) = LX is not faithful because in all the previous cases LX7 = 0.
ρ : g → End(g)
Remark. Let us consider the polarization of the product ∇(X, Y ) = X · Y . We put
µ(X, Y ) = ∇(X, Y ) − ∇(Y, X), s(X, Y ) = ∇(X, Y ) + ∇(Y, X).
Since K is of characteristic not 2, then
∇(X, Y ) =
s(X, Y ) + µ(X, Y )
2
and µ is a Lie bracket. The applications µ and s are related by the affine condition
A(X, Y, Z) = µ(µ(X, Y ), Z) + µ(s(Y, Z), X) − µ(s(Z, X), Y ) + 2s(µ(X, Y ), Z)
−s(µ(Y, Z), X) + s(µ(X, Z), Y ) − s(s(Y, Z), X) + s(s(X, Z), Y ) = 0.
We have also
A(X, Y, Z) + A(Y, Z, X) + A(Z, X, Y ) = 2(µ(s(X, Y ), Z) + µ(s(Y, Z), X) + µ(s(Z, X), Y )).
3. Filiform Lie algebras of dimension 9
3.1. The variety F il9. Using a similar approach as in dimension 8, we obtain
16
ELISABETH REMM
Proposition 21. Any 9-dimensional filiform Lie algebra over K is given in a Vergne basis
by
(14)
µ(X0, Xi) = Xi+1, 1 ≤ i ≤ 7,
µ(X1, X6) = a2X8,
µ(X2, X5) = a4X8,
µ(X1, X5) = (a2 + a4)X7 + a5X8,
µ(X3, X4) = a6X8
µ(X2, X4) = (a4 + a6)X7 + a7X8,
µ(X1, X4) = (a2 + 2a4 + a6)X6 + (a5 + a7)X7 + a8X8,
µ(X2, X3) = (a4 + a6)X6 + a7X7 + a9X8,
µ(X1, X3) = (a2 + 3a4 + 2a6)X5 + (a5 + 2a7)X6 + (a8 + a9)X7 + a10X8,
µ(X1, X2) = (a2 + 3a4 + 2a6)X4 + (a5 + 2a7)X5 + (a8 + a9)X6 + a10X7 + a11X8,
4 + 2a2
6 + 2a2a6 + a4a6 = 0.
with −3a2
We denote by F il9 the set of Lie algebras described above. It is clear that F il9, the open
set of 9-dimensional filiform Lie algebras, is the orbit of F il9 in N ilp9 associated with the
action of the linear group GL(9, K). This reduces the study of F il9 to F il9. Let V 9 be the
9-dimensional K-vector space characterized by the structure constants {ai}2≤i≤11, i 6= 3.
Proposition 22. F il9 is a 8-dimensional irreducible algebraic subvariety of V 9.
This algebraic variety F il9 has only one singular point corresponding to ai = 0 for all i.
In all the other points, the tangent space Tµ(F il9) to F il9 is identified to the vector space
of 2-cocycles:
(15)
ϕ(X0, Xi) = 0, 1 ≤ i ≤ 7,
ϕ(X1, X6) = u2X8,
ϕ(X2, X5) = u4X8,
ϕ(X1, X5) = (u2 + u4)X7 + u5X8,
ϕ(X3, X4) = u6X8,
ϕ(X2, X4) = (u4 + u6)X7 + u7X8,
ϕ(X1, X4) = (u2 + 2u4 + u6)X6 + (u5 + u7)X7 + u8X8,
ϕ(X2, X3) = (u4 + u6)X6 + u7X7 + u9X8,
ϕ(X1, X3) = (u2 + 3u4 + 2u6)X5 + (u5 + 2u7)X6 + (u8 + u9)X7 + u10X8,
ϕ(X1, X2) = (u2 + 3u4 + 2u6)X4 + (u5 + 2u7)X5 + (u8 + u9)X6 + u10X7 + u11X8,
2a6 − 3a4) + u6(2a6 + a2 + 1
with u2a6 + u4( 1
2a4) = 0. Let f ∈ gl(9, K) be an endomorphism.
We put f (X0) = P αiXi and f (X1) = P βiXi. Assume that δf (X0, Xi) = 0. Then
f (Xi+1) = µ(f (X0), Xi) + µ(X0, f (Xi)) for i = 1,· · · , 7. The other components of δf are
4 + 6a2a4 + 5a4a6),
2 + 9a2
6 + 3a2a4 + 7a2a6 + 11a4a6),
v2 = a2(β1 − 2α0), v4 = a4(β1 − 2α0), v6 = a6(β1 − 2α0),
v5 = a5(β1 − 3α0) − α1(2a2
v7 = a7(β1 − 3α0) − α1(7a2
v8 = a8(β1 − 4α0) − α1((5a2 + 11a4 + 5a6)a5 + (6a2 + 19a4 + 10a6)a7) − 2β3a4,
v9 = a9(β1 − 4α0) − α1((3a4 + 4a6)a5 + (4a2 + 9a4 + 8a6)a7) − 2β3a6,
3 (a5, a7)) + α3P 2
v10 = a10(β1 − 5α0) − α1(P 1
4 (a2, a4, a6)
−3β4(a4 + a6) − 2β3a7,
v11 = a11(β1 − 6α0) − α1P 2
α4P 2
5 (a2, a4, a5, a6, a7, a8, a9, a10) + α3P 2
7 (a2, a4, a6) − 2β3a9 − 3β4a7 − 2β5(2a4 + a6),
1 (a8, a9)P 1
2 (a2, a4, a6) + P 2
6 (a2, a4, a5, a6, a7)
ON FILIFORM LIE ALGEBRAS
17
CR the restricted
CR(µ, µ) ≥ 1 for any
i are an homogeneous polynomials of degree k. If we denote by H ∗
where P k
Chevalley cohomology of Lie algebras belonging to F il9, we have dim H 2
µ ∈ F il9. We deduce
Proposition 23. None of the 9-dimensional filiform K-Lie algebras is rigid in F il9 and also
in N ilp9, and also in Lie9.
Proof. In fact for any µ ∈ F il9, dim H 2CR(µ, µ) 6= 0. Since the affine schema associated
with F il9 is reduced, any rigid Lie algebra in this variety has a trivial cohomology. Then no
µ ∈ F il9 is rigid.
Now we determinate the Lie algebras µ such that dim H 2
sume that a6(a4 + a6)(2a4 + a6) 6= 0. Then we can find δf such that
CR(µ, µ) is the smallest one. As-
u2 + v2 = u5 + v5 = u9 + v9 = u10 + v10 = u11 + v11 = 0.
In this case, the parameters α0, α1, β3, β4, β5 are fixed and the others relations ui + vi cannot
be reduced to 0. Then there exists a represent ant ϕ in the cohomological classe with
u2 = u5 = u9 = u10 = u11 = 0. Since u2a6 + u4( 1
2a4) = 0, for a
such Lie algebra we have dim H 2
Theorem 24. The variety F il9 is the closure of the orbit of a rigid 2-parameters family.
Let us consider the family T 9
1, a6 = 1, a8 = u, a9 = 0, a10 = 0, a11 = 0) with t 6= 0,−1,− 1
theorem.
, a4 = t, a5 =
2. This family answers to this
t of Lie algebras µ defined by (a2 = 3t2−t−2
CR(µ, µ) = 2 and it is the lower bound. We deduce
2 a6 − 3a4) + u6(2a6 + a2 + 1
2
3.2. Contact 9-dimensional filiform Lie algebras. Let g be a 9-dimensional filiform Lie
algebra. Let {ω0,· · · , ω8} the dual basis of {X0,· · · , X8}. From Proposition 4, g is a contact
Lie algebra if and only if ω8 is a contact form. This is equivalent to
where ai are the constant structures of g described in (14). We deduce:
a2a4a6 6= 0
Proposition 25. A 9-dimensional filiform Lie algebra admit a contact form if and only it is
isomorphic to a Lie algebra of F il9 whose structure constant given in (14) satisfy a2a4a6 6= 0.
3.3. Come back on 8-dimensional symplectic filiform Lie algebras. The previous
theorem is the result announced in Proposition 18. As we have say, the determination of
contact 9-dimensional filiform Lie algebras permits a quickly determination of the class of
symplectic filiform 8-dimensional Lie algebra.
From Proposition 25, we can highlight a model of 9-dimensional filiform contact Lie algebra,
that is a Lie algebra such as any 9-dimensional filiform contact Lie algebra is isomorphic to
a deformation of this model. We consider the Lie algebras ga2,a4,a6 given by
a5 = a7 = a8 = a9 = a10 = a11 = 0
and
4 + 2a2
−3a2
(a4 − a6)(3a4 + 2a6)
2a6
Since a6 6= 0, a2 =
a4a6(a4 − a6)(3a4 + 2a6) 6= 0.
6 + 2a2a6 + a4a6 = 0, a2a4a6 6= 0.
and the condition a2a4a6 6= 0 is equivalent to
18
ELISABETH REMM
Proposition 26. Any 9-dimensional filiform contact Lie algebra is isomorphic to a linear
deformation of a Lie algebra ga2,a4,a6.
Corollary 27. Any 8-dimensional symplectic filiform Lie algebra is isomorphic to a linear
symplectic deformation of a Lie algebra of the following family
µ(X0, Xi) = Xi+1, 1 ≤ i ≤ 6,
µ(X1, X5) = b2X7
µ(X2, X4) = b4X7,
µ(X1, X4) = (b2 + b4)X6,
µ(X2, X3) = b4X6,
µ(X1, X3) = (b2 + 2b4)X5,
µ(X1, X2) = (b2 + 2b4)X4,
with b2 = b4 = 0 or b4(b2 + b4)(2b2 − b4)(b2 + 2b4) 6= 0.
3.4. The varieties F il10 and F il11.
Proposition 28. Any 10-dimensional filiform Lie algebra over K is given in a Vergne basis
by
(16)
µ(X0, Xi) = Xi+1, 1 ≤ i ≤ 8,
µ(X2, X7) = a1X9,
µ(X1, X7) = a1X8 + a2X9,
µ(X3, X6) = −a1X9,
µ(X2, X6) = a4X9,
µ(X1, X6) = a1X7 + (a2 + a4)X8 + a5X9,
µ(X4, X5) = a1X9,
µ(X3, X5) = a7X9,
µ(X2, X5) = (a4 + a7)X8 + a8X9,
µ(X1, X5) = a1X6 + (a2 + 2a4 + a7)X7 + (a5 + a8)X8 + a9X9,
µ(X3, X4) = a7X8 + a10X9
µ(X2, X4) = (a4 + 2a7)X7 + (a8 + a10)X8 + a11X9,
µ(X1, X4) = a1X5 + (a2 + 3a4 + 3a7)X6 + (a5 + 2a8 + a10)X7 + (a9 + a11)X8 + a12X9,
µ(X2, X3) = (a4 + 2a7)X6 + (a8 + a10)X7 + a11X8 + a13X9,
µ(X1, X3) = a1X4 + (a2 + 4a4 + 5a7)X5 + (a5 + 3a8 + 2a10)X6 + (a9 + 2a11)X7+
(a12 + a13)X8 + a14X9,
µ(X1, X2) = a1X3 + (a2 + 4a4 + 5a7)X4 + (a5 + 3a8 + 2a10)X5 + (a9 + 2a11)X6+
(a12 + a13)X7 + a14X8 + a15X9,
with the conditions
a1(2a2 + 7a4 + 7a7) = 0,
3a2
4 + 3a4a7 − 2a2a7 = 0,
a1(2a9 + 5a11) − 2a2a10 + a4(7a8 − 2a10) + a7(−3a5 + 2a8 − 7a10) = 0.
We denote the set of this multiplications by F il10. If 2a2 + 7a4 + 7a7 = 0, then
3a2
4 + 3a4a7 − 2a2a7 = 3a2
4 + 10a4a7 + 7a2
7 = (a4 + a7)(3a4 + 7a7).
We deduce
ON FILIFORM LIE ALGEBRAS
19
Proposition 29. The set F il10 of 10-dimensional filiform Lie algebras is the union of the
algebraic components
(1) F il10(1) = O(F il10(1)) where F il10(1) is the set of multiplication µ ∈ F il10 satisfying
a1 = 0, 3a2
(2) F il10(2) = O(F il10(2)) where F il10(2) is the set of multiplication µ ∈ F il10 satisfying
4 + 3a4a7 − 2a2a7 = 0,−2a2a10 + a4(7a8 − 2a10) + a7(−3a5 + 2a8 − 7a10) = 0.
a1, a2 = 0, a4 = −a7, a1(2a9 + 5a11) + a4(3a5 + 5a8 + 5a10) = 0,
(3) F il10(3) = O(F il10(3)) where F il10(3) is the set of multiplication µ ∈ F il10 satisfying
a2 = −2a4, 3a4 = −7a7, a1(2a9 + 5a11) + a4(cid:18) 9
7
a5 +
43
7
a8 + 5a10(cid:19) = 0.
Proposition 30. Any 11-dimensional filiform Lie algebra over K is given in a Vergne basis
by
(17)
µ(X0, Xi) = Xi+1, 1 ≤ i ≤ 9,
µ(X1, X8) = a2X10,
µ(X2, X7) = a4X10,
µ(X1, X7) = (a2 + a4)X9 + a5X10,
µ(X3, X6) = a7X10,
µ(X2, X6) = (a4 + a7)X9 + a8X10,
µ(X1, X6) = (a2 + 2a4 + a7)X8 + (a5 + a8)X9 + a9X10,
µ(X4, X5) = a10X10,
µ(X3, X5) = (a7 + a10)X9 + a11X10,
µ(X2, X5) = (a4 + 2a7 + a10)X8 + (a8 + a11)X9 + a12X10,
µ(X1, X5) = (a2 + 3a4 + 3a7 + a10)X7 + (a5 + 2a8 + a11)X8 + (a9 + a12)X9 + a13X10,
µ(X3, X4) = (a7 + a10)X8 + a11X9 + a14X10,
µ(X2, X4) = (a4 + 3a7 + 2a10)X7 + (a8 + 2a11)X8 + (a12 + a14)X9 + a15X10,
µ(X1, X4) = (a2 + 4a4 + 6a7 + 3a10)X6 + (a5 + 3a8 + 3a11)X7 + (a9 + 2a12 + a14)X8
+(a13 + a15)X9 + a16X10,
µ(X2, X3) = (a4 + 3a7 + 2a10)X6 + (a8 + 2a11)X7 + (a12 + a14)X8 + a15X9 + a17X10,
µ(X1, X3) = (a2 + 5a4 + 9a7 + 5a10)X5 + (a5 + 4a8 + 5a11)X6 + (a9 + 3a12 + 2a14)X7
+(a13 + 2a15)X8 + (a16 + a17)X9 + a18X10,
µ(X1, X2) = (a2 + 5a4 + 9a7 + 5a10)X4 + (a5 + 4a8 + 5a11)X5 + (a9 + 3a12 + 2a14)X6
+(a13 + 2a15)X7 + (a16 + a17)X8 + a18X9 + a19X10,
with the conditions
2 + 3z2z3 − 2z1z3 = 0,
(J1) : 3z2
(J2) : z7(2z1 + 2z2 + z3) + z3(3z5 + z6) − 7z2z6 = 0,
(J3) : z4(2z1 + 7(z2 + z3)) − 2z3(2z2 + z3) = 0,
(J4) : z4(2z8 + 5z9) − z10(2z1 + 9z2 + 12z3) − z7(3z5 + 7z6 − z7) + 4z2
−2z3(2z8 + 7z9) + 8z9(z2 + 2z3) = 0.
6
where z1 = a2 + a4, z2 = a4 + a7, z3 = a7 + a10, z4 = a10, z5 = a5 + a11, z6 = a8 + a11, z7 =
a11, z8 = a9 + a12, z9 = a12 + a14, z10 = a14.
20
ELISABETH REMM
This system determines an irreducible component in F il11. This component is the Zariski
closure of the open set whose elements are the filiform contact Lie algebra.
3.5. Contact and symplectic structures. Let g be a 11-dimensional filiform Lie algebra
belonging to F il11. Let {ω1,· · · , ω10} the dual basis of the Vergne basis of g. Assume that
g is a contact Lie algebra. Then, from [16], the form ω10 is also a contact form. Then g is a
contact algebra if and only if ω10 is a contact form in g1. We deduce
Proposition 31. A 11-dimensional filiform Lie algebra. is a contact Lie algebra if and only
if it is isomorphic to a Lie algebra of F il11 with a2a4a7a10 6= 0.
We deduce that a model of contact 11-dimensional filiform Lie algebra is given by the
family ga2,a4,a7,a10 of Lie algebras of F il11 corresponding to
a5 = a8 = a9 = a11 = a12 = a13 = a14 = a15 = a16 = a17 = a18 = a19 = 0
with the conditions,
3z2
z4(2z1 + 7z2 + 7z3) − 2z3(2z2 + z3) = 0,
a2a4a7a10 = z4(z3 − z4)(z2 − z3 + z4)(z1 − z2 + z3 − z4) 6= 0.
2 + 3z2z3 − 2z1z3 = 0,
and z1 = a2 + a4, z2 = a4 + a7, z3 = a4 + a10, z4 = a10. Let us note that the algebraic
variety determined by the two first equations is not reduced to 0, for example the point
(a2, a4, a7, a10) = (1,−1, 1,−1) belongs to this algebraic set. Let us note also that the linear
form ω10 where {ωi} is the dual basis of {Xi} satisfies
dω10 = −ω0 ∧ ω9 − a2ω1 ∧ ω8 − a4ω2 ∧ ω7 − a7ω3 ∧ ω6 − a10ω4 ∧ ω5
and is a contact form.
Let A be the open set of 11-dimensional filiform contact Lie algebra. Then this open set is
the orbit of the family of Lie algebras ga2,a4,a7,a10 and satisfying a2a4a7a10 6= 0. Moreover
Proposition 32. None of the 11-dimensional filiform Lie algebras is rigid.
F il11 = A.
Proof. The proof is similar to the 9-dimensional case. The affine scheme being reduced, it is
sufficient to prove that the dimension of the second space of cohomology H 2
CR(µ, µ) of any
Lie algebras of (17) is not zero. It is clear that, if µ is a Lie algebra of (17), the dimension
of the 2-cocycles ψ such that µ + tψ belongs to 17 is of dimension 16 parametrized by the
1 βiXi,
δf (X1, X8) = v2X10, δf (X2, X7) = v4X10, δf (X3, X6) = v7X10, δf (X4, X5) = v10X10, we
have
ai. If f is an endomorphism of K11, then putting f (e0) = P10
0 αiXi and f (e1) = P10
v2 = a2(β1 − 2α0)
v4 = a4(β1 − 2α0)
v7 = a7(β1 − 2α0)
v10 = a10(β1 − 2α0)
We deduce that the dimension of the space of deformations is greater or equal to 3.
Consequence: Determination of the symplectic 10-dimensional filiform Lie alge-
bras. From the Proposition 4 we deduce:
Proposition 33. Any 10-dimensional symplectic filiform Lie algebra is isomorphic to
ON FILIFORM LIE ALGEBRAS
21
µ(X0, Xi) = Xi+1, 1 ≤ i ≤ 8,
µ(X1, X7) = (a2 + a4)X9,
µ(X2, X6) = (a4 + a7)X9,
µ(X1, X6) = (a2 + 2a4 + a7)X8 + (a5 + a8)X9,
µ(X3, X5) = (a7 + a10)X9,
µ(X2, X5) = (a4 + 2a7 + a10)X8 + (a8 + a11)X9,
µ(X1, X5) = (a2 + 3a4 + 3a7 + a10)X7 + (a5 + 2a8 + a11)X8 + (a9 + a12)X9,
µ(X3, X4) = (a7 + a10)X8 + a11X9,
µ(X2, X4) = (a4 + 3a7 + 2a10)X7 + (a8 + a11)X8 + (a12 + a14)X9,
µ(X1, X4) = (a2 + 4a4 + 6a7 + 3a10)X6 + (a5 + 3a8 + a11)X7 + (a9 + a12 + a14)X8
+(a13 + a15)X9,
µ(X2, X3) = (a4 + 3a7 + 2a10)X6 + (a8 + 2a11)X7 + (a12 + a14)X8 + a15X9,
µ(X1, X3) = (a2 + 5a4 + 9a7 + 5a10)X5 + (a5 + 4a8 + 5a11)X6 + (a9 + 3a12 + 2a14)X7
+(a13 + 2a15)X8 + (a16 + a17)X9,
µ(X1, X2) = (a2 + 5a4 + 9a7 + 5a10)X4 + (a5 + 4a8 + 5a11)X5 + (a9 + 3a12 + 2a14)X6
+(a13 + 2a15)X7 + (a16 + a17)X8 + a18X9
with a2a4a7a10 6= 0 and if (z1 = a2 +a4, z2 = a4 +a7, z3 = a7 +a10, z4 = a10, z5 = a5 +a11, z6 =
a8 + a11, z7 = a11,
2 + 3z2z3 − 2z1z3 = 0,
(cid:26) 3z2
z7(2z1 + 2z2 + z3) + z3(3z5 + z6) − 7z2z6 = 0.
We deduce, from the definition of a symplectic model:
Corollary 34. The symplectic models of 10-dimensional filiform symplectic Lie algebras are
the Lie algebra corresponding to
a5 = a8 = a11 = a9 = a12 = a14 = a12 = a14 = a15 = a16 = a17 = a18 = 0.
4. Contact and symplectic filiform Lie algebras
4.1. (2p+1)-dimensional contact filiform Lie algebras. Let {X0,· · · , X2n} be a Vergne
basis of a (2p + 1)-dimensional filiform Lie algebra g and let us denote by C k
i,j the constant
structures related to this basis. We have seen in dimension 11 or smaller, but this remains
trivially true for greater dimension, that the structure constants of g related to this basis are
linear combinations of the (p − 1)2 structure constants ai,j = C 2p
i,j for 1 ≤ i < j ≤ 2p − 1 − i.
Using the same notation as the previous section, we deduce that F il2p+1 is an algebrac
variety embedded in K(p−1)2
. In fact, all the other structure constants are defined by the
linear equation
[X0, [Xi, Xj]] = [Xi+1, Xj] + [Xi, Xj+1]
as soon as j + 1 ≤ 2p. More precisely, we have
C 2p−k
2p−1−j−k,j = Pj αja2p−1−j,j, 2j > 2p − 1 − k, k = 1,· · · , 2p − 4
C 2p−k
2p−2−j−k,j = Pj βja2p−2−j,j, 2j > 2p − 2 − k, k = 1,· · · , 2p − 5
· · ·
C 2p−k
2p−(2p−3)−j−k,j = C 2p−k
3−j−k,j = a1,2,
22
ELISABETH REMM
The coefficients αj, βj are described in [9]. Let {ω0, ω1,· · · , ω2p} be the dual basis. All the
Jacobi conditions are given writing that d(dω2p) = 0. This gives
(p − 3)2 + (p − 4)(p − 5) + (p − 6)2 + · · · + ε
(18)
equations where ε = 2 if p ≡ 0 (mod 3), ε = 1 if p ≡ 1 (mod 3) and ε = 22 if p ≡ 2 (mod 3).
This shows that, as soon as the dimension exceeds 19 the number of polynomial equations
is greater than the number of parameters ai,j.
If g admits a contact form, then from [16] the form ω2p is also a contact form. We deduce
Proposition 35. A (2p + 1)-dimensional filiform Lie algebra admits a contact form if and
only if the structure constants related to the Vergne basis satisfyy:
a1,2p−2 · a2,2p−3 · · · ai,2p−1−i · · · ap−1,p 6= 0.
Since any deformation of a contact Lie algebra is also a contact Lie algebra, we deduce that
the set of (2p+1)-dimensional filiform contact Lie algebra is a Zariski open set in F il2p+1. Let
us consider the family A contained in this open set and corresponding to Lie algebras whose
structure constants satisfy ai,j = 0 except a1,2p−2, a2,2p−3,· · · , ai,2p−1−i,· · · , ap−1,p which are
supposed different to 0. It is clear that any contact filiform (2p+1)-dimensional Lie algebra is
a deformation of a Lie algebra of this family. We remark that this family is parametrized by
the (p − 1) constant structures a1,2p−2, a2,2p−3, ai,2p−1−i, ap−1,p but the system of polynomial
equations deduced from the Jacobi conditions, which is a consequence of d(dω2p) = 0 is
composed, when p is greater than 7, of a number of equations greater than p − 1. This
number depend to p mod(3). For example, if p = 3k + 1, we have 3k parameters and
3k2 − 3k + 1 polynomial equations and 3k2 − 3k + 1 > 3k as soon as k ≥ 2.
Let us note also that A is not reduced to 0. In fact the Lie algebra corresponding to
(19)
a1,2p−2 = 1, a2,2p−3 = −1,· · · , ai,2p−1−i = (−1)i+1,· · · , ap−1,p = (−1)p
belongs to this family and more generally, if
a1,2p−2 = λ, a2,2p−3 = −λ,· · · , ai,2p−1−i = (−1)i+1λ,· · · , ap−1,p = (−1)pλ
then the corresponding Lie algebras are in A. This implies that dimA ≥ 1. Let us denote by
g0 the Lie algebra of A corresponding to (19) and let us compute the space of deformations.
As we have seen in previous work ([13]), it is sufficient to compute the cocycles of g0 which
preserves the Vergne's basis. For example, in case of dimension 9, from (14), the space of
such cocycles is of dimension 8. Let us compute the space of coboundaries. It is generated
by the δf satisfying δf (X0, Y ) = 0 and f ∈ gl(9, K). This last condition implies that f is
determined when we know f (X0) = α0X0 + · · · + α8X8 and f (X1) = β1X1 + · · · + β8X8. We
obtain
and
δf (X1, X6) = (−2α0 + β1)X8 = −δf (X2, X5) = δf (X3X4)
δf (X1, X4) = −δf (X2, X3) = −2β3X8, δf (X1, X2) = −2β5X8,
if not δf (Xi, Xj) = 0. We deduce that the space of deformations is of dimension 4.
In
dimension 2p, the space of cocycles which preserves the Vergne's basis is embedded in a
vector space of dimension (p − 1)2 parametrized by the structure constants
{a1,2p−2,· · · , a1,2, a2,p−3,· · · , a2,3,· · · , ap−2,p, ap−2,p−1, ap−1,p}
ON FILIFORM LIE ALGEBRAS
23
that is ai,j with 1 ≤ i < j ≤ 2p − 2, 3 ≤ i + j ≤ 2p − 1 and with ai,j = C 2p
i,j. If f is a
linear endomorphism of gl(2p, K), then δf (X0, Xi) = 0 implies that f (Xi) is determined for
2 ≤ i ≤ 2p by f (X0) = α0X0 + · · · + α2pX2p and f (X1) = β1X1 + · · · + β2pX2p. This implies
δf (X1, X2p−2) = (−2α0 + β1)X2p = −δf (X2, X2p−3) = · · · = (−1)pδf (Xp−1Xp),
and the other non zero δf (Xi, Xj) are
δf (X1, X2i) = 2β2p−2i−1X2p, 1 ≤ i ≤ p − 2
δf (X2, X2i+1) = −2β2p−2i−3X2p, 1 ≤ i ≤ p − 3
δf (X3, X2i) = 2β2p−2i−5X2p, 2 ≤ i ≤ p − 2
· · ·δf (Xp−2, Xp−1) = β3X2p.
We remark also than the parameters a1,2, a1,3, a2,3, a1,4, a2,4, a1,5 do not appear in the poly-
nomial Jacobi equations because the forms ωi∧ ωj which appear in the expression of dω2p are
closed for (i, j) ∈ {(1, 2), (1, 3), (2, 3), (1, 4), (2, 4), (1, 5)}. From the previous computations of
the coboundaries, we have in particular
(cid:26) δf (X1, X2) = 2β2p−3X2p, δf (X1, X4) = 2β2p−5X2p, δf (X1, X3) = δf (X1, X5) = 0,
δf (X2, X3) = 2β2p−5X2p, δf (X2, X4) = 0.
Then we can consider that the parameters a1,2, a1,4 are orbital parameters and a1,3, a2,3, a2,4,
a1,5 are parameters of non trivial deformations. To end this work, we compute the space of
deformation of g0. It remains to compute the dimension of the space of cocycles. We have
seen that it is embedded in a vector space of dimension (p−1)2 and the number of polynomial
equations given by the Jacobi relations was (18). But the affine scheme associated to this
polynomial equations is not reduced. We can find relations between these equations from
the following remark which we illustrate in dimension 11.
In this dimension the Jacobi
polynomial equations is constituted of 4 equations corresponding to J(Xi, Xj, Xk) = 0 (J
for the Jacobi condition related to triple (Xi, Xj, Xk). To simplify we denote by (i, j, k)
the polynomial J(Xi, Xj, Xk) and let p = i + j + k be its weight.
In this case we have
4 parameters (a1,8 = a2, a2,7 = a4, a3,6 = a7, a4,5 = a10) and 4 equation corresponding to
(p = 6, (i, j, k) = (1, 2, 3)), (p = 7, (i, j, k) = (1, 2, 4)), (p = 8, (i, j, k) = (1, 2, 5), (1, 3, 4)).
But we have
[X0, (1, 2, 3)] = (1, 2, 4)
[X0, (1, 2, 4)] = (1, 3, 4) + (1, 2, 5).
Thus the system of Jacobi equations can be reduced to the system
(1, 2, 3) = 0, (1, 3, 4) = 0
and the corresponding affine scheme is reduced. Then we have 4 parameters which have to
satisfy 2 independent relations. The space of parameters is then of dimension 2. Let us come
back to the general model g0. The linear space of parameters is of dimension (p − 1) and it
is generated by the structure constants
a1,2p−2, a2,2p−3,· · · , ap−2,p+1, ap−1,p.
The weights take its values in (6, 7,· · · , 2p − 2) and concern the Jacobi equation:
(1, 2, 3); (1, 2, 4); (1, 2, 5), (1, 3, 4);··· ; (1, p − 2, p − 1), (1, p − 3, p),· · · , (1, 2, 2p − 5),
(2, p − 1, p + 1),· · · , (2, 3, 2p − 7),· · · ,}
24
ELISABETH REMM
the last term in this ordered sequence depends to (p mod3), more precisely, if 2p − 2 ≡ 2
(mod 3), then the last term is (k − 1, k + 1, k + 2) with 2p − 2 = 3k + 2, if 2p − 2 ≡ 1
(mod 3), then the last term is (k − 1, k, k + 2), and if 2p− 2 ≡ 0 (mod 3), then the last term
is (k − 1, k, k + 2). We have seen (18) that the number of Jacobi polynomial equations is
(p − 3)2 + (p − 4)(p − 5) + (p − 6)2 + · · · + ε
where ε = 2 if p ≡ 0 (mod 3), ε = 1 if p ≡ 1 (mod 3) and ε = 22 if p ≡ 2 (mod 3). This
scheme is not reduced. To reduce it we consider the relations
[X0, (i, j, k)] = (i + 1, j, k) + (i, j + 1, k) + (i, j, k + 1).
Putting 2p − 2 = 3m + r with 0 ≤ r ≤ 2, we can write the reduced number Nr of equations:
• If m = 2h and r = 0, then Nr = 3h2 − 3h + 1,
• If m = 2h + 1 and r = 1, then Nr = 3h2 + 3h,
• If m = 2h and r = 2, then Nr = 3h2 − h.
We can see than, as soon as n ≥ 14, that is p = 7, m = 4, r = 2 then the number of
parameters is 6 and Nr = 10. In the same way, we can reduce this new polynomial system
using the identity
[X1, (i, j, k)] = ((i + 2, j, k)C i+2
1,i + (i, j + 2, k)C j+2
1,j + (i, j, k + 2)C k+2
1,k )Xi+j+k+2.
which is a direct consequence of the natural grading of g0. We deduce in particular
[X1, (1, 2, 3)] = (−(1, 3, 4)C 4
1,2 + (1, 2, 5)C 5
1,3)X10.
To end this section, we can look the case n = 14, this case corresponding to Nr > p − 1.
We have 6 coefficients and 7 relations after the reduction of the first type. We can choose
as generating relations, the relation (1, 2, i) for i = 3, 5, 6, 7, 8, 9 and (3, 4, 5). The relation of
second type concerning these equations are, where (1, (i, j, k) is the coefficient of [X1, (i, j, k)],
1(1, 2, 3) = (1, 2, 5)C 5
1(1, 2, 4) = (1, 2, 6)C 6
1(1, 2, 5) = (1, 4, 5)C 4
1(1, 2, 6) = (1, 4, 6)C 4
1(1, 2, 7) = (1, 4, 7)C 4
1(2, 3, 5) = −(3, 4, 5)C 4
1,3,
1,4,
1,2 + (1, 2, 7)C 7
1,2 + (1, 2, 8)C 8
1,2 + (1, 2, 9)C 9
1,5,
1,6,
1,7,
1,2 + (2, 3, 7)C 7
1,5,
1,4C 7
1,3C 6
1,2C 5
If C 4
1,7 6= 0, then the Jacobi polynomial system is reduced only to one
equation (1, 2, 3). In this open set, the space of parameters of deformations of the models is
of dimension greater or equal to 5.
1,5C 8
1,6C 9
4.2. Filiform symplectic algebras. From the previous study, we have that the (2p)-
dimensional symplectic filiform Lie algebras are isomorphic to a quotient of a contact (2p+1)-
dimensional filiform Lie algebra g2p+1 by its center K{X2p}. Then its can be written with
the structure constants of g2p+1 with the condition a1,2p−2a2,2p−4 · · · ap−1,p 6= 0.
ON FILIFORM LIE ALGEBRAS
25
References
[1] Ancochea-Berm´udez, Jose Maria; G´omez-Martin, Jose Ramon; Valeiras, Gerardo; Goze, Michel. Sur les
composantes irr´eductibles de la vari´et´e des lois d'alg`ebres de Lie nilpotentes. J. Pure Appl. Algebra 106
(1996), no. 1, 11-22.
[2] Ancochea-Berm´udez, Jose Maria; Goze, Michel. Sur la classification des alg`ebres de Lie nilpotentes de
dimension 7. C. R. Acad. Sci. Paris 302 (1986), 611–613.
[3] Ancochea-Berm´udez, Jose Maria; Goze, Michel. Classification des alg`ebres de Lie filiformes de dimension
8. Arch. Math. (Basel) 50 (1988), no. 6, 511-525
[4] Ancochea, Jose Maria; Campoamor Stursberg, Rutwig. Characteristically nilpotent Lie algebras: a
survey. Extracta Math. 16 (2001), no. 2, 153–210.
[5] Bahturin, Yuri; Goze, Michel; Remm, Elisabeth. Group gradings on filiform Lie algebras. . Comm.
Algebra 44 (2016), no. 1, 40-62.
[6] Bouyakoub, Abdelkader; Goze, Michel. Sur les alg`ebres de Lie munies d'une forme symplectique.
(French) [On Lie algebras equipped with a symplectic form] Rend. Sem. Fac. Sci. Univ. Cagliari. 57
(1987), no. 1, 85-97.
[7] Burde, Dietrich. Left-symmetric algebras, or pre-Lie algebras in geometry and physics. Cent. Eur. J.
Math. 4 (2006), no. 3, 323-357 (electronic).
[8] G´omez, Jose Ramon; Jimen´ez-Merch´an, Antonio; Khakimdjanov, Yusupjan. Low-dimensional filiform
Lie algebras. J. Pure Appl. Algebra 130 (1998), no. 2, 133-158.
[9] Goze, Michel; Khakimdjanov, Yusupjan. Nilpotent and solvable Lie algebras. Handbook of algebra, Vol.
2, 615-663, North-Holland, Amsterdam, 2000.
[10] Goze, Michel; Khakimdjanov, Yusupjan. Some nilpotent Lie algebras and its applications. Algebra and
operator theory (Tashkent, 1997), 49-64, Kluwer Acad. Publ., Dordrecht, 1998.
[11] Goze, Michel; Khakimdjanov, Yusupjan. Sur les alg`ebres de Lie nilpotentes admettant un tore de
d´erivations. Manuscripta Math. 84 (1994), no. 2, 115-124.
[12] Goze, Michel; Remm, Elisabeth. Alg`ebres de Lie r´eelles ou complexes. Preprint, Universit´e de Haute
Alsace. 2012 (www.livresmathematiques.fr).
[13] Goze, Michel; Remm, Elisabeth. k-step nilpotent Lie algebras. Georgian Math. J. 22 (2015), no. 2,
219-234.
[14] Goze, Michel; Remm, Elisabeth. Contact and Frobeniusian forms on Lie groups. Differential Geom.
Appl. 35 (2014), 74-94.
[15] Goze, Michel; Remm, Elisabeth. Non existence of complex structures on filiform Lie algebras. Comm.
Algebra 30 (2002), no. 8, 3777–3788.
[16] Khakimdjanov, Yusupjan; Goze, Michel; Medina, Alberto. Symplectic or contact structures on Lie
groups. Differential Geom. Appl. 21 (2004), no. 1, 41-54.
[17] Remm, Elisabeth; Goze, Michel. Affine structures on abelian Lie groups. Linear Algebra Appl. 360
(2003), 215-230.
[18] Remm, Elisabeth. Vinberg algebras associated to some nilpotent Lie algebras. Non-associative algebra
and its applications, 347-364, Lect. Notes Pure Appl. Math., 246, Chapman - Hall/CRC, Boca Raton,
FL, 2006.
[19] Remm, Elisabeth. Breadth and characteristic sequence of nilpotent Lie algebras. Comm. Algebra 45
(2017), no. 7, 2956-2966.
[20] Vergne, Mich`ele. Cohomologie des alg`ebres de Lie nilpotentes. Application l'´etude de la vari´et´e des
alg`ebres de Lie nilpotentes. C. R. Acad. Sci. Paris , Sr. A-B 267 (1968).
Universit´e de Haute Alsace, LMIA, 6 rue des Fr`eres Lumi`ere, 68093 Mulhouse
E-mail address: [email protected]
|
1612.05297 | 1 | 1612 | 2016-12-15T22:42:04 | H-standard cohomology for Courant-Dorfman algebras and Leibniz algebras | [
"math.RA"
] | We introduce the notion of H-standard cohomology for Courant-Dorfman algebras and Leibniz algebras, by generalizing Roytenberg's construction. Then we generalize a theorem of Ginot-Grutzmann on transitive Courant algebroids, which was conjectured by Stienon-Xu. The relation between H-standard complexes of a Leibniz algebra and the associated crossed product is also discussed. | math.RA | math | H-STANDARD COHOMOLOGY FOR COURANT-DORFMAN ALGEBRAS
AND LEIBNIZ ALGEBRAS
XIONGWEI CAI
6
1
0
2
Abstract. We introduce the notion of H-standard cohomology for Courant-Dorfman algebras
and Leibniz algebras, by generalizing Roytenberg's construction. Then we generalize a theorem
of Ginot-Grutzmann on transitive Courant algebroids, which was conjectured by Stienon-Xu.
The relation between H-standard complexes of a Leibniz algebra and the associated crossed
product is also discussed.
c
e
D
5
1
]
.
A
R
h
t
a
m
[
1
v
7
9
2
5
0
.
2
1
6
1
:
v
i
X
r
a
1. Introduction
The notion of Leibniz algebras, objects that date back to the work of "D-algebras" by Bloh [2],
is due to Loday [7]. In literature, Leibniz algebras are sometimes also called Loday algebras.
A (left) Leibniz algebra L is a vector space over a field k (k = R or C) equipped with a bracket
◦ : L ⊗ L → L, called the Leibniz bracket, satisfying the (left) Leibniz identity:
x ◦ (y ◦ z) = (x ◦ y) ◦ z + y ◦ (x ◦ z),
∀x, y, z ∈ L.
A concrete example is the omni Lie algebra ol(V ) , gl(V ) ⊕ V , where V is a vector space.
It is first introduced by Weinstein [16] as the linearization of the standard Courant algebroid
T V ∗ ⊕ T ∗V ∗. The Leibniz bracket of ol(V ) is given by:
(A + v) ◦ (B + w) = [A, B] + Aw,
∀A, B ∈ gl(V ), v, w ∈ V.
In [8], Loday and Pirashvili introduced the notions of representations (corepresentations) and
Leibniz homology (cohomology) for Leibniz algebras. They also studied universal enveloping alge-
bras and PBW theorem for Leibniz algebras.
Leibniz algebras can be viewed as a non-commutative analogue of Lie algebras. Some theorems
and properties of Lie algebras are still valid for Leibniz algebras, while some are not. The properties
of Leibniz algebras are under continuous investigation by many authors, we can only mention a
few works here [1, 4, 9, 13, 14].
Leibniz algebras have attracted more interest since the discovery of Courant algebroids, which
can be viewed as the geometric realization of Leibniz algebras in certain sense. Courant algebroids
are important objects in recent studies of Poisson geometry, symplectic geometry and generalized
complex geometry.
The notion of Courant algebroids was first introduced by Liu, Weistein and Xu in [6], as an
answer to an earlier question "what kind of object is the double of a Lie bialgebroid". In their
original definition, a Courant algebroid is defined in terms of a skew-symmetric bracket, now known
as "Courant bracket". In [10], Roytenberg proved that a Courant algebroid can be equivalently
defined in terms of a Leibniz bracket, now known as "Dorfman bracket". And he defined standard
complex and standard cohomology of Courant algebroids in the language of supermanifolds in [12].
In [15], Stienon and Xu defined naive cohomology of Courant algebroids , and conjectured that
Key words and phrases. Courant-Dorfman algebras, Leibniz algebras, H-standard cohomology, crossed product.
1
H-STANDARD COHOMOLOGY FOR COURANT-DORFMAN ALGEBRAS AND LEIBNIZ ALGEBRAS
2
there is an isomorphism between standard and naive cohomology for a transitive Courant algebroid.
Later this conjecture was proved by Ginot and Grutzmann in [5].
In [11], Roytenberg introduced the notion of Courant-Dorfman algebras, as an algebraic analogue
of Courant algebroids. And he defined standard complex and standard cohomology for Courant-
Dorfman algebras. Furthermore, he proved that there is an isomorphism of graded Poisson algebras
between the standard complex of a Courant algebroid E and the associated Courant-Dorfman
algebra E = Γ(E).
The main objective of this article is to develop a similar cohomology theory, the so-called H-
standard cohomology, for Courant-Dorfman algebras as well as Leibniz algebras.
Given a Courant-Dorfman algebra (E, R, h·, ·i, ∂, ◦), and an R-submodule H ⊇ ∂R which is an
isotropic ideal of E, let (V, ∇) be an H-representation of E (a left representation of Leibniz algebra
E such that ∇ is a covariant differential and H acts trivially on V). By generalizing Roytenberg's
construction, we shall define the H-standard complex (C•(E, H, V), d) and H-standard cohomology
H •(E, H, V) (see Theorem 3.3 and Definition 3.5). And when E/H is projective, we shall prove the
following result:
H •(E, H, V) ∼= H •
CE(E/H, V).
Note that we don't require the symmetric bilinear form h·, ·i to be non-degenerate here.
In
particular when E is the space of sections of a transitive Courant algebroid E (over M ), and
H = ρ∗(Ω1(M )), V = C∞(M ), the result above recovers Stienon and Xu's conjecture.
Given a Leibniz algebra L with left center Z, suppose H ⊇ Z is an isotropic ideal of L, and (V, τ )
is an H-representation of L (a left representation of L such that H acts trivially on V ), similarly
we can define the H-standard complex (C•(L, H, V ), d) and H-standard cohomology H •(L, H, V ).
And we have the following result:
H •(L, H, V ) ∼= H •
CE(L/H, V ).
This result can be proved directly, but in this paper we choose a roundabout way. We construct
a Courant-Dorfman algebra structure on L = S•(Z) ⊗ L, and then prove there is an isomorphism
between the H-standard complex of L and the H = S•(Z) ⊗ H-standard complex of L. Finally
based on the result for Courant-Dorfman algebras, we may obtain the result above by inference.
The structure of this paper is organized as follows:
In Section 2, we provide some basic knowledge about Leibniz algebras and Courant-Dorfman
algebras. In Section 3, we give the definition of H-standard complex and cohomology for Courant-
Dorfman algebras and Leibniz algebras.
In Section 4, we prove the isomorphism theorem for
Courant-Dorfman algebras, as a generalization of Stienon and Xu's conjecture. In Section 5, we
associate a Courant-Dorfman algebra structure on L to any Leibniz algebra L, and discuss the
relation between H-standard complexes of them, finally we prove an isomorphism theorem for
Leibniz algebras.
Acknowledgements. This paper is a part of my PhD dissertation, it is funded by the University
of Luxembourg. I would like to thank my advisors, Prof. Martin Schlichenmaier and Prof. Ping
Xu, for their continual encouragement and support. I am particularly grateful to Prof. Zhangju
Liu for instructive discussions and helpful comments during my stay in Peking University.
In this section we list some basic notions and properties about Leibniz algebras and Courant-
Dorfman algebras. For more details we refer to [8, 11].
2. Preliminaries
H-STANDARD COHOMOLOGY FOR COURANT-DORFMAN ALGEBRAS AND LEIBNIZ ALGEBRAS
3
Definition 2.1. A (left) Leibniz algebra is a vector space L over a field k (k = R for our main
interest), endowed with a bilinear map (called Leibniz bracket) ◦ : L ⊗ L → L, which satisfies
(left) Leibniz rule:
e1 ◦ (e2 ◦ e3) = (e1 ◦ e2) ◦ e3 + e2 ◦ (e1 ◦ e3) ∀e1, e2, e3 ∈ L
Example 2.2. Given any Lie algebra g and its representation (V, ρ), the semi-direct product g ⋉ V
with a bilinear operation ◦ defined by
(A + v) ◦ (B + w) , [A, B] + ρ(A)w,
∀A, B ∈ g, v, w ∈ V
forms a Leibniz algebra.
In particular, for any vector space V , gl(V ) ⊕ V is a Leibniz algebra with Leibniz bracket
(A + v) ◦ (B + w) = [A, B] + Aw,
∀A, B ∈ gl(V ), v, w ∈ V.
It is called an omni Lie algebra, and denoted by ol(V ) (see Weinstein [16]).
Definition 2.3. A representation of a Leibniz algebra L is a triple (V, l, r), where V is a vector
space equipped with two linear maps: left action l : L → gl(V ) and right action r : L → gl(V )
satisfying the following equations:
(2.1)
le1◦e2 = [le1 , le2 ], re1◦e2 = [le1 , re2 ], re1 ◦ le2 = −re1 ◦ re2 ,
∀e1, e2 ∈ L,
where the brackets on the right hand side are the commutators in gl(V ).
If V is only equipped with left action l : L → gl(V ) which satisfies le1◦e2 = [le1 , le2], we call
(V, l) a left representation of L.
For (V, l, r)(or (V, l)) a representation (or left representation) of L, we call V an L-module (or
left L-module).
Given a left representation (V, l), there are two standard ways to extend V to an L-module.
One is called symmetric extension, with the right action defined as re = −le; the other is called
antisymmetric extension, with the right action defined as re = 0. In this paper, we always take
the symmetric extension (V, l, −l).
Example 2.4. Denote by Z the left center of L, i.e.
It is easily checked that
Z , {e ∈ Le ◦ e′ = 0, ∀e′ ∈ L}.
e1 ◦ e2 + e2 ◦ e1 ∈ Z,
∀e1, e2 ∈ L
and Z is an ideal of L. Moreover, the Leibniz bracket of L induces a left action ρ of L on Z:
ρ(e)z , e ◦ z
∀e ∈ L, z ∈ Z.
Definition 2.5. Given a Leibniz algebra L and an L-module (V, l, r), the Leibniz cohomology of L
with coefficients in V is the cohomology of the cochain complex C n(L, V ) = Hom(⊗nL, V ) (n ≥ 0)
with the coboundary operator d0 : C n(L, V ) → C n+1(L, V ) given by:
(d0η)(e1, · · · , en+1)
(−1)a+1leaη(e1, · · · , ea, · · · , en+1) + (−1)n+1ren+1η(e1, · · · , en)
=
(2.2)
nX
+ X
a=1
(−1)aη(e1, · · · , ea, · · · , eb, ea ◦ eb, · · · , en+1)
1≤a<b≤n+1
The resulting cohomology is denoted by H •(L; V, l, r), or simply H •(L, V ) if it causes no confu-
sion.
H-STANDARD COHOMOLOGY FOR COURANT-DORFMAN ALGEBRAS AND LEIBNIZ ALGEBRAS
4
As a special type of Leibniz algebras, Courant-Dorfman algebras can be viewed as the alge-
braization of Courant algebroids:
Definition 2.6. A Courant-Dorfman algebra (E, R, h·, ·i, ∂, ◦) consists of the following data:
a commutative algebra R over a field k (k = R for our main interest);
an R-module E;
a symmetric bilinear form h·, ·i : E ⊗R E → R;
a derivation ∂ : R → E;
a Dorfman bracket ◦ : E ⊗ E → E.
These data are required to satisfy the following conditions for any e, e1, e2, e3 ∈ E and f, g ∈ R:
(1). e1 ◦ (f e2) = f (e1 ◦ e2) + he1, ∂f ie2;
(2). he1, ∂(e2, e3)i = he1 ◦ e2, e3i + he2, e1 ◦ e3i
(3). e1 ◦ e2 + e2 ◦ e1 = ∂he1, e2i;
(4). e1 ◦ (e2 ◦ e3) = (e1 ◦ e2) ◦ e3 + e2 ◦ (e1 ◦ e3);
(5). ∂f ◦ e = 0;
(6). h∂f, ∂gi = 0.
Given a Courant-Dorfman algebra E, we can recover the anchor map
ρ : E → X1 = Der(R, R)
from the derivation ∂ by setting:
(2.3)
ρ(e) · f , he, ∂f i.
Let Ω1 be the R-module of Kahler differentials with the universal derivation dR : R → Ω1. By
the universal property of Ω1, there is a unique homomorphism of R-modules ρ∗ : Ω1 → E such
that
(2.4)
ρ∗ is called the coanchor map of E. When the bilinear form of E is non-degenerate, ρ∗ can be
equivalently defined by
∀f ∈ R
ρ∗(dRf ) , ∂f,
∀α ∈ Ω1, e ∈ E,
where h·, ·i on the right handside is the natural pairing of Ω1 and X1.
hρ∗α, ei = hα, ρ(e)i,
In the following of this section, we always assume
e ∈ E, α ∈ Ω1, f ∈ R.
Given a Courant-Dorfman algebra E, denote by C n(E, R) the space of all sequences ω =
2 ]), where ωk is a linear map from (⊗n−2kE) ⊗ (⊙kΩ1) to R, ∀k, satisfying the fol-
(ω0, · · · , ω[ n
lowing conditions:
1). Weak skew-symmetricity in arguments of E.
∀k, ωk is weakly skew-symmetric up to ωk+1:
ωk(e1, · · · ea, ea+1, · · · en−2k; α1, · · · αk) + ωk(e1, · · · ea+1, ea, · · · en−2k; α1, · · · αk)
= −ωk+1(e1, · · · bea, dea+1, · · · en−2k; dRhea, ebi, α1, · · · αk),
2). Weak R-linearity in arguments of E.
∀k, ωk is weakly R-linear up to ωk+1:
ωk(e1, · · · f ea, · · · en−2k; α1, · · · αk)
= f ωk(e1, · · · ea, · · · en−2k; · · · ) + X
(−1)b−ahea, ebiωk+1(e1, · · · bea, · · · beb, · · · en−2k; dRf, · · · ),
b>a
H-STANDARD COHOMOLOGY FOR COURANT-DORFMAN ALGEBRAS AND LEIBNIZ ALGEBRAS
5
3). R-linearity in arguments of Ω1.
∀k, ωk is R-linear in arguments of Ω1:
ωk(e1, · · · en−2k; α1, · · · f αl, · · · αk) = f ωk(e1, · · · en−2k; α1, · · · αl, · · · αk),
Then C•(E, R) = Ln C n(E, R) becomes a cochain complex, with coboundary map d given for
any ω ∈ C n(E, R) by:
a
i
(−1)aωk(· · · bea, · · · beb, ea ◦ eb, · · · ; · · · )
(dω)k(e1, · · · , en+1−2k; α1, · · · , αk)
= X
(−1)a+1ρ(ea)ωk(· · · bea, · · · ; · · · ) + X
+X
ωk−1(ρ∗(αi), e1, · · · , en+1−2k; α1, · · · bαi, · · · αk)
+X
(−1)aωk(· · · , bea, · · · ; · · · , bαi, ιρ(ea)dRαi, · · · ).
a<b
a,i
Definition 2.7. The cochain complex (C•(E, R), d) is called the standard (cochain) complex of
the Courant-Dorfman algebra E, the resulting cohomology is called the standard cohomology of E,
and denoted by H •
st(E).
3. H-Standard cohomology
In this section, we give the definition of H-standard cohomology for Courant-Dorfman alge-
bras and Leibniz algebras respectively. The inspiration comes from Roytenberg's construction of
standard cohomology for Courant-Dorfman algebras [11].
3.1. For Courant-Dorfman algebras. Given a Courant-Dorfman algebra (E, R, h·, ·i, ∂, ◦), sup-
pose H is an R-submodule as well as an isotropic ideal of E containing ∂R, and (V, ∇) is an
H-representation of E, which is defined as follows:
Definition 3.1. An H-trivial representation (or H-representation for short) of a Courant-Dorfman
algebra (E, R, h·, ·i, ∂, ◦) is a pair (V, ∇), where V is an R-module, and ∇ : E → Der(V) is a
homomorphism of Leibniz algebras such that:
∇αv = 0
∇f ev = f ∇ev
∇e(f v) = (ρ(e)f )v + f (∇ev)
∀α ∈ H, e ∈ E, f ∈ R, v ∈ V.
Example 3.2. Since
ρ(α)f = hα, ∂f i = 0,
∀α ∈ H, f ∈ R,
H is actually an R-submodule of kerρ. It is easily checked that (R, ρ) is an H-representation of E.
In the following of this subsection, we always assume
Denote by C n(E, H, V) the space of all sequences
e ∈ E, α ∈ H, f ∈ R.
where the k-linear maps
satisfy the following conditions:
ω = (ω0, · · · , ω[ n
2 ]),
ωk : (⊗n−2kE) ⊗ (⊙kH) → V
H-STANDARD COHOMOLOGY FOR COURANT-DORFMAN ALGEBRAS AND LEIBNIZ ALGEBRAS
6
1). Weak skew-symmetricity in arguments of E.
∀k, ωk is weakly skew-symmetric up to ωk+1:
ωk(e1, · · · ea, ea+1, · · · en−2k; α1, · · · αk) + ωk(e1, · · · ea+1, ea, · · · en−2k; α1, · · · αk)
= −ωk+1(e1, · · · bea, dea+1, · · · en−2k; ∂hea, ebi, α1, · · · αk),
2). Weak R-linearity in arguments of E.
∀k, ωk is weakly R-linear up to ωk+1:
ωk(e1, · · · f ea, · · · en−2k; α1, · · · αk)
= f ωk(e1, · · · ea, · · · en−2k; · · · ) + X
b>a
3). R-linearity in arguments of H.
∀k, ωk is R-linear in arguments of H:
(−1)b−ahea, ebiωk+1(e1, · · · bea, · · · beb, · · · en−2k; ∂f, · · · ),
ωk(e1, · · · en−2k; α1, · · · f αl, · · · αk) = f ωk(e1, · · · en−2k; α1, · · · αl, · · · αk).
Theorem 3.3. C•(E, H, V) , Ln C n(E, H, V) is a cochain complex under the coboundary map
d = d0 + δ + d′, where d0 is the coboundary map (Equation 2.2) corresponding to the Leibniz
cohomology of E with coefficients in (V, ∇, −∇), and δ, d′ are defined for any ω ∈ C n(E, H, V)
respectively by:
(δω)k(e1, · · · en+1−2k; α1, · · · αk) , X
(d′ω)k(e1, · · · en+1−2k; α1, · · · αk) , X
i
a,i
ωk−1(αi, e1, · · · , en+1−2k; · · · bαi, · · · ),
(−1)a+1ωk(· · · bea · · · ; · · · bαi, αi ◦ ea · · · ).
Lemma 3.4. C•(E, H, V) is closed under d = d0 + δ + d′.
Proof. ∀ω ∈ C n(E, H, V), we need to prove that dω ∈ C n+1(E, H, V).
First, we prove the weak skew-symmetricity in arguments of E. We will calculate d0, δ, d′ parts
separately. The calculations are straightforward from the definitions but rather tedious. To save
space, we omit the details, and only list the results of calculations here.
(3.1)
(d0ω)k(e1, · · · ei, ei+1 · · · en+1−2k; α1, · · · αk) + (d0ω)k(e1, · · · ei+1, ei · · · ; · · · )
= −(d0ω)k+1(e1, · · · bei, dei+1, · · · ; ∂hei, ei+1i, · · · ) − ωk(∂hei, ei+1i, e1, · · · , bei, dei+1, · · · ; · · · ),
(δω)k(e1, · · · ei, ei+1 · · · en+1−2k; α1, · · · αk) + (δω)k(e1, · · · ei+1, ei · · · ; · · · )
= −(δω)k+1(e1, · · · , bei, dei+1, · · · ; ∂hei, ei+1i, · · · ) + ωk(∂hei, ei+1i, e1, · · · , bei, dei+1, · · · ; · · · ),
(3.2)
(d′ω)k(e1 · · · ei, ei+1 · · · en+1−2k; α1, · · · αk) + (d′ω)k(e1, · · · ei+1, ei · · · ; · · · )
= −(d′ω)k+1(e1, · · · , bei, dei+1, · · · ; ∂hei, ei+1i, · · · ).
The sum of the three equations above tells:
(dω)k(e1, · · · ei, ei+1 · · · en+1−2k; α1, · · · αk) + (dω)k(e1, · · · ei+1, ei · · · ; · · · )
= −(dω)k+1(e1, · · · , bei, dei+1, · · · ; ∂hei, ei+1i, · · · ),
i.e. (dω)k is weakly skew-symmetric up to (dω)k+1.
H-STANDARD COHOMOLOGY FOR COURANT-DORFMAN ALGEBRAS AND LEIBNIZ ALGEBRAS
7
Next, we prove the weak R-linearity in arguments of E. By direct calculations, we have the
following:
(3.3)
(3.4)
(d0ω)k(e1, · · · f ei, · · · en+1−2k; α1, · · · αk) − f (d0ω)k(e1, · · · ei, · · · ; · · · )
= X
+X
a>i
(−1)a−ihei, eai(d0ω)k+1(· · · bei, · · · bea, · · · ; ∂f, · · · )
(−1)a−ihei, eaiωk(∂f, e1, · · · bei, · · · bea, · · · ; · · · )
i<a
(δω)k(e1, · · · f ei, · · · en+1−2k; α1, · · · αk) − f (δω)k(e1, · · · ei, · · · ; · · · )
= X
−X
i<a
(−1)a−ihei, eai(δω)k+1(e1, · · · bei, · · · bea, · · · ; ∂f, · · · )
(−1)a−ihei, eaiωk(∂f, e1, · · · bei, · · · bea, · · · ; · · · )
i<a
(d′ω)k(e1, · · · f ei, · · · en+1−2k; α1, · · · αk) − f (d′ω)k(e1, · · · ei, · · · ; · · · )
(−1)a−ihei, eai(d′ω)k+1(e1, · · · bei, · · · bea, · · · ; ∂f, · · · )
= X
a>i
= X
a>i
The sum of the three equations above is:
(dω)k(e1, · · · f ei, · · · en+1−2k; α1, · · · αk) − f (dω)k(e1, · · · ei, · · · ; · · · )
(−1)a−ihei, eai(dω)k+1(e1, · · · bei, · · · bea, · · · ; ∂f, · · · )
i.e. (dω)k is weakly R-linear up to (dω)k+1.
Finally, we need to prove the R-linearity in arguments of H. Again by direct calculations, we
have the following:
(d0ω + d′ω)k(e1, · · · en+1−2k; α1, · · · f αl, · · · αk) − f (d0ω + d′ω)k(· · · ; · · · αl, · · · )
= X
(−1)a+1hαl, eaiωk(· · · bea, · · · ; ∂f, · · · bαl, · · · ),
a
(δω)k(e1, · · · en+1−2k; α1, · · · f αl, · · · αk) − f (δω)k(· · · ; · · · αl, · · · )
= X
(−1)ahαl, eaiωk(· · · bea, · · · ; ∂f, · · · bαl, · · · ),
a
(3.5)
and
so
(dω)k(e1, · · · en+1−2k; α1, · · · f αl, · · · αk) − f (dω)k(· · · ; · · · αl, · · · ) = 0.
The lemma is proved.
Now we turn to the proof of Theorem 3.3:
Proof. By the lemma above, we only need to prove d2 = 0.
d2 can be divided into six parts
d2 = d2
0 + δ2 + (d0 ◦ δ + δ ◦ d0) + (d0 ◦ d′ + d′ ◦ d0) + (δ ◦ d′ + d′ ◦ δ) + d′2.
The first part equals 0, so we only need to compute the other five parts. By direct calculations, we
have the following:
H-STANDARD COHOMOLOGY FOR COURANT-DORFMAN ALGEBRAS AND LEIBNIZ ALGEBRAS
8
(δ2ω)k(e1, · · · , en+2−2k; α1, · · · , αk) = 0,
((d0 ◦ δ + δ ◦ d0)ω)k(e1, · · · ; α1, · · · ) = −X
((d0 ◦ d′ + d′ ◦ d0)ω)k(e1, · · · ; α1, · · · ) = X
((δ ◦ d′ + d′ ◦ δ)ω)k(e1, · · · ; α1, · · · ) = X
j,a<c
ωk−1(· · · bea, αi ◦ ea, · · · ; · · · bαi, · · · ),
i,a
(−1)a+cωk(· · · bea, · · · bec, · · · ; · · · αj ◦ (ea ◦ ec), · · · ),
(−1)a+1ωk−1(αj ◦ ea, · · · bea, · · · ; · · ·cαj, · · · ),
j,a
(d′2ω)k(e1, · · · ; α1, · · · ) = X
(−1)a+bωk(· · · bea, · · · beb, · · · ; · · · bαi, (αi ◦ eb) ◦ ea − (αi ◦ ea) ◦ eb, · · · ),
a<b,i
The sum of the above equations is:
(d2ω)k(e1, · · · en+2−2k; α1, · · · αk)
ωk−1(· · · αi ◦ ea, · · · ; · · · bαi, · · · )
i,a
j,a
j,a<b
= X
(−1)a+1ωk−1(αj ◦ ea, · · · bea, · · · ; · · ·cαj, · · · ) − X
+ X
(−1)a+bωk(· · · bea, · · · beb, · · · ; · · · αj ◦ (ea ◦ eb), · · · )
+ X
(−1)a+bωk(· · · bea, · · · beb, · · · ; · · · bαi, (αi ◦ eb) ◦ ea − (αi ◦ ea) ◦ eb, · · · )
= X
(−1)a+b(cid:0)ωk−1(· · · αi ◦ ea, eb, · · · bea, · · · ; · · · bαi, · · · )
+ωk−1(· · · eb, αi ◦ ea, · · · bea, · · · ; · · · bαi, · · · )(cid:1)
+ X
(−1)a+bωk(· · · bea, · · · beb, · · · ; · · · hea, αi ◦ ebi, · · · )
a<b,i
i,a<b
i,b<a
= 0
The proof is finished.
Definition 3.5. With the notations above, (C•(E, H, V), d) is called H-standard complex of E
with coefficients in V. The resulting cohomology, denoted by H •(E, H, V), is called H-standard
cohomology of E with coefficients in V.
Let's consider the standard cohomology in lower degrees:
Degree 0:
H 0(E, H, V) is the submodule of V consisting of all invariants, i.e.
H 0(E, H, V) = {v ∈ V∇ev = 0, ∀e ∈ E}.
Degree 1:
A cocycle ω in C1(E, H, V) is a map ω0 : E → V satisfying:
ω0(e1 ◦ e2) = ∇e1 ω0(e2) − ∇e2 ω0(e1),
∀e1, e2 ∈ E
and
ω0(α) = 0,
∀α ∈ H.
The first equation above tells that ω0 is a derivation from E to V, while the second equation
tells that ω0 induces a map from E/H to V.
H-STANDARD COHOMOLOGY FOR COURANT-DORFMAN ALGEBRAS AND LEIBNIZ ALGEBRAS
9
η ∈ C1(E, H, V) is a coboundary iff there exists v ∈ V such that:
η0(e) = ∇ev,
∀e ∈ E,
i.e. η0 is an inner derivation from E to V.
Thus H 1(E, H, V) is the space of "outer derivations": {derivations}/{inner derivations} from E
to V which act trivially on H. Or equivalently, H 1(E, H, V) is the space of outer derivations from
E/H to V.
Degree 2:
ω = (ω0, ω1) ∈ C2(E, H, V) is a 2-cocycle iff:
(3.6)
and
(3.7)
∇e1 ω0(e2, e3) − ∇e2 ω0(e1, e3) + ∇e3 ω0(e1, e2)
−ω0(e1 ◦ e2, e3) − ω0(e2, e1 ◦ e3) + ω0(e1, e2 ◦ e3) = 0
∇eω1(α) + ω0(α, e) + ω1(α ◦ e) = 0
∀e, e1, e2, e3 ∈ E, α ∈ H.
Equation 3.6 holds iff the bracket on ¯E , E ⊕ V defined for any e1, e2 ∈ E, v1, v2 ∈ V by:
(e1 + v1)¯◦(e2 + v2) , e1 ◦ e2 + (cid:0)∇e1 v2 − ∇e2 v1 + ω0(e1, e2)(cid:1)
is a Leibniz bracket. Furthermore, if Equation 3.7 also holds, it is easily checked that ( ¯E, R, h·, ·i, ¯∂, ¯◦)
is a Courant-Dorfman algebra, where h·, ·i and ¯∂ are defined as:
he1 + v1, e2 + v2i = he1, e2i
¯∂f = ∂f − ω1(∂f ).
Actually Equation 3.6 implies that
is an ideal of ¯E.
¯H , {α − ω1(α)α ∈ H}
In a summation, 2-cocycles are in 1-1 correspondence with central extensions of Courant-
Dorfman algebras which are split as metric R-modules:
0 → V → ¯E → E → 0
such that ¯H is an ideal of ¯E.
The central extensions determined by 2-cocycles ω1, ω2 are isomorphic iff ω1 − ω2 = dλ, for some
λ ∈ C1(E, H, V).
Thus H 2(E, H, V) classifies isomorphism classes of central extensions of Courant-Dorfman alge-
bras which are split as metric R-modules:
such that ¯H is an ideal of ¯E.
0 → V → ¯E → E → 0
3.2. For Leibniz algebras. Assume L is a Leibniz algebra with left center Z. There is a sym-
metric bilinear product (·, ·) : L ⊗ L → Z defined as:
It is easily checked that such defined bilinear product is invariant, i.e.
(e1, e2) , e1 ◦ e2 + e2 ◦ e1,
∀e1, e2 ∈ L.
ρ(e1)(e2, e3) = (e1 ◦ e2, e3) + (e2, e1 ◦ e3),
∀e1, e2, e3 ∈ L.
H-STANDARD COHOMOLOGY FOR COURANT-DORFMAN ALGEBRAS AND LEIBNIZ ALGEBRAS
10
Let H ⊇ Z be an isotropic ideal in L. Let (V, τ ) be an H-representation of L, which is defined
as follows:
Definition 3.6. An H-trivial left representation (or H-representation for short) of a Leibniz
algebra L is a pair (V, τ ), where V is vector space, and τ : L → gl(V ) is a homomorphism of
Leibniz algebras such that:
τ (h)v = 0,
∀h ∈ H, v ∈ V.
Example 3.7. Since
ρ(h)z = h ◦ z = (h, z) = 0,
∀h ∈ H, z ∈ Z,
(Z, ρ) is an H-representation.
Denote by C n(L, H, V ) the space of all sequences ω = (ω0, · · · , ω[ n
2 ]), where ωk is a linear map
from (⊗n−2kL) ⊗ (⊙kH) to V , ∀k, and is weakly skew-symmetric in arguments of L up to ωk+1:
ωk(e1, · · · ei, ei+1, · · · en−2k; h1, · · · hk) + ωk(e1, · · · ei+1, ei, · · · en−2k; h1, · · · hk)
= −ωk+1(· · · bei, dei+1, · · · ; (ei, ei+1), · · · )
∀ej ∈ L, hl ∈ H.
Theorem 3.8. C•(L, H, V ) , Ln C n(L, H, V ) is a cochain complex, under the coboundary map
d = d0 + δ + d′, where d0 is the coboundary map (Equation 2.2) corresponding to the Leibniz
cohomology of L with coefficients in (V, τ, −τ ), and δ, d′ are defined for any ω ∈ C n(L, H, V )
respectively by:
(δω)k(e1, · · · , en+1−2k; h1, · · · hk) , X
(d′ω)k(e1, · · · en+1−2k; h1, · · · hk) , X
j
a,j
ωk−1(αj, e1, · · · , en+1−2k; h1, · · · bhj, · · · hk)
(−1)a+1ωk(· · · bea, · · · ; · · · bhj, hj ◦ ea, · · · )
∀ea ∈ L, hi ∈ H. ((δω)0 is defined to be 0.)
Proof. The proof of this theorem is quite similar to that of Theorem 3.3, so we omit it here.
Definition 3.9. (C•(L, H, V ), d) is called the H-standard complex of L with coefficients in V .
The resulting cohomology, denoted by H •(L, H, V ) is called the H-standard cohomology of L with
coefficients in V .
The H-standard cohomology of L in degree 0, 1, 2 have similar interpretations to the case of
Courant-Dorfman algebras:
H 0(L, H, V ) is the submodule of V consisting of all invariants.
H 1(L, H, V ) is the space of outer derivations from L to V acting trivially on H.
H 2(L, H, V ) classfies the equivalence classes of abelian extensions of L by V :
such that ¯H is an ideal of ¯L.
0 → V → ¯L → L → 0
H-STANDARD COHOMOLOGY FOR COURANT-DORFMAN ALGEBRAS AND LEIBNIZ ALGEBRAS
11
4. Isomorphism theorem for Courant-Dorfman algebra
In this section, we present one of our main results in this paper, which is a generalization of a
theorem of Ginot-Grutzmann (conjectured by Stienon-Xu) for transitive Courant algebroids.
Let E, H, V be as described in the last section. Since H is an ideal in E containing ∂R, it is
easily checked that E/H is a Lie-Rinehart algebra with the induced anchor map(still denoted by
ρ):
and induced bracket:
ρ([e])f , ρ(e)f,
∀e ∈ E, f ∈ R,
[e1] ◦ [e2] , [e1 ◦ e2],
∀e1, e2 ∈ E.
Moreover, V becomes a representation of E/H with the induced action (still denoted by τ ):
τ ([e])v , τ (e)v,
∀e ∈ E, v ∈ V.
As a result, we have the Chevalley-Eilenberg complex (C•
in V, and the corresponding cohomology H •
CE(E/H, V).
CE(E/H, V), dCE) of E/H with coefficients
Theorem 4.1. Given a Courant-Dorfman algebra (E, R, h·, ·i, ∂, ◦), an R-submodule H ⊇ ∂R
which is an isotropic ideal of E, and an H-representation (V, ∇). If the quotient module E/H is
projective, then we have:
H •(E, H, V) ∼= H •
CE(E/H, V).
Before proof of this theorem, we prove the following two lemmas first.
Lemma 4.2. (C•
CE(E/H, V), dCE) is isomorphic to the following subcomplex of (C•(E, H, V), d):
C•
nv(E, H, V) , {ω ∈ C•(E, H, V)ωk = 0, ∀k ≥ 1, ιαω0 = 0, ∀α ∈ H}
Proof. Obviously C•
nv(E, H, V) is a subcomplex of C•(E, H, V).
And it is easily checked that the following two maps ϕ, φ are well-defined cochain maps and
invertible to each other:
ϕ : C•
ϕ(η)([e1], · · · [en]) , η(e1, · · · en) ∀η ∈ C n
nv(E, H, V) → C•
CE(E/H, V)
nv(E, H, V), ea ∈ E,
and
φ : C•
CE(E/H, V) → C•
nv(E, H, V)
φ(ζ)(e1, · · · en) , ζ([e1], · · · [en]) ∀ζ ∈ C•
CE(E/H, V), ea ∈ E.
Lemma 4.3. Given any ω ∈ C n(E, H, V), if (dω)k = 0, ∀k ≥ 1, then there exists η ∈ C n
and λ ∈ C n−1(E, H, V) such that ω = η + dλ.
nv(E, H, V)
Proof. Since the quotient E/H is a projective module, there exists an R-module decomposition:
E = H ⊕ X . We will give an inductive construction of λ and β. The construction depends on the
decomposition, but the cohomology class of β doesn't depend on the decomposition.
Suppose n = 2m or 2m − 1, we will define λm−1, λm−2, · · · , λ0 one by one, so that each λp :
⊗n−1−2pE ⊗ ⊙pH → V, 0 ≤ p ≤ m − 1 satisfies the following conditions, which we call "Lambda
Conditions":
1). λp is weakly skew-symmetric in arguments of E up to λp+1,
2). λp is weakly R-linear in arguments of E up to λp+1,
3). λp is R-linear in arguments of R,
H-STANDARD COHOMOLOGY FOR COURANT-DORFMAN ALGEBRAS AND LEIBNIZ ALGEBRAS
12
4). ωp+1 = (dλ)p+1,
5). Pi(ωp − d0λp − d′λp)(αi, e1, · · · en−1−2p; α1, · · · bαi, · · · αp+1) = 0, ∀αj ∈ H, ea ∈ E.
The construction of λm−1, λm−2, · · · , λ0 is done in the following four steps.
Step 1:
Construction of λm−1:
When n = 2m − 1 is odd, let
When n = 2m is even, let
λm−1(α1, · · · αm−1) = 0,
∀αj ∈ H.
λm−1(β; α1, · · · αm−1) =
1
m
ωm(β, α1, · · · αm−1), ∀β, αj ∈ H
and
λm−1(x; α1, · · · αm−1) = 0, ∀x ∈ X , αj ∈ H.
It is obvious that λm−1 defined above satisfies Lambda Conditions 1) - 4). So we only need to
prove Lambda Condition 5):
When n = 2m − 1,
(ωm−1 − d0λm−1 − d′λm−1)(αi; α1, · · · bαi, · · · αm) = (dω)m(α1, · · · αm) = 0.
When n = 2m, the left hand side in condition 3) equals
= (δω)m(e; α1, · · · , αm) + X
(ωm−1 − d0λm−1 − d′λm−1)(αi, e; α1, · · · bαi, · · · αm)
∇eλm−1(αi; · · · bαi, · · · ) + X
i
i
λm−1(αi ◦ e; · · · bαi, · · · )
+X
j6=i
(−1)λm−1(e; · · · bαi, · · · αj ◦ αi, · · · ) + X
λm−1(αi; · · · bαi, · · · , αj ◦ e, · · · )
m X
ωm(α1, · · · αi ◦ e, · · · αm)
j6=i
1
= (δω)m(e; α1, · · · αm) + ∇eωm(α1, · · · αm) +
i
X
i
X
i
+X
i<j
(−1)λm−1(e; αi ◦ αj + αj ◦ αi, · · · bαi, · · ·cαj, · · · ) +
ωm(· · · αj ◦ e, · · · )
1
X
m X
)X
i6=j
j
i
ωm(· · · αi ◦ e, · · · )
= (δω)m(e; α1, · · · αm) + (d0ω)m(e; α1, · · · αm) + (
1
m
+
= (dω)m(e; α1, · · · αm)
= 0
m − 1
m
Step 2:
Suppose λm−1, · · · , λk(k > 0) are already defined so that they satisfy Lambda Conditions, we
will construct λk−1, so that it also satisfies Lambda Conditions.
By k-linearity and the decomposition E = H ⊕ X , in order to determine λk−1, we only need to
define the value of λk−1(e1, · · · en+1−2k; α1, · · · αk) in which each ea is either in H or in X .
First we let
(4.1)
,
λk−1(β1, · · · βl, x1, · · · xn+1−2k−l; α1, · · · αk−1)
1
k + l − 1 X
1≤j≤l
(−1)j+1(ωk − d0λk − d′λk)(β1, · · · bβj, · · · βl, · · · ; βj, · · · )
H-STANDARD COHOMOLOGY FOR COURANT-DORFMAN ALGEBRAS AND LEIBNIZ ALGEBRAS
13
∀βr, αs ∈ H, xa ∈ X .
(We call (β, · · · β, x, · · · x) a regular permutation. )
Note that if l = 0, we simply let
λk−1(x1, · · · , xn+1−2k; α1, · · · αk−1) = 0.
For a general permutation σ, an x ∈ X in σ is called an irregular element iff there exists at least
one element of H standing behind x in σ. The value of λk−1(σ; · · · ) is determined inductively as
follows:
If the number of irregular elements in σ is 0, σ is a regular permutation. So the value of
λk−1(σ; · · · ) is determined by Equation 4.1.
Suppose the value of λk−1(σ; · · · ) is already determined for σ with irregular elements less than
t (t ≥ 1). Now for a permutation σ with t irregular elements, assume the last of them is y ∈ X ,
and σ = (•, y, β1, · · · βr, x1, · · · xs), βi ∈ H, xj ∈ X . Switching y with β1, · · · βr one by one, finally
we will get a permutation σ = (•, β1, · · · βr, y, x1, · · · , xs), which has t − 1 irregular elements. The
value of λk−1(σ; · · · ) is already determined. By weak skew-symmetricity we let
λk−1(σ; · · · ) , (−1)rλk−1(σ; · · · ) + X
1≤i≤r
(−1)iλk(•, β1, · · · bβi, · · · βr, x1, · · · xs; ∂hy, βii, · · · ).
As a summary, we have extended λk−1 from regular permutations to general permutations by
weak skew-symmetricity. The extension could be written as a formula:
λk−1(σ; · · · ) = (±1)λk−1(¯σ; · · · ) + X(±1)λk(•; •),
where ¯σ is the regular permutation corresponding to σ.We observe that, for different k, if we do
exactly the same switchings, then the extension formulas should be similar (each summand has
the same sign, with the subscripts of λ modified correspondingly). For example, if we have an
extension formula for k:
λk(σ; · · · ) = (±1)λk(¯σ; · · · ) + X(±1)λk+1(•; •),
then for k − 1, we have similar formula:
λk−1(β, σ, x; · · · ) = (±1)λk−1(β, ¯σ, x; · · · ) + X(±1)λk(β, •, x; •),
∀β ∈ H, x ∈ X .
Step 3:
We need to prove that λk−1 constructed above satisfies Lambda Conditions:
Proof of Lambda Condition 1):
First we prove that λk−1 for regular permutations is weakly skew-symmetric up to λk for the
arguments in H and X respectively.
When the number of arguments in H is 0, the result is obvious.
Otherwise, for the arguments in H,
H-STANDARD COHOMOLOGY FOR COURANT-DORFMAN ALGEBRAS AND LEIBNIZ ALGEBRAS
14
λk−1(β1, · · · βr, βr+1, · · · x1, · · · ; α1, · · · αk−1) + λk−1(β1, · · · βr+1, βr, · · · x1, · · · ; · · · )
=
1
k + l − 1
+
k + l − 1
1
1
=
=
= 0
((−1)r+1 + (−1)r)(ωk − d0λk − d′λk)(· · ·cβr, βr+1 · · · ; βr, · · · )
((−1)r + (−1)r+1)(ωk − d0λk − d′λk)(· · · βr, [βr+1, · · · ; βr+1, · · · )
1
+
j6=r,r+1
j6=r,r+1
(−1)j
(−1)j+1{(ωk − d0λk − d′λk)(· · · bβj, · · · βr, βr+1 · · · ; βj, · · · )
k + l − 1 X
+(ωk − d0λk − d′λk)(· · · bβj, · · · βr+1, βr · · · ; βj, · · · )}
k + l − 1 X
{(d0λk)(· · · bβj, · · · βr, βr+1 · · · ; βj, · · · ) + (d0λk)(· · · bβj, · · · βr+1, βr · · · ; βj, · · · )
+(d′λk)(· · · bβj, · · · βr, βr+1 · · · ; βj, · · · ) + (d′λk)(· · · bβj, · · · βr+1, βr · · · ; βj, · · · )}
k + l − 1 X
−λk(∂hβr, βr+1i, · · · bβj, · · ·cβr, [βr+1, · · · ; βj, · · · )
−d′λk+1(· · · bβj, · · ·cβr, [βr+1, · · · ; ∂hβr, βr+1i, βj, · · · )}
(−1)j{−d0λk+1(· · · bβj, · · ·cβr, [βr+1, · · · ; ∂hβr, βr+1i, βj, · · · )
(by equation 3.1 and 3.2)
j6=r,r+1
1
For the arguments in X ,
λk−1(β1 · · · βl, x1 · · · xa, xa+1 · · · ; · · · ) + λk−1(β1 · · · βl, x1 · · · xa+1, xa · · · ; · · · )
=
=
(−1)j+1{(ωk − d0λk − d′λk)(· · · bβj, · · · xa, xa+1 · · · ; βj, · · · )
1
+λk(β1 · · · βl, x1, · · ·cxa, [xa+1, · · · ; ∂hxa, xa+1i, · · · )
k + l − 1 X
+(ωk − d0λk − d′λk)(· · · bβj, · · · xa+1, xa, · · · ; βj, · · · )}
k + l X
+
1
j
(−1)j+1(ωk+1 − d0λk+1 − d′λk+1)(· · · bβj, · · ·cxa, [xa+1, · · · ; βj, ∂hxa, xa+1i, · · · )
j
(−1)j+1{−ωk+1(· · · bβj, · · ·cxa, [xa+1, · · · ; βj, ∂hxa, xa+1i, · · · )
(by equation 3.1 and 3.2)
j
1
k + l − 1 X
+d0λk+1(· · · bβj, · · ·cxa, [xa+1, · · · ; βj, ∂hxa, xa+1i, · · · )
+λk(hxa, xa+1i, · · · bβj, · · ·cxa, [xa+1, · · · ; βj, · · · )
+d′λk+1(· · · bβj, · · ·cxa, [xa+1, · · · ; βj, ∂hxa, xa+1i, · · · )}
k + l X
1
+
(−1)j+1(ωk+1 − d0λk+1 − d′λk+1)(· · · bβj, · · ·cxa, [xa+1, · · · ; βj, ∂hxa, xa+1i, · · · )
j
H-STANDARD COHOMOLOGY FOR COURANT-DORFMAN ALGEBRAS AND LEIBNIZ ALGEBRAS
15
(−1)j(ωk+1 − d0λk+1 − d′λk+1)(· · · bβj, · · ·cxa, [xa+1, · · · ; βj, ∂hxa, xa+1i, · · · )
=
=
j
1
+
1
1
j
j
(−1)j+1λk(∂hxa, xa+1i, · · · bβj, · · ·cxa, [xa+1, · · · ; βj, · · · )
(k + l − 1)(k + l) X
k + l − 1 X
(k + l − 1)(k + l) X
(ωk+1 − d0λk+1 − d′λk+1)(∂hxa, xa+1i, · · · bβi, · · · bβj, · · ·cxa, [xa+1, · · · ; βi, βj, · · · )
(k + l − 1)(k + l) X
(ωk+1 − d0λk+1 − d′λk+1)(∂hxa, xa+1i, · · · bβj, · · · bβi, · · ·cxa, [xa+1, · · · ; βi, βj, · · · )
(−1)j+1 X
(−1)j+1X
(−1)i+1
(−1)i
j
i>j
i<j
+
1
= 0
Next, for general permutation σ, we give the proof in the following three cases:
(1). λk−1(σ1, β1, β2, σ2; · · · ) + λk−1(σ1, β2, β1, σ2; · · · ) = 0, ∀β1, β2 ∈ H
If every element in σ1 is in H, then
λk−1(σ1, β1, β2, σ2; · · · ) + λk−1(σ1, β2, β1, σ2; · · · )
= (±1)λk−1(σ1, β1, β2, ¯σ2; · · · ) + X(±1)λk(σ1, β1, β2, •; •)
+(±1)λk−1(σ1, β2, β1, ¯σ2; · · · ) + X(±1)λk(σ1, β2, β1, •; •)
= (±1)(cid:0)λk−1(σ1, g1, g2, ¯σ2; · · · ) + λk−1(σ1, g2, g1, ¯σ2; · · · )(cid:1)
+X(±1)(cid:0)λk(σ1, g1, g2, •; •) + λk(σ1, g2, g1, •; •)(cid:1)
= 0
Now suppose (1) holds for σ1 containing at most m elements in X , consider the case when σ1
contains m + 1 elements in X , suppose x is the last one of them, move x to the last position in σ1
and denote the elements in front of x as σ1, σ1 contains m elements in X.
λk−1(σ1, β1, β2, σ2; · · · ) + λk−1(σ1, β2, β1, σ2; · · · )
= (±1)λk−1(σ1, β1, β2, ¯σ2; · · · ) + X(±1)λk(σ1, β1, β2, •; •)
+(±1)λk−1(σ1, β2, β1, ¯σ2; · · · ) + X(±1)λk(σ1, β2, β1, •; •)
= (±1)(cid:0)λk−1(σ1, β1, β2, ¯σ2; · · · ) + λk−1(σ1, β2, β1, ¯σ2; · · · )(cid:1)
= (±1)(cid:0)(±1)λk−1( σ1, x, β1, β2, ¯σ2; · · · ) + X(±1)λk(•, β1, β2, ¯σ2; •)
+(±1)λk−1( σ1, x, β2, β1, ¯σ2; · · · ) + X(±1)λk(•, β2, β1, ¯σ2; •)(cid:1)
= (±1)(cid:0)λk−1( σ1, x, β1, β2, ¯σ2; · · · ) + λk−1( σ1, x, β2, β1, ¯σ2; · · · )(cid:1)
= (±1)(cid:0) − λk( σ1, β2, ¯σ2; ∂hx, β1i, · · · ) + λk( σ1, β1, ¯σ2; ∂hx, β2i, · · · ) + λk−1( σ1, β1, β2, x, ¯σ2; · · · )
= (±1)(cid:0)λk−1( σ1, β1, β2, x, ¯σ2; · · · ) + λk−1( σ1, β2, β1, x, ¯σ2; · · · )(cid:1)
−λk( σ1, β1, ¯σ2; ∂hx, β2i, · · · ) + λk( σ1, β2, ¯σ2; ∂hx, β1i, · · · ) + λk−1( σ1, β2, β1, x, ¯σ2; · · · )(cid:1)
= 0
By mathematical induction, (1) is proved.
H-STANDARD COHOMOLOGY FOR COURANT-DORFMAN ALGEBRAS AND LEIBNIZ ALGEBRAS
16
(2). λk−1(σ1, β, y, σ2; · · · ) + λk−1(σ1, y, β, σ2; · · · ) = −λk(σ1, σ2; ∂hβ, yi, · · · ), ∀β ∈ H, y ∈ X
λk−1(σ1, β, y, σ2; · · · ) + λk−1(σ1, y, β, σ2; · · · )
= (±1)λk−1(σ1, β, y, ¯σ2; · · · ) + X(±1)λk(σ1, β, y, •; •)
+(±1)λk−1(σ1, y, β, ¯σ2; · · · ) + X(±1)λk(σ1, β, g, •; •)
= (±1)(cid:0)λk−1(σ1, β, y, ¯σ2; · · · ) + λk−1(σ1, y, β, ¯σ2; · · · )(cid:1)
+X(±1)(cid:0)λk(σ1, β, y, •; •) + λk(σ1, y, β, •; •)(cid:1)
= (±1)(cid:0)λk−1(σ1, β, y, ¯σ2; · · · ) + (−λk(σ1, ¯σ2; ∂hβ, yi, · · · ) − λk−1(σ1, β, y, ¯σ2; · · · ))(cid:1)
−X(±1)λk+1(σ1, •; ∂hβ, yi, •)
= −(cid:0)(±1)λk(σ1, ¯σ2; ∂hβ, yi, · · · ) + X(±1)λk+1(σ1, •; ∂hβ, yi, •)(cid:1)
= −λk(σ1, σ2; ∂hβ, yi, · · · )
(3). λk−1(σ1, y1, y2, σ2; · · · ) + λk−1(σ1, y2, y1, σ2; · · · ) = −λk(σ1, σ2; ∂hy1, y2i, · · · ), ∀y1, y2 ∈ X.
Suppose ¯σ2 = (β1, · · · βa, x1, · · · xb), then
(−1)jλk+1(σ1, β1, · · · bβj, · · · bβi, · · · βa, x1, · · · xb; ∂hy1, βji, ∂hy2, βii, · · · )
(−1)j+1λk+1(σ1, β1, · · · bβi, · · · bβj, · · · βa, x1 · · · xb; ∂hy1, βji, ∂hy2, βii, · · · )
j>i
= (cid:0) X
1≤i≤a
= X
λk−1(σ1, y1, y2, ¯σ2; · · · ) + λk−1(σ1, y2, y1, ¯σ2; · · · )
(−1)iλk(σ1, y1, β1, · · · bβi, · · · βa, x1, · · · xb; ∂hy2, βii, · · · )
i
1≤j<i
1≤j≤a
(−1)jλk(σ1, y2, β1, · · · bβj, · · · βa, x1, · · · xb; ∂hy1, βji, · · · )
+(−1)aλk−1(σ1, y1, β1, · · · βa, y2, x1, · · · xb; · · · )(cid:1)
+(cid:0) X
+(−1)aλk−1(σ1, y2, β1, · · · βa, y1, x1, · · · xb; · · · )(cid:1)
(−1)i(cid:0) X
+X
+(−1)a+1λk(σ1, β1, · · · bβi, · · · βa, y1, x1, · · · xb; ∂hy2, βii, · · · )(cid:1)
+(−1)a(cid:0) X
+(−1)aλk−1(σ1, β1, · · · βa, y1, y2, x1, · · · xb; · · · )(cid:1)
+X
+X
+(−1)a+1λk(σ1, β1, · · · bβj, · · · βa, y2, x1, · · · xb; ∂hy1, βji, · · · )(cid:1)
+(−1)a(cid:0) X
+(−1)aλk−1(σ1, β1, · · · βa, y2, y1, x1, · · · xb; · · · )(cid:1)
(−1)j(cid:0) X
1≤j≤a
j
1≤i<j
1≤i≤a
(−1)iλk(σ1, β1, · · · bβi, · · · βa, y1, x1, · · · xb; ∂hy2, βii, · · · )
(−1)jλk(σ1, β1, · · · bβj, · · · βa, y2, x1, · · · xb; ∂hy1, βji, · · · )
(−1)iλk+1(σ1, β1, · · · bβi, · · · bβj, · · · βa, x1, · · · xb; ∂hy1, βji, ∂hy2, βii, · · · )
(−1)i+1λk+1(σ1, β1, · · · bβj, · · · bβi, · · · βa, x1, · · · xb; ∂hy1, βji, ∂hy2, βii, · · · )
i>j
H-STANDARD COHOMOLOGY FOR COURANT-DORFMAN ALGEBRAS AND LEIBNIZ ALGEBRAS
17
= λk−1(σ1, β1, · · · βa, y1, y2, x1, · · · xb; · · · ) + λk−1(σ1, β1, · · · βa, y2, y1, x1, · · · xb; · · · )
So
λk−1(σ1, y1, y2, σ2; · · · ) + λk−1(σ1, y2, y1, σ2; · · · )
= (±1)λk−1(σ1, y1, y2, ¯σ2; · · · ) + X(±1)λk(σ1, y1, y2, •; •)
+(±1)λk−1(σ1, y2, y1, ¯σ2; · · · ) + X(±1)λk(σ1, y2, y1, •; •)
= (±1)(cid:0)λk−1(σ1, y1, y2, ¯σ2; · · · ) + λk−1(σ1, y2, y1, ¯σ2; · · · )(cid:1)
+X(±1)(cid:0)λk(σ1, y1, y2, •; •) + λk(σ1, y2, y1, •; •)(cid:1)
−X(±1)λk+1(σ1, by1, by2, •; ∂hy1, y2i, •)
(denote by µthe permutation (σ1, β1, · · · , βa))
= (±1)(cid:0)λk−1(σ1, β1, · · · βa, y1, y2, x1, · · · xb; · · · ) + λk−1(σ1, β1, · · · βa, y2, y1, x1, · · · xb; · · · )(cid:1)
= (±1)(cid:0)(±1)λk−1(¯µ, y1, y2, x1, · · · xb; · · · ) + X(±1)λk(•, y1, y2, x1, · · · xb; •)
+(±1)λk−1(¯µ, y2, y1, x1, · · · , xb; · · · ) + X(±1)λk(•, y2, y1, x1, · · · , xb; •)(cid:1)
−X(±1)λk+1(σ1, by1, by2, •; ∂hy1, y2i, •)
−X(±1)λk+1(σ1, by1, by2, •; ∂hy1, y2i, •)
= −(±1)(cid:0)(±1)λk(¯µ, by1, by2, x1, · · · ; ∂hy1, y2i, · · · ) + X(±1)λk+1(•, by1, by2, x1, · · · ; ∂hy1, y2i, •)(cid:1)
= −(cid:0)(±1)λk(µ, by1, by2, x1, · · · xb; ∂hy1, y2i, · · · ) + X(±1)λk+1(σ1, by1, by2, •; ∂hy1, y2i, •)(cid:1)
= −(cid:0)(±1)λk(σ1, by1, by2, ¯σ2; ∂hy1, y2i, · · · ) + X(±1)λk+1(σ1, by1, by2, •; ∂hy1, y2i, •)(cid:1)
= −λk(σ1, by1, by2, σ2; ∂hy1, y2i, · · · )
Combining (1)(2)(3) above, Lambda condition 1) for λk−1 is proven.
Proof of Lambda Condition 2):
Since the weak skew-symmetricity of λk−1 is already proven, we only need to prove that λk−1
is R-linear in the last argument of E.
When the last argument is in X :
λk−1(σ, f x; · · · ) − f λk−1(σ, x; · · · )
= (cid:0)(±1)λk−1(¯σ, f x; · · · ) + X(±1)λk(•, f x; •)(cid:1) − f(cid:0)(±1)λk−1(¯σ, x; · · · ) + X(±1)λk(•, x; •)(cid:1)
=
j
(suppose ¯σ = (β1, · · · βl, x1, · · · xn−2k−l))
±1
(−1)j+1{(ωk − d0λk − d′λk)(· · · bβj, · · · f x; βj, · · · )
k + l − 1 X
−f (ωk − d0λk − d′λk)(· · · bβj, · · · x; βj, · · · )}
= 0 (By Equation 3.3 and 3.4).
When the last argument is in H:
λk−1(σ, f β; · · · ) − f λk−1(σ, β; · · · )
= (cid:0)(±1)λk−1(¯σ, f β; · · · ) + X(±1)λk(•, f β; •)(cid:1) − f(cid:0)(±1)λk−1(¯σ, β; · · · ) + X(±1)λk(•, β; •)(cid:1)
(suppose ¯σ = (β1, · · · βl−1, x1, · · · xn+1−2k−l))
H-STANDARD COHOMOLOGY FOR COURANT-DORFMAN ALGEBRAS AND LEIBNIZ ALGEBRAS
18
= (±1)(cid:0)(−1)n+1+lλk−1(β1, · · · βl−1, f β, x1, · · · ; · · · )
a
a
+
j
a
±1
(±1)(−1)n+1+l
(By Equation 3.3, 3.4 and 3.5)
(−1)a+n+lλk(β1, · · · βl−1, x1, · · ·cxa, · · · xn+1−2k−l; ∂hxa, f βi, · · · )(cid:1)
(−1)a+n+lλk(β1, · · · βl−1, x1, · · ·cxa, · · · xn+1−2k−l; ∂hxa, βi, · · · )(cid:1)
(−1)j+1(cid:0)(ωk − d0λk − d′λk)(· · · bβj, · · · f β, · · · ; βj, · · · )
+X
−(±1)f(cid:0)(−1)n+1+lλk−1(β1, · · · βl−1, β, x1, · · · ; · · · )
+X
k + l − 1 X
−f (ωk − d0λk − d′λk)(· · · bβj, · · · β, · · · ; βj, · · · )(cid:1)
k + l − 1 (cid:0)(ωk − d0λk − d′λk)(· · · bβ, · · · ; f β, · · · ) − f (ωk − d0λk − d′λk)(· · · bβ, · · · ; β, · · · )(cid:1)
(±1)(−1)n
+(±1)X
(−1)a+n+lhxa, βiλk(· · · bβ, · · ·cxa, · · · ; ∂f, · · · )
(−1)n+l+j+ahβ, xai(cid:0)(ωk+1 − d0λk+1 − d′λk+1)(· · · bβj, · · · bβ, · · ·cxa, · · · ; ∂f, βj, · · · )
k + l − 1 X
−λk(∂f, · · · bβj, · · · bβ, · · ·cxa, · · · ; βj, · · · )(cid:1)
k + l − 1 X
(−1)n+1+l+ahβ, xaiλk(· · · bβ, · · ·cxa, · · · ; ∂f, · · · )
k + l − 1 X
(−1)a+n+l+j+1hβ, xai(ωk+1 − d0λk+1 − d′λk+1)(· · · bβj, · · · bβ, · · ·cxa, · · · ; ∂f, βj, · · · )
k + l − 1 X
X
(k + l − 1)2 X
(cid:0) X
+ X
(−1)i+j(ωk+1 − d0λk+1 − d′λk+1)(β0, · · · bβi, · · · bβj, · · ·cxa, · · · ; βi, βj, · · · )
(−1)i+j+1(ωk+1 − d0λk+1 − d′λk+1)(β0, · · · bβj, · · · bβi, · · ·cxa, · · · ; βi, βj, · · · )(cid:1)
(−1)n+l+j+a+1hβ, xaiλk(β0(, ∂f ), · · · bβj, · · · βl−1, · · ·cxa, · · · ; βj, · · · )
(−1)n+l+a+1hβ, xai
a
0≤j≤l−1
+
+
0≤i<j≤l−1
±1
±1
a
j,a
±1
±1
0≤j<i≤l−1
j,a
a
=
=
=
=
= 0
Proof of Lambda Condition 3):
=
λk−1(β1, · · · βl, x1, · · · xn+1−2k−l; f α1, · · · αk−1) − f λk−1(· · · ; α1, · · · αk−1)
1
(−1)j+1(cid:0)(ωk − d0λk − d′λk)(· · · bβj, · · · ; βj, f α1, · · · )
k + l − 1 X
−f (ωk − d0λk − d′λk)(· · · bβj, · · · ; βj, α1, · · · )(cid:1)
j
(By Equation 3.5)
H-STANDARD COHOMOLOGY FOR COURANT-DORFMAN ALGEBRAS AND LEIBNIZ ALGEBRAS
19
(−1)j X
(−1)l+a+1hxa, α1iλk(· · · bβj, · · ·cxa, · · · ; βj, ∂f,cα1, · · · )
a
(−1)l+a+1hxa, α1i
(−1)j+i+1(ωk+1 − d0λk+1 − d′λk+1)(· · · bβi, · · · bβj, · · ·cxa, · · · ; βi, βj, ∂f,cα1, · · · )
(−1)j+i(ωk+1 − d0λk+1 − d′λk+1)(· · · bβj, · · · bβi, · · ·cxa, · · · ; βi, βj, ∂f,cα1, · · · )(cid:1)
i>j
=
=
j
1
1
k + l − 1 X
(k + l − 1)2 X
(cid:0)X
+X
i<j
a
= 0
Proof of Lambda Condition 4):
For regular permutations:
= X
(δλ)k(β1, · · · βl, x1, · · · xn−2k−l; α1, · · · αk)
λk−1(αi, β1, · · · βl, x1, · · · xn−2k−l; · · · bαi, · · · )
i
(ωk − d0λk − d′λk)(β1, · · · βl, · · · ; α1, · · · αk)
1
k + l X
k + l X
1
i
+
i,j
k
k + l
1
+
k + l X
j
=
=
=
(−1)j(ωk − d0λk − d′λk)(αi, β1 · · · bβj, · · · βl · · · ; βj, · · · bαi, · · · )
(By Lambda Condition 5) for λk)
(ωk − d0λk − d′λk)(β1, · · · βl, · · · ; α1, · · · αk)
(−1)j(−1)(ωk − d0λk − d′λk)(βj, β1 · · · bβj, · · · βl · · · ; α1, · · · αk)
k + l
k + l
(ωk − d0λk − d′λk)(β1, · · · βl, · · · ; α1, · · · αk)
So ωk = (dλ)k for regular permutation (β1, · · · βl, x1, · · · xn−2k−l).
Since ωk and (dλ)k are both weakly skew-symmetric up to ωk+1 = (dλ)k+1, so ωk = (dλ)k holds
for general permutations.
Proof of Lambda Condition 5):
We only need to prove the following:
X
i
(ωk−1 − d0λk−1 − d′λk−1)(αi, e1, · · · , en+1−2k; α1, · · · bαi, · · · αk)
= (dω)k(e1, · · · , en+1−2k; α1, · · · αk)
or equivalently,
(d0ωk + d′ωk)(e1, · · · en+1−2k; α1, · · · αk) + X
i
(d0λk−1 + d′λk−1)(αi, e1, · · · ; · · · bαi, · · · )
= (d0ωk + d′ωk)(e1, · · · en+1−2k; α1, · · · αk)
+X
i,a
(−1)a∇ea λk−1(αi, · · · bea, · · · ; · · · bαi, · · · ) + X
i,a
(−1)λk−1(· · · bea, αi ◦ ea, · · · ; · · · bαi, · · · )
H-STANDARD COHOMOLOGY FOR COURANT-DORFMAN ALGEBRAS AND LEIBNIZ ALGEBRAS
20
i,a<b
+ X
+X
(−1)a+1λk−1(αi, · · · bea, · · · ea ◦ eb, · · · ; · · · bαi, · · · ) + X
λk−1(· · · ; · · · bαi, · · · αj ◦ αi, · · · ) + X
= (d0ωk + d′ωk)(e1, · · · en+1−2k; α1, · · · αk) + X
λk−1(· · · ; · · · bαi, · · · αj ◦ αi, · · · )
(−1)aλk−1(αi, · · · bea, · · · ; · · · bαi, · · · αj ◦ ea, · · · )
(−1)a∇ea (ωk − d0λk − d′λk)(· · · bea, · · · ; · · · )
a,j6=i
j6=i
j6=i
a
λk−1(· · · bea, αi ◦ ea, · · · ; · · · bαi, · · · )
a,i
i<j
a<b
(−1)a+1(ωk − d0λk − d′λk)(· · · bea, · · · ea ◦ eb, · · · ) − X
λk−1(e1, · · · en+1−2k; αi ◦ αj + αj ◦ αi, · · · bαi, · · ·cαj, · · · )
(−1)a(ωk − d0λk − d′λk)(· · · bea, · · · ; · · · αj ◦ ea, · · · )
(−1)aλk−1(αj ◦ ea, · · · bea, · · · ; · · ·cαj, · · · )(cid:1)
+X
+X
+(cid:0)X
−X
0 + d0 ◦ d′)λ)k(e1, · · · ; α1, · · · ) − X
+((d′ ◦ d0 + d′2)λ)k(e1, · · · ; α1, · · · ) + X
a,j
a,j
a,i
= ((d2
λk−1(· · · bea, αi ◦ ea, · · · ; · · · bαi, · · · )
(−1)a+1λk−1(αj ◦ ea, · · · bea, · · · ; · · ·cαj, · · · )
a,j
= ((d0 + d′ + δ)2λ)k(e1, · · · en+1−2k; α1, · · · αk)
= 0
Thus λk−1 satisfies all Lambda Conditions.
Step 4:
By mathematical induction, finally we obtain (λ0, · · · , λm−1) with each λp satisfying Lambda
Conditions.
Let λ , (λ0, · · · , λm−1), Lambda Conditions 1), 2), 3) imply that λ is a cochain in C n−1(E, H, V).
Let η , ω − dλ ∈ C n(E, H, V).
By Lambda Condition 4), η = (ω0 − (dλ)0, 0, · · · , 0) = (ω0 − d0λ0 − d′λ0, 0, · · · , 0).
By Lambda Condition 5), (ω0 − d0λ0 − d′λ0)(α, e1, · · · , en−1) = 0, ∀α ∈ H, ei ∈ E.
So ιαη0 = ια(ω0 − d0λ0 − d′λ0) = 0, ∀α ∈ H, which implies that η ∈ C n
nv(E, H, V).
Thus ω = η + dλ, η ∈ C n
nv(E, H, V), λ ∈ C n−1(E, H, V), the proof is finished.
Now we turn to the proof of Theorem 4.1:
Proof. By Lemma 4.2, we only need to prove that the inclusion map
induces an isomorphism
ψ : Cnv(E, H, V) → C(E, H, V)
ψ∗ : H •
nv(E, H, V) → H •(E, H, V).
1). ψ∗ is surjective.
Given any [ω] ∈ H n(E, H, V), since ω is closed, by Lemma 4.3, there exists η ∈ C n
nv(E, H, V) and
λ ∈ C n−1(E, H, V) such that ω = η + dλ. We see that η = ω − dλ is closed, so ψ∗([η]) = [η] = [ω].
2). ψ∗ is injective.
Suppose ζ ∈ C n
prove that ζ is exact in C n
nv(E, H, V).
nv(E, H, V) is exact in C n(E, H, V), i.e. ∃ω ∈ C n−1(E, H, V), dω = ζ, we need to
H-STANDARD COHOMOLOGY FOR COURANT-DORFMAN ALGEBRAS AND LEIBNIZ ALGEBRAS
21
Since (dω)k = ζk = 0, ∀k > 0, by Lemma 4.3, there exists η ∈ C n−1
nv (E, H, V) and λ ∈
C n−2(E, H, V) such that ω = η + dλ. So
ζ = dω = dη
is exact in C n
nv(E, H, V).
The proof is finished.
Remark 4.4. When ρ∗ is injective (in this case we call E a transitive Courant-Dorfman algebra),
and H = ρ∗(Ω1), V = R, Definition 3.5 recovers the ordinary standard cohomology (Definition
2.7). Moreover, if E = Γ(E) is the space of sections of a transitive Courant algebroid E, Theorem
4.1 recovers the isomorphism between the standard cohomology and naive cohomology of E, as
conjectured by Stienon-Xu [15] and first proved by Ginot-Grutzmann [5]. So Theorem 4.1 is a
generalization of their result.
Example 4.5. Suppose G is a bundle of quadratic Lie algebras on M . Given a standard Courant
algebroid structure (see Chen, Stienon and Xu [3]) on
E = T M ⊕ G ⊕ T ∗M,
let E = Γ(E). As mentioned in the remark above, if we take H = Γ(T ∗M ) = Ω1(M ) and
V = C∞(M ), the Ω1(M )-standard cohomology H •(E, Ω1(M ), C∞(M )) coincides with the standard
cohomology of E, and is isomorphic to the cohomology of Lie algebroid T M ⊕ G with coefficients
in C∞(M ).
Now suppose K is an isotropic ideal in G, then H = Γ(K ⊕ T ∗M ) ⊇ Γ(T ∗M ) is an isotropic ideal
in E. Given a Γ(K ⊕ T ∗M )-representation V (e.g. C∞(M )), we have the Γ(K ⊕ T ∗M )-standard
cohomology H •(E, Γ(K ⊕ T ∗M ), C∞(M )). By Theorem 4.1, it is isomorphic to the cohomology of
Lie algebroid T M ⊕ (G/K) with coefficients in V.
5. Crossed products of Leibniz algebras
In this section, we associate a Courant-Dorfman algebra to any Leibniz algebra and consider
the relation between H-standard complexes of them. At last we prove an isomorphism theorem
for Leibniz algebras.
Given a Leibniz algebra L with left center Z, let S•(Z) be the algebra of symmetric tensors of
Z. We construct a Courant-Dorfman algebra structure on the tensor product
L , S•(Z) ⊗ L
as follows:
let R be S•(Z);
let the S•(Z)-module structure of L be given by multiplication of S•(Z), i.e.
f1 · (f2 ⊗ e) , (f1f2) ⊗ e,
∀f1, f2 ∈ S•(Z), e ∈ L;
(For simplicity, we will write f ⊗ e as f e from now on.)
let the symmetric bilinear form h·, ·i of L be the S•(Z)-bilinear extension of the symmetric
product (·, ·) of L, i.e.
hf1e1, f2e2i , f1f2(e1, e2),
∀f1, f2 ∈ S•(Z), e1, e2 ∈ L
(since he1, e2i = (e1, e2), in the following we always use the notation h·, ·i);
let the derivation ∂ : S•(Z) → L be the extension of the inclusion map Z ֒→ L by Leibniz rule,
i.e.
∂(z1 · · · zk) , X
1≤i≤k
(z1 · · · bzi · · · zk)∂zi,
∀zi ∈ Z;
H-STANDARD COHOMOLOGY FOR COURANT-DORFMAN ALGEBRAS AND LEIBNIZ ALGEBRAS
22
let the Dorfman bracket on L, still denoted by ◦, be the extension of the Leibniz bracket of L:
(5.1)
∀f1, f2 ∈ S•(Z), e1, e2 ∈ L.
f1e1 ◦ f2e2 , f1f2(e1 ◦ e2) + he1, e2if2∂f1 + he1, ∂f2if1e2 − he2, ∂f1if2e1
Proposition 5.1. With the above notations, (L, S•(Z), h·, ·i, ∂, ◦) becomes a Courant-Dorfman
algebra (called the crossed product of L).
Proof. We need to check all the six conditions in Definition 2.6.
1). f1e1 ◦ f (f2e2) = f (f1e1 ◦ f2e2) + hf1e1, ∂f if2e2
f1e1 ◦ f (f2e2)
= f f1f2(e1 ◦ e2) + he1, e2if f2∂f1 + he1, ∂(f f2)if1e2 − he2, ∂f1if f2e1
= f f1f2(e1 ◦ e2) + he1, e2if f2∂f1 + f he1, ∂f2if1e2 + f2he1, ∂f if1e2 − he2, ∂f1if f2e1
= f (f1e1 ◦ f2e2) + hf1e1, ∂f if2e2.
2). hf1e1, ∂hf2e2, f3e3ii = hf1e1 ◦ f2e2, f3e3i + hf2e2, f1e1 ◦ f3e3i
hf1e1, ∂hf2e2, f3e3ii
= f1he1, ∂(f2f3he2, e3i)i
= f1f2f3he1, ∂he2, e3ii + f1f2he2, e3ihe1, ∂f3i + f1f3he2, e3ihe1, ∂f2i
= f1f2f3(cid:0)he1 ◦ e2, e3i + he2, e1 ◦ e3i(cid:1) + f1f2he2, e3ihe1, ∂f3i + f1f3he2, e3ihe1, ∂f2i
= f1f2f3he1 ◦ e2, e3i + f2f3he1, e2ihe3, ∂f1i + f1f3he2, e3ihe1, ∂f2i − f2f3he1, e3ihe2, ∂f1i
+f1f2f3he2, e1 ◦ e3i + f2f3he1, e3ihe2, ∂f1i + f1f2he2, e3ihe1, ∂f3i − f2f3he1, e2ihe3, ∂f1i
= hf1f2(e1 ◦ e2) + he1, e2if2∂f1 + he1, ∂f2if1e2 − he2, ∂f1if2e1, f3e3i
+hf2e2, f1f3(e1 ◦ e3) + he1, e3if3∂f1 + he1, ∂f3if1e3 − he3, ∂f1if3e1i
= hf1e1 ◦ f2e2, f3e3i + hf2e2, f1e1 ◦ f3e3i.
3). f1e1 ◦ f2e2 + f2e2 ◦ f1e1 = ∂hf1e1, f2e2i
f1e1 ◦ f2e2 + f2e2 ◦ f1e1
= f1f2(e1 ◦ e2) + he1, e2if2∂f1 + he1, ∂f2if1e2 − he2, ∂f1if2e1
+f1f2(e2 ◦ e1) + he1, e2if1∂f2 + he2, ∂f1if2e1 − he1, ∂f2if1e2
= f1f2∂he1, e2i + he1, e2if2∂f1 + he1, e2if1∂f2
= ∂hf1e1, f2e2i.
Combining 1) and 3), we get the following:
f (f1e1) ◦ f2e2
= (f (f1e1) ◦ f2e2 + f2e2 ◦ f (f1e1)) − f2e2 ◦ f (f1e1)
= ∂hf (f1e1), f2e2i − (f (f2e2 ◦ f1e1) + hf2e2, ∂f if1e1
= hf1e1, f2e2i∂f + f ∂hf1e1, f2e2i − f (f2e2 ◦ f1e1) − hf2e2, ∂f if1e1
= f (f1e1 ◦ f2e2) + hf1e1, f2e2i∂f − hf2e2, ∂f if1e1
4). h∂f, ∂f ′i = 0.
We only need to consider the case of monomials, suppose
f = z1z2 · · · zk, f ′ = z′
1z′
2 · · · z′
l, zi, z′
j ∈ Z,
H-STANDARD COHOMOLOGY FOR COURANT-DORFMAN ALGEBRAS AND LEIBNIZ ALGEBRAS
23
h∂f, ∂f ′i
= hX
= X
i
(z1 · · · bzi · · · zk)∂zi,X
1 · · · bz′
(z1 · · · bzi · · · zkz′
j
i,j
= 0
l)∂z′
ji
j · · · z′
1 · · · bz′
(z′
l)(∂zi ◦ ∂z′
j · · · z′
j + ∂z′
j ◦ ∂zi)
5). ∂f ◦ (f ′e) = 0
First we prove that ∂f ◦ e = 0, ∀f ∈ S•(Z), e ∈ L.
We only need to consider the case of monomials: suppose f = z1z2 · · · zk, zi ∈ Z.
When k = 1, i.e. f = z1 ∈ Z, the equation is trivial.
Now suppose the equation holds for any k ≤ m, let's consider the case of k = m + 1.
∂(z1z2 · · · zm+1) ◦ e
= (cid:0)(z1 · · · zm)∂zm+1 + zm+1∂(z1 · · · zm)(cid:1) ◦ e
= (z1 · · · zm)(∂zm+1 ◦ e) + h∂zm+1, ei∂(z1 · · · zm) − he, ∂(z1 · · · zm)i∂zm+1
+zm+1(∂(z1 · · · zm) ◦ e) + h∂(z1 · · · zm), ei∂zm+1 − he, ∂zm+1i∂(z1 · · · zm)
= 0
Thus by induction, ∂f ◦ e = 0 holds for any f ∈ S•(Z).
Then combining 1) and 4),
∂f ◦ (f ′e) = f ′(∂f ◦ e) + h∂f, ∂f ′ie = 0
6). f1e1 ◦ (f2e2 ◦ f3e3) = (f1e1 ◦ f2e2) ◦ f3e3 + f2e2 ◦ (f1e1 ◦ f3e3)
First we prove the equation for the case when f2 = f3 = 1:
(f1e1 ◦ e2) ◦ e3 + e2 ◦ (f1e1 ◦ e3)
= (cid:0)f1(e1 ◦ e2) + he1, e2i∂f1 − he2, ∂f1ie1(cid:1) ◦ e3 + e2 ◦ (cid:0)f1(e1 ◦ e3) + he1, e3i∂f1 − he3, ∂f1ie1(cid:1)
= (cid:0)f1((e1 ◦ e2) ◦ e3) + he1 ◦ e2, e3i∂f1 − he3, ∂f1i(e1 ◦ e2)(cid:1)
+(cid:0)he1, e2i(∂f1 ◦ e3) + h∂f1, e3i∂he1, e2i − he3, ∂he1, e2ii∂f1(cid:1)
−(cid:0)he2, ∂f1i(e1 ◦ e3) + he1, e3i∂he2, ∂f1i − he3, ∂he2, ∂f1iie1(cid:1)
+(cid:0)f1(e2 ◦ (e1 ◦ e3) + he2, ∂f1i(e1 ◦ e3)(cid:1) + (cid:0)he1, e3i(e2 ◦ ∂f1) + he2, ∂he1, e3ii∂f1(cid:1)
−(cid:0)he3, ∂f1i(e2 ◦ e1) + he2, ∂he3, ∂f1iie1(cid:1)
+(cid:0)he3, ∂he2, ∂f1ii − he2, ∂he3, ∂f1ii(cid:1)e1 + he1, e3i(cid:0)e2 ◦ ∂f1 − ∂he2, ∂f1i(cid:1)
+he3, ∂f1i(cid:0)∂he1, e2i − e1 ◦ e2 − e2 ◦ e1(cid:1) + he1, e2i(∂f1 ◦ e3)
= f1((e1 ◦ e2) ◦ e3) + f1(e2 ◦ (e1 ◦ e3) + (cid:0)he1 ◦ e2, e3i − he3, ∂he1, e2ii + he2, ∂he1, e3ii(cid:1)∂f1
+he2, ∂f1i(e1 ◦ e3) − he2, ∂f1i(e1 ◦ e3)
= f1(e1 ◦ (e2 ◦ e3)) + (cid:0)he2, ∂he1, e3ii − he2 ◦ e1, e3i(cid:1)∂f1 − he2 ◦ e3, ∂f1ie1
= f1e1 ◦ (e2 ◦ e3)
Then we prove the equation for the case when f3 = 1:
H-STANDARD COHOMOLOGY FOR COURANT-DORFMAN ALGEBRAS AND LEIBNIZ ALGEBRAS
24
(f1e1 ◦ f2e2) ◦ e3 + f2e2 ◦ (f1e1 ◦ e3)
(let x1 , f1e1)
= (cid:0)f2(x1 ◦ e2) + hx1, ∂f2ie2(cid:1) ◦ e3 + f2(e2 ◦ (x1 ◦ e3)) + he2, x1 ◦ e3i∂f2 − hx1 ◦ e3, ∂f2ie2
= (cid:0)f2((x1 ◦ e2) ◦ e3) + hx1 ◦ e2, e3i∂f2 − he3, ∂f2i(x1 ◦ e2)(cid:1)
+(cid:0)hx1, ∂f2i(e2 ◦ e3) + he2, e3i∂hx1, ∂f2i − he3, ∂hx1, ∂f2iie2(cid:1)
+f2(e2 ◦ (x1 ◦ e3)) + he2, x1 ◦ e3i∂f2 − hx1 ◦ e3, ∂f2ie2
= f2((x1 ◦ e2) ◦ e3) + f2(e2 ◦ (x1 ◦ e3)) + hx1, ∂f2i(e2 ◦ e3) + he2, e3i∂hx1, ∂f2i
+(cid:0)hx1 ◦ e2, e3i + he2, x1 ◦ e3i(cid:1)∂f2 − (cid:0)he3, ∂f2i(x1 ◦ e2) + (he3, ∂hx1, ∂f2ii + hx1 ◦ e3, ∂f2i)e2(cid:1)
= x1 ◦ (cid:0)f2(e2 ◦ e3) + he2, e3i∂f2 − he3, ∂f2ie2(cid:1)
= f1e1 ◦ (f2e2 ◦ e3).
Finally,
(f1e1 ◦ f2e2) ◦ f3e3 + f2e2 ◦ (f1e1 ◦ f3e3)
(let x1 , f1e1, x2 , f2e2)
= f ((x1 ◦ x2) ◦ e3) + hx1 ◦ x2, ∂f3ie3
+f (x2 ◦ (x1 ◦ e3)) + hx2, ∂f3i(x1 ◦ e3) + hx1, ∂f3i(x2 ◦ e3) + hx2, ∂hx1, ∂f3iie3
= x1 ◦ (cid:0)f3(x2 ◦ e3) + hx2, ∂f3ie3(cid:1)
= f1e1 ◦ (f2e2 ◦ f3e3)
Thus the proposition is proved.
By Equation 2.3, the anchor map
ρ : L → Der(S•(Z), S•(Z))
can be defined as follows:
ρ(f e)(z1 · · · zk) , f X
i
(z1 · · · bzi · · · zk)(ρ(e)zi),
∀f ∈ S•(Z), e ∈ L, zi ∈ Z.
Proposition 5.2. Suppose H ⊇ Z is an isotropic ideal in L, and (V, τ ) is an H-representation of
L, let V , S•(Z) ⊗ V , then
1). H , S•(Z) ⊗ H is an isotropic ideal in L
2). (V, τ ) induces an H-representation (V, ∇) of L, where ∇ : L → Der(V) is defined as follows:
∇f1e(f2v) , f1(cid:0)he, ∂f2iv + f2(τ (e)v)(cid:1), ∀f1, f2 ∈ S•(Z), e ∈ L, v ∈ V.
Proof. 1). Since
hf1h1, f2h2i = f1f2hh1, h2i = 0,
∀f1, f2 ∈ S•(Z), h1, h2 ∈ H,
H is isotropic in L. And it is easily observed from Equation 5.1 that H is an ideal.
2). From the definition of ∇, it is obvious that
∇f1h(f2v) = 0
∇f1x(f2v) = f1∇x(f2v)
∇x(f1(f2v)) = (ρ(x)f1)(f2v) + f1∇x(f2v)
H-STANDARD COHOMOLOGY FOR COURANT-DORFMAN ALGEBRAS AND LEIBNIZ ALGEBRAS
25
∀f1, f2 ∈ S•(Z), h ∈ H, x ∈ L, v ∈ V. So we only need to prove that ∇ is a homomorphism of
Leibniz algebras:
[∇f1 e1 , ∇f2e2 ](f v)
= ∇f1e1(cid:0)f2he2, ∂f iv + f2f τ (e2)v(cid:1) − ∇f2e2(cid:0)f1he1, ∂f iv + f1f τ (e1)v(cid:1)
= (cid:0)f1he1, ∂(f2he2, ∂f i)iv + f1f2he2, ∂f iτ (e1)v + f1he1, ∂(f2f )iτ (e2)v + f1f2f τ (e1)τ (e2)v(cid:1)
−(cid:0)f2he2, ∂(f1he1, ∂f i)iv + f2f1he1, ∂f iτ (e2)v + f2he2, ∂(f1f )iτ (e1)v + f2f1f τ (e2)τ (e1)v(cid:1)
= f1f2(cid:0)he1, ∂he2, ∂f ii − he2, ∂he1, ∂f ii(cid:1)v + f1f2f (τ (e1)τ (e2)v − τ (e2)τ (e1)v)
+f1he2, ∂f ihe1, ∂f2iv − f2he1, ∂f ihe2, ∂f1iv + f1f he1, ∂f2iτ (e2)v − f2f he2, ∂f1iτ (e1)v
= f1f2∇e1◦e2 (f v) + he1, ∂f2if1∇e2 (f v) − he2, ∂f1if2∇e1 (f v)
= ∇f1f2(e1◦e2)+he1,∂f2if1e2−he2,∂f1if2e1 (f v)
= ∇f1e1◦f2e2 (f v)
The proof is finished.
Obviously V with the restriction of ∇ to L ⊆ L is still an H-representation of L, we still denote
it by (V, ∇).
In the following, we always assume that
f ∈ S•(Z), e ∈ L, x ∈ L, h ∈ H, α ∈ H.
Theorem 5.3. The H-standard complex of L with coefficients in V is isomorphic to the H-standard
complex of L with coefficients in V, i.e.
C•(L, H, V) ∼= C•(L, H, V).
1 , d1) be C•(L, H, V), and (C•
Proof. For simplicity, let (C•
we can obtain an associated cochain in C n
obviously a cochain map.
Next, given any ω ∈ C n
for the degree 2 arguments, extend ω from H to H by S•(Z)-linearity;
for the degree 1 arguments, extend ω from L to L, from the last argument to the first argument
2 , d2) be C•(L, H, V). Given any η ∈ C n
2 ,
1 by restriction, denote this restriction map by ψ. ψ is
1 , we can extend it to a cochain ϕω ∈ C n
2 as follows:
one by one, by the equation of weak S•(Z)-linearity:
(ϕω)k(e1, · · · ea−1, f ea, xa+1, · · · xn−2k; α1, · · · αk)
= f (ϕω)k(e1, · · · ea−1, ea, xa+1, · · · xn−2k; α1, · · · αk)
(−1)b−ahea, xbi(ϕω)k+1(e1, · · · ea−1, bea, xa+1, · · · bxb, · · · xn−2k; ∂f, α1, · · · αk).
+X
b>a
The proof that ϕω is a cochain in C n
Obviously, ψ ◦ ϕ = idC •
1 , ϕ ◦ ψ = idC •
2 is left to the lemma below 5.4.
2 . And ϕ is also a cochain map:
ϕ(d1ω) = ϕ(d1(ψ(ϕω))) = ϕ(ψ(d2(ϕω))) = d2(ϕω), ∀ω ∈ C•
1
The proof is finished.
Lemma 5.4. η , ϕω as defined above is a cochain in C n
2 .
H-STANDARD COHOMOLOGY FOR COURANT-DORFMAN ALGEBRAS AND LEIBNIZ ALGEBRAS
26
Proof. η is S•(Z)-linear in the arguments of H by definition. So we only need to prove weak
skew-symmetricity and weak S•(Z)-linearity in the arguments of L.
Proof of weak skew-symmetricity:
(5.2)
ηk(x1, · · · xa, xa+1 · · · xn−2k; α1, · · · αk) + ηk(x1, · · · xa+1, xa · · · xn−2k; · · · )
= −ηk+1(x1, · · ·cxa, [xa+1, · · · xn−2k; ∂hxa, xa+1i, α1, · · · αk).
Suppose xb = fbeb, ∀b. First we prove Equation 5.2 for the case when x1, · · · xa−1 ∈ L:
(5.3)
ηk(e1, · · · ea−1, xa, xa+1, · · · xn−2k; · · · ) + ηk(e1, · · · ea−1, xa+1, xa, · · · xn−2k; · · · )
= faηk(e1, · · · ea, fa+1ea+1, · · · xn−2k; · · · ) − hea, fa+1ea+1iηk+1(· · · bea, dea+1, · · · ; ∂fa, · · · )
(−1)b+ahea, xbiηk+1(· · · bea, fa+1ea+1, · · · bxb, · · · ; ∂fa, · · · )
+ X
b>a+1
+fa+1ηk(· · · ea+1, faea, · · · ; · · · ) − hea+1, faeaiηk+1(· · · dea+1, bea, · · · ; ∂fa+1, · · · )
+ X
(−1)b+ahea+1, xbiηk+1(· · · dea+1, faea, · · · bxb, · · · ; ∂fa+1, · · · )
b>a+1
= fafa+1ηk(. . . ea, ea+1, · · · ) + fa X
(−1)b+a+1hea+1, xbiηk+1(· · · ea, dea+1, · · · bxb, · · · ; ∂fa+1, · · · )
b>a+1
b>a+1
a+1<b<c
a+1<c<b
(−1)b+ahea, xbiηk+1(· · · bea, ea+1, · · · bxb, · · · ; ∂fa, · · · )
(−1)c+a(−1)b+a+1hea, xcihea+1, xbiηk+2(· · · bea, dea+1, · · · bxb, · · · bxc, · · · ; ∂fa, ∂fa+1, · · · )
(−1)c+a(−1)b+ahea, xcihea+1, xbiηk+2(· · · bea, dea+1, · · · bxc, · · · bxb, · · · ; ∂fa, ∂fa+1, · · · )
(−1)b+a+1hea, xbiηk+1(· · · ea+1, bea, · · · bxb, · · · ; ∂fa, · · · )
+fa+1 X
+ X
+ X
+fafa+1ηk(· · · ea+1, ea, · · · ) + fa+1 X
+fa X
(−1)b+ahea+1, xbiηk+1(· · · dea+1, ea, · · · bxb, · · · ; ∂fa+1, · · · )
+ X
(−1)c+a(−1)b+a+1hea+1, xcihea, xbiηk+2(· · · dea+1, bea, · · · bxb, · · · bxc, · · · ; ∂fa, ∂fa+1, · · · )
+ X
(−1)c+a(−1)b+ahea+1, xcihea, xbiηk+2(· · · dea+1, bea, · · · bxc, · · · bxb, · · · ; ∂fa, ∂fa+1, · · · )
−(cid:0)fa+1hea, ea+1iηk+1(· · · bea, dea+1, · · · ; ∂fa, · · · ) + fahea+1, eaiηk+1(· · · dea+1, bea, · · · ; ∂fa+1, · · · )(cid:1)
−(cid:0)fa+1hea, ea+1iηk+1(· · · bea, dea+1, · · · ; ∂fa, · · · ) + fahea+1, eaiηk+1(· · · dea+1, bea, · · · ; ∂fa+1, · · · )(cid:1)
= fafa+1(cid:0)ηk(. . . , ea, ea+1, · · · ; · · · ) + ηk(. . . , ea+1, ea, · · · ; · · · )(cid:1)
a+1<c<b
b>a+1
a+1<b<c
b>a+1
= −ηk+1(· · · bea, dea+1, · · · , xn−2k; ∂hfaea, fa+1ea+1i, · · · )
We will prove Equation 5.2 by mathematical induction.
If n = 2l is even, when k = l − 1, Equation 5.2 is equivalent to 5.3.
If n = 2l + 1 is odd, when k = l − 1, 5.2 is the combination of 5.3 and the following:
H-STANDARD COHOMOLOGY FOR COURANT-DORFMAN ALGEBRAS AND LEIBNIZ ALGEBRAS
27
ηl−1(f1e1, f2e2, f3e3; α1, · · · αl−1) + ηl−1(f1e1, f3e3, f2e2; α1, · · · αl−1)
= f1ηl−1(e1, f2e2, f3e3; · · · ) − he1, f2e2iηl(f3e3; ∂f1, · · · ) + he1, f3e3iηl(f2e2; ∂f1, · · · )
+f1ηl−1(e1, f3e3, f2e2; · · · ) − he1, f3e3iηl(f2e2; ∂f1, · · · ) + he1, f2e2iηl(f3e3; ∂f1, · · · )
= −f1ηl(e1; ∂hf2e2, f3e3i, · · · )
= −ηl(f1e1; ∂hf2e2, f3e3i, · · · ).
Now suppose Equation 5.2 holds for k > m, consider the case when k = m. By Equation 5.3,
we can further suppose that 5.2 holds for x1, · · · xi ∈ L, i < a. We will prove 5.2 for the case when
k = m and x1, · · · , xi−1 ∈ L:
ηm(e1, · · · ei−1, fiei, · · · xa, xa+1, · · · ; · · · ) + ηm(e1, · · · ei−1, fiei, · · · xa+1, xa, · · · ; · · · )
= fiηm(· · · ei, · · · xa, xa+1, · · · ) + X
fiηm(· · · ei, · · · xa+1, xa, · · · ) + X
b>i6=a,a+1
b>i6=a,a+1
(−1)b−ihei, xbiηm+1(· · · bei, · · · bxb, · · · xa, xa+1, · · · ; ∂fi · · · )
(−1)b−ihei, xbiηm+1(· · · bei, · · · bxb, · · · xa+1, xa, · · · ; ∂fi · · · )
+((−1)a−i + (−1)a+1−i)hei, xaiηm+1(· · · bei, · · ·cxa, xa+1, · · · ; ∂fi · · · )
+((−1)a+1−i + (−1)a−i)hei, xa+1iηm+1(· · · bei, · · · xa, [xa+1, · · · ; ∂fi · · · )
= fi(cid:0)ηm(· · · ei, · · · xa, xa+1, · · · ; · · · ) + ηm(· · · ei, · · · xa+1, xa, · · · ; · · · )(cid:1)
+(cid:0) X
(−1)b−ihei, xbiηm+1(· · · bei, · · · bxb, · · · xa, xa+1, · · · ; ∂fi · · · )
+ X
(−1)b−ihei, xbiηm+1(· · · bei, · · · bxb, · · · xa+1, xa, · · · ; ∂fi · · · )(cid:1)
b>i6=a,a+1
b>i6=a,a+1
= −fiηm+1(· · · ei, · · ·cxa, [xa+1, · · · ; ∂hxa, xa+1i, · · · )
(−1)b−ihei, xbiηm+2(· · · bei, · · · bxb, · · ·cxa, [xa+1, · · · ; ∂hxa, xa+1i, ∂fi · · · )
− X
b>i6=a,a+1
= −ηm+1(e1, · · · ei−1, fiei, · · ·cxa, [xa+1, · · · ; ∂hxa, xa+1i, · · · )
By induction, 5.2 is proved.
Proof of weak S•(Z)-linearity:
(5.4)
ηk(x1, · · · xi−1, f xi, · · · xn−2k; α1, · · · αk)
= f ηk(· · · xi, · · · ; α1, · · · αk) + X
When x1, · · · , xi−1 ∈ L,5.4 holds:
a>i
(−1)a−ihxi, xaiηk+1(· · · bxi, · · ·cxa, · · · ; ∂f, α1, · · · αk)
ηk(e1, · · · , ei−1, f fiei, · · · ; · · · )
= f fiηk(· · · ei, · · · ; · · · ) + X
= f (ηk(· · · fiei, · · · ; · · · ) − X
a>i
(−1)a−ihei, xaiηk+1(· · · bxi, · · ·cxa, · · · ; ∂(f fi), · · · )
(−1)a−ihei, xaiηk+1(· · · bxi, · · ·cxa, · · · ; ∂fi, · · · ))
a>i
+X
a>i
(−1)a−ihei, xaiηk+1(· · · bxi, · · ·cxa, · · · ; ∂(f fi), · · · )
H-STANDARD COHOMOLOGY FOR COURANT-DORFMAN ALGEBRAS AND LEIBNIZ ALGEBRAS
28
= f ηk(· · · fiei, · · · ; · · · ) + X
a>i
(−1)a−ihfiei, xaiηk+1(· · · bxi, · · ·cxa, · · · ; ∂f, · · · )
Now suppose 5.4 holds for any k > m, and for the case when x1, · · · , xj ∈ L(j < i), k = m as
well. Consider the case when x1, · · · , xj−1 ∈ L, k = m:
ηk(e1, · · · ej−1, fjej, · · · , f xi, · · · ; · · · )
= fjηk(· · · ej, · · · f xi, · · · ; · · · ) + (−1)i−jhej, f xiiηk+1(· · · bej, · · · bxi, · · · ; ∂fj, · · · )
(−1)b+jhej, xbiηk+1(· · · bej, · · · bxb, · · · , f xi, · · · ; ∂fj, · · · )
+ X
b>j,b6=i
= fj(cid:0)f ηk(· · · ej, · · · xi, · · · ; · · · ) + X
(−1)a+ihxi, xaiηk+1(· · · ej, · · · bxi, · · ·cxa, · · · ; ∂f, · · · )(cid:1)
a>i
a>i
j<b<i
+ X
(−1)b+jhej, xbi(cid:0)f ηk+1(· · · bej, · · · bxb, · · · , xi, · · · ; ∂fj, · · · )
+X
(−1)a+ihxi, xaiηk+2(· · · bej, · · · bxb, · · · bxi, · · · ,cxa, · · · ; ∂f, ∂fj, · · · )(cid:1)
+X
(−1)b+jhej, xbi(cid:0)f ηk+1(· · · bej, · · · , xi, · · · bxb, · · · ; ∂fj, · · · )
+ X
+X
(−1)a+ihxi, xaiηk+2(· · · bej, · · · bxi, · · ·cxa, · · · bxb, · · · ; ∂f, ∂fj, · · · )
(−1)a+i+1hxi, xaiηk+2(· · · bej, · · · bxi, · · · bxb, · · ·cxa, · · · ; ∂f, ∂fj, · · · )(cid:1)
i<a<b
b<a
b>i
b>i
j<b<i
= f(cid:0)fjηk(· · · ej, · · · , xi, · · · ; · · · ) + (−1)i+jhej, xiiηk+1(· · · bej, · · · bxi, · · · ; ∂fj, · · · )
+(−1)i+jhej, f xiiηk+1(· · · bej, · · · bxi, · · · ; ∂fj, · · · )
+ X
(−1)b+jhej, xbiηk+1(· · · bej, · · · bxb, · · · , xi, · · · ; ∂fj, · · · )
+X
(−1)b+jhej, xbiηk+1(· · · bej, · · · xi, · · · bxb, · · · ; ∂fj, · · · )(cid:1)
+X
(−1)a+ihxi, xai(cid:0)fjηk+1(· · · bej, · · · bxi, · · ·cxa, · · · ; ∂f, · · · )
+ X
+ X
+X
(−1)b+jhej, xbiηk+2(· · · ej, · · · bxb, · · · bxi, · · · xa, · · · ; ∂f, ∂fj, · · · )
(−1)b+j+1hej, xbiηk+2(· · · bej, · · · bxi, · · · bxb, · · ·cxa, · · · ; ∂f, ∂fj, · · · )
(−1)b+jhej, xbiηk+2(· · · bej, · · · bxi, · · ·cxa, · · · bxb, · · · ; ∂f, ∂fj, · · · )(cid:1)
i<b<a
j<b<i
b>a
a>i
= f ηk(· · · fjej, · · · xi, · · · ; · · · ) + X
(−1)a−ihxi, xaiηk+1(· · · fjej, · · · bxi, · · · ,cxa, · · · ; ∂f, · · · )
a>i
By mathematical induction, Equation 5.4 holds for any k.
The proof is finished.
It is easily proved that the Chevalley-Eilenberg complex of the Lie algebra L/H with coefficients
in V is isomorphic to the Chevalley-Eilenberg complex of the Lie-Rinehart algebra L/H with
H-STANDARD COHOMOLOGY FOR COURANT-DORFMAN ALGEBRAS AND LEIBNIZ ALGEBRAS
29
coefficients in V. Whence, combining Theorem 5.3 and 4.1 (the quotient L/H ∼= S•(Z) ⊗ (L/H)
is a free module), we have the following:
Corollary 5.5. With the notations above,
H •(L, H, V) ∼= H •
CE(L/H, V).
Actually this result is true for the H-representation (V, τ ):
Theorem 5.6. With the notations above,
H •(L, H, V ) ∼= H •
CE(L/H, V ).
Proof. Consider a subspace of C•(L, H, V):
0 (L, H, V) , M
C•
{ω ∈ C n(L, H, V)ωk(e1, · · · en−2k; h1, · · · hk) ∈ V, ∀k, ∀e ∈ L, h ∈ H}.
n
It is easily checked from the definition of d = d0 + δ + d′ that C•
Given any ω ∈ C•
0 (L, H, V) satisfying the condition of Lemma 4.3, we see that λ and η as
0 (L, H, V). Then by similar arguments to the
0 (L, H, V) is a subcomplex.
constructed in the proof of this lemma are both in C•
proof of Theorem 4.1,
H •
0 (L, H, V) , H •(C•
0 (L, H, V), d)
is isomorphic to the cohomology of the following subcomplex of C•(L, H, V)
nv 0(L, H, V) , M
C•
{ω ∈ C n(L, H, V)ωk = 0, ∀k ≥ 1, ιαω0 = 0, ∀α ∈ H,
n
ω0(e1, · · · en) ∈ V, ∀e ∈ L},
which is again isomorphic to H •
CE(L/H, V ).
On the other hand, in the proof of Theorem 5.3, if we restrict ψ from C•(L, H, V) to C•
0 (L, H, V)
and ϕ from C•(L, H, V) to C•(L, H, V ), we can get mutually invertible cochain maps between
C•(L, H, V ) and C•
0 (L, H, V).
Thus
H •(L, H, V) ∼= H •
0 (L, H, V) ∼= H •
CE(L/H, V).
Example 5.7. If L is the omni-Lie algebra gl(V ) ⊕ V , V is the only isotropic ideal of L containing
the left center V , and (V, τ ) is a V -representation with τ being the standard action of gl(V ) on V .
As introduced by Weinstein [16], the omni Lie algebra gl(V ) ⊕ V can be viewed as the linearization
of the standard Courant algebroid T V ∗ ⊕ T ∗V ∗, where gl(V ) is identified with the space of linear
vector fields and V is identified with the space of constant 1-forms. If we ignore the difference
between S•(V ) and C∞(V ∗), the crossed product L as constructed in Proposition 5.1 can be
viewed as a Courant-Dorfman subalgebra of Γ(T V ∗ ⊕ T ∗V ∗), in the sense that L consists of all
polynomial vector fields (excluding constant ones) and polynomial 1-forms. By Theorem 5.3 and
Corollary 5.5, the standard cohomology of L is isomorphic to the cohomology of Lie algebra gl(V )
with coefficients in S•(V ), which is trivial. Whence, although L is different from Γ(T V ∗ ⊕ T ∗V ∗),
the standard cohomology of them are both trivial.
H-STANDARD COHOMOLOGY FOR COURANT-DORFMAN ALGEBRAS AND LEIBNIZ ALGEBRAS
30
References
[1] Barnes, D.W., Some theorems on Leibniz algebras, Communications in Algebra 39.7 (2011), 2463-2472.
[2] Bloh, A., On a generalization of the concept of Lie algebra, Dokl. Akad. Nauk SSSR. Vol. 165. No. 3 (1965),
471-473.
[3] Chen, Z., Stiénon, M. and Xu, P., On regular Courant algebroids, Journal of Symplectic Geometry 11.1 (2013),
1-24.
[4] Covez, S., The local integration of Leibniz algebras, Annales de l'institut Fourier. Association des Annales de
l'institut Fourier, 63(1)(2013), 1-35.
[5] Ginot, G. and Grützmann M., Cohomology of Courant algebroids with split base, Journal of Symplectic Geom-
etry 7.3 (2009), 311-335.
[6] Liu, Z.J., Weinstein A. and Xu P., Manin triples for Lie bialgebroids, J. Differential Geom 45.3 (1997), 547-574.
[7] Loday, J.L., Une version non commutative des algèbres de Lie, L'Ens. Math. (2), 39 (1993), 269-293.
[8] Loday, J.L. and Pirashvili T., Universal enveloping algebras of Leibniz algebras and (co) homology, Mathema-
tische Annalen 296.1 (1993), 139-158.
[9] Patsourakos, A., On Nilpotent Properties of Leibniz Algebras, Communications in Algebra. 35 (12)(2007),
3828 -- 3834.
[10] Roytenberg, D., Courant algebroids, derived brackets and even symplectic supermanifolds, Ph.D. thesis, Uni-
versity of California, Berkeley, 1999.
[11] Roytenberg, D., Courant -- Dorfman algebras and their cohomology, Letters in Mathematical Physics 90.1-3
(2009), 311-351.
[12] Roytenberg, D., On the structure of graded symplectic supermanifolds and Courant algebroids, Contemporary
Mathematics 315 (2002), 169-186.
[13] Sheng, Y.H. and Liu Z.J., Leibniz 2-algebras and twisted Courant algebroids, Communications in Algebra 41.5
(2013), 1929-1953.
[14] Sheng, Y.H. and Liu Z.J., From Leibniz Algebras to Lie 2-algebras, Algebras and Representation Theory 19.1
(2016), 1-5.
[15] Stiénon, M. and Xu P., Modular classes of Loday algebroids, Comptes Rendus Mathematique 346.3 (2008),
193-198.
[16] Weinstein, A., Omni-Lie Algebras, arXiv preprint math.RT/9912190.
Mathematics Research Unit, FSTC, University of Luxembourg, Luxembourg
E-mail address: [email protected]
|
0802.1872 | 2 | 0802 | 2015-03-20T14:07:01 | The realization problem for von Neumann regular rings | [
"math.RA"
] | We survey recent progress on the realization problem for von Neumann regular rings, which asks whether every countable conical refinement monoid can be realized as the monoid of isoclasses of finitely generated projective right $R$-modules over a von Neumann regular ring $R$. | math.RA | math |
September 25, 2018 0:19
WSPC - Proceedings Trim Size: 9in x 6in
realizationWS
1
THE REALIZATION PROBLEM FOR VON NEUMANN
REGULAR RINGS
PERE ARA
Departament de Matem`atiques, Universitat Aut`onoma de Barcelona,
08193, Bellaterra (Barcelona), Spain
E-mail: [email protected]
We survey recent progress on the realization problem for von Neumann regular rings, which asks
whether every countable conical refinement monoid can be realized as the monoid of isoclasses of
finitely generated projective right R-modules over a von Neumann regular ring R.
Keywords: von Neumann regular ring; Leavitt path algebra; refinement monoid.
This survey consists of four sections. Section 1 introduces the realization problem for von
Neumann regular rings, and points out its relationship with the separativity problem of [7].
Section 2 surveys positive realization results for countable conical refinement monoids, including
the recent constructions in [5] and [4]. We analyze in Section 3 the relationship with the
realization problem of algebraic distributive lattices as lattices of two-sided ideals over von
Neumann regular rings. Finally we observe in Section 4 that there are countable conical monoids
which can be realized by a von Neumann regular K-algebra for some countable field K, but
they cannot be realized by a von Neumann regular F -algebra for any uncountable field F .
1. The problem
All rings considered in this paper will be associative, and all the monoids will be commutative.
For a unital ring R, let V(R) denote the monoid of isomorphism classes of finitely generated
projective right R-modules, where the operation is defined by
[P ] + [Q] = [P ⊕ Q].
This monoid describes faithfully the decomposition structure of finitely generated projective
modules. The monoid V(R) is always a conical monoid, that is, whenever x + y = 0, we have
x = y = 0. Recall that an order-unit in a monoid M is an element u in M such that for every
x ∈ M there is y ∈ M and n ≥ 1 such that x+y = nu. Observe that [R] is a canonical order-unit
September 25, 2018 0:19
WSPC - Proceedings Trim Size: 9in x 6in
realizationWS
2
in V(R). By results of Bergman [11, Theorems 6.2 and 6.4] and Bergman and Dicks [12, page
315], any conical monoid with an order-unit appears as V(R) for some unital hereditary ring
R.
A monoid M is said to be a refinement monoid in case any equality x1 + x2 = y1 + y2 admits
a refinement, that is, there are zij, 1 ≤ i, j ≤ 2 such that xi = zi1 + zi2 and yj = z1j + z2j for
all i, j, see e.g. [8]. If R is a von Neumann regular ring, then the monoid V(R) is a refinement
monoid by [20, Theorem 2.8].
The following is still an open problem:
R1. Realization Problem for von Neumann Regular Rings Is every countable conical
refinement monoid realizable by a von Neumann regular ring?
A related problem was posed by K.R. Goodearl in [22]:
FUNDAMENTAL OPEN PROBLEM Which monoids arise as V(R)'s for a von Neumann
regular ring R?
It was shown by Wehrung in [35] that there are conical refinement monoids of size ℵ2 which
cannot be realized. If the size of the monoid is ℵ1 the question is open. Wehrung's approach
is related to Dilworth's Congruence Lattice Problem (CLP), see Section 3. A solution to the
latter problem has recently appeared in [42].
Problem R1 is related to the separativity problem. A class C of modules is called separative
if for all A, B ∈ C we have
A ⊕ A ∼= A ⊕ B ∼= B ⊕ B =⇒ A ∼= B.
A ring R is separative if the class F P (R) of all finitely generated projective right R-modules is
a separative class. Separativity is an old concept in semigroup theory, see [16]. A commutative
semigroup S is called separative if for all a, b ∈ S we have a + a = a + b = b + b =⇒ a = b. An
alternative characterization is that a commutative semigroup is separative if and only if it can
be embedded in a product of monoids of the form G⊔{∞}, where G is an abelian group. Clearly
a ring R is separative if and only if V(R) is a separative semigroup. Separativity provides a
key to a number of outstanding cancellation problems for finitely generated projective modules
over exchange rings, see [7].
Outside the class of exchange rings, separativity can easily fail. In fact it is easy to see that
a commutative ring R is separative if and only if V(R) is cancellative. Among exchange rings,
however, separativity seems to be the norm. It is not known whether there are non-separative
exchange rings. This is one of the major open problems in this area. See [3] for some classes of
exchange rings which are known to be separative. We single out the problem for von Neumann
regular rings. (Recall that every von Neumann regular ring is an exchange ring.)
SP. Is every von Neumann regular ring separative?
September 25, 2018 0:19
WSPC - Proceedings Trim Size: 9in x 6in
realizationWS
3
We have (R1 has positive answer ) =⇒ (SP has a negative answer ). To explain why we
have to recall results of Bergman and Wehrung concerning existence of countable non-separative
conical refinement monoids.
Recall that every monoid M is endowed with a natural pre-order, the so-called algebraic
pre-order, by x ≤ y iff there is z ∈ M such that y = x+z. This is the only order on monoids that
we will consider in this paper. A monoid homomorphism f : M → M ′ is an order-embedding in
case f is one-to-one and, for x, y ∈ M , we have x ≤ y if and only if f (x) ≤ f (y).
Proposition 1.1. (cf. [36]) Let M be a countable conical monoid. Then there is an order-
embedding of M into a countable conical refinement monoid.
Let us apply the above Proposition to the conical monoid M generated by a with the only
relation 2a = 3a. Then
a + a = a + (2a) = (2a) + (2a)
but a 6= 2a in M . By Proposition 1.1 there exists an order-embedding M → M ′, where M ′ is
a countable conical refinement monoid, and M ′ cannot be separative.
Thus if R1 is true we can represent M ′ as V(R) for some von Neumann regular ring and R
will be non-separative.
2. Known cases
It turns out that only a few cases of R1 are known. In this section I will describe the positive
realization results of which I am aware.
The first realization result is by now a classical one. Recall that a monoid M is said to be
unperforated if, for x, y ∈ M and n ≥ 1, the relation nx ≤ ny implies that x ≤ y. A dimension
monoid is a cancellative, refinement, unperforated conical monoid. These are the positive cones
of the dimension groups [20, Chapter 15]. Recall that, by definition, an ultramatricial K-algebra
R is a direct limit of a sequence of finite direct products of matrix algebras over K. Clearly
every ultramatricial algebra is von Neumann regular.
Theorem 2.1. ([18], [19, Theorem 3.17], [20, Theorem 15.24(b)] ) If M is a countable di-
mension monoid and K is any field, then there exists an ultramatricial K-algebra R such that
V(R) ∼= M .
A K-algebra is said to be locally matricial in case it is a direct limit of a directed system
of finite direct products of matrix algebras over K, see [23, Section 1]. It was proved in [23,
Theorem 1.5] that if M is a dimension monoid of size ≤ ℵ1, then it can be realized as V(R)
for a locally matricial K-algebra R. Wehrung constructed in [35] dimension monoids of size
ℵ2 which cannot be realized by regular rings. Indeed the monoids constructed in [35] are the
positive cones of dimension groups which are vector spaces over Q. A refinement of the method
September 25, 2018 0:19
WSPC - Proceedings Trim Size: 9in x 6in
realizationWS
4
used in [35] gave a dimension monoid counterexample of size ℵ2 with an order unit of index
two [41], thus answering a question posed by Goodearl in [21].
Another realization result was obtained by Goodearl, Pardo and the author in [6].
Theorem 2.2.
there is a purely infinite simple regular K-algebra R such that K0(R) ∼= G.
[6, Theorem 8.4]. Let G be a countable abelian group and K any field. Then
Recall that a simple ring R is purely infinite in case it is not a division ring and, for every
nonzero element a ∈ R there are x, y ∈ R such that xay = 1 (see [6, Section 1], especially
Theorem 1.6). Since V(R) = K0(R) ⊔ {0} for a purely infinite simple regular ring [6, Corollary
2.2], we get that all monoids of the form G ⊔ {0}, where G is a countable abelian group, can
be realized.
As Fred Wehrung has kindly pointed out to me, another class of conical refinement monoids
which can be realized by von Neumann regular rings is the class of continuous dimension scales,
see [28, Chapter 3]. All the monoids in this class satisfy the property that every bounded subset
has a supremum, as well as some additional axioms, see [28, Definition 3-1.1]. Indeed, if M is a
commutative monoid, then M ∼= V(R) for some regular, right self-injective ring R if and only if
M is a continuous dimension scale with order-unit [28, Corollary 5-3.15]. These monoids have
unrestricted cardinality, indeed they are usually quite large.
A recent realization result has been obtained by Brustenga and the author in [5]. As we
will note later, the two results mentioned above (Theorem 2.1 and Theorem 2.2) can be seen as
particular cases of the main result in [5]. Before we proceed with the statement of this result,
and in order to put it in the right setting, we need to recall a few monoid theoretic concepts.
Definition 2.1. Let M be a monoid. An element p ∈ M is prime if for all a1, a2 ∈ M ,
p ≤ a1 + a2 implies p ≤ a1 or p ≤ a2. A monoid is primely generated if each of its elements is
a sum of primes.
Proposition 2.1. [14, Corollary 6.8] Any finitely generated refinement monoid is primely
generated.
We have the following particular case of question R1.
R2. Realization Problem for finitely generated refinement monoids: Is every finitely
generated conical refinement monoid realizable by a von Neumann regular ring?
We conjecture that R2 has a positive answer. Our main tool to realize a large class of
finitely generated refinement monoids is the consideration of some regular algebras associated
with quivers.
Recall that a quiver (= directed graph) consists of a 'vertex set' E0, an 'edge set' E1,
together with maps r and s from E1 to E0 describing, respectively, the range and source of
September 25, 2018 0:19
WSPC - Proceedings Trim Size: 9in x 6in
realizationWS
edges. A quiver E = (E0, E1, r, s) is said to be row-finite in case, for each vertex v, the set
s−1(v) of arrows with source v is finite. For a row-finite quiver E, the graph monoid M (E) of
E is defined as the quotient monoid of F = FE, the free abelian monoid with basis E0 modulo
the congruence generated by the relations
5
v = X
r(e)
{e∈E 1s(e)=v}
for every vertex v ∈ E0 which emits arrows (that is s−1(v) 6= ∅).
It follows from Proposition 2.1 and [8, Proposition 4.4] that, for a finite quiver E, the monoid
M (E) is primely generated. Note that this is not always the case for a general row-finite graph
E. An example is provided by the graph:
p0
a
p1
p2
t✐✐✐✐✐✐✐✐✐✐✐✐✐✐✐✐✐✐✐✐✐✐✐
v♠♠♠♠♠♠♠♠♠♠♠♠♠♠♠
}⑤⑤⑤⑤⑤⑤⑤⑤
p3
· · ·
The corresponding monoid M has generators a, p0, p1, . . . and relations given by pi = pi+1 +
a for all i ≥ 0. One can easily see that the only prime element in M is a, so that M is not
primely generated.
Now we have the following result of Brookfield:
Theorem 2.3.
ment monoid. Then M is separative.
[14, Theorem 4.5 and Corollary 5.11(5)] Let M be a primely generated refine-
In fact, primely generated refinement monoids enjoy many other nice properties, see [14]
and also [37].
It follows from Proposition 2.1 and Theorem 2.3 that a finitely generated refinement monoid
is separative. In particular all the monoids associated to finite quivers are separative. For a row-
finite quiver E the result follows by using the fact that the monoid M (E) is the direct limit of
monoids associated to certain finite subgraphs of E, see [8, Lemma 2.4].
Theorem 2.4. ( [5, Theorem 4.2, Theorem 4.4]) Let M (E) be the monoid corresponding to
a finite quiver E and let K be any field. Then there exists a unital von Neumann regular
hereditary K-algebra QK(E) such that V(QK(E)) ∼= M (E). Furthermore, if E is a row-finite
quiver, then there exists a (not necessarily unital) von Neumann regular K-algebra QK(E) such
that V(QK(E)) ∼= ME.
Note that, due to unfortunate lack of convention in this area, the Cuntz-Krieger relations
used in [5] are the opposite to the ones used in [8], which are the ones we are following in
this survey, so that the result in [5, Theorem 4.4] is stated for column-finite quivers instead of
/
/
/
/
}
/
/
v
/
/
t
September 25, 2018 0:19
WSPC - Proceedings Trim Size: 9in x 6in
realizationWS
6
row-finite ones. The regular algebras QK(E) are related to the Leavitt path algebras LK(E) of
[1], [2], [8].
We now observe that Theorem 2.1 and Theorem 2.2 are particular cases of Theorem 2.4.
This follows from the fact that the monoids considered in these theorems are known to be
graph monoids M (E) for suitable quivers E. Indeed, taking into account [8, Theorem 3.5
and Theorem 7.1], we see that the case of dimension monoids follows from [31, Proposition
2.12] and the case of monoids of the form {0} ⊔ G, with G a countable abelian group, follows
from [34, Theorem 1.2]. As Pardo pointed out to me, the quiver E can be chosen to be finite
in case M = {0} ⊔ G for a finitely generated abelian group G. To see this, note that such a
monoid admits a presentation given by a finite number of generators a1, . . . , an, and relations
of the form ai = Pn
j=1 γjiaj, where all γji are strictly positive integers and γii ≥ 2 for all i.
The corresponding finite quiver will have γji arrows from the vertex i to the vertex j.
Now we would like to describe how this construction sheds light on problem R2. The answer
is completely known for the class of antisymmetric finitely generated refinement monoids. The
monoid M is said to be antisymmetric in case the algebraic pre-order is a partial order, that
is, in case x ≤ y and y ≤ x imply that x = y. Note that every antisymmetric monoid is conical.
We say that a monoid M is primitive if it is an antisymmetric primely generated refinement
monoid [30, Section 3.4]. A primitive monoid M is completely determined by its set of primes
P(M ) together with a transitive and antisymmetric relation ✁ on it, given by q ✁ p iff p + q = p.
Indeed given such a pair (P, ✁), the abelian monoid M (P, ✁) defined by taking as a set of
generators P and with relations given by p = p + q whenever q ✁ p, is a primitive monoid,
and the correspondences M 7→ (P(M ), ✁) and (P, ✁) 7→ M (P, ✁) give a bijection between
isomorphism types of primitive monoids and isomorphism types of pairs (P, ✁), where P is a
set and ✁ a transitive antisymmetric relation on P, see [30, Proposition 3.5.2].
Let M be a primitive monoid and p ∈ P(M ). Then p is said to be free in case p ⋪ p.
Otherwise p is regular, see [10, Section 2]. So giving a primitive monoid is equivalent to giving
a poset (P, ≤) which is a disjoint union of two subsets: P = Pfree ⊔ Preg. If M is a finitely
generated primitive monoid then P(M ) is a finite set, indeed P(M ) is the minimal generating
set of M .
We can now describe the finitely generated primitive monoids which are graph monoids.
Recall that a lower cover of an element p of a poset P is an element q in P such that q < p and
[q, p] = {q, p}. The set of lower covers of p in P(M ) is denoted by L(M, p), and Lfree(M, p) and
Lreg(M, p) denote the sets of free and regular elements in L(M, p) respectively.
Theorem 2.5. (cf. [10, Theorem 5.1]) Let M be a finitely generated primitive monoid. Then
the following statements are equivalent:
(1) M is a graph monoid.
(2) M is a direct limit of graph monoids.
(3) Lfree(M, p) ≤ 1 for each p ∈ Pfree(M ).
September 25, 2018 0:19
WSPC - Proceedings Trim Size: 9in x 6in
realizationWS
7
For the monoids as in the statement there is a hereditary, von Neumann regular ring Q(E)
such that V(Q(E)) = ME = M (Theorem 2.4). It is worth mentioning that, in some cases, an
infinite quiver is required in Theorem 2.5. A slightly more general result is indeed presented
in [10, Theorem 5.1]. Namely the same characterization holds when M is a primitive monoid
such that L(M, p) is finite for every p in P(M ).
In view of Theorem 2.5, the simplest primitive monoid which is not a graph monoid is the
monoid
M = hp, a, b p = p + a = p + bi.
a
p
❂❂❂❂❂❂❂
b
Fig. 1. The poset P(M ) for the monoid M = hp, a, b p = p + a = p + bi.
In this example P(M ) = Pfree(M ) = {p, a, b}, and p has two free lower covers a, b. Thus,
by Theorem 2.5, the monoid M is not even a direct limit of graph monoids (with monoid
homomorphisms as connecting maps). However M can be realized as the monoid of a suitable
von Neumann regular ring, as follows. Fix a field K and consider two indeterminates t1, t2 over
K. We consider the regular algebra QK(t2)(S1) over the quiver S1 with two vertices v0,1, v1,1
and two arrows e1, f1 such that r(e1) = s(e1) = v1,1 = s(f1) and r(f1) = v0,1. The picture of
S1 is as shown in Figure 2.
v1,1
v0,1
Fig. 2. The graph S1.
The algebra Q1 := QK(t2)(S1) has a unique non-trivial (two-sided) ideal M1, which coincides
with its socle, so that we get an extension of rings:
0 −−−−→ M1 −−−−→ Q1
π1−−−−→ K(t2)(t1) = K(t1, t2) −−−−→ 0
However the element t1 does not lift to a unit in Q1, rather there are z1, z1 in Q1 such that
z1z1 = 1, but z1z1 6= 1, and π1(z1) = t1. Some additional information on the algebra QF (S1),
September 25, 2018 0:19
WSPC - Proceedings Trim Size: 9in x 6in
realizationWS
8
where F denotes an arbitrary field, can be found in [5, Examples 4.3].
Let S2 be a copy of S1, now with vertices labelled as v0,2, v1,2 and arrows labelled as e2, f2,
and set Q2 = QK(t1)(S2). There is a corresponding diagram
0 −−−−→ M2 −−−−→ Q2
π2−−−−→ K(t1)(t2) = K(t1, t2) −−−−→ 0
Let P be the pullback of the maps π1 and π2, so that P fits in the following commutative
square:
P
ρ2y
Q2
ρ1−−−−→ Q1
y
π1
π2−−−−→ K(t1, t2)
(1)
Then P is a von Neumann regular ring and V(P ) = M , see [4]. Indeed a wide generalization
of this method gives the following realization result:
Theorem 2.6. ( [4, Theorem 2.2]) Let M be a finitely generated primitive monoid such that
all primes of M are free and let K be a field. Then there is a unital regular K-algebra QK(M )
such that V(QK(M )) ∼= M .
Moreover both the regular algebras associated with quivers [5] and the regular algebras
constructed in [4] are given explicitly in terms of generators and relations (including universal
localization [33]).
Recall that a monoid M is strongly separative in case a + a = a + b implies a = b for
a, b ∈ M . A ring R is said to be strongly separative in case V(R) is a strongly separative
monoid, see [7] for background and various equivalent conditions. As we mentioned above,
every primely generated refinement monoid is separative [14, Theorem 4.5]. In particular every
primitive monoid is separative. Moreover, a primitive monoid M is strongly separative if and
only if all the primes in M are free, see [14, Theorem 4.5, Corollary 5.9]. Thus, the class
of monoids covered by Theorem 2.6 coincides exactly with the strongly separative finitely
generated primitive monoids. The case of a general finitely generated primitive monoid remains
open, although it seems amenable to analysis in the light of [5] and [4].
3. Realizing distributive lattices
Let R be a regular ring. Then the lattice Id(R) of all (two-sided) ideals of R is an algebraic
distributive lattice. Here an algebraic lattice means a complete lattice such that each element
is the supremum of all the compact elements below it. The set of compact elements in Id(R) is
the set Idc(R) of all finitely generated ideals of R, and it is a distributive semilattice, see for
example [27]. Here a semilattice means a ∨-semilattice with least element 0. A semilattice is
the same as a monoid M such that x = x + x for every x in M , and a distributive semilattice
September 25, 2018 0:19
WSPC - Proceedings Trim Size: 9in x 6in
realizationWS
9
is just a semilattice satisfying the refinement axiom [27, Lemma 2.3]. Observe that, if M is a
semilattice then x ∨ y = x + y gives the supremum of x, y in M .
The famous Congruence Lattice Problem (CLP) asks whether an algebraic distributive lat-
tice is the congruence lattice of some lattice; equivalently, whether every distributive semilattice
is the semilattice of all the compact congruences of a lattice. This problem has been recently
solved in the negative by Fred Wehrung [42], who constructed for each ℵ ≥ ℵω+1 an algebraic
distributive lattice with ℵ compact elements such that it cannot be represented as the congru-
ence lattice of any lattice, see also [25]. His methods have been refined by Ruzicka [32] to cover
the case ℵ ≥ ℵ2. It is worth to remark that Wehrung had previously shown in [39] that every
algebraic distributive lattice with ≤ ℵ1 compact elements can be realized as the ideal lattice
for some von Neumann regular ring R, and thus is isomorphic to the congruence lattice of the
lattice L(RR) of principal right ideals of R [38, Corollary 4.4]. So the formulation of Ruzicka
is best possible (concerning the number of compact elements). The "hard core" of Wehrung's
proof in [39] is a ring-theoretic amalgamation result proved by P. M. Cohn in [17, Theorem
4.7].
One can ask: What is the relationship between the CLP, or more concretely, the represen-
tation problem of algebraic distributive lattices as lattices of ideals of regular rings, and our
problem R1? The answer is that, for a regular ring R, the lattice Id(R) is only a small piece
of information compared with the information contained in the monoid V(R), in the sense
that V(R) determines Id(R), but generally the structure of V(R) can be much more compli-
cated than the structure of Id(R), e.g. for simple rings. Indeed we have a lattice isomorphism
Id(R) ∼= Id(V(R)), where for a conical monoid M , Id(M ) is the lattice of all order-ideals of M ,
cf. [27, Proposition 7.3]. Recall that an order-ideal of M is a submonoid I of M with the prop-
erty that whenever x ≤ y in M and y ∈ I, we have x ∈ I. If L is an algebraic distributive lattice
which is not the congruence lattice of any lattice and M is any conical refinement monoid such
that Id(M ) ∼= L, then M cannot be realized as V(R) for a regular ring R. For every algebraic
distributive lattice L there is at least one such conical refinement monoid, namely the semi-
lattice Lc of compact elements of L, but we should expect a myriad of such monoids to exist.
Wehrung proved in [39] that if Lc ≤ ℵ1 then the semilattice Lc can be realized as V(R) where
R is a von Neumann regular ring, and he showed in [40] that there is a distributive semilattice
Sω1 of size ℵ1 which is not the semilattice of finitely generated, idempotent-generated ideals
of any exchange ring of finite stable rank. In particular there is no locally matricial K-algebra
A over a field K such that Idc(A) ∼= Sω1 ; see [40] for details. This contrasts with Bergman's
result [13] stating that every distributive semilattice of size ≤ ℵ0 is the semilattice of finitely
generated ideals of an ultramatricial K-algebra, for every field K.
Say that a subset A of a poset P is a lower subset in case q ≤ p and p ∈ A imply q ∈ A. The
set L(P) of all lower subsets of P forms an algebraic distributive lattice, which is a sublattice
of the Boolean lattice 2P. Now if L is a finite distributive lattice, then by a result of Birkhoff
September 25, 2018 0:19
WSPC - Proceedings Trim Size: 9in x 6in
realizationWS
10
( [24, Theorem II.1.9]) there is a finite poset P such that L is the lattice of all lower subsets
of P. In the case of a finite Boolean algebra 2n with n atoms, the poset P is just an antichain
with n points, and 2n = P(P) = L(P). Our construction in [4] gives a realization of L as the
ideal lattice of a regular K-algebra QK(P), where K is an arbitrary fixed field, such that the
monoid V(QK(P)) is the monoid M (P) associated with (P, <), with all elements in M (P) being
free, that is, M (P) is the abelian monoid with generators P and relations given by p = q + p
whenever q < p in P. Moreover QK(P) satisfies the following properties ( [4, Proposition 2.12,
Remark 2.13, Theorem 2.2]):
(a) There is a canonical family of commuting idempotents {e(A) : A ∈ L} such that
(i) e(A)e(B) = e(A ∩ B)
(ii) e(A) + e(B) − e(A)e(B) = e(A ∪ B)
(iii) e(∅) = 0 and e(P) = 1.
(iv) e(A)QK (P)e(A) ∼= QK(A).
(b) Let I(A) be the ideal of QK(P) generated by e(A). Then the assignment
A 7→ I(A)
defines a lattice isomorphism from L = L(P) onto Id(QK(M )).
(c) The map M (P) → V(QK(P)) given by p 7→ [e(P ↓ p)], for p ∈ P, is a monoid isomorphism.
Here P ↓ p = {q ∈ P : q ≤ p} is the lower subset of P generated by p.
The set Idem(R) of idempotents of a ring is a poset in a natural way, by using the order e ≤ f
iff e = f e = ef . This poset is a partial lattice, in the following sense: every two commuting
idempotents e and f have an infimum ef and a supremum e + f − ef in Idem(R). Say that a
map φ : L → Idem(R) from a lattice L to Idem(R) is a lattice homomorphism in case φ(x) and
φ(y) commute and φ(x ∨ y) = φ(x) ∨ φ(y) and φ(x ∧ y) = φ(x) ∧ φ(y), for every x, y ∈ L.
The above results can be paraphrased as follows: The canonical mapping Idem(QK(P)) →
Id(QK(P)) = L(P) sending each element e in Idem(QK(P)) to the ideal generated by e has a
distinguished section e : Id(QK(P)) → Idem(QK(P)), A 7→ e(A), which is a lattice homomor-
phism.
Write Q = QK(P). Observe that we have lattice isomorphisms
L = L(P) ∼= Id(Q) ∼= IdV(Q) ∼= V(Q)/≍.
Here V(Q)/≍ is the maximal semilattice quotient of the monoid V(Q), see [27, Section 2].
The finite distributive lattice L can be represented in many other ways as an ideal lattice of
a regular ring, for instance using ultramatricial algebras [13], but the monoids corresponding
to these ultramatricial algebras have little to do with M (P). Indeed as soon as P is not an
antichain we will have that V(R) is non-finitely generated for every ultramatricial algebra R
such that Id(R) ∼= L(P).
September 25, 2018 0:19
WSPC - Proceedings Trim Size: 9in x 6in
realizationWS
11
4. The dependence on the field
We like to work with von Neumann regular rings which are algebras over a field K. A natural
question is whether the field K plays any role concerning the realization problem. So we ask
the following variant of R1.
R(K). Realization Problem for von Neumann Regular K-algebras Let K be a fixed
field. Is every countable conical refinement monoid realizable by a von Neumann regular K-
algebra?
The answer to this question is known for uncountable fields, thanks to an observation due
to Wehrung. Indeed the basic counter-example comes from a construction due to Chuang and
Lee [15]. Their remarkable example gave a negative answer to five open questions in Goodearl's
book [20] on von Neumann regular rings.
We take this opportunity to present the complete argument, including a result of Goodearl,
generalizing Wehrung's observation, and a version of the Chuang-Lee construction.
We will need the notion of (pseudo-)rank function, as given in [20, Chapter 16]. Recall that
a pseudo-rank function N on a unital regular ring R is a function N : R → [0, 1] such that
(a) N (1) = 1.
(b) N (xy) ≤ N (x) and N (xy) ≤ N (y) for all x, y ∈ R.
(c) N (e + f ) = N (e) + N (f ) for all orthogonal idempotents e, f ∈ R.
A rank function is a pseudo-rank function N such that N (x) > 0 for all nonzero x ∈ R.
Proposition 4.1. (Goodearl) Let (M, u) be a conical refinement monoid with order-unit ad-
mitting a faithful state s, i.e. a monoid homomorphism s : M → R+ such that s(u) = 1 and
s(x) > 0 for every nonzero x in M . Assume that M is not cancelative. Then there is no regular
algebra R over an uncountable field F such that V(R) ∼= M .
Proof. Assume that R is a regular F -algebra over an uncountable field F with V(R) ∼= M .
Clearly we can assume that R is unital and that [1] corresponds to u under the isomorphism
V(R) ∼= M .
By [26, Theorem 2.2], it suffices to prove that there is no uncountable independent family
of nonzero right or left ideals of R. Since V(R) ∼= V(Rop), where Rop is the opposite ring of R,
we see that it suffices to show this fact for right ideals. Indeed, once this is established, we get
from [26, Theorem 2.2] and [7, Proposition 4.12] that R is unit-regular, and thus V(R) must
be cancellative by [7, Theorem 4.5], a contradiction with our hypothesis.
By [20, Proposition 17.12] there exists a pseudo-rank function N on R such that N (x) =
s([xR]) for every x ∈ R. Since s is faithful, we see that N is indeed a rank function. Now
by [20, Proposition 16.11] we get that R contains no uncountable direct sums of nonzero right
ideals, as desired.
September 25, 2018 0:19
WSPC - Proceedings Trim Size: 9in x 6in
realizationWS
12
Now we are going to recall the example of Chuang and Lee [15]. We will give a presentation
which is a little bit more general. Let R be a σ-unital regular ring, that is, a regular ring having
an increasing sequence (en) of idempotents in R such that R = S∞
n=1 enRen. Put Rn = enRen.
Recall that the multiplier ring M(R) is the completion of R with respect to the strict topology;
see [9]. Write
R = {(xn) (xn) is a Cauchy sequence in the strict topology } ⊆
∞
Y
n=1
Rn.
By the continuity of operations, R is a unital subring of Q∞
n=1 Rn. There is an obvious canonical
surjective homomorphism Φ : R → M(R) whose kernel is I = {(xn) xn → 0}, where the
convergence is with respect to the strict topology.
Lemma 4.1. I is always a (non-unital) regular ring. If each enRen is unit-regular, then I is
unit-regular, meaning that eIe is unit-regular for every idempotent e in I.
Proof. Let x = (xn) ∈ I. Choose a sequence of integers m1 < m2 < · · · such that for all
m ≥ mi we have xmei = eixm = 0. Now for mi ≤ m < mi+1, choose a quasi-inverse ym of xm
in (em − ei)R(em − ei). (Note that xm ∈ (em − ei)R(em − ei) for mi ≤ m < mi+1.) We get a
quasi-inverse y = (yn) of x such that yn → 0 strictly, so y ∈ I and I is regular. The last part
is easy, and is left to the reader.
Observe that if Q is any regular ring such that Q ⊆ M(R), then Φ−1(Q) is a regular ring
( [20, Lemma 1.3]) which is a subdirect product of the regular rings (Rn). In particular Φ−1(Q)
is stably finite if each Rn is so.
Now we see that when K is a countable field the regular algebra QK(E) of the quiver E
with E0 = {v0, v1} and E1 = {e, f }, with r(e) = s(e) = s(f ) = v1 and r(f ) = v0 gives an
example that fits in the above picture. (Note that the quiver E is the same as the quiver S1 of
Figure 2.) Since the field K is countable, the algebra QK(E) is also countable.
Write Q = QK(E), and let I be the ideal of Q generated by v0. Then I = Soc(Q) is a simple
(non-unital) ring, and I is countable, so we have I ∼= M∞(K) (because v0Qv0 is isomorphic to
K), see [9, Remark 2.9]. Here M∞(K) denotes the K-algebra of countably infinite matrices with
only a finite number of nonzero entries. Since this is a crucial argument here, let us recall the
details. The ring I is countable and simple with a minimal idempotent v0, so by general theory
there is a dual pair V, W of K-vector spaces such that I ∼= FW (V ), the algebra of all adjointable
operators on V of finite rank. Since I is countable, both V and W are countably dimensional
K-vector spaces. By an old result of G. W. Mackey [29, Lemma 2], there are dual bases (vi)
and (wj ) for V and W respectively, that is, we have hvi, wji = δij for all i, j, which shows that
FW (V ) ∼= M∞(K). Since I is essential in Q we get an embedding of Q into the multiplier algebra
M(I). Observe that M(I) ∼= RCF M (K), the algebra of row-and-column-finite matrices with
coefficients in K, and that Q/I ∼= K(t). So the algebra Q has the same essential properties
September 25, 2018 0:19
WSPC - Proceedings Trim Size: 9in x 6in
realizationWS
13
as the Chuang and Lee algebra, see [15]. Now M∞(K) is clearly σ-unital and unit-regular.
Indeed there is a σ-unit (en) for I consisting of idempotents such that enIen ∼= Mn(K). Now
Lemma 4.1 together with [20, Lemma 1.3] gives that S := Φ−1(Q) is regular, and it is residually
artinian. The ring S is not countable but it can be easily modified to get a countable algebra
with similar properties. Indeed consider the K-subalgebra S0 of S generated by ⊕∞
n=1Mn(K)
and a, b where a, b are elements in S such that Φ(a)Φ(b) = 1 and Φ(b)Φ(a) 6= 1. Observe that
S0 is countable. We can build a sequence of countable K-subalgebras of S:
S0 ⊆ S1 ⊆ S2 ⊆ · · · ⊆ S
such that each element in Si is regular in Si+1 for all i. It follows that S∞ = S Si is a countable,
regular K-algebra, which is embedded in Q∞
n=1 Mn(K). Moreover S∞ cannot be unit-regular
because it has a quotient ring which is not directly finite. It follows that M = V(S∞) is not
cancellative and it is a countable monoid satisfying the hypothesis of Proposition 4.1, because
there is a rank function on Q∞
n=1 2−nNn(xn), where
Nn is the unique rank function on Mn(K). Therefore M gives a counterexample to R(F ) for
uncountable fields F , although by definition it can be realized over some countable field K.
n=1 Mn(K), e.g. the function N ((xn)) = P∞
Acknowledgments
This work has been partially supported by the DGI and European Regional Development Fund,
jointly, through Project MTM2005-00934, and by the Comissionat per Universitats i Recerca
de la Generalitat de Catalunya.
It is a pleasure to thank Gene Abrams, Ken Goodearl, Kevin O'Meara and Enrique Pardo
for their helpful comments. I am specially grateful to Fred Wehrung for his many valuable
comments and suggestions.
References
1. G. Abrams, G. Aranda Pino, The Leavitt path algebra of a graph, J. Algebra, 293 (2005),
319 -- 334.
2. G. Abrams, G. Aranda Pino, Purely infinite simple Leavitt path algebras, J. Pure Appl. Algebra,
207 (2006), 553 -- 563.
3. P. Ara, Stability properties of exchange rings, International Symposium on Ring Theory (Kyongju,
1999), 23 -- 42, Trends Math., Birkhauser Boston, Boston, MA, 2001
4. P. Ara, The regular algebra of a poset, Preprint.
5. P. Ara, M. Brustenga, The regular algebra of a quiver, J. Algebra, 309 (2007), 207 -- 235.
6. P. Ara, K. R. Goodearl, E. Pardo, K0 of purely infinite simple regular rings, K-Theory, 26
(2002), 69 -- 100.
7. P. Ara, K. R. Goodearl, K. C. O'Meara, E. Pardo, Separative cancellation for projective
modules over exchange rings, Israel J. Math., 105 (1998), 105 -- 137.
8. P. Ara, M. A. Moreno, and E. Pardo, Nonstable K-theory for graph algebras, Algebr. Represent.
Theory, 10 (2007), 157 -- 178.
September 25, 2018 0:19
WSPC - Proceedings Trim Size: 9in x 6in
realizationWS
14
9. P. Ara, F. Perera, Multipliers of von Neumann regular rings, Comm. Algebra, 28 (2000), 3359 --
3385.
10. P. Ara, F. Perera, F. Wehrung, Finitely generated antisymmetric graph monoids, Preprint,
HAL ccsd-00156906.
11. G. M. Bergman, Coproducts and some universal ring constructions, Trans. Amer. Math. Soc.,
200 (1974), 33 -- 88.
12. G. M. Bergman, W. Dicks, Universal derivations and universal ring constructions, Pacific J.
Math., 79 (1978), 293 -- 337.
13. G. M. Bergman, Von Neumann regular rings with tailor-made ideal lattices. Unpublished note,
1986.
14. G. Brookfield, Cancellation in primely generated refinement monoids, Algebra Universalis, 46
(2001), 342 -- 371.
15. C. L. Chuang, P. H. Lee, On regular subdirect products of simple Artinian rings, Pacific J.
Math., 142 (1990), 17 -- 21.
16. A. H. Clifford, G. B. Preston, "The algebraic theory of semigroups. Vol. I", Mathematical
Surveys, No. 7 American Mathematical Society, Providence, R.I. 1961.
17. P. M. Cohn, On the free product of associative rings, Math. Z., 71 (1959), 380 -- 398.
18. G. A. Elliott, On the classification of inductive limits of sequences of semisimple finite-
dimensional algebras, J. Algebra, 38 (1976), 29 -- 44.
19. K. R. Goodearl, "Partially ordered abelian groups with interpolation", Mathematical Surveys
and Monographs, 20. American Mathematical Society, Providence, RI, 1986
20. K. R. Goodearl, "Von Neumann Regular Rings", Pitman, London 1979; Second Ed., Krieger,
Malabar, Fl., 1991.
21. K. R. Goodearl, K0 of regular rings with bounded index of nilpotence, Abelian group theory and
related topics (Oberwolfach, 1993), 173 -- 199, Contemp. Math., 171, Amer. Math. Soc., Providence,
RI, 1994.
22. K. R. Goodearl, Von Neumann regular rings and direct sum decomposition problems, Abelian
groups and modules (Padova, 1994), Math. Appl. 343, 249 -- 255, Kluwer Acad. Publ., Dordrecht,
1995.
23. K. R. Goodearl, D. E. Handelman, Tensor products of dimension groups and K0 of unit-regular
rings, Canad. J. Math. 38 (1986), 633 -- 658.
24. G. Gratzer, "General Lattice Theory, Second Edition", Birkhauser Verlag, Basel, 1998.
25. G. Gratzer, Two problems that shaped a century of lattice theory. Notices Amer. Math. Soc. 54
(2007), no. 6, 696 -- 707.
26. K. R. Goodearl, P. Menal, Stable range one for rings with many units, J. Pure Applied Algebra
54 (1988), 261 -- 287.
27. K. R. Goodearl, F. Wehrung, Representations of distributive semilattices in ideal lattices of
various algebraic structures, Algebra Universalis, 45 (2001), 71 -- 102.
28. K. R. Goodearl, F. Wehrung, The complete dimension theory of partially ordered systems with
equivalence and orthogonality. Mem. Amer. Math. Soc., 176 (2005), no. 831.
29. L. G. W. Mackey, Note on a theorem of Murray, Bull. Amer. Math. Soc., 52 (1946), 322 -- 325.
30. R. S. Pierce, Countable Boolean Algebras, in Handbook of Boolean Algebras, edited by J. D.
Monk with R. Bonnet, Elsevier, 1989, 775 -- 876.
31. I. Raeburn, Graph algebras, CBMS Reg. Conf. Ser. Math., vol. 103, Amer. Math. Soc., Provi-
dence, RI, 2005.
32. P. Ruzicka, Free trees and the optimal bound in Wehrung's theorem, to appear in Fund. Math.
September 25, 2018 0:19
WSPC - Proceedings Trim Size: 9in x 6in
realizationWS
15
33. A. H. Schofield, "Representations of Rings over Skew Fields", LMS Lecture Notes Series 92,
Cambridge Univ. Press, Cambridge, UK, 1985.
34. W. Szyma´nski, The range of K-invariants for C∗-algebras of infinite graphs, Indiana Univ. Math.
J., 51 (2002), 239 -- 249.
35. F. Wehrung, Non-measurability properties of interpolation vector spaces, Israel J. Math., 103
(1998), 177 -- 206.
36. F. Wehrung, Embedding simple commutative monoids into simple refinement monoids, Semi-
group Forum, 56 (1998), 104 -- 129.
37. F. Wehrung, The dimension monoid of a lattice, Algebra Universalis, 40 (1998), 247 -- 411.
38. F. Wehrung, A uniform refinement property for congruence lattices, Proc. Amer. Math. Soc.,
127 (1999), 363 -- 370.
39. F. Wehrung, Representation of algebraic distributive lattices with ℵ1 compact elements as ideal
lattices of regular rings, Publicacions Matem`atiques, 44 (2000), 419-435.
40. F. Wehrung, Semilattices of finitely generated ideals of exchange rings with finite stable rank,
Trans. Amer. Math. Soc., 356 (2004), 1957-1970.
41. F. Wehrung, A K0-avoiding dimension group with an order-unit of index two, J. Algebra, 301
(2006), 728 -- 747.
42. F. Wehrung, A solution to Dilworth's congruence lattice problem, Adv. Math., 216 (2007), 610-
625.
|
1908.02009 | 1 | 1908 | 2019-08-06T08:27:10 | On associative operations on commutative integral domains | [
"math.RA"
] | We describe the associative multilinear polynomial functions over commutative integral domains. This extends Marichal and Mathonet's result on infinite integral domains and provides a new proof of Andres's classification of two-element $n$-semigroups. | math.RA | math |
ON ASSOCIATIVE OPERATIONS ON COMMUTATIVE
INTEGRAL DOMAINS
ERKKO LEHTONEN AND FLORIAN STARKE
Abstract. We describe the associative multilinear polynomial functions over
commutative integral domains. This extends Marichal and Mathonet's result
on infinite integral domains and provides a new proof of Andres's classification
of two-element n-semigroups.
1. Introduction
The classical notions of associativity and semigroup are generalized by n-ary
associativity and n-semigroup. Marichal and Mathonet described associative poly-
nomial functions over infinite commutative integral domains [2]. We slightly modify
this result to obtain a description of the associative multilinear polynomial func-
tions over arbitrary commutative integral domains. As a special case, this gives
a classification of the associative operations on a two-element set (Boolean func-
tions), which was first established by Andres [1]. We provide another, elementary
proof of the result for Boolean functions, which is a streamlined version of the proof
presented in [1]. Furthermore, we describe which n-semigroups on a two-element
set are not derivable from any semigroup of smaller arity.
2. Preliminaries
Throughout this paper, we denote the set of nonnegative integers by N. Let A
be an arbitrary set. An operation on A is a map f : An → A for some number
n ∈ N, called the arity of f . An operation f : An → A is associative, if for all
i, j ∈ {0, . . . , n − 1}, the equality
f (a1, . . . , ai, f (ai+1, . . . , ai+n), ai+n+1, . . . , a2n−1)
= f (a1, . . . , aj, f (aj+1, . . . , aj+n), aj+n+1, . . . , a2n−1)
holds for all a1, . . . , a2n−1 ∈ A. For n = 2, this condition is exactly the classical
associative law. Note that every unary operation is associative. An algebra (A; f )
with a single n-ary associative operation f is called an n-semigroup or an n-ary
semigroup. Thus 2-semigroups are just the classical semigroups.
Given an n-ary operation f : An → A and ℓ ∈ N, we define the operation fℓ of
arity N (ℓ) := ℓ(n − 1) + 1 by the following recursion: f0 := idA, and for ℓ ≥ 0,
define fℓ+1 : AN (ℓ+1) → A by the rule
fℓ+1(a1, . . . , aN (ℓ+1)) := fℓ(f (a1, . . . , an), an+1, . . . , aN (ℓ+1)),
for all a1, . . . , aN (ℓ+1) ∈ A. The operations fℓ are said to be derived from f . Note
that f1 = f . In order to emphasize the arity of a derived operation, we will also write
Technische Universitat Dresden, Institut fur Algebra, 01062 Dresden, Germany
1
2
ON ASSOCIATIVE OPERATIONS ON COMMUTATIVE INTEGRAL DOMAINS
f (N (ℓ))
or simply f (N (ℓ)) for fℓ. It is easy to verify that if f : An → A is associative,
ℓ
then every operation derived from f is associative. Not every n-ary associative
operation arises in this way. We say that an associative operation f : An → A is
primitive, if f is not derivable from any associative operation g : Am → A with
m < n.
3. Associative multilinear polynomial functions on integral domains
We would like to modify the following result so as to make it applicable for finite
domains.
Theorem 3.1 (Marichal, Mathonet [2, Main Theorem]). Let R be an infinite
commutative integral domain with identity and n ≥ 2. A polynomial function
p : Rn → R is associative if and only if it is one of the following:
(i) p(x) = c, where c ∈ R,
(ii) p(x) = x1,
(iii) p(x) = xn,
(iv) p(x) = c +Pn
(v) p(x) =Pn
(vi) p(x) = −b + aQn
i=1 xi, where c ∈ R,
i=1 ωi−1xi (if n ≥ 3), where ω ∈ R \ {1} satisfies ωn−1 = 1,
i=1(xi + b), where a ∈ R \ {0} and b is an element of
the field of fractions of R such that abn − b ∈ R and abk ∈ R for every
k ∈ {1, . . . , n − 1}.
Marichal and Mathonet's proof of Theorem 3.1 starts with the observation that
polynomials and polynomial functions over R are in one-to-one correspondence.
Then it is shown in [2, Proposition 2] that for any associative polynomial function
p : Rn → R, the polynomial p must be multilinear, i.e., no variable occurs with an
exponent higher than 1. The remainder of the proof only relies on this multilin-
earity. Thus, restricting ourselves to multilinear polynomials to begin with, we can
also allow finite domains, and we are lead to the following result.
Theorem 3.2. Let R be a commutative integral domain with identity and n ≥ 2.
A multilinear polynomial function p : Rn → R is associative if and only if it is of
one of the forms prescribed in Theorem 3.1.
Since every Boolean function is a multilinear polynomial function over GF(2), a
description of associative Boolean functions follows immediately from Theorem 3.2
(note that item (v) is void in this case). On the other hand, the theorem fails to
capture all associative functions over finite fields with at least three elements. It is
easy to provide examples of n-ary associative operations that are not of any of the
forms listed in Theorem 3.1, such as n-semigroups derived from rectangular bands,
or operations of the form (x1, . . . , xn) 7→ ϕ(x1), where ϕ : A → A is a nonconstant
idempotent map distinct from idA. This leads to an intriguing open problem.
Problem 3.3. Describe the associative operations on a finite set with at least three
elements.
4. Elementary proof of the description of two-element n-semigroups
It is well known that there are eight semigroups on the two-element set {0, 1}.
They are precisely the algebras with one of the following binary operations: constant
operations c0, c1, projections pr(2)
2 , semigroup operations ∨, ∧, and group
operations +, ⊞. These operations are defined in Table 1.
1 , pr(2)
ON ASSOCIATIVE OPERATIONS ON COMMUTATIVE INTEGRAL DOMAINS
3
0
c0
0 0
1 0
∨ 0
0 0
1 1
1
0
0
1
1
1
0
c1
0 1
1 1
∧ 0
0 0
1 0
1
1
1
1
0
1
pr(2)
1
0
0 0
1 1
+ 0
0 0
1 1
1
0
1
1
1
0
pr(2)
2
0
0 0
1 0
⊞ 0
0 1
1 0
1
1
1
1
0
1
Table 1. The semigroup operations on {0, 1}.
For notational simplicity, in what follows, we view tuples over {0, 1} as words
in the free monoid with two generators 0 and 1, and we will concatenate words by
writing them one after the other. We use the shorthand an for a word comprising
n copies of a. In particular, a0 equals the empty word ε, a1 = a, and a2 = aa.
Moreover, we denote an n-ary operation by brackets, i.e., (·) : {0, 1}n → {0, 1},
a1a2 . . . an 7→ (a1a2 . . . an).
Theorem 4.1 (Andres [1, Theorem 3.1]). For n ≥ 2, an n-ary operation on {0, 1}
is associative if and only if it is one of the following: c(n)
0 , c(n)
n , ∨(n),
∧(n), +(n), +(n), where +(n)(a1, . . . , an) := +(n)(a1, . . . , an) + 1.
1 , pr(n)
1 , pr(n)
Proof. It is clear that the operations specified in the statement are associative,
because each one is derived from a binary associative operation, with the exception
of +(n) for odd n. (We have ⊞(n) = +(n) for even n, and ⊞(n) = +(n) for odd n.)
It is easy to verify that also +(n) is associative.
In order to show necessity, assume that (·) : {0, 1}n → {0, 1} is associative. We
consider several cases and subcases.
Case 1: (0n) = 0.
Case 1.1: (10n−1) = 0. Then for all uv ∈ {0, 1}n−2,
(u10v) = (u1(0n)v) = (u(10n−1)0v) = (u00v).
It follows that (·) is completely determined by its values at tuples of the form
0n−k1k with 0 ≤ k ≤ n. More precisely, if a = u01k for some u ∈ {0, 1}n−k−1,
then (a) = (0n−k1k), because the value of (·) does not change if we change any 1
followed by 0 to 0.
Case 1.1.1: (0n−11) = 0. Similarly as above, we obtain for all uv ∈ {0, 1}n−2,
(u01v) = (u(0n)1v) = (u0(0n−11)v) = (u00v).
Thus the value of (·) does not change if we change any 1 preceded by 0 to 0; in
particular, (0n−k1k) = (0n−k+11k−1) for 0 < k < n. Consequently, (a) = 0 for all
a ∈ {0, 1}n \ {1n}. It remains to consider the value of (·) at 1n.
Case 1.1.1.1: (1n) = 0. Then (·) = c(n)
0 .
Case 1.1.1.2: (1n) = 1. Then (·) = ∧(n).
Case 1.1.2: (0n−11) = 1. Then for all uv ∈ {0, 1}n−2,
(u01v) = (u(10n−1)1v) = (u1(0n−11)v) = (u11v).
4
ON ASSOCIATIVE OPERATIONS ON COMMUTATIVE INTEGRAL DOMAINS
Thus (0n−k1k) = (0n−k−11k+1) for 0 < k < n, and we have (·) = pr(n)
n .
Case 1.2: (10n−1) = 1.
Case 1.2.1: (110n−2) = 0. Then for all uv ∈ {0, 1}n−2,
(u00v) = (u(110n−2)0v) = (u1(10n−1)v) = (u11v).
Case 1.2.1.1: (010n−2) = 0. Then for all uv ∈ {0, 1}n−2,
(u00v) = (u(010n−2)0v) = (u0(10n−1)v) = (u01v).
Applying the above identities, we obtain
0 = (0000n−3) = (0010n−3) = (1110n−3) = (1000n−3) = 1,
a contradiction. Thus, this case is not possible.
Case 1.2.1.2: (010n−2) = 1. Then for all uv ∈ {0, 1}n−2,
(u01v) = (u0(10n−1)v) = (u(010n−2)0v) = (u10v).
This means that (·) is symmetric, and the value of (·) at a depends only on the
number of 1's in a. Together with the identity (u00v) = (u11v) established above,
this implies that (a) depends only on the parity of the number of 1's in a. Since
(0n) = 0 and (10n−1) = 1, we have (a) = 0 if and only if the number of 1's in a is
even, in other words, (·) = +(n).
Case 1.2.2: (110n−2) = 1. Then for all uv ∈ {0, 1}n−2,
(u10v) = (u(110n−2)0v) = (u1(10n−1)v) = (u11v).
It follows that (·) is completely determined by its values at tuples of the form
0k1n−k with 0 ≤ k ≤ n. More precisely, if a = 0k1u for some u ∈ {0, 1}n−k−1, then
(a) = (0k1n−k), because the value of (·) does not change if we change any 0 preceded
by 1 to 1. In particular, for any u ∈ {0, 1}n−1, we have (1u) = (1n) = (10n−1) = 1.
Case 1.2.2.1: (01n−1) = 0. Then for all uv ∈ {0, 1}n−2,
(u01v) = (u0(1n−10)v) = (u(01n−1)0v) = (u00v).
Thus (0k1n−k) = (0k+11n−k−1) for 0 < k < n. Consequently, (0k1n−k) = (0n) = 0
for 0 < k ≤ n. Therefore (·) = pr(n)
1 .
Case 1.2.2.2: (01n−1) = 1. Then for all uv ∈ {0, 1}n−2,
(u01v) = (u0(1n)v) = (u(01n−1)1v) = (u11v).
Thus (0k1n−k) = (0k−11n−k+1) for 0 < k < n. Consequently, (0k1n−k) = (1n) = 1
for 0 ≤ k < n. Therefore (a) = 0 if and only if a = 0n, that is, (·) = ∨(n).
Case 2: (0n) = 1. Then for all uv ∈ {0, 1}n−2,
(u01v) = (u0(0n)v) = (u(0n)0v) = (u10v).
Consequently, (·) is symmetric, and the value of (·) at a depends only on the number
of 1's in a.
Case 2.1: (10n−1) = 0. Then for all uv ∈ {0, 1}n−2,
(u00v) = (u(10n−1)0v) = (u1(0n)v) = (u11v).
ON ASSOCIATIVE OPERATIONS ON COMMUTATIVE INTEGRAL DOMAINS
5
Similarly as in Case 1.2.1.2 we conclude that the value of (·) at a depends only on
the parity of the number of 1's in a. Since (0n) = 1 and (10n−1) = 0, we have
(a) = 0 if and only if the number of 1's in a is odd, in other words, (·) = +(n).
Case 2.2: (10n−1) = 1. Then for all uv ∈ {0, 1}n−2,
(u10v) = (u(10n−1)0v) = (u1(0n)v) = (u11v).
Thus (1k0n−k) = (1k+10n−k−1) for 0 < k < n. Since (10n−1) = 1 and (0n) = 1, it
follows that (·) = c(n)
1 .
(cid:3)
Remark 4.2. For n ≥ 2, the only n-ary associative operation on {0, 1} that is not
derivable from a binary associative operation is +(n) for odd n.
Proposition 4.3. For n ≥ 1, the only primitive n-ary associative operations on
{0, 1} are the unary and binary ones and +(n) for n = 2k + 1, k ∈ N.
Proof. For each binary semigroup operation ◦ (see Table 1) and for each n ≥ 3,
the n-ary associative operation ◦(n) is obviously derivable from ◦ and is hence not
It remains to consider operations of the form +(n). The operations
primitive.
derivable from +(m) are, for ℓ ∈ N,
(+(m))(ℓ(m−1)+1)
ℓ
=(+(ℓ(m−1)+1),
+(ℓ(m−1)+1),
if ℓ is odd,
if ℓ is even.
It follows that, for n ≥ 2, +(n) is primitive if and only if n is not of the form
ℓ(m − 1) + 1 for any odd ℓ > 1 and for any m > 1. This is equivalent to n = 2k + 1
for some k ∈ N.
(cid:3)
Remark 4.4. The solution to problem 7 in the 2018 Mikl´os Schweitzer competition
[3] reveals that the self-commuting Boolean functions are the same as the associative
ones with fictitious arguments introduced.
Acknowledgments
The authors would like to thank Robert Baumann, Thomas Quinn-Gregson, and
Nikolaas Verhulst for inspiring discussions and the anonymous reviewer for helpful
comments.
References
[1] Andres, S.D.: Classification of all associative mono-n-ary algebras with 2 elements. Int. J.
Math. Math. Sci. 2009, Art. ID 678987, 16 pp. (2009)
[2] Marichal, J.-L., Mathonet, P.: A description of n-ary semigroups polynomial-derived from
integral domains. Semigroup Forum 83, 241 -- 249 (2011)
[3] "2018 Mikl´os Schweitzer", Art of Problem Solving portal, accessed June 11, 2019,
https://artofproblemsolving.com/community/c771105 2018 mikloacutes schweitzer
|
1209.6256 | 3 | 1209 | 2013-03-10T15:21:34 | Lie nilpotency indices of symmetric elements under oriented involutions in group algebras | [
"math.RA",
"math.GR"
] | Let $G$ be a group and let $F$ be a field of characteristic different from 2. Denote by $(FG)^+$ the set of symmetric elements and by $\mathcal{U}^+(FG)$ the set of symmetric units, under an oriented classical involution of the group algebra $FG$. We give some lower and upper bounds on the Lie nilpotency index of $(FG)^+$ and the nilpotency class of $\mathcal{U}^+(FG)$. | math.RA | math |
LIE NILPOTENCY INDICES OF SYMMETRIC ELEMENTS UNDER
ORIENTED INVOLUTIONS IN GROUP ALGEBRAS
JOHN H. CASTILLO
Abstract. Let G be a group and let F be a field of characteristic different from 2. Denote
by (F G)+ the set of symmetric elements and by U +(F G) the set of symmetric units, under
an oriented classical involution of the group algebra F G. We give some lower and upper
bounds on the Lie nilpotency index of (F G)+ and the nilpotency class of U +(F G).
1. Introduction
Let F G denote the group algebra of a group G over a field F with char(F ) = p 6= 2. A
homomorphism σ : G → {±1} is called an orientation of the group G. Working in the context
of K-theory, Novikov [11], introduced an oriented involution ∗ of F G, given by
X
g∈G
αgg
∗
= X
g∈G
αgσ(g)g−1.
When σ is trivial this involution coincides with the so called classical involution of F G.
We denote (F G)+ = {α ∈ F G : α∗ = α} and (F G)− = {α ∈ F G : α∗ = −α} the set of
symmetric and skew-symmetric elements of F G under ∗, respectively. We denote by N the
kernel of σ. It is obvious that the involution ∗ coincides on the group algebra F N with the
classical involution. It is easy to see that, as an F -module, (F G)+ is generated by the set
S = {g + g−1 : g ∈ N } ∪ {g − g−1 : g ∈ G \ N, g2 6= 1}
and (F G)− is generated by
L = {g + g−1 : g ∈ G \ N } ∪ {g − g−1 : g ∈ N, g2 6= 1}.
Given g1, g2 ∈ G, we define the commutator (g1, g2) = g−1
2 g1g2 and recursively,
(g1, . . . , gn) = ((g1, . . . , gn−1), gn) for n elements g1, . . . , gn of G. By the commutator (X, Y )
of the subsets X and Y of G we mean the subgroup of G generated by all commutators (x, y)
with x ∈ X, y ∈ Y .
In this way, we can define the lower central series of a nonempty
subset H of G by: γ1(H) = H and γn+1(H) = (γn(H), H), for n ≥ 1. We say that
H is nilpotent if γn(H) = 1, for some n. For a nilpotent subset H ⊆ G the number
cl(H) = min{n ∈ N0 : γn+1(H) = 1} is called the nilpotency class of H. It can be proved
that H is a nilpotent set if and only if H satisfies the group identity (g1, . . . , gn) = 1 for some
n ≥ 2.
1 g−1
In an associative ring R, the Lie bracket on two elements x, y ∈ R is defined by [x, y] =
xy − yx. This definition is extended recursively via [x1, . . . , xn+1] = [[x1, . . . , xn], xn+1]. For
X, Y ⊆ R by [X, Y ] we denote the additive subgroup generated by all Lie commutators [x, y]
2010 Mathematics Subject Classification. 16W10, 16U80, 16U60.
Key words and phrases. Involution, symmetric elements, Lie nilpotent, strongly Lie nilpotent, Lie nilpotency
index, nilpotency class.
1
2
J.H. CASTILLO
with x ∈ X, y ∈ Y . The lower Lie central series of a nonempty subset S of R is defined
inductively by setting γ1(S) = S and γn+1(S) = [γn(S), S]. We say that the subset S is
Lie nilpotent if there exists a natural number n, such that γn(S) = 0. The smallest natural
number with the last property, denoted by t(S), is called the Lie nilpotency index of S. It
is possible to show that S is Lie nilpotent if and only if S satisfies the polynomial identity
[x1, . . . , xn] = 0 for some n ≥ 2.
Given a nonempty subset S of R, we let S(1) = R, and then for each i ≥ 2, let S(i) be the
(associative) ideal of R generated by all elements of the form [a, b], with a ∈ S(i−1), b ∈ S. We
say that S is strongly Lie nilpotent if S(i) = 0 for some i. The minimal n for which S(n) = 0
is called the upper Lie nilpotency index and denoted by tL(S). Clearly, strong Lie nilpotence
implies Lie nilpotence and t(S) ≤ tL(S). Denote by U (S) the set of units in the subset S of
R and suppose that it is nonempty. By the equality (x, y) = 1 + x−1y−1[x, y], it is easy to
see that γn(U (S)) ⊆ 1 + S(n) for all n ≥ 2. In consequence, the set of units of a strongly Lie
nilpotent subset S is nilpotent, and
cl(U (S)) < tL(S).
(1)
In 1973, Passi, Passman and Sehgal [12] showed that the group algebra F G is Lie nilpotent
if and only if G is nilpotent and G′ is a finite p-group, where p is the characteristic of F .
Actually, see [14], a group algebra is Lie nilpotent if and only if it is strongly Lie nilpotent.
Next, S.K. Sehgal characterized group algebras which are Lie n-Engel, for some n.
In 1993, Giambruno and Sehgal [6] began the study of Lie nilpotence of symmetric and
skew-symmetric elements under the classical involution. They proved that given a group G
without elements of order 2 and a field F with char(F ) 6= 2, if either (F G)+ or (F G)− is Lie
nilpotent, then F G is Lie nilpotent. This work was completed by G.T. Lee [8], for groups in
general. More specifically, he proved that the Lie nilpotence of the symmetric elements under
the classical involution is equivalent to the Lie nilpotence of F G when the group G does not
contain a copy of Q8, the quaternion group of order 8 and he also characterized the group
algebras such that the set of symmetric elements is Lie nilpotent when G contains a copy of
Q8.
Recently, Castillo and Polcino Milies, see [5], studied Lie properties of the symmetric ele-
ments under an oriented classical involution. They extended some previous results from [6],
[8] and [9]. In particular, they gave some groups algebras such that the Lie nilpotence of the
symmetric set implies the same property in the whole group algebra. Also, they obtained
a complete characterization of the group algebras F G, such that Q8 ⊆ G and (F G)+ is Lie
nilpotent.
Lately, Z. Balogh and T. Juh´asz in [2] and [3] studied the Lie nilpotency index of (F G)+
and the nilpotency class of the U +(F G) under the classical involution in group algebras. They
gave a necessary condition to the numbers t((F G)+) and cl(U +(F G)) be maximal, as possible,
in a nilpotent group algebra. Also, they studied this two numbers to group algebras such that
(F G)+ is Lie nilpotent but F G is not.
In this article we study the Lie nilpotency index of (F G)+ and the nilpotency class of
U +(F G) under an oriented classical involution. In the next section we give some preliminary
results. In the third section we study the numbers t((F G)+) and cl(U +(F G)) in Lie nilpotent
In the fourth section we study the case when Q8 ⊆ G and (F G)+ is Lie
group algebras.
nilpotent.
Throughout this paper F will always denote a field of characteristic not 2, G a group and
σ a nontrivial orientation of G. In a number of places, all over this paper, we use arguments
from [2], [3] and [10]. Some of them are reproduced here for the sake of completeness.
SYMMETRIC ELEMENTS UNDER ORIENTED INVOLUTIONS
3
We recall the following result from [10].
2. Preliminaries
Lemma 2.1. Let R be a ring and S a subset of R. Suppose, for some i ≥ 1, that S(i) ⊆ zR,
where z is central in R. Then for all j > 0, we have S(i+j) ⊆ zS(j). In particular, for any
positive integer m, S(mi) ⊆ zmR.
Proof. The proof is by induction on j. If j = 1, then S(i+1) ⊆ S(i), there is nothing to do.
Assume that S(i+j) ⊆ zS(j). Take a ∈ S(i+j), b ∈ S. So a = za1, for some a1 ∈ S(j). Thus,
[a, b] = [za1, b] = z[a1, b] ∈ zS(i+j), as we want to prove.
To get the second part, notice that
S(2i) = S(i+i) ⊆ zS(i) ⊆ z2R.
Suppose that S((m−1)i) ⊆ zm−1R. So S(mi) = S((m−1)i+i) ⊆ zS((m−1)i) ⊆ zmR.
(cid:3)
Throughout this article we denote by Q8 = (cid:10)x, y : x4 = 1, x2 = y2, xy = x−1(cid:11) the quaternion
group of order 8. Castillo and Polcino Milies [5] characterized the group algebras of groups
containing Q8 and with a nontrivial orientation, such that (F G)+ is Lie nilpotent. Here we
prove that the conditions obtained by them are also satisfied when (F G)+ is strongly Lie
nilpotent.
Theorem 2.1. Let F be a field of characteristic p 6= 2, G a group with a nontrivial orientation
σ and x, y elements of G such that hx, yi ≃ Q8. Then (F G)+ is strongly Lie nilpotent if and
only if either
(i) char(F ) = 0, N ≃ Q8 × E and G ≃ hQ8, gi × E, where E2 = 1 and g ∈ G \ N is such
that (g, x) = (g, y) = 1 and g2 = x2; or,
(ii) char(F ) = p > 2, N ≃ Q8 × E × P , where E2 = 1, P is a finite p-group and there exists
g ∈ G \ N such that G ≃ hQ8, gi × E × P , (g, x) = (g, y) = 1 and g2 = x2.
Proof. If (F G)+ is strongly Lie nilpotent, then (F G)+ is Lie nilpotent and from [5, Theo-
rem 4.2] we get (i) and (ii).
Conversely, assume that P = pn. We claim that, ((F G)+)(2pn) = 0. The proof will be
by induction on n. If n = 0, then G ≃ hQ8, gi × E and thus, from [5, Lemma 4.3], (F G)+
is commutative. Assume that P = pn > 1. Take z ∈ ζ(P ) with o(z) = p, applying our
inductive hypothesis on G = G/ hzi. Then, ((F G)+)(2pn−1) = 0. Thus
((F G)+)(2pn−1) ⊆ ∆(G, hzi) = (z − 1)F G.
((F G)+)(2pn) ⊆ (z − 1)pF G = 0,
By Lema 2.1,
as we claimed.
From the equality, (x, y) = 1 + x−1y−1[x, y] we know that γn(U +(F G)) ⊆ 1 + ((F G)+)(n)
and thus we get the following.
Corollary 2.1. Let F be a field of characteristic different from 2. Assume that Q8 ⊆ G and
(F G)+ is Lie nilpotent. Then, U +(F G) is nilpotent.
We need the following easy observation.
(cid:3)
4
J.H. CASTILLO
Lemma 2.2. Let G be a group, H any subgroup and A a normal subgroup such that A ⊆ N .
If (F G)+ is Lie nilpotent, then so are (F H)+ and (F (G/A))+. Furthermore, t((F H)+) ≤
t((F G)+) and t((F (G/A))+) ≤ t((F G)+).
Proof. Note that (F H)+ is a subset of (F G)+, and thus it has the required properties.
Since A is a normal subgroup contained in the kernel of the orientation σ, we can define in
F (G/A) an induced oriented classical involution from ∗ in F G as follows:
X
¯g∈G/A
⋆
αg ¯g
= X
¯g∈G/A
αgσ(g)¯g−1.
Now, simply observe that the symmetric elements in F (G/A), under ⋆, are linear combina-
tions of terms of the form gA+ σ(g)g−1A, with g ∈ G. That is, every element of (F (G/A))+ is
the homomorphic image of an element of (F G)+ under the natural map εA : F G → F (G/A),
defined by εA(Pg∈G αgg) = Pg∈G αg ¯g.
So assume that (F G)+ is Lie nilpotent, therefore there exists n = t((F G)+) such that
[α1, . . . , αn] = 0 for all αi ∈ (F G)+. Let β1, . . . , βn ∈ (F (G/A))+. Thus
[β1, . . . , βn] = [εA(α1), . . . , εA(αn)]
= εA([α1, . . . , αn]) = εA(0) = 0.
Consequently, t((F (G/A))+) ≤ t((F G)+).
(cid:3)
3. Lie nilpotent group algebras
In this section we assume that F G is Lie nilpotent. By [15], tL(F G) ≤ G′ + 1 and by [4]
the equality holds if and only if G′ is cyclic, or G′ is a noncentral elementary abelian group
of order 4.
Note that a group G of odd finite order has trivial orientation. Indeed, let a be an element
of G. So 1 = σ(aG) = σ(a)G and as G is odd we get that σ(a) = 1. For the last reason
when G is a group of odd finite order, the involution ∗ is the classical involution. In this way,
we can use the following result, that is a combination from [2, Lemma 2] and [3, Lemma 2].
Lemma 3.1. Let G be a finite p-group with a cyclic derived subgroup. Then t((F G)+) ≥
G′ + 1 and cl(U +(F G)) ≥ G′.
We recall that a group G is called p-abelian if G′, the commutator subgroup of G, is a finite
p-group and 0-abelian means abelian.
Theorem 3.1. Let F G be a Lie nilpotent group algebra of odd characteristic and nontrivial
orientation. Then, t((F G)+) = G′ + 1 if and only if G′ is cyclic. Moreover, assuming that
G is a torsion group, cl(U +(F G)) = G′ if and only if G′ is cyclic.
Proof. Assume that t((F G)+) = G′ + 1. As G′ is a finite p-group, if G′ is not cyclic, from
[4], we know that t((F G)+) ≤ tL(F G) < G′ + 1 and we get a contradiction. Thus, G′ is
cyclic.
Conversely, suppose that G′ is cyclic. By the hypotheses, G is a nilpotent p-abelian group
and from [1, Lemma 1] there exists a finite p-group P which is isomorphic to a subgroup of
factor group of G and P ′ ≃ G′. Actually, from the proof of [1, Lemma 1], we know that
P ≃ H/A, where A is a maximal torsion-free central subgroup of G.
SYMMETRIC ELEMENTS UNDER ORIENTED INVOLUTIONS
5
Assume that there exists g ∈ A such that σ(g) = −1. In this way, as G = N ∪ gN , we get
G′ = N ′. Using in F P the classical involution, by lemmas 3.1 and 2.2, we obtain that
G′ + 1 = N ′ + 1 = P ′ + 1 ≤ t((F P )+) ≤ t((F N )+) ≤ t((F G)+).
In the other hand, suppose that A ⊆ N . Then we can define an induced oriented classical
involution in P ≃ H/A, from that one in F G. Consequently,
G′ + 1 = P ′ + 1 ≤ t((F P )+) ≤ t((F G)+).
The proof of the second part is similar.
(cid:3)
4. Groups that contain a copy of Q8
We assume that Q8 ⊆ G and (F G)+ is Lie nilpotent. This means that the group algebra
F G is not Lie nilpotent. Recently, this kind of group algebras was characterized by Castillo
and Polcino Milies [5]. This characterization is the same as in Theorem 2.1, so during this
section we assume that G is as in that result. In this section, we will study the Lie nilpotency
index of the symmetric elements under oriented classical involutions.
It is easy to show that
gm − 1 ≡ m(g − 1)
(mod ∆(G)2).
(2)
for every g ∈ G and any integer m.
We begin with the following result.
Lemma 4.1. Consider F G with an oriented classical involution. Then
((F G)+)(n) ⊆ F G∆(P )n
for all n ≥ 2
Proof. Recall that the symmetric elements are spanned as an F -module by the set
S = {z + z−1 : z ∈ N } ∪ {z − z−1 : z ∈ G \ N }.
If z ∈ N , then z = ah with a ∈ Q8 × E and h ∈ P . Note that if a2h = 1, then h = 1 and
a2 = 1. Thus, a ∈ ζ(Q8 × E). Assuming a2h 6= 1, follows that z + z−1 = ah + a−1h−1 =
ah + a3h−1 = a(h + a2h−1).
Also, if z ∈ G \ N ; we can write z = gah with a ∈ Q8 × E and h ∈ P . If a2h = 1, then
a2 = h = 1. Again, a ∈ ζ(Q8 × E) and thus z − z−1 = gah − g−1a−1h−1 = ga − g−1a =
ga(1 − g2) ∈ ζ(Q8 × E). Now we suppose that a2h 6= 1 and we get the following cases:
(1) If a2 = 1 and h 6= 1, then z − z−1 = gah − g−1a−1h−1 = ag(h − g2h−1).
(2) If a2 6= 1 and h = 1, then z − z−1 = gah − g−1a−1h−1 = ga − g3a3 = ga − ga = 0.
(3) If a2 6= 1 and h 6= 1, then z − z−1 = gah − g−1a−1h−1 = agh − a3g3h−1 = ag(h − h−1),
because a3g3 = ag.
From the above considerations, we obtain that
S = A ∪ B ∪ C ∪ ζ(Q8 × E),
where
A = {a(h + a2h−1) : a ∈ Q8 × E, h ∈ P and a2h 6= 1},
B = {ag(h − g2h−1) : a ∈ Q8 × E, h ∈ P and (a2 = 1 and h 6= 1)},
C = {ag(h − h−1) : a ∈ Q8 × E, h ∈ P and (a2 6= 1 and h 6= 1)}.
6
J.H. CASTILLO
Given a ∈ Q8 × E, such that a2 6= 1 we know that 1 + a2 is symmetric and a2 ∈ ζ(Q8 × E).
In this way,
a(h + a2h−1) + 1 + a2 = a(h − 1) + a3(h−1 − 1) + 1 + a + a2 + a3,
where 1 + a + a2 + a3 is a central element in F G and a(h − 1) + a3(h−1 − 1) ∈ F G∆(P ). It is
clear that, ag(h − h−1) ∈ F G∆(P ). Furthermore, if a2 = 1 and h 6= 1, then ag(h − g2h−1) =
ag(h − 1) − ag3(h−1 − 1) + a(g − g−1) ∈ F G∆(P ) + ζ(F G).
So
also spans (F G)+ as an F -module, where
eS = A′ ∪ B ∪ C ∪ ζ(Q8 × E),
A′ = {a(h + a2h−1) + 1 + a2 : a ∈ Q8 × E, h ∈ P and a2h 6= 1}
and B, C are as above.
In consequence,
(F G)+ ⊆ F G∆(P ) + ζ(F G).
(3)
The proof follows by induction on n. Indeed, if n = 2
[(F G)+, (F G)+] ⊆ [F G∆(P ), F G∆(P )] ⊆ F G∆(P )2.
Suppose that the lemma is true for some n ≥ 2. Take α ∈ ((F G)+)(n) and β ∈ (F G)+. So
[α, β] ∈ [F G∆(P )n, F G∆(P )] ⊆ F G∆(P )n+1.
and we get that ((F G)+)(n+1) ⊆ F G∆(P )n+1 as required.
(cid:3)
Denote by c the central element of Q8 × E, such that (Q8 × E)2 = hci. Given n ≥ 2, we
denote with Mn the F -subspace of the vector space F G generates by the set
{(h1 − h−1
1 ) · · · (hn − h−1
n )(1 − c)a : h1, . . . , hn ∈ P, a ∈ (Q8 × E) \ ζ(Q8 × E)}.
To simplify, we write f1,...,n instead of (h1 − h−1
1 ) · · · (hn − h−1
n ).
Let Sn be the symmetric group of degree n and F Sn its group algebra over the field F . It
is possible to define a group action of Sn on Mn via: for a σ ∈ Sn and a generator element
f1,...,n(1 − c)a of Mn let
σ · f1,...,n(1 − c)a = fσ(1),...,σ(n)(1 − c)a.
Naturally, this group action on a generator set of Mn can be extended linearly to the whole
ασσ ∈ F Sn and
Mn. We extend this group action to a group algebra action: for x = Pσ∈Sn
z ∈ Mn, let
x · z = X
σ∈Sn
ασ(σ · z).
For n ≥ 2 we define the elements x2,n, x3,n, . . . , xn,n of F Sn recursively as:
x2,n = 1 + (2, 1),
xi,n = xi−1,n + xi−1,n(i, i − 1, . . . , 1); for 3 ≤ i ≤ n.
(4)
(5)
Since (F N )+ ⊆ (F G)+, from Lemma 4 and Lemma 5 in [3], we get the following results.
Lemma 4.2. xn,nMn ∈ γn((F G)+)(1 − c) for all n ≥ 2.
Lemma 4.3. If P = pk, then bP (1 − c)a ∈ γk(p−1)((F G)+) for some a ∈ Q8 × E
SYMMETRIC ELEMENTS UNDER ORIENTED INVOLUTIONS
7
We recall that the augmentation ideal ∆(P ) of a finite p-group P is a nilpotent ideal, see
[13, Theorem 6.3.1], we will denote by tnil(P ) its nilpotency index. Also, we remind that a
finite p-group P , is called powerful if P ′ ⊆ P p. Let P be a powerful group. We denote with
Di = Di(F P ) the i-th dimensional subgroup. By Theorem 5.5 in [7], D1 = P and for n > 1,
Dn = D(Dn−1, P ), (D⌈ n
p ⌉)pE .
It can be showed that, (P pi)pj = P pi+j and (P pi, P ) ⊆ P pi+1
pi−1 < n ≤ pi then Dn = P pi
Lemma 4.4. Let P be a powerful group and hi − 1 ∈ ∆(P )ki and hj − 1 ∈ ∆(P )kj , where ki
and kj are positive integers. Then
for every pair i, j. So, if
.
(hi − 1)(hj − 1) ≡ (hj − 1)(hi − 1)
(mod ∆(P )ki+kj+1).
(6)
Proof. First, we prove that (Di, Dj) ⊆ Di+j+1, for every i, j. Take hi ∈ Di and hj ∈ Dj. We
get the following equation
(hi, hj) − 1 = h−1
i h−1
j ((hi − 1)(hj − 1) − (hj − 1)(hi − 1)).
(7)
If either i or j, say i, is not a power of p, then hi ∈ Di = Di+1, so by (7), (hi, hj ) − 1 ∈
∆(P )i+j+1; thus (hi, hj) ∈ Di+j+1. If both i and j are powers of p, then i + j cannot be a
power of p and consequently Di+j = Di+j+1. By (7) follows (hi, hj) ∈ Di+j+1; therefore our
claim is proved.
Let hi − 1 ∈ ∆(P )ki and hj − 1 ∈ ∆(P )kj for some positive integers ki, kj . Then
(hi − 1)(hj − 1) = (hj − 1)(hi − 1) + hjhi((hi, hj) − 1),
and as (hi, hj) ∈ Dki+kj+1, the result follows.
Now we can prove our main result in this section.
(cid:3)
Theorem 4.1. Let F be a field of characteristic p > 2. Consider the group algebra F G with
an oriented classical involution. Assume that Q8 ⊆ G, (F G)+ is Lie nilpotent and the Sylow
p-group P of G is of order pm, with m ≥ 1. Then
(i) 1 + m(p − 1) ≤ t((F G)+) ≤ tL((F G)+) ≤ tnil(P ) and cl(U +(F G)) ≤ tnil(P ) − 1.
(ii) If t((F G)+) = tnil(P ), then cl(U +(F G)) + 1 = t((F G)+).
(iii) If P is powerful, then t((F G)+) = tnil(P ).
(iv) If P is abelian, then, for all k ≥ 2, the F -space γk((F G)+) is generated by the set
Mk ={(h1 − h−1
{g(h1 − h−1
1 ) · · · (hk − h−1
1 ) · · · (hk − h−1
k )(1 − a2)a : hi ∈ P, a ∈ (Q8 × E) \ ζ(Q8 × E)}∪
k )(1 − a2)a : hi ∈ P, a ∈ (Q8 × E) \ ζ(Q8 × E)}.
Proof. From Theorem 2.1, we know that N ≃ Q8 × E × P , where E2 = 1, P is a finite p-group
and there exists g ∈ G \ N such that G ≃ hQ8, gi × E × P , (g, x) = (g, y) = 1 and g2 = x2.
To
By Lemma 4.3, there exists 0 6= bP (1 − c)a ∈ γm(p−1)((F G)+) for some a ∈ Q8 × E. In this
way, 1 + m(p − 1) ≤ t((F G)+). Furthermore, Lemma 4.1 implies that tL((F G)+) ≤ tnil(P ).
elements
show
ui = 1 − ai(1 + a2
) ∈ S, ai ∈ Q8 × E and hi ∈ P . Thus,
ui = 1 + ai(hi − 1) + a3
i − 1) ∈ 1 + F G∆(P ). Since F G∆(P ) is a nilpotent ideal, we get
that 1 + F G∆(P ) is a normal subgroup of U (F G) and in consequence ui is a unit in F G. We
will prove, by induction, that
consider
i ) + xi, where xi = ai(hi + a2
symmetric
i (h−1
i h−1
(ii),
the
i
(u1, u2, . . . , un) ≡ 1 + [x1, x2, . . . , xn]
(mod F G∆(P )n+1).
(8)
8
J.H. CASTILLO
Since u−1
1 u−1
2 ≡ 1 (mod F G∆(P )), Lemma 4.1 implies that
(u1, u2) = 1 + u−1
1 u−1
2 [u1, u2] = 1 + (u−1
(mod F G∆(P )3).
1 u−1
≡ 1 + [u1, u2]
i + a3
2 − 1)[u1, u2] + [u1, u2]
We recall that bai = 1 + ai + a2
F G. So
i and 1 + a2
i , for each ai ∈ Q8 × E, are central elements of
[u1, u2] = [1 − a1(1 + a2
1) + x1, 1 − a2(1 + a2
2) + x2]
= [x1, x2] + [a1(1 + a2
= [x1, x2] + [ba1,ba2] − [x1,ba2] − [ba1, x2]
1), a2(1 + a2
= [x1, x2],
2)] − [x1, a2(1 + a2
2)] − [a1(1 + a2
1), x2]
which proves the congruence (8) when n = 2.
Suppose that (8), is true to n − 1; that is
(u1, u2, . . . , un−1) ≡ 1 + [x1, x2, . . . , xn−1]
(mod F G∆(P )n).
(9)
Then, Lemma 4.1 and as (u1, u2, . . . , un−1)−1u−1
n − 1 ∈ F G∆(P ) imply
(u1, u2, . . . , un)
n − 1)[(u1, u2, . . . , un−1), un] + [(u1, u2, . . . , un−1), un]
= 1 + ((u1, u2, . . . , un−1)−1u−1
≡ 1 + [(u1, u2, . . . , un−1), un]
≡ 1 + [[x1, x2, . . . , xn−1], 1 − an(1 + a2
≡ 1 + [x1, x2, . . . , xn] − [[x1, x2, . . . , xn−1],ban]
(mod F G∆(P )n+1),
≡ 1 + [x1, x2, . . . , xn]
n) + xn]
(mod F G∆(P )n+1)
(mod F G∆(P )n+1)
(mod F G∆(P )n+1)
and the statement (8) is true for all n ≥ 2.
Let n = tnil(P ) − 1.
If t((F G)+) = tnil(P ), then there are x1, . . . , xn ∈ S such that
[x1, . . . , xn] 6= 0. Thus, by the congruence (8), γn(U +(F G)) 6= 1. So n ≤ cl(U +(F G)).
Moreover, we know that cl(U +(F G)) < tL((F G)+) ≤ tnil(P ) = n + 1 and we get (ii).
Assume that P is powerful. Then, by Lemma 4.4, we obtain
xn,nf1,...,n(1 − c)a ≡ 2nf1,...,n(1 − c)a (mod F G∆(P )n+1).
Furthermore, if hi − 1 ∈ ∆(P )ki, then by (2)
hi − h−1
i = (hi − 1) − (h−1
i − 1) ≡ 2(hi − 1)
(mod ∆(P )ki+1),
thus
xn,nf1,...,n(1 − c)a ≡ 2n(h1 − h−1
1 ) · · · (hn − h−1
n )(1 − c)a
≡ 22n(h1 − 1) · · · (hn − 1)(1 − c)a (mod F G∆(P )n+1).
It is clear that, if n < tnil(P ), there exist h1, . . . , hn ∈ P such that Qn
xn,nMn 6= 0. Thus, tnil(P ) ≤ t((F G)+) and (iii) follows.
i=1(hi − 1) 6= 0, and then
Finally, assume that P is abelian. Let a, b ∈ (Q8 × E) \ ζ(Q8 × E), and h1, h2 ∈ P , such
that (a, b) 6= 1. Then
[a(h1 + a2h−1
1 ), b(h2 + b2h−1
2 )] = (h2 + b2h−1
2 )(h1 + a2h−1
= (h2 − h−1
2 )(h1 − h−1
1 )[a, b]
1 )(1 − c)ab.
(10)
SYMMETRIC ELEMENTS UNDER ORIENTED INVOLUTIONS
9
If α ∈ F P , h ∈ P , then
[α(1 − c)a, b(h + b2h−1)] = α(1 − c)(h + b2h−1)[a, b]
= α(h − h−1)(1 − c)2ab = 2α(h − h−1)(1 − c)ab.
[a(h1 + a2h−1
1 ), bg(h2 − h−1
2 )] = (h2 − h−1
= g(h2 − h−1
= g(h2 − h−1
2 )(h1 + a2h−1
2 )(h1 + ch−1
2 )(h1 − h−1
1 )[a, bg]
1 )(1 − c)ab
1 )(1 − c)ab,
and
[ag(h1 − h−1
1 ), bg(h2 − h−1
2 )] = (h2 − h−1
= (h2 − h−1
= −(h2 − h−1
2 )(h1 − h−1
2 )(h1 − h−1
1 )[ag, bg]
1 )g2(1 − c)ab
1 )(1 − c)ab.
2 )(h1 − h−1
(11)
(12)
(13)
The equations (11),
Suppose that
γn−1((F G)+) = Mn−1 for some n ≥ 3. Take α ∈ F P , h ∈ P and a, b ∈ (Q8 × E) \ ζ(Q8 × E),
such that (a, b) 6= 1. We get the following equalities:
(12) and (13) imply that γ2((F G)+) = M2.
and
[α(1 − c)a, gb(h − h−1)] = α(h − h−1)(1 − c)[a, gb]
= gα(h − h−1)(1 − c)[a, b]
= gα(h − h−1)(1 − c)2ab
= 2gα(h − h−1)(1 − c)ab,
[gα(1 − c)a, gb(h − h−1)] = g2α(h − h−1)(1 − c)[a, b]
= cα(h − h−1)(1 − c)[a, b]
= cα(h − h−1)(1 − c)2ab
= −2α(h − h−1)(1 − c)ab.
(14)
(15)
substituting
By
[Mn−1, (F G)+] = Mn and therefore
f1,...,n−1
for
α
in
(11),
(14)
and
(15),
we
get
that
γn((F G)+) = [γn−1((F G)+), (F G)+] = [Mn−1, (F G)+] = Mn,
as we wanted to prove.
(cid:3)
Acknowledgements
The results of this paper are part of the author's Ph.D. thesis, at the Instituto de Matem´a-
tica e Estat´ıstica of the Universidade de Sao Paulo, under the guidance of Prof. C´esar Polcino
Milies. This work was partially supported by CAPES and CNPq. proc. 141857/2011-0 of
Brazil.
References
1. Z. Balogh and T. Juh´asz, Derived lengths of symmetric and skew symmetric elements in group algebras,
JP J. Algebra Number Theory Appl. 12 (2008), no. 2, 191 -- 203. MR 2500081 (2010b:16045)
2.
, Nilpotency class of symmetric units of group algebras, Publ. Math. Debrecen 79 (2011), no. 1-2,
171 -- 180. MR 2850041 (2012h:16048)
10
3.
, Nilpotency indices of symmetric elements of group algebras, Comm. Algebra 40 (2012), no. 11,
J.H. CASTILLO
4283 -- 4294.
4. V. Bovdi and E. Spinelli, Modular group algebras with maximal Lie nilpotency indices, Publ. Math. Debre-
cen 65 (2004), no. 1-2, 243 -- 252. MR 2075267 (2005d:16040)
5. J. H. Castillo and C. Polcino Milies, Lie properties of symmetric elements under oriented involutions,
Comm. Algebra (to appear 2012).
6. A. Giambruno and S. K. Sehgal, Lie nilpotence of group rings, Comm. Algebra 21 (1993), no. 11, 4253 --
4261. MR 1238157 (94g:20008)
7. S. A. Jennings, The structure of the group ring of a p-group over a modular field, Trans. Amer. Math. Soc.
50 (1941), 175 -- 185. MR 0004626 (3,34f)
8. G. T. Lee, Group rings whose symmetric elements are Lie nilpotent, Proc. Amer. Math. Soc. 127 (1999),
no. 11, 3153 -- 3159. MR 1641124 (2000b:16052)
9.
10.
, The Lie n-Engel property in group rings, Comm. Algebra 28 (2000), no. 2, 867 -- 881. MR 1736769
(2001b:16027)
, Group identities on units and symmetric units of group rings, Algebra and Applications, vol. 12,
Springer-Verlag London Ltd., London, 2010. MR 2723223
11. S. P. Novikov, Algebraic construction and properties of Hermitian analogs of K-theory over rings with
involution from the viewpoint of Hamiltonian formalism. Applications to differential topology and the theory
of characteristic classes. I. II, Izv. Akad. Nauk SSSR Ser. Mat. 34 (1970), 253 -- 288; ibid. 34 (1970), 475 -- 500.
MR 0292913 (45 #1994)
12. I. B. S. Passi, D. S. Passman, and S. K. Sehgal, Lie solvable group rings, Canad. J. Math. 25 (1973),
748 -- 757. MR 0325746 (48 #4092)
13. C. Polcino Milies and S. K. Sehgal, An introduction to group rings, Algebras and Applications, vol. 1,
Kluwer Academic Publishers, Dordrecht, 2002. MR 1896125 (2003b:16026)
14. S. K. Sehgal, Topics in group rings, Monographs and Textbooks in Pure and Applied Math., vol. 50, Marcel
Dekker Inc., New York, 1978. MR 508515 (80j:16001)
15. R. K. Sharma and V. Bist, A note on Lie nilpotent group rings, Bull. Austral. Math. Soc. 45 (1992), no. 3,
503 -- 506. MR 1165157 (93g:20011)
John H. Castillo, Departamento de Matem´aticas y Estad´ıstica, Universidad de Narino
E-mail address: [email protected]@gmail.com
|
1511.07683 | 2 | 1511 | 2016-02-15T16:57:46 | Central extensions of null-filiform and naturally graded filiform non-Lie Leibniz algebras | [
"math.RA",
"math.KT"
] | In this paper we describe central extensions of some nilpotent Leibniz algebras. Namely, central extensions of the Leibniz algebra with maximal index of nilpotency are classified. Moreover, non-split central extensions of naturally graded filiform non-Lie Leibniz algebras are described up to isomorphism. It is shown that $k$-dimensional central extensions ($k\geq 5$) of these algebras are split. | math.RA | math |
CENTRAL EXTENSIONS OF NULL-FILIFORM AND NATURALLY GRADED
FILIFORM NON-LIE LEIBNIZ ALGEBRAS.
J.K. ADASHEV, L.M. CAMACHO, AND B.A. OMIROV
Abstract. In this paper we describe central extensions of some nilpotent Leibniz algebras. Namely,
central extensions of the Leibniz algebra with maximal index of nilpotency are classified. Moreover,
non-split central extensions of naturally graded filiform non-Lie Leibniz algebras are described up to
isomorphism. It is shown that k-dimensional central extensions (k ≥ 5) of these algebras are split.
Mathematics Subject Classification 2010: 17A32, 17B30.
Key Words and Phrases: Leibniz algebra, filiform algebra, quasi-filiform algebra, natural grada-
tion, characteristic sequence, 2-cocycles, central extension.
1. Introduction
Central extensions play an important role in quantum mechanics: one of the earlier encounters is by
means of Wigner's theorem which states that a symmetry of a quantum mechanical system determines
a (anti-) unitary transformation of a Hilbert space.
Another area of physics where one encounters central extensions is the quantum theory of conserved
currents of a Lagrangian. These currents span an algebra which is closely related to so called affine
Kac-Moody algebras, which are the universal central extension of loop algebras.
Central extensions are needed in physics, because the symmetry group of a quantized system usually
is a central extension of the classical symmetry group, and in the same way the corresponding symmetry
Lie algebra of the quantum system is, in general, a central extension of the classical symmetry algebra.
Kac-Moody algebras have been conjectured to be a symmetry groups of a unified superstring theory.
The centrally extended Lie algebras play a dominant role in quantum field theory, particularly in
conformal field theory, string theory and in M -theory.
In the theory of Lie groups, Lie algebras and their representations, a Lie algebra extension is an
enlargement of a given Lie algebra g by another Lie algebra h. Extensions arise in several ways.
There is a trivial extension obtained by taking a direct sum of two Lie algebras. Other types are split
extension and central extension. Extensions may arise naturally, for instance, when forming a Lie
algebra from projective group representations.
A central extension and an extension by a derivation, one obtains a Lie algebra which is isomorphic
with a non-twisted affine Kac-Moody algebra ([2], Chapter 19). Using the centrally extended loop
algebra one may construct a current algebra in two spacetime dimensions. The Virasoro algebra is the
universal central extension of the Witt algebra, the Heisenberg algebra is the central extension of a
commutative Lie algebra ([2], Chapter 18), [4], [13].
During last 20 years Leibniz algebras are studied mainly to extend the classical results from Lie
algebras. From structural point of view some classical analogues which are true for nilpotent Lie
algebras are obtained. In particular, papers devoted to the study of naturally graded nilpotent Leibniz
algebras are [1], [5], [6], [7], [8] and many others.
In the present paper we consider a notion of a central extension of Leibniz algebras, which is
similar to the notion of a central extension of Lie algebras [12]. Here we consider central extensions
of some nilpotent Leibniz algebras, such as null-filiform Leibniz algebra and naturally graded non-
Lie filiform Leibniz algebras.
In fact, the method of central extensions of Lie algebras is adapted
for Leibniz algebras in [10]. In works [10]-[11] one-dimensional central extensions of the null-filiform
Leibniz algebra and the naturally graded Lie algebra are described. The difficulties of the description
of one-dimensional central extensions of a naturally graded Lie algebra show that the further progress
in description of filiform Lie algebras by using central extensions is too complicated. That is why in
our study we restrict to non-Lie Leibniz algebras.
The main results of this paper consists of the classification of central extensions of null-filiform
Leibniz algebras and naturally graded filiform non-Lie Leibniz algebras.
1
2
J.K. ADASHEV, L.M. CAMACHO, AND B.A. OMIROV
Throughout the paper all vector spaces and algebras are finite-dimensional over the field of complex
numbers. Moreover, in the table of multiplication of an algebra the omitted products are assumed to
be zero.
Let F be an algebraically closed field of characteristic 0.
2. Preliminaries
Definition 2.1. [9] An algebra L over a field F is called a Leibniz algebra if for any elements x, y, z ∈ L
the so-called Leibniz identity is satisfied:
[x, [y, z]] = [[x, y], z] − [[x, z], y],
where [−,−] is a multiplication in L.
Let L be a finite-dimensional Leibniz algebra. For this algebra we define sequence:
L1 = L, Lk+1 = [Lk, L1],
k ≥ 1.
A Leibniz algebra L is called nilpotent if there exists s ∈ N such that Ls = {0}. The minimal
number s possessing of such property is called index of nilpotency of the algebra L.
Definition 2.2. [1] A Leibniz algebra L of dimension n is called null-filiform if dimLi = (n+1)−i, 1 ≤
i ≤ n + 1.
Evidently, by definition algebra being null-filiform is equivalent that it has maximal possible index
of nilpotency.
Theorem 2.3.
{e1, e2, . . . , en} such that the table of multiplication of the algebra is in the following form:
[1] In arbitrary n-dimensional null-filiform Leibniz algebra there exists a basis
N Fn :
[ei, e1] = ei+1, 1 ≤ i ≤ n − 1.
Definition 2.4. [1] A Leibniz algebra L is called filiform if dimLi = n − i, 2 ≤ i ≤ n.
Note that the nilindex of a filiform Leibniz algebras coincides with its dimension.
Let L be a Leibniz algebra of nilindex s. We put Li = Li/Li+1, 1 ≤ i ≤ s − 1. Then we obtain a
graded algebra GrL = L1 ⊕ L2 ⊕ ··· ⊕ Ls−1, where [Li, Lj] ⊆ Li+j. An algebra L is called a naturally
graded algebra if L ∼= GrL.
In the following theorem we summarize the results of the works [1], [14].
Theorem 2.5. An arbitrary complex n-dimensional naturally graded filiform Leibniz algebra is iso-
morphic to one of the following pairwise non-isomorphic algebras:
F 1
n :
[e1, e1] = e3,
[ei, e1] = ei+1, 2 ≤ i ≤ n − 1,
F 2
n :
[e1, e1] = e3,
[ei, e1] = ei+1, 3 ≤ i ≤ n − 1,
F 3
n :
[ei, e1] = −[e1, ei] = ei+1,
2 ≤ i ≤ n − 1
[ei, en+1−i] = −[en+1−i, ei] = α(−1)i+1en, 2 ≤ i ≤ n − 1,
where α ∈ {0, 1} for even n and α = 0 for odd n.
It is known that the set of filiform Leibniz algebras is decomposed into three disjoint classes, two
n and one family of algebras
families of non-Lie Leibniz algebras which arise from the algebras F 1
(which include filiform Lie algebras) arises from the Lie algebra F 3
n and F 2
n [8].
Since further we consider only filiform non-Lie Leibniz algebras we omit the third family of algebras.
CENTRAL EXTENSIONS OF SOME NILPOTENT LEIBNIZ ALGEBRAS.
3
Theorem 2.6. Any (n + 1)-dimensional complex non-Lie filiform Leibniz algebra, whose naturally
graded algebra is not a Lie algebra, belongs to one of the following two classes:
F 1
n(α4, α5, . . . , αn, θ) :
F 2
n(β4, β5, . . . , βn, γ) :
2 ≤ i ≤ n − 1,
αkek + θen,
αk+1−iek,
2 ≤ i ≤ n − 2,
3 ≤ i ≤ n − 1,
[e1, e1] = e3,
[ei, e1] = ei+1,
[e1, e2] =
[ei, e2] =
n−1Pk=4
nPk=i+2
[e1, e1] = e3,
[ei, e1] = ei+1,
[e1, e2] =
βkek,
[e2, e2] = γen,
nPk=4
nPk=i+2
[ei, e2] =
βk+1−iek,
3 ≤ i ≤ n − 2.
Definition 2.7. An algebra L is called quasi-filiform if Ln−2 6= {0} and Ln−1 = {0}, where dimL = n.
Let x be a nilpotent element of the set L\ L2. For the nilpotent operator of right multiplication Rx
we define a decreasing sequence C(x) = (n1, n2, . . . , nk), which consists of the dimensions of Jordan
blocks of the operator Rx. In the set of such sequences we consider the lexicographic order, that is,
C(x) = (n1, n2, . . . , nk) ≤ C(y) = (m1, m2, . . . , ms) if and only if there exists i ∈ N such that nj = mj
for any j < i and ni < mi.
Definition 2.8. The sequence C(L) = max C(x)x∈L\L2 is called characteristic sequence of the algebra
L.
Definition 2.9. A quasi-filiform non Lie Leibniz algebra L is called algebra of type I, if there exists
an element x ∈ L\L2 such that the operator Rx has the form: (cid:18) Jn−2
0
0
J2 (cid:19) .
Combining the results of the papers [5] and [7] we have the following theorem.
Theorem 2.10. Let L be a naturally graded quasi-filiform Leibniz algebra of type I. Then it is
isomorphic to the one of the following pairwise non-isomorphic algebras:
L1,λ
L3,λ
L1
n :
n :
n :
n
:
L5,λ,µ
[ei, e1] = ei+1,
[e1, en−1] = en + e2,
[ei, en−1] = ei+1,
N Fn−2 ⊕ C2; F 1
1 ≤ i ≤ n − 3,
n−1 ⊕ C;
n : (cid:26) [ei, e1] = ei+1,
[e1, en−1] = en.
L2
2 ≤ i ≤ n − 3.
1 ≤ i ≤ n − 3,
[ei, e1] = ei+1, 1 ≤ i ≤ n − 3
[en−1, e1] = en,
[e1, en−1] = λen, λ ∈ C
[ei, e1] = ei+1, 1 ≤ i ≤ n − 3
[en−1, e1] = en + e2,
[e1, en−1] = λen, λ ∈ {−1, 0, 1}
[ei, e1] = ei+1, 1 ≤ i ≤ n − 3
[en−1, e1] = en + e2,
[e1, en−1] = λen, (λ, µ) = (1, 1) or (2, 4)
[en−1, en−1] = µen,
[ei, e1] = ei+1, 1 ≤ i ≤ n − 3
[en−1, e1] = en,
[e1, en−1] = λen, λ ∈ {0, 1}
[en−1, en−1] = en
[ei, e1] = ei+1, 1 ≤ i ≤ n − 3
[en−1, e1] = en + e2,
[en−1, en−1] = λen, λ 6= 0
[ei, e1] = ei+1, 1 ≤ i ≤ n − 3
[en−1, e1] = en,
[e1, en−1] = −en,
[en−1, en−1] = e2,
[en−1, en] = e3.
L2,λ
L4,λ
n :
n :
n :
L6
Central extensions of Leibniz algebras. It is remarkable that any nilpotent Leibniz algebra
has non-trivial center. Let L be a Leibniz algebra over a field F and Z(L) be its center.
The central extension of a Leibniz algebra L by V is called a short exact sequence of Leibniz algebras
0 −→ V ε−→bL λ−→ L −→ 0,
4
J.K. ADASHEV, L.M. CAMACHO, AND B.A. OMIROV
where Imε = Kerλ and V is the center of the algebra bL.
Two central extensions
0 −→ V ǫ−→bL λ−→ L −→ 0 and 0 −→ V ǫ′
−→bL′ λ′
−→ L −→ 0
are called equivalent if there exists an isomorphism ϕ : bL −→ bL′ such that ϕ ◦ ǫ = ǫ′, λ′ ◦ ϕ = λ.
A bilinear map θ : L × L −→ V satisfying an identity
θ(x, [y, z]) = θ([x, y], z) − θ([x, z], y),
for any elements x, y, z ∈ L is called central 2-cocycle.
We denote by ZL2(L, V ) the set of all central 2-cocycles from L to V . If for linear map ϕ : L −→ V
we have θ(x, y) = ϕ([x, y]), then θ is called 2-coboundary. We denote by BL2(L, V ) the set of all
2-coboundaries from L to V . The quotient space HL2(L, V ) := ZL2(L, V )/BL2(L, V ) is called the
2-nd group of cohomology. If θ1 − θ2 is coboundary, then θ1 and θ2 are called cohomological.
vector space Lθ = L ⊕ V with equipped multiplication:
For a given 2-cocycle θ on L we construct the associated central extension Lθ. By definition it is a
[x + u, y + v] = [x, y]L + θ(x, y),
(1)
where [·,·]L is the bracket on L and x, y ∈ L, u, v ∈ V.
3. Classification of k-dimensional central extensions of null-filiform Leibniz
algebras
In this section, we focus on k-dimensional central extensions of null-filiform Leibniz algebra. First,
we compute a basis of HL2(N Fn, V ).
Proposition 3.1. Let L be the null-filiform Leibniz algebra and let V =< x1, x2, . . . , xk > be an
abelian algebra. Then
• A basis of ZL2(N Fn, V ) is formed by the following cocycles
θi,j (ei, e1) = xj, 1 ≤ i ≤ n, 1 ≤ j ≤ k
• A basis of BL2(N Fn, V ) is formed by the following coboundaries
θi,j(ei, e1) = xj, 1 ≤ i ≤ n − 1, 1 ≤ j ≤ k
• A basis of HL2(N Fn, V ) is formed by the following cocycles
θj(en, e1) = xj, 1 ≤ j ≤ k
Proof. The proof follows directly from the definition of a cocycle.
(cid:3)
Theorem 3.2. A central extension of an n-dimensional complex null-filiform Leibniz algebra is iso-
morphic to one of the following non-isomorphic algebras:
N Fn ⊕ Ck, N Fn+1 ⊕ Ck−1.
Proof. Let {e1, e3, . . . , en} be a basis of N Fn and {x1, . . . , xk} be a basis of V. From the definition of
the product on central extension (1) we conclude that {e1, e3, . . . , en, x1, x2, . . . , xk} forms a basis of
the algebra L = N Fn ⊕ V.
Applying Proposition 3.1, we get that the table of multiplication of the central extension of N Fn
has the following form:
[ei, e1] = ei+1,
[en, e1] =
αixi,
kPi=1
1 ≤ i ≤ n − 1,
If αi = 0 for all 1 ≤ i ≤ k, then we obtain the algebra N Fn ⊕ Ck.
If there exists αi 6= 0 for some i, then setting en+1 =
kPi=1
αixi we get the algebra N Fn+1 ⊕ Ck−1. (cid:3)
CENTRAL EXTENSIONS OF SOME NILPOTENT LEIBNIZ ALGEBRAS.
5
4. Classification of k-dimensional central extensions of naturally graded non-Lie
filiform Leibniz algebras
In this section we present the classification of k-dimensional central extensions of naturally graded
non-Lie filiform Leibniz algebras. Let V =< x1, x2, . . . , xk > be an abelian algebra. First, we compute
a basis of HL2(F i
n, V ) with i = 1, 2.
Proposition 4.1.
• The following cocycles
θi,j (ei, e1) = xj, 1 ≤ i ≤ n, θn+1,j(e1, e2) = xj, θn+2,j(e2, e2) = xj
form a basis of ZL2(F 1
n, V ) and ZL2(F 2
n, V );
• The following coboundaries
form a basis of BL2(F 1
a) θ1,j(e1, e1) = xj, θ1,j(e2, e1) = xj , θi,j(ei+1, e1) = xj, 2 ≤ i ≤ n − 2, 1 ≤ j ≤ k,
b) θ1,j(e1, e1) = xj, θi,j(ei+1, e1) = xj , 2 ≤ i ≤ n − 2, 1 ≤ j ≤ k,
n, V ),
form a basis of BL2(F 2
• Basis of the spaces HL2(F 1
n, V );
n , V ) and HL2(F 2
n , V ) are formed by the following cocycles
θ1,j(e2, e1) = xj , θ2,j(en, e1) = xj , θ3,j(e1, e2) = xj, θ4,j(e2, e2) = xj , 1 ≤ j ≤ k.
Proof. The proof is carried out by straightforward calculations.
(cid:3)
One-dimensional central extensions.
dimensional central extensions of the algebras F 1
n and F 2
n.
In this subsection we study the classification of one-
Theorem 4.2. An arbitrary one-dimensional central extension of the complex algebra F 1
to one of the following pairwise non-isomorphic algebras:
n is isomorphic
F 1
n ⊕ C; F 1
n+1(0, . . . , 0, α4, α3), α3, α4 ∈ {0, 1}; L3,λ
n+1; L4,λ
n+1; L5,λ,µ
n+1 ; L1
n+1.
Proof. From Proposition 4.1 and the construction of central extensions, we derive that the table of
multiplication of one-dimensional extension of F 1
n has the form:
[e1, e1] = e3,
[e2, e1] = e3 + α1x,
[ei, e1] = ei+1,
[en, e1] = α2x,
3 ≤ i ≤ n − 1,
[e1, e2] = α3x,
[e2, e2] = α4x.
We distinguish the following cases:
n ⊕ C.
• If αi = 0 for any 1 ≤ i ≤ 4, then we get algebra F 1
• If there exists i such that αi 6= 0, then we consider the cases.
Case 1. Let α2 6= 0, then one can assume α2 = 1. Note that dimLi = n + 1 − i for 2 ≤ i ≤ n + 1.
Consequently we have (n + 1)-dimensional filiform Leibniz algebra. Making the following change of
basis e′
n+1 = x, we derive α1 = 0. Hence, in this case we obtain the family of
algebras F 1
2 = e2 − α1en and e′
n+1(0, 0, . . . , 0, α4, α3).
In the work [8], the authors show that instead the general change of basis in the family
n+1(α4, α5, . . . , αn+1, θ) it is sufficient to consider the following change of basis:
F 1
e′
1 = Ae1 + Be2,
e′
i = Ai−2(A + B)ei, 4 ≤ i ≤ n + 1,
From [e′
n+1 and [e′
2] = α′
2, e′
1, e′
3e′
e′
2 = (A + B)e2 + B(α3 − α4)en,
2] = α′
4e′
n+1, we obtain
e′
3 = A(A + B)e3 + (ABα3 + B2α4)en+1,
A + B 6= 0.
α′
3 =
Aα3 + Bα4
An−1
, α′
4 =
(A + B)α4
An−1
.
Consider subcases.
a) α4 = 0. Then α′
4 = 0 and α′
3 = α3
An−2 .
If α3 = 0, then α′
3 = 0. Otherwise, putting A = n−2√α3 we have that α′
3 = 1. Therefore, we
have F 1
n+1(0, . . . , 0, 0, α3), where α3 ∈ {0, 1}.
6
J.K. ADASHEV, L.M. CAMACHO, AND B.A. OMIROV
b) α4 6= 0. Putting B = An−1−Aα4
α4
If α3 = α4, then α′
obtain the algebras F 1
, we obtain α′
3 = α3−α4+An−2
3 = 1, otherwise, putting A = n−2√α4 − α3, we get α′
n+1(0, . . . , 0, 1, α3), where α3 ∈ {0, 1}.
4 = 1 and α′
An−2
.
3 = 0. Therefore, we
Thus, we have F 1
n+1(0, . . . , 0, α4, α3), where α3, α4 ∈ {0, 1}.
Case 2. Let α2 = 0. Taking the change of basis
e′
1 = e1, e′
i = ei+1, 2 ≤ i ≤ n − 1, e′
n = e2, e′
n+1 = x,
we get the algebra
[ei, e1] = ei+1,
[en, e1] = e2 + α1en+1,
1 ≤ i ≤ n − 2,
[e1, en] = α3en+1,
[en, en] = α4en+1.
Note that this algebra is (n + 1)-dimensional quasi-filiform Leibniz algebra.
• Case 2.1. α1 6= 0.
If α4 = 0, then we get the algebra which is isomorphic to L3,λ
n+1.
If α3 = 0, α4 6= 0, then obtained algebra is isomorphic to L4,λ
n+1.
If α3 6= 0, α4 6= 0, then derived algebra is isomorphic to L5,λ,µ
n+1 .
• Case 2.2. α1 = 0. Take the change of basis
e′
1 = e1 + en, e′
2 = 2e2 + (α3 + α4)en+1, e′
i = 2ei, 3 ≤ i ≤ n, e′
n+1 = en+1.
Then we obtain the algebra
[ei, e1] = ei+1,
[en, e1] = e2 + (α4 − α3)en+1,
1 ≤ i ≤ n − 2,
[e1, en] = 2(α3 + α4)en+1,
[en, en] = 4α4en+1.
It is easy to see that if α3 6= α4, then we are in the case of α1 6= 0. Therefore, we may assume
that α3 = α4 6= 0. Making the change of basis
e′
i = ei, 1 ≤ i ≤ n − 1, e′
n = e1 − en, e′
n+1 = −4α3en+1,
we deduce the algebra L1
n+1.
Since obtained algebras in the case 1 are filiform and the algebras of the case 2 are quasi-filiform,
(cid:3)
they are non isomorphic.
The next theorem describes one-dimensional central extensions of the algebra F 2
n.
Theorem 4.3. An arbitrary one-dimensional central extension of the algebra F 2
of the following pairwise non-isomorphic algebras:
n is isomorphic to one
n ⊕ C; F 2
F 2
n+1; F 2
n+1(0, . . . , 0, 0, 1); F 2
n+1(0, . . . , 0, 1, 0); L1,λ
n+1; L2,λ
n+1; L2
n+1.
Proof. From Proposition 4.1 and from the construction of the central extensions, we derive that the
table of multiplication of one-dimensional extension of F 2
n has the form:
[e1, e1] = e3,
[e2, e1] = α1x,
3 ≤ i ≤ n − 1,
[e1, e2] = α3x,
If αi = 0 for any 1 ≤ i ≤ 4, then we get split algebra F 2
n ⊕ C.
Let αi 6= 0 for some i. Then we consider the possible cases.
[ei, e1] = ei+1,
[en, e1] = α2x,
[e2, e2] = α4x.
Case 1. Let α2 6= 0. Scaling of basis elements we can assume α2 = 1. Since dimLi = n + 1 − i for
2 ≤ i ≤ n + 1, we have (n + 1)-dimensional filiform Leibniz algebra. Making the change e′
2 = e2 − α1en
and e′
n+1(0, 0, . . . , 0, α3, α4).
Arguing similarly as in the proof of Theorem 4.2 we obtain the following pairwise non-isomorphic
n+1 = x, we obtain α1 = 0. So, in this case we get the family of algebras F 2
algebras:
F 2
n+1, F 2
n+1(0, . . . , 0, 1, 0), F 2
n+1(0, . . . , 0, 0, 1).
Case 2. Let α2 = 0. Making the following change of basis:
i = ei+1, 2 ≤ i ≤ n − 1, e′
e′
1 = e1, e′
n = e2, e′
n+1 = x,
CENTRAL EXTENSIONS OF SOME NILPOTENT LEIBNIZ ALGEBRAS.
7
we get the family of algebras
[ei, e1] = ei+1,
[en, e1] = α1en+1,
1 ≤ i ≤ n − 2,
[e1, en] = α3en+1,
[en, en] = α4en+1.
Note that these algebras are (n + 1)-dimensional quasi-filiform Leibniz algebras of type I.
• Case 2.1. Let α1 6= 0. Then by rescaling basis elements we can assume α1 = 1.
If α4 = 0, then obtained algebra is isomorphic to the algebra L1,λ
If α4 6= 0, then derived algebra is isomorphic to the algebra L2,λ
n+1.
n+1.
• Case 2.2. Let α1 = 0. Let us take the change of basis
2 = e2 + (α3 + α4)en+1, e′
e′
1 = e1 + en, e′
i = ei, 3 ≤ i ≤ n + 1.
Then we obtain the family of algebras:
[ei, e1] = ei+1,
[en, e1] = α4en+1,
1 ≤ i ≤ n − 2,
[e1, en] = (α3 + α4)en+1,
[en, en] = α4en+1.
It is easy to see that if α4 6= 0, then we are in conditions of the case α1 6= 0. Therefore,
we assume that α4 = 0 and α3 6= 0. Then making the change of basis e′
n+1 = α3en+1, we get
L2
n+1.
(cid:3)
Two-dimensional central extensions. This subsection is devoted to the study of classification of
two-dimensional non-split central extensions of the algebras F 1
n and F 2
n.
First, we consider two-dimensional central extensions of the algebra F 1
n. From Proposition 4.1 and
from the definition of central extensions, we conclude that the table of multiplication of two-dimensional
central extension of the algebra F 1
n has the form:
[e1, e1] = e3,
[ei, e1] = ei+1,
[en, e1] = α2x1 + β2x2,
[e2, e1] = e3 + α1x1 + β1x2,
3 ≤ i ≤ n − 1,
[e1, e2] = α3x1 + β3x2, [e2, e2] = α4x1 + β4x2.
Let us take the following change of basis
e′
1 = e1, e′
2 = e3, e′
i = ei+1, 2 ≤ i ≤ n − 1, e′
n = e2, en+1 = x1, en+2 = x2.
Then we obtain:
1 ≤ i ≤ n − 2,
[ei, e1] = ei+1,
[en, e1] = e2 + α1en+1 + β1en+2,
[en−1, e1] = α2en+1 + β2en+2,
[e1, en] = α3en+1 + β3en+2,
[en, en] = α4en+1 + β4en+2.
Theorem 4.4. An arbitrary two-dimensional non-split central extension of the algebra F 1
phic to one of the following pairwise non-isomorphic algebras:
n is isomor-
• L(1, 0, 0, 0), L(1, 1, 0, 0), L(0, 1, 0, 0), L(0, 1, 1, 0), L(1, 0, 0, β4), β4
6= 0, L(0, 1,−1, 0),
L(0, 1, 1, 1), L(0, 1, 2, 4),
where
L(α3, α4, β3, β4)
[ei, e1] = ei+1,
[en+1, e1] = e2 + en+2,
[e1, en+1] = α3en + β3en+2,
[en+1, en+1] = α4en + β4en+2,
1 ≤ i ≤ n − 1,
• M (1, β4), β4 6= 1, M (0, 0), M (0, 1), M (0, 2), M (2, 1), M (2, 2), M (1, 1),
where
M (α4, β4) :
[ei, e1] = ei+1,
[en, e1] = e2 + en+1,
[e1, en] = en+2,
[en, en] = α4en+1 + β4en+2,
1 ≤ i ≤ n − 2,
8
J.K. ADASHEV, L.M. CAMACHO, AND B.A. OMIROV
n+2, L3,λ
n+2, L4,λ
n+2, L5,λ,µ
n+2 and the algebra
• L1
L∗ :
[ei, e1] = ei+1,
[e1, en+1] = e2 + en+2,
[ei, en+1] = ei+1,
[en+1, en+1] = en.
1 ≤ i ≤ n − 1,
2 ≤ i ≤ n − 1,
Proof. We distinguish the following cases:
Case 1. Let (α2, β2) 6= (0, 0). Without loss of generality, one can assume α2 6= 0 and [en−1, e1] = en+1.
Making the change e′
n+1 = en, we obtain the
table of multiplication:
n = en − α1en−1, we obtain α1 = 0. Setting e′
n = en+1, e′
[ei, e1] = ei+1,
[en+1, e1] = e2 + β1en+2,
[e1, en+1] = α3en + β3en+2,
[en+1, en+1] = α4en + β4en+2.
1 ≤ i ≤ n − 1,
Case 1.1. Let βi = 0 for 1 ≤ i ≤ 4. Then we get a split central extension.
Case 1.2. Let β1 6= 0. Putting e′
n+2 = β1en+2, we can assume that β1 = 1.
Since dimL2 = n and dimLi = n + 1 − i for 3 ≤ i ≤ n, we have (n + 2)-dimensional quasi-filiform
Let us take the general change of basis:
Leibniz algebra with characteristic sequence equal to C(L) = (n, 2).
n+2Xi=1
Then we express new basis elements {e′
2 we have
e′
1 =
for e′
Aiei,
e′
n+1 =
n+2Xi=1
Biei.
1, e′
2, . . . , e′
n+2} via basis elements {e1, e2, . . . , en+2}. Then
e′
2 = [e′
1, e′
1] = A1(A1 + An+1)e2 + A1
n−2Xi=2
Aiei+1+
+(A1An−1 + A1An+1α3 + A2
Further, we obtain the expressions for e′
n+1α4)en + (A1An+1 + A1An+1β3 + A2
t with 3 ≤ t ≤ n, namely:
n−t+1Xi=2
Aiei+t−1, 3 ≤ t ≤ n − 1,
1
n = An−1
e′
n+1β4)en+2.
t = At−1
e′
1
(A1 + An+1)et + At−1
1
(A1 + An+1)en.
Therefore, A1(A1 + An+1) 6= 0.
From equations [e′
Since e′
n−1, e′
n+2 = [e′
n+1, e′
n+1] = An−2
1
2, we deduce
1] − e′
(A1 + An+1)B1en = 0 we derive B1 = 0.
e′
n+2 = (A1Bn+1 − A1(A1 + An+1))e2 + A1
n−2Xi=2
(Bi − Ai)ei+1 + [(A1Bn−1 + An+1Bn+1α4)−
(A1An−1+A1An+1α3+A2
n+1α4)]en+[(A1Bn+1+An+1Bn+1β4)−(A1An+1+A1An+1β3+A2
n+1β4)]en+2.
Taking into account that in new basis [e′
n+2, e′
1] = 0, we derive restrictions:
Considering the products
Bn+1 = A1 + An+1, Bi = Ai, 2 ≤ i ≤ n − 2.
[e′
1, e′
n+1] = α′
3e′
n + β′
3e′
n+2,
[e′
n+1, e′
n+1] = α′
4e′
n + β′
4e′
n+2,
,
(1)
(2)
(3)
(4)
CENTRAL EXTENSIONS OF SOME NILPOTENT LEIBNIZ ALGEBRAS.
9
we derive expressions for the new parameters:
A2
1α3 + A1An+1α4 + An+1Bn+1(α3β4 − α4β3) + A1(An−1 − Bn−1)β3 + An+1(An−1 − Bn−1)β4
An−1
1
(A1 + An+1β4 − An+1β3)
Bn+1(A1α4 + An+1(α3β4 − α4β3) + (An−1 − Bn−1)β4)
,
An−1
1
(A1 + An+1β4 − An+1β3)
α′
3 =
α′
4 =
β′
3 =
β′
4 =
(A1β3 + An+1β4)Bn+1
A1(A1 + An+1β4 − An+1β3)
β4B2
n+1
A1(A1 + An+1β4 − An+1β3)
,
,
with condition
Consider now the possible cases.
A1(A1 + An+1)(A1 + An+1β4 − An+1β3) 6= 0.
Case 1.2.1. Let β4 = 0. Then β′
4 = 0 and we obtain that β3 is nullity invariant, that is, β′
3 = 0 if
and only if β3 = 0. Thus, we can distinguish two cases.
a) Let β3 = 0.
a.1) α4 = 0. Then α′
, α′
3 = α3
An−2
3 = 0. Otherwise, putting A1 = n−2√α3. Hence, we obtain α′
4 = 0.
1
3 = 1. In
n+2 and L(1, 0, 0, 0).
If α3 = 0, then α′
this subcase we get the algebras L3,0
, α′
3 = A1α3+An+1α4
1
a.2) α4 6= 0. Then α′
4 = α3−α4
An−2
4 = (A1+An+1)α4
give us one more nullity invariant.
An−1
An−1
α′
3 − α′
If α3 − α4 = 0, then putting An+1 = An−1
L(1, 1, 0, 0).
If α3 − α4 6= 0, then choosing An+1 = − A1α3
0, α′
4 = 1. Therefore, we get the algebra L(0, 1, 0, 0).
1 −A1α4
α4
α4
1
1
. Note that the following expression
, we have α′
3 = α′
4 = 1 and the algebra
and A1 = n−2√α4 − α3, we obtain α′
3 =
b) Let β3 6= 0. Then putting
An−1 − Bn−1 = −
A2
1α3 + A1An+1α4 − An+1Bn+1α4β3
A1β3
(A1 + An+1)β3
A1 − An+1β3
.
.
we obtain
α′
3 = 0, α′
4 =
(A1 + An+1)α4
An−1
1
, β′
3 =
From the last expression we deduce nullity invariant:
A1(β3 + 1)
A1 − An+1β3
β′
3 + 1 =
We distinguish the following subcases:
b.1) β3 6= −1. Putting An+1 = A1(1−β3)
2β3
If α4 = 0, then α′
4 = 0 and we have the algebra L3,1
n+2.
If α4 6= 0, then choosing A1 = n−2q (β3+1)α4
2β3
b.2) β3 = −1. It implies β′
3 = −1, α′
If α4 = 0, then α′
If α4 6= 0, then setting An+1 = An−1
An−1
4 = 0 and the algebra L3,−1
n+2 is obtained.
, we get α′
1 −A1α4
α4
4 = (A1+An+1)α4
.
1
, we obtain β′
3 = 1, α′
4 = (β3+1)α4
2An−2
β3
.
1
, we obtain α′
4 = 1 and the algebra L(0, 1, 1, 0).
4 = 1 and the algebra L(0, 1,−1, 0).
Case 1.2.2. Let β4 6= 0. Then similar as in naturally graded case (see the proof of Theorem 9 in
[7]) the values for (β3, β4) are only following: (0, β4), where β4 6= 0, (1, 1) and (2, 4).
a) Let (β3, β4) = (0, β4) with β4 6= 0. Then from expressions (1) − (4) we get
, An+1 = 0.
A1α4 + (An−1 − Bn−1)β4
, α′
α3
α′
3 =
4 =
An−1
1
An−1
1
10
J.K. ADASHEV, L.M. CAMACHO, AND B.A. OMIROV
Taking An−1 − Bn−1 = − A1α4
If α3 = 0, then we have α′
If α3 6= 0, choosing A1 = n−1√α3, we have α′
, we obtain α′
β4
4 = 0.
b) Let (β3, β4) = (1, 1). Then expressions (1) − (4) imply
3 = 0 and we obtain the algebra L4,λ
n+2 with λ 6= 0.
3 = 1 and the algebra L(1, 0, 0, β4), β4 6= 0.
α′
3 =
A1α3 + An−1 − Bn−1
, α′
4 =
A1α4 + (An−1 − Bn−1)
, An+1 = 0.
An−1
1
An−1
1
Setting An−1 − Bn−1 = −A1α3, we deduce α′
Note that the following expression
3 = 0 and α′
4 = α4−α3
An−2
.
1
α′
3 − α′
4 =
α3 − α4
An−2
1
is nullity invariant.
4 = 0 and the algebra L5,1,1
If α3 − α4 = 0, then we obtain α′
n+2 .
If α3−α4 6= 0, then choosing A1 = n−2√α4 − α3, we have α′
4 = 1 and the algebra L(0, 1, 1, 1).
c) Let (β3, β4) = (2, 4). Then expressions (1) − (4) transform to the following form:
, An+1 = 0.
A1α3 + 2(An−1 − Bn−1)
A1α4 + 4(An−1 − Bn−1)
, α′
α′
3 =
4 =
An−1
1
An−1
1
Putting An−1 − Bn−1 = − A1α3
It is easy to check that the following expression:
, we derive α′
2
3 = 0 and α′
4 = α4−2α3
An−2
.
1
2α′
3 − α′
4 =
2α3 − α4
An−2
1
is nullity invariant.
If α3 − 2α4 = 0, then we have α′
If α3 − 2α4 6= 0, then choosing A1 = n−2√α4 − 2α3, we obtain α′
4 = 0 and the algebra L5,2,4
n+2 .
L(0, 1, 2, 4).
4 = 1 and the algebra
Case 1.3. Let β1 = 0. Let us make the following change of basis:
1 = e1 +(α4−α3)en−1 +en+1, e′
e′
2 = 2e2 +2α4en +(β3 +β4)en+2, e′
Then the table of multiplication of the algebra we have the form:
i = 2ei, 3 ≤ i ≤ n+1, e′
n+2 = en+2.
[ei, e1] = ei+1,
[en+1, e1] = e2 + (β4 − β3)en+2,
[en+1, en+1] = 2α4en + 4β4en+2.
1 ≤ i ≤ n − 1,
[e1, en+1] = (α3 + α4)en + 2(β3 + β4)en+2,
We have β3 = β4 6= 0. Indeed, if β3 6= β4, then we are in the case of β1 6= 0.
Putting e′
n+2 = (α3 + α4)en + 4β3en+2 we obtain an (n + 2)-dimensional quasi-filiform Leibniz
algebra with C(L) = (n, 1, 1):
[ei, e1] = ei+1,
[en+1, e1] = e2,
[e1, en+1] = en+2,
[en+1, en+1] = α4en + en+2.
1 ≤ i ≤ n − 1,
Taking the general change of generator basis elements and applying similar arguments as in Case
1.2 we derive the expression for the new parameter:
α′
4 =
α4
An−2
1
.
Since α4 is nullity invariant, we can easily get that α4 is equal either 0 or 1. Therefore, we have
algebras:
1 ≤ i ≤ n − 1,
[ei, e1] = ei+1,
[en+1, e1] = e2,
[en+1, en+1] = en+2,
[e1, en+1] = en+2,
[ei, e1] = ei+1,
[en+1, e1] = e2,
[en+1, en+1] = en + en+2,
[e1, en+1] = en+2.
1 ≤ i ≤ n − 1,
CENTRAL EXTENSIONS OF SOME NILPOTENT LEIBNIZ ALGEBRAS.
11
The following change of basis:
leads to the algebras L1
e′
i = ei, 1 ≤ i ≤ n, e′
n+2 and L∗.
n+1 = e1 − en+1, e′
n+2 = −en+2.
Case 2. Let (α2, β2) = (0, 0) and (α1, β1) 6= (0, 0). Without loss of generality, we might assume that
α1 6= 0. Therefore, taking e′
n+1 = e2 + α1en+1 + β1en+2, we get [en, e1] = e2 + en+1.
Thus, we have the table of multiplications:
1 ≤ i ≤ n − 2,
[ei, e1] = ei+1,
[en, e1] = e2 + en+1,
[e1, en] = α3en+1 + β3en+2,
[en, en] = α4en+1 + β4en+2.
Note that dimL2 = n, dimLi = n + 1 − i, 3 ≤ i ≤ n.
Case 2.1. Let β3 6= 0. Putting e′
have the family of algebras:
n+2 = α3en+1 + β3en+2, we can assume α3 = 0, β3 = 1. Then we
M (α4, β4) :
[ei, e1] = ei+1,
[en, e1] = e2 + en+1,
[e1, en] = en+2,
[en, en] = α4en+1 + β4en+2.
1 ≤ i ≤ n − 2,
Similar as in Case 1.2 we get the expressions for the new parameters
α′
4 =
β′
4 =
B2
nα4
A2
1 + A1Anβ4 + A1Anα4 + A2
nα4
Bn(A1β4 + Anα4)
A2
1 + A1Anβ4 + A1Anα4 + A2
nα4
,
,
with restriction
Consider
A1(A1 + An)(A2
1 + A1Anβ4 + A1Anα4 + A2
nα4) 6= 0.
• Let α4 = 0. Then α′
4 = 0 and β′
4 = (A1+An)β4
A1+Anβ4
.
If β4 = 0, then we obtain the algebra M (0, 0).
If β4 6= 0, then we have β′
4 − 1 = A1(β4−1)
imply
nullity invariant. Therefore the assumption β4 = 1 deduce the algebra M (0, 1) and in opposite
assumption (that is, β4 6= 1), choosing An = A1(β4−2)
4 = 2 and the algebra
M (0, 2).
. Note that the expression β′
4 = (A1+An)β4
, we obtain β′
A1+Anβ4
A1+Anβ4
β4
• Let α4 6= 0. The following expressions:
α′
4 − β′
4 =
β′
4 − 1 =
A1(A1 + An)(α4 − β4)
1 + A1Anβ4 + A1Anα4 + A2
A2
nα4
,
A2
1(β4 − 1)
A2
1 + A1Anβ4 + A1Anα4 + A2
nα4
β4 6= 1.
give us nullity invariants.
a) α4 − β4 6= 0 and β4 6= 1. Putting An = A1(1−α4)
b) α4 − β4 6= 0 and β4 = 1. Setting An = (α4−2)A1
c) α4 − β4 = 0 and β4 6= 1. Choosing An =
d) α4 − β4 = 0 and β4 = 1. Then we obtain α′
algebra M (2, 2).
α4−β4
4 = β′
α4
, we obtain the algebra M (1, β4), where
, we have α′
A1(−α4+√2α2
4−2α4)
α4
4 = 2 and the algebra M (2, 1).
, we get α′
4 = 2 and the
4 = 1 and the algebra M (1, 1).
Case 2.2. Let β3 = 0 and β4 6= 0. Let us take the change of basis as follows:
e′
1 = A1e1 + Anen, e′
n = (A1 + An)en, with A1(A1 + An) 6= 0.
12
J.K. ADASHEV, L.M. CAMACHO, AND B.A. OMIROV
Similar as in Case 1.2 we get e′
1, e′
On the other hand, we have [e′
n+1 = [e′
n] = (A1 + An)(A1α3 + Anα4)en+1 + (A1 + An)Anβ4en+2.
1 + A1Anα4 − A1Anα3)en+1 + A1Anβ4en+2.
1]− e′
2 = (A2
n, e′
Consider the determinant:
A2
1 + A1Anα4 − A1Anα3
(A1 + An)(A1α3 + Anα4)
A1Anβ4
(A1 + An)Anβ4
(cid:12)(cid:12)(cid:12)(cid:12)(cid:12)
(cid:12)(cid:12)(cid:12)(cid:12)(cid:12)
= A1An(A1 + An)β4(A1 − Anα3 − A1α3) 6= 0.
Therefore, we conclude that e′
n+1 and [e′
1, e′
n] are lineally independent. Therefore, we are in the
conditions of Case 2.1.
Case 2.3. Let (α1, β1) = (0, 0). Taking the following change of basis:
e′
1 = e1 +en, e′
we obtain the family of algebras:
2 = 2e2 +(α3 +α4)en+1 +(β3 +β4)en+2, e′
i = 2ei, 3 ≤ i ≤ n, e′
n+1 = en+1, e′
n+2 = en+2,
[ei, e1] = ei+1,
[en, e1] = e2 + (α4 − α3)en+1 + (β4 − β3)en+2,
[e1, en] = 2(α3 + α4)en+1 + 2(β3 + β4)en+2,
[en, en] = 4α4en+1 + 4β4en+2.
1 ≤ i ≤ n − 2,
Since (α1, β1) = (0, 0), we can assume that (α3, β3) = (α4, β4). Making the change e′
4β3en+2, we obtain a split algebra.
n+1 = 4α3en+1+
(cid:3)
Now, we study the two-dimensional non-split central extensions of the algebra F 2
n. From Proposition
4.1 and definition of central extension, we conclude that the table of multiplication of two-dimensional
central extension of the algebra F 2
n has the following form:
[e1, e1] = e3,
[ei, e1] = ei+1,
[en, e1] = α2x1 + β2x2,
[e2, e1] = α1x1 + β1x2,
3 ≤ i ≤ n − 1,
[e1, e2] = α3x1 + β3x2, [e2, e2] = α4x1 + β4x2.
It we take the change of basis
2 = e3, e′
e′
1 = e1, e′
i = ei+1, 2 ≤ i ≤ n − 1, e′
n = e2, e′
n+1 = x1, e′
n+2 = x2,
then its multiplication have the following form:
1 ≤ i ≤ n − 2,
[ei, e1] = ei+1,
[en, e1] = α1en+1 + β1en+2,
[en−1, e1] = α2en+1 + β2en+2,
[e1, en] = α3en+1 + β3en+2,
[en, en] = α4en+1 + β4en+2.
Theorem 4.5. An arbitrary two-dimensional non-split central extension of the algebra F 2
phic to one of the following pairwise non-isomorphic algebras:
n is isomor-
• N (1, 0, 0, 0), N (0, 1, β3, 0), α4 ∈ {0, 1}, β3 6= 0, N (1, 0, 0, 1), N (0, 1, 0, 0), N (0, 1, 1, 1),
where
N (α3, α4, β3, β4) :
[ei, e1] = ei+1,
[en+1, e1] = en+2,
[e1, en+1] = α3en + β3en+2,
[en+1, en+1] = α4en + β4en+2,
• R(0, 0, 1, 0), R(0, 0, 1, 1), R(0, 1, 1, β4), β4 ∈ C, R(1, 0, 0, 1)
where
1 ≤ i ≤ n − 1,
R(α3, α4, β3, β4) :
[ei, e1] = ei+1,
[en, e1] = en+1,
[e1, en] = α3en+1 + β3en+2,
[en, en] = α4en+1 + β4en+2,
1 ≤ i ≤ n − 2,
CENTRAL EXTENSIONS OF SOME NILPOTENT LEIBNIZ ALGEBRAS.
13
• L1,λ
n+2, L2,λ
n+2, L2
n+2 and the algebra
N ∗ :
[ei, e1] = ei+1,
[e1, en+1] = en+2,
[en+1, en+1] = en.
1 ≤ i ≤ n − 1,
Proof. The proof is similar to the proof of Theorem 4.4.
(cid:3)
Three-dimensional central extensions. In this subsection we study three-dimensional non-split
central extensions of the algebras F 1
n and F 2
n.
From Proposition 4.1 and definition of central extension, we conclude that the table of multiplication
of three-dimensional central extension of the algebra F 1
n has the form:
[e1, e1] = e3,
[e2, e1] = e3 + α1x1 + β1x2 + γ1x3,
[ei, e1] = ei+1,
[en, e1] = α2x1 + β2x2 + γ2x3,
[e1, e2] = α3x1 + β3x2 + γ3x3,
[e2, e2] = α4x1 + β4x2 + γ4x3.
3 ≤ i ≤ n − 1,
Making the change of basis
e′
1 = e1, e′
i = ei+1, 2 ≤ i ≤ n − 1, e′
n = e2, e′
n+j = xj, 1 ≤ j ≤ 3,
we obtain
[ei, e1] = ei+1,
[en, e1] = e2 + α1en+1 + β1en+2 + γ1en+3,
[en−1, e1] = α2en+1 + β2en+2 + γ2en+3,
[e1, en] = α3en+1 + β3en+2 + γ3en+3,
[en, en] = α4en+1 + β4en+2 + γ4en+3.
1 ≤ i ≤ n − 2,
Theorem 4.6. An arbitrary three-dimensional non-split central extension of the algebra F 1
phic to one of the following pairwise non-isomorphic algebras:
n is isomor-
• P (0, 0, 0), P (1, 0, 0), P (0, 0, 1), P (1, 0, 1), P (0, 0, 2), P (1, 0, 2), P (0, 1, γ4), γ4 6= 1, P (0, 2, 1),
P (0, 2, 2), P (0, 1, 1), where
P (α4, β4, γ4) :
[ei, e1] = ei+1,
[en+1, e1] = e2 + en+2,
[e1, en+1] = en+3,
[en+1, en+1] = α4en + β4en+2 + γ4en+3,
1 ≤ i ≤ n − 1,
•
1 ≤ i ≤ n − 2,
P ∗ :
[ei, e1] = ei+1,
[en, e1] = e2 + en+1,
[e1, en] = en+2,
[en, en] = en+3.
Proof. Consider the cases.
Case 1. (α2, β2, γ2) 6= (0, 0, 0). Without loss of generality, one can assume that α2 6= 0 and [en−1, e1] =
en+1. Taking e′
n = en − α1en−1, we get α1 = 0.
Making the following change of basis:
e′
i = ei, 1 ≤ i ≤ n − 1, e′
n = en+1, e′
n+1 = en, e′
n+2 = en+2, e′
n+3 = en+3,
14
J.K. ADASHEV, L.M. CAMACHO, AND B.A. OMIROV
we transform the table of multiplication to the form:
[ei, e1] = ei+1,
[en+1, e1] = e2 + β1en+2 + γ1en+3,
[e1, en+1] = α3en + β3en+2 + γ3en+3,
[en+1, en+1] = α4en + β4en+2 + γ4en+3.
1 ≤ i ≤ n − 1,
If βi = 0 or γi = 0 for any i, then we obtain a split Leibniz algebra. Therefore, we consider the case
of βi 6= 0 and γj 6= 0 for some i and j.
Case 1.1. (β1, γ1) 6= (0, 0). Without loss of generality, one can assume that β1 6= 0 and [en+1, e1] =
e2 + en+2. Hence, we have the following algebra:
[ei, e1] = ei+1,
[en+1, e1] = e2 + en+2,
[e1, en+1] = α3en + β3en+2 + γ3en+3,
[en+1, en+1] = α4en + β4en+2 + γ4en+3,
1 ≤ i ≤ n − 1,
with (γ3, γ4) 6= (0, 0).
• γ3 6= 0. Then we may assume [e1, en+1] = en+3 (by replacing e′
Thus, we have the family of algebras:
n+3 = α3en + β3en+2 + γ3en+3).
[ei, e1] = ei+1,
[en+1, e1] = e2 + en+2,
[e1, en+1] = en+3,
[en+1, en+1] = α4en + β4en+2 + γ4en+3.
1 ≤ i ≤ n − 1,
Taking the general change of generator basis elements and applying similar arguments as in
Case 1.2 of Theorem 4.4 we obtain expressions for the new parameters:
α′
4 =
β′
4 =
γ′
4 =
with restriction
(A1 + An+1)(A1α4 + (An−1 − Bn−1)β4)
1 + A1An+1β4 + A1An+1γ4 + A2
(A2
n+1β4)
An−2
1
,
(A1 + An+1)2β4
A2
1 + A1An+1β4 + A1An+1γ4 + A2
(A1 + An+1)(A1γ4 + An+1β4)
1 + A1An+1β4 + A1An+1γ4 + A2
A2
n+1β4
n+1β4
,
,
A1(A1 + An+1)(A2
1 + A1An+1β4 + A1An+1γ4 + A2
n+1β4) 6= 0.
Consider all possible cases.
a) β4 = 0. Then β′
4 = 0.
a.1) γ4 = 0. Then α′
4 = (A1+An+1)α4
4 = 0, otherwise, putting An+1 = An−1
An−1
.
1
If α4 = 0, then α′
the algebras P (0, 0, 0), P (1, 0, 0).
1 −A1α4
α4
, we get α′
4 = 1, and
a.2) γ4 6= 0. It is remarkable that the following expression γ′
4 − 1 = A1(γ4−1)
A1+An+1γ4
nullity invariant.
If γ4 = 1, then α′
If α4 = 0, then we have the algebra P (0, 0, 1) and when α4 6= 0, putting A1 =
n−2√α4, we get the algebra P (1, 0, 1).
4 = α4
An−2
4 = 1.
, γ′
1
deduce
If γ4 6= 1, then choosing An+1 = A1(γ4−2)
γ4An−2
In the case of α4 = 0, we have the algebra P (0, 0, 2). In the opposite case (that is
α4 6= 0) choosing A1 = n−2q 2α4
4 = 1 and the algebra P (1, 0, 2).
4 = 2 and α′
, we deduce γ′
, we have α′
4 = 2α4
γ4
γ4
.
1
CENTRAL EXTENSIONS OF SOME NILPOTENT LEIBNIZ ALGEBRAS.
15
b) β4 6= 0. Putting An−1 − Bn−1 = − A1α4
β4
we deduce α′
4 = 0. The equalities
4 =
β′
4 − γ′
γ′
4 − 1 =
A1(A1 + An+1)(β4 − γ4)
1 + A1An+1β4 + A1An+1γ4 + A2
A2
n+1β4
,
A2
1(γ4 − 1)
A2
1 + A1An+1β4 + A1An+1γ4 + A2
n+1β4
,
add two more nullity invariants.
Consider the following subcases:
b.1) β4 − γ4 6= 0 and γ4 6= 1. Choosing An+1 = A1(1−β4)
and the algebra P (0, 1, γ4), where γ4 6= 1.
4 +β4−β2
β4γ4−γ 2
β4(1−γ4)
β4−γ4
b.2) β4 − γ4 6= 0 and γ4 = 1. Setting An+1 = (β4−2)A1
β4
4
algebra P (0, 2, 1).
b.3) β4 − γ4 = 0 and γ4 6= 1. Putting An+1 =
b.4) β4 − γ4 = 0 and γ4 = 1. Then we get β′
the algebra P (0, 2, 2).
, we get β′
4 = 1, γ′
4 =
we derive α′
4 = 2, γ′
4 = 1 and the
A1(−β4+√2β2
4 −2β4)
β4
, we obtain β′
4 = 2 and
4 = γ′
4 = 1 and the algebra P (0, 1, 1).
• γ3 = 0. It implies γ4 6= 0. Thus, we have the following family of algebras:
[ei, e1] = ei+1,
[en+1, e1] = e2 + en+2,
[e1, en+1] = α3en + β3en+2,
[en+1, en+1] = α4en + β4en+2 + γ4en+3.
1 ≤ i ≤ n − 1,
Let us take the general change of generator basis elements:
e′
1 = A1e1 + An+1en+1, e′
n+1 = (A1 + An+1)en+1, with A1(A1 + An+1) 6= 0.
Let us take the general change of generator basis elements:
e′
1 = A1e1 + An+1en+1, e′
n+1 = (A1 + An+1)en+1.
Similar to above we derive:
e′
2 = [e′
1, e′
1] = A1(A1 + An+1)e2 + (A1An+1α3 + A2
n+1α4)en+
(A1An+1 + A1An+1β3 + A2
n+1β4)en+2 + A2
n+1γ4en+3
and
Since e′
n+2 = [e′
n+1, e′
1
t = At−1
e′
1] − e′
2, then
(A1 + An+1)et, 3 ≤ t ≤ n.
e′
n+2 = (A1An+1α4 − A1An+1α3)en + (A2
1 + A1An+1β4 − A1An+1β3)en+2 + A1An+1γ4en+3.
We have
[e′
1, e′
n+1] = (A1+An+1)(A1α3+An+1α4)en+(A1+An+1)(A1β3+An+1β4)en+2+(A1+An+1)An+1γ4en+3.
= A1An+1(A1+An+1)γ4(A1−An+1β3−A1β3) 6= 0.
Consider
A2
1 + A1An+1β4 − A1An+1β3
(A1 + An+1)(A1β3 + An+1β4)
A1An+1γ4
(A1 + An+1)An+1γ4
(cid:12)(cid:12)(cid:12)(cid:12)(cid:12)
(cid:12)(cid:12)(cid:12)(cid:12)(cid:12)
Therefore, we conclude that e′
n+2 and [e′
1, e′
n+1] are lineally independent, which satisfy the
conditions of the case γ3 6= 0.
Case 1.2. (β1, γ1) = (0, 0). Note that β3γ4 − β4γ3 6= 0, because otherwise the algebra is split.
Let us make the following change of basis:
e′
1 = e1 + (α4 − α3)en−1 + en+1, e′
2 = 2e2 + 2α4en + (β3 + β4)en+2 + (γ3 + γ4)en+3,
e′
i = 2ei, 3 ≤ i ≤ n + 1, e′
n+2 = en+2, e′
n+3 = en+3.
16
J.K. ADASHEV, L.M. CAMACHO, AND B.A. OMIROV
Then we obtain the following family of algebras:
[ei, e1] = ei+1,
[en+1, e1] = e2 + (β4 − β3)en+2 + (γ4 − γ3)en+3,
[e1, en+1] = (α3 + α4)en + 2(β3 + β4)en+2 + 2(γ3 + γ4)en+3,
[en+1, en+1] = 2α4en + 4β4en+2 + 4γ4en+3,
1 ≤ i ≤ n − 1,
The assumption (β1, γ1) = (0, 0) implies β3 = β4 6= 0 and γ3 = γ4 6= 0. But this is a contradiction
to restriction β3γ4 − β4γ3 6= 0.
Case 2. (α2, β2, γ2) = (0, 0, 0) and det
6= 0. Without loss of generality we can
assume that α1β3γ4 6= 0, from which we have the algebra P ∗.
α1 β1
α3 β3
α4 β4
γ1
γ3
γ4
(cid:3)
Now, we present three-dimensional central extensions of the algebra F 2
n.
Theorem 4.7. Three-dimensional non-split central extension of the algebra F 2
of the following pairwise non-isomorphic algebras:
n is isomorphic to one
• Q(0, 0, 0, 1, 0), Q(1, 0, 0, 1, 0), Q(0, 0, 0, 1, 1), Q(1, 0, 0, 1, 1), Q(0, 0, 1, 1, γ4),
Q(0, 1, 0, 0, 1),
γ4 ∈ C;
where
Q(α4, β3, β4, γ3, γ4) :
•
[ei, e1] = ei+1,
[en+1, e1] = en+2,
[e1, en+1] = β3en+2 + γ3en+3,
[en+1, en+1] = α4en + β4en+2 + γ4en+3,
1 ≤ i ≤ n − 1,
Q∗ :
1 ≤ i ≤ n − 2,
[ei, e1] = ei+1,
[en, e1] = en+1,
[e1, en] = en+2,
[en, en] = en+3.
Proof. The proof is carried out similar to the proof of Theorem 4.6.
(cid:3)
Four-dimensional central extensions. This subsection is devoted to the classification of four-
dimensional non-split central extensions of the algebras F 1
n and F 2
n.
Analogously, the table of multiplication of three-dimensional central extension of the algebra F 1
n
has the form:
[e1, e1] = e3,
[ei, e1] = ei+1,
[e2, e1] = e3 + α1x1 + β1x2 + γ1x3 + δ1x4,
[en, e1] = α2x1 + β2x2 + γ2x3 + δ2x4,
[e1, e2] = α3x1 + β3x2 + γ3x3 + δ3x4,
[e2, e2] = α4x1 + β4x2 + γ4x3 + δ4x4.
3 ≤ i ≤ n − 1,
Making the change of basis
1 = e1, e′
e′
i = ei+1, 2 ≤ i ≤ n − 1, e′
n = e2, e′
n+1 = x1, e′
n+2 = x2, e′
n+3 = x3, e′
n+4 = x4,
we deduce
[ei, e1] = ei+1,
[en, e1] = e2 + α1en+1 + β1en+2 + γ1en+3 + δ1en+4,
[en−1, e1] = α2en+1 + β2en+2 + γ2en+3 + δ2en+4,
[e1, en] = α3en+1 + β3en+2 + γ3en+3 + δ3en+4,
[en, en] = α4en+1 + β4en+2 + γ4en+3 + δ4en+4.
1 ≤ i ≤ n − 2,
CENTRAL EXTENSIONS OF SOME NILPOTENT LEIBNIZ ALGEBRAS.
17
Theorem 4.8. Four-dimensional non-split central extension of the algebra F 1
following algebra:
n is isomorphic to the
1 ≤ i ≤ n − 2,
[ei, e1] = ei+1,
[en−1, e1] = en+2,
[en, e1] = e2 + en+1,
[e1, en] = en+3,
[en, en] = en+4.
Proof. Note that det
α1 β1
α2 β2
α3 β3
α4 β4
γ1
γ2
γ3
γ4
δ1
δ2
δ3
δ4
6= 0. Otherwise we get a split Leibniz algebra.
Without loss of generality, one can assume α1β2γ3δ4 6= 0. Thus, we obtain the algebra of the
(cid:3)
theorem.
Similar to the case of F 1
n, we obtain the following result:
Theorem 4.9. Four-dimensional non-split central extension of the algebra F 2
following algebra:
n is isomorphic to the
1 ≤ i ≤ n − 2,
[ei, e1] = ei+1,
[en−1, e1] = en+2,
[en, e1] = en+1,
[e1, en] = en+3,
[en, en] = en+4.
Finally, we present the description of k-dimensional (k ≥ 5) central extensions of naturally graded
filiform non-Lie Leibniz algebras. In fact, all of them are split.
Theorem 4.10. A k-dimensional (k ≥ 5) central extension of the algebras F 1
algebra.
n and F 2
n is a split
Acknowledgments
The authors were supported by Ministerio de Econom´ıa y Competitividad (Spain), grant MTM2013-
43687-P (European FEDER support included) and by V Plan Propio de Investigaci´on of Sevilla Uni-
versity (VPPI-US). The last named author was also partially supported by the Grant No. 0828/GF4
of Ministry of Education and Science of the Republic of Kazakhstan.
References
[1] Ayupov Sh.A., Omirov B.A. On some classes of nilpotent Leibniz algebras, (Russian) Sibirsk. Mat. Zh., vol. 42 (1),
2001, p. 18 -- 29; translation in Siberian Math. J., vol. 42 (1), 2001, p. 15 -- 24.
[2] Bauerle, G.G.A.; de Kerf, E.A.; ten Kroode, A.P.E. Lie algebras. Part 2. Finite and infinite dimensional Lie algebras
and applications in physics. Edited and with a preface by E. M. de Jager. Studies in Mathematical Physics, 7. North-
Holland Publishing Co., Amsterdam, 1997. x+554 pp. ISBN: 0-444-82836-2
[3] Cabezas J.M., Pastor E. Naturally graded p-filiform Lie algebras in arbitrary finite dimension, Journal of Lie Theory,
vol. 15, 2005, p. 379 -- 391.
[4] Cahen M., Ohn C. Bialgebra structures on the Heisenberg algebra Bulletin de la Classe des Sciencies de L'Academia
Royale de Belgique, vol. 75, 1989, p. 315-321.
[5] Camacho L.M., G´omez J.R., Gonz´alez A.J., Omirov B. A. Naturally graded 2-filiform Leibniz algebras, Comm.
Algebra, vol. 38(10), 2010, p. 3671 -- 3685.
[6] Camacho L.M., G´omez J.R., Gonz´alez A.J., Omirov B. A. The classification of naturally graded p-filiform Leibniz
algebras, Com. Algebra, vol. 39(1), 2011, p. 153 -- 168.
[7] Camacho L.M., G´omez J.R., Gonz´alez A.J., Omirov B. A. Naturally graded quasi-filiform Leibniz algebras, J. Symb.
Comput., vol. 44(5), 2009, p. 527 -- 539.
[8] G´omez, J.R., Omirov, B.A. On classification of complex filiform Leibniz algebras, Alg. Colloq., vol. 22(1), 2015, p.
757 -- 774.
[9] Loday J.L. Une version non commutative des alg`ebres de Lie: les alg`ebres de Leibniz. Ens. Math. vol. 39, 1993, p.
269 -- 293.
[10] Rakhimov I.S., Langari S.L., Langari M.B. On central extensions of null-filiform Leibniz algebras, Inter. J. Algebra,
vol. 3(6), 2009, p. 271 -- 280.
18
J.K. ADASHEV, L.M. CAMACHO, AND B.A. OMIROV
[11] Rakhimov I.S., Hassan M.A. On one-dimensional Leibniz central extensions of a filiform Lie algebra, Bull. Aust.
Math. Soc., vol. 84, 2011, 205 -- 224.
[12] Skjelbred T., Sund T. On the classification of nilpotent Lie algebras, Technical report Mathematisk Institutt,
Universitetet i Oslo, (1977).
[13] Szymczak I, Zakrzewski S. Quantum deformation of the Heisenberg group obtained by geormetric quantization,
Journal of Geometry and Physics, vol. 7, 1990, p. 553-569.
[14] Vergne M. Cohomologie des alg`ebres de Lie nilpotentes. Application `a l'´etude de la variet´e des alg`ebres de Lie
nilpotentes, Bull. Soc. Math. France, vol. 98, 1970, p. 81 -- 116.
[J.K. Adashev] Institute of Mathematics, National University of Uzbekistan, Tashkent, 100125, Uzbek-
istan
E-mail address: [email protected]
[L.M. Camacho] Dpto. Matem´atica Aplicada I. Universidad de Sevilla. Avda. Reina Mercedes, s/n. 41012
Sevilla. (Spain)
E-mail address: [email protected]
[B.A. Omirov] Institute of Mathematics, National University of Uzbekistan, Tashkent, 100125, Uzbekistan
E-mail address: [email protected]
|
1806.10204 | 2 | 1806 | 2018-10-08T00:32:23 | Special Identities for Comtrans Algebras | [
"math.RA",
"math.RT"
] | Comtrans algebras, arising in web geometry, have two trilinear operations, commutator and translator. We determine a Gr\"obner basis for the comtrans operad, and state a conjecture on its dimension formula. We study multilinear polynomial identities for the special commutator $[x,y,z] = xyz-yxz$ and special translator $\langle x, y, z \rangle = xyz-yzx$ in associative triple systems. In degree 3, the defining identities for comtrans algebras generate all identities. In degree 5, we simplify known identities for each operation and determine new identities relating the operations. In degree 7, we use representation theory of the symmetric group to show that each operation satisfies identities which do not follow from those of lower degree but there are no new identities relating the operations. We use noncommutative Gr\"obner bases to construct the universal associative envelope for the special comtrans algebra of $2 \times 2$ matrices. | math.RA | math |
SPECIAL IDENTITIES FOR COMTRANS ALGEBRAS
MURRAY R. BREMNER AND HADER A. ELGENDY
Abstract. Comtrans algebras, arising in web geometry, have two trilinear
operations, commutator and translator. We determine a Grobner basis for
the comtrans operad, and state a conjecture on its dimension formula. We
study multilinear polynomial identities for the special commutator [x, y, z] =
xyz − yxz and special translator hx, y, zi = xyz − yzx in associative triple
In degree 3, the defining identities for comtrans algebras generate
systems.
In degree 5, we simplify known identities for each operation
all identities.
and determine new identities relating the operations.
In degree 7, we use
representation theory of the symmetric group to show that each operation
satisfies identities which do not follow from those of lower degree but there are
no new identities relating the operations. We use noncommutative Grobner
bases to construct the universal associative envelope for the special comtrans
algebra of 2 × 2 matrices.
1. Introduction
Comtrans algebras were introduced by Smith [33, §3] to answer a problem in
web geometry [2, 12] posed by Goldberg [11, Problem X.3.9]: to find the algebraic
structure on the tangent bundle of the coordinate ternary loop of a 4-web [19, §3.7].
Comtrans algebras are a common ternary generalization of Lie algebras, Malcev al-
gebras [24] and Akivis algebras [1]: every such algebra can be given the structure
of a comtrans algebra [30, §5]. A generalization of Lie's Third Fundamental The-
orem connects formal ternary loops with comtrans algebras [33, §5]. For physical
applications of comtrans algebras, see [34, p. 321]: ". . . the Lorentz metric on 4-
dimensional real space-time provides a simple comtrans algebra that extends the
3-dimensional vector triple product comtrans algebra."
Definition 1.1. Smith [33, §3]. A comtrans algebra is a vector space A with two
trilinear operations A × A × A → A, the commutator [x, y, z] and the translator
hx, y, zi, satisfying the following polynomial identities for all x, y, z ∈ A:
(1)
(2)
(3)
[x, y, z] + [y, x, z] = 0,
hx, y, zi + hy, z, xi + hz, x, yi = 0,
[x, y, z] + [z, y, x] = hx, y, zi + hz, y, xi.
2010 Mathematics Subject Classification. Primary 17A40. Secondary 15-04, 15A21, 15A69,
15B36, 18D50, 20C30, 68W30.
Key words and phrases. Comtrans algebras, trilinear operations, polynomial identities, lattice
basis reduction, representation theory of the symmetric group, algebraic operads, Grobner bases,
universal associative enveloping algebras.
The research of the first author was supported by the Discovery Grant Algebraic Operads
from NSERC, the Natural Sciences and Engineering Research Council of Canada. He thanks the
Department of Mathematics at Damietta University in Egypt for its hospitality during May 2018.
1
2
MURRAY R. BREMNER AND HADER A. ELGENDY
The commutator alternates in the first two arguments (1), the translator satisfies
the Jacobi identity (2), and the operations are related by the comtrans identity (3).
Example 1.2. If T is a Lie triple system with bracket [x, y, z] then letting both
commutator and translator equal [x, y, z] gives T the structure of a comtrans algebra
T CT . If T is obtained from a Lie algebra L by [x, y, z] = [[x, y], z] then Shen &
Smith [32, Theorem 3.2] have shown that L is simple if and only if T CT is simple.
Example 1.3. Let Am,n denote the vector space of m × n matrices over F. Fix
matrices p (n × n) and q (m × m) over F. Define a trilinear operation on Am,n by
(x, y, z) = xpytqz. Setting [x, y, z] = (x, y, z) − (y, x, z) and hx, y, zi = (x, y, z) −
(y, z, x) gives Am,n the structure of a comtrans algebra [32, Example 2.1].
Definition 1.4. An associative triple system [25] is a vector space A with a trilinear
operation xyz satisfying (vw(xyz)) = (v(wxy)z) = (vw(xyz)). The special commu-
tator and special translator in A are [x, y, z] = xyz − yxz and hx, y, zi = xyz − yzx.
Remark 1.5. In the terminology of [8], two trilinear operations in QS3 are equiv-
alent if they generate the same left ideal. The special commutator is equivalent
to the q = 2 case of the q-deformed anti-Jordan triple product. The special trans-
lator is the same as the cyclic commutator, and is equivalent to the operation
2xyz − yzx − zxy which represents to the identity matrix I2 in the simple 2-sided
ideal corresponding to partition 2+1.
We restrict consideration to multilinear polynomial identities, since this allows
us to apply the representation theory of the symmetric group. This approach to
polynomial identities was introduced by Malcev [27] and Specht [35]; its computer
implementation was pioneered by Hentzel [20, 21]. This point of view also fits very
naturally into the theory of algebraic operads [6, 26, 28]. For an introduction to
operads which emphasizes connections to homotopical algebra, see Vallette [36].
In §2 we recall basic definitions on triple systems. In §3 we show how algebraic
operads may be used to discuss polynomial identities. In particular, we calculate a
Grobner basis for the comtrans operad, and explain how Grobner bases of operads
may be used to compute polynomial identities. In §4 we show that every identity
in degree 3 for the commutator and translator follows from (1) -- (3). In §5 we use
computer algebra to find explicit generators for the S5-module of identities which
do not follow from those of degree 3. We consider three cases: the commutator
by itself, the translator by itself, and identities in which each term contains both
operations.
In §6 we use a constructive version of the representation theory of
the symmetric group to demonstrate that there are new identities in degree 7 for
each operation separately but no new identities relating the operations.
In §7
we construct the universal associative enveloping algebra of the special comtrans
algebra M CT obtained from the associative triple system M of 2 × 2 matrices.
Our results are valid over any field of characteristic 0. Our computations were
performed using Maple worksheets written by the authors.
2. Preliminaries
Lemma 2.1. In an associative triple system A, the special commutator and special
translator satisfy the relations (1) -- (3).
Proof. Trivial calculation. This result appears in a more general setting and without
any assumption of associativity in [33, (3.4)-(3.6)].
(cid:3)
SPECIAL IDENTITIES FOR COMTRANS ALGEBRAS
3
Definition 2.2. Let M be an associative triple system and let M CT be the com-
trans algebra defined on the underlying vector space by the special commutator and
special translator. We say that a comtrans algebra A is special if there exists an
associative triple system M and an injective morphism A → M CT of comtrans alge-
bras; otherwise, A is exceptional. (This terminology is motivated by the definitions
of special and exceptional Jordan algebras [29].)
Remark 2.3. Tercom (ternary commutator) algebras were introduced by Ross-
manith & Smith [30]. They have trilinear operations λ(x, y, z) and ρ(x, y, z) which
satisfy the following polynomial identities:
λ(x, y, z) + λ(y, x, z) = 0,
ρ(x, y, z) + ρ(x, z, y) = 0,
λ(x, y, z) + λ(y, z, x) + λ(z, x, y) = ρ(x, y, z) + ρ(y, z, x) + ρ(z, x, y).
These identities hold for the left and right commutators λ(x, y, z) = xyz − yxz
and ρ(x, y, z) = xyz − xzy in every associative triple system. Tercom algebras are
equivalent to comtrans algebras in the sense that
[x, y, z] = λ(x, y, z),
hx, y, zi = λ(x, y, z) + ρ(y, x, z),
ρ(x, y, z) = hy, x, zi − [y, x, z].
Remark 2.4. The weakly anticommutative operation [8], which is equivalent to
the operation {x, y, z} = xyz + xzy − 2zyx, satisfies the symmetric sum identity in
every associative triple system:
{xσ, yσ, zσ} = 0.
Xσ∈S3
This allows us to define special comtrans algebras in terms of a single operation:
[x, y, z] = 1
hx, y, zi = 1
{x, y, z} = [y, z, x] + hx, y, zi + hx, z, yi.
2 ({z, x, y} − {z, y, x}),
6 (4{x, z, y} + 2{y, z, x} + {z, x, y} − {z, y, x}),
Remark 2.5. Every special comtrans algebra becomes a special anti-Jordan triple
system using the trilinear operation (x, y, z) = [x, y, z] − hz, y, xi = xyz − zyx.
3. Algebraic operads
We consider operads in the symmetric monoidal category of vector spaces over a
field of characteristic 0; equivalently, Z-graded vector spaces concentrated in degree
0. We say degree instead of arity since our motivation comes from nonassociative
algebra and we never refer to the homological degree (that is, all operations have
homological degree 0).
3.1. Basic definitions. A monomial of weight w (and hence degree d(w) = 2w+1)
in two trilinear operations γ and δ is a sequence of d(w) distinct arguments (usually
identified with a permutation of the variables x1, . . . , xd(w)) into which w operation
symbols have been inserted (each symbol may be either γ or δ). There is a bijection
between monomials of weight w and complete ternary trees with w internal nodes
4
MURRAY R. BREMNER AND HADER A. ELGENDY
each labelled by one of the operation symbols, and d(w) leaves labelled by distinct
arguments. For example, writing ◦ for γ and • for δ,
γ(x3, δ(x1, x4, x5), γ(x9, δ(x6, x8, x7), x2)) ←→
x3
◦
❘❘❘❘❘❘❘❘❘
❧❧❧❧❧❧❧❧
•
◦
✿✿✿✿
✿✿✿✿
☎☎☎☎
☎☎☎☎
x1 x4 x5
x9 • x2
✿✿✿✿
☎☎☎☎
x6 x8 x7
Definition 3.1. Let T be the free weight-graded symmetric operad generated
by ternary operations γ, δ with no symmetry, and let T (w) be the homogeneous
subspace of weight w. Let A be the weight-graded symmetric operad generated by
a ternary operation α with no symmetry satisfying ternary associativity (Definition
1.4), and let A(w) be the homogeneous subspace of weight w. Using the formula
for ternary Catalan numbers, we have
dim T (w) =
2w
2w+1(cid:18)3w
w (cid:19)(2w+1)! ,
dim A(w) = (2w+1)! .
Since we use the weight grading, T (w) and A(w) are modules over S2w+1.
Definition 3.2. Since T is free, a morphism with domain T is determined by its
values on γ and δ. We define the expansion map X: T → A by
X(γ) = α − (12) · α, X(δ) = α − (132) · α,
p · x1x2x3 = xp−1(1)xp−1(2)xp−1(3).
Permutations act on positions of arguments (not subscripts). Thus
X(γ)(x, y, z) = α(x, y, z) − α(y, x, z),
X(δ)(x, y, z) = α(x, y, z) − α(y, z, x).
We use the more convenient notation
γ(x, y, z) = [x, y, z],
δ(x, y, z) = hx, y, zi,
α(x, y, z) = xyz.
Thus X may be written in terms of the special ternary commutator and translator:
[x, y, z]
X−−−−−→ xyz − yxz,
hx, y, zi
X−−−−−→ xyz − yzx.
We write Xw for the restriction of X to T (w).
Lemma 3.3. We have Xw(T (w)) ⊆ A(w) for w ≥ 0. The kernel of Xw is the
S2w+1-submodule of T (w) of multilinear polynomial identities of degree 2w+1 sat-
isfied by the special commutator and translator in every associative triple system.
3.2. A Grobner basis for the comtrans operad. The notion of Grobner basis
is not well-defined in general for ideals in symmetric operads; see for example [6,
Proposition 5.2.2.5].
In order to apply the algorithm for computing a Grobner
basis to a symmetric operad P, we must first apply the forgetful functor from
symmetric operads to shuffle operads; equivalently, we must ignore the symmetric
group actions1. We then compute a Grobner basis of the corresponding ideal in the
shuffle operad Psh; see for example [6, Example 5.3.4.3]. Shuffle operads and their
Grobner bases were introduced by Dotsenko & Khoroshkin [14]; for a more detailed
exposition, see [6, Chapter 5]).
1We thank Vladimir Dotsenko for emphasizing the importance of the forgetful functor from
symmetric operads to shuffle operads in this subsection and the next.
SPECIAL IDENTITIES FOR COMTRANS ALGEBRAS
5
In the free weight-graded symmetric operad T of Definition 3.1, the homogeneous
subspace T (1) has the following ordered basis of ternary operations:
(4)
hx, y, zi,
[x, y, z],
hx, z, yi,
[x, z, y],
hy, x, zi,
[y, x, z],
hy, z, xi,
[y, z, x],
hz, x, yi,
[z, x, y],
hz, y, xi,
[z, y, x].
As a symmetric operad, T (1) has the natural S3-module structure where S3 acts by
permuting the variables x, y, z. The free symmetric operad T is generated by this
12-dimensional S3-module in degree 3 (weight 1). Note also that T (1) is generated
as an S3-module by the two operations hx, y, zi and [x, y, z].
We now apply the forgetful functor to T to obtain the free shuffle operad Tsh.
The homogeneous subspace Tsh(1) has the same basis (4), but we have forgotten
the action of S3: that is, Tsh(1) is not an S3-module, it is simply a vector space.
Hence the free shuffle operad Tsh is generated by the same 12-dimensional space,
but now we must regard each of the basis monomials (4) as a distinct operation.
For the most general statement of this isomorphism, see [6, Corollary 5.3.3.3].
Let I ⊂ T be the (weight-graded) ideal in the free symmetric operad T such
that the quotient operad CT = T /I is the symmetric comtrans operad. The ideal I
is generated by its homogeneous component I(1), which in turn is generated as an
S3-module by relations (1) -- (3) defining comtrans algebras. In order to determine
the homogeneous component Ish(1) of the ideal Ish in the corresponding shuffle
operad Tsh, we must first find a linear basis of I(1).
·
·
·
·
·
·
·
·
·
·
·
·
·
·
·
·
·
·
·
·
·
·
·
·
·
·
·
·
·
·
·
·
·
·
·
·
1 · 1 ·
· 1 ·
1 · 1 ·
·
·
· 1 ·
·
·
· 1 ·
·
·
· 1 · 1
· 1 ·
· 1 · 1
·
·
·
·
·
·
·
·
·
·
·
·
·
·
·
·
·
·
·
·
·
·
·
·
·
·
·
·
·
·
·
·
·
·
·
·
·
·
· 1 1
·
1
· 1 1
· 1 1
1
1
· 1 1
·
·
·
·
·
· 1 1
· 1 1
·
· 1
· 1
·
· 1
· −1
·
·
·
·
−1
·
· −1
·
·
· −1
· −1
·
· −1
· −1
·
· −1
·
· −1
·
−1
· −1
·
·
·
·
· −1
· 1
1 ·
· 1 · 1 ·
·
·
·
· 1 · 1 ·
·
· 1 · 1 ·
·
· 1 · 1 ·
· 1
·
·
·
1 ·
RCF
−−−−→
1 ·
·
· 1 ·
·
· 1 ·
·
· 1
· −1
· −1
· 1 · 1
· 1 1
·
·
·
·
·
·
·
·
·
·
·
·
·
·
·
·
·
·
·
·
·
·
·
· 1 −1
· 1 1 −1
· 1
· −1 −1 1
·
·
·
·
· −1 1
·
1 ·
· 1 1
· 1 1
·
·
·
· −1
·
· 1
Figure 1. Computation of a Grobner basis for the comtrans operad
We form the 18 × 12 matrix (Figure 1, left) whose three 6 × 12 blocks contain
the coefficient vectors of all permutations of relations (1) -- (3) with respect to the
ordered basis (4). The row space of this relation matrix may be identified with the
S3-module I(1). The homogeneous subspace Ish(1) of the shuffle ideal Ish is the
same row space, regarded simply as a vector space, forgetting the action of S3.
Relations (1) -- (3) are linear (weight 1) relations between ternary operations. For
linear relations, the computation of a Grobner basis for the shuffle ideal Ish reduces
to applying Gaussian elimination to the relation matrix. We compute the RCF
(row canonical form) of the relation matrix (Figure 1, right). The nonzero rows
of the RCF are (the coefficient vectors of) the Grobner basis for Ish. These rows
6
MURRAY R. BREMNER AND HADER A. ELGENDY
represent the following relations among the 12 operations (4) which generate the
free shuffle operad Tsh, and this is the required Grobner basis:
hx, y, zi + hz, y, xi + [y, x, z] − [z, y, x],
hx, z, yi − hz, x, yi + hz, y, xi + [y, x, z] + [z, x, y],
hy, x, zi + hz, x, yi − [y, x, z] − [z, x, y],
hy, z, xi + hz, x, yi − hz, y, xi − [y, x, z] + [z, y, x],
[x, y, z] + [y, x, z],
[x, z, y] + [z, x, y],
[y, z, x] + [z, y, x].
Equivalently, in terms of rewrite rules, we have:
(5)
hx, y, zi −→ −hz, y, xi − [y, x, z] + [z, y, x],
hx, z, yi −→ hz, x, yi − hz, y, xi − [y, x, z] − [z, x, y],
hy, x, zi −→ −hz, x, yi + [y, x, z] + [z, x, y],
hy, z, xi −→ −hz, x, yi + hz, y, xi + [y, x, z] − [z, y, x],
[x, y, z] −→ −[y, x, z],
[x, z, y] −→ −[z, x, y],
[y, z, x] −→ −[z, y, x].
If x, y, z are multilinear monomials of arbitrary degree, and x ≺ y ≺ z in the deglex
extension of the operation order, then the rewrite rules show how to move greater
factors to the left, to the extent allowed by the comtrans relations.
Conjecture 3.4. If CT is the symmetric weight-graded comtrans operad then
dim CT (w) =
(3w)!
w! (3! )w · 5w
(w ≥ 0).
Using terminology from nonassociative algebra, this is the dimension of the multi-
linear subspace of degree 2w+1 in the free comtrans algebra on 2w+1 generators.
We implemented the rewrite rules in Maple to verify the first four terms 1, 5,
250, 35000 by computing all multilinear normal forms in degrees 1, 3, 5, 7. The
factor 5w comes from the number of basis operations (4) which are irreducible
with respect to the rewrite rules (5). The other factor is OEIS sequence A025035
which counts (i) set partitions of {1, 2, . . . , 3w} into parts of size 3, (ii) rooted
phylogenetic (non-planar) complete ternary trees with w internal vertices, and (iii)
distinct multilinear monomials in degree 2w+1 for a symmetric ternary operation:
(xσ, yσ, zσ) = (x, y, z) for σ ∈ S3.
3.3. Polynomial identities and operadic Grobner bases. In principle, we
can use Grobner bases to determine polynomial identities satisfied by the spe-
cial commutator and translator in every associative triple system. We describe
the method briefly to illustrate the connection between polynomial identities and
Grobner bases of operads. (We will not use this method; instead, we apply compu-
tational techniques based on the LLL algorithm for lattice basis reduction [9] and
the representation theory of the symmetric group [7].)
We express the expansion map (Definition 3.2) in terms of a single operad. Let
P be the free weight-graded symmetric operad generated by the ternary operations
α, δ, γ with no symmetries. Let O = P/I where I is generated by these relations:
(6)
γ(x, y, z) + γ(y, x, z),
SPECIAL IDENTITIES FOR COMTRANS ALGEBRAS
7
(7)
(8)
(9)
(10)
(11)
(12)
δ(x, y, z) + δ(y, z, x) + δ(z, x, y),
γ(x, y, z) + γ(z, y, x) − δ(x, y, z) − δ(z, y, x),
α(x, y, z) − α(y, x, z) − γ(x, y, z),
α(x, y, z) − α(y, z, x) − δ(x, y, z),
α(v, α(w, x, y), z) − α(α(v, w, x), y, z),
α(v, w, α(x, y, z)) − α(α(v, w, x), y, z).
Relations (6)-(8) are the defining identities for comtrans algebras; (9)-(10) express
the comtrans operations γ, δ in terms of the associative operation α; and (11)-
(12) are the defining identities for associative triple systems. Relations (6)-(10)
are linear (degree 3) and relations (11)-(12) are quadratic (degree 5). Thus I is
generated by its homogeneous components I(1) and I(2). A linear basis for each
of these components is obtained by applying all permutations of the variables and
then applying Gaussian elimination to the resulting relation matrices.
We apply the forgetful functor from symmetric operads to shuffle operads as
described in §3.2 and obtain the ideal Ish ⊂ Psh. We must use an operation order
such as α ≻ γ ≻ δ to ensure that tree monomials containing α are greater than
tree monomials not containing α. This operation order extends to the revdeglex
(reverse degree lexicographical) order on the tree monomials forming a linear basis
of Psh. We apply the algorithm for computing Grobner bases of shuffle ideals to the
relations forming the linear bases of Ish(1) and Ish(2). This computation produces
relations of the form f + g where f consists of the terms containing α (and possibly
also γ, δ) and g consists of the terms not containing α.
If the Grobner basis
algorithm produces a polynomial f + g with f = 0 then Ish contains a polynomial
g involving only γ, δ which is identically 0 in the quotient Osh = Psh/Ish. Hence
g is a relation between γ, δ which belongs to the kernel of the expansion map, and
so g is a polynomial identity relating γ, δ.
4. Identities of degree 3
Lemma 4.1. Every multilinear identity satisfied by the special commutator and
special translator in every associative triple system follows from (1) -- (3).
Proof. We follow [9]. For the expansion map X1: T (1) → A(1) in degree 3, the
permutations xyz, . . . , zyx in lex order form a basis of A(1), and the monomials
[x, y, z], . . . , [z, y, x], hx, y, zi, . . . , hz, y, xi form an ordered basis of T (1). The
matrix representing X1 has rank 5 and nullity 7:
[X1] =
·
·
·
· −1
1
·
·
1
· −1
1 −1
·
·
·
·
· −1
·
·
1
· −1
1
·
·
·
·
·
1
−1
·
·
·
1
·
·
· −1 −1
1
·
·
·
1
· −1
·
1
·
· −1
·
·
·
1
· −1
· −1
·
·
·
·
·
·
·
1
·
·
·
We compute the Hermite normal form H of the transpose [X1]t and an integer
matrix U such that det(U ) = ±1 and U [X1]t = H. Rows 6-12 of H are zero and
8
MURRAY R. BREMNER AND HADER A. ELGENDY
so rows 6-12 of U form a lattice basis N1 for the integer nullspace of [X1]:
N1 =
·
−1
·
·
· −1
·
· −1
·
·
1
·
1 −1 −1
·
· −1
1
·
·
1
·
· 1
1
·
·
·
·
·
·
·
· −1
·
·
· −1
1
1
1
·
·
1
1 −1
·
1 −1
1
·
·
1
·
·
·
·
·
·
·
1
·
· 1
·
·
·
·
·
·
·
· 1
·
·
·
·
·
·
·
·
1
The rows of N1 have squared Euclidean lengths 2, 2, 2, 4, 4, 7, 7 (sorted) with product
6272. We apply the LLL algorithm to the lattice L spanned by the rows of N1,
obtain a reduced basis of L, sort the vectors by increasing length, and multiply
each row by ±1 to make its leading entry positive:
N2 =
·
·
· 1
1
·
·
· 1
·
· 1
·
· 1
·
·
· 1
·
· 1
·
· 1
·
·
·
·
· 1
·
·
·
·
·
·
·
·
1
1
· −1
1
·
·
·
·
· −1
· −1 1 −1 −1
·
·
·
·
·
· −1
· −1
·
·
·
·
·
· −1
·
·
·
·
·
·
·
·
· 1
·
·
·
The rows of N2 have squared lengths 2, 2, 2, 3, 4, 4, 5 with product 1920. The lattice
L is a module over ZS3, where S3 acts by permuting x, y, z in the ordered basis
of T (1). Rows 1, 4, 7 of N2 are the lex-minimal subset which generates L as a
ZS3-module. These rows represent relations (1) -- (3).
(cid:3)
Remark 4.2. There are 40 pairs of trilinear operations xyz ± xpypzp and xyz ±
xqyqzq in an associative triple system where p, q are distinct non-identity permu-
tations of x, y, z. In every case, a single operation generates the same S3-module
as the original pair. Up to equivalence there are four possibilities: the transla-
tor xyz − yzx, the weakly commutative operation xyz − xzy + 2zyx, the weakly
anticommutative operation xyz + xzy − 2zyx, and the associative operation xyz.
5. Identities of degree 5
We first consider each operation separately and then the two operations together.
Lemma 5.1. Every multilinear identity of degree 5 satisfied by the special commu-
tator in every associative triple system follows from relation (1) and
(13)
Tx,z(v, w, y) + Tx,z(w, y, v) + Tx,z(y, v, w) = 0,
where Tx,z(v, w, y) = [[v, w, x], y, z] − [v, w, [x, y, z]]. (This simplifies the q = 2 case
of the deformed anti-Jordan triple product in [8].)
Proof. It is easy to check by hand that (13) is satisfied by the special commutator.
We must verify that every identity in degree 5 satisfied by the special commutator
follows from (1) and (13). We use computer algebra [9].
Let T be the free weight-graded operad generated by one ternary operation
[−, −, −] with no symmetry. The homogeneous space T (2) is isomorphic to the
direct sum of three copies of the regular representation FS5 corresponding to the
three association types [[−, −, −], −, −], [−, [−, −, −], −], [−, −, [−, −, −]] in that
order. Each association type has 120 permutations of v, w, x, y, z in lex order.
SPECIAL IDENTITIES FOR COMTRANS ALGEBRAS
9
Each of these nonassociative monomials expands using the special commutator
to a linear combination of monomials in the operad A of Definition 3.1: for example,
[[v, w, x], y, z]
7→ vwxyz − wvxyz − yvwxz + ywvxz.
We construct the 120 × 360 matrix M whose (i, j) entry is the coefficient of the i-th
associative monomial in the expansion of the j-th nonassociative monomial.
We compute the Hermite normal form H of the transpose M t and an integer
matrix U with det(U ) = ±1 and U M t = H. We find that H has rank 70 so the
bottom 290 rows of H are 0. Hence the bottom 290 rows of U , denoted N , form an
integer basis for the nullspace of M ; that is, the kernel of the expansion map X5.
The largest squared Euclidean length of the rows of N is 49352.
We apply the LLL algorithm to the rows of N and obtain a matrix N ′ whose
rows generate the same lattice. The largest squared Euclidean length of the rows
of N ′ is only 6. Let N ′′ consist of the rows of N ′ sorted by increasing length.
Let I(x, y, z) = [x, y, z] + [y, x, z] denote relation (1). We generate all conse-
quences of I in degree 5 by partial compositions in the operad; that is, substituting
[−, −, −] into I, or substituting I into [−, −, −]:
(14)
I([x, v, w], y, z),
[I(x, y, z), v, w],
I(x, [y, v, w], z),
[v, I(x, y, z), w],
I(x, y, [z, v, w]),
[v, w, I(x, y, z)].
These relations generate the S5-module of identities which follow from (1).
We construct the 720 × 360 matrix C whose rows contain all permutations of the
relations (14). We compute the RCF and find that C has rank 270. Since N has
rank 290, there is a 20-dimensional quotient S5-module of new identities.
For each row i of N ′′, we stack C on top of the matrix containing all permutations
of the relation represented by row i. Row 194 is the first which increases the rank
from 270 to 290. This row is the coefficient vector of relation (13).
(cid:3)
Remark 5.2. The special commutator is equivalent to xyz + xzy − yxz + yzx −
zxy − zyx, case q = 2 of the deformed anti-Jordan triple product [8]. This follows
from the equality of the RCFs of the representation matrices for the operations:
(cid:20) 0,(cid:20) 1
0
2
0 (cid:21), 1 (cid:21)
Relation (13) is much simpler than the relation with 24 terms in [8, equation (55)].
Lemma 5.3. Every multilinear identity in degree 5 satisfied by the special translator
in every associative triple system follows from relation (2) and
(15)
Ry,z(hv, w, xi) = hRy,z(v), w, xi + hv, Ry,z(w), xi + hv, w, Ry,z(x)i,
where Ry,z(u) = hu, y, zi. (Compare the result for the cyclic commutator in [8].)
Proof. Similar to the proof of Lemma 5.1.
(cid:3)
Theorem 5.4. Every multilinear identity in degree 5 relating the special commuta-
tor and the special translator in every associative triple system follows from (1) -- (3),
(13), (15), and the following new identities involving both operations:
(16)
(17)
(18)
[[vwx]yz] + [[xvy]wz] − [hvwxiyz] + [hvywixz] − [hxywivz] = 0,
[[vwx]yz] − [[vyw]xz] + h[wvz]xyi + [hvywixz] + [vwhzxyi] = 0,
[[vwx]yz] + [[vwz]xy] − [[xwy]vz] − [[zwy]xv] + h[wvz]xyi
+ h[xwz]yvi + h[zwy]xvi + [hwvxiyz] + [hwvzixy] − [hwyvixz]
10
MURRAY R. BREMNER AND HADER A. ELGENDY
− hhwvzixyi + hhwyvixzi + hwxhzyvii = 0.
Proof. Let T be the free symmetric operad generated by two ternary operations
[−, −, −] and h−, −, −i with no symmetry. The homogeneous space T (2) is isomor-
phic to the direct sum of 12 copies of the regular representation FS5 corresponding
to the association types ordered as follows:
[[−, −, −], −, −],
[−, [−, −, −], −],
[−, −, [−, −, −]],
h[−, −, −], −, −i,
h−, [−, −, −], −i,
h−, −, [−, −, −i,
[h−, −, −i, −, −],
[−, h−, −, −i, −],
[−, −, h−, −, −i],
hh−, −, −i, −, −i,
h−, h−, −, −i, −i,
h−, −, h−, −, −ii.
Each association type has 120 permutations of v, w, x, y, z in lex order. We con-
struct the 120 × 1440 matrix M in which the (i, j) entry is the coefficient of the
i-th associative monomial in the expansion of the j-th nonassociative monomial.
There are 36 consequences of identities (1) -- (3) since we can substitute either of two
operations. We must exclude these consequences together with all permutations of
relations (13) and (15). The rest is similar to the proof of Lemma 5.1.
(cid:3)
Definition 5.5. A Smith algebra is a comtrans algebra satisfying the five identities
of Lemmas 5.1, 5.3 and Theorem 5.4. (Since these identities have weights 1 and 2,
the corresponding operad is quadratic and hence has a Koszul dual [18].)
Remark 5.6. The weakly anticommutative operation satisfies a 141-dimensional
S5-module of multilinear identities which are not consequences of the symmetric
sum identity (Remark 2.4). If Tyz(u) = {u, y, z} + {u, z, y} then the simplest new
identity that we found (which however does not generate all new identities) is
Tyz({v, w, x}) = {v, Tyz(w), x} + {v, Tyz(w), x} + {v, w, Tyz(x)}.
6. Identities of degree 7
Every special comtrans algebra is a Smith algebra, but the converse is false:
we demonstrate the existence of identities in degree 7 satisfied by every special
comtrans algebra but not by every Smith algebra. There are new identities for each
operation separately, but none relating the operations.
For degree 7, the methods of the previous sections are impractical:
for one
(resp. two) operation(s), there are 12 (resp. 96) association types and 60480 (resp.
483840) multilinear monomials. We use a constructive version of the representation
theory of the symmetric group to decompose the kernel of the expansion map X7
into isotypic components corresponding to the partitions of 7. We provide only an
outline; these methods have been discussed in detail in previous papers [3, 7, 10].
We order the 12 association types for one ternary operation as follows:
(19)
[[[−, −, −], −, −], −, −],
[[−, −, −], [−, −, −], −],
[−, [−, [−, −, −], −], −],
[−, −, [[−, −, −], −, −]],
[[−, [−, −, −], −], −, −],
[[−, −, −], −, [−, −, −]],
[−, [−, −, [−, −, −]], −],
[−, −, [−, [−, −, −], −]],
[[−, −, [−, −, −]], −, −],
[−, [[−, −, −], −, −], −],
[−, [−, −, −], [−, −, −]],
[−, −, [−, −, [−, −, −]]].
Recall that an identity is new if it does not follow from those of lower degree.
Lemma 6.1. The S7-module of new identities for the special commutator is nonzero,
and has the following decomposition into isotypic components:
[43] ⊕ [421]2 ⊕ [413] ⊕ [321]3 ⊕ [322]2 ⊕ [3212]5 ⊕ [314]2 ⊕ [231]2 ⊕ [2213]3 ⊕ [215].
We write [λ]m if the irreducible representation for partition λ has multiplicity m.
SPECIAL IDENTITIES FOR COMTRANS ALGEBRAS
11
Proof. There are six consequences (14) in degree 5 of identity (1). Combining
these with (13), we obtain seven identities J(v, w, x, y, z) each of which produces
eight consequences K(t, . . . , z) in degree 7; these can be expressed using partial
compositions as J ◦k γ (1 ≤ k ≤ 5) and γ ◦k J (1 ≤ k ≤ 3) where γ denotes
the commutator. For each partition λ of 7 we write Rλ: QS7 → Mdλ(Q) for the
corresponding irreducible representation; dλ is the number of standard tableaux of
shape λ. For each p ∈ S7 we use Clifton's algorithm [7, 13] to compute Rλ(p). For
each λ we construct a 56dλ × 12dλ matrix Cλ partitioned into dλ × dλ blocks. Each
consequence Ki (1 ≤ i ≤ 56) is a sum of 12 components K 1
corresponding
to the association types (19). Block (i, j) of Cλ contains Rλ(K j
i ). We compute the
RCF of Cλ; its rank cλ is the multiplicity of representation [λ] in the S7-module of
all consequences of the identities of lower degree (Figure 2).
i , . . . K 12
i
We substitute the identity permutation of t, u, . . . , z into each association type,
and obtain monomials ξ1, . . . , ξ12 which we expand using the special commutator
into a linear combination of eight associative monomials. For each λ we construct
a dλ × 12dλ matrix Eλ: one row of dλ × dλ blocks in which block j is Rλ(ξj )t. We
compute the RCF of Eλ and find its rank eλ; then aλ = 12dλ −eλ is the multiplicity
of [λ] in the kernel of the expansion map (Figure 2). We compute the aλ × 12dλ
matrix Nλ in RCF whose row space is the nullspace of Eλ, and check that the
row space of Nλ contains the row space of Cλ. The difference nλ = aλ − cλ is the
multiplicity of [λ] in the quotient module Nλ/Cλ (we identify each matrix with its
row space) of new identities in degree 7 for partition λ (Figure 2).
(cid:3)
λ
7 61
52 512
43 421 413 321 322 3212 314 231 2213 215 17
dλ
cλ
aλ
nλ
6
35
14
15
14
1
21
12 71 162 173 157 394 225 231 233
12 71 162 173 158 396 226 234 235
0
2
21
20
0
0
0
1
2
1
3
35
15
14
385 166 153
390 168 155
2
5
2
14
152
155
3
6
1
66 11
67 11
0
1
Figure 2. Multiplicities in degree 7 for the ternary commutator
Lemma 6.2. The S7-module of new identities for the special translator is nonzero
and has the following decomposition into isotypic components:
[52] ⊕ [512] ⊕ [43] ⊕ [421]3 ⊕ [413]2 ⊕ [321]2 ⊕ [322]2 ⊕ [3212]3 ⊕ [314] ⊕ [231] ⊕ [2213].
Proof. Similar to the proof of Lemma 6.1; see Figure 3.
(cid:3)
λ
7 61
52 512
43 421 413 321 322 3212 314 231 2213 215 17
dλ
cλ
aλ
nλ
6
35
14
15
14
1
21
12 68 156 168 155 388 222 232 232
12 68 157 169 156 391 224 234 234
2
0
21
20
1
1
1
0
3
2
2
35
15
14
388 168 155
391 169 156
1
1
3
14
156
157
1
6
1
68 12
68 12
0
0
Figure 3. Multiplicities in degree 7 for the ternary translator
For two ternary operations we have 23 · 12 = 96 association types: in each type
for one operation (19), each operation may be replaced by either of two operations.
12
MURRAY R. BREMNER AND HADER A. ELGENDY
Theorem 6.3. Every identity in degree 7 satisfied by the special commutator and
translator in every associative triple system follows from (1) -- (3), (13), (15), and
(16) -- (18), together with the new identities for the commutator and translator sep-
arately whose existence is demonstrated by Lemmas 6.1 and 6.2. That is, special
comtrans algebras satisfy no new identities in which every term has both operations.
λ, D2
Proof. Each identity J in equations (1) -- (3) produces 12 consequences in degree 5:
the partial compositions J ◦k γ, J ◦k δ (1 ≤ k ≤ 3), γ ◦k J, γ ◦k J (1 ≤ k ≤ 3),
where γ, δ denote commutator and translator. Including (13), (15), (16) -- (18), we
obtain 41 consequences K in degree 5. Each of these produces 16 consequences in
degree 7: K ◦k γ, K ◦k δ (1 ≤ k ≤ 5), γ ◦ℓ K, γ ◦ℓ K (1 ≤ ℓ ≤ 3). For each partition
λ of 7, the matrix Cλ representing the consequences has size 656dλ × 96dλ. Let
D1
λ be obtained from the matrices Cλ in the proofs of Lemmas 6.1, 6.2 by
embedding the 12 association types for one operation into the 96 association types
for two operations. The row space of D1
λ) contains the identities satisfied
by the commutator (resp. translator) in degree 7. We stack Cλ on top of D1
λ and
D2
λ to obtain CDλ; we denote its rank by cλ. The expansion matrix Eλ has size
dλ × 96dλ; we denote its rank by eλ, so it has nullity aλ = 96dλ − eλ. Let Nλ be
the matrix whose row space is the nullspace of Eλ. For each partition λ, we find
that cλ = aλ and that the RCFs of Cλ and Nλ coincide.
(cid:3)
λ (resp. D2
7. Enveloping algebras for 2 × 2 matrices
In this final section we use noncommutative Grobner bases [5] to construct the
universal associative enveloping algebras of the nonassociative triple systems AC ,
AT , and ACT , obtained by applying the special commutator, special translator, and
both together, to the associative triple system A = (aij ) of 2 × 2 matrices. This
method has previously been used for the universal envelopes of triple systems [16]
obtained by applying trilinear operations [8] to the 2 × 2 matrices with a11 =
a22 = 0, and for infinite families of simple anti-Jordan triple systems [15, 17]. This
approach to the representation theory of comtrans algebras is based on the special
commutator and special translator in an associative triple system, and therefore
differs essentially from the approach of Im, Shen & Smith [22, 23, 31].
Definition 7.1. Let A be the associative triple system of 2 × 2 matrices. Let B
and X be the basis of matrix units and a set of symbols in bijection with B:
B = {Eij 1 ≤ i, j ≤ 2},
X = {eij 1 ≤ i, j ≤ 2},
η(Eij ) = eij.
Extend η linearly to the injective map η: A → F hXi where F hXi is the free asso-
ciative algebra generated by X. Define three ideals in F hXi as follows:
I C = hGC i,
I T = hGT i,
I CT = hGCT i,
GC = {eijeklest − ekleijest − η([Eij , Ekl, Est])},
GT = {eijeklest − eklesteij − η(hEij , Ekl, Esti)},
GCT = GT ∪ GC .
The corresponding universal associative enveloping algebras are
U (AC ) = F hXi/I C,
U (AT ) = F hXi/I T ,
U (ACT ) = F hXi/I CT .
We write eij ≺ ekl if i < k, or i = k, j < l; and also e11, e12, e21, e22 = a, b, c, d.
Lemma 7.2. The universal enveloping algebra U (AC ) has basis
BC = { 1, a, b, c, d, a2, ab, ca, cb }.
SPECIAL IDENTITIES FOR COMTRANS ALGEBRAS
13
Proof. The set GT contains 64 elements; after putting each generator in standard
form (reverse deglex order and monic), only 24 distinct generators remain:
G1 = ba2−aba,
G5 = ca2−aca−c, G6 = cab−acb−d, G7 = cac−ac2,
G9 = cba−bca+a, G10 = cb2−bcb+b, G11 = cbc−bc2−c, G12 = cbd−bcd−d,
G13 = da2−ada,
G17 = dba−bda,
G21 = dca−cda−c, G22 = dcb−cdb−d, G23 = dc2−cdc,
G15 = dac−adc,
G19 = dbc−bdc+a, G20 = dbd−bd2+b,
G3 = bac−abc+a, G4 = bad−abd+b,
G14 = dab−adb,
G18 = db2−bdb,
G16 = dad−ad2,
G24 = dcd−cd2.
G2 = bab−ab2,
G8 = cad−acd,
There are only 56 distinct normal forms of the compositions of these generators;
the corresponding compositions appear in Figure 4.
S3 = G3ac−baG7,
S1 = G3a2−baG5,
S2 = G3ab−baG6,
S4 = G3ad−baG8,
S6 = G3b2−baG10, S7 = G3bc−baG11, S8 = G3bd−baG12,
S5 = G3ba−baG9,
S9 = G4ba−baG17, S10 = G4b2−baG18, S11 = G4bc−baG19, S12 = G4bd−baG20,
S13 = G4ca−baG21, S14 = G4cb−baG22, S15 = G4c2−baG23, S16 = G4cd−bsG24,
S17 = G7a2−caG5, S18 = G7ab−caG6, S19 = G7ac−caG7, S20 = G7ad−caG8,
S21 = G7ba−caG9, S22 = G7b2−caG10, S23 = G7bc−caG11, S24 = G7bd−caG12,
S25 = G8ba−caG17, S26 = G8b2−caG18, S27 = G8bc−caG19, S28 = G8bd−caG20,
S29 = G9a−cG1,
S33 = G11ba−cbG9, S34 = G11b2−cbG10, S35 = G11bc−cbG11, S36 = G11bd−cbG12,
S37 = G12a2−cbG13, S38 = G12ab−cbG14, S39 = G12ac−cbG15, S40 = G12ad−cbG16,
S41 = G12ba−cbG17, S42 = G12b2−cbG18, S43 = G12bc−cbG19, S44 = G12bd−cbG20,
S45 = G17a−dG1,
S59 = G21c−dG7,
S53 = G22ac−dcG3, S54 = G22ad−dcG4, S55 = G22a−dG9,
S47 = G21a−dG5,
S51 = G22a2−dcG1, S52 = G22ab−dcG2,
S56 = G22b−dG10.
S46 = G17b−dG2,
S50 = G21d−dG8,
S48 = G21b−dG6,
S32 = G9d−cG4,
S31 = G9c−cG3,
S30 = G9b−cG2,
Figure 4. Compositions of the generators for the ideal I C ⊂ F hXi
To compute normal forms we eliminate all occurrences of leading monomials of
the generators. We write ≡ for congruence modulo G1, . . . , G24. For example, to
find the normal form of S1, we eliminate the leading monomials of G5, G1, G3:
S1 = (bac − abc + a)a2 − ba(ca2 − aca − c) = (−abc + a)a2 − ba(−aca − c)
≡ −ab(aca + c) + a3 + (aba)ca + bac
≡ −a((abc − a)a + bc) + a3 + a(abc − a)a + (abc − a) = a3 − a.
Similar calculations produce the normal forms of S2, . . . S56:
a2b − b, a2c, a2d, aba, ab2, abc−a, abd−b, b2a, b3, b2c, b2d, bca−ada−a,
bcb−adb−b, bc2−adc, bcd−ad2, c2a, c2b, c3, c2d, cda, cdb, cdc, cd2, d2a, d2b,
d2c−c, d3−d, da−bc+aa, db−bd+ab, dc−ca+3ac, dd−cb+3ad, bca−a, bcb−b,
bc2, bcd, d2a−bca+ada+a, d2b−bcb+adb+b, d2c−bc2+adc−c, d3−bcd+ad2−d,
bda, bdb, bdc−a, bd2−b, ba, b2, dc−ca+ac, d2−cb+ad, c2, cd,
d2a−3ada+abc−a3, d2b−3adb+abd−a2b, d2c−adc+aca−a2c−c,
d3−ad2+acb−a2d−d, da− 1
3 bc+ 1
3 a2, db− 1
3 bd+ 1
3 ab.
Including these normal forms with the original 24 generators, and then self-reducing,
produces the following 16 generators:
(20)
ac,
ad,
da, db,
bc−a2,
ba,
dc−ca, d2−cb, a3−a,
b2,
bd−ab,
a2b−b,
c2,
ca2−c,
cd,
cab−d.
14
MURRAY R. BREMNER AND HADER A. ELGENDY
All compositions of these 16 generators reduce to 0, and so we have a Grobner basis
for I C. A basis for U (AC ) consists of the cosets of those monomials which are not
divisible by the leading monomial of any element of the Grobner basis.
(cid:3)
Lemma 7.3. In U (AC ) we have the relations
ac = ad = ba = b2 = cd = da = db = c2 = 0,
bc = a2, bd = ab, dc = ca, d2 = cb, a3 = a, a2b = b, ca2 = c, cab = d.
Proof. This follows immediately from the Grobner basis (20).
(cid:3)
Lemma 7.4. The nonzero structure constants of U (AC ) are
a · b = ab,
b · cb = b,
d · d = cb,
a · a2 = a,
a · a = a2,
c · a = ca,
b · ca = a,
d · c = ca,
d · ca = c,
a2 · a2 = a2, a2 · ab = ab, ab · c = a,
ca · a = c,
cb · ca = ca,
ca · b = d,
cb · cb = cb.
a · ab = b,
c · b = cb,
d · cb = d,
ab · d = b,
b · c = a2,
c · a2 = c,
a2 · a = a,
ab · ca = a2, ab · cb = ab,
b · d = ac,
c · ab = d,
a2 · b = b,
ca · a2 = ca, ca · ab = cb, cb · c = c,
ca · d = d,
Proof. This follows immediately from Lemma 7.3.
(cid:3)
Theorem 7.5. The Wedderburn decomposition of U (AC ) is
U (AC ) = Q ⊕ M2×2(Q) ⊕ M2×2(Q),
where M2×2(Q) is the ordinary associative algebra of 2 × 2 matrices.
Proof. Following [4], we first verify that the radical is zero and hence U (AC ) is
semisimple. The center Z(U (AC )) has this basis and structure constants:
z1 · z1 = z1, z1 · z2 = z1, z1 · z3 = z3, z2 · z2 = z3, z2 · z3 = z2, z3 · z3 = z3.
z1 = 1,
z2 = a + d,
z3 = a2 + cb,
The minimal polynomial of z3 is t2 − t and hence Z(U (AC )) = J ⊕ K where
J = hz3 − z1i and K = hz3i. We have dim J = 1 with basis z3 − z1, and dim K = 2
with basis z2, z3. In K the identity element is z3, and the minimal polynomial of z2
is t2 − z3. Hence K splits into 1-dimensional ideals with bases z2 − z3 and z2 + z3.
After scaling, we obtain a basis of orthogonal idempotents for Z(U (AC )):
e1 = z1 − z3,
e2 = 1
2 (z2 − z3),
e3 = 1
2 (z2 + z3).
These elements of the center correspond to the following elements of U (AC ):
e1 = 1 − a2 + cb,
e2 = 1
2 (a + d − a2 + cb),
e3 = 1
2 (a + d + a2 + cb).
The ideals in U (AC ) generated by e1, e2, e3 have dimensions 1, 4, 4 respectively,
which completes the proof. (We omit the construction of an isomorphism between
each simple two-sided ideal in U (AC ) and the corresponding matrix algebra.) (cid:3)
Lemma 7.6. The universal enveloping algebra U (AT ) has basis
BT = { am, b, c, ab, ca, cb, and m ≥ 0, 0 ≤ n ≤ 2 }.
SPECIAL IDENTITIES FOR COMTRANS ALGEBRAS
15
Proof. The set GT contains 64 elements; we put each generator in standard form
(reverse deglex order and monic) and self-reduce the set, obtaining 40 generators:
aba−a2b+b,
bac−acb,
bda−abd+b,
cad−adc,
c2a−ac2,
da2−a2d,
db2−b2d,
dc2−c2d,
ada−a2d,
aca−a2c,
b2a−ab2,
bad−adb,
ca2−a2c−c,
bdb−bbd,
cb2−b2c,
cba−acb,
c2b−bcc,
cda−acd,
dab−abd+b, dac−acd,
dbc−bcd,
dcd−cd2,
dbd−bd2+b, dca−adc−c,
d2b−bd2+b,
d2a−ad2,
ba2−a2b+b,
bca−abc,
cab−abc−d+a,
cbc−bc2−c,
cdb−bcd,
dad−ad2,
bab−ab2,
bcb−b2c−b,
cac−ac2,
cbd−bdc−d+a,
cdc−c2d,
dba−adb,
dcb−bdc−d+a,
d2c−cd2−c.
There are 143 nontrivial compositions of these generators; including their normal
forms and self-reducing produces 16 generators:
(21)
ac,
ba,
da−ad,
cab+a2d−a3−d+a,
b2,
db,
bc+ad−a2,
dc−ca,
bd−ab,
d2−cb−ad,
c2,
a2b−b,
a3d−a4−ad+a2.
cad,
cd,
ca2−c,
All compositions of these generators reduce to 0, so we have a Grobner basis of I T .
A basis for U (AT ) consists of the cosets of those monomials which are not divisible
by the leading monomial of any element of the Grobner basis.
(cid:3)
Lemma 7.7. In U (AT ), we have the relations
ac = ba = b2 = c2 = cd = db = cad = 0,
bc = −ad + a2,
a2b = b,
ca2 = c,
da = ad,
bd = ab,
cab = −a2d + a3 + d − a,
dc = ca,
d2 = cb + ad,
a3d = a4 + ad − a2.
Proof. This follows immediately from the Grobner basis (21).
(cid:3)
Theorem 7.8. For m, l ≥ 0 and 0 ≤ n ≤ 2, the structure constants of U (AT ) are
(22)
am · al = am+n,
am · b = (cid:26)ab
b
(m odd)
(m even),
(23)
am · and = (cid:26)am+nd
am+n+1 + am+n−2d − am+n−1
(m+n < 3)
(m+n ≥ 3),
b · c = −ad + a2,
b · ca = −a2d + a3,
b · cb = b,
b · d = ab,
c · am = (cid:26)ca (m odd)
(m even),
c
c · b = cb,
c · (ab) = −a2d + a3 + d − a,
ab · c = −a2d + a3,
ca · b = −a2d + a3 + d − a,
ca · ab = cb,
ab · ca = −ad + a2,
ab · cb = ab,
cb · c = c,
cb · ca = ca,
cb · cb = cb,
cb · d = −a2d + a3 + d − a,
and · am = (cid:26)am+nd
am+n+1 + am+n−2d − am+n−1
(m+n < 3)
(m+n ≥ 3),
d · c = ca,
d2 = cb + ad,
d · ca = c,
d · cb = −a2d + a3 + d − a,
ad · ca = ac,
ad · a2d = a5 − a3 + a2d,
ad · d = ad2,
(24)
ad · ad = a4 + ad − a2,
a2d · a2d = a6 + ad − a2.
16
MURRAY R. BREMNER AND HADER A. ELGENDY
Proof. The first relation of (22) is clear. For the second we use induction on m.
The claim is clear for 0 ≤ m ≤ 1. For m = 2 we have a2b = b by Lemma 7.7. If m
is odd (resp. even) then m−1 is even (resp. odd) and hence am · b = a(am−1b) = ab
and a(am−1b) = a(ab) = a2b = b by induction. Relation (23) is clear for m+n < 3.
We assume m+n ≥ 3 and use induction on m. For m = 1, Lemma 7.7 gives
For m+n > 3, the inductive hypothesis implies
a · a2d = a3d = a4 + ad − a2.
am · and = a · am−1and = a · (am+n + am+n−3d − am+n−2)
= am+n+1 + am+n−2d − am+n−1.
The proofs of the unnumbered relations are similar. For relations (24) we have
ad · a2d = adaad = aadad = aaadd = a3d2 = (a4 + ad − a2)d
= a4d + a(cb + ad) − a2d = a(a4 + ad − a2) + a(cb + ad) − a2d
= a5 + a2d − a3 + acb = a5 + a2d − a3,
a2d · a2d = aadaad = aaadad = aaaadd = a4d2 = a(a4 + ad − a2)d
= a5d + a2d2 − a3d = a5d + a2(cb + ad) − (a4 + ad − a2)
= a5d + a3d − a4 − ad + a2 = (a2 + 1)a3d − a4 − ad + a2
= (a2 + 1)(a4 + ad − a2) − a4 − ad + a2 = a6 + a3d − a4
= a6 + a4 + ad − a2 − a4 = a6 + ad − a2,
again using Lemma 7.7.
(cid:3)
Theorem 7.9. We have the isomorphism U (ACT ) ∼= U (AC ).
Proof. The set GCT contains 128 elements; we put each generator in standard form
and self-reduce the set, obtaining the following 44 generators:
ada−a2d,
bac−abc+a,
bda−abd+b,
cac−ac2,
cbd−bcd−d,
cdc−c2d,
dba−abd+b,
dcb−bcd−d,
d2c−cd2−c.
acb−abc+a,
bab−ab2,
bcb−b2c−b,
cab−abc−d+a,
cbc−bc2−c,
cdb−bcd,
dad−ad2,
dca−acd−c,
d2b−bd2+b,
adb−abd+b,
bad−abd+b,
bdb−bbd,
cad−acd,
c2a−ac2,
da2−a2d,
db2−b2d,
dc2−c2d,
aba−a2b+b,
adc−acd,
b2a−ab2,
bdc−bcd−a,
cba−abc+a,
c2b−bc2,
dab−abd+b,
dbc−bcd,
dcd−cd2,
aca−a2c,
ba2−a2b+b,
bca−abc,
ca2−a2c−c,
cb2−b2c,
cda−acd,
dac−acd,
dbd−bd2+b,
d2a−ad2,
There are 133 nontrivial compositions (omitted); we include their normal forms,
self-reduce, and obtain 16 generators, coinciding with the Grobner basis (20). (cid:3)
Conjecture 7.10. If C is a finite dimensional comtrans algebra (special or not)
then U (C) is also finite dimensional.
References
[1] M. A. Akivis: The local algebras of a multidimensional three-web. Siberian Math. J. 17, 1
(1976) 3 -- 8.
[2] M. A. Akivis, V. V. Goldberg: Algebraic aspects of web geometry. Comment. Math. Univ.
Carolin. 41, 2 (2000) 205 -- 236.
[3] F. Bagherzadeh, M. Bremner, S. Madariaga: Jordan trialgebras and post-Jordan algebras. J.
Algebra 486 (2017) 360 -- 395.
SPECIAL IDENTITIES FOR COMTRANS ALGEBRAS
17
[4] M. R. Bremner: How to compute the Wedderburn decomposition of a finite-dimensional
associative algebra. Groups Complex. Cryptol. 3, 1 (2011) 47 -- 66.
[5] M. R. Bremner: Free associative algebras, noncommutative Grobner bases, and universal
associative envelopes for nonassociative structures. Comment. Math. Univ. Carolin. 55, 3
(2014) 341 -- 379.
[6] M. R. Bremner, V. Dotsenko: Algebraic Operads: An Algorithmic Companion. CRC Press,
Boca Raton, 2016.
[7] M. R. Bremner, S. Madariaga, L. A. Peresi: Structure theory for the group algebra of the
symmetric group, with applications to polynomial identities for the octonions. Comment.
Math. Univ. Carolin. 57, 4 (2016) 413 -- 452.
[8] M. R. Bremner, L. A. Peresi: Classification of trilinear operations. Comm. Algebra 35, 9
(2007) 2932 -- 2959.
[9] M. R. Bremner, L. A. Peresi: An application of lattice basis reduction to polynomial identities
for algebraic structures. Linear Algebra Appl. 430, 2-3 (2009) 642 -- 659.
[10] M. R. Bremner, L. A. Peresi: Higher identities for the ternary commutator. J. Phys. A 45,
50 (2012) 505201, 12 pages.
[11] O. Chein, H. Pflugfelder, J. D. H. Smith (editors): Quasigroups and Loops: Theory and
Applications. Heldermann, Berlin, 1990.
[12] S. S. Chern: Web geometry. Bull. Amer. Math. Soc. (N.S.) 6, 1 (1982) 1 -- 8.
[13] J. M. Clifton: A simplification of the computation of the natural representation of the sym-
metric group Sn. Proc. Amer. Math. Soc. 83, 2 (1981) 248 -- 250.
[14] V. Dotsenko, A. Khoroshkin: Grobner bases for operads. Duke Math. J. 153, 2 (2010) 363 --
396.
[15] H. A. Elgendy: The universal associative envelope of the anti-Jordan triple system of n × n
matrices. J. Algebra 383 (2013) 1 -- 28.
[16] H. A. Elgendy: Universal associative envelopes of nonassociative triple systems. Comm. Al-
gebra 42, 4 (2014) 1785 -- 1810.
[17] H. A. Elgendy: Representations of simple anti-Jordan triple systems of m × n matrices. J.
Algebra Appl. 16, 5 (2017) 1750093, 28 pages.
[18] V. Ginzburg, M. Kapranov: Koszul duality for operads. Duke Math. J. 76, 1 (1994) 203 -- 272.
[19] V. V. Goldberg: Theory of Multicodimensional (n+1)-Webs. Kluwer, Dordrecht, 1988.
[20] I. R. Hentzel: Applying group representation to nonassociative algebras. Ring Theory (Proc.
Conf., Ohio Univ., Athens, Ohio, 1976), 133 -- 141. Lecture Notes in Pure and Appl. Math.,
Vol. 25. Dekker, New York, 1977.
[21] I. R. Hentzel: Processing identities by group representation. Computers in Nonassociative
Rings and Algebras (Special Session, 82nd Annual Meeting Amer. Math. Soc., San Antonio,
Tex., 1976), 13 -- 40. Academic Press, New York, 1977.
[22] B. Im, J. D. H. Smith: Trilinear products and comtrans algebra representations. Linear
Algebra Appl. 430, 1 (2009) 17 -- 26.
[23] B. Im, J. D. H. Smith: Representation theory for varieties of comtrans algebras and Lie triple
systems. Internat. J. Algebra Comput. 21, 3 (2011) 459 -- 472.
[24] E. N. Kuzmin: Structure and representations of finite dimensional Malcev algebras. Quasi-
groups Related Systems 22, 1 (2014) 97 -- 132.
[25] W. G. Lister: Ternary rings. Trans. Amer. Math. Soc. 154 (1971) 37 -- 55.
[26] J.-L. Loday, B. Vallette: Algebraic Operads. Grundlehren der Mathematischen Wis-
senschaften [Fundamental Principles of Mathematical Sciences], 346. Springer, Heidelberg,
2012.
[27] A. I. Malcev: On algebras defined by identities. [Russian]. Mat. Sbornik N.S. 26(68) (1950)
19 -- 33.
[28] M. Markl, S. Shnider, J. Stasheff: Operads in Algebra, Topology and Physics. Mathematical
Surveys and Monographs, 96. American Mathematical Society, Providence, RI, 2002.
[29] K. McCrimmon: A Taste of Jordan Algebras. Springer, New York, 2004.
[30] R. Rossmanith, J. D. H. Smith: Alternative approaches to alternative algebras. Acta Sci.
Math. (Szeged) 62, 1-2 (1996) 145 -- 151.
[31] X. Shen, J. D. H. Smith: Representation theory of comtrans algebras. J. Pure Appl. Algebra
80, 2 (1992) 177 -- 195.
[32] X. Shen, J. D. H. Smith: Simple multilinear algebras, rectangular matrices and Lie algebras.
J. Algebra 160, 2 (1993) 424 -- 433.
18
MURRAY R. BREMNER AND HADER A. ELGENDY
[33] J. D. H. Smith: Multilinear algebras and Lie's theorem for formal n-loops. Arch. Math.
(Basel) 51, 2 (1988) 169 -- 177.
[34] J. D. H. Smith: Comtrans algebras and their physical applications. Algebraic Methods in
Logic and in Computer Science, 319 -- 326. Polish Acad. Sci. Inst. Math., Warsaw, 1993.
[35] W. Specht: Gesetze in Ringen, I. [German]. Math. Z. 52, (1950) 557 -- 589.
[36] B. Vallette: Algebra + homotopy = operad. Symplectic, Poisson, and Noncommutative Ge-
ometry, 229 -- 290. Math. Sci. Res. Inst. Publ., 62. Cambridge Univ. Press, New York, 2014.
Department of Mathematics and Statistics, University of Saskatchewan, Saskatoon,
Canada
E-mail address: [email protected]
Faculty of Science, Department of Mathematics, Damietta University, Damietta
34517, Egypt
E-mail address: [email protected]
|
1211.0981 | 1 | 1211 | 2012-11-05T19:34:39 | Skew group algebras, invariants and Weyl Algebras | [
"math.RA"
] | The aim of this paper is two fold:
First to study finite groups $G$ of automorphisms of the homogenized Weyl algebra $B_{n}$, the skew group algebra $B_{n}\ast G$, the ring of invariants $B_{n}^{G}$, and the relations of these algebras with the Weyl algebra $A_{n}$, with the skew group algebra $A_{n}\ast G$, and with the ring of invariants $A_{n}^{G}$. Of particular interest is the case $n=1$.
In the on the other hand, we consider the invariant ring $\QTR{sl}{C}[X]^{G}$ of the polynomial ring $K[X]$ in $n$ generators, where $G$ is a finite subgroup of $Gl(n,\QTR{sl}{C}$) such that any element in $G$ different from the identity does not have one as an eigenvalue. We study the relations between the category of finitely generated modules over $\QTR{sl}{C}[X]^{G}$ and the corresponding category over the skew group algebra $\QTR{sl}{C}% [X]\ast G$. We obtain a generalization of known results for $n=2$ and $G$ a finite subgroup of $Sl(2,C)$. In the last part of the paper we extend the results for the polynomial algebra $C[X]$ to the homogenized Weyl algebra $B_{n}$. | math.RA | math |
SKEW GROUP ALGEBRAS, INVARIANTS
AND WEYL ALGEBRAS
ROBERTO MART´INEZ-VILLA AND JERONIMO MONDRAG ´ON
Abstract. The aim of this paper is two fold:
First to study finite groups G of automorphisms of the homogenized Weyl
algebra Bn, the skew group algebra Bn ∗ G, the ring of invariants BG
n , and
the relations of these algebras with the Weyl algebra An, with the skew group
algebra An ∗ G, and with the ring of invariants AG
n . Of particular interest is
the case n = 1.
In the on the other hand, we consider the invariant ring C[X]G of the
polynomial ring K[X] in n generators, where G is a finite subgroup of Gl(n, C)
such that any element in G different from the identity does not have one as an
eigenvalue. We study the relations between the category of finitely generated
modules over C[X]G and the corresponding category over the skew group
algebra C[X] ∗ G. We obtain a generalization of known results for n = 2 and
G a finite subgroup of Sl(2,C ). In the last part of the paper we extend the
results for the polynomial algebra C[X] to the homogenized Weyl algebra Bn.
1. Automorphism of the homogenized Weyl algebra.
In tis section we will assume the reader is familiar with basic results on Weyl
algebras as in [Co] , For results on the homogenized Weyl algebra we refer to [MMo]
Let K be a field of zero characteristic. In this section we consider the it homog-
enized Weyl algebra Bn defined by generators and relations as:
Bn = K < X1, X2, ...Xn, Y1, Y2, ...Yn, Z > /{[Xi, δj] − ∂ijZ 2, [Xi, Xj], [Yi, Yj],
[Xi, Z'], [Yi, Z']}, with K < X1, X2, ...Xn, Y1, Y2, ...Yn, Z >the free algebra in 2n+1
generators, [u, v'] the commutator uv − vu and δij, Kronecker's delta.
It is known Bn has a Poincare-Birkoff basis and it is quadratic, hence by [Li], [GH]
B(K, K).
n be its Yoneda algebra [GM1],[GM2], B!
it is Koszul. Let B!
Extk
n = ⊕k≥0
n has the same quiver as Bn and relations orthogonal with respect
n has the following form:
The algebra B!
to the canonical bilinear form, it is easy to see that B!
n = Kq[X1, X2, ...Xn, Y1, Y2, ...Yn, Z]/{X 2
B!
where Kq[X1, X2, ...Xn, Y1, Y2, ...Yn, Z] denote the quantum polynomial ring.
PXiYi + Z 2},
i , Y 2
j ,
i=0
n
(u, v) denotes the anti commutators uv + vu.
K < X1, X2, ...Xn, Y1, Y2, ...Yn, Z > /{(Xi, Xj), (Yi, Yj), (Xi, Z), (Yi, Z)}. Here
The polynomial algebra Cn = K[X1, X2, ...Xn, Y1, Y2, ...Yn, Z] is a Koszul al-
j }. The
n = Kq[X1, X2, ...Xn, Y1, Y2, ...Yn, Z]/{X 2
gebra with Yoneda algebra C!
i , Y 2
Date: November 5, 2012.
2000 Mathematics Subject Classification. Primary 05C38, 15A15; Secondary 05A15, 15A18.
Key words and phrases. Group invariants, skew group algebras, Weyl algebras.
1
2
ROBERTO MART´INEZ-VILLA AND JERONIMO MONDRAG ´ON
Weyl algebra is defined by An = K < X1, X2, ...Xn, Y1, Y2, ...Yn, > /{[Xi, Yj] −
∂ij, [Xi, Xj]', [Yi, Yj], [Xi, Z'], [Yi, Z]'}.
Observe we obtain Cn as a quotient of Bn and C!
n is a sub algebra of B!
n.
These algebras are related as follows: Bn /ZBn ∼= Cn and Bn /(Z − 1)Bn ∼= An
The algebra B!
In fact we have:
n-module of rank two.
n is a free C!
Proposition 1. There exists a C!
n-module decomposition: B!
n = C!
n ⊕ ZC!
n.
We consider now a finite group G of grade preserving automorphisms of Bn.
It was proved in [MMo ] that the center of Bn is K[Z] and any automorphism
σ ∈ G takes the center to the center, and since σ is grade preserving σ(Z) is an
homogeneous element of degree one, hence σ(Z) = λσZ. We are assuming G has
finite order m, it follows σm(Z) = Z = λm
σ Z , and λσ is an m-th root of unity. If
we denote by m√1 the group of m roots of unity, there is a group homomorphism
η : G → m√1 given by η(σ) = λσ , since m√1 is a cyclic group the image of η is
cyclic and we have a group extension: 1 → N → G → Zk → 0 with Zk the cyclic
group of order k and N is the subgroup of G such that for every σ in G, σ(Z) = Z.
Since Bn is generated in degree one, the action of G on Bn is determined by the
action on M = (Bn)1 =
n
⊕i=1
n
KXi ⊕ (
⊕i=1
KYi) ⊕ KZ.
To determine the structure of G we need to look for automorphisms of M which
leave KZ invariant and preserve the relations: [Xi, δj]=∂ijZ 2, [Xi, Xj]= [Yi, Yj]=
0 .
For any element σof G we have equations:
σ(Xj ) =
A2i−1,2j−1Xi +
A2i,2j−1Yi + µjZ
σ(Yk) =
A2ℓ−1,2kXℓ +
A2ℓ,2kYℓ + ν kZ
n
n
Pi=1
Pℓ=1
n
n
Pi=1
Pℓ=1
σ(Z) = λZ.
We must have equalities:
σ(XjXk−XkXj) = 0 = σ(Xj)σ(Xk) − σ(Xk)σ(Xj).
Using the relations that define Bn, we obtain after cancellation the following
equation involving 2 × 2 determinants:
n
Xi=1
(cid:12)(cid:12)(cid:12)(cid:12)
Since for every i we have XiYi−YiXi=Z2, it follows:
=0.
A2i−1,2j−1 A2i,2j−1
n
A2i−1,2j−1 A2i,2j−1
A2i−1,2k−1 A2i,2k−1 (cid:12)(cid:12)(cid:12)(cid:12)
Xi=1
A2i−1,2k−1 A2i,2k−1 (cid:12)(cid:12)(cid:12)(cid:12)
(cid:12)(cid:12)(cid:12)(cid:12)
(XiYi−YiXi)= 0.
We may assume j < k. Using that a matrix and its transpose have the same
determinant we obtain the equivalent relations:
1)
n
Xi=1
(cid:12)(cid:12)(cid:12)(cid:12)
A2i−1,2j−1 A2i−1,2k−1
A2i,2j−1
A2i,2k−1
(cid:12)(cid:12)(cid:12)(cid:12)
=0, for all j, k with j < k.
In a similar way, for we obtain from the relation:
SKEW GROUP ALGEBRAS AND INVARIANTS
3
σ(Yj Yk−YkYj) = 0 = σ(Yj)σ(Yk) − σ(Yk)σ(Yj), the following equation:
n
2)
Xi=1
A2i−1,2j A2i−1,2k
A2i,2j
A2i,2k
=0, for all j, k with j < k.
From the relation:
σ(Xj Yj−YjXj) = σ(Xj)σ(Yj) − σ(Yj)σ(Xj) = σ(Z 2) = λ2Z 2, the following
equation:
A2i−1,2j−1 A2i−1,2j
A2i,2j−1
A2i,2j
(XiYi−YiXi)=λ2 Z2.
(cid:12)(cid:12)(cid:12)(cid:12)
(cid:12)(cid:12)(cid:12)(cid:12)
n
Xi=1
(cid:12)(cid:12)(cid:12)(cid:12)
(cid:12)(cid:12)(cid:12)(cid:12)
Since for every i we have XiYi−YiXi=Z2, it follows:
n
3)
Xi=1
(cid:12)(cid:12)(cid:12)(cid:12)
A2i−1,2j−1 A2i−1,2j
A2i,2j−1
A2i,2j
= λ2
(cid:12)(cid:12)(cid:12)(cid:12)
With similar calculations, from the equation:
σ(Xj Yk−YkXj) = 0 = σ(Xj )σ(Yk) − σ(Yk)σ(Xj ), for k 6= 0.
we get the equation:
A2i−1,2j−1 A2i−1,2k
A2i,2j−1
A2i,2k
n
Xi=1
(cid:12)(cid:12)(cid:12)(cid:12)
Xi=1
n
(cid:12)(cid:12)(cid:12)(cid:12)
which implies:
4)
A2i−1,2j−1 A2i−1,2k
A2i,2j−1
A2i,2k
We have proved the following:
(XiYi−YiXi) = 0.
(cid:12)(cid:12)(cid:12)(cid:12)
= 0 , for i 6= k.
(cid:12)(cid:12)(cid:12)(cid:12)
Theorem 1. Let Bn be the homogenized Weyl algebra in n+1 generators. Then
an automorphism σ of M =
KYi) ⊕ KZ, with matrix in block
,(cid:21) where ρ is the vector: ρ = (µ1, µ2, ...µn, ν1, ν2...ν n) and λ 6= 0,
KXi ⊕ (
ρ λ
⊕i=1
⊕i=1
n
n
extends to an automorphism of Bn if and only if A satisfies the following equations:
(cid:12)(cid:12)(cid:12)(cid:12)
A2i−1,2j−1 A2i−1,2k−1
A2i,2j−1
A2i,2k−1
A2i−1,2j A2i−1,2k
A2i,2j
A2i,2k
A2i−1,2j−1 A2i−1,2j
A2i,2j−1
A2i,2j
(cid:12)(cid:12)(cid:12)(cid:12)
=0, for all j, k with j < k,
=0, for all j, k with j < k.
= λ2,4)
(cid:12)(cid:12)(cid:12)(cid:12)
n
Xi=1
(cid:12)(cid:12)(cid:12)(cid:12)
A2i−1,2j−1 A2i−1,2k
A2i,2j−1
A2i,2k
= 0 , for
(cid:12)(cid:12)(cid:12)(cid:12)
n
1)
form:(cid:20) A 0
(cid:12)(cid:12)(cid:12)(cid:12)
(cid:12)(cid:12)(cid:12)(cid:12)
(cid:12)(cid:12)(cid:12)(cid:12)
Xi=1
Xi=1
Xi=1
i 6= k.
2)
3)
n
n
(cid:12)(cid:12)(cid:12)(cid:12)
A particular case is obtained when ρ = 0, λ = 1 and the matrix A satisfies
A2i−1,2i−1 A2i−1,2i
= 1 for all i and all remaining 2 × 2 minors of A are zero.
A2i,2i
A2i,2i−1
This is the product of n matrices of seize 2 × 2 and determinant one.
Corollary 1. Let G1, G2, ...Gn be finite subgroups of Sl(2, K),then the product G =
G1 × G2 × ...× Gn acts as automorphism group of the homogenized algebra in n + 1
variables, Bn.
(cid:12)(cid:12)(cid:12)(cid:12)
4
ROBERTO MART´INEZ-VILLA AND JERONIMO MONDRAG ´ON
2. Structure of the homogenized Weyl algebra Bn
and its skew group algebra Bn ∗ G.
In this section we study the structure of the homogenized Weyl algebras Bn.
We will see that they can be obtained from the homogenized algebras Bi, Bj with
i + j = n, as follows: Bn = Bi ⊗K Bj/(Z ⊗ 1 − 1 ⊗ Z)Bi ⊗K Bj. This result is
very similar to the situation of the Weyl algebras for which it is well known [Co]
that for i + j = n, there is an isomorphism of K-algebras: An = Ai ⊗K Aj or the
polynomial algebras Ci = K[X1, X2, ...Xi] , Cj = K[Xi+1, Xi+2, ...Xn] for which
the isomorphism Cn ∼= Ci ⊗ Cn−i is well known.
Theorem 2. For integers n, m > 0 let Bn, Bm and Bn+m be the corresponding
homogenized algebras. Then there exists an isomorphism of graded K-algebras:
Bn+m = Bn ⊗K Bm/(Z ⊗ 1 − 1 ⊗ Z)Bn ⊗K Bm.
Proof. The algebras Bn, Bm and Bn+m can be written as:
[Xi, Z'], [Yi, Z']}
[Xi, Xj], [Yi, Yj ], [Xi, Z'], [Yi, Z']}
[Yi, Yj], [Xi, Z'], [Yi, Z']}.
by: ϕi(Xj) = Xj, ϕi(Yj) = Yj, ϕi(Z) = Z and i = 1, 2.
Bn = K < X1, X2, ...Xn, Y1, Y2, ...Yn, Z > /{[Xi, δj] − ∂ijZ 2, [Xi, Xj], [Yi, Yj],
Bm = K < Xn+1, Xn+2, ...Xn+m, Yn+1, Yn+2, ...Yn+m, Z > /{[Xi, δj] − ∂ijZ 2,
Bn+m = K < X1, X2, ...Xn+m, Y1, Y2, ...Yn+m, Z > /{[Xi, δj] − ∂ijZ 2, [Xi, Xj],
Then there are natural inclusions: ϕ1 : Bn → Bn+m and ϕ2 : Bm → Bn+m given
Let j1 : Bn → Bn+m and j2 : Bm → Bn+m, be the inclusions j1(b1) = b1 ⊗ 1,
and j2(b2) = 1 ⊗ b2. Since multiplication is bilinear we have a vector space map:
ϕ : Bn ⊗K Bm → Bn+m given by ϕ(b1 ⊗ b2) = ϕ1(b1)ϕ2(b2), such that the diagram:
Bn
*)
j1→ Bn ⊗K Bm
ϕ1 ց
↓ ϕ
Bn+m
j2←− Bm
ւ ϕ2
commutes.
Since ϕ1(b1)ϕ2(b2) = ϕ2(b2)ϕ1(b1) with b1 ∈ Bn, b2 ∈ Bm , ϕ is an algebra
homomorphism. It is clear ϕ is surjective and ϕ(Z ⊗ 1 − 1 ⊗ Z) = 0, hence (Z ⊗
1 − 1 ⊗ Z)Bn ⊗K Bm is contained in the kernel of ϕ. We will prove that they are
actually equal.
Let b be an element of degree t in the kernel of ϕ.
The element b has the following form:
Xα+β+k+,ℓ+α′+β′=t
Y β = Y β1
2 ...Y βn
1 Y β2
Cα,β,k,X αY βZ k⊗Bα′,β′,ℓ,X α′
n+2...X α′
m = X α′
n+1X α′
and X α′
1
2
m Y β′
m Z ℓ, with X α = X α1
1 X α2
2 ...X αn
n ,
m
n+m, Y β ′
m = Y β′
n+1Y β ′
n+2...Y β′
n+m.
m
1
2
X α′
m Y β ′
n
Assume k > 0.
Then X αY βZ k ⊗ X α′
m Y β ′
By induction,
Cα,β,k,X αY βZ k ⊗ Bα′,β′,ℓ,X α′
Then
m Y β ′
Bα′,β ′,ℓ,X α′
m Z ℓ(Z ⊗ 1 − 1 ⊗ Z) + X αY βZ k−1 ⊗ X α′
m Z ℓ = X αY βZ k−1 ⊗ X α′
m Y β ′
m Y β ′
m Z ℓ+1.
m Z ℓ(Z ⊗ 1) = X αY βZ k−1 ⊗
m Z k+ℓ and g(X, Y, Z) an expression in X, Y , Z..
m Y β ′
m Z ℓ=g(X, Y, Z)(Z ⊗ 1− 1⊗ Z)+Cα,β,k,X αY β ⊗
r=k+ℓ
r Xα+β+α′+β′=t−r
X(
n Y β ′
Y β1
1 Y β2
2 ...Y βn
It follows:
Xα+β+α′ +β′ =t−r
r=k+ℓ
Therefore:
1
n+1Y β ′
n+2...Y β ′
2
m
n+m)Z r = 0.
Cα,β,kBα′,β ′,ℓX αY β ⊗ X α′
m Y β ′
m = 0.
SKEW GROUP ALGEBRAS AND INVARIANTS
5
Xα+β+k+,ℓ+α′+β′=t
Z)+X(
r Xα+β+α′ +β′ =t−r
Cα,β,k,X αY βZ k ⊗ Bα′,β ′,ℓ,X α′
m Y β ′
Cα,β,kBα′,β ′,ℓX αY βZ k ⊗ X α′
m Z ℓ = G(X, Y, Z) (Z ⊗ 1 − 1 ⊗
m Y β′
m )(1 ⊗ Z r).
r=k+ℓ
Applying ϕ to the above expression we get:
Cα,β,kBα′,β′,ℓX α1
ϕ(b) = 0 = X(
n+2...X α′
n+1X α′
X α′
Using the fact Xn+jYi = YiXn+j for 1 6 i 6 m we obtain:
r Xα+β+α′+β′=t−r
n+mY β ′
n+2...Y β ′
2 ...X αn
n+m)Z r.
n+1Y β′
1 X α2
r=k+ℓ
m
m
1
1
2
2
n Y β1
1 Y β2
2 ...Y βn
n
Cα,β,kBα′,β′,ℓX α1
1 X α2
2 ...X αn
n X α′
n+1X α′
n+2...X α′
1
2
n+m
m
Xα+β+k+,ℓ+α′+β′=t
is an element of (Z ⊗ 1 − 1 ⊗ Z)Bn ⊗K Bm , as claimed.
Cα,β,k,X αY βZ k⊗Bα′,β ′,ℓ,X α′
m Y β ′
m Z ℓ = G(X, Y, Z) (Z⊗1−1⊗Z)
(cid:3)
Let Λ, Γ be K-algebras and G, H finite group of automorphisms of Λ, Γ, respec-
tively. Assume the characteristic of K does not divide neither the order of G nor
the order of H. By the universal property of the coproduct, given σ ∈ G, τ ∈ H,
there is a commutative diagram:
Λ
σ ↓
Λ
j1→ Λ ⊗K Γ
(σ, τ ) ↓
j1→ Λ ⊗K Γ
j2←− Γ
↓ τ
j2←− Γ
with (σ, τ )(λ⊗ γ) = σ(λ)⊗ τ (γ), the homomorphism (σ, τ ) is an automorphism,
such that (σ, τ ) = 1 implies σ = 1 and τ = 1.Hence G × H embeds faithfully in the
automorphism group of Λ ⊗K Γ.
The following proposition is well known: (see for example [CR]).
Proposition 2. Let K be a field and G, H, finite groups. Then there is a natural
isomorphism of finite dimensional algebras: K(G × H) ∼= KG ⊗K K(H).
Proof. We define a map ϕ : K(G×H) → KG⊗KK(H) as ϕ(g, h) = g⊗h and extend
it linearly. The map sends basis to basis, hence is a vector space isomorphism. It
is easy to see it is also a ring isomorphism.
(cid:3)
We use this result to prove the following:
Theorem 3. Let Λ, Γ be graded K-algebras and G, H finite group of (grade pre-
serving) automorphisms of Λ, Γ, respectively. Assume the characteristic of K does
not divide neither the order of G nor the order of H. Denote by Λ∗ G and Γ∗ H the
corresponding skew group algebras. Then there is a natural isomorphism of (graded)
K-algebras: Λ ⊗K Γ ∗ (G × H) ∼= Λ ∗ G ⊗K Γ ∗ H.
6
ROBERTO MART´INEZ-VILLA AND JERONIMO MONDRAG ´ON
Proof. We have vector space isomorphisms: Λ⊗K Γ∗ (G× H) ∼= Λ⊗K Γ⊗K K(G×
H) ∼= Λ ⊗K Γ ⊗K KG ⊗K K(H) ∼= Λ ⊗K KG ⊗K Γ ⊗K K(H) ∼= Λ ∗ G ⊗K Γ ∗ H.
We look to the action of G × H on Λ ⊗K Γ.
Let (σ, τ ) be an element of G × H and λ ⊗ γ ∈ Λ ⊗K Γ.Then (σ, τ )λ ⊗ γ =
λσσ ⊗ γτ τ = λσ ⊗ γτ (σ, τ ) = λσ ⊗ γτ ⊗ σ ⊗ τ = λσ ⊗ σ ⊗ γτ ⊗ τ = (σ ⊗ τ )(λ ⊗ γ).
It follows Λ ⊗K Γ ∗ (G × H) ∼= Λ ∗ G ⊗K Γ ∗ H as (graded) algebras.
(cid:3)
Corollary 2. Let K- be a field of zero characteristic, Bn, Bm homogenized Weyl
algebras and G, H, finite groups of grade preserving automorphisms of Bn and
Bm,respectively. Assume for all σ ∈ G and τ ∈ H, σ(Z) = Z and τ (Z) = Z. Then
G× H acts on Bn+m and there is an isomorphism of graded algebras: Bn+m ∗ (G×
H) ∼= Bn ∗ G ⊗K Bm ∗ H/(Z ⊗ 1 − 1 ⊗ Z)Bn ∗ G ⊗K Bm ∗ H.
n+m , BG
m the rings of invariants, then we have an iso-
morphism of algebras: BG×H
m /Z ⊗ 1- 1 ⊗ Z)BG
n ⊗K BH
Proof. Given σ ∈ G, τ ∈ H, we have a commutative diagram:
n , BH
n+m ∼= BG
If we denote by BG×H
n ⊗K BH
m.
Bn
σ ↓
Bn
j1→ Bn ⊗K Bm
σ ⊗ τ ↓
j1→ Bn ⊗K Bm
ϕ1 ց
↓ ϕ
Bn+m
j2←− Bm
↓ τ
j2←− Bm
ւ ϕ2
Let
n
Pi=1
bi ⊗ b′
i be an element of the kernel of ϕσ ⊗ τ . Then ϕ(
n
bi ⊗ τ b′
i = g(X, Y, Z)(Z ⊗ 1-1 ⊗ Z).
Pσ
i=1
i) = 0.
n
Pσ
i=1
bi ⊗ τ b′
By that above description of Kerϕ,
n
Therefore:
bi ⊗ b′
i =((σ−1, τ −1)(g(X, Y, Z))(σ−1Z ⊗ 1-1⊗ τ −1Z)=g′(X, Y, Z)(Z ⊗ 1-1⊗ Z).
Pi=1
It follows Kerϕσ⊗τ =(Z⊗1-1⊗Z)Bn⊗K Bm and the map ϕσ⊗τ factors through
Bn+m,denote by σ ⊗ τ the induced map, which is clearly an automorphism of Bn+m
such that the diagram:
Bn
σ ↓
Bn
ϕ1→ Bn+m
σ ⊗ τ ↓
ϕ1→ Bn+m
ϕ2←− Bm
↓ τ
ϕ2←− Bm
The above diagram *) induces a commutative diagram of graded algebras and
homomorphisms:
Bn ∗ G
j1→ Bn ⊗K Bm(G × H)
ϕ1 ց
Bn+m ∗ (G × H)
↓ ϕ
j2←− Bm ∗ H
ւ ϕ2
There is an exact sequence:
0→ (Z⊗1- 1⊗Z)Bn⊗K Bm(G×H) → Bn⊗K Bm(G×H) → Bn+m∗(G×H) →0
From the isomorphism (Bn ⊗K Bm) ∗ (G × H) ∼= Bn ∗ G ⊗K Bm ∗ H, it follows
Bn+m ∗ (G × H) ∼= Bn ∗ G ⊗K Bm ∗ H/(Z ⊗ 1-1 ⊗ Z)Bn ∗ G ⊗K Bm ∗ H.
Now let e = 1/ GP g
h be elements of KG and KH,
respectively. The elements e and f are idempotents, it is well known and easy
to prove that the rings BG
m are isomorphic e(Bn ∗ G)e and f (Bm ∗ H)f ,
respectively.
, and f = 1/ H Ph∈H
n and BH
g∈G
SKEW GROUP ALGEBRAS AND INVARIANTS
7
get an isomorphism:
G)e ⊗K f (Bm ∗ H)f.
(g, h) correspond to e ⊗ f.
Under the isomorphism K(G × H) → KG ⊗K KH the idempotent (e, f ) =
1/ G × H P(g,h)∈G×H
Then from the isomorphism: (Bn ⊗K Bm) ∗ (G × H) ∼= Bn ∗ G ⊗K Bm ∗ H we
(e, f )(Bn ⊗K Bm) ∗ (G × H)(e, f ) ∼= e ⊗ f (Bn ∗ G ⊗K Bm ∗ H)e ⊗ f ∼= e(Bn ∗
Therefore (Bn ⊗K Bm)G×H ∼= BG
It follows (e, f )(Bn+m)∗ (G× H)(e, f ) ∼= BG×H
From this we have the isomorphism of algebras:
n+m ∼= BG
BG×H
n ⊗K BH
m.
For Weyl algebras we have the following analogous of the previous theorem.
n ⊗K BH
m.
n+m ∼= (e⊗ f )(Bn∗ G⊗K Bm∗ H)(e⊗
f )/(Z ⊗ 1 − 1 ⊗ Z) ∼= e(Bn ∗ G)e ⊗K f (Bm ∗ H)f /(Z ⊗ 1 − 1 ⊗ Z).
m /(Z ⊗ 1- 1 ⊗ Z)BG
n ⊗K BH
(cid:3)
Theorem 4. Given finite groups of automorphisms G, H of the Weyl algebras
An and Am, G × H acts as a group of automorphisms of An+m and there is an
isomorphism of K-algebras, AG×H
m, where AG×H
n+m , AG
n , AH
m.
n+m ∼= AG
n ⊗K AH
3. The Weyl algebras B1, and A1 and their skew group algebras
B1 ∗ G and A1 ∗ G, with G a finite subgroup of Sl(2, C).
In this section we describe the basic algebras Morita equivalent to B1 ∗ G and
A1 ∗ G by quivers and relations. To achieve this we will make use of the following
result, [AR], [L], [C-B]:
Theorem 5. For any subgroup G of Sl(2, C) the skew group algebra C[X, Y ] ∗ G
is Morita equivalent to the preprojective algebra of an Euclidean diagram.
of finite subgroups Gi of Sl(2, C) acting on Bn and on An.
We will end the section sketching the situation for a product G = G1×G2×...Gn
We start by recalling the situation of C [X, Y ] ∗ G, following the approach of
Recall the construction of a McKay quiver of a finite subgroup G of the linear
[GuM] and take the opportunity to correct some inaccuracies.
group Gl(n, C). [ Mc]
Let S1, S2, ...Sn be the non isomorphic irreducible representations of G and M
the representation corresponding to the inclusion of G in Gl(n, C). Tensoring M
with some Sj we obtain a decomposition in irreducible representations: M ⊗K Sj ∼=
aij Si.
⊕i
The McKay quiver of G has vertices v1, v2,... vn, with each vertex vi corre-
sponding to an irreducible representation Si and we put aij arrows from vi to vj.
For the proof we will make use of the following well known result from ordinary
group representations:
Theorem 6. [Mu] Let G be a finite group and L a complex irreducible representa-
tion. Then the dimension of L as C-vector space divides the order of G.
We reproduce here the proof given in [Ste] of the following:
Theorem 7. Let M be a C-vector space of dimension 2 and G a finite subgroup
of the special linear group Sl(2, M ).Then the McKay quiver has no loops.
8
ROBERTO MART´INEZ-VILLA AND JERONIMO MONDRAG ´ON
Proof. For the proof of the theorem we have two cases:
0
λ2
0
0
µ λ′ (cid:21)(cid:20) λ
0
phism.
µ = 0 and g is in the center of G.
µ λ′ (cid:21), with λλ′ = 1 , and (cid:20) λ
1) M is an irreducible representation.
According to the previous theorem, 2 divides the order of G and by Sylow
theorems, there exists an element g ∈ G of order 2. Since g ∈ sl(2, M ), by a
µ λ′ (cid:21) =
change of basis P gP −1 = (cid:20) λ
λµ + λ′µ λ′2 (cid:21) = 1
(cid:20)
Therefore: λ2 = 1 = (λ′)2 and either λ = λ′ = 1 or λ = λ? = −1, in any case
Let S be another irreducible representation of G and ϕ : CG → S an epimor-
Let´s suppose g − 1 is not in the kernel of ϕ. Then there exists s′ ∈ S with
(g− 1)s′ 6= 0. Since S is simple, for any s ∈ S, s 6= 0, there exists ρ ∈ CG such that
ρ(g − 1)s′ = s and ρ(g − 1) = (g − 1)ρ. It follows (g + 1)s = (g + 1)(g − 1)ρs′ =
(g2 − 1)ρs′ = 0 and gs = −s.
This means that g acts on S either as the identity or as −1.
If g acts as −1 on M , and as 1 on S, then g acts as -1 on M ⊗K S and S can
not appear as a summand of M ⊗K S and if g acts as −1 on M and as −1 on S,
then it acts as 1 on M ⊗K S and again S can not be a summand of M ⊗K S.
We have proved that in this case the McKay quiver has no loops.
2) The representation M is reducible, this is: M = M1 ⊕ M2 and dimC M1 =
dimC M2 = 1. Let´s say that M1 is generated by m1and M2 is generated by m2.
If the order of G is n, then for any g ∈ G, g(m1) = λ1m1 and g(m2) = λ2m2 with
λ1and λ2 n-th roots of unity.
The element g has form: g = (cid:20) λ1
unity, n√1 given by g → λ1, hence G is cyclic and all irreducible representations
There exists an injective homomorphism from G to the group of nth roots of
λ2 (cid:21), with λ1. λ2=1.
0
0
have dimension one.
Let Cs = S be an irreducible representation G =< g > and gs = ts.
Suppose M ⊗K S = S1 ⊕ S2. Then M1 ⊗ S = S1 and M2 ⊗ S = S2.
Assume S = S1, then m1 ⊗ s = s′, s′ ∈ S and s′ = rs. It follows g(m1 ⊗ s) =
g(m1) ⊗ g(s) = λ1m1 ⊗ ts = g(rs) = rts = tλ1(m1 ⊗ s) = tλ1rs.Therefore λ1 = 1
and g is the identity.
We have proved the McKay quiver has no loops.
(cid:3)
We sketch here the arguments used in [GuM] to prove the theorem.
Let G be a finite subgroup of Sl(2, C), which extends to a group of grade pre-
serving automorphisms of C[X, Y ]. The algebra C[X, Y ] is Koszul with Yoneda
algebra the exterior algebra Λ = C < X, Y > /{X 2, Y 2, XY + Y X}, it was proved
in [MV1] that G acts in a natural way as an automorphism group of the Yoneda
algebra, hence G is a sub group of the automorphisms group of Λ.
It was also proved in [MV1], that the skew group algebra C[X, Y ] ∗ G is Koszul
with Yoneda algebra Λ ∗ G.
The algebra Λ is selfinjective of radical cube zero. It follows from [RR], Λ ∗ G
is selfinjective of radical cube zero. Hence C[X, Y ] ∗ G is Artin-Schelter regular of
global dimension two [Sm],[MV4]. It is also clear that C[X, Y ] ∗ G is noetherian.
SKEW GROUP ALGEBRAS AND INVARIANTS
9
Consider now an arbitrary Koszul C-algebra Λ and G a finite group of automor-
phisms of Λ. Given a complete set of primitive orthogonal idempotents e1, e2,... en
of CG, they are also a complete set of orthogonal idempotents of Λ ∗ G and taking
e =
ei, the algebra e(Λ ∗ G)e is basic and Morita equivalent to Λ ∗ G.
n
Pi=1
There is a natural isomorphism:
Extk
Λ∗G(CGe, CGe) ∼= eExtk
Λ∗G(CG, CG)e
Ge,−) induces an equivalence of categories GrΛ∗G ∼= GreΛ∗Ge .
By the Morita theorems applied to graded algebras [MV4], the functor HomΛ∗G(Λ∗
It follows that for each k ≥ 0, there is an isomorphism:
Extk
Λ∗G(CGe, CGe) ∼= Extk
eΛ∗Ge(eCGe, eCGe).
Using this two isomorphisms and adding up, we get an isomorphism of graded
K-algebras.
e( ⊕k≥0
Extk
Λ∗G(CG, CG))e ∼= ⊕k≥0
Extk
eΛ∗Ge(eCGe, eCGe)
In particular when Λ is the exterior algebra in two generators we have:
C[X, Y ] ∗ G ∼= ⊕k≥0
e(C[X, Y ] ∗ G)e ∼= ⊕k≥0
Extk
Λ∗G(CG, CG) and
Extk
eΛ∗Ge(eCGe, eCGe).
Yoneda algebra the basic noetherian algebra e(C[X, Y ] ∗ G)e.
diagram Q.
The algebra e(Λ ∗ G)e is basic Koszul selfinjective of radical cube zero with
It follows from [GMT] that the separated quiver of e(Λ ∗ G)e is an Euclidean
It is easy to check that the quiver of e(C[X, Y ] ∗ G)e is the McKay quiver of
G, by Theorem ?, this quiver does not have loops. By the properties of Koszul
algebras e(C[X, Y ] ∗ G)e and e(Λ ∗ G)e have the same quiver
∧
Q is a translation quiver with translation τ the Nakayama
permutation, but soc(Λ ∗ G)ei = J 2 ∗ Gei , and since Λ has simple socle J 2 = K.
Then soc(Λ ∗ G)ei = KGei = top(Λ ∗ G)ei and τ is the identity.
We know by [GM T ]
∧
Q .
∧
Q is the complete quiver of an Euclidean diagram, this means
∧
Q0 =
1 , with Q an Euclidean diagram. For each arrow α : i → j in
Q1 = Qi ∪ Qop
∧
The quiver
Q0 and
j
∧
Q we have an arrow α−1 : j → i in
∧
Q :
i
α ր
β ց
k
ց α−1
ր β−1
i
.
Since (Λ ∗ G)ei has simple socle, for any pair of arrows α : i →, β : i → k, there
is a non zero c ∈ K, such that α−1α = cβ−1β. If we assume Q is a tree then we
can change the maps b : (Λ ∗ G)ei → (Λ ∗ G)ej corresponding to the arrow β by
cb and we get an arrow which we denote again by β such that α−1α = β−1β and
we obtain an isomorphism e(Λ ∗ G)e ∼= K
relations: α−1α − β−1β , αα−1 − ββ−1 and αδ if δ 6= α−1, δα if δ 6= α−1.
∧
Q/I, where I is the ideal generated by
10
ROBERTO MART´INEZ-VILLA AND JERONIMO MONDRAG ´ON
From this is clear that e(C[X, Y ]∗G)e ∼= K
ց α−1
ր α−1
α1 ր
αk ց
i → j2
jk
→ i
j1
k
by mesh relations:
∧
Q/I ⊥, where I ⊥ is the ideal generated
1
, this is
k
Pi=1
α−1
i αi ∈ I ⊥.
We have proved e(C[X, Y ] ∗ G)e is isomorphic to the preprojective algebra.
The case
∼
Anis the skew group algebra corresponding to the cyclic group Zn and
∼
has to be considered separately, since for
Anit is not clear that we could choose the
arrows in the algebra e(Λ ∗ G)e in such a way that for any pair of arrows α, β with
the same origin, α−1α = β−1β.
We also need to prove that all preprojective algebras appear in this way.
A full description of the quiver of a preprojective algebra and its relations with
the McKay graph for finite subgroups of Sl(2, C) has appeared in several papers
by authors like: Crawley-Boevey or Lenzing [L], [C-B].
n
We now consider finite groups of automorphisms G of the homogenized Weyl
algebra Bn such that for all σ ∈ G, σ(Z) = Z and for all 1 ≤ i ≤ n, σ(Xi),
σ(Yi) ∈
CXi ⊕
Given a Koszul algebra Λ with Yoneda algebra Γ and a finite group of auto-
morphisms G of Λ, we recall from [MV1] , how the action transfers to a group of
automorphisms of Γ:
⊕i=1
⊕i=1
CYi.
n
Let M be a Λ-module and σ ∈ G, we define M σ as the module with M σ = M
as vector space and multiplication given as follows: for λ ∈ Λ, m ∈ M σwe define
λ ∗ m = λσm.
In case M is a G-module, there is an isomorphism: ϕσ : M → M σgiven by
ϕσ(m) = σm.Then ϕσ(λm) = σ(λm) = λσσm = λ ∗ ϕσ(m).
Now given an extension δ ∈ Extk
Λ(Si, Sj), with Si, Sj graded simple modules:
δ : 0 → Sj → E1 → E2 → ...Ek → Si → 0
we define
σ(δ) : 0 → Sσ
j → Eσ
1 → Eσ
2 → ...Eσ
k → Sσ
i → 0.
It is clear that the modules Sσ
In this way we have an isomorphism of C-algebras: σ : ⊕k≥0
Extk
Λ(Λ/J, Λ/J) given by σ(δ1, δ2, ...δn) = (σ(δ1), σ(δ2), ...σ(δn)).
i are again graded simple and σ(δ)∈Extk
i ,Sσ
j ).
Λ(Λ/J, Λ/J) →
Λ(Sσ
Extk
j ,Sσ
⊕k≥0
n we want to see that for any group
of automorphisms G of Bn such that any element σof G has matrix form σ =
In the particular case Λ = Bn and Γ = B!
(cid:20) A 0
every σ ∈ G has matrix form σ = (cid:20) ∗ 0
1 (cid:21), with A a n − 1 × n − 1 matrix, the action of G on (cid:0)B!
0 1 (cid:21), with ∗ a n − 1 × n − 1 matrix.
n(cid:1)1 is such that
0
The element Z of B!
n corresponds to the extension:
SKEW GROUP ALGEBRAS AND INVARIANTS
11
→ C → 0
0 → J →
0 → J/J 2 →
→ C → 0
0 → CZ → C[Z]/(Z 2) → C → 0
0 → C → C[Z]/(Z 2) → C → 0
Bn
↓
E
↓
↓
↓ 1
↓ 1
↓ 1
↓
↓
↓∼=
n
n
The extension δ = Z : 0 → C z→ CZ/(Z 2) → C → 0 is such that both C and
CZ/(Z 2) are G-modules with trivial action. Then for any σ ∈ G, (CZ/(Z 2))σ =
CZ/(Z 2) and Cσ = C . Moreover σ(δ) = δ . We have proved that in B!
n, σ(Z) = Z
for all σ ∈ G.
Bn (C, C) ∼= HomBn (J, C) ∼= HomBn (J/J 2, C),
We have an isomorphism: Ext1
where J/J 2 ∼=
⊕i=1
CY i ⊕ CZ.
CX i ⊕
The element Xi of B!
⊕i=1
∼=→ K
n corresponds to the map: f : J → J/J 2 → CX i
∼=→ C.
applying the homomorphism σwe get a map: f σ : J σ → (J/J 2)σ → (CX i)σ
But by hypothesis (X i)σ = P AijXj + P BijYj ,hence in the expansion of f σ
does not appear Z.
For Yi ∈ B!
σ = (cid:20) [∗] 0
n the situation is similar and the automorphism σ of (B!
1 (cid:21), with [∗] a n − 1 × n − 1 matrix.
n)1has form
0
We want to use the previous remarks to describe B!
of Sl(2, C) with action on B1 as above.
1 ∗ G for G a finite subgroup
k
ei,
1 ∗ Ge, with C!
Let e1, e2,... ek be a complete set of orthogonal idempotents of CG and e =
Pi=1
B1/ZB1 ∼= C1, the polynomial algebra in two variables.
We saw in previous section eC1 ∗ Ge is the preprojective algebra and its Yoneda
algebra is the selfinjective radical cube zero algebra eC!
1 the exterior
algebra in two variables. and for any pair of arrows α : i → j , β : i → k in the
1 ∗ Ge there exists arrows α−1 : j → i and β−1 : k → i such that
quiver of eC!
α−1α = β−1β and δα = 0 if δ is an arrow different from α−1.
n ∼=
We also saw in Section 1, that there is an isomorphism of C!
n ⊕ ZC!
C!
n.
If we assume G acts on Bn in such a way that for σ ∈ G, σ(Z) = Z and
σ(Xi), σ(Yi)is contained in the vector space generated by the X´s and the Y´s, then
Gacts on the same way on B!
n ∗ G ⊕
ZC!
n ∗ G-bimodules.
n ∗ G)e of
1∗G)e.
the Jacobson
1 ∗ G
, hence
n ∗ G of C!
which induces an isomorphism: e(B!
n∗G)e-bimodules. In particular, for n = 1, e(B!
The Jacobson radical of B!
1 is of the form: J = JC 1
1. It follows the Jacobson radical of B!
1∗G)e ∼= e(C!
1 ⊕ZC!
1, with JC 1
1 ∗ G is J ∗ G = JC 1
1∗G)e⊕Ze(C!
1 ∗ G⊕ ZC!
n and we have an isomorphism: B!
n ∗ G)e ∼= e(C!
n ∗ G)e ⊕ Ze(C!
n ∗ G ∼= C!
n-bimodules: B!
e(C!
1 ∗ G)+ Z 2C!
1 ∗ G . But Z 2 = −XY ∈ J 2
C 1
1
1
radical of C!
and (J ∗ G)2 = J 2
(J ∗ G)2 = J 2
C 1
C 1
1 ∗ G + Z(JC 1
1 ∗ G).
1 ∗ G + Z(JC 1
Therefore: J ∗ G/J 2 ∗ G ∼= J/J 2 ∗ G ∼= JC 1
/J 2
1 ∗ G⊕ Z(C!
C 1
G ⊕ Z(CG). Then e( J ∗ G/J 2 ∗ G)e ∼= e/JC 1
/J 2
C 1
1 ≤ i ≤ n, ei( J ∗ G/J 2 ∗ G)ej ∼= ei(JC 1
1 ∗ G)ej ⊕ Zei(CG)ej .
1 ∗ G) ∼= JC 1
1 ∗
1 ∗ G)e ⊕ Ze(CG)e. Hence for
/J 2
C 1
/J 2
C 1
1JC 1
1
1
1
1
12
ROBERTO MART´INEZ-VILLA AND JERONIMO MONDRAG ´ON
Since the preprojective algebra has no loops, ei(JC 1
1
/J 2
C 1
1 ∗ G) ei = 0. In the
We have proved that the quiver of e(B!
other hand, for i 6= j, ei(CG)ej = 0 and ei( J ∗ G/J 2 ∗ G)ei = ZCei.
1 ∗ G)e has the same vertices and arrows
as a preprojective algebra corresponding to an Euclidean diagram, and in addition
a loop for each vertex.
We shall next find the relations.
Given an idempotent e′
and an isomorphism ϕ : CGe′
ejγe′
j /∈ {e1, e2,... ek}, there exists one of the idempotents ej
j) = e′
jρej and ϕ−1(ej) =
jρejγe′
j.
j → CGej, in particular ϕ(e′
j = e′
, and G acts trivially on XY .
j with ρ, γ ∈ CG . Then ej = ejγe′
Note that CXY = J 2
C 1
1
Let g ∈ G, g(X) = aX + bY , g(Y ) = cX + dY.Hence, g(XY ) = g(X)g(Y ) =
(aX + bY )(cX + dY ) = caX 2 + adXY + bcY X + bdY 2 = adX − bcY = XY , since
det(g) = 1.In consequence, XY ei = eiXY = −eiZ 2.
jρej and e′
n
It follows XY ei = eiXY ei =
eiXe′
jY ei.
Consider paths of the following form:
Pi=1
α
i
k
j
ր
ց
ց
ր
α−1
i
eiXe′
jρej
ejγe′
iY ei
jρejγe′
Where eiXe′
Since each indecomposable projective of e(B!
jY ei = eiXe′
iY ei.
1∗G)e
is a radical cube zero algebra, there are constants cj ∈ C such that cjα−1α =
eiXe′
1∗G)e has simple socle and e(B!
jY ei = eiXe′
Therefore: eiXY = (
cj)α−1α = tiα−1α = −eiZ 2, with ti ∈ C − {0}. This is
iY ei.
n
jρejγe′
Pj=1
tiα−1α + Z 2
i = 0, with Zi = Zei.
In a similar way we obtain relations tkαα−1 + Z 2
k = 0, α−1α and αα−1are paths
in e(C!
1 ∗ G)e .
The element Z of B!
1 anticommutes with X, Y and commutes with all the ele-
ments g of G. Then Z anticommutes with α and α−1. This is Zkα = −αZi.
i , then αα−1α + (Z 2
αα−1α + αZ 2
It follows: 0 = tkαα−1α + Z 2
kα = tiαα−1α + αZ 2
i /ti.
k/tk)α =
(Z 2
an arrow. By connectivity, tk = ti for all k, i.
We have the following equalities: αZ 2
i /ti = (αZ)Zi/ti = −ZαZ/ti = (Z 2α)/ti =
k/tk)α = (Z 2α)/tk. Therefore tk = ti for ani pair of vertices i, k connected by
Assume every element of K has a square root and let c = 2√tk, we can make a
change of variables, zk = Z/c to get relations: {zkα + αzi, for every arrow α, and
αα−1 + z2
i , α−1α + z2
We have proved the algebra e(B!
R , with
∧
Q 0 = Q0 and
∧
∧
1 ∗ G)e is isomorphic to the quiver algebra K
Q /
1 ∪{zi i ∈ Q0} and Q an Euclidean diagram.
∧
Q / R⊥ , with
1 ∪ {zi i ∈ Q0} and Q an Euclidean diagram and R⊥
∧
∧
Q 0 = Q0 and
Using Koszulity, e(B1 ∗ G)e is isomorphic to the quiver algebra K
i , P α−1
the orthogonal relations: R⊥ = {zkα − αzi, P αjα−1
Q 1 = Qi ∪ Qop
j − z2
Q 1 = Qi∪ Qop
j αj − z2
k}.
k} = R
SKEW GROUP ALGEBRAS AND INVARIANTS
13
For the first Weyl algebra we have: e(A1 ∗ G)e = e(B1 ∗ G)e/(z − 1)e(B1 ∗ G)e.
∧
It follows A1∗G is Morita equivalent to an algebra with quiver
Q 0 =
j − ek,
P α−1
1 and Q an Euclidean diagram and relations {P αjα−1
j αj − ei}. This means, e(A1 ∗ G)e is the deformed preprojective algebra.
We state the previous results as a theorem:
Q 1 = Qi ∪ Qop
∧
Q, such that
Q0 and
∧
Theorem 8. Let G be a finite subgroup of Sl(2, K),with K a field of zero charac-
teristic and containing the square root of each element. Let A1 be the first Weyl
algebra and B1 its homogenized algebra. The group G acts as a grade preserving
automorphism group of B1, fixing Z and sending X and Y to a linear combination
of X and Y . Then G acts in the same way on B!
1, the Yoneda algebra of B1, and
G acts as a group of automorphisms of A1. The skew group algebras B1 ∗ G, B!
1 ∗ G
and A1 ∗ G are Morita equivalent to the algebras defined by quivers and relations
as follows:
i , α−1α + z2
∧
Q 0 = Q0 and
∧
i) The skew group algebra B!
∧
1 ∗ G is Morita equivalent to the quiver algebra K
Q
1 ∪ {zi vi ∈ Q0} and Q an Euclidean
Q 1 = Qi ∪ Qop
/ R, with
diagram and R = {zkα + αzi, for every arrow α, and αα−1 + z2
∧
Q /
1 ∪{zi vi ∈ Q0} and Q an Euclidean diagram
j αj − z2
k}.
∧
Q/I,
k} .
ii) The skew group algebra B1∗G is Morita equivalent to the quiver algebra K
and R⊥ the orthogonal relations: R⊥ = {zkα− αzi, P αjα−1
iii) The skew group algebra A1 ∗ G is Morita equivalent to the algebra K
I = {P αjα−1
We want to sketch the situation for groups of the form G = G1×G2×...×Gn, such
that Gi is a finite subgroup of Sl(2, C) acting as subgroups of the automorphism
group of the homogenized Weyl algebra Bn.
1 and Q an Euclidean diagram and relations
Q 1 = Qi ∪ Qop
j αj − ei}.
j − ek, P α−1
j − z2
i , P α−1
Q 1 = Qi∪Qop
∧
Q 0 = Q0 and
∧
R⊥ , with
∧
Q 0 = Q0 and
∧
such that
We will recall first the description by quivers and relations of the tensor product
of two quiver algebras.
Let K be a field of zero characteristic and let Λ = KQ1/I1 and Γ = KQ2/I2 be
two graded quiver algebras, Q the quiver of Λ ⊗K Γ and I an admissible ideal of
KQ such that KQ/I ∼= Λ ⊗K Γ. We next describe Q and I.
Let { e1, e2,... en} and { f1, f2,... fm} be complete sets of orthogonal primitive
idempotents of Λ and Γ, respectively. Then {ei ⊗ fj 1 ≤ i ≤ n, 1 ≤ j ≤ m} is a
complete set of primitive orthogonal idempotents of Λ ⊗K Γ. This means that the
quiver Q has vertices Q0 = (Q1)0 × (Q2)0, where the pair (vi, wj) corresponds to
the idempotent ei ⊗ fj and vi is the vertex corresponding to ei and wj the vertex
corresponding to fj.
The arrows αk : vi → vj in (Q1)1 correspond to elements ak ∈ ejr∆ei − ejr2
∆ei
and the arrows βk : wi → wj of (Q2)1 correspond to elements bk ∈ fjrΓfi − fjr2
Γfi.
We have in Q arrows Q1 = (Q1)1 × (Q2)0 ∪ (Q1)0 ∪ (Q2)1. This is to an arrow
α : vi → vj in (Q1)1 and a vertex wk of (Q2)0 corresponds an arrow (α, wk) :
(vi, wk) → (vj , wk) of Q1 associated to a ⊗ fk ∈ ei ⊗ fk(rΛ⊗K Γ)ej ⊗ fk − ei ⊗
fk(r2
Λ⊗K Γ)ej ⊗ fk. Similarly, to an arrow β : wi → wj of (Q2)1 and a vertex vk
of (Q1)0 corresponds the arrow (vk, β) : (vk, wi) → (vk, wj) of Q1 associated to
ek ⊗ b ∈ ek ⊗ fi(rΛ⊗K Γ)ek ⊗ fj − ek ⊗ fi(r2
Λ⊗K Γ)ek ⊗ fj.
14
ROBERTO MART´INEZ-VILLA AND JERONIMO MONDRAG ´ON
Given paths α1α2...αr : vi → vj and β1β2...βs : wk → wℓ in Q1 and Q2,
respectively, there are paths (α1α2...αr, w): (vi, w) → (vj , w) and (v, β1β2...βs) :
(v, wk) → (v, wℓ). Now it is clear that given readable relations ρ = P ckγk in I1,
and ρ′ = P bkγ′
k in I2 there are induced readable relations (ρ, w) = P ck(γk, w)
and (v, ρ´) = P bk(v, γ′
If we have two arrows α : v2 → v1 and β : w2 → w1 in Q1 and Q2, respectively,
k) in KQ.
the we have maps:
Λ ⊗K Γe1 ⊗ f1
↓ e1 ⊗ b
Λ ⊗K Γe2 ⊗ f1
a⊗f1→ Λ ⊗K Γe2 ⊗ f1
a⊗f2→ Λ ⊗K Γe2 ⊗ f2
↓ e2 ⊗ b
∧
∧
ℓ
Pi=1
k
Pi=1
ei and f =
acting in the way above described as an automorphism group of B2.
making the diagram commute.
Hence we have in I a commutative relation ζα,β = (v2, β)(α1, w1)−(α1, w2)(v1, β).
Then I is the ideal generated by the relations { {ρ}×(Q2)0 ρ a readable relation
in I1}∪{ (Q1)0 ×{ρ′} ρ′a readable relation in I2}∪{ ζα,β α ∈ (Q1)1, β ∈ (Q2)1}.
We obtained the description of Q and I such that KQ/I ∼= Λ ⊗K Γ.
We return to the case G = G1 × G2 with G1, G2 finite subgroups of Sl(2, C)
Taking complete sets of primitive orthogonal idempotents { e1, e2,... ek} and
{ f1, f2,... fℓ} of KG1 and KG2, respectively, Letting e, f be the idempotents
fi we saw above e(B1∗G1)e and f (B1∗G2)f are preprojective
e =
∧
Q2/I2
1 )1 ∪{zi vi ∈
2 )1 ∪ {zi wi ∈ (Q2)0} with Q1
Then e ⊗ f (B1 ∗ G1 ⊗K B1 ∗ G2)e ⊗ f = e(B1 ∗ G1)e ⊗K f (B1 ∗ G2)f ∼= K
∧
Q2/I2 is the tensor product of two algebras related to preprojective
algebras of Euclidean diagrams, e(B1∗G1)e ∼= K
∧
∧
Q1)0 = (Q1)0, (
Q1is of the form (
the quiver
∧
∧
Q2)1 = (Q2)1 ∪ (Qop
Q2)0 = (Q2)0, (
(Q1)0} and (
and Q2 Euclidean diagrams. and I1, I2 the ideals described above
Q1/Ii and f (B1∗G2)f ∼= K
Q1)1 = (Q1)1 ∪ (Qop
Zei ⊗ fj, hence (Z ⊗ 1 − 1 ⊗ Z) is
ei ⊗ Zfj and Z ⊗ 1 = Pi,j
Q1/Ii ⊗ K
algebras of Euclidean diagrams, as described above.
We have 1 ⊗ Z = Pi,j
the ideal generated by readable relations ei ⊗ Zfj − Zei ⊗ fj.
Therefore in (e, f )(B2∗ (G1× G2))(e, f ) ∼= (e⊗ f (B1∗ G1⊗K B1∗ G2)e⊗ f )/(Z ⊗
1 − 1 ⊗ Z) we are identifying the two loops ei ⊗ Zfj and Zei ⊗ fj.
Then B2 ∗ (G1 × G2) can be completely described by quiver and relations: it has
the same quiver as the tensor product of two preprojective algebras of Euclidean
diagrams and in addition a loop in each vertex, the relations are naturally induced
from those in each preprojective algebra as studied in detail above. We leave to the
reader to write explicitly the quiver and relations as well as the induced relations
in (e, f )A2(G1 × G2)(e, f ) ∼= ((e, f )(B2 ∗ (G1 × G2))(e, f ))/(Z − 1). Observe that
from this could also give a full description of the relations in BG1×G2
and AG1×G2
.
By induction a full description by quivers and relations of the basic algebras
Morita equivalent to Bn ∗ (G1 × G2 × ...Gn) can be given, the algebras come from
iterated tensor products of preprojective algebras of Euclidean diagrams and addi-
tional loops for each vertex and naturally induced relations.
∧
2
2
SKEW GROUP ALGEBRAS AND INVARIANTS
15
4. The algebras C[X] ∗ G and C[X]G, with G a finite subgroup of
Gl(n, C), such that no σ ∈ G, σ 6= 1, has a fixed point.
In this section we study the relations between the categories of finitely generated
modules modC[X]∗G and modC[X]G, where C[X] ∗ G is the skew group algebra and
C[X]G the invariant ring of a finite subgroup of Gl(n, C), such that no σ ∈ G,
σ 6= 1, has a fixed point.
More precisely, we prove that after killing the modules of finite length, the cat-
egories modC[X]∗G and modC[X]G are equivalent. These results generalize known
results for C[X, Y ] and finite subgroups of Sl(2, C), [C-B].
Lemma 1. Assume G is a finite subgroup of Gl(n, C). Then G acts naturally on
the polynomial ring in n variables C[X] and the invariant ring C[X]G is the center
of C[X] ∗ G.
Proof. It is clear C[X]G is contained in the center of C[X]∗ G. Let v ∈ Z(C[X]∗ G)
be an element in the center, v = Pg∈G
For any Xiv = vXi for any Xi. Then Xiv = Pg∈G
If cg 6= 0, then X g
i = Xi . Since Xi is arbitrary, g = 1 and v ∈ C[X].
Now, gv = vgg = vg implies v = vg for all g ∈ G and v ∈ C[X]G.
cgg and cg ∈ C[X].
cgXig = Pg∈G
cggXi = Pg∈G
cgX g
i g.
(cid:3)
It is well known [Be], [Stu], C[X]G is an affine algebra and C[X]G → C[X] an in-
tegral extension,, in particular, dim C[X]G = dim C[X] and the maximal spectrums
max specC[X]G and max specC[X] are isomorphic [Ku].
n
⊕i=1
Moreover, it is known [Be] that given a prime p of C[X]G and primes P, Q of
K[X] above p, this is: P∩C[X]G = Q ∩ C[X]G = p, there exists σ ∈ G such that
P σ = Q and for any σ ∈ G, P σ ∩ C[X]G = P ∩ C[X]G = p.
In particular for m a maximal ideal of C[X]G and maximal ideals n 1, n 2 of C[X]
with n 1 ∩ C[X]G = n 2∩ C[X]G = m, there exists σ ∈ G with n 2,=n σ
1 and for any
σ ∈ G, n σ
1 ∩ C[X]G = n 1∩ C[X]G = m.
1 is a maximal ideal with n σ
The points v of V =
CXi correspond to maximal ideals n v of C[X] .
In
particular, 0 ∈ V corresponds to the irrelevant maximal ideal n0 = (X1, X2, ...Xn)
of C[X].
By hypothesis, for v 6= 0, the orbit of v under G, O (v)= {σ(v) σ ∈ G} has
order O(v) = G .The point v corresponds to a maximal ideal n such that n∩
C[X]G = m and there are G ideals above m and they are precisely {n σ σ ∈ G}.
This is m is an unramified prime.
In the case v = 0, for any σ ∈ G, σ(v) = v and the maximal ideal n0 satisfies
nσ
0 = n0 for all σ ∈ G.
The maximal ideal n 0∩ C[X]G = m0 has a unique maximal ideal, n0, above it.
We consider the two cases separately.
a) The ideal m of C[X]G is maximal and different of m0.
Let n σwith σ ∈ G be the set of all maximal ideals of C[X] above m0, hence
n σ, the radical of mC[X].
mC[X] ⊆ ∩σ∈G
n σ. It follows pmC[X] = ∩σ∈G
It is clear that for any τ ∈ G, ( ∩σ∈G
n σ)τ = ∩σ∈G
n στ = ∩σ∈G
G-invariant.
n σ and pmC[X] is
16
ROBERTO MART´INEZ-VILLA AND JERONIMO MONDRAG ´ON
By the Chinese Remainder Theorem, C[X]/pmC[X]= C[X]/ ∩σ∈G
Since C[X]/n ∼= Cv, Qσ∈G
Cvσ, with 1 = Pσ∈G
C[X]/n σ ∼= Qσ∈G
complete set of primitive orthogonal idempotents.
n σ= Qσ∈G
C[X]/n σ.
vσ, and { vσ}σ∈G is a
C.
cσ,vτ vτ ,then vρ
sσvρσσ =
cστ vσ =
cρτ −1 vρ. It follows
cσvσ = Pσ∈G
vτ = c11.
σ = Pτ ∈G
sσσvρ = Pσ∈G
sσσ. Each sσ has form sσ = Pτ ∈G
vρsσσ = Pσ∈G
cσ,ρvρσ = zvρ = Pσ∈G
cσ,ρσvρσσ. Therefore: cσ,ρσvρσ = cσ,ρvρ for all ρ, σand cσ,ρσ = cσ,ρ = 0 for
For the skew group algebra we have the following isomorphism:
Cvσ)∗G.
cσvσ and cσ ∈ C.Hence, τ z = Pσ∈G
cσvστ = Pσ∈G
C[X]∗ G/pmC[X] ∗ G ∼=(C[X]/pmC[X])∗G ∼=( Qσ∈G
Cvσsemisimple, it follows by [], C[X]∗ G/pmC[X] ∗ G is semismple.
Since Qσ∈G
Cvσ, the group G acts transitively on the basis { vσ}σ∈G of S.
Let S be S = Qσ∈G
We claim S ∗ G is simple, it will be enough to prove that the center of S ∗ G is
It is clear that C is contained in the center. Let z ∈ Z(S∗G) be an element of the
cσ,τ vρvτ =
center, z = Pσ∈G
cσ,ρvρ.
It follows vρz = Pσ∈G
Pσ∈G
σ 6= 1.
It follows z ∈ S, this is z = Pσ∈G
cσvστ = zτ and Pσ∈G
cσvστ τ = Pσ∈G
Pσ∈G
cρτ −1 = cρ for all ρ, τ , in particular, c1 = cτ and z = c1 Pτ ∈G
Since SG ⊆ Z(S ∗ G) we have proved SG = Z(S ∗ G) = C.
The element e = 1/ G Pg∈G
tent of (C[X]/pmC[X])∗G and e((C[X]/pmC[X])∗G) 6= 0.
Being the algebra (C[X]/pmC[X])∗G simple and the ideal:
((C[X]/pmC[X])∗G)e((C[X]/pmC[X])∗G) non zero
(C[X]/pmC[X])∗G =((C[X]/pmC[X])∗G)e((C[X]/pmC[X])∗G).
In this case we have : C[X]∗ G/pm0C[X] ∗ G ∼=(C[X]/n0)∗G ∼= CG.
two sided ideals C[X] ∗ G containing pm0C[X] ∗ G. In particular, C[X] ∗ G/Li ∼=
CG/Li ∼= C and
Let {m i}i∈I be the set of maximal ideals of C[X]G different from m 0, we have
Ψ : C[X]∗ G → Qi∈I
rgg) =
rgg +pmiC[X])∗G), Pg∈G
(( Pg∈G
G)∩ (n0 ∗ G) = ( ∩i∈I
of all maximal ideals of C[X]. The map Ψ is an injective ring homomorphism.
Denote by Σ the product Qi∈I
C[X]/pmiC[X])∗G× C[X]∗ G/n0∗ G, given by Ψ( Pg∈G
( ∩σ∈G
n σ
i )∗
n σ
i )∩ n0) is the intersection
Let L1, L2,... Ls the two sided maximal ideals of CG and L1, L2,... Ls maximal
rgg + n0 ∗ G), which has kernel ker Ψ = ∩i∈I
( ∩σ∈G
n σ
i )∩ n0)∗ G = 0, since ( ∩i∈I
( ∩σ∈G
g is an idempotent of KG, hence a non zero idempo-
Si ∗ G = Σ and Σ = Σ × CG.
Case 2.
s
∩i=1
Li = n0 ∗ G.
a natural homomorphism:
SKEW GROUP ALGEBRAS AND INVARIANTS
17
g, into
∧
e = ((e, e), where e =
The map Ψsends the idempotent e = 1/ G Pg∈G
e +pmiC[X])∗G).
e Σ = (Qi∈I
Si ∗ GeSi ∗ G) × CGeCG = Qi∈I
Then Σ
∧
Si ∗ G × Ce=Σ × Ce.
∧
e Σ ∩ Ψ(C[X] ∗ G) =
(Σ × Ce) ∩ Ψ(C[X] ∗ G).
It is clear that Ψ(C[X] ∗ GeC[X] ∗ G) is contained in Σ
We want to prove they are equal.
Let ((ri), ce) be an element of (Σ × Ce) ∩ Ψ(C[X] ∗ G). This means that there
exists an element r = Pg∈G
rgg ∈ C[X] ∗ G such that r = ri for all i ∈ I and r = ce
in C[X] ∗ Ge/n0 ∗ Ge ∼= CGe.Then Pg∈G
rgg = ce + he with h a polynomial without
constant term. This implies for all g ∈ G, rg = c + h and Pg∈G
rgg = (c + h)e ∈
C[X] ∗ GeC[X] ∗ G. It follows (Σ × Ce) ∩ Ψ(C[X] ∗ G) = Ψ(C[X] ∗ GeC[X] ∗ G).
We have a commutative exact diagram:
0
↓
C[X] ∗ GeC[X] ∗ G
↓
C[X] ∗ G
0 →
0 →
↓
↓
0
0
↓
Si ∗ G × Ce
Si ∗ G × CG
→ Qi∈I
→ Qi∈I
↓
↓
↓
0
0 → C[X] ∗ G/C[X] ∗ GeC[X] ∗ G →
CG(1 − e)
subcategory of M odR. [P]
it is a finite dimensional C-algebra.
It follows C[X]∗G/C[X]∗GeC[X]∗G is a subalgebra of a semisimple algebra, hence
We can identify the category of C[X]∗G/C[X]∗GeC[X]∗G-modules with the cat-
egory of C[X] ∗ G-modules M with eM = 0.
Let´s assume more generally that R is a ring and e an idempotent of R. Then
there is a functor HomR(Re,−) : M odR → M ode Re, which is exact and it has a
left adjoint Re⊗e Re- such that HomR(Re, Re⊗e ReM ) = M.
It follows HomR(Re,−) is a dense functor with kernel M odR/ Re R.
The category A =M odR/ Re R = {M ∈ M odR eM = 0} is a dense (Serre)
Given an exact sequence of R-modules: 0 → L → M → N → 0 applying
the exact functor HomR(Re,−) we obtain an exact sequence of e Re -modules:
0 → eL → eM → eN → 0, hence eM = 0 if and only if eN = eL = 0.
Given a dense subcategory A of M odR we define the multiplicative system of all
maps f : M → N such that the kernel and the cokernel of f are in A.
We define a quotient category M odR/A = (M odR)Σ (see [Ga], [P], [Mi]).
The category (M odR)Σ is abelian and the quotient functor π : M odR → (M odR)Σ
is exact an it has the following universal property: Given an abelian category C
and an exact functor F : M odR → C such that F (X) = 0 for all X ∈ A, there is a
unique exact functor H : (M odR)Σ → C such that Hπ = F .
functors:
M odR
↓ π
M odR/A
→
H→
M ode Re
↓∼=
M ode Re
Proof. 1) The functor H is dense.
Since HomR(Re,−) is dense, it follows H is dense.
2) The functor H is full.
Let M and N be R-modules and Re⊗e ReeM
0 → L → Re⊗e ReeM
→ M → M/ Re M → 0 .
From the exact sequence:
0 → eL → e(Re⊗e ReeM )
eµ
µ
µ
isomorphisms eL = 0 = e(M/ Re M ).
cation, taking the kernel and the cokernel of µ we get an exact sequence:
→ M the map given by multipli-
18
ROBERTO MART´INEZ-VILLA AND JERONIMO MONDRAG ´ON
Theorem 9. Let R be a ring and e an idempotent of R and A the category A
={M ∈ M odR eM = 0}. Then there is a commutative diagram of categories and
HomR(Re,−)
, with H an equivalence.
→ eM → e(M/ Re M ) → 0 and the fact eµ is an
Hence the multiplication maps Re⊗e ReeM
isomorphisms in M odR/A.
µ
→ M and Re⊗e ReeN
µ′
→ N are
Given a map f : eM → eN the "roof"
µ ւ
M
Re⊗e ReeM
ց µ′1 ⊗ f
N
is
a map in M odR/A such that when we apply the functor HomR(Re,−) we obtain
ց HomR(Re, µ′1 ⊗ f )
maps:
,
HomR(Re, Re⊗e ReeM )
HomR(Re, µ) ւ
HomR(Re, M )
HomR(Re, N )
with HomR(Re, µ), HomR(Re, µ′) isomorphisms.
Hence H(µ′1 ⊗ f µ−1) = f.
3) The functor H is faithful.
W
Let
s ւ
M
ց f
N
be a map in M odR/A.Then the kernel and the cok-
ernel of s are in A. Assume H(f s−1) = 0.
Hence, H(f s−1) = HomR(Re, f )HomR(Re, s)−1 = 0 and HomR(Re, s) an iso-
morphism implies HomR(Re, f ) = ef = 0.
0 → eW
ef ↓
0 → eN
j
→ W
f ↓
→ N
j
From the commutative diagram:
we get f j = 0.But
j : eW → W is an isomorphism in M odR/A. Therefore: f = 0 and f s−1 = 0 in
M odR/A.
(cid:3)
We come back now to the case of a finite subgroup G of Gl(n, C) such that
no σ ∈ G different from the identity has a fixed point, G acting as a group of
automorphism of the polynomial ring C[X] in n variables the idempotent e =
1/ G Pg∈G
g of CG is an idempotent of the skew group algebra C[X] ∗ G such that
eC[X] ∗ Ge is isomorphic to the group of invariants C[X]G.The algebra C[X]G
is a sub algebra of C[X] and C[X] is finitely generated as C[X]G-module. Let
C[X]G → C[X] given
f1, f2,... fm be the generators. The epimorphism ρ : ⊕m
SKEW GROUP ALGEBRAS AND INVARIANTS
19
∧
ρ : ⊕m
C[X]Ge → C[X]e of left
γifie . Let eλ = λe = eλe ∈ C[X]Ge.
γifie)λ =
γiλfie = (
m
m
m
Pi=1
Pi=1
Pi=1
by ρ(γ1,γ2,...γm) =
γifi , extends to a map:
C[X]G-modules
∧
ρ(γ1e,γ2e,...γme) =
m
Pi=1
Then
∧
ρ(γ1eλ,γ2eλ,...γmeλ) =
∧
ρ(γ1λe,γ2λe,...γmλe) =
m
∧
∧
(
Pi=1
γifie)eλ and
ρ is a map of C[X]G−eC[X]∗Ge bimodules. If M is a eC[X]∗Ge-
C[X]Ge⊗eC[X]∗Ge
module of finite dimensional over C, from the epimorphism:
M → C[X]e ⊗eC[X]∗GeM and the isomorphism: C[X]Ge ⊗eC[X]∗Ge M ∼= eC[X] ∗
Ge⊗eC[X]∗Ge M ∼= M we obtain that C[X]e ⊗eC[X]∗GeM = C[X]∗ Ge ⊗eC[X]∗GeM
is finite dimensional over C.
It is also clear that if M is a C[X] ∗ G-module finite dimensional, then eM =
HomC[X]∗G( C[X] ∗ Ge, M ) is finite dimensional over C.
To simplify the notation we will write R = C[X]∗G and T = eC[X]∗Ge ∼= C[X]G
and denote by SR and ST the categories of finite dimensional R and T -modules,
respectively.
ρ⊗1 : ⊕m
The categories SR and ST are dense subcategories of the categories of finitely
generated, modR, modT , R and T -modules, respectively.
Then we have:
Theorem 10. Let G be a finite subgroup of Gl(n, C) such that no σ ∈ G different
from the identity has a fixed point, G acting as a group of automorphism of the
polynomial ring C[X] in n variables and e the idempotent e = 1/ G Pg∈G
g of the
skew group algebra C[X] ∗ G.Writing R = C[X] ∗ G and T = eC[X] ∗ Ge ∼= C[X]G
and denoting by SR and ST the categories of finite dimensional R and T -modules,
respectively and by modR, modT , the categories of finitely generated, R and T
-modules, respectively. Then there is a commutative diagram of categories and
functors:
modR
↓ πR
modR /SR
HomR(Re,−)
→
H→
modT
↓ πT
modT /ST
, with H an equivalence.
Before proving the theorem, observe that there is a graded version.
The ring e(C[X] ∗ G)e is a positively graded C-algebra with grading ( e(C[X] ∗
G)e)k = e((C[X] ∗ G)k)e, hence the isomorphism of C-algebras e(C[X] ∗ G)e ∼=
C[X]G induces a positive grading on C[X]G. In what follows we will assume C[X]G
has this grading and denote by R and T the positively graded C-algebras C[X]∗ G
and C[X]G, respectively. If we denote by grR and grT the categories of finitely
generated graded R and T modules, respectively, and degree zero maps. Following
[], [], we call the quotient categories grR/SR and grT / ST the categories of tails
tailsR and tailsT , another notation used in [MV2] is QgrR and QgrT.
Theorem 11. Let G be a finite subgroup of Gl(n, C) such that no σ ∈ G different
from the identity has a fixed point, G acting as a group of automorphism of the
polynomial ring C[X] in n variables and e the idempotent e = 1/ G Pg∈G
g of the
skew group algebra C[X] ∗ G.Writing R = C[X] ∗ G and T = eC[X] ∗ Ge ∼= C[X]G
and considering both as positively graded algebras, denoting by SR and ST the cat-
egories of finite dimensional graded R and T -modules, respectively and by gR, grT ,
20
ROBERTO MART´INEZ-VILLA AND JERONIMO MONDRAG ´ON
the categories of finitely generated, R and T -modules, respectively and by tailsR
and tailsT the quotient categories grR/SR and grT / ST . Then there is a commuta-
tive diagram of categories and functors:
equivalence.
grR
↓ πR
tailsR
HomR(Re,−)
→
H→
grT
↓ πT
tailsT
, with H an
We will prove only the ungraded case, the other follows with the same line of
arguments.
Proof. 1) The functor H is dense.
It follows as above from the fact HomR(Re,−) is dense.
2) The functor H is full.
Let X, Y be objects in modR /SR and consider a map ϕ : H(X) → H(Y ) in
modT /ST , with H(X) = HomR(Re, X), H(Y ) = HomR(Re, Y ) and ϕ a "roof"
W
f
eY
ց
s ւ
eX
length.
, where the kernel L, and the cokernel N , of s are of finite
Applying the tensor functor Re⊗e Re− and composing with multiplication we
obtain the maps:
Re⊗e Res
Re⊗e ReeX
ւ
Re⊗e ReW
ց
Re⊗e ReeY
Re⊗e Ref
ց
µ′
Y
µ
X
ւ
Re ⊗e Rej
We also have exact sequences:
Re⊗e ReL
→ Re⊗e ReW
Re⊗e Re Im s → Re⊗e ReeX → Re⊗e ReN → 0
By the above observation, Re⊗e ReL and Re⊗e ReN are finite dimensional C-
We have a commutative exact diagram:
vector spaces, then U = Im Re⊗e Rej = Ker Re⊗e Res is finite dimensional.
→ Re⊗e Re Im s → 0
Re ⊗e Res
0 → U → Re ⊗ W → Re ⊗ Ims →
0 → V → Re ⊗ W → Re ⊗ eX → Re ⊗ N → 0
Re ⊗ N → 0
→ Re ⊗ N
↓ 1
0
0
↓
↓
↓
Z
↓
0
0
↓
Z
↓
↓
↓
↓
0
0
↓
↓
↓
0
Applying the functor HomR(Re,−) to the diagram we obtain the commutative
exact diagram:
SKEW GROUP ALGEBRAS AND INVARIANTS
21
0
↓
0
↓
eZ
↓
s
↓ 1
0 → eU → W → Ims → 0
↓
↓
↓
0
0 → eV → W → eX → N → 0
0 → N → N → 0
ց ↓ i
↓
↓
0
↓
↓
eZ
↓
0
Since the map i is a monomorphism eZ=0 and by Theorem ? Z is finite dimen-
sional, but both U and Z finite dimensional implies V is finite dimensional.
and both the kernel and cokernel of Re⊗e Res are in SR.
=fs−1.
By the above remark, N of finite dimension implies Re⊗N is of finite dimension
It is clear that H(µ(Re⊗f)(Re ⊗s)−1=HomR Re,µ(Re⊗f))HomR(Re,Re⊗s)−1
3) The functor H is faithful.
Let
s ւ
X
W
g
Y
ց
be a map in modR /SR, where s has kernel L and
cokernel N,both finite dimensional. Assume H(gs−1 = 0.
There exist an exact sequence: 0→ eL→ eW es→ eX→ eN→0 and L, N ∈ SR
implies eL, eN ∈ ST .Then we have in modT /ST the map:
eW
eg
ց
es ւ
eX
which by assumption satisfies (eg)(es)−1 = H(gs−1) = 0.
Then we have a commutative diagram:
eY
,
eW
es ւ ↑
Z
eX
տ ↓
V
t′
eg
r1 ց
eY
r2 ր 0
with esr1 = t′r2 having kernel and cokernel in ST and egr1 = 0.
We want to see first that r1 has kernel and cokernel in ST .
We have a commutative exact diagram:
0
↓
↓
0 → B ′ → Z
1 ↓
0 → B → Z
r1→ eW → D′ → 0
esr1→ eX → D → 0
u ↓
0
↓
eL
↓
es ↓
↓eN
↓0
22
ROBERTO MART´INEZ-VILLA AND JERONIMO MONDRAG ´ON
with B, D, eL, eN in ST . We need to check that B′ and D′ are also in ST .
From the commutative exact diagram:
0
↓
↓
0 → B′ → Z → E′ → 0
0 → B → Z → E → 0
1 ↓
↓
↓0
We obtain the following commutative exact diagram:
0
↓
0 −→ E′′ −→ eL −→ D′′ −→ 0
↓
0 → E′ → eW → D′ → 0
↓
0 → E → eX → D → 0
↓
0 → eN → eN → 0
0
↓
↓
es ↓
↓
↓
0
0
↓
↓
u ↓
↓
↓
0
Since D′′ is a quotient of e L and Im u a submodule of D it follows both D′′ and
Im u are in ST .Therefore: D′ is in ST .
From the equality egr1 = 0 it follows eg factors through the finite dimensional
module D′.Tensoring with Re, the map: Re⊗eg : Re⊗e ReeW → Re⊗e ReeY fac-
tors through a finite dimensional R-module.
µ
→ W
↓ g
→ Y
We have a commutative exact diagram:
Re⊗e ReeW
Re⊗eg ↓
Re⊗e ReeY
where the multiplication maps µ and µ′ have kernels and cokernels in SR. It
follows that a the nap gµ factors through a module of finite length. This implies
gµ = 0 in modR /SR and µ an isomorphism in modR /SR implies g = 0 and
s−1g = 0.
(cid:3)
µ′
For the benefit of the reader we prove the last claim in more detail.
Lemma 2. Assume H : L → M is a map in modR which factors through a finite
dimensional module. Then h = 0 in modR /SR.
Proof. h = vu with u : L → U , v : U → M homomorphisms and U a R-module of
finite dimension. Changing U by Im u, we can assume u is an epimorphism.
We have a commutative exact diagram:
SKEW GROUP ALGEBRAS AND INVARIANTS
23
0
↓
U′
↓
u→ U → 0
v ↓
h→ M
0 → B′ → L
1 ↓
→ L
j
0
↓
↓
0 → B
↓U′
↓0
Since u is an epimimorphism, then Im v = Im h and U ′ are of finite dimension.
Therefore the injective map j : B → L has kernel and cokernel in SR.
We have a commutative diagram:
L
1 ւ ↑
B
L
տ ↓
L
h
j ց
M
j ր 0
t′
We have proved h = 0 in modR /SR.
(cid:3)
We comeback to the last claim of the theorem
Let s : Z → L be a map in modR whose kernel and cokernel are finite dimen-
sional, h : L → M a map such that hs factors through a module of finite dimension
and let j : B → Z be the kernel of hs. Then j has kernel and cokernel of finite di-
mension and sj has kernel and cokernel of finite dimension. We have a commutative
diagram:
L
1 ւ ↑
B
L
տ ↓
Z
s
h
sj ց
M
j ր 0
It follows h = 0 in modR /SR.
Remark 1. In our case R = C[X] ∗ G, T = e(C[X] ∗ G)e ∼= C[X]G, the algebra
C[X]G is affine this is, the is a polynomial ring C[Y1, Y2,... Ym] = C[Y ] and an
ideal I of C[Y ] such that C[Y ]/I ∼= C[X]G . The ring C[X]G has the grading
induced by the isomorphism e(C[X] ∗ G)e ∼= C[X]G, but in general I is not an
homogeneous ideal in the natural grading of C[Y ].
Changing notation, from the graded version of the theorem we have an equiva-
lence of categories: QgrC[X]∗G ∼= QgrC[X]G. This equivalence induces at the level
of bounded derived equivalences: Db( QgrC[X]∗G) ∼= Db(QgrC[X]G).
As remarked above, the exterior algebra in n variables, Λn is the Yoneda algebra
of C[X] , G acts as an automorphism group of Λn and the Yoneda algebra of
C[X] ∗ G is Λn ∗G.
By a theorem of [M S and by [M M ] there is an equivalence of triangulated cat-
egories: grΛn∗G ∼= Db( QgrC[X]∗G), where grΛn∗G denotes the stable category of
finitely generated graded modules. Therefore: there is an equivalence of triangu-
lated categories: grΛn∗G ∼= Db(QgrC[X]G).
24
ROBERTO MART´INEZ-VILLA AND JERONIMO MONDRAG ´ON
We have proved the following:
Corollary 3. Let G be a finite subgroup of Gl(n, C) such that no σ ∈ G different
from the identity has a fixed point, G acting as a group of automorphism of the
polynomial ring C[X] in n variables and e the idempotent e = 1/ G Pg∈G
g of the
skew group algebra C[X]∗ G and let Λn be the exterior algebra in n variables. Then
there are isomorphisms of triangulated categories: grΛn∗G ∼= Db(QgrC[X]G) ∼= Db(
QgrC[X]∗G). In particular, the categories Db(QgrC[X]G) and Db( QgrC[X]∗G) have
Auslander-Reiten triangles and they are of the form ZAn.
Proof. For the proof we use the fact Λn ∗ G is selfinjective Koszul and results from
[MZ] .
(cid:3)
4.1. Invariants for the Weyl algebra An and the homogenized Weyl alge-
bra Bn.. In this subsection we study the ring of invariants of the Weyl algebra AG
n
with G a finite subgroup of the automorphism group of An.We prove AG
n is a simple
algebra Morita equivalent to the skew group algebra An ∗ G . We then consider the
homogenized Weyl algebra Bn and a subgroup G of the group of automorphisms
of Bn satisfying some mild conditions, and prove for the invariant group BG
n and
the skew group algebra Bn ∗ G there is a a theorem relating the category of finitely
generated left BG
n modules and the category of finitely generated Bn ∗ G modules,
similar to the last theorem of the previous section.
i
Theorem 12. Let Λ be a simple algebra over a field K and, G a finite group of
automorphisms of Λ.Then the skew group algebra Λ ∗ G is simple.
Proof. Let I be a non zero two sided ideal of Λ ∗ G and r = a0 + a1σ1 + ...atσtwith
ai ∈ Λ and σi ∈ G, if some ai 6= 0 the element rσ−1
6= 0 is in I and the coefficient
of the identity is no zero, hence we can assume all ai in the expression of r are
non zero and call t + 1 = ℓ(r) the length of r. We can choose r to be of minimal
length among the non zero elements of I. Consider the set L0 defined by L0 = {b0
b0 + b1σ1 + ...btσt ∈ I} ∪ {0} . By definition L0 is non zero, we prove it is a two
sided ideal of Λ.
Let r1 = b0 + b1σ1 + ...btσt and r2 = c0 + c1σ1 + ...ctσt be two elements of I.Then
r1 + r2 = (b0 + c0) + (b1 + c1)σ1 + ...(bt + ct)σt is in I and ℓ(r1 + r3) ≤ ℓ(r1), by
minimality either r1 + r2 = 0 or ℓ(r1 + r3) = ℓ(r1) in any case b0 + c0 ∈ L0.
Let a be a non zero element of Λ. Then ar1 = ab0 + ab1σ1 + ...abtσt and
r1a = b0a + b1aσ1 σ1 + ...btaσtσt are in I with ab0 and b0a in L0.Hence L0 is a two
sided ideal.
It follows L0 = Λ and there is an expression of minimal length 1 + b1σ1 + ...btσt
in I.
Multiplying by σ−1
i + ...btσtσ−1
If a0 + a1σ1 + ...atσt is another non zero expression in I, then (a0b1−a1)σ1 +
..(a0bt. − at)σt ∈ I and by minimality a0bi − ai = 0 for i 6= 0. This is: a0 + a1σ1 +
...atσt = a0(1 + b1σ1 + ...btσt).
i +
i belongs to I is non zero and of minimal length. We define
i +
i ∈ I}. As before Li is a non zero two sided ideal of An.Then there is
in I
i +
bi+1σi+1σ−1
as above the set Li = { a0 a0 + a1σσ−1
...atσtσ−1
an expression 1 + c1σσ−1
and bi + b1σσ−1
i we obtain the expression bi + b1σσ−1
i + ci+1σi+1σ−1
i + ...btσtσ−1
i + ...ctσtσ−1
i = bi(1 + c1σσ−1
i + ...bi−1σi−1 + σ−1
i + bi+1σi+1σ−1
i + ...ai−1σi−1 + aiσ−1
i + ai+1σi+1σ−1
i + ...bi−1σi−1 + σ−1
i + ...ci−1σi−1 + ciσ−1
i
SKEW GROUP ALGEBRAS AND INVARIANTS
25
...ci−1σi−1 + ciσ−1
).In particular, bici = 1 and bi ∈
C−{0} is a unity. Since the element was arbitrary all coefficients in 1+b1σ1 +...btσt
are units, this is they are non zero complex numbers.
i + ci+1σi+1σ−1
i + ...ctσtσ−1
i
Assume the length t > 0.
Let { Xi}i∈Φ be a set of algebra generators of Λ as K-algebra. Then Xi +
Xib1σ1 + ...Xibtσt and Xi +b1X σ1
i σt are elements of I and b1(Xi −
i )σ1 + ...bt(Xi − X σt
X σ1
for an arbitrary Xi.
But if σ1 fixes all the generators of Λ. then σ1 = 1, a contradiction. It follows t = 0
and Λ = {a ∈ Λ a1 ∈ I}.Therefore 1 ∈ I and I = Λ ∗ G.
(cid:3)
i )σt is in I.By minimality Xi = X σ1
i σ1 + ...btX σt
i
As a corollary we obtain the following:
Corollary 4. Let An=C<X1,X2,...Xn,Y1,Y2,...Yn>/{[Xi,Xj],[Yi,Yj],[Xi,Yj]-δij}
be a Weyl algebra and, G a finite group of automorphisms of An.Then the skew
group algebra Λ ∗ G is simple.
Another consequence of the theorem is the following:
Theorem 13. Let Λ be a simple algebra over a field K and, G a finite group of
automorphisms of Λ.Then the algebra of invariants ΛG is simple.
The ideal Λ∗ GeIeΛ∗ G of Λ∗ G is non zero otherwise, eΛ∗ GeIeΛ∗ Ge = eIe =
Proof. Let I be a non zero ideal of ΛG.Then Ie = eIe is a non zero ideal of
ΛGe = eΛ ∗ Ge.
0.Therefore Λ ∗ GeIeΛ ∗ G = Λ ∗ G and eIe = eΛ ∗ GeIeΛ ∗ G e = eΛ ∗ Ge.
Corollary 5. Let An=C<X1,X2,...Xn,Y1,Y2,...Yn>/{[Xi,Xj],[Yi,Yj],[Xi,Yj]-δij}
be a Weyl algebra and, G a finite group of automorphisms of An.Then the algebra
of invariants AG
(cid:3)
n is simple.
We can prove now the following:
Theorem 14. Let Λ be a simple algebra over a field K and, G a finite group of
automorphisms of Λ. Then the algebra of invariants ΛG and the skew group algebra
Λ ∗ G are Morita equivalent.
Proof. Consider the idempotent of Λ ∗ G, e = 1/ G Pσ∈G
functor F = HomΛ∗G(Λ ∗ Ge,−) : M odΛ∗G → M odeΛ∗Ge has zero kernel.
A Λ∗ G-module M is in the kernel of F if and only if eM = 0,this is: if and only
if Λ ∗ GeΛ ∗ GM = 0 . But Λ ∗ G simple implies Λ ∗ G = Λ ∗ GeΛ ∗ G and M is in
the kernel of F if and only if M = 0.
σ. We prove first that the
The functor F is always dense.
Let f : M → N be a map of Λ ∗ G-modules and ef : eM → eN the restriction.
There is a commutative diagram:
Λ ∗ Ge ⊗ ⊗eΛ∗GeeM
µ
↓ Λ ∗ Ge ⊗ ef
Λ ∗ Ge ⊗eΛ∗Ge eN
→ M
f ↓
→ N
µ′
with µ and µ′ multiplication. The kernels of µ and µ′ are annihilated by e, hence
they are zero and the cokernel of µ is M/Λ ∗ GeΛ ∗ GM = 0.
Similarly for µ′.It follows both µ and µ′ are isomorphisms.
Therefore ef = 0 implies f = 0 and F is faithful.
26
ROBERTO MART´INEZ-VILLA AND JERONIMO MONDRAG ´ON
Let g : M ′ → N ′be a map of eΛ ∗ Ge-modules. Since F is dense, there exists
Λ ∗ G-modules M and N such that eM ∼= M ′and eN ∼= N ′and g can be identified
with a map g : eM → eN.There are maps:
µ
Λ ∗ Ge ⊗ ⊗eΛ∗GeeM
↓ Λ ∗ Ge ⊗ g
→ M
µ′
Λ ∗ Ge ⊗eΛ∗Ge eN
→ N
with µ, µ′isomorphisms and the map f = µ′Λ ∗ Ge ⊗ gµ−1is such that
HomΛ∗G(Λ ∗ Ge, f ) = ef = g.
Therefore HomΛ∗G(Λ ∗ Ge,−) is full.
(cid:3)
Corollary 6. Let An=C<X1,X2,...Xn,Y1,Y2,...Yn>/{[Xi,Xj],[Yi,Yj],[Xi,Yj]-δij}
be a Weyl algebra and, G a finite group of automorphisms of An.Then the algebra
of invariants AG
n and the skew group algebra An ∗ G are Morita equivalent.
Corollary 7. Let G be a finite subgroup of Sl(2, C) acting as automorphism group
of A1. Then A1 ∗ G and AG
1 are simple and Morita equivalent to the deformed
preprojective algebra. In particular the deformed preprojective algebra is simple.
For more results on the deformed preprojective algebra we refer to [C-BH].
We analyze next the relations between modBn∗G and modBG
for a special class of
subgroups of automorphisms of Bn, which include finite products of finite subgroups
of Sl(2, C).
n
We will need the following:
Theorem 15. Let Λ be a noetherian algebra over a field K and, G a finite group
of automorphisms of Λ such that Λ is finitely generated as left module over the ring
of invariants ΛG. Then ΛG is a noetherian algebra.
nj
Proof. Let I be a left ideal of ΛG. Then ΛI is a left ideal and by hypothesis, ΛI
is finitely generated as Λ-module and Λ finitely generated over ΛG implies ΛI is a
finitely generated ΛG-module. Let {
i ∈ I} be a set of
generators of ΛI. Then {bj
i ∈ I} is also a set of generators
of ΛI,after re indexing we have a set of generators {biui, 1 ≤ i ≤ n bi ∈ Λ, ui ∈ I}
. Let x be an element of I. Then for 1 ≤ i ≤ n, there exists ci ∈ ΛG such
i ui and
that x =
m
i , 1 ≤ j ≤ m bj
i ∈ Λ, uj
bj
i uj
Pi=1
i , 1 ≤ j ≤ m bj
i ∈ Λ, uj
Pi=1
Pi=1
i bσ
cσ
i =
i uσ
cibσ
i uj
m
m
m
m
cibiui. Then for σ ∈ G , x = xσ =
Pi=1
Pi=1
Pi=1
x = 1/ G Pσ∈G
with ci, ti ∈ ΛG.
Therefore: I is finitely generated as ΛG-module.
i ui = 1/ G
Pi=1 Pσ∈G
i ui =
cibσ
cibσ
m
m
ci(1/ G Pσ∈G
bσ
i )ui =
Pi=1
citiui,
(cid:3)
Proposition 3. Let G be a finite group of grade preserving automorphisms of Bn
fixing Z.Then Bn is a finitely generated left (right) BG
Proof. Since G is a group of grade preserving automorphisms BG
graded ring with (BG
j1 : BG
n is a positively
n )i = {b ∈ (Bn)i σ(b) = b for all σ ∈ G} and the inclusion
Let Cn be the polynomial ring in 2n variables, G acts as an automorphism group
n → Cn the inclusion as a graded
n → Bn is a homomorphism of graded C-algebras.
n be the ring of invariants and j0 : CG
n -module.
of Cn, let CG
subring.
SKEW GROUP ALGEBRAS AND INVARIANTS
27
We have a commutative exact diagram:
0 → ZBG
n → CG
n → BG
n → 0
↓ j0
↓ j1
↓ j2
0 → ZBn → Bn → Cn → 0
We know Cn is a finitely generated CG
n -module. [] and we can choose homoge-
n -module . Let b1, b2,... bt be homogeneous
neous generators c1, c2,... ct of Cn as CG
elements of Bn such that bi + ZBn = ci for 1 ≤ i ≤ t.
Let b be an element of Bn of degree d.Then b + ZBn =
t
n and b =
r0
i bi + Zµ1 .
r0
i ∈ BG
Since b, b1, b2,... bt are homogeneous we can choose r0
i +degree bi= d and degreeµ1 = d − 1.
elements with degreer0
Pi=1
t
r0
i bi + ZBn, with
t
Pi=1
i and µ1 homogeneous
Then µ1 =
r1
i bi + Zµ2 with µ2 homogeneous of degree d − 2.Continuing by
induction we obtain µi homogeneous of degree d − i, in particular µd has degree
zero, which mans it is a constant and we get:
Pi=1
d−1
t
b =
r0
i bi+
t
Pi=1
Z dk, with Z jrj
Z jrj
i bi+Z dk, k a complex number. Then b =
Pj=1
Pi=1
n and 1, b1, b2,... bt generate Bn as left BG
i ∈ BG
t
Pi=1
d−1
(
Pj=0
Z jrj
i )bi+
n -module.
(cid:3)
Corollary 8. Let G be a finite group of grade preserving automorphisms of Bn
fixing Z.Then BG
n is a noetherian algebra.
(Z − c)Bn = 0.
Lemma 3. Let Bn be the homogenized Weyl K-algebra over an infinite field H.
Then ∩c∈K
Proof. Since Bn has a Poincare-Birkoff basis with Z in the center, any element of
Bn is a polynomial g/X, Y, Z) in X´s, Y´s and Z. Given d ∈ K Z − d divides g if
and only if g(X, Y, d) = 0.
It is clear that g(X, Y, Z) = q(X, Y, Z)(Z − d) implies g(X, Y, d) = 0. We can
write g as a polynomial in Z, g(X, Y, Z) = g0(X, Y ) + g1(X, Y )Z + g2(X, Y )Z 2 +
...gt(X, Y )Z t with gi(X, Y ) polynomials in X´s and Y´s.
Assume now 0 6= h ∈ ∩c∈K
Then gi(X, Y )Z i = gi(X, Y )((Z − d) + d)i = g′
i(X, Y, Z)(Z − d) + gi(X, Y )di and
g(X, Y, Z) = q(X, Y, Z)(Z − d) + g(X, Y, d).Therefore g(X, Y, d) = 0 implies Z − d
divides g.
(Z − c)Bn. Then h = (Z − c1)q1 = (Z − c2)f with
c1 6= c2.Then (c2−c1)q1(X, Y, c2) = (c2−c2)f = 0 implies Z−c2 divides q1(X, Y, Z)
and h = (Z − c1)(Z − c2)q2.
Assume h is a polynomial in Z of degree m.Continuing by induction we obtain
h = (Z − c1)(Z − c2)...(Z − cm+1)qm+1 a contradiction.
(cid:3)
Therefore: ∩c∈K
(Z − c)Bn = 0.
In what remains of the section we want to extend the theorems of the previous
subsection to the homogenized Weyl algebras, the following result will be crucial:
Proposition 4. Let Bn =C<X1,X2,...Xn,Y1,Y2,...Yn,Z>/{[Xi,Xj],[Yi,Yj],[Xi,Yj]-
δijZ2,[Xi,Z],[Yi,Z]} be the homogenized Weyl algebra in 2n+1 variables, G a finite
28
ROBERTO MART´INEZ-VILLA AND JERONIMO MONDRAG ´ON
σ. Moreover, assume Bn satisfies the following conditions:
group of grade preserving automorphisms of Bn and e the idempotent of Bn ∗ G,
e = 1/ G Pσ∈G
i) For all σ ∈ G, σ(Z) = Z.
ii) The C-subspace V =
⊕i=1
iii) If v ∈ V , v 6= 0 and σ(v) = v, then σ = 1.
Then Bn ∗ G/Bn ∗ GeBn ∗ G is a finite dimensional C-algebra.
CYj of Bn is G- invariant.
CXi ⊕
⊕j=1
n
n
Proof. We will use the commutative exact diagram:
0 → ZBG
n → CG
n → BG
n → 0
↓ j0
↓ j1
↓ j2
0 → ZBn → Bn → Cn → 0
Let {mi}i∈I be the set of maximal ideals of CG
n , for each mi an ideal ni of Cn
such that Cn ∩ ni = mi. Let n0= (X1, X2,... Xn, Y1, Y2,... Yn) be the maximal
irrelevant ideal of Cn and m0 = Cn ∩ n0.
We saw in Proposition ?, that given any ideal mi 6= m0 , {nσ
i }σ∈G , such that
i = nτ
nσ
i implies σ = τ , is the set of maximal ideals of Cn above mi. In particular,
is a G-invariant ideal of Cn, and √m0Cn = n0.
the radical √miCn = ∩σ∈G
By the isomorphism theorems, there exists maximal ideals { mi}i∈I of BG
nσ
i
n , such that mi/ZBG
taining ZBG
ZBn such that ni/ZBn ∼= ni.
get ni ∩ BG
G-invariant ideal of Bn, with L0 = n0.
From the isomorphisms: ni/ZBn ∩ BG
n = mi and nσ
i ∩ BG
n con-
n ∼= mi and maximal ideals {ni}i∈I of Bn containing
n = mi/ZBG
n , we
nσ
is a
i
n = mi for all σ ∈ G. It follows Li = ∩σ∈G
n = ni ∩ BG
n /ZBG
n /ZBG
We have ZBn ⊆ ∩i∈I ∩σ∈G
nσ
i = ∩i∈I
Li and ∩i∈I
Li/ZBn = ∩i∈I ∩σ∈G
nσ
i = 0.Therefore:
nσ
nσ
i ∩
i = ZBn.
∩c∈C−{0}
(Z − c)B = ∩c∈C
∩i∈I ∩σ∈G
c)Bn ∼= An, hence each (Z − c)Bn is a maximal ideal of Bn.
It was proved in [], that for any c ∈ C − {0} there is an isomorphism Bn/(Z −
Then we have: ∩i∈I ∩σ∈G
We have a ring homomorphism:
ψ : Bn∗ G → Yi∈I−{0}
Bn∗ G/(Z− c)Bn∗ G
By the isomorphism theorems: Bn/Li ∼= Bn/ZBn/Li/ZBn ∼= Cn /√miCn and
Bn/n0 ∼= Bn/ZBn/n0/ZBn ∼= Cn/n0 ∼= C.
Therefore: Bn ∗ G/n0 ∗ G ∼= CG and for each i 6=0, Si ∗ G = Bn ∗ G/Li ∗ G ∼=
Cn/√miCn ∗ G is a simple finite dimensional algebra, as we proved in Proposition
Bn∗ G/Li∗ G× Bn∗ G/n0∗ G× Yc∈C−{0}
(Z − c)B = 0.
whose kernel is zero.
?.
Then we have an injective ring homomorphism:
ψ : Bn ∗ G → Yi∈I−{0}
By the simplicity of Si∗G and An∗G for the idempotent e we have: Si∗GeSi∗G =
Si ∗ G × CG × Yc∈C−{0}
An ∗ G
Si ∗ G and An ∗ GeAn ∗ G = An ∗ G.
∧
ψ(e) =
e = ((e), e, (e)) is an idempotent of Yi∈I−{0}
Si ∗ G × CG × Yc∈C−{0}
An ∗ G
such that:
SKEW GROUP ALGEBRAS AND INVARIANTS
29
(Yi∈I−{0}
(Yi∈I−{0}
Si*G×CG× Yc∈C−{0}
Si*G×CGe× Yc∈C−{0}
∧
An*G)
e (Yi∈I−{0}
Si*G×CG× Yc∈C−{0}
An*G)=
An*G and as in Therorem ? there is an injective
ring homomorphism:
(cid:3)
Bn ∗ G/Bn ∗ GeBn ∗ G → CG/CGe.
It follows Bn ∗ G/Bn ∗ GeBn ∗ G is a finite dimensional C-algebra.
We have all the ingredients to prove for the homogenized Weyl algebras, theorem
analogous to Theorem9 and Theorem 10, what was essential in the proof of those
theorems was the fact Cn ∗ G/Cn ∗ GeCn ∗ G is a finite dimensional C-algebra and
the fact Cn is a finitely generated CG
n -module. Then the proof of the next two
theorems follows by similar arguments to those used in the proof of Theorem 9 and
Theorem 10 and we will skip it.
Theorem 16. Let Bn =C<X1,X2,...Xn,Y1,Y2,...Yn,Z>/{[Xi,Xj],[Yi,Yj],[Xi,Yj]-
δijZ2,[Xi,Z],[Yi,Z]} be the homogenized Weyl algebra in 2n+1 variables, G a finite
group of grade preserving automorphisms of Bn and e the idempotent of Bn ∗ G,
e = 1/ G Pσ∈G
i) For all σ ∈ G, σ(Z) = Z.
ii) The C-subspace V =
⊕i=1
iii) If v ∈ V , v 6= 0 and σ(v) = v, then σ = 1.
Then: writing R = Bn ∗ G and T = eBn ∗ Ge ∼= BG
n and denoting by SR and ST
the categories of finite dimensional R and T -modules, respectively and by modR,
modT , the categories of finitely generated, R and T -modules, respectively. Then
there is a commutative diagram of categories and functors:
σ. Moreover, assume Bn satisfies the following conditions:
CYj of Bn is G- invariant.
CXi ⊕
⊕j=1
n
n
modR
↓ πR
modR /SR
HomR(Re,−)
→
H→
We have also the graded version:
modT
↓ πT
modT /ST
, with H an equivalence.
σ. Moreover, assume Bn satisfies the following conditions:
Theorem 17. Let Bn =C<X1,X2,...Xn,Y1,Y2,...Yn,Z>/{[Xi,Xj],[Yi,Yj],[Xi,Yj]-
δijZ2,[Xi,Z],[Yi,Z]} be the homogenized Weyl algebra in 2n+1 variables, G a finite
group of grade preserving automorphisms of Bn and e the idempotent of Bn ∗ G,
e = 1/ G Pσ∈G
i) For all σ ∈ G, σ(Z) = Z.
ii) The C-subspace V =
⊕i=1
iii) If v ∈ V , v 6= 0 and σ(v) = v, then σ = 1. Then: writing R = Bn ∗ G and
T = eBn ∗ Ge ∼= BG
n and considering both as positively graded algebras, denoting by
SR and ST the categories of finite dimensional graded R and T -modules, respectively
and by gR, grT , the categories of finitely generated, R and T -modules, respectively
and by tailsR and tailsT the quotient categories grR/SR and grT / ST . Then there is
CYj of Bn is G- invariant.
CXi ⊕
⊕j=1
n
n
a commutative diagram of categories and functors:
with H an equivalence.
grR
↓ πR
tailsR
HomR(Re,−)
→
H→
grT
↓ πT
tailsT
,
30
ROBERTO MART´INEZ-VILLA AND JERONIMO MONDRAG ´ON
n
We also have as in Theorem 10:
Changing notation, from the graded version of the theorem we have an equiv-
. This equivalence induces at the level of
alence of categories: QgrBn∗G ∼= QgrBG
bounded derived equivalences: Db( QgrBn∗G) ∼= Db(QgrBG
).
As remarked above, G acts as an automorphism group of B!
n , the Yoneda algebra
The algebra B!
of Bn and the Yoneda algebra of Bn ∗ G is B!
there is an equivalence of triangulated categories: grB!
grB!
there is an equivalence of triangulated categories: grB!
n ∗G is Koszul selfinjective. By a theorem of [M S and by [M M ]
n∗G ∼= Db( QgrBn∗G), where
n∗G denotes the stable category of finitely generated graded modules. Therefore:
n ∗G.
n∗G ∼= Db(QgrBG
).
n
We have proved the following:
n
σ. Moreover, assume Bn satisfies the following conditions:
Corollary 9. Let Bn =C<X1,X2,...Xn,Y1,Y2,...Yn,Z>/{[Xi,Xj],[Yi,Yj],[Xi,Yj]-
δijZ2,[Xi,Z],[Yi,Z]} be the homogenized Weyl algebra in 2n+1 variables, G a finite
group of grade preserving automorphisms of Bn and e the idempotent of Bn ∗ G,
e = 1/ G Pσ∈G
i) For all σ ∈ G, σ(Z) = Z.
ii) The C-subspace V =
⊕i=1
iii) If v ∈ V , v 6= 0 and σ(v) = v, then σ = 1.
Let B!
CYj of Bn is G- invariant.
n be the Yoneda algebra of Bn. Then there are isomorphisms of triangulated
) ∼= Db( QgrBn∗G).In particular, the categories
) and Db( QgrBn∗G) have Auslander-Reiten triangles and they are of the
n∗G ∼= Db(QgrBG
CXi ⊕
⊕j=1
n
n
n
categories: grB!
Db(QgrBG
form ZAn.
n
Proof. As before, the proof uses the fact B!
from [MZ] .
n ∗ G is selfinjective Koszul and results
(cid:3)
References
[AR] Auslander, M.; Reiten, I.; McKay quivers and extended Dynkin diagrams.
Trans. Amer. Math. Soc. 293 (1986), no. 1, 293 -- 301
[Be] Benson D,; Polynomial Invariants of Finite Groups. London Mathematical
Society Lecture Notes Series 190, Cambridge University Press, 1993.
[CR] Curtis Ch. W. Reiner I. Representation Theory of Finite Groups and
Associative Algebras, Pure and Applied Mathematics Vol XI, Interscience 1962
[C-BH] Crawley-Boevey, W.; Holland, M. P. Noncommutative deformations of
Kleinian singularities. Duke Math. J. 92 (1998), no. 3, 605 -- 635.
[C-B] Crawley-Boevey W. DMV Lectures on Representations of quivers, prepro-
jective algebras and deformation of quotient singularities.(preprint Leeds)
[Co] Coutinho S.C. A Primer of Algebraic D-modules, London Mathematical
Society, Students Texts 33, 1995
[Ga] Gabriel, Pierre Des cat´egories ab´eliennes. Bull. Soc. Math. France 90 1962
323 -- 448.
[GH] Green, E.; Huang, R. Q. Projective resolutions of straightening closed
algebras generated by minors. Adv. Math. 110 (1995), no. 2, 314 -- 333.
[GM1] Green, E. L.; Mart´ınez Villa, R.; Koszul and Yoneda algebras. Represen-
tation theory of algebras (Cocoyoc, 1994), 247 -- 297, CMS Conf. Proc., 18, Amer.
Math. Soc., Providence, RI, 1996.
SKEW GROUP ALGEBRAS AND INVARIANTS
31
[GM2] Green, E. L.; Mart´ınez-Villa, R.; Koszul and Yoneda algebras. II. Al-
gebras and modules, II (Geiranger, 1996), 227 -- 244, CMS Conf. Proc., 24, Amer.
Math. Soc., Providence, RI, 1998.
[GMT] Guo, Jin Yun; Mart´ınez-Villa, R.; Takane, M.; Koszul generalized Aus-
lander regular algebras. Algebras and modules, II (Geiranger, 1996), 263 -- 283, CMS
Conf. Proc., 24, Amer. Math. Soc., Providence, RI, 1998
[GuM] Guo, Jin Yun; Mart´ınez-Villa, R.; Algebra pairs associated to McKay
quivers. Comm. Algebra 30 (2002), no. 2, 1017 -- 1032.
[Ku] Kunz E.; Introduction to Commutative Algebra and Algebraic Geometry,
Birkhauser, 1985
[Le] Lenzing H.;, Polyhedral groups and the geometric study of tame hereditary
algebras (preprint Paderborn).
[Li] Li Huishi. Noncommutative Groebner Bases and Filtered-Graded Transfer,
Lecture Notes in Mathematics 1795, Springer, 2002
[MM] Mart´ınez-Villa, R., Martsinkovsky; A. Stable Projective Homotopy Theory
of Modules, Tails, and Koszul Duality, (aceptado 11 de septiembre 2009), Comm.
Algebra 38 (2010), no. 10, 3941 -- 3973.
[MS] Mart´ınez Villa, Roberto; Saor´ın, M.; Koszul equivalences and dualities.
Pacific J. Math. 214 (2004), no. 2, 359 -- 378.
[MZ] Mart´ınez-Villa, Roberto; Zacharia, Dan; Approximations with modules
having linear resolutions. J. Algebra 266 (2003), no. 2, 671 -- 697.
[MV1] Mart´ınez-Villa, R.; Skew group algebras and their Yoneda algebras. Math.
J. Okayama Univ. 43 (2001), 1 -- 16.
[MV2] Mart´ınez-Villa, R.; Serre duality for generalized Auslander regular alge-
bras. Trends in the representation theory of finite-dimensional algebras (Seattle,
WA, 1997), 237 -- 263, Contemp. Math., 229, Amer. Math. Soc., Providence, RI,
1998.
[MV3] Mart´ınez-Villa, R.; Applications of Koszul algebras: the preprojective
algebra. Representation theory of algebras (Cocoyoc, 1994), 487 -- 504, CMS Conf.
Proc., 18, Amer. Math. Soc., Providence, RI, 1996.
[MV4] Martinez-Villa, R. Graded, Selfinjective, and Koszul Algebras, J. Algebra
215, 34-72 1999
[MMo] Martinez-Villa, R. Mondragon J. On the homogeneized Weyl Algebra.
(preprint 2011).
[Mi] Miyachi, Jun-Ichi; Derived Categories with Applications to Representations
of Algebras, Chiba Lectures, 2002.
[Mu] Muller W. Darstellungstheorie von eindlichen Gruppen, Teubner Studi-
enbucher, 1980
[Mc] McKay J.;, Graphs, singularities and finite groups, Proc. Sympos. Pure
Math., vol. 37, Amer. Math. Soc., Providence, R. I., 1980, pp. 183-186.
[P] Popescu N.; Abelian Categories with Applications to Rings and Modules;,
L.M.S. Monographs 3, Academic Press 1973.
[RR] Reiten, I.; Riedtmann, Ch. Skew group algebras in the representation
theory of Artin algebras. J. Algebra 92 (1985), no. 1, 224 -- 282.
[Sm] Smith, P.S.; Some finite dimensional algebras related to elliptic curves,
Rep. Theory of Algebras and Related Topics, CMS Conference Proceedings, Vol.
19, 315-348, Amer. Math. Soc 1996.
32
ROBERTO MART´INEZ-VILLA AND JERONIMO MONDRAG ´ON
[Ste] Steinberg R.,; Finite subgroups of SU2, Dynkin Diagrams and Affine Cox-
eter elements. Pacific Journal of Mathematics, Vol. 118, No. 2, 1985.
[Stu] Sturmfels B.; Algorithms in Invariant Theory, Texts and Monographs in
Symbolic Computation, Springer-Verlag, 1993
(A. One and A. Two) Centro de Ciencias Matem´aticas, UNAM, Morelia, Mich. M´exico
E-mail address, A. One: [email protected]
URL: http://www.matmor.unam.mx
Current address, A. Two: Centro de Ciencias Matem´aticas, UNAM, Morelia, Mich. M´exico
E-mail address, A. Two: [email protected]
URL: http://www.matmor.unam.mx
|
1407.1750 | 2 | 1407 | 2015-12-18T11:51:30 | Non-abelian tensor product and homology of Lie Superalgebras | [
"math.RA"
] | We introduce the non-abelian tensor product of Lie superalgebras, study some of its properties including nilpotency, solvability and Engel, and we use it to describe the universal central extensions of Lie superalgebras. We present the low-dimensional non-abelian homology of Lie superalgebras and establish its relationship with the cyclic homology of associative superalgebras. We also define the non-abelian exterior product and give an analogue of Miller's theorem, Hopf formula and a six-term exact sequence for the homology of Lie superalgebras. | math.RA | math |
NON-ABELIAN TENSOR PRODUCT AND HOMOLOGY OF LIE
SUPERALGEBRAS
X. GARCÍA-MARTÍNEZ, E. KHMALADZE, M. LADRA
Abstract. We introduce the non-abelian tensor product of Lie superalgebras
and study some of its properties. We use it to describe the universal central
extensions of Lie superalgebras. We present the low-dimensional non-abelian
homology of Lie superalgebras and establish its relationship with the cyclic
homology of associative superalgebras. We also define the non-abelian exterior
product and give an analogue of Miller's theorem, Hopf formula and a six-term
exact sequence for the homology of Lie superalgebras.
1. Introduction
In [1], Brown and Loday introduced the non-abelian tensor product of groups
in the context of an application in homotopy theory. Analogous theories of non-
abelian tensor product have been developed in other algebraic structures such as
Lie algebras [8] and Lie -- Rinehart algebras [4]. In [8], Ellis investigated the main
properties of the non-abelian tensor product of Lie algebras and its relation to the
low-dimensional homology of Lie algebras. In particular, he described the universal
central extension of a perfect Lie algebra via the non-abelian tensor product. In [7],
the non-abelian exterior product of Lie algebras is introduced and a six-term exact
sequence relating low-dimensional homologies is obtained. In [10], using the non-
abelian tensor product, Guin defined the non-abelian low-dimensional homology
of Lie algebras and compared these groups with the cyclic homology and Milnor
additive K-theory of associative algebras.
The theory of Lie superalgebras, also called Z2-graded Lie algebras, has aroused
much interest both in mathematics and physics. Lie superalgebras play a very
important role in theoretical physics since they are used to describe supersymmetry
in a mathematical framework. A comprehensive description of the mathematical
theory of Lie superalgebras is given in [14], containing the complete classification
of all finite-dimensional simple Lie superalgebras over an algebraically closed field
of characteristic zero. In the last few years, the theory of Lie superalgebras has
experienced a remarkable evolution obtaining many results on representation theory
and classification, most of them extending well-known facts on Lie algebras.
In this paper we develop the non-abelian tensor product and the low-dimensional
non-abelian homology of Lie superalgebras, generalizing the corresponding notions
for Lie algebras, with applications in universal central extensions and homology of
Lie superalgebras and cyclic homology of associative superalgebras.
2010 Mathematics Subject Classification. 17B55, 17B30, 17B60.
Key words and phrases. Lie superalgebras, associative superalgebras, non-abelian tensor and
exterior products, non-abelian homology, cyclic homology, Hopf formula, crossed module.
1
2
X. GARCÍA-MARTÍNEZ, E. KHMALADZE, M. LADRA
The organization of this paper is as follows: after this introduction, in Section 2
we give some definitions and necessary well-known results for the development of
the paper. We also introduce actions and crossed modules of Lie superalgebras. In
Section 3 we introduce the non-abelian tensor product of Lie superalgebras, we es-
tablish its principal properties such as right exactness and relation with the tensor
product of supermodules. We describe the universal central extension of a perfect
Lie superalgebra via the non-abelian tensor product (Theorem 4.1). In particular,
applying this theorem, we obtain that st(m, n, A) is the universal central extension
of sl(m, n, A), for m + n ≥ 5, where A is a unital associative superalgebra. We
also study nilpotency and solvability of the non-abelian tensor product of Lie su-
peralgebras (Theorem 3.9). Using the non-abelian tensor product, in Section 5 we
introduce the low-dimensional non-abelian homology of Lie superalgebras with co-
efficients in crossed modules. We show that, if the crossed module is a supermodule,
then the non-abelian homology is the usual homology of Lie superalgebras. Then
we apply this non-abelian homology to relate cyclic homology and Milnor cyclic
homology of associative superalgebras, extending the results of [10]. Finally, in the
last section we construct the non-abelian exterior product of Lie superalgebras and
we use it to obtain Miller's type theorem for free Lie superalgebras, Hopf formula
and a six-term exact sequence in the homology of Lie superalgebras.
Conventions and notations. Throughout this paper we denote by K a unital
commutative ring unless otherwise stated. All modules and algebras are defined
¯0 = 1
over K. We write Z2 = {¯0, ¯1} and use its standard field structure. We put (−1)
and (−1)
¯1 = −1.
By a supermodule M we mean a module endowed with a Z2-gradation: M =
M¯0 ⊕ M¯1. We call elements of M¯0 (resp. M¯1) even (resp. odd). Non-zero elements
of M¯0 ∪ M¯1 will be called homogeneous. For a homogeneous m ∈ M ¯α, ¯α ∈ Z2,
its degree will be denoted by m . We adopt the convention that whenever the
degree function occurs in a formula, the corresponding elements are supposed to be
homogeneous. By a homomorphism of supermodules f ∶ M → N of degree f ∈ Z2
we mean a linear map satisfying f (M ¯α) ⊆ N ¯α+ f . In particular, if f = ¯0, then the
homomorphism f will be called of even grade (or even linear map).
By a superalgebra A we mean a supermodule A = A¯0 ⊕ A¯1 equipped with a
bilinear multiplication satisfying A ¯αA ¯β ⊆ A ¯α+ ¯β, for ¯α, ¯β ∈ Z2.
2. Preliminaries on Lie Superalgebras
In this section we review some terminology on Lie superalgebras and recall no-
tions used in the paper. We mainly follow [2, 18], although with some modifications.
We also introduce notions of actions and crossed modules of Lie superalgebras.
2.1. Definition and some examples of Lie superalgebras.
Definition 2.1. A Lie superalgebra is a superalgebra M = M¯0⊕M¯1 with a multipli-
cation denoted by [ , ], called bracket operation, satisfying the following identities:
[x, y] = −(−1) x y [y, x],
x,[y, z](cid:6) =[x, y], z(cid:6) +(−1) x y y,[x, z](cid:6),
[m¯0, m¯0] = 0,
for all homogeneous elements x, y, z ∈ M and m¯0 ∈ M¯0.
NON-ABELIAN TENSOR PRODUCT AND HOMOLOGY OF LIE SUPERALGEBRAS
3
Note that the last equation is an immediate consequence of the first one in the
case 2 has an inverse in K. Moreover, it can be easily seen that the second equation
is equivalent to the graded Jacobi identity
(−1) x z x,[y, z](cid:6) +(−1) y x y,[z, x](cid:6) +(−1) z y z,[x, y](cid:6) = 0.
For a Lie superalgebra M = M¯0 ⊕ M¯1, the even part M¯0 is a Lie algebra. Hence,
if M¯1 = 0, then M is just a Lie algebra. A Lie superalgebra M without even part,
i. e., M¯0 = 0, is an abelian Lie superalgebra, that is, [x, y] = 0 for all x, y ∈ M .
of even grade such that f[x, y] =[f(x), f(y)] for all x, y ∈ M .
A Lie superalgebra homomorphism f ∶ M → M ′ is a supermodule homomorphism
Example 2.2.
(i) Any associative superalgebra A can be considered as a Lie superalgebra with
the bracket
[a, b] = ab −(−1) a b ba.
(ii) Let m, n be positive integers and A a unital associative superalgebra. Con-
and with the usual product of matrices. A Z2-gradation is defined as follows: ho-
sider the algebra M(m, n, A) of all (m + n) ×(m + n)-matrices with entries in A
mogeneous elements are matrices Eij(a) having the homogeneous element a ∈ A at
the position (i, j) and zero elsewhere, and Eij(a) = i + j + a , where i = ¯0 if
1 ≤ i ≤ m and i = ¯1 if m + 1 ≤ i ≤ m + n. With this gradation, M(m, n, A) turns
denoted by gl(m, n, A).
(iii) Let V = V¯0 ⊕ V¯1 be a supermodule. Then the supermodule EndK(V) of
out to be an associative superalgebra. The corresponding Lie superalgebra will be
all linear endomorphisms V → V (of both degrees 0 and 1) has a structure of an
associative superalgebra with respect to composition (see [2]) and hence becomes
a Lie superalgebra.
In particular, if the ground ring K is a field, and m, n are
dimensions of V¯0 and V¯1 respectively, then choosing a homogeneous basis of V
ordered such that even elements stand before odd, the elements of EndK(V) can
be seen as (m + n) ×(m + n)-square matrices
a b
d
c
where a, b, c and d are respectively m×m, m×n, n×m and n×n matrices with entries
in K. The even elements are the matrices with b = c = 0 and the odd elements are
matrices with a = d = 0.
Let M and N be two submodules of a Lie superalgebra P . We denote by [M, N]
the submodule of P spanned by all elements [m, n] with m ∈ M and n ∈ N . A
Z2-graded submodule M is a graded ideal of P if [M, P] ⊆ M . In particular, the
submodule Z(P) = {c ∈ P ∶ [c, p] = 0 for all p ∈ P} is a graded ideal and it is called
the centre of P . Clearly if M and N are graded ideals of P , then so is [M, N].
Let M be a Lie superalgebra and D ∈ EndK(M). We say that D is a derivation
if for all x, y ∈ M
D([x, y]) =[D(x), y] +(−1) D x [x, D(y)].
We denote by DerK(M) ¯α
One verifies that the supermodule of derivations
the set of homogeneous derivations of degree ¯α ∈ Z2.
DerK(M) = DerK(M)¯0 ⊕ DerK(M)¯1
4
X. GARCÍA-MARTÍNEZ, E. KHMALADZE, M. LADRA
is a subalgebra of the Lie superalgebra EndK(M).
2.2. Actions and crossed modules of Lie superalgebras.
Definition 2.3. Let P and M be two Lie superalgebras. By an action of P on M
we mean a K-bilinear map of even grade,
P × M → M,
(p, m) ↦ pm,
such that
(i)
[p,p′]m = p(p′
m) −(−1) p p′ p′(pm),
(ii) p[m, m′] =[pm, m′] +(−1) p m [m, pm′],
for all homogeneous p, p′ ∈ P and m, m′ ∈ M .
The action is called trivial if pm = 0 for all p ∈ P and m ∈ M .
For example, if M is a graded ideal and P is a subalgebra of a Lie superalgebra
Q, then the bracket in Q induces an action of P on M .
Note that the action of P on M is the same as a Lie superalgebra homomorphism
P → DerK(M).
Remark. If M is an abelian Lie superalgebra enriched with an action of a Lie
superalgebra P , then M has a structure of a supermodule over P (P -supermodule,
for short) (see e. g. [18]), that is, there is a K-bilinear map of even grade P ×M → M ,
(p, m) ↦ pm, such that
for all homogeneous p, p′ ∈ P and m ∈ M .
[p, p′]m = p(p′m) −(−1) p p′ p′(pm),
Note that a P -supermodule M is the same as a K-supermodule M together with
a Lie superalgebra homomorphism P → EndK(M).
Definition 2.4. Given two Lie superalgebras M and P with an action of P on
M , we can define the semidirect product M ⋊ P with the underlying supermodule
M ⊕ P endowed with the bracket given by
[(m, p),(m′, p′)] =([m, m′] + pm′ −(−1) m p′ (p′
m),[p, p′]).
Now we are ready to introduce the following notion of crossed modules of Lie
superalgebras (see also [23, Definition 5]).
Definition 2.5. A crossed module of Lie superalgebras is a homomorphism of Lie
superalgebras ∂∶ M → P with an action of P on M satisfying
(i) ∂(pm) =[p, ∂(m)],
(ii) ∂(m)m′ =[m, m′],
for all p ∈ P and m, m′ ∈ M .
Example 2.6. There are some standard examples of crossed modules:
(i) The inclusion M ↪ P of a graded ideal M of a Lie superalgebra P is a crossed
module of Lie superalgebras.
(ii) If P is a Lie superalgebra and M is a P -supermodule, the trivial map 0∶ M →
P is a crossed module of Lie superalgebras.
(iii) A central extension of Lie superalgebras ∂∶ M ↠ P (i.e., Ker ∂ ⊆ Z(M)) is
pm =[ ¯m, m], where ¯m ∈ M is any element of ∂−1(p).
a crossed module of Lie superalgebras. Here the action of P on M is given by
NON-ABELIAN TENSOR PRODUCT AND HOMOLOGY OF LIE SUPERALGEBRAS
5
(iv) The homomorphism of Lie superalgebras ∂∶ M → DerK(M) which sends m ∈
M to the inner derivation ad(m) ∈ DerK(M), defined by ad(m)(m′) = [m, m′],
together with the action of DerK(M) on M given by Dm = D(m), is a crossed
module of Lie superalgebras.
Lemma 2.7. Let ∂∶ M → P be a crossed module of Lie superalgebras. Then the
following conditions are satisfied:
(i) The kernel of ∂ is in the centre of M .
(ii) The image of ∂ is a graded ideal of P .
(iii) The Lie superalgebra Im ∂ acts trivially on the centre Z(M), and so trivially
on Ker ∂. Hence Ker ∂ inherits an action of P~ Im ∂ making Ker ∂ a P~ Im ∂-
supermodule.
Proof. This is an immediate consequence of Definition 2.5.
(cid:3)
2.3. Free Lie superalgebra and enveloping superalgebra of a Lie superal-
gebra.
Definition 2.8. The free Lie superalgebra on a Z2-graded set X = X¯0 ∪ X¯1 is a Lie
is any Lie superalgebra and j∶ X → M is a degree zero map, then there is a unique
superalgebra F(X) together with a degree zero map i∶ X → F(X) such that if M
Lie superalgebra homomorphism h∶ F(X) → M with j = h ○ i.
The existence of free Lie superalgebras is guaranteed by an analogue of Witt's
theorem (see [18, Theorem 6.2.1]). In the sequel we need the following construction
of the free Lie superalgebra.
Construction 2.9. Let X = X¯0 ∪ X¯1 be a Z2-graded set. Denote by mag(X) the
free magma over the set X. The free superalgebra on X, denoted by alg(X), has
as elements the finite sums ∑i λivi, where λi ∈ K and xi are elements of mag(X)
and the multiplication in alg(X) extends the multiplication in mag(X). Note that
the grading is naturally defined in alg(X). The free Lie superalgebra F(X) is the
quotient alg(X)~I, where I is the graded ideal generated by the elements
xy +(−1) x y yx,
(−1) x z x(yz) +(−1) y x y(zx) +(−1) z y z(xy),
x¯0x¯0,
for all homogeneous x, y, z ∈ X and x¯0 ∈ X¯0.
Definition 2.10. The universal enveloping superalgebra of a Lie superalgebra M
is a pair (U(M), σ), where U(M) is a unital associative superalgebra and σ∶ M →
U(M) is an even linear map satisfying
σ[x, y] = σ(x)σ(y) −(−1) x y σ(y)σ(x),
(1)
for all homogeneous x, y ∈ M , such that the following universal property holds: for
any other pair (A, σ′), where A is a unital associative superalgebra and σ′∶ M → A
f ∶ U(M) → A such that f ○ σ = σ′.
is an even linear map satisfying (1), there is a unique superalgebra homomorphism
Now we need to recall (see e. g. [21]) that, given two supermodules M and N ,
the tensor product of modules M ⊗K N has a natural supermodule structure with
6
X. GARCÍA-MARTÍNEZ, E. KHMALADZE, M. LADRA
Z2-grading given by
(M ⊗K N) ¯α = ࣷ
¯β+¯γ = ¯α(M ¯β ⊗K N¯γ).
tensor superalgebra.
In particular, the tensor power M ⊗n, n ≥ 2, has the induced Z2-grading. Hence the
tensor algebra T(M) has the Z2-grading extending that of M . We call T(M) the
Construction 2.11. Let M be a Lie superalgebra and T(M) the tensor superalge-
bra over the underlying supermodule of M . Consider the two-sided ideal J(M) of
T(M) generated by all elements of the form
m ⊗ m′ −(−1) m m′ m′ ⊗ m −[m, m′],
for all homogeneous m, m′ ∈ M . Then the quotient U(M) = T(M)~J(M) is a
unital associative superalgebra. By composing the canonical inclusion M → T(M)
with the canonical projection T(M) → U(M) we get the canonical even linear map
σ∶ M → U(M). Then the pair (U(M), σ) is the universal enveloping superalgebra
of M (see [2]).
Note that, as in the Lie algebra case, the universal enveloping superalgebra turns
out to be a very useful tool for the representation theory of Lie superalgebras. In
particular, by the universal property, it follows that a Lie supermodule over a Lie
superalgebra M is the same as a Z2-graded (left) U(M)-module (see [21, Chapter
1]).
Let us consider K with Z2-grading concentrated in degree zero, that is, with
K¯1 = 0. Then the trivial map from a Lie superalgebra M into K gives rise to a
unique homomorphism of superalgebras ε∶ U(M) → K. The kernel of ε, denoted by
Ω(M), is called the augmentation ideal of M . Obviously, Ω(M) is just the graded
ideal of U(M) generated by σ(M).
2.4. Homology of Lie superalgebras. Now we briefly recall from [18, 22] the
definition of homology of Lie superalgebras.
The Grassmann algebra of a Lie superalgebra P , denoted by ⋀K(P), is defined
to be the quotient of the tensor superalgebra T(P) of P by the ideal generated by
the elements
x ⊗ y +(−1) x y y ⊗ x,
for all homogeneous x, y ∈ P . Note that ⋀K(P) = ࣷn>0 ⋀n
image of P ⊗n in ⋀K(P), has an induced P -supermodule structure given by
K(P), where ⋀n
i=1(−1) x ∑k<i xk (x1 ∧ ⋯ ∧[x, xi] ∧ ⋯ ∧ xn).
x(x1 ∧ ⋯ ∧ xn) =
n
Q
K(P) is the
Let M be a P -supermodule and consider the chain complex (C∗(P, M), d∗) de-
fined by Cn(P, M) = ⋀n
K(P) ⊗K M , for n ≥ 0, with boundary maps dn∶ Cn(P, M) →
Cn−1(P, M) defined on generators by
dn(x1 ∧ ⋯ ∧ xn ⊗ y) =
i=1(−1)i+ xi ∑k>i xk (x1 ∧ ⋯ ∧ xi ∧ ⋯ ∧ xn ⊗ xiy)
n
Q
i<j(−1)i+j+ xi ∑k<i xk + xj ∑l<j xl + xi xj ([xi, xj] ∧ ⋯ ∧ xi ∧ ⋯ ∧ xj ∧ ⋯ ∧ xn ⊗ y).
+ Q
NON-ABELIAN TENSOR PRODUCT AND HOMOLOGY OF LIE SUPERALGEBRAS
7
The n-th homology of the Lie superalgebra P with coefficients in the P -supermodule
M , Hn(P, M), is the n-th homology of the chain complex (C∗(P, M), d∗), i.e.
Hn(P, M) =
Ker dn
Im dn+1
.
If K is regarded as a trivial P -supermodule, we write Hn(P) for Hn(P, K).
In the case when the ground ring K is a field, there is a relation between Tor
functor and the homology (see [18]) given by
By analogy to Lie algebras (see e. g. [11]), we have the following isomorphisms
Hn(P, M) ≅ TorU(P )
n
(K, M).
H0(P, M) ≅ CokerΩ(P) ⊗U(P ) M Ð→ M,
H1(P, M) ≅ KerΩ(P) ⊗U(P ) M Ð→ M.
(2)
(3)
3. Non-abelian tensor product of Lie superalgebras
In this section we introduce a non-abelian tensor product of Lie superalgebras,
which generalizes the non-abelian tensor product of Lie algebras [8], and study its
properties.
3.1. Construction of the non-abelian tensor product.
Definition 3.1. Let M and N be two Lie superalgebras with actions on each
other. Let XM,N be the Z2-graded set of all symbols m ⊗ n, where m ∈ M¯0 ∪ M¯1,
n ∈ N¯0 ∪ N¯1 and the Z2-gradation is given by m ⊗ n = m + n . We define the non-
abelian tensor product of M and N , denoted by M ⊗ N , as the Lie superalgebra
generated by XM,N and subject to the relations:
(i) λ(m ⊗ n) = λm ⊗ n = m ⊗ λn,
(ii) (m + m′) ⊗ n = m ⊗ n + m′ ⊗ n,
m ⊗(n + n′) = m ⊗ n + m ⊗ n′,
(iii) [m, m′] ⊗ n = m ⊗ m′
m ⊗[n, n′] =(−1) n′ ( m + n )(n′
(iv) [m ⊗ n, m′ ⊗ n′] = −(−1) m n (nm ⊗ m′
n −(−1) m m′ (m′ ⊗ mn),
n′),
for every λ ∈ K, m, m′ ∈ M¯0 ∪ M¯1 and n, n′ ∈ N¯0 ∪ N¯1.
m ⊗ n) −(−1) m n (nm ⊗ n′),
where m, m′ have the same grade,
where n, n′ have the same grade,
Let us remark that if m = m¯0 + m¯1 is any element of M and n = n¯0 + n¯1 is any
element of N , then under the notation m ⊗ n we mean the sum
m¯0 ⊗ n¯0 + m¯0 ⊗ n¯1 + m¯1 ⊗ n¯0 + m¯1 ⊗ n¯1.
If M = M¯0 and N = N¯0 then M ⊗ N is the non-abelian tensor product of Lie
algebras introduced and studied in [8] (see also [12]).
Definition 3.2. Actions of Lie superalgebras M and N on each other are said to
be compatible if
(i) (nm)n′ = −(−1) m n [mn, n′],
(ii) (mn)m′ = −(−1) m n [nm, m′],
for all m, m′ ∈ M¯0 ∪ M¯1 and n, n′ ∈ N¯0 ∪ N¯1.
For example, if M and N are two graded ideals of some Lie superalgebra, the
actions induced by the bracket are compatible.
8
X. GARCÍA-MARTÍNEZ, E. KHMALADZE, M. LADRA
Proposition 3.3. Let M and N be Lie superalgebras acting compatibly on each
other. Then there is a natural isomorphism of Lie superalgebras
where D(M, N) is the submodule of the supermodule M ⊗K N generated by the
elements
M ⊗ N ≅
M ⊗K N
D(M, N) ,
n +(−1) m m′ (m′ ⊗ mn),
m ⊗ n) +(−1) m n (nm ⊗ n′),
(i) [m, m′] ⊗ n − m ⊗ m′
(ii) m ⊗[n, n′] −(−1) n′ ( m + n )(n′
(iii) (nm) ⊗(mn), with m = n
(iv) (−1) m n (nm) ⊗(m′
(v) ↺
n′) +(−1)( m + n )( m′ + n′ )+ m′ n′ (n′
(−1)( m + n )( m′′ + n′′ )+ m n + m′ n′ [nm, n′
for all m, m′, m′′ ∈ M¯0 ∪ M¯1 and n, n′, n′′ ∈ N¯0 ∪ N¯1, where ↺
(m,n),(m′,n′),(m′′,n′′)
m′) ⊗(mn),
m′] ⊗(m′′
n′′),
denotes the cyclic
summation with respect to x, y, z.
Proof. There is a Lie superalgebra structure on the supermodule(M ⊗KN)~D(M, N)
given on generators by the following bracket
x,y,z
[m ⊗ n, m′ ⊗ n′] = −(−1) m n (nm ⊗ m′
n′),
for all m, m′ ∈ M¯0 ∪ M¯1, n, n′ ∈ N¯0 ∪ N¯1 and extended by linearity.
tine to check that this bracket is compatible with the defining relations of (M ⊗K
N)~D(M, N) and it indeed defines a Lie superalgebra structure. Then the canon-
ical homomorphism M ⊗ N → (M ⊗K N)~D(M, N), m ⊗ n ↦ m ⊗ n, is an isomor-
It is rou-
(cid:3)
phism.
The proof of the following proposition is a routine calculation.
Proposition 3.4. Let M and N be two Lie superalgebras acting compatibly on
each other.
(i) The following morphisms
µ∶ M ⊗ N → M, m ⊗ n ↦ −(−1) m n (nm),
ν∶ M ⊗ N → N, m ⊗ n ↦ mn,
are Lie superalgebra homomorphisms.
(ii) There are actions of M and N on M ⊗ N given by
m′(m ⊗ n) =[m′, m] ⊗ n +(−1) m m′ m ⊗(m′
n),
n′(m ⊗ n) =(n′
m) ⊗ n +(−1) n n′ m ⊗[n′, n],
for m, m′ ∈ M¯0 ∪ M¯1, n, n′ ∈ N¯0 ∪ N¯1 and extended by linearity. Moreover,
with these actions µ and ν are crossed modules of Lie superalgebras.
Lemma 2.7(ii) is a graded ideal of M (resp. N ) generated by the elements of the
We will denote by [M, N]M (resp. [M, N]N ) the image of µ (resp. ν), which by
form nm (resp. mn) for m ∈ M and n ∈ N . Note that by Lemma 2.7(iii) Ker(µ)
resp. Ker(ν) is an M~[M, N]M -supermodule (resp. N~[M, N]N -supermodule).
NON-ABELIAN TENSOR PRODUCT AND HOMOLOGY OF LIE SUPERALGEBRAS
9
3.2. Some properties of the non-abelian tensor product. The obvious ana-
logues of Brown and Loday results [1] hold for Lie superalgebras. In the following
two propositions immediately below we show that sometimes the non-abelian ten-
sor product of Lie superalgebras can be expressed in terms of the tensor product
of supermodules.
Proposition 3.5. Let M and N be Lie superalgebras acting on each other. Then
the canonical map M ⊗K N → M ⊗ N , m ⊗ n ↦ m ⊗ n, is an even, surjective
homomorphism of supermodules. In addition, if M and N act trivially on each
other, then M ⊗ N is an abelian Lie superalgebra and there is an isomorphism of
supermodules
where M ab = M~[M, M] and N ab = N~[N, N].
M ⊗ N ≅ M ab ⊗K N ab,
Proof. It is straightforward by the identities (iv), (iii) of Definition 3.1.
(cid:3)
Proposition 3.6. Let P be a Lie superalgebra and M a P -supermodule considered
as an abelian Lie superalgebra acting trivially on P . Then there is an isomorphism
of supermodules
P ⊗ M ≅ Ω(P) ⊗U(P ) M.
Proof. By Proposition 3.3 there is an isomorphism of supermodules
P ⊗ M ≅
P ⊗K M
W
,
where W is the submodule of P ⊗K M generated by all elements of the form
[p, p′] ⊗ m − p ⊗ p′m +(−1) p p′ p′ ⊗ pm
for all p, p′ ∈ P¯0 ∪ P¯1 and m ∈ M¯0 ∪ M¯1. Now by using Construction 2.11 and by
repeating the respective part of the proof of [3, Proposition 13], it is easy to see
that there is an isomorphism of supermodules
P ⊗K M
W
≅ Ω(P) ⊗U(P ) M,
which completes the proof.
(cid:3)
The non-abelian tensor product of Lie superalgebras is symmetric, in the sense
of the following proposition.
Proposition 3.7. The Lie superalgebra homomorphism
M ⊗ N → N ⊗ M, m ⊗ n ↦ −(−1) m n (n ⊗ m),
is an isomorphism.
Proof. This can be checked readily.
(cid:3)
Let us consider the category SLie2
K whose objects are ordered pairs of Lie superal-
gebras (M, N) acting compatibly on each other, and the morphisms are pairs of Lie
superalgebra homomorphisms (φ∶ M → M ′, ψ∶ N → N ′) which preserve the actions,
i.e., φ(nm) = ψ(n)φ(m) and ψ(mn) = φ(m)ψ(n). For such a pair (φ, ψ) we have a
homomorphism of Lie superalgebras φ⊗ψ∶ M ⊗N → M ′ ⊗N ′, m⊗n ↦ φ(m)⊗ψ(n).
Therefore, ⊗ is a functor from SLie2
K to the category of Lie superalgebras.
10
X. GARCÍA-MARTÍNEZ, E. KHMALADZE, M. LADRA
Given an exact sequence in SLie2
K
(0, 0) Ð→(K, L) (i,j)ÐÐ→(M, N) (φ,ψ)ÐÐÐ→(P, Q) Ð→(0, 0),
(4)
by Proposition 3.4(ii) there is a Lie superalgebra homomorphism M ⊗ L → L and
an action of N on K ⊗ N . Thus, there is an action of M ⊗ L on K ⊗ N , so we can
form the semidirect product (K ⊗ N) ⋊(M ⊗ L), and we have the following obvious
analogue of [8, Proposition 9].
Proposition 3.8. Given the short exact sequence (4), there is an exact sequence
of Lie superalgebras
(K ⊗ N) ⋊(M ⊗ L) αÐÐ→ M ⊗ N
φ⊗ψÐÐ→ P ⊗ Q Ð→ 0.
In particular, given a Lie superalgebra M and a graded ideal K of M , there is
an exact sequences of Lie superalgebras
(K ⊗ M) ⋊(M ⊗ K) → M ⊗ M →(M~K) ⊗(M~K) → 0.
(5)
3.3. Nilpotency, solvability and Engel of the non-abelian tensor product.
The results from [20] on nilpotency, solvability and Engel of the non-abelian tensor
product on Lie algebras can be easily extended to the case of Lie superalgebras.
The notions of nilpotency and solvability of Lie superalgebras are given in [18]. As
they are very similar to the respective notions for Lie algebras, we omit them. We
say that a Lie superalgebra M is n-Engel if it satisfies ad(x)n = 0 for all x ∈ M .
The proof of the following result is similar to the proof of [20, Theorem 2.2].
Theorem 3.9. Let M and N be two Lie superalgebras acting compatibly on each
other. Then,
(i) If [M, N]M is nilpotent, then M ⊗N and [M, N]N are nilpotent too. More-
over, if the nilpotency class of [M, N]M is cl([M, N]M), then
cl([M, N]M) ≤ cl(M ⊗ N) ≤ cl([M, N]M) + 1,
cl([M, N]N) ≤ cl([M, N]M) + 1.
(ii) If [M, N]M is solvable, then M ⊗ N and [M, N]N are solvable too. More-
over, if the derived length of [M, N]M is ℓ([M, N]M), then
ℓ([M, N]M) ≤ ℓ(M ⊗ N) ≤ ℓ([M, N]M) + 1,
ℓ([M, N]N) ≤ ℓ([M, N]M) + 1.
(iii) If [M, N]M is Engel, then M ⊗ N and [M, N]N are Engel too. Moreover,
if [M, N]M is n-Engel, then M ⊗ N and [M, N]N are (n + 1)-Engel.
4. Universal central extensions of Lie superalgebras
Now we use the non-abelian tensor product of Lie superalgebras to describe
universal central extensions of Lie superalgebras. Recall that a central extension
u∶ U ↠ P is universal if for any other central extension f ∶ M ↠ P there is a unique
homomorphism θ∶ U → M such that f ○ θ = u.
It is shown in [19] that a Lie
superalgebra P admits a universal central extension if and only if P is perfect, i.e.
P =[P, P].
P ↠[P, P], u(p ⊗ p′) =[p, p′], is a central extension of the Lie superalgebra [P, P].
It follows from Proposition 3.4 and Lemma 2.7(i) that the homomorphism u∶ P ⊗
NON-ABELIAN TENSOR PRODUCT AND HOMOLOGY OF LIE SUPERALGEBRAS
11
Theorem 4.1. If P is a perfect Lie superalgebra, then the central extension u∶ P ⊗
P ↠ P is the universal central extension.
Proof. Let f ∶ M ↠ P be a central extension of P . Since Ker f is in the centre of M ,
we get a well-defined homomorphism of Lie superalgebras θ∶ P ⊗ P → M given by
θ(p⊗p′) =[mp, mp′], where mp and mp′ are any preimages of p and p′, respectively.
Obviously θ ○ f = u. Since P is perfect, then by relation (iv) of Definition 3.1, so is
P ⊗ P . Then by [19, Lemma 1.4] the homomorphism θ is unique.
(cid:3)
the kernel of the universal central extension is isomorphic to the second homology
Remark. If P is a perfect Lie superalgebra, then H2(P) ≈ Ker(P ⊗ P
H2(P) (see [19]).
sl(n, A), where A is a unital associative algebra, is the Steinberg algebra st(n, A),
It is a classical result that the universal central extension of the Lie algebra
when n ≥ 5 (see e. g. [16]). Recently, in [5, 9], this result has been extended to Lie
superalgebras. Below, using the non-abelian tensor product of Lie superalgebras,
we propose an alternative proof of the same result.
uÐ→ P), since
First we recall from [5] that, given a unital associative superalgebra A, the Lie
a perfect Lie superalgebra. This guarantees the existence of the universal central
superalgebra sl(m, n, A), m + n ≥ 3, is defined to be the subalgebra of the Lie
superalgebra gl(m, n, A) see Example 2.2 (ii) generated by the elements Eij(a),
1 ≤ i ≠ j ≤ m + n, a ∈ A¯0 ∪ A¯1. It is shown in [5, Lemma 3.3] that sl(m, n, A) is
extension of sl(m, n, A).
The Steinberg Lie superalgebra st(m, n, A) is defined for m + n ≥ 3 to be the Lie
superalgebra generated by the homogeneous elements Fij(a), where 1 ≤ i ≠ j ≤ m+n,
a ∈ A is a homogeneous element and the Z2-grading is given by Fij(a) = i + j + a ,
subject to the following relations:
a ↦ Fij(a) is a K-linear map,
[Fij(a), Fjk(b)] = Fik(ab), for distinct i, j, k,
[Fij(a), Fkl(b)] = 0, for j ≠ k, i ≠ l.
Theorem 4.2 ([5]). If m + n ≥ 5, then the canonical epimorphism
st(m, n, A) ↠ sl(m, n, A), Fij(a) ↦ Eij(a),
is the universal central extension of the perfect Lie superalgebra sl(m, n, A).
Proof. We claim that there is an isomorphism of Lie superalgebras
st(m, n, A) ≅ st(m, n, A) ⊗ st(m, n, A).
Indeed, one can readily check that the maps
st(m, n, A) Ð→ st(m, n, A) ⊗ st(m, n, A), Fij(a) ↦ Fik(a) ⊗ Fkj(1) for k ≠ i, j,
st(m, n, A) ⊗ st(m, n, A) Ð→ st(m, n, A), Fij(a) ⊗ Fkl(b) ↦[Fij(a), Fkl(b)],
inverses to each other. Since st(m, n, A) is a perfect Lie superalgebra, then Theo-
are well-defined homomorphisms of Lie superalgebras if m + n ≥ 5, and they are
rem 4.1 and [19, Corollary 1.9] complete the proof.
(cid:3)
12
X. GARCÍA-MARTÍNEZ, E. KHMALADZE, M. LADRA
5. Non-abelian homology of Lie superalgebras
The low-dimensional non-abelian homology of Lie algebras with coefficients in
crossed modules was defined in [10] and it was extended to all dimensions in [12]. In
this section we extend to Lie superalgebras the construction of zero and first non-
abelian homologies. We also relate the non-abelian homology of Lie superalgebras
with the cyclic homology of associative superalgebras studied in [13, 15].
5.1. Construction of the non-abelian homology and some properties. Let
Lie superalgebras over P (crossed P -modules, for short), whose objects are crossed
P be a Lie superalgebra. We denote by Cross(P) the category of crossed modules of
modules (M, ∂) ≡ (∂∶ M → P) and a morphism from (M, ∂) to (N, ∂′) is a Lie
superalgebra homomorphism f ∶ M → N such that f(pm) = pf(m) for all p ∈ P ,
m ∈ M and ∂′ ○ f = ∂. By an exact sequence (L, ∂′′) fÐ→ (M, ∂) gÐ→ (N, ∂′) in
Cross(P) we mean that the sequence of Lie superalgebras L
gÐ→ N is exact.
Lemma 5.1. Given a short exact sequence in Cross(P)
fÐ→ M
0 →(L, ∂′′) fÐ→(M, ∂) gÐ→(N, ∂′) → 0,
the morphism ∂′′∶ L → P is trivial and L is an abelian Lie superalgebra.
and N act compatibly on each other via the action of P . Thus, we can construct
the non-abelian tensor product of Lie superalgebras M ⊗ N . Moreover, we have
Proof. Clearly ∂′′ = ∂′ ○ g ○ f = 0 and [l, l′] = ∂ ′′(l)l′ = 0, for all l, l′ ∈ L.
If (M, ∂) and (N, ∂′) are two crossed P -modules, then the Lie superalgebras M
an action of P on M ⊗ N defined by p(m ⊗ n) = pm ⊗ n +(−1) p m m ⊗ pn, and
straightforward computations show that η∶ M ⊗ N → P , m ⊗ n ↦ [∂(m), ∂′(n)], is
Proposition 5.2. Let (M, ∂) be a crossed P -module. There is a right exact functor
(M ⊗ −)∶ Cross(P) → Cross(P) given, for any crossed P -module (N, ∂′), by
a crossed P -module.
(cid:3)
(M ⊗ −)(N, ∂′) =(M ⊗ N, η).
Proof. It is an immediate consequence of Proposition 3.8
(cid:3)
Definition 5.3. Let (M, ∂) be a crossed P -module. We define the zero and first
non-abelian homologies of P with coefficients in M by setting
H0(P, M) = Coker ν
and
H1(P, M) = Ker ν,
where ν∶ P ⊗ M → M , p ⊗ m ↦ pm, is the Lie superalgebra homomorphism as in
Proposition 3.4.
If we consider the crossed P -module (P, idP) we have that
H0(P, P) =
P
[P, P] ≅ H1(P).
In addition, if P is perfect, by Theorem 4.1 we have that H1(P, P) ≅ H2(P).
The zero and first non-abelian homologies generalize respectively the zero and
first homologies of Lie superalgebras in the sense of the following proposition.
NON-ABELIAN TENSOR PRODUCT AND HOMOLOGY OF LIE SUPERALGEBRAS
13
Proposition 5.4. Let the ground ring K be a field. Let P be a Lie superalgebra
and M a P -supermodule thought as a crossed P -module (M, 0). Then there are
isomorphisms of super vector spaces
H0(P, M) ≅ H0(P, M)
and H1(P, M) ≅ H1(P, M).
Proof. This is a direct consequence of Proposition 3.6 and the isomorphisms (2)
and (3).
(cid:3)
Proposition 5.5. Given a short exact sequence in Cross(P)
0 →(L, 0) →(M, ∂) →(N, ∂′) → 0
we have an exact sequence of supermodules
H1(P, L) → H1(P, M) → H1(P, N) → H0(P, L) → H0(P, M) → H0(P, N) → 0.
Proof. The proof is an immediate consequence of the snake lemma applied to the
diagram obtained from Proposition 5.2
P ⊗ L
/ P ⊗ M
/ P ⊗ N
/ 0
0
/ L
/ M
/ N
/ 0.
(cid:3)
5.2. Application to the cyclic homology of associative superalgebras. Now
we recall from [15] and [13] the definition of cyclic homology of associative superal-
gebras. Let A be an associative superalgebra and (C′
n(A) = A⊗K(n+1) and the boundary map dn∶ C′
∗) denote its Hochschild
n−1(A) is
n(A) → C′
∗(A), d′
complex, that is C′
given by
d′
n(a0 ⊗ ⋯ ⊗ an) =
n−1
Q
i=0(−1)ia0 ⊗ ⋯ ⊗ aiai+1 ⊗ ⋯ ⊗ an
+(−1)n+ an ( a0 +⋯+ an−1 )ana0 ⊗ ⋯ ⊗ an−1.
Now the cyclic group Z~(n + 1)Z acts on A⊗K(n+1) via
tn(a0 ⊗ ⋯ ⊗ an) =(−1)n+ an ∑k<n ak an ⊗ a0 ⊗ ⋯ ⊗ an−1,
where tn = 1 + (n + 1)Z ∈ Z~(n + 1)Z. For each n ≥ 0, consider the quotient
Cn(A) = A⊗K(n+1)~ Im(1−tn) which is the module of coinvariants of C ′
n(A) under the
Z~(n + 1)Z-action. Then d′
n induces a well-defined map dn∶ Cn(A) → Cn−1(A) and
there is an induced chain complex (C∗(A), d∗), which is called the Connes complex
superalgebra A, denoted by HCn(A), n ≥ 0.
Easy calculations show that, given an associative superalgebra A, HC1(A) is the
Its homologies are, by definition, the cyclic homologies of the associative
kernel of the homomorphism of supermodules
of A.
(A ⊗K A)~ I(A) →[A, A],
a ⊗ b ↦ ab −(−1) a b ba,
where[A, A] is the graded submodule of A generated by the elements ab−(−1) a b ba
and I(A) is the graded submodule of the supermodule A ⊗K A generated by the
/
/
/
/
/
/
/
14
X. GARCÍA-MARTÍNEZ, E. KHMALADZE, M. LADRA
elements
a ⊗ b +(−1) a b b ⊗ a,
ab ⊗ c − a ⊗ bc +(−1) c ( a + b )ca ⊗ b,
for all homogeneous a, b, c ∈ A.
Now let us consider A as a Lie superalgebra see Example 2.2(i). Then there
is a Lie superalgebra structure on (A ⊗K A)~ I(A) given by
[a ⊗ b, a′ ⊗ b′] =[a, b] ⊗[a′, b′]
for all a, a′, b, b′ ∈ A. We denote this Lie superalgebra by V(A). In fact, V(A) is
by the elements x ⊗ y +(−1) x y y ⊗ x and xy ⊗ z − x ⊗ yz +(−1) z ( x + y )zx ⊗ y, for
the quotient of the non-abelian tensor product A ⊗ A by the graded ideal generated
all homogeneous x, y, z ∈ A.
Proposition 5.6. Let A be a Lie superalgebra. Then the following assertions hold:
other.
(i) There are compatible actions of the Lie superalgebras A and V(A) on each
(ii) The map µ∶ V(A) → A given by x ⊗ y ↦ [x, y], together with the action of
(iii) The action of A on V(A) induces the trivial action of A on HC1(A).
(iv) There is a short exact sequence in the category Cross(A)
0 →(HC1(A), 0) →(V(A), µ) →([A, A], i) → 0,
A on V(A), is a crossed module of Lie superalgebras.
where i∶[A, A] → A is the inclusion.
Proof.
(i) The action of A on V(A) is induced by the action of A on A ⊗ A given in
Proposition 3.4(ii), that is
a(x ⊗ y) =[a, x] ⊗ y +(−1) a x x ⊗[a, y]
= ax ⊗ y +(−1) a ( x + y )x ⊗ ya −(−1) x a x ⊗ ay −(−1) x a xa ⊗ y
= a ⊗ xy −(−1) x y a ⊗ yx
= a ⊗[x, y],
whilst the action of V(A) on A is defined by
x⊗ya =[x, y], a(cid:6)
for all homogeneous a, x, y ∈ A. Straightforward calculations show that these are
indeed (compatible) actions of Lie superalgebras.
(ii) Since the crossed module of Lie superalgebras A ⊗ A → A, x ⊗ y ↦ [x, y],
given in Proposition 3.4, vanishes on the elements of the form x ⊗ y +(−1) x y y ⊗ x
and xy ⊗ z − x ⊗ yz +(−1) z ( x + y )zx ⊗ y, then µ is well defined and obviously it is
(iii) If ∑i λi(xi ⊗ yi) ∈ HC1(A), i.e. ∑i λi[xi, yi] = 0, then for all a ∈ A we have
a crossed module of Lie superalgebras.
a Q
i
λi(xi ⊗ yi) = Q
i
λi(a ⊗[xi, yi]) = a ⊗ Q
λi[xi, yi] = 0.
i
(iv) This is an immediate consequence of the assertions above.
(cid:3)
NON-ABELIAN TENSOR PRODUCT AND HOMOLOGY OF LIE SUPERALGEBRAS
15
By Proposition 5.5 we have the following exact sequence of supermodules
H1A, HC1(A)
H0A, HC1(A)
H1A, V(A)
H0A, V(A)
H1(A,[A, A])
H0(A,[A, A])
(6)
0.
Below, we will calculate some of the terms of this exact sequence. At first, by
analogy to the Dennis-Stein generators [6], we give a definition of the first Milnor
cyclic homology for associative superalgebras.
Definition 5.7. Let A be an associative superalgebra. We define the first Milnor
cyclic homology HCM
the graded ideal generated by the elements
1 (A) of A to be the quotient of the supermodule A ⊗K A by
a ⊗ b +(−1) a b b ⊗ a,
ab ⊗ c − a ⊗ bc +(−1) c ( a + b )ca ⊗ b,
a ⊗ bc −(−1) b c a ⊗ cb,
for all homogeneous a, b, c ∈ A.
1 (A).
Lemma 5.8. We have the following equalities and isomorphisms
It is clear that if A is supercommutative, that is, ab = (−1) a b ba, for all homo-
geneous a, b ∈ A, then HC1(A) ≅ HCM
(i) H0A, HC1(A) = HC1(A),
(ii) H1A, HC1(A) ≅ A~[A, A] ⊗K HC1(A),
(iii) H0(A,[A, A]) =[A, A]~A,[A, A](cid:6),
(iv) H0A, V(A) ≅ HCM
(i) Since A acts trivially on HC1(A), we have that CokerA⊗HC1(A) → HC1(A) =
HC1(A).
(ii) Since HC1(A) is abelian, by Proposition 3.5 we have that KerA⊗HC1(A) →
HC1(A) ≅ A~[A, A] ⊗K HC1(A).
1 (A).
Proof.
(iii) and (iv) are straightforward.
(cid:3)
It follows that the exact sequence (6) can be written as in the following theorem.
Theorem 5.9. If A is a unital associative superalgebra. Then there is an exact
sequence of supermodules
A
[A, A] ⊗K HC1(A)
HC1(A)
H1A, V(A)
1 (A)
HCM
H1(A,[A, A])
[A, A]
A,[A, A](cid:6)
0.
Corollary 5.10. If A is perfect as a Lie superalgebra, we have an exact sequence
0 → H1A, V(A) → H2(A) → HC1(A) → 0,
16
X. GARCÍA-MARTÍNEZ, E. KHMALADZE, M. LADRA
where H2(A) is the usual second homology of the Lie superalgebra A. If in addition
H2(A) = 0, then all terms of the exact sequence in the previous theorem are trivial.
Proof. Since A is perfect we know that H1(A, A) ≅ H2(A), A~[A, A]⊗K HC1(A) = 0
and the map A ⊗ V(A) → V(A) is surjective.
(cid:3)
6. Non-abelian exterior product of Lie superalgebras
In this section we extend to Lie superalgebras the definition of the non-abelian
exterior product of Lie algebras introduced in [8]. Then we use it to derive the Hopf
formula for the second homology of a Lie superalgebra and to construct a six-term
exact homology sequence of Lie superalgebras.
6.1. Construction of the non-abelian exterior product. Let P be a Lie su-
peralgebra and (M, ∂) and (N, ∂ ′) two crossed P -modules. We consider the actions
of M and N on each other via P .
Lemma 6.1. Let M ◻ N be the graded submodule of M ⊗ N generated by the
elements
(a) m ⊗ n +(−1) m′ n′ m′ ⊗ n′, where ∂(m) = ∂ ′(n′) and ∂(m′) = ∂ ′(n),
(b) m¯0 ⊗ n¯0, where ∂(m¯0) = ∂ ′(n¯0),
with m, m′ ∈ M¯0 ∪ M¯1, n, n′ ∈ N¯0 ∪ N¯1, m¯0 ∈ M¯0 and n¯0 ∈ N¯0. Then, M ◻ N is a
graded ideal in the centre of M ⊗ N .
Proof. Given an element m ⊗ n +(−1) m′ n′ m′ ⊗ n′ of the form (a), suppose that
m′ = n , then we have
[x ⊗ y, m ⊗ n +(−1) m′ n′ m′ ⊗ n′] = −(−1) x y (yx) ⊗mn +(−1) m′ n′ (m′
n′)
= −(−1) x y (yx) ⊗∂(m)n +(−1) m′ n′ (∂(m′)n′)
= −(−1) x y (yx) ⊗∂ ′(n′)n +(−1) m′ n′ (∂ ′(n)n′)
= −(−1) x y (yx) ⊗[n′, n] +(−1) n n′ [n, n′]
= 0.
This is also true when m′ ≠ n . Indeed, if m′ ≠ n , since ∂, ∂ ′ are even maps, the
equality ∂(m) = ∂ ′(n′) holds if and only if ∂(m) = 0 = ∂ ′(n′). Now take an element
m¯0 ⊗ n¯0 of the form (b). Then we have
[x ⊗ y, m¯0 ⊗ n¯0] = −(−1) x y (yx) ⊗(m¯0 n¯0)
= −(−1) x y (yx) ⊗(∂(m¯0)n¯0)
= −(−1) x y (yx) ⊗(∂(n¯0)n¯0)
= −(−1) x y (yx) ⊗[n¯0, n¯0]
= 0,
for any x ⊗ y ∈ M ⊗ N . This completes the proof.
(cid:3)
Definition 6.2. Let P be a Lie superalgebra and (M, ∂) and (N, ∂ ′) two crossed
P -modules. The non-abelian exterior product M ∧ N of the Lie superalgebras M
and N is defined by
M ∧ N =
M ⊗ N
M ◻ N
.
The equivalence class of m ⊗ n will be denoted by m ∧ n.
NON-ABELIAN TENSOR PRODUCT AND HOMOLOGY OF LIE SUPERALGEBRAS
17
Note that if M = M¯0 and N = N¯0 then M ∧ N coincides with the non-abelian
exterior product of Lie algebras [8].
Reviewing Section 3, one can easily check that most of results on the non-abelian
tensor product are fulfilled for the non-abelian exterior product. In particular, there
are homomorphisms of Lie superalgebras M ∧ N → M , M ∧ N → N and actions of
M and N on M ∧N , induced respectively by the homomorphisms and actions given
in Proposition 3.4. It is also satisfied the isomorphism M ∧ N ≅ N ∧ M . Further,
given a short exact sequence of Lie superalgebras 0 → K → M → P → 0, as an
exterior analogue of the exact sequence (5), we get the following exact sequence of
Lie superalgebras
(7)
Given a Lie superalgebra M , since id∶ M → M is a crossed module, we can
K ∧ M → M ∧ M → P ∧ P → 0.
consider M ∧ M . It is the quotient of M ⊗ M by the following relations
m ∧ m′ = −(−1) m m′ m′ ∧ m,
m¯0 ∧ m¯0 = 0,
for all m, m′ ∈ M¯0 ∪ M¯1 and m¯0 ∈ M¯0. In the particular case when M is perfect,
it is easy to see that M ◻ M = 0, so M ∧ M ≅ M ⊗ M and in Theorem 4.1 we can
replace M ⊗ M by M ∧ M .
6.2. A six term exact homology sequence. In [7], the non-abelian exterior
product of Lie algebras is used to construct a six-term exact sequence of homology
of Lie algebras. In this section we will extend these results to Lie superalgebras.
First of all, we prove an analogue of Miller's theorem [17] on free Lie superalge-
bras extending the similar result obtained in [7] for Lie algebras.
Then the homomorphism F ∧ F → F , x ∧ y ↦ xy is injective.
Proposition 6.3. Let F = F(X) be the free Lie superalgebra on a graded set X.
Proof. Let us prove that [F, F] ≅ F ∧ F . Using the same notations as in Construc-
tion 2.9, we define a map φ∶ alg(X) ∗ alg(X) → F ∧ F by ∑i λixiyi ↦ ∑i λi(xi ∧ yi),
where alg(X) ∗ alg(X) is the free product of superalgebras. It is easy to see that φ
is a K-superalgebra homomorphism since [x ∧ y, x′ ∧ y′] = xy ∧ x′y′. The ideal I is
contained in alg(X) ∗ alg(X) and by using the defining relations of F ∧ F it is not
difficult to check that φ vanishes on I. So we have an induced map from [F, F] to
F ∧ F , which is inverse to the homomorphism F ∧ F →[F, F], x ∧ y ↦ xy.
Let P be a Lie superalgebra and take the quotient supermodule (P ∧K P)~ Im d3,
Here K is considered as a trivial P -module. We define a bracket in (P ∧K P)~ Im d3
K(P) is the boundary map in the homology complex(C∗(P, K), d∗).
(cid:3)
where d3∶ ⋀3
K(P) → ⋀2
by setting
[x ∧ y, x′ ∧ y′] =[x, y] ∧[x′, y′]
for all x, y ∈ P . As a particular case of the exterior analogue of Proposition 3.3 we
have
Lemma 6.4. There is an isomorphism of Lie superalgebras
P ∧K P
Im d3
≈ P ∧ P.
Corollary 6.5.
18
X. GARCÍA-MARTÍNEZ, E. KHMALADZE, M. LADRA
(i) For any Lie superalgebra P there is an isomorphism of supermodules
H2(P) ≅ Ker(P ∧ P → P).
(ii) H2(F) = 0 if F is a free Lie superalgebra.
(iii) (Hopf Formula) Given a free presentation 0 → R → F → P → 0 of a Lie
superalgebra P , there is an isomorphism of supermodules
H2(P) ≅
Proof.
R ∩[F, F]
[F, R]
.
(i) This follows immediately from Lemma 6.4.
(ii) This is a consequence of (i) and Proposition 6.3.
(iii) Since F ∧ F ≅[F, F], using the exact sequence (7), we have
P ∧ P ≅ [F, F]
[F, R] .
/ 0
/ 0.
Then Lemma 6.4 completes the proof.
(cid:3)
Theorem 6.6. Let M be a graded ideal of a Lie superalgebra P . Then there is an
exact sequence
Ker(P ∧ M → P) → H2(P) → H2(P~M) →
M
[P, M] → H1(P) → H1(P~M) → 0.
Proof. By using the exact sequence (7) we have the following commutative diagram
of Lie superalgebras with exact rows
M ∧ P
/ P ∧ P
P
M
∧
P
M
0
/ M
/ P
P
M
Since Coker(M ∧ P ≅ P ∧ M → M) ≅ M~[P, M] and Coker(P ∧ P → P) ≅ P~[P, P] ≅
H1(P), then the assertion follows by using snake lemma and Corollary 6.5(i). (cid:3)
In particular, if P is a Lie algebra and M is an ideal of P , then this sequence
coincides with the six-term exact sequence in the homology of Lie algebras obtained
in [7].
Acknowledgements
The authors wish to thank the anonymous referee for his help in improving the
presentation of this paper. The authors were supported by Ministerio de Economía
y Competitividad (Spain), grant MTM2013-43687-P (European FEDER support
included). The first and third authors were also supported by Xunta de Gali-
cia, grant GRC2013-045 (European FEDER support included). The first author
was also supported by FPU scholarship, Ministerio de Educación, Cultura y De-
porte (Spain). The second author was also supported by Xunta de Galicia, grant
EM2013/016 (European FEDER support included) and Shota Rustaveli National
Science Foundation, grant DI/12/5-103/11.
/
/
/
/
/
/
/
/
/
NON-ABELIAN TENSOR PRODUCT AND HOMOLOGY OF LIE SUPERALGEBRAS
19
References
[1] R. Brown, J.-L. Loday, Van Kampen theorems for diagrams of spaces, Topology 26 (3) (1987)
311 -- 335, with an appendix by M. Zisman.
[2] C. Carmeli, L. Caston, R. Fioresi, Mathematical foundations of supersymmetry, EMS Series
of Lectures in Mathematics, European Mathematical Society (EMS), Zürich, 2011.
[3] J. M. Casas, M. Ladra, Perfect crossed modules in Lie algebras, Comm. Algebra 23 (5) (1995)
1625 -- 1644.
[4] J. Castiglioni, X. García-Martínez, M. Ladra, Universal central extensions of Lie -- Rinehart
algebras, arXiv:1403.7159 (2014).
[5] H. Chen, J. Sun, Universal central extensions of slm n over Z~2Z-graded algebras, J. Pure
Appl. Algebra 219 (2015) 4278 -- 4294.
[6] R. K. Dennis, M. R. Stein, K2 of discrete valuation rings, Advances in Math. 18 (2) (1975)
182 -- 238.
[7] G. J. Ellis, Nonabelian exterior products of Lie algebras and an exact sequence in the homol-
ogy of Lie algebras, J. Pure Appl. Algebra 46 (2-3) (1987) 111 -- 115.
[8] G. J. Ellis, A nonabelian tensor product of Lie algebras, Glasgow Math. J. 33 (1) (1991)
101 -- 120.
[9] X. García-Martínez, M. Ladra, Universal central extensions of sl(m, n, a) of small rank over
associative superalgebras, arXiv:1405.4035 (2014).
[10] D. Guin, Cohomologie des algèbres de Lie croisées et K-théorie de Milnor additive, Ann. Inst.
Fourier (Grenoble) 45 (1) (1995) 93 -- 118.
[11] P. J. Hilton, U. Stammbach, A course in homological algebra, vol. 4 of Graduate Texts in
Mathematics, 2nd ed., Springer-Verlag, New York, 1997.
[12] N. Inassaridze, E. Khmaladze, M. Ladra, Non-abelian homology of Lie algebras, Glasg. Math.
J. 46 (2) (2004) 417 -- 429.
[13] K. Iohara, Y. Koga, Second homology of Lie superalgebras, Math. Nachr. 278 (9) (2005)
1041 -- 1053.
[14] V. G. Kac, Lie superalgebras, Advances in Math. 26 (1) (1977) 8 -- 96.
[15] C. Kassel, A Künneth formula for the cyclic cohomology of Z~2-graded algebras, Math. Ann.
275 (4) (1986) 683 -- 699.
[16] C. Kassel, J.-L. Loday, Extensions centrales d'algèbres de Lie, Ann. Inst. Fourier (Grenoble)
32 (4) (1982) 119 -- 142.
[17] C. Miller, The second homology group of a group; relations among commutators, Proc. Amer.
Math. Soc. 3 (1952) 588 -- 595.
[18] I. M. Musson, Lie superalgebras and enveloping algebras, vol. 131 of Graduate Studies in
Mathematics, American Mathematical Society, Providence, RI, 2012.
[19] E. Neher, An introduction to universal central extensions of Lie superalgebras, in: Groups,
rings, Lie and Hopf algebras (St. John's, NF, 2001), vol. 555 of Math. Appl., Kluwer Acad.
Publ., Dordrecht, 2003, pp. 141 -- 166.
[20] A. R. Salemkar, H. Tavallaee, H. Mohammadzadeh, B. Edalatzadeh, On the non-abelian
tensor product of Lie algebras, Linear Multilinear Algebra 58 (3-4) (2010) 333 -- 341.
[21] M. Scheunert, The theory of Lie superalgebras. An introduction, vol. 716 of Lecture Notes in
Mathematics, Springer, Berlin, 1979.
[22] J. Tanaka, On homology and cohomology of Lie superalgebras with coefficients in their finite-
dimensional representations, Proc. Japan Acad. Ser. A Math. Sci. 71 (3) (1995) 51 -- 53.
[23] T. Zhang, Z. Liu, Omni-Lie superalgebras and Lie 2-superalgebras, Front. Math. China 9 (5)
(2014) 1195 -- 1210.
Xabier García-Martínez: Department of Algebra, IMAT, University of Santiago
de Compostela, 15782 Santiago de Compostela, Spain
E-mail address: [email protected]
Emzar Khmaladze: A. Razmadze Mathematical Institute of Tbilisi State University,
Tamarashvili Str. 6, 0177 Tbilisi, Georgia
E-mail address: [email protected]
20
X. GARCÍA-MARTÍNEZ, E. KHMALADZE, M. LADRA
Manuel Ladra: Department of Algebra, IMAT, University of Santiago de Com-
postela, 15782 Santiago de Compostela, Spain
E-mail address: [email protected]
|
1611.10143 | 1 | 1611 | 2016-11-30T13:39:40 | Horadam Octonions | [
"math.RA"
] | In this paper, first we define Horadam octonions by Horadam sequence which is a generalization of second order recurrence relations. Also, we give some fundamental properties involving the elements of that sequence. Then, we obtain their Binet-like formula, ordinary generating function and Cassini identity. | math.RA | math |
HORADAM OCTONIONS
Adnan KARATAS¸ and Serpil HALICI
Abstract. In this paper, first we define Horadam octonions by Horadam
sequence which is a generalization of second order recurrence relations.
Also, we give some fundamental properties involving the elements of that
sequence. Then, we obtain their Binet-like formula, ordinary generating
function and Cassini identity.
Mathematics Subject Classification (2010). 11B39, 17A20.
Keywords. Fibonacci Numbers and Generalization, Octonions.
1. Introduction
Four dimensional quaternion algebra is defined by W. R. Hamilton in 1843.
After the definition of quaternion algebra, J. T. Graves asked Hamilton about
higher dimensional algebras. He tried to define eight and sixteen dimensional
algebras, but Graves encountered zero divisors on sixteen dimensional al-
gebra. Because of zero divisors, Graves defined only eight dimensional real
octonion algebra in 1843. Besides Graves' definition, A. Cayley defined also
octonions in 1845. For detailed knowledge reader referred to [2]. Octonion
algebra is eight dimensional, non-commutative, non-associative and normed
division algebra. This algebra is used in various physical problems such as;
supersymmetry, super gravity and super strings. In addition octonions are
used in many subjects such as, unified fields of dyons and gravito-dyons,
electrodynamic, electromagnetism . For background the reader can find their
properties in [2, 4, 5, 6, 7, 8].
The octonion algebra O form an eight-dimensional real algebra with a basis
{1, e1, e2, . . . , e7}. Addition operation is component-wise and multiplication
operation of base elements can be made by Cayley-Dickson process with
quaternions or Fano plane or multiplication table. For detailed information,
2
Adnan KARATAS¸ and Serpil HALICI
see [2].
Now, we give some fundamental definitions for the octonion algebra. Let
x be any octonion, i.e;
x =
7
Xi=0
kiei,
ki ∈ R
where e0, e1, . . . , e7 are base elements. Conjugate of any octonion x can be
defined as
x = k0 −
7
Xi=1
kiei.
One can easily see that the conjugate of an octonion x is an involution of
first kind [19]. With the help of the conjugate, norm of an octonion x can be
defined as follows,
N r(x) = x ◦ x = x ◦ x = k2
0 + k2
1 + ··· + k2
7 .
To reader who wants to make further readings on octonions, we refer to [2,
7, 20].
One can find many studies involving sequences with positive integers which
satisfy a recursive relation in the literature. And many paper are dedicated
to Fibonacci sequences and their generalization.
In this study we generalize Fibonacci octonions and we investigate their fun-
damental properties. In first section, we give definition of octonions. In second
section, we give definition of Horadam numbers which is generalization of Fi-
bonacci numbers. In third section, we introduce Horadam octonions which
contains Fibonacci, Lucas, Pell, Jacobsthal and all other second order se-
quences. For this newly defined sequence we give the generating function,
norm value, Cassini identity, a summation formula. As we give identities and
properties mentioned above we present connections to earlier studies.
2. HORADAM NUMBERS
The famous Fibonacci numbers are second order recursive relation and used
in various disciplines. Some lesser known second order recursive relations are
Lucas numbers, Pell numbers, Jacobsthal numbers, etc.. Many mathemati-
cians tried to generalize these second order recursive relations. Then Horadam
generalized these relations and this generalization is named as Horadam se-
quence. Horadam sequences are firstly defined for generalization of Fibonacci
and Lucas recurrences as Un = Un−1 + Un−2 in [13] and this version of Ho-
radam numbers studied by Koshy in [18]. Then this definition is altered such
that it includes other integer sequences like Jacobstal numbers. We note that
there are many identities and properties about Horadam sequence [14, 15, 16].
HORADAM OCTONIONS
3
Now, let us recall the definition of Horadam numbers. In [15], Horadam num-
bers are defined as
{wn(a, b; p, q)} : wn = pwn−1 + qwn−2; w0 = a, w1 = b,
(n ≥ 2)
(2.1)
where a, b, p, q are integers. Let us give four important properties that are
needed. Firstly, Binet formula of Horadam sequence can be calculated using
its characteristic equation. The roots of characteristic equation are
α =
p +pp2 + 4q
2
,
β =
p −pp2 + 4q
2
.
(2.2)
Using these roots and the recurrence relation Binet formula can be given as
follows
wn =
Aαn − Bβn
α − β
,
A = b − aβ and B = b − aα.
(2.3)
(2.4)
(2.5)
Secondly, the generating function for Horadam numbers is
g(t) =
w0 + (w1 − pw0)t
.
1 − pt − qt2
Thirdly, the Cassini identity for Horadam numbers is
n = qn−1(pw0w1 − w2
Lastly, a summation formula for Horadam numbers is
wn+1wn−1 − w2
1 − w2
0q).
n
Xi=0
wi =
w1 − w0(p − 1) + qwn − wn+1
1 − p − q
.
(2.6)
Horadam firstly defined Horadam numbers on R and then defined Horadam
numbers on C and H [14]. In addition, Halici gave a very complete survey
about Horadam quaternions in [9].
More recently octonions have been studied by many authors. For example,
the Fibonacci octonions, Pell octonions and Modified Pell octonions appeared
in [3, 17, 22]. There are many studies about Fibonacci numbers over dual oc-
tonions and generalized octonions [11, 21, 23]. Also, some representations of
Fibonacci octonions are considered, for example see [12].
In the next section we define a new octonion sequence with Horadam com-
ponents which is generalization of earlier studies, for example, see [3, 11, 12,
17, 21, 22, 23]. And then we give some fundamental properties and identities
related with this sequences.
3. HORADAM OCTONIONS
In this section we introduce Horadam octonions. Horadam octonions are octo-
nions with coefficients from Horadam sequence and we inspired the idea from
[9]. In [9], author studied Binet formula, generating function, Cassini iden-
tity, summation formula and norm value for Horadam quaternions. Similarly
we define Horadam octonions and we investigate Binet formula, generating
4
Adnan KARATAS¸ and Serpil HALICI
function, Cassini identity, summation formula and norm value for Horadam
octonions. Horadam octonions can be shown as follows
OGn = wne0 + wn+1e1 + ··· + wn+7e7
where wn as in equation (2.1). Because of its coefficients this new octonion
can be called Horadam octonion. After some necessary calculations we acquire
the following recurrence relation;
OGn+1 = pOGn + qOGn−1.
In this section we give Binet formula, generating function, Cassini identities,
summation and norm of these octonions. Binet formula is very useful for find-
ing desired Horadam octonion and this formula takes part at many theorems'
proof.
The Cassini identity yields many fascinating by-products and this formula is
used to establish interesting results concerning with some sequences.
Theorem 3.1. The Binet formula for Horadam octonions is
OGn =
Aααn − Bββn
α − β
,
(3.1)
where α = 1e0 + αe1 + α2e2 +···+ α7e7 and β = 1e0 + βe1 + β2e2 +···+ β7e7.
Proof. Binet formula for Horadam octonions can be calculated similar to
Binet formula for Horadam sequence. By using characteristic equation
The roots of characteristic equation is
t2 − pt − q = 0.
α =
p +pp2 + 4q
2
,
β =
p −pp2 + 4q
2
.
(3.2)
(3.3)
Using these roots and the recurrence relation Binet formula can be written
as follows
OGn =
Aααn − Bββn
α − β
(3.4)
where
α = 1e0 +αe1 +α2e2 +···+α7e7 and β = 1e0 +βe1 +β2e2 +···+β7e7. (3.5)
In the Binet formula of Horadam octonions if we take A = B = 1 and
calculating the value of α, β then we have (3.2)
(cid:3)
M Pn =
.
(3.6)
ααn + ββn
α − β
We should note that the equation (3.6) is Binet formula of Modified Pell
octonions which is given by Catarino in [3]. Also, if we take A = B = 1, and
calculate the value α, β with respect to its recurrence relation, we have Binet
HORADAM OCTONIONS
5
formula for Fibonacci octonions which is given by Ke¸cilioglu and Akku¸s in
[17].
In the following theorem we give the generating function for Horadam octo-
nions.
Theorem 3.2. The generating function for Horadam octonions is
OG0 + (OG1 − pOG0)t
1 − pt − qt2
.
(3.7)
Proof. To prove this claim, firstly, we need to write generating function for
Horadam octonions;
g(t) = OG0t0 + OG1t + OG2t2 + ··· + OGntn + . . .
(3.8)
Secondly, we need to calculate ptg(t) and qt2g(t) as the following equations;
ptg(t) =
∞
Xn=0
pOGntn+1 and qt2g(t) =
∞
Xn=0
qOGntn+2.
(3.9)
Finally, if we made necessary calculations, then we get the following equation
which is the generating function for Horadam octonions
g(t) =
OG0 + (OG1 − pOG0)t
1 − pt − qt2
(3.10)
(cid:3)
Using the initial values for Modified Pell octonions and the equation
(3.7), we obtain
g(t) =
M P O0 + (M P O1 − 2M P O0)t
.
(3.11)
1 − 2t − t2
In fact the formula (3.11) is the generating function for Modified Pell oc-
tonions given by Catarino in [3]. In addition if we take initial values of
Fibonacci octonions and calculate equation (3.7) accordingly then we get
generating function for Fibonacci octonions. So, generating function for Ho-
radam octonions generalizes Fibonacci octonions, Lucas, octonions, Modified
Pell octonions, etc..
In the following theorem, we state two different Cassini identities which occur
from non-commutativity of octonion multiplication.
Theorem 3.3. For Horadam octonions the Cassini formulas are as follows
i) OGn−1 ◦ OGn+1 − OG2
n =
ii) OGn+1 ◦ OGn−1 − OG2
n =
AB(αβ)n−1(βαβ − αβα)
,
(3.12)
AB(αβ)n−1(ββα − ααβ)
.
(3.13)
α − β
α − β
6
Adnan KARATAS¸ and Serpil HALICI
Proof. We use Binet formula in order to prove equation (3.12)
Aααn−1 − Bββn−1
=
α − β
−(cid:18) Aααn − Bββn
(cid:19)2
.
α − β
OGn+1 ◦ OGn−1 − OG2
Aααn+1 − Bββn+1
n =
α − β
If necessary calculations are made, we obtain
OGn+1 ◦ OGn−1 − OG2
n =
AB(αβ)n−1(ββα − ααβ)
α − β
which is desired. In a similar way, the equation (3.13) can be easily obtain. (cid:3)
Cassini identities are studied for Fibonacci octonions in [12, 17] and for
Modified Pell octonions in [3].
If we consider as a = 0, b = 1 and p = 1, q = 1 in the equation (3.12)
and (3.13) we have the Cassini identities of Fibonacci octonions. Hence, we
conclude that Horadam octonions are a generalization of all the Fibonacci-
like octonions.
Summation formula for the first n + 1 Horadam octonions can be given as
follows.
Theorem 3.4. The sum formula for Horadam octonions is follows,
n
Xi=0
OGi =
1
α − β(cid:0)
Bββn+1
1 − β −
Aααn+1
1 − α (cid:1) + K,
(3.14)
where K is as follows,
K =
Aα(1 − β) − Bβ(1 − α)
(α − β)(1 − α)(1 − β)
.
Proof. Using the Binet formula we can calculate the summation formula as
follows
=
n
Xi=0
OGi =
Aααn − Bββn
Aα
α − β
From geometric series we get the following
1 − αn+1
1 − α −
Aα
α − β
n
Xi=0
α − β
n
Xi=0
OGi =
n
Xi=0
αn −
Bβ
α − β
n
Xi=0
βn
Bβ
α − β
n
Xi=0
1 − βn+1
1 − β
.
After direct calculations we will get the explicit result as
n
Xi=0
OGi =
1
α − β(cid:0)
Bββn+1
1 − β −
Aααn+1
1 − α (cid:1) +
Aα(1 − β) − Bβ(1 − α)
(α − β)(1 − α)(1 − β)
.
HORADAM OCTONIONS
7
(cid:3)
Summation formula for Horadam octonions is a generalization of sum-
mation formula for Fibonacci octonions which can be calculated according to
its initial values as follows.
n
Xi=0
Oi =
1
√5(cid:18) ββn+1
1 − β −
ααn+1
1 − α (cid:19) −
α(1 − β) − β(1 − α)
√5
.
Using the initial values and roots of characteristic equation we state the
norm of nth Horadam octonion as follows.
Theorem 3.5. The norm of nth Horadam octonion is
A2α2n(1 + α2 + α4 + · · · + α14) + B 2β 2n(1 + β 2 + β 4 + · · · + β 14)
(α − β)2
−L
N r(OGn) =
where L is
(3.15)
.
2AB(−q)n(a + (−q) + ··· + (−q)7)
L =
(α − β)2
Proof. We defined nth Horadam octonion as
OGn = wne0 + wn+1e1 + ··· + wn+7e7.
So, norm of nth Horadam octonion is
N r(OGn) = OGnOGn = OGnOGn = w2
n + w2
n+1 + ··· + w2
n+7.
Making necessary calculations and using the equations α + β = p and αβ =
−q, we will get the explicit form of desired result as
N r(OGn) =
A2α2n(1 + α2 + α4 + ··· + α14)
+
B2β2n(1 + β2 + β4 + ··· + β14)
(α − β)2
−
(α − β)2
2AB(−q)n(a + (−q) + ··· + (−q)7)
(α − β)2
.
(cid:3)
One can observe that the norm of Horadam octonions is generalization
of Pell octonions and Pell-Lucas octonions. To get desired norms on [3] one
can take values a = 0, b = 1, p = 2, q = 1 and a = 2, b = 2, p = 2, q = 1,
respectively.
4. Conclusion
In this paper, we define Horadam octonions which is studied for the first
time. Moreover, we give their Binet formula and some identities related with
them. We demonstrate that our results are generalization of the other studies
in this area. It should be noted that the further studies can be to investigated
to get new identities about Horadam octonions.
8
Adnan KARATAS¸ and Serpil HALICI
References
[1] Akku¸s I. and O. Ke¸cilioglu Split Fibonacci and Lucas octonions. Adv. Appl.
Clifford Algebr. doi 10 (2015): s00006-014.
[2] Baez, J. The octonions. Bulletin of the American Mathematical Society 39.2
(2002): 145-205.
[3] Catarino, P. The Modified Pell and Modified k-Pell Quaternions and Octonions.
Advances in Applied Clifford Algebras 26, (2016):577-590.
[4] Chanyal, B. C. Split octonion reformulation of generalized linear gravitational
field equations. Journal of Mathematical Physics 56(5), (2015): 051702.
[5] Chanyal, B. C., P. S. Bisht and O. P. S. Negi Generalized split-octonion electro-
dynamics. International Journal of Theoretical Physics 50.6, (2011): 1919-1926.
[6] Chanyal, B. C., S. K. Chanyal, O. Bekta¸s and S. Yuce A new approach on
electromagnetism with dual number coefficient octonion algebra. International
Journal of Geometric Methods in Modern Physics 13.09, (2016): 1630013.
[7] Gursey, F. and C. Tze. On the role of division, Jordan and related algebras in
particle physics. World Scientific, 1996.
[8] Gunaydin, M. and F. Gursey. Quark structure and octonions. Journal of Math-
ematical Physics 14.11 (1973): 1651-1667.
[9] Halici, S. On a Generalization for Quaternion Sequences, arXiv:1611.07660.
[10] Halici, S. On Fibonacci quaternions. Advances in Applied Clifford Algebras
22.2, (2012): 321-327.
[11] Halici, S. On dual Fibonacci octonions. Advances in Applied Clifford Algebras
25.4 (2015): 905-914.
[12] Halici, S. and A. Karata¸s. Some Matrix Representations of Fibonacci Quater-
nions and Octonions. Advances in Applied Clifford Algebras: 1-10.
[13] Horadam, A. F. A generalized Fibonacci sequence. The American Math. Mont.
68.5, (1961):455-459.
[14] Horadam, A. F. Complex Fibonacci numbers and Fibonacci quaternions. The
American Mathematical Monthly 70.3, (1963):289-291.
[15] Horadam, A. F. Generating functions for powers of a certain generalized se-
quence of numbers. Duke Math. J 32, (1965): 437-446.
[16] Horadam, A. F. Special Properties of the Sequence Wn (a, b; p, q). Fibonacci.
Quart. 5(5), (1967):424-434.
[17] Ke¸cilioglu O. and I. Akku¸s The fibonacci octonions. Advances in Applied Clif-
ford Algebras 25.1 (2015): 151-158.
[18] Koshy, T. Fibonacci and Lucas numbers with applications. Vol. 51. John Wiley
and Sons, 2011.
[19] Lewis, D. W. Quaternion algebras and the algebraic legacy of Hamiltons quater-
nions. Irish Math. Soc. Bull 57 (2006): 41-64.
[20] Lounesto, P. Clifford Algebras and Spinors vol. 286. Cambridge university
press, Cambridge 2001.
[21] Savin, D. Some properties of Fibonacci numbers, Fibonacci octonions, and gen-
eralized Fibonacci-Lucas octonions. Advances in Difference Equations 2015.1
(2015): 1.
HORADAM OCTONIONS
9
[22] Szynal-Liana, A. and I. Wloch The Pell quaternions and the Pell octonions.
Advances in Applied Clifford Algebras 26.1, (2016):435-440.
[23] Unal, Z., U. Toke¸ser and G. Bilgici Some Properties of Dual Fibonacci and Dual
Lucas Octonions. Advances in Applied Clifford Algebr. doi:10.1007/s00006-
016-0724-4
Adnan KARATAS¸
Pamukkale University,
Faculty of Arts and Sciences,
Department of Mathematics,
Denizli/TURKEY
e-mail: [email protected]
Serpil HALICI
Pamukkale University,
Faculty of Arts and Sciences,
Department of Mathematics,
Denizli/TURKEY
e-mail: [email protected]
|
1704.04970 | 1 | 1704 | 2017-04-17T14:14:08 | On cyclic essential extensions of simple modules over differential operator rings | [
"math.RA"
] | In this paper we discuss under which conditions cyclic essential extensions of simple modules over a differential operator ring R[z;d] are Artinian. In particular, we study the case when R is either d-simple or d-primitive. Furthermore, we obtain important results when R is an affine algebra of Kull dimension 2. As an application we characterize the differential operator rings C[x,y][z;d] for which cyclic essential extensions of simple modules are Artinian. | math.RA | math |
ON CYCLIC ESSENTIAL EXTENSIONS OF SIMPLE MODULES OVER
DIFFERENTIAL OPERATOR RINGS
ALVERI SANT'ANA AND ROBSON VINCIGUERRA
Abstract. In this paper we discuss under which conditions cyclic essential extensions of
simple modules over a differential operator ring R[θ; δ] are Artinian. In particular, we study
the case when R is either δ-simple or δ-primitive. Furthermore, we obtain important results
when R is an affine algebra of Kull dimension 2. As an application we characterize the
differential operator rings C[x, y][θ; δ] for which cyclic essential extensions of simple modules
are Artinian.
Introduction
A Noetherian ring S whose cyclic essential extensions of simple S-modules are Artinian are
known as rings which satisfy the property (⋄). Such rings began to be studied independently
around 1974 when Jategaonkar used this property to answer the Jacobson's conjecture in the
affirmative, for fully bounded Noetherian rings. The property (⋄) has been investigated in
several classes of rings, for example: Noetherian commutative rings [21] and [22]; Noetherian
domains of Krull dimension one (see [18, Theorem 10]); the enveloping algebra U (sl2(K)) [6];
Noetherian down-up algebras [4]; the quantum plane and the quantized Weyl algebras [5].
The first example of a Noetherian domain not satisfying this property was published
by Musson in 1980 [24]. The same author showed in [23], that differential operator rings
K[x][θ, xn∂x] not satisfy (⋄) whenever n > 0. In a recent paper by Carvalho, Hatipoglu and
Lomp [3], it was completely characterized the rings K[x][θ, δ] which have the property (⋄)
(see [3, Corollary 4.1]). Furthermore, in the same paper it was also shown that if δ is a locally
nilpotent derivation, then R[θ; δ] satisfies (⋄) (see [3, Proposition 2.1]).
Our main purpose in this paper is to provide necessary and sufficient conditions for R[θ; δ]
to satisfy the property (⋄) in the following cases: R is δ-simple; R[θ; δ] is a primitive ring and
also whenever R has Krull dimension 2.
We organize the paper as follows. In the first section, we fix some notations and we recall
some basic concepts involving derivations and differential operator rings that will be useful
throughout the text.
In the second section, we show that under certain conditions, if a Noetherian integral
domain R is δ-simple, then R[θ; δ] does not satisfy (⋄) (see Theorem 2.4). As usual, the Krull
dimension of R will be denoted by K.dim(R). When R is δ-simple and K.dim(R) ≤ 1, R[θ; δ]
has the property (⋄) (see Proposition 2.5). As a consequence, if a commutative K-algebra R is
either a unique factorization domain (UFD, for short) δ-simple or an affine domain δ-simple,
we show that R[θ; δ] satisfies (⋄) if and only if K.dim(R) ≤ 1 (see Theorem 2.6).
Key words and phrases. Differential operator rings; Cyclic essential extensions; Simple modules; δ-simple
rings; δ-primitive rings.
The last author was partially supported by CNPq, Brazil. Some results were obtained during the visit of
the last author at University of Porto, Portugal. He would like to thanks the hospitality of the mathematical
department of University of Porto.
1
2
ALVERI SANT'ANA AND ROBSON VINCIGUERRA
In the third section, we show that if a commutative K-algebra R is an affine domain, then
R is δ-primitive (where δ 6= 0) if and only if R[θ; δ] is a primitive ring (see Theorem 3.2).
As a consequence, in the case when R[θ; δ] is primitive (or equivalently, R is δ-primitive), we
obtain that R[θ; δ] satisfies (⋄) if and only if R is δ-simple and K.dim(R) ≤ 1 (see Corollary
3.4). Moreover, when K.dim(R) = 1, we characterize the differential operator rings R[θ; δ]
which satisfy the property (⋄) (see Theorem 3.9) and we present an example of a differential
operator ring R[θ; δ] satisfying (⋄), where the derivation δ is not locally nilpotent (see Remark
3.11).
In the last section, we consider R a UFD of Krull dimension 2 which is an affine K-algebra.
Whenever R does not have maximal ideals which are δ-ideals, we provide necessary and
sufficient conditions under which R[θ; δ] satisfies (⋄) (see Theorem 4.3). In this case when R
has maximal ideals which are δ-ideals, we give sufficient conditions for that R[θ; δ] does not
satisfy (⋄) (see Proposition 4.4). Moreover, the differential operator rings C[x, y][θ; δ] that are
primitives were characterized (see Proposition 4.11). As an application we provide necessary
and sufficient conditions for C[x, y][θ; δ] to satisfy (⋄) (see Theorem 4.12).
1. Preliminaries
Throughout this paper K will denote a field of characteristic zero and R will be a commu-
tative ring. An additive map δ : R → R is called a derivation of R if
δ(ab) = δ(a)b + aδ(b)
for all a, b ∈ R. If R is a K-algebra and δ is a derivation of R such that δ(αa) = αδ(a) for
all α ∈ K and a ∈ R, we say that δ is a K-derivation. If R = K[x1, . . . , xn], the polynomial
ring in n variables over K, any K-derivation of K[x1, . . . , xn] is of the form
δ = a1∂x1 + · · · + an∂xn,
where ∂xi = ∂/∂xi is the partial derivative with respect to xi and a1, . . . , an ∈ R (see [25,
Theorem 1.2.1]).
An ideal I of R is said to be a δ-ideal if δ(I) ⊆ I. Of course, 0 and R are δ-ideals of R and
in the case when these are the only such ideals of R, we say that R is δ-simple. The set
Rδ = {r ∈ R δ(r) = 0}
is a subring of R known as the ring of constants of R. Note that if a K-algebra R is δ-simple,
then Rδ is a field and R is a domain (see [25, Propositions 13.1.1 and 13.1.2]).
An element a ∈ R is called a Darboux element with respect to δ if δ(a) = ba for some b ∈ R.
If the context is clear we simply refer to a as a Darboux element. Note that a is a Darboux
element if and only if Ra is a non-zero δ-ideal of R. Furthermore, whenever a K-algebra R is
a UFD and a ∈ R is a Darboux element, it is easy to check that every irreducible factor of a
is also a Darboux element.
Let I be a ideal of R and set
(I : δ) = {r ∈ R δn(r) ∈ I, ∀ n ≥ 0}.
It is easy to see that (I : δ) is the biggest δ-ideal of R contained in I. Moreover, if R is a
K-algebra and P is a prime ideal of R, (P : δ) is a prime δ-ideal (see [11, Proposition 1.1]).
We say that an ideal P of R is δ-prime if P is a δ-ideal such that P 6= R and for all I and J
δ-ideals of R with IJ ⊆ P , then I ⊆ P or J ⊆ P . The ring R is said to be δ-prime whenever
0 is a δ-prime ideal. Note that if P is a prime ideal that is a δ-ideal, then P is δ-prime. In
ON CYCLIC ESSENTIAL EXTENSIONS OF SIMPLE MODULES
3
the case when R is a Noetherian K-algebra, the δ-prime ideals of R are precisely the prime
ideals of R that are δ-ideals (see [11, Corollary 1.4]).
Let A be a multiplicative system of R, that is, A is a multiplicatively closed set, 0 6∈ A
and 1 ∈ A. We will denote by RA−1 the localization of R at A and by π : R → RA−1 the
canonical homomorphism r 7→ r1−1. If A = {xi i ∈ N}, for some x ∈ R non-zero-divisor, we
denote RA−1 by Rx and if A = R \ P , for some prime ideal P of R, we write RP for RA−1
and PP for the extended ideal P A−1. Note that any derivation δ of R extends uniquely to a
derivation (still denoted by δ) of RA−1 via the quotient's rule δ(cid:0)rx−1(cid:1) = (δ(r)x − rδ(x))x−2
for all rx−1 ∈ RA−1. Furthermore, if I is a δ-ideal of R, the extended ideal IA−1 is a δ-ideal
of RA−1. Moreover, it is easy to check that RA−1 is δ-simple whenever R is δ-simple.
The differential operator ring of R with respect to δ, denoted by S = R[θ; δ], is a free left
(and right) R-module with basis {1, θ, θ2, . . .}. The elements of S are polynomials in θ with
coefficients in R whose addition is the usual of polynomials and the multiplication extends
i=0 aiθi be a non-zero polynomial
of S with an 6= 0. The integer n is called degree of f and will be denoted by deg(f ). As usual,
we say that the degree of the zero of S is −∞. Moreover, the following identities are true:
from R via the rule θa = aθ + δ(a) for all a ∈ R. Let f =Pn
(−1)i(cid:18)n
Xi=0
i(cid:19)θn−iδi(a)
i(cid:19)δn−i(a)θi
Xi=0(cid:18)n
for all a ∈ R.
θna =
and
aθn =
n
n
Let I be a δ-ideal of R and δ the derivation on R/I induced by δ, that is, δ(a + I) = δ(a) + I
for all a ∈ R. In this case
S/SI ≃ R/I[θ; δ].
If R is a Noetherian K-algebra and P is any prime ideal of S, then P = P ∩ R is a prime
δ-ideal of R satisfying either P = SP or δ(R) ⊆ P (see [9, Lemma 3.22]). Furthermore,
whenever δ is non-zero and P 6= 0, we must have also P = P ∩ R 6= 0 (see [9, Lemma 3.18]).
2. Commutative Noetherian δ-simple Rings
We say that R is δ-primitive if R contains a maximal ideal which does not contain any non-
zero δ-ideals. In [3], Carvalho, Hatipoglu and Lomp showed that given a differential operator
ring R[θ; δ] over a Noetherian integral domain without Z-torsion R, if R is δ-primitive and
R[θ; δ] satisfy (⋄), then R is δ-simple. This fact motivated us to study property (⋄) whenever
R is δ-simple. In this section, we shall give necessary and sufficient conditions for R[θ; δ] to
satisfy (⋄) whenever R is δ-simple. We begin by showing some lemmas that will be useful to
construct a non-Artinian cyclic essential extension of a simple R[θ; δ]-module.
Lemma 2.1. Let R be an integral domain with derivation δ and S = R[θ; δ]. If x ∈ R \ {0},
then R ∩ Sθx = 0.
Proof. Suppose that there exists a = (Pn
Then
n
i=0 biθi)θx ∈ R ∩ Sθx with a, bi ∈ R and bn 6= 0.
ThusPn
a =
biθi+1x = bnθn+1x +
biθi+1x = bnxθn+1 + g
Xi=0
n−1
Xi=0
for some g ∈ S such that deg(g) < n + 1. By comparing the coefficients of all monomials in
θ, we obtain bnx = 0. Since R is a domain, we must have x = 0, which is a contradiction.
(cid:3)
i=0 biθi = 0, and therefore a = 0.
4
ALVERI SANT'ANA AND ROBSON VINCIGUERRA
Lemma 2.2. Let R be a K-algebra which is an integral domain with a derivation δ and
S = R[θ; δ]. Suppose that there exists x ∈ R such that δ(x) is an invertible element in R
or x is a prime element but not Darboux. Then S/Sθx is an essential extension of the left
S-submodule Sx/Sθx.
Proof. We claim that if f =Pn
that
xn+1f − rx ∈ Sθx.
i=0 aiθi ∈ S \ {0} and x ∤ an, then there exists r ∈ R \ {0} such
We will argue by induction on n = deg(f ). If n = 0, then f = a0 ∈ R and we can consider
r = a0 6= 0. Then xf − rx = xa0 − a0x = 0 ∈ Sθx.
Suppose that n > 0 and assume the result is valid for every g ∈ S such that deg(g) < n.
Then we have
xn+1f = xn+1 n
Xi=0
aiθi! = xn+1(anθn + an−1θn−1 + g1)
for some g1 ∈ S such that deg(g1) < n − 1. Thus
xn+1f = xn[an(xθn) + xan−1θn−1 + xg1]
= xn[an(θnx − nδ(x)θn−1 + g2) + xan−1θn−1 + xg1]
for some g2 ∈ S such that deg(g2) < n − 1. Consequently,
xn+1f = (xnanθn−1)θx + xn[(xan−1 − nanδ(x))θn−1 + ang2 + xg1
deg < n−1
].
{z
}
If δ(x) is an invertible element in R, then x ∤ nδ(x)an because nδ(x) is invertible in R and
x ∤ an. Hence, x ∤ (xan−1 − nanδ(x)). On the other hand, if x is a prime element such that
x ∤ δ(x), then x ∤ nδ(x)an and so x ∤ (xan−1 − nanδ(x)). Using the induction hypothesis, we
can write
xn[(xan−1 − nanδ(x))θn−1 + ang2 + xg1] = hθx + rx
for some h ∈ S and r ∈ R \ {0}. Thus
xn+1f = (xnanθn−1)θx + hθx
+rx
∈ Sθx
{z
}
and xn+1f − rx ∈ Sθx with r ∈ R \ {0}, as claimed.
an 6= 0.
Let U be any non-zero left S-submodule of S/Sθx. We will show that U ∩ (Sx/Sθx) 6= 0.
i=0 aiθi with
In fact, let f + Sθx be a non-zero element of minimal degree of U , say f = Pn
If n = 0, then f = a0 ∈ R \ {0}. Thus xf = xa0 ∈ Sx \ Sθx since xa0 ∈ R \ {0} and, by
Lemma 2.1, R ∩ Sθx = 0. Hence, 0 6= xf + Sθx ∈ U ∩ (Sx/Sθx) and so U ∩ (Sx/Sθx) 6= 0.
If n > 0, then x ∤ an. Otherwise, an = bx for some b ∈ R and f = b(xθn) + g1 where g1 ∈ S
and deg(g1) < n. Then
f = b(θnx + g2) + g1
where g2 ∈ S with deg(g2) < n. Hence,
f = (bθn−1)θx + bg2 + g1.
This implies that 0 6= f + Sθx = (bg2 + g1) + Sθx ∈ U with deg (bg2 + g1) < n, contradicting
our minimality assumption. Therefore x ∤ an.
ON CYCLIC ESSENTIAL EXTENSIONS OF SIMPLE MODULES
5
Now, by the above statement, there exists r ∈ R \ {0} such that xn+1f − rx ∈ Sθx. Hence,
xn+1f + Sθx = rx + Sθx ∈ U ∩ (Sx/Sθx).
Note that, by Lemma 2.1, rx + Sθx 6= 0. Therefore U ∩ (Sx/Sθx) 6= 0 as desired.
(cid:3)
Let R be a ring with a derivation δ and S = R[θ; δ]. Then R becomes a left S-module
under the action
for all b ∈ R andPn
i=0 aiθi ∈ S. The map φ : S → R defined by
aiθi! · b =
n
Xi=0
aiθi! = n
Xi=0
φ n
Xi=0
n
aiδi(b)
Xi=0
aiθi! · 1 = a0
is an epimorphism of left S-modules with ker(φ) = Sθ. Hence, S/Sθ ≃ R as left S-modules
and the S-submodules of S/Sθ are precisely the left δ-ideals of R.
As in [10], we define the δ-Krull dimension of R, denoted by δ-K.dim(R), as being the
Krull dimension of left module SR, that is, δ-K.dim(R) = K.dimS(R).
Lemma 2.3. Let R be a Noetherian K-algebra which is an integral domain with a derivation
δ and S = R[θ; δ]. If R is δ-simple and P is a non-maximal prime ideal of R, then S/SP is
a non-Antinian left S-module.
Proof. First assume that K.dim(R/P ) is finite. Since P a non-maximal prime ideal of R,
there exists a maximal ideal M such that M ) P . Note that SM is a proper left ideal of S,
otherwise we would have 1 ∈ SM ∩ R = M , which is a contradiction. Hence, S/SM 6= 0, and
therefore K.dimS(S/SM ) ≥ 0. This implies that
m = max{K.dimS(S/SQ) Q ∈ Spec(R) and Q ) P } ≥ 0.
By [10, Proposition 2.7], we obtain K.dimS(S/SP ) = m+1 ≥ 1, so that S/SP is non-Antinian.
Suppose now that K.dim(R/P ) is infinite. Since R is δ-simple, SR is simple because the S-
submodules of SR are precisely the δ-ideals of R. Hence, δ-K.dim(R) = K.dimS(R) = 0. By
[10, Proposition 4.2], we obtain K.dimS(S/SP ) = K.dim(R/P ) is infinite. Again, it follows
that S/SP is non-Antinian.
(cid:3)
Now we are able to give sufficient conditions to obtain non-Artinian cyclic essential exten-
sion of a simple R[θ; δ]-module, in the case when R is δ-simple.
Theorem 2.4. Let R be a Noetherian K-algebra which is an integral domain with a derivation
δ such that R is δ-simple. If there exists x ∈ R satisfying the following two conditions:
(1) there exists a non-maximal prime ideal P ⊆ R such that x ∈ P ,
(2) δ(x) is an invertible element in R or x is a prime element of R,
then S = R[θ; δ] does not satisfy (⋄).
Proof. First note that x ∤ δ(x), otherwise Rx would be a non-trivial δ-ideal of R, a contra-
diction. Since either δ(x) is an invertible element in R or x is a prime element in R, by
Lemma 2.2, it follows that S/Sθx is an essential extension of the left S-submodule Sx/Sθx.
Moreover, using the fact that R is δ-simple, it follows that S/Sθ ≃ Sx/Sθx is a simple left
S-module.
6
ALVERI SANT'ANA AND ROBSON VINCIGUERRA
Now, since P is a non-maximal prime ideal of R and R is δ-simple, it follows that S/SP
is a non-Antinian left S-module, by Lemma 2.3. As x ∈ P , we have Sθx ⊆ SP ⊆ S, and
consequently S/Sθx is a non-Artinian essential extension of the simple module Sx/Sθx. (cid:3)
Proposition 2.5. Let R be a Noetherian integral domain without Z-torsion and δ be a deriva-
tion of R such that R is δ-simple. If K.dim(R) ≤ 1, then S = R[θ; δ] satisfies (⋄).
Proof. Suppose first that R is a field. In this case, since 0 is a maximal ideal of R which is a δ-
ideal, from [10, Theorem 2.10] it follows that K.dim(S) = 1. Now, assume that K.dim(R) = 1
and consider M any maximal ideal of R. Since R is δ-simple, δ(M ) * M . Moreover, as R
is an integral domain without Z-torsion, if char(R/M ) = p > 0, then Rp ⊆ M is a non-zero
δ-ideal, a contradiction. Hence, char(R/M ) = 0 and so K.dim(S) = 1, by [10, Theorem 2.10].
Therefore in both cases we have that S is a Noetherian domain of Krull dimension 1, and
consequently S satisfies (⋄) (see [18, Theorem 10]).
(cid:3)
We will apply the general results obtained so far to UFDs and also to affine domains. For
such rings, we get a full description of when R[θ; δ] satisfies (⋄) provided R is also δ-simple.
We start by recalling some basic facts of commutative algebra.
Given a commutative Noetherian ring R and I an ideal of R, the grade (or depth) of
I, denoted by G(I), is the length of a maximal regular sequence contained in I (see [17]).
In general, we have G(I) ≤ ht(I), where ht(I) is the height of I (also called rank of I or
codimension of I) (see [17, Theorem 132]). A ring R is Cohen-Macaulay if G(M ) = ht(M )
for any maximal ideal M of R. A local Noetherian ring R with maximal ideal M is said to
be regular if the minimal number of generators of M is equal to the Krull dimension of R.
In such a case, any minimal set of generators for M form a regular sequence, and therefore
G(M ) = ht(M ) (see [8, Corollary 10.15]). This shows that every regular local ring is a Cohen-
Macaulay ring. Finally, a ring R is said to be regular if RP is a regular local ring for any
prime ideal P of R.
Theorem 2.6. Let R be a commutative K-algebra which is a Noetherian UFD or an affine
domain with a derivation δ such that R is δ-simple. Then
R[θ; δ] satisfies (⋄) if and only if K.dim(R) ≤ 1.
Proof. (⇒) Suppose first that R is a UFD and K.dim(R) > 1, then every non-zero prime
ideal contains a prime element. Hence, R[θ; δ] does not satisfy (⋄), by Theorem 2.4.
Assume now that R is an affine domain such that K.dim(R) = n > 1. Thus there exists a
maximal ideal M of R such that ht(M ) = n. Since R is a δ-simple affine algebra, it follows
from [27, Theorems 3 and 5] that R is a regular ring. Hence, RM is a regular local ring, and
therefore a Cohen-Macaulay ring. Then
Moreover, by [17, Theorem 135], we obtain
G(MM ) = ht(MM ) = ht(M ) = n.
G(M ) = G(MM ) = n > 1.
Applying now [7, Theorem 2], we obtain that M has a minimal set {p1, . . . , pm} of generators
such that {pi} generates a prime ideal for any 1 ≤ i ≤ m (since G(M ) > 1). Set x = pk for
some 1 ≤ k ≤ m and P = Rx. By Principal Ideal Theorem (see [8, Theorem 10.2]), we have
1 < ht(M ) ≤ m, and consequently P ( M . Since x is a prime element and P = Rx is a
non-maximal prime ideal of R, it follows from Theorem 2.4 that R[θ; δ] does not satisfy (⋄).
(cid:3)
(⇐) It follows from Proposition 2.5.
ON CYCLIC ESSENTIAL EXTENSIONS OF SIMPLE MODULES
7
The above result has the following immediate consequence.
Corollary 2.7. Let R = K[x1, . . . , xn] or R = K[x±1
that R is δ-simple. Then R[θ; δ] satisfies (⋄) if and only if n = 1.
1 , . . . , x±1
n ] with a K-derivation δ such
Examples of UFD of Krull dimension 1 that are δ-simple for some derivation δ are given
below.
Examples 2.8. (1) Given α ∈ K \ {0}, it is easy to see that K[x] and K[[x]] are δ-simple
with respect to the K-derivation δ = α∂x.
(2) Consider the local ring K[x]M , where M = K[x]x. Since K[x] is δ-simple with the
K-derivation δ = α∂x, where α ∈ K \ {0}, we obtain that K[x]M is also δ-simple.
In the literature there are several examples of δ-simple UFDs with Krull dimension bigger
than 1.
Examples 2.9. (1) [25, Example 13.4.5] The polynomial ring K[x, y] is δ-simple with the
K-derivation
δ = (x2 + y2 + 2)∂x + (x2 − y2)∂y.
(2) [25, Example 13.4.1] The polynomial ring R = K[x1, . . . , xn], where n > 2, is δ-simple
with the following K-derivation
δ = (1 − x1x2)∂x1 + x3
1∂x2 +
xi−1∂xi.
n
Xi=3
(3) [1, Teorema 2.5.9] The affine C-algebra R = C[x1, x2, y1, y2]/hx2
1 + y2
1 − 1, x2
2 + y2
2 − 1i
with the C-derivation
δ = ay1∂x1 − ax1∂y1 + by2∂x2 − bx2∂y2
is δ-simple if and only if a/b is a non rational number. Note that R ≃ C[t1, t−1
the isomorphism
1 , t2, t−1
2 ] via
x1 + iy1 7→ t1,
x2 + iy2 7→ t2,
x1 − iy1 7→ t−1
1
x2 − iy2 7→ t−1
2 .
(4) [2, Proposition 3.3] The local ring R = K[x, y]hx,yi is δ-simple with the K-derivation
where n is a positive integer and β ∈ Q \ {0}.
δ = ∂x + (βyn + 1)∂y,
The next examples give us an application of the above theorem for affine domains.
Example 2.10. Let R = K[x, y]/hx2 + y2 − 1i with derivation δ = y∂x − x∂y. Then R is
δ-simple (see [30, Example 2.5]). Note that R is an affine domain of Krull dimension 1. By
Theorem 2.6, we concluded that R[θ; δ] satisfies (⋄).
Example 2.11. If R = K[x, y, z]/hx2 +yz −1i with derivation δ = (x2y −z)∂x +2x∂y −2x3∂z,
R is δ-simple (see [1, Theorem 2.5.23]). Since R is an affine domain of Krull dimension > 1,
by Theorem 2.6, R[θ; δ] does not satisfy (⋄).
8
ALVERI SANT'ANA AND ROBSON VINCIGUERRA
We finish this section with a result on locally nilpotent derivations. Recall that a derivation
δ of R is called locally nilpotent if, for every a ∈ R, there exists k > 0 such that δk(a) = 0.
For example, let R = K[x1, . . . , xn]. A class of locally nilpotent derivations of these rings
are the so called triangular derivations, that is, derivations satisfying δ(x1) ∈ K and δ(xi) ∈
K[x1, . . . , xi−1] for all i > 1. Carvalho, Hatipoglu and Lomp showed that R[θ; δ] satisfies
(⋄) whenever R is a commutative affine K-algebra with a locally nilpotent derivation ([3,
Proposition 2.1]). We will show now that if R is a δ-simple ring with respect to a locally
nilpotent derivation, then we have
K.dim(R) ≤ K.dim(R[θ; δ]) ≤ 1.
Given K ⊂ L an extension of fields, denote by tr.degK(L) the transcendence degree of L
over K and F(R) the field of fractions of a ring R. If R is an affine K-algebra which is an
integral domain, it is known that the Krull dimension of R coincides with the transcendence
degree of F(R) over K, that is, K.dim(R) = tr.degK (F(R)) (see [28, Corollary 14.29]). In
general, we have the following lemma whose proof is due to Manuel Reyes in MathOverflow
[26]. Since we were not aware of any reference for it in the literature we reproduce his proof
here.
Lemma 2.12. [26] Let R be a K-algebra which is an integral domain. Then
K.dim(R) ≤ tr.degK(F(R)).
Proof. Let P0 ⊂ P1 ⊂ . . . ⊂ Pm be a strictly ascending chain of prime ideals of R. We will show
that m ≤ tr.degK(F(R)). For each i ∈ {1, . . . , m}, we consider an element xi ∈ Pi \ Pi−1.
Let R′ ⊆ R be the K-subalgebra generated by {x1, . . . , xm} and set P ′
i = R′ ∩ Pi for all
0 ⊂ P ′
0 ≤ i ≤ m. Note that P ′
m is a strictly ascending chain of prime ideals
of R′ because xi ∈ P ′
i \ P ′
i−1. Since R′ is a affine K-algebra, it follows from [28, Corollary
14.29] that K.dim(R′) = tr.degK(F(R′)). Therefore m ≤ K.dim(R′) = tr.degK(F(R′)) ≤
tr.degK(F(R)), as desired.
(cid:3)
1 ⊂ . . . ⊂ P ′
Proposition 2.13. Let R be a Noetherian integral domain without Z-torsion and let δ be a
derivation of R. If R is δ-simple and δ is locally nilpotent, then
K.dim(R) ≤ K.dim(R[θ; δ]) ≤ 1.
Proof. First assume that R is a field. Since 0 is a maximal ideal of R which is δ-ideal,
we can apply [10, Theorem 2.10] to obtain K.dim(R[θ; δ]) = 1. Thus 0 = K.dim(R) <
K.dim(R[θ; δ]) = 1.
On the other hand, suppose that R is not a field and let M be any maximal ideal of R.
Since R is δ-simple, we must have δ(M ) * M . Furthermore, as R has no Z-torsion, we have
also char(R/M ) = 0. By [10, Theorem 2.10], K.dim(R) = K.dim(R[θ; δ]). Now, consider
the subring of constants Rδ of R which is a field because R is δ-simple. Set K = Rδ and
it follows from Lemma 2.12 that K.dim(R) ≤ tr.degK(F(R)). Since δ is locally nilpotent, it
follows from [20, Lemma 4] that tr.degK(F(R)) = 1. Therefore K.dim(R[θ; δ]) = K.dim(R) ≤
tr.degK(F(R)) = 1.
(cid:3)
3. Commutative Noetherian δ-primitive rings
In [3], Carvalho, Hatipoglu and Lomp studied the property (⋄) in R[θ; δ] whenever R is a
δ-primitive ring. In particular, they showed the following.
ON CYCLIC ESSENTIAL EXTENSIONS OF SIMPLE MODULES
9
Theorem 3.1. [3, Theorem 3.5] Let R be a Noetherian integral domain without Z-torsion with
non-zero derivation δ such that R is δ-primitive. If R[θ; δ] satisfies (⋄), then R is δ-simple.
In this section, we will show that if R is an affine K-algebra that is an integral domain
with non-zero derivation δ such that R is δ-primitive, then R[θ; δ] satisfies (⋄) if and only if R
is δ-simple and K.dim(R) ≤ 1. Moreover, we will characterize the differential operator rings
R[θ; δ] satisfying (⋄) when R has Krull dimension 1.
A ring R is called δ-G-ring if R is δ-prime and
\{P ⊳ R P is non-zero δ-prime} 6= 0.
In [11], Goodearl and Warfield showed that R[θ; δ] is a primitive ring if and only if δ 6= 0 and
R is either δ-primitive or a δ-G-ring (see [11, Theorem 3.7]). In the case when R is an affine
K-algebra which is an integral domain, we have the following stronger result.
Theorem 3.2. Let R be an affine K-algebra which is an integral domain with a derivation
δ. If R is a δ-G-anel, then R is δ-primitive. Consequently
R[θ; δ] is primitive if and only if δ 6= 0 and R is δ-primitive.
Proof. Assume that R is a δ-G-ring but is not δ-primitive and take M an arbitrary maximal
ideal of R. Then there must exist a non-zero δ-ideal I such that M ⊇ I. Consider (M : δ)
the largest δ-ideal of R contained in M . Note that (M : δ) 6= 0 because (M : δ) ⊇ I. Since
char(R/M ) = 0 (because R is a K-algebra) and M is a prime ideal, it follows that (M : δ) is
a prime ideal of R. Thus
0 6= \{P ⊳ R P is non-zero δ-prime} ⊆
⊆
(M : δ)
\M ∈ Max(R)
\M ∈ Max(R)
M
= J (R).
But, as R is an affine K-algebra which is an integral domain, by [19, Theorem 5.3], we obtain
J (R) = 0, a contradiction. Therefore R is δ-primitive.
The last statement is now a consequence of [11, Theorem 3.7].
(cid:3)
Taking into account the Propositions 2.5 and 3.1 it follows easily the following.
Corollary 3.3. Let R be a Noetherian integral domain of Krull dimension ≤ 1 without Z-
torsion. If R is δ-primitive for some non-zero derivation δ, then
R[θ; δ] satisfies (⋄) if and only if R is δ-simple.
Corollary 3.4. Let R be an affine K-algebra which is an integral domain with a non-zero
derivation δ such that R[θ; δ] is primitive (or equivalently, R is δ-primitive). Then
R[θ; δ] satisfies (⋄) if and only if R is δ-simple and K.dim(R) ≤ 1.
Proof. (⇒) Suppose that R[θ; δ] satisfies (⋄). Since R[θ; δ] is primitive, by Theorem 3.2, it
follows that R is δ-primitive and, by Theorem 3.1, R is δ-simple. The Theorem 2.6 implies
that K.dim(R) ≤ 1.
(⇐) It follows from Theorem 2.6.
(cid:3)
The following is a straightforward consequence of Corollary 3.4.
10
ALVERI SANT'ANA AND ROBSON VINCIGUERRA
Corollary 3.5. Let R = K[x1, . . . , xn] or K[x±1
that R[θ; δ] is primitive. Then
1 , . . . , x±1
n ] with non-zero K-derivation δ such
R[θ; δ] satisfies (⋄) if and only if R is δ-simple and n = 1.
Corollary 3.6. Let K be an algebraically closed field, R be an affine K-algebra which is
an integral domain and S = R[θ; δ]. If δ is a locally nilpotent derivation of R, then every
primitive quotients of S have Krull dimension ≤ 1.
Proof. Suppose that δ is a locally nilpotent derivation of R. It follows from [3, Proposition
2.1] that S satisfies (⋄). Hence, so does its primitive quotients. Let P be any primitive
ideal of S. By [9, Theorem 3.22], either P = S(P ∩ R) or S/P is commutative. The latter
case implies that S/P is a field, that is, K.dim(S/P) = 0. If P = S(P ∩ R), then S/P =
S/S(P ∩ R) = R/(P ∩ R)[θ; δ], where δ is the derivation of R/(P ∩ R) induced by δ. Since
R/(P ∩ R)[θ; δ] is primitive and satisfies (⋄), it follows from Corollary 3.4 that R/(P ∩ R) is δ-
simple and K.dim(R/(P ∩ R)) ≤ 1. By [10, Theorem 2.10], we have K.dim(R/(P ∩ R)[θ; δ]) =
K.dim(R/(P ∩ R)) ≤ 1.
(cid:3)
The following result provides a sufficient condition for a ring R to be δ-primitive when R
has Krull dimension one.
Lemma 3.7. Let R be a K-algebra which is an integral domain of Krull dimension 1 and δ
a derivation of R. If there exists a maximal ideal M of R such that δ(M ) * M , then R is
δ-primitive.
Proof. Suppose that there exists a maximal ideal M of R such that δ(M ) * M . We claim
that M does not contain non-zero δ-ideals. Otherwise, let I be a non-zero δ-ideal contained
in M and consider the δ-ideal (M : δ). Since M is not a δ-ideal, we must have (M : δ) ( M .
Thus
0 ( I ⊆ (M : δ) ( M.
But, as M is prime and char(R/M ) = 0, we obtain that (M : δ) is prime ([11, Proposition
1.1]). Hence, ht(M ) > 1, a contradiction. Therefore M does not contain non-zero δ-ideals
and the result follows.
(cid:3)
The following lemma is known in the literature and its proof can be obtained in [1, Theorem
2.3.8] or in [10, Corollary 2.12]. We will denote by Max(R) the maximal spectrum of a ring
R.
Lemma 3.8. [1, Theorem 2.3.8] Let R be a commutative affine K-algebra with K-derivation
δ and M ∈ Max(R). Then δ(M ) ⊆ M if and only if δ(R) ⊆ M .
The next result characterize differential operator rings R[θ; δ] satisfying (⋄) when R is an
affine K-algebra which is an integral domain of Krull dimension 1.
Theorem 3.9. Let R be an affine K-algebra which is an integral domain of Krull dimension
1 with non-zero K-derivation δ. Then R is δ-primitive. Consequently,
R[θ; δ] satisfies (⋄) if and only if R is δ-simple.
Proof. We claim that there exists a maximal ideal which is not a δ-ideal. Otherwise, δ(M ) ⊆
M for all M ∈ Max(R). Since δ is a K-derivation, by Lemma 3.8, we would have δ(R) ⊆ M
for all M ∈ Max(R), and so
δ(R) ⊆\{M ⊳ R M ∈ Max(R)} = J (R).
ON CYCLIC ESSENTIAL EXTENSIONS OF SIMPLE MODULES
11
But, by [19, Theorem 5.3], J (R) = 0, and therefore δ = 0, a contradiction. Therefore there
exists a maximal ideal which is not δ-ideal and, by Lemma 3.7, the result follows.
The result is now a consequence of Corollary 3.4.
(cid:3)
As an application we completely classify the differential operator rings K[x, x−1][θ; δ] which
satisfy the property (⋄).
Corollary 3.10. For any non-zero K-derivation δ of K[x, x−1] such that δK[x] is a K-
derivation of K[x], the following statements are equivalent:
(1) K[x, x−1][θ; δ] satisfies (⋄);
(2) δ(x) = αxn for some α ∈ K \ {0} and n ≥ 0;
(3) K[x, x−1] is δ-simple.
Proof. Let R = K[x, x−1] = K[x]A−1 = K[x]x, where A = {xi i ≥ 0}, and recall that any
K-derivation of K[x] is of the form δ = a∂x for some a ∈ K[x]. Moreover, note that any
derivation δ of K[x] extends uniquely to a derivation of R via the quotient's rule δ(cid:0)rs−1(cid:1) =
(δ(r)s − rδ(s))s−2 for all rs−1 ∈ R.
(1) ⇔ (3) Follows from Theorem 3.9.
(2) ⇒ (3) Suppose that δ(x) = αxn for some α ∈ K \ {0} and n ≥ 0. If n = 0, that is,
δ(x) = α, where α ∈ K \ {0}, then R = K[x]x is δ-simple. Now consider the case when n > 0.
Note that the non-zero δ-prime ideals of K[x] are precisely the maximal ideals of K[x] that
are δ-ideals ([11, Corollary 1.4]). Hence, by [11, Proposition 2.8], it is enough to show that
every maximal ideal of K[x] which is a δ-ideal contains x. Given any maximal ideal M which
is a δ-ideal, it follows that M = K[x]p, for some irreducible element p ∈ K[x], and p δ(p).
But p δ(p) = αxn∂x(p) implies p x, and therefore x ∈ K[x]p = M .
(3) ⇒ (2) Suppose that δ(x) = a, for some a ∈ K[x] \ {0}, such that a 6= αxn for all
α ∈ K \ {0} and n ≥ 0. Consider the δ-ideal I = K[x]a. Note that I ∩ A = ∅, and as a 6= 0,
we obtain IA−1 is a non-trivial δ-ideal of R. This shows that R is not δ-simple.
(cid:3)
Remark 3.11. Let K be an algebraically closed field of characteristic zero and let R be a
commutative affine K-algebra with derivation δ. In [3, Proposition 2.1], Carvalho, Hatipoglu
and Lomp showed that δ being locally nilpotent implies that R[θ; δ] satisfies (⋄). The converse
Indeed, consider R = K[x, x−1] with the K-derivation δ = x∂x.
is not true in general.
Corollary 3.10 shows that R[θ; δ] satisfies (⋄), but δ is not locally nilpotent, since δn(x) = x 6= 0
for all n ≥ 0.
4. Affine Algebras of Krull Dimension 2
In the present section, we drop the condition of primitivity in S = R[θ; δ]. If S satisfies
(⋄), so does its primitive quotients. Given a commutative Noetherian K-algebra R with a
derivation δ and P a primitive ideal of S, either P = (P ∩ R)S or S/P is commutative (see [9,
Theorem 3.22]). If P is primitive such that P = (P ∩ R)S, conditions for S/P to satisty (⋄)
have been presented in the previous section. In some cases, the study of (⋄) can be reduced
to the study of (⋄) on its primitives quotients.
Lemma 4.1. [13, Lemma 2.5] Suppose that S is a Noetherian algebra such that every primitive
ideal P of S contains an ideal Q ⊆ P which has a normalizing sequence of generators and
S/Q satisfies (⋄). Then S satisfies (⋄).
12
ALVERI SANT'ANA AND ROBSON VINCIGUERRA
In this section, we consider the case when R is an affine K-algebra of Krull dimension 2.
The main result is the Theorem 4.12 when R = C[x, y]. We will denote by Fδ the set of all
δ-ideals of R, that is, Fδ = {I ⊳ R I is a δ-ideal}. We begin by considering the case when R
does not have maximal ideals which are δ-ideals, that is, Max(R) ∩ Fδ = ∅.
Lemma 4.2. Let R be a commutative Noetherian K-algebra of Krull dimension 2 with deriva-
tion δ and S = R[θ; δ]. If P is a non-zero prime δ-ideal of R which is not contained in any
maximal ideal which is a δ-ideal, then K.dimS(S/SP ) = 1.
Proof. Let P be as in the hypothesis. Note that P is not maximal because P is a δ-ideal. By
[10, Proposition 2.7], we have K.dimS(S/SP ) = m + 1, where
m = max{K.dimS(S/SQ) Q ∈ Spec(R) and Q ) P }.
Since R has Krull dimension 2, the only prime ideals that properly contain P are the maximal
ones. Given any maximal ideal M of R containing P , by hypotheses, we must have δ(M ) * M .
Moreover, as char(R/M ) = 0, it follows as in the proof of [12, Lemma 2.4] that S/SM is a
simple left S-module, and so K.dimS(S/SM ) = 0. This shows that m = 0. Therefore
K.dimS(S/SP ) = 1, as desired.
(cid:3)
Recall that an element a in a ring S is called a normal element if Sa = aS. Note that if
S = R[θ; δ] and a ∈ R is a Darboux element, then a is a normal element in S. In fact, as
δ(a) = ba for some b ∈ R, we have θa = aθ + δ(a) = a(θ + b). Thus θna = a(θ + b)n and also
aθn = (θ − b)na, showing that Sa = aS.
The next result shows that the property (⋄) also can be related to the primitiveness of S
in our setting.
Theorem 4.3. Let R be a UFD of Krull dimension 2 which is an affine K-algebra. Suppose
that δ is a derivation of R such that Max(R) ∩ Fδ = ∅. Then the following conditions are
equivalent:
(1) S = R[θ; δ] satisfies (⋄);
(2) R is not δ-primitive;
(3) S is not primitive.
Proof. (1) ⇒ (2) Suppose that S satisfies (⋄). Since K.dim(R) = 2, it follows from Corollary
3.4 that R is not δ-primitive.
(2) ⇔ (3) Since Max(R) ∩ Fδ = ∅, it follows that δ 6= 0. The equivalence is now a
consequence of Theorem 3.2.
(3) ⇒ (1) Assume that S is not primitive. Let P be any primitive ideal of S. Thus
It follows from [9, Lemma 3.19 and Theorem 3.22 (a)]
P is a non-zero prime ideal of S.
that P = P ∩ R is a non-zero prime δ-ideal of R (note that δ 6= 0). Moreover, by [9,
Theorem 3.22 (b)], either P = SP or δ(R) ⊆ P . The later case does not apply.
In fact,
suppose that δ(R) ⊆ P and consider a maximal ideal M of R such that M ⊇ P . Hence,
δ(M ) ⊆ δ(R) ⊆ P ⊆ M , contradicting the fact that Max(R) ∩ Fδ = ∅. Thus P = SP and,
by Lemma 4.2, S/P = S/SP is a Noetherian domain of Krull dimension 1, and therefore
satisfies (⋄).
Now, since P 6∈ Max(R), we must have P = Rp for some irreducible Darboux elements
p ∈ R. This shows that P = SP = S(Rp) = Sp is generated by a normal element. Applying
Lemma 4.1 we conclude that S satisfies (⋄).
(cid:3)
ON CYCLIC ESSENTIAL EXTENSIONS OF SIMPLE MODULES
13
Proposition 4.4. Let R be an affine K-algebra which is an integral domain and let δ be a
K-derivation of R. Suppose that there exist M ∈ Max(R) ∩ Fδ and a prime δ-ideal P such
that M ⊇ P and ht(P ) = K.dim(R) − 1. If S = R[θ; δ] satisfy (⋄), then δ(R) ⊆ P .
Proof. Suppose that δ(R) * P . Then δ induces a non-zero K-derivation δ in R/P . Since P is
a δ-ideal and δ(R) * P , P is not maximal by Lemma 3.8. Thus P ( M and δ(M ) ⊆ M implies
that M/P is a non-trivial δ-ideal of R/P , and therefore R/P is not δ-simple. Moreover, by
[8, Corollary 13.4], K.dim(R/P ) = K.dim(R) − ht(P ) and, since ht(P ) = K.dim(R) − 1, we
obtain K.dim(R/P ) = 1. Using Theorem 3.9, we conclude that S/SP ≃ R/P [θ; δ] does not
satisfy (⋄). Therefore S does not satisfy (⋄).
(cid:3)
Next, we present an example of a ring R which is not δ-primitive and such that R[θ; δ] does
not satisfy (⋄).
Example 4.5. Suppose that K is an algebraically closed field and consider R = K[x, y] with
the K-derivation δ = x∂x + y∂y. Since K is an algebraically closed field, every maximal ideal
of R is of the form M = R(x − λ) + R(y − µ) for some λ, µ ∈ K. If λ = 0, then M contains
the Darboux element x. In the case when λ 6= 0, we have
µx − λy = µx − µλ + λµ − λy = µ(x − λ) − λ(y − µ) ∈ M
and µx − λy is a Darboux element. This shows that R is not δ-primitive. On the other
hand, we have that Rx + Ry ∈ Max(R) ∩ Fδ and contains the irreducible Darboux element
x. Moreover δ(R) * Rx, since y = δ(y) ∈ δ(R) and y 6∈ Rx. By Proposition 4.4, we conclude
that R[θ; δ] does not satisfy (⋄).
The existence of Darboux elements plays, not only an importante role in the study of the
property (⋄), but also in the study in δ-G-rings as next propositions show.
Proposition 4.6. Let R be a Noetherian UFD with derivation δ. If there are infinitely many
non-associated irreducible Darboux elements in R, then R is not a δ-G-ring.
Proof. By [11, Proposition 2.9], it is enough to show that Rc is not δ-simple for all c ∈ R\{0}.
Given any non-zero element c of R, then c can be written as a finite product of irreducible
elements. Since there are infinitely many irreducible Darboux elements, there must be an
irreducible Darboux element p that does not divide c. Let π : R → Rc be the canonical map
a 7→ a1−1. Thus π(p) is not invertible in the localization Rc, and so Rcπ(p) is a non-trivial
δ-ideal, that is, Rc is not δ-simple, as desired.
(cid:3)
Proposition 4.7. Let R be a UFD of Krull dimension 2 which is an affine K-algebra and
let δ be a non-zero K-derivation of R. Then there are only a finite number of non-associated
irreducible Darboux elements in R if and only if R is a δ-G-ring.
Proof. (⇒) First note that, since R has Krull dimension 2, given any non-zero δ-prime ideal
of R, it is either an ideal generated by an irreducible Darboux element or a maximal ideal
which is a δ-ideal. Thus considering
A = {Rp p is irreducible Darboux} and B = Max(R) ∩ Fδ,
since δ 6= 0. It follows from Lemma 3.8 that 0 6= δ(b) ∈ M for all M ∈ B, that is, 0 6= δ(b) ∈
we have T{P ⊳ R P is non-zero δ-prime} =TP ∈A∪B P . Moreover, δ(b) 6= 0 for some b ∈ R,
TM ∈B M .
14
ALVERI SANT'ANA AND ROBSON VINCIGUERRA
If R has no irreducible Darboux elements, that is, A = ∅, then
0 6= δ(b) ∈ \M ∈B
M =\{P ⊳ R P is non-zero δ-prime}.
In the case when there are only a finite number s > 0 of non-associated irreducible Darboux
elements in R, namely p1, . . . , ps, we have
0 6= p1 · · · psδ(b) ∈ \P ∈A∪B
P =\{P ⊳ R P is non-zero δ-prime}.
Hence, in both cases, we must have T{P ⊳ R P is non-zero δ-prime} 6= 0. Furthermore,
since R is a domain, we have that R is δ-prime. Therefore R is a δ-G-ring.
(⇐) Follows from Proposition 4.6.
(cid:3)
Combining Proposition 4.7 with Theorem 3.2, we obtain easily the following.
Proposition 4.8. Let R be a UFD of Krull dimension 2 which is an affine K-algebra and
let δ be a non-zero K-derivation of R. If there are only a finite number of non-associated
irreducible Darboux elements in R, then R is δ-primitive.
Let R = K[x, y] with a K-derivation δ. Using Lemma 3.8, we can conclude that
Max(R) ∩ Fδ = ∅ if and only if R = hδ(x), δ(y)i.
A trivial example of K-derivations δ of R = K[x, y] satisfying R = hδ(x), δ(y)i are those such
that δ(x) ∈ K \ {0} or δ(y) ∈ K \ {0}.
A K-derivation δ of K[x, y] is said to be a Shamsuddin derivation if δ is of the form
δ = ∂x + (ay + b)∂y, where a, b ∈ K[x]. For these derivations, R[θ; δ] does not satisfy (⋄).
Example 4.9. Let R = K[x, y] with the K-derivation δ = ∂x + (ay + b)∂y, where a, b ∈ K[x]
and a 6= 0. Then R[θ; δ] does not satisfy (⋄). In fact, if R is δ-simple, the result follows from
Corollary 2.7. In the case when R is not δ-simple, it follows from [2, Theorem 4.1, (b)] that
there exists a unique polynomial c ∈ K[x] such that δ(c) = ac + b and P = R(y − c) is the
unique non-zero prime δ-ideal of R. This shows that y − c is the unique irreducible Darboux
element of R. By Proposition 4.8, we conclude that R is δ-primitive. Hence, Theorem 4.3
shows that S does not satisfy (⋄).
Example 4.10. Suppose that K is algebraically closed and consider R = K[x, y] with the
K-derivation
δ = ∂x + xm(y + γ)n∂y,
where m, n ∈ N and γ ∈ K. Then S = R[θ; δ] satisfies (⋄) if and only if n 6= 1. Indeed, if
n = 1, the result follows from Example 4.9. Conversely, if n = 0, then δ = ∂x + xm∂y is a
triangular derivation. Thus δ is locally nilpotent, and so S satisfies (⋄). Assume n > 1, and
consider the elements
p = y + γ
and
qω = (xm+1 + ω)(y + γ)n−1 +
(m + 1)
(n − 1)
, where ω ∈ K.
We have
δ(p) = xm(y + γ)n−1p
and
δ(qω) = (n − 1)xm(y + γ)n−1qω,
and so they are Darboux elements. Now, given any maximal ideal M of R. Since K is
If µ = −γ, then
algebraically closed, we have M = hx − λ, y − µi for some λ, µ ∈ K.
ON CYCLIC ESSENTIAL EXTENSIONS OF SIMPLE MODULES
15
M = hx − λ, y + γi contains p = y + γ. In the case when µ 6= −γ, consider the Darboux
element
qω =(cid:18)xm+1 − λm+1 −
(m + 1)
(n − 1)(µ + γ)n−1(cid:19) (y + γ)n−1 +
(m + 1)
(n − 1)
for ω = −λm+1 −
(m+1)
(n−1)(µ+γ)n−1 . We have
qω = (y + γ)n−1(xm+1 − λm+1) −
= a(x − λ) + b(y − µ) ∈ M,
(m + 1)
(n − 1)(µ + γ)n−1(cid:0)(y + γ)n−1 − (µ + γ)n−1(cid:1)
where
a = (y + γ)n−1 m
Xi=0
λixm−i! and b = −
(m + 1)
(n − 1)(µ + γ)n−1 n−2
Xi=0
(µ + γ)i(y + γ)n−2−i! .
Thus any maximal ideal of R contains a Darboux element, and consequently a non-zero δ-
ideal. Therefore R is not δ-primitive. Moreover, R = hδ(x), δ(y)i because δ(x) = 1 ∈ K \ {0}.
Hence, Max(R) ∩ Fδ = ∅ and the result follows from Theorem 4.3.
From now on, we assume that K = C, the field of complex numbers, and R = C[x, y] with
a C-derivation δ. Consider L = C(x, y) the field of fractions of R. Then δ extends uniquely
to L. Let
Lδ = {p/q ∈ L δ(p/q) = 0}
be the subring of constants of L. The main purpose here, is to provide necessary and sufficient
conditions for the differential operator rings C[x, y][θ; δ] to satisfy (⋄).
Proposition 4.11. Let R = C[x, y] with non-zero C-derivation δ. The following statements
are equivalent:
(1) R[θ; δ] is primitive;
(2) R is a δ-G-ring;
(3) There are only a finite number of non-associated irreducible Darboux elements in R;
(4) R is δ-primitive;
(5) There is a maximal ideal that does not contain irreducible Darboux elements;
(6) Lδ = C.
Proof. The equivalences (1) ⇔ (4) and (2) ⇔ (3) are a consequence of Theorem 3.2 and
Proposition 4.7, respectively. The statement (3) ⇒ (4) follows from Proposition 4.8 and (4)
⇒ (5) is obvious.
(5) ⇒ (6) Suppose that Lδ 6= C. We will show that every maximal ideal of R contains
an irreducible Darboux element. Let p, q ∈ R be such that p/q ∈ Lδ \ C. Without loss of
generality, we can assume mdc(p, q) = 1. Since δ(p/q) = (δ(p)q − pδ(q))/q2 = 0, we have
δ(p)q = pδ(q), and so p δ(p)q. As mdc(p, q) = 1, we obtain p δ(p). Thus there exists c ∈ R
such that δ(p) = cp, and therefore δ(q) = cq. In this case, the elements of the form αp − βq
are Darboux elements for all α, β ∈ C, not both zero.
Let M = hx − λ, y − µi, where λ, µ ∈ C, be any maximal ideal of R. Note that
M = {r(x, y) ∈ R r(λ, µ) = 0}.
If p(λ, µ) = 0, then M contains the Darboux element p. In the case when p(λ, µ) 6= 0, set
α = q(λ, µ) and β = p(λ, µ) 6= 0. Thus r = αp − βq ∈ R is a Darboux element such that
r ∈ M .
16
ALVERI SANT'ANA AND ROBSON VINCIGUERRA
The above argument shows that every maximal ideal M contains a Darboux element. Since
the irreducible factors of a Darboux element are also Darboux elements, M must contains an
irreducible Darboux element.
(6) ⇒ (3) By [29, Darboux's Theorem, p. 686], if there are infinity non-associated irre-
ducible Darboux elements in R, then there exists w ∈ L \ C such that δ(w) = 0, that is,
Lδ 6= C.
(cid:3)
Theorem 4.12. Let R = C[x, y] with non-zero C-derivation δ and let S = R[θ; δ]. Then S
satisfies (⋄) if and only if S is not primitive and for any M ∈ Max(R) ∩ Fδ and p ∈ M , where
p is an irreducible Darboux element, we have δ(R) ⊆ Rp.
Proof. Suppose that S satisfaz (⋄).
It follows from Corollary 3.5 that S is not primitive.
Assume, in addition, that Max(R) ∩ Fδ 6= ∅. By contradiction, suppose that there exist M ∈
Max(R) ∩ Fδ and p ∈ M , where p is an irreducible Darboux element, such that δ(R) * Rp.
Applying Proposition 4.4, we would have that S does not satisfy (⋄), a contradiction.
Conversely, if S is not primitive and Max(R) ∩ Fδ = ∅, it follows from Theorem 4.11 that
S satisfies (⋄). Suppose now that S is not primitive and for any M ∈ Max(R) ∩ Fδ and every
irreducible Darboux element p contained in M , we have δ(R) ⊆ Rp. In this case, we will
show that every primitive ideal P of S contains an ideal Q of S such that Q is generated by
normal elements and S/Q satisfies (⋄). In fact, let P be any primitive ideal of S. Since S is
not primitive, P is a non-zero prime ideal of S. Thus P = P ∩ R is a non-zero prime δ-ideal
of R and either P = SP or δ(R) ⊆ P . We will study these two cases below.
Case δ(R) ⊆ P : If P is not maximal, then P = Rp for some irreducible Darboux element
p ∈ R. In this case, set Q = SP = Sp ⊆ P and note that Q is an ideal of S generated by the
normal element p. Since δ(R) ⊆ P , we obtain that
S/Q = S/SP ≃ R/P [θ; δ] = R/P [θ]
is a commutative Noetherian ring, and therefore satisfies (⋄). On the other hand, when P
is a maximal ideal, as S is not primitive, it follows from Proposition 4.11 that there exists a
irreducible Darboux element p such that p ∈ P . Set Q = Sp ⊆ P and note that again Q is an
ideal of S generated by the normal element p. Since P ∈ Max(R) ∩ Fδ and p is an irreducible
Darboux element contained in P , by hypothesis, we have δ(R) ⊆ Rp. Thus
S/Q = S/Sp ≃ R/Rp[θ; δ] = R/Rp[θ]
is again a commutative Noetherian ring, and so it satisfies (⋄).
Case δ(R) * P : In this case, we have P = SP . Moreover, P is not maximal. Otherwise, P
would be a maximal ideal which is a δ-ideal and, by Lemma 3.8, we would have δ(R) ⊆ P ,
which does not occur. Hence, P = Rp for some irreducible Darboux element p ∈ R, and so
P = SP = Sp is generated by the normal element p. Furthermore, any maximal ideal M
of R containing P = Rp is not a δ-ideal because, otherwise by hypothesis, we would have
δ(R) ⊆ Rp = P . It follows from Lemma 4.2 that K.dim(S/P) = K.dim(S/SP ) = 1, and
therefore S/P satisfies (⋄). In this case, it is enough to set Q = P.
By Lemma 4.1, we conclude that S satisfies (⋄).
(cid:3)
Corollary 4.13. Let R = C[x, y] with C-derivation δ such that mdc(δ(x), δ(y)) = 1. Then
R[θ; δ] satisfies (⋄) if and only if S is not primitive and Max(R) ∩ Fδ = ∅.
Proof. Since mdc(δ(x), δ(y)) = 1 and δ(x), δ(y) ∈ δ(R), we have δ(R) * Rp, for any irre-
ducible Darboux element p ∈ R. By Theorem 4.12, the result follows.
(cid:3)
ON CYCLIC ESSENTIAL EXTENSIONS OF SIMPLE MODULES
17
Recall that a polynomial is called homogeneous of degree n if all its monomials have the
same degree n. A derivation δ of C[x, y] is said to be homogeneous derivation of degree n if
δ(x) and δ(y) are both homogeneous of the same degree n.
Example 4.14. Let R = C[x, y] with C-derivation δ. Suppose that δ is homogeneous of
degree n and mdc(δ(x), δ(y)) = 1. Then R[θ; δ] satisfies (⋄) if and only if n = 0. In fact,
assume that n > 0 and set M = hx, yi. Since δ(x) and δ(y) are homogeneous polynomials
of the same degree n > 0, we have that δ(x) and δ(y) does not have constant terms. This
implies that δ(x), δ(y) ∈ M , and consequently M is a maximal ideal of R which is a δ-ideal.
Thus Max(R) ∩ Fδ 6= ∅. By Corollary 4.13, S does not satisfy (⋄). Conversely, it is enough
to observe that if n = 0, δ is locally nilpotent.
Remark 4.15. Let R = C[x, y] with C-derivation δ such that mdc(δ(x), δ(y)) = c ∈ R \ C,
and so δ = c(a∂x + b∂y) = cδ′, where a, b ∈ R, mdc(a, b) = 1 e δ′ = a∂x + b∂y. In this case,
we have:
(1) Max(R) ∩ Fδ 6= ∅.
In fact, if Max(R) ∩ Fδ = ∅, then R = hδ(x), δ(y)i, and therefore mdc(δ(x), δ(y)) =
1, a contradiction.
(2) If R[θ; δ] satisfies (⋄), then c ∈T{M ∈ Max(R) δ(M ) ⊆ M }.
Indeed, suppose that there is a maximal ideal M of R which is a δ-ideal and c 6∈ M .
Since R[θ; δ] satisfies (⋄), R is not δ-primitive and, by Proposition 4.11, there exists
an irreducible Darboux element p ∈ R such that p ∈ M . As c 6∈ M , we have p ∤ c.
Moreover, mdc(a, b) = 1 implies that either p ∤ a or p ∤ b. Thus either p ∤ ca = δ(x)
or p ∤ cb = δ(y). This shows that either δ(x) 6∈ Rp or δ(y) 6∈ Rp, and therefore
δ(R) * Rp. Applying now Theorem 4.12, we obtain that R[θ; δ] does not satisfy (⋄),
a contradiction.
(3) Using (2) above, we can construct examples of differential operator rings R[θ; δ] which
do not satisfy (⋄). For instance, choose the polynomials a and b belonging to a maximal
ideal M such that c 6∈ M . For example, set c = x. We know that c 6∈ M = hx − 1, yi.
Now, set a = y ∈ M and b = x − 1 ∈ M . Thus M is a maximal ideal of R which is a
δ-ideal because δ(x − 1) = δ(x) = ca ∈ M and δ(y) = cb ∈ M . As c 6∈ M , the remark
above shows that C[x, y][θ; x(y∂x + (x − 1)∂y] does not satisfy (⋄).
In the next example, we also have that Max(R) ∩ Fδ 6= ∅. In this case, we will show that S
is not primitive and for any M ∈ Max(R) ∩ Fδ and p ∈ M , where p is an irreducible Darboux
element, we have δ(R) ⊆ Rp, and apply Theorem 4.12.
Example 4.16. Let R = C[x, y] with the C-derivation δ = [(x + 1)y − 1]y(∂x − y2∂y).
We start by describing the irrebucible Darboux elements of R. Set δ′ = ∂x − y2∂y, and so
δ = [(x + 1)y − 1]yδ′. Note that if p is an irreducible Darboux element of δ, from p δ(p) =
[(x + 1)y − 1]yδ′(p), it follows that p (x + 1)y − 1, p y or p δ′(p). In the latter case, p is an
irreducible Darboux element of δ′. Let δ′
= ∂x be the ordinary derivation of polynomials
in C[x] and use the classical notation for derivatives, that is, δ′(a) = a′ for all a ∈ C[x].
Suppose that p = Pn
i=0 aiyi is a non-constant Darboux element of δ′ with ai ∈ C[x] and
an 6= 0. Thus n > 0. Otherwise p ∈ C[x] would be a Darboux elements and as the only
Darboux elements of C[x] are the non-zero constants, p would be constant. By computing
C[x]
Since p is a Darboux element of δ′, there exists c ∈ R such that δ′(p) = cp. Moreover, as p
is a polynomial of degree n in y that divides δ′(p), we must have that c is a polynomial of
degree 1 in y, and so c = b0 + b1y with b0, b1 ∈ C[x] and b1 6= 0. Hence,
cp =
b0aiyi +
b1aiyi+1 = b0a0 + (b0a1 + b1a0)y +
(b0ai + b1ai−1)yi + b1anyn+1.
n
Xi=0
n
Xi=0
n
Xi=2
18
ALVERI SANT'ANA AND ROBSON VINCIGUERRA
δ′(p), we obtain
δ′(p) =
a′
iyi −
iaiyi+1 = a′
0 + a′
1y +
(a′
i − (i − 1)ai−1)yi − nanyn+1.
n
Xi=0
n
Xi=1
n
Xi=2
Comparing the coefficients of the monomials in y in both expressions, we have
(1)
(2)
(3)
(4)
a′
0 = b0a0
a′
1 = b0a1 + b1a0
a′
i − (i − 1)ai−1 = b0ai + b1ai−1
∀ 2 ≤ i ≤ n
−nan = b1an.
From the last equation it follows that b1 = −n ∈ C \ {0}. We will show that b0 is zero.
Suppose b0 6= 0. By (1), a0 is a Darboux element in C[x], and so a0 ∈ C. But then
0 = δ(a0) = a′
0 = b0a0 and a0 = 0. Let 1 ≤ i ≤ n be the least index with ai 6= 0. From (3),
it follows that a′
i = b0ai, that is, ai is a Darboux element in C[x], and again ai ∈ C. Thus
0 = a′
i = b0ai, and consequently b0 = 0 or ai = 0, a contradiction. Therefore b0 = 0, c = −ny
and (−ny)p = d(p). From equations (1) to (3), we have
a′
0 = 0,
a′
1 = −na0,
a′
i = (i − 1 − n)ai−1,
∀ 2 ≤ i ≤ n.
Consider a = an. Then, for any 1 ≤ k ≤ n,
an−k =
−1
k
a′
n−(k−1) =
−1
k
(−1)k−1
(k − 1)!
a(k) =
(−1)k
k!
a(k).
Hence, the Darboux elements of δ′ have the form p =Pn
(−1)k
k! a(k)yn−k for some polynomial
a ∈ C[x] . Since a(n+1) = (−1)nn!a′
0 = 0, we have deg(a) ≤ n. On the other hand, it is easy
to check that for any element p of the above form, we have δ′(p) = (−ny)p, and so also is a
Darboux element of δ′. Therefore, a non-constant element p ∈ R is a Darboux element of δ′
if and only if
k=0
p =
n
Xk=0
(−1)k
k!
a(k)yn−k
for some non-zero polynomial a ∈ C[x] with deg(a) ≤ n. Factoring out the leading coefficient
of a, one can assume that a is monic.
k=0
Let p = Pn
(−1)k
k! a(k)yn−k be a non-constant Darboux element of δ′ of degree n in y.
Suppose a ∈ C[x] is a monic polynomial such that 2 ≤ deg(a) ≤ n. Since C is algebraically
closed, a = uv for some u, v ∈ C[x, y] such that u is monic of degree 1. Now, taking into
account that u′ = 1, for any k > 0,
a(k) = (uv)(k) =
k
Xi=0(cid:18)k
i(cid:19)u(i)v(k−i) = uv(k) + kv(k−1).
ON CYCLIC ESSENTIAL EXTENSIONS OF SIMPLE MODULES
19
Using the fact that v is a polynomial of degree at most n − 1, v(n) = 0 and
n
(−1)k
p = uvyn +
(−1)k
Xk=1
= uy n−1
Xk=0
= (uy − 1) n
Xk=1
k!
k! (cid:16)uv(k) + kv(k−1)(cid:17) yn−k
v(k)yn−k−1! +
(−1)k
(k − 1)!
n
Xk=1
v(k−1)yn−k! .
(−1)k−1
(k − 1)!
v(k−1)yn−k
Hence, if p is an irreducible Darboux element of δ′, deg(a) < 2. If a is constant, then a = 1 and
p = yn. Being irreducible forces p = y. If a has degree one, then a′ = 1 and p = ayn − yn−1.
Since p is irreducible, p = (x + µ)y − 1 for some µ ∈ C. This shows that every irreducible
Darboux elements of δ′ (and therefore of δ) is a scalar multiple of one of the following:
p = y
or
qµ = (x + µ)y − 1
for some µ ∈ C.
Note that qµ1 = (x + µ1)y − 1 and qµ2 = (x + µ2)y − 1 are non-associated whenever
µ1 6= µ2. Therefore there are infinitely many non-associated irreducible Darboux elements of
δ. By Proposition 4.11, we have that S is not primitive.
We show now that if there exists an irreducible Darboux element p in M ∈ Max(R) ∩ Fδ,
then δ(R) ⊆ Rp. Let M = hx − α, y − βi be a maximal ideal which is a δ-ideal for some
α, β ∈ C. Thus δ(x − α) = [(x + 1)y − 1]y ∈ M , and so (x + 1)y − 1 ∈ M or y ∈ M . If
If y ∈ M , β = 0 and
M = hx − α, yi. Therefore, the maximal ideals which are δ-ideals are of the form
(x + 1)y − 1 ∈ M , (α + 1)β − 1 = 0 and M = Dx − α, y − 1
α+1E.
α + 1(cid:29) .
M =(cid:28)x − α, y −
M = hx − α, yi
or
1
If M = hx − α, yi, as qµ(x, y) = (x + µ)y − 1 6∈ M , the only irreducible Darboux elements
contained in M are the scalar multiples of p(x, y) = y. Note that δ(R) ⊆ Ry, since δ(x), δ(y) ∈
Ry. If M =Dx − α, y − 1
qµ(x, y) = (x + µ)y − 1 ∈ M ⇔ qµ(cid:18)α,
α+1E, obviously p(x, y) = y 6∈ M . Moreover,
1
α + 1(cid:19) = (α + µ)
α + 1
1
− 1 = 0 ⇔ µ = 1.
Hence, the only irreducible Darboux elements contained in M are the scalar multiples of
q1(x, y) = (x + 1)y − 1. Note that δ(R) ⊆ R[(x + 1)y − 1] because δ(x), δ(y) ∈ R[(x + 1)y − 1].
Therefore, it follows from Theorem 4.12 that R[θ; δ] satisfies (⋄).
Acknowledgments
This paper is part of the doctoral thesis of the second author at the Federal University of
Rio Grande do Sul, Brazil. The second author wishes to thank his co-supervisor Paula A. A.
B. Carvalho for her guidance, assistance and understanding. Both authors would also like to
thank Christian Lomp for all of his comments and suggestions. Finally, they would like to
thank Rene Baltazar for fruitful discussions and Marcelo Escudeiro Hernandes for suggesting
several references needed for this work.
20
ALVERI SANT'ANA AND ROBSON VINCIGUERRA
References
[1] Archer, J., Derivations on commutative rings and projective modules over skew polynomial rings, PhD
thesis, Leeds University, 1981.
[2] Brumatti, P., Lequain, Y., Levcovitz, D., Differential simplicity in polynomial rings and algebraic inde-
pendence of power series, J. London Math. Soc. 68, (2), 615-630, 2003.
[3] Carvalho, P. A. A. B., Hatipoglu, C., Lomp, C., Injective hulls of simple modules over differential operator
rings, Communications in Algebra 10, 4221-4230, 2015.
[4] Carvalho, P. A. A. B., Lomp, C., Pusat-Yilmaz, D., Injective modules over down-up algebras, Glasg. Math.
J. 52, (A), 53-59, 2010.
[5] Carvalho, P. A. A. B., Musson, I. M., Monolithic modules over Noetherian rings, Glasg. Math. J. 53, (3),
683-692, 2011.
[6] Dahlberg, R. P., Injective Hulls of Simple sl(2, C) Modules are Locally Artinian, Proc. Amer. Math. Soc.
107, (1), 35-37, 1989.
[7] Davis, E. D., Prime ideals and prime sequences in polynomial rings, Proc. Amer. Math. Soc. 72, 33-38,
1978.
[8] Eisenbud, D., Commutative Algebra with a view toward Algebraic Geometry, Graduate Texts in Mathe-
matics (Springer), Vol. 150, 1995.
[9] Goodearl, K. R., Warfield R. B. Jr., An introduction to noncommutative Noetherian rings, LMS Student
Texts Vol. 61, Cambridge University Press, 2004.
[10] Goodearl, K. R., Warfield R. B. Jr., Krull dimension of differential operator rings, Proc. London Math.
Soc. 45, (3), 49-70, 1982.
[11] Goodearl, K. R., Warfield R. B. Jr., Primitivity in differential operator rings, Math. Z. 180, 503-523,
1982.
[12] Hart, R., Krull dimension and global dimension of simple Ore-extensions, Math. Z. 121, 341-345, 1971.
[13] Hatipoglu, C., Lomp, C., Injective hulls of simple modules over finite dimensional nilpotent complex Lie
superalgebras, J. Algebra 361, 79 -- 91, 2012.
[14] Jacobson, N., Structure of rings, American Mathematical Society, Colloquium Publications, Vol. 37. 190
Hope Street, Prov., R. I., 1956.
[15] Jategaonkar, A. V., Jacobson's conjecture and modules over fully bounded Noetherian rings, J. Algebra
30, 103-121, 1974.
[16] Jordan, D. A., Differentially simple rings with no invertible derivatives, Quart. J. Math. Oxford 32, (2),
417-424, 1981.
[17] Kaplansky, I., Commutative rings, Allyn and Bacon, Boston, 1970.
[18] Krause, G. On the Krull-dimension of left noetherian left Matlis-rings, Math. Z., 118, 207-214, 1970.
[19] Lam, T. Y., A first course in noncommutative rings, Graduate Texts in Mathematics, Springer-Verlag,
New York, 1991.
[20] Limanov, L. M., Locally nilpotent derivations, a new ring invariant and applications, Lecture notes, Bar-
Ilan University, 1998. Avail. at http://www.math.wayne.edu/∼lml/.
[21] Matlis, E., Injective modules over Noetherian rings, Pacific J. Math. 8, 511-528, 1958.
[22] Matlis, E., Modules with descending chain conditions, Trans. Amer. Math. Soc. 97, (3), 495-508, 1960.
[23] Musson, I. M., Finitely generated, non-Artinian monolithic modules, New Trends in Noncommutative
Algebra, Contemp. Math., Amer. Math. Soc., 562, 211 -- 220, 2012.
[24] Musson, I. M., Some examples of modules over Noetherian rings, Glasg. Math. J. 23, 9-13, 1982.
[25] Nowicki, A., Polynomial derivations and their rings of constants, Nicolaus Copernicus University, Torun,
1994.
[26] Reyes, M. (http://mathoverflow.net/users/778/manny-reyes), Krull dimension less or equal than tran-
scendence degree?, MathOverflow, URL: http://mathoverflow.net/q/79974 (version: 2012-11-30).
[27] Seidenberg, A., Differential ideals in rings of finitely generated type, Amer. J. Math. 89, 22-42, 1967.
[28] Sharp, R. Y., Steps in commutative algebra, Cambridge University Press, 2000.
[29] Singer, M. F., Liouvillian first integrals of differential equations, Trans. Amer. Math. Soc. 333, (2),
673 -- 688, 1992.
[30] Voskoglou, M. G., Differential simplicity and dimension of a commutative ring, Riv. Mat. Univ. Parma
4, (6), 111-119, 2001.
ON CYCLIC ESSENTIAL EXTENSIONS OF SIMPLE MODULES
21
Alveri Sant'Ana, Instituto de Matem´atica, Universidade Federal do Rio Grande do Sul,
Brazil
E-mail address: [email protected]
Robson Vinciguerra, Instituto de Matem´atica, Universidade Federal do Rio Grande do Sul,
Brazil
E-mail address: [email protected]
|
1701.06329 | 1 | 1701 | 2017-01-23T10:58:36 | Invariant Theory of finite general linear groups modulo Frobenius powers | [
"math.RA",
"math.CO",
"math.RT"
] | We prove some cases of a conjecture of Lewis, Reiner and Stanton regarding Hilbert series corresponding to the action of $Gl_n(\mathbb{F}_q)$ on a polynomial ring modulo Frobenius powers. We also give a few conjectures about the invariant ring for certain cases that we don't prove completely. | math.RA | math | Invariant theory of finite general linear groups modulo Frobenius
powers
Pallav Goyal
Abstract
We prove some cases of a conjecture of Lewis, Reiner and Stanton regarding Hilbert series
corresponding to the action of Gln(Fq) on a polynomial ring modulo Frobenius powers. We also
give a few conjectures about the invariant ring for certain cases that we don't prove completely.
1
Introduction
The subject of this paper is a conjecture inspired by the following celebrated result by L.E. Dickson
[2]:
Theorem 1.1 (Dickson). When G = Gln(Fq) acts via invertible linear substitution of variables on
the polynomial ring S = Fq[x1, x2, · · · , xn], the G-invariant ring is given as:
SG = Fq[Dn,0, Dn,1, · · · , Dn,n−1].
Here, the Dn,i's are the Dickson polynomials defined via the identity:
Y(t + l(x)) =
n
Xi=0
Dn,itqi
where the product is taken over all linear functionals l(x) over S. Building on this, we wish to
consider the action of G on the ring Q = S/m[qm] where m[qm] = (xqm
n ). As m[qm]
remains invariant under the action of G, the action of G on Q is well defined. We wish to describe
the structure of the invariant ring QG and work towards proving the following conjecture by J.
Lewis, V. Reiner and D. Stanton [3]:
2 , · · · , xqm
1 , xqm
Conjecture 1.2. The Hilbert series for the above action of Gln(Fq) on Q is given by:
7
1
0
2
n
a
J
3
2
]
.
A
R
h
t
a
m
[
1
v
9
2
3
6
0
.
1
0
7
1
:
v
i
X
r
a
Hilb(QG, t) =
min(n,m)
Xk=0
t(n−k)(qm−qk)(cid:20)m
k(cid:21)q,t
1−tqm −qi
1−tqk −qi .
where (cid:2)m
k(cid:3)q,t =
k−1
Qi=0
The layout of the paper is as follows. Section 2 deals with the fairly easy case of the problem
when m = 1. Although this case has already been dealt with in [3], it will set the stage for
the following sections to flow more naturally. Section 3 tackles the case m = 2 in 2 stages, first
1
assuming that n = 2, and then, generalising for an arbitrary n. Section 4 uses the ideas developed
in Section 3 to talk about the 'k = 1' case of the Hilbert series. Section 5 is about the action of
Steenrod operations on the invariant polynomials when m = 2. Section 6 develops certain ideas
that approach a solution for the case n = k = m − 1. Section 7 records certain simplifications
that arise when we take F2 as the underlying field. Finally, Section 8 deals with the parabolic
generalization of the conjecture with focus on the case m = 2.
1.1 Preliminaries
Before starting off, we state a few results which will be used multiple times throughout the paper.
Lemma 1.3. A polynomial is invariant under the action of Gln(Fq) if it is invariant under diagonal
matrices, permutations of the variables, and the substitution x1 → x1 + x2.
Proof. First, we note that the group Gln(Fq) is generated by diagonal matrices, permutation ma-
trices and elementary matrices. Hence, by symmetry, checking that a given polynomial is invariant
amounts to checking its invariance under the former two and the substitution x1 → x1 + x2.
Lemma 1.4. Q has a basis of monomials of the form xi1
n where 0 ≤ ij ≤ qn − 1 as an
Fq-vector space. Furthermore, if y ∈ QG, then y is a linear combination of monomials where each
ij is divisible by q − 1.
1 · · · xin
Proof. The first claim of the lemma is rather obvious. For the second claim, we note that for y to
be invariant, it must be invariant under the action of the diagonal matrices.
Finally, we state a well known result from number theory. For proof, see [5, Page 126].
Theorem 1.5 (Lucas' Theorem). Given positive integers m and n, if n = n0+n1p+n2p2+· · ·+ndpd
and m = m0 + m1p + m2p2 + · · · + mdpd are their respective base p expansions, then:
(cid:18) n
m(cid:19) ≡(cid:18) n0
m0(cid:19)(cid:18) n1
m1(cid:19) · · ·(cid:18) nd
md(cid:19) (mod p).
2 The case m = 1
We have Q = Fq[x1, x2, · · · , xn]/m[q] where m[q] = (xq
n). Let y be an element of the
invariant ring QG. Then, by Lemma 1.4, y must be a linear combination of elements of the form
xi1
1 xi2
n be a monomial occurring in y
where b ∈ Fq \ {0}.
n where ij = 0 or q − 1 for all j. Now, let bxj1
2, · · · , xq
2 · · · xjn
2 · · · xin
1 xj2
1, xq
I claim that j1 = j2 = · · · = jn. Suppose not. Without loss of generality, let j1 = j2 = · · · =
1 xq−1
· · · xq−1
jr = q − 1 and jr+1 = · · · = jn = 0 where 0 < r < n. Hence, the monomial is bxq−1
.
2
If we perform the substitution xr → xr + xr+1 in y, the coefficient of xq−1
· · · xq−2
r xr+1 in y
becomes non-zero. But this isn't possible as y does not contain any such monomial except possibly
when q = 2. In that particular case, the coefficient of xq−1
r xr+1 = x1x2 · · · xr−1xr+1 in
y either changes from 1 to 0 or from 0 to 1, which isn't allowed in the invariant y. Hence, we have
a contradiction.
1 xq−1
1 xq−1
· · · xq−2
Therefore, y is of the form c0 + c1xq−1
. It is trivial to see that such a y is invariant
under the action of G. Hence, the Hilbert series is 1 + tn(q−1), which agrees with what has been
conjectured.
1 xq−1
· · · xq−1
2
2
n
r
2
2
3 The case m = 2
We first solve the problem for the case when n = 2, and then show how the solution can be
generalised for arbitrary n.
3.1 m = 2, n = 2
We have Q = Fq[x1, x2]/m[q2] where m[q2] = (xq2
linear combination of elements of the form xi1
1 xi2
We'll now prove the following theorem which is in accord with the conjecture at hand.
1 , xq2
2 ). By Lemma 1.4, we see that any y ∈ QG is a
2 where 0 ≤ i1, i2 ≤ q2 − 1 and (q − 1)i1, (q − 1)i2.
Theorem 3.1. The Hilbert series for QG is
1 + tq2−q(1 + tq−1 + t2(q−1) + · · · + tq2−q) + t2(q2−1).
A basis of the invariant ring as an Fq-vector space is given as
1, xk(q−1)
1
xk(q−1)
2
for 0 ≤ k ≤ q.
x(q−k+1)(q−1)
1
xq−1
1 − xq−1
2
2
− x(q−k+1)(q−1)
and xq2−1
1
xq2−1
2
In order to prove this, we'll show that there exist unique invariant polynomials in Q of degrees
q2 − q + c(q − 1) and 2(q2 − 1) where 0 ≤ c ≤ q, and that there do not exist any other invariant
polynomials (up to scalar multiples). Without loss of generality, let us assume that y is a homoge-
neous G-invariant polynomial. Then, the degree d of y must be some multiple of q − 1 between 1
and 2(q2 − 1). Let d = k(q − 1), where 0 ≤ k ≤ 2(q + 1).
Lemma 3.2. If k < q or k = 2q + 1, we don't have any invariant polynomials y of degree k(q − 1).
Proof. First, consider the case q = 2. If k < q, we have k = 1, and so, y is of the form c0x1 + c1x2.
If k = 2q + 1 = 5, y is of the form c0x3
2. Neither of these can be invariant under the
substitutions x1 → x1 + x2 and x2 → x1 + x2 unless c0 = c1 = 0.
2 + c1x2
1x2
1x3
Now, suppose q > 2 and k < q. Let k = ptr where p ∤ r. A general expression of y is the form:
y = c0xk(q−1)
1
+ c1x(k−1)(q−1)
1
2 + · · · + ckxk(q−1)
xq−1
2
.
Consider the substitution x1 → x1 + x2. Then, the coefficient of xk(q−1)−pt
1
xpt
2
is equal to
pt
pt (cid:1) =(cid:0)ptr(q−1)
(cid:0)k(q−1)
(cid:1) ≡ r(q − 1) 6≡ 0 mod p (where the second equality follows by Lucas's Theorem).
Hence, the coefficient becomes non-zero after the substitution if c0 6= 0. So, let c0 = 0. By exactly
the same argument, we get that cj = 0 for all j. Hence, y = 0 contradicting the fact that its degree
is non-zero. Thus, such a y can not be invariant. Hence, k ≥ q.
+ xq2−q
Next, let k = 2q + 1. In this case, y = c(xq2−1
x1 → x1 + x2 gives a non-zero coefficient to xq2−2
invariant of degree (2q + 1)(q − 1).
1
1
xq2−q
2
xq2−q+1
2
xq2−1
2
). But here, the substitution
1
unless c = 0. Thus, we do not have an
The cases left to investigate are q ≤ k ≤ 2q and k = 2(q + 1). For k = 2(q + 1), there's a
. It is easy to see that this y remains invariant under all
unique choice for y, i.e. y = cxq2−1
xq2−1
2
the possible transformations from G.
1
3
Lemma 3.3. For q ≤ k ≤ 2q, we have unique (up to scalar multiples) invariant polynomials of
degree k(q − 1).
Proof. Fix k and let k′ = k − q. Then, a general expression of an invariant polynomial y of degree
k is of the form:
y = c0xq2−1
1
x(k′−1)(q−1)
2
+ c1xq2−q
1
xk′(q−1)
2
+ · · · + cq−k′+2x(k′−1)(q−1)
1
xq2−1
2
.
1
x(k′−1)(q−1)+1
2
I claim that c0 = 0 and c1 6= 0. If c0 6= 0, then on the substitution x1 → x1 + x2, the coefficient
becomes c0(q2 − 1) 6= 0, which isn't allowed. Thus, c0 = 0. Next suppose
In the above polynomial, ci is the coefficient
. Let q − i + 1 = ptr where p ∤ r. As i 6= 1, pt < q. Then, the
which is again
of xq2−2
c1 = 0. Choose the smallest i such that ci 6= 0.
of x(q−i+1)(q−1)
same substitution gives a non-zero coefficient to x(q−i+1)(q−1)−pt
impossible. Hence, c1 6= 0. Without loss of generality, c1 = 1. I claim that
x(k′+i−1)(q−1)+pt
2
x(k′+i−1)(q−1)
2
1
1
y = yk′ := xq2−q
1
xk′(q−1)
2
+ x(q−1)2
1
x(k′+1)(q−1)
2
+ · · · + xk′(q−1)
1
xq2−q
2
.
That yk′ is invariant, will be proven in a more general setting in Proposition 4.1. So, we only
need to check that this choice of y is forced, and thus, unique. But, this is easy to see. Having
x(k′+i−1)(q−1)+pt
fixed c1 = 1, after the substitution x1 → x1 + x2, the coefficient of x(q−i+1)(q−1)−pt
2
should be equal to 0 (except in the case when q = 2 and k = 1 when this coefficient should be 1),
and using this, the values of the other ci's gets uniquely determined.
1
We illustrate the last step of the proof with an example.
1x3
1x6
2 + c4x3
1x9
2 + c3x6
1 + c2x9
Example 3.4. Consider the case q = 4 and k′ = 0. Then, a general expression for y is of the
form c1x12
2 . As above, we'll assume c1 = 1 and perform the
substitution x1 → x1 + x2. Then, the coefficient of x8
1(cid:1) which should be 0.
1(cid:1) 6= 0. Next, we try to find c3. As 26 and 4 ∤ 6, we consider the coefficient
2(cid:1) 6= 0,
2 becomes c1(cid:0)12
2(cid:1) which must be 0. Again, as (cid:0)6
of x6−2
c3 gets uniquely determined. Proceeding similarly, considering the coefficient of x2
2 , we get the
value of c4, and then, c5 = c1 by symmetry. Hence, c1 determines all the other coefficients uniquely.
2. This is equal to c1(cid:0)12
This determines c2 as (cid:0)9
8(cid:1) + c2(cid:0)9
4(cid:1) + c2(cid:0)9
5(cid:1) + c3(cid:0)6
2 + c5x12
2 = x8
1 x6+2
1x10
1x4
1x4
xk′(q−1)
2
Remark 3.5. Up till now, we've only described the invariant ring QG in terms of a linear basis
as an Fq-vector space. Some multiplicative properties can be easily observed. Recall that yk′ =
xq2−q
for 0 ≤ k′ ≤ q. It is a matter of simple
1
computations to check that y2
2 = 0. All the products yiyj, apart
from these 3, are zero because of degree arguments. For example, y0y1 has degree 2q2 − q − 1, but
we know that there aren't any invariant polynomials of this degree, and so, y0y1 = 0.
+ · · · + xk′(q−1)
0 = yq, y0y2 = −xq2−1
xq2−q
2
xq2−1
2
x(k′+1)(q−1)
2
+ x(q−1)2
and y2
1
1
1
With this, we have now proven Theorem 3.1 and have a complete description of the structure
of the invariant ring.
3.2 m = 2, arbitrary n
Now, we work towards solving the problem for a general n when m = 2. We have that Q =
Fq[x1, x2, · · · , xn]/m[q2] where m[q2] = (xq2
n ). The aim of this section will be to gener-
2 , · · · , xq2
1 , xq2
4
alise the ideas from the case n = 2. We define the polynomials:
zn :=
xq2−1
i
n
Yi=1
and
yn,k :=
X
i1+i2+···+in=(n−1)q+k
0≤i1,i2,··· ,in≤q
(cid:18) n
Yj=1
x
ij(q−1)
j
(cid:19)
for 0 ≤ k ≤ q. In fact, yn,k can be defined for k > q using the same definition. But in those cases,
the bounds on the ij's would imply that the sum is an empty sum, and so, yn,k = 0 for k > q.
Lemma 3.6. zn and yn,k remain invariant under the action of G.
Proof. The proof is by induction. The case when n = 1 is easily verified and the case n = 2 has
already been solved and acts as base case of the induction.
zn can be trivially seen to be invariant under the action of Gln(Fq) on Q. Now, fixing k, we
consider yn,k. As this expression is symmetric in the xi's, it remains invariant under the action of
the permutation matrices. Also, as all the exponents are divisible by q − 1, diagonal matrices too
act trivially on yn,k. Therefore, by Lemma 1.3, we only need to check the action of the substitution
x1 → x1 + x2 on yn,k. Here, we make the observation that yn,k can be written as:
X
i1+i2+···+in=(n−1)q+k
0≤i1,i2,··· ,in≤q
(cid:18) n
Yj=1
xij(q−1)
j
(cid:19)
(cid:18) n−1
Yj=1
X
i1+i2+···+in−1=(n−1)q+k−in
0≤i1,i2,··· ,in−1≤q
xij (q−1)
j
(cid:19)!xin(q−1)
n
yn,k =
=
=
=
q
q
Xin=0
Xin=0
Xi=k
q
yn−1,q+k−inxin(q−1)
n
yn−1,q+k−ixi(q−1)
n
where, in the last step, we have discarded some of the initial terms of the sum, because yn−1,q+k−in
is non-zero only when q + k − in ≤ q. When we apply the transformation x1 → x1 + x2 to the
final expression obtained above, yn−1,q+k−i remains invariant by the induction hypothesis and xn
remains invariant as n > 2. Hence, yn,k remains invariant under the action of G.
Now, we state the main result of this section:
Theorem 3.7. The Hilbert series for QG is given as:
Hilb(QG, t) = 1 + t(n−1)(q2−q)(1 + tq−1 + t2(q−1) + · · · + tq2−q) + tn(q2−1)
when n ≥ 2. For n = 1, the Hilbert series is 1−tq2
Fq-vector space is given as 1, yn,k and zn for 0 ≤ k ≤ q.
1−tq−1
+q−2
. A basis for the invariant ring QG as an
5
As the case when n = 1 is trivially true, we'll assume n ≥ 2 here onwards. By Lemma 3.6 we
only need to check that these are the only invariants in Q up to scalar multiples. The proof will
be done in 2 steps. Firstly, we'll try to find the possible degrees for an invariant polynomial. Next,
for those degrees, we'll prove that there exist unique invariant polynomials up to scalar multiples.
As we have already proven the existence of at least one such polynomial for each degree, we'll be
done.
Step 1: Let y ∈ QG be a homogeneous polynomial of degree r. Express y as:
y = c0 + c1xq−1
n + c2x2(q−1)
n
+ · · · + ckxk(q−1)
n
(1)
for some k where ci ∈ Fq[x1, x2, · · · , xn−1]/(xq2
If y is to
be invariant under the action of G, each of the ci's must also be invariants. By the induction
hypothesis, any such non-scalar ci must have degree greater than or equal to (n − 2)(q2 − q).
Therefore, degree of y is greater than or equal to (n − 2)(q2 − q). We claim that r = n(q2 − 1) or
(n − 1)(q2 − q) ≤ r ≤ n(q2 − q). The following lemmas prove this by showing that r can not take
any other values.
n−1) for all i and ck 6= 0.
2 , · · · , xq2
1 , xq2
Lemma 3.8. For an invariant polynomial y, the degree r can not be less that (n − 1)(q2 − q).
Proof. As we have already shown that r ≥ (n−2)(q2−q), suppose (n−2)(q2−q) ≤ r < (n−1)(q2−q).
We can assume that none of the ci's is a scalar multiple of zn−1, because due to symmetry, we would
need to multiply it by xq2−1
which would exceed the assumed bound on degrees. Also, we must
have that at least one of the ci's is non-scalar, otherwise, y won't be symmetric. In any such ci,
the maximum degree of any xj for j ≤ n − 1 is q2 − q. So, the maximum degree of xn in y, which
is k(q − 1), should also be equal to q2 − q, and thus, k = q.
n
Next, ck = cq can not be scalar, because that would imply that the degree of y is equal to
q2 − q < (n − 2)(q2 − q) except when n = 3 (in which case, we would need to have an invariant of
degree q2 − q which is not possible), and thus, give a contradiction. So, cq is not scalar, and so,
the degree of cqxq2−q
is greater than or equal to (n − 2)(q2 − q) + (q2 − q) = (n − 1)(q2 − q) which
exceeds the bound on the degree of y, and so, such a y can't exist.
n
Lemma 3.9. For an invariant polynomial y, if r is the degree, we can't have n(q2 − q) < r <
n(q2 − 1).
Proof. Let us assume the contrary. Again writing y in the form as in Equation (1), we get the
following similar observations: k = q, at least one of the ci's must be non-scalar and none of the
ci's is a scalar multiple of zn−1. This implies that deg(y) = r = degree of cqxq(q−1)
≤ (n − 1)(q2 −
q) + (q2 − q) = n(q2 − q) which is again a contradiction. Hence, an invariant polynomial can't have
its degree in the given range.
n
Step 2: Step 1 makes it clear that the degree of y is either 0, n(q2 − 1) or a multiple of q − 1
between (n − 1)(q2 − q) and n(q2 − q) (both included). Now, we verify the uniqueness. The fact is
trivially true for an invariant of degree 0 or n(q2 − 1).
Now, suppose we have an invariant y of degree ((n−1)q+k)(q−1) for some k such that 0 ≤ k ≤ q.
First, we express y as in Equation (1).
In this sum, each ci must be some scalar multiple of
yn−1,q+k−i. Without loss of generality, let cq = yn−1,k. Now, each of the ci's contain some monomial
(say Mi) having xq2−q
as a factor. As y is assumed to be invariant, applying the transformation
1
6
x1 → xn, xn → x1, Mi transforms into a monomial occurring in cqxq2−q
. All such
monomials have scalar coefficient equal to 1 by induction. Hence, the scalar coefficient of Mi must
be 1 too. This implies that for each i, ci = yn−1,q+k−i, and thus, given its degree, y is determined
uniquely up to scalar multiples. This completes the proof of the theorem.
= yn−1,kxq2−q
n
n
4 The case k = 1
After having dealt with the case m = 2, it becomes easier to make some predictions for the invariant
ring in the general case. Recalling the conjectured Hilbert series for the invariant ring:
Hilb(QG, t) =
t(n−k)(qm−qk)(cid:20)m
k(cid:21)q,t
.
min(n,m)
Xk=0
q−1 , where:
yk′ := xqm−q
1
When k = 1, the summand in the above expression is equal to t(n−1)(qm−q)(cid:2)m
discussion above, we claim that we can find invariant polynomials whose degrees correspond to this
'k = 1' term of the Hilbert series. For simplicity, first we assume that n = 2.
Proposition 4.1. Taking Q = Fq[x1, x2]/(xqm
k′ ≤ qm−q
2 ), the polynomial yk′ is G-invariant for 0 ≤
1(cid:3). Inspired by our
1 , xqm
xk′(q−1)
2
+ xqm−2q+1
1
x(k′+1)(q−1)
2
+ · · · + x(k′+1)(q−1)
1
xqm−2q+1
2
+ xk′(q−1)
1
xqm−q
2
.
Proof. In order to prove this, we need to check that each yk′ remains invariant under the transfor-
mation x1 → x1 + x2. First, let k′ = 0. Then, after the above substitution, y0 becomes:
+ xqm−q
(x1 + x2)qm−q + (x1 + x2)qm−2q+1xq−1
2 + · · · + (x1 + x2)q−1xqm−2q+1
(2)
2
2
.
The coefficient of xt
where those binomial coefficients which do not make sense are assumed to be zero. This sum is
also the coefficient of xt in:
in the substituted expression is(cid:0)qm−q
t (cid:1)+(cid:0)qm−2q+1
(cid:1)+· · ·+(cid:0)q−1
t (cid:1)+(cid:0)0
t(cid:1)
1xqm−q−t
2
t
(1 + x)qm−q + (1 + x)qm−2q+1 + · · · + (1 + x)q−1 + 1 =
=
=
=
=
(1 + x)(qm−1) − 1
(1 + x)q−1 − 1
(1+x)qm
1+x − 1
(1+x)q
(1+x) − 1
1+xqm
1+x − 1
1+xq
1+x − 1
xqm
− x
xq − x
xqm−1 − 1
xq−1 − 1
= 1 + xq−1 + x2(q−1) + · · · + xqm−q.
And so, the coefficient of xt
in the substituted expression is 1 if and only if q − 1t and is
0 otherwise. Thus, the substituted expression is equal to y0, and thus, y0 remains invariant under
the action of G.
2
1xqm−q−t
7
Remark 4.2. Up till now, we haven't used the fact that xqm
invariant in S itself.
1 = xqm
2 = 0. Hence, for each m, y0 is
For any other k′, the transformation x1 → x1 + x2 turns yk′ to:
(x1 + x2)qm−qxk′(q−1)
2
+ (x1 + x2)qm−2q+1x(k′+1)(q−1)
2
+ · · · + (x1 + x2)k′(q−1)xqm−q
2
.
(3)
Here, we need to check the coefficient of xt
for (k′ − 1)(q − 1) ≤ t ≤ qm − q.
This is because, for smaller t, the exponent of x2 exceeds qm − 1, and so, the term becomes zero.
1xqm+(k′−1)(q−1)−1−t
2
For t > (k′ − 1)(q − 1), the coefficient is equal to (cid:0)qm−q
coefficient of xt in
t (cid:1) +(cid:0)(qm−2q+1
(1 + x)qm−q + (1 + x)qm−2q+1 + · · · + (1 + x)k′(q−1).
t
(cid:1) + · · · +(cid:0)k′(q−1)
t
(cid:1) which is the
(4)
In fact, we can continue the above sum up to 1 as the addition of these terms won't contribute
to a coefficient of xt. Hence, working similarly as when k′ = 0, we get the desired coefficients. For
t = (k′ − 1)(q − 1), we get coefficient equal to 1 because of the addition of the term (1 + x)(k′−1)(q−1)
to the above sum. Subtracting this, we get its coefficient to be zero, and so, we conclude that y′
k
remains invariant under the given transformation.
This proposition also provides us the proof of the invariance of the polynomials we talked about
in Section 3.1.
Now, for a general n, mimicking the strategy followed by us in Section 3.2, we have an easy
corollary to the above proposition.
Corollary 4.3.
am,n,k′ :=
X
i1+i2+···+in=(n−1) qm −q
0≤i1,i2,··· ,in≤ qm−q
q−1 +k′
q−1
(cid:18) n
Yj=1
xij(q−1)
j
(cid:19)
for 0 ≤ k′ ≤ qm−q
q−1 are invariant polynomials in the ring Q for any n, m.
The proof of this corollary uses exactly the same idea as used in Lemma 3.6 after the following
observation:
am,n,k′ =
=
=
X
i1+i2+···+in=(n−1) qm −q
0≤i1,i2,··· ,in≤ qm−q
q−1 +k′
q−1
(cid:18) n
Yj=1
xij (q−1)
j
(cid:19)
qm−q
q−1
Xin=0
X
i1+i2+···+in−1=(n−1) qm−q
0≤i1,i2,··· ,in−1≤ qm−q
q−1
q−1 +k′−in
(cid:18) n−1
Yj=1
xij(q−1)
j
(cid:19)!xin(q−1)
n
qm−q
q−1
Xin=0
am,n−1,q+k′−inxin(q−1)
n
8
=
am,n−1,q+k′−ixi(q−1)
n
.
qm−q
q−1
Xi=k′
1 , xqm
Hence, the proven proposition and corollary provide us a set of invariant polynomials in the
ring Fq[x1, x2, · · · , xn]/(xqm
n ) which correspond to the the case 'k = 1' as for given
values of m and n, we have homogeneous invariant polynomials am,n,k′ for 0 ≤ k′ ≤ qm−q
q−1 , where
the degree of am,n,k′ is equal to (n − 1)(qm − q) + k′(q − 1). The Hilbert series for the polynomials is
thus given as t(n−1)(qm−q) + t(n−1)(qm−q)+q−1 + · · · + tn(qm−q) = t(n−1)(qm−q) 1−tqm−1
which is exactly the conjectured value.
2 , · · · , xqm
1−tq−1 = t(n−1)(q−1)(cid:2)m
1(cid:3),
5 Action of Steenrod operators
Now that we are done with the analysis of the 'k = 1' case of the Hilbert series, before moving on
to further cases, we study the dependencies of the invariants we have talked about, viewed from
the perspective of Steenrod operations. Throughout this section, we'll assume that n = 2.
Recall (see [4]) that for a polynomial f (x1, x2, · · · , xn) ∈ Fq[x1, x2, · · · , xn], we define the action
of Steenrod operators P i on f as follows:
P (ξ)(f ) : = f (x1 + xq
1ξ, x2 + xq
2ξ, · · · , xn + xq
nξ)
P i(f )ξi.
=
∞
Xi=0
It is easy to see that for any f , P 0(f ) = f . Before moving towards the results of this section,
we prove a lemma that motivates the study of these operators in the context of our problem.
Lemma 5.1. The Steenrod operators preserve the ideal m[qm] = (xqm
have a well-defined action on Q.
2 , · · · , xqm
1 , xqm
n ), and hence,
Proof. For any polynomial f and g ∈ m[qm], by the definition of the operators, we have
P d(f g) =
d
Xi=0
P i(f )P d−i(g)
for any d. Thus, we only need to verify our claim for the generators of the ring: xqm
, xqm+1
This is easy to check as P (ξ)(xqm
and xqm
) = (xi + xq
= xqm
ξqm
i + xqm+1
i
i ξ)qm
i
i
i
i
for 1 ≤ i ≤ n.
∈ m[qm].
Remark 5.2. The action of the Steenrod operators on the Dickson invariants Dn,i can be found
(when q = p) in [6, §II].
In this section, we try to find a minimal additive basis for QG with respect to the action of the
Steenrod operators (For a connection to the 'hit problem' for Steenrod algebra, see [1])
9
5.1 m = 2
When m = 2, we had found the invariants in Section 3.1 and they were named yk′ where 0 ≤ k′ ≤ q.
Using the same notation, and taking k = k′ + q, we check the action of the first Steenrod operation
on each yk′. For compactness, we write yk′ =
q
Pi=k′
q
x(k−i)(q−1)
1
xi(q−1)
2
. Then, we have:
P (yk′) =
(x1 + xq
1ξ)(k−i)(q−1)(x2 + xq
2ξ)i(q−1).
Xi=k′
The action of the first Steenrod operator on yk equals the coefficient of ξ in the above expression.
Hence,
P 1(yk′) =
=
q
q
Xi=k′
Xi=k′
{x(k−i)(q−1)
1
xi(q−1)−1
2
2i(q − 1) + x(k−i)(q−1)−1
xq
1
1(k − i)(q − 1)xi(q−1)
xq
2
}
{−i.x(k−i)(q−1)
1
x(i+1)(q−1)
2
+ (i − k).x(k−i+1)(q−1)
1
xi(q−1)
2
}
q
= −kx(k−k′+1)(q−1)
1
xk′(q−1)
2
− k
q
{x(k−i+1)(q−1)
1
Xi=k′+1
xi(q−1)
2
}
+
Xi=k′
= −kx(k−k′+1)(q−1)
= (1 − k)yk′+1.
1
{i(x(k−i+1)(q−1)
1
xi(q−1)
2
− x(k−i)(q−1)
1
x(i+1)(q−1)
2
)}
xk′(q−1)
2
− kyk′+1 + k′x(k−k′+1)(q−1)
1
xk′(q−1)
2
+ yk′+1
Thus, we have proved the following proposition:
Proposition 5.3. For all k′ such that 0 ≤ k′ < q, P 1(yk′) = (1 − k)yk′+1, and hence, when
(1 − k) 6≡ 0 (mod p), we can use P 1 to obtain yk′+1 from yk′.
Next, we prove a proposition that essentially will demonstrate that the action of the Steenrod
operators on just 2 of the invariant polynomials is sufficient for generating all the other invariants.
Proposition 5.4. By applying the Steenrod operations P1, P2, · · · , Pq−2 on y2, we get non-zero
scalar multiples of y3, y4, · · · , yq respectively.
Using the same notation as defined earlier:
q
P (ξ)(y2) =
=
{P (ξ)(x1)}(q+2−i)(q−1){P (ξ)(x2)}i(q−1)
(x1 + xq
1ξ)(q−i)(q−1)(x2 + xq
2ξ)(i+2)(q−1).
q−2
Xi=2
Xi=0
For 1 ≤ r ≤ q − 2, P r(y2) is the coefficient of ξr in P (y2) which is
q−2
Xi=0
r
Xj=0(cid:18)(q − i)(q − 1)
j
(cid:19)x(q−i+j)(q−1)
1
(cid:18)(i + 2)(q − 1)
r − j
(cid:19)x(i+2+r−j)(q−1)
2
.
(5)
Then, the following lemma proves the above proposition.
10
Lemma 5.5. P r(y2) =(cid:0) q−1
q−2
Xi=0
r
q−r−1(cid:1)yr+2 or
Xj=0(cid:18)(q − i)(q − 1)
=(cid:18) q − 1
q − r − 1(cid:19)
j
(cid:19)x(q−i+j)(q−1)
1
(cid:18)(i + 2)(q − 1)
r − j
(cid:19)x(i+2+r−j)(q−1)
2
q−r−2
Xi=0
x(q−i)(q−1)
1
x(i+r+2)(q−1)
2
.
Proof. In order to prove this identity, we first transform it to a simpler form. Firstly, cancel
xq(q−1)
by z1 and z2 respectively, and next,
1
replace z2/z1 by z. Thus, the identity to be proven transforms into:
from both sides. Next, replace xq−1
x(r+2)(q−1)
2
and xq−1
1
2
q−2
Xi=0
r
Xj=0(cid:18)(q − i)(q − 1)
j
(cid:19)(cid:18)(i + 2)(q − 1)
r − j
(cid:19)zi−j =(cid:18) q − 1
Now, as we are in Fq, we have (cid:0)(q−i)(q−1)
j
Finally, putting k = i − j and suitably modifying the limits for summation, we get:
r−j
(cid:1) =(cid:0)i
k(cid:19)(cid:18) q − i − 2
j(cid:1) and (cid:0)(i+2)(q−1)
q − 2 − r − k(cid:19)zk =(cid:18) q − 1
q − r − 1(cid:19)
q−2
Xi=0
i
Xk=i−r(cid:18) i
q−r−2
(6)
zi.
Xi=0
q − r − 1(cid:19)
(cid:1) =(cid:0)q−i−2
r−j (cid:1) by Lucas's Theorem.
Xk=0
q−r−2
zk.
(7)
Next, we notice that for k > i and for k < i − r, the summand on the LHS becomes 0. So,
we can sum over k from 0 to q − 2. Then, the coefficient of zk on the LHS becomes equal to
q−2
k(cid:1)(cid:0) q−i−2
q−2−r−k(cid:1). (For k > q − r − 2, this is zero exactly as in the RHS). This is the coefficient of
akbq−2−r−k in:
Pi=0(cid:0) i
q−2
(1 + a)i(1 + b)q−i−2 = (1 + a)q−2
Xi=0
= (1 + a)q−2
q−2
1 + a(cid:19)q−i−2
Xi=0(cid:18) 1 + b
1+a(cid:19)q−1
1 −(cid:18) 1+b
1 − 1+b
1+a
=
=
=
(1 + a)q−1 − (1 + b)q−1
a − b
q−1
a − b
Pi=0(cid:0)q−1
Xi=0(cid:18)q − 1
i (cid:1)(ai − bi)
i (cid:19) i−1
Xj=0
q−1
ajbi−j−1.
The coefficient of akbq−2−r−k in the above expression is exactly (cid:0) q−1
coefficient of zk in the RHS of the identity we are trying to prove. Hence the two expressions are
equal in Fq.
q−r−1(cid:1) which is exactly the
11
An important observation here is that (cid:0) q−1
q−r−1(cid:1) = (cid:0)q−1
Lucas' Theorem). This signifies that given y0 and y2, we can generate all the other yk's (up to
scalar multiples) by the action of the Steenrod operations on these.
r (cid:1) 6= 0 for the given values of r (again by
5.2 m ≥ 2
Now, we look at the case when n is still 2 and m is arbitrary. Then, corresponding to the 'k = 1'
case, from Section 4, we have the invariants am,2,k′ where 0 ≤ k′ ≤ qm−q
q−1 . Then, we make the
following conjecture about the relationships between these invariants by means of the Steenrod
operations.
: k′ = 0, 1 + qt−1
Conjecture 5.6. For a fixed m, consider the set A := {am,2,k′
q−1 } and the set
q−1 for 1 ≤ t ≤ m − 1}. Then, it is possible to generate all the
B := {am,2,k′
elements of A by the action of Steenrod operations on the elements of the set B. am,2,0 generates
itself and am,2,1. For any t such that 1 ≤ t < m − 1 and k′ = 1 + qt−1
q−1 , am,2,k′ generates am,2,l for
l = k′, k′ +1, k′ +2, · · · , k′ +qt. For t = m−1, am,2,k′ generates am,2,l for l = k′, k′ +1, · · · , k′ +qt −1.
: 0 ≤ k′ ≤ qm−q
What we have already proven agrees with this conjecture for the case m = 2. The proof in
the general case should involve the verification of identities involving binomial coefficients just like
the case m = 2. If we proceed in the same manner as above, it all boils down to proving that the
expression
qm−qt−2q+2
q−1
Xi=0
(cid:18)(q − 1)( qm−q
q−1 − i)
j
(cid:19)(cid:18)(q − 1)( qt+q−2
r − j
q−1 + i)
(cid:19) (mod p)
is independent of j when 0 ≤ j ≤ r.
6 The case n = k = m − 1
Recall that the conjectured Hilbert series is given as:
Hilb(QG, t) =
min(n,m)
Xk=0
t(n−k)(qm−qk)(cid:20)m
k(cid:21)q,t
.
In this section, we'll investigate the case n = k = m − 1 = 2 first, and then try to tackle the
problem for arbitrary n. The reason for treating different values of k separately stems from our
belief that the invariants can be systematically categorised on the basis of the value of k to which
they correspond. For n = 2 and m = 3, the conjectured series is:
t2(q3−1) + t(q3−q) 1 − tq3−1
1 − tq−1 +
(1 − tq3−1)(1 − tq3−q)
(1 − tq2−1)(1 − tq2−q)
.
The first two terms in the above sum are easy to account for: the first one corresponds to the
invariant polynomial xq3−1
1
xq3−1
2
whereas the second one was dealt with in Section 4.
Let the third term be referred to as f (t) which is a polynomial of degree 2(q3 − q2). An
important observation is that f (1/t) = t−2(q3−q2)f (t). This implies that the coefficient of ta equals
12
the coefficient of t2(q3−q2)−a in f (t) where 0 ≤ a ≤ 2(q3 −q2). Now, we try to see how the coefficients
of f (t) look like. Now,
f (t) =
(1 − tq3−1)(1 − tq3−q)
(1 − tq2−1)(1 − tq2−q)
= (1 − tq3−1)(1 − tq3−q)(1 + tq2−1 + t2(q2−1) + · · · )(1 + tq2−q + t2(q2−q) + · · · ).
Lemma 6.1. For any a between 0 and q3 − q2, the coefficient of ta in f (t) is exactly 1 if a is a
non-negative integer linear combination of q2 − 1 and q2 − q, and 0 otherwise.
Proof. For 0 ≤ a ≤ q3 − q2, let us look at the coefficient of ta in the above product. Now, the first
2 terms of the product will have to contribute 1 (as q3 − 1 and q3 − q are greater than a). Hence,
a must be of the form m(q2 − 1) + n(q2 − q) for some m, n ∈ N0. In fact, for a given a, such a
representation is unique. This is because if we have 2 pairs (m1, n1) and (m2, n2) such that
a = m1(q2 − 1) + n1(q2 − q) = m2(q2 − 1) + n2(q2 − q),
then,
(m1 − m2)(q + 1) = (n2 − n1)q.
And so q(m1 − m2).
If m1 6= m2, one of them will have to be ≥ q and that would imply
a ≥ q(q2 − 1) which as a contradiction as a ≤ q3 − q2. Hence, m1 = m2 and n1 = n2.
For q3 − q2 ≤ a ≤ 2(q3 − q2), the coefficient of ta equals the coefficient of t2(q3−q2)−a, and as,
0 ≤ 2(q3 − q2) − a ≤ q3 − q2, we have already determined this coefficient. Hence, all the coefficients
in f (t) are either 0 or 1. Now, recall that S = Fq[x1, x2] and Q = S/(xq3
1 , xq3
2 ).
Proposition 6.2. Corresponding to the term f (t) in the Hilbert series, we can find invariant
polynomials in Q which are images of Dickson invariant polynomials in S.
Proof. The invariant ring SG is generated by the polynomials D2,0 and D2,1 having degrees q2 − 1
and q2 − q respectively. Let their images in Q be denoted by S0 and S1 respectively.
Now, by manual computation, one sees that f (t) = 1 + tq2−q + tq2−1+ higher degree terms. As
degree(S0) = q2 − q and degree(S1) = q2 − 1, these 2 polynomials correspond to the 2nd and 3rd
monomials of f (t) respectively. For any other a such that 0 ≤ a ≤ q3 − q2, the coefficient of ta in
f (t) is 1 iff a = m(q2 − 1) + n(q2 − q) for some m, n ∈ N0. Then, for such an a, the polynomial
1 Sn
Sm
0 6= 0.
An easy observation that will be used later is that, if 0 ≤ a < q3 − q2, then m ≤ q and n ≤ q − 1.
Next, we talk about a when q3 − q2 < a ≤ 2(q3 − q2).
0 is an invariant polynomial in Q of degree a, and as a < q3 −1, we can be sure that Sm
1 Sn
As will be proved in Lemma 6.3, Sq
6= 0. Then, for any a in the above range, if the coefficient
of ta is non-zero in f (t), then 2(q3 − q2) − a = m(q2 − 1) + n(q2 − q) for some m, n ∈ N0. Then,
Sq−m
is an invariant polynomial of degree (q − m)(q2 − 1) + (q − n − 1)(q2 − q) = a. It is well
1
defined as m ≤ q and n ≤ q − 1. Also, Sq−m
6= 0 as Sq−m
Sq−1−n
0
1Sq−1
6= 0.
0
0 = Sq
1Sq−1
0
Sq−1−n
0
Sq−1−n
0
1
Lemma 6.3. Using the same notation as in the previous proposition, Sq
6= 0.
× Sm
1 Sn
1Sq−1
0
1
13
Proof. For this proof, we need the explicit form of S0 and S1. S1 is easy to determine from the
definition of D2,1 whereas for S0, we use Remark 4.2. So, we have:
S0 = xq2−q
S1 = xq2−q
1
1
+ x(q−1)2
2 + · · · + xq2−q
xq−1
1
2 + x(q−1)2
xq−1
x2(q−1)
2
2
1
+ · · · + xq−1
1 xq2−q
2
.
Easy to see that S0 = xq2
1
−1
−1
−xq2
xq−1
1 −xq−1
2
2
. Then, the coefficient of xq3−q
and S1 = S0xq−1
in Sq
2 − xq2−1
1Sq−1
0
2
. Then, Sq
1Sq−1
0 = S2q−1
0
xq2−q
2
−
1
xq3−2q2+q
2
xq2−q
in S2q−1
2
in S2q−1
0
0
2
0 xq3−q
Sq−1
= the coefficient of xq3−q
= the coefficient of xq3−q
= the coefficient of xq3−q
1
1
1
xq3−2q2+q
2
xq3−3q2+2q
2
xq3−3q2+2q
2
= the coefficient of tq3−q in (cid:16) 1−tq2
= the coefficient of tq2+q in (cid:16) 1−tq+1
= −1.
1
−1
in (cid:16) xq2
1−tq−1 (cid:17)2q−1
1−t (cid:17)2q−1
Thus, as the coefficient is non-zero, Sq
1Sq−1
0
6= 0.
−1
−1
−xq2
xq−1
1 −xq−1
2
2
(cid:17)2q−1
(by substituting x1 = tx2)
We also believe that none of the invariant polynomials talked about in the above proposition
are scalar multiples of the invariant polynomials corresponding to the 'k = 1' term of the Hilbert
series. If that is true, then we have computed invariants corresponding to every term of the Hilbert
series and would not expect any other invariant polynomials to exist.
Conjecture 6.4. None of the polynomials described by us in Proposition 6.2 are scalar multiples
of the invariant polynomials corresponding to the 'k = 1' case.
Now, we make a brief comment in the case when n = k = m − 1 for an arbitrary n. Then, the
corresponding term of the Hilbert series is:
(cid:20)n + 1
n (cid:21)q,t
=
n−1
Yi=0
1 − tqn+1−qi
1 − tqn−qi
.
In the denominator, we see terms having degrees qn − qi for 0 ≤ i ≤ n − 1. This is exactly the
set of degrees of the Dickson invariant polynomials Dn,0, Dn,1, · · · , Dn,n−1 in Fq[x1, x2, · · · , xn]. On
computation, one sees that (cid:2)n+1
n (cid:3) is again a polynomial all of whose coefficients are 0 or 1. Thus,
inspired by our work in the case n = 2, we make the following conjecture.
Conjecture 6.5. For arbitrary n and m = n + 1, if we consider the k = n term of the conjectured
Hilbert series of QG, then one can find invariant polynomials corresponding to this term of the
series that are images of invariant polynomials in S.
Remark 6.6. The ideas developed above for the case n = k = m − 1 = 2 can be used to work
further on the case n = 2 for an arbitrary m. In such a case, the k = 2 term of the Hilbert series is
2(cid:3)q,t = (1−tqm −1)(1−tqm −q)
(cid:2)m
. Once again, we can talk about non-negative integer linear combinations
of q2 − 1 and q2 − q. However, complications start to arise because the coefficients of ta in this
series are not necessarily just 0 or 1.
(1−tq2−1)(1−tq2 −q)
14
7 The case q = 2
Now, we record some results about the structure of the invariant ring when q = 2. We assume that
n = 2 and consider the conjectured Hilbert series given as (taking m ≥ 3):
t2(2m−1) + t2m−2 1 − t2m−1
1 − t
+
(1 − t2m−1)(1 − t2m−2)
(1 − t3)(1 − t2)
.
The 3rd term in the above sum has been discussed comprehensively in the previous section.
Another conjecture that could be coupled with Conjecture 6.5 is that this term corresponds to the
images of invariant polynomials in S. Throughout this section, we refer to polynomials of the ring
SG as Dickson invariant polynomials.
Conjecture 7.1. For q = n = 2 and arbitrary m, the 'k = 3' term of the conjectured Hilbert series
corresponds to invariant polynomials of Q that are images of Dickson invariant polynomials in S.
However, in this section, we are more concerned about the 2nd term of the Hilbert series. The
+ · · · + t2m+1−4. From Proposition 4.1, we have that the
2nd term is equal to t2m−2 + t2m−1 + t2m
invariant polynomials corresponding to this are given by yk′ where 0 ≤ k′ ≤ 2m − 2 and:
yk′ = x2m−2
1
2 + x2m−3
xk′
1
xk′+1
2 + · · · + xk′+1
1
x2m−3
2
+ xk′
1 x2m−2
2
.
The degrees of y0 and y1 are 2m − 2 and 2m − 1 respectively. As these are less than 2m, the
polynomials are actually invariant in S itself. Hence, they are Dickson polynomials. We claim that
1 +x1x2 +x2
each y′
2.
Thus,
k is the image of a Dickson polynomial. To show this, we first note that D2,0 = x2
D2,0 × yk′ = (x2
1 + x1x2 + x2
2)
x2m−2+k′−i
1
xi
2
x2m+k′−i
1
x2m−1+k′−i
1
xi+1
2 +
x2m−2+k′−i
1
xi+2
2
=
=
=
=
2m−2
2m−2
Xi=k′
Xi=k′+2
2m−2
Xi=k′+1
Xi=k′+2
= yk′+2.
2m−2
2m−2
xi
2 +
Xi=k′
Xi=k′
Xi=k′+1
2 + x2m−1
xk′+1
xi
2 +
2m−3
1
2m−2
2m−4
Xi=k′
Xi=k′
x2m+k′−i
1
x2m−1+k′−i
1
xi+1
2 +
x2m−2+k′−i
1
xi+2
2
+ x2m−1
2m−3
1
xk′+1
2 + xk′+1
2m−3
1
x2m−1
2
+ xk′+1
1
x2m−1
2
2m−2
x2m+k−1′−i
1
xi+1
2 +
Xi=k′+1
x2m−1+k′−i
1
xi+1
2 +
Xi=k′+2
x2m+k′−i
1
xi
2
x2m+k′−i
1
xi
2
Thus, D2,0yk′ = yk′+2, and so, as y0 and y1 are images of Dickson invariant polynomials, we
can conclude the same for yk′ for all k′.
Also, the polynomial x2m+1−2
+x2m+1−3
1
1
in SG (as a corollary of Remark 4.2). Its image in Q is x2m−1
x2 +· · ·+x1x2m+1−3
x2m−1
2
1
2
+x2m+1−2
is an invariant polynomial
2
which is the invariant polynomial
15
corresponding to the first term of the Hilbert series. Hence, that too is an image of a Dickson
invariant polynomial. We conclude with the following theorem:
Theorem 7.2. Assuming Conjecture 7.1, all the polynomials in the invariant ring QG are images
of Dickson invariant polynomials, i.e. images of polynomials in SG, when the underlying field is
F2 and n = 2.
8 The parabolic conjecture: m = 2
In this section, we discuss a generalisation of Conjecture 1.2 to parabolic subgroups (see [3]). A
parabolic subgroup Pα of G specified by a composition α = (α1, α2, · · · , αl) of n, so that α :=
α1 + α2 + · · · + αl = n, is defined to be the subgroup consisting of block upper-triangular matrices
of the form:
∗
g2
.
.
.
0
· · ·
· · ·
.
.
.
· · ·
g1
0
.
.
.
0
∗
∗
.
.
.
gl
where each diagonal block gi is an αi × αi matrix. For such subgroups of G, we wish to study the
structure of QPα where Q = Fq[x1, x2, · · · , xn]/(xqm
1 , xqm
n ). Denoting partial sums of the
components of α by Ai := α1 + α2 + · · · + αi, we define:
2 , · · · , xqm
(cid:20)n
α(cid:21) :=
n−1
αi−1
Qj=1
Qj=0
l
Qi=1
(1 − tqn−qj
)
(1 − tqAi −qAi−1+j
.
)
For two compositions α and β both of length l, we define a partial order such that β ≤ α iff βi ≤ αi
for all i. With this definition, we are ready to state the conjecture for the Hilbert series for QPα as
mentioned in [3]. Define Bi = β1 + β2 + · · · + βi.
Conjecture 8.1. Fixing m, n and a composition α of n, we have
Hilb(QPα, t) = Xβ:β≤α
β≤m
te(m,α,β)(cid:20) m
β, m − β(cid:21)
(αi − βi)(qm − qBi).
where, e(m, α, β) :=
l
Pi=1
The n = 1 case of Conjectures 1.2 and 8.1 coincide, and so, we assume n ≥ 2 here onwards.
The m = 1 case of this conjecture has already been dealt with in [3]. In this section, we analyse
the m = 2 case.
Our aim is to show that for arbitrary n and α, we can find invariant polynomials corresponding
In the summation over β in the conjecture, the
to each term in the conjectured Hilbert series.
condition that β ≤ m = 2 implies that we have 4 cases:
16
• β = (0, 0, · · · , 0)
• β = er = (0, 0, · · · , 0, 1, 0, · · · , 0) where 1 is in the rth position, 1 ≤ r ≤ l
• β = er + es where 1 ≤ r, s ≤ l and r 6= s
• β = 2er where 1 ≤ r ≤ l (provided αr ≥ 2 as β ≤ α)
We'll deal with each case separately and show that invariant polynomials can be found corre-
sponding to each term in the summation. The idea will be to fiddle with the polynomials already
constructed in Section 3. Before we start, we state an easily proved counterpart of Lemma 1.3 for
parabolic subgroups.
Lemma 8.2. A polynomial is invariant under the action of Pα if it is invariant under diagonal
matrices, permutations of variables within a block, and the substitutions xi → xi + xj whenever
1 ≤ j < i ≤ n.
Here, by a permutation within a block, we mean any permutation of the variables xAi−1+1, xAi−1+2, . . . , xAi
for a fixed i between 1 and l.
Case 1: Suppose β = (0, 0, · · · , 0). Then, e(m, α, β) = α(qm − 1) = n(q2 − 1), and (cid:2) 2
te(m,α,β)(cid:2) m
β,m−β(cid:3) = tn(q2−1). Consider the polynomial
Yj=1
an :=
xq2−1
n
.
j
β,2(cid:3) = 1. So,
We already know that an is invariant under the action of G. In particular, it is invariant under the
action of Pα. As its degree is n(q2 − 1), it corresponds to the tn(q2−1) term of the Hilbert series.
Case 2: Suppose β = er for some r. Then, e(m, α, β) = Ar−1(q2 − 1) + (n − Ar−1 − 1)(q2 − q) and
−1
β,2(cid:3) = 1−tq2
(cid:2) 2
1−tq−1 = 1+tq−1+t2(q−1)+· · ·+tq2−q. So, te(m,α,β)(cid:2) m
tq−1 + t2(q−1) + · · · + tq2−q). For each r, consider the polynomials
β,m−β(cid:3) = tAr−1(q2−1)t(n−Ar−1−1)(q2−q)(1+
br,k :=
xq2−1
j
×
Ar−1
Yj=1
iAr−1+1+iAr−1+2+···+in=(n−Ar−1−1)q+k
0≤iAr−1+1,iAr−1+2,··· ,in≤q
X
(cid:18)
n
Yj=Ar−1+1
x
ij(q−1)
j
(cid:19)
where 0 ≤ k ≤ q. Let the 2 multiplicands in the above product be Fr,k and Gr,k respec-
tively. Then, degree of Fr,k is Ar−1(q2 − 1) and degree of Gr,k is {(n − Ar−1 − 1)q + k}(q − 1).
Thus, if it is invariant, the polynomial br,k corresponds to the tAr−1(q2−1)t{(n−Ar−1−1)q+k}(q−1) =
tAr−1(q2−1)t(n−Ar−1−1)(q2−q)tk(q−1) term of the series. Hence, if we let k vary from 0 to q, we can
account for all the terms in te(m,α,β)(cid:2) m
β,m−β(cid:3).
Now, we check the invariance of br,k. As, all exponents are multiples of q − 1, diagonal matri-
ces leave br,k invariant. Any permutation of the block i when 1 ≤ i ≤ r − 1 affects only Fr,k and
leaves it invariant, and r ≤ i ≤ n affects only Gr,k and leaves it invariant (both due to symmetry).
17
Now, we check the action of substitutions. For xi → xi + xj where j < i ≤ r − 1, the substitution
affects only Fr,k and invariance is trivial as powers greater than or equal to q2 become 0. Next, for
xi → xi + xj where r ≤ j < i, the substitution affects only Gr,k and we know from Section 3.2 that
this substitution leaves it invariant. Finally, consider the substitution xi → xi + xj where j < r ≤ i.
This again leaves br,k invariant because the exponent of xj is already q2 − 1 and any higher powers
would become zero. Thus, we conclude that br,k is an invariant polynomial.
Case 3: Suppose β = er + es where 1 ≤ r < s ≤ l. Then, e(m, α, β) = Ar−1(q2 − 1) + (As−1 −
Ar−1 − 1)(q2 − q) and (cid:2) 2
tAr−1(q2−1)t(As−1−Ar−1−1)(q2−q)(1 + tq−1 + t2(q−1) + · · · + tq2−q). For each such r and s, consider the
polynomials
1−tq−1 = 1 + tq−1 + t2(q−1) + · · · + tq2−q. So, te(m,α,β)(cid:2) m
β,m−β(cid:3) =
β,2(cid:3) = 1−tq2
−1
cr,s,k :=
xq2−1
j
×
Ar−1
Yj=1
iAr−1+1+iAr−1+2+···+iAs−1 =(As−1−Ar−1−1)q+k
0≤iAr−1+1,iAr−1+2,··· ,iAs−1 ≤q
X
(cid:18)
As−1
Yj=Ar−1+1
xij (q−1)
j
(cid:19)
where 0 ≤ k ≤ q. Let the 2 multiplicands in the above product be Sr,s,k and Tr,s,k respectively.
Then, degree of Sr,s,k is Ar−1(q2 − 1) and degree of Tr,s,k is {(As−1 − Ar−1 − 1)q + k}(q − 1). Thus,
if it is invariant, the polynomial cr,s,k corresponds to the tAr−1(q2−1)t{(As−1−Ar−1−1)q+k}(q−1) =
tAr−1(q2−1)t(As−1−Ar−1−1)(q2−q)tk(q−1) term of the series. Hence, if we let k vary from 0 to q, we can
account for all the terms in te(m,α,β)(cid:2) m
as in the previous case.
β,m−β(cid:3). The proof that cr,s,k is invariant is exactly the same
Case 4: Suppose β = 2er for some r such that αr ≥ 2. Then, e(m, α, β) = Ar−1(q2 − 1) and
(cid:2) 2
β,2(cid:3) = 1. So, te(m,α,β)(cid:2) m
β,m−β(cid:3) = tAr−1(q2−1). For each r, consider the polynomial
Ar−1
dr :=
xq2−1
j
.
Yj=1
We claim that each dr is invariant under Pα, and thus, corresponds to the tAr−1(q2−1) term of the
Hilbert series. dr is trivially invariant under diagonal matrices. Also, as we are taking product
for all variables corresponding to the first r − 1 blocks, dr is invariant under the permutation of
variables within a block. Finally, for any j, if we substitute xj → xj + xk where k < j, we end up
with dr again as exponents greater than or equal to q2 become zero. Thus, our claim is true.
The following example helps illustrate the above cases:
Example 8.3. Consider the case when q = 3, n = 6 and m = 2 and the composition α = (2, 1, 3)
of 6. In Case 1, we take β = (0, 0, 0). The corresponding term of the Hilbert series is t48 and the
associated invariant polynomial is a6 = x8
In Case 2, β = (1, 0, 0) corresponds to the term t30(1 + t2 + t4 + t6). The associated invariant
polynomials are the polynomials 4 polynomials that are invariant under the action of Gl6(F3) (i.e.,
the y6,k's for 0 ≤ k ≤ 3 in the notation of Section 3.2). Next, taking β = (0, 1, 0) the corresponding
term of the Hilbert series is t16t18(1 + t2 + t4 + t6). The associated invariant polynomials are ob-
tained by taking the product of the monomial x8
2 with polynomials in x3, x4, x5, and x6 that are
1x8
1x8
2x8
3x8
4x8
5x8
6.
18
3(cid:16)x6
6(cid:17).
invariant under the action of Gl4(F3) (i.e., the y4,k's for 0 ≤ k ≤ 4). Next, take β = (0, 0, 1). The
corresponding term in the series is t24t12(1 + t2 + t4 + t6). The associated invariant polynomials will
be products of the monomial x8
3 with polynomials in x4, x5, and x6 that are invariant under
the action of Gl3(F3) (i.e., the y3,k's for 0 ≤ k ≤ 4). For example,
6)x6
b3,1 = x8
6 + (x6
6 + (x6
5 + x4
5 + x4
5 + x2
1x8
2x8
1x8
2x8
4x6
5x2
4x4
4x6
5)x4
4x2
4x4
4x6
Now we move on to Case 3. β = (1, 1, 0) corresponds to the term t6(1 + t2 + t4 + t6). The
associated invariant polynomials, the c1,2,k's are the polynomials in x1 and x2 that are invariant
under the action of Gl2(F3) (i.e., the y2,k's for 0 ≤ k ≤ 3). Next, taking β = (1, 0, 1), the corre-
sponding term of the Hilbert series is t24(1 + t2 + t4 + t6) and the associated invariant polynomials
are the polynomials in x1, x2, x3 that are invariant under the action of Gl3(F3) (i.e., the y3,k's for
0 ≤ k ≤ 3). Finally, when β = (0, 1, 1), the term of the series is t16(1 + t2 + t4 + t6). The associated
invariant polynomials are x8
1x8
2, x8
1x8
2x2
3, x8
1x8
2x4
3 and x8
1x8
2x6
3.
Finally, we look at Case 4. When β = (2, 0, 0), the corresponding term of the Hilbert series is
1 and the associated polynomial is d1 = 1. For β = (0, 0, 2), the corresponding term is t24 and the
associated invariant polynomial is d3 = x8
3. This finishes our analysis of all the cases.
1x8
2x8
We conclude with the following conjecture:
Conjecture 8.4. Given m = 2 and arbitrary n and fixing a composition α of n, the polynomi-
als an, br,k's, cr,s,k's and dr's provide an additive basis for the invariant ring QPα, thus proving
Conjecture 8.1 when m = 2.
As we have already checked the invariance of the above polynomials, the proof of the above
conjecture hinges on showing that, up to scalar multiples, these are the only homogeneous invariant
polynomials in Q.
Remark 8.5. Among the polynomials defined above, the polynomials an and b1,k for 0 ≤ k ≤ q are
the ones that are invariant under the action of the entire group G. In the notation of Section 3.2,
an is equal to zn and b1,k is equal to yn,k for all k.
9 Acknowledgements
I would like to thank the S.N. Bose Scholars Program, specifically Dr. Aseem Z. Ansari, Depart-
ment of Biochemistry, University of Wisconsin-Madison, and the entire team of Winstep Forward,
IUSSTF and SERB for having provided me with this peerless opportunity of getting a hands-on re-
search experience and for facilitating my stay in the United States. I would also like to acknowledge
the crucial role played by my mentor Dr. Steven V. Sam, Department of Mathematics, University
of Wisconsin-Madison, who provided me a better insight into the field as well as the problem that
I was working on, and elaborate discussions with whom helped me systematize my working style.
Finally, I wish to thank my fellow Bose-Khorana scholars at UW Madison, spending time with
whom was an immense pleasure and a source of great inspiration.
19
References
[1] R.M.W. Wood, Problems in the Steenrod Algebra. Bull. London Math. Soc. (1998) 30 (5):
449-517.
[2] L.E. Dickson, A fundamental system of invariants of the general modular linear group with a
solution of the form problem. Trans. Amer. Math. Soc. 12 (1911), 75–98.
[3] J. Lewis, V. Reiner, D. Stanton, Invariants of Gln(Fq) in polynomials mod Frobenius powers.
Proc. Roy. Soc. Edinburgh A to appear., arXiv:1403.6521v3
[4] Larry Smith, Polynomial invariants of finite groups: A survey of recent developments. Bull.
Amer. Math. Soc. (N.S.) 34 (1997), no. 3, 211–250.
[5] Tianxin Cai, The Book of Numbers, World Scientific Publishing Co. Pte. Ltd.
[6] Clarence Wilkerson, A primer on the Dickson invariants. Contemporary Math. 19 (1983), 421–
434.
20
|
1606.01982 | 1 | 1606 | 2016-06-07T00:31:14 | Self-dual nonsymmetric operads with two binary operations | [
"math.RA",
"math.CT"
] | We consider nonsymmetric operads with two binary operations satisfying relations in arity 3; hence these operads are quadratic, and so we can investigate Koszul duality. We first consider operations which are nonassociative (not necessarily associative) and then specialize to the associative case. We obtain a complete classification of self-dual quadratic nonsymmetric operads with two (associative or nonassociative) binary operations. These operads generalize associativity for one operation to the setting of two operations. | math.RA | math |
SELF-DUAL NONSYMMETRIC OPERADS
WITH TWO BINARY OPERATIONS
MURRAY BREMNER AND JUANA S ´ANCHEZ-ORTEGA
Abstract. We consider nonsymmetric operads with two binary operations
satisfying relations in arity 3; hence these operads are quadratic, and so we
can investigate Koszul duality. We first consider operations which are nonasso-
ciative (not necessarily associative) and then specialize to the associative case.
We obtain a complete classification of self-dual quadratic nonsymmetric oper-
ads with two (associative or nonassociative) binary operations. These operads
generalize associativity for one operation to the setting of two operations.
1. Introduction
1.1. Associativity for one operation. Let O be the free nonsymmetric operad
generated by one binary operation • over the field F:
dimF O(w) =
O(w),
1
w+1(cid:18)2w
w(cid:19).
O = Mw≥0
A basis of O(w) consists of all complete rooted binary trees with w internal nodes,
and hence w+1 leaves. (Note that we are indexing by weight, not arity.) We inter-
pret these trees as association types (placements of parentheses) for the composition
of w+1 arguments: the internal nodes represent the operation • and the leaves rep-
resent the arguments x1, . . . , xw+1. Since O is a nonsymmetric operad, we do not
need to specify the arguments: the i-th argument is always xi for 1 ≤ i ≤ w+1. We
call this set of trees the monomial basis of O(w); its size is the Catalan number.
Any (nonzero) element of O(w) is called a relation of weight w. To illustrate, we
list the monomial bases for 0 ≤ w ≤ 3, using dash to represent an argument:
w = 1 : − • −
w = 2 :
(− • −) • −, − • (− • −)
w = 0 : −
w = 3 :
((− • −) • −) • −,
(− • (− • −)) • −,
− • ((− • −) • −), − • (− • (− • −)).
(− • −) • (− • −),
Given basis monomials m1, m2 of weights w1, w2 we define the composition m1◦im2
for 1 ≤ i ≤ w1+1 to be the result of substituting m2 for the i-th argument of m1;
in terms of trees, we are identifying the root of m2 with the i-th leaf of m1.
A quadratic relation (w = 2) has the form am1 +bm2 ≡ 0 with a, b ∈ F (not both
0) where m1 = (− • −) • −, m2 = − • (− • −). We define a symmetric bilinear
form h , i on O(2) by hm1, m1i = 1, hm2, m2i = −1, hm1, m2i = 0; see Loday [5,
Proposition B.3]. With respect to this form, the orthogonal complement of the
subspace spanned by am1 + bm2 has basis bm1 + am2. The operad is (Koszul)
2010 Mathematics Subject Classification. Primary 18D50. Secondary 13P10, 15A54, 16S37,
17A30, 68W30.
Key words and phrases. Algebraic operads, Koszul duality, Grobner bases, computer algebra.
1
2
MURRAY BREMNER AND JUANA S ´ANCHEZ-ORTEGA
b
self-dual (see §2.2) if and only if these two subspaces coincide; equivalently, the
matrix R =(cid:20)a
b a(cid:21) has rank 1. This holds if and only if det R = a2 − b2 = 0 (and
a, b are not both 0). Up to scalar multiples, the only solutions are (a, b) = (1, 1)
and (a, b) = (1,−1), which define the anti-associative operad m1 + m2 ≡ 0 and the
associative operad m1 − m2 ≡ 0.
By a result of Osborn [8, Corollary 2] we know that a homogeneous polynomial
identity is satisfied by a unital algebra if and only if the sum of its coefficients is 0.
This condition distinguishes associativity from anti-associativity: only the former
is satisfied by unital algebras. We have therefore classified self-dual nonsymmetric
operads with one binary operation, and those which define unital algebras.
Our goal in this paper is to extend this classification to operads with two binary
operations. This allows us to determine generalizations of associativity for these
operads, in the sense that the relations define a self-dual nonsymmetric operad, and
in every relation the sum of coefficients is 0. We use an approach based on computer
algebra; our main tools are linear algebra over polynomial rings and Grobner bases
for polynomial ideals [1, Chs. 7-10]. Throughout, we assume that all vector spaces
are over the field F which is algebraically closed of characteristic 0.
1.2. Operads with two binary operations. To motivate our study of structures
with two binary operations, we recall the most important examples. In his encyclo-
pedia of algebras, Zinbiel (Loday) [9] mentions a number of algebraic operads with
two binary operations satisfying relations which are quadratic (each monomial has
two operations and three arguments) and nonsymmetric (each monomial has the
identity permutation of the arguments). Some are (Koszul) self-dual, but most are
not. In most cases, both operations are associative; we denote them by ⊢ and ⊣.
Definition 1.1. Two-associative algebras satisfy only associativity:
(a ⊢ b) ⊢ c ≡ a ⊢ (b ⊢ c),
(a ⊣ b) ⊣ c ≡ a ⊣ (b ⊣ c).
Dual two-associative algebras satisfy associativity and these relations:
(a ⊢ b) ⊣ c ≡ 0,
(a ⊣ b) ⊢ c ≡ 0,
a ⊢ (b ⊣ c) ≡ 0,
a ⊣ (b ⊢ c) ≡ 0.
Duplicial algebras satisfy associativity and inner associativity:
Dual duplicial algebras satisfy associativity, inner associativity, and:
(a ⊢ b) ⊣ c ≡ a ⊢ (b ⊣ c).
(a ⊣ b) ⊢ c ≡ 0,
a ⊣ (b ⊢ c) ≡ 0.
Completely associative algebras satisfy the following relations which include
associativity and define a self-dual operad:
(a ∗ b) ∗′ c ≡ a ∗ (b ∗′ c),
∗,∗′ ∈ {⊢,⊣}.
Two-compatible algebras satisfy associativity and the relation which states that
any linear combination of the operations is associative:
(a ⊢ b) ⊣ c + (a ⊣ b) ⊢ c ≡ a ⊢ (b ⊣ c) + a ⊣ (b ⊢ c).
Dual two-compatible algebras satisfy the relations of completely associative and
two-compatible algebras. Diassociative algebras (or associative dialgebras) satisfy
associativity, inner associativity, and the left and right bar relations:
a ⊣ (b ⊢ c) ≡ a ⊣ (b ⊣ c).
(a ⊢ b) ⊢ c ≡ (a ⊣ b) ⊢ c,
SELF-DUAL NONSYMMETRIC OPERADS
3
The dual operad defines dendriform algebras which have nonassociative operations
satisfying inner associativity and these relations:
(a ⊣ b) ⊣ c ≡ a ⊣ (b ⊣ c) + a ⊣ (b ⊢ c),
a ⊢ (b ⊢ c) ≡ (a ⊢ b) ⊢ c + (a ⊣ b) ⊢ c.
2. Binary operations and Koszul duality
2.1. Binary operations. We write O for the free nonsymmetric operad generated
by two (nonassociative) binary operations ⊢ and ⊣ which form a basis of the space
O(1) of all binary operations. For w ≥ 0, a basis of O(w) consists of all complete
rooted binary trees with w internal nodes each labelled by an operation.
Lemma 2.1. We have
dimO(w) =
2w
w+1(cid:18)2w
w(cid:19)
Proof. The factor 2w represents the choices of operation symbols.
Example 2.2. We have dim O(0) = 1 with basis { − } (the argument symbol),
and dim O(1) = 2 with ordered basis { − ⊢ −, − ⊣ − }. Every quadratic relation
is an element of O(2) which has dimension 8 and the ordered basis in Table 1.
(cid:3)
(− ⊢ −) ⊢ −,
− ⊢ (− ⊢ −),
(− ⊢ −) ⊣ −,
− ⊢ (− ⊣ −),
(− ⊣ −) ⊢ −,
− ⊣ (− ⊢ −),
(− ⊣ −) ⊣ −,
− ⊣ (− ⊣ −).
Table 1. Ordered basis of quadratic space O(2) for two binary operations
Definition 2.3. A (nonzero) element ρ ∈ O(2) is called a quadratic relation,
and a subspace R ⊆ O(2) is a space of quadratic relations. The operad ideal
(R) generated by a subspace R ⊆ O(w) is the smallest subspace of O which contains
R and is closed under composition by arbitrary elements of O.
The elements of a space R of quadratic relations are satisfied by the quotient
operad Q = O/(R) where (R) is the operad ideal generated by R. If dim R = r
then R is the row space of a unique r × dimO(2) matrix denoted [R] which has full
rank and is in row canonical form (RCF); the columns are labelled by the ordered
basis in Table 1. Conversely, the row space of any matrix with 8 columns can be
regarded as a space of quadratic relations.
Definition 2.4. The matrix [R] is the relation matrix of the quadratic operad
Q = O/(R), and its rank r is the relation rank of Q.
2.2. Koszul duality. Loday [5, Proposition B.3], see also [6, Chapter 7], has shown
that Koszul duality for binary operations can be defined in elementary terms, using
a nondegenerate inner product h−,−i on O(2). For n-ary operations, see [7, §2].
Definition 2.5. For all •1,•2 ∈ {⊢,⊣} we define the symmetric bilinear form h−,−i
on basis monomials in Table 1 as follows:
h (− •1 −) •2 −, − •1 (− •2 −) i = 0,
h (− •1 −) •2 −, (− •1 −) •2 − i = 1,
h − •1 (− •2 −), − •1 (− •2 −) i = −1.
Notation 2.6. For any subspace R ⊆ O(2) we write R± for its orthogonal comple-
ment with respect to the symmetric bilinear form h−,−i of Definition 2.5. We write
R⊥ for its orthogonal complement with respect to the Euclidean inner product for
which the monomials in Table 1 are an orthonormal basis.
4
MURRAY BREMNER AND JUANA S ´ANCHEZ-ORTEGA
Definition 2.7. If Q = O/(R) then its Koszul dual is Q! = O/(R±). We say
that Q is self-dual if Q = Q! (equivalently R± = R).
Lemma 2.8. We have dim R + dim R± = 8. If Q = Q! then dim R = 4.
Proof. h−,−i is nondegenerate and if Q = Q! then dim R = dim R±.
Remark 2.9. The relation matrices [R] have entries in the polynomial ring Φ =
F[x1, . . . , xp], so we regard operads as modules over Φ, not vector spaces over F.
(cid:3)
Loday has shown that computing R± can be reduced to computing the Euclidean
orthogonal complement of a modified space; see Table 2.
Input : The relation matrix [R] for the quadratic operad Q = O/(R) where
R ⊆ O(2) and the entries of [R] belong to Φ.
Output : The relation matrix [R±] for the Koszul dual Q! = O/(R±).
Algorithm:
(1) Since the last 4 monomials in Table 1 have association type 2, we multiply
columns 5 -- 8 of [R] by −1 to obtain the matrix [R′].
(2) Since any leading 1 of [R] in position (i, j) for j ≥ 5 becomes −1 in [R′],
we multiply any such rows of [R′] by −1 to obtain [R′′].
(3) To find a basis for R± = (R′′)⊥, we solve the linear system [R′′]X = 0.
If dim R = r then there are 8−r free variables; we set them equal to the 8−r
standard basis vectors in F8−r and solve for the leading variables.
(4) Construct the (8−r) × 8 matrix [R±] whose rows are the basis vectors for
(R′′)⊥ computed in step (3).
To find conditions for self-duality when r = 4 we do two more steps:
(5) Stack [R] onto [R±], and use the leading 1s of the upper block [R] to reduce
the rows of the lower block [R±], obtaining the matrix [T ]:
(cid:20) R
R± (cid:21) [R] reduces [R±] to [T ]
−−−−−−−−−−−−−−−→(cid:20) R
T (cid:21)
(6) The operad is self-dual if and only if R = R±; that is, [T ] = 0. Since the
entries of [T ] are elements of Φ, we find the zero set of the ideal generated by
[T ]; these values of the parameters x1, . . . , xp define self-dual operads.
Table 2. Loday's algorithm for the Koszul dual relation matrix
3. Self-duality for two nonassociative operations
By nonassociative in this section we mean not necessarily associative: we do not
explicitly assume associativity, but we will in §4.
3.1. Computational methods. In a matrix in RCF, the entries above, below, and
to the left of each leading 1 are 0, and the remaining entries are free parameters.
For an r × n matrix, there are (cid:0)n
In particular, for r = 4 and n = 8 we have 70 cases.
Definition 3.1. We define the lex order on subsets by {j1, . . . , jr} ≺ {j′
if and only if jk < j′
r(cid:1) choices of columns j1 < ··· < jr for leading 1s.
1, . . . , j′
r}
k where k is the least index for which jk 6= j′
k.
SELF-DUAL NONSYMMETRIC OPERADS
5
Lemma 3.2. The number of parameters as a function of {j1, . . . , jr} is:
r
p = p(j1, . . . , jr) =
Xi=1(cid:2)(n − ji) − (r − i)(cid:3).
Proof. Add the number of entries to the right of each leading 1, and subtract the
number of entries which belong to the column of another leading 1.
(cid:3)
Lemma 3.3. For the 35 subsets {j1, . . . , j4} not containing 1 the matrices [R] do
not define self-dual operads for any values of the parameters.
Proof. Starting with [R], we compute [R′], [R′′], [R±], [T ]. Since column 1 of [R] is
0, column 1 of [R±] contains a leading 1. This leading 1 remains unchanged when
we compute [T ]. But if [T ] contains 1 then its entries generate the unit ideal Φ. (cid:3)
Example 3.4. For {j1, . . . , j4} = {2, 3, 4, 5} we obtain:
[R] =
[R′′] =
0 1 0 0 0 A B C
0 0 1 0 0 D E F
0 0 0 1 0 G H I
0 0 0 0 1 J K L
0 1 0 0 0 −A −B −C
0 0 1 0 0 −D −E −F
0 0 0 1 0 −G −H −I
J K L
0 0 0 0 1
[R′] =
[R±] =
The resulting matrix [T ] is
0 −A −B −C
0 1 0 0
0 −D −E −F
0 0 1 0
0 −G −H −I
0 0 0 1
0 0 0 0 −1 −J −K −L
1 0 0 0
0 0 0 0
0 A D G −J 1 0 0
0 B E H −K 0 1 0
0 C F I −L 0 0 1
0
1 0 0 0 0
0 0 0 0 0 J 2−G2−D2−A2+1
JK−GH−DE−AB JL−GI−DF−AC
0 0 0 0 0 JK−GH−DE−AB K 2−H 2−E2−B2+1 KL−HI−EF−BC
0 0 0 0 0 JL−GI−DF−AC KL−HI−EF−BC L2−I 2−F 2−C2+1
0
0
Since [T ] contains 1 as an entry, no values of the parameters give [T ] = 0. Note that
the lower right 3×3 block is I−M where M = (vi·vj) for (a1, . . . , a4)·(b1, . . . , b4) =
a1b1 + a2b2 + a3b3 − a4b4 and v1, v2, v3 are the last 3 columns of [R].
3.2. Cases 1 to 35. We now consider the subsets {j1, . . . , j4} for which j1 = 1.
Lemma 3.5. Of the 35 subsets {1, j2, j3, j4} there are 21 for which the ideal gener-
ated by the entries of [T ] equals Φ: in lex order, cases 5, 9, 12 -- 15, 19, 22 -- 35. For
these cases, [R] does not define a self-dual operad for any values of the parameters.
Proof. We give details for case 5; the computations in the other cases are similar.
With leading 1s in columns 1, 2, 3, 8 we obtain:
(cid:20) R
R± (cid:21) =
I
0
1
0
0
0
0
1
0
0
0 A B C D 0
0 E F G H 0
1
J K L 0
0
1
0
−A −E −I
B
J
C
D
1
F
0
G K 0
L 0
H
0
0
0
1
0
0
1
0
0
1
0
0
0
0
0
0
0
0
6
MURRAY BREMNER AND JUANA S ´ANCHEZ-ORTEGA
Reducing [R±] using [R] produces the matrix [T ] whose nonzero columns are:
IJ+EF +AB
I 2+E2+A2+1
IL+EH+AD
−IJ−EF−AB −J 2−F 2−B2+1 −JK−F G−BC −JL−F H−BD
−IK−EG−AC −JK−F G−BC −K 2−G2−C2+1 −KL−GH−CD
−IL−EH−AD −JL−F H−BD −KL−GH−CD −L2−H 2−D2+1
IK+EG+AC
[T ] = 0 if and only if there exist 4 vectors in F3, namely the columns of the 3 × 4
block of [R] containing the parameters, satisfying these equations with respect to the
Euclidean inner product: v1·v1 = −1, v2·v2 = v3·v3 = v4·v4 = 1, vi·vj = 0 (i 6= j).
This says that there exist 4 orthogonal nonzero vectors in F3; contradiction.
(cid:3)
We can also prove Lemma 3.5 using computer algebra; see Table 3.
Input: A monomial order ≺ on Φ = F[x1, . . . , xp], together with a subset
G = {f1 ≺ ··· ≺ fn} ⊂ Φ generating the ideal I ⊆ Φ.
Output: The Grobner basis of I with respect to ≺.
Algorithm:
(1) Set G0 ← ∅ and G1 ← G.
(2) Set k ← 1.
(3) While Gk−1 6= Gk do:
(a) Self-reduce Gk: For each fi ∈ Gk do:
(b) Compute the set of S-polynomials:
ements f1 ≺ ··· ≺ fi−1.
monicform(N (fi)).
• Compute the normal form N (fi) with respect to the previous el-
• If N (fi) = 0 then remove fi from Gk, otherwise replace fi by
• Sort Gk with respect to ≺.
• Set H ← ∅.
• For all 1 ≤ i < j ≤ Gk compute hij = S(fi, fj) and its normal
form N (hij) with respect to Gk; if N (hij ) 6= 0 then set H ←
H ∪ {monicform(N (hij ))}.
(c) Set Gk+1 ← Gk ∪ H. Sort Gk+1 with respect to ≺.
(d) Set k ← k + 1.
Table 3. Algorithm to compute a Grobner basis for a polynomial ideal
Remark 3.6. We apply the algorithm of Table 3 to the ideal generated by the
entries of the matrix [T ] from the proof of Lemma 3.5:
k = 1: The original set G1 of 10 generators is already self-reduced; it produces 24
k = 3: The set G3 has 262 elements but self-reduction eliminates 114. The remain-
k = 2: The set G2 has 34 elements but self-reduction eliminates 4. The remaining
S-polynomials with N (h) 6= 0.
30 generators produce 232 S-polynomials with N (h) 6= 0.
ing 148 generators produce 6916 S-polynomials h with N (h) 6= 0.
remaining 444 generators produce 92 S-polynomials h with N (h) 6= 0.
ing 13 generators are the 12 parameters A, . . . , L together with 1.
k = 4: The set G4 has 7064 elements but self-reduction eliminates 6620. The
k = 5: The set G5 has 536 elements but self-reduction eliminates 523. The remain-
SELF-DUAL NONSYMMETRIC OPERADS
7
k = 6: The set G6 has 13 elements but self-reduction eliminates 12 and leaves {1}.
The algorithm terminates with the Grobner basis {1}.
We did these calculations in Maple with the graded reverse lex order (A ≺ ··· ≺ L).
Definition 3.7. Let S be the set of 14 cases corresponding to subsets {1, j2, j3, j4}
for which the entries of [T ] generate a proper ideal in Φ:
S = { 1, 2, 3, 4, 6, 7, 8, 10, 11, 16, 17, 18, 20, 21 }.
1 0 0 0 W1 X1 Y1 Z1
0 1 0 0 W2 X2 Y2 Z2
0 0 1 0 W3 X3 Y3 Z3
0 0 0 1 W4 X4 Y4 Z4
1 0 0 W1 X1 Y1 0 Z1
0 1 0 W2 X2 Y2 0 Z2
0 0 1 W3 X3 Y3 0 Z3
0 0 1 Z4
0 0 0 0
By Lemma 3.5, these cases have self-dual operads defined by the parameter values
in the zero set of the ideal generated by [T ]. The corresponding relation matrices
[R] are displayed in Table 4. We write P for the 4× 4 parameter matrix obtained
by deleting columns 1, j2, j3, j4 from [R]. In case 1, P = [W, X, Y, Z] where W =
(W1, W2, W3, W4)t, etc. In the other cases, we obtain P by setting some entries to
0 in [W, X, Y, Z]; these entries are a subset of {W2, W3, W4, X3, X4, Y4}.
1
4
8
16
20
2
6
10
17
21
1 0 W1 0 0 X1 Y1 Z1
0 1 W2 0 0 X2 Y2 Z2
0 0 0 1 0 X3 Y3 Z3
0 0 0 0 1 X4 Y4 Z4
1 W1 0 0 X1 0 Y1 Z1
0 0 1 0 X2 0 Y2 Z2
0 0 0 1 X3 0 Y3 Z3
0 0 0 0 0 1 Y4 Z4
1 0 0 W1 X1 0 Y1 Z1
0 1 0 W2 X2 0 Y2 Z2
0 0 1 W3 X3 0 Y3 Z3
0 1 Y4 Z4
0 0 0 0
1 0 W1 0 X1 0 Y1 Z1
0 1 W2 0 X2 0 Y2 Z2
0 0 0 1 X3 0 Y3 Z3
0 0 0 0 0 1 Y4 Z4
1 0 W1 X1 0 Y1 0 Z1
0 1 W2 X2 0 Y2 0 Z2
0 0 0
0 1 Y3 0 Z3
0 0 0 1 Z4
0 0 0
1 W1 0 0 X1 Y1 0 Z1
0 0 1 0 X2 Y2 0 Z2
0 0 0 1 X3 Y3 0 Z3
0 0 0 0 0 0 1 Z4
1 W1 0 X1 0 Y1 0 Z1
0 0 1 X2 0 Y2 0 Z2
0 0 0 0 1 Y3 0 Z3
0 0 0 0 0 0 1 Z4
1 0 W1 0 X1 Y1 0 Z1
0 1 W2 0 X2 Y2 0 Z2
0 0 0 1 X3 Y3 0 Z3
0 0 0 0 0 0 1 Z4
1 W1 0 0 0 X1 Y1 Z1
0 0 1 0 0 X2 Y2 Z2
0 0 0 1 0 X3 Y3 Z3
0 0 0 0 1 X4 Y4 Z4
1 W1 0 X1 0 0 Y1 Z1
0 0 1 X2 0 0 Y2 Z2
0 0 0 0 1 0 Y3 Z3
0 0 0 0 0 1 Y4 Z4
3
7
11
18
1 0 0 W1 0 X1 Y1 Z1
0 1 0 W2 0 X2 Y2 Z2
0 0 1 W3 0 X3 Y3 Z3
0 0 0 0 1 X4 Y4 Z4
1 0 W1 X1 0 0 Y1 Z1
0 1 W2 X2 0 0 Y2 Z2
0 0 0
0 1 0 Y3 Z3
0 0 1 Y4 Z4
0 0 0
Table 4. The 14 relation matrices [R] defining self-dual operads
Remark 3.8. As the parameters range over F, each matrix in Table 4 defines a
Schubert cell in the Grassmannian GF(4, 8) of 4-dimensional subspaces of F8. The
nonzero entries of P (rotated 90◦ counter-clockwise) form a Young diagram for a
partition of the number of parameters. For example, case 1 gives 16 = 4+4+4+4,
and case 21 gives 10 = 4+3+2+1. For more information, see Fulton [2], Hiller [3].
Lemma 3.9. Assume that F = R. Let D and E be n × n diagonal matrices with
nonzero entries ±1. Every solution of AtEA = D has the form A = √DC√E for
some orthogonal matrix C where bar denotes complex conjugate.
8
MURRAY BREMNER AND JUANA S ´ANCHEZ-ORTEGA
Proof. If D 6= I or E 6= I then we extend scalars to C so that we can form √D
and √E. Let S be any real symmetric matrix with the same signature as E. By
Sylvester's Law of Inertia there is an orthogonal matrix C for which C tSC = E.
Thus S has the square root √S = C√EC t and hence √S
= √S. It follows that
B = √SC = C√E is a solution of BtB = E and every solution can be obtained
this way for some orthogonal matrix C. Clearly, B satisfies BtB = E if and only if
A = (√D)−1B satisfies AtDA = E. Finally, note that (√D)−1 = √D.
(cid:3)
t
Remark 3.10. If A = (aij) is a solution of AtEA = D then the columns U1, . . . , Un
of A are basis of Rn such that hUi, Uii = dii and hUi, Uji = 0 (i 6= j) where the
diagonal entries of E are the signature: h(v1, . . . , vn), (w1, . . . , wn)i =Pn
i=1 eiiviwi.
Theorem 3.11. Assume that F = R. For every case in S, the parameter values
defining self-dual operads are the solutions of the equation P tDP = E, where P is
the parameter matrix and the diagonal matrices D, E are as follows:
• Case 1: D = E = I4.
• Cases 2, 3, 4, 6, 8, 16, 17, 18: D = diag(1, 1, 1,−1), E = diag(−1, 1, 1, 1).
• Cases 10, 11, 20, 21: D = diag(1, 1,−1,−1), E = diag(−1,−1, 1, 1).
Thus P = √DC√E where C is orthogonal and has zeros in the same entries as P .
Proof. We verify the claims case-by-case.
• Case 1 : This case can be solved in terms of 4-dimensional Euclidean geometry.
The relation matrix is [ R ] = [ I4 P ] where P = [W, X, Y, Z]. Clearly [ R′ ] =
[ R′′ ] = [ I4 −P ], and hence [ R± ] = [ P t I4 ]. We obtain
−−−−−−−−−−−→" I4 P
I4 # [R] reduces [R±]
R± (cid:21) =" I4 P
(cid:20) R
O T ′ # =(cid:20) R
T (cid:21)
P t
In this and the remaining cases we simplify T by making its diagonal entries monic:
we divide row i by the leading coefficient of the i-th diagonal entry for 1 ≤ i ≤ 4.
With these sign changes, T ′ becomes T ′′ where
W · W − 1 W · X
X · X − 1
X · Y
X · Z
W · X
W · Y
W · Z
W · Y
X · Y
Y · Y − 1
Y · Z
W · Z
X · Z
Y · Z
Z · Z − 1
T ′′ = −T ′ = P tP − I =
Hence, in order for the matrix P of parameter values to belong to the zero set of
the ideal generated by T ′′, it is necessary and sufficient that its columns W, X, Y, Z
form an orthonormal basis of R4.
The Grobner basis for the ideal generated by T ′′ has 141 elements, the greatest
of which in lex order is this polynomial of degree 5 with 14 terms:
W4X4Y3Z1Z2 − W3X4Y4Z1Z2 − W4X4Y1Z2Z3 + W1X4Y4Z2Z3 − W4X2Y3Z1Z4
+ W3X2Y4Z1Z4 + W3X4Y1Z2Z4 − W1X4Y3Z2Z4 + W4X2Y1Z3Z4
− W1X2Y4Z3Z4 − W3X2Y1Z 2
It is easier to find the zero set from the generators than the Grobner basis.
4 + W3X2Y1 − W1X2Y3.
4 + W1X2Y3Z 2
SELF-DUAL NONSYMMETRIC OPERADS
9
• Case 2 : We define the symmetric bilinear form hU, V i to have signature equal to
the diagonal entries of D. We obtain
T ′′ =
hW, Xi
hX, Xi−1
hX, Y i
hX, Zi
hW, Zi
hX, Zi
hY, Zi
hZ, Zi−1
hW, Y i
hX, Y i
hY, Y i−1
hY, Zi
W = [W1, W2, W3, 0]t
hW, Wi+1
hW, Xi
hW, Y i
hW, Zi
= P tDP − E.
Hence P = diag(1, 1, 1,∓i) C diag(±i, 1, 1, 1) for some orthogonal matrix C. Since
W4 = 0 we require that P (and hence C) has 0 in the lower left corner.
The Grobner basis for the ideal generated by the entries of T ′′ has 112 elements,
the greatest of which in lex order is this polynomial of degree 7 with 36 terms:
4 Z 2
3
3 − W2X1Y3Y 2
4 Z3Z4 + W1X3Y4Z2Z 2
4 Z 2
4 Z2Z4 − 2W3X1Y2Y3Y4Z3Z4
3 Z4
4 − W1X2Y3Y 2
4 Z 2
4
4 Z2Z3 + W3X1Y2Y 2
3 Y4Z2Z4 + W1X3Y 3
4 Z2Z3 − W1X4Y 3
4 Z 2
3 − W1X3Y 2
3 Y4Z3Z4 + W1X2Y 3
3 Y4Z3Z4 + W1X2Y 2
3 Z 2
3 Z4 + W3X1Y2Y 2
4 − W2X1Y 3
4 + W3X1Y2Y 2
W1X3Y3Y 2
− W1X2Y3Y 2
+ 2W2X1Y 2
− W1X2Y4Z 3
− W1X3Y3Z2Z3Z 2
+ W1X4Y2Z3Z 3
+ W2X1Y3Y 2
− W1X3Y2Z 2
+ W1X2Y4Z3Z4 − W1X3Y2Z 2
4 − 2W1X4Y4Z2Z3Z 2
4 + W1X3Y3Z2Z3 − W1X4Y4Z2Z3 + W3X1Y2Z 2
3
3 − W2X1Y3Z 2
3 + W1X3Y4Z2Z4 + W1X4Y2Z3Z4
3 Z 2
3 − W2X1Y 3
4 − W1X3Y2Z 4
4 + W2X1Y3Z 2
4 + W1X2Y3Z 2
4 + W2X1Y3.
3 Z 2
3
4 + 2W1X3Y4Z2Z 3
4
• Case 3 : Define D, E as in Case 2. Then T ′′ has the same form as Case 2 but
W = [W1, W2, W3, 0]t and X = [X1, X2, X3, 0]t. The Grobner basis for the ideal
generated by T ′′ has 63 elements; the greatest is
W3X3Y1Y2 − W2X3Y1Y3 − W3X1Y2Y3 + W2X1Y 2
3 + W2X1Z 2
4 .
• Case 4 : Define D, E as in Case 2. Then T ′′ has the same form as Case 2 but
W = [W1, W2, W3, 0]t, X = [X1, X2, X3, 0]t, Y = [Y1, Y2, Y3, 0]t. The Grobner basis
for the ideal generated by T ′′ has 31 elements; the greatest is
W3X3Y1Y2 − W2X3Y1Y3 − W3X1Y2Y3 + W2X1Y 2
3 − W2X1.
• Case 6 : Define D, E as in Case 2. Then T ′′ has the same form as Case 2 but
W = [W1, W2, 0, 0]t. The Grobner basis for the ideal generated by T ′′ has 72
elements; the greatest is
3 − X1Y2Y4Z 2
4 + X1Z2Z 3
3 + X4Z1Z2Z 2
4 − X4Y1Y2 + X1Y2Y4 + X1Z2Z4.
X4Y1Y2Z 2
− X4Z1Z2Z 2
3 Z4 − X4Y1Y2Z 2
3 − X1Z2Z 2
4 + X1Y2Y4Z 2
4
• Case 7 : Define D, E as in Case 2. Then T ′′ has the same form as Case 2 but W =
[W1, W2, 0, 0]t and X = [X1, X2, X3, 0]t. The Grobner basis for the ideal generated
by T ′′ has 50 elements; the greatest is X1X2Y 2
• Case 8 : Define D, E as in Case 2. Then T ′′ has the same form as Case 2 but W =
[W1, W2, 0, 0]t, X = [X1, X2, X3, 0]t, Y = [Y1, Y2, Y3, 0]t. The Grobner basis for the
ideal generated by T ′′ has 50 elements; the greatest is X1X2Y 2
3 − X1X2.
3 + X1X2Z 2
4 .
3 + Y1Y2Y 2
3 + Y1Y2Y 2
3 + Y1Y2Z 2
10
MURRAY BREMNER AND JUANA S ´ANCHEZ-ORTEGA
• Case 10 : We now have D = diag(1, 1,−1,−1) and E = diag(−1,−1, 1, 1). We
define hU, V i to have signature equal to the diagonal entries of D. We obtain
T ′′ =
hW, Xi
hX, Xi+1
hX, Y i
hX, Zi
hW, Zi
hX, Zi
hY, Zi
hZ, Zi−1
hW, Y i
hX, Y i
hY, Y i−1
hY, Zi
W = [W1, W2, 0, 0]t,
X = [X1, X2, 0, 0]t
hW, Wi+1
hW, Xi
hW, Y i
hW, Zi
= P tDP−E.
2 − W1.
Lemma 3.9 shows that P = diag(1, 1,∓i,∓i) C diag(±i,±i, 1, 1) for some orthog-
onal matrix C. Since W3 = W4 = X3 = X4 = 0 we require that P (and hence
C) has a 2 × 2 zero block in the lower left corner. The Grobner basis for the ideal
generated by T ′′ has 16 elements; the greatest is W2X1X2 − W1X 2
• Case 11 : Define D, E as in Case 10. Then T ′′ has the same form as Case 10 but
W = [W1, W2, 0, 0]t, X = [X1, X2, 0, 0]t, Y = [Y1, Y2, Y3, 0]t. The Grobner basis for
the ideal generated by T ′′ has 13 elements; the greatest is W2X1X2 − W1X 2
2 − W1.
• Case 16 : Define D, E as in Case 2. Then T ′′ has the same form as Case 2
but W = [W1, 0, 0, 0]t. The Grobner basis for the ideal generated by T ′′ has 31
elements; the greatest is X4Y4Z2Z3 − X3Y4Z2Z4 − X4Y2Z3Z4 + X3Y2Z 2
4 + X3Y2.
• Case 17 : Define D, E as in Case 2. Then T ′′ has the same form as Case 2
but W = [W1, 0, 0, 0]t and X = [X1, X2, X3, 0]t. The Grobner basis for the ideal
generated by T ′′ has 23 elements; the greatest is Y2Y3Z 2
• Case 18 : Define D, E as in Case 2. Then T ′′ has the same form as Case 2 but W =
[W1, 0, 0, 0]t, X = [X1, X2, X3, 0]t, Y = [Y1, Y2, Y3, 0]t. We have T ′′ = P tDP−E
and the rest is the same as in Case 2 except that W2 = W3 = W4 = X4 = Y4 = 0.
The Grobner basis for the ideal generated by the entries of T ′′ has 33 elements, the
greatest of which is X3Y2Y3 − X2Y 2
• Case 20 : Define D, E as in Case 10. Then T ′′ has the same form as Case 10
but W = [W1, 0, 0, 0]t and X = [X1, X2, X3, 0]t. The Grobner basis for the ideal
generated by T ′′ has 13 elements; the greatest is Y4Z3Z4 − Y3Z 2
• Case 21 : Define D, E as in Case 10. Then T ′′ has the same form as Case 10 but
W = [W1, 0, 0, 0]t, X = [X1, X2, 0, 0]t, Y = [Y1, Y2, Y3, 0]t. The Grobner basis for
the ideal generated by the entries of T ′′ has 10 elements; the greatest is W 2
1 + 1. (cid:3)
4 + Z2Z3Z 2
4 − Y3.
4 + Y2Y3.
3 + X2.
Example 3.12. In case 21, where D, E are as in case 10, the Grobner basis is
Z3, Z2, Z1, Y2, Y1, X1, Z 2
2 + 1, W 2
From this we immediately obtain the following zero set for the ideal:
3 + 1, X 2
4 + 1, Y 2
1 + 1.
W = [±i, 0, 0, 0], X = [0,±i, 0, 0], Y = [0, 0,±i, 0], Z = [0, 0, 0,±i].
The corresponding relations between the operations ⊢,⊣ are:
(a ⊢ b) ⊢ c = ∓i (a ⊢ b) ⊣ c,
a ⊢ (b ⊢ c) = ∓i a ⊢ (b ⊣ c),
(a ⊣ b) ⊢ c = ∓i (a ⊣ b) ⊣ c,
a ⊣ (b ⊢ c) = ∓i a ⊣ (b ⊣ c).
"Whenever the second operation changes, the multiplier ±i appears."
Example 3.13. In case 18, for which D, E are as in case 2, the relation matrix is
1 W1
0
0
0
0
0
0
0
1
0
0
0 X1 Y1
0 X2 Y2
1 X3 Y3
0
0
0
0 Z1
0 Z2
0 Z3
1 Z4
SELF-DUAL NONSYMMETRIC OPERADS
11
and the Grobner basis is
Z3, Z2, Z1, Y1, X1, Z 2
X2X3 + Y2Y3, X 2
4 + 1, Y 2
2 + Y 2
3 − 1, X2Y2 + X3Y3, X 2
3 + Y 2
3 − 1,
2 − Y 2
3 , W 2
1 + 1, X3Y2Y3 − X2Y 2
3 + X2.
From this we obtain the following one-parameter family of solutions:
W1 = ±i, X1 = 0, X2 = free, X3 = ±q1 − X 2
Y2 = ∓q1 − X 2
Y3 = X2, Z1 = 0, Z2 = 0, Z3 = 0, Z4 = ±i.
Y1 = 0,
2 ,
2 ,
The second solution is obtained by changing the signs of Y2 and Y3. The first
one-parameter family gives this relation matrix (writing λ for X2):
1 ±i 0
1
0
0
0
0
0
0
0
0
0
0
0
0
λ
0
1 ±√1−λ2
∓√1−λ2
0
λ
0
0
0
0
0
0
0
1 ±i
We leave it to the reader to write down the relations between the operations ⊢,⊣.
4. Self-duality for two associative operations
Definition 4.1. The operad with two associative operations has the relation matrix
[A] whose row space is a 2-dimensional subspace A ⊂ O(2):
[A] =(cid:20) 1 0 0 0 −1 0 0
0 0 0 −1(cid:21)
0 0 0 1
0
(x1 ⊢ x2) ⊢ x3 − x1 ⊢ (x2 ⊢ x3) ≡ 0
(x1 ⊣ x2) ⊣ x3 − x1 ⊣ (x2 ⊣ x3) ≡ 0
Linear combinations of ⊢,⊣ are associative if and only if the operations satisfy the
compatibility relation (Definition 1.1).
Let [R] be a relation matrix of rank 4 for an operad with two binary operations.
[R] is in RCF with leading 1s in columns j1, j2, j3, j4. Since the operations are
associative, A ⊂ R, and stacking [R] on top of [A] gives a matrix of rank 4. Since
A has leading 1s in columns 1 and 4, we have {1, 4} ⊂ {j1, j2, j3, j4}. There are
(cid:0)6
2(cid:1) = 15 cases for the other two columns; see Table 5.
Definition 4.2. To obtain necessary and sufficient conditions for the operations to
be associative, we stack [R] on top of [A], and use the leading 1s in [R] to eliminate
the nonzero entries in columns j1, . . . , j4 in [A]. Then rows 5, 6 are zero if and only
if R + A = R, that is A ⊂ R. The equations obtained by setting the entries in rows
5, 6 to zero are the associativity conditions on the parameters.
Example 4.3. Consider the relation matrix [R] in case 1:
(cid:20) R
A (cid:21) =
1 0 0 0 W1 X1 Y1 Z1
0 1 0 0 W2 X2 Y2 Z2
0 0 1 0 W3 X3 Y3 Z3
0 0 0 1 W4 X4 Y4 Z4
1 0 0 0 −1 0 0 0
0 0 −1
0 0 0 1 0
−→
Y1
Y2
Y3
Y4
X1
X2
X3
X4
1 0 0 0 W1
Z1
0 1 0 0 W2
Z2
0 0 1 0 W3
Z3
0 0 0 1 W4
Z4
0 0 0 0 −W1−1 −X1 −Y1 −Z1
0 0 0 0 −W4 −X4 −Y4 −Z4−1
Hence A ⊂ R if and only if W1, Z4 = −1 and X1, Y1, Z1, W4, X4, Y4 = 0, which
means that rows 1 and 4 of [R] coincide with the two rows of [A].
12
MURRAY BREMNER AND JUANA S ´ANCHEZ-ORTEGA
1 0 0 0 W1 X1 Y1 Z1
0 1 0 0 W2 X2 Y2 Z2
0 0 1 0 W3 X3 Y3 Z3
0 0 0 1 W4 X4 Y4 Z4
1 0 W1 0 X1 Y1 0 Z1
0 1 W2 0 X2 Y2 0 Z2
0 0 0 1 X3 Y3 0 Z3
0 0 0 0 0 0 1 Z4
1 W1 0 0 X1 0 Y1 Z1
0 0 1 0 X2 0 Y2 Z2
0 0 0 1 X3 0 Y3 Z3
0 0 0 0 0 1 Y4 Z4
1 W1 X1 0 0 0 Y1 Z1
0 1 0 0 Y2 Z2
0 0
0 0 1 0 Y3 Z3
0 0
0 0
0 0 0 1 Y4 Z4
1
4
7
10
13
2
5
8
11
14
1 0 W1 0 0 X1 Y1 Z1
0 1 W2 0 0 X2 Y2 Z2
0 0 0 1 0 X3 Y3 Z3
0 0 0 0 1 X4 Y4 Z4
1 0 W1 0 X1 Y1 Z1 0
0 1 W2 0 X2 Y2 Z2 0
0 0 0 1 X3 Y3 Z3 0
0 0 0 0 0 0 0 1
1 W1 0 0 X1 Y1 0 Z1
0 0 1 0 X2 Y2 0 Z2
0 0 0 1 X3 Y3 0 Z3
0 0 0 0 0 0 1 Z4
1 W1 X1 0 0 Y1 0 Z1
0 1 0 Y2 0 Z2
0 0
0 0 1 Y3 0 Z3
0 0
0 0
0 0 0 0 1 Z4
3
6
9
12
15
1 0 W1 0 X1 0 Y1 Z1
0 1 W2 0 X2 0 Y2 Z2
0 0 0 1 X3 0 Y3 Z3
0 0 0 0 0 1 Y4 Z4
1 W1 0 0 0 X1 Y1 Z1
0 0 1 0 0 X2 Y2 Z2
0 0 0 1 0 X3 Y3 Z3
0 0 0 0 1 X4 Y4 Z4
1 W1 0 0 X1 Y1 Z1 0
0 0 1 0 X2 Y2 Z2 0
0 0 0 1 X3 Y3 Z3 0
0 0 0 0 0 0 0 1
1 W1 X1 0 0 Y1 Z1 0
0 1 0 Y2 Z2 0
0 0
0 0 1 Y3 Z3 0
0 0
0 0
0 0 0 0 0 1
1 W1 X1 0 Y1 0 0 Z1
0 1 Y2 0 0 Z2
0 0
0 0 0 1 0 Z3
0 0
0 0
0 0 0 0 1 Z4
Table 5. Relation matrices for two associative operations
1 W1 X1 0 Y1 0 Z1 0
0 1 Y2 0 Z2 0
0 0
0 0 0 1 Z3 0
0 0
0 0
0 0 0 0 0 1
1 W1 X1 0 Y1 Z1 0 0
0 1 Y2 Z2 0 0
0 0
0 0 0 0 1 0
0 0
0 0
0 0 0 0 0 1
The relation matrices obtained after applying the associativity conditions appear
in Table 6. The number of parameters in each case has dropped.
It remains to use Loday's algorithm (see Table 2) to determine which values of
the parameters produce self-dual operads.
Lemma 4.4. For cases 6, . . . , 15 no values of the parameters imply self-duality.
Proof. In these cases T contains a nonzero scalar; the rest follows Lemma 3.3. (cid:3)
0
0
0
0
λ
0
(λ ∈ F)
1 0 0 0 −1
0 1 0 0 0
0 0 1 0 0 ±√1−λ2
0 0 0 1 0
±λ ±√1−λ2 0
0
−1
Theorem 4.5. The quadratic nonsymmetric operad Q with two associative binary
operationsis self-dual if and only if its relation matrix [R] is one of:
1 0 0 0 −1 0 0 0
0 1 ±i 0 0 0 0 0
0 0 0 1 0 0 0 −1
0 0 0 0 0 1 ±i 0
Proof. By Lemma 4.4, self-dual associative operads exist only in cases 1, . . . , 5.
• Case 1: We obtain (omitting zero columns from −T ):
0 −1
0
0 W2 X2 Y2 Z2
0 W3 X3 Y3 Z3
0 −1
1
0
0
0
0
1
0
1
0
1
0
0
0
0
1
0
0
−1 W2 W3
0 X2 X3
0
Y3
0
−→(cid:20)R
T(cid:21) ,
(cid:20) R
R±(cid:21) =
0
0
Y2
0
Z2 Z3 −1
0
0
0
1
0
0
0
1
0
0
0
0
1
0
0
0
SELF-DUAL NONSYMMETRIC OPERADS
13
1
4
7
10
13
1 0 0 0 −1 0 0 0
0 1 0 0 W2 X2 Y2 Z2
0 0 1 0 W3 X3 Y3 Z3
0 0 −1
0 0 0 1 0
1 0 0 0 −1 0 0 0
0 1 W2 0 X2 Y2 0 Z2
0 0 0 1 0 0 0 −1
0 0 0 0 0 0 1 Z4
1 0 0 0 −1 0 0 0
0 0 1 0 X2 0 Y2 Z2
0 0 0 1 0 0 0 −1
0 0 0 0 0 1 Y4 Z4
1 0 0 0 0 0 Y1 Z1
0 0 0 1 0 0 0 −1
0 0 0 0 1 0 Y1 Z1
0 0 0 0 0 1 Y4 Z4
1 0 0 0 −1 0 0 0
0 0 0 1 0 0 0 −1
0 0 0 0 0 1 0 Z3
0 0 0 0 0 0 1 Z4
2
1 0 0 0 0 X1 Y1 Z1
0 1 W2 0 0 X2 Y2 Z2
0 0 0 1 0 0 0 −1
0 0 0 0 1 X1 Y1 Z1
1 0 0 0 −1 0 0 0
5
0 1 W2 0 X2 Y2 Z2 0
0 0 0 1 0 0 0 0
0 0 0 0 0 0 0 1
1 0 0 0 −1 0 0 0
8
0 0 1 0 X2 Y2 0 Z2
0 0 0 1 0 0 0 −1
0 0 0 0 0 0 1 Z4
11
1 0 0 0 0 Y1 0 Z1
0 0 0 1 0 0 0 −1
0 0 0 0 1 Y1 0 Z1
0 0 0 0 0 0 1 Z4
1 0 0 0 −1 0 0 0
14
0 0 0 1 0 0 0 0
0 0 0 0 0 1 Z3 0
0 0 0 0 0 0 0 1
3
6
9
12
15
1 0 0 0 −1 0 0 0
0 1 W2 0 X2 0 Y2 Z2
0 0 0 1 0 0 0 −1
0 0 0 0 0 1 Y4 Z4
1 0 0 0 0 X1 Y1 Z1
0 0 1 0 0 X2 Y2 Z2
0 0 0 1 0 0 0 −1
0 0 0 0 1 X1 Y1 Z1
1 0 0 0 −1 0 0 0
0 0 1 0 X2 Y2 Z2 0
0 0 0 1 0 0 0 0
0 0 0 0 0 0 0 1
1 0 0 0 0 Y1 Z1 0
0 0 0 1 0 0 0 0
0 0 0 0 1 Y1 Z1 0
0 0 0 0 0 0 0 1
1 0 0 0 −1 0 0 0
0 0 0 1 0 0 0 0
0 0 0 0 0 0 1 0
0 0 0 0 0 0 0 1
Table 6. Relation matrices after applying associativity conditions
−T =
W 2
2 +W 2
3
X 2
3−1
W2X2+W3X3
W2Y2+W3Y3
X2Y2+X3Y3
W2Z2+W3Z3 X2Z2+X3Z3
W2X2+W3X3 W2Y2+W3Y3 W2Z2+W3Z3
X2Y2+X3Y3 X2Z2+X3Z3
2 +Y 2
Y 2
3 −1
Y2Z2+Y3Z3
Y2Z2+Y3Z3
2 +X 2
Z 2
2 +Z 2
3
The entries of [T ] generate an ideal I in the polynomial ring Φ[P] where P is the
set of 8 parameters. The zero set V (I) consists of the parameter values for which
the relation matrix [R] defines a self-dual operad. The Grobner basis for I is
3 − 1,
3 − 1, X2Y2 + X3Y3, X 2
3 + Y 2
Z3, Z2, W3, W2,
X2X3 + Y2Y3, X 2
Y 2
2 + Y 2
3 , X3Y2Y3 − X2Y 2
2 − Y 2
3 + X2.
We obtain a one-parameter set of solutions; the signs may be chosen independently:
W2 = 0, W3 = 0, X2 = ±Y3, X3 = ±q1 − Y 2
Y2 = ±q1 − Y 2
Y3 = free, Z2 = 0, Z3 = 0.
3 ,
3 ,
With these values the relation matrix [R1] takes the indicated form.
• Case 2: In this case, after deleting the columns which are zero, we obtain
[T ] =
W2Y2
2 + 1 W2X2
W 2
−W2X2 −X 2
−W2Y2
−W2Z2 −X2Z2
W2Z2
2 + 1 −X2Y2 −X2Z2
2 + 1 −Y2Z2
−Z 2
−X2Y2 −Y 2
−Y2Z2
2
The ideal generated by the entries of [T ] has Grobner basis {1}: no solutions.
14
MURRAY BREMNER AND JUANA S ´ANCHEZ-ORTEGA
• Case 3: After deleting the zero columns, we obtain
[T ] =
2 + 1 W2X2
−X 2
2
W 2
−W2X2
−W2Y2 −X2Y2 −Y 2
−W2Z2 −X2Z2 −Y2Z2 + Y4Z4
W2Y2
−X2Y2
2 + Y 2
4 + 1 −Y2Z2 + Y4Z4
W2Z2
−X2Z2
−Z 2
2 + Z 2
4
The Grobner basis for the ideal generated by the entries is
4 + 1, W 2
Y2, X2, Y 2
Z4, Z2,
2 + 1.
With the zero set of this ideal, the relation matrix [R3] takes the indicated form.
• Cases 4, 5: The ideal generated by the entries of [T ] has Grobner basis {1}. (cid:3)
Acknowledgements
Murray Bremner was supported by a Discovery Grant from NSERC, the Natural
Sciences and Engineering Research Council of Canada. Juana S´anchez-Ortega was
supported by a CSUR grant from the NRF, the National Research Foundation of
South Africa. The authors thank Vladimir Dotsenko for pointing out the connection
between relation matrices and Schubert calculus on Grassmannians.
References
[1] M. R. Bremner, V. Dotsenko: Algebraic Operads: An Algorithmic Companion. CRC Press,
Boca Raton, 2016.
[2] W. Fulton: Young Tableaux, with Applications to Representation Theory and Geometry.
London Mathematical Society Student Texts, 35. Cambridge University Press, 1997.
[3] H. Hiller: Geometry of Coxeter Groups. Research Notes in Mathematics, 54. Pitman (Ad-
vanced Publishing Program), Boston & London, 1982.
[4] J.-L. Loday: Alg`ebres ayant deux op´erations associatives (dig`ebres). C. R. Acad. Sci. Paris
S´er. I Math. 321 (1995), no. 2, 141 -- 146.
[5] J.-L. Loday: Dialgebras. Dialgebras and Related Operads, pages 7 -- 66. Lecture Notes in
Mathematics, 1763. Springer, Berlin, 2001.
[6] J.-L. Loday, B. Vallette: Algebraic Operads. Grundlehren der mathematischen Wis-
senschaften, 346. Springer, Heidelberg, 2012.
[7] M. Markl, E. Remm: (Non-)Koszulness of operads for n-ary algebras, galgalim and other
curiosities. J. Homotopy Relat. Struct. 10 (2015), no. 4, 939 -- 969.
[8] J. M. Osborn: Identities of non-associative algebras. Canad. J. Math. 17 (1965) 78 -- 92.
[9] G. W. Zinbiel (J.-L. Loday): Encyclopedia of types of algebras 2010. Operads and Universal
Algebra, pages 217 -- 297. Nankai Ser. Pure Appl. Math. Theoret. Phys., 9. World Sci. Publ.,
Hackensack, NJ, 2012.
Department of Mathematics and Statistics, University of Saskatchewan, Saskatoon,
Canada
E-mail address: [email protected]
Department of Mathematics and Applied Mathematics, University of Cape Town,
South Africa
E-mail address: [email protected]
|
1706.04270 | 1 | 1706 | 2017-06-13T22:20:05 | The Reticulation of a Universal Algebra | [
"math.RA"
] | The reticulation of an algebra $A$ is a bounded distributive lattice ${\cal L}(A)$ whose prime spectrum of filters or ideals is homeomorphic to the prime spectrum of congruences of $A$, endowed with the Stone topologies. We have obtained a construction for the reticulation of any algebra $A$ from a semi-degenerate congruence-modular variety ${\cal C}$ in the case when the commutator of $A$, applied to compact congruences of $A$, produces compact congruences, in particular when ${\cal C}$ has principal commutators; furthermore, it turns out that weaker conditions than the fact that $A$ belongs to a congruence-modular variety are sufficient for $A$ to have a reticulation. This construction generalizes the reticulation of a commutative unitary ring, as well as that of a residuated lattice, which in turn generalizes the reticulation of a BL-algebra and that of an MV-algebra. The purpose of constructing the reticulation for the algebras from ${\cal C}$ is that of transferring algebraic and topological properties between the variety of bounded distributive lattices and ${\cal C}$, and a reticulation functor is particularily useful for this transfer. We have defined and studied a reticulation functor for our construction of the reticulation in this context of universal algebra. | math.RA | math |
The Reticulation of a Universal Algebra
George GEORGESCU and Claudia MURES¸AN∗
University of Bucharest
Faculty of Mathematics and Computer Science
Academiei 14, RO 010014, Bucharest, Romania
Emails: [email protected]; [email protected], [email protected]
October 15, 2018
Abstract
The reticulation of an algebra A is a bounded distributive lattice L(A) whose prime spectrum of filters
or ideals is homeomorphic to the prime spectrum of congruences of A, endowed with the Stone topologies.
We have obtained a construction for the reticulation of any algebra A from a semi -- degenerate congruence --
modular variety C in the case when the commutator of A, applied to compact congruences of A, produces
compact congruences, in particular when C has principal commutators; furthermore, it turns out that weaker
conditions than the fact that A belongs to a congruence -- modular variety are sufficient for A to have a retic-
ulation. This construction generalizes the reticulation of a commutative unitary ring, as well as that of a
residuated lattice, which in turn generalizes the reticulation of a BL -- algebra and that of an MV -- algebra.
The purpose of constructing the reticulation for the algebras from C is that of transferring algebraic and
topological properties between the variety of bounded distributive lattices and C, and a reticulation functor
is particularily useful for this transfer. We have defined and studied a reticulation functor for our construction
of the reticulation in this context of universal algebra.
2010 Mathematics Subject Classification: primary: 08B10; secondary: 08A30, 06B10, 06F35, 03G25.
Keywords: (congruence -- modular, congruence -- distributive) variety, commutator, (prime, compact) congru-
ence, reticulation.
1
Introduction
The reticulation of a commutative unitary ring R is a bounded distributive lattice L(R) whose prime spectrum
of ideals is homeomorphic to the prime spectrum of ideals of R. Its construction has appeared in [32], but it
has been extensively studied in [52], where it has received the name reticulation. The mapping R 7→ L(R) sets
a covariant functor from the category of commutative unitary rings to that of bounded distributive lattices,
through which properties can be transferred between these categories. In [7], the reticulation has been defined
and studied for non -- commutative unitary rings and it has been proven that such a ring has a reticulation (with
the topological definition above) iff it is quasi -- commutative.
Over the past two decades, reticulations have been constructed for orderred algebras related to logic: MV --
algebras [8, 9], BL -- algebras [37, 20, 38], residuated lattices [41, 42, 43, 44, 45, 46], 0 -- distributive lattices [49],
almost distributive lattices [50], Hilbert algebras [13], hoops [16]. All these algebras posess a "prime spectrum"
which is homeomorphic to the prime spectrum of filters or ideals of a bounded distributive lattice; their retic-
ulations consist of such bounded distributive lattices, whose study involves obtaining a construction for them
and using that construction to transfer properties between these classes of algebras and bounded distributive
lattices.
The purpose of the present paper is to set the problem of constructing a reticulation in a universal algebra
framework and providing a solution to this problem in a case as general as possible, that includes the cases of the
varieties above and generalizes the constructions which have been obtained in those particular cases. Apart from
the novelty of using commutator theory [18, 39] for the study of the reticulation, essentially, the tools needed
for obtaining reticulations in this very general setting are quite similar to those which have been put to work for
∗Corresponding author.
the classes of algebras above, and it turns out that many types of results that hold for their reticulations can be
generalized to our setting. In order to obtain strong generalizations, we have worked with hypotheses as weak
as possible; all our results in this paper hold for semi -- degenerate congruence -- modular varieties whose members
have the sets of compact congruences closed with respect to the commutator, with just a few exceptions that
necessitate, moreover, principal commutators.
The present paper is structured as follows: Section 2 presents the notations and basic results we use in what
follows; Section 3 collects a set of results from commutator theory which we use in the sequel; in Section 4,
we present the standard construction of the Stone topologies on prime spectra, specifically the prime spectrum
of ideals of a bounded distributive lattice and the prime spectrum of congruences of a universal algebra whose
commutator fulfills certain conditions. The results in the following sections that are not cited from other papers,
or mentioned as being either known or quite simple to obtain, are new and original.
In Section 5, we construct the reticulation for universal algebras whose commutators fulfill certain conditions,
prove that this construction has the desired topological property and obtain some related results.
In Section 6, we provide some examples of reticulations, study particular cases, such as the congruence --
distributive case, show that our construction generalizes constructions for the reticulation which have been
obtained for particular varieties, and prove that our construction preserves finite direct products of algebras
without skew congruences.
In Section 7, we obtain some arithmetical properties on commutators that we need in what follows, as well
as algebraic properties regarding the behaviour of surjections with respect to commutators and to certain types
of congruences.
In Section 8 we study the behaviour of Boolean congruences with respect to the reticulation, in the general
case, but also in particular ones, such as the case of associative commutators or that of semiprime algebras.
In Section 9, we define a reticulation functor; our definition is not ideal, as it only acts on surjections;
extending it to all morphisms remains an open problem. In this final section, we also show that the reticulation
preserves quotients, and that it is a Boolean lattice exactly in the case of hyperarchimedean algebras, which we
also characterize by several other conditions on their reticulation. These characterizations serve as an example
for the transfer of properties to and from the category of bounded distributive lattices which the reticulation
makes possible.
We intend to further pursue the study of the reticulation in this universal algebra setting and use it to
transfer more properties between the variety of bounded distributive lattices and the kinds of varieties that
allow a construction for the reticulation. A theme for a potentially extensive future study is characterizing those
varieties with the property that the reticulations of their members cover the entire class of bounded distributive
lattices.
2 Preliminaries
In this section, we recall some properties on lattices and congruences in universal algebras. For a further study
of the following results on universal algebras, we refer the reader to [1], [12], [27], [34]. For those on lattices, we
recommend [5], [11], [17], [26], [51].
We shall denote by N the set of the natural numbers and by N∗ = N\{0}. For any set M , P(M ) shall be the
set of the subsets of M , idM : M → M shall be the identity map, and we shall denote by ∆M = {(x, x) x ∈ M}
and ∇M = M 2. For any family (Mi)i∈I of sets and any M ⊆ Y
Mi, whenever there is no danger of confusion,
by a = (ai)i∈I ∈ M we mean ai ∈ Mi for all i ∈ I, such that a ∈ M . For any sets M , N and any function
f : M → N , we shall denote by Ker(f ) = {(x, y) ∈ M 2 f (x) = f (y)}, and the direct and inverse image of f in
the usual way; we shall denote, simply, f = f 2 : P(M 2) → P(N 2) and f ∗ = (f 2)−1 : P(N 2) → P(M 2); so, for
any X ⊆ M 2 and any Y ⊆ N 2, f (X) = {(f (a), f (b)) (a, b) ∈ X} and f ∗(Y ) = {(a, b) ∈ M 2 (f (a), f (b)) ∈ Y },
thus Ker(f ) = f ∗(∆N ). Also, if Xi ⊆ M 2
i for all i ∈ I, then the direct product of (Xi)i∈I as a family of binary
relations shall be denoted just as the one for sets, because there will be no danger of confusion when using
this notation: Y
Xi = {((ai)i∈I , (bi)i∈I ) (∀ i ∈ I) ((ai, bi) ∈ Xi)} ⊆ M 2. Unless mentioned otherwise, the
operations and order relation of a (bounded) lattice shall be denoted in the usual way, and the complementation
of a Boolean algebra shall be denoted by ¬ .
Throughout this paper, whenever there is no danger of confusion, any algebra shall be designated by its
i∈I
i∈I
support set. All algebras shall be considerred non -- empty; by trivial algebra we shall mean one -- element algebra,
and by non -- trivial algebra we shall mean algebra with at least two distinct elements. Any direct product of
algebras and any quotient algebra shall be considerred with the operations defined canonically. For brevity, we
shall denote by A ∼= B the fact that two algebras A and B of the same type are isomorphic.
Let L be a bounded lattice. By Id(L) we shall denote the set of the ideals of L, that is the non -- empty subsets
of L which are closed with respect to the join and to lower bounds. By Filt(L) we shall denote the set of the filters
of L, that is the ideals of the dual of L: the non -- empty subsets of L which are closed with respect to the meet
and to upper bounds. For any M ⊆ L and any a ∈ L, (M ], respectively [M ), shall denote the ideal, respectively
the filter of L generated by M , and the principal ideal, ({a}] = {x ∈ L a ≥ x}, respectively the principal filter,
[{a}) = {x ∈ L a ≤ x}, generated by a shall also be denoted by (a], respectively [a); whenever we need to
specify the lattice L, we shall denote [M )L, (M ]L, [a)L and (a]L instead of [M ), (M ], [a) and (a], respectively.
It is well known that (Id(L),∨,∩,{0}, L) and (Filt(L),∨,∩,{1}, L) are bounded lattices, with J ∨ K = (J ∪ K]
and F ∨ G = [F ∪ G) for all J, K ∈ Id(L) and all F, G ∈ Filt(L), and they are distributive iff L is distributive;
moreover, they are complete lattices, with _
Fi] for any families (Ji)i∈I ⊆ Id(L)
and (Fi)i∈I ⊆ Filt(L). Obviously, for any a, b ∈ L, (a] ∨ (b] = (a ∨ b], (a] ∩ (b] = (a ∧ b], [a) ∨ [b) = [a ∧ b) and
[a) ∩ [b) = [a∨ b). If L is a complete lattice, then, for any family (ai)i∈I ⊆ L, _
ai],
_
[ai) = [^
ai). By PId(L), respectively PFilt(L), we shall denote the set of the principal
Ji) and _
ai) and \
(ai] = (_
(ai] = (^
[ai) = [_
Fi = ([
Ji = [[
ai], \
i∈I
i∈I
i∈I
i∈I
i∈I
i∈I
i∈I
i∈I
i∈I
i∈I
i∈I
i∈I
ideals, respectively the principal filters of L. We shall denote by MaxId(L), respectively MaxFilt(L), the set of
the maximal ideals, respectively the maximal filters of L, that is the maximal elements of the set of proper ideals
of L, Id(L) \ {L}, respectively that of proper filters of L, Filt(L) \ {L}. By SpecId(L) we shall denote the set of
the prime ideals of L, that is the proper ideals P of L such that, for any x, y ∈ L, x ∧ y ∈ P implies x ∈ P or
y ∈ P . Dually, SpecFilt(L) shall denote the set of the prime filters of L, that is the proper filters P of L such
that, for any x, y ∈ L, x ∨ y ∈ P implies x ∈ P or y ∈ P .
For any algebra A, Con(A) shall denote the set of the congruences of A, and Max(A) shall denote the
set of the maximal congruences of A, that is the maximal elements of the set of proper congruences of A:
Con(A) \ {∇A}. Let θ ∈ Con(A), a ∈ A, M ⊆ A and X ⊆ A2, arbitrary. Then a/θ shall denote the congruence
class of a with respect to θ, M/θ = {x/θ x ∈ M}, pθ : A → A/θ shall be the canonical surjective morphism:
pθ(a) = a/θ for all a ∈ A, X/θ = {(x/θ, y/θ) (x, y) ∈ X} and CgA(X) shall be the congruence of A generated
It is well known that (Con(A),∨,∩, ∆A,∇A) is a bounded lattice, orderred by set inclusion, where
by X.
φ ∨ ψ = CgA(φ ∪ ψ) for all φ, ψ ∈ Con(A); moreover, this is a complete lattice, in which _
φi) for
any family (φi)i∈I ⊆ Con(A). For any a, b ∈ A, the principal congruence CgA({(a, b)}) shall also be denoted by
CgA(a, b). The set of the principal congruences of A shall be denoted by PCon(A). K(A) shall denote the set of
the finitely generated congruences of A, which coincide to the compact elements of the lattice Con(A). Clearly,
PCon(A) ⊆ K(A) and ∆A ∈ PCon(A), because ∆A = CgA(x, x) for any x ∈ A.
Throughout the rest of this paper, τ shall be a universal algebras signature, C shall be an equational class
of τ -- algebras A and B shall be algebras from C and f : A → B shall be a morphism in C. Unless mentioned
otherwise, by morphism we shall mean τ -- morphism. We recall that A is said to be congruence -- modular, re-
spectively congruence -- distributive, iff the lattice Con(A) is modular, respectively distributive, and that C is said
to be congruence -- modular, respectively congruence -- distributive, iff every algebra in C is congruence -- modular,
respectively congruence -- distributive.
Remark 2.1. If β ∈ Con(B), then f ∗(β) ∈ Con(A); thus Ker(f ) = f ∗(∆B) ∈ Con(A). Also, f ∗(β) ⊇ f ∗(∆B) =
Ker(f ) and f (f ∗(β)) = β ∩ f (A2), thus, if f is surjective, then f (f ∗(β)) = β.
If α ∈ Con(A) such that α ⊇ Ker(f ), then f (α) ∈ Con(f (A)), so, if f is surjective, then f (α) ∈ Con(B).
Thus, for any α ∈ Con(A), we have f (α∨Ker(f )) ∈ Con(f (A)), so, if f is surjective, then f (α∨Ker(f )) ∈ Con(B).
Moreover, α 7→ f (α) is an order isomorphism from [Ker(f )) ∈ PFilt(Con(A)) to Con(f (A)), thus to Con(B) if
f is surjective, having the corresponding restriction of f ∗ as inverse.
For any θ ∈ Con(A), clearly, Ker(pθ) = θ. By the above, for all α ∈ Con(A) such that α ⊇ θ, α/θ = pθ(α) =
φi = CgA([
{(a/θ, b/θ) (a, b) ∈ α} ∈ Con(A/θ), and α 7→ α/θ is a bijection from [θ) to Con(A/θ).
i∈I
i∈I
3 The Commutator
This section is composed of results on the commutator in arbitrary and in congruence -- modular varieties, which
are either previously known of very easy to derive from previously known results. For a further study of these
results, see [1], [21], [34], [48].
Out of the various definitions for commutator operations on congruence lattices, we have chosen to work with
the term condition commutator, from the following definition. Recall that, in algebras from congruence -- modular
varieties, all definitions for the commutator give the same commutator operation. For any term t over τ , we
shall denote by tA the derivative operation of A associated to t.
Definition 3.1. [39] Let α, β ∈ Con(A). For any µ ∈ Con(A), by C(α, β; µ) we denote the fact that the
if (ai, bi) ∈ α for all
following condition holds:
i ∈ 1, n and (cj, dj) ∈ β for all j ∈ 1, k, then (tA(a1, . . . , an, c1, . . . , ck), tA(a1, . . . , an, d1, . . . , dk)) ∈ µ iff
(tA(b1, . . . , bn, c1, . . . , ck), tA(b1, . . . , bn, d1, . . . , dk)) ∈ µ. We denote by [α, β]A = T{µ ∈ Con(A) C(α, β; µ)};
we call [α, β]A the commutator of α and β in A.
Remark 3.2. Let α, β ∈ Con(A). Clearly, C(α, β;∇A). Since Con(A) is a complete lattice, it follows that
[α, β]A ∈ Con(A). Furthermore, according to [39, Lemma 4.4,(2)], for any family (µi)i∈I ⊆ Con(A), if C(α, β; µi)
for all i ∈ I, then C(α, β; \
µi). Hence C(α, β; [α, β]A), and thus [α, β]A = min{ µ ∈ Con(A) C(α, β; µ)},
for all n, k ∈ N and any term t over τ of arity n + k,
i∈I
which is exactly the definition of the commutator from [40].
Definition 3.3. The operation [·,·]A : Con(A) × Con(A) → Con(A) is called the commutator of A.
Theorem 3.4. [21] If C is congruence -- modular, then, for each member M of C, [·,·]M is the unique binary
operation on Con(M ) such that, for all α, β ∈ Con(M ), [α, β]M = min{µ ∈ Con(M ) µ ⊆ α ∩ β and, for any
member N of C and any surjective morphism h : M → N in C, µ∨Ker(h) = h∗([h(α∨Ker(h)), h(β∨Ker(h))]N )}.
Theorem 3.5. [31] If C is congruence -- distributive, then, in each member of C, the commutator coincides to the
intersection of congruences.
For brevity, most of the times, we shall use the remarks in this paper without referencing them, and the
same goes for the lemmas and propositions that state basic results.
Proposition 3.6. [39, Lemma 4.6,Lemma 4.7,Theorem 8.3] The commutator is:
• increasing in both arguments, that is, for all α, β, φ, ψ ∈ Con(A), if α ⊆ β and φ ⊆ ψ, then [α, φ]A ⊆
[β, ψ]A;
• smaller than its arguments, so, for any α, β ∈ Con(A), [α, β]A ⊆ α ∩ β.
If C is congruence -- modular, then the commutator is also:
• commutative, that is [α, β]A = [β, α]A for all α, β ∈ Con(A);
• distributive in both arguments with respect to arbitrary joins, that is, for any families (αi)i∈I and (βj)j∈J
of congruences of A, [_
βj]A = _
αi, _
[αi, βj]A.
_
i∈I
j∈J
i∈I
j∈J
Remark 3.7. Assume that [·,·]A is commutative. Then the distributivity of [·,·]A in both arguments w.r.t.
arbitrary joins is equivalent to its distributivity in one argument w.r.t. arbitrary joins, which in turn is equivalent
to its distributivity w.r.t. the join in the case when Con(A) is finite, in particular when A is finite.
Obviously, if [·,·]A equals the intersection and it is distributive w.r.t. the join (by Proposition 3.6, the latter
holds if C is congruence -- modular), then A is congruence -- distributive.
Lemma 3.8. [21] If C is congruence -- modular and S is a subalgebra of A, then, for any α, β ∈ Con(A), [α ∩
S2, β ∩ S2]S ⊆ [α, β]A ∩ S2.
Proposition 3.9. [48, Theorem 5.17, p. 48] Assume that C is congruence -- modular, and let n ∈ N∗, M1, . . . , Mn
be algebras from C, M =
[αi, βi]Mi.
Mi and, for all i ∈ 1, n, αi, βi ∈ Con(Mi). Then: [
βi]M =
nY
nY
αi,
i=1
nY
i=1
i=1
nY
i=1
Remark 3.10. By Theorem 3.4 and Remark 2.1, if C is congruence -- modular, α, β, θ ∈ Con(A) and f is
surjective, then [f (α∨Ker(f )), f (β∨Ker(f ))]B = f ([α, β]A∨Ker(f )), thus [(α∨θ)/θ, (β∨θ)/θ]B = ([α, β]A∨θ)/θ,
hence, if θ ⊆ [α, β]A, then [α/θ, β/θ]A/θ = [α, β]A/θ.
Definition 3.11. [21] Let φ be a proper congruence of A. Then φ is called a prime congruence of A iff, for
all α, β ∈ Con(A), [α, β]A ⊆ φ implies α ⊆ φ or β ⊆ φ. φ is called a semiprime congruence of A iff, for all
α ∈ Con(A), [α, α]A ⊆ φ implies α ⊆ φ.
The set of the prime congruences of A shall be denoted by Spec(A). Spec(A) is called the (prime) spectrum
of A and Max(A) is called the maximal spectrum of A.
Following [34], we say that C is semi -- degenerate iff no non -- trivial algebra in C has one -- element subalgebras.
For instance, the class of unitary rings and any class of bounded orderred structures is semi -- degenerate.
Lemma 3.12. [1, Theorem 5.3] If C is congruence -- modular and semi -- degenerate, then:
• any proper congruence of A is included in a maximal congruence of A;
• any maximal congruence of A is prime.
Remark 3.13. By Lemma 3.12, if A is non -- trivial and C is congruence -- modular and semi -- degenerate, then A
has maximal congruences, thus it has prime congruences.
Proposition 3.14. [34] C is semi -- degenerate iff, for all members M of C, ∇M ∈ K(M ).
Proposition 3.15. [21, Theorem 8.5, p. 85] If C is congruence -- modular, then the following are equivalent:
(i) for any algebra M from C, [∇M ,∇M ]M = ∇M ;
(ii) for any algebra M from C and any θ ∈ Con(M ), [θ,∇M ]M = θ;
(iii) C has no skew congruences, that is, for any algebras M and N from C, Con(M × N ) = {θ × ζ θ ∈
Con(M ), ζ ∈ Con(N )}.
Lemma 3.16.
(i) If C is congruence -- modular and semi -- degenerate, then C fulfills the equivalent conditions
from Proposition 3.15.
(ii) If C is congruence -- distributive, then C fulfills the equivalent conditions from Proposition 3.15.
Proof. (i) This is exactly [1, Lemma 5.2].
(ii) Clear, from Theorem 3.5.
Lemma 3.17. [4, Lemma 1.11], [53, Proposition 1.2] If f is surjective, then, for any a, b ∈ A, any X ⊆ A2,
any θ ∈ Con(A) and any α, β ∈ [Ker(f )):
(i) f (θ ∨ Ker(f )) = CgB(f (θ)); f (α ∨ β) = f (α) ∨ f (β);
(ii) f (CgA(a, b) ∨ Ker(f )) = CgB(f (a), f (b)); f (CgA(X) ∨ Ker(f )) = CgB(f (X));
(iii) (CgA(a, b) ∨ θ)/θ = CgA/θ(a/θ, b/θ); (CgA(X) ∨ θ)/θ = CgA/θ(X/θ).
We say that A has principal commutators iff, for all α, β ∈ PCon(A), we have [α, β]A ∈ PCon(A), that is iff
PCon(A) is closed with respect to the commutator of A. Following [1], we say that C has principal commutators
iff each member of C has principal commutators. We say that C has associative commutators iff, for each member
M of C, the commutator of M is an associative binary operation on Con(M ).
Remark 3.18. K(A) = {CgA(∅)} ∪ {CgA({(a1, b1), . . . , (an, bn)}) n ∈ N∗, a1, b1, . . . , an, bn ∈ A} = {∆A} ∪
n_
CgA(ai, bi) n ∈ N∗, a1, b1, . . . , an, bn ∈ A}, since
{
∆A ∈ PCon(A). From this, it is immediate that K(A) is closed with respect to finite joins, and, if A has
principal commutators and [·,·]A is commutative and distributive w.r.t. the join (for instance if C is congruence --
modular), then K(A) is also closed with respect to the commutator of A.
CgA(ai, bi) n ∈ N∗, a1, b1, . . . , an, bn ∈ A} = {
i=1
n_
i=1
Remark 3.19. If C is congruence -- distributive, then, as shown by Theorem 3.5:
• C has principal commutators iff C has the principal intersection property (PIP);
• K(M ) is closed with respect to the commutator for each member M of C iff C has the compact intersection
property (CIP).
As a particular case of Remark 3.18, if C is congruence -- distributive and has the PIP, then C has the CIP.
Example 3.20. [1], [10], [25], [31, Theorem 2.8], [33], [36] As shown by Theorem 3.5, any congruence -- distributive
variety has associative commutators. The variety of commutative unitary rings is semi -- degenerate, congruence --
modular, with principal commutators and associative commutators, and it is not congruence -- distributive. Out of
the semi -- degenerate congruence -- distributive varieties with the CIP, we mention semi -- degenerate filtral varieties.
Out of the semi -- degenerate congruence -- distributive varieties with the PIP, we mention: bounded distributive lat-
tices, residuated lattices (a variety which includes Godel algebras, product algebras, MTL -- algebras, BL -- algebras,
MV -- algebras) and semi -- degenerate discriminator varieties (out of which we mention Boolean algebras, n -- valued
Post algebras, n -- valued Lukasiewicz algebras, n -- valued MV -- algebras, n -- dimensional cylindric algebras, Godel
residuated lattices).
4 The Stone Topologies on Prime and Maximal Spectra
In what follows, we present the Stone topologies on the prime and maximal spectra of ideals and filters of a
bounded distributive lattice and those of congruences of an algebra with the greatest congruence compact from
a congruence -- modular variety; in particular, the following hold for algebras from semi -- degenerate congruence --
modular varieties. The results in this section are either previously known or very easy to derive from previously
known results; see, for instance, [30].
Let L be a bounded distributive lattice. For any I ∈ Id(L) and any a ∈ L, we shall denote by VId,L(I) =
SpecId(L) ∩ [I) = {P ∈ SpecId(L) I ⊆ P}, DId,L(I) = SpecId(L) \ VId,L(I) = {Q ∈ SpecId(L) I * Q},
VId,L(a) = VId,L((a]) = {P ∈ SpecId(L) a ∈ P} and DId,L(a) = DId,L((a]) = SpecId(L) \ VId,L(a) = {Q ∈
SpecId(L) a /∈ Q}. By replacing SpecId(L) with SpecFilt(L), in the same way we can define VFilt,L(F ),
DFilt,L(F ), VFilt,L(a) and DFilt,L(a) for any F ∈ Filt(L) and any a ∈ L.
Remark 4.1. The following hold, and their duals hold for filters:
Ji) = \
VId,L(Ji) and DId,L(_
• for any J, K ∈ Id(L), VId,L(J ∩ K) = VId,L(J) ∪ VId,L(K) and DId,L(J ∩ K) = DId,L(J) ∩ DId,L(K);
• for any family (Ji)i∈I ⊆ Id(L), VId,L(_
• thus, for any a, b ∈ L, VId,L(a ∧ b) = VId,L(a) ∪ VId,L(b), DId,L(a ∧ b) = DId,L(a) ∩ DId,L(b), VId,L(a ∨ b) =
VId,L(a) ∩ VId,L(b) and DId,L(a ∨ b) = DId,L(a) ∪ DId,L(b);
• if L is a complete lattice, then, for any family (ai)i∈I ⊆ L, VId,L(_
[
ai) = \
VId,L(ai) and DId,L(_
Ji) = [
DId,L(Ji);
DId,L(Ji);
Ji) =
i∈I
i∈I
i∈I
i∈I
i∈I
i∈I
i∈I
i∈I
• if I ∈ Id(L), then: DId,L(I) = SpecId(L) iff VId,L(I) = ∅ iff I = L;
• DId,L({0}) = ∅ and VId,L({0}) = SpecId(L);
• if L is distributive (so that the Prime Ideal Theorem holds in L and, hence, any ideal of L equals the
intersection of the prime ideals that include it) and I ∈ Id(L), then: DId,L(I) = ∅ iff VId,L(I) = SpecId(L)
iff I = {0}.
As shown by Remark 4.1, {DId,L(I) I ∈ Id(L)} is a topology on SpecId(L), called the Stone topology,
having {DId,L(a) a ∈ L} as a basis and, obviously, {VId,L(I) I ∈ Id(L)} as the family of closed sets and
{VId,L(a) a ∈ L} as a basis of closed sets. Since MaxId(L) ⊆ SpecId(L), {DId,L(I) ∩ MaxId(L) I ∈ Id(L)}
is a topology on MaxId(L), which is also called the Stone topology, and it has {DId,L(a) ∩ MaxId(L) a ∈ L}
as a basis, {VId,L(I) ∩ MaxId(L) I ∈ Id(L)} as the family of closed sets and {VId,L(a) ∩ MaxId(L) a ∈ L}
as a basis of closed sets. Dually, we have the Stone topologies on SpecFilt(L) and MaxFilt(L). SpecId(L),
MaxId(L), SpecFilt(L) and MaxFilt(L) are called the (prime) spectrum of ideals, maximal spectrum of ideals,
(prime) spectrum of filters and maximal spectrum of filters of L, respectively.
Throughout the rest of this section, we shall assume that [·,·]A is commutative and distributive w.r.t. ar-
bitrary joins. For each θ ∈ Con(A), we shall denote by VA(θ) = Spec(A) ∩ [θ) = {φ ∈ Spec(A) θ ⊆ φ} and
by DA(θ) = Spec(A) \ VA(θ) = {ψ ∈ Spec(A) θ * ψ}. We shall also denote, for any a, b ∈ A, by VA(a, b) =
VA(CgA(a, b)) = {φ ∈ Spec(A) (a, b) ∈ φ} and by DA(a, b) = DA(CgA(a, b)) = {ψ ∈ Spec(A) (a, b) /∈ ψ}.
The proof of the following result is straightforward.
Proposition 4.2. [1] (Spec(A),{DA(θ) θ ∈ Con(A)}) is a topological space, having {DA(a, b) a, b ∈ A} as a
basis and in which, for all α, β ∈ Con(A) and any family (αi)i∈I ⊆ Con(A), the following hold:
(i) DA(∆A) = ∅ and DA(∇A) = Spec(A); VA(∆A) = Spec(A) and VA(∇A) = ∅;
(ii) DA([α, β]A) = DA(α ∩ β) = DA(α) ∩ DA(β) =; VA([α, β]A) = VA(α ∩ β) = VA(α) ∪ VA(β);
(iii) DA(_
DA(αi); VA(_
αi) = [
αi) = \
VA(αi).
i∈I
i∈I
i∈I
i∈I
{DA(θ) θ ∈ Con(A)} is called the Stone topology on Spec(A). Obviously, its family of closed sets is
{VA(θ) θ ∈ Con(A)}, and {VA(a, b) a, b ∈ A} is a basis of closed sets for this topology. The Stone topology
on Spec(A) induces the Stone topology on Max(A), namely {DA(θ) ∩ Max(A) θ ∈ Con(A)}.
Remark 4.3. Let α, β ∈ Con(A). Then, clearly:
• VA(α) ⊆ VA(β) iff Spec(A) \ DA(α) ⊆ Spec(A) \ DA(β) iff DA(β) ⊆ DA(α);
• if α ⊆ β, then VA(β) ⊆ VA(α) and DA(α) ⊆ DA(β).
Proposition 4.4. If C is congruence -- modular and semi -- degenerate, then, for any α ∈ Con(A): DA(α) =
Spec(A) iff VA(α) = ∅ iff α = ∇A.
Proof. DA(α) = Spec(A) iff Spec(A) \ DA(α) = ∅ iff VA(α) = ∅. Since Spec(A) ⊆ Con(A) \ {∇A}, we have
VA(∇A) = ∅, which was also part of Proposition 4.2. If α 6= ∇A, then, according to Lemma 3.12, there exists a
φ ∈ Spec(A) such that α ⊆ φ, that is VA(α) 6= ∅.
Remark 4.5. Recall that, if f is surjective, then the map α 7→ f (α) is a lattice isomorphism from [Ker(f )) to
Con(B). Now assume that C is congruence -- modular.
Then this map is an order isomorphism from Max(A) ∩ [Ker(f )) to Max(B). Furthermore, this map is an
order isomorphism from Spec(A)∩ [Ker(f )) to Spec(B) (see also [1], [25], [47]). Hence, if Ker(f ) ⊆ α ∈ Con(A),
then VB(f (α)) = f (VA(α)) and [f (α)) ∩ Max(B) = f ([α) ∩ Max(A)).
Therefore, for all θ ∈ Con(A), the map α 7→ α/θ is a lattice isomorphism from [θ) to Con(A/θ), an order
isomorphism from Max(A)∩ [θ) to Max(A/θ) and an order isomorphism from Spec(A)∩ [θ) to Spec(A/θ); hence,
if θ ⊆ α ∈ Con(A), then VA/θ(α/θ) = {ψ/θ ψ ∈ VA(α)} and [α/θ) ∩ Max(A/θ) = {ψ/θ ψ ∈ [α) ∩ Max(A)}.
5 The Construction of the Reticulation of a Universal Algebra and
Related Results
Throughout this section, we shall assume that [·,·]A is commutative and distributive w.r.t. arbitrary joins, and
that ∇A ∈ K(A). For every θ ∈ Con(A), we shall denote by ρA(θ) the radical of θ, that is the intersection of the
prime congruences of A which include θ: ρA(θ) = \{φ ∈ Spec(A) θ ⊆ φ} = \
φ.
φ∈VA(θ)
Remark 5.1. Let α, β ∈ Con(A) and φ ∈ Spec(A). Then, clearly:
(i) VA(∇A) = ∅, and thus ρA(∇A) = ∇A;
(ii) ρA(φ) = φ; moreover, ρA(α) = α iff α is the intersection of a family of prime congruences of A;
(iii) if α ⊆ β, then VA(α) ⊇ VA(β), hence ρA(α) ⊆ ρA(β);
(iv) if α ⊆ φ, then ρA(α) ⊆ φ, since φ ∈ VA(α).
A, [α, β]n
A]A, and by (α, β]1
Following [1], for any α, β ∈ Con(A) and every n ∈ N∗, we denote by [α, β]1
[[α, β]n
Lemma 5.2. For all n ∈ N∗, any α, β ∈ Con(A) and any family (αi)i∈I ∈ Con(A):
(i) α ⊆ ρA(α);
(ii) VA(α) = VA(ρA(α));
A = [α, β]A and (α, β]n+1
A = (α, (α, β]n
A]A.
A = [α, β]A and [α, β]n+1
A =
(iii) VA(_
αi) = VA(_
ρA(αi));
i∈I
i∈I
(iv) VA([α, β]n
(v) VA([α, α]n
A) = VA([α, β]A) = VA(α ∩ β) = VA(α) ∪ VA(β);
A) = VA([α, α]A) = VA(α).
Proof. (i) Trivial.
(ii) By (i) and Remark 4.3, VA(ρA(α)) ⊆ VA(α). If φ ∈ VA(α), then φ ∈ VA(ρA(α)), according to Remark 5.1,
(iv), thus VA(α) ⊆ VA(ρA(α)). Hence VA(α) = VA(ρA(α)).
(iii) By (ii) and Proposition 4.2, (iii), VA(_
(iv) By Proposition 4.2, (ii). VA([α, β]A) = VA(α ∩ β) = VA(α) ∪ VA(β) Now we prove that VA([α, β]n
VA(α) ∪ VA(β) by induction on n ∈ N∗. VA([α, β]1
such that VA([θ, ζ]n
A, [α, β]n
VA([α, β]n
(v) By (iv).
A) =
A) = VA([α, β]A) = VA(α) ∪ VA(β). Now let n ∈ N∗
A]A) =
A) = VA(θ) ∪ VA(ζ) for all θ, ζ ∈ Con(A). Then VA([α, β]n+1
VA(ρA(αi)) = VA(_
A) = VA(α) ∪ VA(β).
A) ∪ VA([α, β]n
VA(αi) = \
A ) = VA([[α, β]n
A) = VA([α, β]n
αi) = \
ρA(αi)).
i∈I
i∈I
i∈I
i∈I
Proposition 5.3. For all α, β, θ ∈ Con(A), the following hold:
(i) ρA(α) ⊆ ρA(β) iff α ⊆ ρA(β) iff VA(α) ⊇ VA(β);
(ii) ρA(α) = ρA(β) iff VA(α) = VA(β);
(iii) if θ ⊆ α, then ρA/θ(α/θ) = ρA(α)/θ;
(iv) ρA/θ(∆A/θ) = ρA(θ)/θ;
(v) ρA/θ((α ∨ θ)/θ) = ρA(α ∨ θ)/θ.
Proof. (i) Clearly, if VA(α) ⊇ VA(β), then ρA(α) ⊆ ρA(β). If ρA(α) ⊆ ρA(β), then, since α ⊆ ρA(α), it follows
that α ⊆ ρA(β). Finally, if α ⊆ ρA(β), then VA(α) ⊇ VA(ρA(β)) = VA(β), by Remark 5.1, (iii), and Lemma 5.2,
(ii).
(ii) By (i).
(iii) If θ ⊆ α, then we may write: ρA/θ(α/θ) = \
φ/θ = ( \
ψ = \
φ)/θ = ρA(α)/θ.
ψ∈VA/θ (α/θ)
φ∈VA(α)
φ∈VA(α)
(iv) By (iii), ρA/θ(∆A/θ) = ρA/θ(θ/θ) = ρA(θ)/θ.
(v) By (iii).
Proposition 5.4. For any n ∈ N∗, any α ∈ Con(A) and any family (αi)i∈I ⊆ Con(A):
(i) if C is congruence -- modular and semi -- degenerate, then: ρA(α) = ∇A iff α = ∇A;
(ii) ρA([α, β]n
(iii) ρA([α, α]n
A) = ρA([α, β]A) = ρA(α ∩ β) = ρA(α) ∩ ρA(β);
A) = ρA([α, α]A) = ρA(α);
(iv) ρA(ρA(α)) = ρA(α);
ρA(αi)) = ρA(_
(v) ρA(_
(vi) if C is congruence -- modular and semi -- degenerate, then: _
αi);
i∈I
i∈I
i∈I
ρA(αi) = ∇A iff _
i∈I
αi = ∇A.
Proof. (i) By Lemma 5.2, (i), ∇A ⊆ ρA(∇A), thus ρA(∇A) = ∇A. If α 6= ∇A, then there exists φ ∈ VA(α), thus
ρA(α) ⊆ φ ( ∇A.
(ii) By Remark 5.1, (iii), Lemma 5.2, (iv), and Proposition 4.2, (ii), ρA([α, β]n
A) = ρA([α, β]A) = ρA(α ∩ β) =
\
φ =
\
= \
φ ∩ \
φ = ρA(α) ∩ ρA(β).
φ∈VA(α)∪VA(β)
φ∈VA(α∩β)
(iii) By (ii).
(iv) By Remark 5.1, (iii), and Lemma 5.2, (ii).
φ∈VA(α)
φ∈VA(β)
(v) By Remark 5.1, (iii), and Lemma 5.2, (iii), ρA(_
(vi) By (v) and (i), _
ρA(αi) = ∇A iff ρA(_
ρA(αi)) = ρA(_
ρA(αi)) = ∇A iff ρA(_
i∈I
i∈I
i∈I
i∈I
αi).
αi) = ∇A iff _
i∈I
i∈I
αi = ∇A.
The radical congruences of A are the congruences α of A such that α = ρA(α). Let us denote by RCon(A)
the set of the radical congruences of A.
Remark 5.5. By Remark 5.1, (ii), Spec(A) ⊆ RCon(A); moreover, the elements of RCon(A) are exactly the
intersections of prime congruences of A.
Remark 5.6. RCon(A) = {α ∈ Con(A) α = ρA(α)} = {ρA(α) α ∈ Con(A)}. Indeed, the first of these
equalities is the definition of RCon(A) and the second equality follows from Proposition 5.4, (iv).
Proposition 5.7. If the commutator of A equals the intersection, in particular if C is congruence -- distributive,
then RCon(A) = Con(A).
Proof. By [1, Lemma 1.6], the radical congruences of A coincide to its semiprime congruences, that is the
congruences θ of A such that, for all α ∈ Con(A), [α, α]A ⊆ θ implies α ⊆ θ. Clearly, if [·,·]A = ∩, then every
congruence of A is semiprime, and thus radical.
Most of the previous results on the radicals of congruences are known, but, for the sake of completeness, we
∨ β = ρA(α ∨ β). For any
have provided short proofs for them. For any α, β ∈ Con(A), let us denote by α
family (αi)i∈I ⊆ Con(A), we shall denote by
αi = ρA(_
•_
αi).
•
i∈I
i∈I
Proposition 5.8. (RCon(A),
•
∨,∩, ρA(∆A), ρA(∇A) = ∇A) is a bounded lattice, orderred by set inclusion. More-
over, it is a complete lattice, in which the arbitrary join is given by the
•_ defined above.
•
Proof. Of course, ∩ is idempotent, commutative and associative, and, clearly,
∨ is commutative. Now let
α, β, γ ∈ Con(A) and R = {ρA(α), ρA(β), ρA(γ)} ⊆ RCon(A); we shall use Proposition 5.4, (ii), (iv) and
∨ is idempotent;
(v): α, β, γ ∈ Con(A), ρA(α)
∨ ρA(β ∨ γ) = ρA(ρA(α) ∨ ρA(ρA(β ∨ γ))) =
ρA(α)
•_
θ, thus, by
ρA(ρA(α) ∨ ρA(β ∨ γ)) = ρA(α ∨ (β ∨ γ)) = ρA(α ∨ β ∨ γ) = ρA(ρA(α) ∨ ρA(β) ∨ ρA(γ)) =
∨ ρA(α) = ρA(ρA(α) ∨ ρA(α)) = ρA(α ∨ α) = ρA(α), so
∨ ρA(ρA(β) ∨ ρA(γ)) = ρA(α)
∨ ρA(γ)) = ρA(α)
∨ (ρA(β)
•
•
•
•
•
•
the commutativity of
•
∨, we also have (ρA(α)
•
∨ ρA(β))
•
∨ ρA(γ) = ρA(γ)
•
∨ (ρA(α)
•
∨ ρA(β)) =
θ, hence
•
•
•
•
•
•
∨ (ρA(β)
∨ is associative; ρA(α)
ρA(α)
ρA(α ∩ β) = ρA(ρA(α) ∨ ρA(α ∩ β)) = ρA(α ∨ (α ∩ β)) = ρA(α) and ρA(α) ∩ (ρA(α)
∨ ρA(γ)) = (ρA(α)
∨ ρA(γ), so
∨ ρA(β))
∨ (ρA(α) ∩ ρA(β)) = ρA(α)
∨ ρA(β)) = ρA(α ∩ (ρA(α)
•
•
•
∨
∨
θ∈R
•_
θ∈R
•
ρA(β))) = ρA(α∩ρA(ρA(α)∨ρA(β))) = ρA(α∩ρA(α∨β)) = ρA(ρA(α∩ρA(α∨β))) = ρA(ρA(α))∩ρA(ρA(α∨β))) =
ρA(α) ∩ ρA(α ∨ β)) = ρA(α ∩ (α ∨ β)) = ρA(α), so the absorption laws hold. Of course, for all θ, ζ ∈ RCon(A),
θ ∩ ζ = θ iff θ ⊆ ζ. Therefore (RCon(A),
∨,∩) is a lattice, orderred by set inclusion. From Remark 5.1, (iii) and
(i), we obtain that this lattice has ρA(∆A) as first element and ρA(∇A) = ∇A as last element.
Now let us consider a family (αi)i∈I ⊆ Con(A), M = {ρA(αi) i ∈ I} ⊆ RCon(A) and let us denote by
•_
αi), by Proposition 5.4, (v). Then θ ∈ RCon(A) and ρA(αi) ⊆ θ for
ρA(αi) ⊆ ζ, so, by Remark 5.1, (iii), and
all i ∈ I. Now, if ζ ∈ RCon(A) and ρA(αi) ⊆ ζ for all i ∈ I, then _
Proposition 5.4, (v), ζ = ρA(ζ) ⊇ ρA(_
ρA(αi) = θ. Therefore θ = sup(M ) in the bounded lattice
ρA(αi)) = ρA(_
ρA(αi) = ρA(_
ρA(αi)) =
•_
θ =
i∈I
i∈I
i∈I
i∈I
i∈I
RCon(A), hence this lattice is complete.
i∈I
Let us define a binary relation ≡A on Con(A) by: α ≡A β iff ρA(α) = ρA(β), for any α, β ∈ Con(A).
≡A ∩(K(A))2 shall also be denoted by ≡A.
Remark 5.9. Clearly, ≡A is an equivalence on Con(A), thus also on K(A). On RCon(A), ≡A coincides to the
equality, that is to ∆RCon(A), because, for any α, β ∈ Con(A), ρA(α) ≡A ρA(β) iff ρA(ρA(α)) = ρA(ρA(β)) iff
ρA(α) = ρA(β). So, trivially, ≡A is a congruence of the lattice RCon(A).
•W over arbitrary families of congruences,
∨, even W and
in particular it is a congruence of the lattice Con(A). Indeed, if α, α′, β, β′ ∈ Con(A) such that α ≡A α′ and
β ≡A β′, that is ρA(α) = ρA(α′) and ρA(β) = ρA(β′), then, by Proposition 5.4, (ii), ρA([α, β]A) = ρA(α ∩ β) =
ρA(α) ∩ ρA(β) = ρA(α′) ∩ ρA(β′) = ρA(α′ ∩ β′) = ρA([α′, β′]A), thus [α, β]A ≡A [α′, β′]A ≡A α ∩ β ≡A α′ ∩ β′.
Now, if (αi)i∈I ⊆ Con(A) and (α′
i), then,
by Proposition 5.4, (iv) and (v), ρA(
On Con(A), ≡A preserves the commutator, ∩, ∨ and
i)i∈I ⊆ Con(A) such that, for all i ∈ I, αi ≡A α′
i, that is ρA(αi) = ρA(α′
ρA(αi)) = ρA(_
αi) = ρA(ρA(_
ρA(α′
i)) =
•_
•
i∈I
ρA(_
i) = ρA(ρA(_
α′
α′
i)) = ρA(
•_
i∈I
i∈I
i∈I
i∈I
α′
i), hence
•_
i∈I
αi)) = ρA(_
•_
i ≡A _
α′
i∈I
αi) = ρA(_
αi ≡A _
i∈I
α′
i.
i∈I
i∈I
i∈I
αi ≡A
i∈I
Moreover, as shown by Proposition 5.4, (ii), (iv) and (v), just as in the calculations above, for all α, β ∈
Con(A) and all (αi)i∈I ⊆ Con(A), [α, β]A ≡A α ∩ β and
•_
i∈I
αi ≡A _
i∈I
αi.
Note, also, that, for all α ∈ Con(A), α ≡A ρA(α), by Proposition 5.4, (iv).
For all α ∈ Con(A), let us denote by bα the equivalence class of α with respect to ≡A, and let L(A) =
K(A)/≡A = {bθ θ ∈ K(A)}. Let λA : Con(A) → Con(A)/≡A be the canonical surjection: λA(θ) = bθ for all
θ ∈ Con(A); we denote in the same way its restriction to K(A), with its co -- domain restricted to L(A), that is
the canonical surjection λA : K(A) → L(A). Let us define the following operations on Con(A), where the second
equalities follow from Remark 5.9, as does the fact that these operations are well defined:
• for all α, β ∈ Con(A), bα ∨ bβ = \α ∨ β =
• 0 = d∆A = \ρA(∆A) and 1 = d∇A = \ρA(∇A).
•
\
α
∨ β and bα ∧ bβ = \α ∩ β = \[α, β]A;
Remark 5.10. By Proposition 5.4, (i), if C is congruence -- modular and semi -- degenerate, then, for any α ∈
Con(A), bα = 1 iff α = ∇A.
Lemma 5.11. (Con(A)/≡A,∨,∧, 0, 1) is a bounded distributive lattice and λA : Con(A) → Con(A)/≡A is a
bounded lattice morphism. Moreover, Con(A)/≡A is a complete lattice, in which _
i∈I bαi =
\\
αi for any family (αi)i∈I ⊆ Con(A), and the meet is completely distributive with respect to the join, thus
Con(A)/≡A is a frame.
αi and ^
i∈I bαi =
\_
i∈I
i∈I
Proof. By Remark 5.9, ≡A is a congruence of the bounded lattice Con(A), hence (Con(A)/≡A,∨,∧, 0, 1) is
a bounded lattice and the canonical surjection λA : Con(A) → Con(A)/≡A is a bounded lattice morphism,
in particular it is order -- preserving. It is straightforward, from the fact that the lattice Con(A) is complete
and the surjectivity of the lattice morphism λA, that the lattice Con(A)/≡A is complete and its joins and
meets of arbitrary families of elements have the form in the enunciation. By Proposition 3.6, for any families
(αi)i∈I and (βj )j∈J of congruences of A, (_
i∈I bαi) ∧ (_
βj]A)b =
(_
_
(bαi ∧ bβj), that is the meet is completely distributive with respect
to the join in Con(A)/≡A, thus Con(A)/≡A is a frame, in particular it is a bounded distributive lattice.
[αi, βj]A)b = _
j∈J bβj) = ((_
βj))b = ([_
\[αi, βj]A = _
αi) ∩ (_
αi, _
_
i∈I
j∈J
i∈I
j∈J
_
i∈I
j∈J
i∈I
j∈J
i∈I
j∈J
We shall denote by ≤ the partial order of the lattice Con(A)/≡A.
Proposition 5.12. (RCon(A),
•
∨,∩, ρA(∆A), ρA(∇A) = ∇A) is a frame, isomorphic to Con(A)/≡A.
Proof. Let ϕ : Con(A)/≡A → RCon(A), for all α ∈ Con(A), ϕ(bα) = ρA(α). If α, β ∈ Con(A), then the following
equivalences hold: bα = bβ iff α ≡A β iff ρA(α) = ρA(β) iff ϕ(bα) = ϕ(bβ), hence ϕ is well defined and injective.
By Remark 5.6, ϕ is surjective. By Proposition 5.4, (ii) and (v), for all α, β ∈ Con(A), ϕ(bα ∧ bβ) = ϕ(\α ∩ β) =
ρA(α ∩ β) = ρA(α) ∩ ρA(β) and ϕ(bα ∨ bβ) = ϕ(\α ∨ β) = ρA(α ∨ β) = ρA(α) ∨ ρA(β) (actually, Proposition 5.4,
(v), and Lemma 5.11 show that ϕ preserves arbitrary joins). Therefore ϕ is a lattice isomorphism, thus an order
isomorphism, hence it preserves arbitrary joins and meets. From this and Lemma 5.11 we obtain that RCon(A)
is a frame and ϕ is a frame isomorphism.
Throughout the rest of this section, we shall assume that K(A) is closed with respect to the commutator.
Proposition 5.13. L(A) is a bounded sublattice of Con(A)/≡A, thus it is a bounded distributive lattice.
Proof. Since ∇A ∈ K(A), we have 1 = d∇A ∈ L(A). By Remark 3.18, ∆A ∈ K(A), thus 0 = d∆A ∈ L(A).
If K(A) is closed with respect to the commutator, then, for each α, β ∈ K(A), we have [α, β]A ∈ K(A), thus
bα ∧ bβ = \[α, β]A ∈ L(A) Again by Remark 3.18, for each α, β ∈ K(A), bα ∨ bβ = \α ∨ β ∈ L(A). Hence L(A) is a
bounded sublattice of Con(A)/≡A, which is distributive by Lemma 5.11, thus L(A) is a bounded distributive
lattice.
For any θ ∈ Con(A) and any I ∈ Id(L(A)), we shall denote by:
• θ∗ = {bα α ∈ K(A), α ⊆ θ} = λA(K(A) ∩ (θ]) ⊆ L(A), where (θ] = (θ]Con(A) ∈ PId(Con(A));
• I∗ = _{α ∈ K(A) bα ∈ I} = _
and d∆A = \ρA(∆A) = 0 ∈ I.
Lemma 5.14. For all θ ∈ Con(A):
α ∈ Con(A); note that λ−1
α∈λ−1
A (I)
A (I) is non -- empty, because ∆A ∈ K(A)
• θ∗ ⊆ (bθ ]Con(A)/≡A ∩ L(A) and θ∗ ∈ Id(L(A));
• if θ ∈ K(A), then θ∗ = (bθ ]Con(A)/≡A ∩ L(A) = (bθ ]L(A) ∈ PId(L(A)).
Proof. Let θ ∈ Con(A), and, in this proof, let us denote by hbθi = (bθ ]Con(A)/≡A
and, in the case when θ ∈ K(A),
by (bθ ] = (bθ ]L(A). θ∗ = {bα α ∈ (θ] ∩ K(A)}.
For all α ∈ (θ] ∩ K(A), we have bα ∈ L(A) and α ⊆ θ, thus bα ≤ bθ in Con(A)/≡A, hence bα ∈ hbθi ∩ L(A),
therefore θ∗ ⊆ hbθi ∩ L(A). ∆A ∈ K(A) and ∆A ⊆ θ, thus d∆A ∈ θ∗, so θ∗ is non -- empty. Since K(A) is closed
[·,·]A, α ∨ β, [α, β]A ∈ K(A) for any α, β ∈ K(A). Let x, y ∈ θ∗, which means that x = bα and y = bβ
for some α, β ∈ K(A) ∩ (θ]. Then α ∨ β ∈ K(A) ∩ (θ], thus x ∨ y = bα ∨ bβ = \α ∨ β ∈ θ∗. Now let x ∈ θ∗
and y ∈ L(A) such that x ≥ y, so that y = x ∧ y. Then x = bα for some α ∈ K(A) ∩ (θ] and y = bβ for
w.r.t.
some β ∈ K(A). Thus [α, β]A ∈ K(A) and [α, β]A ⊆ α ∩ β ⊆ α ⊆ θ, hence [α, β]A ∈ K(A) ∩ (θ], therefore
y = x ∧ y = bα ∧ bβ = \[α, β]A ∈ θ∗. Hence θ∗ ∈ Id(L(A)).
Now assume that θ ∈ K(A), so that bθ ∈ L(A). By the above, θ∗ ⊆ hbθi ∩ L(A) = (bθ ]. Let x ∈ (bθ ], so
that there exists an α ∈ K(A) with bα = x ≤ bθ, thus \[α, θ]A = bα ∩ bθ = bα = x. But [α, θ]A ∈ K(A) ∩ (θ], so
x = \[α, θ]A ∈ θ∗. Therefore we also have (bθ ] ⊆ θ∗, hence θ∗ = (bθ ] ∈ PId(L(A)).
By the above, we have two functions:
• θ ∈ Con(A) 7→ θ∗ ∈ Id(L(A));
• I ∈ Id(L(A)) 7→ I∗ ∈ Con(A).
A (I) ⊆ λ−1
A (J), thus I∗ ⊆ J∗.
Lemma 5.15. The two functions above are order -- preserving.
Proof. For any θ, ζ ∈ Con(A) such that θ ⊆ ζ, we have (θ] ⊆ (ζ], hence θ∗ ⊆ ζ∗. For any I, J ∈ Id(L(A)) such
that I ⊆ J, we have λ−1
Lemma 5.16. Let α ∈ K(A) and I ∈ Id(L(A)). Then: α ⊆ I∗ iff bα ∈ I.
Proof. "⇐:" If bα ∈ I, then α ∈ λ−1
"⇒:" If α ⊆ I∗ = _{β ∈ K(A) bβ ∈ I}, then, since α ∈ K(A), it follows that there exist an n ∈ N∗ and
β1, . . . , βn ∈ K(A) such that cβ1, . . . , cβn ∈ I and α ⊆
(i) For any θ ∈ Con(A), θ ⊆ (θ∗)∗.
n_
i=1 bβi ∈ I, thus bα ∈ I.
A (I), thus α ⊆ I∗.
βi, hence bα ⊆
Lemma 5.17.
n_
i=1
\n_
βi =
i=1
(ii) For any I ∈ Id(L(A)), I = (I∗)∗.
Proof. (i) Let θ ∈ Con(A). For any (a, b) ∈ θ, CgA(a, b) ∈ PCon(A) ⊆ K(A) and CgA(a, b) ⊆ θ, thus CgA(a, b) ∈
K(A) ∩ (θ], hence \CgA(a, b) ∈ θ∗, therefore CgA(a, b) ⊆ (θ∗)∗ by Lemmas 5.14 and 5.16, so (a, b) ∈ (θ∗)∗. Hence
θ ⊆ (θ∗)∗.
(ii) For any x ∈ L(A), by Lemma 5.16, the following equivalences hold: x ∈ (I∗)∗ iff there exists an α ∈ K(A)
such that α ⊆ I∗ and x = bα iff there exists an α ∈ K(A) such that bα ∈ I and x = bα iff x ∈ I. Therefore
(I∗)∗ = I.
Proposition 5.18.
(i) The map I ∈ Id(L(A)) 7→ I∗ ∈ Con(A) is injective.
(ii) The map θ ∈ Con(A) 7→ θ∗ ∈ Id(L(A)) is surjective.
Proof. (i) Let I, J ∈ Id(L(A)) such that I∗ = J∗. Then (I∗)∗ = (J∗)∗, so I = J by Lemma 5.17, (ii).
(ii) Let I ∈ Id(L(A)), and denote θ = I∗ ∈ Con(A). Then θ∗ = (I∗)∗ = I by Lemma 5.17, (ii).
Lemma 5.19. For any φ ∈ Spec(A), φ = (φ∗)∗.
Proof. Let φ ∈ Spec(A). Then φ ⊆ (φ∗)∗ by Lemma 5.17, (i).
Now let β ∈ K(A) such that bβ ∈ φ∗ = {bα α ∈ K(A), α ⊆ φ}, which means that bβ = bα for some α ∈ K(A) with
α ⊆ φ. Since bβ = bα, we have ρA(β) = ρA(α), while α ⊆ φ gives us ρA(α) ⊆ ρA(φ) = φ, where the last equality
follows from the fact that φ ∈ Spec(A). Hence β ⊆ ρA(β) ⊆ φ. Therefore (φ∗)∗ = _{γ ∈ K(A) bγ ∈ φ∗} ⊆ φ.
Lemma 5.20. For any φ ∈ Spec(A), we have φ∗ ∈ SpecId(L(A)).
Proof. Let φ ∈ Spec(A). Then φ∗ ∈ Id(L(A)) = Id(K(A)/≡A ). Let α, β ∈ K(A) such that \[α, β]A = bα∧bβ ∈ φ∗ =
{bγ γ ∈ K(A), γ ⊆ φ}. Then there exists a γ ∈ K(A) such that γ ⊆ φ and bγ = \[α, β]A, thus ρA(γ) = ρA([α, β]A)
and ρA(γ) ⊆ ρA(φ) = φ since φ ∈ Spec(A). Hence [α, β]A ⊆ ρA([α, β]A) ⊆ φ, hence α ⊆ φ or β ⊆ φ since
φ ∈ Spec(A). But this means that bα ∈ φ∗ or bβ ∈ φ∗. Therefore φ∗ ∈ SpecId(L(A)).
Lemma 5.21. For any P ∈ SpecId(L(A)), we have P∗ ∈ Spec(A).
Hence φ = (φ∗)∗.
Proof. Let P ∈ SpecId(L(A)). Then P∗ ∈ Con(A). Let α, β ∈ PCon(A) such that [α, β]A ⊆ P∗. Then
α, β ∈ K(A), so that [α, β]A ∈ K(A), and [α, β]A ⊆ _{γ ∈ K(A) bγ ∈ P}, hence there exist an n ∈ N∗
n_
i=1 bγi ∈ P , hence
and γ1, . . . , γn ∈ K(A) such that bγ1, . . . ,cγn ∈ P and [α, β]A ⊆
bα ∧ bβ = \[α, β]A ∈ P , thus bα ∈ P or bβ ∈ P since P ∈ SpecId(L(A)). By Lemma 5.16, it follows that α ⊆ P∗ or
β ⊆ P∗. Therefore P∗ ∈ Spec(A).
γi. But then
[n_
n_
i=1
γi =
i=1
By Lemmas 5.20 and 5.21, we have these restrictions of the functions defined above:
• u : Spec(A) → SpecId(L(A)), for all φ ∈ Spec(A), u(φ) = φ∗;
• v : SpecId(L(A)) → Spec(A), for all P ∈ SpecId(L(A)), v(P ) = P∗.
Proposition 5.22. u and v are homeomorphisms, inverses of each other, between the prime spectrum of A and
the prime spectrum of ideals of L(A), endowed with the Stone topologies.
Proof. By Lemma 5.17, (ii), for all P ∈ SpecId(L(A)), we have u(v(P )) = P . By Lemma 5.19, for all φ ∈ Spec(A),
we have v(u(φ)) = φ. Thus u and v are bijections and they are inverses of each other.
Let θ ∈ Con(A) and φ ∈ VA(θ), that is φ ∈ Spec(A) and θ ⊆ φ. Then, by Lemmas 5.21 and 5.15,
φ∗ ∈ SpecId(L(A)) and θ∗ ⊆ φ∗, so φ∗ ∈ VId,L(A)(θ∗), and we have u(φ) = φ∗. Hence u(VA(θ)) ⊆ VId,L(A)(θ∗).
Now let P ∈ VId,L(A)(θ∗), that is P ∈ SpecId(L(A)) and θ∗ ⊆ P . Then, by Lemma 5.17, (i), and Lemmas
5.15 and 5.21, θ ⊆ (θ∗)∗ ⊆ P∗ ∈ Spec(A), thus P∗ ∈ VA(θ), and we have u(P∗) = u(v(P )) = P . Hence
VId,L(A)(θ∗) ⊆ u(VA(θ)). Therefore u(VA(θ)) = VId,L(A)(θ∗), thus u is closed, hence u is open, so v is continuous.
Now let I ∈ Id(L(A)). Then, according to Proposition 5.18, (ii), I = θ∗ for some θ ∈ Con(A). By the above,
u(VA(θ)) = VId,L(A)(θ∗) = VId,L(A)(I), hence v(VId,L(A)(I)) = v(u(VA(θ))) = VA(θ), therefore v is closed, hence
v is open, thus u is continuous.
Hence u and v are homeomorphisms.
Corollary 5.23 (existence of the reticulation). L(A) is a reticulation for the algebra A.
Proposition 5.24. [5],[26] If L and M are bounded distributive lattices whose prime spectra of ideals, endowed
with the Stone topologies, are homeomorphic, then L and M are isomorphic.
Corollary 5.25 (uniqueness of the reticulation). The reticulation of A is unique up to a lattice isomorphism.
Corollary 5.26. If C is congruence -- modular and semi -- degenerate, then u and v induce homeomorphisms,
inverses of each other, between the maximal spectrum of A and the maximal spectrum of ideals of L(A), endowed
with the Stone topologies.
A ⊆ θ)}, so ρA(∆A) = {(a, b) ∈ A2 (∃ n ∈ N∗) ([CgA(a, b), CgA(a, b)]n
Proof. By Lemma 3.12, Proposition 5.22 and the fact that, as Lemma 5.15 ensures us, u and v are order --
preserving, and hence they are order isomorphisms between the posets (Spec(A),⊆) and (SpecId(L(A)),⊆).
Proposition 5.27. [1, Proposition 4.1] For any θ ∈ Con(A), ρA(θ) = {(a, b) ∈ A2 (∃ n ∈ N∗) ([CgA(a, b),
CgA(a, b)]n
Proposition 5.28. For any θ ∈ Con(A), (θ∗)∗ = ρA(θ).
Proof. For every β ∈ K(A) such that bβ ∈ θ∗ = {bγ γ ∈ K(A), γ ⊆ θ}, there exists an α ∈ K(A) such that
α ⊆ θ and bα = bβ, thus β ⊆ ρA(β) = ρA(α) ⊆ ρA(θ). Therefore (θ∗)∗ = _{γ ∈ K(A) bγ ∈ θ∗} ⊆ ρA(θ). Now
let (a, b) ∈ ρA(θ), so that, according to Proposition 5.27, Lemma 5.17, (i), and Lemma 5.16, for some n ∈ N∗,
[CgA(a, b), CgA(a, b)]n
A) =
ρA(CgA(a, b)), thus \CgA(a, b) = ([CgA(a, b), CgA(a, b)]n
A)b ∈ θ∗, hence (a, b) ∈ CgA(a, b) ⊆ (θ∗)∗ by Lemma
5.16. Therefore ρA(θ) ⊆ (θ∗)∗. Hence (θ∗)∗ = ρA(θ).
A)b ∈ θ∗. But ρA([CgA(a, b), CgA(a, b)]n
A ⊆ θ ⊆ (θ∗)∗, hence ([CgA(a, b), CgA(a, b)]n
A = ∆A)}.
Corollary 5.29.
(i) For all θ ∈ Con(A), ρA(θ)∗ = θ∗.
(ii) For all I ∈ Id(L(A)), ρA(I∗) = I∗.
Proof. (i) By Lemma 5.17, (ii), and Proposition 5.28, θ∗ = ((θ∗)∗)∗ = ρA(θ)∗.
(ii) By Proposition 5.28 and Lemma 5.17, (ii), we have ρA(I∗) = ((I∗)∗)∗ = I∗.
Corollary 5.30. The maps:
• θ ∈ RCon(A) 7→ θ∗ ∈ Id(L(A)),
• I ∈ Id(L(A)) 7→ I∗ ∈ RCon(A)
are frame isomorphisms and inverses of each other.
Proof. By Corollary 5.29, (ii), for all I ∈ Id(L(A)), we have I∗ ∈ RCon(A), hence the second map above is
well defined. By Lemma 5.17, (ii), for all I ∈ Id(L(A)), (I∗)∗ = I. By Proposition 5.28, for all θ ∈ RCon(A),
θ = ρA(θ) = (θ∗)∗. Hence these functions are inverses of each other, thus they are bijections. By Lemma 5.15,
these maps are order -- preserving, thus they are order isomorphisms, hence they preserve arbitrary joins and
meets, therefore they are frame isomorphisms.
6 Some Examples, Particular Cases and Preservation of Finite Di-
rect Products
A = ∆A for some n ∈ N∗; nilpotent iff (∇A,∇A]n
Throughout this section, we shall assume that [·,·]A is commutative and distributive w.r.t. arbitrary joins and
∇A ∈ K(A). These hypotheses are sufficient for the following results we cite from other works to hold. We
shall denote by HSP(A) the variety generated by A. In the following examples, we determine the prime spectra
by using [1, Proposition 1.2], which says that, for each proper congruence φ of A: φ is prime iff φ is meet --
irreducible and semiprime. So, if we know that HSP(A) is congruence -- modular, then we only have to calculate
[α, α]A for every α ∈ Con(A). The complete tables of the commutators for the following algebras show that
their commutators are commutative and distributive w.r.t. the join. Of course, since each of the algebras M
from the following examples is finite, we have ∇M ∈ K(M ). We have used the method in [40] to calculate the
commutators, excepting those in groups, where we have used the commutators on normal subgroups; recall that
the variety of groups is congruence -- modular [39]. Following [1], we say that A is: Abelian iff [∇A,∇A]A = ∆A;
solvable iff [∇A,∇A]n
A = ∆A for some n ∈ N∗. For any n ∈ N∗,
we shall denote by Ln the n -- element chain. By ⊕ we shall denote the ordinal sum of bounded lattices.
Remark 6.1. If Spec(A) = ∅, then ρA(α) = ∇A for all α ∈ Con(A), hence ≡A= ∇K(A), thus L(A) =
K(A)/∇K(A) ∼= L1. If Spec(A) = {φ} for some φ ∈ Con(A) \ {∇A}, then: ρA(θ) = φ = ρA(∆A) for all θ ∈ (φ],
and ρA(θ) = ∇A = ρA(∇A) for all θ ∈ Con(A) \ (φ], therefore, since ∆A,∇A ∈ K(A), L(A) = K(A)/ ≡A∼= L2.
Obviously, if A is Abelian, then A is nilpotent and solvable and Spec(A) = ∅. Moreover, by [1, Proposition
1.3], if A is solvable or nilpotent, then Spec(A) = ∅. Thus, if A is solvable or nilpotent, in particular if A is
Abelian, then L(A) ∼= L1. For instance, according to [39], any Abelian group is an Abelian algebra, hence its
reticulation is trivial.
If A is simple, that is Con(A) = {∆A,∇A} ⊆ K(A) ⊆ Con(A), so that K(A) = Con(A) = {∆A,∇A}, thus
L(A) = {0, 1}, so we are situated in one of the following two cases: either A is Abelian, so that L(A) ∼= L1, or
the commutator of A equals the intersection, so that Spec(A) = {∆A} and thus L(A) ∼= L2.
Proposition 6.2. If the commutator of A equals the intersection, in particular if C is congruence -- distributive,
then K(A) is a bounded sublattice of the bounded distributive lattice Con(A) and λA : K(A) → L(A) is a lattice
isomorphism, thus we may take L(A) = K(A).
Proof. Assume that [·,·]A = ∩. ∆A ∈ PCon(A) ⊆ K(A). By Remark 3.18, K(A) is closed w.r.t. the join, and we
are under the assumptions that ∇A ∈ K(A) and K(A) is closed w.r.t. the commutator, so w.r.t. the intersection.
Hence K(A) is a bounded sublattice of Con(A). By Proposition 5.7, ≡A= ∆K(A), thus L(A) = K(A)/∆K(A) ∼=
K(A) and the canonical surjection λA : K(A) → L(A) is a lattice isomorphism.
Remark 6.3. If Con(A) = K(A), in particular if A is finite, then L(A) = Con(A)/ ≡A, so, if, furthermore, the
commutator of A equals the intersection, in particular if C is congruence -- distributive, then L(A) ∼= Con(A) by
Proposition 6.2, thus we may take L(A) = Con(A).
As a fact that may be interesting by its symmetry, if A is finite and its commutator equals the intersection,
so that Con(A) is a finite distributive lattice, then L(Con(A)) = Con(Con(A)) = Con(L(A)). It might also be
interesting to find weaker conditions on A under which L(Con(A)) ∼= Con(L(A)).
Remark 6.4. By Proposition 6.2, if A is a residuated lattice, then L(A) = K(A). If we denote by Filt(A) the
set of the filters of A and by PFilt(A) the set of the principal filters of A, then, since Con(A) ∼= Filt(A) and
the finitely generated filters of A are principal filters [22], [28], it follows that L(A) = K(A) ∼= PFilt(A), which
is the dual of the reticulation of a residuated lattice obtained in [41], [42], [43], where the reticulation has the
prime spectrum of filters homeomorphic to the prime spectrum of filters, thus to that of congruences of A by
the above, so this duality to the construction of L(A) from Section 5 was to be expected.
Remark 6.5. If A is a commutative unitary ring and Id(A) is its lattice of ideals, then it is well known that
Id(A) ∼= Con(A). If, for all I ∈ Id(A), we denote by √I the intersection of the prime filters of A which include
I, then [7, Lemma, p. 1861] shows that, for any J ∈ Id(A), there exists a finitely generated ideal K of A such
that √J = √K. From this, it immediately follows that the lattice L(A) is isomorphic to the reticulation of A
constructed in [7].
Remark 6.6. Let n, k ∈ N∗ and assume that C is congruence -- modular, S is a subalgebra of A, α, β ∈ Con(A),
M1, . . . , Mn are algebras from C, M =
Mi and, for all i ∈ 1, n, αi, βi ∈ Con(Mi).
nY
i=1
From Lemma 3.8, it is immediate that [α ∩ S2, β ∩ S2]k
A ∩ S2 and (α ∩ S2, β ∩ S2]k
S ⊆ (α, β]k
A ∩ S2.
Hence, if A is Abelian or solvable or nilpotent, then S is Abelian or solvable or nilpotent, respectively.
S ⊆ [α, β]k
nY
M =
βi]k
i=1
nY
i=1
nY
i=1
[αi, βi]k
Mi and (
αi,
βi]k
M =
(αi, βi]k
Mi .
nY
i=1
nY
i=1
nY
i=1
From Proposition 3.9, it is immediate that [
αi,
From this, it is easy to prove that: M is Abelian or solvable or nilpotent iff M1, . . . , Mn are Abelian or solvable
or nilpotent, respectively.
Example 6.7. For any group (G,·), any x ∈ G and any normal subgroup H of G, let us denote by hxi the
subgroup of G generated by x and by ≡H the congruence of G associated to H: ≡H = {(y, z) ∈ G2 yz−1 ∈ H}.
As shown by the following commutators calculations, the cuaternions group, C8 = {1,−1, i,−i, j,
−j, k,−k}, is a solvable algebra which is not Abelian, while the group S3 = {1, t, u, v, c, d} of the permuta-
tions of the set 1, 3, where 1 = id1,3, t = (1 2), u = (1 3), v = (2 3), c = (1 2 3) and d = c ◦ c, has Spec(S3) = ∅,
without being solvable or nilpotent. The following are the subgroups of C8, respectively S3, all of which are
normal, and the proper ones are cyclic, thus Abelian: h1i, h−1i, hii, hji, hki and C8, respectively h1i, hti, hui,
hvi, hci and S3, so C8 and S3 have the following congruence lattices and commutators, which suffice to conclude
that Spec(C8) = Spec(S3) = ∅, since we are in a congruence -- modular variety, and thus L(C8) ∼= L(S3) ∼= L1, by
Remark 6.1:
≡hii
r
❅❅
r
∇C8 =≡C8
❅
❅❅
≡hji ≡hki
≡h−1i
❅
r
r
r
r
∆C8 =≡h1i
r
r
✑✑
◗◗
∇S3 =≡S3
◗
✑
✁
❆
✑
✁
✑
❆
✁
≡hti ≡hui
❆
◗
✁
❆
◗
◗
✑
✁
❆
∆S3 =≡h1i
◗
◗
❆
≡hvi ≡hci
✁
✑
✑
◗◗
✑✑
r
r
r
r
θ
[θ, θ]C8
∆C8
≡h−1i
≡hii
≡hji
≡hki
∇C8
∆C8
∆C8
≡h−1i
≡h−1i
≡h−1i
≡h−1i
θ
∆S3
≡hti
≡hui
≡hvi
≡hci
∇S3
[θ, θ]S3
∆S3
∆S3
∆S3
∆S3
∆S3
∇S3
Notice, also, that C8 is solvable, as we have announced, thus, according to Remark 6.6, so is any finite direct
product whose factors are subgroups of C8, which, of course, is Abelian if all those subgroups are proper.
Example 6.8. This is the algebra from [2, Example 6.3] and [3, Example 4.2]: U = ({0, a, b, c, d}, +), with + de-
fined by the following table, which has the congruence lattice represented below, where U/α = {{0, a},{b, c, d}},
U/β = {{0, b},{a, c, d}}, U/γ = {{0, c, d},{a, b}} and U/δ = {{0},{a},{b},{c, d}}:
β
α
δ
+ 0
0
a
a
b
x
b
c
y
z
d
a
a
0
c
b
b
b
b
c
0
a
a
c
c
b
a
0
0
d
d
b
a
0
0
α
r
❅
❅
r
γ
r
∇U
❅
β
δ
❅
r
r
r
∆U
γ
[·,·]U ∆U
∇U
∆U ∆U ∆U ∆U ∆U ∆U
∆U
α
α
∆U
β
∆U
β
γ
∆U
γ
δ
∆U
δ
∇U
∇U
∆U
δ
δ
δ
∆U
δ
δ
δ
δ
δ
α
δ
δ
δ
δ
β
δ
δ
γ
δ
γ
U is not Abelian, nor is it solvable or nilpotent, as shown by the table of [·,·]U above, but Spec(A) = ∅, thus
L(U ) ∼= L1 by Remark 6.1.
Example 6.9. Let M = ({a, b, x, y, z}, +) and N = ({a, b, c, x, y}, +), with + defined by the following tables.
Then Con(M ) and Con(N ) have the Hasse diagrams below, where:
• M/α = {{a, b},{x, y, z}}, M/β = {{a, b},{x, y},{z}}, M/γ = {{a, b},{x, z},{y}}, M/δ = {{a, b},{x},
{y, z}} and M/ε = {{a, b},{x},{y},{z}};
• N/χ = {{a, b, c},{x, y}}, N/χ1 = {{a, b, c},{x},{y}}, N/ξ = {{a, b},{c},{x, y}}, N/ξ1 = {{a, b},{c},{x},{y}},
N/ψ = {{a},{b, c},{x, y}}, N/ψ1 = {{a},{b, c},{x},{y}} and N/φ = {{a},{b},{c},{x, y}}.
+ a
a
a
b
b
x
x
y
y
z
z
b
b
b
x
y
z
x
a
b
x
y
z
y
a
b
x
y
z
z
a
b
x
y
z
β
δ
+ a
a
a
b
b
c
c
x
x
y
y
b
b
b
c
x
y
c
c
c
c
x
y
x
a
b
c
x
y
y
a
b
c
x
y
r
❅
❅
r
r
r
∇M
α
❅
γ
ε
r
r
❅
r
∆M
r
r
r
∇N
χ
❅
χ1
❅
φ
∆N
r
r
❅
❅
r
ξ
❅
❅r
ξ1
ψ
r
❅
❅
r
ψ1
Note that, despite the fact that M is congruence -- modular and N is congruence -- distributive, neither HSP(M ),
nor HSP(N ) is congruence -- modular, because S = ({a, b}, +) ∼= (L2, max) ∼= (Z2,·) is a subalgebra of both M
and N , and it can be easily checked that S2 is not congruence -- modular. Thus neither HSP(M ), nor HSP(N )
is semidegenerate, which is also obvious from the fact that ({a}, +) is a subalgebra of both M and N .
We have: [θ, ζ]M = ε for all θ, ζ ∈ [ε) and, of course, [∆M , θ]M = [θ, ∆M ]M = ∆M for all θ ∈ Con(M ), hence
Spec(M ) = {∆M} and thus L(M ) ∼= L2, while [·,·]N is given by the following table, thus Spec(N ) = {ψ, ξ}, so
ρN is defined as follows and hence L(M ) ∼= L2
2:
ξ1
ρN (θ)
ψ1
ψ
χ
φ
ξ
ψ1
ψ1
ψ1 ∆N ∆N ∆N
ψ1 ∆N ∆N ∆N
χ1 ∇N
[·,·]N ∆N
∆N ∆N ∆N ∆N ∆N ∆N ∆N ∆N ∆N
∆N
ψ1
ψ1
∆N
ψ
∆N
ψ1
ψ1
ψ1
∆N ∆N ∆N ∆N ∆N ∆N ∆N ∆N ∆N
φ
ξ
ξ1
∆N ∆N ∆N ∆N
∆N ∆N ∆N ∆N
ξ1
ξ1
χ1
ψ1 ∆N
∆N
χ
χ1
ψ1 ∆N
∆N
χ1
∇N
∆N
ψ1 ∆N
χ1
ξ1
ξ1
χ1
χ1
χ1
ξ1
ξ1
χ1
χ1
χ1
ξ1
ξ1
ξ1
ξ1
ξ1
ξ1
ξ1
ξ1
ξ1
ξ1
ψ1
ψ1
ψ1
ψ1
ψ1
θ
∆N
ψ
ψ1
φ
ξ
ξ1
χ
χ1
∇N
φ
ψ
ψ
φ
ξ
ξ
∇N
∇N
∇N
bξ
r
❅
❅
r
bψ
1
❅
r
r
❅
0
L(N )
Example 6.10. Here are some finite congruence -- distributive examples, thus in which the reticulations are
isomorphic to the congruence lattices. Regarding the preservation properties fulfilled by the reticulation, these
examples show that there is no embedding relation between the reticulation of an algebra and those of its
if E is the following bounded lattice, then, for instance, {0, x, y, 1} = L4 = L2 ⊕ L2 ⊕ L2, D =
subalgebras:
{0, a, x, b, 1} and P = {0, a, x, y, 1} are bounded sublattices of E. We have: L(E) ∼= Con(E) = {∆E , µ,∇E} ∼= L3,
where E/µ = {{0},{a},{x, y},{b},{1}}, L(L4) ∼= Con(L4) = Con(L2 ⊕ L2 ⊕ L2) ∼= Con(L2)3 ∼= L3
2, L(D) ∼=
Con(D) = {∆D,∇D} ∼= L2 and L(P) ∼= Con(P) = {∆P , α, β, γ,∇P} ∼= L2⊕L2
2, where P/α = {{0, x, y},{a, 1}},
P/β = {{0, a},{x, y, 1}} and P/γ = {{0},{a},{x, y},{1}}:
∇E
µ
β
a
b
r
r
r
r
r
r
r
∇P
❅α
❅
γ
r
r
E :
❅
r
1
❅
y
x
❅
0
r
r
❅
r
∆E
r
∆P
Remark 6.11. By [24, Lemma 3.3], in any variety, arbitrary intersections commute with arbitrary direct
products of congruences. If C is congruence -- modular and M is an algebra from C such that A × M has no skew
congruences, then Spec(A × M ) = {φ × ∇M φ ∈ Spec(A)} ∪ {∇A × ψ ψ ∈ Spec(M )}. This follows from
Proposition 3.9 in the same way as in the congruence -- distributive case, treated in [24, Proposition 3.5,(ii)].
Proposition 6.12 (the reticulation preserves finite direct products without skew congruences). Let M be an
algebra from C such that the direct product A × M has no skew congruences. Then:
(i) for all ∅ 6= X ⊆ A2 and all ∅ 6= Y ⊆ M 2, CgA×M (X×Y ) = CgA(X)×CgM (Y ), and the map (α, µ) 7→ α×µ
is a lattice isomorphism from Con(A) × Con(M ) to Con(A × M );
(ii) PCon(A × M ) = {α × µ α ∈ PCon(A), µ ∈ PCon(M )} and K(A × M ) = {α × µ α ∈ K(A), µ ∈ K(M )}.
If C is congruence -- modular and ∇M ∈ K(M ), then:
• for all α ∈ Con(A) and all µ ∈ Con(M ), ρA×M (α × µ) = ρA(α) × ρM (µ);
• ≡A×M =≡A × ≡M and L(A × M ) ∼= L(A) × L(M ).
Proof. A × M has no skew congruences, that is Con(A × M ) = {α × µ α ∈ Con(A), µ ∈ Con(M )}.
(i) By Remark 6.11, CgA×M (X × Y ) = T{θ ∈ Con(A × M ) X × Y ⊆ θ} = T{α × µ α ∈ Con(A), µ ∈
Con(M ), X × Y ⊆ α × µ} = T{α × µ α ∈ Con(A), µ ∈ Con(M ), X ⊆ α, Y ⊆ µ} = (T{α ∈ Con(A) X ⊆
α}) × (T{µ ∈ Con(M ) Y ⊆ µ}) = CgA(X) × CgM (Y ).
This also shows that the map (α, µ) 7→ α× µ is a lattice isomorphism from Con(A)× Con(M ) to Con(A× M ),
because it is clearly injective, it is surjective by the above, it preserves the intersection by Remark 6.11 and, for all
α, β ∈ Con(A) and all µ, ν ∈ Con(M ), (α×µ)∨(β×ν) = CgA×M ((α×µ)∪(β×ν)) ⊆ CgA×M ((α∪β)×(µ∪ν)) =
CgA(α ∪ β) × CgM (µ ∪ ν) = (α ∨ β) × (µ ∨ ν), since, clearly, (α × µ) ∪ (β × ν) ⊆ (α ∪ β) × (µ ∪ ν), but, also,
(α × µ) ∨ (β × ν) ∈ Con(A × M ), thus (α × µ) ∪ (β × ν) ⊆ (α × µ) ∨ (β × ν) = γ ∨ σ for some γ ∈ Con(A)
and σ ∈ Con(M ), so α × µ ⊆ γ ∨ σ and β × ν ⊆ γ ∨ σ, hence α ⊆ γ, β ⊆ γ, µ ⊆ σ and ν ⊆ σ, so
(α ∨ β) ⊆ γ and (µ ∨ ν) ⊆ σ, hence (α ∨ β) × (µ ∨ ν) ⊆ γ ∨ σ = (α × µ) ∨ (β × ν) ⊆ (α ∨ β) × (µ ∨ ν), therefore
(α ∨ β) × (µ ∨ ν) = (α × µ) ∨ (β × ν).
(ii) By (i), for all a, b ∈ A and all u, v ∈ M , CgA×M ((a, u), (b, v)) = CgA(a, b)× CgM (u, v), hence the expression
of PCon(A × M ) in the enunciation. From this and the second statement in (i), we obtain: K(A × M ) =
{CgA×M ({(a1, u1), . . . , (an, un)}) n ∈ N∗, a1, . . . , an ∈ A, u1, . . . , un ∈ M} = {
CgA×M (ai, ui) n ∈
n_
i=1
N∗, a1, . . . , an ∈ A, u1, . . . , un ∈ M} = {
(CgA(ai) × CgM (ui)) n ∈ N∗, a1, . . . , an ∈ A, u1, . . . , un ∈
n_
i=1
n_
n_
i=1
i=1
CgA(ai)) × (
CgM (ui)) n ∈ N∗, a1, . . . , an ∈ A, u1, . . . , un ∈ M} = {CgA(a1, . . . , an) ×
M} = {(
CgM (u1, . . . , un) n ∈ N∗, a1, . . . , an ∈ A, u1, . . . , un ∈ M} = {α × µ α ∈ K(A), µ ∈ K(M )}, since the
above also hold if some of the elements a1, . . . , an or u1, . . . , un coincide.
Now assume that C is congruence -- modular and ∇M ∈ K(M ). Then, by Remark 6.11, for any α ∈ Con(A) and
any µ ∈ Con(M ), ρA×M (α×µ) = T{χ ∈ Spec(A×M ) α×µ ⊆ χ} = T{φ×∇M φ ∈ Spec(A), α×µ ⊆ φ×∇M}∩
T{∇A× ψ ψ ∈ Spec(M ), α× µ ⊆ ∇A× ψ} = T{φ×∇M φ ∈ Spec(A), α ⊆ φ}∩T{∇A× ψ ψ ∈ Spec(M ), µ ⊆
ψ} = (T{φ φ ∈ Spec(A), α ⊆ φ}×∇M )∩(∇A×T{ψ ψ ∈ Spec(M ), µ ⊆ ψ}) = (ρA(α)×∇M )∩(∇A×ρM (µ)) =
(ρA(α) ∩ ∇A) × (∇M ∩ ρM (µ)) = ρA(α) × ρM (µ). Hence, for all θ, ζ ∈ Con(A × M ), we have: θ = α × µ
and ζ = β × ν for some α, β ∈ Con(A) and µ, ν ∈ Con(M ), and thus: θ ≡A×M ζ iff ρA×M (θ) = ρA×M (ζ) iff
ρA×M (α× µ) = ρA×M (β× ν) iff ρA(α)× ρM (µ) = ρA(β)× ρM (ν) iff ρA(α) = ρA(β) = ρA(β) and ρM (µ) = ρM (ν)
iff α ≡A β and µ ≡M ν.
Now let ϕ : L(A)×L(M ) → L(A×M ), for all α ∈ K(A) and all µ ∈ K(M ), ϕ(bα,bµ) = \α × µ. By (ii), ϕ is well
defined and surjective and fulfills: ϕ((bα,bµ)∨ (bβ,bν)) = ϕ(bα ∨ bβ,bµ∨bν) = ϕ(\α ∨ β, [µ ∨ ν) = ((α∨ β)× (µ∨ ν))∧ =
((α × µ)∨ (β × ν))∧ = \(α × µ)∨ \(β × ν) = ϕ(bα,bµ)∨ ϕ(bβ,bν) and, similarly, ϕ((bα,bµ)∧ (bβ,bν)) = ϕ(bα,bµ)∧ ϕ(bβ,bν).
By the form of ≡A×M above, ϕ is injective. Hence ϕ is a lattice isomorphism.
Example 6.13. Let V be the variety generated by the variety of lattices and that of groups. Then, according
to [15, Theorem 1, Lemma 1, Proposition 3] and [35], V is congruence -- modular and any algebra M from V is of
the form M = (L,∨,∧)× (G,·, ⋆), where (L,∨,∧) is a lattice, (G,·) is a group and x ⋆ y = x−1 · y for all x, y ∈ G,
and the direct product above has no skew congruences, thus, by Proposition 6.12, Con(M ) ∼= Con(L) × Con(G)
and L(M ) ∼= L(L) × L(G), since each congruence of the group G also preserves the operation ⋆. Thus, for
instance, in we consider the lattice P from Example 6.10 and the group (S3,◦) from Example 6.7, and we
denote σ ⋆ τ = σ−1 ◦ τ for all σ, τ ∈ S3, and M = (P,∨,∧) × (S3,◦, ⋆), then M is a finite algebra from V
which is not congruence -- distributive, because Con(M ) ∼= Con(P) × Con(S3) and Con(S3) is not distributive,
and L(M ) ∼= L(P) × L(S3) ∼= L(P) × L1 ∼= L(P) ∼= Con(P) ∼= L2 ⊕ L2
2.
7 Further Results on The Commutator
Throughout this section, we shall assume that [·,·]A is commutative and distributive w.r.t. arbitrary joins and
∇A ∈ K(A).
Lemma 7.1. For all n ∈ N∗ and all α, β ∈ Con(A), [α, β]n+1
Proof. Let α, β ∈ Con(A). We proceed by induction on n. By its definition, [α, β]2
Now let n ∈ N∗ such that [α, β]n+1
[[α, β]n+1
Lemma 7.2. If the commutator of A is associative, then, for any n ∈ N∗ and all α, β ∈ Con(A), [α, β]n+1
[[α, α]n
A. Then, by the induction hypothesis, [α, β]n+2
A]A = [[α, β]A, [α, β]A]n+1
A .
A = [[α, β]A, [α, β]A]n
A, [[α, β]A, [α, β]A]n
A = [[α, β]A, [α, β]A]A.
A =
A ]A = [[[α, β]A, [α, β]A]n
A = [[α, β]A, [α, β]A]n
A.
A , [α, β]n+1
A =
A, [β, β]n
A]A.
A ]A.
A, [β, β]n
A, [β, β]n
A]A, [[α, α]n
A , [β, β]n+1
A , [α, β]n+1
A]A]A = [[[α, α]n
A = [[α, β]n+1
A]A]A = [[α, α]n+1
A]A. Then [α, β]n+2
A, [β, β]n
A]A, [[β, β]n
A = [[α, α]n
A, [α, α]n
Proof. Assume that the commutator of A is associative, and let us also use its commutativity, along with Lemma
7.1. Let α, β ∈ Con(A). We apply induction on n. For n = 1, [α, β]2
A = [[α, β]A, [α, β]A]A = [[α, α]A, [β, β]A]A.
Now let n ∈ N∗ such that [α, β]n+1
A ]A = [[[α, α]n
A,
[β, β]n
Lemma 7.3. For all n, k ∈ N∗ and all α, β, φ, ψ, α1, α2, . . . , αk ∈ Con(A):
(i) if α ⊆ β and φ ⊆ ψ, then [α, φ]n
A ⊆ [β, ψ]n
A;
(ii) if k ≤ n, then [α, β]n
A ⊆ [α, β]k
A;
(iii) if k ≥ 2 and n ≥ 2, then [α, β]k·n
A ⊆ [[α, β]k
(iv) [α ∨ β, α ∨ β]n
(v) [α ∨ β, α ∨ β]n·k
(vi) [α ∨ β, α ∨ β]n2
(vii) [α1 ∨ . . . ∨ αk, α1 ∨ . . . ∨ αk]nk
A ⊆ α ∨ [β, β]n
A;
A ⊆ [α, α]k
A ⊆ [α, α]n
A ∨ [β, β]n
A;
A ∨ [β, β]n
A;
A ∨ . . . ∨ [αk, αk]n
A.
A ⊆ [α1, α1]n
A, [α, β]k
A]n
A;
A ]n
A
A
A
.
A, [α, β]p
A, [α, β]2
A]n
A , [α, β]k+1
A = [α, β]n+2
A = [[α, β]p
A = [α, β]k·n+1
⊇ [α, β]k·n+n
A]A ⊆ [α, β]p
A, [[α, β]A, [α, β]A]k
A = [[[α, β]A, [α, β]A]k
A ⊇ [α, β]2n
= [α, β](k+1)·n
A =
A . Now take a k ≥ 2 that fulfills the inclusion in
A ⊇
Proof. (i) By Proposition 3.6, through induction on n.
(ii) For all p ∈ N∗, [α, β]p+1
A, hence the inclusion in the enunciation.
(iii) Assume that n ≥ 2. We apply induction on k, (ii) and Lemma 7.1. For k = 2, we have: [[α, β]2
[[[α, β]A, [α, β]A]A, [[α, β]A, [α, β]A]A]n
the enunciation for all α, β ∈ Con(A). Then [[α, β]k+1
[[α, β]A, [α, β]A]k·n
(iv) We apply induction on n. For n = 1 and all α, β ∈ Con(A), we have [α ∨ β, α ∨ β]A = [α, α]A ∨ [α, β]A ∨
A ⊆ α ∨ [β, β]n
[β, α]A ∨ [β, β]A ⊆ α ∨ [β, β]A. Now let n ∈ N∗ such that [α ∨ β, α ∨ β]n
A for all α, β ∈ Con(A).
Then, by the induction hypothesis and the case n = 1, we have, for all α, β ∈ Con(A):
A =
A, [β, β]n
A]A ⊆ α ∨ [[β, β]n
A, [α ∨ β, α ∨ β]n
[[α ∨ β, α ∨ β]n
(v) We apply Lemma 7.3. For n = 1, [α, α]k
For k ≥ 2 and n ≥ 2, [α, α]k
[α ∨ β, α ∨ β]k·n
A .
(vi) Take k = n in (v).
(vii) We apply induction on k. The statement is trivial for k = 1. Let k ∈ N∗ that fulfills the equality in the
enunciation for any congruences of A, and let α1, . . . , αk, αk+1 ∈ Con(A). By (v), it follows that [α1 ∨ . . .∨ αk ∨
αk+1, α1∨. . .∨αk∨αk+1]nk+1
A ∨ =
[αk+1, αk+1]n
For all θ, ζ ∈ Con(A), we shall denote by θ → ζ = _{α ∈ Con(A) [θ, α]A ⊆ ζ} and by θ⊥ = θ → ∆A =
[α ∨ β, α ∨ β]n+1
A ⊇ [α∨β, α∨β]n
A.
A ⊇
= [(α1∨. . .∨αk)∨αk+1, (α1∨. . .∨αk)∨αk+1]nk·n
A]A ⊆ [α ∨ [β, β]n
A ∨ [β, β]n
A ⊆ [α1∨. . .∨αk, α1∨. . .∨αk]nk
A. For k = 1, [α, α]A∨[β, β]n
A∨[β, β]A ⊇ [α∨β, α∨β]k
A]A = α ∨ [β, β]n+1
A .
A ⊇ [[α ∨ β, α ∨ β]k
A ∨ [αk+1, αk+1]n
A.
A ∨ . . . ∨ [αk, αk]n
A, [α ∨ β, α ∨ β]k
A ∨ β, [α, α]k
A ⊆ [α1, α1]n
A, α ∨ [β, β]n
A ⊇ [[α, α]k
A ∨ β]n
A]n
A]n
A
_{α ∈ Con(A) [θ, α]A = ∆A}.
Remark 7.4. For all θ, ζ ∈ Con(A), θ → ζ = max{α ∈ Con(A) [θ, α]A ⊆ ζ}, because, if we denote by
M = {α ∈ Con(A) [θ, α]A ⊆ ζ}, then [θ, θ → ζ]A = [θ, _
Lemma 7.5. For all α, β, γ ∈ Con(A), [α, β]A ⊆ γ iff α ⊆ β → γ.
Proof. "⇒:" β → γ = _{θ ∈ Con(A) [β, θ]A ⊆ γ}. Since [β, α]A = [α, β]A ⊆ γ, it follows that α ⊆ β → γ.
"⇐:" We have α ⊆ β → γ = _{θ ∈ Con(A) [β, θ]A ⊆ γ}, hence [β, α]A = [α, β]A ⊆ [β, β → γ]A = [β,_{θ ∈
Con(A) [β, θ]A ⊆ γ}]A = _{[β, θ]A θ ∈ Con(A), [β, θ]A ⊆ γ} ⊆ γ.
[θ, α]A ⊆ ζ, hence θ → ζ ∈ M .
α]A = _
α∈M
α∈M
For the following results, recall, also, the equivalences in Proposition 3.15.
Lemma 7.6. For all α, β ∈ Con(A) such that [α,∇A]A = α: α → β = ∇A iff α ⊆ β.
Proof. α → β = ∇A iff ∇A ⊆ α → β iff α = [∇A, α]A ⊆ β, according to Lemma 7.5.
Remark 7.7. By the above, if [·,·]A is associative, then (Con(A),∨,∩, [·,·]A,→, ∆A,∇A) is a residuated lattice,
and, if a C is a congruence -- distributive variety, then (Con(A),∨,∩, [·,·]A,→, ∆A,∇A) is, moreover, a Godel
algebra.
Proposition 7.8. If [θ,∇A]A = θ for all θ ∈ Con(A), for any α, β, γ ∈ Con(A):
A ∨ [β, β]n
A = ∇A for all n ∈ N∗.
(i) if α ∨ β = ∇A, then [α, β]A = α ∩ β;
(ii) if α ∨ β = α ∨ γ = ∇A, then α ∨ [β, γ]A = α ∨ (β ∩ γ) = ∇A;
(iii) if α ∨ β = ∇A, then [α, α]n
Proof. (i) Assume that α ∨ β = ∇A. Since (α ∩ β) → [α, β]A = _{θ ∈ Con(A) [α ∩ β, θ]A ⊆ [α, β]A} and
[α ∩ β, β]A ⊆ [α, β]A and [α, α ∩ β]A ⊆ [α, β]A, it follows that α ⊆ (α ∩ β) → [α, β]A and β ⊆ (α ∩ β) → [α, β]A,
hence ∇A = α ∨ β ⊆ (α ∩ β) → [α, β]A, therefore (α ∩ β) → [α, β]A = ∇A, thus α ∩ β ⊆ [α, β]A by Lemma 7.6.
Since the converse inclusion always holds, it follows that α ∩ β = [α, β]A.
(ii) Assume that α ∨ β = α ∨ γ = ∇A, so that ∇A = [∇A,∇A]A = [α ∨ β, α ∨ γ]A = [α, α]A ∨ [β, α]A ∨ [α, γ]A ∨
[β, γ]A ⊆ α ∨ [β, γ]A ⊆ α ∨ (β ∩ γ) ⊆ ∇A, hence α ∨ [β, γ]A = α ∨ (β ∩ γ) = ∇A.
(iii) We apply induction on n. Assume that α∨β = ∇A, so that, by (ii), α∨[β, β]A = ∇A, thus [α, α]A∨[β, β]A =
∇A, hence the implication holds in the case n = 1. Now, if n ∈ N∗ fulfills the implication in the enunciation for
all α, β ∈ Con(A), and assume that α ∨ β = ∇A, so that [α, α]n
A = ∇A. Then, by the case n = 1, it
follows that [α, α]n+1
A ∨ [β, β]n
A = [[α, α]n
A, [α, α]n
A ∨ [β, β]n+1
A]A ∨ [[β, β]n
A, [β, β]n
A]A = ∇A.
Lemma 7.9. If [γ,∇A]A = γ for all γ ∈ Con(A), then, for all α ∈ B(Con(A)) and all θ ∈ Con(A), [α, θ]A = α∩θ.
Proof. Let θ ∈ Con(A) and α ∈ B(Con(A)), so that there exists a β ∈ Con(A) with α∨ β = ∇A and α∩ β = ∆A.
[α, θ]A ⊆ α ∩ θ = [∇A, α ∩ θ]A = [α ∨ β, α ∩ θ]A = [α, α ∩ θ]A ∨ [β, α ∩ θ]A ⊆
Then the following hold:
[α, α ∩ θ]A ∨ (β ∩ α ∩ θ) ⊆ [α, θ]A ∨ ∆A = [α, θ]A, hence [α, θ]A = α ∩ θ. We have followed the argument from
[29, Lemma 4].
Remark 7.10. By Lemma 7.9, if [γ,∇A]A = γ for all γ ∈ Con(A), then, in B(Con(A)), the commutator of A
equals the intersection, in particular the intersection in B(Con(A)) is distributive with respect to the join.
Lemma 7.11.
(i) If f is surjective, then:
• f (PCon(A) ∩ [Ker(f ))) ⊆ f ({α ∨ Ker(f ) α ∈ PCon(A)}) = PCon(B);
• f (K(A) ∩ [Ker(f ))) ⊆ f ({α ∨ Ker(f ) α ∈ K(A)}) = K(B);
• if C is congruence -- modular and semi -- degenerate, then f (B(Con(A))∩[Ker(f ))) ⊆ f ({α∨Ker(f ) α ∈
B(Con(A))}) ⊆ B(Con(B)).
(ii) For all θ ∈ Con(A):
• {α/θ α ∈ PCon(A) ∩ [θ)} ⊆ {(α ∨ θ)/θ α ∈ PCon(A)}) = PCon(A/θ);
• {α/θ α ∈ K(A) ∩ [θ)} ⊆ {(α ∨ θ)/θ α ∈ K(A)}) = K(A/θ);
• if C is congruence -- modular and semi -- degenerate, then {α/θ α ∈ B(Con(A))∩ [θ)} ⊆ {(α∨ θ)/θ α ∈
B(Con(A))}) ⊆ B(Con(A/θ)).
Proof. The first inclusion in each statement is trivial.
(i) By Lemma 3.17, (ii), for the statements on principal and on compact congruences. Now let α ∈ B(Con(A)),
so that α ∨ β = ∇A and [α, β]A = ∆A for some β ∈ Con(A), hence, by Lemma 3.17, (i), and Remark 3.10,
f (α∨ Ker(f ))∨ f (β∨ Ker(f )) = f (α∨ Ker(f )∨ β∨ Ker(f )) = f (∇A) = ∇B and [f (α∨ Ker(f )), f (β∨ Ker(f )]B =
f ([α, β]A ∨ Ker(f )) = f (∆A) = ∆B, therefore f (α ∨ Ker(f )) ∈ B(Con(B)).
(ii) By (i) for f = pθ.
Proposition 7.12.
(i) Assume that f is surjective. Then: if ∇A ∈ PCon(A), then ∇B ∈ PCon(B), while, if
∇A ∈ K(A), then ∇B ∈ K(B).
(ii) ∇A ∈ PCon(A) iff ∇A/θ ∈ PCon(A/θ) for all θ ∈ Con(A). ∇A ∈ K(A) iff ∇A/θ ∈ K(A/θ) for all
θ ∈ Con(A).
Proof. (i) By Lemma 7.11, (i).
(ii) By (i) for the direct implications, and the fact that A/∆A is isomorphic to A, for the converse implications.
Lemma 7.13. If C is congruence -- modular, then, for all n ∈ N∗ and any α, β ∈ Con(A):
(i) if f is surjective, then [f (α ∨ Ker(f )), f (β ∨ Ker(f ))]n
A/θ = ([α, β]n
(ii) for any θ ∈ Con(A), [(α ∨ θ)/θ, (β ∨ θ)/θ]n
(iii) for any θ ∈ Con(A) and any X, Y ∈ P(A2), [CgA/θ(X/θ), CgA/θ(Y /θ)]n
B = f ([α, β]n
A ∨ θ)/θ;
A/θ = ([CgA(X), CgA(Y )]n
A ∨ Ker(f ));
A ∨ θ)/θ.
Proof. (i) We proceed by induction on n. For n = 1, this holds by Remark 3.10. Now take an n ∈ N∗ such
that [f (α ∨ Ker(f )), f (β ∨ Ker(f ))]n
A ∨ Ker(f )). Then, by the induction hypothesis and Remark
3.10, [f (α ∨ Ker(f )), f (β ∨ Ker(f ))]n+1
B]B =
[f ([α, β]n
(ii) Take f = pθ in (i).
(iii) Take α = CgA(X) and β = CgA(Y ) in (ii) and apply Lemma 3.17, (iii).
B = f ([α, β]n
B = [[f (α ∨ Ker(f )), f (β ∨ Ker(f ))]n
B, [f (α ∨ Ker(f )), f (β ∨ Ker(f ))]n
A ∨ Ker(f )) = f ([α, β]n+1
A]n
A ∨ Ker(f ))]B = f ([[α, β]n
A ∨ Ker(f )), f ([α, β]n
A, [α, β]n
A ∨ Ker(f )).
8 Boolean Congruences versus the Reticulation
Throughout this section, we shall assume that [·,·]A is commutative and distributive w.r.t. arbitrary joins and
∇A ∈ K(A). We call A a semiprime algebra iff ρA(∆A) = ∆A. So A is semiprime iff ∆A ∈ RCon(A).
Remark 8.1. By Proposition 5.7, if the commutator of A equals the intersection, then A is semiprime, hence,
if C is congruence -- distributive, then every member of C is semiprime.
Proposition 8.2. A/ρA(∆A) is semiprime.
Proof. By Proposition 5.3, (iv), and Proposition 5.4, (iv), ρA/ρA(∆A)(∆A/ρA(∆A)) = ρA(ρA(∆A))/ρA(∆A) =
ρA(∆A)/ρA(∆A) = ∆A/ρA(∆A).
Lemma 8.3. If A is semiprime, then, for all α, β ∈ Con(A):
• λA(α) = 0 iff α = ∆A;
• [α, β]A = ∆A iff α ∩ β = ∆A.
Proof. Let α, β ∈ Con(A). Since λA(∆A) = 0 and [α, β]A ⊆ α ∩ β, the converse implications always hold. Now
assume that A is semiprime. If λA(α) = 0 = λA(∆A), then α ⊆ ρA(α) = ρA(∆A) = ∆A, thus α = ∆A. If
[α, β]A = ∆A, then λA(α ∩ β) = λA([α, β]A) = λA(∆A) = 0, hence α ∩ β = ∆A by the above.
Lemma 8.4. For any θ ∈ Con(A), the following hold:
(i) ρA(θ) = _{α ∈ Con(A) (∃ k ∈ N∗) ([α, α]k
A ⊆ θ)} = _{α ∈ K(A) (∃ k ∈ N∗) ([α, α]k
A ⊆ θ)} = _{α ∈
PCon(A) (∃ k ∈ N∗) ([α, α]k
A ⊆ θ)};
A ⊆ θ.
A ⊆ θ)} = _{α ∈ PCon(A) (∃ k ∈ N∗) ([α, α]k
A ⊆ θ)}, where the last inclusion holds because, for any α ∈ Con(A), if k ∈ N∗ is such that [α, α]k
A ⊆ θ)} ⊆ _{α ∈ Con(A) (∃ k ∈ N∗) ([α, α]k
A ⊆ θ, thus α = _
(ii) for any α ∈ K(A), α ⊆ ρA(θ) iff there exists a k ∈ N∗ such that [α, α]k
Proof. (i) By Proposition 5.27 and the fact that PCon(A) ⊆ K(A) ⊆ Con(A), ρA(θ) = _{CgA(a, b) (a, b) ∈
A ⊆ θ)} ⊆ _{α ∈
A2, (∃ k ∈ N∗) ([CgA(a, b), CgA(a, b)]k
A ⊆ θ)} ⊆ _{α ∈ PCon(A) (∃ k ∈
K(A) (∃ k ∈ N∗) ([α, α]k
N∗) ([α, α]k
A ⊆ θ,
CgA(a, b) ⊆ _{CgA(a, b) (a, b) ∈
then, for any (a, b) ∈ α, [CgA(a, b), CgA(a, b)]k
A2, (∃ k ∈ N∗) ([CgA(a, b), CgA(a, b)]k
(ii) The converse implication follows directly from (i).
non -- empty families (βj)j∈J ⊆ K(A) and (kj)j∈J ⊆ N∗ such that α ⊆ _
J. Since α ∈ K(A), it follows that there exist an n ∈ N∗ and j1, . . . , jn ∈ J such that α ⊆
j = max{j1, . . . , jn} ∈ N∗. Then [βji , βji]k
[α, α]kn
For the direct implication, from (i) it follows that, for any α ∈ K(A) such that α ⊆ ρA(θ), there exist
A ⊆ θ for all j ∈
A ⊆ θ for each i ∈ 1, n, thus, by Lemma 7.3, (vii),
A ⊆ θ)}. Hence the equalities in the enunciation.
βj and [βj, βj]kj
A ⊆ [α, α]k
βji . Let
n_
(a,b)∈α
j∈J
i=1
A ⊆ [βji , βji ]ki
A ⊆ θ.
n_
n_
A ⊆ [
n_
i=1
i=1
A = ∆A)} = _{α ∈ K(A) (∃ k ∈
βji,
i=1
A ⊆
[βji , βji ]k
βji ]kn
(i) ρA(∆A) = _{α ∈ Con(A) (∃ k ∈ N∗) ([α, α]k
A = ∆A)};
A = ∆A)} = _{α ∈ PCon(A) (∃ k ∈ N∗) ([α, α]k
Proposition 8.5.
N∗) ([α, α]k
(ii) for any α ∈ K(A), α ⊆ ρA(∆A) iff there exists a k ∈ N∗ such that [α, α]k
Proof. By Lemma 8.4.
Corollary 8.6. A is semiprime iff, for any α ∈ K(A) and any k ∈ N∗, if [α, α]k
A = ∆A.
A = ∆A, then α = ∆A.
Throughout the rest of this section, we shall assume that K(A) is closed w.r.t. the commutator of A and
[θ,∇A]A = θ for all θ ∈ Con(A); see also Proposition 3.15.
For any bounded lattice L, we shall denote by B(L) the set of the complemented elements of L.
If L is
distributive, then B(L) is the Boolean center of L. Although Con(A) is not necessarily distributive, we shall call
B(Con(A)) the Boolean Center of Con(A). So B(Con(A)) is the set of the α ∈ Con(A) such that there exists a
β ∈ Con(A) which fulfills α ∨ β = ∇A and α ∩ β = ∆A, thus also [α, β]A = ∆A.
Remark 8.7. Obviously, ∆A,∇A ∈ B(Con(A)).
Lemma 8.8. B(Con(A)) ⊆ K(A).
Proof. Let α ∈ B(Con(A)), so that α ∨ β = ∇A and α ∩ β = ∆A for some β ∈ Con(A). Now let ∅ 6= (αi)i∈I ⊆
Con(A) such that α ⊆ _
αij for some n ∈ N∗ and
αi = ∇A ∈ K(A), thus ∇A = β ∨
αi, so that β ∨ _
n_
i∈I
i∈I
j=1
some i1, . . . , in ⊆ I, hence, by Proposition 7.8, (i), α = [α,∇A]A = [α, β ∨
∆A ∨ [α,
n_
j=1
αij ]A = [α,
n_
j=1
αij ]A ⊆
n_
j=1
αij , hence α ∈ K(A).
Proposition 8.9. If K(A) = B(Con(A)), then L(A) = B(L(A)).
n_
j=1
αij ]A = [α, β]A ∨ [α,
n_
j=1
αij ]A =
α∈M
α∈M
α∈M
α∈M
α =
α∈M
α∈M∩K(A)
α∈M∩PCon(A)
_
α∈M
α∈M
(a,b)∈α
CgA(a, b) ⊆
α = _
α = _
_
α, therefore θ⊥ = _
[α, θ]A = _
γ ⊆ _
_
α. Note, also, that [θ⊥, θ]A = [ _
Proof. If K(A) = B(Con(A)), then L(A) = λA(K(A)) = λA(B(Con(A))) ⊆ B(L(A)) ⊆ L(A) by Lemma 8.11,
thus L(A) = B(L(A)).
Lemma 8.10. For any σ, θ ∈ Con(A): θ⊥ = _{α ∈ PCon(A) [α, θ]A = ∆A} = _{α ∈ K(A) [α, θ]A =
∆A} = _{α ∈ Con(A) [α, θ]A = ∆A} = max{α ∈ Con(A) [α, θ]A = ∆A}, thus: σ ⊆ θ⊥ iff [σ, θ]A = ∆A.
Proof. Let M = {α ∈ Con(A) [α, θ]A = ∆A}. For all α ∈ M and all (a, b) ∈ α, [CgA(a, b), θ]A ⊆ [α, θ]A = ∆A,
thus CgA(a, b) ∈ M ∩ PCon(A). Hence θ⊥ = _
γ ⊆
_
_
[σ, θ]A = ∆A, and conversely: if [σ, θ]A = ∆A, then σ ∈ M , thus σ ⊆ max(M ) = θ⊥.
Lemma 8.11. λA(B(Con(A))) = B(Con(A))/≡A ⊆ B(L(A)) ⊆ B(Con(A)/≡A) and λA B(Con(A)): B(Con(A))
→ B(L(A)) is a Boolean morphism.
Proof. λA(B(Con(A))) = B(Con(A))/≡A . Now we use Lemma 8.8. Let α ∈ B(Con(A)) ⊆ K(A), so that
λA(α) ∈ L(A) and, for some β ∈ B(Con(A)) ⊆ K(A), we have α ∨ β = ∇A and α ∩ β = ∆A. Then λA(β) ∈
L(A), 1 = λA(∇A) = λA(α ∨ β) = λA(α) ∨ λA(β) and 0 = λA(∆A) = λA(α ∩ β) = λA(α) ∧ λA(β), hence
λA(α) ∈ B(L(A)). Therefore λA(B(Con(A))) ⊆ B(L(A)). Since L(A) is a bounded sublattice of the bounded
distributive lattice Con(A)/≡A, it follows that B(L(A)) is a Boolean subalgebra of B(Con(A)/≡A). Hence
λA(B(Con(A))) = B(Con(A))/≡A ⊆ B(L(A)) ⊆ B(Con(A)/≡A). λA : Con(A) → Con(A)/≡A is a (surjective)
bounded lattice morphism. Hence λA B(Con(A)): B(Con(A)) → B(L(A)) is well defined and it is a bounded
lattice morphism, thus it is a Boolean morphism.
∆A = ∆A, hence θ⊥ ∈ M , thus θ⊥ = max(M ). If σ ⊆ θ⊥, then [σ, θ]A ⊆ [θ⊥, θ]A = ∆A, thus
α, θ]A =
γ∈M∩PCon(A)
γ∈M∩K(A)
Throughout the rest of this section, C shall be congruence -- modular and semi -- degenerate.
Proposition 8.12.
(i) The Boolean morphism λA B(Con(A)): B(Con(A)) → B(L(A)) is injective.
(ii) If the commutator of A is associative, then λA(B(Con(A))) = B(L(A)) = B(Con(A))/≡A ⊆ B(Con(A)/≡A)
and λA B(Con(A)): B(Con(A)) → B(L(A)) is a Boolean isomorphism.
(iii) If A is semiprime, then λA(B(Con(A))) = B(L(A)) = B(Con(A))/≡A = B(Con(A)/≡A) and λA B(Con(A)):
B(Con(A)) → B(L(A)) is a Boolean isomorphism.
Proof. (i) By Lemma 8.11, λA B(Con(A)): B(Con(A)) → B(L(A)) is a Boolean morphism. By Remark 5.10,
λA(α) = 1 iff α = ∇A, hence this Boolean morphism is injective.
(ii) Assume that A is semiprime, and let x ∈ B(Con(A)/≡A), so that x ∨ y = 1 and x ∧ y = 0 for some
y ∈ B(Con(A)/≡A ). Hence there exist α, β ∈ Con(A) such that x = λA(α) and y = λA(β), thus 1 = x ∨ y =
λA(α)∨ λA(β) = λA(α∨ β) and 0 = x∧ y = λA(α)∧ λA(β) = λA(α∩ β), therefore α∨ β = ∇A and α∩ β = ∆A,
by Remark 5.10 and Lemma 8.3. Hence α ∈ B(Con(A)), thus x = λA(α) ∈ λA(B(Con(A))) = B(Con(A))/≡A,
therefore, by Lemma 8.11, B(Con(A)/≡A) ⊆ λA(B(Con(A))) = B(Con(A))/≡A ⊆ B(L(A)) ⊆ B(Con(A)/≡A),
hence λA(B(Con(A))) = B(Con(A))/≡A = B(L(A)) = B(Con(A)/≡A ). Therefore λA B(Con(A)): B(Con(A)) →
B(L(A)) is surjective, so, by (i), it is a Boolean isomorphism.
(iii) Assume that the commutator of A is associative, and let x ∈ B(L(A)) ⊆ L(A) = λA(K(A)), so that x∨y = 1
and x ∧ y = 0 for some y ∈ B(L(A)) and there exist α, β ∈ K(A) such that x = λA(α) and y = λA(β). Then
λA(α ∨ β) = λA(α) ∨ λA(β) = x ∨ y = 1 = λA(∇A), hence α ∨ β = ∇A by Remark 5.10. We also have
λA([α, β]A) = λA(α) ∧ λA(β) = x ∧ y = 0 = λA(∆A), thus [α, β]A ⊆ ρA(α ∩ β) = ρA(∆A), and, since K(A)
is closed with respect to the commutator, we have [α, β]A ∈ K(A), thus, according to Proposition 8.5, (ii),
A]A = [α, β]k+1
[[α, α]k
A = ∆A for some k ∈ N∗; we have applied Lemmas 7.1 and 7.2.
But α∨β = ∇A, thus [α, α]k
A]A = ∆A by Proposition 7.8,
(iii) and (i). Therefore [α, α]k
A) ∈ λA(B(Con(A))), hence B(L(A)) ⊆
λA(B(Con(A))), thus B(L(A)) ⊆ λA(B(Con(A))) = B(Con(A))/≡A ⊆ B(L(A)) ⊆ B(Con(A)/≡A) by Lemma
8.11, therefore λA(B(Con(A))) = B(Con(A))/≡A = B(L(A)) ⊆ B(Con(A)/≡A). Therefore λA B(Con(A)):
B(Con(A)) → B(L(A)) is surjective, so, by (i), it is a Boolean isomorphism.
A = [[α, β]A, [α, β]A]k
A∨[β, β]k
A ∈ B(Con(A)), thus x = λA(α) = λA([α, α]k
A = ∇A, hence [α, α]k
A∩[β, β]k
A = [[α, α]k
A, [β, β]k
A, [β, β]k
Lemma 8.13. If A is semiprime and α ∈ Con(A), then: α ∈ B(Con(A)) iff λA(α) ∈ B(L(A)).
Proof. We apply Lemma 8.11, which, first of all, gives us the direct implication. For the converse, assume that
λA(α) ∈ B(L(A)) = B(Con(A)/≡A), so that there exists a β ∈ Con(A) with λA(α ∨ β) = λA(α) ∨ λA(β) = 1 =
λA(∇A) and λA(α ∩ β) = λA(α) ∧ λA(β) = 0, thus α ∨ β = ∇A and α ∩ β = ∆A by Remark 5.10 and Lemma
8.3. Therefore α ∈ B(Con(A)).
For any Ω ⊆ Con(A), let us consider the property:
(A, Ω)
for all α, β ∈ Ω and all n ∈ N∗, there exists a k ∈ N∗ such that [[α, α]k
A, [β, β]k
A]A ⊆ [α, β]n
A
Remark 8.14. By Lemma 7.2, if the commutator of A is associative, then (A, Con(A)) holds.
Notice, from the proof of statement (iii) from Proposition 8.12, that this statement, and thus the fact that
λA B(Con(A)): B(Con(A)) → B(L(A)) is a Boolean isomorphism, also hold if property (A,K(A)) is fulfilled,
instead of the associativity of the commutator of A.
Open problem 8.15. Under the current context, determine whether (A,K(A)) always holds; if it doesn't, then
determine whether (A,K(A)) is equivalent to the associativity of the commutator of A.
Lemma 8.16. (B(Con(A)),∨, [·,·]A = ∩,⊥, ∆A,∇A) is a Boolean algebra.
Proof. We follow, in part, the argument from [29, Lemma 4]. Let α, β ∈ B(Con(A)), so that there exist
α, β ∈ B(Con(A)) such that α ∨ α = β ∨ β = ∇A and α ∩ α = β ∩ β = ∆A. Then, by Remark 7.10, the following
hold: (α ∨ β) ∩ α ∩ β = (α ∩ α ∩ β) ∨ (β ∩ α ∩ β) = ∆A ∨ ∆A = ∆A and, since α ∩ β ⊆ β, it follows that
α∨ β∨ (α∩ β) = α∨ β∨ (α∩ β)∨ (α∩ β) = α∨ β∨ (α∩ (β∨ β)) = α∨ β∨ (α∩∇A) = α∨ β∨ α = ∇A. Analogously,
(α∨ β)∩ α∩ β = ∆A and α∨ β ∨ (α∩ β) = ∇A. Hence α∨ β, α∩ β ∈ B(Con(A)). Clearly, ∆A,∇A ∈ B(Con(A)).
Therefore B(Con(A)) is a bounded sublattice of Con(A). By Remark 7.10, it follows that (B(Con(A)),∨, [·,·]A =
∩, ∆A,∇A) is a bounded distributive lattice, and, by its definition, it is also complemented, thus it is a Boolean
lattice. By a well -- known characterization of the complement in a Boolean lattice, for any θ ∈ B(Con(A)), the
complement of θ in B(Con(A)) is θ = max{α ∈ B(Con(A)) α ∩ θ = ∆A} = max{α ∈ B(Con(A)) [α, θ]A =
∆A} ⊆ max{α ∈ Con(A) [α, θ]A = ∆A} = θ⊥ according to Lemma 8.10, thus ∇A = θ ∨ θ ⊆ θ ∨ θ⊥, so
θ ∨ θ⊥ = ∇A. Again by Lemma 8.10, ∆A = [θ, θ⊥]A = θ ∩ θ⊥. Therefore θ⊥ ∈ B(Con(A)) and θ⊥ is the
complement of θ in B(Con(A)).
For any bounded lattice L and any I ∈ Id(L), we shall denote by Ann(I) the annihilator of I in L: Ann(I) =
{a ∈ L (∀ x ∈ I) (a ∧ x = 0)}. It is immediate that, if L is distributive, then Ann(I) ∈ Id(L). Throughout the
rest of this paper, all annihilators shall be considerred in the bounded distributive lattice L(A), so they shall be
ideals of the lattice L(A). Recall that L(A) = λA(K(A)).
Lemma 8.17. For any α ∈ K(A):
• Ann(α∗) = {λA(β) β ∈ K(A), λA([α, β]A) = 0};
• if A is semiprime, then Ann(α∗) = {λA(β) β ∈ K(A), [α, β]A = ∆A}.
Proof. By Lemma 5.14, Ann(α∗) = Ann((λA(α)]) = {λA(β) β ∈ K(A), (∀ x ∈ (λA(α)]) (x ∧ λA(β) = 0)} =
{λA(β) β ∈ K(A), λA(α) ∧ λA(β) = 0)} = {λA(β) β ∈ K(A), λA([α, β]A) = 0}. By Lemma 8.3, if A
is semiprime, then, for any β ∈ K(A), λA([α, β]A) = 0 iff [α, β]A = ∆A, hence the second equality in the
enunciation.
Lemma 8.18. For any α ∈ Con(A) and any I ∈ Id(L(A)), if Ann(α∗) ⊆ I, then α⊥ ⊆ I∗. If A is semiprime
and α ∈ K(A), then the converse implication holds, as well.
Proof. For the direct implication, assume that Ann(α∗) ⊆ I and let β ∈ K(A) such that [α, β]A = ∆A, hence
λA(α) ∧ λA(β) = λA([α, β]A) = λA(∆A) = 0. Now let x ∈ α∗, so that x = λA(γ) for some γ ∈ K(A) with
γ ⊆ α. Then x = λA(γ) ≤ λA(α), hence x ∧ λA(β) = λA(γ) ∧ λA(β) ≤ λA(α) ∧ λA(β) = 0, so x ∧ λA(β) = 0,
thus λA(β) ∈ Ann(α∗) ⊆ I, therefore β ⊆ I∗ by Lemma 5.16. According to Lemma 8.10, α⊥ = _{β ∈
K(A) [α, β]A = ∆A} ⊆ I∗.
For the converse implication, assume that A is semiprime, α ∈ K(A) and α⊥ ⊆ I∗, and let x ∈ Ann(α∗),
which means that x = λA(β) for some β ∈ K(A) with [α, β]A = ∆A, according to Lemma 8.17. Hence, by
Lemmas 8.10 and 5.16, β ⊆ α⊥ ⊆ I∗, thus x = λA(β) ∈ I, therefore Ann(α⊥) ⊆ I.
Proposition 8.19. For any θ ∈ Con(A):
(i) (θ⊥)∗ ⊆ Ann(θ∗);
(ii) if A is semiprime, then (θ⊥)∗ = Ann(θ∗).
Proof. (θ⊥)∗ = {λA(α) α ∈ K(A), α ⊆ θ⊥} = {λA(α) α ∈ K(A), [α, θ]A = ∆A}, by Lemma 8.10.
Ann(θ∗) = {λA(α) α ∈ K(A), (∀ x ∈ θ∗) (λA(α) ∧ x = λA(∆A))} = {λA(α) α ∈ K(A), (∀ β ∈ K(A)) (β ⊆
θ ⇒ λA([α, β]A) = λA(α) ∧ λA(β) = λA(∆A))} = {λA(α) α ∈ K(A), (∀ β ∈ K(A)) (β ⊆ θ ⇒ ρA([α, β]A) =
ρA(∆A))}.
(i) Let α ∈ K(A) such that λA(α) ∈ (θ⊥)∗, which means that [α, θ]A = ∆A. Then, for any β ∈ K(A) fulfilling
β ⊆ θ, we have [α, β]A ⊆ [α, θ]A = ∆A, so [α, β]A = ∆A, thus ρA([α, β]A) = ρA(∆A)), hence λA(α) ∈ Ann(θ∗).
Therefore (θ⊥)∗ ⊆ Ann(θ∗).
(ii) Assume that A is semiprime and α ∈ K(A) such that λA(α) ∈ Ann(θ∗), which means that, for all β ∈ K(A)
CgA(a, b) ⊆ _{β ∈
such that β ⊆ θ, [α, β]A ⊆ ρA([α, β]A) = ρA(∆A) = ∆A, so [α, β]A = ∆A. θ = _
K(A) β ⊆ θ} ⊆ θ, thus θ = _{β ∈ K(A) β ⊆ θ}, so [α, θ]A = [α,_{β ∈ K(A) β ⊆ θ}]A = _{[α, β]A β ∈
K(A), β ⊆ θ} = _{∆A β ∈ K(A), β ⊆ θ} = _{∆A} = ∆A, therefore λA(α) ∈ (θ⊥)∗, hence Ann(θ∗) ⊆ (θ⊥)∗,
(a,b)∈θ
thus Ann(θ∗) = (θ⊥)∗ by (i).
Proposition 8.20. For any I ∈ Id(L(A)):
(i) (I∗)⊥ ⊆ Ann(I)∗;
(ii) if A is semiprime, then (I∗)⊥ = Ann(I)∗.
Proof. (I∗)⊥ = W{α ∈ K(A) [α, I∗]A = ∆A} = W{α ∈ K(A) [α,W{β ∈ K(A) λA(β) ∈ I}]A = ∆A} =
W{α ∈ K(A) W{[α, β]A ∈ K(A) β ∈ K(A), λA(β) ∈ I} = ∆A} = W{α ∈ K(A) (∀ β ∈ K(A)) (λA(β) ∈ I ⇒
[α, β]A = ∆A)}. (Ann(I))∗ = W{α ∈ K(A) λA(α) ∈ Ann(I)} = W{α ∈ K(A) (∀ β ∈ K(A)) (λA(β) ∈ I ⇒
λA(α) ∧ λA(β) = 0)} = W{α ∈ K(A) (∀ β ∈ K(A)) (λA(β) ∈ I ⇒ λA([α, β]A) = λA(∆A))}.
(i) For all α, β ∈ Con(A), if [α, β]A = ∆A, then λA([α, β]A) = λA(∆A), hence (I∗)⊥ ⊆ Ann(I)∗.
(ii) If A is semiprime, then, for every α, β ∈ Con(A), λA([α, β]A) = λA(∆A) implies [α, β]A ⊆ ρA([α, β]A) =
ρA(∆A) = ∆A, thus [α, β]A = ∆A, hence Ann(I)∗ ⊆ (I∗)⊥. By (i), it follows that (I∗)⊥ = Ann(I)∗.
We call A a hyperarchimedean algebra iff, for all α ∈ PCon(A), there exists an n ∈ N∗ such that [α, α]n
A ∈
B(Con(A)).
Remark 8.21. If α ∈ Con(A) and n ∈ N∗ are such that [α, α]n
[[α, α]n
A, thus [α, α]k
A]A = [α, α]n
A = [α, α]n
A, [α, α]n
A ∩ [α, α]n
A ∈ B(Con(A)), then, by Remark 7.10, [α, α]n+1
A =
A for all k ∈ N such that k ≥ n.
A = [α, α]n
Remark 8.22. If [α, α]A ∈ B(Con(A)) for all α ∈ PCon(A), then A is hyperarchimedean. Thus, if PCon(A) ⊆
B(Con(A)) and A has principal commutators, then A is hyperarchimedean. If the commutator of A equals the
intersection, for instance if C is congruence -- distributive, then: A is hyperarchimedean iff PCon(A) ⊆ B(Con(A)).
By Lemmas 8.16 and 8.8, the following equivalences hold: PCon(A) ⊆ B(Con(A)) iff K(A) ⊆ B(Con(A)) iff
K(A) = B(Con(A)).
Remark 8.23. By Lemma 8.16, the lattice Con(A) is Boolean iff Con(A) = B(Con(A)), which implies that
the commutator of A equals the intersection, according to Remark 7.10, and thus, since PCon(A) ⊆ Con(A) =
B(Con(A)), A is hyperarchimedean, while Remark 8.1 ensures us that A is semiprime. From Lemma 8.8, we
obtain the following equivalences: Con(A) is a Boolean lattice iff B(Con(A)) = Con(A) iff B(Con(A)) = K(A) =
Con(A). Of course, since L(A) is a bounded distributive lattice, L(A) is a Boolean algebra iff L(A) = B(L(A)).
Proposition 8.24.
(i) If A is semiprime, then: K(A) = B(Con(A)) iff L(A) = B(L(A)).
(ii) If Con(A) is a Boolean lattice, then A is hyperarchimedean and semiprime and L(A) is isomorphic to
Con(A), in particular L(A) is a Boolean lattice, as well.
Proof. (i) The direct implication is Proposition 8.9. For the converse, let α ∈ K(A), so that λA(α) ∈ L(A) =
B(L(A)), thus α ∈ B(Con(A)) by Lemma 8.13. Hence K(A) ⊆ B(Con(A)), thus K(A) = B(Con(A)) by Lemma
8.8.
(ii) By Lemma 8.23, we obtain that A is hyperarchimedean and semiprime, and B(Con(A)) = K(A) = Con(A),
hence L(A) = B(L(A)) by Proposition 8.9, and thus λA : Con(A) = B(Con(A)) → B(L(A)) = L(A) is a Boolean
isomorphism, according to Proposition 8.12.
Lemma 8.25. If A is hyperarchimedean, then A/θ is hyperarchimedean for all θ ∈ Con(A).
Proof. Let θ ∈ Con(A). For any a, b ∈ A, there exists an n ∈ N∗ such that [CgA(a, b), CgA(a, b)]n
B(Con(A)). Then, according to Lemma 7.13, (iii), and Lemma 7.11, (ii), [CgA/θ(a/θ, b/θ), CgA/θ(a/θ, b/θ)]n
([CgA(a, b), CgA(a, b)]n
A ∨ θ)/θ ∈ B(Con(A/θ)), therefore A/θ is hyperarchimedean.
A ∈
A/θ =
Lemma 8.26. If A is hyperarchimedean, then L(A) is a Boolean lattice.
Proof. Let θ ∈ K(A), so that θ = α1 ∨ . . . ∨ αn for some n ∈ N∗ and α1, . . . , αn ∈ PCon(A). Since A is
hyperarchimedean, there exists a k ∈ N∗ such that, for all i ∈ 1, n, [αi, αi]k
A ∈ B(Con(A)), thus λA(αi) =
λA([αi, αi]k
A) ∈ λA(B(Con(A))) ⊆ B(L(A)) by Lemma 8.11, so that λA(θ) = λA(α1) ∨ . . . ∨ λA(αn) ∈ B(L(A)).
Hence λA(K(A)) = L(A) ⊆ B(L(A)), thus L(A) = B(L(A)), so L(A) is a Boolean lattice.
9 A Reticulation Functor
Throughout this section, C shall be congruence -- modular and semi -- degenerate and such that, in each of its
members, the set of the compact congruences is closed w.r.t. the commutator. Also, the morphism f : A → B
shall be surjective, so that the map ϕf : Con(A) → Con(B), ϕf (α) = f (α ∨ Ker(f )) for all α ∈ Con(A), is well
defined.
Remark 9.1. By Lemma 7.11, (i), ϕf (K(A)) = K(B).
For any algebra M from C and any X ⊆ M 2, let us denote VM (X) = VM (CgM (X)). Then, by the proof of
[1, Proposition 2.1] and Lemma 3.17, (i), for all α ∈ Con(A), {f (φ) φ ∈ VA(α)} = f (VA(α)) = VB(f (α)) =
VB(CgB(f (α))) = VB(f (α ∨ Ker(f ))) = VB(ϕf (α)).
ϕf
✲
Con(B)
Con(A)
S
K(A)
λA❄
L(A)
ϕf
L(f )
✲
✲
S
K(B)
λB❄
L(B)
Let us define L(f ) : L(A) → L(B), for all α ∈ K(A), L(f )(bα) = \ϕf (α), that is L(f )(λA(α)) = λB(f (α ∨
Ker(f ))).
Proposition 9.2. L(f ) is well defined and it is a surjective lattice morphism.
Proof. By Remark 9.1, the restriction ϕf K(A): K(A) → K(B) is well defined and surjective. Let α, β ∈ K(A)
such that λA(α) = λA(β), so that ρA(α) = ρA(β), thus VA(α) = VA(β), hence VB(ϕf (α)) = f (VA(α)) =
f (VA(β)) = VB(ϕf (β)), thus ρB(ϕf (α)) = ρB(ϕf (β)), so λB(ϕf (α)) = λB(ϕf (β)), that is L(f )(λA(α)) =
L(f )(λA(β)); we have used Proposition 5.3, (ii), and Remark 9.1. Hence L(f ) is well defined. λB : K(B) → L(B)
ϕf K(A): K(A) → K(B) are surjective, thus so is their composition, and, since L(f ) ◦ λA = λB ◦ ϕf , it follows
that L(f ) is surjective.
By Remark 3.10, Lemma 3.17, (ii), and Proposition 5.4, (ii) and (v), for all α, β ∈ K(A), the following
hold: L(f )(bα ∧ bβ) = L(f )(λA(α) ∧ λA(β)) = L(f )(λA([α, β]A)) = λB(ϕf ([α, β]A)) = λB(f ([α, β]A ∨ Ker(f ))) =
λB([f (α ∨ Ker(f )), f (β ∨ Ker(f ))]B)) = λB(f (α ∨ Ker(f ))) ∧ λB(f (β ∨ Ker(f ))) = λB(ϕf (α)) ∧ λB(ϕf (β)) =
L(f )(λA(α)) ∧ L(f )(λA(β)) = L(f )(bα) ∧ L(f )(bβ) and L(f )(bα ∨ bβ) = L(f )(λA(α) ∨ λA(β)) = L(f )(λA(α ∨ β)) =
λB(ϕf (α ∨ β)) = λB(f (α ∨ β ∨ Ker(f ))) = λB(f (α ∨ Ker(f ) ∨ β ∨ Ker(f ))) = λB(f (α ∨ Ker(f ))) ∨ λB(f (β ∨
Ker(f ))) = λB(ϕf (α)) ∨ λB(ϕf (β)) = L(f )(λA(α)) ∨ L(f )(λA(β)) = L(f )(bα) ∨ L(f )(bβ). Therefore L(f ) is a
lattice morphism.
Con(A)
S
K(A)
λA❄
L(A)
✲
Con(A/θ)
S
K(A/θ)
λA/θ❄
L(A/θ)
ϕpθ
L(pθ)
✲
✲
Remark 9.3. Clearly, if C is an algebra from C and g : B → C is a surjective morphism in C, then L(g ◦ f ) =
L(g) ◦ L(f ). Hence we have defined a covariant functor L from the partial category of C whose morphisms are
exactly the surjective morphisms from C to the partial category of the category D01 of bounded distributive
lattices whose morphisms are exactly the surjective morphisms from D01.
Open problem 9.4. Extend the definition of L to the whole category C, with the image in D01, of course.
Remark 9.5. By Proposition 6.2, if C is congruence -- distributive, then we may take L(f ) = ϕf K(A): K(A) →
K(B), with K(A) and K(B) bounded sublattices of Con(A) and Con(B), respectively.
For any bounded lattice morphism h : L → M , let us denote by KerId(h) = h−1({0}) = {x ∈ L h(x) =
0} ∈ Id(L), so that L/KerId(h) ∼= h(L) by the Main Isomorphism Theorem (for lattices and lattice ideals).
Proposition 9.6 (the reticulation preserves quotients). For any θ ∈ Con(A), the lattices L(A/θ) and L(A)/θ∗
are isomorphic.
Proof. Recall that θ∗ = λA(K(A) ∩ (θ]) = {bα α ∈ K(A), α ⊆ θ} ∈ Id(L(A)). pθ : A → A/θ is a surjective
morphism in C, so we can apply the construction above:
ϕpθ
For all α ∈ Con(A), ϕpθ (α) = pθ(α ∨ Ker(pθ)) = (α ∨ θ)/θ, so, for all α ∈ K(A), L(pθ)(bα) = \(α ∨ θ)/θ ∈
L(A/θ). Thus, for any α ∈ K(A): bα ∈ KerId(L(pθ)) iff L(pθ)(bα) = [∆A/θ iff \(α ∨ θ)/θ = dθ/θ, that is λA/θ((α ∨
θ)/θ) = λA/θ(θ/θ), iff ρA/θ((α∨ θ)/θ) = ρA/θ(θ/θ) iff ρA(α∨ θ)/θ = ρA(θ)/θ iff ρA(α∨ θ) = ρA(θ) iff ρA(α∨ θ) ⊆
ρA(θ) iff α∨ θ ⊆ ρA(θ) iff α ⊆ ρA(θ) iff bα ∈ (ρA(θ))∗ = θ∗, hence KerId(L(pθ)) = θ∗; we have applied Proposition
5.3, (iii), Remark 5.1, (iii), Proposition 5.3, (i), and Corollary 5.29, (i). Proposition 9.2 ensures us that the lattice
morphism L(pθ) is surjective, so, from the Main Isomorphism Theorem, we obtain: L(A/θ) ∼= L(A)/θ∗.
Proposition 9.7. The lattices L(A) and L(A/ρA(∆A)) are isomorphic.
Proof. By Corollary 5.29, (i), and Proposition 9.6, ρA(∆A)∗ = ∆∗
to L(A)/ρA(∆A))∗ = L(A)/∆∗
A/∆A and A are isomorphic.
A, hence the lattice L(A/ρA(∆A)) is isomorphic
A, which, in turn, is isomorphic to L(A/∆A), and thus to L(A), since the algebras
Remark 9.8. Propositions 8.2 and 9.7 show that the reticulation of any algebra M from a semi -- degenerate
congruence -- modular variety, such that K(M ) is closed with respect to the commutator of M and ∇M ∈ K(M ),
is isomorphic to the reticulation of a semiprime algebra from the same variety.
Corollary 9.9. B(L(A)) and B(Con(A/ρA(∆A))) are isomorphic Boolean algebras.
Proof. By Propositions 8.2, 8.12 and 9.7, A/ρA(∆A) is semiprime, thus the Boolean algebra B(Con(A/ρA(∆A)))
is isomorphic to B(L(A/ρA(∆A))), which in turn is isomorphic to B(L(A)).
Recall the well -- known Nachbin's Theorem, which states that, given a bounded distributive lattice L, we
have: L is a Boolean algebra iff MaxId(L) = SpecId(L) iff MaxFilt(L) = SpecFilt(L).
Proposition 9.10. The following are equivalent:
(i) A is hyperarchimedean;
(ii) A/ρA(∆A) is hyperarchimedean;
(iii) Max(A) = Spec(A);
(iv) L(A) is a Boolean lattice;
(v) the lattice L(A) is isomorphic to B(Con(A));
(vi) the lattice L(A) is isomorphic to B(Con(A/ρA(∆A))).
Proof. By Nachbin's Theorem, Proposition 5.22 and Corollary 5.26, (iii) is equivalent to (iv). Trivially, (vi)
implies (iv), while the converse holds by Corollary 9.9.
If A is semiprime, that is ρA(∆A) = ∆A, so that A/ρA(∆A) = A/∆A is isomorphic to A, then (i) is equiv-
alent to (ii) and (v) is equivalent to (vi). Now let us drop the condition that A is semiprime. But A/ρA(∆A)
is semiprime, according to Proposition 8.2, hence, by the above, (ii) is equivalent to Max(A/ρA(∆A)) =
Spec(A/ρA(∆A)) and to the fact that L(A/ρA(∆A)) is a Boolean lattice, which, in turn, is equivalent to (iv)
by Proposition 9.7. But, as shown by Lemma 3.12 and Remark 4.5, Max(A/ρA(∆A)) = Spec(A/ρA(∆A)) iff
Max(A) ∩ [ρA(∆A)) = Spec(A) ∩ [ρA(∆A)) iff Max(A) = Spec(A), since Max(A) ⊆ Spec(A) ⊆ [ρA(∆A)).
References
[1] P. Agliano, Prime Spectra in Modular Varieties, Algebra Universalis 30 (1993), 581 -- 597.
[2] P. Agliano, A. Ursini, On Subtractive Varieties, II: General Properties, Algebra Universalis 36, Issue 2 (June
1996), 222 -- 259.
[3] P. Agliano, A. Ursini, On Subtractive Varieties, III: from Ideals to Congruences, Algebra Universalis 37
(1997), 296 -- 333.
[4] K. A. Baker, Primitive Satisfaction and Equational Problems for Lattices and Other Algebras, Trans. Amer.
Math. Soc. 190 (1974), 125 -- 150.
[5] R. Balbes, P. Dwinger, Distributive Lattices, University of Missouri Press, Columbia, Missouri, 1974.
[6] L. P. Belluce, Semisimple Algebras of Infinite Valued Logic and Bold Fuzzy Set Theory, Can. J. Math. 38,
No. 6 (1986), 1356 -- 1379.
[7] L. P. Belluce, Spectral Spaces and Non -- commutative Rings, Communications in Algebra 19, Issue 7 (1991),
1855 -- 1865.
[8] L. P. Belluce, Semisimple Algebras of Infinite Valued Logic and Bold Fuzzy Set Theory, Canadian Journal
of Mathematics 38 (1986), 1356 -- 1379.
[9] L. P. Belluce, A. DiNola, A. Lettieri, Subalgebras, Direct Products and Associated Lattices of MV -- algebras,
Glasgow Math. J. 34 (1992), 301 -- 307.
[10] W. J. Blok, D. Pigozzi, On the Structure of Varieties with Equationally Definable Principal Congruences I,
Algebra Universalis 15 (1982), 195 -- 227.
[11] T. S. Blyth, Lattices and Ordered Algebraic Structures, Springer -- Verlag London Limited, 2005.
[12] S. Burris, H. P. Sankappanavar, A Course in Universal Algebra, Graduate Texts in Mathematics, 78,
Springer -- Verlag, New York -- Berlin (1981).
[13] D. Bu¸sneag, D. Piciu, The Belluce -- lattice Associated with a Bounded Hilbert Algebra, Soft Computing 19,
Issue 11 (November 2015), 3031 -- 3042.
[14] J. L. Castiglioni, M. Menni, W. J. Zuluaga Botero, A Representation Theorem for Integral Rings and Its
Applications to Residuated Lattices, Journal of Pure and Applied Algebra 220 (2016), 3533 -- 3566.
[15] I. Chajda, A congruence Modular Variety that Is Neither Congruence Distributive Nor 3-permutable, Soft
Computing 17 (2013), 1467 -- 1469.
[16] D. Cheptea, Reticulation of the Hoops, in preparation.
[17] P. Crawley, R. P. Dilworth, Algebraic Theory of Lattices, Prentice Hall, Englewood Cliffs (1973).
[18] J. Czelakowski, The Equationally -- defined Commutator. A Study in Equational Logic and Algebra,
Birkhauser Mathematics, 2015.
[19] J. Czelakowski, Additivity of the Commutator and Residuation, Reports on Mathematical Logic 43 (2008),
109 -- 132.
[20] A. DiNola, G. Georgescu, L. Leu¸stean, Boolean Products of BL -- algebras, Journal of Mathematical Analysis
and Applications 251, Issue 1 (November 2000), 106 -- 131.
[21] R. Freese, R. McKenzie, Commutator Theory for Congruence -- modular Varieties, London Mathematical
Society Lecture Note Series 125, Cambridge University Press, 1987.
[22] N. Galatos, P. Jipsen, T. Kowalski, H. Ono, Residuated Lattices: An Algebraic Glimpse at Substructural Log-
ics, Studies in Logic and The Foundations of Mathematics 151, Elsevier, Amsterdam/ Boston /Heidelberg
/London /New York /Oxford /Paris /San Diego/ San Francisco /Singapore /Sydney /Tokyo, 2007.
[23] G. Georgescu, I. Voiculescu, Some Abstract Maximal Ideal -- like Spaces, Algebra Universalis 26 (1989),
90 -- 102.
[24] G. Georgescu, C. Mure¸san, Congruence Boolean Lifting Property, to appear in Journal of Multiple -- valued
Logic and Soft Computing, arXiv:1502.06907 [math.LO].
[25] G. Georgescu, C. Mure¸san, Going Up and Lying Over in Congruence -- modular Algebras, arXiv:1608.04985
[math.RA].
[26] G. Gratzer, General Lattice Theory, Birkhauser Akademie -- Verlag, Basel -- Boston -- Berlin (1978).
[27] G. Gratzer, Universal Algebra, Second Edition, Springer Science+Business Media, LLC, New York, 2008.
[28] P. Jipsen, C. Tsinakis, A Survey of Residuated Lattices, Ordered Algebraic Structures, Kluwer Academic
Publishers, Dordrecht, 2002, 19 -- 56.
[29] P. Jipsen, Generalization of Boolean Products for Lattice -- orderred Algebras, Annals of Pure and Applied
Logics 161, Issue 2 (November 2009), 224 -- 234.
[30] P. T. Johnstone, Stone Spaces, Cambridge Studies in Advanced Mathematics 3, Cambridge University Press,
Cambridge/London/New York/New Rochelle/Melbourne/Sydney, 1982.
[31] B. J´onsson, Congruence -- distributive Varieties, Math. Japonica 42, No. 2 (1995), 353 -- 401.
[32] A. Joyal, Le Th´eor`eme de Chevalley − Tarski et Remarques sur l'alg`ebre Constructive, Cahiers Topol.
G´eom. Diff´er., 16 (1975), 256 -- 258.
[33] J. Kaplansky, Commutative Rings, First Edition: University of Chicago Press, 1974; Second Edition: Polyg-
onal Publishing House, 2006.
[34] J. Koll´ar, Congruences and One -- element Subalgebras, Algebra Universalis 9, Issue 1 (December 1979),
266 -- 267.
[35] T. Kowalski, A. Ledda, F. Paoli, On Independent Varieties and Some Related Notions, Algebra Universalis
70, Issue 2 (October 2013), 107 -- 136.
[36] A. Ledda, F. Paoli, C. Tsinakis, Lattice -- theoretic Properties of Algebras of Logic, J. Pure Appl. Algebra
218, No. 10 (2014), 1932 -- 1952.
[37] L. Leu¸stean, The Prime and Maximal Spectra and the Reticulation of BL -- algebras, Central European
Journal of Mathematics 1 (2003), No. 3, 382 -- 397.
[38] L. Leu¸stean, Representations of Many -- valued Algebras, Editura Universitara, Bucharest, 2010.
[39] R. McKenzie and J. Snow, Congruence Modular Varieties: Commutator Theory and Its Uses, in Structural
Theory of Automata, Semigroups, and Universal Algebra, Springer, Dordrecht, 2005.
[40] W. DeMeo, The Commutator as Least Fixed Point of a Closure Operator, arXiv:1703.02764 [math.LO].
[41] C. Mure¸san, The Reticulation of a Residuated Lattice, Bull. Math. Soc. Sci. Math. Roumanie 51 (99), No.
1 (2008), 47 -- 65.
[42] C. Mure¸san, Algebras of Many -- valued Logic. Contributions to the Theory of Residuated Lattices, Ph. D.
Thesis, 2009.
[43] C. Mure¸san, Characterization of the Reticulation of a Residuated Lattice, Journal of Multiple -- valued Logic
and Soft Computing 16, No. 3 -- 5 (2010), Special Issue: Multiple -- valued Logic and Its Algebras, 427 -- 447.
[44] C. Mure¸san, Dense Elements and Classes of Residuated Lattices, Bull. Math. Soc. Sci. Math. Roumanie 53
(101), No. 1 (2010), 11 -- 24.
[45] C. Mure¸san, Further Functorial Properties of the Reticulation, Journal of Multiple-valued Logic and Soft
Computing 16, No. 1 -- 2 (2010), 177 -- 187.
[46] C. Mure¸san, Co -- Stone Residuated Lattices, Annals of the University of Craiova, Mathematics and Computer
Science Series 40 (2013), 52 -- 75.
[47] C. Mure¸san, Taking Prime, Maximal and Two -- class Congruences Through Morphisms, submitted,
arXiv:1607.06901 [math.RA].
[48] P. Ouwehand, Commutator Theory and Abelian Algebras, arXiv:1309.0662 [math.RA].
[49] Y. S. Pawar, Reticulation of a 0-distributive Lattice, Acta Universitatis Palackianae Olomucensis. Facultas
Rerum Naturalium. Mathematica 54, Issue 1 (2015), 121 -- 128.
[50] Y. S. Pawar, I. A. Shaikh, Reticulation of an Almost Distributive Lattice, Asian -- European Journal of
Mathematics 08, Issue 04 (December 2015).
[51] E. T. Schmidt, A Survey on Congruence Lattice Representations, Teubner -- Texte zur Mathematik, Leipzig
(1982).
[52] H. Simmons, Reticulated Rings, Journal of Algebra 66, Issue 1 (September 1980), 169 -- 192.
[53] A. Ursini, On Subtractive Varieties, V: Congruence Modularity and the Commutator, Algebra Universalis
43 (2000), 51 -- 78.
|
1509.02422 | 1 | 1509 | 2015-09-08T15:53:16 | Idempotent ideals and the Igusa-Todorov functions | [
"math.RA",
"math.RT"
] | Let $\Lambda$ be an artin algebra and $\mathfrak{A}$ a two-sided idempotent ideal of $\Lambda$, that is, $\mathfrak{A}$ is the trace of a projective $\Lambda$-module $P$ in $\Lambda$. We consider the categories of finitely generated modules over the associated rings $\Lambda/\mathfrak{A}, \Lambda$ and $\Gamma=\mathrm{End}_{\Lambda}(P)^{op}$ and study the relationship between their homological properties via the Igusa-Todorov functions. | math.RA | math | IDEMPOTENT IDEALS AND
THE IGUSA-TODOROV FUNCTIONS
A. GATICA, M. LANZILOTTA, M. I. PLATZECK
Abstract. Let Λ be an artin algebra and A a two-sided idempotent
ideal of Λ, that is, A is the trace of a projective Λ-module P in Λ. We
consider the categories of finitely generated modules over the associated
rings Λ/A, Λ and Γ = EndΛ(P )op and study the relationship between
their homological properties via the Igusa-Todorov functions.
]
.
A
R
h
t
a
m
[
1
v
2
2
4
2
0
.
9
0
5
1
:
v
i
X
r
a
1. Introduction
Throughout this paper we assume that Λ is an artin algebra and all Λ-
modules are in modΛ, the category of finitely generated left Λ-modules.
In [8] Igusa and Todorov introduced two functions φ and ψ which turned
out to be powerful tools to study the finitistic dimension of some classes of
algebras. On the other hand, associated to an idempotent ideal A of Λ, there
eP−−→ modΓ, where P is a
is an exact sequence of categories modΛ/A → modΛ
projective module such that A = τP Λ is the trace of P in Λ, Γ = EndΛ(P )op
and eP = HomΛ(P, −) is the evaluation functor. In [1] the authors studied
the relation between the homological properties of the three categories in-
volved: modΛ/A, modΛ and modΓ. Our objective in this paper is to study
the behaviour of the Igusa-Todorov functions in this situation. For a finitely
generated Λ-module X, we will denote φ(X) by φΛ
l (X), and the supremum of
these numbers for X in modΛ is the φl dimension of Λ, denoted by φldim(Λ).
Additionally, addX denotes the full subcategory of modΛ consisting of sum-
mands of finite direct sums of X.
First we consider the inclusion of modΛ/A in modΛ. To compare the values
of the Igusa-Todorov functions in a Λ/A-module X in both categories we need
the further assumption that the idempotent ideal A is a strong idempotent
ideal, in the sense defined in [1]. We recall that the ideal A is a strong
idempotent ideal if the morphism Exti
Λ(X, Y ) induced by
Λ/A(X, Y ) → Exti
Key words and phrases.
Idempotent ideals, Igusa-Todorov functions, homological di-
mensions, artin algebras.
2010 Mathematics Subject Classification. 16E10, 16G10.
The authors thank the financial support received from Universidad de la Rep´ublica, Mon-
tevideo, Uruguay, from Universidad Nacional del Sur, Bah´ıa Blanca, Argentina and from
CONICET, Argentina.
1
2
A. GATICA, M. LANZILOTTA, M. I. PLATZECK
the canonical isomorphism HomΛ/A(X, Y ) → HomΛ(X, Y ) is an isomorphism
for all i ≥ 0 and all X, Y in modΛ/A. We prove that φΛ/A
l (X) for
all X ∈ modΛ/A, whenever A is a strong idempotent ideal of finite projective
dimension. Thus in this case the φl dimension of Λ/A is bounded by the φl
dimension of Λ.
(X) ≤ φΛ
l
In order to compare the behaviour of the Igusa-Todorov functions under the
eP−−→ modΓ, we recall that eP induces an equivalence between the
functor modΛ
full subcategory of modΛ consisting of the Λ-modules X having a presentation
in addP , and modΓ. We prove that both functions φ and ψ are preserved
under eP for modules having a resolution in addP . As a consequence we
obtain that when all Λ-modules with a presentation in addP have also a
resolution in addP , then φldimΓ ≤ φldimΛ and ψldimΓ ≤ ψldimΛ (Theorem
4.7).
Then we obtain information about the φ dimension of Λ from the φ di-
mensions of the algebras Λ/A and Γ. We prove several inequalities, which are
interesting when either the global dimension of Λ/A or the global dimension
of Γ are finite.
To prove these results we use, in one hand, the characterization of the
Igusa-Todorov function φ in terms of the bifunctor Ext(−, −) given in [5]. On
the other hand, the full subcategory T of modΛ introduced in [1] consisting
of the modules T such that Exti
Λ(Λ/A, T ) = 0 for all i ≥ 1, is very useful for
our purposes. Consider the full subcategories P0 and P∞ of modΛ, where P0
consists of the modules whose projective cover is in addP , and P∞ of those
having a projective resolution in addP . We use the fact, proven in section 3,
that (P0, modΛ/A) is a torsion pair in modΛ whose properties are inherited
by the pair (P∞, modΛ/A) in the category T dual of T.
2. Preliminaries
Let Λ be an artin algebra, M and N in modΛ. We denote by τM N the
trace of M in N , that is, the submodule of N generated by the homomorphic
images of maps from M to N . Moreover, P0(M ), I0(M ) denote the projective
cover and injective envelope of M , and Ωn(M ), Ω−n(M ) the nth syzygy and
the nth cosyzygy of M , respectively. Finally, pdM denotes the projective
dimension of M and gldΛ stands for the global dimension of Λ.
We start by recalling some definitions and results from [1] which will be used
throughout the paper. Let A be an idempotent ideal of Λ, P0 the projective
cover of A, and P = Λe where e is an idempotent element of Λ such that
addP = addP0. Then A = ΛeΛ = τP Λ is the trace of P in Λ, mod Λ/A is
a Serre subcategory of modΛ and this inclusion induces an exact sequence of
eP−−→ modΓ, where Γ = EndΛ(P )op and eP =
categories modΛ/A → modΛ
HomΛ(P, −) is the evaluation functor.
IDEMPOTENT IDEALS AND IGUSA-TODOROV FUNCTIONS
3
Let P/rP ≃ S1 ⊕ · · ·⊕ Sr, with Si simple for all i, so that P = P0(S1 ⊕ · · ·⊕
Sr), and let I = I0(S1 ⊕ · · · ⊕ Sr). To compare the homological properties of
modΛ and modΓ, full subcategories Pk and Ik were introduced in [1] for any
k ≥ 0. These subcategories will be useful for our purposes, and are defined
as follows: Pk is the full subcategory of modΛ consisting of the Λ-modules X
having a projective resolution · · · → P1 → P0 → X → 0 with Pi in addP for
0 ≤ i ≤ k. The full subcategory Ik is defined dually.
Then HomΛ(P, −) induces equivalences P1 → modΓ and I1 → modΓ.
Λ(X, Y ) →
Moreover, the morphism of connected sequences of functors Exti
Exti
whenever X ∈ P(k+1) or Y ∈ I(k+1) (Theorem 3.2, [1]).
Γ((P, X), (P, Y )) induced by HomΛ(P, −) is an isomorphism for i = 1, · · · , k,
We next turn our attention to the definition of the Igusa-Todorov functions,
defined in [8]. Let K0 denote the abelian group generated by all symbols [M ],
where M in modΛ, modulo the relations a) [C] = [A] + [B] if C ≃ A ⊕ B
and b) [P ] = 0 if P is projective. That is, K0 is the free abelian group
generated by the isomorphism classes of indecomposable finitely generated
nonprojective Λ-modules. Let Ω : K0 → K0 denote the group homomorphism
induced by the syzygy, that is, Ω([M ]) := [Ω(M )], and let < addM > be the
subgroup of K0 generated by the indecomposable sumands of M . When we
apply the homomorphism Ω to this subgroup the rank does not increase: rank
Ω(< addM >) ≤ rank < addM >, and there is then an integer n such that
Ω : Ωs(< addM >) → Ωs+1(< addM >) is an isomorphism for all s ≥ n.
Then the Igusa-Todorov functions φ and ψ are defined as follows: φ(M ) is
the smallest non-negative integer n with this property, and ψ(M ) := φ(M ) +
sup{pdX X is a direct summand of Ωφ(M)(M ) with pdX < ∞}. Since we
will need also the dual notions, we will denote the Igusa-Todorov functions φ
and ψ by φl and ψl, respectively. Using the cozyzygy we can define φr(M ) and
ψr(M ) in an analogous way. Then φr(M ) = φl(DM ) and ψr(M ) = ψl(DM ),
for any M in modΛ.
Let φldimΛ = sup{φl(M ) M in modΛ} and ψldimΛ = sup{ψl(M ) M in
modΛ}. Moreover, for a subcategory X of modΛ we indicate by φldimX and
ψldimX the supremum of the sets {φl(X) X in X } and {ψl(X) X in X },
respectively. Analogous notions are defined for φr and ψr.
We will also need the characterization of the function φ in terms of the
Λ(−, −) given in [5]. We recall first that a pair (X, Y ) of objects
bifunctor Exti
in addM is called d-division of M if the following three conditions hold:
(a) add(X) ∩ add(Y ) = 0
(b) Extd
(c) Extd+1
Dually, a pair (X, Y ) of objects in addM is called d-injective division of M
Λ(X, −) 6≃ Extd
Λ (X, −) ≃ Extd+1
Λ (Y, −) in modΛ.
Λ(Y, −) in modΛ
if (a) and the following two conditions hold:
(b') Extd
Λ(−, X) 6≃ Extd
Λ(−, Y ) in modΛ
4
A. GATICA, M. LANZILOTTA, M. I. PLATZECK
Λ (−, X) ≃ Extd+1
(c') Extd+1
Then φl(M ) =max ({d ∈ N :there is a d-division of M }∪{0}) ([5], Theorem
3.6), and φr(M ) =max ({d ∈ N :there is a d-injective division of M } ∪ {0}).
Λ (−, Y ) in modΛ.
3. Torsion theories associated to an idempotent ideal
It is interesting to notice that the idempotent ideal A determines two tor-
sion pairs (modΛ/A, I0) and (P0, modΛ/A) in modΛ, in the sense defined by
Dickson in [4], as we state in the following proposition.
Proposition 3.1. Let A be an idempotent ideal of Λ, A = τP Λ, where P is
a projective Λ-module. Then
(a) (modΛ/A, I0) is a torsion pair in modΛ .
(b) (P0, modΛ/A) is a torsion pair in modΛ.
Proof. (a) To prove this we observe that I0 consists of the modules with socle
in add(S1 ⊕ · · · ⊕ Sr), and that a Λ-module M is in modΛ/A if and only
if S1, · · · , Sr are not composition factors of M . Then HomΛ(M, Y ) = 0 for
M ∈ modΛ/A and Y ∈ I0. Moreover, if HomΛ(M, Y ) = 0 for all Y ∈ I0,
then in particular HomΛ(M, I0(S1 ⊕ · · · ⊕ Sr)) = 0, so that S1, · · · , Sr are
not composition factors of M, and M is thus a Λ/A-module. Finally, suppose
that HomΛ(M, Y ) = 0 for each M ∈ modΛ/A. Then HomΛ(S, Y ) = 0 for any
simple S not isomorphic to S1, · · · , Sr. Thus the only simples in the socle of Y
are amongst S1, · · · , Sr, and therefore Y ∈ I0. This shows that (modΛ/A, I0)
is a torsion pair in modΛ.
The statement (b) follows by duality.
(cid:3)
In the sequel we will consider the full subcategory T of modΛ introduced
and studied in section 5 of [1], consisting of the modules T such that the group
Λ(Λ/A, T ) = 0 for all i ≥ 1. Dually, we define the subcategory T = D(TΛ)
Exti
consisting of the Λ-modules X such that Exti
Λ(X, D(Λ/AΛ)) = 0 for all i ≥ 1.
The notion of torsion pairs in abelian categories defined by Dickson was
extended to pretriangulated categories by Beligianis and Reiten (see [2], Ch.
II, Definition 3.1). Additive categories with kernels and cokernels are examples
of pretriangulated categories, as shown in section 1, Example 2 of the same
paper, and in this case torsion pairs are defined as follows.
Definition 3.2. ([2]) A pair of subcategories (X , Y) in an additive category
C with kernels and cokernels and closed under isomorphisms is a torsion pair
if the following conditions hold:
T1) HomC(X, Y ) = 0 for all X ∈ X , Y ∈ Y
T2) For every C ∈ C there is an exact sequence 0 → XC → C → YC → 0
with XC ∈ X , YC ∈ Y (glueing sequence).
IDEMPOTENT IDEALS AND IGUSA-TODOROV FUNCTIONS
5
We observe that for the torsion pairs (modΛ/A, I0) and (P0, modΛ/A)
above considered the glueing sequences for a module X in modΛ are 0 →
τΛ/AX → X → X/τΛ/AX → 0 and 0 → τAX → X → X/τAX → 0 respec-
tively.
We now turn our attention to the subcategories T and T of modΛ, and
study the restriction of these torsion pairs to T and T respectively, under the
assumption that the ideal A is strong idempotent.
Proposition 3.3. Let A be a strong idempotent ideal of Λ. Then
(a) I∞ = I0 ∩ T, and the pair (modΛ/A, I∞) of subcategories of T satisfies
conditions T1) and T2) of Definition 3.2 of torsion pair. Moreover,
for T in T the glueing sequence is 0 → τΛ/AT → T → T /τΛ/AT → 0.
(b) P∞ = P0 ∩ T, and the pair (modΛ/A, P∞) of subcategories of T satis-
fies conditions T1) and T2) of Definition 3.2 of torsion pair. More-
over, for T ∈ T the glueing sequence is 0 → τAT → T → T /τAT → 0.
Proof. Since I∞ ⊆ I0 then condition T1) in the definition of torsion pair holds.
Assume now that A is a strong idempotent ideal. Then modΛ/A ⊂ T. In
fact, if X ∈ modΛ/A then Exti
Λ/A(Λ/A, X) = 0 for all i ≥ 0,
so X ∈ T. On the other hand, we know that a module Y is in I∞ if and only
if Exti
Λ(Λ/A, Y ) = 0 for all i ≥ 0, by [1], Proposition 2.6. Thus, it follows
from the definition of T that I∞ = I0 ∩ T.
Λ(Λ/A, X) ∼= Exti
Therefore, for T ∈ T the exact sequence 0 → τΛ/AT → T → T /τΛ/AT → 0
has τΛ/AT in modΛ/A and T /τΛ/AT in I∞ and is then a glueing sequence for
T . This proves condition T2) and ends the proof of (a). The proof of (b) is
similar.
(cid:3)
Remark 3.4. Though we do not know wether the subcategories T and T have
kernels and cokernels, we observe that the category T is not in general abelian
as the following simple example shows. Let Λ be the path algebra of the quiver
1 → 2, and P = S2, the simple projective module associated to the vertex
2. Then, A = τP Λ ≃ S2 ⊕ S2 is a projective Λ module and therefore it is
a strong idempotent ideal. Moreover, T = add { S1
S2 , S1 } and Λ/A ≃ S1.
f
Consider the exact sequence 0 → S2 → S1
−→ S1 → 0. Then f is a map in
S2
T, and KerT(f ) = 0, because there are no nonzero maps from objects in T to
S2. Thus CokerT(KerT(f )) = CokerT(0 → S1
S2 ). However,
KerT(CokerT(f )) = KerT(S1 → 0) = (S1
S2 ) = ( S1
id−→ S1).
id−→ S1
S2
In connection with the torsion pairs and subcategories above considered we
prove two technical lemmas which will be useful throughout the paper.
Lemma 3.5. Let A be an idempotent ideal. Then
6
A. GATICA, M. LANZILOTTA, M. I. PLATZECK
(a) HomΛ(P, X) ≃ HomΛ(P, X/τΛ/AX) for all X ∈ modΛ.
(b) Extj
Λ(−, X)P∞ ≃ Extj
Λ(−, X/τΛ/AX)P∞ for all j ≥ 0 and X ∈
modΛ.
(c) HomΛ(X, I) ≃ HomΛ(τAX, I) for all X ∈ modΛ.
(d) Extj
Λ(X, −)I∞ ≃ Extj
Λ(τAX, −)I∞ for all j ≥ 0 and X ∈ modΛ.
Proof. (a) The result follows directly by applying the exact functor HomΛ(P, −)
to the exact sequence 0 → τΛ/AX → X → X/τΛ/AX → 0.
(b) We recall from [1], Theorem 3.2 c), that there is a functorial isomor-
Γ((P, −), (P, Y ))P∞ for all Y in modΛ. The result
Λ(−, Y )P∞ ≃ Extj
phism Extj
follows now using (a).
By duality we obtain the statements (c) and (d).
(cid:3)
When we further assume that the ideal A is strong idempotent we get the
following result.
Lemma 3.6. Let A be a strong idempotent ideal. Then
(a) Extj
(b) Extj
Extj
(c) Extj
(d) Extj
Extj
Λ(−, X2) with j ≥ 0 implies
Λ(−, X/τΛ/AX)modΛ/A = 0 for all X ∈ T and j ≥ 1.
Λ(−, X1) ≃ Extj
Λ(−, X1/τΛ/AX1)T ≃ Extj
Λ(τAX, −) = 0 for all X ∈ T and j ≥ 1.
Λ(X1, −) ≃ Extj
Λ(τAX1, −)T ≃ Extj
Λ(τAX2, −)T for all X1, X2 ∈ T.
Λ(X2, −) with j ≥ 0 implies
Λ(−, X2/τΛ/AX2)T for all X1, X2 ∈ T.
Proof. (a) Let X in T and Z in modΛ/A. Then X/τΛ/AX ∈ I∞ and using (d)
of the previous lemma we conclude that Extj
0, since τAZ = 0 because Z is a Λ/A-module.
Λ(Z, X/τΛ/AX) ≃ Extj
Λ(τAZ, X/τΛ/AX) =
(b) Let X1, X2 ∈ T be such that Extj
Λ(−, X1) ≃ Extj
Λ(−, X2). So L1 =
X1/τΛ/AX1, L2 = X2/τΛ/AX2 ∈ I∞. Then we obtain from (a) that Extj
0 for i = 1, 2 and for all j ≥ 1.
Let Z ∈ T. Applying the functor HomΛ(−, Li) to the glueing sequence
Λ(−, Li)modΛ/A =
0
/ τAZ
/ Z
/ Z/τAZ
/ 0
the corresponding long exact sequence yields isomorphisms
Extj
Λ(Z, Li)≃Extj
Λ(τAZ, Li),
for i = 1, 2 and j ≥ 1.
On the other hand, by (b) of the previous lemma we know that Extj
Λ(τAZ, Li) for i = 1, 2, j ≥ 1 because A is a strong idempotent ideal, so Z
Extj
in T implies that Z is a Λ/A-module. Then, in the commutative diagram
Λ(τAZ, Xi) ≃
/
/
/
/
IDEMPOTENT IDEALS AND IGUSA-TODOROV FUNCTIONS
7
Extj(Z, L1)
Extj(τAZ, L1)
Extj(Z, L2)
/ Extj(τAZ, L2)
the horizontal arrows and the right vertical arrow are isomorphisms when
j ≥ 1. This proves the left vertical arrow is also an isomorphism in this case,
as desired.
Finally, statements (c) and (d) follow from (a) and (b) by duality.
(cid:3)
4. Main results
Next we turn our attention to the functions φ and ψ defined by Igusa and
Todorov. We are going to use the characterization of these functions in terms
of the functor Ext given in [5]. We start with two lemmas comparing the
behaviour of this functor in modΛ and in modΛ/A.
Lemma 4.1. Let A be an idempotent ideal such that pd(ΛΛ/A) = r < ∞.
Let X1, X2 ∈ modΛ/A and t ≥ 1. Then Extt
Λ/A(X2, −)
implies Extt+r
Λ/A(X1, −) ≃ Extt
Λ (X1, −) ≃ Extt+r
Λ (X2, −).
Proof. Note first that Extt+r
Λ(Xj, Ω−r(−)), for j = 1, 2. Since
Ω−r(modΛ) ⊂ T, from Lemma 5.5, [1], using Proposition 1.1, [1], it follows
that Exti
Λ(Xj, Ω−r(−)), for all 1 ≤ i, and j =
1, 2.
Λ/A(Xj, τΛ/A(Ω−r(−))) ≃ Exti
Λ (Xj, −) ≃ Extt
Λ/A(X1, τΛ/A(Ω−r(−))) ≃
(cid:3)
Λ (X1, −) ≃ Extt
Now Extt+r
Λ/A(X2, τΛ/A(Ω−r(−))) ≃ Extt+r
Extt
Λ(X1, Ω−r(−)) ≃ Extt
Λ (X2, −)
Lemma 4.2. Let A be an idempotent ideal such that pd(Λ/AΛ) = r < ∞.
Let X1, X2 ∈ modΛ/A and t ≥ 1. Then Extt
Λ/A(−, X2)
implies Extt+r
Λ/A(−, X1) ≃ Extt
Λ (−, X1) ≃ Extt+r
Λ (−, X2).
Proof. It follows from Lemma 4.1 by duality, using that Extj
Extj
Λ/Aop (DY, D(−)).
Λ/A(−, Y ) ≃
(cid:3)
We prove next that when the ideal A is a strong idempotent ideal of fi-
nite projective dimension then the φ dimension of the factor algebra Λ/A is
bounded by the φ dimension of Λ.
Theorem 4.3. Let A be a strong idempotent ideal of Λ such that pd(ΛΛ/A) =
r < ∞. Then
(a) φΛ/A
(b) φldimΛ/A ≤ φldimΛ.
l (X) for all X ∈ modΛ/A.
(X) ≤ φΛ
l
/
/
/
8
A. GATICA, M. LANZILOTTA, M. I. PLATZECK
Λ/A(X1, −) 6≃ Extd
Λ/A(X2, −). Since A is a strong idempotent ideal then Extd
Λ(X2, −). On the other hand, we know that φΛ
Proof. Let d a positive integer and assume that X = X1 ⊕X2 is a d-division of
the Λ/A-module X. This is, Extd
Extd+1
Λ(X1, −) 6≃
Extd
l (X) = max({d ∈ N : there
is a d-division of X in modΛ}∪{0}), by Theorem 3.6 in [5]. Thus to prove that
l (X) ≥ d, it is enough to find l > d such that Extl
φΛ
Λ(X2, −).
In fact, if l0 is minimal with this property, then X = X1 ⊕ X2 is an (l0 − 1)-
division of the Λ-module X.
Λ/A(X2, −) and Extd+1
Λ(X1, −) ≃ Extl
Λ/A(X1, −) ≃
Since we assumed that Extd+1
4.1, we obtain that Extd+r+1
Λ
Hence, l = d + r + 1 > d satisfies Extl
Λ(X1, −) ≃ Extl
Λ
Λ/A(X1, −) ≃ Extd+1
(X1, −) ≃ Extd+r+1
(X2, −).
Λ/A(X2, −) then, using Lemma
found l as required, proving that φΛ
immediately.
Λ(X2, −). So we
l (X) ≥ d. This proves (a), and (b) follows
(cid:3)
We observe that φl can be replaced by φr in the previous theorem, since
φr(X) = φl(DX).
These results apply to any convex subcategory ∆ of a quiver algebra Λ =
kQ/I, where Q is a finite quiver and I is an admissible ideal of the path
algebra kQ. That is, ∆ = kQ′/(kQ′ ∩ I), where Q′ is a full convex subquiver
of Q. In this case ∆ = Λ/A, where A is the trace of the projective module
Pi. In this situation it is known that A is a strong idempotent
ideal ([9], Ch. II, Lemma 3.7) and we obtain the following corollary.
P = Li /∈Q′
0
Corollary 4.4. Let ∆ be a full convex subcategory of the quiver algebra Λ,
and let A be the idempotent ideal such that ∆ = Λ/A. If A has finite projective
dimension then φldim∆ ≤ φldimΛ.
Now we turn our attention to Γ-modules. We study the behaviour of both
Igusa-Todorov functions φ and ψ under the functor HomΛ(P, −) : modΛ →
modΓ restricted to the subcategories P∞ and I∞ of modΛ.
Proposition 4.5. For a Λ-module Y ∈ P∞ the following properties hold:
(a) there exists a d-division of Y if and only if there exists a d-division of
HomΛ(P, Y ).
l (Y ) = φΓ
l (Y ) = ψΓ
(b) φΛ
(c) ψΛ
l (HomΛ(P, Y )).
l (HomΛ(P, Y )).
Proof. (a) Since HomΛ(P, −) : modΛ → modΓ induces an equivalence of cat-
egories P1 → modΓ and Y ∈ P∞ ⊆ P1, it follows that Y = Y1 ⊕ Y2 if and only
if HomΛ(P, Y ) = HomΛ(P, Y1)⊕HomΛ(P, Y2). The statement follows from the
fact that Y1, Y2 ∈ P∞ implies Exti
Γ(HomΛ(P, Yj ), HomΛ(P, −)),
for j = 1, 2 and for all i ≥ 0 ([1],Theorem 3.2,(c)).
Λ(Yj , −) ≃ Exti
IDEMPOTENT IDEALS AND IGUSA-TODOROV FUNCTIONS
9
(b) This is a direct consequence of (a), using that φΛ
l (Y ) = n if and only if
n =max({ d ∈ N : there exists a d-division of Y } ∪ 0) by Theorem 3.6 in [5].
l (X) and l = pdZ1,
where Z1 is the largest summand of Ωn(X) of finite projective dimension. We
write Ωn(X) = Z1 ⊕ Z2.
l (X) = n + l, with n = φΓ
(c) Let X ∈ modΓ. Then ψΓ
Let Y ∈ P∞ ⊆ P1 be such that X = HomΛ(P, Y ). Since φΛ
l (X),
we only need to prove that l is the projective dimension of largest summand
Y1 of Ωn(Y ) of finite projective dimension. Let Ωn(Y ) = Y1 ⊕ Y2 and let
l (Y ) = φΓ
· · · Pn+1 → Pn → · · · → P0 → Y → 0
be a minimal projective resolution of Y in modΛ. Since Y ∈ P∞ then
· · · → HomΛ(P, Pn) → · · · → HomΛ(P, P0) → HomΛ(P, Y ) → 0
is a minimal projective resolution of X = HomΛ(P, Y ) in modΓ, as follows
from [1], Lemma 3.1.
Therefore Ωn(X) ≃ HomΛ(P, Ωn(Y )). This is, Z1 ⊕ Z2 ≃ HomΛ(P, Y1) ⊕
HomΛ(P, Y2). Since Y ∈ P∞, then Ωn(Y ) and all direct summands of Ωn(Y )
are also in P∞. Thus pdL = pdHomΛ(P, L) for any summand L of Ωn(Y )
(see [1], Corollary 3.3). From this we conclude that the projective dimensions
of the largest summands of finite projective dimension of Ωn(X) and Ωn(Y )
coincide, as desired.
(cid:3)
We state the corresponding result for the φ-injective dimension in the next
proposition.
Proposition 4.6. For a Λ-module Y ∈ I∞ the following properties hold:
(a) there exists a d-division of Y if and only if there exists a d-division of
HomΛ(P, Y ).
r (Y ) = φΓ
r (Y ) = ψΓ
(b) φΛ
(c) ψΛ
r (HomΛ(P, Y )).
r (HomΛ(P, Y )).
Proof. The result follows using that HomΛ(P, −) : I1 → modΓ is an equiva-
lence of categories, the fact that Exti
Γ(HomΛ(P, −), HomΛ(P, Y ))
for all Y ∈ I∞ and i ≥ 0 ([1], Theorem 3.2, (b)), and dualizing arguments in
the proof of the previous proposition.
(cid:3)
Λ(−, Y ) ≃ Exti
Since HomΛ(P, −) : modΛ → modΓ induces equivalences of categories
P1 → modΓ and I1 → modΓ, the previous propositions yield the following
result.
Theorem 4.7.
ψldimΛ.
(a) If P1 = P∞ then φldimΓ ≤ φldimΛ and ψldimΓ ≤
(b) If I1 = I∞ then φrdimΓ ≤ φrdimΛ and ψldimΓ ≤ ψldimΛ.
Our next objective is to find bounds for the φ dimension of Λ in terms of
the φ dimensions of Λ/A and Γ.
10
A. GATICA, M. LANZILOTTA, M. I. PLATZECK
We start with the case when the global dimension of Γ is finite. Let T be
the subcategory of modΛ considered in the previous section, consisting of the
modules T such that Exti
Λ(Λ/A, T ) = 0 for all i ≥ 1. We observe first that
bounds for the function φ in the subcategory T will give us bounds for φ in
mod Λ, as the following lemma shows.
Lemma 4.8. Let A be an idempotent ideal. Then
(a) φrdim(Λ) ≤ pdΛ(Λ/A) + φrdim(T)
(b) φldim(Λ) ≤ pd(Λ/A)Λ + φldim(T).
Proof. If pdΛ(Λ/A) = ∞ there is nothing to prove. Assume pdΛ(Λ/A) = t <
∞ and let X ∈ modΛ. Then by Lemma 5.5 in [1], Ω−t(X) ∈ T. The lemma
follows now by repeated use of the dual of the inequality of Lemma 1.3 in [7],
φr(X) ≤ t + φr(Ω−t(X) . This proves (a), and (b) follows by duality.
(cid:3)
Observe now that when A is a strong idempotent ideal then A is in P∞ ([1],
Theorem 2.1'), so pdΛ(A) ≤ pdΓ(HomΛ(P, A)) ≤ gldΓ. Thus pdΛ(Λ/A) ≤
gldΓ + 1. Since being a strong idempotent ideal is a symmetric condition we
obtain that pd(Λ/A)Λ ≤ gldΓ + 1, as observed in [1] at the end of section 5.
Proposition 4.9. Let A be a strong idempotent ideal. Then
φrdim(T) ≤ max{ gld(Γ) + 1, φrdim(Λ/A) + pd(Λ/A)Λ }.
Proof. Let now r = pd(Λ/A)Λ, T in T and consider the glueing sequence
0 → τΛ/AT → T → T /τΛ/AT → 0 in T. Since T /τΛ/AT is in I∞, we know
by [1], Corollary 3.3 b) that injdimΛT /τΛ/AT = injdimΓHomΛ(P, T /τΛ/AT ).
Then, the corresponding long exact sequence of functors yields isomorphisms
of functors δi : Exti
Λ(−, T ) for i > gldΓ + 1.
Λ(−, τΛ/AT ) → Exti
r (T ) ≤ φΛ/A
r
(τΛ/AT ) + r, whenever φΛ
It is enough to show that φΛ
r (T ) >
gldΓ + 1. With this purpose we assume that d = φΛ
r (T ) > gldΓ + 1 and
that T = T1 ⊕ T2 is a d-injective-division in modΛ. We start by proving that
τΛ/AT = τΛ/AT1 ⊕ τΛ/AT2 is a j-injective division in modΛ/A, for some j such
that d + 1 ≤ j + r. Since T = T1 ⊕ T2 is a d-injective-division of T , then
Extd
Λ (−, T2). Therefore,
since T1 and T2 are also in T we have that Extd
Λ(−, τΛ/AT2)
and Extd+1
Λ (−, τΛ/AT2) in modΛ, by the isomorphisms
above.
Λ (−, τΛ/AT1) ≃ Extd+1
Λ (−, T1) ≃ Extd+1
Λ(−, τΛ/AT1) 6≃ Extd
Λ(−, T1) 6≃ Extd
Λ(−, T2) and Extd+1
Now, since A is a strong idempotent ideal we deduce from the last isomor-
phism that Extd+1
Λ/A(−, τΛ/AT1) ≃ Extd+1
Λ/A(−, τΛ/AT2) in modΛ/A.
On the other hand, pd(Λ/A)Λ ≤ gldΓ + 1 as we observed just before the
statement of the proposition. Since we assumed that gld(Γ) + 1 < d we
obtain that pd(Λ/A)Λ + 1 = r + 1 ≤ d. We conclude then, from Lemma 4.2
that Extd−r
Λ/A(−, τΛ/AT2) in modΛ/A. This fact and the
isomorphism Extd+1
Λ/A(−, τΛ/AT2) obtained above imply
Λ/A(−, τΛ/AT1) ≃ Extd+1
Λ/A(−, τΛ/AT1) 6≃ Extd−r
IDEMPOTENT IDEALS AND IGUSA-TODOROV FUNCTIONS
11
that τΛ/AT = τΛ/AT1 ⊕ τΛ/AT2 is a j-injective division in modΛ/A, for some
j such that d − r ≤ j ≤ d. Thus d ≤ j + r ≤ φrdimΛ/A + r. This proves that
φr(T ) ≤ φrdimΛ/A + r, provided gldΓ + 1 > d = φr(T ), and ends the proof
of the proposition.
(cid:3)
Proposition 4.10. Let A be a strong idempotent ideal. Then
φrdim(Λ) ≤ pdΛ(Λ/A) + max{ gld(Γ) + 1, pd(Λ/A)Λ + φrdim(Λ/A) }.
Proof. The result follows from Lemma 4.8 and Proposition 4.9.
(cid:3)
Corollary 4.11. Let A be a strong idempotent ideal. Assume gld(Γ) < ∞,
then φrdim(Λ/A) is finite if and only if φrdim(Λ) is finite.
Since convex subalgebras of Λ are obtained as factors of Λ by a strong
idempotent ideal, the previous results apply to them. In particular, we obtain
the following corollary.
Corollary 4.12. Let ∆ be a full convex subalgebra of Λ, and let A = τP (Λ)
be the idempotent ideal such that ∆ = Λ/A. If Γ has finite global dimension,
then φrdim(∆) is finite if and only if φrdim(Λ) is finite.
Example 4.13. Let Λ = (cid:18) A 0
M B(cid:19), where A and B are artin algebras,
gld(B) < ∞ and M is a B-A-bimodule. Then φrdim(Λ) is finite if and
only if φrdim(A) is finite.
In particular we obtain that a one point co-extension of A has finite φrdim
if and only if A does.
Next we illustrate the previous Proposition with the following example.
Example 4.14. Let Λ be the algebra given by the quiver Q
β
γ
@
3
θ
1
δ
/ 4
α
µ
❃❃❃❃❃❃❃❃
ǫ
2
5
with relations αβ = βα = 0, µγ = 0, δγ = 0, ǫµ = 0. Let P = P3 ⊕ P4 ⊕ P5,
and let A = τP (Λ). Then A ≃ P3 ⊕P 3
5, so pdA = 1. Moreover, since
there is an exact sequence 0 → P4 → P5 → S5 → 0 we obtain that S5 ∈ P∞,
so A ∈ P∞ and is thus a strong idempotent. Then the quiver of Λ/A is
4 ⊕P5 ⊕S2
β
α
/ 2
1
/
/
o
o
@
/
o
o
/
o
o
12
A. GATICA, M. LANZILOTTA, M. I. PLATZECK
with radical square zero, so Λ/A is selfinjective and therefore φrdimΛ/A = 0,
by [6]. On the other hand Γ is the hereditary algebra with quiver
and we get
3
θ
/ 4
5
ǫ
φrdim(Λ) ≤ 2 + max{ 1 + 1, 2 + 1 } = 5.
We now turn our attention to the case when gld(Λ/A) is finite.
Proposition 4.15. Let A be a strong idempotent ideal and assume that
gld(Λ/A) is finite. Then
(a) φr(T ) ≤ max{ gld(Λ/A) + pd(Λ/A)Λ + 1, φrdim(T/τΛ/AT) } for any
T ∈ modΛ.
(b) φrdim(T) ≤ max{ gld(Λ/A) + pd(Λ/A)Λ + 1, φrdim(Γ) }.
Proof. Let s = gld(Λ/A), r = pd(Λ/A)Λ and let X be a Λ/A-module.
We claim that Exti
we know that Extj
Lemma 4.2, for j ≥ s + 1. So the claim holds.
Λ(−, X) = 0 for all i ≥ s+r+1. In fact, since gldΛ/A = s
Λ (−, X) = 0 by
Λ/A(−, X) = 0 for all j ≥ s + 1. Then Extj+r
Let now T in modΛ and consider the sequence 0 → τΛ/AT → T →
T /τΛ/AT → 0. The corresponding long exact sequence of functors yields
an isomorphism of functors δi : Exti
Λ(−, T /τΛ/AT ) for each
i ≥ s + r + 1.
Λ(−, T ) → Exti
We prove next that a d-division of T in modΛ yields a d-division T /τΛ/AT
in modΛ.
Λ(−, T1) 6≃ Extd
Λ(−, T2) and Extd+1
In fact, let d = φrdim(T ) and let T = T1 ⊕ T2 be a d-division of T .
This is, Extd
Λ (−, T2).
Assume now d ≥ s + r + 1. The functorial isomorphisms Exti
Λ(−, Tk) ≃
Exti
Λ(−, Tk/τΛ/ATk), k = 1, 2 and i ≥ s + r + 1 induced by δi show that
T /τΛ/AT = T1/τΛ/AT1 ⊕ T2/τΛ/AT2 is a d-division of T /τΛ/AT in modΛ.
Therefore, φΛ
l (T /τΛ/AT ) ≥ d. This ends the proof of (a).
Λ (−, T1) ≃ Extd+1
Assume finally that T ∈ T. Therefore T /τΛ/AT ∈ I∞. We know by Propo-
r HomΛ(P, T /τΛ/AT ) ≥ d . This proves that φrdimT ≤
(cid:3)
sition 4.6 (b) that φΓ
max {s + r + 1, φrdimΓ}.
Proposition 4.16. Let A be a strong idempotent ideal. Then φrdim(Λ) ≤
pdΛΛ/A + max{ gld(Λ/A) + pd(Λ/A)Λ + 1, φrdim(Γ) }
Proof. If gldΛ/A = ∞ there is nothing to prove. If gldΛ/A is finite, the result
follows from the previous proposition and Lemma 4.8
(cid:3)
Next we obtain another bound for φldim(Λ) with different methods.
/
o
o
IDEMPOTENT IDEALS AND IGUSA-TODOROV FUNCTIONS
13
Lemma 4.17. Let A be a strong idempotent ideal. Let 0 → X → Y → Z → 0
be an exact sequence in modΛ with X ∈ P∞ and Z ∈ modΛ/A such that
pdΛ/AZ is finite. Then
(a) Ωn(Y ) ∈ P∞ for n ≥ pdΛ/AZ.
(b) φΛ
l (Y ) ≤ pdΛ/AZ + φldimΓ.
Proof. (a) Let · · · Pn → · · · → P2 → P1 → P0 → X → 0 and · · · Qn → · · · →
Q2 → Q1 → Q0 → Z → 0 be minimal projective resolutions in modΛ.
Since A is a strong idempotent ideal then · · · Qn/AQn → · · · → Q2/AQ2 →
Q1/AQ1 → Q0/AQ0 → Z → 0 is a minimal projective resolution of Z in
modΛ/A (Theorem 1.6 iii) in [1]). Then Qn/AQn = 0 for n > s = pdΛ/AZ.
So Qn = AQn = τP Qn is in addP for n > s and therefore Ωn(Z) ∈ P∞ for
n > s.
On the other hand, we assumed that X is in P∞, so that Ωn(X) is also in
P∞. Let n > s. Since · · · Pr ⊕ Qr → · · · → P1 ⊕ Q1 → P0 ⊕ Q0 → Y → 0 is a
projective resolution of Y and Pr ⊕ Qr ∈ addP for r ≥ n, then Ωn(Y ) ∈ P∞.
This proves (a).
(b) Let s = pdΛ/AZ. By a) we know that Ωn(Y ) is in P∞, for all n ≥ s.
l (HomΛ(P, Ωs(Y )))+s ≤ φldimΓ+s, where
Thus φΛ
the first inequality is given by Lemma 1.3 in [7], and the equality follows from
Proposition 4.5 (b).
(cid:3)
l (Ωs(Y ))+s = φΓ
l (Y ) ≤ φΛ
Proposition 4.18. Let A be a strong idempotent ideal. Then
(a) φldim(T) ≤ gld(Λ/A) + φldimΓ.
(b) φldim(Λ) ≤ pd(Λ/A)Λ + gld(Λ/A) + φldim(Γ).
Proof. (a) Taking supremum on T ∈ T and using (b) of the previous lemma
applied to the glueing sequence 0 → τAT → T → T /τAT → 0, we get
φldim(T) ≤ φldimΓ + gld(Λ/A).
(b) If gldΛ/A = ∞ there is nothing to prove.
If gldΛ/A is finite, by
Lemma 4.8 (b), we get that φldim(Λ) ≤ pd(Λ/A)Λ + φldim(T) ≤ pd(Λ/A)Λ +
gld(Λ/A) + φldim(Γ).
(cid:3)
References
[1] M. Auslander, M. I. Platzeck, G. Todorov. Homological theory of idempotent ideals.
Trans Am. Math. Soc., vol. 332, n. 2, 667-692, (1992).
[2] A. Beligiannis, I. Reiten. Homological and homotopical aspects of torsion theories.
Mem. Amer. Math. Soc., vol. 188, n. 883, viii+207 pp., (2007).
[3] H. Cartan, S. Eilenberg. Homological algebra. Princeton Landmarks in Mathematics.
Princeton University Press, xvi+390 pp., (1999).
[4] S. E. Dickson. A torsion theory for abelian categories. Trans. Amer. Math. Soc. 121,
223-235, (1966).
[5] S. Fernandes, M. Lanzilotta, O. Mendoza. The φ-dimension: a new homological mea-
sure. Algebras and Representation Theory vol. 18,(2), 463-476, (2015).
14
A. GATICA, M. LANZILOTTA, M. I. PLATZECK
[6] F. Huard, M. Lanzilotta, Self-injective right artinian rings and Igusa-Todorov func-
tions, Algebras and Representation Theory, 16 (3), pp. 765-770, (2012).
[7] F. Huard, M. Lanzilotta, O. Mendoza. An approach to the Finitistic Dimension Con-
jecture. J. of Algebra 319, 3918-3934, (2008).
[8] K. Igusa, G. Todorov. On the finitistic global dimension for artin algebras. Represen-
tation of algebras and related topics, Fields Institute Communications 45 (American
Mathematical Society, Providence, RI), 201-204, (2005).
[9] S. Michelena. Sobre un problema de clasificaci´on y cohomolog´ıa de Hochschild de ex-
tensiones locales, Tesis Doctor en Matem´atica, Universidad Nacional del Sur, (1998).
[10] D. Xu. Homological dimensions and strongly idempotent ideals. J. Algebra 414, 175-
189, (2014).
Mar´ıa Andrea Gatica:
Instituto de Matem´atica de Bah´ıa Blanca,
Universidad Nacional del Sur,
Av. Alem 1253, B8000CPB,
Bah´ıa Blanca, ARGENTINA.
[email protected]
Marcelo Lanzilotta:
Instituto de Matem´atica y Estad´ıstica Rafael Laguardia (IMERL),
Universidad de la Rep´ublica.
J. Herrera y Reissig 565 C.P. 11300, Montevideo, URUGUAY.
[email protected]
Mar´ıa In´es Platzeck:
Instituto de Matem´atica de Bah´ıa Blanca,
Universidad Nacional del Sur,
Av. Alem 1253, B8000CPB,
Bah´ıa Blanca, ARGENTINA.
[email protected]
|
1206.0309 | 1 | 1206 | 2012-06-01T21:20:02 | N-derivations for finitely generated graded Lie algebras | [
"math.RA"
] | $N$-derivation is the natural generalization of derivation and triple derivation. Let ${\cal L}$ be a finitely generated Lie algebra graded by a finite dimensional Cartan subalgebra. In this paper, a sufficient condition for Lie $N$-derivation algebra of ${\cal L}$ coinciding with Lie derivation algebra of ${\cal L}$ is given. As applications, any $N$-derivation of Schr\"{o}dinger-Virasoro algebra, generalized Witt algebras, Kac-Moody algebras and their Borel subalgebras, is a derivation. | math.RA | math |
N-derivations for finitely generated
graded Lie algebras
School of Mathematics and Physics, Fujian University of Technology, Fuzhou 350108, China
Cui Chen
Email: [email protected]
Haifeng Lian
Mathematics Department, Fujian Agriculture and Forestry University, Fuzhou 350002, China
Email: [email protected]
Abstract: N-derivation is the natural generalization of derivation and triple deriva-
tion. Let L be a finitely generated Lie algebra graded by a finite dimensional Cartan
subalgebra. In this paper, a sufficient condition for Lie N-derivation algebra of L co-
inciding with Lie derivation algebra of L is given. As applications, any N-derivation
of Schrodinger-Virasoro algebra, generalized Witt algebras, Kac-Moody algebras
and their Borel subalgebras, is a derivation.
Keywords: Derivation, Cartan subalgebra, Virasoro algebra, Kac-Moody algebra
2000 MR Subject Classification 17B40, 17B67, 17B68
1
Introduction
Let L be a Lie algebra over an arbitrary field F . Recall that an F -linear mapping
ψ : L → L is called a Lie derivation of L if
ψ([x, y]) = [ψ(x), y] + [x, ψ(y)],
∀x, y ∈ L.
Let N ≥ 2 be a positive integer. We say ϕ is an N-derivation of L, if ϕ is a
linear map from L to itself, and satisfies
ϕ([x1, · · · , xN −1, xN ]) =
NX
i=1
[x1, · · · , xi−1, ϕ(xi), xi+1, · · · , xN ]
(∀x1, · · · , xN ∈ L),
where [x1, · · · , xN −1, xN ] = [x1 · · · [xN −2, [xN −1, xN ]] · · · ].
The set of Lie N-derivation is clearly a Lie algebra under the usual bracket and
will be denoted by Der(N )L. By the definition, we have Der(2)L = DerL, where DerL
1
is the Lie derivation algebra of L. A Lie derivation is obviously a Lie N-derivation,
which implies DerL ⊆ Der(N )L. In general, the derivation algebra DerL is a proper
subalgebra of Der(N )L for N ≥ 3. It would be interesting to know when these two
algebras coincide.
For N = 3, Lie N-derivation is called Lie triple derivation which has received
a fair amount of attentions (see [4, 6, 8, 12, 13, 14]). In [13], Wang and Yu showed
that a linear map on Borel subalgebra of finite-dimensional simple Lie algebra over
an algebraically closed field F of characteristic zero is a Lie triple derivation, if and
only if it is an inner derivation. Which are examples for Lie triple derivation algebra
coinciding with Lie derivation algebra, and the Lie algebras concerned are graded
by a finite-dimensional Cartan subalgebra.
The Schrodinger-Virasoro algebra sv introduced by Henkel in [3], during his
study on the invariance of the free Schrodinger equation, is an infinite-dimensional
Lie algebra with C-basis {Ln, Mn, Yn+ 1
, C n ∈ Z} subject to the following Lie
brackets:
2
[Lm, Ln] = (n − m)Ln+m + δm+n,0
n3 − n
12
C,
[Lm, Mn] = nMn+m,
[Lm, Yn+ 1
] = (n +
)Ym+n+ 1
2
,
[Ym+ 1
2
, Yn+ 1
2
] = (n − m)Mm+n+1,
1 − m
2
2
[Mm, Mn] = [Mm, Yn+ 1
2
] = 0,
[sv, C] = {0}.
Due to its important applications in many areas of Mathematics and Physics, the
structure and representation theory of sv have been extensively studied (see [2, 7, 10,
11]. For example, in [10], Rosen and Unterberger presented detailed cohomological
study and determined that sv has three linear independent outer derivations. This
follows that not all derivation of sv is an inner derivation.
In this paper, we consider Lie N-derivations for finitely generated graded Lie
algebra L. In section 2, we prove that the Lie N-derivation algebra of L is graded,
which generalizes the result obtained by Farnsteiner (see [1]). In section 3, we assume
L is graded by a finite-dimensional Cartan subalgebra and is over an algebraically
closed field F of characteristic zero. We show that Der(N )L = DerL for N ≥ 3, if
L satisfies condition (P) (see theorem 3.2). As applications, in section 4, we show
that the Lie N-derivation algebras of Schrodinger-Virasoro algebra, generalized Witt
algebras and Kac-Moody algebras are coinciding with the derivation algebras. In
particularly, any N-derivation of finite-dimensional semisimple complex Lie algebra
is an inner derivation.
2 N -derivation of finitely generated graded Lie al-
gebra
Let G be an abelian group, and let L = ⊕α∈GLα be a G-graded Lie algebra. For
x1, · · · , xn ∈ L, set
[x1, · · · , xn−1, xn] = [x1 · · · [xn−2, [xn−1, xn]] · · · ].
2
Let N ≥ 2 be a positive integer. As above, denote Der(N )(L), Der(L) the set
of all N-derivations of L and the set of all derivations of L, respectivly. We have
Der(L) ⊆ Der(N )(L). An N-derivation ϕ is called an N-derivation of homogeneous
degree α if ϕ(Lβ) ⊆ Lα+β, for α, β ∈ G. Set
Der(N )
α (L) = {ϕ ∈ Der(N )(L) deg ϕ = α}.
Lemma 2.1 Let G be an abelian group. For any finitely generated G-graded Lie
algebra L = Lα∈G
Lα, we have
Der(N )(L) = M
α∈G
Der(N )
α (L).
Proof
finite subset generating L. Set
For α ∈ G, let ρα : L → Lα denote the canonical projection. Let S be a
Y := S ∪ {[x1, · · · , xm]x1, · · · , xm ∈ S, m = 2, · · · , N − 1},
Y is also a finite subset of L. Clearly, Y generates L as an N-Lie algebra (i.e., L is
the smallest subspace containing Y and stable under taking iterated brackets of the
form [x1, · · · , xN ]). For ϕ ∈ Der(N )(L), there is a finite set K ⊆ G such that
Y ∪ ϕ(Y ) ⊂ X
α∈K
Lα.
(1)
For α ∈ G, set ϕα := Pβ∈G ρα+βϕρβ. Since for xβi ∈ Lβi (i = 1, · · · , N), we
ϕα([xβ1, · · · , xβN ])
have
= ρα+β1+···+βN ϕ([xβ1, · · · , xβN ])
= ρα+β1+···+βN (PN
= PN
= PN
i=1[xβ1, · · · , xβi−1, ϕ(xβi), xβi+1, · · · , xβN ])
i=1[xβ1, · · · , xβi−1, ρα+βiϕ(xβi), xβi+1, · · · , xβN ]
i=1[xβ1, · · · , xβi−1, ϕα(xβi), xβi+1, · · · , xβN ],
which follows that ϕα ∈ Der(N )
α (L).
Let T := {α − βα, β ∈ K}, then T is finite. For y ∈ Y , we obtain
ϕ(y)
(a)
(b)
= Pα,β∈K ραϕρβ(y)
= Pα,β∈K ρα−β+βϕρβ(y)
= Pβ∈K Pγ∈K−β ργ+βϕρβ(y)
= Pβ∈K Pγ∈T ργ+βϕρβ(y)
= Pγ∈T Pβ∈K ργ+βϕρβ(y)
= Pγ∈T Pβ∈G ργ+βϕρβ(y)
= Pγ∈T ϕγ(y),
(c)
3
where (a), (b) and (c) follow from (1). This shows that the Lie N-derivation ϕ
and Pγ∈T ϕγ coincide on Y . By the construction of Y and the definition of Lie
N-derivation, we obtain ϕ = Pγ∈T ϕγ.
✷
Remark 2.2 In case N = 2, Der(N )(L) = Der(L), the above lemma is contained
in the proposition 1.1 of [1]. In what follows, we assume N ≥ 3.
3 Main theorem
Let L be a finitely generated Lie algebra graded by a nontrivial finite dimensional
Cartan subalgebra H. Let H∗ be the dual space of H. For α ∈ H∗,
Lα = {x ∈ L[h, x] = α(h)x for all h ∈ H}
is the root space associated to α. Let R = {α ∈ H∗Lα 6= {0}}, R is called the root
system of L with respect to H. Thus we have
L = ⊕α∈RLα,
and L0 = H.
Set R× = {α ∈ Rα 6= 0}, R± = {α ∈ R× − α ∈ R}.
Definition 3.1 We say L satisfying property (P) if for α ∈ R±, every element
xα ∈ Lα satisfies (P1) or (P2), where
(P1) xα = [yα−β, yβ] for some yα−β ∈ Lα−β and yβ ∈ Lβ, with β 6∈ {0, α}.
(P2)
[xα, xα, x−α] 6= 0 for some x−α ∈ L−α.
Let L be a finitely generated Lie algebra with a nontrivial finite
Theorem 3.2
dimensional Cartan subalgebra H, and let N ≥ 3 be a positive integer. If N is even
or if L satisfying property (P), then we have
Der(N )(L) = Der(L).
Proof Assume H, R as above, let Q = ZR be the abelian group. Clearly, L is a
finitely generated Q-graded Lie algebra. Using lemma 2.1, we have
Der(N )(L) = M
α∈Q
Der(N )
α (L).
Now let ϕ ∈ Der(N )
divide the argument into two cases.
γ
(L), γ ∈ Q. In what follows we will prove ϕ is a derivation. We
Case one: γ = 0.
For x, y ∈ H, we have ϕ([x, y]) = 0 = [ϕ(x), y] + [x, ϕ(y)] as required. Suppose
0 6= y ∈ Lα for some α ∈ R×, then there is hα ∈ H such that α(hα) = 1. Taking ϕ
on
y = [hα, · · · , hα
, y],
4
{z
N −1
}
we have ϕ(y) = (N − 1)α(ϕ(hα))y + ϕ(y), which implies α(ϕ(hα)) = 0. Thus, for
x ∈ L, taking ϕ on
we have ϕ([x, y]) = [ϕ(x), y] + [x, ϕ(y)] as required.
[x, y] = [x, hα, · · · , hα
, y],
{z
N −2
}
Case two: γ 6= 0.
Since γ 6= 0, there exists hγ ∈ H, such that γ(hγ) = 1. For h ∈ H, taking ϕ on
0 = [hγ, · · · , hγ
, h],
{z
N −1
}
we have 0 = [ϕ(hγ), h] + ϕ(h), that is ϕ(h) = (−adϕ(hγ))(h). Set ψ = ϕ + adϕ(hγ),
ψ is an N-derivation and ψ(H) = {0}. In what follows, we use two subcases to show
ψ(xα) = 0,
for 0 6= xα ∈ Lα, α ∈ R×.
Thus ϕ = −adϕ(hγ) is a derivation.
Subcase one: N is even or γ 6= −2α.
Since N is even or γ 6= −2α, there is ¯h ∈ H such that α(¯h)N −1 6= (α + γ)(¯h)N −1.
Taking ψ on
α(¯h)N −1xα = [¯h, · · · , ¯h
{z }
N −1
, xα],
we have α(¯h)N −1ψ(xα) = (α + γ)(¯h)N −1ψ(xα), which implies ψ(xα) = 0, as required.
Subcase two: N is odd and γ = −2α.
Since α 6= 0, then there is hα ∈ H such that α(hα) = 1. If L−α = {0}, then
we have ψ(xα) = 0, as required. Suppose L−α 6= {0}, xα ∈ Lα. If xα satisfies (P1),
then xα = [yα−β, yβ] for some yα−β ∈ Lα−β and yβ ∈ Lβ, with β 6∈ {0, α}. Using
subcase one and taking ψ on
xα = [hα, · · · , hα
, yα−β, yβ],
{z
N −2
}
then we have ψ(xα) = 0.
x−α ∈ L−α. Firstly, using subcase one, we have ψ(x−α) = 0. Next, taking ψ on
If xα satisfies (P2), then [xα, xα, x−α]
6= 0 for some
0 = [hα, · · · , hα
, x−α, xα],
{z
N −2
}
then we have (−2)N −2[x−α, ψ(xα)] = 0, which implies [x−α, ψ(xα)] = 0. Finally,
since N is odd, setting h′ = [xα, x−α] and taking ψ on
α(h′)N −2xα = [h′, · · · , h′
, xα, x−α, xα],
}
{z
N −3
5
then we have
α(h′)N −2ψ(xα)
= [h′, · · · , h′
, ψ(xα), x−α, xα]
{z
N −3
}
= (−α)(h′)N −2ψ(xα)
= −α(h′)N −2ψ(xα).
Since [xα, xα, x−α] 6= 0, which implies α(h′) 6= 0, we have ψ(xα) = 0 as required.
Therefore, every N-derivations of homogeneous degree γ are derivations. This
✷
completes the proof.
Remark 3.3 If L is a finitely generated Lie algebra graded by a nontrivial fi-
nite dimensional Cartan subalgebra H, but doesn't satisfy the property (P), then
Der(2N +1)(L) = Der(L) doesn't holds in general for N ≥ 1. See example 4.1.3 in the
following section.
4 Applications
4.1 Schrodinger-Virasoro algebra
The Schrodinger-Virasoro algebra sv is an infinite-dimensional Lie algebra with C-
basis {Ln, Mn, Yn+ 1
, C n ∈ Z} subject to the following Lie brackets:
2
[Lm, Ln] = (n − m)Ln+m + δm+n,0
n3 − n
12
C,
[Lm, Mn] = nMn+m,
1 − m
[Lm, Yn+ 1
] = (n +
2
2
[Mm, Mn] = [Mm, Yn+ 1
2
)Ym+n+ 1
2
,
[Ym+ 1
2
, Yn+ 1
2
] = (n − m)Mm+n+1,
] = 0,
[sv, C] = {0}.
It is easy to see the following facts about sv :
sv is generated by L1, L2, L−2, M1 and Y− 1
2
.
sv = ⊕p∈ 1
h = CL0 ⊕ CM0 ⊕ CC, where, svn = CLn ⊕ CMn for n ∈ Z×(= Z\{0}), svn+ 1
CYn+ 1
for n ∈ Z, and sv0 = h .
Z graded Lie algebra according to the Cartan algebra
=
Zsvp is a 1
2
2
2
(i)
(ii)
2
(iii)
[Ln, L−n, Ln] = −2n2Ln, [L2n+1, Y−n− 1
] = (−2n−1)Yn+ 1
2
, [L−2n−1, M3n+1]
2
= (3n + 1)Mn, for n ∈ Z×.
Therefore, sv is a finitely generated Lie algebra graded by a nontrivial finite
dimensional Cartan subalgebra h and satisfies property (P). Using theorem 3.2, we
have corollary 4.1.1.
Corollary 4.1.1 Der(N )(sv) = Der(sv)
(N ≥ 3).
6
Let K = spanC{L0, M1, M−1} be the subalgebra of sv. K is a
Example 4.1.2
finitely generated Lie algebra with Cartan subalgebra H = CL0, but doesn't satisfy
property (P). Let ϕ : K → K be a linear map such that
ϕ(L0) = ϕ(M−1) = 0, ϕ(M1) = M−1.
Taking ϕ on [L0, M1] = M1, we get ϕ is not a derivation of K. But one can check
that ϕ is an N-derivation of K for any odd integer N ≥ 3.
4.2 Generalied Witt algebra
Let A be the Laurent polynomial ring C[t±1
d ] with commuting variables,
and let wd be the Lie algebra of derivations of A. wd is called generalied Witt
algebra, and is also the Lie algebra of diffeomorphisms of torus T d (see [9]). When
d = 1, wd is the Witt algebra and its universal central extension is called the Virasoro
algebra. When d ≥ 2, wd has no nontrivial central extension.
2 , · · · , t±1
1 , t±1
Let Zd = Zε1 ⊕ · · · ⊕ Zεd. For n = n1ε1 + · · · + ndεd ∈ Zd, let tn = tn1
2 · · · tnd
d
. Then wd = spanC{Dj(n)n ∈ Zd, j = 1, 2 · · · , d} with the
1 tn2
∂
and let Dj(n) = tntj
∂tj
following Lie structure:
[Dj(n), Dk(m)] = mjDk(n + m) − nkDj(n + m), ∀n, m ∈ Zd, j, k = 1, 2 · · · , d.
It is easy to see the following facts about wd:
(i) wd is generated by {Di(±εj), Di(±2εj)i, j = 1, 2 · · · , d}.
(ii) wd = ⊕n∈Zd(wd)n is a Zd-graded Lie algebra respects to the Cartan sub-
algebra H = spanC{D1(0), · · · , Dd(0)}, where (wd)n = spanC{D1(n), · · · , Dd(n)}.
(iii)
[Di(−(2ni + 1)εi), Di(n + (2ni + 1)εi)] = (5ni + 2)Di(n).
Therefore, wd is a finitely generated Lie algebra graded by a nontrivial finite
dimensional Cartan subalgebra H and satisfies property (P). Using theorem 3.2, we
have corollary 4.2.
Corollary 4.2 Der(N )(wd) = Der(wd) for N ≥ 3.
4.3 Kac-Moody algebra
Recall that a matrix A = (aij)n
the following conditions:
i,j=1 is called a generalized Cartan matrix if it satisfies
(C1)
(C2)
(C3)
aii = 2 for i = 1, · · · , n;
aij are nonpositive integers for i 6= j;
aij = 0 implies aji = 0.
Let A = (aij)n
i,j=1 be a generalized Cartan matrix of rank l. A realization
of A is a triple (h, Π, Π∨), where h is a 2n − l dimensional complex vector space,
7
Π = {α1, · · · , αn} ⊂ h∗ and Π∨ = {α∨
subsets in h∗ and h, respectively, satisfying
1 , · · · , α∨
n} ⊂ h are linearly independent
αj(α∨
i ) = aij
for
i, j = 1, · · · , n.
Let g(A) be the Kac-Moody algebra associated to A. One can see from [5] that
g(A) is a complex Lie algebra generated by h, e1, · · · , en, f1, · · · , fn with following
defining relations:
[ei, fj] = δijα∨
i ,
[h, h′] = 0,
[h, ei] = αi(h)ei, [h, fi] = −αi(h)fi,
(adei)1−aij ej = (adfi)1−aij fj = 0,
(i, j = 1, · · · , n);
(h, h′ ∈ h);
(i = 1, · · · , n; h ∈ h);
(i 6= j).
Let Q = Pn
i=1
Zαi be the root lattice, and Q+ = Pn
g(A) = (cid:16) M
g−α(cid:17) ⊕ h ⊕(cid:16) M
i=1
Nαi. Then
gα(cid:17)
α∈Q+,α6=0
α∈Q+,α6=0
is a Q-graded Lie algebra related to Cartan subalgebra h. Here, gα = {x ∈
g(A)[h, x] = α(h)x for all h ∈ h} is a root space attached to α. For α > 0 (resp.
α < 0), gα is the linear span of the elements of the form
[ei1, ei2, · · · , eis]
(resp. [fi1, fi2, · · · , fis])
such that αi1 + · · · + αis = α (resp. = −α). Moreover, we have
(i)
(ii)
(iii)
(iv)
g−αi = Cfi,
gαi = Cei,
gsαi = {0},
[ei, fi, ei] = 2ei 6= 0,
[ei1 , ei2, · · · , eis] = [ei1, [ei2, · · · , eis]],
[fi1, fi2, · · · , fis] = [fi1, [fi2, · · · , fis]],
[fi, ei, fi] = 2fi 6= 0,
for i ∈ {1, · · · , n};
for i ∈ {1, · · · , n}, s > 1;
for i ∈ {1, · · · , n};
for i1, · · · , is ∈ {1, · · · , n}, s ≥ 2.
Let b± = Lα∈Q+ g±α. b+, b− are Borel subalgebras of g(A). One can check
that g(A) and b± are finitely generated Lie algebras graded by a nontrivial finite
dimensional Cartan subalgebra h and satisfy property (P). Using theorem 3.2, we
have corollary 4.3.1.
Corollary 4.3.1
as above. For N ≥ 3,
Let A = (aij)n
i,j=1 be a generalized Cartan matrix, g(A) and b±
Der(N )(g(A)) = Der(g(A)), Der(N )(b±) = Der(b±).
Let g be a semisimple complex Lie algebra, and A be the Cartan
Remark 4.3.2
matrix of g. Then g(A) = g. Since any derivation of g is an inner derivation,
by corollary 4.3.1, any N-derivation of semisimple complex Lie algebra is an inner
derivation for N ≥ 3.
8
References
[1] Farnsteiner R. Derivations and central extensions of finitely generated graded
Lie algebras, J. Algebra, 1988, 118: 33-45.
[2] Gao S, Jiang C and Pei Y. Structure of the extended Schrodinger-Virasoro Lie
algebra esv, Alg. Colluq., 2009, 16(4): 549-566.
[3] Henkel M. Schrodinger invariance and strongly anisotropic critical systems, J.
Stat. Phys., 1994, 75: 1023-1029.
[4] Ji P, Wang L. Lie triple derivations of TUHF algebras, Linear Algebra and its
Appl, 2005, 403: 399-408.
[5] Kac V. Infinite dimensional Lie algebras, Cambridge: Cambridge University
Press, 1990.
[6] Lu F. Lie triple derivations on nest algebras, Math. Nachr., 2007, 8: 882-887.
[7] Li J, Su Y. Representations of the Schrodinger-Virasoro algebras, J. Math.
Phys., 2008, 49(5), 053512: 1-14.
[8] Miers C. Lie triple derivations of von Neumann algebras, Proc. Amer. Math.
Soc., 1978, 71: 57-61.
[9] Ramos E, Sah C, and Shrock R. Algebras of diffeomorphisms of the N-torus,
J. Math. Phys., 1990, 8: 1805-1816.
[10] Roger C, Unterberger J. The Schrodinger-Virasoro Lie group and algebra: from
geometry to representation thery, Ann. Henri Poincare, 2006, 7: 1477-1529.
[11] Unterberger J. The Schrodinger-Virasoro Lie algebra: a mathematical struc-
ture between conformal field theory and non-equilibrium dynamics, Journal of
Physics: Conference Series, 2006, 40: 156-162.
[12] Wang H, Li Q. Lie triple derivations of the Lie algebras of strictly upper trian-
gular matrix over a commutative ring, Linear Algebra Appl, 2009, 430: 66-77.
[13] Wang D, Yu X. Lie triple derivations on the parabolic subalgebras of simple
Lie algebras. Linear and Multilinear Algebra, 2011, 59(8): 837-840.
[14] Zhao J, Li H and Fang L. Lie triple derivations for the parabolic subalgebras
of gl(n, R), Lecture Notes in Computer Science, 2011, 6729: 457-464.
9
|
1412.8042 | 2 | 1412 | 2015-09-04T08:47:22 | Group Algebra and Coding Theory | [
"math.RA"
] | Group algebras have been used in the context of Coding Theory since the beginning of the latter, but not in its full power. The work of Ferraz and Polcino Milies entitled Idempotents in group algebras and minimal abelian codes (Finite Fields and their Applications, 13, (2007) 382-393) gave origin to many thesis and papers linking these two subjects. In these works, the techniques of group algebras are mainly brought into play for the computing of the idempotents that generate the minimal codes and the minimum weight of such codes. In this paper I summarize the main results of the work done by doctorate students and research partners of Polcino Milies and Ferraz. | math.RA | math |
Group algebras and Coding Theory
Marines Guerreiro ∗†
September 7, 2015
Abstract
Group algebras have been used in the context of Coding Theory
since the beginning of the latter, but not in its full power. The work
of Ferraz and Polcino Milies entitled Idempotents in group algebras
and minimal abelian codes (Finite Fields and their Applications, 13,
(2007) 382-393) gave origin to many thesis and papers linking these
two subjects.
In these works, the techniques of group algebras are
mainly brought into play for the computing of the idempotents that
generate the minimal codes and the minimum weight of such codes. In
this paper I summarize the main results of the work done by doctorate
students and research partners of Polcino Milies and Ferraz.
Keywords: idempotents, group algebra, coding theory.
Dedicated to Prof. C´esar Polcino Milies on the occasion of his
seventieth birthday.
1
Introduction
The origins of Information Theory and Error Correcting Codes Theory are
in the papers by Shannon [64] and Hamming [35], where they settled the
theoretical foundations for such theories.
∗M. Guerreiro is with Departamento de Matem´atica, Universidade Federal de Vi¸cosa,
CEP 36570-000 - Vi¸cosa-MG (Brasil), supported by CAPES, PROCAD 915/2010 and
FAPEMIG, APQ CEX 00438-08 and PCE-00151-12(Brazil). E-mail: [email protected].
†This work was presented at Groups, Rings and Group Rings 2014 on the occasion of
the 70th Birthday of C´esar Polcino Milies, Ubatuba-SP (Brasil) 21-26 July, 2014.
1
For a non empty finite set A, called alphabet, a code C of length n is
simply a proper subset of An and an n-tuple (a0, a1, . . . , an−1) ∈ C is called
a word of the code C.
If A = Fq is a finite field with q elements, then a linear code C of length
q . If dim C = k (k < n), then the number of
n is a proper subspace of Fn
words in C is qk.
We shall call “cyclic shift” the linear map π : Fn
q −→ Fn
q such that
π(a0, a1, . . . , an−1) = (an−1, a0, a1, . . . , an−2).
A linear cyclic code is a linear code C that is invariant under the cyclic
shift. This structure gives rise to fast-decoding algorithms, which is a con-
siderable aspect regarding the conditions on communication.
Consider the quotient ring Rn =
and denote by [f (x)] the
class of the polynomial f (x) in Rn. There is a natural vector space isomor-
phism ϕ : Fn
q −→ Rn given by
Fq[x]
< xn − 1 >
ϕ(a0, a1, . . . , an−1) = [a0 + a1x + · · · + an−1xn−1].
Linear cyclic codes are often realized as ideals in Rn and the cyclic shift
is equivalent, via the isomorphism ϕ, to the multiplication by the class of x
in Rn.
Group algebras may be defined in a more general setting, that is, for any
group and over any field. However, we restrict the definitions and results
below to finite groups and finite fields because this is the context for coding
theory.
Let G be a finite group written multiplicatively and Fq a finite field. The
group algebra of G over Fq is the set of all formal linear combinations
α =Xg∈G
αgg and β =Xg∈G
Given α =Xg∈G
αgg, where αg ∈ Fq.
βgg we have
α = β ⇐⇒ αg = βg,
for all g ∈ G.
The support of an element α ∈ FqG is the set of elements of G effectively
appearing in α; i.e.,
supp(α) = {g ∈ G ag 6= 0}.
2
We define
For λ in Fq, we define
(αg + βg)g.
(αgβh)gh.
αgg! + Xg∈G
Xg∈G
Xg∈G
αgg! Xh∈G
λ Xg∈G
βgg! =Xg∈G
βhh! = Xg,h∈G
αgg! =Xg∈G
(λαg)g.
It is easy to see that, with the operations above, FqG is an algebra over
the field Fq.
The weight of an element α =Pg∈G agg ∈ FqG is the number of elements
in its support; i.e.
w(α) = {g ag 6= 0}.
For an ideal I of FqG, we define the minimum weight of I as:
w(I) = min{w(α) α ∈ I, α 6= 0}.
Let Cn = hai denote a cyclic finite group of order n generated by an
element a. MacWilliams [44] was the first one to consider cyclic codes as
ideals of the group ring FqCn which is easily proved to be isomorphic to
Fq[x]
<xn−1> . In FqCn, the cyclic shift is equivalent to the multiplication of the
elements of the code by a.
The following diagram helps us to understand the cyclic shift in these
three different ways of considering a cyclic code.
cyclic
shift
C ⊂ Fn
q
ϕ
−→ Rn = Fq[x]
<xn−1>
∼=−→ FqCn = Fq < a >
↓
¯x ↓
a ↓
C ⊂ Fn
q
ϕ
−→ Rn = Fq[x]
<xn−1>
∼=−→ FqCn = Fq < a >
Extending these ideas, Berman [9, 10] and, independently, MacWilliams
[45] defined abelian codes as ideals in finite abelian group algebras and,
3
more generally, a group (left) code was defined as an (left) ideal in a finite
group algebra. Group codes were then studied using ring and character-
theoretical results.
From now on, for a finite group G and a finite field Fq, we treat ideals in
a group algebra FqG as codes. In this approach, the length of the code is the
order of the group G and the dimension of a code I is its dimension as an
Fq-subspace in FqG. Length, dimension and minimum weight are the three
parameters that define a linear code.
A group code is called minimal if the corresponding ideal is minimal
in the set of ideals of the group algebra. Keralev and Sol´e in [40] showed
that many important codes can be realized as ideals in a group algebra,
for example, the generalized Reed-Muller codes and generalized quadratic
residue codes. These results are included in Section 9.1 of [39]. There is also
a good treatment on the subject in [22].
A word of warning is necessary here, because the expression “group code”
may also have some other meanings. For example, in Computer Science,
sometimes group codes consist of n linear block codes which are subgroups
of Gn, where G is a finite abelian group, as in [12, 29].
Usually in the papers that present techniques to compute the idempo-
tents that generate the codes, character theory is used in the context of
polynomials, as it can be seen in [1, 2, 4, 5, 6, 50, 56, 57, 65]. Sometimes
the expressions for the idempotents are not very “reader friendly”. More-
over, the character theory and polynomial approaches in the computation of
idempotents did not fully explore the structure of the group underneath the
group algebra that defines the underlying set for the codes.
Summarizing the work of Ferraz and Polcino Milies [28], in Section 2 we
define the idempotents using subgroups of the group and establish the basic
theorems that are used in this work. We emphasize that in [28], they gave
simpler proofs for computing idempotents, dimension and minimum weight
for minimal cyclic and abelian codes of length 2kpn, generalizing and com-
paring their results with the ones in [1, 2]. Full details of Ferraz and Polcino
Milies´ paper are described in the Master´s Dissertation of Luchetta [43].
In Section 3 we discuss some topics of non-abelian group codes, including
some equivalence questions. In Section 4 we describe some results on codes
of length 2n, for a natural n ≥ 1 and in Section 5 we treat some aspects
of equivalence of abelian codes.
In Section 6 we summarize some results
on cyclic and abelian codes of length pnqm, for p and q distinct primes and
4
n, m ≥ 1 and in Section 7 we explore some facts on codes over rings.
2 Subgroups and idempotents
We recall that an element in the group algebra FqG is called central if
it commutes with every other element of the algebra. A non-zero central
idempotent e is called primitive if it cannot be decomposed in the form
e = e′ + e′′, where e′ and e′′ are both non-zero central idempotents such that
e′e′′ = e′′e′ = 0. For char(Fq) 6
G, the group algebra FqG is semisimple and
the primitive central idempotents are the generators of the minimal two-sided
ideals. Two idempotents e′, e′′ are orthogonal if e′e′′ = e′′e′ = 0.
The primitive central idempotents of the rational group algebra QG were
computed in [34, Theorem VII.1.4] in the case G abelian; in [37, Theorem
2.1] when G is nilpotent; in [53, Theorem 4.4] in a more general context and
in [13, Theorem 7] an algorithm to compute the primitive idempotents is
given.
In what follows, we shall establish a correspondence between primitive
idempotents of FqG and certain subgroups of an abelian group G.
Let G be a finite (abelian) group and Fq a field such that char(Fq) 6
G.
Given a subgroup H of G, denote
1
HXh∈H
h
(1)
bH =
3.6.7]).
which is an idempotent of FqG and, for an element x ∈ G, setbx = chxi.
It is known that the idempotent bG is always primitive (see [55, Proposition
Definition 2.1. Let G be an abelian group. A subgroup H of G is called a
co-cyclic subgroup if the factor group G/H 6= {1} is cyclic.
We use the notation
Scc(G) = {H H is a co-cyclic subgroup of G}.
For a finite group G, denote by exp(G) the exponent of G which is the
smallest positive integer t such that gt = 1, for all g ∈ G. A group G is called
a p-group if its exponent is a power of a given prime p. In particular, this
means that the order of every element of G is itself a power of p.
5
Let G be a finite abelian p-group and Fq a field such that char(Fq) 6
G.
For each co-cyclic subgroup H of G, we can construct an idempotent of FqG.
In fact, we remark that, since G/H is a cyclic p-group, there exists a unique
subgroup H ♯ of G containing H such that H ♯/H = p. Then eH = bH −cH ♯
is an idempotent and we consider the set
{bG} ∪ {eH = bH −cH ♯ H ∈ Scc(G)}.
We recall the following results that are used throughout this paper.
In the case of a rational abelian group algebra QG, the set (2) is the set
(2)
of all primitive central idempotents [34, Theorem 1.4].
Theorem 2.2. [28, Lemma 5] Let p be a prime integer and G a finite abelian
group of exponent pn and Fq a finite field with q elements such that p 6 q.
Then (2) is a set of pairwise orthogonal idempotents of FqG whose sum is
equal to 1, i.e.,
eH ,
(3)
1 = bG + XH∈Scc(G)
where 1 also denotes the identity element in FqG.
In our next statement, we denote by U(Zpn) the set of invertible elements
of the ring Zpn of integers modulo pn; ¯q denotes the class of the integer q in
Zpn and, when it is invertible, o(¯q) denotes its multiplicative order; i.e., the
least positive integer m such that ¯qm = ¯1.
Theorem 2.3. [28, Theorem 4.1] Under the same hypotheses of Theorem 2.2,
the set (2) is the set of all primitive idempotents of FqG if and only if o(¯q) =
φ(pn) in U(Zpn), where φ denotes Euler’s totient function.
For positive integers r and m, we shall denote by ¯r ∈ Zm the image of r in
the ring of integers modulo m. Then, for an element g in a group G, define
Gg = {gr gcd(r, o(g)) = 1} = {gr ¯r ∈ U(Zo(g))}. The following theorem
gives us conditions on the exponent e of the group G and the size q of the
finite field that satisfy Theorem 2.3.
Corollary 2.4. [46, Teorema 7.10] Let Fq be a finite field with q elements
and G a finite abelian group with exponent e such that gcd (q, G) = 1. Then
Cg = Gg, for all g ∈ G, if only if one of following conditions holds, where φ
denotes Euler´s totient function:
6
(a) e = 2 and q is odd;
(b) e = 4 and q ≡ 3(mod 4);
(c) e = pn and o(q) = φ(pn) in U (Zpn);
(d) e = 2pn and o(q) = φ(pn) in U (Z2pn).
Theorem 2.5. [28, Lemma 3] Let G = hgi be a cyclic group with order pn
and Fq a finite field with q elements such that q generates U (Zpn). Consider
G = G0 ⊃ G1 ⊃ ... ⊃ Gn = {1}
the descending chain of all subgroups of G. Then a complete set of primitive
idempotents in FqG is:
e0 = bG =
1
pnXg∈G
g
with Gi =< gpi >, for 1 ≤ i ≤ n.
and ei =cGi − dGi−1, for 1 ≤ i ≤ n,
(4)
As the authors comment in [28], a straightforward computation shows
that these are the same idempotents given in [2, Theorem 3.5], though there
they are expressed in terms of cyclotomic cosets.
The idempotent generators of minimal ideals in the case of cyclic groups
of order 2pn now follow easily from the previous results.
Theorem 2.6. [2, Theorem 2.6] Let Fq be a finite field with q elements and
G a cyclic group of order 2pn, p an odd prime, such that o(¯q) = φ(pn) in
U(Z2pn). Write G = C × A, where A is the p-Sylow subgroup of G and
C = {1, t} is its 2-Sylow subgroup. If ei, for 0 ≤ i ≤ n, denote the primitive
idempotents of FqA, then the primitive idempotents of FqG are
1 + t
2
ei,
1 − t
2
ei,
0 ≤ i ≤ n.
(5)
More generally,
Theorem 2.7. [28, Theorem 4.2] Let p be an odd prime, A be an abelian
p-group of exponent 2pr and Fq be a finite field with q elements such that
o(q) = φ(pr) in U (Z2pr ). Write A = E × B, where E is an elementary
abelian 2-group and B a p-group. Then the primitive idempotents of FqA are
products of the form e · f , where e is a primitive idempotent of FqE and f
a primitive idempotent of FqB.
7
Section 5 of [28] is devoted to the computation of dimension and minimum
weight of codes generated by the idempotents presented in previous theorems.
For non-cyclic abelian groups, we may also apply the ideas above to construct
idempotents. In [27], the following results are presented in details.
For a finite abelian group G, we write G = Gp1 × · · · × Gpt, where Gpi
denotes the pi-Sylow subgroup of G, for the distinct prime numbers p1, . . . , pt.
Lemma 2.8. [27, Lemma II.5] Let G = Gp1 × · · · × Gpt be a finite abelian
group and H ∈ Scc(G). Write H = Hp1 ×· · · ×Hpt, where Hpi is the pi-Sylow
subgroup of H. Then each subgroup Hpi is co-cyclic in Gpi, 1 ≤ i ≤ t.
With the notation above, for each H ∈ Scc(G), define an idempotent
eH ∈ FqG as follows. For each 1 ≤ i ≤ t, either Hpi = Gpi or there exists
a unique subgroup H ♯
pi : Hpi] = pi. Thus, let eHpi
pi such that [H ♯
eHpi
=dHpi −dH ♯
pi, respectively, and define
eH = eHp1
eHp2
· · · eHpt .
(6)
= cGpi or
For any other K ∈ Scc(G), with K 6= H, we have Kpi
6= Hpi, for some
= 0, hence eH eK = 0. It is easy to
eKpi
1 ≤ i ≤ t, and, by Theorem 2.2, eHpi
see that bGeH = 0, for all H ∈ Scc(G).
Thus, we have the following.
Proposition 2.9. [27, Proposition II.6] Let G be a finite abelian group and
Fq a field such that char(Fq) 6
G. Then
is a set of orthogonal idempotents of FqG, where eH is defined as in (6).
B = {eH H ∈ Scc(G)} ∪ {bG}
(7)
A similar construction of idempotents for rational group algebras of abelian
groups is given in [34, Section VII.1]. For the rational case, these idempotents
are primitive while for finite fields this is usually not true.
Now, we extend Theorem 2.2 to finite abelian groups.
Lemma 2.10. [27, Lemma II.7] Let G be a finite abelian group and Fq a
field such that char(Fq) 6
G. Then, in the group algebra FqG, we have
1 = bG + XH∈Scc(G)
eH .
8
(8)
The following lemma starts the discussion about the relation between
idempotents and certain subgroups of the abelian group, which we elaborate
in more details in Section 5.
Lemma 2.11. [27, Lemma II.8] Let G be a finite abelian group and Fq a
field such that char(Fq) 6
there exists a unique H ∈ Scc(G) such that e · eH = e. Also, e · eK = 0, for
any other K ∈ Scc(G).
G. For each primitive idempotent e ∈ FqG, e 6= bG,
3 Non Abelian Codes
3.1 Dihedral and Quaternion Codes
As a natural way to proceed, Theorem 2.2 is used by Dutra [23, 25] in her
Ph.D. thesis to compute idempotents for non abelian group codes, particu-
larly, for dihedral and quaternion groups.
For n ≥ 1, Dutra considered the semisimple group algebras FqDn of the
dihedral groups Dn = ha, b an = b2 = 1, bab = a−1i over a finite field Fq
and gave conditions under which the number of its simple components is
minimum, that is, the same as for the rational group algebra QDn. These
conditions are stated in the following theorem.
Theorem 3.1. [23, Teorema 2.2] Let Fq be a field with q elements and Dn
the dihedral group with 2n elements such that gcd(q, 2n) = 1. Let p, p1 and
p2 be distinct odd primes and m, m1 and m2 be positive integers. Then FqDn
and QDn have the same number of simple components if and only if one of
the following conditions occurs:
(i) n = 2 or 4 and q is odd.
(ii) n = 2m, with m ≥ 3 and congruent to 3 or 5 modulo 8.
(iii) n = pm and the class ¯q generates the group of units U(Zpm).
(iv) n = pm, the class ¯q generates the group U 2(Zpm) = {x2 x ∈ U(Zpm)}
and −1 is not a square modulo pm.
(v) n = 2pm and the class ¯q generates the group of units U(Zpm).
(vi) n = 2pm, ¯q generates the group U 2(Zpm) = {x2 x ∈ U(Zpm)} and
−1 is not a square modulo pm.
(vii) n = 4pm, 4 divides φ(pm) and the class ¯q generates the group U(Zpm).
(viii) n = 4pm, 4 does not divide φ(pm), q ≡ 1(mod 4) and the class ¯q
generates the group U(Zpm).
9
(ix) n = 4pm, 4 does not divide φ(pm), q ≡ −1(mod 4) and the class ¯q
φ(pm1
has order φ(pm)/2.
1 pm2
(x) n = pm1
1 pm2
(xi) n = 2pm1
1 pm2
φ(pm1
2 )/2 modulo pm1
1 pm2
2 .
2 )/2 modulo pm1
1 pm2
2 .
2 , with gcd(φ(pm1
1 ), φ(pm2
2 )) = 2 and q or −q has order
1 pm2
2 , with gcd(φ(pm1
1 ), φ(pm2
2 )) = 2 and q or −q has order
Under such conditions, Dutra computed the set of minimal codes of FqDn,
their dimensions, minimum weights and bases for these codes as follows.
Theorem 3.2. [23, Proposi¸cao 3.1] Let q and n be integers related as in
conditions (i) and (ii) of Theorem 3.1. If C is a dihedral code of length 2n
generated by the idempotent e, then C has dimension and minimum weight
described in the table below.
e
bbba
(1 −bb)ba
bb(ba2 −ba)
(1 −bb)(ba2 −ba)
(ca2i − da2i−1)
dimFqC w(C)
2m+1
2m+1
2m+1
2m+1
2m−i+1
1
1
1
1
2i
Theorem 3.3. [23, Proposi¸cao 3.2] Let q and n be integers related as in
conditions (iii) and (iv) of Theorem 3.1. If C is a dihedral code of length 2n
generated by the idempotent e, then C has dimension and minimum weight
described in the table below.
e
1
1
dimFqC w(C)
2pm
2pm
bbba
(1 −bb)ba
(capi − dapi−1) 2φ(pi) 2pm−i
Theorem 3.4. [23, Proposi¸cao 3.3] Let q and n be integers related as in
conditions (v) to (ix) of Theorem 3.1. For n = pm1
2 with p1 = 2, m1 = 1
or 2 and p2 an odd prime, if C is a dihedral code of length 2n generated by
the idempotent e1e2, then C has dimension and minimum weight described in
1 pm2
10
the table below.
e1
m1
1
m1
1
bb[Cp
(1 −bb)[Cp
bb( \Cp
(1 −bb)( \Cp
m1−1
1
m1−1
1
− [Cp
m1
1
)
− [Cp
m1
1
m1
1
[Cp
− \Cp
− \Cp
m1−i+1
1
m1−i+1
1
m1−i
1
\Cp
\Cp
m1−i
1
m2
2
e2
[Cp
[Cp
[Cp
) [Cp
m2
2
m2
2
m2
2
\Cp
m2−j
2
\Cp
m2−j
2
m2−j+1
2
− \Cp
[Cp
− \Cp
m2
2
m2−j+1
2
dimFqC
1
1
1
1
2φ(pj
2)
2φ(pj
1)
1)2φ(pj
2φ(pj
w(C)
2pm1
1 pm2
2
1 pm2
2pm1
2
1 pm2
2pm1
2
2pm1
1 pm2
2
1 pm2−j
2pm1
2
2pm1−i
pm2
2
pm2−j
2) 4pm1−i
2
1
1
Theorem 3.5. [23, Proposi¸cao 3.4] Let q and n = pm1
2 , with p1 and p2 odd
distinct prime numbers, integers related as in condition (x) of Theorem 3.1.
If C is a dihedral code of length 2n generated by the idempotent e1e2, then C
has dimension and minimum weight described in the table below.
1 pm2
e1
m1
1
m1
1
bb[Cp
(1 −bb)[Cp
[Cp
− \Cp
− \Cp
m1
1
m1−i+1
1
m1−i+1
1
m1−i
1
\Cp
\Cp
m1−i
1
m2
2
m2
2
e2
[Cp
[Cp
− \Cp
[Cp
− \Cp
m2
2
m2−j+1
2
m2−j+1
2
\Cp
m2−j
2
\Cp
m2−j
2
dimFqC
1
1
2φ(pj
2)
2φ(pj
1)
1)2φ(pj
2φ(pj
w(C)
2pm1
1 pm2
2
1 pm2
2pm1
2
1 pm2−j
2pm1
2
2pm1−i
pm2
2
pm2−j
2) 4pm1−i
2
1
1
Theorem 3.6. [23, Proposi¸cao 3.5] Let q and n = 2pm1
2 , with p1 and
p2 odd distinct prime numbers, integers related as in condition (xi) of The-
orem 3.1. If C is a dihedral code of length 2n generated by the idempotent
e0e1e2, then C has dimension and minimum weight described in the table
1 pm2
11
e1
e2
dimFq C
p
p
\C
m2
2
\C
m2
2
\C
m2
2
\C
m2
2
p
p
\C
m2
2
p
p
− \C
m2−j+1
2
− \C
m2−j+1
2
p
\C
m2
2
p
1
1
1
1
2φ(pj
2)
2φ(pj
1)
2φ(p
j
2)
2φ(pj
1)
w(C)
m1
1 p
m1
1 p
m1
1 p
m1
1 p
m2
2
m2
2
m2
2
m2
2
4p
4p
4p
4p
4p
m1
1 p
m2 −j
2
4p
m1 −i
1
p
m2
2
4p
m1
1 p
m2 −j
2
4p
m1 −i
1
p
m2
2
\C
p
m2−j
2
\C
p
m2−j
2
\C
p
m2−j
2
\C
p
m2−j
2
below.
e0
bb cC2
(1 − bb) cC2
bb(1 − cC2)
(1 − bb)(1 − cC2)
cC2
cC2
(1 − cC2)
p
p
\C
m1
1
\C
m1
1
\C
m1
1
\C
m1
1
p
p
\C
m1
1
p
\C
p
m1−i
1
− \C
m1−i+1
1
p
\C
m1
1
p
(1 − cC2) \C
p
m1−i
1
− \C
p
m1−i+1
1
cC2
\C
p
m1−i
1
(1 − cC2) \C
p
m1−i
1
− \C
p
m1−i+1
1
− \C
p
m1−i+1
1
− \C
p
m2−j+1
2
− \C
p
m2−j+1
2
2φ(pj
1)2φ(pj
2)
8p
m1 −i
1
p
m2−j
2
2φ(p
j
1)2φ(p
j
2)
8p
m1 −i
1
p
m2−j
2
Similar results were obtained by Dutra [23, Cap´ıtulos 4 e 5] for group
codes over the quaternion groups.
3.2 Metacyclic Codes and Equivalence Questions
A group G is metacyclic if it contains a normal cyclic subgroup H such
that G/H is also cyclic. It is easy to prove that a finite metacyclic group has
the following presentation
G =(cid:10)a, b am = 1, bn = as, bab−1 = ai(cid:11) ,
(9)
with a and b such that H = hai and G/H = hbHi, for m, n ∈ N and
1 ≤ s, i ≤ m such that sm, ms(i − 1), i < m, gcd(i, m) = 1. For s = m,
we say that G is a split metacyclic group and, in this case, G is the
semi-direct product G = hai ⋊ hbi.
Earlier approaches on non-abelian metacyclic codes include results ob-
tained by Sabin [61] and Sabin and Lomonaco [62], where we also find the
following definition of equivalence of codes.
Definition 3.7. Let G and H be two finite groups of the same order and
Fq a field. A combinatorial equivalence is a vector space isomorphism
ψ : FqG −→ FqH induced by a bijection ψ : G −→ H.
Two codes C ⊂ FqG and C ⊂ FqH are said to be combinatorially
equivalent if there exists a combinatorial equivalence ψ : FqG −→ FqH
such that ψ(C) = C.
12
For G a metacyclic finite group such that gcd(q, G) = 1, Sabin and
Lomonaco [62], by using group representation theory, proved that codes
generated by central idempotents in FqG are combinatorially equivalent to
abelian codes. This motivated the search for left minimal codes in FqG.
Considering the group algebra of a non-abelian split metacyclic group
G over a finite field Fq, Assuena [3] in his Ph.D. thesis, found a necessary
condition under which FqG has the minimum number of simple components.
Theorem 3.8. [3, Teorema 2.1.16] Let G be a metacyclic group and Fq a
finite field with q elements such that gcd(q, G) = 1. If the number of simple
components of the group algebra FqG is minimal, then U(Zn) = h¯qi and
U(Zm) = h¯iih¯qi.
In his thesis, Assuena used the structure of the group to determine the
minimal metacyclic codes for a non abelian split metacyclic group of order
pmℓn, with p and ℓ odd prime numbers, under the conditions that FqG is
semisimple and the number of simple components of FqG is minimum.
For Dpm, the dihedral group of order 2pm, and Fq a finite field such that
gcd(q, 2pm) = 1, he constructs left minimal codes that are not combina-
torially equivalent to abelian codes and also exhibits one case where a left
minimal code is more efficient then the abelian ones of the same length,
giving a positive answer to a conjecture of Sabin and Lomonaco [62].
Further studies on group codes are given in [11], where it is defined a
(left) G-code as any linear (left) code of length n over a field Fq which is
the image of a (left) ideal of a group algebra via an isomorphism FqG −→ Fn
q
which maps the finite group G of order n to the standard basis of Fn
q . Their
ideas are used in [63] to study two-sided and abelian group ring codes and
in [30], where Garc´ıa Pillado et al. first communicated an example of a non-
abelian S4-code over F5. The full proof of this computacional construction
was given later in [31]. New examples of non-abelian G-codes are given in [32]
and, particularly, using the group SL(2; F3) instead of the symmetric group,
they prove, without using a computer for it, that there is a code over F2
of length 24, dimension 6 and minimal weight 10. This code has greater
minimum distance than any abelian group code having the same length and
dimension over F2, and, moreover, it has the greatest minimum weight among
all binary linear codes with the same length and dimension.
In [24] Elia and Garc´ıa Pillado give an overview of the properties of ideal
group codes defined as principal ideals in the group algebra of a finite group
13
G over a finite field Fq and present their encoding and syndrome decoding.
They also describe in detail a correction of a single error, using syndromes.
4 Cyclic codes of length 2m
Codes are usually considered over the binary field F2. For cyclic codes of
length 2m, with a natural m ≥ 1, over a field of odd size, the results obtained
using a polynomial approach by Bakshi and Raka [4], Pruthi [59], Sharma
et al. [67], Sharma et al. [66] and using the group algebra approach, by
Prado [58] in her Ph.D. thesis, are essentially the same. In Chapter 2, Prado
states the general facts:
Theorem 4.1. [58, Lema 2.1.1] Let G = hai be a finite cyclic group of order
2m, m ≥ 1 and Fq a finite field of odd characteristic. Let
G = G0 ⊃ G1 ⊃ · · · ⊃ Gm = {1}
be the descending chain of all subgroups of G, with Gi = ha2ii and Gi =
2m−i. Then the elements e0 = bG and ei =cGi − dGi−1, with 1 ≤ i ≤ m, form
a set of orthogonal idempotents of FqG such that e0 + e1 + · · · + em = 1.
Theorem 4.2. [58, Lema 2.2.1] Under the same hypothesis of Theorem 4.1,
let Ii = FqGei, with 1 ≤ i ≤ m, be the ideals of FqG generated by the
idempotents ei of Theorem 4.1. Then
dim(I0) = 1,
dim(Ii) = 2i−1,
d(I0) = G = 2m
d(Ii) = G = 2m−i+1, for 1 ≤ i ≤ m.
The notion of a visible code was given by Ward [74], where he defines
a visible basis for a code as a basis where all its elements have the same
weight. Prado also proved the following for codes of length 2m.
Theorem 4.3. [58, Proposi¸cao 2.3.1] Under the same hypothesis of Theo-
rem 4.1, for 1 ≤ i ≤ m, the set
Bi = {ei, aei, a2ei, . . . , a2i−1−1ei}
is a visible basis for the code Ii = FqGei.
14
In her thesis [58], Prado studied in details the minimal codes generated
by primitive idempotents in FqC2m, with q odd. She considered four cases:
q ≡ 1(mod 8), q ≡ 3(mod 8), q ≡ 5(mod 8) and q ≡ 7(mod 8). The order of
q(mod 2m), the number of simple components of FqC2m and the computation
of idempotents are different for each one of these cases. For q ≡ 3(mod 8) and
q ≡ 5(mod 8) a complete discussion is presented in the thesis and the other
cases are exemplified with particular examples. Here is the case q ≡ 3(mod 8).
Theorem 4.4. [58, Proposi¸cao 3.1.1] Let Fq be a field with q elements such
that q ≡ 3(mod 8) and G = ha a2m = 1i be a cyclic group of order 2m. The
following elements of the group algebra FqG
1 + a + a2 + · · · + a2m−1
2m
1 − a + a2 − · · · − a2m−1
2m
1 − a2 + a4 − · · · − a2m−2
e0 =
e1 =
e2 =
e3 = (1 − a4)
3 = (1 − a4)
e′
e4 = (1 − a8)
e′
4 = (1 − a8)
2m−1
(1 + a23 + · · · + a2m−23)(2 + αa + αa3)
(1 + a23 + · · · + a2m−23)(2 − αa − αa3)
2m
2m
(1 + a24 + · · · + a2m−24)(2 + αa2 + αa3·2)
(1 + a24 + · · · + a2m−24)(2 − αa2 − αa3·2)
2m−1
2m−1
. . . ,
em−1 = (1 − a2m−2)
m−1 = (1 − a2m−2)
e′
(1 + a2m−1)(2 + αa2m−4 + αa3·2m−4)
(1 + a2m−1)(2 − αa2m−4 − αa3·2m−4)
24
em = (1 − a2m−1)
m = (1 − a2m−1)
e′
24
(2 + αa2m−3 + αa3·2m−3)
23
(2 − αa2m−3 − αa3·2m−3)
23
form a complete set of primitive idempotents of FqG, with α2 = −2 in Fq.
15
In Chapter 4 of her thesis, Prado simplifies results of Poli [56] in order to
obtain a clearer description of the principal nilpotent ideals of a group algebra
of finite abelian groups in a modular case (i.e., when char(Fq) divides the
order of the group G). She also exemplifies the process of lifting idempotents
modulo a nilpotent ideal.
5 More on equivalence of abelian codes
The question of equivalence in Coding Theory has many approaches.
In
Section 3.2, we have defined combinatorial equivalence. In [50], we found the
following for abelian codes. Here G stands for a finite abelian group and Fq
is a finite field with q elements.
Definition 5.1. Two abelian codes I1 and I2 are G-equivalent if there exists
an automorphism θ of G whose linear extension to FqG maps I1 on I2.
The following statements also appeared in [50].
Theorem A [50, Theorem 3.6] Let G be a finite abelian group of odd order
and exponent n and denote by τ (n) the number of divisors of n. Then there
exist precisely τ (n) non G-equivalent minimal abelian codes in F2G.
Theorem B [50, Theorem 3.9] Let G be a finite abelian group of odd order.
Then two minimal abelian codes in F2G are G-equivalent if and only if they
have the same weight distribution.
Unfortunately both statements are not correct. The errors arise from the
assumption, implicit in the last paragraph of [50, p. 167], that if e and f are
primitive idempotents of F2Cm and F2Cn, respectively, then ef is a primitive
idempotent of F2[Cm × Cn]. To the best of our knowledge, these results have
not been used in a wrong way in the literature.
We first communicated the following counterexamples to both Theorems
A and B in [26].
Proposition 5.2. [26, Proposition 3.1] Let p be an odd prime such that ¯2
generates U(Zp2) and G = hai × hbi an abelian group, with o(a) = p2 and
o(b) = p. Then F2G has four inequivalent minimal codes, namely, the ones
16
generated by the idempotents e0 = bG, e1 =bb − \hapi × hbi, e2 =ba − bG and
e3 = \hapi × hbi − bG.
Also all minimal codes of F2G are described in Table 1 with their dimen-
sion and weight. Moreover, the minimal inequivalent codes I2 and I3 have
Code Primitive Idempotent Dimension Minimum Weight
I0
I1
I1j
I2
I2i
I3
1
p2 − p
p2 − p
p − 1
p − 1
p − 1
p3
2p
2p
2p2
2p2
2p2
j = 1, . . . , p − 1
e0 =babb = bG
e1 =bb − \hapi × hbi
e1j = dajpb − \hapi × hbi
e2 =ba − bG
e2i = cabi − bG
e3 = \hapi × hbi − bG
i = 1, . . . , p − 1
Table 1: Minimal codes in F2(Cp2 × Cp)
the same weight distribution.
In [26, Proposition 4.2] we showed that Theorem A holds in the special
case of minimal codes in F2(Cpn × Cpn) and, in [27, Theorem V.3], we gen-
eralize these result for G a direct product of m ≥ 2 copies of a cyclic group
Cpn, as follows.
Proposition 5.3. [27, Proposition V.3] Let m and r be positive integers and
p a prime number. If G = (Cpr)m is a finite abelian p-group and Fq is a field
of char(Fq) 6= p. Then a primitive idempotent of FqG, different from bG, is
of the form bK · eh, where K is a subgroup of G isomorphic to (Cpr)m−1 and
eh is a primitive idempotent of Fqhhi, where h ∈ G is such that G = hhi × K
and hhi ∼= Cpr.
This result can be applied as follows.
Corollary 5.4. [27, Corollary V.4] Let m and r be positive integers, p a
prime number, a finite abelian p-group G = (Cpr)m and Fq a finite field with
q elements such that o(¯q) = φ(pr) in U(Zpr ). Then the minimal abelian codes
17
in FqG are as follows, where h and K are as in Proposition 5.3.
Primitive Idempotent Dimension
Weight
· · ·
bG
bK(bhp −bh)
bK(chp2 − bhp)
bK(chp3 − chp2)
bK(chpi − [hpi−1)
bK(1 − [hpr−1)
· · ·
1
p − 1
p(p − 1)
p2(p − 1)
· · ·
prm
2pr(m−1)+(r−1)
2pr(m−1)+(r−2)
2pr(m−1)−(r−3)
pi−1(p − 1) 2pr(m−1)−(r−i)
· · ·
pr−1(p − 1)
2pr(m−1)
Consequently, the number of non G-equivalent minimal abelian codes is
r + 1 = τ (pr).
Corollary 5.5. [27, Corollary V.5] Let n, m ≥ 2 be integers, G = (Cn)m
an abelian group and Fq a finite field such that gcd(q, n) = 1. Then the
primitive idempotents of FqG are of the form bK · eh, where K is a subgroup
of G isomorphic to (Cn)m−1, h ∈ G is such that G = K × hhi and eh is a
primitive idempotent of Fqhhi.
Theorem 5.6. [27, Theorem V.6] Let G = C n be a direct product of cyclic
groups isomorphic to one another, of exponent n, and Fq a finite field such
that char(Fq) 6
G. Then, the number of non G-equivalent minimal abelian
codes is precisely τ (n).
We fully discussed the G-equivalence of abelian codes and established
in [27, Section III] a relation between the classes of equivalence of G-equivalent
codes and some classes of isomorphisms of subgroups of G, as follows.
We say that two subgroups H and K of a group G are G-isomorphic if
there exists an automorphism ψ ∈ Aut(G) such that ψ(H) = K.
Notice that isomorphic subgroups are not necessarily G-isomorphic. For
example, for a prime p, if G = hai × hbi with o(a) = p2 and o(b) = p,
then hapi and hbi are isomorphic, as they are both cyclic groups of order p.
However, they are not G-isomorphic, since hbi is contained, as a subgroup
of index p, only in hapi × hbi while hapi is contained in hai and in haibi, for
18
all 1 ≤ i ≤ p − 1. An automorphism of G carrying one to the other would
preserve also inclusions.
We shall denote by P(FqG) the set of all primitive idempotents of FqG.
Recall the notion of co-cyclic subgroup (Definition 2.1). Then, under the
same hypotheses of Lemma 2.11, the following map is well-defined
(10)
Φ : P(FqG) −→ Scc(G) ∪ {G}
7−→ Φ(e) = He,
7−→
G
e 6= bG
bG
where He is the unique co-cyclic subgroup of G such that e · eHe = e.
Theorem 5.7. [27, Theorem II.9] Let G be a finite abelian group, Fq a field
such that char(Fq) 6
G and H ∈ Scc(G). Then eH is the sum of all primitive
idempotents e ∈ P(FqG) such that Φ(e) = H.
The study of the G-equivalence of ideals involves to know how the group
of automorphisms Aut(G) acts on the lattice of the subgroups of G and hence
on the idempotents in the group algebra which arise from these subgroups.
From now on, we use the same notation for an automorphism of the group
G and its linear extension to the group algebra FqG. The following results
from [27] relate subgroups in G and idempotents in FqG.
Lemma 5.8. [27, Lemma III.1] Let G be a finite abelian group, H ∈ Scc(G)
and eH its corresponding idempotent defined as in (6). Then, for any ψ ∈
Aut(G), we have ψ(eH ) = eψ(H) and ψ(bG) = bG.
For finite abelian groups, Propositions 5.9, 5.10 and 5.11 below estab-
lish a correspondence between G-equivalent minimal ideals in FqG and G-
isomorphic subgroups of G.
Proposition 5.9. [27, Proposition III.2] Let G be a finite abelian group and
G. If e, e′ ∈ P(FqG) are such that ψ(e) = e′,
Fq a field such that char(Fq) 6
for some automorphism ψ ∈ Aut(G) linearly extended to FqG, then
ψ(He) = Hψ(e) = He′,
i.e., He and He′ are G-isomorphic.
We set LAut(G) = {ψ ∈ Aut(G) ψ(H) = H, for all H ≤ G}.
19
Proposition 5.10. [27, Proposition III.7] Let G be a finite abelian group
G. If e′, e′′ ∈ P(FqG) are both different
and Fq a field such that char(Fq) 6
from bG and He′ = He′′, then there exists an automorphism ψ ∈ LAut(G)
whose linear extension to FqG maps e′ to e′′.
The following is the converse of Proposition 5.9.
Proposition 5.11. [27, Proposition III.8] Let G be a finite abelian group
G. If e′, e′′ ∈ P(FqG), both different
and Fq a field such that char(Fq) 6
from bG, are such that ψ(He′) = He′′, for some ψ ∈ Aut(G), then there exists
an automorphism θ ∈ Aut(G) whose linear extension to FqG maps e′ to e′′,
i.e., the ideals of FqG generated by e′ and e′′ are G-equivalent.
As an application of Propositions 5.9 and 5.11, in [27, Section IV] we
consider the minimal codes in F2(Cpn × Cp), for an odd prime p and n ≥ 3.
Its proof is similar to the proof of Proposition 5.2. This gives a whole family
of counterexamples to Theorem A.
Proposition 5.12. [27, Theorem IV.3] Let n ≥ 3 be a positive integer and p
an odd prime such that ¯2 generates U(Zpn) and G = hai × hbi be an abelian
group, with o(a) = pn and o(b) = p. Then the minimal codes of F2G are
described in Table 2. Moreover, there are 2n inequivalent minimal codes in
F2(Cpn × Cp).
In the first column of Table 3 we give a complete list of representatives
of classes of G-isomorphisms of subgroups of Cpn × Cp and, in the second
column, we list the corresponding representatives of G-equivalent classes of
minimal codes of F2(Cpn × Cp).
6 Cyclic and abelian codes of length pnqm
6.1 Binary abelian codes
In [15], we considered finite abelian groups of type G = Gp × Gq, for distinct
primes p and q such that Gp is a p-group, Gq is a q-group satisfying the
following conditions which will allow us to use the results in [28]:
(i)
(ii)
(iii)
gcd(p − 1, q − 1) = 2,
¯2 generates the groups of units U(Zp2) and U(Zq2)
gcd(p − 1, q) = gcd(p, q − 1) = 1.
(11)
20
Code
Dimension Weight
i = 1, . . . , p − 1
. . .
i = 0, . . . , p − 1
I2 = h \hap2i × hbi − \hapi × hbii
I0 = hbabbi = hbGi
I1 = h \hapi × hbi − bGi
I1i = hcabi − bGi
I2i = hdapbi − \hapi × hbii
Ik = h \(cid:10)apk(cid:11) × hbi − \(cid:10)apk−1(cid:11) × hbii
Iki = h\apk−1bi − \(cid:10)apk−1(cid:11) × hbii
In−1 = hchbi − \hapn−2i × hbii
In−1,i = h \apn−1bi − \hapn−2i × hbii
i = 1, . . . , p − 1
i = 1, . . . , p − 1
. . .
pn+1
2pn
2pn
2pn−1
2pn−1
1
p − 1
p − 1
p(p − 1)
p(p − 1)
. . .
pk−1(p − 1)
pk−1(p − 1)
2pn−k+1
2pn−k+1
. . .
p(n−1)(p − 1)
p(n−1)(p − 1)
2p
2p
Table 2: Minimal codes in F2(Cpn × Cp)
The hypothesis (i) above implies that at least one of the primes p and q
is congruent to 3 (mod 4). In this section, to fix notations, we shall always
assume that q ≡ 3 (mod 4). As a code (ideal) generated by a primitive
idempotent is isomorphic to a field, condition (i) also helps us to have some
control on the number of simple components that appear in the group algebra
Fq(Gp × Gq), because of the following elementary facts of Number Theory.
Lemma 6.1. Let ℓ be a positive prime number and r, s ∈ N∗. Then
Fℓr ⊗Fℓ Fℓs ∼= gcd(r, s) · Fℓlcm(r,s).
Lemma 6.2. Let r, s ∈ N be non-zero elements such that gcd(r, s) = 2. Let
u ∈ F2r and v ∈ F2s be elements satisfying the equation x2 + x + 1 = 0.
Then
(12)
and e1 = (u ⊗ v) + (u2 ⊗ v2) and e2 = (u ⊗ v2) + (u2 ⊗ v) are the primitive
idempotents generating to the simple components of (12).
F2r ⊗F2 F2s ∼= F2
2 ⊕ F2
rs
rs
2
21
Subgroups
Codes
. . .
. . .
hapbi
G
hai
hapi × hbi
I2 = h \hap2i × hbi − \hapi × hbii
I0 = hbGi
I11 = hba − bGi
I1 = h \hapi × hbi − bGi
I21 = hcapb − \hapi × hbii
Dap2E × hbi
DapkbE
Ik+1,1 = hdapkb − \(cid:10)apk(cid:11) × hbii
Dapk+1E × hbi Ik+1 = h \hapk+1i × hbi − \(cid:10)apk(cid:11) × hbii
In−1 = hbb − \hapn−1i × hbii
. . .
. . .
hbi
Table 3:
Methods to determine idempotent generators for minimal cyclic codes
were given in [5, 6, 70] using representation theory. We develop our results
without appealing to representation theory, working inside the group algebra.
For two co-cyclic subgroups H of Gp and K of Gq, consider the respective
idempotents eH = bH − cH ∗ in F2Gp and eK = bK − cK ∗ in F2Gq. Clearly
cGp ·cGq = \Gp × Gq is a primitive idempotent of F2G = F2(Gp × Gq).
It is ease to prove that idempotents of the form cGp · eK and eH · cGq
argument. For eH = bH − cH ∗, set a ∈ H ∗ \ H (hence aH is a generator of
are primitive in F2G. We proved that each idempotent of the form eH · eK
decomposes as the sum of two primitive idempotents in F2G, by the following
H ∗/H). Set
and
1 + a20 + a22 + · · · + a2p−3,
u = (cid:26) a20 + a22 + · · · + a2p−3,
u′ = (cid:26) a2 + a23 + · · · + a2p−2,
1 + a2 + a23 + · · · + a2p−2,
if p ≡ 1(mod 4) or
if p ≡ 3(mod 4)
if p ≡ 1(mod 4) or
if p ≡ 3(mod 4)
(13)
(14)
replacing a by b. As gcd(pr−1(p−1), qs−1(q−1)) = 2 we can apply Lemma 6.2
For eK = bK −cK ∗, set b ∈ K ∗ \ K and define v and v′ as in (13) and (14)
22
to see that
e1(H, K) = ubH · vbK + u′bH · v′bK and
e2(H, K) = ubH · v′bK + u′bH · vbK
are primitive orthogonal idempotents such that e1 + e2 = eHeK.
Hence, we have shown the following.
Theorem 6.3. [15, Theorem III.1] Let Gp and Gq be abelian p and q-groups,
respectively satisfying the conditions in (11). For a group G, denote by S(G)
the set of subgroups N of G such that G/N 6= 1 is cyclic. Then the set of
primitive idempotents in F2[Gp × Gq] is:
cGp ·cGq,
cGp · eK,
eH ·cGq,
e1(H, K), e2(H, K),
K ∈ S(Gq),
H ∈ S(Gp),
H ∈ S(Gp), K ∈ S(Gq).
Particularly, in [15, Section IV] we compute, for each minimal code of
F2(Cp × Cq), the generating primitive idempotent, its dimension and give
explicitly a basis for it over F2.
In [15, Theorem IV.7] we presented the
results on minimum weight for these codes. In [15, Theorem V.1] we deal
with the case F2(Cpm × Cqn), for m ≥ 2, n ≥ 2 and extend this technique for
three primes as follows.
Theorem 6.4. [15, Theorem IV.10] Let p1, p2 and p3 be three distinct
positive odd prime numbers such that gcd(pi − 1, pj − 1) = 2, for 1 ≤ i 6=
j ≤ 3, and ¯2 generates the groups of units U(Zpi). Then the primitive
idempotents of the group algebra F2G for the finite abelian group G = Cp1 ×
Cp2 × Cp3, with Cp1 =< a >, Cp2 =< b > and Cp2 =< c >, are
e0 = abc, e1 = ab(1 − c), e2 = a(1 − b)c, e3 = (1 − a)bc,
e4 = (uv + u2v2)c, e5 = (u2v + uv2)c e6 = (uw + u2w2)b,
e7 = (u2w + uw2)b e8 = (vw + v2w2)a, e9 = (v2w + vw2)a
e10 = (1 − a)(1 − b)(1 − c) + u2v2w + uvw2
e11 = (1 − a)(1 − b)(1 − c) + u2v2w2 + uvw
e12 = (1 − a)(1 − b)(1 − c) + u2vw + uv2w2 and
e13 = (1 − a)(1 − b)(1 − c) + uv2w + u2vw2,
where u = u(a), v = v(b), w = w(c) are defined as in (13).
23
Comparing and using both the group algebra techniques of [15, 28] with
the polynomial techniques of [5], Bastos and Guerreiro [7, 8] improved the
presentation of minimal idempotents of length pnq given in [41], correcting
some coefficients in their expressions.
6.2 Codes of length pn also for non-cyclic abelian groups
Let Fq be a finite field with q elements and G a cyclic group of order pn
generated by a such that gcd(q, p) = 1. Then the group algebra FqG is
semisimple and each of its ideals is a direct sum of minimal ones. Under
the conditions (b) and (c) of Corollary 2.4, the minimal ideals (codes) are
generated by the primitive idempotents given by Theorem 2.5.
In her thesis [48], Melo first considered all cyclic codes of FqG, that is,
not only the minimals and computed dimension and minimum weights of
these codes, using the following result.
Lemma 6.5. [25, Proposi¸cao 2.1] Under the hypothesis above and considering
Ii the minimal ideal of FqG generated by the primitive idempotent ei, as
in (4), for 1 ≤ i ≤ n, we have
d(Ii) = 2Gi = 2pn−i
and
dimFq Ii = φ(pi) = pi − pi−1,
and a basis for Ii is
Bi = {a(1 − b)cGi a ∈ A, 1 6= b ∈ B},
with A a transversal of Gi in Gi−1 and B a transversal of Gi in G. For the
minimal code I0 = (FqG)e0, we have
w(I0) = pn
and
dimFq I0 = 1.
Considering that the dimension of a direct sum of ideals is the sum of
their dimensions, Melo [48, 49] focused her attention on computing minimum
weight of the direct sum of minimal ideals as follows.
Theorem 6.6. Under the hypothesis of this section and of Lemma 6.5, we
have:
(i) [48, Lema 2.3] if 0 < i < j, then w(Ii ⊕ Ij) = 2Gj = 2pn−j.
(ii) [48, Lema 2.4] If 1 < j, then w(I0 ⊕ Ij) = 2Gj = 2pn−j.
(iii) [48, Lema 2.5] If I = I0 ⊕ I1, then w(I) = G1 = pn−1.
24
Gt = pn−t.
(iv) [48, Lema 2.6] If I =Lt
(v) [48, Lema 2.7] If I =Lt
i=0(FqG)ei, then I = (FqG)cGt and w(I) =
k=0(FqG)eik, with 0 ≤ i1 < i2 < · · · < it and
ei1 + ei2 + · · · + eit 6= e0 + e1 + · · · + et, then w(I) = 2Git = 2pn−it.
Melo [48, Section 2.3] also considered the distribution of weights for these
cyclic codes. Furthermore, in [48, Chapter 3], she briefly compared cyclic
and non-cyclic abelian codes of length p2, fully exploring some examples
using GAP Wedderga package.
For the group G = Cp × Cp = hai × hbi and Fq a finite field of q elements
such that ¯q generates U(Zp), the idempotents of FqG are
e0 = bG, e1 =ba − bG, e2 =bb − bG, fi = cabi − bG, with 1 ≤ i ≤ p − 1.
Note that if H and K are any among the subgroups hai, hbi,habii, with
1 ≤ i ≤ p − 1, then G = H × K. For the idempotents e = bH − bG and
e = bK − bG associated to H and K, respectively, and considering the ideal
I = (FqG)e ⊕ (FqG)f , Melo proved:
Theorem 6.7. [48, Teorema 3.2.1] The minimum weight of the ideal I is
d(I) = 2p − 2 and its dimension is dimFq I = 2p − 2.
6.3 Essential idempotents an one weight cyclic codes
In [16], a special type of idempotent elements in the semisimple group algebra
of a finite abelian group is considered, the so called essencial idempotents.
These idempotents were previously considered by Bakshi, Raka and Sharma
in [6], where they were called non-degenerate, in the special case of group
algebras of cyclic groups over finite fields.
Definition 6.8. In a semisimple group algebra FqG of a finite group G, a
primitive idempotent e is an essential idempotent if ebH = 0, for every
subgroup H 6= {1} in G. A minimal ideal of FqG is called an essential
ideal if it is generated by an essential idempotent.
The following is a characterization of essential idempotents.
Proposition 6.9. [16, Proposition 2.3] Let e ∈ FqG be a primitive central
idempotent. Then e is essential if and only if the map π : G −→ Ge is a
group isomorphism.
25
Corollary 6.10. [16, Corollary 2.4] If G is an abelian group and FqG con-
tains an essential idempotent, then G is cyclic.
For cyclic groups, Chalom, Ferraz and Polcino Milies [16] proved the
existence of a non-zero central idempotent which is the sum of all essential
idempotents. They also give a criteria to determine essential idempotents
using the well-known Galois descent method and, as a consequence, compute
the number of these idempotents in FqCn, for Cn a cyclic group of order n.
In [16, Section 3] they show that the coeficients of the primitive idem-
potents of a semisimple group algebra FqA, for A is a finite abelian group,
can be easily computed as a concatenation of the coeficients of an essential
idempotent in the group algebras of a cyclic factor of A. In terms of coding
theory, this will imply that every minimal abelian code generated by a non
essential idempotent is a repetition code: their elements can be written as
repetitions of the coeficients of elements in a cyclic code generated by an
essential idempotent. In particular, one application of this is to determine
the weight distribution of all codes when the weight distributions of codes
generated by essential idempotents are known.
Nascimento, in her Ph.D. Thesis [51], uses this notion of essential idem-
potents to state conditions for a cyclic code in FqCn to be a one-weight code.
Besides, she describes precisely the form of the elements on such a code and
determines the number of one-weight codes in FqCn. She also constructs
examples of two weight codes in Fq(Cn × Cn) and gives conditions to ensure
that a code is of constant weight in FqA, for A an abelian group. Her work
simplifies many of the proofs given by Vega [72] for the same facts. In the
literature there is also an interesting paper by Wood [78] on linear codes of
constant weight.
7 Codes over rings
In the 1990’s many papers on cyclic codes over rings started to appear,
motivated by the fact that good non linear binary codes were related to
linear codes over Z4 (see, for example, [17, 38, 54]). The paper [36] by
Hammons et al. was even the best paper award for Information Theory of
the IEEE-IT Society in the 1996 Symposium of IT - Whistler (Canad´a).
Wood [76] addressed the problem of duality for modules over finite chain
rings and applied it to equivalence of codes and to the extension theorem of
MacWilliams.
26
In [14] Carlderbank e Sloane determine the structure of cyclic codes over
Zpm. Later on, in [38] Kanwar and L´opez-Permouth did the same, but with
different proofs. With the same techniques, Wan [73] extended the results
from [38] to cyclic codes over Galois rings. Em 1999, Norton and Salagean-
Mandache in [52] extended results of [14, 38] to cyclic codes over finite chain
rings and later on, in 2004, Dinh and L´opez-Permouth in [19] prove the same
results in a different way.
Codes over rings developed even more in the beginning of the 21st century
that they deserved a CIMPA Summer School in 2008 [71]. Further works can
be found in [18], [42], [47]. A small survey on the subject is [33].
In his thesis [68, 69], Silva used group ring approach to characterize cyclic
codes over chain rings, their duals and some conditions on self-dual codes,
simplifying the proofs and improving results given in [19].
Let R be a finite commutative chain ring with unity such that R = qk,
for a prime q. For M the maximal ideal of R, the quotient R = R
M is a field
and we work under the hypotesis that q ∤ G , for a finite cyclic group G.
Under these conditions, the group ring RG is a principal ideal ring, as Silva
proves in [68, Teorema 2.1.9], after characterizing all the ideals in RG. The
following general fact is a basis for all this work.
Theorem 7.1. [68, Teorema 2.1.2] Let R be a local ring, with maximal ideal
M =< a > and R = qk, and G a cyclic group of order n such that q ∤ n.
If {e0, ..., em} is a full set of primitive orthogonal idempotents in RG, then
{e0, ..., em} is a full set of primitive orthogonal idempotents in RG.
The next theorem characterizes all cyclic codes of length n over the local
ring RGei (see [68, Corollary 11.31]), for R a chain ring and ei a primitive
orthogonal idempotent, translating results of [19] to the group ring setting.
To simplify the notation we write (RG)ajei as hajeii.
Theorem 7.2. [68, Teorema 2.1.3] Let R be a commutative finite chain ring
with unity, R = qk, M = hai the maximal ideal of R and t the nilpotency
index o ´ındice de nilpotencia of a in R. Let G = Cn such that q ∤ n. If I is
an ideal of RGei, then I is of the form I =(cid:10)akiei(cid:11), with 0 ≤ ki ≤ t.
Corollary 7.3. [68, Corollary 2.1.4] Under the same hypothesis of Theo-
rem 7.2, the ideal RGei is indecomposable in RG and the code hat−1eii is
minimal.
From this we have a characterization of all cyclic codes of length n over
chain rings.
27
Theorem 7.4. Let R be a commutative finite chain ring with unity, R =
qk, M = hai the maximal ideal of R and t the nilpotency index of a in R. Let
0 = 1i be such that q ∤ n and {e0, ..., em} be a full set of primitive
G = hg0 / gn
orthogonal idempotents of RG. Then:
(i) [68, Teorema 2.1.5] If I is an ideal of RG, then I is of the form
I = I0 ⊕ ... ⊕ Im, with Ii =(cid:10)akiei(cid:11), for 0 ≤ ki ≤ t.
(t + 1)m+1.
(ii) [68, Teorema 2.1.8] The number of such codes of length n over R is
One important data in a code is its number of words. Next theorem gives
this number for cyclic codes over finite chain rings. We have
RG = RGe0 ⊕ ... ⊕ RGem ≃
R[x]
hxn − 1i
≃
R[x]
hf0i
⊕ ... ⊕
R[x]
hfmi
,
where fi are irreducible factors of xn − 1 and , after reordering the indexes
if necessary, we have RGei ≃ R[x]
hfii . Hence, RGei = R wi, for wi = deg(fi).
Theorem 7.5. [68, Teorema 2.1.7] Under the same hypothesis of Theo-
rem 7.2, let C be a cyclic code of the form C = (cid:10)aki1 ei1(cid:11) ⊕ ... ⊕(cid:10)akir eir(cid:11)
(t − kis)wis
in RG. The the number of words in C is C = R
.
rXs=1
Considering ∗ : RG −→ RG the classical involution, Silva also gives a
description of the dual cyclic codes in RG as follows.
Theorem 7.6. [68, Teorema 2.2.3] Under the same hypothesis of Theo-
rem 7.4, the dual code of a cyclic code C = (cid:10)ak0e0(cid:11) ⊕ ... ⊕(cid:10)akmem(cid:11), with
0 ≤ ki ≤ t, is C ⊥ = ⊕Pm
As in [19], Silva in [68, Section 2.2] states the conditions for the ring R
r=0(cid:10)at−kr er
∗(cid:11).
under which the group ring RG admits self-dual codes.
Chapter 3 of [68] is dedicated to codes over chain rings of length pn, for a
prime p, extending the results of Ferraz and Milies [28] and of Melo [48] to this
context. Silva also proves in [68, Teorema 3.0.14] some facts about the size
of such codes and computes minimum weight of these codes [68, Teoremas
3.0.15 to 3.0.18], similarly to Theorem 6.6. He also discusses about free codes
in RG in [68, Section 3.1] and about MDS codes of length pn over R in [68,
Section 3.2]. Finally, in [68, Chapter 4], Silva proves all such results for cyclic
codes of length 2pn over finite chain rings.
28
There are also interesting discussion on equivalence of linear codes over
rings in [20, 21, 75, 77].
References
[1] S.K. Arora, M. Pruthi, Minimal codes of prime power length. Finite Fields
and their Applications 3 (1997) 99-113.
[2] S.K. Arora, M. Pruthi, Minimal cyclic codes of length 2pn. Finite Fields and
their Applications 5 (1999) 177-187.
[3] S. Assuena, C´odigos Metac´ıclicos. Tese de Doutorado,
Matem´atica e Estat´ıstica da Universidade de Sao Paulo, 2013.
Instituto de
[4] G.K. Bakshi and M. Raka, Minimal cyclic codes of length 2n. Ranchi Univer-
sity Math. Journal 33 (2002) 1-18.
[5] G.K. Bakshi and M. Raka, Minimal cyclic codes of length pnq. Finite Fields
and their Applications 9 (2003) 432-448.
[6] G.K. Bakshi, M. Raka and A.Sharma, Idempotent generators of Irreducible
Cyclic Codes. Number Theory and Discrete Geometry, RMS Lecture Notes
Series 6 (2008), 13-18.
[7] G.T. Bastos, M. Guerreiro, Compara¸cao de t´ecnicas para o c´alculo de idem-
potentes geradores de c´odigos c´ıclicos. Atas do CNMAC 2014 (to appear).
[8] G.T. Bastos, M. Guerreiro, Idempotents generators for minimal cyclic codes
of length pnq. Proceedings of the 4th International Castle Meeting on Coding
Theory and Applications, Palmela, Portugal, September 2014 (to appear).
[9] S.D. Berman, Semisimple cyclic and abelian codes, II. Kybernetika 3 (1967)
21-30.
[10] S. D. Berman, On the theory of group codes. Kibernetika, 3 (1967) 31-39.
[11] J.J. Bernal, A. del Rio, J.J. Sim´on, An intrinsical description of group codes.
Designs, Codes and Cryptography 51 (3) (2009) 289-300.
[12] E. Biglieri and M. Elia, On the construction of group block codes. Annales des
T´el´ecommunications, 50 Issue 9-10 (1995) 817-823.
[13] O. Broche and A. del Rio, Wedderburn decomposition of finite group algebras.
Finite Fields and their Applications 13 (2007) 71-79.
29
[14] A. R. Calderbank and N. J. A. Sloane, Modular and p-adic codes. Designs,
Codes and Cryptography 6 (1995) 21-35.
[15] G. Chalom, R. Ferraz, M. Guerreiro and C. Polcino Milies, Minimal binary
abelian codes of length pnqn. Preprint in arXiv:1205.5699v1 [cs.IT] 2012.
[16] G. Chalom, R.A. Ferraz and C. Polcino Milies, Essencial idempotents in group
algebras and minimal cyclic codes. Preprint.
[17] J. H. Conway and N. J. A. Sloane, Self-dual codes over the integers modulo
4. J. Combinatorial Theory Series A, 62 (1993) 30-45.
[18] H. Q. Dinh, Complete distances of all negacyclic codes of length 2s over Z2a.
IEEE Transactions on Information Theory 53 (2007) 147-161.
[19] H. Q. Dinh and S. R. L´opez-Permouth, Cyclic and negacyclic codes over finite
chain rings. IEEE Transactions on Information Theory, 50 (2004) 1728-1744.
[20] H. Q. Dinh and S. R. L´opez-Permouth, On the equivalence of codes over finite
rings. AAECC 15 (2004) No. 1 37-50.
[21] H. Q. Dinh and S. R. L´opez-Permouth, On the equivalence of codes over rings
and modules. Finite Fields and their Applications 10 (2004) No. 4 615-625.
[22] V. Drensky and P. Lakatos, Monomial ideals, group algebras and error-
correcting codes. Applied Algebra, Algebraic Algorithms and Error-Correcting
Codes, Lecture Notes in Computer Science 357 (1989) 181-188.
[23] F.S. Dutra, Sobre c´odigos diedrais e quat´ernios. Tese de Doutorado, Univer-
sidade Federal de Minas Gerais, 2006.
[24] M. Elia and C. Garc´ıa Pillado, Ideal Group codes and their Syndrome De-
coding. Proceedings of the 21st International Symposium on Mathematical
Theory of Networks and Systems, Groningen, The Netherlands, July 2014.
[25] R.A. Ferraz, F.S. Dutra and C.Polcino Milies, Semisimple group codes and
dihedral codes. Algebra and Discrete Mathematics 3 (2009) 28-48.
[26] R.A. Ferraz, M. Guerreiro and C. Polcino Milies, Minimal codes in binary
abelian group algebras. Information Theory Workshop (ITW) IEEE (2011)
225-228.
[27] R.A. Ferraz, M. Guerreiro, C. Polcino Milies, G-equivalence in group algebras
and minimal abelian codes. IEEE Transactions on Information Theory 60 (1)
(2014) 252-260.
30
[28] R.A. Ferraz and C. Polcino Milies, Idempotents in group algebras and minimal
abelian codes. Finite Fields and their Applications 13 (2007) 382-393.
[29] G.D. Forney and M. Trott, The dynamics of group codes : state spaces, trel-
lis diagrams and canonical encoders. IEEE Trans. Inform. Theory 39 (1993)
1491-1593.
[30] C. Garc´ıa Pillado, S. Gonz´alez, V. T. Markov, C. Mart´ınez and A. A. Nechaev,
When are all group codes of a noncommutative group abelian (a computational
approach)?. Journal of Mathematical Sciences 186 No. 4 (2012) 578-585.
[31] C. Garc´ıa Pillado, S. Gonz´alez, V. T. Markov, C. Mart´ınez and A. A. Nechaev,
Group codes over non-abelian groups. Journal of Algebra and its Applications
12 No. 7 (2013) 20 pages.
[32] C. Garc´ıa Pillado, S. Gonz´alez, V. T. Markov, C. Mart´ınez and A. A. Nechaev,
New examples of non-abelian group codes. Proceedings of the 4th Interna-
tional Castle Meeting on Coding Theory and Applications, Palmela, Portugal,
September 2014 (to appear).
[33] M. Greferath, An introduction to ring-linear coding theory.In: Ed. M. Sala et
al. Grobner Bases, Coging and Cryptography, Springer-Verlag, Berlin Heidel-
berg, 2009.
[34] E.G. Goodaire, E. Jespers and C. Polcino Milies, Alternative Loop Rings.
North-Holland Mathematics Studies 184, Amsterdam, Holland: Elsevier,
1996.
[35] R. W. Hamming, Error detecting and error correcting codes. The Bell System
Technical Journal, 26 (1950) 147-160.
[36] A.R. Hammons Jr., P.V. Kumar, A.R. Calderbank, N.J.A. Sloane and P. Sol´e,
The Z4-linearity of Kerdock, Preparata, Goethals and related codes. IEEE
Transactions on Information Theory 40 (1994) 301-319.
[37] E. Jespers, G. Leal and A. Paques, Central idempotents in the rational group
algebra of a finite nilpotent group. Journal of Algebra and its Applications 2
No. 1 (2003) 57-62.
[38] P. Kanwar, S.R. L´opez-Permouth, Cyclic codes over the integers modulo pm.
Finite Fields and its Applications 3 (1997) 334-352.
[39] A.V. Kelarev, Ring constructions and applications. River Edge-NJ, World
Scientific, 2002.
31
[40] A.V. Kelarev and P. Sol´e, Error-correcting codes as ideals in group rings.
Contemporary Math., 273 (2001), 11-18.
[41] P. Kumar, S.K. Arora, λ-mapping and primitive idempotents in semisimple
rings Rm. Communications in Algebra 41 (10) (2013) 3679-3694.
[42] Z. H. Liu, Notes on linear codes over finite chain rings. Acta Mathematicae
Applicatae Sinica, 27 (2011) 141-148.
[43] V.O. J. Luchetta, C´odigos c´ıclicos como ideais em ´algebras de grupo. Dis-
serta¸cao de Mestrado, Instituto de Matem´atica e Estat´ıstica da Universidade
de Sao Paulo, 2005.
[44] F.J. MacWilliams, Codes and ideals in group algebra. Combinatorial Mathe-
matics and its Applications, University of North Carolina Press, 1969.
[45] F.J. MacWilliams, Binary codes which are ideals in the group algebra of an
abelian group. Bell System Tech. Journal 44 (1970) 987-1011.
[46] P.A. Martin, Grupos, Corpos e Teoria de Galois. Editora Livraria da F´ısica,
Sao Paulo, 2010.
[47] E. Mart´ınez-Moro and I. F. R´ua, On repeated-root multivariable codes over a
finite chain ring. Des. Codes Cryptogr. 45 (2007) 219-227.
[48] F.D. de Melo, Sobre c´odigos c´ıclicos e abelianos. Tese de Doutorado, Instituto
de Matem´atica e Estat´ıstica da Universidade de Sao Paulo, 2012.
[49] F.D. de Melo and C. Polcino Milies, On Cyclic and Abelian Codes. IEEE
Transactions on Information Theory 59 (11) (2013) 7314-7319.
[50] R.L. Miller, Minimal codes in abelian group algebras. Journal of Combinatorial
Theory, Series A 26 (1979) 166-178.
[51] R. Nascimento, C´odigos de peso constante. Tese de Doutorado, Instituto de
Matem´atica e Estat´ıstica da Universidade de Sao Paulo, 2014.
[52] G. Norton and A. Salagean-Mandache, On the structure of linear cyclic codes
over finite chain rings. Appl. Algebra Eng. Commun. Comput. 10 (2000)
489-506.
[53] A. Olivieri, A. del R´ıo and J.J. Sim´on, On monomial characters and central
idempotents of rational group algebras. Comm. Algebra 32 (4) (2004) 1531-
1550.
32
[54] V. Pless, Z. Qian, Cyclic codes and quadratic residue codes over Z4. IEEE
Transactions on Information Theory 42 (1996 ) 1594-1600.
[55] C. Polcino Milies and S.K. Sehgal, An Introduction to Group Rings. Kluwer
Academic Publishers, Dordrecht, 2002.
[56] A. Poli, Ideaux principaux nilpotents de dimension maximale dans l´algebre
Fq[X] d´um groupe abelien fini G. Communications in Algebra, 12 (4) (1984)
391-401.
[57] A. Poli, Important algebraic calculations for n-variables polynomial codes. Dis-
crete Mathematics 56 (1985) 255-263.
[58] J. do Prado, Idempotentes geradores de c´odigos minimais. Tese de Doutorado,
Instituto de Matem´atica e Estat´ıstica da Universidade de Sao Paulo, 2010.
[59] M. Pruthi, Cyclic codes of length 2m. Proc. Indian Acad. Sci. (Math. Sci.)
111 (2001) 371-379.
[60] J.J. Rotman, An Introduction to the Theory of Groups, 4th Ed., Springer-
Verlag, New York, 1995.
[61] R.E. Sabin, On row-cyclic codes with algebraic structure. Designs, Codes and
Cryptography 4 (1994) 145-155.
[62] R.E. Sabin and S.J. Lomonaco, Metacyclic Error-Correcting Codes. AAECC
6 (1995) 191-210.
[63] A. Schafer, Two-sided and abelian group ring codes. Master of Science Thesis,
Aachen University, Germany, 2012.
[64] C. Shannon, A Mathematical Theory of Communication, The Bell System
Technical Journal, bf 27 (1948) 379-423 July and 623-656 October.
[65] A. Sharma, G.K. Bakshi, V.C. Dumir and M. Raka, Cyclotomic numbers and
− 1 >. Finite Fields Appl.
primitive idempotents in the ring GF (q)[x]/ < xpn
10 (2004) 653-673.
[66] A. Sharma, G.K. Bakshi and M. Raka, The weight distribution of irreducible
cyclic codes of length 2m. Finite Fields and Applications 13 (2007) 1086-1095.
[67] A. Sharma, G.K. Bakshi, V.C. Dumir and M. Raka, Irreducible cyclic codes
of length 2n. Ars Combinatoria 86 (2008) 133-146.
[68] A.T. Silva, C´odigos c´ıclicos sobre an´eis de cadeia. Tese de Doutorado, Insti-
tuto de Matem´atica e Estat´ıstica da Universidade de Sao Paulo, 2012.
33
[69] A.T. Silva and C. Polcino Milies, On cyclic codes over finite chain rings.
Preprint.
[70] R. Singh and M. Pruthi, Primitive idempotents of irreducible quadratic residue
cyclic codes of length pnqm. International Journal of Algebra, 5 N.6 (2011)
285 - 294.
[71] P. Sol´e (Editor), Codes over Rings. Series on Coding Theory and Cryptology
6. Proceedings of the CIMPA Summer School, Ankara, Turkey, 2008.
[72] G. Vega, Determining the number of one-weight cyclic codes when length and
dimension are given. Lecture Notes in Computer Science 4547 (2007) 284-
293.
[73] Z. Wan, Cyclic codes over Galois rings. Alg. Colloquium 6 (1999) 291-304.
[74] H.N. Ward, Visible codes. Archiv der Mathematik 54 Issue 3 (1990) 307-312.
[75] H.N. Ward and J.A. Wood, Characters and the equivalence of codes. J. Comb.
Theory Series A 73 (1996) No. 2 348-352.
[76] J.A. Wood, Duality for modules over finite rings and applications to coding
theory. American Journal of Mathematics 121 (1999) 555-575.
[77] J.A. Wood Code equivalence characterizes finite Frobenius rings. Proc. Amer.
Math. Soc. 136 (2008) 699-706.
[78] J.A. Wood, The structure of linear codes of constant weight. Trans. Amer.
Math. Soc. 354 (2002) No. 3 1007-1026.
34
|
1004.2467 | 4 | 1004 | 2011-04-28T16:00:43 | The linear preservers of non-singularity in a large space of matrices | [
"math.RA"
] | Let K be an arbitrary (commutative) field, and V be a linear subspace of M_n(K) such that codim V<n-1. Using a recent generalization of a theorem of Atkinson and Lloyd, we show that every linear embedding of V into M_n(K) which strongly preserves non-singularity must be M->PMQ or M->PM^TQ for some pair (P,Q) of non-singular matrices of M_n(K), unless n=3, codim V=1 and K is isomorphic to F_2. This generalizes a classical theorem of Dieudonn\'e with a similar strategy of proof. Weak linear preservers are also discussed, as well as the exceptional case of a hyperplane of M_3(F_2). | math.RA | math | The linear preservers of non-singularity in a large
space of matrices
Cl´ement de Seguins Pazzis∗†
November 5, 2018
Abstract
Let K be an arbitrary (commutative) field, and V be a linear subspace
of Mn(K) such that codim V < n−1. Using a recent generalization of a the-
orem of Atkinson and Lloyd [11], we show that every linear embedding of V
into Mn(K) which strongly preserves non-singularity must be M 7→ P M Q
or M 7→ P M T Q for some pair (P, Q) of non-singular matrices of Mn(K),
unless n = 3, codim V = 1 and K ≃ F2. This generalizes a classical theorem
of Dieudonn´e with a similar strategy of proof. Weak linear preservers are
also discussed, as well as the exceptional case of a hyperplane of M3(F2).
AMS Classification : 15A86; 15A30
Keywords : linear preservers, non-singular matrices, dimension, codimension.
1
Introduction
1.1 Notations and goals
Here, K will denote an arbitrary (commutative) field and n a positive integer.
By a line in a vector space, we will always mean a 1-dimensional linear subspace
of it.
We let Mn,p(K) denote the set of matrices with n rows, p columns and entries
in K, and GLn(K) the set of non-singular matrices in the algebra Mn(K) of square
∗Professor of Mathematics at Lyc´ee Priv´e Sainte-Genevi`eve, 2, rue de l'´Ecole des Postes,
78029 Versailles Cedex, FRANCE.
†e-mail address: [email protected]
1
1
1
0
2
r
p
A
8
2
]
.
A
R
h
t
a
m
[
4
v
7
6
4
2
.
4
0
0
1
:
v
i
X
r
a
matrices of order n. The entries of a matrix M ∈ Mn,p(K) are always denoted
by small letters i.e. M = (mi,j). The rank of M ∈ Mn,p(K) is denoted by rk M .
We denote by sln(K) the linear hyperplane of Mn(K) consisting of matrices
with trace zero. We make the group GLn(K) × GLp(K) act on the set of linear
subspaces of Mn,p(K) by
(P, Q).V := P V Q−1.
Two linear subspaces of the same orbit will be called equivalent (this means
that they represent, in a change of basis, the same set of linear transformations
from a p-dimensional vector space to an n-dimensional vector space).
For P and Q in GLn(K), we define
uP,Q :(Mn(K) −→ Mn(K)
7−→ P M Q
M
and vP,Q :(Mn(K) −→ Mn(K)
7−→ P M T Q.
M
Any map of the previous kind will be called a Frobenius automorphism.
It will be noteworthy to remark that the set of Frobenius automorphisms is a
subgroup of the general linear group of the vector space Mn(K).
One of the earliest linear preserver problems was Dieudonn´e's determina-
tion of the linear bijections f of Mn(K) which satisfy f (GLn(K)) ⊂ GLn(K):
using the structure of linear subspaces of singular matrices of Mn(K) with max-
imal dimension, he showed that the solutions were precisely the Frobenius au-
tomorphisms (see the recent [12] for a full classification of non-invertible linear
preservers). More recently, the determination of the linear preservers of non-
singularity was successfully carried out in many other contexts (e.g. Banach
spaces [8], spaces of triangular matrices [5], spaces of symmetric matrices [2]).
Here, we wish to extend Dieudonn´e's theorem to linear subspaces of Mn(K)
with a small codimension. This question arose when we observed that a lin-
ear subspace of Mn(K) is automatically spanned by its non-singular elements
provided its codimension is small enough (see Corollary 6 in [11]).
More precisely, we will prove the following results:
Theorem 1. Let V be a linear subspace of Mn(K) such that codim V < n − 1.
Let f : V ֒→ Mn(K) be a linear embedding such that
∀M ∈ V, f (M ) ∈ GLn(K) ⇔ M ∈ GLn(K).
Then f extends to a Frobenius automorphism of Mn(K) unless n = 3, codim V =
1 and K ≃ F2.
2
The above theorem would normally be called a strong linear preserver the-
orem. We will also prove the following two theorems, which are more in tune
with what the reader is used to (i.e. weak linear preservers):
Theorem 2. Let V be a linear subspace of Mn(K) such that codim V < n − 1.
Let f : V → V be a linear bijection such that f(cid:0)V ∩ GLn(K)(cid:1) ⊂ GLn(K). Then
f extends to a Frobenius automorphism of Mn(K) unless n = 3, codim V = 1
and K ≃ F2.
Theorem 3. Assume K is infinite. Let V be a linear subspace of Mn(K) such
that codim V < n − 1, and f : V ֒→ Mn(K) be a linear embedding such that
f(cid:0)V ∩ GLn(K)(cid:1) ⊂ GLn(K). Then f extends to a Frobenius automorphism of
Mn(K).
Whether the last theorem still holds for finite fields remains an exciting open
problem.
Before proving those results, we wish to show that the upper bound n − 1
is tight provided n ≥ 3 (the case n = 2 and codim V = 1 will be dealt with in
Section 6). Consider indeed the subspace
and the linear bijection:
Hn :=(cid:26)(cid:20)M C
a(cid:21) (M, C, a) ∈ Mn−1(K) × Mn−1,1(K) × K(cid:27)
Φ :(cid:20)M C
0(cid:3)T
where e1 :=(cid:2)1 0 · · ·
Since the matrix (cid:20)M C
a(cid:21) is non-singular if and only if M is non-singular and
a(cid:21) 7→(cid:20)M C + m2,2.e1
(and of course m2,2 is M 's entry at the (2, 2) spot).
a 6= 0, it follows that Φ is a strong preserver of non-singularity. However, Φ
does not extend to a Frobenius automorphism of Mn(K) since it is not a rank
preserver: indeed, taking M = E2,2 (the matrix with entry 1 at the spot (2, 2),
(cid:21)
a
and zero entries elsewhere), one has rk(cid:20)M 0
0(cid:21) = 1 whereas rk Φ(cid:20)M 0
0(cid:21) = 2.
0
0
0
0
0
0
1.2 Strategy of proof and structure of the article
Our strategy for the proof of Theorem 1 is essentially similar to that of Dieudonn´e
[6]: given a linear embedding f : V ֒→ Mn(K) which strongly preserves non-
singularity, we study the preimages of subspaces of singular matrices of Mn(K)
3
with maximal dimension. To understand the structure of those preimages, we
will use our recent generalization [11] of a theorem of Atkinson and Lloyd
[1]. From there, we will show (leaving aside a technical problem in the case
codim V = n − 2, which will be tackled in Section 3) that the situation may
be reduced to the one where f preserves the image of any matrix of V . We
will then use the so-called representation lemma of [11] (Theorem 8) to show
that this property forces f to have the form M 7→ M Q for some Q ∈ GLn(K),
which will conclude the proof.
In Section 4, we will derive Theorems 2 and
3 from Theorem 1: this is trivial in the case of a finite field, and will involve
considerations of polynomials in the case K is infinite (we will prove that every
polynomial on V which vanishes on its singular elements must be a multiple of
the determinant restricted to V : this will show that the weak preservation of
non-singularity implies the strong one for a one-to-one linear map).
The remaining two sections will be devoted to the inspection of special cases:
• In Section 5, we will show that there is a linear hyperplane V of M3(F2)
and an embedding which do not satisfy the conclusion of Theorem 2, and
we will also determine which linear hyperplanes of M3(F2) do satisfy this
conclusion for any embedding. Naturally, this is related to the special case
in the generalized Atkinson-Lloyd theorem, see Theorem 2 of [11].
• In Section 6, we will show that the conclusions of Theorems 1 to 3 still
hold in the case n = 2 and V is a linear hyperplane of M2(K). This is
interesting because it shows that the result holds for linear hyperplanes
regardless of n, e.g. for sln(K) (in that case, even if K ≃ F2, see Section
5).
2 Preimage of large singular subspaces
2.1 A review of large subspaces of singular matrices
Definition 1. A linear subspace V of Mn(K) is called singular when all its
matrices are singular. It is said to have rank k when k = max(cid:8)rk M M ∈ V }.
Notation 2. We set E := Kn and let P(E) denote the projective space asso-
ciated to E, i.e. the set of lines in E. We equip E with the non-degenerate
symmetric bilinear form (X, Y ) 7→ X T Y . Given D ∈ P(E), the linear hyper-
4
plane D⊥ = {X ∈ Kn : X T D = 0} is the annihilator of D, and we set
MD :=(cid:8)M ∈ Mn(K) : D ⊂ Ker M(cid:9) and MD :=(cid:8)M ∈ Mn(K) : Im M ⊂ D⊥(cid:9).
D = MD and1 (MD)T = MD, and that MD and
Remark 1. Notice that MT
MD are singular subspaces of Mn(K) with codimension n. Classically (see [6], or
prove it directly), these are maximal singular subspaces of Mn(K) (i.e. maximal
in the set of the singular subspaces of Mn(K), ordered by inclusion).
Notation 3. Let (s, t) ∈ [[0, n]] × [[0, p]]. Set then
R(s, t) :=(cid:26)(cid:20)M N
0(cid:21) M ∈ Ms,t(K), N ∈ Ms,p−t(K), P ∈ Mn−s,t(K)(cid:27) ⊂ Mn,p(K)
P
(notice that we understate n and p in this notation; however, no confusion should
arise when we use it).
With the above notations, we may reformulate a theorem of Atkinson and
Lloyd [1] recently generalized in [11] to an arbitrary field:
Theorem 4. Let V be a singular subspace of Mn(K) such that codim V ≤ 2n−2.
Then one and only one of the following three conditions holds, unless n = 3,
codim V = 1 and #K = 2:
(i) V ⊂ MD for a unique D ∈ P(E);
(ii) V ⊂ MD for a unique D ∈ P(E);
(iii) codim V = 2n − 2 and V is equivalent to R(n − 2, 1) or to R(1, n − 2).
Remark 2. In [11], the incompatibility between (i) and (ii) was not proven, nor
was the uniqueness of D in the case V is equivalent to a subspace of R(n − 1, 0)
or R(0, n − 1). However, the proof is essentially similar to that of [1].
2.2 Reduction to the case of an image-preserving map
In this paragraph, we let V be a linear subspace of Mn(K) with codimension
lesser than n − 1, and f : V ֒→ Mn(K) be a linear embedding such that
f −1(GLn(K)) = V ∩ GLn(K). We discard the case n = 3, codim V = 1 and
#K = 2. We also assume n ≥ 3, since V = M2(K) if n = 2, in which case the
1For V ⊂ Mn(K), we write V T := {M T M ∈ V }.
5
result we claim is already known (see [6]). Our aim is to prove that, by pre and
post-composing f with well-chosen Frobenius automorphisms, we may obtain a
linear map (necessarily one-to-one) which preserves the image for any matrix of
V . Following Dieudonn´e [6], the basic idea is to study the subspaces f −1(MD)
and f −1(MD) for every D ∈ P(E).
Let D ∈ P(E). Then MD has codimension n in Mn(K), hence the rank theo-
rem shows that codimV f −1(MD) = codimf (V ) f (V )∩MD ≤ codimMn(K) MD =
n, hence codimMn(K) f −1(MD) ≤ 2n − 2 since codimMn(K) V ≤ n − 2. How-
ever, since MD is a maximal singular subspace of Mn(K), f is one-to-one and
f −1(GLn(K)) = V ∩GLn(K), it is clear that f −1(MD) is a maximal singular sub-
space of V . A similar argument shows that f −1(MD) has the same properties,
hence the following result:
Claim 1. For every D ∈ P(E), the linear subspaces f −1(MD) and f −1(MD)
are maximal singular subspaces of V with codimension ≤ 2n − 2 in Mn(K).
Using Theorem 4, we deduce:
Claim 2. For any D ∈ P(E), one and only one of the following conditions holds:
(i) There is a unique D′ ∈ P(E) such that f −1(MD) = V ∩ MD′;
(ii) There is a unique D′ ∈ P(E) such that f −1(MD) = V ∩ MD′;
(iii) The subspace f −1(MD) is equivalent to R(n − 2, 1) or R(1, n − 2), and
codim V = n − 2.
A similar result also holds for f −1(MD) instead of f −1(MD).
For the rest of the paragraph, we will admit the following lemma, the proof
of which is tedious and will only be given in Section 3:
Lemma 5. Let V be a linear subspace of codimension n − 2 in Mn(K), and
g : V ֒→ Mn(K) be a linear embedding such that g−1(GLn(K)) = V ∩ GLn(K).
Assume that (n, #K) 6= (3, 2). Let D ∈ P(E). Then g−1(MD) is equivalent
neither to R(n − 2, 1) nor to R(1, n − 2).
This yields:
Claim 3. For every D ∈ P(E), there is a unique D′ ∈ P(E) such that f −1(MD) =
V ∩ MD′ or f −1(MD) = V ∩ MD′
, and only one of those two conditions holds.
6
Here is our next claim:
1) ∈ P(E)2 such that f −1(MD1) =
. Then, for every D ∈ P(E), there is a unique D′ ∈ P(E) such that
Claim 4. Assume there is a pair (D1, D′
V ∩ MD′
f −1(MD) = V ∩ MD′.
1
Proof. Let D2 ∈ P(E) r {D1}. We may then choose non-zero vectors x1 ∈ D1,
x2 ∈ D2 and extend (x1, x2) into a basis (x1, . . . , xn) of E. Set Di := span(xi)
for i ∈ [[3, n]]. For every i ∈ [[2, n]], we may find a (unique) D′
i ∈ P(E) such
that f −1(MDi) = V ∩ MD′
i. Define I as the set of
those i ∈ [[1, n]] such that f −1(MDi) = V ∩ MD′
, and J := [[1, n]] r I. Set also
i, and notice that dim F + dim G ≤ n. Note that
or f −1(MDi) = V ∩ MD′
D′
D′
i
i
MDi(cid:19) = \1≤i≤n
V ∩ f −1(MDi) = V ∩\i∈I
MD′
i
∩\j∈J
MD′
j .
is the set of matrices M ∈ Mn(K) such that F ⊂ Ker M , and
i and G := Pi∈J
MDi = {0}, hence
F := Pi∈I
T1≤i≤n
{0} = f −1(cid:18) \1≤i≤n
However, Ti∈I
Tj∈J
dim"\i∈I
MD′
MD′
MD′
∩\j∈J
i
i
j the set of matrices M ∈ Mn(K) such that Im M ⊂ G⊥, hence
MD′
j# = (n − dim F ) (n − dim G) ≥ dim G (n − dim G).
Assume finally that J 6= ∅. Then 1 ≤ dim G ≤ n−1 hence (dim G) (n−dim G) ≥
n − 1. Since codim V < n − 1, this yields
V ∩\i∈I
MD′
i
∩\j∈I
MD′
j 6= {0},
in contradiction with a previous result. We deduce that J = ∅.
With a similar proof, or by applying the above results to M 7→ f (M T )T , we
also have:
Claim 5. Assume there is a pair (D1, D′
V ∩ MD′
f −1(MD) = V ∩ MD′.
1) ∈ P(E)2 such that f −1(MD1) =
1. Then, for every D ∈ P(E), there is a unique D′ ∈ P(E) such that
7
We now lose no generality making the following additional assumption:
There is a pair (D1, D′
1) ∈ P(E)2 such that f −1(MD1) = V ∩ MD′
1.
Indeed, in the case this does not hold, we still have some pair (D1, D′
such that f −1(MD1) = V ∩MD′
which satisfies the preceding assumption.
1) ∈ P(E)2
1, and we may then replace f with M 7→ f (M )T ,
Now, Claim 4 applied to both f and f −1 : f (V ) ֒→ Mn(K) shows there is
a bijective map ϕ : P(E) → P(E) such that f (V ∩ MD) = f (V ) ∩ Mϕ(D) for
every D ∈ P(E). Let D ∈ P(E). If f (V ∩ MD) = f (V ) ∩ MD′ for some line D′,
then f (V ∩ MD) = f (V ∩ Mϕ−1(D′)) and therefore V ∩ MD = V ∩ Mϕ−1(D′),
contradicting the uniqueness in Theorem 4. Therefore f (V ∩ MD) = V ∩ MD′
for some D′ ∈ P(E). Claim 4 applied to both f and f −1 then shows there is a
bijective map ψ : P(E) → P(E) such that f (V ∩ MD) = f (V ) ∩ Mψ(D) for every
D ∈ P(E).
Claim 6. The map ϕ is a projective automorphism of P(E).
Proof. First notice that ϕ preserves alinement on the projective space P(E).
Indeed, let D1, D2 and D3 be three distinct lines of E and assume that D1 +
D2 + D3 has dimension 2 and ϕ(D1) + ϕ(D2) + ϕ(D3) has dimension 3. Notice
that
MDi has codimension 2n in Mn(K) whereas
Mϕ(Di) has codimension
contradicting the definition of ϕ.
By the fundamental theorem of projective geometry (recall that dim E ≥ 3), we
deduce that there is a semi-linear automorphism u of E such that ϕ(D) = u(D)
for every D ∈ P(E). The same line of reasoning shows there is a semi-linear
automorphism v of E such that ψ(D) = v(D) for every D ∈ P(E).
It only remains to prove that u is linear. Consider an arbitrary non-zero vec-
tor Y0 ∈ E r {0}, notice that {XY T
X ∈ E} is an n-dimensional linear
0
subspace of Mn(K), hence we may find two linearly independent vectors X1
8
3Ti=1
3n. It follows that
whereas
dim(cid:20)V ∩
3Ti=1
Mϕ(Di)(cid:21) ≤ n(n − 3),
3\i=1
dim(cid:20)f (V ) ∩
MDi(cid:21) ≥ n(n − 2) − n + 2 > n(n − 3),
3\i=1
0 and X2Y T
and X2 in E such that X1Y T
linear automorphism of E, we find that there is a non-zero vector Y ′
that, for every X ∈ E such that XY T
some X ′ ∈ E:
perplane {Y0}⊥, notice then that
0 belong to V . Since v is a semi-
0 ∈ E such
0)T for
indeed, we may consider a basis (Y2, . . . , Yn) of the linear hy-
Mv(span(Yi)) is the set of matrices which
0 ∈ V , one has f (XY T
0 ) = X ′(Y ′
nTi=2
2(Y ′
2)(Y ′
1 + βX ′
1 + βX ′
1 and X ′
0 ) = X ′
0 ) = X ′
0)T and f (X2Y T
0 ) = (αX ′
1 + βX ′
vanish on the hyperplane span(v(Yi))2≤i≤n and then choose a non-zero vector
Y ′
0 in its orthogonal subspace. We recover two non-zero vectors X ′
2 such
0)T . Let now (α, β) ∈ K2.
1(Y ′
that f (X1Y T
Then f ((αX1 + βX2)Y T
0)T since f is linear. We deduce
that αX ′
2 is orthogonal to u(X) for every X orthogonal to αX1 + βX2.
We then choose two linearly independent vectors Z1 and Z2 in E such that
i Zj = δi,j for every (i, j) ∈ {1, 2}2, and let λ : K → K denote the field
X T
automorphism associated to the semi-linear map u. Then αX ′
2 is orthog-
onal to u(βZ1 − αZ2) = λ(β) u(Z1) − λ(α) u(Z2). In particular, X ′
1⊥u(Z2) and
X ′
2)T u(Z2),
and the special case α = 1 then yields:
2⊥u(Z1). Taking β = 1 then shows that α (X ′
1)T u(Z1) = λ(α) (X ′
∀α ∈ K, (α − λ(α))(X ′
1)T u(Z1) = 0.
Notice finally that X1Y T
0
shows that (X ′
1(Y ′
1)T u(Z1) 6= 0. We deduce that λ = idK.
6∈ Mspan(Z1) hence X ′
0)T 6∈ Mspan(u(Z1)) which
Denote then by P the non-singular matrix of Mn(K) such that ϕ(X) = P X
for every X ∈ E. Then the map f ′ : M 7→ P T f (M ) satisfies all the assumptions
of Theorem 1 with the additional property:
For every D ∈ P(E), one has f ′(V ∩ MD) = f ′(V ) ∩ MD.
We may now conclude this section by summing up the above results, still
assuming Lemma 5 holds:
Proposition 6. Let V be a linear subspace of Mn(K) such that codim V < n − 1
and n ≥ 3. Let f : V ֒→ Mn(K) be a linear embedding such that
∀M ∈ V,
f (M ) ∈ GLn(K) ⇔ M ∈ GLn(K).
Unless (n, codim V, #K) = (3, 1, 2), there are two Frobenius automorphisms2 u
and v, together with a linear subspace V ′ of Mn(K) with dim V ′ = dim V , and a
linear embedding f ′ : V ′ ֒→ Mn(K) such that:
2One may even take (u, V ′) = (idV , V ) or (u, V ′) = (X 7→ X T , V T ), whilst v : M 7→ QM
for some Q ∈ GLn(K).
9
(i) u(V ) = V ′;
(ii) f = v ◦ f ′ ◦ uV ;
(iii) for every M ∈ V ′, one has Im f ′(M ) = Im M .
2.3
Image-preserving linear embeddings
We will now prove the following result, which completes the proof of Theorem
1 modulo the proof of Lemma 5.
Proposition 7. Let V be a linear subspace of Mn(K) such that codim V < n−1.
Let f : V ֒→ Mn(K) be a linear embedding such that Im f (M ) = Im M for every
M ∈ V . Then f coincides with uIn,Q on V for some Q ∈ GLn(K).
This is direct consequence of the following lemma, which was recently proven
in [11] (Theorem 8):
Lemma 8 (Representation lemma). Let (n, p, r) ∈ N3. Let V be a linear sub-
space of Mn,r(K) such that dim V ≥ nr − n + 2. Let ϕ : V → Mn,p(K) be a linear
map such that Im ϕ(M ) ⊂ Im M for every M ∈ V .
Then there exists C ∈ Mr,p(K) such that ∀M ∈ V, ϕ(M ) = M C.
Proof of Proposition 7. Applying Lemma 8 to V and f (with p = r = n), we
find a matrix Q ∈ Mn(K) such that ∀M ∈ V, f (M ) = M Q.
Since f is one-to-one, we deduce that V contains no non-zero matrix which van-
ishes on Im Q. If Im Q ( Kn, this would yield codimMn(K) V ≥ n, contradicting
our assumptions. Therefore Q is non-singular and f = uIn,Q.
3 A (very) technical lemma
This entire section is devoted to the proof of Lemma 5, which is the last obsta-
cle for proving Theorem 1. In the whole proof, we denote by (e1, . . . , en) the
canonical basis of E = Kn.
3.1 Starting the proof
We use a reductio ad absurdum. Let V and g be as in Lemma 5, and assume
that there is a line D such that g−1(MD) is equivalent to R(1, n − 2) (notice
that the case where g−1(MD) is equivalent to R(n − 2, 1) may be reduced to this
10
one by pre-composing g with M 7→ M T ). By composing g with a well-chosen
Frobenius automorphism, we may also assume that D = span(en). We finally
lose no generality assuming that
g−1(MD) = R(1, n − 2).
(1)
a b
To make things clearer, those assumptions mean that: V contains every matrix
of the form M =(cid:20) ?
N 0 0(cid:21) with (a, b) ∈ K2 and N ∈ Mn−1,n−2(K), for such
a matrix M we always have g(M ) = (cid:2)N ′ 0(cid:3) for some N ′ ∈ Mn,n−1(K), and
R(1, n − 2) is precisely the set of matrices in V whose images by g have 0 as last
column.
Notation 4. In the rest of the proof, we set V ′ := R(0, n − 2), i.e. V ′ is the set
of all matrices of Mn(K) with zero as (n − 1)-th and n-th column.
We will first investigate the structure of g(V ∩ MD1) for an arbitrary line
D1 ⊂ span(en−1, en).
3.2 Sorting out the structure of g(V ∩ MD1) (I)
Notation 5. For an arbitrary line D1 ⊂ span(en−1, en), we set
HD1 := g(V ∩ MD1).
Claim 7. Let D1 and D2 be distinct lines in P(span(en−1, en)). Then HD1 6=
HD2.
Proof. Notice indeed that V ′ = V ∩ MD1 ∩ MD2 has a codimension greater
than or equal to 2n in Mn(K). On the other hand, we know from the inclusion
R(1, n − 2) ⊂ V that V ∩ MD1 has a codimension lesser than 2n in Mn(K),
which shows V ∩ MD1 ∩ MD2 6= V ∩ MD1 and proves our claim since g is
one-to-one.
Claim 8. Let D1 be a line of span(en−1, en). Then there is no line D′
HD1 ⊂ MD′
or HD1 ⊂ MD′
1.
1
1 such that
Proof. Assumption (1), together with the conclusions of the present claim, are
unchanged should we choose P ∈ GLn(K) which leaves span(e1, . . . , en−2) and
span(en−1, en) invariant and replace V and g respectively with V P −1 and g ◦
uIn,P . Therefore we lose no generality assuming that D1 = span(en). In this
11
1
or HD1 ⊂ MD′
1.
case, we use a reductio ad absurdum and assume there is a line D′
HD1 ⊂ MD′
1 = D, then V ∩ MD1 ⊂ g−1(MD) =
Assume first that HD1 ⊂ MD′
R(1, n−2), and we deduce that every matrix of V ∩MD1 has zero as last column
and, starting from the second one, all its entries on the (n−1)-th last column are
zero, therefore codimMn(K)(V ∩MD1) ≥ 2n−1, which contradicts the assumption
codimMn(K) V ≤ n − 2. Therefore D′
) =
1 6= D, which yields dim(MD ∩ MD′
1 such that
If D′
.
1
1
V ∩ R(1, n − 2) ∩ MD1 = R(1, n − 2) ∩ MD1 has dimension n(n − 2) + 1, which
contradicts the fact that g is one-to-one.
We deduce that HD1 ⊂ MD′
n(n − 2), and it follows that dim(cid:0)g(V ) ∩ MD ∩ MD′
span(en). In particular, any matrix of V ′ is mapped by g to (cid:20)N 0
1(cid:1) ≤ n(n − 2). However
0(cid:21) for some
1, and we lose no generality assuming that D′
N ∈ Mn−1(K).
Pick now a second line D2 ⊂ span(en−1, en) (different from D1) and consider the
subspace g(V ∩ MD2): by the previous line of reasoning, it may not be included
2 ⊂ E. Then D′
in any MD′
2
must be different from D′
1 by Claim 7. Then, since R(1, n − 2) ⊂ V , one has
. Assume it is included in MD′
2 for some line D′
1 =
0
2
g(V ′) = g(MD1 ∩ MD2 ∩ R(1, n − 2)) ⊂ MD′
1 ∩ MD′
2 ∩ MD
which shows that codim g(V ′) ≥ n + 2(n − 1) > 2n, contradicting codim V ′ = 2n.
Now, we may apply Claim 2 to the map g−1, and deduce that HD2 is equivalent
either to R(1, n − 2) or R(n − 2, 1).
Assume first that HD2 is equivalent to R(1, n−2). Then there is a 2-dimensional
subspace P of E such that, for every x ∈ P , one has dim HD2x ≤ 1 (where
HD2x = (cid:8)M x M ∈ HD2(cid:9)). We may then choose a non-zero vector x in
P ∩ span(e1, . . . , en−1), which shows that
codim(HD2 ∩ MD′
1 ∩ MD) ≥ (n − 2) + (n − 1) + n = 3n − 3.
However, since R(1, n − 2) ⊂ V ,
HD2 ∩ MD′
1 ∩ MD = g(MD2 ∩ MD1 ∩ R(1, n − 2)) = g(V ′)
hence HD2 ∩ MD′
1 ∩ MD has codimension 2n in Mn(K). Notice that a similar
line of reasoning holds in the case HD2 is equivalent to R(n − 2, 1), so this yields
a contradiction if n > 3.
Assume finally that n = 3. In this case, we lose no generality assuming that e2
12
belongs to P . Then HD2e2 = span(y) for some y ∈ K3 r {0}, and HD2 contains
the 3-dimensional space Z of all matrices M ∈ M3(K) such that Im(M ) ⊂
span(y). However g(V ′) ⊂ HD1 ⊂ Mspan(e3). If span(y) = span(e3), then we
would have V ′ ∩ g−1(Z) = {0}, which is not possible since V ′ and g−1(Z) are
both 3-dimensional subspaces of the 5-dimensional space V ∩ MD2.
Therefore we lose no generality assuming that HD2e2 = span(e1), in which case
we find that g maps any matrix of the form
. Set
? ? 0
? 0 0
0 0 0
? 0 0
? 0 0
? 0 0
to a matrix of the form
G :=(
a b 0
c 0 0
0 0 0
(a, b, c) ∈ K3).
Let D3 ∈ P(span(e2, e3)) r {D1}. Then HD3 contains g(V ′) = G. However, the
previous considerations apply with D2 replaced by D3, hence HD3 is equivalent
to R(1, 1). The fact that G ⊂ HD3 then yields HD3 = R(1, 1):
indeed, G has
two obvious 2-dimensional linear subspaces of rank 1, their sum has rank 2, so
each one must be contained in one and only one of the two 3-dimensional rank
1 linear subspaces of HD3, which forces those subspaces to be
a b
c
0 0 0
0 0 0
(
(a, b, c) ∈ K3(cid:27) and (
a 0 0
b 0 0
c 0 0
(a, b, c) ∈ K3).
Finally, if we choose D3 different from D2, then we recover HD3 = R(1, 1) =
HD2, contradicting Claim 7.
3.3 Sorting out the structure of g(V ∩ MD1) (II)
Applying Claim 2 to g−1, we deduce from Claim 8:
Claim 9. For any line D1 ⊂ span(en−1, en), the subspace HD1 is equivalent to
R(1, n − 2) or to R(n − 2, 1).
We now prove:
Claim 10. Let D1 ⊂ span(en−1, en) be a line. Then HD1 is equivalent to
R(1, n − 2).
13
Proof. This follows directly from the preceding claim when n = 3. Assume
now that n ≥ 4. As in the beginning of the proof of Claim 8, we lose no
generality assuming that D1 = span(en). We use another reductio ad absurdum
by assuming that HD1 is equivalent to R(n − 2, 1). Then there is a unique
linear subspace F with codimension 2 in E such that ∀x ∈ E r {0}, HD1x =
F or HD1x = E. We lose no generality assuming F = span(e1, . . . , en−2) (we
may reduce the general case to this one by composing g with uP,In for some
well-chosen P ∈ GLn(K)). Then the set of vectors x ∈ E such that HD1x ⊂ F
is a linear hyperplane G of E.
We may then find a linear subspace G′ ⊂ G such that G′ ∩ span(en) = {0} and
dim G′ = n−2. It then easily follows that g(MD1 ∩R(1, n−2)) has a codimension
greater than or equal to n + 2(n − 2) in Mn(K). Since MD1 ∩ R(1, n − 2) has
codimension 2n − 1 in Mn(K) and n > 3, this yields a contradiction.
Using the definition of R(1, n − 2), we deduce that there is a unique 2-
dimensional linear subspace PD1 ⊂ E such that dim(HD1x) = 1 for every x ∈
PD1 r {0}, whilst HD1x = E for every x ∈ E r PD1; moreover the line
D′
1 := HD1x
is independent from x ∈ PD1 r {0} (indeed, given a pair (P, Q) ∈ GLn(K)2 such
that HD1 = P R(1, n − 2) Q, one simply has PD1 = Q−1 span(en−1, en−2) and
D′
1 = P span(e1)).
Claim 11. The plane PD1 is independent from the choice of D1, and it contains
en.
Proof. Consider the linear subspace F := span(en) +
PD1. As-
D1∈P(span(en−1,en))
P
sume dim F ≥ 3 and extend en into a linearly independent triple (en, x, y) in
F . Setting Y := g(V ′), we find that Y en = 0, dim(Y x) ≤ 1 and dim(Y y) ≤ 1,
hence codimMn(K) Y ≥ n + 2(n − 1) > 2n, which contradicts the fact that
dim V ′ = n(n − 2). We deduce that dim F ≤ 2, which proves that all the planes
PD1 are equal and contain en.
We may now assume:
PD1 = span(en−1, en)
for every D1 ∈ P(span(en−1, en)).
(2)
The situation is indeed unchanged should we replace g with uIn,Q ◦ g for any
Q ∈ GLn(K) such that Qen = en.
14
Claim 12. One has g(V ′) = V ′.
Proof. Choose two arbitrary distinct lines D1 and D2 in P(span(en−1, en)), and
notice that V ′ = V ∩ MD1 ∩ MD2 hence g(V ′) = HD1 ∩ HD2 = V ′ since
D′
1 6= D′
2.
Claim 13. The sum P of all lines D′
dimensional subspace of E.
1, for D1 in P(span(en−1, en)), is a 2-
1 6= D′
1 = span(e1) and D′
Proof. Set D1 := span(en−1) and D2 := span(en). Note again that (1) and (2)
are unchanged should g be replaced by uP,In ◦ g for an arbitrary P ∈ GLn(K),
so we lose no generality whatsoever assuming that D′
2 =
span(e2) (recall that D′
2 since HD1 6= HD2). For (i, j) ∈ [[1, n]], denote
by Ei,j the elementary matrix of Mn(K) with entry 1 at the spot (i, j) and zero
elsewhere. Then E1,n ∈ R(1, n − 2) ∩ MD1, hence g(E1,n) ∈ HD1 ⊂ R(2, n − 2).
Similarly g(E1,n−1) ∈ HD2 ⊂ R(2, n − 2). Since g(V ′) = V ′, we deduce that
g maps R(1, n − 2) into R(2, n − 2). Let finally D3 be an arbitrary line in
P(span(en−1, en)). Some non-trivial linear combination A of E1,n−1 and E1,n
must then belong to MD3. Note that A ∈ R(1, n − 2) r V ′, which shows
that g(A) ∈ R(2, n − 2) r V ′ since g is one-to-one and g(V ′) = V ′. On the
other hand g(A) ∈ HD3 hence g(A)x ∈ D′
3 for any x ∈ span(en−1, en). Since
g(A) 6∈ V ′, we may then choose x such that g(A)x 6= 0, which shows that
D′
3 ⊂ span(e1, e2) = D′
2 and proves our claim.
1+D′
2. This shows P = D′
1+D′
Notice in particular that g(V ) contains every rank 1 matrix with image D′
1,
for D1 in P(span(en−1, en)), hence it contains any matrix M ∈ Mn(K) such that
Im(M ) ⊂ P.
As in the beginning of the proof of Claim 13, we lose no generality assuming
that P = span(e1, e2) and D′
1 = span(e1) for D1 := span(en).
Note then that, for D := D1 = span(en), one has g(V ∩ MD) = R(1, n − 2)
hence (g−1)−1(MD) = R(1, n − 2) and g−1(MD) = R(1, n − 2), and moreover
g−1(V ′) = V ′. Since g(V ) contains both R(1, n − 2) and the space of all matrices
M with Im M ⊂ span(e1, e2), one has R(2, n − 2) ⊂ g(V ).
Now we replace (g, V ) with (g−1, g(V )). Notice that (1) and (2) are
preserved, but we now have the additional fact:
V contains the linear subspace R(2, n − 2).
(3)
15
Note that the reductions of the present section preserve (3) hence we lose no
generality assuming that3:
g(V ∩ Mspan(en)) = R(1, n − 2)
(4)
and
HD1 ⊂ R(2, n − 2)
Noting that R(2, n−2) ⊂
2)(cid:1) ⊂ R(2, n − 2), therefore:
for every line D1 ⊂ span(en−1, en).
(5)
P
MD1, this yields the inclusion g(cid:0)R(2, n−
D1∈P(span(en−1,en))
g(cid:0)R(2, n − 2)(cid:1) = R(2, n − 2).
(6)
3.4 Sorting out the action of g on R(2, n − 2)
?
?
Since g stabilizes both V ′ and R(2, n − 2), we deduce that there is a linear
automorphism ϕ of M2(K) such that, for every M =(cid:20)? A
one has g(M ) = (cid:20)? ϕ(A)
0(cid:21) with A ∈ M2(K),
0 (cid:21). Since g preserves non-singularity, it follows that
ϕ must also preserve non-singularity, hence Dieudonn´e's theorem shows that
it is a Frobenius automorphism. However, we know that g maps R(1, n − 2)
into the set of matrices with zero as last column, hence ϕ may not be some
uP,Q. Hence ϕ = vP,Q for some pair (P, Q) ∈ GL2(K)2. The initial assumption
g−1(Mspan(en)) = R(1, n−2) then yields Qe2 ∈ span(e2), whilst the intermediate
one g(V ∩ Mspan(en)) = R(1, n − 2) yields P e1 ∈ span(e1). We thus lose no
generality assuming:
ϕ(A) = AT
for every A ∈ M2(K)
(7)
(indeed, in the general case, replace g with g′ := uP ′,Q′◦g with P ′ :=(cid:20)P −1
and Q′ := (cid:20)In−2
Let L ∈ M2,n−2(K) and set ML :=(cid:20)L 0
In−2(cid:21)
Q−1(cid:21), and check that g′ satisfies assumptions (1), (2), (4),
0(cid:21). Let N ∈ GLn−2(K), and denote by
(5), (6) and (7)).
0
0
0
0
0
3(4) and (5) are respectively (1) and the result of Claim 10 for the "old" pair (g, V ), i.e. for
(g−1, g(V )) with our new notations.
16
M the matrix of R(2, n−2) such that g(M ) =(cid:20) 0
for some L′ ∈ Mn−2(K). However M + ML is non-singular, hence(cid:20)
N 0(cid:21). Then g(ML) =(cid:20) ?
is non-singular, which shows that N + L′ is non-singular. It follows that N + L′
is non-singular for every N ∈ GLn−2(K), and the next lemma shows that L′ = 0.
L′ 0(cid:21)
0(cid:21)
N + L′
I2
I2
0
?
Lemma 9. Let A ∈ Mp(K) be such that ∀P ∈ GLp(K), A + P ∈ GLp(K). Then
A = 0.
Proof. Using the equivalence of matrices, we lose no generality assuming that
If q > 0, then taking P := −In yields a
A = (cid:20)Iq 0
0(cid:21) for some q ∈ [[0, p]].
contradiction. Hence q = 0 and A = 0.
0
0
We now deduce that g stabilizes the subspace of all matrices of the form
0(cid:21) with L ∈ M2,n−2(K). Since g also stabilizes V ′, it follows that there is
B 0(cid:21) with
(cid:20)L 0
an automorphism ψ of Mn−2(K) such that any matrix of the form(cid:20) ?
B ∈ Mn−2(K) is mapped by g to(cid:20) ?
ψ(B) 0(cid:21).
0
0
3.5 The final contradiction
The final contradiction will now come by considering the structure of the sub-
space H := g(V ∩ Mspan(e1)).
Claim 14. There is no line D1 ⊂ E such that H ⊂ MD1.
0
0
L1 L2
B 0
form
Proof. Set H ′ := R(2, n − 2) ∩ Mspan(e1), i.e. H ′ is the set of matrices of the
with L1 ∈ M1,n−2(K), L2 ∈ M1,2(K) and B ∈ Mn−2(K). On
the one hand, applying g to those matrices with L1 = 0 and B = 0 shows that
the subspace g(H ′)E contains span(e1, e2) (by (7)). On the other hand, ψ is
an automorphism hence applying g to those matrices with L1 = 0 and L2 = 0
shows that the projection of g(H ′)E on span(e3, . . . , en) alongside span(e1, e2)
is onto. This shows that g(H ′)E = E, hence g(H ′) may not be included in any
MD1, which proves our claim since H ′ ⊂ V ∩ Mspan(e1).
17
Claim 15. There is no line D1 ⊂ E such that H ⊂ MD1.
Proof. Assume that there is a line D1 ∈ P(E) such that H ⊂ MD1.
• Assume first that D1 ⊂ span(en−1, en). Then, applying Theorem 4 to
g−1 : g(V ) ֒→ Mn(K), we would have g(V ∩ Mspan(e1)) = g(V ) ∩ MD1,
hence g−1(g(V ) ∩ MD1) = V ∩ Mspan(e1). However, g−1 satisfies condition
(1) so we may apply Claim 8 to it and obtain a contradiction.
• We deduce that D1 6⊂ span(en−1, en). Since g stabilizes V ′, it follows that
every matrix of g(V ′ ∩ Mspan(e1)) vanishes on the 3-dimensional subspace
D1 + span(en−1, en), hence codim g(V ′ ∩ Mspan(e1)) ≥ 3n, contradicting
the fact that V ′ ∩ Mspan(e1) has codimension 3n − 2.
?
Applying the previous claims together with Claim 2, we deduce that H =
g(V ∩ Mspan(e1)) is equivalent to R(1, n − 2) or to R(n − 2, 1). In any case, H
is spanned by its rank 1 matrices, which will yield a final contradiction, as we
shall see.
Let M ∈ H, and write M = (cid:20)?
? α(M )(cid:21) with α(M ) ∈ Mn−2,2(K). Since
g(V ) contains R(2, n − 2), we deduce that g(V ) contains the matrix(cid:20)0
0 α(M )(cid:21).
Assume α(M ) has rank 1 and let (cid:20)a
b(cid:21) be a non-zero vector in its kernel. Then
g(V ), which contains R(2, n − 2), also contains the matrix(cid:20)0 A
0(cid:21) for every sin-
gular matrix A ∈ M2(K) with kernel K(cid:20)a
b(cid:21). Setting D := K(cid:2)0 . . .
0 a b(cid:3)T
we deduce that the intersection of g(V ) with the set of all matrices M such that
span(e1, . . . , en−2) ⊕ D ⊂ Ker M has a dimension greater than 2. This however
yields a contradiction because it would show that codim(g(V ) ∩ MD) < 2n − 2,
whereas Claim 9 applies to g−1 and shows that codim g(V ) ∩ MD = 2n − 2.
0
0
,
We deduce that if rk M = 1, then α(M ) = 0. Since α is linear and H
is spanned by its rank 1 matrices, we deduce that α = 0. This shows that
H ⊂ R(2, n − 2). However g(R(2, n − 2)) = R(2, n − 2) and g is one-to-one
hence V ∩ Mspan(e1) ⊂ R(2, n − 2). It follows that V contains no matrix of the
0 C(cid:21) for some C ∈ Mn−2,2(K) r {0} hence codim V ≥ 2(n − 2) > n − 2, a
form(cid:20)0
0
18
final contradiction. Thus Lemma 5 is proven at last, which completes the proof
of Theorem 1.
4 The weak preservers of non-singular matrices
In this section, we turn to the proof of Theorems 2 and 3. First, notice that
Theorem 2 trivially derives from Theorem 1 when K is finite:
indeed, in this
case, if f : V → V is one-to-one and stabilizes V ∩ GLn(K), then we have
f −1(GLn(K)) = V ∩ GLn(K) since V ∩ GLn(K) is finite. In the case K is infinite,
Theorem 2 will be deduced from Theorem 3.
We now try to derive Theorem 3 from Theorem 1. It obviously suffices to
prove the following proposition:
Proposition 10. Assume K is infinite. Let V be a linear subspace of Mn(K)
such that codim V < n − 1, and f : V ֒→ Mn(K) be a linear embedding such that
f (V ∩ GLn(K)) ⊂ GLn(K). Then f −1(GLn(K)) = V ∩ GLn(K).
In order to show this, we will generalize a method of [3] by considering
polynomial functions over the K-vector space V . Since K is infinite, these can
be treated as algebraic polynomials. Notice in particular that if V is a linear
subspace of Mn(K), then detV , the restriction of the determinant to V , is a
homogeneous polynomial of degree n.
In order to establish Proposition 10, we successively prove the following two
results:
Proposition 11. Assume K is infinite. Let V be a linear subspace of Mn(K)
such that codim V ≤ max(n − 2, 0). Then detV is irreducible.
Proposition 12. Assume K is infinite. Let V be a linear subspace of Mn(K)
such that codim V ≤ max(n − 2, 0), and p : V → K be a polynomial function
such that p(M ) = 0 whenever M ∈ V is singular. Then p is a multiple of detV .
Before proving those results, let us see right away how they may help us
prove Proposition 10:
Proof of Proposition 10. Consider the polynomial function p := detV ◦f −1 on
f (V ). Then p is homogeneous with degree n. The assumptions on f show
that f −1(M ) is singular whenever M ∈ f (V ) is singular, hence Proposition 12
applied to f (V ) shows that p is a multiple of detf (V ). However, since detf (V )
19
also has degree n, we deduce that p = λ detf (V ) for some λ ∈ K. This yields
det(M ) = λ det(f (M )) for every M ∈ V . Since detV is irreducible, it is non-zero
hence λ 6= 0. This shows that f −1(GLn(K)) = V ∩ GLn(K).
In order to prove Propositions 11 and 12, we first reduce the situation to a
more elementary one. For M ∈ Mn(K), write M =(cid:20)?
? K(M )(cid:21) with K(M ) ∈
?
Mn−1(K).
Given a linear subspace V of Mn(K), we denote by V ′ the linear subspace of
matrices of V with a zero first column. For (i, j) ∈ [[1, n]]2, denote by Ei,j the
elementary matrix with entry 1 at the spot (i, j) and 0 elsewhere. Assume that
span(E1,2, . . . , E1,n) 6⊂ V . Then the rank theorem shows that
codimMn−1(K) K(V ′) ≤ codimMn(K) V − 1
therefore
codimMn−1(K) K(V ) ≤ codimMn(K) V − 1.
We may now state the basic lemma that we will use:
Lemma 13. Let V be a linear subspace of Mn(K) such that codim V ≤ max(n −
2, 0). Then V is equivalent to a linear subspace W which contains E1,1 and for
which codim K(W ′) ≤ max(n − 3, 0).
Proof. The result is trivial when codim V = 0. Assume codim V > 0. Then
there must be an index i ∈ [[1, n]] such that span(Ei,1, . . . , Ei,n) 6⊂ V . Using row
operations, we lose no generality assuming i = 1. However, since codim V < n,
we have span(E1,1, . . . , E1,n) ∩ V 6= {0}. Using a series of column operations, we
may then assume furthermore that E1,1 ∈ V , whereas span(E1,2, . . . , E1,n) 6⊂ V .
With the above inequalities, this leads to codim K(V ′) ≤ n − 3.
Proof of Proposition 11. We use an induction on n. The result is trivial when
n = 1. Set an arbitrary integer n > 0 and assume that the result holds for n − 1.
Let V ⊂ Mn(K) be a linear subspace such that codim V ≤ max(n − 2, 0). Notice
that the problem is essentially unchanged should V be replaced with u(V ) for
some Frobenius automorphism. By Lemma 13, we lose no generality assuming
that V contains E1,1 and codim K(V ′) ≤ max(n − 2, 0). Assume detV = p q for
some non-constant polynomial functions p and q. For any M ∈ V ′, we then have
p(E1,1 + M ) q(E1,1 + M ) = det K(M ).
20
However, the induction hypothesis shows that the homogeneous polynomial
function detK(V ′) is irreducible, hence the homogeneous polynomial function
M 7→ det K(M ) on V ′ also is. We then lose no generality assuming that
M 7→ p(E1,1 + M ) is a non-zero scalar multiple of M 7→ det K(M ), hence has
total degree n − 1. Since detV is homogeneous, p and q are also homogeneous
hence q must have degree 1, i.e. q is a linear form. It follows that every matrix
of Ker q is singular. However, codimMn(K) Ker q ≤ codimMn(K) V + 1 ≤ n − 1
hence the Dieudonn´e theorem [6] shows that Ker q must contain a non-singular
matrix. This is a contradiction, which shows that detV is irreducible.
Proof of Proposition 12. As in the proof of Proposition 11, we lose no generality
assuming that the linear subspace V contains E1,1 and that codim K(V ) ≤
max(n − 2, 0). Define now V ′′ :=(cid:8)M ∈ V : m1,1 = 0(cid:9) so that V = V ′′ ⊕ KE1,1
and K(V ) = K(V ′′). Development of the determinant along the first column
yields a polynomial function q : V ′′ → K such that
∀(x, M ) ∈ K × V ′′, det(xE1,1 + M ) = x det K(M ) + q(M ).
Using the Euclidian algorithm with respect to the indeterminate x, we may then
find two polynomial functions r : K × V ′′ → K and s : V ′′ → K, together with a
positive integer N such that
∀(x, M ) ∈ K×V ′′, (det K(M ))N p(xE1,1 +M ) = det(xE1,1 +M ) r(x, M )+s(M )
and we may even assume that s is a multiple of the polynomial function M 7→
det K(M ) (on V ′′). Let M ∈ V ′′ such that det K(M ) 6= 0. Then we may find
some x ∈ K such that det(xE1,1 + M ) = 0, hence p(xE1,1 + M ) = 0 and we
deduce that s(M ) = 0.
This shows that s = 0, hence detV divides the polynomial function M 7→
(det K(M ))N p(M ) on V . However, we know from Proposition 11 that both
detV and M 7→ det K(M ) are irreducible homogeneous polynomial functions
on V , with respective degrees n and n − 1. Therefore detV may not divide the
latter, which shows that it divides p.
5 The exceptional case of linear hyperplanes of M3(F2)
5.1 Reduction to the case of an internal linear preserver
In this section, we wish to examine more closely the situation of linear hyper-
planes of M3(F2). The major obstruction for proving Theorems 1 and 2 in this
21
case is the counterexample in the Atkinson-Lloyd theorem. Recall from Theorem
2 of [11] that every 5-dimensional singular linear subspace V of M3(F2) satisfies
one of the mutually exclusive conditions:
(i) V ⊂ MD for a (unique) line D ⊂ F3
2;
(ii) V ⊂ MD for a (unique) line D ⊂ F3
2;
(iii) V is equivalent to R(1, 1);
(iv) V is equivalent to the subspace
J3(F2) :=(
(a, b, c, d, e) ∈ F5
2)
0
0
a 0
c
b
d e a + b
i.e. to the subspace of lower triangular matrices with trace zero.
This last case is one major obstacle both in the proof of Lemma 5 and in that
of Claim 3. Notice however that if the result of Claim 3 holds for some linear
embedding f : V ֒→ M3(F2) of a hyperplane V such that f strongly preserves
non-singularity, then the rest of the proof from Section 2 applies and shows that
f extends to a Frobenius automorphism of M3(F2).
We reduce the study to three cases. Using the non-degenerate symmetric
bilinear form (A, B) 7→ tr(AB) on Mn(K), we see that orbits of hyperplanes of
Mn(K) are classified by the orbits of their orthogonal subspace (which is always
a line), i.e. by the rank of the non-zero matrices in their orthogonal subspace.
It follows that there are exactly n orbits of hyperplanes of Mn(K) under equiva-
lence, and in the case at hand, every hyperplane of M3(F2) is equivalent to one
and only one of the three particular hyperplanes:
(a) V1(F2) :=(cid:26)(cid:20)M C
(b) V2(F2) :=(cid:26)(cid:20)M C
L 0(cid:21) (M, C, L) ∈ M2(F2) × M2,1(F2) × M1,2(F2)(cid:27);
L a(cid:21) (M, C, L, a) ∈ sl2(F2) × M2,1(F2) × M1,2(F2) × F2(cid:27);
(c) sl3(F2).
Therefore, we lose no generality assuming that V is one of these three hy-
perplanes, and we will actually study the three cases separately. In order to do
this, it will be convenient to reduce the situation to the case where f (V ) = V .
This is done thanks to the next result:
22
Proposition 14. Let V and V ′ be linear hyperplanes of M3(F2) and f : V → V ′
be a linear isomorphism such that f −1(GL3(F2)) = V ∩ GL3(F2). Then V and
V ′ are equivalent.
The case one of the hyperplanes V and V ′ is equivalent to V1(F2) is easy:
indeed, V1(F2) contains a 6-dimensional singular subspace. However, if a linear
hyperplane V ′′ of M3(F2) contains such a subspace, then the Dieudonn´e theorem
on singular subspaces shows that MD ⊂ V ′′ or MD ⊂ V ′′ for some line D ⊂ F3
2,
and this proves that every matrix of (V ′′)⊥ has a rank lesser than or equal to
1, hence V ′′ is equivalent to V1(F2). We deduce that if V or V ′ is equivalent to
V1(F2), then so is the other one.
The remaining cases rely upon a counting argument: we show that #(cid:0)V2(F2)∩
GL3(F2)(cid:1) 6= #(cid:0)sl3(F2) ∩ GL3(F2)(cid:1), which clearly yields Proposition 14.
Proposition 15. The space sl3(F2) has 80 non-singular elements.
Proof. A matrix M ∈ M3(F2) belongs to sl3(F2) ∩ GL3(F2) if and only if its
characteristic polynomial is t3 + t + 1 or t3 + 1. Recall that # GL3(F2) =
(23 − 1)(23 − 2)(23 − 22) = 7 × 6 × 4.
• Note that t3 +t+1 is irreducible in F2[t] hence the matrices of M3(F2) with
characteristic polynomial t3 + t + 1 form a single orbit under similarity, and
the companion matrix of t3 +t+1 is one of them. Moreover, the centralizer
of this companion matrix in the algebra M3(F2) is F2[t]/(t3 + t + 1) ≃ F8
since t3 + t + 1 is irreducible: therefore this centralizer contains exactly
7 non-singular matrices. It follows that there are 6 × 4 = 24 matrices of
M3(F2) with characteristic polynomial t3 + t + 1.
• We may factorize t3 + 1 = (t + 1)(t2 + t + 1).
If a matrix has t3 + 1
as characteristic polynomial, then it must also have t3 + 1 as minimal
polynomial hence it is similar both to the companion matrix of t3 + 1
1 0 0
0 0 1
0 1 1
must stabilize Ker(A − I3) and Im(A − I3), therefore the centralizer of A
and to the matrix A =
in M3(F2) is the set of matrices of the form (cid:20)a
B ∈ M2(F2) commutes with(cid:20)0 1
. Any matrix that commutes with A
0 B(cid:21) where a ∈ F2 and
1 1(cid:21). For such a matrix to be non-singular,
it is necessary and sufficient that a = 1 and B be non-singular, which leaves
0
23
3 possibilities (notice that the centralizer of the companion matrix(cid:20)0 1
1 1(cid:21)
in M2(F2) is isomorphic to F2[t]/(t2 + t + 1) ≃ F4). We conclude that there
are 7 × 2 × 4 = 56 matrices in M3(F2) with characteristic polynomial t3 + 1.
Proposition 16. The space V2(F2) has 88 non-singular elements.
Proof. Let M ∈ sl2(F2). We count the triples (L, C, x) ∈ M1,2(F2)×M2,1(F2)×F2
such that(cid:20)M C
L x(cid:21) is non-singular. If M = 0, then there is no such triple. As-
sume M is non-singular. Then the former matrix has determinant LfM C − x,
where fM denotes the transpose of the matrix of cofactors of M . Hence there
are 24 well-suited triples (we choose L and C freely, and then x accordingly).
Since sl2(F2) contains exactly 4 non-singular matrices, we find 26 non-singular
matrices in V2(F2) of the former type.
Assume finally that rk M = 1. Then M is nilpotent and we thus lose no gen-
0 0(cid:21). However, given a 5-tuple (a, b, c, d, e) ∈ K5,
= bc, hence there are 23 well-suited triples (L, C, x) for M .
erality assuming that M =(cid:20)0 1
one has (cid:12)(cid:12)(cid:12)(cid:12)(cid:12)(cid:12)
0 1 a
0 0 b
c d e
(cid:12)(cid:12)(cid:12)(cid:12)(cid:12)(cid:12)
Since sl3(F2) contains exactly three rank 1 matrices, we find that V2(F2) contains
exactly 26 + 3 × 23 = 88 non-singular matrices.
5.2 The case of V1(F2)
Here, we prove the following result:
Proposition 17. There exists a linear automorphism f of V1(F2) which (strongly)
preserves non-singularity but does not extend to a Frobenius automorphism of
M3(F2).
Proof. Let α : M1,2(F2) → M2(F2) and β : M2,1(F2) → M2(F2) be arbitrary
linear maps. We will show that we may choose α and β so that the linear
automorphism
f :(cid:20)M C
L 0(cid:21) 7−→(cid:20)M + α(L) + β(C) C
0(cid:21)
L
has the claimed properties.
24
• In order to do this, we first study on what conditions on α and β the
map f may be extended to a Frobenius automorphism. A sufficient con-
dition is easy to find: if α : L 7→(cid:20)a L
b L(cid:21) and β : C 7→(cid:2)c C d C(cid:3) for some
0
(a, b, c, d) ∈ F4
2, then f (M ) is simply obtained from M by performing a
series of row and column operations (that is independent from M ), hence
f clearly extends to a Frobenius automorphism.
Conversely, assume that f = uP,Q or f = vP,Q for some (P, Q) ∈ GL3(F2)2.
Notice for every M ∈ M2(F2) that uP,Q fixes the matrix(cid:20)M 0
0(cid:21) or maps
it to its transpose, i.e. uP,Q fixes or transposes every matrix with im-
age span(e1, e2) and kernel span(e3), where (e1, e2, e3) is the canonical
basis of F3
2. It easily follows that P stabilizes span(e1, e2) and Q stabilizes
span(e3), hence there are matrices C1 ∈ M2,1(F2), L1 ∈ M1,2(F2) and ma-
trices P1 and Q1 in GL2(F2) such that P =(cid:20)P1 C1
1(cid:21) and Q =(cid:20)Q1 0
1(cid:21).
Computing the image by f of the previous matrices shows that ∀M ∈
M2(F2), P1M Q1 = M or ∀M ∈ M2(F2), P1M Q1 = M T .
In any case,
taking M = I2 yields Q1 = P −1
In the first case, P1 commutes with every matrix of M2(F2), which shows
that P1 = I2 = Q1, and we then notice that α and β have the aforemen-
tioned form.
However, the second case leads to a contradiction by taking every M with
zero as second column.
L1
.
1
0
We now prove that α and β may be chosen so that f is not a Frobenius auto-
morphism although it is a determinant preserver. Let (cid:20)M C
L 0(cid:21) ∈ V1(F2). Its
determinant is LfM C (recall that fM denotes the transpose of the matrix of
cofactors of M ). However, M 7→ fM is linear. It follows that f is a determinant
preserver if (and only if)
∀(L, C) ∈ M1,2(F2) × M2,1(F2), L(]α(L) + ]β(C)) C = 0.
Taking
α : (cid:2)l1
0
l2 0(cid:21)
l2(cid:3) 7→(cid:20) 0
and β : (cid:20)c1
c2(cid:21) 7→(cid:20) 0
c1 0(cid:21) ,
0
it is obvious from the above necessary condition that f is not a Frobenius au-
25
tomorphism, However, for every L =(cid:2)l1
M2,1(F2), one has
l2(cid:3) ∈ M1,2(F2) and every C =(cid:20)c1
c2(cid:21) ∈
L(]α(L) + ]β(C)) C = l2(l2 + c1)c1) = l2
2c1 + l2c2
1 = 2l2c1 = 0,
and therefore f is a determinant preserver.
5.3 The case of V2(F2)
Here, we let f : V2(F2) → V2(F2) be a linear transformation which preserves non-
singularity. We wish to prove that Claim 3 holds in this situation. We do this by
analyzing the 5-dimensional singular subspaces of V2(F2). Recall from section
5.1 (or prove this elementary fact directly) that V2(F2) ∩ MD and V2(F2) ∩ MD
have dimension 5 for any line D ⊂ F3
2. In order to simplify the discourse, we
will say that a 5-dimensional singular subspace V of M3(F2) is:
• maximal of the first kind if equivalent to R(1, 1);
• maximal of the second kind if equivalent to J3(F2);
• non-maximal if V ⊂ MD of V ⊂ MD for some line D ⊂ F3
2.
This terminology stems from the problem of maximality in the set of singular
linear subspaces of M3(F2) ordered by the inclusion of subsets. Since F3
2 has 7
non-zero vectors, and therefore 7 one-dimensional subspaces, V2(F2) has exactly
fourteen non-maximal 5-dimensional singular subspaces. Let us now consider
the maximal ones.
Claim 16. The linear subspace
0 0 a
0 0 b
c d e
(a, b, c, d, e) ∈ F5
2)
F :=(
0(cid:21) ∈ M3(F2). Let V be a 5-dimensional maximal singular sub-
set J2 :=(cid:20)I2 0
0
is the sole 5-dimensional maximal singular subspace of the first kind in V2(F2).
Proof. Clearly, F is equivalent to R(1, 1) and is included in V2(F2). Conversely,
space in V2(F2) of the first kind. Then there are two non-zero vectors X1 and X2
26
2 such that V contains X1Y T and Y X T
in F3
2 and V is actually
spanned by those matrices. Writing that those matrices are orthogonal to J2
for the symmetric bilinear form (A, B) 7→ tr(AB), we find that J2X1 = 0 and
2 for every Y ∈ F3
X T
2 J2 = 0, which shows that X1 and X2 are both scalar multiples of
shows that V = F.
. This
0
0
1
Claim 17. There are exactly three 5-dimensional maximal singular subspaces of
the second kind in V2(F2). One of them is
G :=(
0
0
a c
0 b
a + b d e
(a, b, c, d, e) ∈ F5
2)
and the two other ones may be obtained by conjugating G with(cid:20)P 0
1(cid:21) for some
0
P ∈ GL2(F2).
Proof. Obviously, G is equivalent to J3(F2) and is a linear subspace of V2(F2).
Also, the number p of 5-dimensional maximal singular subspaces of the second
kind in an hyperplane V which is equivalent to V2(F2) is independent from the
given V . We now resort to a counting argument. Notice that the orthogonal
subspace of J3(F2) (for (A, B) 7→ tr(AB)) is the subspace
a 0 0
b a 0
d c a
(
(a, b, c, d) ∈ F4
2) :
it contains exactly two rank 2 matrices, hence J3(F2) is contained in exactly
two linear hyperplanes that are equivalent to V2(F2) (being equivalent to V2(F2)
being the same, for a hyperplane of M3(F2), as being orthogonal to a rank 2
matrix). It follows from a standard counting argument that p n1 = 2 n2, where
n1 denotes the number of hyperplanes in M3(F2) which are equivalent to V2(F2),
and n2 the number of 5-dimensional maximal singular subspaces of the second
kind in M3(F2).
• Clearly, n1 is the number of rank 2 matrices of M3(F2), hence n1 = 7×7×6
(there are 7 possibilities for the kernel of such a matrix and 7 × 6 ones for
a linearly independent 2-tuple in F3
2).
27
• Denote by (e1, e2, e3) the canonical basis of F3
2. Let X ∈ F3
2 r {0} . Set
2(cid:9). Notice then
RX := (cid:8)Y X T Y ∈ F3
2(cid:9) and RX := (cid:8)XY T Y ∈ F3
that RX ∩ J3(F2) has dimension 0 if X ∈ F3
2 r span(e1, e2), dimension 1 if
X ∈ span(e1, e2) r span(e1), and dimension 2 if X ∈ span(e1). Similarly,
RX ∩ J3(F2) has dimension 0 if X ∈ F3
2 r span(e2, e3), dimension 1 if
X ∈ span(e2, e3) r span(e3), and dimension 2 if X ∈ span(e3).
This shows that if some pair (P, Q) ∈ GL3(F2)2 satisfies P J3(F2) Q−1 =
J3(F2), then P must stabilize span(e3) and span(e2, e3), whilst QT must
stabilize span(e1) and span(e1, e2), hence both P and Q are lower triangu-
lar. Conversely P J3(F2) Q−1 = J3(F2) for every pair (P, Q) of lower trian-
gular matrices in GL3(F2). Since there are 82 such pairs and # GL3(F2) =
7 × 6 × 4, we deduce that
n2 =
72 × 62 × 42
82
= 72 × 32.
The previous formulae then yield p = 3.
Let finally M be a non-zero nilpotent matrix of M2(F2). Set B :=(cid:20)0 1
P ∈ GL2(F2) be such that P BP −1 = M and set Q = (cid:20)P 0
0 0(cid:21). Let
1(cid:21). Then QGQ−1
is a 5-dimensional maximal singular subspace of V2(F2) of the second kind and
the projection of G onto the first 2 × 2 block is span(M ). Since 3 distinct lines
of sl2(F2) may be obtained in this manner (there are three non-zero nilpotent
matrices in sl2(F2)), we deduce that this yields three 5-dimensional maximal
singular subspaces of V2(F2) of the second kind, hence we have found them
all.
0
In the rest of the proof, we denote by (e1, e2, e3) the canonical basis of F3
2.
Claim 18. Let V be a 5-dimensional singular subspace of V2(F2). Then:
(a) Either V ⊂ MD or V ⊂ MD for some line D ⊂ F3
2 not included in
span(e1, e2); then dim(V ∩ V ′) ≤ 3 for every other 5-dimensional singular
subspace V ′ of V2(F2);
(b) Or there exists a 5-dimensional singular subspace V ′ of V2(F2) such that
dim(V ∩ V ′) = 4.
28
Proof. In this proof, we will simply write V2 instead of V2(F2) to lighten the
burden of notations.
Assume first that V ⊂ MD or V ⊂ MD for some line D ⊂ F3
2 not included in
span(e1, e2). By transposing, we lose no generality assuming that V ⊂ MD. We
also lose no generality assuming that D = span(e3).
• Let D′ ⊂ span(e1, e2, e3) be an arbitrary line distinct from D. A straight-
forward computation shows that dim(V2 ∩MD ∩MD′) = 2 (notice that we
lose no generality assuming D + D′ = span(e2, e3) for this computation).
• Let D′ ⊂ span(e1, e2, e3) be an arbitrary line. Write every matrix M of
MD) = G(V2) is a hyperplane of M3,2(F2) and its orthogonal subspace for
M3(F2) as (cid:2)G(M ) ?(cid:3) with G(M ) ∈ M3,2(F2). On the one hand G(V2 ∩
b : (A, B) 7→ tr(AT B) is span(cid:20)I2
0(cid:21). On the other hand, the orthogonal
subspace of G(MD′ ) contains no rank 2 matrix, hence G(MD′) 6⊂ G(V2),
and it follows that G(V2 ∩ MD ∩ MD′
) is a hyperplane
of G(MD′
) = G(V2) ∩ G(MD′
) hence
dim(V2 ∩ MD ∩ MD′
) = dim G(V2 ∩ MD ∩ MD′
) = 4 − 1 = 3.
• Obviously dim(V ∩ F) = 2 and dim(V ∩ G) = 2. Claims 16 and 17 then
entail that dim(V ∩ V ′) = 2 for every 5-dimensional maximal singular
subspace V ′ of V2.
Assume now that V = V2 ∩ MD for some line D ⊂ span(e1, e2). Then we lose no
generality assuming that D = span(e1). In this case, we have dim(V ∩ F) = 4.
The same obviously holds when V = V2(F2)∩MD for some line D ⊂ span(e1, e2).
Assume finally that V = F or V = G. Then clearly dim(V ∩ MD1) = 4 for
D1 = span(e1). Using Claim 17, this finishes the proof of Claim 18.
In the course of the above proof, we have also obtained the following result:
Claim 19. Let D be a line included in F3
2 but not in span(e1, e2). Set V :=
V2(F2) ∩ MD (resp. V := V2(F2) ∩ MD) and let V ′ be a 5-dimensional singular
subspace of V2(F2). Then dim(V ∩ V ′) = 3 if and only if V ′ = V2(F2) ∩ MD′
for
some line D′ ⊂ F3
2 (resp. V ′ = V2(F2) ∩ MD′ for some line D′ ⊂ F3
2).
29
Recall now that f : V2(F2) → V2(F2) is a linear bijection which (strongly)
preserves non-singularity. Then f permutes the 5-dimensional singular subspaces
of V2(F2). Set
X :=(cid:8)V2(F2)∩Mspan(x) x ∈ F3
2rspan(e1, e2)(cid:9)[(cid:8)V2(F2)∩Mspan(x) x ∈ F3
2rspan(e1, e2)(cid:9).
1, D2, D′
Then Claim 18 clearly entails that f must stabilize X . We then lose no generality
(left-composing f with M 7→ M T if necessary) assuming that there are four lines
D1, D′
2 × {0}, such
and f (V2(F2) ∩ MD2) = V2(F2) ∩ MD′
that f (V2(F2) ∩ MD1) = V2(F2) ∩ MD′
.
Claim 3 then easily follows from Claim 19, and then the rest of Section 2 shows
that f extends to a Frobenius automorphism of M3(F2).
2, with D1 6= D2, and none of them included in F2
2
1
5.4 The case of sl3(F2)
Here, we let f : sl3(F2) → sl3(F2) be a bijective linear transformation which
preserves non-singularity. Note again that f is a determinant preserver. Our
aim is to prove Claim 3 in this situation. This has the following three steps:
Lemma 18. No linear subspace of sl3(F2) is equivalent to R(1, 1).
Proof. Indeed, if there were such a linear subspace, then there would be a 2-
dimensional linear subspace P of F3
2 such that sl3(F2) contains every matrix
which vanishes on P , one of which has a non-zero trace.
Using the line of reasoning from Section 2, it thus suffice to prove the follow-
ing:
Claim 20. For every line D ⊂ F3
to J3(F2).
2, neither f −1(MD) nor f −1(MD) is equivalent
To prove this, we establish two lemmas:
Lemma 19. There are five rank 1 matrices in J3(F2).
For any D ∈ F3
the same holds for MD ∩ sl3(F2).
2, there are more than five rank 1 matrices in MD ∩ sl3(F2), and
Lemma 20. The map f is a rank preserver.
Clearly, combining those lemmas yields Claim 20, hence the rest of the proof
from Section 2 applies with no restriction.
30
Proof of Lemma 19. The first claim is straightforward (notice that a matrix of
J3(F2) has rank 1 only if its diagonal is zero).
For the second one, we lose no generality assuming that D is spanned by the first
vector of the canonical basis and by only considering the case of MD ∩ sl3(F2).
Then MD∩sl3(F2) is the set of all matrices of the form(cid:20)0 L
0 M(cid:21) with L ∈ M2,1(K)
and M ∈ sl2(F2). Taking M = 0 and an arbitrary L 6= 0 yields three rank 1
matrices, then taking L = 0 and an arbitrary nilpotent matrix M yields three
others.
Proof of Lemma 20. We start by using the fact that f is a determinant preserver
on sl3(F2). The Newton formulae show that tr A3 = 3 det A = det A for every
A ∈ sl3(F2). It follows that f preserves the form b(A, B) = det(A+B)−det(A)−
det(B) = tr(A2B) + tr(B2A). Fixing A and computing b(A, B + C) − b(A, B) −
b(A, C) for an arbitrary pair (B, C), we deduce:
∀(A, B, C) ∈ sl3(F2)3, tr(cid:0)[f (A), f (B)]f (C)(cid:1) = tr(cid:0)[A, B]C(cid:1),
where [−, −] denotes the standard Lie bracket on M3(F2). For A ∈ M3(F2),
[A, M ] = 0} its centralizer and set C′(A) :=
denote by C(A) := {M ∈ M3(F2) :
C(A) ∩ sl3(F2). Notice then that identity (8) yields:
(8)
∀A ∈ sl3(F2), f (C′(A)) = C′(f (A)).
Indeed tr(I 2
3 ) = 1, hence the symmetric bilinear form (A, B) 7→ tr(AB) is non-
degenerate on the orthogonal sl3(F2) of span(I3). However, since I3 6∈ sl3(F2),
we may write C(A) = span(I3) ⊕ C′(A) for any A ∈ sl3(F2), which yields:
∀A ∈ sl3(F2), dim C(A) = dim C(f (A))
It now suffices to prove that f
i.e. f preserves the dimension of centralizers.
preserves the set of rank 1 matrices of sl3(F2). In order to do this, we characterize
the rank 1 matrices in sl3(F2) in terms of their centralizer, in the next lemma.
Lemma 21. Let A ∈ sl3(F2). Then the following conditions are equivalent:
(i) rk A = 1;
(ii) dim C(A) = 5 and C′(A) ∩ sl3(F2) contains only singular matrices.
31
Proof. Assume rk A = 1. Then A is nilpotent since tr A = 0, and we lose no
generality assuming that A is the elementary matrix E1,3. A straightforward
computation then yields
C(A) =(
a b
c
0 d e
0 0 a
(a, b, c, d, e) ∈ F5
2) and C′(A) =(
a b
c
0 0 e
0 0 a
(a, b, c, e) ∈ F4
2)
hence A satisfies condition (ii).
Conversely, assume condition (ii) holds and rk A 6= 1. The condition dim C(A) =
5 shows, using the Frobenius formula for the dimension of the centralizer (see
Theorem 19 p.111 of [7]), that A is a linear combination of I3 and a rank 1
matrix B. However rk A 6= 1 and A 6= I3 hence A = I3 + B. We deduce that
tr B = 1 hence we lose no generality assuming that A =
the non-singular matrix
. Then
commutes with A and has trace 0, which
0 1 0
1 1 0
0 0 1
1 0 0
0 1 0
0 0 0
contradicts condition (ii).
This finishes the proof of Lemma 20 and shows that f extends to a Frobenius
automorphism of M3(F2).
5.5 Conclusion
We may now sum up the previous results:
Theorem 22. Let V be a linear hyperplane of M3(F2) which is not equivalent to
V1(F2), and f : V ֒→ M3(F2) be a linear embedding such that ∀M ∈ V, f (M ) ∈
GL3(F2) ⇔ M ∈ GL3(F2). Then f extends to a Frobenius automorphism.
6 The case of linear hyperplanes of M2(K)
In this final section, we show that the result from Theorem 1 still holds in the
case n = 2 and V is a linear hyperplane of M2(K), and we also investigate the
question of weak preservers. Using the same line of reasoning as in section 5.1,
we see that, up to equivalence, the only linear hyperplanes of M2(K) are T +
2 (K)
(the set of upper triangular matrices of M2(K)) and sl2(K). Let f : V ֒→ M2(K)
32
be a linear embedding such that f (V ∩GL2(K)) ⊂ GL2(K). We lose no generality
assuming that both V and f (V ) belong to(cid:8)T +
Let us assume first that V = f (V ).
2 (K), sl2(K)(cid:9).
• If V = sl2(K), then f is a linear automorphism of sl2(K) which (weakly)
preserves nilpotency, hence the Botta-Pierce-Watkins theorem [4] (or clas-
sical results on projective conics) shows that f extends to uλ P,P −1 for some
pair (λ, P ) ∈ K∗ × GL2(K).
• If V = T +
2 (K), then a theorem of Chooi and Lim [5] shows that f extends
to a Frobenius automorphism of M2(K).
Assume now only that f (V ∩ GL2(K)) ⊂ GL2(K).
• If f −1(GL2(K)) = V ∩ GL2(K), then V is equivalent to f (V ) since T +
2 (K)
contains a 2-dimensional singular subspace whereas sl2(K) does not. For
the same reason, f (V ) is equivalent to V if V = sl2(K).
This proves the following results:
Proposition 23. Let V be a linear hyperplane of M2(K), and f : V ֒→ M2(K)
be a linear embedding such that f −1(GL2(K)) = V ∩ GL2(K). Then f extends
to a Frobenius automorphism of M2(K).
Proposition 24. Let V be a linear hyperplane of M2(K) which is equivalent to
sl2(K), and f : V ֒→ M2(K) be a linear embedding such that f (V ∩ GL2(K)) ⊂
GL2(K). Then f extends to a Frobenius automorphism of M2(K).
Let us finally examine whether there exists a linear bijective map f : T +
2 (K) →
sl2(K) which maps non-singular matrices to non-singular matrices. Assume such
a map exists. Then f −1 maps any singular matrix of sl2(K) to a singular matrix.
2 (K)) associated
to f −1. The set of rank 1 matrices of sl2(K) yields a non-degenerate projective
Denote by u the projective isomorphism from P(cid:0)sl2(K)(cid:1) to P(T +
conic C of P(cid:0)sl2(K)(cid:1) and u maps C into the union of two distinct projective lines
D1 and D2 of P(T +
2 (K)). This is impossible if #K ≥ 4 since a non-degenerate
projective conic has at most 2 common points with every line and #C = #K + 1.
However, if #K ≤ 3, then C has at most 4 points, hence we may find two dis-
tinct projective lines D′
2 such that C ⊂ D′
2: we may then choose
a projective transformation v : P(sl2(K)) → P(T +
2 (K)) which maps respectively
D′
2 (K) associated to
2 to D2. Given a linear map g : sl2(K) → T +
1 to D1 and D′
1 and D′
1 ∪ D′
33
u, we then find that g−1 is a weak linear preserver of non-singularity but does
not extend to a Frobenius automorphism of M2(K). We may now generalize
Proposition 24 as follows:
Proposition 25. Let V be a linear hyperplane of M2(K) and f : V ֒→ M2(K)
be a linear embedding such that f (V ∩ GL2(K)) ⊂ GL2(K). Then f extends to a
Frobenius automorphism of M2(K) unless V is equivalent to T +
2 (K) and #K ≤ 3.
We conclude by summing up the previous results in the case of a linear
hyperplane of Mn(K) (the case n = 1 being trivial).
Theorem 26. Let V be a linear hyperplane of Mn(K), and f : V → V be a
linear automorphism such that f (V ∩ GLn(K)) ⊂ GLn(K).
Then f extends to a Frobenius automorphism unless n = 3, K ≃ F2 and V is
equivalent to V1(F2).
Theorem 27. Let f : sln(K) → sln(K) be a linear automorphism such that
f (sln(K) ∩ GLn(K)) ⊂ GLn(K). Then there exists P ∈ GLn(K) and a non-zero
scalar λ such that
∀M ∈ sln(K), f (M ) = λ P M P −1
or ∀M ∈ sln(K), f (M ) = λ P M T P −1.
Note that we find exactly the linear preservers of nilpotency (the common
ground being the case n = 2, as we have just seen)!
Proof of Theorem 27. Using Theorem 26, it suffices to show that a Frobenius au-
tomorphism which stabilizes sln(K) must be of the aforementioned form. Since
sln(K) is stable under transposition, it suffices to fix an arbitrary (P, Q) ∈
GLn(K)2 such that uP,Q stabilizes sln(K) and prove that Q is a scalar multiple
of P −1. However, for every M ∈ sln(K), one has tr(QP M ) = tr(P M Q) = 0
hence QP is a scalar multiple of In, which proves our claim.
References
[1] M.D. Atkinson, S. Lloyd, Large spaces of matrices of bounded rank, Quart.
J. Math. Oxford (2), 31 (1980) 253-262.
[2] L.B. Beasley, R. Loewy, Rank preservers on spaces of symmetric matrices,
Lin. Multilin. Alg., 43 (1997), 63-86.
34
[3] E.P. Botta, Linear maps that preserve singular and nonsingular matrices,
Linear Algebra Appl., 20 (1978) 45-49.
[4] E.P. Botta, S. Pierce, W. Watkins, Linear transformations that preserve
nilpotent matrices, Pacific J. Math., 104 (1983) 39-46.
[5] W.L. Chooi, M.H. Lim, Linear preservers on triangular matrices, Linear
Algebra Appl., 269 (1998) 241-255.
[6] J. Dieudonn´e, Sur une g´en´eralisation du groupe orthogonal `a quatre vari-
ables, Arch. Math., 1 (1949) 282-287.
[7] N. Jacobson, Lectures in Abstract Algebra (II), The University Series in
Higher Mathematics, Van Nostrand, 1953.
[8] A.A. Jafarian, A.R. Sourour, Spectrum-preserving linear maps, Journal of
Functional Analysis, 66 (1986-2) 255-261.
[9] R. Meshulam, On the maximal rank in a subspace of matrices, Quart. J.
Math. Oxford (2), 36 (1985) 225-229.
[10] C. de Seguins Pazzis, The affine preservers of non-singular matrices, Arch.
Math., 95 (2010) 333-342.
[11] C. de Seguins Pazzis, The classification of large spaces of matrices with
bounded rank, ArXiv preprint http://arxiv.org/abs/1004.0298
[12] C. de Seguins Pazzis, The singular linear preservers of non-singular matri-
ces, Linear Algebra Appl., 433 (2010) 483-490.
35
|
1907.03845 | 1 | 1907 | 2019-07-08T20:11:25 | On Higher {g_n, h_n}-derivations | [
"math.RA"
] | In this article, we introduce the concepts of higher {g_n, h_n}-derivation and Jordan higher {g_n, h_n}-derivation, and then we give a characterization of higher {g_n, h_n}-derivations in terms of {g, h}-derivations. Using this result, we prove that every Jordan higher {g_n, h_n}-derivation on a semiprime algebra is a higher {g_n, h_n}-derivation. | math.RA | math | ON HIGHER {gn, hn}-DERIVATIONS
AMIN HOSSEINI∗ AND NADEEM UR REHMAN
Abstract. In this article, we introduce the concepts of higher {gn, hn}-derivation
and Jordan higher {gn, hn}-derivation, and then we give a characterization of
higher {gn, hn}-derivations in terms of {g, h}-derivations. Using this result, we
prove that every Jordan higher {gn, hn}-derivation on a semiprime algebra is
a higher {gn, hn}-derivation.
9
1
0
2
l
u
J
8
]
.
A
R
h
t
a
m
[
1
v
5
4
8
3
0
.
7
0
9
1
:
v
i
X
r
a
1. Introduction and preliminaries
Recently, in 2016 Bresar [1] introduced the notion of {g, h}-derivation. Let A
be an algebra over a field F with char(F) 6= 2, and let f, g, h : A → A be linear
maps. We say that f is a {g, h}-derivation if f (ab) = g(a)b+ah(b) = h(a)b+ag(b)
for all a, b ∈ A, and f is called a Jordan {g, h}-derivation if f (a ◦ b) = g(a) ◦
b + a ◦ h(b) for all a, b ∈ A, where a ◦ b = ab + ba. We call a ◦ b the Jordan
product of a and b. It is evident that a ◦ b = b ◦ a for all a, b ∈ A. The notion of a
Jordan {g, h}-derivation is a generalization of what is called a Jordan generalized
derivation in [8]. Recall that a linear mapping f : A → A is called a Jordan
generalized derivation if there exists a linear mapping d : A → A such that
f (a ◦ b) = f (a) ◦ b + a ◦ d(b) for all a, b ∈ A, where d is called an associated linear
map of f . It is clear that f (a ◦ b) = d(a) ◦ b + a ◦ f (b) for all a, b ∈ A. Obviously,
the definition of a Jordan generalized derivation is generally not equivalent to the
ordinary Jordan case of generalized derivations. For more details in this regard,
see [1, 8] and references therein.
As an important result, Bresar [1, Theorem 4.3] established that every Jordan
{g, h}-derivation of a semiprime algebra A is a {g, h}-derivation. He also showed
that every Jordan {g, h}-derivation of the tensor product of a semiprime and a
commutative algebra is a {g, h}-derivation. Obviously, every {g, h}-derivation is
a Jordan {g, h}-derivation, but the converse is in general not true. For instance
in this regard see [1, Example 2.1].
In this study, we introduce the concept of a higher {gn, hn}-derivation, and then
we characterize it on algebras. Throughout this paper, A denotes an algebra over
a field of characteristic zero, and I denotes the identity mapping on A. Let f be
a {g, h}-derivation on an algebra A. An easy induction argument implies that
f n(ab) =Pn
k=0(cid:16)n
k(cid:17)gn−k(a)hk(b) =Pn
k=0(cid:16)n
k(cid:17)hn−k(a)gk(b) (Leibniz rule) for each
Date: Received: xxxxxx; Revised: yyyyyy; Accepted: zzzzzz.
∗ Corresponding author.
2010 Mathematics Subject Classification. 47B47; 16W10.
Key words and phrases. derivation; {g, h}-derivation; higher {gn, hn}-derivation; Jordan
higher {gn, hn}-derivation.
1
2
AMIN HOSSEINI AND NADEEM UR REHMAN
a, b ∈ A and each non-negative integer n, where f 0 = g0 = h0 = I. If we define
the sequences {fn}, {gn} and {hn} of linear mappings on A by f0 = g0 = h0 = I,
and fn = f n
n! , then it follows from the Leibniz rule that
fn's, gn's and hn's satisfy
n! and hn = hn
n! , gn = gn
fn(ab) =
n
Xk=0
gn−k(a)hk(b) =
n
Xk=0
hn−k(a)gk(b),
(1.1)
for each a, b ∈ A and each non-negative integer n. This is our motivation to
consider the sequences {fn}, {gn} and {hn} of linear mappings on an algebra
A satisfying (1.1). A sequence {fn} of linear mappings on A is called a higher
{gn, hn}-derivation if there exist two sequences {gn} and {hn} of linear mappings
k=0 hn−k(a)gk(b) for any a, b ∈ A
and any non-negative integer n. Additionally, a sequence {fn} of linear mappings
on A is called a Jordan higher {gn, hn}-derivation if
on A satisfying fn(ab) =Pn
k=0 gn−k(a)hk(b) =Pn
fn(a ◦ b) =
n
Xk=0
gn−k(a) ◦ hk(b),
(1.2)
for each a, b ∈ A and each non-negative integer n. Notice that if {fn} is a
higher {fn, fn}-derivation (resp. Jordan higher {fn, fn}-derivation), then it is an
ordinary higher derivation (resp. Jordan higher derivation). We know that if f is
a {g, h}-derivation, then {fn = f n
n! } is a higher {gn, hn}-derivation, where gn = gn
n! ,
hn = hn
n! and f0 = g0 = h0 = I. We call this kind of higher {gn, hn}-derivation an
ordinary higher {gn, hn}-derivation, but this is not the only example of a higher
{gn, hn}-derivation.
In 2010, Miravaziri [9] characterized all higher derivations on an algebra A in
terms of the derivations on A. In this article, by getting idea from [9], our aim
is to characterize all higher {gn, hn}-derivations on an algebra A in terms of the
{g, h}-derivations on A. Indeed, we show that each higher {gn, hn}-derivation is
a combination of compositions of {g, h}-derivations. As the main result of this
article, we prove that if {fn} is a higher {gn, hn}-derivation on an algebra A with
f0 = g0 = h0 = I, then there is a sequence {Fn} of {Gn, Hn}-derivations on A
such that
j=1
fn =Pn
gn =Pn
hn =Pn
i=1 PPi
i=1 PPi
i=1 PPi
j=1 rj=n(cid:16)Qi
j=1 rj=n(cid:16)Qi
j=1 rj =n(cid:16)Qi
j=1
j=1
1
1
rj +...+ri(cid:17)Fr1...Fri!,
rj+...+ri(cid:17)Gr1...Gri!,
rj+...+ri(cid:17)Hr1...Hri!,
1
ON HIGHER {gn, hn}-DERIVATIONS
3
where the inner summation is taken over all positive integers rjwithPi
j=1 rj = n.
The importance of this result is to transfer the problems such as the charac-
terization of Jordan higher {gn, hn}-derivations on semiprime algebras and auto-
matic continuity of higher {gn, hn}-derivations into the same problems concerning
{g, h}-derivations. Let A be an algebra, D be the set of all higher derivations
{dn}n=0,1,... on A with d0 = I and ∆ be the set of all sequences {δn}n=0,1,...
of derivations on A with δ0 = 0. As an application of the main result of this
article, we investigate Jordan higher {gn, hn}-derivations on algebras. It is a clas-
sical question in which algebras (and rings) a Jordan derivation is necessarily a
derivation. Let us give a brief background in this issue.
In 1957, Herstein [7]
achieved a result which asserts any Jordan derivation on a prime ring of char-
acteristic different from two is a derivation. A brief proof of Herstein's result
can be found in [3]. In 1975, Cusack [5] generalized Herstein's result to 2-torsion
free semiprime rings (see also [2] for an alternative proof). Moreover, Vukman
[12] investigated generalized Jordan derivations on semiprime rings and he proved
that every generalized Jordan derivation of a 2-torsion free semiprime ring is a
generalized derivation. Recently, the first name author along with Ajda Fosner
[6] have studied the same problem for (σ, τ )-derivations from a C ∗-algebra A into
a Banach A-module M. In this paper, we show that if {fn} is a Jordan higher
{gn, hn}-derivation of a semiprime algebra A with f0 = g0 = h0 = I, then it is a
higher {gn, hn}-derivation.
2. characterization of higher {gn, hn}-derivations on algebras
Throughout the article, A denotes an algebra over a field of characteristic zero,
and I is the identity mapping on A. Let f, g, h : A → A be linear maps. We
say that f is a {g, h}-derivation if f (ab) = g(a)b + ah(b) = h(a)b + ag(b) for all
a, b ∈ A, and f is called a Jordan {g, h}-derivation if f (a ◦ b) = g(a) ◦ b + a ◦ h(b)
for all a, b ∈ A, where a ◦ b = ab + ba.
We begin with the following definition.
Definition 2.1. A sequence {fn} of linear mappings on A is called a higher
{gn, hn}-derivation if there are two sequences {gn} and {hn} of linear mappings on
k=0 hn−k(a)gk(b) for each a, b ∈ A
k=0 gn−k(a)hk(b) = Pn
A such that fn(ab) = Pn
and each non-negative integer n.
We begin our results with the following lemma which will be used extensively to
prove the main theorem of this article. The following lemma has been motivated
by [9].
Lemma 2.2. Let {fn} be a higher {gn, hn}-derivation on an algebra A with f0 =
g0 = h0 = I. Then there is a sequence {Fn} of {Gn, Hn}-derivations on A such
4
that
AMIN HOSSEINI AND NADEEM UR REHMAN
(n + 1)fn+1 =Pn
(n + 1)gn+1 =Pn
(n + 1)hn+1 =Pn
k=0 Fk+1fn−k,
k=0 Gk+1gn−k,
k=0 Hk+1hn−k
for each non-negative integer n.
Proof. Using induction on n, we prove this lemma. Let n = 0. We know that
f1(ab) = g1(a)b + ah1(b) = h1(a)b + ag1(b) for all a, b ∈ A. Thus, if F1 = f1, G1 =
g1 and H1 = h1, then F1 is a {G1, H1}-derivation on A and further, (0 + 1)f0+1 =
k=0 Hk+1h0−k.
As induction assumption, suppose that Fk is a {Gk, Hk}-derivation for any k ≤ n
and further
k=0 Gk+1g0−k and (0 + 1)h0+1 =P0
k=0 Fk+1f0−k, (0 + 1)g0+1 =P0
P0
(r + 1)fr+1 =Pr
(r + 1)gr+1 =Pr
(r + 1)hr+1 =Pr
k=0 Fk+1fr−k,
k=0 Gk+1gr−k,
k=0 Hk+1hr−k
for r = 0, 1, ..., n − 1. Put Fn+1 = (n + 1)fn+1 −Pn−1
k=0 Gk+1gn−k and Hn+1 = (n + 1)hn+1 −Pn−1
1)gn+1 −Pn−1
k=0 Fk+1fn−k, Gn+1 = (n +
k=0 Hk+1hn−k. Our next
task is to show that Fn+1 is a {Gn+1, Hn+1}-derivation on A. For a, b ∈ A, we
have
Fn+1(ab) = (n + 1)fn+1(ab) −
Fk+1fn−k(ab)
n−1
Xk=0
= (n + 1)
n+1
Xk=0
gk(a)hn+1−k(b) −
n−1
Xk=0
Fk+1(cid:16)
n−k
Xl=0
gl(a)hn−k−l(b)(cid:17).
So, we have
Fn+1(ab) =
=
n+1
n+1
Xk=0
Xk=0
(n + 1)gk(a)hn+1−k(b) −
n−1
Xk=0
(k + n + 1 − k)gk(a)hn+1−k(b) −
n−k
Fk+1(cid:16)
gl(a)hn−k−l(b)(cid:17)
Xl=0
Fk+1(cid:16)
Xk=0
Xl=0
n−1
n−k
gl(a)hn−k−l(b)(cid:17).
Since Fk is a {Gk, Hk}-derivation for each k = 1, 2, ..., n,
Fn+1(ab) =
−
n+1
n−1
Xk=0
Xk=0
n−k
kgk(a)hn+1−k(b) +
gk(a)(n + 1 − k)hn+1−k(b)
n+1
Xk=0
Xl=0 hGk+1(cid:0)gl(a)(cid:1)hn−k−l(b) + gl(a)Hk+1(cid:0)hn−k−l(b)(cid:1)i.
ON HIGHER {gn, hn}-DERIVATIONS
5
Letting
G =
H =
n+1
Xk=0
n+1
Xk=0
kgk(a)hn+1−k(b) −
n−1
Xk=0
n−k
Xl=0
Gk+1(cid:0)gl(a)(cid:1)hn−k−l(b),
gk(a)(n + 1 − k)hn+1−k(b) −
n−1
Xk=0
n−k
Xl=0
gl(a)Hk+l(cid:0)hn−k−l(b)(cid:1),
we have Fn+1(ab) = G + H. Here, we compute G and H.
In the summation
l=0 , we have 0 ≤ k + l ≤ n and k 6= n. Thus if we put r = k + l, then
n+1
G =
Pn−1
k=0Pn−k
we can write it as the form Pn
Xk=0
Xk=0
kgk(a)hn+1−k(b) −
kgk(a)hn+1−k(b) −
n−1
n+1
=
n
r=0Pk+l=r,k6=n. Putting l = r − k, we find that
Xr=0 X0≤k≤r,k6=n
Xr=0
Gk+1(cid:0)gr−k(a)(cid:1)hn−r(b)
Xk=0
Gk+1(cid:0)gr−k(a)(cid:1)hn−r(b) −
Gk+1(cid:0)gn−k(a)(cid:1)b.
Xk=0
n−1
r
It means that
G +
n−1
Xk=0
Gk+1(cid:0)gn−k(a)(cid:1)b =
n+1
Xk=0
kgk(a)hn+1−k(b) −
n−1
Xr=0
r
Xk=0
Gk+1(cid:0)gr−k(a)(cid:1)hn−r(b).
Putting r + 1 instead of k in the first summation of above, we have
n−1
(r + 1)gr+1(a)hn−r(b) −
G +
=
=
n
Gk+1(cid:0)gn−k(a)(cid:1)b
Xk=0
Xr=0
Xr=0h(r + 1)gr+1(a) −
n−1
n−1
r
Xk=0
Gk+1(cid:0)gr−k(a)(cid:1)hn−r(b)
Xr=0
Gk+1(cid:0)gr−k(a)(cid:1)ihn−r(b) + (n + 1)gn+1(a)b.
r
Xk=0
According to the induction hypothesis, (r + 1)gr+1(a) =Pr
r = 0, ..., n − 1. So, it is obtained that
k=0 Gk+1(cid:0)gr−k(a)(cid:1) for
G =h(n + 1)gn+1(a) −
n−1
Xk=0
Gk+1(cid:0)gn−k(a)(cid:1)ib = Gn+1(a)b.
Like above, we achieve that
H = ah(n + 1)hn+1(b) −
n−1
Xk=0
Hk+1(cid:0)hn−k(b)(cid:1)i = aHn+1(b).
6
AMIN HOSSEINI AND NADEEM UR REHMAN
Therefore, we have Fn+1(ab) = G + H = Gn+1(a)b + aHn+1(b). In the next step,
we will show that Fn+1(ab) = Hn+1(a)b + aGn+1(b). We have
Fn+1(ab) = (n + 1)fn+1(ab) −
Fk+1fn−k(ab)
n−1
Xk=0
= (n + 1)
n+1
Xk=0
n+1
hk(a)gn+1−k(b) −
n−1
Xk=0
Fk+1(cid:16)
n−k
Xl=0
hl(a)gn−k−l(b)(cid:17).
(n + 1)hk(a)gn+1−k(b) −
(k + n + 1 − k)hk(a)gn+1−k(b) −
n−1
Xk=0
n−k
Fk+1(cid:16)
hl(a)gn−k−l(b)(cid:17)
Xl=0
Fk+1(cid:16)
Xk=0
Xl=0
n−1
n−k
hl(a)gn−k−l(b)(cid:17).
So,
Fn+1(ab) =
=
n+1
Xk=0
Xk=0
Since Fk is a {Gk, Hk}-derivation for each k = 1, 2, ..., n,
Fn+1(ab) =
−
n+1
n−1
Xk=0
Xk=0
n−k
khk(a)gn+1−k(b) +
hk(a)(n + 1 − k)gn+1−k(b)
n+1
Xk=0
Xl=0 hHk+1(cid:0)hl(a)(cid:1)gn−k−l(b) + hl(a)Gk+1(cid:0)gn−k−l(b)(cid:1)i.
Letting
G =
n+1
Xk=0
n+1
khk(a)gn+1−k(b) −
n−1
Xk=0
n−k
Xl=0
Hk+1(cid:0)hl(a)(cid:1)gn−k−l(b),
n−1
n−k
we have Fn+1(ab) = G + H. The next step is to compute G and H.
In the
l=0 , we have 0 ≤ k + l ≤ n and k 6= n. Thus if we put
H =
hk(a)(n + 1 − k)gn+1−k(b) −
Xk=0
summation Pn−1
k=0Pn−k
r = k + l, then we can write it as the form Pn
we find that
Xk=0
n
r=0Pk+l=r,k6=n. Putting l = r − k,
Xl=0
hl(a)Gk+l(cid:0)gn−k−l(b)(cid:1),
G =
=
n+1
n+1
Xk=0
Xk=0
It means that
khk(a)gn+1−k(b) −
khk(a)gn+1−k(b) −
n−1
Xr=0 X0≤k≤r,k6=n
Xr=0
Hk+1(cid:0)hr−k(a)(cid:1)gn−r(b)
Xk=0
Hk+1(cid:0)hr−k(a)(cid:1)gn−r(b) −
Xk=0
n−1
r
Hk+1(cid:0)hn−k(a)(cid:1)b.
G +
n−1
Xk=0
Hk+1(cid:0)hn−k(a)(cid:1)b =
n+1
Xk=0
khk(a)gn+1−k(b) −
n−1
Xr=0
r
Xk=0
Hk+1(cid:0)hr−k(a)(cid:1)gn−r(b).
ON HIGHER {gn, hn}-DERIVATIONS
7
Putting r + 1 instead of k in the first summation of above, we have
n−1
(r + 1)hr+1(a)gn−r(b) −
G +
=
=
n
Hk+1(cid:0)hn−k(a)(cid:1)b
Xk=0
Xr=0
Xr=0h(r + 1)hr+1(a) −
n−1
n−1
r
Xk=0
Hk+1(cid:0)hr−k(a)(cid:1)gn−r(b)
Xr=0
Hk+1(cid:0)hr−k(a)(cid:1)ign−r(b) + (n + 1)hn+1(a)b.
r
Xk=0
According to the induction hypothesis, (r + 1)hr+1(a) =Pr
r = 0, ..., n − 1. So, it is obtained that
k=0 Hk+1(cid:0)hr−k(a)(cid:1) for
n−1
Xk=0
n−1
Xk=0
G =h(n + 1)hn+1(a) −
Hk+1(cid:0)hn−k(a)(cid:1)ib = Hn+1(a)b.
By a reasoning like above, we get that
H = ah(n + 1)gn+1(b) −
Gk+1(cid:0)gn−k(b)(cid:1)i = aGn+1(b).
Therefore, we have Fn+1(ab) = G + H = Hn+1(a)b + aGn+1(b). Consequently,
Fn+1 is a {Gn+1, Hn+1}-derivation and the proof is complete.
(cid:3)
Example 2.3. Using Lemma 2.2, the first five terms of a higher {gn, hn}-derivation
{fn} are
f0 = I,
f1 = F1,
2f2 = F1f1 + F2f0 = F1F1 + F2,
f2 =
1
2
F 2
1 +
1
2
F2,
3f3 = F1f2 + F2f1 + F3f0 = F1(
1
2
1
3
4f4 = F1f3 + F2f2 + F3f1 + F4f0
F1F2 +
F2F1 +
f3 =
1 +
F 3
1
6
1
3
1
6
F3,
F 2
1 +
1
2
F2) + F2F1 + F3,
= F1(cid:16) 1
F 3
1 +
6
F 4
1
24
1 +
f4 =
1
6
1
24
F1F2 +
F 2
1 F2 +
1
12
1
3
F2F1 +
1
3
F3(cid:17) + F2(cid:16) 1
2
F1F2F1 +
F1F3 +
1
12
F 2
1
2
F2F 2
1 +
1
8
F2(cid:17) + F3F1 + F4,
1 +
F 2
2 +
F3F1 +
1
8
1
4
1
4
F4.
Now the main result of this paper reads as follows:
Theorem 2.4. Let {fn} be a higher {gn, hn}-derivation on an algebra A with
f0 = g0 = h0 = I. Then there is a sequence {Fn} of {Gn, Hn}-derivations on A
such that
8
AMIN HOSSEINI AND NADEEM UR REHMAN
j=1
fn =Pn
gn =Pn
hn =Pn
i=1 PPi
i=1 PPi
i=1 PPi
j=1 rj=n(cid:16)Qi
j=1 rj=n(cid:16)Qi
j=1 rj =n(cid:16)Qi
j=1
j=1
1
1
rj +...+ri(cid:17)Fr1...Fri!,
rj+...+ri(cid:17)Gr1...Gri!,
rj+...+ri(cid:17)Hr1...Hri!,
1
where the inner summation is taken over all positive integers rjwith Pi
ar1,...,ri = Qi
Proof. First, we show that if fn, gn and hn are of the above form, then they
satisfy the recursive relations of Lemma 2.2. Simplifying the notation, we put
. Note that if r1 + ... + ri = n + 1, then (n + 1)ar1,...,ri =
n+1. According to the aforementioned assumptions,
ar2,...,ri. Furthermore, an+1 = 1
we have
j=1 rj = n.
rj+...+ri
j=1
1
fn+1 =
=
j=1 rj=n+1
n+1
Xi=2 XPi
Xi=2 XPi
n+1
j=1 rj=n+1
ar1,...riFr1...Fri! + an+1Fn+1
ar1,...riFr1...Fri! +
Fn+1
n + 1
.
(n + 1)ar1,...riFr1...Fri! + Fn+1
ar2,...riFr1...Fri! + Fn+1
ar2,...riFr2...Fri! + Fn+1
ar2,...riFr2...Fri! + Fn+1
j=2 rj=n−(r1−1)
So,
(n + 1)fn+1 =
=
=
=
=
=
n+1
n+1
n+1
j=1 rj =n+1
j=1 rj =n+1
Xi=2 XPi
Xi=2 XPi
Xi=2 n+2−i
Fr1 XPi
Xr1=1
Xi=2 XPi
Xr1=1
Xr1=1
Xk=0
Fr1fn−(r1−1) + Fn+1
Fk+1fn−k.
n−(r1−1)
Fr1
n
n
n
j=2 rj=n+1−r1
ON HIGHER {gn, hn}-DERIVATIONS
9
Using a reasoning like above, we get that
for each non-negative integer n. Putting n + 1 = m, we find that mfm =
k=0 Fk+1fm−1−k + Fm, and consequently
(n + 1)gn+1 =Pn
(n + 1)hn+1 =Pn
k=0 Gk+1gn−k,
k=0 Hk+1hn−k,
k=0 Fk+1fm−1−k =Pm−2
Pm−1
Fm = mfm −
Similarly, we have
Fk+1fm−1−k.
m−2
Xk=0
Therefore, we can define Fn, Gn, Hn : A → A by F0 = G0 = H0 = 0 and
Gm = mgm −Pm−2
Hm = mhm −Pm−2
Fn = nfn −Pn−2
Gn = ngn −Pn−2
Hn = nhn −Pn−2
k=0 Gk+1gm−1−k,
k=0 Hk+1hm−1−k.
k=0 Fk+1fn−1−k,
k=0 Gk+1gn−1−k,
k=0 Hk+1hn−1−k,
for each positive integer n. It follows from Lemma 2.2 that {Fn} is a sequence of
{Gn, Hn}-derivations. In addition, we prove that if fn, gn and hn are of the form
(n + 1)fn+1 =Pn
(n + 1)gn+1 =Pn
(n + 1)hn+1 =Pn
k=0 Fk+1fn−k,
k=0 Gk+1gn−k,
k=0 Hk+1hn−k,
where {Fn} is a sequence of {Gn, Hn}-derivations, then {fn} is a higher {gn, hn}-
derivation on A with f0 = g0 = h0 = I. To see this, we use induction on n.
For n = 0, we have f0(ab) = ab = g0(a)h0(b) = h0(a)g0(b). As the inductive
hypothesis, assume that
fk(ab) =
k
Xi=0
gi(a)hk−i(b) =
k
Xi=0
hi(a)gk−i(b) f or k ≤ n.
10
AMIN HOSSEINI AND NADEEM UR REHMAN
Therefore, we have
n
(n + 1)fn+1(ab) =
=
=
n−k
n
n
Fk+1
Fk+1fn−k(ab)
Xk=0
Xi=0
Xk=0
Xi=0 n−i
Gk+1gn−k−i(a)!hi(b) +
Xk=0
gi(a)hn−k−i(b)
n
Xi=0
gi(a) n−i
Xk=0
Hk+1hn−k−i(b)!.
According to the above-mentioned recursive relations, we continue the previous
expressions as follows:
(n − i + 1)gn−i+1(a)hi(b) +
gi(a)(n − i + 1)hn−i+1(b)
n
Xi=0
igi(a)hn+1−i(b) +
(n − i + 1)gi(a)hn+1−i(b)
(n + 1)fn+1(ab) =
=
n
n+1
Xi=0
Xi=1
n
Xi=0
= (n + 1)
gi(a)hn+1−i(b),
n+1
Xi=0
which means that fn+1(ab) = Pn+1
we can obtain that fn+1(ab) = Pn+1
i=0 gi(a)hn+1−i(b). By an argument like above,
i=0 hi(a)gn+1−i(b). Thus, {fn} is a higher
{gn, hn}-derivation on A which is characterized by the sequence {Fn} of {Gn, Hn}-
derivations. This completes the proof.
(cid:3)
In the next example, using the above theorem, we characterize term f4 of a
higher {gn, hn}-derivation {fn}.
Example 2.5. We compute the coefficients ar1,...,ri for the case n = 4. First, note
that 4 = 1 + 3 = 3 + 1 = 2 + 2 = 1 + 1 + 2 = 1 + 2 + 1 = 2 + 1 + 1 = 1 + 1 + 1 + 1.
ON HIGHER {gn, hn}-DERIVATIONS
11
Based on the definition of ar1,...,ri we have
a4 =
1
4
,
a1,3 =
a3,1 =
a2,2 =
1
1 + 3
1
3 + 1
1
2 + 2
=
=
=
,
1
12
1
4
1
8
,
,
.
.
1
3
1
1
1
2
1
.
a1,1,2 =
a1,2,1 =
a2,1,1 =
a1,1,1,1 =
1 + 1 + 2
1
1 + 2 + 1
1
2 + 1 + 1
1
.
.
.
1
1 + 2
1
2 + 1
1
1 + 1
.
.
.
1
2
1
1
1
1
.
,
,
1
24
1
12
1
8
,
=
=
=
1
1 + 1 + 1 + 1
1 + 1 + 1
.
1
1 + 1
.
1
1
=
1
24
.
Therefore, f4 is characterized as follows:
f4 =
1
4
F4 +
1
12
F1F3 +
1
4
F3F1 +
1
8
F2F2 +
1
24
F1F1F2 +
1
12
F1F2F1
+
1
8
F2F1F1 +
1
24
F1F1F1F1.
In the following there are some immediate consequences of the above theorem.
Before it, recall that a sequence {fn} of linear mappings on A is called a Jordan
higher {gn, hn}-derivation if there exist two sequences {gn} and {hn} of linear
k=0 gn−k(a) ◦ hk(b) holds for each a, b ∈ A
and each non-negative integer n. Since the Jordan product is commutative, we
have
mappings on A such that fn(a ◦ b) =Pn
n
fn(a ◦ b) = fn(b ◦ a) =
=
=
gn−k(b) ◦ hk(a)
gk(b) ◦ hn−k(a)
hn−k(a) ◦ gk(b).
n
Xk=0
Xk=0
Xk=0
n
So, it is observed that if {fn} is a Jordan higher {gn, hn}-derivation, then
fn(a ◦ b) =
for all a, b ∈ A.
gn−k(a) ◦ hk(b) =
n
Xk=0
hn−k(a) ◦ gk(b),
n
Xk=0
12
AMIN HOSSEINI AND NADEEM UR REHMAN
Corollary 2.6. Let {fn} be a Jordan higher {gn, hn}-derivation on a semiprime
algebra A with f0 = g0 = h0 = I. Then {fn} is a higher {gn, hn}-derivation.
Proof. Using the proof of Theorem 2.4, we can show that if {fn} is a Jordan
higher {gn, hn}-derivation on an algebra A with f0 = g0 = h0 = I, then there
exists a sequence {Fn} of Jordan {Gn, Hn}-derivations on A such that
fn =
n
i
Xi=1 (cid:0) XPi
j=1 rj=n(cid:0)
Yj=1
1
rj + ... + ri(cid:1)Fr1...Fri(cid:1),
where the inner summation is taken over all positive integers rj withPi
Since A is a semiprime algebra, [1, Theorem 4.3] proves the corollary.
j=1 rj = n.
(cid:3)
Remark 2.7. We know that the notion of a Jordan {g, h}-derivation is a gen-
eralization of a Jordan generalized derivation (see Introduction). A sequence
{fn} of linear mappings on an algebra A is called a Jordan higher generalized
derivation if there exists a sequence {dn} of linear mappings on A such that
k=0 fn−k(a) ◦ dk(b) for all a, b ∈ A. Obviously, if {fn} is a Jordan
higher generalized derivation associated with a sequence {dn} of linear mappings
on A, then it is a Jordan higher {fn, dn}-derivation. So, Corollary 2.6 is also
valid for Jordan higher generalized derivations.
fn(a ◦ b) = Pn
Theorem 2.8. If {fn}n=0,1,... is a higher {gn, hn}-derivation on A with f0 = g0 =
h0 = I, then there is a sequence {Fn}n=0,1,... of {Gn, Hn}-derivations with F0 =
G0 = H0 = 0 characterizing the higher {gn, hn}-derivation {fn}n=0,1,.... As well
as if {Fn}n=0,1,... is a sequence of {Gn, Hn}-derivations with F0 = G0 = H0 = 0,
then there exists a higher {gn, hn}-derivation {fn} with f0 = g0 = h0 = I which
is characterized by the sequence {Fn}n=0,1,....
Proof. Let {fn} be a higher {gn, hn}-derivation on A with f0 = g0 = h0 = I.
We are going to obtain a sequence {Fn}n=0,1,... of {Gn, Hn}-derivations with F0 =
G0 = H0 = 0 that characterizes the higher {gn, hn}-derivation {fn}. Define
Fn, Gn, Hn : A → A by F0 = G0 = H0 = 0 and
k=0 Fk+1fn−1−k,
k=0 Gk+1gn−1−k,
Fn = nfn −Pn−2
Gn = ngn −Pn−2
Hn = nhn −Pn−2
k=0 Hk+1hn−1−k,
for each positive integer n. Then it follows from Lemma 2.2 that {Fn} is a
sequence of {Gn, Hn}-derivations characterizing the higher {gn, hn}-derivation
{fn}. Now suppose that {Fn}n=0,1,... is a sequence of {Gn, Hn}-derivations with
F0 = G0 = H0 = 0. We will show that there exists a higher {gn, hn}-derivation
{fn} with f0 = g0 = h0 = I which is characterized by the sequence {Fn}n=0,1,....
ON HIGHER {gn, hn}-DERIVATIONS
13
We define fn, gn, hn : A → A by f0 = g0 = h0 = I and
i=1 PPi
j=1 rj=n(cid:16)Qi
fn =Pn
i=1 PPi
j=1 rj=n(cid:16)Qi
gn =Pn
i=1 PPi
j=1 rj =n(cid:16)Qi
hn =Pn
(n + 1)fn+1 =Pn
(n + 1)gn+1 =Pn
(n + 1)hn+1 =Pn
j=1
j=1
j=1
1
1
rj +...+ri(cid:17)Fr1...Fri!,
rj+...+ri(cid:17)Gr1...Gri!,
rj+...+ri(cid:17)Hr1...Hri!.
1
k=0 Fk+1fn−k,
k=0 Gk+1gn−k,
k=0 Hk+1hn−k.
By Theorem 2.4, fn, gn and hn satisfy the following recursive relations:
Based on the last part of the proof of Theorem 2.4, {fn} is a higher {gn, hn}-
(cid:3)
derivation on A with f0 = g0 = h0 = I. This yields the desired result.
An immediate corollary of the precede theorem reads as follows:
Corollary 2.9. If {dn}n=0,1,... with d0 = I is a higher derivation on A, then there
is a sequence {δn}n=0,1,... of derivations with δ0 = 0 characterizing the higher
derivation {dn}n=0,1,.... As well as if {δn}n=0,1,... is a sequence of derivations
with δ0 = 0, then there exists a higher derivation {dn} with d0 = I which is
characterized by the sequence {δn}n=0,1,....
References
1. M. Bresar, Jordan {g, h}-derivations on tensor products of algebras, Linear Multilinear
Algebra. 64 (2016), no. 11, 2199-2207.
2. M. Bresar, Jordan derivations on semiprime rings, Proc. Amer. Math. Soc. 140 (1988), no.
4, 1003 -- 1006.
3. M. Bresar and J. Vukman, Jordan derivations on prime rings, Bull. Austral. Math. Soc. 37
(1988), 321 -- 322.
4. M. Bresar, Characterizations of derivations on some normed algebras with involution, J.
Algebra 152 (1992), 454 -- 462.
5. J. Cusack, Jordan derivations on rings, Proc. Amer. Math. Soc. 53 (1975), 1104 -- 1110.
6. A. Hosseini and A. Fosner, Identities related to (σ, τ )-Lie derivations and (σ, τ )-derivations,
Boll. Unione Mat. Ital. DOI 10.1007/s40574-017-0141-1, (to appear).
7. I. N. Herstein, Jordan derivations of prime rings, Proc.Amer. Math. Soc. 8 (1957), 1104 --
1110.
8. Y. Li and D. Benkovic. Jordan generalized derivations on triangular algebras. Linear Mul-
tilinear Algebra. 59 (2011), 841849.
9. M. Mirzavaziri, Characterization of higher derivations on algebras, Communications in
algebra. 38 (2010), no. 3, 981 -- 987.
10. G. J. Murphy, C ∗-Algebras and Operator Theory. Boston Academic Press, (1990).
14
AMIN HOSSEINI AND NADEEM UR REHMAN
11. M. Takesaki, Theory of Operator Algebras , Springer-Verlag Berlin Heidelberg New York,
2001.
12. J. Vukman, A note on generalized derivations of semiprime rings, Taiwanese J. Math. 11
(2007), 367-370.
Amin Hosseini, Department of Mathematics, Kashmar Higher Education Institute-
Kashmar- Iran
E-mail address: [email protected], [email protected]
Nadeem Ur Rehman, Department of Mathematics, Aligarh Muslim University,
Aligarh-202002 India
E-mail address: [email protected]
|
1304.7824 | 1 | 1304 | 2013-04-30T00:47:10 | Geometrical Structures of the Endomorphism Semiring of a Finite Chain: Simplices, Strings and Triangles | [
"math.RA"
] | We establish new results concerning endomorphisms of a finite chain if the cardinality of the image of such endomorphism is no more than some fixed number k. The semiring of all such endomorphisms can be seen as a k - simplex whose vertices are the constant endomorphisms. We explore the properties of these k - simplices and find some results for arbitrary k. For k = 1 and 2 we give a full description of simplices called strings and triangles, respectively. | math.RA | math |
where m < k, m 6= 0.
{z }
n
Geometrical Structures of the Endomorphism Semiring
of a Finite Chain: Simplices, Strings and Triangles
Ivan Trendafilov
Department of Algebra and Geometry, Faculty of Applied Mathematics and
Informatics, Technical University of Sofia, Sofia, Bulgaria,
e-mail: [email protected]
We establish new results concerning endomorphisms of a finite chain if the cardinality
of the image of such endomorphism is no more than some fixed number k. The
semiring of all such endomorphisms can be seen as a k -- simplex whose vertices are
the constant endomorphisms. We explore the properties of these k -- simplices and find
some results for arbitrary k. For k = 1 and 2 we give a full description of simplices
called strings and triangles, respectively.
1.
Introduction and Preliminaries
The endomorphism semiring of a finite semilattice is well studied in [1] -- [7]. In the
present paper we give a new treatment of the subsemirings of endomorphism semiring bECn
of a finite chain. We investigate endomorphisms α ∈ bECn such that Im(α) ≤ k, where
k is a positive integer and k ≤ n. The set of these endomorphisms is a k -- simplex with
≀ and proper sides all the m -- simplices,
vertices constant endomorphisms a = ≀ a, . . . , a
The paper is organized as follows. After the introduction and preliminaries, in the
second section we study k -- simplices for any natural k. Although we do not speak about
any distance here, we define discrete neighborhoods with respect to any vertex of the
simplex. In the main result of the section Theorem 5, we prove that the biggest discrete
neighborhoods of the least and biggest vertex of the simplex are the semirings of a special
type. In the third section we start the inductive study of the simplices and consider 1-
simplices called strings. Section 4 is devoted to the basic properties of triangle △(n){a, b, c}.
Here we prove that any element of the interior of a triangle is a sum of the elements of
two strings of this triangle. In the next section we consider the set of endomorphisms of
△(n){a, b, c} such that some of elements a, b and c occur just m times, 0 ≤ m < n, of
the image of the endomorphism. These sets are called layers of the triangle with respect
to some vertex. When two boundary elements of the layer are idempotents, we call this
layer a basic layer and prove that all basic layers with respect to a and c are semirings
isomorphic to well-defined strings.
In the last section are placed the main results of the paper. Here we construct a new
subsemiring of △(n){a, b, c}, the so-called idempotent triangle, containing all the right
identities of triangle △(n){a, b, c}. In Theorem 33 we prove that the idempotent triangle is
a disjoint union of two subsmirings. The first one is the semiring of the right identities of
△(n){a, b, c} and the second is the maximal ideal of the idempotent triangle. In this section
we describe the subsemirings of a -- nilpotent, b -- nilpotent and c -- nilpotent endomorphisms
as geometric figures: trapezoids and parallelogram. In this section is also investigated
two subsemirings which are geometric parallelograms. In Theorem 40 we show that any
triangle △(n){a, b, c} is a disjoint union of eight subsemirings.
2
IVAN TRENDAFILOV
Since the terminology for semirings is not completely standardized, we say what our
conventions are. An algebra R = (R, +, .) with two binary operations + and · on R, is
called a semiring if:
• (R, +) is a commutative semigroup,
• (R, ·) is a semigroup,
• both distributive laws hold x · (y + z) = x · y + x · z and (x + y) · z = x · z + y · z for
any x, y, z ∈ R.
Let R = (R, +, .) be a semiring. If a neutral element 0 of the semigroup (R, +) exists
and 0x = 0, or x0 = 0, it is called a left or a right zero, respectively, for all x ∈ R. If
0 · x = x · 0 = 0 for all x ∈ R, then it is called zero. An element e of a semigroup (R, ·) is
called a left (right) identity provided that ex = x, or xe = x, respectively, for all x ∈ R.
If a neutral element 1 of the semigroup (R, ·) exists, it is called identity.
A nonempty subset I of R is called an ideal if I + I ⊆ I, R I ⊆ I and I R ⊆ I.
The facts concerning semirings can be found in [8]. For semilattices we refer to [9].
For a join-semilattice (M, ∨) set EM of the endomorphisms of M is a semiring with
respect to the addition and multiplication defined by:
• h = f + g when h(x) = f (x) ∨ g(x) for all x ∈ M,
• h = f · g when h(x) = f (g(x)) for all x ∈ M.
This semiring is called the endomorphism semiring of M. In this article all semilattices
are finite chains. Following [5] and [6], we fix a finite chain Cn = ({0, 1, . . . , n − 1} , ∨)
Subsemirings E a
α(a) = a are considered in [4].
and denote the endomorphism semiring of this chain with bECn. We do not assume that
α(0) = 0 for arbitrary α ∈ bECn. So, there is not a zero in endomorphism semiring bECn.
Cn, where a ∈ Cn, of bECn consisting of all endomorphisms α with property
If α ∈ bECn such that f (k) = ik for any k ∈ Cn we denote α as an ordered n -- tuple
≀ i0, i1, i2, . . . , in−1 ≀. Note that mappings will be composed accordingly, although we shall
usually give preference to writing mappings on the right, so that α · β means "first α, then
β". The identity i = ≀ 0, 1, . . . , n − 1 ≀ and all constant endomorphisms κi = ≀ i, . . . , i ≀ are
obviously (multiplicatively) idempotents.
Let a ∈ Cn. For every endomorphism a = ≀ a a . . . a ≀ the elements of
N [a]
n = {α α ∈ bECn, αna = a for some natural number na}
endomorphisms. An important
result
for
called a -- nilpotent
are
endomorphisms is
a -- nilpotent
Theorem (Theorem 3.3 from [3]), For any natural n, n ≥ 2, and a ∈ Cn the set of
is a subsemiring of bECn. The order of this semiring is
a -- nilpotent endomorphisms N [a]
n
Another useful result is
n (cid:12)(cid:12)(cid:12) = Ck.Cn−k−1, where Ck is the k -- th Catalan number.
(cid:12)(cid:12)(cid:12)N [k]
Theorem (Theorem 9 from [2]), The subset of bECn, n ≥ 3, of all
s−1Ym=1
endomorphisms with s fixed points k1, . . . , ks, 1 ≤ s ≤ n − 1,
(km+1 − km).
For definitions and results concerning simplices we refer to [10] and [11].
idempotent
is a semiring of order
GEOMETRICAL STRUCTURES OF THE ENDOMORPHISM SEMIRING
3
2. Simplices
Let us fix elements a0, a1, . . . , ak−1 ∈ Cn, where k ≤ n, a0 < a1 < . . . < ak−1, and let
A = {a0, a1, . . . , ak−1}. We consider endomorphisms α ∈ bECn such that Im(α) ⊆ A. Let us
call k -- simplex (or only a simplex ) the set of all such endomorphisms α. We denote k --
simplex by σ(n)
k (A) = σ(n){a0, a1, . . . , ak−1}.
It is easy to see that Im(α) ⊆ A and Im(β) ⊆ A imply Im(α+β) ⊆ A and Im(α·β) ⊆ A.
Hence, we find
Proposition 1. For any A = {a0, a1, . . . , ak−1} ⊆ Cn k -- simplex σ(n)
k (A) is a
subsemiring of bECn.
The number k is called a dimension of k -- simplex σ(n)
k (A). Endomorphisms aj, where
j = 0, . . . , k − 1, such that aj(i) = aj for any i ∈ Cn are called vertices of k -- simplex
σ(n)
k (A).
Any simplex σ(n){b0, b1, . . . , bℓ−1}, where b0, . . . , bℓ−1 ∈ A, is called a face of k -- simplex
σ(n)
k (A). If ℓ < k, face σ(n){b0, b1, . . . , bℓ−1} is called a proper face.
The proper faces of k -- simplex σ(n)
• 0 -- simplices, which are vertices a0, . . . , ak.
• 1 -- simplices, which are called strings. They are denoted by ST R(n){a, b}, where
k (A) = σ(n){a0, a1, . . . , ak−1} are:
a, b ∈ A.
• 2 -- simplices, which are called triangles. They are denoted by △(n){a, b, c}, where
a, b, c ∈ A.
• 3 -- simplices, which are called tetrahedra. They are denoted by T ET R(n){a, b, c, d},
where a, b, c, d ∈ A.
• The last proper faces are k − 1 -- simplices σ(n)
The boundary of k -- simplex σ(n)
k−1(B), where B = {b0, . . . , bk−2} ⊂ A.
k (A) is a union of all its proper faces and is denoted
by BD(cid:16)σ(n)
k (A)(cid:17). The set IN T (cid:16)σ(n)
k (A)(cid:17) = σ(n)
k (A)\BD(cid:16)σ(n)
k (A)(cid:17) is called an interior
of k -- simplex σ(n)
semirings.
k (A). The boundary and the interior of k -- simplex are, in general, not
The interior of this simplex consists of endomorphisms α, such that Im(α) = Cn. Since
For any natural n, endomorphism semiring bECn is n -- simplex with vertices 0, . . . , n − 1.
the latter is valid only for identity i = ≀ 0, 1, . . . n − 1 ≀, it follows that IN T (cid:16)bECn(cid:17) = i.
There is a partial ordering of the faces of dimension k − 1 of k -- simplex by the
following way: least face does not contain the vertex ak and biggest face does not contain
the vertex a0.
is σ(n)
n−1{1, . . . , n−1} = E (0)
The biggest face of the n -- simplex bECn is the (n−1) -- simplex σ(n)
bECn\σ(n)
Cn which is a subsemiring of bECn. Similarly, the least face of bECn
n−1{0, . . . , n − 2}. Then bECn\σ(n)
bECn. The other faces of bECn, where n ≥ 3, do not have this property. Indeed, one middle side
n−1{0, . . . , k −1, k +1, . . . , n−1}. But set R = bECn\σ(n)
is σ(n)
n−1{0, . . . , k −1, k +1, . . . , n−1} is
not a semiring because for any n ≥ 3 and any k ∈ {1, . . . , n − 2} if α = ≀ 0, . . . , 0, k ≀ ∈ R,
then α2 = 0 /∈ R.
n−1{0, . . . , n − 2} = E (n−1)
which is also a subsemiring of
n−1{1, . . . , n−1}. Now
Cn
4
IVAN TRENDAFILOV
Let us fix vertex am, where m = 0, . . . , k − 1, of simplex σ(n){a0, a1, . . . , ak−1}. The
set of all endomorphisms α ∈ σ(n)
k (A) such that α(i) = am just for s elements i ∈ Cn is
called s-th layer of k -- simplex with respect to am, where s = 0, . . . , n − 1. We denote the
s-th layer of the simplex with respect to am by Ls
layer with respect to any vertex of the k -- simplex is a face of the k -- simplex, hence, it is
a semiring. In the general case, the s-th layer Ls
s = 1, . . . , n − 2, is not a subsemiring of k -- simplex.
am(cid:0)σ(n){a0, a1, . . . , ak−1}(cid:1). So, the 0 --
am(cid:0)σ(n){a0, a1, . . . , ak−1}(cid:1), where s ∈ Cn,
am (cid:0)σ(n){a0, a1, . . . , ak−1}(cid:1) is a discrete
am (cid:0)σ(n){a0, a1, . . . , ak−1}(cid:1) and, in general, DN t
From topological point of view, set {am} ∪ Ln−1
am(cid:0)σ(n){a0, a1, . . . , ak−1}(cid:1), where m = 0, . . . , k − 1, t = 1, . . . , n.
neighborhood consisting of the "nearest points to point" am. We denote this set by DN 1
m.
Similarly, we define DN 2
m =
m = DN 1
m ∪Ln−2
{am} ∪
n−1[ℓ=n−t
Lℓ
Proposition 2. Let am, where m = 0, . . . , k − 1, be a vertex of the simplex
σ(n){a0, a1, . . . , ak−1} and Ln−1
-- simplex with respect to am. Then the set DN 1
m = {am} ∪ Ln−1
where m = 0, . . . , k − 1, is a subsemiring of σ(n){a0, a1, . . . , ak−1}.
am (cid:0)σ(n){a0, a1, . . . , ak−1}(cid:1) be the (n − 1)-th layer of the k
am (cid:0)σ(n){a0, a1, . . . , ak−1}(cid:1),
Proof. We consider three cases.
Case 1. Let m = 0. Then elements of DN 1
0 are endomorphisms:
a0, (a0)n−1a1 = ≀ a0, . . . , a0
, a1 ≀, . . . , (a0)n−1ak−1 = ≀ a0, . . . , a0
, ak−1 ≀ .
n−1
{z
}
n−1
{z
}
Since a0 < (a0)n−1a1 < · · · < (a0)n−1ak−1, it follows that set DN 1
addition.
0 is closed under the
We find (a0)n−1ai · a0 = a0 · (a0)n−1ai = a0 for all i = 1, . . . , k − 1. Also we have
(a0)n−1ai · (a0)n−1aj = (a0)n−1aj · (a0)n−1ai = a0 for all i, j ∈ {1, . . . , k − 1} with the only
exception when ak−1 = n − 1. Now ((a0)n−1(n − 1))2 = (a0)n−1(n − 1), (a0)n−1(n − 1) ·
(a0)n−1ai = (a0)n−1ai and (a0)n−1ai · (a0)n−1(n − 1) = a0. Hence, DN 1
0 is a semiring.
Case 2. Let m = k − 1. Then elements of DN 1
k−1 are endomorphisms:
a0(ak−1)n−1 = ≀ a0, ak−1, . . . , ak−1
≀, . . . , ak−2(ak−1)n−1 = ≀ ak−2, ak−1, . . . , ak−1
≀, ak−1.
n−1
{z
}
n−1
{z
}
Since a0(ak−1)n−1 < · · · < ak−2(ak−1)n−1 < ak−1, it follows that the set DN 1
under the addition.
k−1 is closed
We find ai(ak−1)n−1 · ak−1 = ak−1 · ai(ak−1)n−1 = ak−1 for all i = 1, . . . , k − 1. Also we
have ai(ak−1)n−1·aj(ak−1)n−1 = aj(ak−1)n−1·ai(ak−1)n−1 = ak−1 for all i, j ∈ {0, . . . , k − 2}
with also the only exception when a0 = 0. We have (0(ak−1)n−1)2 = 0(ak−1)n−1, 0(ak−1)n−1·
ai(ak−1)n−1 = ai(ak−1)n−1 and ai(ak−1)n−1·0(ak−1)n−1 = ak−1 Hence, DN 1
k−1 is a semiring.
Case 3. Let 0 < m < k − 1. Then elements of DN 1
m are endomorphisms:
a0(am)n−1 = ≀ a0, am, . . . , am
≀, . . . , am−1(am)n−1 = ≀ am−1, am, . . . , am
≀, am,
(am)n−1am+1 = ≀ am, . . . , am
, am+1 ≀, . . . , (am)n−1ak−1 = ≀ am, . . . , am
, ak−1 ≀ .
n−1
{z
{z
n−1
}
}
n−1
{z
{z
n−1
}
}
GEOMETRICAL STRUCTURES OF THE ENDOMORPHISM SEMIRING
5
Since a0(am)n−1 < · · · < am−1(am)n−1 < am < (am)n−1am+1 < · · · < (am)n−1ak−1, it
follows that set DN 1
m is closed under the addition.
Now there are four possibilities:
3.1. Let 0 < a0 and ak−1 < n − 1. Then
ai(am)n−1 · aj(am)n−1 = aj(am)n−1 · ai(am)n−1 = am for any i, j = 0, . . . m − 1,
(am)n−1ai · (am)n−1aj = (am)n−1aj · (am)n−1ai = am for any i, j = m + 1, . . . k − 1,
ai(am)n−1 · (am)n−1aj = (am)n−1aj · ai(am)n−1 = am
for any i = 0, . . . , m − 1 and j = m + 1, . . . k − 1.
Since ai(am)n−1 · am = am · ai(am)n−1 = am for i = 1, . . . , m − 1 and, in a similar way,
(am)n−1aj · am = am · (am)n−1aj = am for j = m + 1, . . . , k − 1 and also (am)2 = am, it
follows that DN 1
m is a commutative semiring.
3.2. Let a0 = 0 and ak−1 < n − 1. Then (0(am)n−1)2 = 0(am)n−1,
0(am)n−1 · ai(am)n−1 = ai(am)n−1, ai(am)n−1 · 0(am)n−1 = am for any i = 1, . . . m − 1 and
0(am)n−1 · (am)n−1aj = (am)n−1aj · 0(am)n−1 = am for any j = m + 1, . . . , k − 1.
We also observe that am · 0(am)n−1 = 0(am)n−1 · am = am. All the other equalities
between the products of the elements of DN 1
m are the same as in 3.1.
3.3. Let a0 > 0 and ak−1 = n − 1. Then ((am)n−1(n − 1))2 = (am)n−1(n − 1),
(am)n−1(n − 1) · ai(am)n−1 = ai(am)n−1 · (am)n−1(n − 1) = am for any i = 1, . . . m − 1 and
(am)n−1(n − 1) · (am)n−1aj = (am)n−1aj, (am)n−1aj · (am)n−1(n − 1) = am
for any j = m + 1, . . . , k − 1.
We also observe that am · (am)n−1(n − 1) = (am)n−1(n − 1) · am = am. All the other
equalities between the products of the elements of DN 1
m are the same as in 3.1.
3.4. Let a0 = 0 and ak−1 = n − 1. Now all equalities between the products of the
(cid:3)
m are the same as in 3.1., 3.2. and 3.3. So, DN 1
m is a semiring.
elements of DN 1
Any simplex σ(n){b0, b1, . . . , bℓ−1} which is a face of simplex σ(n){a0, a1, . . . , ak−1}
is called internal of the simplex σ(n){a0, a1, . . . , ak−1} if a0 /∈ σ(n){b0, b1, . . . , bℓ−1} and
ak−1 /∈ σ(n){b0, b1, . . . , bℓ−1}. Similarly simplex σ(n){a0, a1, . . . , ak−1}, which is a face of
n -- simplex bECn, is called internal simplex if 0 /∈ σ(n){a0, a1, . . . , ak−1} and n − 1 /∈
σ(n){a0, a1, . . . , ak−1}.
Immediately from the proof of Proposition 2 follows
Corollary 3. For any internal simplex σ(n){a0, a1, . . . , ak−1} semirings DN 1
m are
commutative and all their elements are am -- nilpotent, where m = 0, . . . , k − 1.
Proposition 4. Let am, where m = 0, . . . , k − 1, be a vertex of internal simplex
σ(n){a0, a1, . . . , ak−1}. Then the set DN 2
m ∪ Ln−2
m = 0, . . . , k − 1, is a subsemiring of σ(n){a0, a1, . . . , ak−1}.
m = DN 1
am (cid:0)σ(n){a0, a1, . . . , ak−1}(cid:1), where
6
IVAN TRENDAFILOV
Proof. The elements of DN 2
m are: am, ai(am)n−1, where i = 0, . . . , m − 1, (am)n−1aj,
where j = m + 1, . . . , k − 1, apaq(am)n−2, where p, q = 0, . . . , m − 1, p ≤ q, (am)n−2aras,
where r, s = m + 1, . . . , k − 1, r ≤ s, and ap(am)n−2as, where p = 0, . . . , m − 1, s =
m + 1, . . . , k − 1.
Since DN 1
m is closed under the addition in order to prove the same for DN 2
m, we
consider:
ai(am)n−1+apaq(am)n−2 =(cid:26) apaq(am)n−2
(am)n−1aj + (am)n−2aras =(cid:26) (am)n−2aras
(am)n−2araj
aiaq(am)n−2
if i ≤ p
if i > p
, ai(am)n−1+(am)n−2aras = (am)n−2aras,
if j ≤ s
if j > s
, (am)n−1aj + apaq(am)n−2 = (am)n−1aj,
ai(am)n−1 + ap(am)n−2as =(cid:26) ap(am)n−2as
(am)n−1aj + ap(am)n−2as =(cid:26) ap(am)n−2as
ap(am)n−2aj
ai(am)n−2as
if i ≤ p
if i > p
if j ≤ s
if j > s
,
,
apaq(am)n−2 + (am)n−2aras = (am)n−2aras,
ap0aq0(am)n−2 + ap(am)n−2as =(cid:26) ap(am)n−2as
(am)n−2ar0as0 + ap(am)n−2as =(cid:26) a(am)n−2ar0as
(am)n−2ar0as0
ap0(am)n−2as
if p0 ≤ p
if p0 > p
if s0 ≤ s
if s0 > s
,
,
ap(am)n−2as + ap0(am)n−2as0 =
ap(am)n−2as
ap(am)n−2as0
ap0(am)n−2as
ap0(am)n−2as0
if p ≤ p0, s ≤ s0
if p ≤ p0, s > s0
if p > p0, s ≤ s0
if p > p0, s > s0
,
am + apaq(am)n−2 = am, am + (am)n−2aras = (am)n−2aras, am + ap(am)n−2as = (am)n−1as,
where i, p, q, p0, q0 = 0, 1, . . . , m − 1, p ≤ q, p0 < q0, j, r, s, r0, s0 = m + 1, . . . , k − 1, r ≤ s,
r0 < s0. So, we prove that DN 2
m is closed under the addition.
Now we consider three cases, where, for the indices, the upper restrictions are fulfilled.
Case 1. Let am = 1. We shall show that all endomorphisms of DN 2
1 are 1 -- nilpotent
with the only exception when ak−1 = n − 2. When ak−1 < n − 2, since 1 is the least image
of any endomorphism, there are only a few equalities: 1n−2aras · 1n−2ar0as0 = 1,
1n−1aj · 1n−2aras = 1n−2aras · 1n−1aj = 1, 1 · 1n−2aras = 1n−2aras · 1 = 1.
Hence, it follows that DN 2
1 is a commutative semiring with trivial multiplication.
If ak−1 = n−2 it is easy to see that endomorphism 1n−2(n−2)2 is the unique idempotent
1 (see [2]). Now we find 1n−2(n − 2)2 · 1n−2aras = 1n−2(ar)2, 1n−1(n − 2) · 1n−2aras =
of DN 2
1n−1ar, 1n−2aras · 1n−1(n − 2) = 1. Hence, DN 2
1 is a semiring.
Case 2. Let am = n − 2. We shall show that all the endomorphisms of DN 2
n−2 are
1 -- nilpotent with the only exception when a0 = 1. When a0 > 1 we find:
apaq(n − 2)n−2 · ap0aq0(n − 2)n−2 = n − 2,
ai(n − 2)n−1 · apaq(n − 2)n−2 = apaq(n − 2)n−2 · ai(n − 2)n−1 = n − 2,
GEOMETRICAL STRUCTURES OF THE ENDOMORPHISM SEMIRING
7
n − 2 · apaq(n − 2)n−2 = apaq(n − 2)n−2 · n − 2 = n − 2.
If a0 = 1 the only idempotent is 12(n − 2)n−2 and we find:
12(n − 2)n−2 · apaq(n − 2)n−2 = (aq)2(n − 2)n−2,
1(n − 2)n−1 · apaq(n − 2)n−2 = aq(n − 2)n−1, apaq(n − 2)n−2 · 1(n − 2)n−1 = n − 2,
Hence, DN 2
n−2 is a semiring.
Case 3. Let 1 < a0 and ak−1 < n − 2. We find the following trivial equalities, which
are grouped by duality:
apaq(am)n−2 · ap0aq0(am)n−2 = am, (am)n−2aras · (am)n−2ar0as0 = am,
apaq(am)n−2 · ap0(am)n−2as0 = ap0(am)n−2as0 · apaq(am)n−2 = am,
(am)n−2aras · ap0(am)n−2as0 = ap0(am)n−2as0 · (am)n−2aras = am,
apaq(am)n−2 · (am)n−2aras = (am)n−2aras · apaq(am)n−2 = am,
ai(am)n−1 · apaq(am)n−2 = apaq(am)n−2 · ai(am)n−1 = am,
(am)n−1aj · apaq(am)n−2 = apaq(am)n−2 · (am)n−1aj = am,
ai(am)n−1 · ap(am)n−2as = ap(am)n−2as · ai(am)n−1 = am,
(am)n−1aj · ap(am)n−2as = ap(am)n−2as · (am)n−1aj = am,
ai(am)n−1 · (am)n−2aras = (am)n−2aras · ai(am)n−1 = am,
(am)n−1aj · (am)n−2aras = a(am)n−2aras · (am)n−1aj = am,
am · apaq(am)n−2 = apaq(am)n−2 · am = am,
am · ap(am)n−2as = ap(am)n−2as · am = am,
am · (am)n−2aras = (am)n−2aras · am = am.
Case 4. Let a0 = 1 and ak−1 < n−2. Then 12(am)n−2 is the only idempotent in DN 2
m.
Additionally to the equalities of the previous case we find:
12(am)n−2 · apaq(n − 2)n−2 = (aq)2(am)n−2, 1(am)n−1 · apaq(am)n−2 = aq(am)n−1.
Case 5. Let 1 < a0 and ak−1 = n − 2. Now the only idempotent endomorphism in
m is (am)n−2(n − 2)2. We additionally find the following equalities:
DN 2
(am)n−2(n − 2)2 · (am)n−2aras = (am)n−2(ar)2, (am)n−1(n − 2) · (am)n−2aras = (am)n−1ar.
Case 6. Let a0 = 1 and ak−1 = n − 2. Now, in DN 2
m, there are two idempotents:
12(am)n−2 and (am)n−2(n − 2)2. Here the equalities from cases 4 and 5 are valid and also
all the equalities from case 3, under the respective restrictions for the indices, are fulfilled.
(cid:3)
Hence, DN 2
Theorem 5. Let σ(n)
m is a semiring.
k (A) = σ(n){a0, a1, . . . , ak−1} be a simplex.
k (A) ∩ E (a0)
Cn .
= σ(n)
a. For the least vertex a0 it follows DN n−a0−1
b. For the biggest vertex ak−1 it follows DN ak−1
0
k−1 = σ(n)
k (A) ∩ E (ak−1)
Cn
.
8
IVAN TRENDAFILOV
a0
Proof. a. Since a0
is the least vertex of the simplex,
La0+1
where a0 + 1 + p1 + · · · + pk−1 = n, i.e. α(0) = a0, . . ., α(a0) = a0. All the layers
Lℓ
fixed point. So, DN n−a0−1
Conversely, let α ∈ σ(n)
(cid:0)σ(n){a0, a1, . . . , ak−1}(cid:1) consists of endomorphisms α = (a0)a0+1(a1)p1 . . . (ak−1)pk−1,
a0(cid:0)σ(n){a0, a1, . . . , ak−1}(cid:1), where ℓ ≥ a0 + 1, consist of endomorphisms having a0 as a
a0(cid:0)σ(n){a0, a1, . . . , ak−1}(cid:1),
simplex, we have α(0) = . . . = α(a0 − 1) = a0, that is α ∈ Lℓ
where ℓ ≥ a0 + 1. Hence, DN n−a0−1
Cn . Then α(a0) = a0. Since a0 is the least vertex of the
k (A) ∩ E (a0)
Cn .
k (A) ∩ E (a0)
k (A) ∩ E (a0)
Cn .
⊆ σ(n)
it follows that layer
0
it
ak−1
ak−1
vertex
Since
layer
is
Ln−ak−1
simplex,
of
of
the
consists
(cid:0)σ(n){a0, a1, . . . , ak−1}(cid:1)
b.
follows
endomorphisms
that
α = (a0)p0 . . . (ak−2)pk−2(ak−1)n−ak−1, where p0 + · · · + pk−2 + n − ak−1 = n. So,
p0 + · · · + pk−2 = ak−1 implies that the images of 0, . . ., ak−1 − 1 are not equal to ak−1,
but α(ak−1) = ak−1. For all the endomorphisms of layers Lℓ
where ℓ ≥ n − ak−1, we have p0 + · · · + pk−2 = ak−1. Hence, the elements of these layers
have ak−1 as a fixed point and DN ak−1
k (A) ∩ E (ak−1)
. Then α(ak−1) = ak−1. Since ak−1 is the
biggest vertex of the simplex, we have α(ak−1 + 1) = . . . = α(n − 1) = ak−1. Thus,
k (A)∩E (ak−1)
α ∈ Lℓ
.
(cid:3)
ak−1(cid:0)σ(n){a0, a1, . . . , ak−1}(cid:1), where ℓ ≥ n−ak−1. Hence, DN ak−1
ak−1(cid:0)σ(n){a0, a1, . . . , ak−1}(cid:1),
k (A) ∩ E (ak−1)
let α ∈ σ(n)
k−1 = σ(n)
Conversely,
⊆ σ(n)
Cn
Cn
Cn
.
0
= σ(n)
biggest
0
the
Remark 6. What is the least ℓ, such that the discrete neighborhood DN ℓ
m of the
vertex am of simplex σ(n){a0, a1, . . . , ak−1}, where m 6= 0 and m 6= k − 1, is a semiring?
Since 132n−2 is an 1 -- nilpotent element of any simplex σ(n){1, 2, a2 . . . , ak−1}, it follows
that ℓ = 2.
From the last theorem it follows that all the a0 -- nilpotent elements of the simplex
. But there are elements of DN n−a0−1
σ(n){a0, a1, . . . , ak−1} are from semiring DN n−a0−1
which are not a0 -- nilpotent. For instance, endomorphism α ∈ La0+1
such that α(i) = am, where m = 1, . . . , k − 1, for any i > a0 is an idempotent. In order
to separate a0 -- nilpotent elements from all the other elements of DN n−a0−1
, we consider
the following
(cid:0)σ(n){a0, a1, . . . , ak−1}(cid:1)
a0
0
0
0
Proposition 7. The endomorphism α ∈ DN n−a0−1
0
is a0 -- nilpotent if
α(0) = · · · = α(a0) = · · · = α(a1) = a0, α(i) < i, for a1 < i ≤ n − 1.
Proof. Let us suppose that for some i ≥ a0+1 follows α(i) ≥ i. Then αm(i) ≥ i ≥ a0 +1
for any natural m, which contradicts that α is a0 -- nilpotent endomorphism. Hence, α(i) < i
for i ≥ a0 + 1. In particular α(as) < as, for any s = 1, . . . , k and then α(a1) = a0.
(cid:3)
that
it
in
follows
From the
proposition
immediately
layer
last
endomorphisms
(cid:0)σ(n){a0, a1, . . . , ak−1}(cid:1).
elements of
is a proper
subsemiring of
(cid:0)σ(n){a0, a1, . . . , ak−1}(cid:1)
a0 -- nilpotent
DN n−a0−1
0
La0+1
a0
the
α = (a0)a0+1(a1)p1 . . . (ak−1)pk−1 ∈ La0+1
that αm = (a0)a0+1(a1)q1 . . . (ak−1)qk−1 ∈ La0+1
So, all the elements of this layer are idempotents or roots of idempotents of the same
layer. Obviously, the layer is closed under the addition. So, we prove
(cid:0)σ(n){a0, a1, . . . , ak−1}(cid:1), it follows by induction
(cid:0)σ(n){a0, a1, . . . , ak−1}(cid:1) for any natural m.
follows
am -- nilpotent
layer are idempotents or
La0+1
a0
simplex,
any
that
elements.
idempotents. But
not
Since
in layer
So,
if
roots of
the
not
there
there
this
are
are
it
a0
a0
GEOMETRICAL STRUCTURES OF THE ENDOMORPHISM SEMIRING
9
Proposition 8. For any simplex σ(n){a0, a1, . . . , ak−1} layer
(cid:0)σ(n){a0, a1, . . . , ak−1}(cid:1) is a subsemiring of the simplex.
La0+1
a0
3. Strings
Let us denote the elements of semiring ST R(n){a, b} by akbn−k, where k = 0, . . . , n is
the number of the elements of Cn with an image equal to a, i.e.
akbn−k = ≀ a, . . . , a
≀.
{z }
k
, b, . . . , b
{z }
n−k
In particular, we denote anb0 = a and a0bn = b.
Let us consider the following subset of ST R(n){a, b}:
N [a](cid:16)ST R(n){a, b}(cid:17) = {a, . . . , ab+1bn−b−1}.
For any endomorphism α ∈ N [a](cid:16)ST R(n){a, b}(cid:17), that is α = aℓ bn−ℓ, where b+ 1 ≤ ℓ ≤ n,
we have α(b) = a. Hence α, β ∈ N [a](cid:16)ST R(n){a, b}(cid:17) implies α · β = a. Since α2 = a for
any α ∈ N [a](cid:16)ST R(n){a, b}(cid:17), it follows, see [3], that
N [a](cid:16)ST R(n){a, b}(cid:17) = N [a]
n ∩ ST R(n){a, b}.
From Theorem 3.3 of [3], see section 1, follows
Proposition 9. The set N [a](cid:16)ST R(n){a, b}(cid:17) is a subsemiring of ST R(n){a, b}
consisting of all a -- nilpotent elements of this string.
The order of this semiring is n − b.
The next subset of ST R(n){a, b} is:
Id(cid:16)ST R(n){a, b}(cid:17) = {abbn−b, . . . , aa+1bn−a−1}.
For any endomorphism of Id(cid:16)ST R(n){a, b}(cid:17) elements a and b are fixed points. From
Corollary 3 of [2] we find that all the elements of Id(cid:16)ST R(n){a, b}(cid:17) are idempotents and
Proposition 10. The set Id(cid:16)ST R(n){a, b}(cid:17) is a subsemiring of ST R(n){a, b}
from Theorem 9 of [2], see section 1, it follows
consisting of all idempotent elements of this string different from a and b. The order
of this semiring is b − a.
The last considered subset of ST R(n){a, b} is:
N [b](cid:16)ST R(n){a, b}(cid:17) = {aabn−a, . . . , b}.
10
IVAN TRENDAFILOV
For any endomorphism α = aℓ bn−ℓ, where 0 ≤ ℓ ≤ a it follows α(a) = b. Hence α, β ∈
N [b](cid:16)ST R(n){a, b}(cid:17) implies α · β = b. Since α2 = b for any α ∈ N [b](cid:16)ST R(n){a, b}(cid:17) it
follows, see [2], that
N [b](cid:16)ST R(n){a, b}(cid:17) = N [b]
n ∩ ST R(n){a, b}.
From Theorem 3.3 of [3], see section 1, we have
Proposition 11. The set N [b](cid:16)ST R(n){a, b}(cid:17) is a subsemiring of ST R(n){a, b}
consisting of all b -- nilpotent elements of this string.
The order of this semiring is a + 1.
Proposition 12. Let a, b ∈ Cn, a < b and akbn−k ∈ ST R(n){a, b}, where k = 0, . . . , n.
Then
akbn−k · α = a,
akbn−k · α = akbn−k,
if α ∈ N [a](cid:16)ST R(n){a, b}(cid:17)
if α ∈ Id(cid:16)ST R(n){a, b}(cid:17)
if α ∈ N [b](cid:16)ST R(n){a, b}(cid:17)
Proof. For any i ∈ Cn and α ∈ N [a](cid:16)ST R(n){a, b}(cid:17) it follows
akbn−k · α = b,
.
(akbn−k · α)(i) = α(akbn−k(i)) =(cid:26) α(a),
α(b),
if 0 ≤ i ≤ k
if k + 1 ≤ i ≤ n − 1
= a
which means that akbn−k · α = a.
For any i ∈ Cn and α ∈ Id(cid:16)ST R(n){a, b}(cid:17) it follows
(akbn−k · α)(i) = α(akbn−k(i)) =(cid:26) α(a),
(cid:26) a,
α(b),
b,
if 0 ≤ i ≤ k
if k + 1 ≤ i ≤ n − 1
=
if 0 ≤ i ≤ k
if k + 1 ≤ i ≤ n − 1
= akbn−k(i)
which means that akbn−k · α = akbn−k.
For any i ∈ Cn and α ∈ N [b](cid:16)ST R(n){a, b}(cid:17) it follows
(akbn−k · α)(i) = α(akbn−k(i)) =(cid:26) α(a),
α(b),
which means that akbn−k · α = b.
Immediately follows
if 0 ≤ i ≤ k
if k + 1 ≤ i ≤ n − 1
= b
(cid:3)
Corollary 13. The idempotent endomorphisms of semiring ST R(n){a, b}, different
from a and b, are right identities.
Corollary 14. Any two different strings are nonisomorphic semirings.
GEOMETRICAL STRUCTURES OF THE ENDOMORPHISM SEMIRING
11
Using the fact that the strings are faces of any k -- simplex for arbitrary k ≥ 2, the
last corollary implies
Corollary 15. Any two different k -- simplices are nonisomorphic semirings.
Remark 16. a. From Propopsition 1.4 we actually observe that the multiplicative
structure of arbitrary string ST R(n){a, b} is very clear: first, we find n−b endomorphisms
(all the a -- nilpotent elements) which are square roots of a or a, then b − a idempotents,
which are right identities and, in the end, a + 1 elements (all the b -- nilpotent elements)
which are square roots of b or b.
b. The union of semirings N [a](cid:16)ST R(n){a, b}(cid:17) and Id(cid:16)ST R(n){a, b}(cid:17) is also a semi-
ring because
N [a](cid:16)ST R(n){a, b}(cid:17) ∪ Id(cid:16)ST R(n){a, b}(cid:17) = ST R(n){a, b} ∩ E (a)
Cn .
Similarly,
N [b](cid:16)ST R(n){a, b}(cid:17) ∪ Id(cid:16)ST R(n){a, b}(cid:17) = ST R(n){a, b} ∩ E (b)
Cn
is a subsemiring of ST R(n){a, b}.
Two strings ST R(n){a, b} and ST R(n){x, y} are called consecutive if they have
a common vertex. So, strings ST R(n){a, b} and ST R(n){b, c}, ST R(n){a, b} and
ST R(n){a, c}, ST R(n){a, c} and ST R(n){b, c} (when a < b < c) are the three possibilities
of the pairs of consecutive strings.
Let akbn−k ∈ ST R(n){a, b}, where k = 0, . . . , n, and bℓcn−ℓ ∈ ST R(n){b, c}, where
ℓ = 0, . . . , n. Since akbn−k < bℓcn−ℓ, then akbn−k + bℓcn−ℓ = bℓcn−ℓ. By similar arguments,
for any amcn−m ∈ ST R(n){a, c}, we can construct amcn−m + bℓcn−ℓ = brcn−r, where
r = min{ℓ, m}. But when we add endomorphisms akbn−k and amcn−m, where k < m, the
sum is akbm−kcn−m, so, the set of these three strings is not closed under the addition.
In the next proposition we examine the product of endomorphisms of two (not
necessarily consecutive) strings.
Proposition 17. Let akbn−k ∈ ST R(n){a, b}, where k = 0, . . . , n, and xℓ yn−ℓ ∈
ST R(n){x, y}, where ℓ = 0, . . . , n. Then
akbn−k · xℓ yn−ℓ = x,
akbn−k · xℓ yn−ℓ = xkyn−k,
akbn−k · xℓ yn−ℓ = y,
if b + 1 ≤ ℓ ≤ n
if a + 1 ≤ ℓ ≤ b
if 0 ≤ ℓ ≤ a
.
Proof. For any i ∈ Cn it follows
(akbn−k · xℓ yn−ℓ) (i) = xℓ yn−ℓ (akbn−k) (i) =(cid:26) xℓ yn−ℓ(a),
xℓ yn−ℓ(b),
if 0 ≤ i ≤ k
if k + 1 ≤ i ≤ n − 1
.
If b + 1 ≤ ℓ ≤ n, we have xℓ yn−ℓ(b) = x and then xℓ yn−ℓ(a) = x. So, for any i ∈ Cn
we find that (akbn−k · xℓ yn−ℓ) (i) = x and hence akbn−k · xℓ yn−ℓ = x.
12
IVAN TRENDAFILOV
If a + 1 ≤ ℓ ≤ b, it follows xℓ yn−ℓ(a) = x and xℓ yn−ℓ(b) = y. Thus, we obtain
(akbn−k · xℓ yn−ℓ) (i) =(cid:26) x,
y,
if 0 ≤ i ≤ k
if k + 1 ≤ i ≤ n − 1
= xkyn−k(i).
Hence, akbn−k ·xℓ yn−ℓ = xkyn−k. If 0 ≤ ℓ ≤ a, we have xℓ yn−ℓ(a) = y and then xℓ yn−ℓ(b) =
y. So, for any i ∈ Cn we obtain (akbn−k · xℓ yn−ℓ) (i) = y and hence akbn−k · xℓ yn−ℓ = y. (cid:3)
Immediately it follows
Corollary 18. For any a, b, c ∈ Cn, a < b < c, the set consisting of all the elements of
the consecutive strings ST R(n){a, b} and ST R(n){b, c} is a semiring.
A subsemiring S of endomorphism semiring bECn is called a trivial semiring if for any
two elements α, β ∈ S it follows α · β = ι, where ι is a fixed element of S. If semiring S
is a trivial, then there exists a unique idempotent ι ∈ S such that the product of any two
elements of S is equal to ι. If this idempotent is the biggest (least) element of the trivial
semiring S, the S is called an upper (lower) trivial semiring.
Proposition 1.4) is an upper trivial semiring.
the proof of Proposition 1.4) is a lower trivial semiring.
Example 19. a) The semiring of a -- nilpotent elements N [a](cid:16)ST R(n){a, b}(cid:17) (using
b) The semiring of b -- nilpotent elements N [b](cid:16)ST R(n){a, b}(cid:17) (using the proof of
that the union of semirings N [b](cid:16)ST R(n){a, b}(cid:17) and N [b](cid:16)ST R(n){b, c}(cid:17) is a trivial
semiring. Since b is the biggest element of N [b](cid:16)ST R(n){a, b}(cid:17) and the least element
of N [b](cid:16)ST R(n){b, c}(cid:17) it follows that the considered trivial semiring is neither upper
c) Let us consider the semiring from Corollary 2.2. Then from Proposition 2.1 follows
trivial nor lower trivial.
Now we shall construct some useful subsemirings of a given string, some of which are
trivial semirings. We consider the following subset of string ST R(n){a, b}:
Ar = {a, an−1b, . . . , arbn−r},
where r = 1, . . . , n. Since Ar is a chain for any r, it is closed under the addition. If
r ≥ b + 1, then Ar ⊆ N [a](cid:16)ST R(n){a, b}(cid:17), so, Ar is a lower trivial semiring. If r ≤ a, then
Ar∩N [b](cid:16)ST R(n){a, b}(cid:17) 6= ∅ which implies that Ar is not closed under the multiplication,
i.e. it is not a semiring. From Remark 1.7 b. it follows that
Aa+1 = N [a](cid:16)ST R(n){a, b}(cid:17) ∪ Id(cid:16)ST R(n){a, b}(cid:17)
is the biggest between sets Ar, which is a semiring. Since every element of semiring
Id(cid:16)ST R(n){a, b}(cid:17) is a right identity of string ST R(n){a, b}, it follows that for every
r = a + 1, . . . , n the set Ar is a semiring.
Using the same idea, we consider the subset of ST R(n){a, b}:
Bs = {b, abn−1, . . . , asbn−s},
GEOMETRICAL STRUCTURES OF THE ENDOMORPHISM SEMIRING
13
where s = 0, . . . , n−1. The set Bs is a chain for any s, so, it is closed under the addition. If
s ≤ a, then Bs ⊆ N [b](cid:16)ST R(n){a, b}(cid:17), so, Bs is an upper trivial semiring. If s ≥ b+1, then
Bs ∩N [a](cid:16)ST R(n){a, b}(cid:17) 6= ∅ which means that Bs is not closed under the multiplication,
so, Bs is not a semiring. Also from Remark 1.7 b. it follows that
Bb = N [b](cid:16)ST R(n){a, b}(cid:17) ∪ Id(cid:16)ST R(n){a, b}(cid:17)
is the biggest from sets Bs which is a semiring. By the same way, considering that every
element of semiring Id(cid:16)ST R(n){a, b}(cid:17) is a right identity of string ST R(n){a, b} it follows
that for every s = 0, . . . , b set Bs is a semiring.
4. Triangles
Let a, b, c ∈ Cn, a < b < c, are fixed elements. The set of endomorphisms α such that
α(0) = · · · = α(k − 1) = a, α(k) = · · · = α(k + ℓ − 1) = b, α(k + ℓ) = · · · = α(n − 1) = c
or briefly α = akbℓcn−k−ℓ, where 0 ≤ k ≤ n − 1, 0 ≤ ℓ ≤ n − 1 and 0 ≤ n − k − ℓ ≤ n − 1
2 (cid:19).
is actually the triangle △(n){a, b, c}. Obviously, the order of this semiring is (cid:18)n + 2
The strings ST R(n){a, b}, ST R(n){a, c} and ST R(n){b, c} are called strings of
△(n){a, b, c}.
Let R be a subsemiring of bECn and α, β ∈ R, α 6= β. These endomorphisms are called
right-similar (left-similar ) if for any γ ∈ R we have α · γ = β · γ (γ · α = γ · β). We denote
this by α ∼r β (α ∼ℓ β). In the next sections we shall answer the question: Are there
right-similar (left-similar) elements in △(n){a, b, c} ?
Example 20. The biggest side of the least tetrahedron T ET R(4){0, 1, 2, 3} is triangle
△(4){1, 2, 3}. The elements of this semiring can be arranged as in the following scheme
(fig.1):
3
13
3
233
13
2 2
123
2
23
2 2
13
3
123
2
123
2
23
3
1
12
3
12
22
12
3
2
Figure 1.
It is easy to see that the interior of this triangle is set Int = {1223, 1223, 1232}. Since
(1232)2 = 233 /∈ Int, it follows that the interior of △(4){1, 2, 3} is not a semiring. Since
14
IVAN TRENDAFILOV
1223 = 1222 + 133, 1223 = 123 + 133 and 1232 = 123 + 1232, it follows that every element
of the interior of the triangle can be represented as a sum of an element of the least side
of the triangle and an element of the middle side of the triangle.
In this triangle there are many left-similar endomorphisms: 123 ∼ℓ 2, 133 ∼ℓ 233 ∼ℓ 3,
1232 ∼ℓ 2232, 1223 ∼ℓ 233. The endomorphism 1223 is a right-identity, so there are not
right-similar endomorphisms in △(4){1, 2, 3}.
Proposition 21. Any element of the interior of △(n){a, b, c} can be uniquely
represented as a sum of the elements of strings ST R(n){a, b} and ST R(n){a, c}.
Proof. We easily calculate
an−1c + akbn−k = akbn−k−1c, where k = 0, . . . , n − 2.
By the same argument, for any j = 1, . . . , n − 1 we find
an−jcj + akbn−k = akbn−k−jcj, where k = 0, . . . , n − j − 1.
So, we prove more: all the elements of the interior of △(n){a, b, c} and all the elements of
the interior of ST R(n){b, c} are sums of the elements of ST R(n){a, b} and ST R(n){a, c}.
From the construction we observe that endomorphisms a and c do not occur in these sums
and every representation of this type is unique.
(cid:3)
Corollary 22. The boundary of an arbitrary triangle △(n){a, b, c} is a multiplicative
semigroup but not a semiring.
Proof. Since the boundary of △(n){a, b, c} is a union of strings ST R(n){a, b},
ST R(n){a, c} and ST R(n){b, c},
from Proposition 8 it follows that this set is a
multiplicative semigroup. From the last proposition, it follows that the boundary of
△(n){a, b, c} is not closed under the addition.
(cid:3)
Corollary 23. The interior of an arbitrary triangle △(n){a, b, c}, where n ≥ 4, is an
additive semigroup but not a semiring.
Proof. From the last proposition and the fact that ST R(n){a, b} and ST R(n){a, c} are
semirings, it follows that the interior of the triangle is closed under the addition. If a > 0,
it follows (abbcn−b−1)2 = bb+1cn−b−1. If a = 0 follows (0bb−1cn−b)2 = 0cn−1. So, in all the
cases the interior of the triangle is not a semiring.
(cid:3)
When n = 3, the interior of the least triangle △(3){0, 1, 2} is one-element semiring and
this element is an identity i = 123.
Let a, b, c, x, y, z ∈ Cn, a < b < c and x < y < z. We consider the map
Φ : △(n){a, b, c} → △(n){x, y, z}
such that Φ (akbℓcn−k−ℓ) = xkyℓzn−k−ℓ, where 0 ≤ k ≤ n − 1, 0 ≤ k ≤ n − 1 and
0 ≤ n − k − ℓ ≤ n − 1. Obviously, Φ is order-preserving. Hence, the additive semigroups
of any two triangles △(n){a, b, c} and △(n){x, y, z} are isomorphic. But △(n){a, b, c} and
△(n){x, y, z} are nonisomorphic semirings.
Proposition 24. In arbitrary triangle △(n){a, b, c}, where n > 3, there is at least one
right identity and there are not any left identities.
GEOMETRICAL STRUCTURES OF THE ENDOMORPHISM SEMIRING
15
Proof. The least idempotent of ST R(n){a, b} is endomorphism abbn−b and the least
idempotent of ST R(n){a, c} is endomorphism accn−c. Their sum is ε = abbc−bcn−c. Let
α ∈ ST R(n){a, b}. If α(a) = a, or α(a) = b, or α(a) = c, then it follows (α · ε)(a) = a, or
(α · ε)(a) = b, or (α · ε)(a) = a, respectively. The same is valid if we replace a with b, or
a with c. So, endomorphism ε = abbc−bcn−c is a right identity of triangle △(n){a, b, c}.
If b > a + 1, endomorphism aa+1bn−a−1 is another (in all the cases, the biggest)
idempotent of ST R(n){a, b}. Now, in a similar way as above, it is easy to check that sum
aa+1bn−a−1 + accn−c = aa+1bc−a−1cn−c is another right identity of △(n){a, b, c}.
Let us, similarly, suppose that c > b + 1. Then endomorphism ab+1cn−b−1 is another,
different from accn−c, idempotent of ST R(n){a, c}. Now sum abbn−b + ab+1cn−b−1 =
abbcn−b−1 is a right identity of △(n){a, b, c}.
If there are two right identities of the triangle, it implies that there is not a left
identity in this semiring. So, we consider the case when there is only one idempotent
of ST R(n){a, b} and there is only one idempotent of ST R(n){a, c}. It is possible when
b = a + 1 and c = a + 2. Thus, it is enough to prove that there is not a left identity in
any triangle △(n){a, a + 1, a + 2}. We consider two cases.
Case 1. Let a ≥ 1 and α = (a + 1)(a + 2)n−1. Then α(a) = α(a + 1) = α(a + 2) = a + 2.
Hence, for any β ∈ △(n){a, a + 1, a + 2} we find β · α = a + 2. Since β · a + 2 = a + 2 it
follows that α ∼ℓ a + 2. So, there are two left-similar elements of △(n){a, a + 1, a + 2}
and hence there is not a left identity.
Case 2. Let a = 0. In semiring △(n){0, 1, 2} we choose endomorphism α = 1n−12. Then
for any β ∈ △(n){0, 1, 2} it follows β · α = 1. Since β · 1 = 1, we show that α ∼ℓ 1. So,
there are two left-similar elements of △(n){0, 1, 2} and there is not a left identity in this
triangle.
(cid:3)
Since there is a right identity in semiring △(n){a, b, c}, it follows
Corollary 25. In an arbitrary triangle △(n){a, b, c}, where n > 3, there are not right-
similar endomorphisms.
Immediately it follows
Corollary 26. Only in the least triangle △(3){0, 1, 2} there is an identity i = 123.
5. Layers in a Triangle
Since any triangle △(n){a, b, c} is a 2 -- simplex, we define layers as in section 2. Any layer
of a triangle is a chain. So, the elements of layer Lk
are the following n − k + 1 endomorphisms:
a(cid:0)△(n){a, b, c}(cid:1), where k = 0, . . . , n − 1,
akbn−k < akbn−k−1c < · · · < akcn−k.
Similarly, we can represent the elements of Lk
where k = 0, . . . , n − 1.
When the least and the biggest element of the layer are idempotents, we call this
layer a basic layer . Let us consider layer Lk
idempotents of ST R(n){a, b} are endomorphisms abbn−b, . . . , aa+1bn−a−1 and, similarly,
idempotents of ST R(n){a, c} are accn−c, . . . , aa+1cn−a−1, it follows that Lk
b(cid:0)△(n){a, b, c}(cid:1) and Lk
c(cid:0)△(n){a, b, c}(cid:1),
a(cid:0)△(n){a, b, c}(cid:1) with respect to vertex a.. Since
a(cid:0)△(n){a, b, c}(cid:1)
16
IVAN TRENDAFILOV
is a basic layer only if k = a + 1, . . . , b. All the elements of the layer are endomorphisms
α = akbn−k−ici, where i = 0, . . . , n − k. It is easy to see that if i ≤ n − c − 1, then
α(b) = α(c) = b. Such endomorphisms are called left elements of the layer. If n − c ≤ i ≤
n − b − 1, then α(b) = b and α(c) = c, so, α is an idempotent which is a right identity
of the triangle. If i ≥ n − b, then α(b) = α(c) = c. Such endomorphisms are called right
elements of the layer. Let α and β be left elements of the layer. If α(x) = b, where x ∈ Cn,
then (α · β)(x) = β(b) = b. If α(x) = c, where x ∈ Cn, then (α · β)(x) = β(c) = b. Hence,
α · β = akbn−k. Similarly, if α and β are right elements of the layer, then α · β = akcn−k.
Let α be a left element and β be a right element of the layer. If α(x) = b, where x ∈ Cn,
then (α · β)(x) = β(b) = c. If α(x) = c, where x ∈ Cn, then (α · β)(x) = β(c) = c. Hence,
α · β = akcn−k. Similarly, β · α = akbn−k. Thus we obtain
Proposition 27. Any basic layer Lk
a(cid:0)△(n){a, b, c}(cid:1), k = a + 1, . . . , b, is a semiring.
a(cid:0)△(n){a, b, c}(cid:1) the first
From the proof of last proposition it follows that in layer Lk
n − c elements α are left elements, i.e. α2 = akbn−k, the next c − b endomorphisms are
idempotents and the last b − k + 1 elements α are right elements, i.e. α2 = akcn−k. If
there is a string ST R(n0){a0, b0} with n − c a0 -- nilpotent elements, c − b idempotents and
b − k + 1 b0 -- nilpotent elements are b − k + 1, then two semirings Lk
a(cid:0)△(n){a, b, c}(cid:1) and
we find n0 = n−k,
a0 = b − k and b0 = c − k. So, we prove
ST R(n0){a0, b0} will be isomorphic. From system(cid:12)(cid:12)(cid:12)(cid:12)(cid:12)(cid:12)
a(cid:0)△(n){a, b, c}(cid:1) and ST R(n−k){b − k, c − k} are isomorphic.
Lk
n0 − b0 = n − c
b0 − a0 = c − b
a0 + 1 = b − k + 1
Proposition 28. For any n ≥ 3, a, b ∈ Cn, a < b and k = a + 1, . . . , b, semirings
layer Lk
The basic layers Lk
Now let us consider
case. For instance, see fig,1, where the layer L2
middle element 1223 we find (1223)2 = 233 /∈ L2
b(cid:0)△(n){a, b, c}(cid:1) are not closed under the multiplication in the general
2(cid:0)△(4){1, 2, 3}(cid:1) is a basic layer, but for his
2(cid:0)△(4){1, 2, 3}(cid:1).
c(cid:0)△(n){a, b, c}(cid:1) with respect
c(cid:0)△(n){a, b, c}(cid:1) is a basic layer only if
to vertex c. Since
idempotents of ST R(n){a, c} are accn−c, . . . , aa+1cn−a−1 and idempotents of ST R(n){b, c}
are bccn−c, . . . , bb+1cn−b−1,
k = n − c, . . . , n − b − 1. All the elements of the layer are endomorphisms α = aibn−k−ick,
where i = 0, . . . , n − k. If b + 1 ≤ i ≤ n − k, then α(a) = α(b) = a. We call these
endomorphisms (as in the previous case) left elements of the layer. When a + 1 ≤ i ≤ b,
it follows α(a) = a, α(b) = b, that is α is an idempotent which is a right identity of the
triangle. If 0 ≤ i ≤ a, then α(a) = α(b) = b. These endomorphisms are the right elements
of the layer. By the same way, as for the basic layers with respect to vertex a, we prove
here that:
it follows that Lk
1. If α and β are left elements of the layer, then α · β = an−kck.
2. If α and β are right elements of the layer, then α · β = bn−kck.
3. If α is a left but β is a right element of the layer, then α·β = bn−kck and β·α = an−kck
So, we obtain
Proposition 29. Any basic layer Lk
a semiring.
c(cid:0)△(n){a, b, c}(cid:1), where k = n − c, . . . , n − b − 1, is
GEOMETRICAL STRUCTURES OF THE ENDOMORPHISM SEMIRING
17
Now we search a string ST R(n0){a0, b0} with n − k − b ao -- nilpotent elements, b − a
idempotents and a + 1 b0 -- nilpotent elements. It is easy to find that n0 = n − k, a0 = a
and b0 = b. Thus we prove
Proposition 30. For any n ≥ 3, b, c ∈ Cn, b < c and k = n − c, . . . , n − b − 1, semirings
Lk
c(cid:0)△(n){a, b, c}(cid:1) and ST R(n−k){a, b} are isomorphic.
0(cid:0)△(n){0, b, c}(cid:1). The
In triangle △(4){1, 2, 3} (fig. 1) we obtain that
there are many left-similar
endomorphisms. In order to prove that there are such elements in any triangle △(n){a, b, c}
we first consider the case a > 0. Then acn−1 ∼ℓ c. Actually, if α = acn−1, easily follows
that α(a) = α(b) = α(c) = c. Thus, for any β ∈ △(n){a, b, c} we have β · α = c = β · c,
that is α ∼ℓ c. Note that α and c are not right identities of the triangle.
Let a = 0. In △(n){0, b, c} we consider the biggest basic layer L1
elements of this layer are 0bn−1 < · · · < 0bcn−2 < 0cn−1. Now, if b > 1, it follows that
0bcn−2 ∼ℓ 0cn−1. Indeed, let α = 0bcn−1. Since α(0) = 0, α(b) = α(c) = c, it follows that
if β ∈ △(n){0, b, c}, then β · α = β · 0cn−1, i.e. α ∼ℓ 0cn−1. Note that α and 0cn−1 are not
right identities of the triangle.
Now let us consider triangle △(n){0, 1, c} and endomorphisms 01n−1 and 01n−2. Let
c < n − 1. Then for α = 01n−2c we have α(0) = 0, α(b) = α(c) = 1. So, for any
β ∈ △(n){0, 1, c}, it follows β · α = β · 01n−1, that is 01n−1c ∼ℓ 01n−1. Note that these
endomorphisms are not right identities.
Finally let c = n − 1. We consider △(n){0, 1, n − 1}, where n > 3. Note, that in
the least triangle △(3){0, 1, 2} there are not left-similar elements since there is an identity.
Now let us consider endomorphisms α = 0n−11 and β = 0n−212. We have α(0) = β(0) = 0,
α(1) = β(1) = 0, α(n − 1) = β(n − 1) = 1. So, for any γ ∈ △(n){0, 1, n − 1}, it follows
γ · α = γ · β, i.e. α ∼ℓ β. Note that α and β are not right identities. Hence, we prove
Proposition 31. For any n > 3 the triangle △(n){a, b, c} contains a pair of elements
which are left-similar endomorphisms and are not right identities.
6. Idempotents and Nilpotent Elements of a Triangle
By a boundary idempotent of triangle △(n){a, b, c} we understand an idempotent
of any of the strings of this triangle. The idempotent of the interior of the triangle is
called interior idempotent. From Proposition 17, it follows that the set of all boundary
idempotents is closed under the multiplication.
Proposition 32. The interior idempotents of the triangle △(n){a, b, c} are just the
right identities of the triangle.
Proof. As we know,
[2], endomorphism α ∈ bECn with s fixed points k1, . . . , ks,
1 ≤ s ≤ n − 1, is an idempotent if and only if Im(α) = {k1, . . . , ks}. So, for any interior
idempotent of △(n){a, b, c}, it follows that a, b and c are fixed points of α. Then, for any
β ∈ △(n){a, b, c} and x ∈ Cn easily follows (β · α)(x) = α(β(x)) = β(x). Hence, α is a
right identity of △(n){a, b, c}.
Conversely, if α is a right identity, then, obviously, α is an idempotent. If we assume
that α is a boundary idempotent, say α = akbn−k, where k = a+1, . . . , b, then c·akbn−k = b.
This contradicts the choice that α is a right identity.
(cid:3)
18
IVAN TRENDAFILOV
For any triangle △(n){a, b, c} endomorphism aa+1cn−a−1
The set of right identities of △(n){a, b, c} is denoted by RI(cid:0)△(n){a, b, c}(cid:1).
it follows that aa+1cn−a−1 + bb+1cn−b−1 = ba+1cn−a−1 ∈ N [a](cid:16)ST R(n){a, b}(cid:17) since
is an idempotent of
ST R(n){a, c} and endomorphism bb+1cn−b−1 is an idempotent of ST R(n){b, c}. Now,
(ba+1cn−a−1)2 = c. So, the set of idempotents of any triangle is not a semiring. But
this does not mean that adding idempotents is a "bad idea".
Any discrete neighborhood of a vertex of the triangle △(n){a, b, c} (see fig. 2) can be
represented as a triangle in a geometrical sense. So, such subset of △(n){a, b, c} is called
geometric triangle. In fig. 2 the endomorphisms an−mcm, bn−mcm and c are the "vertices"
and subsets of ST R(n){a, c} and ST R(n){b, c} consisting of n − m + 1 endomorphisms
and also the layer Ln−m
geometric triangles are called subsets depicted in figures 3 and 4.
(cid:0)△(n){a, b, c}(cid:1) are the "sides" of this geometric triangle. Similarly,
c
c
a b c
k+m n-k m
a b cn-k-
i k
i
a b c
k
i+m n-k-i-m
a cn-m m
b cn-m m
a bn-k
k
a bk-m n-k-m
a b c
n-k-i-m k+m
i
Figure 2
Figure 3
Figure 4
Geometric triangles are, in general, not semirings. But some of them are semirings,
b = {b, abn−1, bn−1c}
a = {a, an−1b, an−1c}, DN n−1
for example, geometric triangles DN n−1
and DN n−1
c = {c, acn−1, bcn−1} are subsemirings of △(n){a, b, c} from Proposition 2.
a
a
(cid:0)△(n){a, b, c}(cid:1) is semiring DN n−a−1
From Theorem 5 it follows that the discrete neighborhood of vertex a of △(n){a, b, c}
= △(n){a, b, c}∩E (a)
containing the biggest layer La+1
Cn .
This semiring is a geometric triangle whose "vertices" the idempotent endomorphisms a,
aa+1bn−a−1 and aa+1cn−a−1. The "sides" of this geometric triangle are also semirings, since
La+1
type Aa+1 (see the end of section 3).
(cid:0)△(n){a, b, c}(cid:1) is a semiring (see Proposition 27) and other "sides" are semirings of
(cid:0)△(n){a, b, c}(cid:1) is semiring DN c
Similarly, from Theorem 5 we know that the discrete neighborhood of the vertex
c =
Cn . This semiring is also a geometric triangle whose "vertices" are the
c of △(n){a, b, c} containing the biggest layer Ln−c
△(n){a, b, c} ∩ E (c)
idempotent endomorphisms accn−c, bccn−c and c.
a
c
a
∩ DN c
The intersection DN n−a−1
c is a semiring consisting of all the endomorphisms
with fixed points a and b. Thus we construct a new geometric triangle whose "vertices"
are the idempotent endomorphisms accn−c, aa+1bc−a−1cn−c and aa+1cn−a−1. This semiring
Theorem 33. For any triangle △(n){a, b, c}, n ≥ 3, the set of right identities
is called an idempotent triangle of △(n){a, b, c} and is denoted by IT (cid:0)△(n){a, b, c}(cid:1).
RI(cid:0)△(n){a, b, c}(cid:1) is a subsemiring of IT (cid:0)△(n){a, b, c}(cid:1) of order (b − a)(c − a).
The set IT (cid:0)△(n){a, b, c}(cid:1) \RI(cid:0)△(n){a, b, c}(cid:1) is a subsemiring of IT (cid:0)△(n){a, b, c}(cid:1)
2((c − b)2 + (b − a)2 + c − a). The semirings Id(cid:16)ST R(n){a, c}(cid:17) and
IT (cid:0)△(n){a, b, c}(cid:1) \RI(cid:0)△(n){a, b, c}(cid:1) are ideals of IT (cid:0)△(n){a, b, c}(cid:1).
Proof. Let α = akbn−k−jcj be a right identity. From the last proposition it follows that
of order
1
GEOMETRICAL STRUCTURES OF THE ENDOMORPHISM SEMIRING
19
a, b and c are the fixed points of α. Since α(a) = a, it follows k ≥ a + 1. Since α(c) = c, we
have j ≥ n − c. Finally, since α(b) = b, it follows a + 1 ≤ k ≤ b, n − c ≤ j ≤ n − a − 1 and
n − k − j ≥ 1. Hence, akbn−k−jcj = akbn−k + an−jcj, where akbn−k ∈ Id(cid:16)ST R(n){a, b}(cid:17)
and an−jcj ∈ Id(cid:16)ST R(n){a, c}(cid:17). Actually we showed that
akbn−k−jcj = Lk
a(cid:0)△(n){a, b, c}(cid:1) ∩ Lℓ
c(cid:0)△(n){a, b, c}(cid:1) ,
the multiplication.
Let α ∈ Lk1
where k = a + 1, . . . , b and ℓ = n − c, . . . , n − b − 1. Thus we prove that any right identity
is an intersection of basic layer with respect to a and basic layer with respect to c. Since
there are b − a basic layers with respect to a and c − b basic layers with respect to c, it
prove in Proposition 29 that all the left elements from some basic layer with respect to
a (cid:0)△(n){a, b, c}(cid:1), β ∈ Lk2
a (cid:0)△(n){a, b, c}(cid:1) and α, β ∈ Lℓ
a (cid:0)△(n){a, b, c}(cid:1) ∩ Lℓ2
a (cid:0)△(n){a, b, c}(cid:1) ∩ Lℓ2
where k = a+1, . . . , b and ℓ = n−c, . . . , n−b−1. If we assume that k1 ≤ k2, then α+β = α.
Similarly, if α and β are endomorphisms of the same layer with respect to a and α ∈
Lℓ1
Lk1
where k1 ≤ k2 and ℓ1 ≤ ℓ2. Then we take γ = Lk1
follows that all right identities are (b − a)(c − b). Since all elements of RI(cid:0)△(n){a, b, c}(cid:1)
are right identities of △(n){a, b, c}, it follows that the set RI(cid:0)△(n){a, b, c}(cid:1) is closed under
c(cid:0)△(n){a, b, c}(cid:1), where
c (cid:0)△(n){a, b, c}(cid:1), where ℓ1 ≥ ℓ2, then α + β = α. Finally, let α =
c (cid:0)△(n){a, b, c}(cid:1), β ∈ Lℓ2
c (cid:0)△(n){a, b, c}(cid:1),
a (cid:0)△(n){a, b, c}(cid:1) ∩ Lℓ1
c (cid:0)△(n){a, b, c}(cid:1) and β = Lk2
c (cid:0)△(n){a, b, c}(cid:1)
and α + β = γ. Hence, RI(cid:0)△(n){a, b, c}(cid:1) is a subsemiring of IT (cid:0)△(n){a, b, c}(cid:1) of order
Since IT (cid:0)△(n){a, b, c}(cid:1) ⊂ DN n−a−1
and c -- nilpotent endomorphisms of IT (cid:0)△(n){a, b, c}(cid:1). Since IT (cid:0)△(n){a, b, c}(cid:1) ⊂ DN c
follows that there are not any a -- nilpotent endomorphisms of IT (cid:0)△(n){a, b, c}(cid:1). But we
c are square roots of endomorphisms of Id(cid:16)ST R(n){a, c}(cid:17). So, the idempotent triangle
consist of all the right identities, all the elements of Id(cid:16)ST R(n){a, c}(cid:17) and all the square
roots of elements of Id(cid:16)ST R(n){a, c}(cid:17). The vertices of the idempotent triangle are accn−c,
2(c − a)(c − a + 1) endomorphisms of IT (cid:0)△(n){a, b, c}(cid:1). Thus
it follows that the elements of set IT (cid:0)△(n){a, b, c}(cid:1) \RI(cid:0)△(n){a, b, c}(cid:1) are
Now we consider a partition of idempotent triangle IT (cid:0)△(n){a, b, c}(cid:1) into three parts.
aa+1bc−a−1cn−c and aa+1cn−a−1. So, any "side" of this geometric triangle consists of c − a
elements. Then there are 1
, it follows that there are not any b -- nilpotent
c it
The first one is a geometric triangle with "vertices" accn−c, ab+1bc−b−1cn−c and ab+1cn−b−1.
This triangle is a subset of DN c
c and there are not any common elements of triangle and
the basic layers with respect to a. Since all the endomorphisms of the triangle with an
exception of elements Id(cid:16)ST R(n){a, c}(cid:17) are left elements of the basic layers with respect
to c, the triangle is denoted by L△. The second part of IT (cid:0)△(n){a, b, c}(cid:1) is a geometric
triangle with "vertices" abcn−b, aa+1bb−a−1cn−b and aa+1cn−a−1. This triangle is a subset of
DN n−a−1
and there are not any common elements of the triangle and the basic layers with
respect to c. Since all the endomorphisms of the triangle with an exception of elements
Id(cid:16)ST R(n){a, c}(cid:17) are right elements of the basic layers with respect to a, the triangle is
denoted by R△. The third part of the idempotent triangle is semiring RI(cid:0)△(n){a, b, c}(cid:1)
1
2 (c − a)(c − a + 1) − (c − b)(b − a) = 1
2((c − b)2 + (b − a)2 + c − a).
(b − a)(c − a).
a
a
semiring
20
IVAN TRENDAFILOV
whose elements are the intersections of all the basic layers with respect to a and c. So,
L△ ∪ R△ = IT (cid:0)△(n){a, b, c}(cid:1) \RI(cid:0)△(n){a, b, c}(cid:1).
Let α ∈ L△ and β ∈ R△. Then α = ambℓ−mcn−ℓ, where ℓ = b + 1, . . . , c, ℓ − m ≥ 0
and α(a) = α(b) = a. Similarly, β = ambℓ−mcn−ℓ, where m = a + 1, . . . , b, ℓ − m ≥ 0 and
β(b) = β(c) = c.
So,
the elements of
Now we calculate:
1. akcn−k · α = akcn−k for α ∈ L△.
2. akcn−k · β = akcn−k for β ∈ R△.
for k = a + 1, . . . , b
for k = b + 1, . . . , c
an−jcj
. Thus, in all
semiring Id(cid:16)ST R(n){a, c}(cid:17) are left zeroes of
Let akcn−k ∈ Id(cid:16)ST R(n){a, c}(cid:17). Then k = a + 1, . . . , c. We find:
3. akcn−k · ε = akcn−k for ε ∈ RI(cid:0)△(n){a, b, c}(cid:1).
IT (cid:0)△(n){a, b, c}(cid:1).
4. α · akcn−k = ambℓ−mcn−ℓ · akcn−k = aℓcn−ℓ ∈ Id(cid:16)ST R(n){a, c}(cid:17).
5. β · akcn−k = ambℓ−mcn−ℓ · akcn−k = amcn−m ∈ Id(cid:16)ST R(n){a, c}(cid:17).
6. For any ε = atbn−t−jcj ∈ RI(cid:0)△(n){a, b, c}(cid:1), where t = a + 1, . . . , b and
j = n − c, . . . , n − a − 1 it follows ε · akcn−k =(cid:26) atcn−t
cases ε · akcn−k ∈ Id(cid:16)ST R(n){a, c}(cid:17). Hence, the semiring Id(cid:16)ST R(n){a, c}(cid:17) is an ideal
of IT (cid:0)△(n){a, b, c}(cid:1).
that IT (cid:0)△(n){a, b, c}(cid:1) \RI(cid:0)△(n){a, b, c}(cid:1) is an ideal of
IT (cid:0)△(n){a, b, c}(cid:1) we find:
b + 1, . . . , c, it follows α · α1 = aℓcn−ℓ ∈ Id(cid:16)ST R(n){a, c}(cid:17).
a + 1, . . . , b, it follows β · β1 = amcn−m ∈ Id(cid:16)ST R(n){a, c}(cid:17).
and m = a + 1, . . . , b, it follows α · β = amcn−m ∈ Id(cid:16)ST R(n){a, c}(cid:17).
and m = a + 1, . . . , b, it follows β · α = aℓcn−ℓ ∈ Id(cid:16)ST R(n){a, c}(cid:17).
11. For any α ∈ L△, β ∈ R△ and ε ∈ RI(cid:0)△(n){a, b, c}(cid:1) it follows α · ε = α and
12. For any ε = atbn−t−jcj ∈ RI(cid:0)△(n){a, b, c}(cid:1), where t = a + 1, . . . , b and
8. For any β = ambℓ−mcn−ℓ ∈ R△ and β1 = am1bℓ1−m1cn−ℓ1 ∈ R△, where m, m1 =
7. For any α = ambℓ−mcn−ℓ ∈ L△ and α1 = am1bℓ1−m1cn−ℓ1 ∈ L△, where ℓ, ℓ1 =
10. For any α = ambℓ−mcn−ℓ ∈ L△ and β = am1bℓ1−m1cn−ℓ1 ∈ r△, where ℓ = b+ 1, . . . , c
9. For any α = ambℓ−mcn−ℓ ∈ L△ and β = am1bℓ1−m1cn−ℓ1 ∈ r△, where ℓ = b + 1, . . . , c
j = n − c, . . . , n − a − 1, α = akbℓ−kcn−ℓ ∈ L△, where ℓ = b + 1, . . . , c and
β = ambℓ−mcn−ℓ ∈ R△, where m = a + 1, . . . , b, it follows
In order
to prove
β · ε = β.
ε · α = atbn−t−jcj · akbℓ−kcn−ℓ = an−jcj ∈ Id(cid:16)ST R(n){a, c}(cid:17) and
ε · β = atbn−t−jcj · ambℓ−mcn−ℓ = atcn−t ∈ Id(cid:16)ST R(n){a, c}(cid:17) .
The endomorphisms α of L△ are characterized in triangle △(n){a, b, c} by equalities:
α(a) = α(b) = a, α(c) = c. So, if α, β ∈ L△, then (α + β)(a) = (α + β)(b) = a and
(α + β)(c) = c, i.e. α + β ∈ L△. Similar reasonings we can have for triangle R△.
GEOMETRICAL STRUCTURES OF THE ENDOMORPHISM SEMIRING
21
The biggest endomorphism of L△ is the vertex of this triangle ab+1cn−b−1. The least
endomorphism of R△ is the vertex of the triangle abcn−b So, for any α ∈ L△ and β ∈ R△
it follows α < ab+1cn−b−1 < abcn−b < β. Hence, IT (cid:0)△(n){a, b, c}(cid:1) \RI(cid:0)△(n){a, b, c}(cid:1) is
an ideal of IT (cid:0)△(n){a, b, c}(cid:1).
(cid:3)
From the proof of the last theorem it follows
Corollary 34. The geometric triangles L△ and R△ are subsemirings of triangle
△(n){a, b, c}.
An immediate consequence of fact above is
Corollary 35. For any triangle △(n){a, b, c}, n ≥ 3, the idempotent triangle is a
n ∩ △(n){a, b, c}.
c
c
c
a
n ∩ △(n){a, b, c}, similarly to Proposition 9, it follows
"vertices" are endomorphisms a, ab+1bn−b−1, ab+1bc−bcn−c−1 and ac+1cn−c−1 and whose
Similarly to geometric triangles, we can consider geometric parallelograms and
disjoint union of subsemirings L△, R△ and RI(cid:0)△(n){a, b, c}(cid:1).
geometric trapezoids. For example, semiring RI(cid:0)△(n){a, b, c}(cid:1) can be represented as a
geometric parallelogram whose "vertices" are endomorphisms abbc−bcn−c, aa+1bc−a−1cn−c,
aa+1bb−acn−b−1 and abbcn−b−1. (Note that exactly the last endomorphism is a boundary
between the triangles L△ and R△.) The "sides" of this parallelogram are the idempotent
parts of basic layers Ln−c
Lb
(cid:0)△(n){a, b, c}(cid:1), Ln−b−1
(cid:0)△(n){a, b, c}(cid:1), La+1
Now we consider the a -- nilpotent elements of triangle △(n){a, b, c}. The set of
the left part of Lb+1
endomorphisms β from the left part of Ln−c−1
(cid:0)△(n){a, b, c}(cid:1) and
a(cid:0)△(n){a, b, c}(cid:1).
all a -- nilpotent elements of this triangle is denoted by N [a](cid:0)△(n){a, b, c}(cid:1). Since
N [a](cid:0)△(n){a, b, c}(cid:1) = N [a]
Proposition 36. The set N [a](cid:0)△(n){a, b, c}(cid:1) is a subsemiring of △(n){a, b, c}.
The semiring N [a](cid:0)△(n){a, b, c}(cid:1) can be represented as a geometric trapezoid whose
"sides" are semiring N [a](cid:16)ST R(n){a, b}(cid:17), the subsetset of n − c endomorphisms α from
(cid:0)△(n){a, b, c}(cid:1) such that α(b) = a, α(c) = b, the subsetset of c − b + 1
(cid:0)△(n){a, b, c}(cid:1) such that α(b) = a, α(c) = b
and semiring N [a](cid:16)ST R(n){a, c}(cid:17). We find, as in the proof of the last theorem, that the
order of this semiring is (cid:12)(cid:12)N [a](cid:0)△(n){a, b, c}(cid:1)(cid:12)(cid:12) = 1
N [b](cid:0)△(n){a, b, c}(cid:1) = N [b]
The semiring N [b](cid:0)△(n){a, b, c}(cid:1) can be represented as a geometric parallelogram
"sides" are semiring N [b](cid:16)ST R(n){a, b}(cid:17), semiring N [b](cid:16)ST R(n){b, c}(cid:17), the subsetset
(cid:0)△(n){a, b, c}(cid:1) such that
a(cid:0)△(n){a, b, c}(cid:1) such that β(a) = β(b) = β(c) = b. The order of this semiring is
(cid:12)(cid:12)N [b](cid:0)△(n){a, b, c}(cid:1)(cid:12)(cid:12) = (a + 1)(n − c).
Finally, the semiring N [c](cid:0)△(n){a, b, c}(cid:1) can be represented as a geometric trapezoid
of a + 1 endomorphisms α from the right part of Ln−c−1
α(a) = α(b) = α(c) = b and the subset of n − c endomorphisms β from the left part
of La
n ∩ △(n){a, b, c} and N [c](cid:0)△(n){a, b, c}(cid:1) = N [c]
whose "vertices" are endomorphisms aabn−a, b, bc+1cn−c−1 and aabc−a+1cn−c−1 and whose
In the same way we can construct the following semirings:
2(n − c)(n + c − 2b + 1).
a
c
22
IVAN TRENDAFILOV
whose "vertices" are endomorphisms bbcn−b, c, aacn−a and aabb−acn−b and whose "sides" are
c
1
2 (a + 1)(2b − a + 2).
Proposition 37. The
endomorphisms α from the right part of La
and the subset of a + 1 endomorphisms β from the right part of Ln−b
semiring N [c](cid:16)ST R(n){b, c}(cid:17), semiring N [c](cid:16)ST R(n){a, c}(cid:17), the subsetset of b − a + 1
a(cid:0)△(n){a, b, c}(cid:1) such that α(b) = α(c) = c
(cid:0)△(n){a, b, c}(cid:1) such
that β(a) = b, β(b) = β(c) = c. The order of this semiring is (cid:12)(cid:12)N [c](cid:0)△(n){a, b, c}(cid:1)(cid:12)(cid:12) =
N [c](cid:0)△(n){a, b, c}(cid:1) are trivial.
Proof. Let α ∈ N [a](cid:0)△(n){a, b, c}(cid:1). Then a is a unique fixed point of α. If we assume
a. Now α(c) = b, or α(c) = a, which means that α = a. For any β ∈ N [a](cid:0)△(n){a, b, c}(cid:1) we
Hence, α · β = a and N [a](cid:0)△(n){a, b, c}(cid:1) is a trivial semiring. The same argument shows
that N [b](cid:0)△(n){a, b, c}(cid:1) and N [c](cid:0)△(n){a, b, c}(cid:1) are also trivial semirings.
semirings N [a](cid:0)△(n){a, b, c}(cid:1), N [b](cid:0)△(n){a, b, c}(cid:1)
that α(b) = c, then c ≥ α(c) ≥ α(b) = c implies α(c) = c, which is impossible. So, α(b) =
find (α · β)(a) = a, (α · β)(b) = β(α(b)) = β(a) = a and (α · β)(c) = β(α(c)) = β(b) = a.
and
(cid:3)
Example 38. Let us consider triangle △(6){1, 3, 4}. Figure 5 illustrates the semirings
of 1 -- nilpotent, 3 -- nilpotent and 4 -- nilpotent endomorphisms, an idempotent triangle
and right identities. Here we observe that △(6){1, 3, 4} can be represented as a union
of the following semirings: N [1](cid:0)△(6){1, 3, 4}(cid:1), N [3](cid:0)△(6){1, 3, 4}(cid:1), N [4](cid:0)△(6){1, 3, 4}(cid:1),
4(cid:0)△(6){1, 3, 4}(cid:1). But this union is not disjoint
1(cid:0)△(6){1, 3, 4}(cid:1), L3
L2
since the right identities are intersections of the basic layers of the triangle.
1(cid:0)△(6){1, 3, 4}(cid:1) and L2
4
14
5
34
5
idempotent triangle
14
2 4
134
4
34
2 4
4 - nilpotent elements
right
identities
14
3 3
134
2
3
134
2 3
34
3 3
14
4 2
134
3
2
134
2 2 2
134
2
3
34
4 2
14
5
134
4
134
3 2
134
2 3
134
4
34
5
1
13
5
13
4 2
13
3 3
13
2 4
13
5
3
1 - nilpotent elements
3 - nilpotent elements
Figure 5.
In order to represent △(n){a, b, c} as a disjoint union of its subsemirings, we look for
new subsemirings. Now we consider the set of all the left elements of the basic layers with
GEOMETRICAL STRUCTURES OF THE ENDOMORPHISM SEMIRING
23
respect to a. This set can be represented as a geometric parallelogram whose "vertices"
are endomorphisms abbn−b, aa+1bn−a−1, aa+1bc−acn−c−1 and abbc−b+1cn−c−1. The "sides" of
this parallelogram are semiring Id(cid:16)ST R(n){a, b}(cid:17), the set of the left elements of the
(cid:0)△(n){a, b, c}(cid:1), the subset of b − a endomorphisms α of the layer
(cid:0)△(n){a, b, c}(cid:1) with fixed points a and b and the set of the left elements of the
a(cid:0)△(n){a, b, c}(cid:1). We denote this parallelogram by Lpar. Then Lpar =
biggest basic layer La+1
Ln−c−1
least basic layer Lb
(b − a)(n − c).
a
c
Similarly, we consider the set of all the right elements of the basic layers with respect
to c. This set can also be represented as a geometric parallelogram whose "vertices"
are endomorphisms aabc−acn−c, bccn−c, bb+1cn−b+1 and aabb−a+1cn−b−1. The "sides" of
parallelogram are the set of right elements of the biggest basic layer Ln−c
the semiring Id(cid:16)ST R(n){b, c}(cid:17), the set of the right elements of
(cid:0)△(n){a, b, c}(cid:1) and the subset of c − b endomorphisms α of layer La
Ln−b−1
with fixed points b and c. We denote this parallelogram by Rpar. Then Rpar = (a + 1)(c −
b).
a
least basic layer
(cid:0)△(n){a, b, c}(cid:1),
a(cid:0)△(n){a, b, c}(cid:1)
c
Proposition 39. The geometric parallelograms Lpar and Rpar are subsemirings of
triangle △(n){a, b, c}.
Proof. We shall prove only that Lpar is a semiring, since the proof for Rpar is the same.
Let α ∈ Lpar. Then α(a) = a and α(b) = α(c) = b. It is evident that (α + β)(a) = a
and (α + β)(b) = (α + β)(c) = b. For products we show (α · β)(a) = a, (α · β)(b) = b and
(α · β)(c) = β(α(c)) = β(b) = b.
(cid:3)
As we have seen in the last proofs, any endomorphism α of some subsemiring of the
triangle can be characterized by ordered triple (x, y, z), where α(a) = x, α(b) = y and
α(c) = z and x, y, z ∈ {a, b, c}. This triple is called a type of semiring.
Now we can summarize the results of Theorem 33, corollaries 34 and 35 and
propositions 36 and 39 and arrange the following "puzzle" -- fig. 6, where we register
the type of semirings.
RV
a,c,c
( )
[ ]c
N
b,c,c
( )
LV
RI
a,a,c
( )
a,b,c
( )
Rpar
b,b,c
( )
[ ]a
N
Lpar
[ ]b
N
a,a,b
( )
a,b,b
( )
b,b,b
( )
Figure 6.
Actually we prove the following
24
IVAN TRENDAFILOV
Theorem 40. Any triangle △(n){a, b, c}, n ≥ 3, is a disjoint union of the following
subsemirings of the triangle: N [a](cid:0)△(n){a, b, c}(cid:1), N [b](cid:0)△(n){a, b, c}(cid:1), N [c](cid:0)△(n){a, b, c}(cid:1),
Lpar, Rpar, L△, R△ and RI(cid:0)△(n){a, b, c}(cid:1).
is a disjoint union of semirings N [a](cid:0)△(n){a, b, c}(cid:1), Lpar, Rpar and RI(cid:0)△(n){a, b, c}(cid:1), that
As a direct consequence of the last theorem, it follows that semiring △(n){a, b, c} ∩ E (b)
Cn
is this semiring can be represented as a geometric parallelogram consisting of the four
parallelograms corresponding to these subsemiring -- fig. 6.
References
[1] J. Jezek, T. Kepka and M. Mar`oti, "The endomorphism semiring of a semilattice",
Semigroup Forum, 78, pp. 21 -- 26, 2009.
[2] I. Trendafilov and D. Vladeva, "Idempotent Elements of the Endomorphism
Semiring of a Finite Chain", ISRN Algebra, vol. 2013, Article ID 120231, 9 pages, 2013.
[3] I. Trendafilov and D. Vladeva, "Nilpotent elements of the endomorphism semiring
of a finite chain and Catalan numbers", Proceedings of the Forty Second Spring Conference
of the Union of Bulgarian Mathematicians, Borovetz, April 2 -- 6, pp. 265 -- 271, 2013.
[4] I. Trendafilov and D. Vladeva, "Endomorphism semirings without zero of a finite
chain", Proceedings of the Technical University of Sofia, vol. 61, no. 2, pp. 9 -- 18, 2011.
[5] I. Trendafilov and D. Vladeva, "The endomorphism semiring of a finite chain",
Proceedings of the Technical University of Sofia, vol. 61, no. 1, pp. 9 -- 18, 2011.
[6] I. Trendafilov and D. Vladeva, "Subsemirings of the endomorphism semiring of a
finite chain", Proceedings of the Technical University of Sofia, vol. 61, no. 1, pp. 19 -- 28,
2011.
[7] J. Zumbragel, "Classification of finite congruence-simple semirings with zero,"
Journal of Algebra and Its Applications, vol. 7, no. 3, pp. 363 -- 377, 2008.
[8] J. Golan, Semirings and Their Applications, Kluwer, Dordrecht, 1999.
[9] G. Gratzer, Lattice Theory: Foundation, Birkhauser Springer Basel AG, 2011.
[10] D. Ferrario and R. Piccinini, Simplicial Structures in Topology, Springer, New
York, USA, 2011.
[11] M. Desbrun, A, Hirani, M. Leok, J. Marsden, "Discrete Exterior Calculus",
arXiv:math/0508341v2 [math.DG] 18 Aug 2005.
|
1911.08282 | 2 | 1911 | 2019-11-21T02:06:32 | Classification of left octonion modules | [
"math.RA"
] | It is natural to study octonion Hilbert spaces as the recently swift development of the theory of quaternion Hilbert spaces. In order to do this, it is important to study first its algebraic structure, namely, octonion modules. In this article, we provide complete classification of left octonion modules.
In contrast to the quaternionic setting, we encounter some new phenomena. That is, a submodule generated by one element $m$ may be the whole module and may be not in the form $\O m$. This motivates us to introduce some new notions such as associative elements, conjugate associative elements, cyclic elements. We can characterize octonion modules in terms of these notions.
It turns out that octonions admit two distinct structures of octonion modules, and moreover, the direct sum of their several copies exhaust all octonion modules with finite dimensions. | math.RA | math |
Classification of left octonion modules
Qinghai Huo, Yong Li, Guangbin Ren
Abstract
It is natural to study octonion Hilbert spaces as the recently swift development of the theory
of quaternion Hilbert spaces. In order to do this, it is important to study first its algebraic
structure, namely, octonion modules. In this article, we provide complete classification of left
octonion modules. In contrast to the quaternionic setting, we encounter some new phenomena.
That is, a submodule generated by one element m may be the whole module and may be not in
the form Om. This motivates us to introduce some new notions such as associative elements,
conjugate associative elements, cyclic elements. We can characterize octonion modules in terms
of these notions. It turns out that octonions admit two distinct structures of octonion modules,
and moreover, the direct sum of their several copies exhaust all octonion modules with finite
dimensions.
Keywords: Octonion module; associative element; cyclic element; Cℓ7-module.
AMS Subject Classifications: 17A05
Contents
1 introduction
2 Preliminaries
2.1 The algebra of the octonions O . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
2.2 Universal Clifford algebra . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
3 O-modules
4 The structure of left O-moudles
4.1 Finite dimensional O-modules . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
4.2 Structure of general left O-modules . . . . . . . . . . . . . . . . . . . . . . . . . . . .
4.3 Cyclic elements in left O-module . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
1
3
3
4
6
10
10
13
15
1
introduction
The theory of quaternion Hilbert spaces brings the classical theory of functional analysis into the
non-commutative realm (see [10, 16, 17, 20, 21]).
It arises some new notions such as spherical
spectrum, which has potential applications in quantum mechanics (see [4, 6]). All these theories
are based on quaternion vector spaces, or more precisely, quaternion modules, and quaternion
bimodules. A systematic study of quaternion modules is given by Ng [15]. It turns out that the
1
category of (both one-sided and two-sided) quaternion Hilbert spaces is equivalent to the category
of real Hilbert spaces.
It is a natural question to study the theory of octonion spaces. Goldstine and Horwitz in 1964
[8] initiated the study of octonion Hilbert spaces; more recently, Ludkovsky [12, 13] studied the
algebras of operators in octonion Banach spaces and spectral representations in octonion Hilbert
spaces. Although there are few results about the theory of octonion Hilbert spaces, it is not full
developed since it even lacks of coherent definition of octonion Hilbert spaces.
In contrast to the complex or quaternion setting, some new phenomena occur in the setting of
octonions (see Example 4.13, 4.14):
maybe the whole module.
• If m is an element of an octonion module, then Om is not an octonion sub-module in general.
• If m is an element of an octonion module, then the octonion sub-module generated by m
This means that the structure of octonion module is more involved. We point out that some
gaps appear in establishing the octonionic version of Hahn-Banach Theorem by taking Om as a
submodule ([12, Lemma 2.4.2]). The submodule generated by a submodule Y and a point x is not
of the form {y + px y ∈ Y, p ∈ O}, this is wrong even for the case Y = {0}. It means the proof
can not repeat the way in canonical case. The involved structure of octonion module accounts for
the slow developments of octonion Hilbert spaces.
In the study of octonion Hilbert space and Banach space, it heavily depends on the direct sum
structure of the space under considered sometimes, which always brings the question back to the
classic situation. For example, in the proof of [12, Theorem 2.4.1], it declares that every O-vector
space is of the following form:
X = X0 ⊕ X1e1 ⊕ ··· ⊕ X7e7.
Note that therein the definition of O-vector space is actually a left O-module with an irrelevant
right O-module structure. We thus can only consider the left O-module structure of it. We show
that the assertion above does not always work. In order to study octonion Hilbert spaces, we need
to provide its solid algebraic foundation by studying the deep structure of one-sided O-modules
and O-bimodules. We only consider the left O-modules in this paper, the bimodule case will be
discussed in a later paper. In this paper, we characterize the structure of O-modules completely.
We remark that Eilenberg [5] initiated the study of bimodules over non-associative rings. Jacob-
son studied the structures of bimodules over Jordan algebra and alternative algebra [11]. One-sided
modules over octonion was investigated in [8] for studying octonion Hilbert spaces. However for
the classification of O-modules, the problem is untouched.
In this article, we shall give the classification of O-modules. It turns out that the set O admits
two distinct O-module structures. One is the canonical one, denoted by O; the other is denoted by
O (see Example 3.5). To characterize any O-module M , we need to introduce some new notions,
called associative element and conjugate associative element. Their collections are denoted
by A (M ) and A −(M ) respectively. The ordered pair of their dimensions as real vector spaces is
called the type of M . These concepts are crucial in the classification of left O-modules.
Our first main result is about the characterization of the left O-modules.
Theorem 1.1. Let M be a left O-module. Then
If dimR M < ∞, then
M = OA (M ) ⊕ OA −(M ).
M ∼= On1 ⊕ O
n2 ,
2
where (n1, n2) is the type of M .
Its proof depends heavily on the isomorphism between the category O-Mod and the category
Cℓ7-Mod. By the matrix realization of Cℓ7,
Cℓ7 ∼= M (8, R) ⊕ M (8, R),
there are only two kinds of simple left O-module, namely, O and O up to isomorphism. Hence
the structure of finite dimension O-modules follows by Wedderburn's Theorem for central simple
algebras [18]. The general case relies on some elementary properties of A (M ) and A −(M ), along
with an important fact that every element of O-module generates a finite dimensional submodule.
By the way, we found that the octonions O can be endowed with both Cℓ7-module structure and
Cℓ6-structure, using it we get an irreducible complex representation of Spin(7).
Our second topic is about the cyclic elements. In contrast to the setting of complex numbers
and quaternions, cyclic elements play key roles in the study of octonion sub-modules. An element
m in a given octonion module M is called cyclic elements if the submodule generated by it is
exactly Om. The collection of these elements is denoted by C (M ). It turns out that the cyclic
elements are determined by the associative elements A (M ) and the conjugate associative elements
A −(M ) completely.
Theorem 1.2. For any left O-module M , we have
C (M ) =
[p∈O
In view of Theorems 1.1 and Theorem 1.2, we find
[
p · A (M )
[p∈O
p · A −(M )
.
M = SpanRC (M ).
This means that the module M is determined completely by its cyclic elements in the form of real
linear combination. For any element m in a left O-module M , there exist m± ∈ SpanRC ±(M ) such
that m = m+ + m−. We can decomposite m± into a combination of real linearly independent cyclic
elements. Denote by l±
m the minimal length of the decompositions of m±. We conjecture that,
If the conjecture is right, then the structure of the submodule generated by one element is completely
clear.
hmiO ∼= Ol+
m ⊕ Ol−
m.
2 Preliminaries
2.1 The algebra of the octonions O
The algebra of the octonions O is a non-associative, non-commutative, normed division algebra
over the R. Let e1, . . . , e7 be its natural basis throughout this paper, i.e.,
eiej + ejei = −2δij,
i, j = 1, . . . , 7.
3
For convenience, we denote e0 = 1.
In terms of the natural basis, an element in octonions can be written as
x = x0 +
7
Xi=1
xiei,
xi ∈ R.
The conjugate octonion of x is defined by x = x0 −P7
√xx ∈ R, the real part of x is Re x = x0 = 1
i=1 xiei, and the norm of x equals x =
The full multiplication table is conveniently encoded in the Fano mnemonic graph (see [3, 22]).
In the Fano mnemonic graph, the vertices are labeled by 1, . . . , 7 instead of e1, . . . , e7. Each of the
7 oriented lines gives a quaternionic triple. The product of any two imaginary units is given by the
third unit on the unique line connecting them, with the sign determined by the relative orientation.
2 (x + x).
Fig.1 Fano mnemonic graph
The associator of three octonions is defined as
[x, y, z] = (xy)z − x(yz)
for any x, y, z ∈ O, which is alternative in its arguments and has no real part. That is, O is an
alternative algebra and hence it satisfies the so-called R. Monfang identities [19]:
(xyx)z = x(y(xz)), z(xyx) = ((zx)y)x, x(yz)x = (xy)(zx).
The commutator is defined as
[x, y] = xy − yx.
2.2 Universal Clifford algebra
We shall use the Clifford algebra Cℓ7 to study left O-modules. In this subsection, we review some
basic facts for universal Clifford algebras. The Clifford algebras are introduced by Clifford in 1882.
For its recent development, we refer to [2, 7, 14].
Definition 2.1. Let A be an associative algebra over R with unit 1 and let v : Rn → A be an
R-linear embedding. The pair (A, v) is said to be a Clifford algebra over Rn, if
(i). A is generated as an algebra by {v(x) x ∈ Rn} and {λ1 λ ∈ R};
4
(ii). (v(x))2 = − x2 ,∀ x ∈ Rn.
We need some notations and conventions.
i=1 x2
i .
• x = (x1, . . . , xn) ∈ Rn,x2 =Pn
• Let {fi}n
with 1 in the i-th slot.
i=1 be the canonical orthonormal basis of Rn, gi = v(fi) ∈ A, and fi = (0, . . . , 0, 1, 0, . . . , 0)
• Let P(n) be the collection of all the subsets of {1, . . . , n}.
• For any α ∈ P(n), if α 6= ∅, we write α = {α1, . . . , αk} with 1 ≤ α1 < ··· < αk ≤ n and we
set gα = gα1 ··· gαk . Otherwise, we denote g∅ = 1.
The Clifford algebra A can be described alternatively with the above notations as
(i). A is R-linearly generated by {gα α ∈ P(n)};
(ii). gigj + gjgi = −2δij for any i, j = 1, . . . , n.
It is well-known that dimR A ≤ 2n. The Clifford algebra (A, v) over Rn may be not unique (up
to isomorphism of algebras); see for example [7]. But the universal Clifford algebra Cℓn over Rn is
unique up to isomorphism.
Definition 2.2. A Clifford algebra (A, v) is said to be a universal Clifford algebra, if for each
Clifford algebra (B, µ) over Rn, there exists an algebra homomorphism β : A → B, such that
µ = β ◦ v and β(1A) = 1B. Namely, the following diagram commutes.
A
v
B
β
µ
Rn
We recall some equivalent descriptions of universal Clifford algebra Cℓn.
Theorem 2.3 ([7]). (A, v) is a Clifford algebra over Rn, the following are equivalent:
(i). (A, v) is a universal Clifford algebra Cℓn over Rn;
(ii). dimRA = 2n;
(iii). g1 ··· gn /∈ R.
At the last of this subsection, we give the following algebra isomorphism of Cℓn (see for example
[2]).
• Cℓn+8 ∼= Cℓn ⊗ M (16, R) ∼= M (16, Cℓn);
• for n = 0, . . . , 7 we have table
5
n
0
1
2
3
4
5
6
Cℓn
R
C
H
HL H
M (2, H)
M (4, C)
M (8, R)
Here, we denote by M (k, F) the collection of all k × k matrices with each entry in the algebra F.
7 M (8, R)L M (8, R)
3 O-modules
We set up in this section some preliminary definitions and results on left O-modules.
Definition 3.1. A real vector space M is called a (left) O-module, equipped with a scalar multi-
plication O × M → M , denoted by
(q, m) 7→ qm,
such that the following axioms hold for all q, q1, q2 ∈ O, λ ∈ R and all m, m1, m2 ∈ M :
(i). (λq)m = λ(qm) = q(λm);
(ii). (q1 + q2)m = q1m + q2m, q(m1 + m2) = qm1 + qm2;
(iii). [q1, q2, m] = −[q2, q1, m];
(iv). 1m = m.
Here, the left associator is defined by
[q1, q2, m] := (q1q2)m − q1(q2m).
Note that this definition is equivalent to the definition given in [12, 13], wherein the axiom (iii)
is replaced by
q2m = q(qm), for all q ∈ O, m ∈ M.
The proof is trivial by polarizing the above relation. It also agrees with the one given in [8] wherein
M needs to satisfy an additional axiom: p(p−1x) = x, which can be deduced from the equation
p(px) = p2x directly.
Let M be a left O-module. The definition of the terms submodule, homomorphism, isomorphism,
kernel of a homomorphism, which are familiar from the study of associative modules, do not invole
associativity of multiplication and are thus immediately applicable to the case in general. Let
HomO(M, M ′) denote the set of all O-homomorphisms from M to M ′ as usual (M ′ being arbitrary
O-module). So is the notation JN for the subset of M spaned by all products rn with r ∈ J and
n ∈ N (J being arbitrary nonempty subset of O and N being arbitrary nonempty subset of M ),
here we must of course distinguish between J1(J2N ) and (J1J2)N . Let hNiO denote the minimal
submodule which contains N as before. An element m ∈ M is said to be associative if
[p, q, m] = 0,
∀p, q ∈ O.
6
Denote by A (M ) the set of all associative elements in M :
One useful identity which holds in any O-module M is
A (M ) := {m ∈ M [p, q, m] = 0, ∀p, q ∈ O}.
[p, q, r]m + p[q, r, m] = [pq, r, m] − [p, qr, m] + [p, q, rm],
(3.1)
where p, q, r ∈ O, m ∈ M . The proof is by striaghtforward calculations. It follows that:
Lemma 3.2. For all associative element m ∈ A (M ), we have
[p, q, rm] = [p, q, r]m, for all p, q, r ∈ O.
The following elementary property will be useful in the sequel. The proof is trivial and will be
omitted here.
Proposition 3.3. If f ∈ HomR(M, N ), then f ([p, q, x]) = [p, q, f (x)] for all p, q ∈ O, x ∈ M .
Therefore f (A (M )) ⊆ A (N ).
We give several elemetary left O-module examples.
Example 3.4. It is easy to see the real vector spaces O, On, M (n, O) with the obvious mul-
tiplication are all left O-module. Clearly, the sets of associative elements on these modules are
R, Rn, M (n, R) respectively.
We can define a different O-module structure on the octonions O itself.
Example 3.5 ( O). Define:
It's easy to check this is a left O-module. Indeed
p·x := px, ∀p ∈ O, x ∈ O.
We shall denote this O-module by O. By direct calculations, we obtain:
p2·x = p2x = p·(p·x).
[p, q, x]O = [p, q, x] + [p, q]x.
This implies that A (O) = {0}. Note that Proposition 3.3 ensures that A (M ) ∼=R A (N ) when
M ∼=O N , and therefore O ≇ O.
However, there is a special subset in O, that is, the real subspace R. To describe such elements,
we introduce a new notion of conjugate associative element.
Definition 3.6. An element m ∈ M is said to be conjugate associative if
∀p, q ∈ O.
Denote by A −(M ) the set of all conjugate associative elements.
(pq)m = q(pm),
Lemma 3.7. A −(O) = R.
7
Proof. Suppose x ∈ A −(O), then for any p, q ∈ O,
0 = (pq)·x − q·(p·x) = (pq)·x − (qp)·x + [q, p, x]O = [p, q]·x − [p, q, x]O
Hence we obtain
[p, q]x − [p, q, x] − [p, q]x = [q, p, x] = 0
this implies that x ∈ R. Clearly R ⊆ A −(O), thus A −(O) = R.
Lemma 3.8. For any left O-module M , we have A (M ) ∩ A −(M ) = {0} .
Proof. Obviously 0 ∈ A (M ) ∩ A −(M ). Let x ∈ A (M ) ∩ A −(M ), then for any p, q ∈ O,
[p, q]x = (pq)x − (qp)x = (pq)x − p(qx) = [p, q, x] = 0.
This implies x = 0 since we can choose p, q ∈ O such that [p, q] 6= 0. This proves the lemma.
Remark 3.9. Clearly, both A (M ) and A −(M ) are real vector spaces. If M is of finite dimension,
we call the ordered pair (dimR A (M ), dimR A −(M )) type of M . We shall use these notions to
describe the structure of left O-modules. It turns out that type is a complete invariant in the finite
dimensional case.
Let M be a left O-module. We shall establish some properties of associative elements and
conjugate associative elements which will be used in the seuel.
Lemma 3.10. Let {xi}n
i=1 be an R-linearly independent set of associative elements of M . If
n
Xi=1
rixi = 0, ri ∈ O for each i = 1, . . . , n,
then ri = 0 for each i = 1, . . . , n.
Proof. The proof is by induction on n. For the case n = 1, we have rx = 0.
x ∈ A (M ), it follows that
0 = r−1(rx) = (r−1r)x = x,
If r 6= 0, since
a contrdiction with our assumption x 6= 0. Assume the lemma holds for degree k, we will prove it
for k + 1. Suppose Pk+1
i=1 rixi = 0 and rk+1 6= 0. Denote si = −rir−1
k+1. Therefore
k
Since xk+1 ∈ A (M ), thus for all p, q,∈ O,
xk+1 =
sixi.
Xi=1
0 = [p, q, xk+1] =
[p, q, sixi] =
k
Xi=1
[p, q, si]xi
k
Xi=1
where we heve used Lemma 3.2 in the last eqution. Hence by induction hypothesis we conclude
that,
[p, q, si] = 0, for each i ∈ {1, . . . , k} and for all p, q ∈ O.
8
This implies si ∈ R. Note that
k
Xi=1
which contradicts the hypothesis that {xi}n
sixi = xk+1,
i=1 is R-linearly independent, we thus prove the lemma.
Note that we have actually proved the following property.
Corollary 3.11. Let S ⊆ A (M ). Then that S is O-linearly independent if and only if it is
R-linearly independent.
Lemma 3.12. Under the assumptions of Lemma 3.10, if y = Pn
ri ∈ R for each i ∈ {1, . . . , n}.
Proof. Since y ∈ A (M ), we have for all p, q ∈ O,
i=1 rixi ∈ A (M ), then we have
0 = [p, q,
rixi] =
[p, q, ri]xi.
n
Xi=1
n
Xi=1
By Lemma 3.10, we conclude [p, q, ri] = 0, i = 1, . . . , n. This yields ri ∈ R.
We next consider the properties of conjugete elements. It turns out that similar statements hold
for S ⊆ A −(M ).
i=1 ⊆ A −(M ) be an R-linearly independent set. If
Lemma 3.13. Let S = {xi}n
Xi=1
rixi = 0, ri ∈ O for each i = 1, . . . , n,
n
i=1 rixi ∈ A −(M ), then ri ∈ R for each i = 1, . . . , n.
Proof. Induction on n as before. The case of n = 1 is trivial. Assume the lemma holds for degree
k+1. Therefore
i=1 rixi = 0 and rk+1 6= 0. Denote si = −rir−1
then ri = 0 for each i = 1, . . . , n.
Moreover, if y =Pn
k, we will prove it for k + 1. SupposePk+1
Since xk+1 ∈ A −(M ), thus for all p, q,∈ O,
xk+1 =
sixi.
k
Xi=1
0 = (qp)xk+1 − p(qxk+1)
k
k
=
=
Xi=1
Xi=1
Xi=1
= −
k
(qp)(sixi) − p(q(sixi))
(si(qp))xi − p((siq)xi)
([si, q, p]) xi
9
Hence by induction hypothesis we obtain that si ∈ R. The rest of the proof runs much the same
as in Lemma 3.10.
4 The structure of left O-moudles
In this section, we are in a position to formulate the structure of any left O-module. We will first
concerned with the finite dimensional case and then the general case follows.
4.1 Finite dimensional O-modules
It is well-known (for example, [3, 9]) that the octonions have a very close relationship with spinors
in 7, 8 dimensions. In particular, multiplication by imaginary octonions is equivalent to Clifford
multiplication on spinors in 7 dimensions.
It turns out that the category of left O-modules is
isomorphic to the category of left Cℓ7-modules.
For any left O-module M , according to [ei, ej, x] = −[ej, ei, x], we get the left multiplication
operator L satisfies:
Hence A := SpanR{Lei i = 0, 1, . . . , 7} is a Clifford algebra over R7. This yields a Cℓ7-module
structure on M , because from the universal properties of Cl7, we have a non-trivial ring homomor-
phism
LeiLej + Lej Lei = −2δijId.
Denote this Cℓ7-module by Cℓ7M , or just M , and the Cℓ7-scalar multiplication is given by
ρ : Cℓ7 → A → EndR(M ).
for any α ∈ P(n). Here gα = gα1 ··· gαk := Leα1 ··· Leαk
O-homomorphism, where M, M ′ are two left O-modules. Then
. Let f ∈ HomO(M, M ′) be a left
gαm := eα1(eα2 (··· (eαk m)))
f (Lei x) = f (eix) = eif (x) = Lei f (x).
and hence f (gαx) = gαf (x). This means f ∈ HomCℓ7(M, M ′). Conversely, for any left Cℓ7-module
M , let {gi}7
i=1 be a basis in R7. Define:
ei·x := gix.
Then
ei·(ej·x) = gi(gjx) = (gigj)x = (−2δij − gjgi)x = −2δijx − ej·(ei·x)
This implies that
by transposition of terms, we obtain
ei·(ej·x) + ej·(ei·x) = (eiej + ejei)·x
[ei, ej, x] = −[ej, ei, x].
This yields for any p, q ∈ O, [p, q, x] = −[q, p, x]. Consequently M is a left O-module. For any
f ∈ HomCℓ7(M, M ′),
f (ei·x) = f (gix) = gif (x) = ei·f (x)
Therefore f ∈ HomO(M, M ′). In summary, we get the following important result:
Theorem 4.1. The category of left O-module is isomorphic to the category of left Cℓ7-module.
Moreover, the only two kinds of simple O-module are O and O up to isomorphsim.
10
Proof. Naturaly we have two categories O-Mod and Cℓ7-Mod.
T : O-Mod −→ Cℓ7-Mod
OM 7→ Cℓ7M
and for any morphism ϕ ∈ HomO(M, N ), it maps to T (ϕ) : Cℓ7M → Cℓ7N , which is given by
T (ϕ) : m 7→ ϕ(m).
Clearly, this is an isomorphism by above discussion. As is well known, Cℓ7 is a semi-simple
algebra and
Cℓ7 ∼= M (8, R) ⊕ M (8, R),
the only simple Cℓ7-module is (R8, 0) and (0, R8) up to isomorphism [7]. Hence there also only
exist two kinds of simple O-module. In view of Example 3.5, we thus conclude that O and O are
the two different kinds of simple O-modules. This completes the proof.
Corollary 4.2. If M is of finite real dimension, then
M ∼= On1 ⊕ O
n2 .
where (n1, n2) is the type of M . In particular, dimRM = 8(n1 + n2).
Proof. By Theorem 4.1, we can regard M as a Cℓ7-module Cℓ7M .
semi-simple algebra, we have
Cℓ7 ∼= M (8, R)M ⊕ M (8, R).
(4.1)
It is known that Cℓ7 is a
This means that Cℓ7 has exactly two isomorphism classes of simple Cℓ7- modules (see for example
[1, Chap5, Thm10]). In view of Wedderburn's Theorem for central simple algebras therefore, we
conclude
Cℓ7M ∼= Cℓ7Sn1
1 ⊕ Cℓ7Sn2
2
where S1, S2 are the representation of the two isomorphism classes of simple Cℓ7-modules. Thus
by Theorem 4.1,
OM ∼= OSn1
1 ⊕ OSn2
2 .
In particular, let S1 = O, S2 = O, then we get the conclusion as desired. We next show that
(n1, n2) is just the type. By definition, one can prove A (M ⊕ N ) = A (M ) ⊕ A (N ) and A −(M ⊕
n2 ) = n1 and
N ) = A −(M ) ⊕ A −(N ). Then it follows from Lemma 3.7 that dimR A (On1 ⊕ O
dimR A −(On1 ⊕ O
n2) = n2. This completes the proof.
We can give a complete description of the set of homomorphisms between two finite dimensional
left O-modules.
Theorem 4.3. Let M, N be two left O-modules of finite dimension. Suppose the type of M and N
are (m1, m2), (n1, n2) respectively. Then
HomO(M, N ) ∼= Mn1×m1 (R)M Mn2×m2 (R).
(4.2)
11
Proof. It follows from Corollary 4.2 that,
We claim each homomorphism of HomO(M, N ) is of the form
M ∼= Om1 ⊕ O
m2 ,
N ∼= On1 ⊕ O
n2 .
(4.3)
θ11
...
θn11
0
0
0
. . .
. . .
. . .
0
0
0
θ1m1
...
θn1m1
0
0
0
0
0
0
0
0
ψ11
...
ψn21
0
0
0
0
. . . ψ1m2
...
. . .
. . . ψn2m2
.
Inneed, we have HomO(O, O) = R IdO, HomO(O, O) = R IdO, HomO(O, O) ∼= {0}. Notice that for
any ϕ ∈ HomO(O, O) we have
ϕ(q) = qϕ(1),
we only need to show that ϕ(1) ∈ R. Propositon 3.3 then yields the conclusion as required. As for
the case HomO(O, O) ∼= R. Given f ∈ HomO(O, O), write f (1) = r, then
f (x) = f (x·1) = x·f (1) = xr.
Since f is an O-homomorphism, it follows that for all p, x ∈ O,
(p·x)r = f (p·x) = p·f (x) = p·(xr).
This yields [p, x, r] = 0 for all p, x ∈ O, consequently r ∈ R as desired. Finally, it follows from
Schur's Lemma that
HomO(O, O) ∼= HomCℓ7(Cℓ7 O, Cℓ7 O) = {0}.
This completes the proof.
At last, we point out another relation between Spin(7) and octonions. More precisely, we can
provide a realization of spinor space over Spin(7) in terms of octonions O.
Here we refer to the concept of (Weyl) spinor spaces as the irreducible complex representations
of Spin(p, q); see [14, Chap. I sec. 4]. By definition,
Spin(7) = {v(x1)v(x2)··· v(x2k) : xi ∈ R7,xi = 1, i = 1, . . . , 2k}
It is well-known that
where Cℓ+
7 is R-linearly generated by {gα α ∈ P(n),α = 2k}.
Let Cl7 = Cℓ7 ⊗ C be the complexification of Cℓ7. It is a complex Clifford algebra and there
exists a C-algebra isomorphism
Spin(7) ⊂ Cℓ+
7 ,
Notice that Cl7
Cl7 ∼= M (8, C) ⊕ M (8, C).
+ is C-linearly generated by Spin(7) and
+ ∼= Cl6 ∼= M (8, C)
Cl7
12
+ is a simple algebra. Therefore, Spin(7) has only one irreducible
as C-algebra. This means that Cl7
representation. We denote by S6 the irreducible representation of Spin(7). It is also a Cl6-module.
As seen before, O has a Cℓ7-module structure. Thanks to
Le1 ··· Le7 = −Id
(4.4)
it admits a Cℓ6-module structure Cℓ6 O, too. Indeed, let A be the algebra generated by {Lei i =
1, . . . , 7}. In virtue of (4.4), we know that A is not the universal Clifford algebra over R7, so that
it is isomorphic to Cℓ6; see [7]. Consequently, we have a ring homomorphism
Cℓ6 ∼= A ֒→ EndR(O),
which provides a Cℓ6-module structure Cℓ6 O on O.
As a result, Cℓ6 O ⊗ C is a simple Cl6-module, which gives the realization of spinor space over
Spin(7).
4.2 Structure of general left O-modules
In this subsection, we proceed to study the characterization of the general left O-modules. We first
introduce the notion of basis on O-modules.
Definition 4.4. Let M be a left O-module. A subset S ⊆ M is called a basis if S is O-linearly
independent and M = OS.
Theorem 4.5. Each left O-module M has a basis S included by A (M ) ∪ A −(M ). In particular,
M = OA (M ) ⊕ OA −(M ).
Moreover, let Λ1, Λ2 be two index sets satisfying Λ1 = S ∩ A (M ), Λ2 = S ∩ A −(M ), here
S stands for the cardinality of S, then M ∼= (⊕i∈Λ1 O)L(⊕i∈Λ2 O).
Lemma 4.6. Let M be a left O-module, then hmiO is finite dimensional for any m ∈ M . More
precisey, the dimension is at most 128.
Its proof will depend on the following lemma.
Proof. hmiO is such module generated by ei1(ei2 (··· (ein m))), where ik ∈ {1, 2, . . . , 7}, n ∈ N. Note
that
ei(ejm) + ej(eim) = (eiej + ejei)m = −2δijm,
, e7m, e1(e2)m,
7 + C 1
7 + ··· + C 7
hence the element defined by ei1 (ei2 (··· (ein m))) for n > 7 can be reduced. Thus the vectors
{m, e1m,
, e1(e2(··· (e7m)))} will generate hmiO, we conclude that
. . .
dimR hmiO 6 C 0
Remark 4.7. In fact, this property has already appeared in [8]. However, it is worth stressing the
essentiality of this property. It enables us to characterize the structure of general left O-modules in
terms of finite case.
. . .
7 = 128.
13
Proof of Theorem 4.5. For the case dimR M < ∞, by Corallory 4.2, M ∼= On ⊕ O
for some
nonnegative integers (n, n′). It's easy to see that there is a basis {ǫi}n+n′
i=1 ⊆ A (M ) ∪ A −(M ).
As for general case, let S + be a basis of the real vector space A (M ) and S− a basis of A −(M ),
It follows from Lemma 3.8 that A (M ) ∩ A −(M ) = {0}, S is
let S := S + ∪ S− ⊆ C (M ).
clearly R-linearly independent. In view of Lemma 3.10 and Lemma 3.13, we conclude that S is
also O-linearly independent. We next show that OS = M . If not, there exists a nonzero element
i=1 ⊆ A (hmiO)∪ A −(hmiO) ⊆
m ∈ M \ OS. It follows by Lemma 4.6 that hmiO has a basis {xi(m)}n
A (M ) ∪ A −(M ), hence we can assume
n′
m =
n
Xi=1
rixi(m),
ri ∈ O.
Suppose xi(m) =Pj rij sj, sj ∈ S, and it follows from Lemma 3.12 and Lemma 3.13 that rij ∈ R,
then we obtain
which contradicts our assumption.
m =X(ririj )sj,
ririj ∈ O, sj ∈ S,
Let N denote the O-module (⊕i∈Λ1 O)L(⊕i∈Λ2 O). Canonically we can choose a basis {ǫi}i∈Λ1∪Λ2
and satisfies ǫi ∈ A (N ) when i ∈ Λ1, ǫj ∈ A −(N ) when j ∈ Λ2. Let S = {si}i∈Λ1∪Λ2 be a basis of
M satisfying the hypothesis in theorem. We define:
Then for any p ∈ O,
f p XΛ1∪Λ2
p(risi)!
(rip)si!
p(risi) +XΛ2
(pri)si +XΛ2
f : M → N, X risi 7→X riǫi.
risi! = f XΛ1
= f XΛ1
=XΛ1
=XΛ1
= pf XΛ1∪Λ2
(pri)ǫi +XΛ2
p(riǫi) +XΛ2
risi!
(rip)ǫi
p(riǫi)
This shows f ∈ HomO(M, N ), similarly we can also define g ∈ HomO(N, M ) such that
f g = IdN , gf = IdM .
Hence we get the conclusion as desired.
Remark 4.8. In the study of octonion Hilbert space and Banach space, it heavily depends on the
direct sum structure of the space under considered sometimes, which always brings the question back
14
to the classic situation. For example, in the proof of Hanh-Banach Theorem in [12, Theorem 2.4.1],
it declares that every O-vector space is of the following form:
X = X0 ⊕ X1e1 ⊕ ··· ⊕ X7e7.
Note that therein the definition of O-vector space is actually a left O-module with an irrelevant
right O-module structure. We thus can only consider the left O-module structure of it. In view of
Theorem 4.5, of course the assertion does not always work.
4.3 Cyclic elements in left O-module
Let M be a left O-module throughout this subsetion. The submodule hmiO generated by one point
will be very different from the classical case. As is known that Om is not always a submodule
(see Example 4.13, Example 4.14). We introduce the notion of cyclic element, which generates a
simple submodule, to describe this phenomenon. It turns out that every element is a real linear
combination of cyclic elements, although the quantity of cyclic elements is much less than others.
Definition 4.9. An element m ∈ M is said to be cyclic if hmiO = Om. Denote by C (M ) the set
of all cyclic elements in M .
Proposition 4.10. An element m is cyclic if and only if for all r, p ∈ O, there exists q ∈ O, such
that [r, p, m] = qm.
Proof. Assume m satisfies the hypothesis above, then it's easy to check that Om is a submodule,
that is, m is cyclic. Assume m is cyclic, thus for all r, p ∈ O, r(pm) ∈ Om and hence there exists
s ∈ O, such that r(pm) = sm, then [r, p, m] = (rp − s)m. This proves the lemma.
Our first observation is the following lemma which is useful sometimes.
Lemma 4.11. Let M be a left O-module, then Sp∈O p · A (M ) ⊆ C (M ).
Proof. Let 0 6= x ∈ A (M ), we want to prove that px ∈ C (M ). To see this, take r, s ∈ O arbitrarily,
then using Lemma 3.2, we obtain:
[r, s, px] = [r, s, p]x
= [r, s, p]((p−1p)x)
= ([r, s, p]p−1)(px) − [[r, s, p], p−1, px]
= ([r, s, p]p−1)(px) − [[r, s, p], p−1, p]x
= ([r, s, p]p−1)(px)
In view of Proposition 4.10, we thus get px ∈ C (M ) as desired.
Remark 4.12. Note that the set Sp∈O p · A (M ) is not OA (M ). Actually,
pixi pi ∈ O, xi ∈ A (M ), n ∈ N) .
OA (M ) =( n
Xi=1
In fact, it is easy to know the sum of two cyclic elements may be not a cyclic element anymore.
15
Now let us first consider the cyclic elements in case of finite dimensional O-modules. It had
been seen that (e1, e2, 0) will generat a 16 dimensional real space and (e1, e2, e3) will generat a 24
dimensional real space, which means that both are not cyclic elements (see [8]). In fact, this is a
general phenomenon. In case M = O2, we actually have:
Example 4.13. Suppose x = (x1, x2) ∈ M , then hxiO = M if and only if x1, x2 are real linearly
indenpendent.
In particular, in view of the structure of finite dimensional O-modules (see Corollary 4.2), we
obtain:
(ii). An element in O2 generates the whole space if and only if it is not cyclic.
(i). C (O2) =Sp∈O p · R2.
Proof. Let x = (x1, x2) /∈ C (M ), it is easy to see x1, x2 are real linearly indenpendent. Indeed,
if x1 = rx2 for some r ∈ R, then x = x2(r, 1). Note that (r, 1) is an associative element of O2,
it follows from Lemma 4.11 that x ∈ C (M ), a contradiction. Hence both x1, x2 are not zero and
(x1)−1x2 /∈ R. Therefore,
(x1)−1x = (1, (x1)−1x2), (x1)−1x2 /∈ R.
Thus we can choose p, q ∈ O, such that [p, q, (x1)−1x2] 6= 0 (if not, we would have [p, q, (x1)−1x2] = 0
for any p, q, which means (x1)−1x2 ∈ R). However,
[p, q, (x1)−1x] =(cid:0)[p, q, 1], [p, q, (x1)−1x2](cid:1) =(cid:0)0, [p, q, (x1)−1x2](cid:1) ∈ hxiO ,
we thus obtain
Similar arguments apply to (x2)−1x, we get the conclusion as desired.
{(0, p) p ∈ O} ⊆ hxiO .
2
An analogous statement holds for O
. We next consider the case M = O ⊕ O.
Example 4.14. h(1, 1)iO = O ⊕ O
Since
e1(1, 1) = (e1,−e1),
e2(e3(1, 1)) = e2(e3,−e3) = (e1, e1),
we conclude that both (e1, 0) and (0, e1) lie in h(1, 1)iO, which yields that the submodules {(0, p)
p ∈ O} and {(p, 0) p ∈ O} both lie in h(1, 1)iO, too. Therefore h(1, 1)iO = O ⊕ O.
In fact, Lemma 4.11, Example 4.13 and Example 4.14 are all specific to a general result that
will be proved later. The following lemma is crucial to set up this result.
Lemma 4.15. Let x be any given nonzero element in M . Then
x ∈ C (M ) ⇐⇒ dimR hxiO = 8 ⇐⇒ hxiO ∼= O or O.
Proof. Let x ∈ C (M ), then hxiO = Ox and hence dimR hxiO 6 8. On the other hand, since hxiO is
a nonzero O-module of finite dimension, thus dimR hxiO > 8, therefore dimR hxiO = 8, this means
hxiO is a simple O-module and hence hxiO ∼= O or O. Suppose hxiO ∼= O or O. Assume hxiO ∼= O
16
first. Let ϕ denote an isomorphism: ϕ : hxiO → O. For any m ∈ hxiO, suppose ϕ(x) = p, ϕ(m) = q.
Then
ϕ(m) = q = qp−1p = qp−1ϕ(x) = ϕ((qp−1)x),
according to that ϕ is isomorphism, we get m = (qp−1)x, hence hxiO = Ox which means x ∈ C (M ).
If hxiO ∼= O, still let ϕ denote the isomorphism: ϕ : hxiO → O. For any m ∈ hxiO, suppose
ϕ(x) = p, ϕ(m) = q. Then
ϕ(m) = q = (qp−1)·p = (qp−1)·ϕ(x) = ϕ((qp−1)x),
then we get m = (qp−1)x, hence x ∈ C (M ).
According to above lemma, we define
C +(M ) := {x ∈ C (M ) hxiO ∼= O} ∪ {0},
C −(M ) := {x ∈ C (M ) hxiO ∼= O} ∪ {0}.
Therefore C (M ) = C +(M ) ∪ C −(M ). We shall show that all the cyclic elements are determined
by the associative subset A (M ) and the conjugate associative subset A −(M ).
Theorem 4.16. Let M be a left O-module, then:
(i). C +(M ) =Sp∈O p · A (M );
(ii). C −(M ) =Sp∈O p · A −(M ).
Proof. We prove assertion (i). We first show Sp∈O p · A (M ) ⊆ C +(M ). Given any x ∈ A (M ).
Without loss of generality we can assume x 6= 0. Define a map φ : hxiO → O such that φ(px) = p
for p ∈ O. This is a homomorphism in HomO(hxiO , O), since
φ(q(px)) = φ((qp)x) = qp = qφ(px).
Define ϕ : O → hxiO by ϕ(p) = px. Then
ϕ(pq) = (pq)x = p(qx) = pϕ(q).
Hence ϕ ∈ HomO(O,hxiO) and φϕ = id, ϕφ = id and thus hxiO ∼= O. This proves x ∈ C +(M ).
Because px ∈ hxiO and x = p−1(px) ∈ hpxiO for p 6= 0, that is, hxiO = hpxiO whenever p 6= 0.
This implies Sp∈O p · A (M ) ⊆ C +(M ). On the contary, let 0 6= x ∈ C +(M ), hence there is an
isomorphism φ ∈ HomO(O,hxiO). Suppose φ(1) = y ∈ hxiO, since φ is an isomorphism, there is
0 6= r ∈ O such that y = rx. However in view of Proposition 3.3, y = φ(1) ∈ A (hxiO) ⊆ A (M ),
thus x = r−1y ∈Sp∈O p · A (M ). This proves assertion (i).
We prove assertion (ii). Easy to show that hxiO = Ox for x ∈ A −(M ). Let x ∈ A −(M ).
Without loss of generality we can assume x 6= 0. Define a map φ : hxiO → O such that φ(px) = p
for p ∈ O. This is a homomorphism in HomO(hxiO , O), since
φ(q(px)) = φ((pq)x) = pq = q·p = qφ(px).
17
Define ϕ : O → hxiO by ϕ(p) = px. Then
ϕ(p·q) = ϕ(pq) = (pq)x = p(qx) = pϕ(q).
Hence ϕ ∈ HomO(O,hxiO) and φϕ = id, ϕφ = id and thus hxiO ∼= O. This proves x ∈ C −(M ).
Hence we conclude from the fact hxiO = hpxiO that Sp∈O p · A −(M ) ⊆ C −(M ). On the contary,
let 0 6= x ∈ C −(M ), hence there is an isomorphism φ ∈ HomO(O,hxiO). Suppose φ(1) = y ∈ hxiO,
since φ is an isomorphism, there is 0 6= r ∈ O such that y = rx. Note that for any p, q ∈ O,
(pq)y = (pq)φ(1) = φ((pq)·1) = φ(q·p) = qφ(p·1) = q(py).
This shows y ∈ A −(M ) and thus x = r−1y ∈Sp∈O p · A −(M ). This proves assertion (ii).
In view of Theorem 4.5, we conclude an important consequence of the above Theorem:
Corollary 4.17. For each left O-module M , we have
M = SpanRC +(M ) ⊕ SpanRC −(M ) = SpanRC (M ).
Remark 4.18. The corollary shows that for any element m ∈ M , there exist some real linear
independent cyclic elements mi ∈ C (M ), i = 1, . . . , n, such that
m =
mi.
n
Xi=1
choose m1 = (1, 0, 0), m2 = (0, 1 + e1, 0), m3 = (0, 0, e1) in C (M ), it clearly holds m =P3
However, this decomposition is not unique. For example, let M = O3 and m = (1, 1 + e1, e1). Then
i=1 mi and
they are real linear independent. On the other hand, we can choose m′
2 = e1(0, 1, 1)
in C (M ), this is also a decomposition of m. It is worth noticing the length of a decomposition of
m must be no less than 2 and no more than 3. Loosely speaking, the length of the decomposition
of one element reflects the size of the submodule generated by it. The accurate relation deserves
further study.
1 = (1, 1, 0), m′
For any element m in a left O-module M , it follows from Corollary 4.17 that there exist m± ∈
SpanRC ±(M ) such that m = m+ + m−. We can decomposite m± into a combination of real linearly
independent cyclic elements. Denote by l±
m the minimal length of the decompositions of m±. We
conjecture that,
hmiO ∼= Ol+
m ⊕ Ol−
m.
It holds true in above example. In fact, in a similar manner as in Example 4.13, we can prove
h(1, 1 + e1, e1)iO = O · (1, 1, 0) ⊕ O · e1(0, 1, 1) ∼= O2.
If the conjecture is right, then the structure of the submodule generated by one element is completely
clear.
18
References
[1] J. L. Alperin and Rowen B. Bell. Groups and representations, volume 162 of Graduate Texts
in Mathematics. Springer-Verlag, New York, 1995.
[2] M. F. Atiyah, R. Bott, and A. Shapiro. Clifford modules. Topology, 3(suppl, suppl. 1):3 -- 38,
1964.
[3] John C. Baez. The octonions. Bull. Amer. Math. Soc. (N.S.), 39(2):145 -- 205, 2002.
[4] Fabrizio Colombo, Irene Sabadini, and Daniele C. Struppa. Noncommutative functional cal-
culus, volume 289 of Progress in Mathematics. Birkhauser/Springer Basel AG, Basel, 2011.
Theory and applications of slice hyperholomorphic functions.
[5] Samuel Eilenberg. Extensions of general algebras. Ann. Soc. Polon. Math., 21:125 -- 134, 1948.
[6] Riccardo Ghiloni, Valter Moretti, and Alessandro Perotti. Continuous slice functional calculus
in quaternionic Hilbert spaces. Rev. Math. Phys., 25(4):1350006, 83, 2013.
[7] John E. Gilbert and Margaret A. M. Murray. Clifford algebras and Dirac operators in harmonic
analysis, volume 26 of Cambridge Studies in Advanced Mathematics. Cambridge University
Press, Cambridge, 1991.
[8] H. H. Goldstine and L. P. Horwitz. Hilbert space with non-associative scalars. I. Math. Ann.,
154:1 -- 27, 1964.
[9] F. Reese Harvey. Spinors and calibrations, volume 9 of Perspectives in Mathematics. Academic
Press, Inc., Boston, MA, 1990.
[10] L. P. Horwitz and A. Razon. Tensor product of quaternion Hilbert modules. In Classical and
quantum systems (Goslar, 1991), pages 266 -- 268. World Sci. Publ., River Edge, NJ, 1993.
[11] N. Jacobson. Structure of alternative and Jordan bimodules. Osaka Math. J., 6:1 -- 71, 1954.
[12] S. V. Ludkovsky. Algebras of operators in Banach spaces over the quaternion skew field and
the octonion algebra. Sovrem. Mat. Prilozh., (35):98 -- 162, 2005.
[13] S. V. Ludkovsky and W. Sprossig. Spectral representations of operators in Hilbert spaces over
quaternions and octonions. Complex Var. Elliptic Equ., 57(12):1301 -- 1324, 2012.
[14] Alan McIntosh. Book Review: Clifford algebra and spinor-valued functions, a function theory
for the Dirac operator. Bull. Amer. Math. Soc. (N.S.), 32(3):344 -- 348, 1995.
[15] Chi-Keung Ng. On quaternionic functional analysis. Math. Proc. Cambridge Philos. Soc.,
143(2):391 -- 406, 2007.
[16] A. Razon and L. P. Horwitz. Uniqueness of the scalar product in the tensor product of
quaternion Hilbert modules. J. Math. Phys., 33(9):3098 -- 3104, 1992.
[17] Aharon Razon and L. P. Horwitz. Projection operators and states in the tensor product of
quaternion Hilbert modules. Acta Appl. Math., 24(2):179 -- 194, 1991.
19
[18] Joseph J. Rotman. Advanced modern algebra. Part 2, volume 180 of Graduate Studies in
Mathematics. American Mathematical Society, Providence, RI, third edition, 2017. With a
foreword by Bruce Reznick.
[19] Richard D. Schafer. An introduction to nonassociative algebras. Dover Publications, Inc., New
York, 1995. Corrected reprint of the 1966 original.
[20] A. Soffer and L. P. Horwitz. B∗-algebra representations in a quaternionic Hilbert module. J.
Math. Phys., 24(12):2780 -- 2782, 1983.
[21] K. Viswanath. Normal operations on quaternionic Hilbert spaces. Trans. Amer. Math. Soc.,
162:337 -- 350, 1971.
[22] Haiyan Wang and Guangbin Ren. Octonion analysis of several variables. Commun. Math.
Stat., 2(2):163 -- 185, 2014.
20
|
Subsets and Splits
No community queries yet
The top public SQL queries from the community will appear here once available.